Merge git://github.com/Paragon-Software-Group/linux-ntfs3

Merge NTFSv3 filesystem from Konstantin Komarov:
 "This patch adds NTFS Read-Write driver to fs/ntfs3.

  Having decades of expertise in commercial file systems development and
  huge test coverage, we at Paragon Software GmbH want to make our
  contribution to the Open Source Community by providing implementation
  of NTFS Read-Write driver for the Linux Kernel.

  This is fully functional NTFS Read-Write driver. Current version works
  with NTFS (including v3.1) and normal/compressed/sparse files and
  supports journal replaying.

  We plan to support this version after the codebase once merged, and
  add new features and fix bugs. For example, full journaling support
  over JBD will be added in later updates"

Link: https://lore.kernel.org/lkml/20210729134943.778917-1-almaz.alexandrovich@paragon-software.com/
Link: https://lore.kernel.org/lkml/aa4aa155-b9b2-9099-b7a2-349d8d9d8fbd@paragon-software.com/

* git://github.com/Paragon-Software-Group/linux-ntfs3: (35 commits)
  fs/ntfs3: Change how module init/info messages are displayed
  fs/ntfs3: Remove GPL boilerplates from decompress lib files
  fs/ntfs3: Remove unnecessary condition checking from ntfs_file_read_iter
  fs/ntfs3: Fix integer overflow in ni_fiemap with fiemap_prep()
  fs/ntfs3: Restyle comments to better align with kernel-doc
  fs/ntfs3: Rework file operations
  fs/ntfs3: Remove fat ioctl's from ntfs3 driver for now
  fs/ntfs3: Restyle comments to better align with kernel-doc
  fs/ntfs3: Fix error handling in indx_insert_into_root()
  fs/ntfs3: Potential NULL dereference in hdr_find_split()
  fs/ntfs3: Fix error code in indx_add_allocate()
  fs/ntfs3: fix an error code in ntfs_get_acl_ex()
  fs/ntfs3: add checks for allocation failure
  fs/ntfs3: Use kcalloc/kmalloc_array over kzalloc/kmalloc
  fs/ntfs3: Do not use driver own alloc wrappers
  fs/ntfs3: Use kernel ALIGN macros over driver specific
  fs/ntfs3: Restyle comment block in ni_parse_reparse()
  fs/ntfs3: Remove unused including <linux/version.h>
  fs/ntfs3: Fix fall-through warnings for Clang
  fs/ntfs3: Fix one none utf8 char in source file
  ...

33 files changed, 31018 insertions, 0 deletions

diff --git a/Documentation/filesystems/index.rst b/Documentation/filesystems/index.rst
index 1a2dd4d35717..c0ad233963ae 100644
--- a/Documentation/filesystems/index.rst
+++ b/Documentation/filesystems/index.rst
@@ -101,6 +101,7 @@ Documentation for filesystem implementations.
    nilfs2
    nfs/index
    ntfs
+   ntfs3
    ocfs2
    ocfs2-online-filecheck
    omfs
diff --git a/Documentation/filesystems/ntfs3.rst b/Documentation/filesystems/ntfs3.rst
new file mode 100644
index 000000000000..ffe9ea0c1499
--- /dev/null
+++ b/Documentation/filesystems/ntfs3.rst
@@ -0,0 +1,106 @@
+.. SPDX-License-Identifier: GPL-2.0
+
+=====
+NTFS3
+=====
+
+
+Summary and Features
+====================
+
+NTFS3 is fully functional NTFS Read-Write driver. The driver works with
+NTFS versions up to 3.1, normal/compressed/sparse files
+and journal replaying. File system type to use on mount is 'ntfs3'.
+
+- This driver implements NTFS read/write support for normal, sparse and
+  compressed files.
+- Supports native journal replaying;
+- Supports extended attributes
+	Predefined extended attributes:
+	- 'system.ntfs_security' gets/sets security
+			descriptor (SECURITY_DESCRIPTOR_RELATIVE)
+	- 'system.ntfs_attrib' gets/sets ntfs file/dir attributes.
+		Note: applied to empty files, this allows to switch type between
+		sparse(0x200), compressed(0x800) and normal;
+- Supports NFS export of mounted NTFS volumes.
+
+Mount Options
+=============
+
+The list below describes mount options supported by NTFS3 driver in addition to
+generic ones.
+
+===============================================================================
+
+nls=name		This option informs the driver how to interpret path
+			strings and translate them to Unicode and back. If
+			this option is not set, the default codepage will be
+			used (CONFIG_NLS_DEFAULT).
+			Examples:
+				'nls=utf8'
+
+uid=
+gid=
+umask=			Controls the default permissions for files/directories created
+			after the NTFS volume is mounted.
+
+fmask=
+dmask=			Instead of specifying umask which applies both to
+			files and directories, fmask applies only to files and
+			dmask only to directories.
+
+nohidden		Files with the Windows-specific HIDDEN (FILE_ATTRIBUTE_HIDDEN)
+			attribute will not be shown under Linux.
+
+sys_immutable		Files with the Windows-specific SYSTEM
+			(FILE_ATTRIBUTE_SYSTEM) attribute will be marked as system
+			immutable files.
+
+discard			Enable support of the TRIM command for improved performance
+			on delete operations, which is recommended for use with the
+			solid-state drives (SSD).
+
+force			Forces the driver to mount partitions even if 'dirty' flag
+			(volume dirty) is set. Not recommended for use.
+
+sparse			Create new files as "sparse".
+
+showmeta		Use this parameter to show all meta-files (System Files) on
+			a mounted NTFS partition.
+			By default, all meta-files are hidden.
+
+prealloc		Preallocate space for files excessively when file size is
+			increasing on writes. Decreases fragmentation in case of
+			parallel write operations to different files.
+
+no_acs_rules		"No access rules" mount option sets access rights for
+			files/folders to 777 and owner/group to root. This mount
+			option absorbs all other permissions:
+			- permissions change for files/folders will be reported
+				as successful, but they will remain 777;
+			- owner/group change will be reported as successful, but
+				they will stay as root
+
+acl			Support POSIX ACLs (Access Control Lists). Effective if
+			supported by Kernel. Not to be confused with NTFS ACLs.
+			The option specified as acl enables support for POSIX ACLs.
+
+noatime			All files and directories will not update their last access
+			time attribute if a partition is mounted with this parameter.
+			This option can speed up file system operation.
+
+===============================================================================
+
+ToDo list
+=========
+
+- Full journaling support (currently journal replaying is supported) over JBD.
+
+
+References
+==========
+https://www.paragon-software.com/home/ntfs-linux-professional/
+	- Commercial version of the NTFS driver for Linux.
+
+almaz.alexandrovich@paragon-software.com
+	- Direct e-mail address for feedback and requests on the NTFS3 implementation.
diff --git a/MAINTAINERS b/MAINTAINERS
index 5ffe43730e43..e4bdea045e85 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -13340,6 +13340,15 @@ T:	git git://git.kernel.org/pub/scm/linux/kernel/git/aia21/ntfs.git
 F:	Documentation/filesystems/ntfs.rst
 F:	fs/ntfs/
 
+NTFS3 FILESYSTEM
+M:	Konstantin Komarov <almaz.alexandrovich@paragon-software.com>
+L:	ntfs3@lists.linux.dev
+S:	Supported
+W:	http://www.paragon-software.com/
+T:	git https://github.com/Paragon-Software-Group/linux-ntfs3.git
+F:	Documentation/filesystems/ntfs3.rst
+F:	fs/ntfs3/
+
 NUBUS SUBSYSTEM
 M:	Finn Thain <fthain@linux-m68k.org>
 L:	linux-m68k@lists.linux-m68k.org
diff --git a/fs/Kconfig b/fs/Kconfig
index b11bd4b387e1..d8207a1b8c44 100644
--- a/fs/Kconfig
+++ b/fs/Kconfig
@@ -136,6 +136,7 @@ menu "DOS/FAT/EXFAT/NT Filesystems"
 source "fs/fat/Kconfig"
 source "fs/exfat/Kconfig"
 source "fs/ntfs/Kconfig"
+source "fs/ntfs3/Kconfig"
 
 endmenu
 endif # BLOCK
diff --git a/fs/Makefile b/fs/Makefile
index 354e2ba3ee67..2f21300851ae 100644
--- a/fs/Makefile
+++ b/fs/Makefile
@@ -101,6 +101,7 @@ obj-$(CONFIG_CIFS)		+= cifs/
 obj-$(CONFIG_SMB_SERVER)	+= ksmbd/
 obj-$(CONFIG_HPFS_FS)		+= hpfs/
 obj-$(CONFIG_NTFS_FS)		+= ntfs/
+obj-$(CONFIG_NTFS3_FS)		+= ntfs3/
 obj-$(CONFIG_UFS_FS)		+= ufs/
 obj-$(CONFIG_EFS_FS)		+= efs/
 obj-$(CONFIG_JFFS2_FS)		+= jffs2/
diff --git a/fs/ntfs3/Kconfig b/fs/ntfs3/Kconfig
new file mode 100644
index 000000000000..6e4cbc48ab8e
--- /dev/null
+++ b/fs/ntfs3/Kconfig
@@ -0,0 +1,46 @@
+# SPDX-License-Identifier: GPL-2.0-only
+config NTFS3_FS
+	tristate "NTFS Read-Write file system support"
+	select NLS
+	help
+	  Windows OS native file system (NTFS) support up to NTFS version 3.1.
+
+	  Y or M enables the NTFS3 driver with full features enabled (read,
+	  write, journal replaying, sparse/compressed files support).
+	  File system type to use on mount is "ntfs3". Module name (M option)
+	  is also "ntfs3".
+
+	  Documentation: <file:Documentation/filesystems/ntfs3.rst>
+
+config NTFS3_64BIT_CLUSTER
+	bool "64 bits per NTFS clusters"
+	depends on NTFS3_FS && 64BIT
+	help
+	  Windows implementation of ntfs.sys uses 32 bits per clusters.
+	  If activated 64 bits per clusters you will be able to use 4k cluster
+	  for 16T+ volumes. Windows will not be able to mount such volumes.
+
+	  It is recommended to say N here.
+
+config NTFS3_LZX_XPRESS
+	bool "activate support of external compressions lzx/xpress"
+	depends on NTFS3_FS
+	help
+	  In Windows 10 one can use command "compact" to compress any files.
+	  4 possible variants of compression are: xpress4k, xpress8k, xpress16k and lzx.
+	  If activated you will be able to read such files correctly.
+
+	  It is recommended to say Y here.
+
+config NTFS3_FS_POSIX_ACL
+	bool "NTFS POSIX Access Control Lists"
+	depends on NTFS3_FS
+	select FS_POSIX_ACL
+	help
+	  POSIX Access Control Lists (ACLs) support additional access rights
+	  for users and groups beyond the standard owner/group/world scheme,
+	  and this option selects support for ACLs specifically for ntfs
+	  filesystems.
+	  NOTE: this is linux only feature. Windows will ignore these ACLs.
+
+	  If you don't know what Access Control Lists are, say N.
diff --git a/fs/ntfs3/Makefile b/fs/ntfs3/Makefile
new file mode 100644
index 000000000000..279701b62bbe
--- /dev/null
+++ b/fs/ntfs3/Makefile
@@ -0,0 +1,36 @@
+# SPDX-License-Identifier: GPL-2.0
+#
+# Makefile for the ntfs3 filesystem support.
+#
+
+# to check robot warnings
+ccflags-y += -Wint-to-pointer-cast \
+	$(call cc-option,-Wunused-but-set-variable,-Wunused-const-variable) \
+	$(call cc-option,-Wold-style-declaration,-Wout-of-line-declaration)
+
+obj-$(CONFIG_NTFS3_FS) += ntfs3.o
+
+ntfs3-y :=	attrib.o \
+		attrlist.o \
+		bitfunc.o \
+		bitmap.o \
+		dir.o \
+		fsntfs.o \
+		frecord.o \
+		file.o \
+		fslog.o \
+		inode.o \
+		index.o \
+		lznt.o \
+		namei.o \
+		record.o \
+		run.o \
+		super.o \
+		upcase.o \
+		xattr.o
+
+ntfs3-$(CONFIG_NTFS3_LZX_XPRESS) += $(addprefix lib/,\
+		decompress_common.o \
+		lzx_decompress.o \
+		xpress_decompress.o \
+		)
\ No newline at end of file
diff --git a/fs/ntfs3/attrib.c b/fs/ntfs3/attrib.c
new file mode 100644
index 000000000000..34c4cbf7e29b
--- /dev/null
+++ b/fs/ntfs3/attrib.c
@@ -0,0 +1,2093 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ *
+ * Copyright (C) 2019-2021 Paragon Software GmbH, All rights reserved.
+ *
+ * TODO: Merge attr_set_size/attr_data_get_block/attr_allocate_frame?
+ */
+
+#include <linux/blkdev.h>
+#include <linux/buffer_head.h>
+#include <linux/fs.h>
+#include <linux/hash.h>
+#include <linux/nls.h>
+#include <linux/ratelimit.h>
+#include <linux/slab.h>
+
+#include "debug.h"
+#include "ntfs.h"
+#include "ntfs_fs.h"
+
+/*
+ * You can set external NTFS_MIN_LOG2_OF_CLUMP/NTFS_MAX_LOG2_OF_CLUMP to manage
+ * preallocate algorithm.
+ */
+#ifndef NTFS_MIN_LOG2_OF_CLUMP
+#define NTFS_MIN_LOG2_OF_CLUMP 16
+#endif
+
+#ifndef NTFS_MAX_LOG2_OF_CLUMP
+#define NTFS_MAX_LOG2_OF_CLUMP 26
+#endif
+
+// 16M
+#define NTFS_CLUMP_MIN (1 << (NTFS_MIN_LOG2_OF_CLUMP + 8))
+// 16G
+#define NTFS_CLUMP_MAX (1ull << (NTFS_MAX_LOG2_OF_CLUMP + 8))
+
+static inline u64 get_pre_allocated(u64 size)
+{
+	u32 clump;
+	u8 align_shift;
+	u64 ret;
+
+	if (size <= NTFS_CLUMP_MIN) {
+		clump = 1 << NTFS_MIN_LOG2_OF_CLUMP;
+		align_shift = NTFS_MIN_LOG2_OF_CLUMP;
+	} else if (size >= NTFS_CLUMP_MAX) {
+		clump = 1 << NTFS_MAX_LOG2_OF_CLUMP;
+		align_shift = NTFS_MAX_LOG2_OF_CLUMP;
+	} else {
+		align_shift = NTFS_MIN_LOG2_OF_CLUMP - 1 +
+			      __ffs(size >> (8 + NTFS_MIN_LOG2_OF_CLUMP));
+		clump = 1u << align_shift;
+	}
+
+	ret = (((size + clump - 1) >> align_shift)) << align_shift;
+
+	return ret;
+}
+
+/*
+ * attr_must_be_resident
+ *
+ * Return: True if attribute must be resident.
+ */
+static inline bool attr_must_be_resident(struct ntfs_sb_info *sbi,
+					 enum ATTR_TYPE type)
+{
+	const struct ATTR_DEF_ENTRY *de;
+
+	switch (type) {
+	case ATTR_STD:
+	case ATTR_NAME:
+	case ATTR_ID:
+	case ATTR_LABEL:
+	case ATTR_VOL_INFO:
+	case ATTR_ROOT:
+	case ATTR_EA_INFO:
+		return true;
+	default:
+		de = ntfs_query_def(sbi, type);
+		if (de && (de->flags & NTFS_ATTR_MUST_BE_RESIDENT))
+			return true;
+		return false;
+	}
+}
+
+/*
+ * attr_load_runs - Load all runs stored in @attr.
+ */
+int attr_load_runs(struct ATTRIB *attr, struct ntfs_inode *ni,
+		   struct runs_tree *run, const CLST *vcn)
+{
+	int err;
+	CLST svcn = le64_to_cpu(attr->nres.svcn);
+	CLST evcn = le64_to_cpu(attr->nres.evcn);
+	u32 asize;
+	u16 run_off;
+
+	if (svcn >= evcn + 1 || run_is_mapped_full(run, svcn, evcn))
+		return 0;
+
+	if (vcn && (evcn < *vcn || *vcn < svcn))
+		return -EINVAL;
+
+	asize = le32_to_cpu(attr->size);
+	run_off = le16_to_cpu(attr->nres.run_off);
+	err = run_unpack_ex(run, ni->mi.sbi, ni->mi.rno, svcn, evcn,
+			    vcn ? *vcn : svcn, Add2Ptr(attr, run_off),
+			    asize - run_off);
+	if (err < 0)
+		return err;
+
+	return 0;
+}
+
+/*
+ * run_deallocate_ex - Deallocate clusters.
+ */
+static int run_deallocate_ex(struct ntfs_sb_info *sbi, struct runs_tree *run,
+			     CLST vcn, CLST len, CLST *done, bool trim)
+{
+	int err = 0;
+	CLST vcn_next, vcn0 = vcn, lcn, clen, dn = 0;
+	size_t idx;
+
+	if (!len)
+		goto out;
+
+	if (!run_lookup_entry(run, vcn, &lcn, &clen, &idx)) {
+failed:
+		run_truncate(run, vcn0);
+		err = -EINVAL;
+		goto out;
+	}
+
+	for (;;) {
+		if (clen > len)
+			clen = len;
+
+		if (!clen) {
+			err = -EINVAL;
+			goto out;
+		}
+
+		if (lcn != SPARSE_LCN) {
+			mark_as_free_ex(sbi, lcn, clen, trim);
+			dn += clen;
+		}
+
+		len -= clen;
+		if (!len)
+			break;
+
+		vcn_next = vcn + clen;
+		if (!run_get_entry(run, ++idx, &vcn, &lcn, &clen) ||
+		    vcn != vcn_next) {
+			/* Save memory - don't load entire run. */
+			goto failed;
+		}
+	}
+
+out:
+	if (done)
+		*done += dn;
+
+	return err;
+}
+
+/*
+ * attr_allocate_clusters - Find free space, mark it as used and store in @run.
+ */
+int attr_allocate_clusters(struct ntfs_sb_info *sbi, struct runs_tree *run,
+			   CLST vcn, CLST lcn, CLST len, CLST *pre_alloc,
+			   enum ALLOCATE_OPT opt, CLST *alen, const size_t fr,
+			   CLST *new_lcn)
+{
+	int err;
+	CLST flen, vcn0 = vcn, pre = pre_alloc ? *pre_alloc : 0;
+	struct wnd_bitmap *wnd = &sbi->used.bitmap;
+	size_t cnt = run->count;
+
+	for (;;) {
+		err = ntfs_look_for_free_space(sbi, lcn, len + pre, &lcn, &flen,
+					       opt);
+
+		if (err == -ENOSPC && pre) {
+			pre = 0;
+			if (*pre_alloc)
+				*pre_alloc = 0;
+			continue;
+		}
+
+		if (err)
+			goto out;
+
+		if (new_lcn && vcn == vcn0)
+			*new_lcn = lcn;
+
+		/* Add new fragment into run storage. */
+		if (!run_add_entry(run, vcn, lcn, flen, opt == ALLOCATE_MFT)) {
+			/* Undo last 'ntfs_look_for_free_space' */
+			down_write_nested(&wnd->rw_lock, BITMAP_MUTEX_CLUSTERS);
+			wnd_set_free(wnd, lcn, flen);
+			up_write(&wnd->rw_lock);
+			err = -ENOMEM;
+			goto out;
+		}
+
+		vcn += flen;
+
+		if (flen >= len || opt == ALLOCATE_MFT ||
+		    (fr && run->count - cnt >= fr)) {
+			*alen = vcn - vcn0;
+			return 0;
+		}
+
+		len -= flen;
+	}
+
+out:
+	/* Undo 'ntfs_look_for_free_space' */
+	if (vcn - vcn0) {
+		run_deallocate_ex(sbi, run, vcn0, vcn - vcn0, NULL, false);
+		run_truncate(run, vcn0);
+	}
+
+	return err;
+}
+
+/*
+ * attr_make_nonresident
+ *
+ * If page is not NULL - it is already contains resident data
+ * and locked (called from ni_write_frame()).
+ */
+int attr_make_nonresident(struct ntfs_inode *ni, struct ATTRIB *attr,
+			  struct ATTR_LIST_ENTRY *le, struct mft_inode *mi,
+			  u64 new_size, struct runs_tree *run,
+			  struct ATTRIB **ins_attr, struct page *page)
+{
+	struct ntfs_sb_info *sbi;
+	struct ATTRIB *attr_s;
+	struct MFT_REC *rec;
+	u32 used, asize, rsize, aoff, align;
+	bool is_data;
+	CLST len, alen;
+	char *next;
+	int err;
+
+	if (attr->non_res) {
+		*ins_attr = attr;
+		return 0;
+	}
+
+	sbi = mi->sbi;
+	rec = mi->mrec;
+	attr_s = NULL;
+	used = le32_to_cpu(rec->used);
+	asize = le32_to_cpu(attr->size);
+	next = Add2Ptr(attr, asize);
+	aoff = PtrOffset(rec, attr);
+	rsize = le32_to_cpu(attr->res.data_size);
+	is_data = attr->type == ATTR_DATA && !attr->name_len;
+
+	align = sbi->cluster_size;
+	if (is_attr_compressed(attr))
+		align <<= COMPRESSION_UNIT;
+	len = (rsize + align - 1) >> sbi->cluster_bits;
+
+	run_init(run);
+
+	/* Make a copy of original attribute. */
+	attr_s = kmemdup(attr, asize, GFP_NOFS);
+	if (!attr_s) {
+		err = -ENOMEM;
+		goto out;
+	}
+
+	if (!len) {
+		/* Empty resident -> Empty nonresident. */
+		alen = 0;
+	} else {
+		const char *data = resident_data(attr);
+
+		err = attr_allocate_clusters(sbi, run, 0, 0, len, NULL,
+					     ALLOCATE_DEF, &alen, 0, NULL);
+		if (err)
+			goto out1;
+
+		if (!rsize) {
+			/* Empty resident -> Non empty nonresident. */
+		} else if (!is_data) {
+			err = ntfs_sb_write_run(sbi, run, 0, data, rsize);
+			if (err)
+				goto out2;
+		} else if (!page) {
+			char *kaddr;
+
+			page = grab_cache_page(ni->vfs_inode.i_mapping, 0);
+			if (!page) {
+				err = -ENOMEM;
+				goto out2;
+			}
+			kaddr = kmap_atomic(page);
+			memcpy(kaddr, data, rsize);
+			memset(kaddr + rsize, 0, PAGE_SIZE - rsize);
+			kunmap_atomic(kaddr);
+			flush_dcache_page(page);
+			SetPageUptodate(page);
+			set_page_dirty(page);
+			unlock_page(page);
+			put_page(page);
+		}
+	}
+
+	/* Remove original attribute. */
+	used -= asize;
+	memmove(attr, Add2Ptr(attr, asize), used - aoff);
+	rec->used = cpu_to_le32(used);
+	mi->dirty = true;
+	if (le)
+		al_remove_le(ni, le);
+
+	err = ni_insert_nonresident(ni, attr_s->type, attr_name(attr_s),
+				    attr_s->name_len, run, 0, alen,
+				    attr_s->flags, &attr, NULL);
+	if (err)
+		goto out3;
+
+	kfree(attr_s);
+	attr->nres.data_size = cpu_to_le64(rsize);
+	attr->nres.valid_size = attr->nres.data_size;
+
+	*ins_attr = attr;
+
+	if (is_data)
+		ni->ni_flags &= ~NI_FLAG_RESIDENT;
+
+	/* Resident attribute becomes non resident. */
+	return 0;
+
+out3:
+	attr = Add2Ptr(rec, aoff);
+	memmove(next, attr, used - aoff);
+	memcpy(attr, attr_s, asize);
+	rec->used = cpu_to_le32(used + asize);
+	mi->dirty = true;
+out2:
+	/* Undo: do not trim new allocated clusters. */
+	run_deallocate(sbi, run, false);
+	run_close(run);
+out1:
+	kfree(attr_s);
+out:
+	return err;
+}
+
+/*
+ * attr_set_size_res - Helper for attr_set_size().
+ */
+static int attr_set_size_res(struct ntfs_inode *ni, struct ATTRIB *attr,
+			     struct ATTR_LIST_ENTRY *le, struct mft_inode *mi,
+			     u64 new_size, struct runs_tree *run,
+			     struct ATTRIB **ins_attr)
+{
+	struct ntfs_sb_info *sbi = mi->sbi;
+	struct MFT_REC *rec = mi->mrec;
+	u32 used = le32_to_cpu(rec->used);
+	u32 asize = le32_to_cpu(attr->size);
+	u32 aoff = PtrOffset(rec, attr);
+	u32 rsize = le32_to_cpu(attr->res.data_size);
+	u32 tail = used - aoff - asize;
+	char *next = Add2Ptr(attr, asize);
+	s64 dsize = ALIGN(new_size, 8) - ALIGN(rsize, 8);
+
+	if (dsize < 0) {
+		memmove(next + dsize, next, tail);
+	} else if (dsize > 0) {
+		if (used + dsize > sbi->max_bytes_per_attr)
+			return attr_make_nonresident(ni, attr, le, mi, new_size,
+						     run, ins_attr, NULL);
+
+		memmove(next + dsize, next, tail);
+		memset(next, 0, dsize);
+	}
+
+	if (new_size > rsize)
+		memset(Add2Ptr(resident_data(attr), rsize), 0,
+		       new_size - rsize);
+
+	rec->used = cpu_to_le32(used + dsize);
+	attr->size = cpu_to_le32(asize + dsize);
+	attr->res.data_size = cpu_to_le32(new_size);
+	mi->dirty = true;
+	*ins_attr = attr;
+
+	return 0;
+}
+
+/*
+ * attr_set_size - Change the size of attribute.
+ *
+ * Extend:
+ *   - Sparse/compressed: No allocated clusters.
+ *   - Normal: Append allocated and preallocated new clusters.
+ * Shrink:
+ *   - No deallocate if @keep_prealloc is set.
+ */
+int attr_set_size(struct ntfs_inode *ni, enum ATTR_TYPE type,
+		  const __le16 *name, u8 name_len, struct runs_tree *run,
+		  u64 new_size, const u64 *new_valid, bool keep_prealloc,
+		  struct ATTRIB **ret)
+{
+	int err = 0;
+	struct ntfs_sb_info *sbi = ni->mi.sbi;
+	u8 cluster_bits = sbi->cluster_bits;
+	bool is_mft =
+		ni->mi.rno == MFT_REC_MFT && type == ATTR_DATA && !name_len;
+	u64 old_valid, old_size, old_alloc, new_alloc, new_alloc_tmp;
+	struct ATTRIB *attr = NULL, *attr_b;
+	struct ATTR_LIST_ENTRY *le, *le_b;
+	struct mft_inode *mi, *mi_b;
+	CLST alen, vcn, lcn, new_alen, old_alen, svcn, evcn;
+	CLST next_svcn, pre_alloc = -1, done = 0;
+	bool is_ext;
+	u32 align;
+	struct MFT_REC *rec;
+
+again:
+	le_b = NULL;
+	attr_b = ni_find_attr(ni, NULL, &le_b, type, name, name_len, NULL,
+			      &mi_b);
+	if (!attr_b) {
+		err = -ENOENT;
+		goto out;
+	}
+
+	if (!attr_b->non_res) {
+		err = attr_set_size_res(ni, attr_b, le_b, mi_b, new_size, run,
+					&attr_b);
+		if (err || !attr_b->non_res)
+			goto out;
+
+		/* Layout of records may be changed, so do a full search. */
+		goto again;
+	}
+
+	is_ext = is_attr_ext(attr_b);
+
+again_1:
+	align = sbi->cluster_size;
+
+	if (is_ext) {
+		align <<= attr_b->nres.c_unit;
+		if (is_attr_sparsed(attr_b))
+			keep_prealloc = false;
+	}
+
+	old_valid = le64_to_cpu(attr_b->nres.valid_size);
+	old_size = le64_to_cpu(attr_b->nres.data_size);
+	old_alloc = le64_to_cpu(attr_b->nres.alloc_size);
+	old_alen = old_alloc >> cluster_bits;
+
+	new_alloc = (new_size + align - 1) & ~(u64)(align - 1);
+	new_alen = new_alloc >> cluster_bits;
+
+	if (keep_prealloc && is_ext)
+		keep_prealloc = false;
+
+	if (keep_prealloc && new_size < old_size) {
+		attr_b->nres.data_size = cpu_to_le64(new_size);
+		mi_b->dirty = true;
+		goto ok;
+	}
+
+	vcn = old_alen - 1;
+
+	svcn = le64_to_cpu(attr_b->nres.svcn);
+	evcn = le64_to_cpu(attr_b->nres.evcn);
+
+	if (svcn <= vcn && vcn <= evcn) {
+		attr = attr_b;
+		le = le_b;
+		mi = mi_b;
+	} else if (!le_b) {
+		err = -EINVAL;
+		goto out;
+	} else {
+		le = le_b;
+		attr = ni_find_attr(ni, attr_b, &le, type, name, name_len, &vcn,
+				    &mi);
+		if (!attr) {
+			err = -EINVAL;
+			goto out;
+		}
+
+next_le_1:
+		svcn = le64_to_cpu(attr->nres.svcn);
+		evcn = le64_to_cpu(attr->nres.evcn);
+	}
+
+next_le:
+	rec = mi->mrec;
+
+	err = attr_load_runs(attr, ni, run, NULL);
+	if (err)
+		goto out;
+
+	if (new_size > old_size) {
+		CLST to_allocate;
+		size_t free;
+
+		if (new_alloc <= old_alloc) {
+			attr_b->nres.data_size = cpu_to_le64(new_size);
+			mi_b->dirty = true;
+			goto ok;
+		}
+
+		to_allocate = new_alen - old_alen;
+add_alloc_in_same_attr_seg:
+		lcn = 0;
+		if (is_mft) {
+			/* MFT allocates clusters from MFT zone. */
+			pre_alloc = 0;
+		} else if (is_ext) {
+			/* No preallocate for sparse/compress. */
+			pre_alloc = 0;
+		} else if (pre_alloc == -1) {
+			pre_alloc = 0;
+			if (type == ATTR_DATA && !name_len &&
+			    sbi->options.prealloc) {
+				CLST new_alen2 = bytes_to_cluster(
+					sbi, get_pre_allocated(new_size));
+				pre_alloc = new_alen2 - new_alen;
+			}
+
+			/* Get the last LCN to allocate from. */
+			if (old_alen &&
+			    !run_lookup_entry(run, vcn, &lcn, NULL, NULL)) {
+				lcn = SPARSE_LCN;
+			}
+
+			if (lcn == SPARSE_LCN)
+				lcn = 0;
+			else if (lcn)
+				lcn += 1;
+
+			free = wnd_zeroes(&sbi->used.bitmap);
+			if (to_allocate > free) {
+				err = -ENOSPC;
+				goto out;
+			}
+
+			if (pre_alloc && to_allocate + pre_alloc > free)
+				pre_alloc = 0;
+		}
+
+		vcn = old_alen;
+
+		if (is_ext) {
+			if (!run_add_entry(run, vcn, SPARSE_LCN, to_allocate,
+					   false)) {
+				err = -ENOMEM;
+				goto out;
+			}
+			alen = to_allocate;
+		} else {
+			/* ~3 bytes per fragment. */
+			err = attr_allocate_clusters(
+				sbi, run, vcn, lcn, to_allocate, &pre_alloc,
+				is_mft ? ALLOCATE_MFT : 0, &alen,
+				is_mft ? 0
+				       : (sbi->record_size -
+					  le32_to_cpu(rec->used) + 8) /
+							 3 +
+						 1,
+				NULL);
+			if (err)
+				goto out;
+		}
+
+		done += alen;
+		vcn += alen;
+		if (to_allocate > alen)
+			to_allocate -= alen;
+		else
+			to_allocate = 0;
+
+pack_runs:
+		err = mi_pack_runs(mi, attr, run, vcn - svcn);
+		if (err)
+			goto out;
+
+		next_svcn = le64_to_cpu(attr->nres.evcn) + 1;
+		new_alloc_tmp = (u64)next_svcn << cluster_bits;
+		attr_b->nres.alloc_size = cpu_to_le64(new_alloc_tmp);
+		mi_b->dirty = true;
+
+		if (next_svcn >= vcn && !to_allocate) {
+			/* Normal way. Update attribute and exit. */
+			attr_b->nres.data_size = cpu_to_le64(new_size);
+			goto ok;
+		}
+
+		/* At least two MFT to avoid recursive loop. */
+		if (is_mft && next_svcn == vcn &&
+		    ((u64)done << sbi->cluster_bits) >= 2 * sbi->record_size) {
+			new_size = new_alloc_tmp;
+			attr_b->nres.data_size = attr_b->nres.alloc_size;
+			goto ok;
+		}
+
+		if (le32_to_cpu(rec->used) < sbi->record_size) {
+			old_alen = next_svcn;
+			evcn = old_alen - 1;
+			goto add_alloc_in_same_attr_seg;
+		}
+
+		attr_b->nres.data_size = attr_b->nres.alloc_size;
+		if (new_alloc_tmp < old_valid)
+			attr_b->nres.valid_size = attr_b->nres.data_size;
+
+		if (type == ATTR_LIST) {
+			err = ni_expand_list(ni);
+			if (err)
+				goto out;
+			if (next_svcn < vcn)
+				goto pack_runs;
+
+			/* Layout of records is changed. */
+			goto again;
+		}
+
+		if (!ni->attr_list.size) {
+			err = ni_create_attr_list(ni);
+			if (err)
+				goto out;
+			/* Layout of records is changed. */
+		}
+
+		if (next_svcn >= vcn) {
+			/* This is MFT data, repeat. */
+			goto again;
+		}
+
+		/* Insert new attribute segment. */
+		err = ni_insert_nonresident(ni, type, name, name_len, run,
+					    next_svcn, vcn - next_svcn,
+					    attr_b->flags, &attr, &mi);
+		if (err)
+			goto out;
+
+		if (!is_mft)
+			run_truncate_head(run, evcn + 1);
+
+		svcn = le64_to_cpu(attr->nres.svcn);
+		evcn = le64_to_cpu(attr->nres.evcn);
+
+		le_b = NULL;
+		/*
+		 * Layout of records maybe changed.
+		 * Find base attribute to update.
+		 */
+		attr_b = ni_find_attr(ni, NULL, &le_b, type, name, name_len,
+				      NULL, &mi_b);
+		if (!attr_b) {
+			err = -ENOENT;
+			goto out;
+		}
+
+		attr_b->nres.alloc_size = cpu_to_le64((u64)vcn << cluster_bits);
+		attr_b->nres.data_size = attr_b->nres.alloc_size;
+		attr_b->nres.valid_size = attr_b->nres.alloc_size;
+		mi_b->dirty = true;
+		goto again_1;
+	}
+
+	if (new_size != old_size ||
+	    (new_alloc != old_alloc && !keep_prealloc)) {
+		vcn = max(svcn, new_alen);
+		new_alloc_tmp = (u64)vcn << cluster_bits;
+
+		alen = 0;
+		err = run_deallocate_ex(sbi, run, vcn, evcn - vcn + 1, &alen,
+					true);
+		if (err)
+			goto out;
+
+		run_truncate(run, vcn);
+
+		if (vcn > svcn) {
+			err = mi_pack_runs(mi, attr, run, vcn - svcn);
+			if (err)
+				goto out;
+		} else if (le && le->vcn) {
+			u16 le_sz = le16_to_cpu(le->size);
+
+			/*
+			 * NOTE: List entries for one attribute are always
+			 * the same size. We deal with last entry (vcn==0)
+			 * and it is not first in entries array
+			 * (list entry for std attribute always first).
+			 * So it is safe to step back.
+			 */
+			mi_remove_attr(NULL, mi, attr);
+
+			if (!al_remove_le(ni, le)) {
+				err = -EINVAL;
+				goto out;
+			}
+
+			le = (struct ATTR_LIST_ENTRY *)((u8 *)le - le_sz);
+		} else {
+			attr->nres.evcn = cpu_to_le64((u64)vcn - 1);
+			mi->dirty = true;
+		}
+
+		attr_b->nres.alloc_size = cpu_to_le64(new_alloc_tmp);
+
+		if (vcn == new_alen) {
+			attr_b->nres.data_size = cpu_to_le64(new_size);
+			if (new_size < old_valid)
+				attr_b->nres.valid_size =
+					attr_b->nres.data_size;
+		} else {
+			if (new_alloc_tmp <=
+			    le64_to_cpu(attr_b->nres.data_size))
+				attr_b->nres.data_size =
+					attr_b->nres.alloc_size;
+			if (new_alloc_tmp <
+			    le64_to_cpu(attr_b->nres.valid_size))
+				attr_b->nres.valid_size =
+					attr_b->nres.alloc_size;
+		}
+
+		if (is_ext)
+			le64_sub_cpu(&attr_b->nres.total_size,
+				     ((u64)alen << cluster_bits));
+
+		mi_b->dirty = true;
+
+		if (new_alloc_tmp <= new_alloc)
+			goto ok;
+
+		old_size = new_alloc_tmp;
+		vcn = svcn - 1;
+
+		if (le == le_b) {
+			attr = attr_b;
+			mi = mi_b;
+			evcn = svcn - 1;
+			svcn = 0;
+			goto next_le;
+		}
+
+		if (le->type != type || le->name_len != name_len ||
+		    memcmp(le_name(le), name, name_len * sizeof(short))) {
+			err = -EINVAL;
+			goto out;
+		}
+
+		err = ni_load_mi(ni, le, &mi);
+		if (err)
+			goto out;
+
+		attr = mi_find_attr(mi, NULL, type, name, name_len, &le->id);
+		if (!attr) {
+			err = -EINVAL;
+			goto out;
+		}
+		goto next_le_1;
+	}
+
+ok:
+	if (new_valid) {
+		__le64 valid = cpu_to_le64(min(*new_valid, new_size));
+
+		if (attr_b->nres.valid_size != valid) {
+			attr_b->nres.valid_size = valid;
+			mi_b->dirty = true;
+		}
+	}
+
+out:
+	if (!err && attr_b && ret)
+		*ret = attr_b;
+
+	/* Update inode_set_bytes. */
+	if (!err && ((type == ATTR_DATA && !name_len) ||
+		     (type == ATTR_ALLOC && name == I30_NAME))) {
+		bool dirty = false;
+
+		if (ni->vfs_inode.i_size != new_size) {
+			ni->vfs_inode.i_size = new_size;
+			dirty = true;
+		}
+
+		if (attr_b && attr_b->non_res) {
+			new_alloc = le64_to_cpu(attr_b->nres.alloc_size);
+			if (inode_get_bytes(&ni->vfs_inode) != new_alloc) {
+				inode_set_bytes(&ni->vfs_inode, new_alloc);
+				dirty = true;
+			}
+		}
+
+		if (dirty) {
+			ni->ni_flags |= NI_FLAG_UPDATE_PARENT;
+			mark_inode_dirty(&ni->vfs_inode);
+		}
+	}
+
+	return err;
+}
+
+int attr_data_get_block(struct ntfs_inode *ni, CLST vcn, CLST clen, CLST *lcn,
+			CLST *len, bool *new)
+{
+	int err = 0;
+	struct runs_tree *run = &ni->file.run;
+	struct ntfs_sb_info *sbi;
+	u8 cluster_bits;
+	struct ATTRIB *attr = NULL, *attr_b;
+	struct ATTR_LIST_ENTRY *le, *le_b;
+	struct mft_inode *mi, *mi_b;
+	CLST hint, svcn, to_alloc, evcn1, next_svcn, asize, end;
+	u64 total_size;
+	u32 clst_per_frame;
+	bool ok;
+
+	if (new)
+		*new = false;
+
+	down_read(&ni->file.run_lock);
+	ok = run_lookup_entry(run, vcn, lcn, len, NULL);
+	up_read(&ni->file.run_lock);
+
+	if (ok && (*lcn != SPARSE_LCN || !new)) {
+		/* Normal way. */
+		return 0;
+	}
+
+	if (!clen)
+		clen = 1;
+
+	if (ok && clen > *len)
+		clen = *len;
+
+	sbi = ni->mi.sbi;
+	cluster_bits = sbi->cluster_bits;
+
+	ni_lock(ni);
+	down_write(&ni->file.run_lock);
+
+	le_b = NULL;
+	attr_b = ni_find_attr(ni, NULL, &le_b, ATTR_DATA, NULL, 0, NULL, &mi_b);
+	if (!attr_b) {
+		err = -ENOENT;
+		goto out;
+	}
+
+	if (!attr_b->non_res) {
+		*lcn = RESIDENT_LCN;
+		*len = 1;
+		goto out;
+	}
+
+	asize = le64_to_cpu(attr_b->nres.alloc_size) >> sbi->cluster_bits;
+	if (vcn >= asize) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	clst_per_frame = 1u << attr_b->nres.c_unit;
+	to_alloc = (clen + clst_per_frame - 1) & ~(clst_per_frame - 1);
+
+	if (vcn + to_alloc > asize)
+		to_alloc = asize - vcn;
+
+	svcn = le64_to_cpu(attr_b->nres.svcn);
+	evcn1 = le64_to_cpu(attr_b->nres.evcn) + 1;
+
+	attr = attr_b;
+	le = le_b;
+	mi = mi_b;
+
+	if (le_b && (vcn < svcn || evcn1 <= vcn)) {
+		attr = ni_find_attr(ni, attr_b, &le, ATTR_DATA, NULL, 0, &vcn,
+				    &mi);
+		if (!attr) {
+			err = -EINVAL;
+			goto out;
+		}
+		svcn = le64_to_cpu(attr->nres.svcn);
+		evcn1 = le64_to_cpu(attr->nres.evcn) + 1;
+	}
+
+	err = attr_load_runs(attr, ni, run, NULL);
+	if (err)
+		goto out;
+
+	if (!ok) {
+		ok = run_lookup_entry(run, vcn, lcn, len, NULL);
+		if (ok && (*lcn != SPARSE_LCN || !new)) {
+			/* Normal way. */
+			err = 0;
+			goto ok;
+		}
+
+		if (!ok && !new) {
+			*len = 0;
+			err = 0;
+			goto ok;
+		}
+
+		if (ok && clen > *len) {
+			clen = *len;
+			to_alloc = (clen + clst_per_frame - 1) &
+				   ~(clst_per_frame - 1);
+		}
+	}
+
+	if (!is_attr_ext(attr_b)) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	/* Get the last LCN to allocate from. */
+	hint = 0;
+
+	if (vcn > evcn1) {
+		if (!run_add_entry(run, evcn1, SPARSE_LCN, vcn - evcn1,
+				   false)) {
+			err = -ENOMEM;
+			goto out;
+		}
+	} else if (vcn && !run_lookup_entry(run, vcn - 1, &hint, NULL, NULL)) {
+		hint = -1;
+	}
+
+	err = attr_allocate_clusters(
+		sbi, run, vcn, hint + 1, to_alloc, NULL, 0, len,
+		(sbi->record_size - le32_to_cpu(mi->mrec->used) + 8) / 3 + 1,
+		lcn);
+	if (err)
+		goto out;
+	*new = true;
+
+	end = vcn + *len;
+
+	total_size = le64_to_cpu(attr_b->nres.total_size) +
+		     ((u64)*len << cluster_bits);
+
+repack:
+	err = mi_pack_runs(mi, attr, run, max(end, evcn1) - svcn);
+	if (err)
+		goto out;
+
+	attr_b->nres.total_size = cpu_to_le64(total_size);
+	inode_set_bytes(&ni->vfs_inode, total_size);
+	ni->ni_flags |= NI_FLAG_UPDATE_PARENT;
+
+	mi_b->dirty = true;
+	mark_inode_dirty(&ni->vfs_inode);
+
+	/* Stored [vcn : next_svcn) from [vcn : end). */
+	next_svcn = le64_to_cpu(attr->nres.evcn) + 1;
+
+	if (end <= evcn1) {
+		if (next_svcn == evcn1) {
+			/* Normal way. Update attribute and exit. */
+			goto ok;
+		}
+		/* Add new segment [next_svcn : evcn1 - next_svcn). */
+		if (!ni->attr_list.size) {
+			err = ni_create_attr_list(ni);
+			if (err)
+				goto out;
+			/* Layout of records is changed. */
+			le_b = NULL;
+			attr_b = ni_find_attr(ni, NULL, &le_b, ATTR_DATA, NULL,
+					      0, NULL, &mi_b);
+			if (!attr_b) {
+				err = -ENOENT;
+				goto out;
+			}
+
+			attr = attr_b;
+			le = le_b;
+			mi = mi_b;
+			goto repack;
+		}
+	}
+
+	svcn = evcn1;
+
+	/* Estimate next attribute. */
+	attr = ni_find_attr(ni, attr, &le, ATTR_DATA, NULL, 0, &svcn, &mi);
+
+	if (attr) {
+		CLST alloc = bytes_to_cluster(
+			sbi, le64_to_cpu(attr_b->nres.alloc_size));
+		CLST evcn = le64_to_cpu(attr->nres.evcn);
+
+		if (end < next_svcn)
+			end = next_svcn;
+		while (end > evcn) {
+			/* Remove segment [svcn : evcn). */
+			mi_remove_attr(NULL, mi, attr);
+
+			if (!al_remove_le(ni, le)) {
+				err = -EINVAL;
+				goto out;
+			}
+
+			if (evcn + 1 >= alloc) {
+				/* Last attribute segment. */
+				evcn1 = evcn + 1;
+				goto ins_ext;
+			}
+
+			if (ni_load_mi(ni, le, &mi)) {
+				attr = NULL;
+				goto out;
+			}
+
+			attr = mi_find_attr(mi, NULL, ATTR_DATA, NULL, 0,
+					    &le->id);
+			if (!attr) {
+				err = -EINVAL;
+				goto out;
+			}
+			svcn = le64_to_cpu(attr->nres.svcn);
+			evcn = le64_to_cpu(attr->nres.evcn);
+		}
+
+		if (end < svcn)
+			end = svcn;
+
+		err = attr_load_runs(attr, ni, run, &end);
+		if (err)
+			goto out;
+
+		evcn1 = evcn + 1;
+		attr->nres.svcn = cpu_to_le64(next_svcn);
+		err = mi_pack_runs(mi, attr, run, evcn1 - next_svcn);
+		if (err)
+			goto out;
+
+		le->vcn = cpu_to_le64(next_svcn);
+		ni->attr_list.dirty = true;
+		mi->dirty = true;
+
+		next_svcn = le64_to_cpu(attr->nres.evcn) + 1;
+	}
+ins_ext:
+	if (evcn1 > next_svcn) {
+		err = ni_insert_nonresident(ni, ATTR_DATA, NULL, 0, run,
+					    next_svcn, evcn1 - next_svcn,
+					    attr_b->flags, &attr, &mi);
+		if (err)
+			goto out;
+	}
+ok:
+	run_truncate_around(run, vcn);
+out:
+	up_write(&ni->file.run_lock);
+	ni_unlock(ni);
+
+	return err;
+}
+
+int attr_data_read_resident(struct ntfs_inode *ni, struct page *page)
+{
+	u64 vbo;
+	struct ATTRIB *attr;
+	u32 data_size;
+
+	attr = ni_find_attr(ni, NULL, NULL, ATTR_DATA, NULL, 0, NULL, NULL);
+	if (!attr)
+		return -EINVAL;
+
+	if (attr->non_res)
+		return E_NTFS_NONRESIDENT;
+
+	vbo = page->index << PAGE_SHIFT;
+	data_size = le32_to_cpu(attr->res.data_size);
+	if (vbo < data_size) {
+		const char *data = resident_data(attr);
+		char *kaddr = kmap_atomic(page);
+		u32 use = data_size - vbo;
+
+		if (use > PAGE_SIZE)
+			use = PAGE_SIZE;
+
+		memcpy(kaddr, data + vbo, use);
+		memset(kaddr + use, 0, PAGE_SIZE - use);
+		kunmap_atomic(kaddr);
+		flush_dcache_page(page);
+		SetPageUptodate(page);
+	} else if (!PageUptodate(page)) {
+		zero_user_segment(page, 0, PAGE_SIZE);
+		SetPageUptodate(page);
+	}
+
+	return 0;
+}
+
+int attr_data_write_resident(struct ntfs_inode *ni, struct page *page)
+{
+	u64 vbo;
+	struct mft_inode *mi;
+	struct ATTRIB *attr;
+	u32 data_size;
+
+	attr = ni_find_attr(ni, NULL, NULL, ATTR_DATA, NULL, 0, NULL, &mi);
+	if (!attr)
+		return -EINVAL;
+
+	if (attr->non_res) {
+		/* Return special error code to check this case. */
+		return E_NTFS_NONRESIDENT;
+	}
+
+	vbo = page->index << PAGE_SHIFT;
+	data_size = le32_to_cpu(attr->res.data_size);
+	if (vbo < data_size) {
+		char *data = resident_data(attr);
+		char *kaddr = kmap_atomic(page);
+		u32 use = data_size - vbo;
+
+		if (use > PAGE_SIZE)
+			use = PAGE_SIZE;
+		memcpy(data + vbo, kaddr, use);
+		kunmap_atomic(kaddr);
+		mi->dirty = true;
+	}
+	ni->i_valid = data_size;
+
+	return 0;
+}
+
+/*
+ * attr_load_runs_vcn - Load runs with VCN.
+ */
+int attr_load_runs_vcn(struct ntfs_inode *ni, enum ATTR_TYPE type,
+		       const __le16 *name, u8 name_len, struct runs_tree *run,
+		       CLST vcn)
+{
+	struct ATTRIB *attr;
+	int err;
+	CLST svcn, evcn;
+	u16 ro;
+
+	attr = ni_find_attr(ni, NULL, NULL, type, name, name_len, &vcn, NULL);
+	if (!attr) {
+		/* Is record corrupted? */
+		return -ENOENT;
+	}
+
+	svcn = le64_to_cpu(attr->nres.svcn);
+	evcn = le64_to_cpu(attr->nres.evcn);
+
+	if (evcn < vcn || vcn < svcn) {
+		/* Is record corrupted? */
+		return -EINVAL;
+	}
+
+	ro = le16_to_cpu(attr->nres.run_off);
+	err = run_unpack_ex(run, ni->mi.sbi, ni->mi.rno, svcn, evcn, svcn,
+			    Add2Ptr(attr, ro), le32_to_cpu(attr->size) - ro);
+	if (err < 0)
+		return err;
+	return 0;
+}
+
+/*
+ * attr_load_runs_range - Load runs for given range [from to).
+ */
+int attr_load_runs_range(struct ntfs_inode *ni, enum ATTR_TYPE type,
+			 const __le16 *name, u8 name_len, struct runs_tree *run,
+			 u64 from, u64 to)
+{
+	struct ntfs_sb_info *sbi = ni->mi.sbi;
+	u8 cluster_bits = sbi->cluster_bits;
+	CLST vcn = from >> cluster_bits;
+	CLST vcn_last = (to - 1) >> cluster_bits;
+	CLST lcn, clen;
+	int err;
+
+	for (vcn = from >> cluster_bits; vcn <= vcn_last; vcn += clen) {
+		if (!run_lookup_entry(run, vcn, &lcn, &clen, NULL)) {
+			err = attr_load_runs_vcn(ni, type, name, name_len, run,
+						 vcn);
+			if (err)
+				return err;
+			clen = 0; /* Next run_lookup_entry(vcn) must be success. */
+		}
+	}
+
+	return 0;
+}
+
+#ifdef CONFIG_NTFS3_LZX_XPRESS
+/*
+ * attr_wof_frame_info
+ *
+ * Read header of Xpress/LZX file to get info about frame.
+ */
+int attr_wof_frame_info(struct ntfs_inode *ni, struct ATTRIB *attr,
+			struct runs_tree *run, u64 frame, u64 frames,
+			u8 frame_bits, u32 *ondisk_size, u64 *vbo_data)
+{
+	struct ntfs_sb_info *sbi = ni->mi.sbi;
+	u64 vbo[2], off[2], wof_size;
+	u32 voff;
+	u8 bytes_per_off;
+	char *addr;
+	struct page *page;
+	int i, err;
+	__le32 *off32;
+	__le64 *off64;
+
+	if (ni->vfs_inode.i_size < 0x100000000ull) {
+		/* File starts with array of 32 bit offsets. */
+		bytes_per_off = sizeof(__le32);
+		vbo[1] = frame << 2;
+		*vbo_data = frames << 2;
+	} else {
+		/* File starts with array of 64 bit offsets. */
+		bytes_per_off = sizeof(__le64);
+		vbo[1] = frame << 3;
+		*vbo_data = frames << 3;
+	}
+
+	/*
+	 * Read 4/8 bytes at [vbo - 4(8)] == offset where compressed frame starts.
+	 * Read 4/8 bytes at [vbo] == offset where compressed frame ends.
+	 */
+	if (!attr->non_res) {
+		if (vbo[1] + bytes_per_off > le32_to_cpu(attr->res.data_size)) {
+			ntfs_inode_err(&ni->vfs_inode, "is corrupted");
+			return -EINVAL;
+		}
+		addr = resident_data(attr);
+
+		if (bytes_per_off == sizeof(__le32)) {
+			off32 = Add2Ptr(addr, vbo[1]);
+			off[0] = vbo[1] ? le32_to_cpu(off32[-1]) : 0;
+			off[1] = le32_to_cpu(off32[0]);
+		} else {
+			off64 = Add2Ptr(addr, vbo[1]);
+			off[0] = vbo[1] ? le64_to_cpu(off64[-1]) : 0;
+			off[1] = le64_to_cpu(off64[0]);
+		}
+
+		*vbo_data += off[0];
+		*ondisk_size = off[1] - off[0];
+		return 0;
+	}
+
+	wof_size = le64_to_cpu(attr->nres.data_size);
+	down_write(&ni->file.run_lock);
+	page = ni->file.offs_page;
+	if (!page) {
+		page = alloc_page(GFP_KERNEL);
+		if (!page) {
+			err = -ENOMEM;
+			goto out;
+		}
+		page->index = -1;
+		ni->file.offs_page = page;
+	}
+	lock_page(page);
+	addr = page_address(page);
+
+	if (vbo[1]) {
+		voff = vbo[1] & (PAGE_SIZE - 1);
+		vbo[0] = vbo[1] - bytes_per_off;
+		i = 0;
+	} else {
+		voff = 0;
+		vbo[0] = 0;
+		off[0] = 0;
+		i = 1;
+	}
+
+	do {
+		pgoff_t index = vbo[i] >> PAGE_SHIFT;
+
+		if (index != page->index) {
+			u64 from = vbo[i] & ~(u64)(PAGE_SIZE - 1);
+			u64 to = min(from + PAGE_SIZE, wof_size);
+
+			err = attr_load_runs_range(ni, ATTR_DATA, WOF_NAME,
+						   ARRAY_SIZE(WOF_NAME), run,
+						   from, to);
+			if (err)
+				goto out1;
+
+			err = ntfs_bio_pages(sbi, run, &page, 1, from,
+					     to - from, REQ_OP_READ);
+			if (err) {
+				page->index = -1;
+				goto out1;
+			}
+			page->index = index;
+		}
+
+		if (i) {
+			if (bytes_per_off == sizeof(__le32)) {
+				off32 = Add2Ptr(addr, voff);
+				off[1] = le32_to_cpu(*off32);
+			} else {
+				off64 = Add2Ptr(addr, voff);
+				off[1] = le64_to_cpu(*off64);
+			}
+		} else if (!voff) {
+			if (bytes_per_off == sizeof(__le32)) {
+				off32 = Add2Ptr(addr, PAGE_SIZE - sizeof(u32));
+				off[0] = le32_to_cpu(*off32);
+			} else {
+				off64 = Add2Ptr(addr, PAGE_SIZE - sizeof(u64));
+				off[0] = le64_to_cpu(*off64);
+			}
+		} else {
+			/* Two values in one page. */
+			if (bytes_per_off == sizeof(__le32)) {
+				off32 = Add2Ptr(addr, voff);
+				off[0] = le32_to_cpu(off32[-1]);
+				off[1] = le32_to_cpu(off32[0]);
+			} else {
+				off64 = Add2Ptr(addr, voff);
+				off[0] = le64_to_cpu(off64[-1]);
+				off[1] = le64_to_cpu(off64[0]);
+			}
+			break;
+		}
+	} while (++i < 2);
+
+	*vbo_data += off[0];
+	*ondisk_size = off[1] - off[0];
+
+out1:
+	unlock_page(page);
+out:
+	up_write(&ni->file.run_lock);
+	return err;
+}
+#endif
+
+/*
+ * attr_is_frame_compressed - Used to detect compressed frame.
+ */
+int attr_is_frame_compressed(struct ntfs_inode *ni, struct ATTRIB *attr,
+			     CLST frame, CLST *clst_data)
+{
+	int err;
+	u32 clst_frame;
+	CLST clen, lcn, vcn, alen, slen, vcn_next;
+	size_t idx;
+	struct runs_tree *run;
+
+	*clst_data = 0;
+
+	if (!is_attr_compressed(attr))
+		return 0;
+
+	if (!attr->non_res)
+		return 0;
+
+	clst_frame = 1u << attr->nres.c_unit;
+	vcn = frame * clst_frame;
+	run = &ni->file.run;
+
+	if (!run_lookup_entry(run, vcn, &lcn, &clen, &idx)) {
+		err = attr_load_runs_vcn(ni, attr->type, attr_name(attr),
+					 attr->name_len, run, vcn);
+		if (err)
+			return err;
+
+		if (!run_lookup_entry(run, vcn, &lcn, &clen, &idx))
+			return -EINVAL;
+	}
+
+	if (lcn == SPARSE_LCN) {
+		/* Sparsed frame. */
+		return 0;
+	}
+
+	if (clen >= clst_frame) {
+		/*
+		 * The frame is not compressed 'cause
+		 * it does not contain any sparse clusters.
+		 */
+		*clst_data = clst_frame;
+		return 0;
+	}
+
+	alen = bytes_to_cluster(ni->mi.sbi, le64_to_cpu(attr->nres.alloc_size));
+	slen = 0;
+	*clst_data = clen;
+
+	/*
+	 * The frame is compressed if *clst_data + slen >= clst_frame.
+	 * Check next fragments.
+	 */
+	while ((vcn += clen) < alen) {
+		vcn_next = vcn;
+
+		if (!run_get_entry(run, ++idx, &vcn, &lcn, &clen) ||
+		    vcn_next != vcn) {
+			err = attr_load_runs_vcn(ni, attr->type,
+						 attr_name(attr),
+						 attr->name_len, run, vcn_next);
+			if (err)
+				return err;
+			vcn = vcn_next;
+
+			if (!run_lookup_entry(run, vcn, &lcn, &clen, &idx))
+				return -EINVAL;
+		}
+
+		if (lcn == SPARSE_LCN) {
+			slen += clen;
+		} else {
+			if (slen) {
+				/*
+				 * Data_clusters + sparse_clusters =
+				 * not enough for frame.
+				 */
+				return -EINVAL;
+			}
+			*clst_data += clen;
+		}
+
+		if (*clst_data + slen >= clst_frame) {
+			if (!slen) {
+				/*
+				 * There is no sparsed clusters in this frame
+				 * so it is not compressed.
+				 */
+				*clst_data = clst_frame;
+			} else {
+				/* Frame is compressed. */
+			}
+			break;
+		}
+	}
+
+	return 0;
+}
+
+/*
+ * attr_allocate_frame - Allocate/free clusters for @frame.
+ *
+ * Assumed: down_write(&ni->file.run_lock);
+ */
+int attr_allocate_frame(struct ntfs_inode *ni, CLST frame, size_t compr_size,
+			u64 new_valid)
+{
+	int err = 0;
+	struct runs_tree *run = &ni->file.run;
+	struct ntfs_sb_info *sbi = ni->mi.sbi;
+	struct ATTRIB *attr = NULL, *attr_b;
+	struct ATTR_LIST_ENTRY *le, *le_b;
+	struct mft_inode *mi, *mi_b;
+	CLST svcn, evcn1, next_svcn, lcn, len;
+	CLST vcn, end, clst_data;
+	u64 total_size, valid_size, data_size;
+
+	le_b = NULL;
+	attr_b = ni_find_attr(ni, NULL, &le_b, ATTR_DATA, NULL, 0, NULL, &mi_b);
+	if (!attr_b)
+		return -ENOENT;
+
+	if (!is_attr_ext(attr_b))
+		return -EINVAL;
+
+	vcn = frame << NTFS_LZNT_CUNIT;
+	total_size = le64_to_cpu(attr_b->nres.total_size);
+
+	svcn = le64_to_cpu(attr_b->nres.svcn);
+	evcn1 = le64_to_cpu(attr_b->nres.evcn) + 1;
+	data_size = le64_to_cpu(attr_b->nres.data_size);
+
+	if (svcn <= vcn && vcn < evcn1) {
+		attr = attr_b;
+		le = le_b;
+		mi = mi_b;
+	} else if (!le_b) {
+		err = -EINVAL;
+		goto out;
+	} else {
+		le = le_b;
+		attr = ni_find_attr(ni, attr_b, &le, ATTR_DATA, NULL, 0, &vcn,
+				    &mi);
+		if (!attr) {
+			err = -EINVAL;
+			goto out;
+		}
+		svcn = le64_to_cpu(attr->nres.svcn);
+		evcn1 = le64_to_cpu(attr->nres.evcn) + 1;
+	}
+
+	err = attr_load_runs(attr, ni, run, NULL);
+	if (err)
+		goto out;
+
+	err = attr_is_frame_compressed(ni, attr_b, frame, &clst_data);
+	if (err)
+		goto out;
+
+	total_size -= (u64)clst_data << sbi->cluster_bits;
+
+	len = bytes_to_cluster(sbi, compr_size);
+
+	if (len == clst_data)
+		goto out;
+
+	if (len < clst_data) {
+		err = run_deallocate_ex(sbi, run, vcn + len, clst_data - len,
+					NULL, true);
+		if (err)
+			goto out;
+
+		if (!run_add_entry(run, vcn + len, SPARSE_LCN, clst_data - len,
+				   false)) {
+			err = -ENOMEM;
+			goto out;
+		}
+		end = vcn + clst_data;
+		/* Run contains updated range [vcn + len : end). */
+	} else {
+		CLST alen, hint = 0;
+		/* Get the last LCN to allocate from. */
+		if (vcn + clst_data &&
+		    !run_lookup_entry(run, vcn + clst_data - 1, &hint, NULL,
+				      NULL)) {
+			hint = -1;
+		}
+
+		err = attr_allocate_clusters(sbi, run, vcn + clst_data,
+					     hint + 1, len - clst_data, NULL, 0,
+					     &alen, 0, &lcn);
+		if (err)
+			goto out;
+
+		end = vcn + len;
+		/* Run contains updated range [vcn + clst_data : end). */
+	}
+
+	total_size += (u64)len << sbi->cluster_bits;
+
+repack:
+	err = mi_pack_runs(mi, attr, run, max(end, evcn1) - svcn);
+	if (err)
+		goto out;
+
+	attr_b->nres.total_size = cpu_to_le64(total_size);
+	inode_set_bytes(&ni->vfs_inode, total_size);
+
+	mi_b->dirty = true;
+	mark_inode_dirty(&ni->vfs_inode);
+
+	/* Stored [vcn : next_svcn) from [vcn : end). */
+	next_svcn = le64_to_cpu(attr->nres.evcn) + 1;
+
+	if (end <= evcn1) {
+		if (next_svcn == evcn1) {
+			/* Normal way. Update attribute and exit. */
+			goto ok;
+		}
+		/* Add new segment [next_svcn : evcn1 - next_svcn). */
+		if (!ni->attr_list.size) {
+			err = ni_create_attr_list(ni);
+			if (err)
+				goto out;
+			/* Layout of records is changed. */
+			le_b = NULL;
+			attr_b = ni_find_attr(ni, NULL, &le_b, ATTR_DATA, NULL,
+					      0, NULL, &mi_b);
+			if (!attr_b) {
+				err = -ENOENT;
+				goto out;
+			}
+
+			attr = attr_b;
+			le = le_b;
+			mi = mi_b;
+			goto repack;
+		}
+	}
+
+	svcn = evcn1;
+
+	/* Estimate next attribute. */
+	attr = ni_find_attr(ni, attr, &le, ATTR_DATA, NULL, 0, &svcn, &mi);
+
+	if (attr) {
+		CLST alloc = bytes_to_cluster(
+			sbi, le64_to_cpu(attr_b->nres.alloc_size));
+		CLST evcn = le64_to_cpu(attr->nres.evcn);
+
+		if (end < next_svcn)
+			end = next_svcn;
+		while (end > evcn) {
+			/* Remove segment [svcn : evcn). */
+			mi_remove_attr(NULL, mi, attr);
+
+			if (!al_remove_le(ni, le)) {
+				err = -EINVAL;
+				goto out;
+			}
+
+			if (evcn + 1 >= alloc) {
+				/* Last attribute segment. */
+				evcn1 = evcn + 1;
+				goto ins_ext;
+			}
+
+			if (ni_load_mi(ni, le, &mi)) {
+				attr = NULL;
+				goto out;
+			}
+
+			attr = mi_find_attr(mi, NULL, ATTR_DATA, NULL, 0,
+					    &le->id);
+			if (!attr) {
+				err = -EINVAL;
+				goto out;
+			}
+			svcn = le64_to_cpu(attr->nres.svcn);
+			evcn = le64_to_cpu(attr->nres.evcn);
+		}
+
+		if (end < svcn)
+			end = svcn;
+
+		err = attr_load_runs(attr, ni, run, &end);
+		if (err)
+			goto out;
+
+		evcn1 = evcn + 1;
+		attr->nres.svcn = cpu_to_le64(next_svcn);
+		err = mi_pack_runs(mi, attr, run, evcn1 - next_svcn);
+		if (err)
+			goto out;
+
+		le->vcn = cpu_to_le64(next_svcn);
+		ni->attr_list.dirty = true;
+		mi->dirty = true;
+
+		next_svcn = le64_to_cpu(attr->nres.evcn) + 1;
+	}
+ins_ext:
+	if (evcn1 > next_svcn) {
+		err = ni_insert_nonresident(ni, ATTR_DATA, NULL, 0, run,
+					    next_svcn, evcn1 - next_svcn,
+					    attr_b->flags, &attr, &mi);
+		if (err)
+			goto out;
+	}
+ok:
+	run_truncate_around(run, vcn);
+out:
+	if (new_valid > data_size)
+		new_valid = data_size;
+
+	valid_size = le64_to_cpu(attr_b->nres.valid_size);
+	if (new_valid != valid_size) {
+		attr_b->nres.valid_size = cpu_to_le64(valid_size);
+		mi_b->dirty = true;
+	}
+
+	return err;
+}
+
+/*
+ * attr_collapse_range - Collapse range in file.
+ */
+int attr_collapse_range(struct ntfs_inode *ni, u64 vbo, u64 bytes)
+{
+	int err = 0;
+	struct runs_tree *run = &ni->file.run;
+	struct ntfs_sb_info *sbi = ni->mi.sbi;
+	struct ATTRIB *attr = NULL, *attr_b;
+	struct ATTR_LIST_ENTRY *le, *le_b;
+	struct mft_inode *mi, *mi_b;
+	CLST svcn, evcn1, len, dealloc, alen;
+	CLST vcn, end;
+	u64 valid_size, data_size, alloc_size, total_size;
+	u32 mask;
+	__le16 a_flags;
+
+	if (!bytes)
+		return 0;
+
+	le_b = NULL;
+	attr_b = ni_find_attr(ni, NULL, &le_b, ATTR_DATA, NULL, 0, NULL, &mi_b);
+	if (!attr_b)
+		return -ENOENT;
+
+	if (!attr_b->non_res) {
+		/* Attribute is resident. Nothing to do? */
+		return 0;
+	}
+
+	data_size = le64_to_cpu(attr_b->nres.data_size);
+	alloc_size = le64_to_cpu(attr_b->nres.alloc_size);
+	a_flags = attr_b->flags;
+
+	if (is_attr_ext(attr_b)) {
+		total_size = le64_to_cpu(attr_b->nres.total_size);
+		mask = (sbi->cluster_size << attr_b->nres.c_unit) - 1;
+	} else {
+		total_size = alloc_size;
+		mask = sbi->cluster_mask;
+	}
+
+	if ((vbo & mask) || (bytes & mask)) {
+		/* Allow to collapse only cluster aligned ranges. */
+		return -EINVAL;
+	}
+
+	if (vbo > data_size)
+		return -EINVAL;
+
+	down_write(&ni->file.run_lock);
+
+	if (vbo + bytes >= data_size) {
+		u64 new_valid = min(ni->i_valid, vbo);
+
+		/* Simple truncate file at 'vbo'. */
+		truncate_setsize(&ni->vfs_inode, vbo);
+		err = attr_set_size(ni, ATTR_DATA, NULL, 0, &ni->file.run, vbo,
+				    &new_valid, true, NULL);
+
+		if (!err && new_valid < ni->i_valid)
+			ni->i_valid = new_valid;
+
+		goto out;
+	}
+
+	/*
+	 * Enumerate all attribute segments and collapse.
+	 */
+	alen = alloc_size >> sbi->cluster_bits;
+	vcn = vbo >> sbi->cluster_bits;
+	len = bytes >> sbi->cluster_bits;
+	end = vcn + len;
+	dealloc = 0;
+
+	svcn = le64_to_cpu(attr_b->nres.svcn);
+	evcn1 = le64_to_cpu(attr_b->nres.evcn) + 1;
+
+	if (svcn <= vcn && vcn < evcn1) {
+		attr = attr_b;
+		le = le_b;
+		mi = mi_b;
+	} else if (!le_b) {
+		err = -EINVAL;
+		goto out;
+	} else {
+		le = le_b;
+		attr = ni_find_attr(ni, attr_b, &le, ATTR_DATA, NULL, 0, &vcn,
+				    &mi);
+		if (!attr) {
+			err = -EINVAL;
+			goto out;
+		}
+
+		svcn = le64_to_cpu(attr->nres.svcn);
+		evcn1 = le64_to_cpu(attr->nres.evcn) + 1;
+	}
+
+	for (;;) {
+		if (svcn >= end) {
+			/* Shift VCN- */
+			attr->nres.svcn = cpu_to_le64(svcn - len);
+			attr->nres.evcn = cpu_to_le64(evcn1 - 1 - len);
+			if (le) {
+				le->vcn = attr->nres.svcn;
+				ni->attr_list.dirty = true;
+			}
+			mi->dirty = true;
+		} else if (svcn < vcn || end < evcn1) {
+			CLST vcn1, eat, next_svcn;
+
+			/* Collapse a part of this attribute segment. */
+			err = attr_load_runs(attr, ni, run, &svcn);
+			if (err)
+				goto out;
+			vcn1 = max(vcn, svcn);
+			eat = min(end, evcn1) - vcn1;
+
+			err = run_deallocate_ex(sbi, run, vcn1, eat, &dealloc,
+						true);
+			if (err)
+				goto out;
+
+			if (!run_collapse_range(run, vcn1, eat)) {
+				err = -ENOMEM;
+				goto out;
+			}
+
+			if (svcn >= vcn) {
+				/* Shift VCN */
+				attr->nres.svcn = cpu_to_le64(vcn);
+				if (le) {
+					le->vcn = attr->nres.svcn;
+					ni->attr_list.dirty = true;
+				}
+			}
+
+			err = mi_pack_runs(mi, attr, run, evcn1 - svcn - eat);
+			if (err)
+				goto out;
+
+			next_svcn = le64_to_cpu(attr->nres.evcn) + 1;
+			if (next_svcn + eat < evcn1) {
+				err = ni_insert_nonresident(
+					ni, ATTR_DATA, NULL, 0, run, next_svcn,
+					evcn1 - eat - next_svcn, a_flags, &attr,
+					&mi);
+				if (err)
+					goto out;
+
+				/* Layout of records maybe changed. */
+				attr_b = NULL;
+				le = al_find_ex(ni, NULL, ATTR_DATA, NULL, 0,
+						&next_svcn);
+				if (!le) {
+					err = -EINVAL;
+					goto out;
+				}
+			}
+
+			/* Free all allocated memory. */
+			run_truncate(run, 0);
+		} else {
+			u16 le_sz;
+			u16 roff = le16_to_cpu(attr->nres.run_off);
+
+			run_unpack_ex(RUN_DEALLOCATE, sbi, ni->mi.rno, svcn,
+				      evcn1 - 1, svcn, Add2Ptr(attr, roff),
+				      le32_to_cpu(attr->size) - roff);
+
+			/* Delete this attribute segment. */
+			mi_remove_attr(NULL, mi, attr);
+			if (!le)
+				break;
+
+			le_sz = le16_to_cpu(le->size);
+			if (!al_remove_le(ni, le)) {
+				err = -EINVAL;
+				goto out;
+			}
+
+			if (evcn1 >= alen)
+				break;
+
+			if (!svcn) {
+				/* Load next record that contains this attribute. */
+				if (ni_load_mi(ni, le, &mi)) {
+					err = -EINVAL;
+					goto out;
+				}
+
+				/* Look for required attribute. */
+				attr = mi_find_attr(mi, NULL, ATTR_DATA, NULL,
+						    0, &le->id);
+				if (!attr) {
+					err = -EINVAL;
+					goto out;
+				}
+				goto next_attr;
+			}
+			le = (struct ATTR_LIST_ENTRY *)((u8 *)le - le_sz);
+		}
+
+		if (evcn1 >= alen)
+			break;
+
+		attr = ni_enum_attr_ex(ni, attr, &le, &mi);
+		if (!attr) {
+			err = -EINVAL;
+			goto out;
+		}
+
+next_attr:
+		svcn = le64_to_cpu(attr->nres.svcn);
+		evcn1 = le64_to_cpu(attr->nres.evcn) + 1;
+	}
+
+	if (!attr_b) {
+		le_b = NULL;
+		attr_b = ni_find_attr(ni, NULL, &le_b, ATTR_DATA, NULL, 0, NULL,
+				      &mi_b);
+		if (!attr_b) {
+			err = -ENOENT;
+			goto out;
+		}
+	}
+
+	data_size -= bytes;
+	valid_size = ni->i_valid;
+	if (vbo + bytes <= valid_size)
+		valid_size -= bytes;
+	else if (vbo < valid_size)
+		valid_size = vbo;
+
+	attr_b->nres.alloc_size = cpu_to_le64(alloc_size - bytes);
+	attr_b->nres.data_size = cpu_to_le64(data_size);
+	attr_b->nres.valid_size = cpu_to_le64(min(valid_size, data_size));
+	total_size -= (u64)dealloc << sbi->cluster_bits;
+	if (is_attr_ext(attr_b))
+		attr_b->nres.total_size = cpu_to_le64(total_size);
+	mi_b->dirty = true;
+
+	/* Update inode size. */
+	ni->i_valid = valid_size;
+	ni->vfs_inode.i_size = data_size;
+	inode_set_bytes(&ni->vfs_inode, total_size);
+	ni->ni_flags |= NI_FLAG_UPDATE_PARENT;
+	mark_inode_dirty(&ni->vfs_inode);
+
+out:
+	up_write(&ni->file.run_lock);
+	if (err)
+		make_bad_inode(&ni->vfs_inode);
+
+	return err;
+}
+
+/*
+ * attr_punch_hole
+ *
+ * Not for normal files.
+ */
+int attr_punch_hole(struct ntfs_inode *ni, u64 vbo, u64 bytes, u32 *frame_size)
+{
+	int err = 0;
+	struct runs_tree *run = &ni->file.run;
+	struct ntfs_sb_info *sbi = ni->mi.sbi;
+	struct ATTRIB *attr = NULL, *attr_b;
+	struct ATTR_LIST_ENTRY *le, *le_b;
+	struct mft_inode *mi, *mi_b;
+	CLST svcn, evcn1, vcn, len, end, alen, dealloc;
+	u64 total_size, alloc_size;
+	u32 mask;
+
+	if (!bytes)
+		return 0;
+
+	le_b = NULL;
+	attr_b = ni_find_attr(ni, NULL, &le_b, ATTR_DATA, NULL, 0, NULL, &mi_b);
+	if (!attr_b)
+		return -ENOENT;
+
+	if (!attr_b->non_res) {
+		u32 data_size = le32_to_cpu(attr->res.data_size);
+		u32 from, to;
+
+		if (vbo > data_size)
+			return 0;
+
+		from = vbo;
+		to = (vbo + bytes) < data_size ? (vbo + bytes) : data_size;
+		memset(Add2Ptr(resident_data(attr_b), from), 0, to - from);
+		return 0;
+	}
+
+	if (!is_attr_ext(attr_b))
+		return -EOPNOTSUPP;
+
+	alloc_size = le64_to_cpu(attr_b->nres.alloc_size);
+	total_size = le64_to_cpu(attr_b->nres.total_size);
+
+	if (vbo >= alloc_size) {
+		/* NOTE: It is allowed. */
+		return 0;
+	}
+
+	mask = (sbi->cluster_size << attr_b->nres.c_unit) - 1;
+
+	bytes += vbo;
+	if (bytes > alloc_size)
+		bytes = alloc_size;
+	bytes -= vbo;
+
+	if ((vbo & mask) || (bytes & mask)) {
+		/* We have to zero a range(s). */
+		if (frame_size == NULL) {
+			/* Caller insists range is aligned. */
+			return -EINVAL;
+		}
+		*frame_size = mask + 1;
+		return E_NTFS_NOTALIGNED;
+	}
+
+	down_write(&ni->file.run_lock);
+	/*
+	 * Enumerate all attribute segments and punch hole where necessary.
+	 */
+	alen = alloc_size >> sbi->cluster_bits;
+	vcn = vbo >> sbi->cluster_bits;
+	len = bytes >> sbi->cluster_bits;
+	end = vcn + len;
+	dealloc = 0;
+
+	svcn = le64_to_cpu(attr_b->nres.svcn);
+	evcn1 = le64_to_cpu(attr_b->nres.evcn) + 1;
+
+	if (svcn <= vcn && vcn < evcn1) {
+		attr = attr_b;
+		le = le_b;
+		mi = mi_b;
+	} else if (!le_b) {
+		err = -EINVAL;
+		goto out;
+	} else {
+		le = le_b;
+		attr = ni_find_attr(ni, attr_b, &le, ATTR_DATA, NULL, 0, &vcn,
+				    &mi);
+		if (!attr) {
+			err = -EINVAL;
+			goto out;
+		}
+
+		svcn = le64_to_cpu(attr->nres.svcn);
+		evcn1 = le64_to_cpu(attr->nres.evcn) + 1;
+	}
+
+	while (svcn < end) {
+		CLST vcn1, zero, dealloc2;
+
+		err = attr_load_runs(attr, ni, run, &svcn);
+		if (err)
+			goto out;
+		vcn1 = max(vcn, svcn);
+		zero = min(end, evcn1) - vcn1;
+
+		dealloc2 = dealloc;
+		err = run_deallocate_ex(sbi, run, vcn1, zero, &dealloc, true);
+		if (err)
+			goto out;
+
+		if (dealloc2 == dealloc) {
+			/* Looks like the required range is already sparsed. */
+		} else {
+			if (!run_add_entry(run, vcn1, SPARSE_LCN, zero,
+					   false)) {
+				err = -ENOMEM;
+				goto out;
+			}
+
+			err = mi_pack_runs(mi, attr, run, evcn1 - svcn);
+			if (err)
+				goto out;
+		}
+		/* Free all allocated memory. */
+		run_truncate(run, 0);
+
+		if (evcn1 >= alen)
+			break;
+
+		attr = ni_enum_attr_ex(ni, attr, &le, &mi);
+		if (!attr) {
+			err = -EINVAL;
+			goto out;
+		}
+
+		svcn = le64_to_cpu(attr->nres.svcn);
+		evcn1 = le64_to_cpu(attr->nres.evcn) + 1;
+	}
+
+	total_size -= (u64)dealloc << sbi->cluster_bits;
+	attr_b->nres.total_size = cpu_to_le64(total_size);
+	mi_b->dirty = true;
+
+	/* Update inode size. */
+	inode_set_bytes(&ni->vfs_inode, total_size);
+	ni->ni_flags |= NI_FLAG_UPDATE_PARENT;
+	mark_inode_dirty(&ni->vfs_inode);
+
+out:
+	up_write(&ni->file.run_lock);
+	if (err)
+		make_bad_inode(&ni->vfs_inode);
+
+	return err;
+}
diff --git a/fs/ntfs3/attrlist.c b/fs/ntfs3/attrlist.c
new file mode 100644
index 000000000000..fa32399eb517
--- /dev/null
+++ b/fs/ntfs3/attrlist.c
@@ -0,0 +1,460 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ *
+ * Copyright (C) 2019-2021 Paragon Software GmbH, All rights reserved.
+ *
+ */
+
+#include <linux/blkdev.h>
+#include <linux/buffer_head.h>
+#include <linux/fs.h>
+#include <linux/nls.h>
+
+#include "debug.h"
+#include "ntfs.h"
+#include "ntfs_fs.h"
+
+/*
+ * al_is_valid_le
+ *
+ * Return: True if @le is valid.
+ */
+static inline bool al_is_valid_le(const struct ntfs_inode *ni,
+				  struct ATTR_LIST_ENTRY *le)
+{
+	if (!le || !ni->attr_list.le || !ni->attr_list.size)
+		return false;
+
+	return PtrOffset(ni->attr_list.le, le) + le16_to_cpu(le->size) <=
+	       ni->attr_list.size;
+}
+
+void al_destroy(struct ntfs_inode *ni)
+{
+	run_close(&ni->attr_list.run);
+	kfree(ni->attr_list.le);
+	ni->attr_list.le = NULL;
+	ni->attr_list.size = 0;
+	ni->attr_list.dirty = false;
+}
+
+/*
+ * ntfs_load_attr_list
+ *
+ * This method makes sure that the ATTRIB list, if present,
+ * has been properly set up.
+ */
+int ntfs_load_attr_list(struct ntfs_inode *ni, struct ATTRIB *attr)
+{
+	int err;
+	size_t lsize;
+	void *le = NULL;
+
+	if (ni->attr_list.size)
+		return 0;
+
+	if (!attr->non_res) {
+		lsize = le32_to_cpu(attr->res.data_size);
+		le = kmalloc(al_aligned(lsize), GFP_NOFS);
+		if (!le) {
+			err = -ENOMEM;
+			goto out;
+		}
+		memcpy(le, resident_data(attr), lsize);
+	} else if (attr->nres.svcn) {
+		err = -EINVAL;
+		goto out;
+	} else {
+		u16 run_off = le16_to_cpu(attr->nres.run_off);
+
+		lsize = le64_to_cpu(attr->nres.data_size);
+
+		run_init(&ni->attr_list.run);
+
+		err = run_unpack_ex(&ni->attr_list.run, ni->mi.sbi, ni->mi.rno,
+				    0, le64_to_cpu(attr->nres.evcn), 0,
+				    Add2Ptr(attr, run_off),
+				    le32_to_cpu(attr->size) - run_off);
+		if (err < 0)
+			goto out;
+
+		le = kmalloc(al_aligned(lsize), GFP_NOFS);
+		if (!le) {
+			err = -ENOMEM;
+			goto out;
+		}
+
+		err = ntfs_read_run_nb(ni->mi.sbi, &ni->attr_list.run, 0, le,
+				       lsize, NULL);
+		if (err)
+			goto out;
+	}
+
+	ni->attr_list.size = lsize;
+	ni->attr_list.le = le;
+
+	return 0;
+
+out:
+	ni->attr_list.le = le;
+	al_destroy(ni);
+
+	return err;
+}
+
+/*
+ * al_enumerate
+ *
+ * Return:
+ * * The next list le.
+ * * If @le is NULL then return the first le.
+ */
+struct ATTR_LIST_ENTRY *al_enumerate(struct ntfs_inode *ni,
+				     struct ATTR_LIST_ENTRY *le)
+{
+	size_t off;
+	u16 sz;
+
+	if (!le) {
+		le = ni->attr_list.le;
+	} else {
+		sz = le16_to_cpu(le->size);
+		if (sz < sizeof(struct ATTR_LIST_ENTRY)) {
+			/* Impossible 'cause we should not return such le. */
+			return NULL;
+		}
+		le = Add2Ptr(le, sz);
+	}
+
+	/* Check boundary. */
+	off = PtrOffset(ni->attr_list.le, le);
+	if (off + sizeof(struct ATTR_LIST_ENTRY) > ni->attr_list.size) {
+		/* The regular end of list. */
+		return NULL;
+	}
+
+	sz = le16_to_cpu(le->size);
+
+	/* Check le for errors. */
+	if (sz < sizeof(struct ATTR_LIST_ENTRY) ||
+	    off + sz > ni->attr_list.size ||
+	    sz < le->name_off + le->name_len * sizeof(short)) {
+		return NULL;
+	}
+
+	return le;
+}
+
+/*
+ * al_find_le
+ *
+ * Find the first le in the list which matches type, name and VCN.
+ *
+ * Return: NULL if not found.
+ */
+struct ATTR_LIST_ENTRY *al_find_le(struct ntfs_inode *ni,
+				   struct ATTR_LIST_ENTRY *le,
+				   const struct ATTRIB *attr)
+{
+	CLST svcn = attr_svcn(attr);
+
+	return al_find_ex(ni, le, attr->type, attr_name(attr), attr->name_len,
+			  &svcn);
+}
+
+/*
+ * al_find_ex
+ *
+ * Find the first le in the list which matches type, name and VCN.
+ *
+ * Return: NULL if not found.
+ */
+struct ATTR_LIST_ENTRY *al_find_ex(struct ntfs_inode *ni,
+				   struct ATTR_LIST_ENTRY *le,
+				   enum ATTR_TYPE type, const __le16 *name,
+				   u8 name_len, const CLST *vcn)
+{
+	struct ATTR_LIST_ENTRY *ret = NULL;
+	u32 type_in = le32_to_cpu(type);
+
+	while ((le = al_enumerate(ni, le))) {
+		u64 le_vcn;
+		int diff = le32_to_cpu(le->type) - type_in;
+
+		/* List entries are sorted by type, name and VCN. */
+		if (diff < 0)
+			continue;
+
+		if (diff > 0)
+			return ret;
+
+		if (le->name_len != name_len)
+			continue;
+
+		le_vcn = le64_to_cpu(le->vcn);
+		if (!le_vcn) {
+			/*
+			 * Compare entry names only for entry with vcn == 0.
+			 */
+			diff = ntfs_cmp_names(le_name(le), name_len, name,
+					      name_len, ni->mi.sbi->upcase,
+					      true);
+			if (diff < 0)
+				continue;
+
+			if (diff > 0)
+				return ret;
+		}
+
+		if (!vcn)
+			return le;
+
+		if (*vcn == le_vcn)
+			return le;
+
+		if (*vcn < le_vcn)
+			return ret;
+
+		ret = le;
+	}
+
+	return ret;
+}
+
+/*
+ * al_find_le_to_insert
+ *
+ * Find the first list entry which matches type, name and VCN.
+ */
+static struct ATTR_LIST_ENTRY *al_find_le_to_insert(struct ntfs_inode *ni,
+						    enum ATTR_TYPE type,
+						    const __le16 *name,
+						    u8 name_len, CLST vcn)
+{
+	struct ATTR_LIST_ENTRY *le = NULL, *prev;
+	u32 type_in = le32_to_cpu(type);
+
+	/* List entries are sorted by type, name and VCN. */
+	while ((le = al_enumerate(ni, prev = le))) {
+		int diff = le32_to_cpu(le->type) - type_in;
+
+		if (diff < 0)
+			continue;
+
+		if (diff > 0)
+			return le;
+
+		if (!le->vcn) {
+			/*
+			 * Compare entry names only for entry with vcn == 0.
+			 */
+			diff = ntfs_cmp_names(le_name(le), le->name_len, name,
+					      name_len, ni->mi.sbi->upcase,
+					      true);
+			if (diff < 0)
+				continue;
+
+			if (diff > 0)
+				return le;
+		}
+
+		if (le64_to_cpu(le->vcn) >= vcn)
+			return le;
+	}
+
+	return prev ? Add2Ptr(prev, le16_to_cpu(prev->size)) : ni->attr_list.le;
+}
+
+/*
+ * al_add_le
+ *
+ * Add an "attribute list entry" to the list.
+ */
+int al_add_le(struct ntfs_inode *ni, enum ATTR_TYPE type, const __le16 *name,
+	      u8 name_len, CLST svcn, __le16 id, const struct MFT_REF *ref,
+	      struct ATTR_LIST_ENTRY **new_le)
+{
+	int err;
+	struct ATTRIB *attr;
+	struct ATTR_LIST_ENTRY *le;
+	size_t off;
+	u16 sz;
+	size_t asize, new_asize, old_size;
+	u64 new_size;
+	typeof(ni->attr_list) *al = &ni->attr_list;
+
+	/*
+	 * Compute the size of the new 'le'
+	 */
+	sz = le_size(name_len);
+	old_size = al->size;
+	new_size = old_size + sz;
+	asize = al_aligned(old_size);
+	new_asize = al_aligned(new_size);
+
+	/* Scan forward to the point at which the new 'le' should be inserted. */
+	le = al_find_le_to_insert(ni, type, name, name_len, svcn);
+	off = PtrOffset(al->le, le);
+
+	if (new_size > asize) {
+		void *ptr = kmalloc(new_asize, GFP_NOFS);
+
+		if (!ptr)
+			return -ENOMEM;
+
+		memcpy(ptr, al->le, off);
+		memcpy(Add2Ptr(ptr, off + sz), le, old_size - off);
+		le = Add2Ptr(ptr, off);
+		kfree(al->le);
+		al->le = ptr;
+	} else {
+		memmove(Add2Ptr(le, sz), le, old_size - off);
+	}
+	*new_le = le;
+
+	al->size = new_size;
+
+	le->type = type;
+	le->size = cpu_to_le16(sz);
+	le->name_len = name_len;
+	le->name_off = offsetof(struct ATTR_LIST_ENTRY, name);
+	le->vcn = cpu_to_le64(svcn);
+	le->ref = *ref;
+	le->id = id;
+	memcpy(le->name, name, sizeof(short) * name_len);
+
+	err = attr_set_size(ni, ATTR_LIST, NULL, 0, &al->run, new_size,
+			    &new_size, true, &attr);
+	if (err) {
+		/* Undo memmove above. */
+		memmove(le, Add2Ptr(le, sz), old_size - off);
+		al->size = old_size;
+		return err;
+	}
+
+	al->dirty = true;
+
+	if (attr && attr->non_res) {
+		err = ntfs_sb_write_run(ni->mi.sbi, &al->run, 0, al->le,
+					al->size);
+		if (err)
+			return err;
+		al->dirty = false;
+	}
+
+	return 0;
+}
+
+/*
+ * al_remove_le - Remove @le from attribute list.
+ */
+bool al_remove_le(struct ntfs_inode *ni, struct ATTR_LIST_ENTRY *le)
+{
+	u16 size;
+	size_t off;
+	typeof(ni->attr_list) *al = &ni->attr_list;
+
+	if (!al_is_valid_le(ni, le))
+		return false;
+
+	/* Save on stack the size of 'le' */
+	size = le16_to_cpu(le->size);
+	off = PtrOffset(al->le, le);
+
+	memmove(le, Add2Ptr(le, size), al->size - (off + size));
+
+	al->size -= size;
+	al->dirty = true;
+
+	return true;
+}
+
+/*
+ * al_delete_le - Delete first le from the list which matches its parameters.
+ */
+bool al_delete_le(struct ntfs_inode *ni, enum ATTR_TYPE type, CLST vcn,
+		  const __le16 *name, size_t name_len,
+		  const struct MFT_REF *ref)
+{
+	u16 size;
+	struct ATTR_LIST_ENTRY *le;
+	size_t off;
+	typeof(ni->attr_list) *al = &ni->attr_list;
+
+	/* Scan forward to the first le that matches the input. */
+	le = al_find_ex(ni, NULL, type, name, name_len, &vcn);
+	if (!le)
+		return false;
+
+	off = PtrOffset(al->le, le);
+
+next:
+	if (off >= al->size)
+		return false;
+	if (le->type != type)
+		return false;
+	if (le->name_len != name_len)
+		return false;
+	if (name_len && ntfs_cmp_names(le_name(le), name_len, name, name_len,
+				       ni->mi.sbi->upcase, true))
+		return false;
+	if (le64_to_cpu(le->vcn) != vcn)
+		return false;
+
+	/*
+	 * The caller specified a segment reference, so we have to
+	 * scan through the matching entries until we find that segment
+	 * reference or we run of matching entries.
+	 */
+	if (ref && memcmp(ref, &le->ref, sizeof(*ref))) {
+		off += le16_to_cpu(le->size);
+		le = Add2Ptr(al->le, off);
+		goto next;
+	}
+
+	/* Save on stack the size of 'le'. */
+	size = le16_to_cpu(le->size);
+	/* Delete the le. */
+	memmove(le, Add2Ptr(le, size), al->size - (off + size));
+
+	al->size -= size;
+	al->dirty = true;
+
+	return true;
+}
+
+int al_update(struct ntfs_inode *ni)
+{
+	int err;
+	struct ATTRIB *attr;
+	typeof(ni->attr_list) *al = &ni->attr_list;
+
+	if (!al->dirty || !al->size)
+		return 0;
+
+	/*
+	 * Attribute list increased on demand in al_add_le.
+	 * Attribute list decreased here.
+	 */
+	err = attr_set_size(ni, ATTR_LIST, NULL, 0, &al->run, al->size, NULL,
+			    false, &attr);
+	if (err)
+		goto out;
+
+	if (!attr->non_res) {
+		memcpy(resident_data(attr), al->le, al->size);
+	} else {
+		err = ntfs_sb_write_run(ni->mi.sbi, &al->run, 0, al->le,
+					al->size);
+		if (err)
+			goto out;
+
+		attr->nres.valid_size = attr->nres.data_size;
+	}
+
+	ni->mi.dirty = true;
+	al->dirty = false;
+
+out:
+	return err;
+}
diff --git a/fs/ntfs3/bitfunc.c b/fs/ntfs3/bitfunc.c
new file mode 100644
index 000000000000..ce304d40b5e1
--- /dev/null
+++ b/fs/ntfs3/bitfunc.c
@@ -0,0 +1,134 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ *
+ * Copyright (C) 2019-2021 Paragon Software GmbH, All rights reserved.
+ *
+ */
+
+#include <linux/blkdev.h>
+#include <linux/buffer_head.h>
+#include <linux/fs.h>
+#include <linux/nls.h>
+
+#include "debug.h"
+#include "ntfs.h"
+#include "ntfs_fs.h"
+
+#define BITS_IN_SIZE_T (sizeof(size_t) * 8)
+
+/*
+ * fill_mask[i] - first i bits are '1' , i = 0,1,2,3,4,5,6,7,8
+ * fill_mask[i] = 0xFF >> (8-i)
+ */
+static const u8 fill_mask[] = { 0x00, 0x01, 0x03, 0x07, 0x0F,
+				0x1F, 0x3F, 0x7F, 0xFF };
+
+/*
+ * zero_mask[i] - first i bits are '0' , i = 0,1,2,3,4,5,6,7,8
+ * zero_mask[i] = 0xFF << i
+ */
+static const u8 zero_mask[] = { 0xFF, 0xFE, 0xFC, 0xF8, 0xF0,
+				0xE0, 0xC0, 0x80, 0x00 };
+
+/*
+ * are_bits_clear
+ *
+ * Return: True if all bits [bit, bit+nbits) are zeros "0".
+ */
+bool are_bits_clear(const ulong *lmap, size_t bit, size_t nbits)
+{
+	size_t pos = bit & 7;
+	const u8 *map = (u8 *)lmap + (bit >> 3);
+
+	if (pos) {
+		if (8 - pos >= nbits)
+			return !nbits || !(*map & fill_mask[pos + nbits] &
+					   zero_mask[pos]);
+
+		if (*map++ & zero_mask[pos])
+			return false;
+		nbits -= 8 - pos;
+	}
+
+	pos = ((size_t)map) & (sizeof(size_t) - 1);
+	if (pos) {
+		pos = sizeof(size_t) - pos;
+		if (nbits >= pos * 8) {
+			for (nbits -= pos * 8; pos; pos--, map++) {
+				if (*map)
+					return false;
+			}
+		}
+	}
+
+	for (pos = nbits / BITS_IN_SIZE_T; pos; pos--, map += sizeof(size_t)) {
+		if (*((size_t *)map))
+			return false;
+	}
+
+	for (pos = (nbits % BITS_IN_SIZE_T) >> 3; pos; pos--, map++) {
+		if (*map)
+			return false;
+	}
+
+	pos = nbits & 7;
+	if (pos && (*map & fill_mask[pos]))
+		return false;
+
+	return true;
+}
+
+/*
+ * are_bits_set
+ *
+ * Return: True if all bits [bit, bit+nbits) are ones "1".
+ */
+bool are_bits_set(const ulong *lmap, size_t bit, size_t nbits)
+{
+	u8 mask;
+	size_t pos = bit & 7;
+	const u8 *map = (u8 *)lmap + (bit >> 3);
+
+	if (pos) {
+		if (8 - pos >= nbits) {
+			mask = fill_mask[pos + nbits] & zero_mask[pos];
+			return !nbits || (*map & mask) == mask;
+		}
+
+		mask = zero_mask[pos];
+		if ((*map++ & mask) != mask)
+			return false;
+		nbits -= 8 - pos;
+	}
+
+	pos = ((size_t)map) & (sizeof(size_t) - 1);
+	if (pos) {
+		pos = sizeof(size_t) - pos;
+		if (nbits >= pos * 8) {
+			for (nbits -= pos * 8; pos; pos--, map++) {
+				if (*map != 0xFF)
+					return false;
+			}
+		}
+	}
+
+	for (pos = nbits / BITS_IN_SIZE_T; pos; pos--, map += sizeof(size_t)) {
+		if (*((size_t *)map) != MINUS_ONE_T)
+			return false;
+	}
+
+	for (pos = (nbits % BITS_IN_SIZE_T) >> 3; pos; pos--, map++) {
+		if (*map != 0xFF)
+			return false;
+	}
+
+	pos = nbits & 7;
+	if (pos) {
+		u8 mask = fill_mask[pos];
+
+		if ((*map & mask) != mask)
+			return false;
+	}
+
+	return true;
+}
diff --git a/fs/ntfs3/bitmap.c b/fs/ntfs3/bitmap.c
new file mode 100644
index 000000000000..831501555009
--- /dev/null
+++ b/fs/ntfs3/bitmap.c
@@ -0,0 +1,1493 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ *
+ * Copyright (C) 2019-2021 Paragon Software GmbH, All rights reserved.
+ *
+ * This code builds two trees of free clusters extents.
+ * Trees are sorted by start of extent and by length of extent.
+ * NTFS_MAX_WND_EXTENTS defines the maximum number of elements in trees.
+ * In extreme case code reads on-disk bitmap to find free clusters.
+ *
+ */
+
+#include <linux/blkdev.h>
+#include <linux/buffer_head.h>
+#include <linux/fs.h>
+#include <linux/nls.h>
+
+#include "debug.h"
+#include "ntfs.h"
+#include "ntfs_fs.h"
+
+/*
+ * Maximum number of extents in tree.
+ */
+#define NTFS_MAX_WND_EXTENTS (32u * 1024u)
+
+struct rb_node_key {
+	struct rb_node node;
+	size_t key;
+};
+
+struct e_node {
+	struct rb_node_key start; /* Tree sorted by start. */
+	struct rb_node_key count; /* Tree sorted by len. */
+};
+
+static int wnd_rescan(struct wnd_bitmap *wnd);
+static struct buffer_head *wnd_map(struct wnd_bitmap *wnd, size_t iw);
+static bool wnd_is_free_hlp(struct wnd_bitmap *wnd, size_t bit, size_t bits);
+
+static struct kmem_cache *ntfs_enode_cachep;
+
+int __init ntfs3_init_bitmap(void)
+{
+	ntfs_enode_cachep =
+		kmem_cache_create("ntfs3_enode_cache", sizeof(struct e_node), 0,
+				  SLAB_RECLAIM_ACCOUNT, NULL);
+	return ntfs_enode_cachep ? 0 : -ENOMEM;
+}
+
+void ntfs3_exit_bitmap(void)
+{
+	kmem_cache_destroy(ntfs_enode_cachep);
+}
+
+static inline u32 wnd_bits(const struct wnd_bitmap *wnd, size_t i)
+{
+	return i + 1 == wnd->nwnd ? wnd->bits_last : wnd->sb->s_blocksize * 8;
+}
+
+/*
+ * wnd_scan
+ *
+ * b_pos + b_len - biggest fragment.
+ * Scan range [wpos wbits) window @buf.
+ *
+ * Return: -1 if not found.
+ */
+static size_t wnd_scan(const ulong *buf, size_t wbit, u32 wpos, u32 wend,
+		       size_t to_alloc, size_t *prev_tail, size_t *b_pos,
+		       size_t *b_len)
+{
+	while (wpos < wend) {
+		size_t free_len;
+		u32 free_bits, end;
+		u32 used = find_next_zero_bit(buf, wend, wpos);
+
+		if (used >= wend) {
+			if (*b_len < *prev_tail) {
+				*b_pos = wbit - *prev_tail;
+				*b_len = *prev_tail;
+			}
+
+			*prev_tail = 0;
+			return -1;
+		}
+
+		if (used > wpos) {
+			wpos = used;
+			if (*b_len < *prev_tail) {
+				*b_pos = wbit - *prev_tail;
+				*b_len = *prev_tail;
+			}
+
+			*prev_tail = 0;
+		}
+
+		/*
+		 * Now we have a fragment [wpos, wend) staring with 0.
+		 */
+		end = wpos + to_alloc - *prev_tail;
+		free_bits = find_next_bit(buf, min(end, wend), wpos);
+
+		free_len = *prev_tail + free_bits - wpos;
+
+		if (*b_len < free_len) {
+			*b_pos = wbit + wpos - *prev_tail;
+			*b_len = free_len;
+		}
+
+		if (free_len >= to_alloc)
+			return wbit + wpos - *prev_tail;
+
+		if (free_bits >= wend) {
+			*prev_tail += free_bits - wpos;
+			return -1;
+		}
+
+		wpos = free_bits + 1;
+
+		*prev_tail = 0;
+	}
+
+	return -1;
+}
+
+/*
+ * wnd_close - Frees all resources.
+ */
+void wnd_close(struct wnd_bitmap *wnd)
+{
+	struct rb_node *node, *next;
+
+	kfree(wnd->free_bits);
+	run_close(&wnd->run);
+
+	node = rb_first(&wnd->start_tree);
+
+	while (node) {
+		next = rb_next(node);
+		rb_erase(node, &wnd->start_tree);
+		kmem_cache_free(ntfs_enode_cachep,
+				rb_entry(node, struct e_node, start.node));
+		node = next;
+	}
+}
+
+static struct rb_node *rb_lookup(struct rb_root *root, size_t v)
+{
+	struct rb_node **p = &root->rb_node;
+	struct rb_node *r = NULL;
+
+	while (*p) {
+		struct rb_node_key *k;
+
+		k = rb_entry(*p, struct rb_node_key, node);
+		if (v < k->key) {
+			p = &(*p)->rb_left;
+		} else if (v > k->key) {
+			r = &k->node;
+			p = &(*p)->rb_right;
+		} else {
+			return &k->node;
+		}
+	}
+
+	return r;
+}
+
+/*
+ * rb_insert_count - Helper function to insert special kind of 'count' tree.
+ */
+static inline bool rb_insert_count(struct rb_root *root, struct e_node *e)
+{
+	struct rb_node **p = &root->rb_node;
+	struct rb_node *parent = NULL;
+	size_t e_ckey = e->count.key;
+	size_t e_skey = e->start.key;
+
+	while (*p) {
+		struct e_node *k =
+			rb_entry(parent = *p, struct e_node, count.node);
+
+		if (e_ckey > k->count.key) {
+			p = &(*p)->rb_left;
+		} else if (e_ckey < k->count.key) {
+			p = &(*p)->rb_right;
+		} else if (e_skey < k->start.key) {
+			p = &(*p)->rb_left;
+		} else if (e_skey > k->start.key) {
+			p = &(*p)->rb_right;
+		} else {
+			WARN_ON(1);
+			return false;
+		}
+	}
+
+	rb_link_node(&e->count.node, parent, p);
+	rb_insert_color(&e->count.node, root);
+	return true;
+}
+
+/*
+ * rb_insert_start - Helper function to insert special kind of 'count' tree.
+ */
+static inline bool rb_insert_start(struct rb_root *root, struct e_node *e)
+{
+	struct rb_node **p = &root->rb_node;
+	struct rb_node *parent = NULL;
+	size_t e_skey = e->start.key;
+
+	while (*p) {
+		struct e_node *k;
+
+		parent = *p;
+
+		k = rb_entry(parent, struct e_node, start.node);
+		if (e_skey < k->start.key) {
+			p = &(*p)->rb_left;
+		} else if (e_skey > k->start.key) {
+			p = &(*p)->rb_right;
+		} else {
+			WARN_ON(1);
+			return false;
+		}
+	}
+
+	rb_link_node(&e->start.node, parent, p);
+	rb_insert_color(&e->start.node, root);
+	return true;
+}
+
+/*
+ * wnd_add_free_ext - Adds a new extent of free space.
+ * @build:	1 when building tree.
+ */
+static void wnd_add_free_ext(struct wnd_bitmap *wnd, size_t bit, size_t len,
+			     bool build)
+{
+	struct e_node *e, *e0 = NULL;
+	size_t ib, end_in = bit + len;
+	struct rb_node *n;
+
+	if (build) {
+		/* Use extent_min to filter too short extents. */
+		if (wnd->count >= NTFS_MAX_WND_EXTENTS &&
+		    len <= wnd->extent_min) {
+			wnd->uptodated = -1;
+			return;
+		}
+	} else {
+		/* Try to find extent before 'bit'. */
+		n = rb_lookup(&wnd->start_tree, bit);
+
+		if (!n) {
+			n = rb_first(&wnd->start_tree);
+		} else {
+			e = rb_entry(n, struct e_node, start.node);
+			n = rb_next(n);
+			if (e->start.key + e->count.key == bit) {
+				/* Remove left. */
+				bit = e->start.key;
+				len += e->count.key;
+				rb_erase(&e->start.node, &wnd->start_tree);
+				rb_erase(&e->count.node, &wnd->count_tree);
+				wnd->count -= 1;
+				e0 = e;
+			}
+		}
+
+		while (n) {
+			size_t next_end;
+
+			e = rb_entry(n, struct e_node, start.node);
+			next_end = e->start.key + e->count.key;
+			if (e->start.key > end_in)
+				break;
+
+			/* Remove right. */
+			n = rb_next(n);
+			len += next_end - end_in;
+			end_in = next_end;
+			rb_erase(&e->start.node, &wnd->start_tree);
+			rb_erase(&e->count.node, &wnd->count_tree);
+			wnd->count -= 1;
+
+			if (!e0)
+				e0 = e;
+			else
+				kmem_cache_free(ntfs_enode_cachep, e);
+		}
+
+		if (wnd->uptodated != 1) {
+			/* Check bits before 'bit'. */
+			ib = wnd->zone_bit == wnd->zone_end ||
+					     bit < wnd->zone_end
+				     ? 0
+				     : wnd->zone_end;
+
+			while (bit > ib && wnd_is_free_hlp(wnd, bit - 1, 1)) {
+				bit -= 1;
+				len += 1;
+			}
+
+			/* Check bits after 'end_in'. */
+			ib = wnd->zone_bit == wnd->zone_end ||
+					     end_in > wnd->zone_bit
+				     ? wnd->nbits
+				     : wnd->zone_bit;
+
+			while (end_in < ib && wnd_is_free_hlp(wnd, end_in, 1)) {
+				end_in += 1;
+				len += 1;
+			}
+		}
+	}
+	/* Insert new fragment. */
+	if (wnd->count >= NTFS_MAX_WND_EXTENTS) {
+		if (e0)
+			kmem_cache_free(ntfs_enode_cachep, e0);
+
+		wnd->uptodated = -1;
+
+		/* Compare with smallest fragment. */
+		n = rb_last(&wnd->count_tree);
+		e = rb_entry(n, struct e_node, count.node);
+		if (len <= e->count.key)
+			goto out; /* Do not insert small fragments. */
+
+		if (build) {
+			struct e_node *e2;
+
+			n = rb_prev(n);
+			e2 = rb_entry(n, struct e_node, count.node);
+			/* Smallest fragment will be 'e2->count.key'. */
+			wnd->extent_min = e2->count.key;
+		}
+
+		/* Replace smallest fragment by new one. */
+		rb_erase(&e->start.node, &wnd->start_tree);
+		rb_erase(&e->count.node, &wnd->count_tree);
+		wnd->count -= 1;
+	} else {
+		e = e0 ? e0 : kmem_cache_alloc(ntfs_enode_cachep, GFP_ATOMIC);
+		if (!e) {
+			wnd->uptodated = -1;
+			goto out;
+		}
+
+		if (build && len <= wnd->extent_min)
+			wnd->extent_min = len;
+	}
+	e->start.key = bit;
+	e->count.key = len;
+	if (len > wnd->extent_max)
+		wnd->extent_max = len;
+
+	rb_insert_start(&wnd->start_tree, e);
+	rb_insert_count(&wnd->count_tree, e);
+	wnd->count += 1;
+
+out:;
+}
+
+/*
+ * wnd_remove_free_ext - Remove a run from the cached free space.
+ */
+static void wnd_remove_free_ext(struct wnd_bitmap *wnd, size_t bit, size_t len)
+{
+	struct rb_node *n, *n3;
+	struct e_node *e, *e3;
+	size_t end_in = bit + len;
+	size_t end3, end, new_key, new_len, max_new_len;
+
+	/* Try to find extent before 'bit'. */
+	n = rb_lookup(&wnd->start_tree, bit);
+
+	if (!n)
+		return;
+
+	e = rb_entry(n, struct e_node, start.node);
+	end = e->start.key + e->count.key;
+
+	new_key = new_len = 0;
+	len = e->count.key;
+
+	/* Range [bit,end_in) must be inside 'e' or outside 'e' and 'n'. */
+	if (e->start.key > bit)
+		;
+	else if (end_in <= end) {
+		/* Range [bit,end_in) inside 'e'. */
+		new_key = end_in;
+		new_len = end - end_in;
+		len = bit - e->start.key;
+	} else if (bit > end) {
+		bool bmax = false;
+
+		n3 = rb_next(n);
+
+		while (n3) {
+			e3 = rb_entry(n3, struct e_node, start.node);
+			if (e3->start.key >= end_in)
+				break;
+
+			if (e3->count.key == wnd->extent_max)
+				bmax = true;
+
+			end3 = e3->start.key + e3->count.key;
+			if (end3 > end_in) {
+				e3->start.key = end_in;
+				rb_erase(&e3->count.node, &wnd->count_tree);
+				e3->count.key = end3 - end_in;
+				rb_insert_count(&wnd->count_tree, e3);
+				break;
+			}
+
+			n3 = rb_next(n3);
+			rb_erase(&e3->start.node, &wnd->start_tree);
+			rb_erase(&e3->count.node, &wnd->count_tree);
+			wnd->count -= 1;
+			kmem_cache_free(ntfs_enode_cachep, e3);
+		}
+		if (!bmax)
+			return;
+		n3 = rb_first(&wnd->count_tree);
+		wnd->extent_max =
+			n3 ? rb_entry(n3, struct e_node, count.node)->count.key
+			   : 0;
+		return;
+	}
+
+	if (e->count.key != wnd->extent_max) {
+		;
+	} else if (rb_prev(&e->count.node)) {
+		;
+	} else {
+		n3 = rb_next(&e->count.node);
+		max_new_len = len > new_len ? len : new_len;
+		if (!n3) {
+			wnd->extent_max = max_new_len;
+		} else {
+			e3 = rb_entry(n3, struct e_node, count.node);
+			wnd->extent_max = max(e3->count.key, max_new_len);
+		}
+	}
+
+	if (!len) {
+		if (new_len) {
+			e->start.key = new_key;
+			rb_erase(&e->count.node, &wnd->count_tree);
+			e->count.key = new_len;
+			rb_insert_count(&wnd->count_tree, e);
+		} else {
+			rb_erase(&e->start.node, &wnd->start_tree);
+			rb_erase(&e->count.node, &wnd->count_tree);
+			wnd->count -= 1;
+			kmem_cache_free(ntfs_enode_cachep, e);
+		}
+		goto out;
+	}
+	rb_erase(&e->count.node, &wnd->count_tree);
+	e->count.key = len;
+	rb_insert_count(&wnd->count_tree, e);
+
+	if (!new_len)
+		goto out;
+
+	if (wnd->count >= NTFS_MAX_WND_EXTENTS) {
+		wnd->uptodated = -1;
+
+		/* Get minimal extent. */
+		e = rb_entry(rb_last(&wnd->count_tree), struct e_node,
+			     count.node);
+		if (e->count.key > new_len)
+			goto out;
+
+		/* Replace minimum. */
+		rb_erase(&e->start.node, &wnd->start_tree);
+		rb_erase(&e->count.node, &wnd->count_tree);
+		wnd->count -= 1;
+	} else {
+		e = kmem_cache_alloc(ntfs_enode_cachep, GFP_ATOMIC);
+		if (!e)
+			wnd->uptodated = -1;
+	}
+
+	if (e) {
+		e->start.key = new_key;
+		e->count.key = new_len;
+		rb_insert_start(&wnd->start_tree, e);
+		rb_insert_count(&wnd->count_tree, e);
+		wnd->count += 1;
+	}
+
+out:
+	if (!wnd->count && 1 != wnd->uptodated)
+		wnd_rescan(wnd);
+}
+
+/*
+ * wnd_rescan - Scan all bitmap. Used while initialization.
+ */
+static int wnd_rescan(struct wnd_bitmap *wnd)
+{
+	int err = 0;
+	size_t prev_tail = 0;
+	struct super_block *sb = wnd->sb;
+	struct ntfs_sb_info *sbi = sb->s_fs_info;
+	u64 lbo, len = 0;
+	u32 blocksize = sb->s_blocksize;
+	u8 cluster_bits = sbi->cluster_bits;
+	u32 wbits = 8 * sb->s_blocksize;
+	u32 used, frb;
+	const ulong *buf;
+	size_t wpos, wbit, iw, vbo;
+	struct buffer_head *bh = NULL;
+	CLST lcn, clen;
+
+	wnd->uptodated = 0;
+	wnd->extent_max = 0;
+	wnd->extent_min = MINUS_ONE_T;
+	wnd->total_zeroes = 0;
+
+	vbo = 0;
+
+	for (iw = 0; iw < wnd->nwnd; iw++) {
+		if (iw + 1 == wnd->nwnd)
+			wbits = wnd->bits_last;
+
+		if (wnd->inited) {
+			if (!wnd->free_bits[iw]) {
+				/* All ones. */
+				if (prev_tail) {
+					wnd_add_free_ext(wnd,
+							 vbo * 8 - prev_tail,
+							 prev_tail, true);
+					prev_tail = 0;
+				}
+				goto next_wnd;
+			}
+			if (wbits == wnd->free_bits[iw]) {
+				/* All zeroes. */
+				prev_tail += wbits;
+				wnd->total_zeroes += wbits;
+				goto next_wnd;
+			}
+		}
+
+		if (!len) {
+			u32 off = vbo & sbi->cluster_mask;
+
+			if (!run_lookup_entry(&wnd->run, vbo >> cluster_bits,
+					      &lcn, &clen, NULL)) {
+				err = -ENOENT;
+				goto out;
+			}
+
+			lbo = ((u64)lcn << cluster_bits) + off;
+			len = ((u64)clen << cluster_bits) - off;
+		}
+
+		bh = ntfs_bread(sb, lbo >> sb->s_blocksize_bits);
+		if (!bh) {
+			err = -EIO;
+			goto out;
+		}
+
+		buf = (ulong *)bh->b_data;
+
+		used = __bitmap_weight(buf, wbits);
+		if (used < wbits) {
+			frb = wbits - used;
+			wnd->free_bits[iw] = frb;
+			wnd->total_zeroes += frb;
+		}
+
+		wpos = 0;
+		wbit = vbo * 8;
+
+		if (wbit + wbits > wnd->nbits)
+			wbits = wnd->nbits - wbit;
+
+		do {
+			used = find_next_zero_bit(buf, wbits, wpos);
+
+			if (used > wpos && prev_tail) {
+				wnd_add_free_ext(wnd, wbit + wpos - prev_tail,
+						 prev_tail, true);
+				prev_tail = 0;
+			}
+
+			wpos = used;
+
+			if (wpos >= wbits) {
+				/* No free blocks. */
+				prev_tail = 0;
+				break;
+			}
+
+			frb = find_next_bit(buf, wbits, wpos);
+			if (frb >= wbits) {
+				/* Keep last free block. */
+				prev_tail += frb - wpos;
+				break;
+			}
+
+			wnd_add_free_ext(wnd, wbit + wpos - prev_tail,
+					 frb + prev_tail - wpos, true);
+
+			/* Skip free block and first '1'. */
+			wpos = frb + 1;
+			/* Reset previous tail. */
+			prev_tail = 0;
+		} while (wpos < wbits);
+
+next_wnd:
+
+		if (bh)
+			put_bh(bh);
+		bh = NULL;
+
+		vbo += blocksize;
+		if (len) {
+			len -= blocksize;
+			lbo += blocksize;
+		}
+	}
+
+	/* Add last block. */
+	if (prev_tail)
+		wnd_add_free_ext(wnd, wnd->nbits - prev_tail, prev_tail, true);
+
+	/*
+	 * Before init cycle wnd->uptodated was 0.
+	 * If any errors or limits occurs while initialization then
+	 * wnd->uptodated will be -1.
+	 * If 'uptodated' is still 0 then Tree is really updated.
+	 */
+	if (!wnd->uptodated)
+		wnd->uptodated = 1;
+
+	if (wnd->zone_bit != wnd->zone_end) {
+		size_t zlen = wnd->zone_end - wnd->zone_bit;
+
+		wnd->zone_end = wnd->zone_bit;
+		wnd_zone_set(wnd, wnd->zone_bit, zlen);
+	}
+
+out:
+	return err;
+}
+
+int wnd_init(struct wnd_bitmap *wnd, struct super_block *sb, size_t nbits)
+{
+	int err;
+	u32 blocksize = sb->s_blocksize;
+	u32 wbits = blocksize * 8;
+
+	init_rwsem(&wnd->rw_lock);
+
+	wnd->sb = sb;
+	wnd->nbits = nbits;
+	wnd->total_zeroes = nbits;
+	wnd->extent_max = MINUS_ONE_T;
+	wnd->zone_bit = wnd->zone_end = 0;
+	wnd->nwnd = bytes_to_block(sb, bitmap_size(nbits));
+	wnd->bits_last = nbits & (wbits - 1);
+	if (!wnd->bits_last)
+		wnd->bits_last = wbits;
+
+	wnd->free_bits = kcalloc(wnd->nwnd, sizeof(u16), GFP_NOFS);
+	if (!wnd->free_bits)
+		return -ENOMEM;
+
+	err = wnd_rescan(wnd);
+	if (err)
+		return err;
+
+	wnd->inited = true;
+
+	return 0;
+}
+
+/*
+ * wnd_map - Call sb_bread for requested window.
+ */
+static struct buffer_head *wnd_map(struct wnd_bitmap *wnd, size_t iw)
+{
+	size_t vbo;
+	CLST lcn, clen;
+	struct super_block *sb = wnd->sb;
+	struct ntfs_sb_info *sbi;
+	struct buffer_head *bh;
+	u64 lbo;
+
+	sbi = sb->s_fs_info;
+	vbo = (u64)iw << sb->s_blocksize_bits;
+
+	if (!run_lookup_entry(&wnd->run, vbo >> sbi->cluster_bits, &lcn, &clen,
+			      NULL)) {
+		return ERR_PTR(-ENOENT);
+	}
+
+	lbo = ((u64)lcn << sbi->cluster_bits) + (vbo & sbi->cluster_mask);
+
+	bh = ntfs_bread(wnd->sb, lbo >> sb->s_blocksize_bits);
+	if (!bh)
+		return ERR_PTR(-EIO);
+
+	return bh;
+}
+
+/*
+ * wnd_set_free - Mark the bits range from bit to bit + bits as free.
+ */
+int wnd_set_free(struct wnd_bitmap *wnd, size_t bit, size_t bits)
+{
+	int err = 0;
+	struct super_block *sb = wnd->sb;
+	size_t bits0 = bits;
+	u32 wbits = 8 * sb->s_blocksize;
+	size_t iw = bit >> (sb->s_blocksize_bits + 3);
+	u32 wbit = bit & (wbits - 1);
+	struct buffer_head *bh;
+
+	while (iw < wnd->nwnd && bits) {
+		u32 tail, op;
+		ulong *buf;
+
+		if (iw + 1 == wnd->nwnd)
+			wbits = wnd->bits_last;
+
+		tail = wbits - wbit;
+		op = tail < bits ? tail : bits;
+
+		bh = wnd_map(wnd, iw);
+		if (IS_ERR(bh)) {
+			err = PTR_ERR(bh);
+			break;
+		}
+
+		buf = (ulong *)bh->b_data;
+
+		lock_buffer(bh);
+
+		__bitmap_clear(buf, wbit, op);
+
+		wnd->free_bits[iw] += op;
+
+		set_buffer_uptodate(bh);
+		mark_buffer_dirty(bh);
+		unlock_buffer(bh);
+		put_bh(bh);
+
+		wnd->total_zeroes += op;
+		bits -= op;
+		wbit = 0;
+		iw += 1;
+	}
+
+	wnd_add_free_ext(wnd, bit, bits0, false);
+
+	return err;
+}
+
+/*
+ * wnd_set_used - Mark the bits range from bit to bit + bits as used.
+ */
+int wnd_set_used(struct wnd_bitmap *wnd, size_t bit, size_t bits)
+{
+	int err = 0;
+	struct super_block *sb = wnd->sb;
+	size_t bits0 = bits;
+	size_t iw = bit >> (sb->s_blocksize_bits + 3);
+	u32 wbits = 8 * sb->s_blocksize;
+	u32 wbit = bit & (wbits - 1);
+	struct buffer_head *bh;
+
+	while (iw < wnd->nwnd && bits) {
+		u32 tail, op;
+		ulong *buf;
+
+		if (unlikely(iw + 1 == wnd->nwnd))
+			wbits = wnd->bits_last;
+
+		tail = wbits - wbit;
+		op = tail < bits ? tail : bits;
+
+		bh = wnd_map(wnd, iw);
+		if (IS_ERR(bh)) {
+			err = PTR_ERR(bh);
+			break;
+		}
+		buf = (ulong *)bh->b_data;
+
+		lock_buffer(bh);
+
+		__bitmap_set(buf, wbit, op);
+		wnd->free_bits[iw] -= op;
+
+		set_buffer_uptodate(bh);
+		mark_buffer_dirty(bh);
+		unlock_buffer(bh);
+		put_bh(bh);
+
+		wnd->total_zeroes -= op;
+		bits -= op;
+		wbit = 0;
+		iw += 1;
+	}
+
+	if (!RB_EMPTY_ROOT(&wnd->start_tree))
+		wnd_remove_free_ext(wnd, bit, bits0);
+
+	return err;
+}
+
+/*
+ * wnd_is_free_hlp
+ *
+ * Return: True if all clusters [bit, bit+bits) are free (bitmap only).
+ */
+static bool wnd_is_free_hlp(struct wnd_bitmap *wnd, size_t bit, size_t bits)
+{
+	struct super_block *sb = wnd->sb;
+	size_t iw = bit >> (sb->s_blocksize_bits + 3);
+	u32 wbits = 8 * sb->s_blocksize;
+	u32 wbit = bit & (wbits - 1);
+
+	while (iw < wnd->nwnd && bits) {
+		u32 tail, op;
+
+		if (unlikely(iw + 1 == wnd->nwnd))
+			wbits = wnd->bits_last;
+
+		tail = wbits - wbit;
+		op = tail < bits ? tail : bits;
+
+		if (wbits != wnd->free_bits[iw]) {
+			bool ret;
+			struct buffer_head *bh = wnd_map(wnd, iw);
+
+			if (IS_ERR(bh))
+				return false;
+
+			ret = are_bits_clear((ulong *)bh->b_data, wbit, op);
+
+			put_bh(bh);
+			if (!ret)
+				return false;
+		}
+
+		bits -= op;
+		wbit = 0;
+		iw += 1;
+	}
+
+	return true;
+}
+
+/*
+ * wnd_is_free
+ *
+ * Return: True if all clusters [bit, bit+bits) are free.
+ */
+bool wnd_is_free(struct wnd_bitmap *wnd, size_t bit, size_t bits)
+{
+	bool ret;
+	struct rb_node *n;
+	size_t end;
+	struct e_node *e;
+
+	if (RB_EMPTY_ROOT(&wnd->start_tree))
+		goto use_wnd;
+
+	n = rb_lookup(&wnd->start_tree, bit);
+	if (!n)
+		goto use_wnd;
+
+	e = rb_entry(n, struct e_node, start.node);
+
+	end = e->start.key + e->count.key;
+
+	if (bit < end && bit + bits <= end)
+		return true;
+
+use_wnd:
+	ret = wnd_is_free_hlp(wnd, bit, bits);
+
+	return ret;
+}
+
+/*
+ * wnd_is_used
+ *
+ * Return: True if all clusters [bit, bit+bits) are used.
+ */
+bool wnd_is_used(struct wnd_bitmap *wnd, size_t bit, size_t bits)
+{
+	bool ret = false;
+	struct super_block *sb = wnd->sb;
+	size_t iw = bit >> (sb->s_blocksize_bits + 3);
+	u32 wbits = 8 * sb->s_blocksize;
+	u32 wbit = bit & (wbits - 1);
+	size_t end;
+	struct rb_node *n;
+	struct e_node *e;
+
+	if (RB_EMPTY_ROOT(&wnd->start_tree))
+		goto use_wnd;
+
+	end = bit + bits;
+	n = rb_lookup(&wnd->start_tree, end - 1);
+	if (!n)
+		goto use_wnd;
+
+	e = rb_entry(n, struct e_node, start.node);
+	if (e->start.key + e->count.key > bit)
+		return false;
+
+use_wnd:
+	while (iw < wnd->nwnd && bits) {
+		u32 tail, op;
+
+		if (unlikely(iw + 1 == wnd->nwnd))
+			wbits = wnd->bits_last;
+
+		tail = wbits - wbit;
+		op = tail < bits ? tail : bits;
+
+		if (wnd->free_bits[iw]) {
+			bool ret;
+			struct buffer_head *bh = wnd_map(wnd, iw);
+
+			if (IS_ERR(bh))
+				goto out;
+
+			ret = are_bits_set((ulong *)bh->b_data, wbit, op);
+			put_bh(bh);
+			if (!ret)
+				goto out;
+		}
+
+		bits -= op;
+		wbit = 0;
+		iw += 1;
+	}
+	ret = true;
+
+out:
+	return ret;
+}
+
+/*
+ * wnd_find - Look for free space.
+ *
+ * - flags - BITMAP_FIND_XXX flags
+ *
+ * Return: 0 if not found.
+ */
+size_t wnd_find(struct wnd_bitmap *wnd, size_t to_alloc, size_t hint,
+		size_t flags, size_t *allocated)
+{
+	struct super_block *sb;
+	u32 wbits, wpos, wzbit, wzend;
+	size_t fnd, max_alloc, b_len, b_pos;
+	size_t iw, prev_tail, nwnd, wbit, ebit, zbit, zend;
+	size_t to_alloc0 = to_alloc;
+	const ulong *buf;
+	const struct e_node *e;
+	const struct rb_node *pr, *cr;
+	u8 log2_bits;
+	bool fbits_valid;
+	struct buffer_head *bh;
+
+	/* Fast checking for available free space. */
+	if (flags & BITMAP_FIND_FULL) {
+		size_t zeroes = wnd_zeroes(wnd);
+
+		zeroes -= wnd->zone_end - wnd->zone_bit;
+		if (zeroes < to_alloc0)
+			goto no_space;
+
+		if (to_alloc0 > wnd->extent_max)
+			goto no_space;
+	} else {
+		if (to_alloc > wnd->extent_max)
+			to_alloc = wnd->extent_max;
+	}
+
+	if (wnd->zone_bit <= hint && hint < wnd->zone_end)
+		hint = wnd->zone_end;
+
+	max_alloc = wnd->nbits;
+	b_len = b_pos = 0;
+
+	if (hint >= max_alloc)
+		hint = 0;
+
+	if (RB_EMPTY_ROOT(&wnd->start_tree)) {
+		if (wnd->uptodated == 1) {
+			/* Extents tree is updated -> No free space. */
+			goto no_space;
+		}
+		goto scan_bitmap;
+	}
+
+	e = NULL;
+	if (!hint)
+		goto allocate_biggest;
+
+	/* Use hint: Enumerate extents by start >= hint. */
+	pr = NULL;
+	cr = wnd->start_tree.rb_node;
+
+	for (;;) {
+		e = rb_entry(cr, struct e_node, start.node);
+
+		if (e->start.key == hint)
+			break;
+
+		if (e->start.key < hint) {
+			pr = cr;
+			cr = cr->rb_right;
+			if (!cr)
+				break;
+			continue;
+		}
+
+		cr = cr->rb_left;
+		if (!cr) {
+			e = pr ? rb_entry(pr, struct e_node, start.node) : NULL;
+			break;
+		}
+	}
+
+	if (!e)
+		goto allocate_biggest;
+
+	if (e->start.key + e->count.key > hint) {
+		/* We have found extension with 'hint' inside. */
+		size_t len = e->start.key + e->count.key - hint;
+
+		if (len >= to_alloc && hint + to_alloc <= max_alloc) {
+			fnd = hint;
+			goto found;
+		}
+
+		if (!(flags & BITMAP_FIND_FULL)) {
+			if (len > to_alloc)
+				len = to_alloc;
+
+			if (hint + len <= max_alloc) {
+				fnd = hint;
+				to_alloc = len;
+				goto found;
+			}
+		}
+	}
+
+allocate_biggest:
+	/* Allocate from biggest free extent. */
+	e = rb_entry(rb_first(&wnd->count_tree), struct e_node, count.node);
+	if (e->count.key != wnd->extent_max)
+		wnd->extent_max = e->count.key;
+
+	if (e->count.key < max_alloc) {
+		if (e->count.key >= to_alloc) {
+			;
+		} else if (flags & BITMAP_FIND_FULL) {
+			if (e->count.key < to_alloc0) {
+				/* Biggest free block is less then requested. */
+				goto no_space;
+			}
+			to_alloc = e->count.key;
+		} else if (-1 != wnd->uptodated) {
+			to_alloc = e->count.key;
+		} else {
+			/* Check if we can use more bits. */
+			size_t op, max_check;
+			struct rb_root start_tree;
+
+			memcpy(&start_tree, &wnd->start_tree,
+			       sizeof(struct rb_root));
+			memset(&wnd->start_tree, 0, sizeof(struct rb_root));
+
+			max_check = e->start.key + to_alloc;
+			if (max_check > max_alloc)
+				max_check = max_alloc;
+			for (op = e->start.key + e->count.key; op < max_check;
+			     op++) {
+				if (!wnd_is_free(wnd, op, 1))
+					break;
+			}
+			memcpy(&wnd->start_tree, &start_tree,
+			       sizeof(struct rb_root));
+			to_alloc = op - e->start.key;
+		}
+
+		/* Prepare to return. */
+		fnd = e->start.key;
+		if (e->start.key + to_alloc > max_alloc)
+			to_alloc = max_alloc - e->start.key;
+		goto found;
+	}
+
+	if (wnd->uptodated == 1) {
+		/* Extents tree is updated -> no free space. */
+		goto no_space;
+	}
+
+	b_len = e->count.key;
+	b_pos = e->start.key;
+
+scan_bitmap:
+	sb = wnd->sb;
+	log2_bits = sb->s_blocksize_bits + 3;
+
+	/* At most two ranges [hint, max_alloc) + [0, hint). */
+Again:
+
+	/* TODO: Optimize request for case nbits > wbits. */
+	iw = hint >> log2_bits;
+	wbits = sb->s_blocksize * 8;
+	wpos = hint & (wbits - 1);
+	prev_tail = 0;
+	fbits_valid = true;
+
+	if (max_alloc == wnd->nbits) {
+		nwnd = wnd->nwnd;
+	} else {
+		size_t t = max_alloc + wbits - 1;
+
+		nwnd = likely(t > max_alloc) ? (t >> log2_bits) : wnd->nwnd;
+	}
+
+	/* Enumerate all windows. */
+	for (; iw < nwnd; iw++) {
+		wbit = iw << log2_bits;
+
+		if (!wnd->free_bits[iw]) {
+			if (prev_tail > b_len) {
+				b_pos = wbit - prev_tail;
+				b_len = prev_tail;
+			}
+
+			/* Skip full used window. */
+			prev_tail = 0;
+			wpos = 0;
+			continue;
+		}
+
+		if (unlikely(iw + 1 == nwnd)) {
+			if (max_alloc == wnd->nbits) {
+				wbits = wnd->bits_last;
+			} else {
+				size_t t = max_alloc & (wbits - 1);
+
+				if (t) {
+					wbits = t;
+					fbits_valid = false;
+				}
+			}
+		}
+
+		if (wnd->zone_end > wnd->zone_bit) {
+			ebit = wbit + wbits;
+			zbit = max(wnd->zone_bit, wbit);
+			zend = min(wnd->zone_end, ebit);
+
+			/* Here we have a window [wbit, ebit) and zone [zbit, zend). */
+			if (zend <= zbit) {
+				/* Zone does not overlap window. */
+			} else {
+				wzbit = zbit - wbit;
+				wzend = zend - wbit;
+
+				/* Zone overlaps window. */
+				if (wnd->free_bits[iw] == wzend - wzbit) {
+					prev_tail = 0;
+					wpos = 0;
+					continue;
+				}
+
+				/* Scan two ranges window: [wbit, zbit) and [zend, ebit). */
+				bh = wnd_map(wnd, iw);
+
+				if (IS_ERR(bh)) {
+					/* TODO: Error */
+					prev_tail = 0;
+					wpos = 0;
+					continue;
+				}
+
+				buf = (ulong *)bh->b_data;
+
+				/* Scan range [wbit, zbit). */
+				if (wpos < wzbit) {
+					/* Scan range [wpos, zbit). */
+					fnd = wnd_scan(buf, wbit, wpos, wzbit,
+						       to_alloc, &prev_tail,
+						       &b_pos, &b_len);
+					if (fnd != MINUS_ONE_T) {
+						put_bh(bh);
+						goto found;
+					}
+				}
+
+				prev_tail = 0;
+
+				/* Scan range [zend, ebit). */
+				if (wzend < wbits) {
+					fnd = wnd_scan(buf, wbit,
+						       max(wzend, wpos), wbits,
+						       to_alloc, &prev_tail,
+						       &b_pos, &b_len);
+					if (fnd != MINUS_ONE_T) {
+						put_bh(bh);
+						goto found;
+					}
+				}
+
+				wpos = 0;
+				put_bh(bh);
+				continue;
+			}
+		}
+
+		/* Current window does not overlap zone. */
+		if (!wpos && fbits_valid && wnd->free_bits[iw] == wbits) {
+			/* Window is empty. */
+			if (prev_tail + wbits >= to_alloc) {
+				fnd = wbit + wpos - prev_tail;
+				goto found;
+			}
+
+			/* Increase 'prev_tail' and process next window. */
+			prev_tail += wbits;
+			wpos = 0;
+			continue;
+		}
+
+		/* Read window. */
+		bh = wnd_map(wnd, iw);
+		if (IS_ERR(bh)) {
+			// TODO: Error.
+			prev_tail = 0;
+			wpos = 0;
+			continue;
+		}
+
+		buf = (ulong *)bh->b_data;
+
+		/* Scan range [wpos, eBits). */
+		fnd = wnd_scan(buf, wbit, wpos, wbits, to_alloc, &prev_tail,
+			       &b_pos, &b_len);
+		put_bh(bh);
+		if (fnd != MINUS_ONE_T)
+			goto found;
+	}
+
+	if (b_len < prev_tail) {
+		/* The last fragment. */
+		b_len = prev_tail;
+		b_pos = max_alloc - prev_tail;
+	}
+
+	if (hint) {
+		/*
+		 * We have scanned range [hint max_alloc).
+		 * Prepare to scan range [0 hint + to_alloc).
+		 */
+		size_t nextmax = hint + to_alloc;
+
+		if (likely(nextmax >= hint) && nextmax < max_alloc)
+			max_alloc = nextmax;
+		hint = 0;
+		goto Again;
+	}
+
+	if (!b_len)
+		goto no_space;
+
+	wnd->extent_max = b_len;
+
+	if (flags & BITMAP_FIND_FULL)
+		goto no_space;
+
+	fnd = b_pos;
+	to_alloc = b_len;
+
+found:
+	if (flags & BITMAP_FIND_MARK_AS_USED) {
+		/* TODO: Optimize remove extent (pass 'e'?). */
+		if (wnd_set_used(wnd, fnd, to_alloc))
+			goto no_space;
+	} else if (wnd->extent_max != MINUS_ONE_T &&
+		   to_alloc > wnd->extent_max) {
+		wnd->extent_max = to_alloc;
+	}
+
+	*allocated = fnd;
+	return to_alloc;
+
+no_space:
+	return 0;
+}
+
+/*
+ * wnd_extend - Extend bitmap ($MFT bitmap).
+ */
+int wnd_extend(struct wnd_bitmap *wnd, size_t new_bits)
+{
+	int err;
+	struct super_block *sb = wnd->sb;
+	struct ntfs_sb_info *sbi = sb->s_fs_info;
+	u32 blocksize = sb->s_blocksize;
+	u32 wbits = blocksize * 8;
+	u32 b0, new_last;
+	size_t bits, iw, new_wnd;
+	size_t old_bits = wnd->nbits;
+	u16 *new_free;
+
+	if (new_bits <= old_bits)
+		return -EINVAL;
+
+	/* Align to 8 byte boundary. */
+	new_wnd = bytes_to_block(sb, bitmap_size(new_bits));
+	new_last = new_bits & (wbits - 1);
+	if (!new_last)
+		new_last = wbits;
+
+	if (new_wnd != wnd->nwnd) {
+		new_free = kmalloc(new_wnd * sizeof(u16), GFP_NOFS);
+		if (!new_free)
+			return -ENOMEM;
+
+		if (new_free != wnd->free_bits)
+			memcpy(new_free, wnd->free_bits,
+			       wnd->nwnd * sizeof(short));
+		memset(new_free + wnd->nwnd, 0,
+		       (new_wnd - wnd->nwnd) * sizeof(short));
+		kfree(wnd->free_bits);
+		wnd->free_bits = new_free;
+	}
+
+	/* Zero bits [old_bits,new_bits). */
+	bits = new_bits - old_bits;
+	b0 = old_bits & (wbits - 1);
+
+	for (iw = old_bits >> (sb->s_blocksize_bits + 3); bits; iw += 1) {
+		u32 op;
+		size_t frb;
+		u64 vbo, lbo, bytes;
+		struct buffer_head *bh;
+		ulong *buf;
+
+		if (iw + 1 == new_wnd)
+			wbits = new_last;
+
+		op = b0 + bits > wbits ? wbits - b0 : bits;
+		vbo = (u64)iw * blocksize;
+
+		err = ntfs_vbo_to_lbo(sbi, &wnd->run, vbo, &lbo, &bytes);
+		if (err)
+			break;
+
+		bh = ntfs_bread(sb, lbo >> sb->s_blocksize_bits);
+		if (!bh)
+			return -EIO;
+
+		lock_buffer(bh);
+		buf = (ulong *)bh->b_data;
+
+		__bitmap_clear(buf, b0, blocksize * 8 - b0);
+		frb = wbits - __bitmap_weight(buf, wbits);
+		wnd->total_zeroes += frb - wnd->free_bits[iw];
+		wnd->free_bits[iw] = frb;
+
+		set_buffer_uptodate(bh);
+		mark_buffer_dirty(bh);
+		unlock_buffer(bh);
+		/* err = sync_dirty_buffer(bh); */
+
+		b0 = 0;
+		bits -= op;
+	}
+
+	wnd->nbits = new_bits;
+	wnd->nwnd = new_wnd;
+	wnd->bits_last = new_last;
+
+	wnd_add_free_ext(wnd, old_bits, new_bits - old_bits, false);
+
+	return 0;
+}
+
+void wnd_zone_set(struct wnd_bitmap *wnd, size_t lcn, size_t len)
+{
+	size_t zlen;
+
+	zlen = wnd->zone_end - wnd->zone_bit;
+	if (zlen)
+		wnd_add_free_ext(wnd, wnd->zone_bit, zlen, false);
+
+	if (!RB_EMPTY_ROOT(&wnd->start_tree) && len)
+		wnd_remove_free_ext(wnd, lcn, len);
+
+	wnd->zone_bit = lcn;
+	wnd->zone_end = lcn + len;
+}
+
+int ntfs_trim_fs(struct ntfs_sb_info *sbi, struct fstrim_range *range)
+{
+	int err = 0;
+	struct super_block *sb = sbi->sb;
+	struct wnd_bitmap *wnd = &sbi->used.bitmap;
+	u32 wbits = 8 * sb->s_blocksize;
+	CLST len = 0, lcn = 0, done = 0;
+	CLST minlen = bytes_to_cluster(sbi, range->minlen);
+	CLST lcn_from = bytes_to_cluster(sbi, range->start);
+	size_t iw = lcn_from >> (sb->s_blocksize_bits + 3);
+	u32 wbit = lcn_from & (wbits - 1);
+	const ulong *buf;
+	CLST lcn_to;
+
+	if (!minlen)
+		minlen = 1;
+
+	if (range->len == (u64)-1)
+		lcn_to = wnd->nbits;
+	else
+		lcn_to = bytes_to_cluster(sbi, range->start + range->len);
+
+	down_read_nested(&wnd->rw_lock, BITMAP_MUTEX_CLUSTERS);
+
+	for (; iw < wnd->nbits; iw++, wbit = 0) {
+		CLST lcn_wnd = iw * wbits;
+		struct buffer_head *bh;
+
+		if (lcn_wnd > lcn_to)
+			break;
+
+		if (!wnd->free_bits[iw])
+			continue;
+
+		if (iw + 1 == wnd->nwnd)
+			wbits = wnd->bits_last;
+
+		if (lcn_wnd + wbits > lcn_to)
+			wbits = lcn_to - lcn_wnd;
+
+		bh = wnd_map(wnd, iw);
+		if (IS_ERR(bh)) {
+			err = PTR_ERR(bh);
+			break;
+		}
+
+		buf = (ulong *)bh->b_data;
+
+		for (; wbit < wbits; wbit++) {
+			if (!test_bit(wbit, buf)) {
+				if (!len)
+					lcn = lcn_wnd + wbit;
+				len += 1;
+				continue;
+			}
+			if (len >= minlen) {
+				err = ntfs_discard(sbi, lcn, len);
+				if (err)
+					goto out;
+				done += len;
+			}
+			len = 0;
+		}
+		put_bh(bh);
+	}
+
+	/* Process the last fragment. */
+	if (len >= minlen) {
+		err = ntfs_discard(sbi, lcn, len);
+		if (err)
+			goto out;
+		done += len;
+	}
+
+out:
+	range->len = (u64)done << sbi->cluster_bits;
+
+	up_read(&wnd->rw_lock);
+
+	return err;
+}
diff --git a/fs/ntfs3/debug.h b/fs/ntfs3/debug.h
new file mode 100644
index 000000000000..31120569a87b
--- /dev/null
+++ b/fs/ntfs3/debug.h
@@ -0,0 +1,52 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ *
+ * Copyright (C) 2019-2021 Paragon Software GmbH, All rights reserved.
+ *
+ * Useful functions for debugging.
+ *
+ */
+
+// clang-format off
+#ifndef _LINUX_NTFS3_DEBUG_H
+#define _LINUX_NTFS3_DEBUG_H
+
+#ifndef Add2Ptr
+#define Add2Ptr(P, I)		((void *)((u8 *)(P) + (I)))
+#define PtrOffset(B, O)		((size_t)((size_t)(O) - (size_t)(B)))
+#endif
+
+#ifdef CONFIG_PRINTK
+__printf(2, 3)
+void ntfs_printk(const struct super_block *sb, const char *fmt, ...);
+__printf(2, 3)
+void ntfs_inode_printk(struct inode *inode, const char *fmt, ...);
+#else
+static inline __printf(2, 3)
+void ntfs_printk(const struct super_block *sb, const char *fmt, ...)
+{
+}
+
+static inline __printf(2, 3)
+void ntfs_inode_printk(struct inode *inode, const char *fmt, ...)
+{
+}
+#endif
+
+/*
+ * Logging macros. Thanks Joe Perches <joe@perches.com> for implementation.
+ */
+
+#define ntfs_err(sb, fmt, ...)  ntfs_printk(sb, KERN_ERR fmt, ##__VA_ARGS__)
+#define ntfs_warn(sb, fmt, ...) ntfs_printk(sb, KERN_WARNING fmt, ##__VA_ARGS__)
+#define ntfs_info(sb, fmt, ...) ntfs_printk(sb, KERN_INFO fmt, ##__VA_ARGS__)
+#define ntfs_notice(sb, fmt, ...)                                              \
+	ntfs_printk(sb, KERN_NOTICE fmt, ##__VA_ARGS__)
+
+#define ntfs_inode_err(inode, fmt, ...)                                        \
+	ntfs_inode_printk(inode, KERN_ERR fmt, ##__VA_ARGS__)
+#define ntfs_inode_warn(inode, fmt, ...)                                       \
+	ntfs_inode_printk(inode, KERN_WARNING fmt, ##__VA_ARGS__)
+
+#endif /* _LINUX_NTFS3_DEBUG_H */
+// clang-format on
diff --git a/fs/ntfs3/dir.c b/fs/ntfs3/dir.c
new file mode 100644
index 000000000000..93f6d485564e
--- /dev/null
+++ b/fs/ntfs3/dir.c
@@ -0,0 +1,599 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ *
+ * Copyright (C) 2019-2021 Paragon Software GmbH, All rights reserved.
+ *
+ *  Directory handling functions for NTFS-based filesystems.
+ *
+ */
+
+#include <linux/blkdev.h>
+#include <linux/buffer_head.h>
+#include <linux/fs.h>
+#include <linux/iversion.h>
+#include <linux/nls.h>
+
+#include "debug.h"
+#include "ntfs.h"
+#include "ntfs_fs.h"
+
+/* Convert little endian UTF-16 to NLS string. */
+int ntfs_utf16_to_nls(struct ntfs_sb_info *sbi, const struct le_str *uni,
+		      u8 *buf, int buf_len)
+{
+	int ret, uni_len, warn;
+	const __le16 *ip;
+	u8 *op;
+	struct nls_table *nls = sbi->options.nls;
+
+	static_assert(sizeof(wchar_t) == sizeof(__le16));
+
+	if (!nls) {
+		/* UTF-16 -> UTF-8 */
+		ret = utf16s_to_utf8s((wchar_t *)uni->name, uni->len,
+				      UTF16_LITTLE_ENDIAN, buf, buf_len);
+		buf[ret] = '\0';
+		return ret;
+	}
+
+	ip = uni->name;
+	op = buf;
+	uni_len = uni->len;
+	warn = 0;
+
+	while (uni_len--) {
+		u16 ec;
+		int charlen;
+		char dump[5];
+
+		if (buf_len < NLS_MAX_CHARSET_SIZE) {
+			ntfs_warn(sbi->sb,
+				  "filename was truncated while converting.");
+			break;
+		}
+
+		ec = le16_to_cpu(*ip++);
+		charlen = nls->uni2char(ec, op, buf_len);
+
+		if (charlen > 0) {
+			op += charlen;
+			buf_len -= charlen;
+			continue;
+		}
+
+		*op++ = '_';
+		buf_len -= 1;
+		if (warn)
+			continue;
+
+		warn = 1;
+		hex_byte_pack(&dump[0], ec >> 8);
+		hex_byte_pack(&dump[2], ec);
+		dump[4] = 0;
+
+		ntfs_err(sbi->sb, "failed to convert \"%s\" to %s", dump,
+			 nls->charset);
+	}
+
+	*op = '\0';
+	return op - buf;
+}
+
+// clang-format off
+#define PLANE_SIZE	0x00010000
+
+#define SURROGATE_PAIR	0x0000d800
+#define SURROGATE_LOW	0x00000400
+#define SURROGATE_BITS	0x000003ff
+// clang-format on
+
+/*
+ * put_utf16 - Modified version of put_utf16 from fs/nls/nls_base.c
+ *
+ * Function is sparse warnings free.
+ */
+static inline void put_utf16(wchar_t *s, unsigned int c,
+			     enum utf16_endian endian)
+{
+	static_assert(sizeof(wchar_t) == sizeof(__le16));
+	static_assert(sizeof(wchar_t) == sizeof(__be16));
+
+	switch (endian) {
+	default:
+		*s = (wchar_t)c;
+		break;
+	case UTF16_LITTLE_ENDIAN:
+		*(__le16 *)s = __cpu_to_le16(c);
+		break;
+	case UTF16_BIG_ENDIAN:
+		*(__be16 *)s = __cpu_to_be16(c);
+		break;
+	}
+}
+
+/*
+ * _utf8s_to_utf16s
+ *
+ * Modified version of 'utf8s_to_utf16s' allows to
+ * detect -ENAMETOOLONG without writing out of expected maximum.
+ */
+static int _utf8s_to_utf16s(const u8 *s, int inlen, enum utf16_endian endian,
+			    wchar_t *pwcs, int maxout)
+{
+	u16 *op;
+	int size;
+	unicode_t u;
+
+	op = pwcs;
+	while (inlen > 0 && *s) {
+		if (*s & 0x80) {
+			size = utf8_to_utf32(s, inlen, &u);
+			if (size < 0)
+				return -EINVAL;
+			s += size;
+			inlen -= size;
+
+			if (u >= PLANE_SIZE) {
+				if (maxout < 2)
+					return -ENAMETOOLONG;
+
+				u -= PLANE_SIZE;
+				put_utf16(op++,
+					  SURROGATE_PAIR |
+						  ((u >> 10) & SURROGATE_BITS),
+					  endian);
+				put_utf16(op++,
+					  SURROGATE_PAIR | SURROGATE_LOW |
+						  (u & SURROGATE_BITS),
+					  endian);
+				maxout -= 2;
+			} else {
+				if (maxout < 1)
+					return -ENAMETOOLONG;
+
+				put_utf16(op++, u, endian);
+				maxout--;
+			}
+		} else {
+			if (maxout < 1)
+				return -ENAMETOOLONG;
+
+			put_utf16(op++, *s++, endian);
+			inlen--;
+			maxout--;
+		}
+	}
+	return op - pwcs;
+}
+
+/*
+ * ntfs_nls_to_utf16 - Convert input string to UTF-16.
+ * @name:	Input name.
+ * @name_len:	Input name length.
+ * @uni:	Destination memory.
+ * @max_ulen:	Destination memory.
+ * @endian:	Endian of target UTF-16 string.
+ *
+ * This function is called:
+ * - to create NTFS name
+ * - to create symlink
+ *
+ * Return: UTF-16 string length or error (if negative).
+ */
+int ntfs_nls_to_utf16(struct ntfs_sb_info *sbi, const u8 *name, u32 name_len,
+		      struct cpu_str *uni, u32 max_ulen,
+		      enum utf16_endian endian)
+{
+	int ret, slen;
+	const u8 *end;
+	struct nls_table *nls = sbi->options.nls;
+	u16 *uname = uni->name;
+
+	static_assert(sizeof(wchar_t) == sizeof(u16));
+
+	if (!nls) {
+		/* utf8 -> utf16 */
+		ret = _utf8s_to_utf16s(name, name_len, endian, uname, max_ulen);
+		uni->len = ret;
+		return ret;
+	}
+
+	for (ret = 0, end = name + name_len; name < end; ret++, name += slen) {
+		if (ret >= max_ulen)
+			return -ENAMETOOLONG;
+
+		slen = nls->char2uni(name, end - name, uname + ret);
+		if (!slen)
+			return -EINVAL;
+		if (slen < 0)
+			return slen;
+	}
+
+#ifdef __BIG_ENDIAN
+	if (endian == UTF16_LITTLE_ENDIAN) {
+		int i = ret;
+
+		while (i--) {
+			__cpu_to_le16s(uname);
+			uname++;
+		}
+	}
+#else
+	if (endian == UTF16_BIG_ENDIAN) {
+		int i = ret;
+
+		while (i--) {
+			__cpu_to_be16s(uname);
+			uname++;
+		}
+	}
+#endif
+
+	uni->len = ret;
+	return ret;
+}
+
+/*
+ * dir_search_u - Helper function.
+ */
+struct inode *dir_search_u(struct inode *dir, const struct cpu_str *uni,
+			   struct ntfs_fnd *fnd)
+{
+	int err = 0;
+	struct super_block *sb = dir->i_sb;
+	struct ntfs_sb_info *sbi = sb->s_fs_info;
+	struct ntfs_inode *ni = ntfs_i(dir);
+	struct NTFS_DE *e;
+	int diff;
+	struct inode *inode = NULL;
+	struct ntfs_fnd *fnd_a = NULL;
+
+	if (!fnd) {
+		fnd_a = fnd_get();
+		if (!fnd_a) {
+			err = -ENOMEM;
+			goto out;
+		}
+		fnd = fnd_a;
+	}
+
+	err = indx_find(&ni->dir, ni, NULL, uni, 0, sbi, &diff, &e, fnd);
+
+	if (err)
+		goto out;
+
+	if (diff) {
+		err = -ENOENT;
+		goto out;
+	}
+
+	inode = ntfs_iget5(sb, &e->ref, uni);
+	if (!IS_ERR(inode) && is_bad_inode(inode)) {
+		iput(inode);
+		err = -EINVAL;
+	}
+out:
+	fnd_put(fnd_a);
+
+	return err == -ENOENT ? NULL : err ? ERR_PTR(err) : inode;
+}
+
+static inline int ntfs_filldir(struct ntfs_sb_info *sbi, struct ntfs_inode *ni,
+			       const struct NTFS_DE *e, u8 *name,
+			       struct dir_context *ctx)
+{
+	const struct ATTR_FILE_NAME *fname;
+	unsigned long ino;
+	int name_len;
+	u32 dt_type;
+
+	fname = Add2Ptr(e, sizeof(struct NTFS_DE));
+
+	if (fname->type == FILE_NAME_DOS)
+		return 0;
+
+	if (!mi_is_ref(&ni->mi, &fname->home))
+		return 0;
+
+	ino = ino_get(&e->ref);
+
+	if (ino == MFT_REC_ROOT)
+		return 0;
+
+	/* Skip meta files. Unless option to show metafiles is set. */
+	if (!sbi->options.showmeta && ntfs_is_meta_file(sbi, ino))
+		return 0;
+
+	if (sbi->options.nohidden && (fname->dup.fa & FILE_ATTRIBUTE_HIDDEN))
+		return 0;
+
+	name_len = ntfs_utf16_to_nls(sbi, (struct le_str *)&fname->name_len,
+				     name, PATH_MAX);
+	if (name_len <= 0) {
+		ntfs_warn(sbi->sb, "failed to convert name for inode %lx.",
+			  ino);
+		return 0;
+	}
+
+	dt_type = (fname->dup.fa & FILE_ATTRIBUTE_DIRECTORY) ? DT_DIR : DT_REG;
+
+	return !dir_emit(ctx, (s8 *)name, name_len, ino, dt_type);
+}
+
+/*
+ * ntfs_read_hdr - Helper function for ntfs_readdir().
+ */
+static int ntfs_read_hdr(struct ntfs_sb_info *sbi, struct ntfs_inode *ni,
+			 const struct INDEX_HDR *hdr, u64 vbo, u64 pos,
+			 u8 *name, struct dir_context *ctx)
+{
+	int err;
+	const struct NTFS_DE *e;
+	u32 e_size;
+	u32 end = le32_to_cpu(hdr->used);
+	u32 off = le32_to_cpu(hdr->de_off);
+
+	for (;; off += e_size) {
+		if (off + sizeof(struct NTFS_DE) > end)
+			return -1;
+
+		e = Add2Ptr(hdr, off);
+		e_size = le16_to_cpu(e->size);
+		if (e_size < sizeof(struct NTFS_DE) || off + e_size > end)
+			return -1;
+
+		if (de_is_last(e))
+			return 0;
+
+		/* Skip already enumerated. */
+		if (vbo + off < pos)
+			continue;
+
+		if (le16_to_cpu(e->key_size) < SIZEOF_ATTRIBUTE_FILENAME)
+			return -1;
+
+		ctx->pos = vbo + off;
+
+		/* Submit the name to the filldir callback. */
+		err = ntfs_filldir(sbi, ni, e, name, ctx);
+		if (err)
+			return err;
+	}
+}
+
+/*
+ * ntfs_readdir - file_operations::iterate_shared
+ *
+ * Use non sorted enumeration.
+ * We have an example of broken volume where sorted enumeration
+ * counts each name twice.
+ */
+static int ntfs_readdir(struct file *file, struct dir_context *ctx)
+{
+	const struct INDEX_ROOT *root;
+	u64 vbo;
+	size_t bit;
+	loff_t eod;
+	int err = 0;
+	struct inode *dir = file_inode(file);
+	struct ntfs_inode *ni = ntfs_i(dir);
+	struct super_block *sb = dir->i_sb;
+	struct ntfs_sb_info *sbi = sb->s_fs_info;
+	loff_t i_size = i_size_read(dir);
+	u32 pos = ctx->pos;
+	u8 *name = NULL;
+	struct indx_node *node = NULL;
+	u8 index_bits = ni->dir.index_bits;
+
+	/* Name is a buffer of PATH_MAX length. */
+	static_assert(NTFS_NAME_LEN * 4 < PATH_MAX);
+
+	eod = i_size + sbi->record_size;
+
+	if (pos >= eod)
+		return 0;
+
+	if (!dir_emit_dots(file, ctx))
+		return 0;
+
+	/* Allocate PATH_MAX bytes. */
+	name = __getname();
+	if (!name)
+		return -ENOMEM;
+
+	if (!ni->mi_loaded && ni->attr_list.size) {
+		/*
+		 * Directory inode is locked for read.
+		 * Load all subrecords to avoid 'write' access to 'ni' during
+		 * directory reading.
+		 */
+		ni_lock(ni);
+		if (!ni->mi_loaded && ni->attr_list.size) {
+			err = ni_load_all_mi(ni);
+			if (!err)
+				ni->mi_loaded = true;
+		}
+		ni_unlock(ni);
+		if (err)
+			goto out;
+	}
+
+	root = indx_get_root(&ni->dir, ni, NULL, NULL);
+	if (!root) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	if (pos >= sbi->record_size) {
+		bit = (pos - sbi->record_size) >> index_bits;
+	} else {
+		err = ntfs_read_hdr(sbi, ni, &root->ihdr, 0, pos, name, ctx);
+		if (err)
+			goto out;
+		bit = 0;
+	}
+
+	if (!i_size) {
+		ctx->pos = eod;
+		goto out;
+	}
+
+	for (;;) {
+		vbo = (u64)bit << index_bits;
+		if (vbo >= i_size) {
+			ctx->pos = eod;
+			goto out;
+		}
+
+		err = indx_used_bit(&ni->dir, ni, &bit);
+		if (err)
+			goto out;
+
+		if (bit == MINUS_ONE_T) {
+			ctx->pos = eod;
+			goto out;
+		}
+
+		vbo = (u64)bit << index_bits;
+		if (vbo >= i_size) {
+			ntfs_inode_err(dir, "Looks like your dir is corrupt");
+			err = -EINVAL;
+			goto out;
+		}
+
+		err = indx_read(&ni->dir, ni, bit << ni->dir.idx2vbn_bits,
+				&node);
+		if (err)
+			goto out;
+
+		err = ntfs_read_hdr(sbi, ni, &node->index->ihdr,
+				    vbo + sbi->record_size, pos, name, ctx);
+		if (err)
+			goto out;
+
+		bit += 1;
+	}
+
+out:
+
+	__putname(name);
+	put_indx_node(node);
+
+	if (err == -ENOENT) {
+		err = 0;
+		ctx->pos = pos;
+	}
+
+	return err;
+}
+
+static int ntfs_dir_count(struct inode *dir, bool *is_empty, size_t *dirs,
+			  size_t *files)
+{
+	int err = 0;
+	struct ntfs_inode *ni = ntfs_i(dir);
+	struct NTFS_DE *e = NULL;
+	struct INDEX_ROOT *root;
+	struct INDEX_HDR *hdr;
+	const struct ATTR_FILE_NAME *fname;
+	u32 e_size, off, end;
+	u64 vbo = 0;
+	size_t drs = 0, fles = 0, bit = 0;
+	loff_t i_size = ni->vfs_inode.i_size;
+	struct indx_node *node = NULL;
+	u8 index_bits = ni->dir.index_bits;
+
+	if (is_empty)
+		*is_empty = true;
+
+	root = indx_get_root(&ni->dir, ni, NULL, NULL);
+	if (!root)
+		return -EINVAL;
+
+	hdr = &root->ihdr;
+
+	for (;;) {
+		end = le32_to_cpu(hdr->used);
+		off = le32_to_cpu(hdr->de_off);
+
+		for (; off + sizeof(struct NTFS_DE) <= end; off += e_size) {
+			e = Add2Ptr(hdr, off);
+			e_size = le16_to_cpu(e->size);
+			if (e_size < sizeof(struct NTFS_DE) ||
+			    off + e_size > end)
+				break;
+
+			if (de_is_last(e))
+				break;
+
+			fname = de_get_fname(e);
+			if (!fname)
+				continue;
+
+			if (fname->type == FILE_NAME_DOS)
+				continue;
+
+			if (is_empty) {
+				*is_empty = false;
+				if (!dirs && !files)
+					goto out;
+			}
+
+			if (fname->dup.fa & FILE_ATTRIBUTE_DIRECTORY)
+				drs += 1;
+			else
+				fles += 1;
+		}
+
+		if (vbo >= i_size)
+			goto out;
+
+		err = indx_used_bit(&ni->dir, ni, &bit);
+		if (err)
+			goto out;
+
+		if (bit == MINUS_ONE_T)
+			goto out;
+
+		vbo = (u64)bit << index_bits;
+		if (vbo >= i_size)
+			goto out;
+
+		err = indx_read(&ni->dir, ni, bit << ni->dir.idx2vbn_bits,
+				&node);
+		if (err)
+			goto out;
+
+		hdr = &node->index->ihdr;
+		bit += 1;
+		vbo = (u64)bit << ni->dir.idx2vbn_bits;
+	}
+
+out:
+	put_indx_node(node);
+	if (dirs)
+		*dirs = drs;
+	if (files)
+		*files = fles;
+
+	return err;
+}
+
+bool dir_is_empty(struct inode *dir)
+{
+	bool is_empty = false;
+
+	ntfs_dir_count(dir, &is_empty, NULL, NULL);
+
+	return is_empty;
+}
+
+// clang-format off
+const struct file_operations ntfs_dir_operations = {
+	.llseek		= generic_file_llseek,
+	.read		= generic_read_dir,
+	.iterate_shared	= ntfs_readdir,
+	.fsync		= generic_file_fsync,
+	.open		= ntfs_file_open,
+};
+// clang-format on
diff --git a/fs/ntfs3/file.c b/fs/ntfs3/file.c
new file mode 100644
index 000000000000..424450e77ad5
--- /dev/null
+++ b/fs/ntfs3/file.c
@@ -0,0 +1,1251 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ *
+ * Copyright (C) 2019-2021 Paragon Software GmbH, All rights reserved.
+ *
+ *  Regular file handling primitives for NTFS-based filesystems.
+ *
+ */
+
+#include <linux/backing-dev.h>
+#include <linux/buffer_head.h>
+#include <linux/compat.h>
+#include <linux/falloc.h>
+#include <linux/fiemap.h>
+#include <linux/nls.h>
+
+#include "debug.h"
+#include "ntfs.h"
+#include "ntfs_fs.h"
+
+static int ntfs_ioctl_fitrim(struct ntfs_sb_info *sbi, unsigned long arg)
+{
+	struct fstrim_range __user *user_range;
+	struct fstrim_range range;
+	struct request_queue *q = bdev_get_queue(sbi->sb->s_bdev);
+	int err;
+
+	if (!capable(CAP_SYS_ADMIN))
+		return -EPERM;
+
+	if (!blk_queue_discard(q))
+		return -EOPNOTSUPP;
+
+	user_range = (struct fstrim_range __user *)arg;
+	if (copy_from_user(&range, user_range, sizeof(range)))
+		return -EFAULT;
+
+	range.minlen = max_t(u32, range.minlen, q->limits.discard_granularity);
+
+	err = ntfs_trim_fs(sbi, &range);
+	if (err < 0)
+		return err;
+
+	if (copy_to_user(user_range, &range, sizeof(range)))
+		return -EFAULT;
+
+	return 0;
+}
+
+static long ntfs_ioctl(struct file *filp, u32 cmd, unsigned long arg)
+{
+	struct inode *inode = file_inode(filp);
+	struct ntfs_sb_info *sbi = inode->i_sb->s_fs_info;
+
+	switch (cmd) {
+	case FITRIM:
+		return ntfs_ioctl_fitrim(sbi, arg);
+	}
+	return -ENOTTY; /* Inappropriate ioctl for device. */
+}
+
+#ifdef CONFIG_COMPAT
+static long ntfs_compat_ioctl(struct file *filp, u32 cmd, unsigned long arg)
+
+{
+	return ntfs_ioctl(filp, cmd, (unsigned long)compat_ptr(arg));
+}
+#endif
+
+/*
+ * ntfs_getattr - inode_operations::getattr
+ */
+int ntfs_getattr(struct user_namespace *mnt_userns, const struct path *path,
+		 struct kstat *stat, u32 request_mask, u32 flags)
+{
+	struct inode *inode = d_inode(path->dentry);
+	struct ntfs_inode *ni = ntfs_i(inode);
+
+	if (is_compressed(ni))
+		stat->attributes |= STATX_ATTR_COMPRESSED;
+
+	if (is_encrypted(ni))
+		stat->attributes |= STATX_ATTR_ENCRYPTED;
+
+	stat->attributes_mask |= STATX_ATTR_COMPRESSED | STATX_ATTR_ENCRYPTED;
+
+	generic_fillattr(mnt_userns, inode, stat);
+
+	stat->result_mask |= STATX_BTIME;
+	stat->btime = ni->i_crtime;
+	stat->blksize = ni->mi.sbi->cluster_size; /* 512, 1K, ..., 2M */
+
+	return 0;
+}
+
+static int ntfs_extend_initialized_size(struct file *file,
+					struct ntfs_inode *ni,
+					const loff_t valid,
+					const loff_t new_valid)
+{
+	struct inode *inode = &ni->vfs_inode;
+	struct address_space *mapping = inode->i_mapping;
+	struct ntfs_sb_info *sbi = inode->i_sb->s_fs_info;
+	loff_t pos = valid;
+	int err;
+
+	if (is_resident(ni)) {
+		ni->i_valid = new_valid;
+		return 0;
+	}
+
+	WARN_ON(is_compressed(ni));
+	WARN_ON(valid >= new_valid);
+
+	for (;;) {
+		u32 zerofrom, len;
+		struct page *page;
+		void *fsdata;
+		u8 bits;
+		CLST vcn, lcn, clen;
+
+		if (is_sparsed(ni)) {
+			bits = sbi->cluster_bits;
+			vcn = pos >> bits;
+
+			err = attr_data_get_block(ni, vcn, 0, &lcn, &clen,
+						  NULL);
+			if (err)
+				goto out;
+
+			if (lcn == SPARSE_LCN) {
+				loff_t vbo = (loff_t)vcn << bits;
+				loff_t to = vbo + ((loff_t)clen << bits);
+
+				if (to <= new_valid) {
+					ni->i_valid = to;
+					pos = to;
+					goto next;
+				}
+
+				if (vbo < pos) {
+					pos = vbo;
+				} else {
+					to = (new_valid >> bits) << bits;
+					if (pos < to) {
+						ni->i_valid = to;
+						pos = to;
+						goto next;
+					}
+				}
+			}
+		}
+
+		zerofrom = pos & (PAGE_SIZE - 1);
+		len = PAGE_SIZE - zerofrom;
+
+		if (pos + len > new_valid)
+			len = new_valid - pos;
+
+		err = pagecache_write_begin(file, mapping, pos, len, 0, &page,
+					    &fsdata);
+		if (err)
+			goto out;
+
+		zero_user_segment(page, zerofrom, PAGE_SIZE);
+
+		/* This function in any case puts page. */
+		err = pagecache_write_end(file, mapping, pos, len, len, page,
+					  fsdata);
+		if (err < 0)
+			goto out;
+		pos += len;
+
+next:
+		if (pos >= new_valid)
+			break;
+
+		balance_dirty_pages_ratelimited(mapping);
+		cond_resched();
+	}
+
+	return 0;
+
+out:
+	ni->i_valid = valid;
+	ntfs_inode_warn(inode, "failed to extend initialized size to %llx.",
+			new_valid);
+	return err;
+}
+
+/*
+ * ntfs_zero_range - Helper function for punch_hole.
+ *
+ * It zeroes a range [vbo, vbo_to).
+ */
+static int ntfs_zero_range(struct inode *inode, u64 vbo, u64 vbo_to)
+{
+	int err = 0;
+	struct address_space *mapping = inode->i_mapping;
+	u32 blocksize = 1 << inode->i_blkbits;
+	pgoff_t idx = vbo >> PAGE_SHIFT;
+	u32 z_start = vbo & (PAGE_SIZE - 1);
+	pgoff_t idx_end = (vbo_to + PAGE_SIZE - 1) >> PAGE_SHIFT;
+	loff_t page_off;
+	struct buffer_head *head, *bh;
+	u32 bh_next, bh_off, z_end;
+	sector_t iblock;
+	struct page *page;
+
+	for (; idx < idx_end; idx += 1, z_start = 0) {
+		page_off = (loff_t)idx << PAGE_SHIFT;
+		z_end = (page_off + PAGE_SIZE) > vbo_to ? (vbo_to - page_off)
+							: PAGE_SIZE;
+		iblock = page_off >> inode->i_blkbits;
+
+		page = find_or_create_page(mapping, idx,
+					   mapping_gfp_constraint(mapping,
+								  ~__GFP_FS));
+		if (!page)
+			return -ENOMEM;
+
+		if (!page_has_buffers(page))
+			create_empty_buffers(page, blocksize, 0);
+
+		bh = head = page_buffers(page);
+		bh_off = 0;
+		do {
+			bh_next = bh_off + blocksize;
+
+			if (bh_next <= z_start || bh_off >= z_end)
+				continue;
+
+			if (!buffer_mapped(bh)) {
+				ntfs_get_block(inode, iblock, bh, 0);
+				/* Unmapped? It's a hole - nothing to do. */
+				if (!buffer_mapped(bh))
+					continue;
+			}
+
+			/* Ok, it's mapped. Make sure it's up-to-date. */
+			if (PageUptodate(page))
+				set_buffer_uptodate(bh);
+
+			if (!buffer_uptodate(bh)) {
+				lock_buffer(bh);
+				bh->b_end_io = end_buffer_read_sync;
+				get_bh(bh);
+				submit_bh(REQ_OP_READ, 0, bh);
+
+				wait_on_buffer(bh);
+				if (!buffer_uptodate(bh)) {
+					unlock_page(page);
+					put_page(page);
+					err = -EIO;
+					goto out;
+				}
+			}
+
+			mark_buffer_dirty(bh);
+
+		} while (bh_off = bh_next, iblock += 1,
+			 head != (bh = bh->b_this_page));
+
+		zero_user_segment(page, z_start, z_end);
+
+		unlock_page(page);
+		put_page(page);
+		cond_resched();
+	}
+out:
+	mark_inode_dirty(inode);
+	return err;
+}
+
+/*
+ * ntfs_sparse_cluster - Helper function to zero a new allocated clusters.
+ *
+ * NOTE: 512 <= cluster size <= 2M
+ */
+void ntfs_sparse_cluster(struct inode *inode, struct page *page0, CLST vcn,
+			 CLST len)
+{
+	struct address_space *mapping = inode->i_mapping;
+	struct ntfs_sb_info *sbi = inode->i_sb->s_fs_info;
+	u64 vbo = (u64)vcn << sbi->cluster_bits;
+	u64 bytes = (u64)len << sbi->cluster_bits;
+	u32 blocksize = 1 << inode->i_blkbits;
+	pgoff_t idx0 = page0 ? page0->index : -1;
+	loff_t vbo_clst = vbo & sbi->cluster_mask_inv;
+	loff_t end = ntfs_up_cluster(sbi, vbo + bytes);
+	pgoff_t idx = vbo_clst >> PAGE_SHIFT;
+	u32 from = vbo_clst & (PAGE_SIZE - 1);
+	pgoff_t idx_end = (end + PAGE_SIZE - 1) >> PAGE_SHIFT;
+	loff_t page_off;
+	u32 to;
+	bool partial;
+	struct page *page;
+
+	for (; idx < idx_end; idx += 1, from = 0) {
+		page = idx == idx0 ? page0 : grab_cache_page(mapping, idx);
+
+		if (!page)
+			continue;
+
+		page_off = (loff_t)idx << PAGE_SHIFT;
+		to = (page_off + PAGE_SIZE) > end ? (end - page_off)
+						  : PAGE_SIZE;
+		partial = false;
+
+		if ((from || PAGE_SIZE != to) &&
+		    likely(!page_has_buffers(page))) {
+			create_empty_buffers(page, blocksize, 0);
+		}
+
+		if (page_has_buffers(page)) {
+			struct buffer_head *head, *bh;
+			u32 bh_off = 0;
+
+			bh = head = page_buffers(page);
+			do {
+				u32 bh_next = bh_off + blocksize;
+
+				if (from <= bh_off && bh_next <= to) {
+					set_buffer_uptodate(bh);
+					mark_buffer_dirty(bh);
+				} else if (!buffer_uptodate(bh)) {
+					partial = true;
+				}
+				bh_off = bh_next;
+			} while (head != (bh = bh->b_this_page));
+		}
+
+		zero_user_segment(page, from, to);
+
+		if (!partial) {
+			if (!PageUptodate(page))
+				SetPageUptodate(page);
+			set_page_dirty(page);
+		}
+
+		if (idx != idx0) {
+			unlock_page(page);
+			put_page(page);
+		}
+		cond_resched();
+	}
+	mark_inode_dirty(inode);
+}
+
+/*
+ * ntfs_file_mmap - file_operations::mmap
+ */
+static int ntfs_file_mmap(struct file *file, struct vm_area_struct *vma)
+{
+	struct address_space *mapping = file->f_mapping;
+	struct inode *inode = mapping->host;
+	struct ntfs_inode *ni = ntfs_i(inode);
+	u64 from = ((u64)vma->vm_pgoff << PAGE_SHIFT);
+	bool rw = vma->vm_flags & VM_WRITE;
+	int err;
+
+	if (is_encrypted(ni)) {
+		ntfs_inode_warn(inode, "mmap encrypted not supported");
+		return -EOPNOTSUPP;
+	}
+
+	if (is_dedup(ni)) {
+		ntfs_inode_warn(inode, "mmap deduplicated not supported");
+		return -EOPNOTSUPP;
+	}
+
+	if (is_compressed(ni) && rw) {
+		ntfs_inode_warn(inode, "mmap(write) compressed not supported");
+		return -EOPNOTSUPP;
+	}
+
+	if (rw) {
+		u64 to = min_t(loff_t, i_size_read(inode),
+			       from + vma->vm_end - vma->vm_start);
+
+		if (is_sparsed(ni)) {
+			/* Allocate clusters for rw map. */
+			struct ntfs_sb_info *sbi = inode->i_sb->s_fs_info;
+			CLST lcn, len;
+			CLST vcn = from >> sbi->cluster_bits;
+			CLST end = bytes_to_cluster(sbi, to);
+			bool new;
+
+			for (; vcn < end; vcn += len) {
+				err = attr_data_get_block(ni, vcn, 1, &lcn,
+							  &len, &new);
+				if (err)
+					goto out;
+
+				if (!new)
+					continue;
+				ntfs_sparse_cluster(inode, NULL, vcn, 1);
+			}
+		}
+
+		if (ni->i_valid < to) {
+			if (!inode_trylock(inode)) {
+				err = -EAGAIN;
+				goto out;
+			}
+			err = ntfs_extend_initialized_size(file, ni,
+							   ni->i_valid, to);
+			inode_unlock(inode);
+			if (err)
+				goto out;
+		}
+	}
+
+	err = generic_file_mmap(file, vma);
+out:
+	return err;
+}
+
+static int ntfs_extend(struct inode *inode, loff_t pos, size_t count,
+		       struct file *file)
+{
+	struct ntfs_inode *ni = ntfs_i(inode);
+	struct address_space *mapping = inode->i_mapping;
+	loff_t end = pos + count;
+	bool extend_init = file && pos > ni->i_valid;
+	int err;
+
+	if (end <= inode->i_size && !extend_init)
+		return 0;
+
+	/* Mark rw ntfs as dirty. It will be cleared at umount. */
+	ntfs_set_state(ni->mi.sbi, NTFS_DIRTY_DIRTY);
+
+	if (end > inode->i_size) {
+		err = ntfs_set_size(inode, end);
+		if (err)
+			goto out;
+		inode->i_size = end;
+	}
+
+	if (extend_init && !is_compressed(ni)) {
+		err = ntfs_extend_initialized_size(file, ni, ni->i_valid, pos);
+		if (err)
+			goto out;
+	} else {
+		err = 0;
+	}
+
+	inode->i_ctime = inode->i_mtime = current_time(inode);
+	mark_inode_dirty(inode);
+
+	if (IS_SYNC(inode)) {
+		int err2;
+
+		err = filemap_fdatawrite_range(mapping, pos, end - 1);
+		err2 = sync_mapping_buffers(mapping);
+		if (!err)
+			err = err2;
+		err2 = write_inode_now(inode, 1);
+		if (!err)
+			err = err2;
+		if (!err)
+			err = filemap_fdatawait_range(mapping, pos, end - 1);
+	}
+
+out:
+	return err;
+}
+
+static int ntfs_truncate(struct inode *inode, loff_t new_size)
+{
+	struct super_block *sb = inode->i_sb;
+	struct ntfs_inode *ni = ntfs_i(inode);
+	int err, dirty = 0;
+	u64 new_valid;
+
+	if (!S_ISREG(inode->i_mode))
+		return 0;
+
+	if (is_compressed(ni)) {
+		if (ni->i_valid > new_size)
+			ni->i_valid = new_size;
+	} else {
+		err = block_truncate_page(inode->i_mapping, new_size,
+					  ntfs_get_block);
+		if (err)
+			return err;
+	}
+
+	new_valid = ntfs_up_block(sb, min_t(u64, ni->i_valid, new_size));
+
+	ni_lock(ni);
+
+	truncate_setsize(inode, new_size);
+
+	down_write(&ni->file.run_lock);
+	err = attr_set_size(ni, ATTR_DATA, NULL, 0, &ni->file.run, new_size,
+			    &new_valid, true, NULL);
+	up_write(&ni->file.run_lock);
+
+	if (new_valid < ni->i_valid)
+		ni->i_valid = new_valid;
+
+	ni_unlock(ni);
+
+	ni->std_fa |= FILE_ATTRIBUTE_ARCHIVE;
+	inode->i_ctime = inode->i_mtime = current_time(inode);
+	if (!IS_DIRSYNC(inode)) {
+		dirty = 1;
+	} else {
+		err = ntfs_sync_inode(inode);
+		if (err)
+			return err;
+	}
+
+	if (dirty)
+		mark_inode_dirty(inode);
+
+	/*ntfs_flush_inodes(inode->i_sb, inode, NULL);*/
+
+	return 0;
+}
+
+/*
+ * ntfs_fallocate
+ *
+ * Preallocate space for a file. This implements ntfs's fallocate file
+ * operation, which gets called from sys_fallocate system call. User
+ * space requests 'len' bytes at 'vbo'. If FALLOC_FL_KEEP_SIZE is set
+ * we just allocate clusters without zeroing them out. Otherwise we
+ * allocate and zero out clusters via an expanding truncate.
+ */
+static long ntfs_fallocate(struct file *file, int mode, loff_t vbo, loff_t len)
+{
+	struct inode *inode = file->f_mapping->host;
+	struct super_block *sb = inode->i_sb;
+	struct ntfs_sb_info *sbi = sb->s_fs_info;
+	struct ntfs_inode *ni = ntfs_i(inode);
+	loff_t end = vbo + len;
+	loff_t vbo_down = round_down(vbo, PAGE_SIZE);
+	loff_t i_size;
+	int err;
+
+	/* No support for dir. */
+	if (!S_ISREG(inode->i_mode))
+		return -EOPNOTSUPP;
+
+	/* Return error if mode is not supported. */
+	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
+		     FALLOC_FL_COLLAPSE_RANGE)) {
+		ntfs_inode_warn(inode, "fallocate(0x%x) is not supported",
+				mode);
+		return -EOPNOTSUPP;
+	}
+
+	ntfs_set_state(sbi, NTFS_DIRTY_DIRTY);
+
+	inode_lock(inode);
+	i_size = inode->i_size;
+
+	if (WARN_ON(ni->ni_flags & NI_FLAG_COMPRESSED_MASK)) {
+		/* Should never be here, see ntfs_file_open. */
+		err = -EOPNOTSUPP;
+		goto out;
+	}
+
+	if (mode & FALLOC_FL_PUNCH_HOLE) {
+		u32 frame_size;
+		loff_t mask, vbo_a, end_a, tmp;
+
+		if (!(mode & FALLOC_FL_KEEP_SIZE)) {
+			err = -EINVAL;
+			goto out;
+		}
+
+		err = filemap_write_and_wait_range(inode->i_mapping, vbo,
+						   end - 1);
+		if (err)
+			goto out;
+
+		err = filemap_write_and_wait_range(inode->i_mapping, end,
+						   LLONG_MAX);
+		if (err)
+			goto out;
+
+		inode_dio_wait(inode);
+
+		truncate_pagecache(inode, vbo_down);
+
+		if (!is_sparsed(ni) && !is_compressed(ni)) {
+			/* Normal file. */
+			err = ntfs_zero_range(inode, vbo, end);
+			goto out;
+		}
+
+		ni_lock(ni);
+		err = attr_punch_hole(ni, vbo, len, &frame_size);
+		ni_unlock(ni);
+		if (err != E_NTFS_NOTALIGNED)
+			goto out;
+
+		/* Process not aligned punch. */
+		mask = frame_size - 1;
+		vbo_a = (vbo + mask) & ~mask;
+		end_a = end & ~mask;
+
+		tmp = min(vbo_a, end);
+		if (tmp > vbo) {
+			err = ntfs_zero_range(inode, vbo, tmp);
+			if (err)
+				goto out;
+		}
+
+		if (vbo < end_a && end_a < end) {
+			err = ntfs_zero_range(inode, end_a, end);
+			if (err)
+				goto out;
+		}
+
+		/* Aligned punch_hole */
+		if (end_a > vbo_a) {
+			ni_lock(ni);
+			err = attr_punch_hole(ni, vbo_a, end_a - vbo_a, NULL);
+			ni_unlock(ni);
+		}
+	} else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
+		if (mode & ~FALLOC_FL_COLLAPSE_RANGE) {
+			err = -EINVAL;
+			goto out;
+		}
+
+		/*
+		 * Write tail of the last page before removed range since
+		 * it will get removed from the page cache below.
+		 */
+		err = filemap_write_and_wait_range(inode->i_mapping, vbo_down,
+						   vbo);
+		if (err)
+			goto out;
+
+		/*
+		 * Write data that will be shifted to preserve them
+		 * when discarding page cache below.
+		 */
+		err = filemap_write_and_wait_range(inode->i_mapping, end,
+						   LLONG_MAX);
+		if (err)
+			goto out;
+
+		/* Wait for existing dio to complete. */
+		inode_dio_wait(inode);
+
+		truncate_pagecache(inode, vbo_down);
+
+		ni_lock(ni);
+		err = attr_collapse_range(ni, vbo, len);
+		ni_unlock(ni);
+	} else {
+		/*
+		 * Normal file: Allocate clusters, do not change 'valid' size.
+		 */
+		err = ntfs_set_size(inode, max(end, i_size));
+		if (err)
+			goto out;
+
+		if (is_sparsed(ni) || is_compressed(ni)) {
+			CLST vcn_v = ni->i_valid >> sbi->cluster_bits;
+			CLST vcn = vbo >> sbi->cluster_bits;
+			CLST cend = bytes_to_cluster(sbi, end);
+			CLST lcn, clen;
+			bool new;
+
+			/*
+			 * Allocate but do not zero new clusters. (see below comments)
+			 * This breaks security: One can read unused on-disk areas.
+			 * Zeroing these clusters may be too long.
+			 * Maybe we should check here for root rights?
+			 */
+			for (; vcn < cend; vcn += clen) {
+				err = attr_data_get_block(ni, vcn, cend - vcn,
+							  &lcn, &clen, &new);
+				if (err)
+					goto out;
+				if (!new || vcn >= vcn_v)
+					continue;
+
+				/*
+				 * Unwritten area.
+				 * NTFS is not able to store several unwritten areas.
+				 * Activate 'ntfs_sparse_cluster' to zero new allocated clusters.
+				 *
+				 * Dangerous in case:
+				 * 1G of sparsed clusters + 1 cluster of data =>
+				 * valid_size == 1G + 1 cluster
+				 * fallocate(1G) will zero 1G and this can be very long
+				 * xfstest 016/086 will fail without 'ntfs_sparse_cluster'.
+				 */
+				ntfs_sparse_cluster(inode, NULL, vcn,
+						    min(vcn_v - vcn, clen));
+			}
+		}
+
+		if (mode & FALLOC_FL_KEEP_SIZE) {
+			ni_lock(ni);
+			/* True - Keep preallocated. */
+			err = attr_set_size(ni, ATTR_DATA, NULL, 0,
+					    &ni->file.run, i_size, &ni->i_valid,
+					    true, NULL);
+			ni_unlock(ni);
+		}
+	}
+
+out:
+	if (err == -EFBIG)
+		err = -ENOSPC;
+
+	if (!err) {
+		inode->i_ctime = inode->i_mtime = current_time(inode);
+		mark_inode_dirty(inode);
+	}
+
+	inode_unlock(inode);
+	return err;
+}
+
+/*
+ * ntfs3_setattr - inode_operations::setattr
+ */
+int ntfs3_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
+		  struct iattr *attr)
+{
+	struct super_block *sb = dentry->d_sb;
+	struct ntfs_sb_info *sbi = sb->s_fs_info;
+	struct inode *inode = d_inode(dentry);
+	struct ntfs_inode *ni = ntfs_i(inode);
+	u32 ia_valid = attr->ia_valid;
+	umode_t mode = inode->i_mode;
+	int err;
+
+	if (sbi->options.no_acs_rules) {
+		/* "No access rules" - Force any changes of time etc. */
+		attr->ia_valid |= ATTR_FORCE;
+		/* and disable for editing some attributes. */
+		attr->ia_valid &= ~(ATTR_UID | ATTR_GID | ATTR_MODE);
+		ia_valid = attr->ia_valid;
+	}
+
+	err = setattr_prepare(mnt_userns, dentry, attr);
+	if (err)
+		goto out;
+
+	if (ia_valid & ATTR_SIZE) {
+		loff_t oldsize = inode->i_size;
+
+		if (WARN_ON(ni->ni_flags & NI_FLAG_COMPRESSED_MASK)) {
+			/* Should never be here, see ntfs_file_open(). */
+			err = -EOPNOTSUPP;
+			goto out;
+		}
+		inode_dio_wait(inode);
+
+		if (attr->ia_size < oldsize)
+			err = ntfs_truncate(inode, attr->ia_size);
+		else if (attr->ia_size > oldsize)
+			err = ntfs_extend(inode, attr->ia_size, 0, NULL);
+
+		if (err)
+			goto out;
+
+		ni->ni_flags |= NI_FLAG_UPDATE_PARENT;
+	}
+
+	setattr_copy(mnt_userns, inode, attr);
+
+	if (mode != inode->i_mode) {
+		err = ntfs_acl_chmod(mnt_userns, inode);
+		if (err)
+			goto out;
+
+		/* Linux 'w' -> Windows 'ro'. */
+		if (0222 & inode->i_mode)
+			ni->std_fa &= ~FILE_ATTRIBUTE_READONLY;
+		else
+			ni->std_fa |= FILE_ATTRIBUTE_READONLY;
+	}
+
+	if (ia_valid & (ATTR_UID | ATTR_GID | ATTR_MODE))
+		ntfs_save_wsl_perm(inode);
+	mark_inode_dirty(inode);
+out:
+	return err;
+}
+
+static ssize_t ntfs_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
+{
+	struct file *file = iocb->ki_filp;
+	struct inode *inode = file->f_mapping->host;
+	struct ntfs_inode *ni = ntfs_i(inode);
+
+	if (is_encrypted(ni)) {
+		ntfs_inode_warn(inode, "encrypted i/o not supported");
+		return -EOPNOTSUPP;
+	}
+
+	if (is_compressed(ni) && (iocb->ki_flags & IOCB_DIRECT)) {
+		ntfs_inode_warn(inode, "direct i/o + compressed not supported");
+		return -EOPNOTSUPP;
+	}
+
+#ifndef CONFIG_NTFS3_LZX_XPRESS
+	if (ni->ni_flags & NI_FLAG_COMPRESSED_MASK) {
+		ntfs_inode_warn(
+			inode,
+			"activate CONFIG_NTFS3_LZX_XPRESS to read external compressed files");
+		return -EOPNOTSUPP;
+	}
+#endif
+
+	if (is_dedup(ni)) {
+		ntfs_inode_warn(inode, "read deduplicated not supported");
+		return -EOPNOTSUPP;
+	}
+
+	return generic_file_read_iter(iocb, iter);
+}
+
+/*
+ * ntfs_get_frame_pages
+ *
+ * Return: Array of locked pages.
+ */
+static int ntfs_get_frame_pages(struct address_space *mapping, pgoff_t index,
+				struct page **pages, u32 pages_per_frame,
+				bool *frame_uptodate)
+{
+	gfp_t gfp_mask = mapping_gfp_mask(mapping);
+	u32 npages;
+
+	*frame_uptodate = true;
+
+	for (npages = 0; npages < pages_per_frame; npages++, index++) {
+		struct page *page;
+
+		page = find_or_create_page(mapping, index, gfp_mask);
+		if (!page) {
+			while (npages--) {
+				page = pages[npages];
+				unlock_page(page);
+				put_page(page);
+			}
+
+			return -ENOMEM;
+		}
+
+		if (!PageUptodate(page))
+			*frame_uptodate = false;
+
+		pages[npages] = page;
+	}
+
+	return 0;
+}
+
+/*
+ * ntfs_compress_write - Helper for ntfs_file_write_iter() (compressed files).
+ */
+static ssize_t ntfs_compress_write(struct kiocb *iocb, struct iov_iter *from)
+{
+	int err;
+	struct file *file = iocb->ki_filp;
+	size_t count = iov_iter_count(from);
+	loff_t pos = iocb->ki_pos;
+	struct inode *inode = file_inode(file);
+	loff_t i_size = inode->i_size;
+	struct address_space *mapping = inode->i_mapping;
+	struct ntfs_inode *ni = ntfs_i(inode);
+	u64 valid = ni->i_valid;
+	struct ntfs_sb_info *sbi = ni->mi.sbi;
+	struct page *page, **pages = NULL;
+	size_t written = 0;
+	u8 frame_bits = NTFS_LZNT_CUNIT + sbi->cluster_bits;
+	u32 frame_size = 1u << frame_bits;
+	u32 pages_per_frame = frame_size >> PAGE_SHIFT;
+	u32 ip, off;
+	CLST frame;
+	u64 frame_vbo;
+	pgoff_t index;
+	bool frame_uptodate;
+
+	if (frame_size < PAGE_SIZE) {
+		/*
+		 * frame_size == 8K if cluster 512
+		 * frame_size == 64K if cluster 4096
+		 */
+		ntfs_inode_warn(inode, "page size is bigger than frame size");
+		return -EOPNOTSUPP;
+	}
+
+	pages = kmalloc_array(pages_per_frame, sizeof(struct page *), GFP_NOFS);
+	if (!pages)
+		return -ENOMEM;
+
+	current->backing_dev_info = inode_to_bdi(inode);
+	err = file_remove_privs(file);
+	if (err)
+		goto out;
+
+	err = file_update_time(file);
+	if (err)
+		goto out;
+
+	/* Zero range [valid : pos). */
+	while (valid < pos) {
+		CLST lcn, clen;
+
+		frame = valid >> frame_bits;
+		frame_vbo = valid & ~(frame_size - 1);
+		off = valid & (frame_size - 1);
+
+		err = attr_data_get_block(ni, frame << NTFS_LZNT_CUNIT, 0, &lcn,
+					  &clen, NULL);
+		if (err)
+			goto out;
+
+		if (lcn == SPARSE_LCN) {
+			ni->i_valid = valid =
+				frame_vbo + ((u64)clen << sbi->cluster_bits);
+			continue;
+		}
+
+		/* Load full frame. */
+		err = ntfs_get_frame_pages(mapping, frame_vbo >> PAGE_SHIFT,
+					   pages, pages_per_frame,
+					   &frame_uptodate);
+		if (err)
+			goto out;
+
+		if (!frame_uptodate && off) {
+			err = ni_read_frame(ni, frame_vbo, pages,
+					    pages_per_frame);
+			if (err) {
+				for (ip = 0; ip < pages_per_frame; ip++) {
+					page = pages[ip];
+					unlock_page(page);
+					put_page(page);
+				}
+				goto out;
+			}
+		}
+
+		ip = off >> PAGE_SHIFT;
+		off = offset_in_page(valid);
+		for (; ip < pages_per_frame; ip++, off = 0) {
+			page = pages[ip];
+			zero_user_segment(page, off, PAGE_SIZE);
+			flush_dcache_page(page);
+			SetPageUptodate(page);
+		}
+
+		ni_lock(ni);
+		err = ni_write_frame(ni, pages, pages_per_frame);
+		ni_unlock(ni);
+
+		for (ip = 0; ip < pages_per_frame; ip++) {
+			page = pages[ip];
+			SetPageUptodate(page);
+			unlock_page(page);
+			put_page(page);
+		}
+
+		if (err)
+			goto out;
+
+		ni->i_valid = valid = frame_vbo + frame_size;
+	}
+
+	/* Copy user data [pos : pos + count). */
+	while (count) {
+		size_t copied, bytes;
+
+		off = pos & (frame_size - 1);
+		bytes = frame_size - off;
+		if (bytes > count)
+			bytes = count;
+
+		frame = pos >> frame_bits;
+		frame_vbo = pos & ~(frame_size - 1);
+		index = frame_vbo >> PAGE_SHIFT;
+
+		if (unlikely(iov_iter_fault_in_readable(from, bytes))) {
+			err = -EFAULT;
+			goto out;
+		}
+
+		/* Load full frame. */
+		err = ntfs_get_frame_pages(mapping, index, pages,
+					   pages_per_frame, &frame_uptodate);
+		if (err)
+			goto out;
+
+		if (!frame_uptodate) {
+			loff_t to = pos + bytes;
+
+			if (off || (to < i_size && (to & (frame_size - 1)))) {
+				err = ni_read_frame(ni, frame_vbo, pages,
+						    pages_per_frame);
+				if (err) {
+					for (ip = 0; ip < pages_per_frame;
+					     ip++) {
+						page = pages[ip];
+						unlock_page(page);
+						put_page(page);
+					}
+					goto out;
+				}
+			}
+		}
+
+		WARN_ON(!bytes);
+		copied = 0;
+		ip = off >> PAGE_SHIFT;
+		off = offset_in_page(pos);
+
+		/* Copy user data to pages. */
+		for (;;) {
+			size_t cp, tail = PAGE_SIZE - off;
+
+			page = pages[ip];
+			cp = copy_page_from_iter_atomic(page, off,
+							min(tail, bytes), from);
+			flush_dcache_page(page);
+
+			copied += cp;
+			bytes -= cp;
+			if (!bytes || !cp)
+				break;
+
+			if (cp < tail) {
+				off += cp;
+			} else {
+				ip++;
+				off = 0;
+			}
+		}
+
+		ni_lock(ni);
+		err = ni_write_frame(ni, pages, pages_per_frame);
+		ni_unlock(ni);
+
+		for (ip = 0; ip < pages_per_frame; ip++) {
+			page = pages[ip];
+			ClearPageDirty(page);
+			SetPageUptodate(page);
+			unlock_page(page);
+			put_page(page);
+		}
+
+		if (err)
+			goto out;
+
+		/*
+		 * We can loop for a long time in here. Be nice and allow
+		 * us to schedule out to avoid softlocking if preempt
+		 * is disabled.
+		 */
+		cond_resched();
+
+		pos += copied;
+		written += copied;
+
+		count = iov_iter_count(from);
+	}
+
+out:
+	kfree(pages);
+
+	current->backing_dev_info = NULL;
+
+	if (err < 0)
+		return err;
+
+	iocb->ki_pos += written;
+	if (iocb->ki_pos > ni->i_valid)
+		ni->i_valid = iocb->ki_pos;
+
+	return written;
+}
+
+/*
+ * ntfs_file_write_iter - file_operations::write_iter
+ */
+static ssize_t ntfs_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
+{
+	struct file *file = iocb->ki_filp;
+	struct address_space *mapping = file->f_mapping;
+	struct inode *inode = mapping->host;
+	ssize_t ret;
+	struct ntfs_inode *ni = ntfs_i(inode);
+
+	if (is_encrypted(ni)) {
+		ntfs_inode_warn(inode, "encrypted i/o not supported");
+		return -EOPNOTSUPP;
+	}
+
+	if (is_compressed(ni) && (iocb->ki_flags & IOCB_DIRECT)) {
+		ntfs_inode_warn(inode, "direct i/o + compressed not supported");
+		return -EOPNOTSUPP;
+	}
+
+	if (is_dedup(ni)) {
+		ntfs_inode_warn(inode, "write into deduplicated not supported");
+		return -EOPNOTSUPP;
+	}
+
+	if (!inode_trylock(inode)) {
+		if (iocb->ki_flags & IOCB_NOWAIT)
+			return -EAGAIN;
+		inode_lock(inode);
+	}
+
+	ret = generic_write_checks(iocb, from);
+	if (ret <= 0)
+		goto out;
+
+	if (WARN_ON(ni->ni_flags & NI_FLAG_COMPRESSED_MASK)) {
+		/* Should never be here, see ntfs_file_open(). */
+		ret = -EOPNOTSUPP;
+		goto out;
+	}
+
+	ret = ntfs_extend(inode, iocb->ki_pos, ret, file);
+	if (ret)
+		goto out;
+
+	ret = is_compressed(ni) ? ntfs_compress_write(iocb, from)
+				: __generic_file_write_iter(iocb, from);
+
+out:
+	inode_unlock(inode);
+
+	if (ret > 0)
+		ret = generic_write_sync(iocb, ret);
+
+	return ret;
+}
+
+/*
+ * ntfs_file_open - file_operations::open
+ */
+int ntfs_file_open(struct inode *inode, struct file *file)
+{
+	struct ntfs_inode *ni = ntfs_i(inode);
+
+	if (unlikely((is_compressed(ni) || is_encrypted(ni)) &&
+		     (file->f_flags & O_DIRECT))) {
+		return -EOPNOTSUPP;
+	}
+
+	/* Decompress "external compressed" file if opened for rw. */
+	if ((ni->ni_flags & NI_FLAG_COMPRESSED_MASK) &&
+	    (file->f_flags & (O_WRONLY | O_RDWR | O_TRUNC))) {
+#ifdef CONFIG_NTFS3_LZX_XPRESS
+		int err = ni_decompress_file(ni);
+
+		if (err)
+			return err;
+#else
+		ntfs_inode_warn(
+			inode,
+			"activate CONFIG_NTFS3_LZX_XPRESS to write external compressed files");
+		return -EOPNOTSUPP;
+#endif
+	}
+
+	return generic_file_open(inode, file);
+}
+
+/*
+ * ntfs_file_release - file_operations::release
+ */
+static int ntfs_file_release(struct inode *inode, struct file *file)
+{
+	struct ntfs_inode *ni = ntfs_i(inode);
+	struct ntfs_sb_info *sbi = ni->mi.sbi;
+	int err = 0;
+
+	/* If we are last writer on the inode, drop the block reservation. */
+	if (sbi->options.prealloc && ((file->f_mode & FMODE_WRITE) &&
+				      atomic_read(&inode->i_writecount) == 1)) {
+		ni_lock(ni);
+		down_write(&ni->file.run_lock);
+
+		err = attr_set_size(ni, ATTR_DATA, NULL, 0, &ni->file.run,
+				    inode->i_size, &ni->i_valid, false, NULL);
+
+		up_write(&ni->file.run_lock);
+		ni_unlock(ni);
+	}
+	return err;
+}
+
+/*
+ * ntfs_fiemap - file_operations::fiemap
+ */
+int ntfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
+		__u64 start, __u64 len)
+{
+	int err;
+	struct ntfs_inode *ni = ntfs_i(inode);
+
+	err = fiemap_prep(inode, fieinfo, start, &len, ~FIEMAP_FLAG_XATTR);
+	if (err)
+		return err;
+
+	ni_lock(ni);
+
+	err = ni_fiemap(ni, fieinfo, start, len);
+
+	ni_unlock(ni);
+
+	return err;
+}
+
+// clang-format off
+const struct inode_operations ntfs_file_inode_operations = {
+	.getattr	= ntfs_getattr,
+	.setattr	= ntfs3_setattr,
+	.listxattr	= ntfs_listxattr,
+	.permission	= ntfs_permission,
+	.get_acl	= ntfs_get_acl,
+	.set_acl	= ntfs_set_acl,
+	.fiemap		= ntfs_fiemap,
+};
+
+const struct file_operations ntfs_file_operations = {
+	.llseek		= generic_file_llseek,
+	.read_iter	= ntfs_file_read_iter,
+	.write_iter	= ntfs_file_write_iter,
+	.unlocked_ioctl = ntfs_ioctl,
+#ifdef CONFIG_COMPAT
+	.compat_ioctl	= ntfs_compat_ioctl,
+#endif
+	.splice_read	= generic_file_splice_read,
+	.mmap		= ntfs_file_mmap,
+	.open		= ntfs_file_open,
+	.fsync		= generic_file_fsync,
+	.splice_write	= iter_file_splice_write,
+	.fallocate	= ntfs_fallocate,
+	.release	= ntfs_file_release,
+};
+// clang-format on
diff --git a/fs/ntfs3/frecord.c b/fs/ntfs3/frecord.c
new file mode 100644
index 000000000000..938b12d56ca6
--- /dev/null
+++ b/fs/ntfs3/frecord.c
@@ -0,0 +1,3257 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ *
+ * Copyright (C) 2019-2021 Paragon Software GmbH, All rights reserved.
+ *
+ */
+
+#include <linux/blkdev.h>
+#include <linux/buffer_head.h>
+#include <linux/fiemap.h>
+#include <linux/fs.h>
+#include <linux/nls.h>
+#include <linux/vmalloc.h>
+
+#include "debug.h"
+#include "ntfs.h"
+#include "ntfs_fs.h"
+#ifdef CONFIG_NTFS3_LZX_XPRESS
+#include "lib/lib.h"
+#endif
+
+static struct mft_inode *ni_ins_mi(struct ntfs_inode *ni, struct rb_root *tree,
+				   CLST ino, struct rb_node *ins)
+{
+	struct rb_node **p = &tree->rb_node;
+	struct rb_node *pr = NULL;
+
+	while (*p) {
+		struct mft_inode *mi;
+
+		pr = *p;
+		mi = rb_entry(pr, struct mft_inode, node);
+		if (mi->rno > ino)
+			p = &pr->rb_left;
+		else if (mi->rno < ino)
+			p = &pr->rb_right;
+		else
+			return mi;
+	}
+
+	if (!ins)
+		return NULL;
+
+	rb_link_node(ins, pr, p);
+	rb_insert_color(ins, tree);
+	return rb_entry(ins, struct mft_inode, node);
+}
+
+/*
+ * ni_find_mi - Find mft_inode by record number.
+ */
+static struct mft_inode *ni_find_mi(struct ntfs_inode *ni, CLST rno)
+{
+	return ni_ins_mi(ni, &ni->mi_tree, rno, NULL);
+}
+
+/*
+ * ni_add_mi - Add new mft_inode into ntfs_inode.
+ */
+static void ni_add_mi(struct ntfs_inode *ni, struct mft_inode *mi)
+{
+	ni_ins_mi(ni, &ni->mi_tree, mi->rno, &mi->node);
+}
+
+/*
+ * ni_remove_mi - Remove mft_inode from ntfs_inode.
+ */
+void ni_remove_mi(struct ntfs_inode *ni, struct mft_inode *mi)
+{
+	rb_erase(&mi->node, &ni->mi_tree);
+}
+
+/*
+ * ni_std - Return: Pointer into std_info from primary record.
+ */
+struct ATTR_STD_INFO *ni_std(struct ntfs_inode *ni)
+{
+	const struct ATTRIB *attr;
+
+	attr = mi_find_attr(&ni->mi, NULL, ATTR_STD, NULL, 0, NULL);
+	return attr ? resident_data_ex(attr, sizeof(struct ATTR_STD_INFO))
+		    : NULL;
+}
+
+/*
+ * ni_std5
+ *
+ * Return: Pointer into std_info from primary record.
+ */
+struct ATTR_STD_INFO5 *ni_std5(struct ntfs_inode *ni)
+{
+	const struct ATTRIB *attr;
+
+	attr = mi_find_attr(&ni->mi, NULL, ATTR_STD, NULL, 0, NULL);
+
+	return attr ? resident_data_ex(attr, sizeof(struct ATTR_STD_INFO5))
+		    : NULL;
+}
+
+/*
+ * ni_clear - Clear resources allocated by ntfs_inode.
+ */
+void ni_clear(struct ntfs_inode *ni)
+{
+	struct rb_node *node;
+
+	if (!ni->vfs_inode.i_nlink && is_rec_inuse(ni->mi.mrec))
+		ni_delete_all(ni);
+
+	al_destroy(ni);
+
+	for (node = rb_first(&ni->mi_tree); node;) {
+		struct rb_node *next = rb_next(node);
+		struct mft_inode *mi = rb_entry(node, struct mft_inode, node);
+
+		rb_erase(node, &ni->mi_tree);
+		mi_put(mi);
+		node = next;
+	}
+
+	/* Bad inode always has mode == S_IFREG. */
+	if (ni->ni_flags & NI_FLAG_DIR)
+		indx_clear(&ni->dir);
+	else {
+		run_close(&ni->file.run);
+#ifdef CONFIG_NTFS3_LZX_XPRESS
+		if (ni->file.offs_page) {
+			/* On-demand allocated page for offsets. */
+			put_page(ni->file.offs_page);
+			ni->file.offs_page = NULL;
+		}
+#endif
+	}
+
+	mi_clear(&ni->mi);
+}
+
+/*
+ * ni_load_mi_ex - Find mft_inode by record number.
+ */
+int ni_load_mi_ex(struct ntfs_inode *ni, CLST rno, struct mft_inode **mi)
+{
+	int err;
+	struct mft_inode *r;
+
+	r = ni_find_mi(ni, rno);
+	if (r)
+		goto out;
+
+	err = mi_get(ni->mi.sbi, rno, &r);
+	if (err)
+		return err;
+
+	ni_add_mi(ni, r);
+
+out:
+	if (mi)
+		*mi = r;
+	return 0;
+}
+
+/*
+ * ni_load_mi - Load mft_inode corresponded list_entry.
+ */
+int ni_load_mi(struct ntfs_inode *ni, const struct ATTR_LIST_ENTRY *le,
+	       struct mft_inode **mi)
+{
+	CLST rno;
+
+	if (!le) {
+		*mi = &ni->mi;
+		return 0;
+	}
+
+	rno = ino_get(&le->ref);
+	if (rno == ni->mi.rno) {
+		*mi = &ni->mi;
+		return 0;
+	}
+	return ni_load_mi_ex(ni, rno, mi);
+}
+
+/*
+ * ni_find_attr
+ *
+ * Return: Attribute and record this attribute belongs to.
+ */
+struct ATTRIB *ni_find_attr(struct ntfs_inode *ni, struct ATTRIB *attr,
+			    struct ATTR_LIST_ENTRY **le_o, enum ATTR_TYPE type,
+			    const __le16 *name, u8 name_len, const CLST *vcn,
+			    struct mft_inode **mi)
+{
+	struct ATTR_LIST_ENTRY *le;
+	struct mft_inode *m;
+
+	if (!ni->attr_list.size ||
+	    (!name_len && (type == ATTR_LIST || type == ATTR_STD))) {
+		if (le_o)
+			*le_o = NULL;
+		if (mi)
+			*mi = &ni->mi;
+
+		/* Look for required attribute in primary record. */
+		return mi_find_attr(&ni->mi, attr, type, name, name_len, NULL);
+	}
+
+	/* First look for list entry of required type. */
+	le = al_find_ex(ni, le_o ? *le_o : NULL, type, name, name_len, vcn);
+	if (!le)
+		return NULL;
+
+	if (le_o)
+		*le_o = le;
+
+	/* Load record that contains this attribute. */
+	if (ni_load_mi(ni, le, &m))
+		return NULL;
+
+	/* Look for required attribute. */
+	attr = mi_find_attr(m, NULL, type, name, name_len, &le->id);
+
+	if (!attr)
+		goto out;
+
+	if (!attr->non_res) {
+		if (vcn && *vcn)
+			goto out;
+	} else if (!vcn) {
+		if (attr->nres.svcn)
+			goto out;
+	} else if (le64_to_cpu(attr->nres.svcn) > *vcn ||
+		   *vcn > le64_to_cpu(attr->nres.evcn)) {
+		goto out;
+	}
+
+	if (mi)
+		*mi = m;
+	return attr;
+
+out:
+	ntfs_set_state(ni->mi.sbi, NTFS_DIRTY_ERROR);
+	return NULL;
+}
+
+/*
+ * ni_enum_attr_ex - Enumerates attributes in ntfs_inode.
+ */
+struct ATTRIB *ni_enum_attr_ex(struct ntfs_inode *ni, struct ATTRIB *attr,
+			       struct ATTR_LIST_ENTRY **le,
+			       struct mft_inode **mi)
+{
+	struct mft_inode *mi2;
+	struct ATTR_LIST_ENTRY *le2;
+
+	/* Do we have an attribute list? */
+	if (!ni->attr_list.size) {
+		*le = NULL;
+		if (mi)
+			*mi = &ni->mi;
+		/* Enum attributes in primary record. */
+		return mi_enum_attr(&ni->mi, attr);
+	}
+
+	/* Get next list entry. */
+	le2 = *le = al_enumerate(ni, attr ? *le : NULL);
+	if (!le2)
+		return NULL;
+
+	/* Load record that contains the required attribute. */
+	if (ni_load_mi(ni, le2, &mi2))
+		return NULL;
+
+	if (mi)
+		*mi = mi2;
+
+	/* Find attribute in loaded record. */
+	return rec_find_attr_le(mi2, le2);
+}
+
+/*
+ * ni_load_attr - Load attribute that contains given VCN.
+ */
+struct ATTRIB *ni_load_attr(struct ntfs_inode *ni, enum ATTR_TYPE type,
+			    const __le16 *name, u8 name_len, CLST vcn,
+			    struct mft_inode **pmi)
+{
+	struct ATTR_LIST_ENTRY *le;
+	struct ATTRIB *attr;
+	struct mft_inode *mi;
+	struct ATTR_LIST_ENTRY *next;
+
+	if (!ni->attr_list.size) {
+		if (pmi)
+			*pmi = &ni->mi;
+		return mi_find_attr(&ni->mi, NULL, type, name, name_len, NULL);
+	}
+
+	le = al_find_ex(ni, NULL, type, name, name_len, NULL);
+	if (!le)
+		return NULL;
+
+	/*
+	 * Unfortunately ATTR_LIST_ENTRY contains only start VCN.
+	 * So to find the ATTRIB segment that contains 'vcn' we should
+	 * enumerate some entries.
+	 */
+	if (vcn) {
+		for (;; le = next) {
+			next = al_find_ex(ni, le, type, name, name_len, NULL);
+			if (!next || le64_to_cpu(next->vcn) > vcn)
+				break;
+		}
+	}
+
+	if (ni_load_mi(ni, le, &mi))
+		return NULL;
+
+	if (pmi)
+		*pmi = mi;
+
+	attr = mi_find_attr(mi, NULL, type, name, name_len, &le->id);
+	if (!attr)
+		return NULL;
+
+	if (!attr->non_res)
+		return attr;
+
+	if (le64_to_cpu(attr->nres.svcn) <= vcn &&
+	    vcn <= le64_to_cpu(attr->nres.evcn))
+		return attr;
+
+	return NULL;
+}
+
+/*
+ * ni_load_all_mi - Load all subrecords.
+ */
+int ni_load_all_mi(struct ntfs_inode *ni)
+{
+	int err;
+	struct ATTR_LIST_ENTRY *le;
+
+	if (!ni->attr_list.size)
+		return 0;
+
+	le = NULL;
+
+	while ((le = al_enumerate(ni, le))) {
+		CLST rno = ino_get(&le->ref);
+
+		if (rno == ni->mi.rno)
+			continue;
+
+		err = ni_load_mi_ex(ni, rno, NULL);
+		if (err)
+			return err;
+	}
+
+	return 0;
+}
+
+/*
+ * ni_add_subrecord - Allocate + format + attach a new subrecord.
+ */
+bool ni_add_subrecord(struct ntfs_inode *ni, CLST rno, struct mft_inode **mi)
+{
+	struct mft_inode *m;
+
+	m = kzalloc(sizeof(struct mft_inode), GFP_NOFS);
+	if (!m)
+		return false;
+
+	if (mi_format_new(m, ni->mi.sbi, rno, 0, ni->mi.rno == MFT_REC_MFT)) {
+		mi_put(m);
+		return false;
+	}
+
+	mi_get_ref(&ni->mi, &m->mrec->parent_ref);
+
+	ni_add_mi(ni, m);
+	*mi = m;
+	return true;
+}
+
+/*
+ * ni_remove_attr - Remove all attributes for the given type/name/id.
+ */
+int ni_remove_attr(struct ntfs_inode *ni, enum ATTR_TYPE type,
+		   const __le16 *name, size_t name_len, bool base_only,
+		   const __le16 *id)
+{
+	int err;
+	struct ATTRIB *attr;
+	struct ATTR_LIST_ENTRY *le;
+	struct mft_inode *mi;
+	u32 type_in;
+	int diff;
+
+	if (base_only || type == ATTR_LIST || !ni->attr_list.size) {
+		attr = mi_find_attr(&ni->mi, NULL, type, name, name_len, id);
+		if (!attr)
+			return -ENOENT;
+
+		mi_remove_attr(ni, &ni->mi, attr);
+		return 0;
+	}
+
+	type_in = le32_to_cpu(type);
+	le = NULL;
+
+	for (;;) {
+		le = al_enumerate(ni, le);
+		if (!le)
+			return 0;
+
+next_le2:
+		diff = le32_to_cpu(le->type) - type_in;
+		if (diff < 0)
+			continue;
+
+		if (diff > 0)
+			return 0;
+
+		if (le->name_len != name_len)
+			continue;
+
+		if (name_len &&
+		    memcmp(le_name(le), name, name_len * sizeof(short)))
+			continue;
+
+		if (id && le->id != *id)
+			continue;
+		err = ni_load_mi(ni, le, &mi);
+		if (err)
+			return err;
+
+		al_remove_le(ni, le);
+
+		attr = mi_find_attr(mi, NULL, type, name, name_len, id);
+		if (!attr)
+			return -ENOENT;
+
+		mi_remove_attr(ni, mi, attr);
+
+		if (PtrOffset(ni->attr_list.le, le) >= ni->attr_list.size)
+			return 0;
+		goto next_le2;
+	}
+}
+
+/*
+ * ni_ins_new_attr - Insert the attribute into record.
+ *
+ * Return: Not full constructed attribute or NULL if not possible to create.
+ */
+static struct ATTRIB *
+ni_ins_new_attr(struct ntfs_inode *ni, struct mft_inode *mi,
+		struct ATTR_LIST_ENTRY *le, enum ATTR_TYPE type,
+		const __le16 *name, u8 name_len, u32 asize, u16 name_off,
+		CLST svcn, struct ATTR_LIST_ENTRY **ins_le)
+{
+	int err;
+	struct ATTRIB *attr;
+	bool le_added = false;
+	struct MFT_REF ref;
+
+	mi_get_ref(mi, &ref);
+
+	if (type != ATTR_LIST && !le && ni->attr_list.size) {
+		err = al_add_le(ni, type, name, name_len, svcn, cpu_to_le16(-1),
+				&ref, &le);
+		if (err) {
+			/* No memory or no space. */
+			return NULL;
+		}
+		le_added = true;
+
+		/*
+		 * al_add_le -> attr_set_size (list) -> ni_expand_list
+		 * which moves some attributes out of primary record
+		 * this means that name may point into moved memory
+		 * reinit 'name' from le.
+		 */
+		name = le->name;
+	}
+
+	attr = mi_insert_attr(mi, type, name, name_len, asize, name_off);
+	if (!attr) {
+		if (le_added)
+			al_remove_le(ni, le);
+		return NULL;
+	}
+
+	if (type == ATTR_LIST) {
+		/* Attr list is not in list entry array. */
+		goto out;
+	}
+
+	if (!le)
+		goto out;
+
+	/* Update ATTRIB Id and record reference. */
+	le->id = attr->id;
+	ni->attr_list.dirty = true;
+	le->ref = ref;
+
+out:
+	if (ins_le)
+		*ins_le = le;
+	return attr;
+}
+
+/*
+ * ni_repack
+ *
+ * Random write access to sparsed or compressed file may result to
+ * not optimized packed runs.
+ * Here is the place to optimize it.
+ */
+static int ni_repack(struct ntfs_inode *ni)
+{
+	int err = 0;
+	struct ntfs_sb_info *sbi = ni->mi.sbi;
+	struct mft_inode *mi, *mi_p = NULL;
+	struct ATTRIB *attr = NULL, *attr_p;
+	struct ATTR_LIST_ENTRY *le = NULL, *le_p;
+	CLST alloc = 0;
+	u8 cluster_bits = sbi->cluster_bits;
+	CLST svcn, evcn = 0, svcn_p, evcn_p, next_svcn;
+	u32 roff, rs = sbi->record_size;
+	struct runs_tree run;
+
+	run_init(&run);
+
+	while ((attr = ni_enum_attr_ex(ni, attr, &le, &mi))) {
+		if (!attr->non_res)
+			continue;
+
+		svcn = le64_to_cpu(attr->nres.svcn);
+		if (svcn != le64_to_cpu(le->vcn)) {
+			err = -EINVAL;
+			break;
+		}
+
+		if (!svcn) {
+			alloc = le64_to_cpu(attr->nres.alloc_size) >>
+				cluster_bits;
+			mi_p = NULL;
+		} else if (svcn != evcn + 1) {
+			err = -EINVAL;
+			break;
+		}
+
+		evcn = le64_to_cpu(attr->nres.evcn);
+
+		if (svcn > evcn + 1) {
+			err = -EINVAL;
+			break;
+		}
+
+		if (!mi_p) {
+			/* Do not try if not enogh free space. */
+			if (le32_to_cpu(mi->mrec->used) + 8 >= rs)
+				continue;
+
+			/* Do not try if last attribute segment. */
+			if (evcn + 1 == alloc)
+				continue;
+			run_close(&run);
+		}
+
+		roff = le16_to_cpu(attr->nres.run_off);
+		err = run_unpack(&run, sbi, ni->mi.rno, svcn, evcn, svcn,
+				 Add2Ptr(attr, roff),
+				 le32_to_cpu(attr->size) - roff);
+		if (err < 0)
+			break;
+
+		if (!mi_p) {
+			mi_p = mi;
+			attr_p = attr;
+			svcn_p = svcn;
+			evcn_p = evcn;
+			le_p = le;
+			err = 0;
+			continue;
+		}
+
+		/*
+		 * Run contains data from two records: mi_p and mi
+		 * Try to pack in one.
+		 */
+		err = mi_pack_runs(mi_p, attr_p, &run, evcn + 1 - svcn_p);
+		if (err)
+			break;
+
+		next_svcn = le64_to_cpu(attr_p->nres.evcn) + 1;
+
+		if (next_svcn >= evcn + 1) {
+			/* We can remove this attribute segment. */
+			al_remove_le(ni, le);
+			mi_remove_attr(NULL, mi, attr);
+			le = le_p;
+			continue;
+		}
+
+		attr->nres.svcn = le->vcn = cpu_to_le64(next_svcn);
+		mi->dirty = true;
+		ni->attr_list.dirty = true;
+
+		if (evcn + 1 == alloc) {
+			err = mi_pack_runs(mi, attr, &run,
+					   evcn + 1 - next_svcn);
+			if (err)
+				break;
+			mi_p = NULL;
+		} else {
+			mi_p = mi;
+			attr_p = attr;
+			svcn_p = next_svcn;
+			evcn_p = evcn;
+			le_p = le;
+			run_truncate_head(&run, next_svcn);
+		}
+	}
+
+	if (err) {
+		ntfs_inode_warn(&ni->vfs_inode, "repack problem");
+		ntfs_set_state(sbi, NTFS_DIRTY_ERROR);
+
+		/* Pack loaded but not packed runs. */
+		if (mi_p)
+			mi_pack_runs(mi_p, attr_p, &run, evcn_p + 1 - svcn_p);
+	}
+
+	run_close(&run);
+	return err;
+}
+
+/*
+ * ni_try_remove_attr_list
+ *
+ * Can we remove attribute list?
+ * Check the case when primary record contains enough space for all attributes.
+ */
+static int ni_try_remove_attr_list(struct ntfs_inode *ni)
+{
+	int err = 0;
+	struct ntfs_sb_info *sbi = ni->mi.sbi;
+	struct ATTRIB *attr, *attr_list, *attr_ins;
+	struct ATTR_LIST_ENTRY *le;
+	struct mft_inode *mi;
+	u32 asize, free;
+	struct MFT_REF ref;
+	__le16 id;
+
+	if (!ni->attr_list.dirty)
+		return 0;
+
+	err = ni_repack(ni);
+	if (err)
+		return err;
+
+	attr_list = mi_find_attr(&ni->mi, NULL, ATTR_LIST, NULL, 0, NULL);
+	if (!attr_list)
+		return 0;
+
+	asize = le32_to_cpu(attr_list->size);
+
+	/* Free space in primary record without attribute list. */
+	free = sbi->record_size - le32_to_cpu(ni->mi.mrec->used) + asize;
+	mi_get_ref(&ni->mi, &ref);
+
+	le = NULL;
+	while ((le = al_enumerate(ni, le))) {
+		if (!memcmp(&le->ref, &ref, sizeof(ref)))
+			continue;
+
+		if (le->vcn)
+			return 0;
+
+		mi = ni_find_mi(ni, ino_get(&le->ref));
+		if (!mi)
+			return 0;
+
+		attr = mi_find_attr(mi, NULL, le->type, le_name(le),
+				    le->name_len, &le->id);
+		if (!attr)
+			return 0;
+
+		asize = le32_to_cpu(attr->size);
+		if (asize > free)
+			return 0;
+
+		free -= asize;
+	}
+
+	/* It seems that attribute list can be removed from primary record. */
+	mi_remove_attr(NULL, &ni->mi, attr_list);
+
+	/*
+	 * Repeat the cycle above and move all attributes to primary record.
+	 * It should be success!
+	 */
+	le = NULL;
+	while ((le = al_enumerate(ni, le))) {
+		if (!memcmp(&le->ref, &ref, sizeof(ref)))
+			continue;
+
+		mi = ni_find_mi(ni, ino_get(&le->ref));
+
+		attr = mi_find_attr(mi, NULL, le->type, le_name(le),
+				    le->name_len, &le->id);
+		asize = le32_to_cpu(attr->size);
+
+		/* Insert into primary record. */
+		attr_ins = mi_insert_attr(&ni->mi, le->type, le_name(le),
+					  le->name_len, asize,
+					  le16_to_cpu(attr->name_off));
+		id = attr_ins->id;
+
+		/* Copy all except id. */
+		memcpy(attr_ins, attr, asize);
+		attr_ins->id = id;
+
+		/* Remove from original record. */
+		mi_remove_attr(NULL, mi, attr);
+	}
+
+	run_deallocate(sbi, &ni->attr_list.run, true);
+	run_close(&ni->attr_list.run);
+	ni->attr_list.size = 0;
+	kfree(ni->attr_list.le);
+	ni->attr_list.le = NULL;
+	ni->attr_list.dirty = false;
+
+	return 0;
+}
+
+/*
+ * ni_create_attr_list - Generates an attribute list for this primary record.
+ */
+int ni_create_attr_list(struct ntfs_inode *ni)
+{
+	struct ntfs_sb_info *sbi = ni->mi.sbi;
+	int err;
+	u32 lsize;
+	struct ATTRIB *attr;
+	struct ATTRIB *arr_move[7];
+	struct ATTR_LIST_ENTRY *le, *le_b[7];
+	struct MFT_REC *rec;
+	bool is_mft;
+	CLST rno = 0;
+	struct mft_inode *mi;
+	u32 free_b, nb, to_free, rs;
+	u16 sz;
+
+	is_mft = ni->mi.rno == MFT_REC_MFT;
+	rec = ni->mi.mrec;
+	rs = sbi->record_size;
+
+	/*
+	 * Skip estimating exact memory requirement.
+	 * Looks like one record_size is always enough.
+	 */
+	le = kmalloc(al_aligned(rs), GFP_NOFS);
+	if (!le) {
+		err = -ENOMEM;
+		goto out;
+	}
+
+	mi_get_ref(&ni->mi, &le->ref);
+	ni->attr_list.le = le;
+
+	attr = NULL;
+	nb = 0;
+	free_b = 0;
+	attr = NULL;
+
+	for (; (attr = mi_enum_attr(&ni->mi, attr)); le = Add2Ptr(le, sz)) {
+		sz = le_size(attr->name_len);
+		le->type = attr->type;
+		le->size = cpu_to_le16(sz);
+		le->name_len = attr->name_len;
+		le->name_off = offsetof(struct ATTR_LIST_ENTRY, name);
+		le->vcn = 0;
+		if (le != ni->attr_list.le)
+			le->ref = ni->attr_list.le->ref;
+		le->id = attr->id;
+
+		if (attr->name_len)
+			memcpy(le->name, attr_name(attr),
+			       sizeof(short) * attr->name_len);
+		else if (attr->type == ATTR_STD)
+			continue;
+		else if (attr->type == ATTR_LIST)
+			continue;
+		else if (is_mft && attr->type == ATTR_DATA)
+			continue;
+
+		if (!nb || nb < ARRAY_SIZE(arr_move)) {
+			le_b[nb] = le;
+			arr_move[nb++] = attr;
+			free_b += le32_to_cpu(attr->size);
+		}
+	}
+
+	lsize = PtrOffset(ni->attr_list.le, le);
+	ni->attr_list.size = lsize;
+
+	to_free = le32_to_cpu(rec->used) + lsize + SIZEOF_RESIDENT;
+	if (to_free <= rs) {
+		to_free = 0;
+	} else {
+		to_free -= rs;
+
+		if (to_free > free_b) {
+			err = -EINVAL;
+			goto out1;
+		}
+	}
+
+	/* Allocate child MFT. */
+	err = ntfs_look_free_mft(sbi, &rno, is_mft, ni, &mi);
+	if (err)
+		goto out1;
+
+	/* Call mi_remove_attr() in reverse order to keep pointers 'arr_move' valid. */
+	while (to_free > 0) {
+		struct ATTRIB *b = arr_move[--nb];
+		u32 asize = le32_to_cpu(b->size);
+		u16 name_off = le16_to_cpu(b->name_off);
+
+		attr = mi_insert_attr(mi, b->type, Add2Ptr(b, name_off),
+				      b->name_len, asize, name_off);
+		WARN_ON(!attr);
+
+		mi_get_ref(mi, &le_b[nb]->ref);
+		le_b[nb]->id = attr->id;
+
+		/* Copy all except id. */
+		memcpy(attr, b, asize);
+		attr->id = le_b[nb]->id;
+
+		/* Remove from primary record. */
+		WARN_ON(!mi_remove_attr(NULL, &ni->mi, b));
+
+		if (to_free <= asize)
+			break;
+		to_free -= asize;
+		WARN_ON(!nb);
+	}
+
+	attr = mi_insert_attr(&ni->mi, ATTR_LIST, NULL, 0,
+			      lsize + SIZEOF_RESIDENT, SIZEOF_RESIDENT);
+	WARN_ON(!attr);
+
+	attr->non_res = 0;
+	attr->flags = 0;
+	attr->res.data_size = cpu_to_le32(lsize);
+	attr->res.data_off = SIZEOF_RESIDENT_LE;
+	attr->res.flags = 0;
+	attr->res.res = 0;
+
+	memcpy(resident_data_ex(attr, lsize), ni->attr_list.le, lsize);
+
+	ni->attr_list.dirty = false;
+
+	mark_inode_dirty(&ni->vfs_inode);
+	goto out;
+
+out1:
+	kfree(ni->attr_list.le);
+	ni->attr_list.le = NULL;
+	ni->attr_list.size = 0;
+
+out:
+	return err;
+}
+
+/*
+ * ni_ins_attr_ext - Add an external attribute to the ntfs_inode.
+ */
+static int ni_ins_attr_ext(struct ntfs_inode *ni, struct ATTR_LIST_ENTRY *le,
+			   enum ATTR_TYPE type, const __le16 *name, u8 name_len,
+			   u32 asize, CLST svcn, u16 name_off, bool force_ext,
+			   struct ATTRIB **ins_attr, struct mft_inode **ins_mi,
+			   struct ATTR_LIST_ENTRY **ins_le)
+{
+	struct ATTRIB *attr;
+	struct mft_inode *mi;
+	CLST rno;
+	u64 vbo;
+	struct rb_node *node;
+	int err;
+	bool is_mft, is_mft_data;
+	struct ntfs_sb_info *sbi = ni->mi.sbi;
+
+	is_mft = ni->mi.rno == MFT_REC_MFT;
+	is_mft_data = is_mft && type == ATTR_DATA && !name_len;
+
+	if (asize > sbi->max_bytes_per_attr) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	/*
+	 * Standard information and attr_list cannot be made external.
+	 * The Log File cannot have any external attributes.
+	 */
+	if (type == ATTR_STD || type == ATTR_LIST ||
+	    ni->mi.rno == MFT_REC_LOG) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	/* Create attribute list if it is not already existed. */
+	if (!ni->attr_list.size) {
+		err = ni_create_attr_list(ni);
+		if (err)
+			goto out;
+	}
+
+	vbo = is_mft_data ? ((u64)svcn << sbi->cluster_bits) : 0;
+
+	if (force_ext)
+		goto insert_ext;
+
+	/* Load all subrecords into memory. */
+	err = ni_load_all_mi(ni);
+	if (err)
+		goto out;
+
+	/* Check each of loaded subrecord. */
+	for (node = rb_first(&ni->mi_tree); node; node = rb_next(node)) {
+		mi = rb_entry(node, struct mft_inode, node);
+
+		if (is_mft_data &&
+		    (mi_enum_attr(mi, NULL) ||
+		     vbo <= ((u64)mi->rno << sbi->record_bits))) {
+			/* We can't accept this record 'cause MFT's bootstrapping. */
+			continue;
+		}
+		if (is_mft &&
+		    mi_find_attr(mi, NULL, ATTR_DATA, NULL, 0, NULL)) {
+			/*
+			 * This child record already has a ATTR_DATA.
+			 * So it can't accept any other records.
+			 */
+			continue;
+		}
+
+		if ((type != ATTR_NAME || name_len) &&
+		    mi_find_attr(mi, NULL, type, name, name_len, NULL)) {
+			/* Only indexed attributes can share same record. */
+			continue;
+		}
+
+		/* Try to insert attribute into this subrecord. */
+		attr = ni_ins_new_attr(ni, mi, le, type, name, name_len, asize,
+				       name_off, svcn, ins_le);
+		if (!attr)
+			continue;
+
+		if (ins_attr)
+			*ins_attr = attr;
+		if (ins_mi)
+			*ins_mi = mi;
+		return 0;
+	}
+
+insert_ext:
+	/* We have to allocate a new child subrecord. */
+	err = ntfs_look_free_mft(sbi, &rno, is_mft_data, ni, &mi);
+	if (err)
+		goto out;
+
+	if (is_mft_data && vbo <= ((u64)rno << sbi->record_bits)) {
+		err = -EINVAL;
+		goto out1;
+	}
+
+	attr = ni_ins_new_attr(ni, mi, le, type, name, name_len, asize,
+			       name_off, svcn, ins_le);
+	if (!attr)
+		goto out2;
+
+	if (ins_attr)
+		*ins_attr = attr;
+	if (ins_mi)
+		*ins_mi = mi;
+
+	return 0;
+
+out2:
+	ni_remove_mi(ni, mi);
+	mi_put(mi);
+	err = -EINVAL;
+
+out1:
+	ntfs_mark_rec_free(sbi, rno);
+
+out:
+	return err;
+}
+
+/*
+ * ni_insert_attr - Insert an attribute into the file.
+ *
+ * If the primary record has room, it will just insert the attribute.
+ * If not, it may make the attribute external.
+ * For $MFT::Data it may make room for the attribute by
+ * making other attributes external.
+ *
+ * NOTE:
+ * The ATTR_LIST and ATTR_STD cannot be made external.
+ * This function does not fill new attribute full.
+ * It only fills 'size'/'type'/'id'/'name_len' fields.
+ */
+static int ni_insert_attr(struct ntfs_inode *ni, enum ATTR_TYPE type,
+			  const __le16 *name, u8 name_len, u32 asize,
+			  u16 name_off, CLST svcn, struct ATTRIB **ins_attr,
+			  struct mft_inode **ins_mi,
+			  struct ATTR_LIST_ENTRY **ins_le)
+{
+	struct ntfs_sb_info *sbi = ni->mi.sbi;
+	int err;
+	struct ATTRIB *attr, *eattr;
+	struct MFT_REC *rec;
+	bool is_mft;
+	struct ATTR_LIST_ENTRY *le;
+	u32 list_reserve, max_free, free, used, t32;
+	__le16 id;
+	u16 t16;
+
+	is_mft = ni->mi.rno == MFT_REC_MFT;
+	rec = ni->mi.mrec;
+
+	list_reserve = SIZEOF_NONRESIDENT + 3 * (1 + 2 * sizeof(u32));
+	used = le32_to_cpu(rec->used);
+	free = sbi->record_size - used;
+
+	if (is_mft && type != ATTR_LIST) {
+		/* Reserve space for the ATTRIB list. */
+		if (free < list_reserve)
+			free = 0;
+		else
+			free -= list_reserve;
+	}
+
+	if (asize <= free) {
+		attr = ni_ins_new_attr(ni, &ni->mi, NULL, type, name, name_len,
+				       asize, name_off, svcn, ins_le);
+		if (attr) {
+			if (ins_attr)
+				*ins_attr = attr;
+			if (ins_mi)
+				*ins_mi = &ni->mi;
+			err = 0;
+			goto out;
+		}
+	}
+
+	if (!is_mft || type != ATTR_DATA || svcn) {
+		/* This ATTRIB will be external. */
+		err = ni_ins_attr_ext(ni, NULL, type, name, name_len, asize,
+				      svcn, name_off, false, ins_attr, ins_mi,
+				      ins_le);
+		goto out;
+	}
+
+	/*
+	 * Here we have: "is_mft && type == ATTR_DATA && !svcn"
+	 *
+	 * The first chunk of the $MFT::Data ATTRIB must be the base record.
+	 * Evict as many other attributes as possible.
+	 */
+	max_free = free;
+
+	/* Estimate the result of moving all possible attributes away. */
+	attr = NULL;
+
+	while ((attr = mi_enum_attr(&ni->mi, attr))) {
+		if (attr->type == ATTR_STD)
+			continue;
+		if (attr->type == ATTR_LIST)
+			continue;
+		max_free += le32_to_cpu(attr->size);
+	}
+
+	if (max_free < asize + list_reserve) {
+		/* Impossible to insert this attribute into primary record. */
+		err = -EINVAL;
+		goto out;
+	}
+
+	/* Start real attribute moving. */
+	attr = NULL;
+
+	for (;;) {
+		attr = mi_enum_attr(&ni->mi, attr);
+		if (!attr) {
+			/* We should never be here 'cause we have already check this case. */
+			err = -EINVAL;
+			goto out;
+		}
+
+		/* Skip attributes that MUST be primary record. */
+		if (attr->type == ATTR_STD || attr->type == ATTR_LIST)
+			continue;
+
+		le = NULL;
+		if (ni->attr_list.size) {
+			le = al_find_le(ni, NULL, attr);
+			if (!le) {
+				/* Really this is a serious bug. */
+				err = -EINVAL;
+				goto out;
+			}
+		}
+
+		t32 = le32_to_cpu(attr->size);
+		t16 = le16_to_cpu(attr->name_off);
+		err = ni_ins_attr_ext(ni, le, attr->type, Add2Ptr(attr, t16),
+				      attr->name_len, t32, attr_svcn(attr), t16,
+				      false, &eattr, NULL, NULL);
+		if (err)
+			return err;
+
+		id = eattr->id;
+		memcpy(eattr, attr, t32);
+		eattr->id = id;
+
+		/* Remove from primary record. */
+		mi_remove_attr(NULL, &ni->mi, attr);
+
+		/* attr now points to next attribute. */
+		if (attr->type == ATTR_END)
+			goto out;
+	}
+	while (asize + list_reserve > sbi->record_size - le32_to_cpu(rec->used))
+		;
+
+	attr = ni_ins_new_attr(ni, &ni->mi, NULL, type, name, name_len, asize,
+			       name_off, svcn, ins_le);
+	if (!attr) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	if (ins_attr)
+		*ins_attr = attr;
+	if (ins_mi)
+		*ins_mi = &ni->mi;
+
+out:
+	return err;
+}
+
+/* ni_expand_mft_list - Split ATTR_DATA of $MFT. */
+static int ni_expand_mft_list(struct ntfs_inode *ni)
+{
+	int err = 0;
+	struct runs_tree *run = &ni->file.run;
+	u32 asize, run_size, done = 0;
+	struct ATTRIB *attr;
+	struct rb_node *node;
+	CLST mft_min, mft_new, svcn, evcn, plen;
+	struct mft_inode *mi, *mi_min, *mi_new;
+	struct ntfs_sb_info *sbi = ni->mi.sbi;
+
+	/* Find the nearest MFT. */
+	mft_min = 0;
+	mft_new = 0;
+	mi_min = NULL;
+
+	for (node = rb_first(&ni->mi_tree); node; node = rb_next(node)) {
+		mi = rb_entry(node, struct mft_inode, node);
+
+		attr = mi_enum_attr(mi, NULL);
+
+		if (!attr) {
+			mft_min = mi->rno;
+			mi_min = mi;
+			break;
+		}
+	}
+
+	if (ntfs_look_free_mft(sbi, &mft_new, true, ni, &mi_new)) {
+		mft_new = 0;
+		/* Really this is not critical. */
+	} else if (mft_min > mft_new) {
+		mft_min = mft_new;
+		mi_min = mi_new;
+	} else {
+		ntfs_mark_rec_free(sbi, mft_new);
+		mft_new = 0;
+		ni_remove_mi(ni, mi_new);
+	}
+
+	attr = mi_find_attr(&ni->mi, NULL, ATTR_DATA, NULL, 0, NULL);
+	if (!attr) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	asize = le32_to_cpu(attr->size);
+
+	evcn = le64_to_cpu(attr->nres.evcn);
+	svcn = bytes_to_cluster(sbi, (u64)(mft_min + 1) << sbi->record_bits);
+	if (evcn + 1 >= svcn) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	/*
+	 * Split primary attribute [0 evcn] in two parts [0 svcn) + [svcn evcn].
+	 *
+	 * Update first part of ATTR_DATA in 'primary MFT.
+	 */
+	err = run_pack(run, 0, svcn, Add2Ptr(attr, SIZEOF_NONRESIDENT),
+		       asize - SIZEOF_NONRESIDENT, &plen);
+	if (err < 0)
+		goto out;
+
+	run_size = ALIGN(err, 8);
+	err = 0;
+
+	if (plen < svcn) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	attr->nres.evcn = cpu_to_le64(svcn - 1);
+	attr->size = cpu_to_le32(run_size + SIZEOF_NONRESIDENT);
+	/* 'done' - How many bytes of primary MFT becomes free. */
+	done = asize - run_size - SIZEOF_NONRESIDENT;
+	le32_sub_cpu(&ni->mi.mrec->used, done);
+
+	/* Estimate the size of second part: run_buf=NULL. */
+	err = run_pack(run, svcn, evcn + 1 - svcn, NULL, sbi->record_size,
+		       &plen);
+	if (err < 0)
+		goto out;
+
+	run_size = ALIGN(err, 8);
+	err = 0;
+
+	if (plen < evcn + 1 - svcn) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	/*
+	 * This function may implicitly call expand attr_list.
+	 * Insert second part of ATTR_DATA in 'mi_min'.
+	 */
+	attr = ni_ins_new_attr(ni, mi_min, NULL, ATTR_DATA, NULL, 0,
+			       SIZEOF_NONRESIDENT + run_size,
+			       SIZEOF_NONRESIDENT, svcn, NULL);
+	if (!attr) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	attr->non_res = 1;
+	attr->name_off = SIZEOF_NONRESIDENT_LE;
+	attr->flags = 0;
+
+	run_pack(run, svcn, evcn + 1 - svcn, Add2Ptr(attr, SIZEOF_NONRESIDENT),
+		 run_size, &plen);
+
+	attr->nres.svcn = cpu_to_le64(svcn);
+	attr->nres.evcn = cpu_to_le64(evcn);
+	attr->nres.run_off = cpu_to_le16(SIZEOF_NONRESIDENT);
+
+out:
+	if (mft_new) {
+		ntfs_mark_rec_free(sbi, mft_new);
+		ni_remove_mi(ni, mi_new);
+	}
+
+	return !err && !done ? -EOPNOTSUPP : err;
+}
+
+/*
+ * ni_expand_list - Move all possible attributes out of primary record.
+ */
+int ni_expand_list(struct ntfs_inode *ni)
+{
+	int err = 0;
+	u32 asize, done = 0;
+	struct ATTRIB *attr, *ins_attr;
+	struct ATTR_LIST_ENTRY *le;
+	bool is_mft = ni->mi.rno == MFT_REC_MFT;
+	struct MFT_REF ref;
+
+	mi_get_ref(&ni->mi, &ref);
+	le = NULL;
+
+	while ((le = al_enumerate(ni, le))) {
+		if (le->type == ATTR_STD)
+			continue;
+
+		if (memcmp(&ref, &le->ref, sizeof(struct MFT_REF)))
+			continue;
+
+		if (is_mft && le->type == ATTR_DATA)
+			continue;
+
+		/* Find attribute in primary record. */
+		attr = rec_find_attr_le(&ni->mi, le);
+		if (!attr) {
+			err = -EINVAL;
+			goto out;
+		}
+
+		asize = le32_to_cpu(attr->size);
+
+		/* Always insert into new record to avoid collisions (deep recursive). */
+		err = ni_ins_attr_ext(ni, le, attr->type, attr_name(attr),
+				      attr->name_len, asize, attr_svcn(attr),
+				      le16_to_cpu(attr->name_off), true,
+				      &ins_attr, NULL, NULL);
+
+		if (err)
+			goto out;
+
+		memcpy(ins_attr, attr, asize);
+		ins_attr->id = le->id;
+		/* Remove from primary record. */
+		mi_remove_attr(NULL, &ni->mi, attr);
+
+		done += asize;
+		goto out;
+	}
+
+	if (!is_mft) {
+		err = -EFBIG; /* Attr list is too big(?) */
+		goto out;
+	}
+
+	/* Split MFT data as much as possible. */
+	err = ni_expand_mft_list(ni);
+	if (err)
+		goto out;
+
+out:
+	return !err && !done ? -EOPNOTSUPP : err;
+}
+
+/*
+ * ni_insert_nonresident - Insert new nonresident attribute.
+ */
+int ni_insert_nonresident(struct ntfs_inode *ni, enum ATTR_TYPE type,
+			  const __le16 *name, u8 name_len,
+			  const struct runs_tree *run, CLST svcn, CLST len,
+			  __le16 flags, struct ATTRIB **new_attr,
+			  struct mft_inode **mi)
+{
+	int err;
+	CLST plen;
+	struct ATTRIB *attr;
+	bool is_ext =
+		(flags & (ATTR_FLAG_SPARSED | ATTR_FLAG_COMPRESSED)) && !svcn;
+	u32 name_size = ALIGN(name_len * sizeof(short), 8);
+	u32 name_off = is_ext ? SIZEOF_NONRESIDENT_EX : SIZEOF_NONRESIDENT;
+	u32 run_off = name_off + name_size;
+	u32 run_size, asize;
+	struct ntfs_sb_info *sbi = ni->mi.sbi;
+
+	err = run_pack(run, svcn, len, NULL, sbi->max_bytes_per_attr - run_off,
+		       &plen);
+	if (err < 0)
+		goto out;
+
+	run_size = ALIGN(err, 8);
+
+	if (plen < len) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	asize = run_off + run_size;
+
+	if (asize > sbi->max_bytes_per_attr) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	err = ni_insert_attr(ni, type, name, name_len, asize, name_off, svcn,
+			     &attr, mi, NULL);
+
+	if (err)
+		goto out;
+
+	attr->non_res = 1;
+	attr->name_off = cpu_to_le16(name_off);
+	attr->flags = flags;
+
+	run_pack(run, svcn, len, Add2Ptr(attr, run_off), run_size, &plen);
+
+	attr->nres.svcn = cpu_to_le64(svcn);
+	attr->nres.evcn = cpu_to_le64((u64)svcn + len - 1);
+
+	err = 0;
+	if (new_attr)
+		*new_attr = attr;
+
+	*(__le64 *)&attr->nres.run_off = cpu_to_le64(run_off);
+
+	attr->nres.alloc_size =
+		svcn ? 0 : cpu_to_le64((u64)len << ni->mi.sbi->cluster_bits);
+	attr->nres.data_size = attr->nres.alloc_size;
+	attr->nres.valid_size = attr->nres.alloc_size;
+
+	if (is_ext) {
+		if (flags & ATTR_FLAG_COMPRESSED)
+			attr->nres.c_unit = COMPRESSION_UNIT;
+		attr->nres.total_size = attr->nres.alloc_size;
+	}
+
+out:
+	return err;
+}
+
+/*
+ * ni_insert_resident - Inserts new resident attribute.
+ */
+int ni_insert_resident(struct ntfs_inode *ni, u32 data_size,
+		       enum ATTR_TYPE type, const __le16 *name, u8 name_len,
+		       struct ATTRIB **new_attr, struct mft_inode **mi,
+		       struct ATTR_LIST_ENTRY **le)
+{
+	int err;
+	u32 name_size = ALIGN(name_len * sizeof(short), 8);
+	u32 asize = SIZEOF_RESIDENT + name_size + ALIGN(data_size, 8);
+	struct ATTRIB *attr;
+
+	err = ni_insert_attr(ni, type, name, name_len, asize, SIZEOF_RESIDENT,
+			     0, &attr, mi, le);
+	if (err)
+		return err;
+
+	attr->non_res = 0;
+	attr->flags = 0;
+
+	attr->res.data_size = cpu_to_le32(data_size);
+	attr->res.data_off = cpu_to_le16(SIZEOF_RESIDENT + name_size);
+	if (type == ATTR_NAME) {
+		attr->res.flags = RESIDENT_FLAG_INDEXED;
+
+		/* is_attr_indexed(attr)) == true */
+		le16_add_cpu(&ni->mi.mrec->hard_links, +1);
+		ni->mi.dirty = true;
+	}
+	attr->res.res = 0;
+
+	if (new_attr)
+		*new_attr = attr;
+
+	return 0;
+}
+
+/*
+ * ni_remove_attr_le - Remove attribute from record.
+ */
+void ni_remove_attr_le(struct ntfs_inode *ni, struct ATTRIB *attr,
+		       struct mft_inode *mi, struct ATTR_LIST_ENTRY *le)
+{
+	mi_remove_attr(ni, mi, attr);
+
+	if (le)
+		al_remove_le(ni, le);
+}
+
+/*
+ * ni_delete_all - Remove all attributes and frees allocates space.
+ *
+ * ntfs_evict_inode->ntfs_clear_inode->ni_delete_all (if no links).
+ */
+int ni_delete_all(struct ntfs_inode *ni)
+{
+	int err;
+	struct ATTR_LIST_ENTRY *le = NULL;
+	struct ATTRIB *attr = NULL;
+	struct rb_node *node;
+	u16 roff;
+	u32 asize;
+	CLST svcn, evcn;
+	struct ntfs_sb_info *sbi = ni->mi.sbi;
+	bool nt3 = is_ntfs3(sbi);
+	struct MFT_REF ref;
+
+	while ((attr = ni_enum_attr_ex(ni, attr, &le, NULL))) {
+		if (!nt3 || attr->name_len) {
+			;
+		} else if (attr->type == ATTR_REPARSE) {
+			mi_get_ref(&ni->mi, &ref);
+			ntfs_remove_reparse(sbi, 0, &ref);
+		} else if (attr->type == ATTR_ID && !attr->non_res &&
+			   le32_to_cpu(attr->res.data_size) >=
+				   sizeof(struct GUID)) {
+			ntfs_objid_remove(sbi, resident_data(attr));
+		}
+
+		if (!attr->non_res)
+			continue;
+
+		svcn = le64_to_cpu(attr->nres.svcn);
+		evcn = le64_to_cpu(attr->nres.evcn);
+
+		if (evcn + 1 <= svcn)
+			continue;
+
+		asize = le32_to_cpu(attr->size);
+		roff = le16_to_cpu(attr->nres.run_off);
+
+		/* run==1 means unpack and deallocate. */
+		run_unpack_ex(RUN_DEALLOCATE, sbi, ni->mi.rno, svcn, evcn, svcn,
+			      Add2Ptr(attr, roff), asize - roff);
+	}
+
+	if (ni->attr_list.size) {
+		run_deallocate(ni->mi.sbi, &ni->attr_list.run, true);
+		al_destroy(ni);
+	}
+
+	/* Free all subrecords. */
+	for (node = rb_first(&ni->mi_tree); node;) {
+		struct rb_node *next = rb_next(node);
+		struct mft_inode *mi = rb_entry(node, struct mft_inode, node);
+
+		clear_rec_inuse(mi->mrec);
+		mi->dirty = true;
+		mi_write(mi, 0);
+
+		ntfs_mark_rec_free(sbi, mi->rno);
+		ni_remove_mi(ni, mi);
+		mi_put(mi);
+		node = next;
+	}
+
+	/* Free base record. */
+	clear_rec_inuse(ni->mi.mrec);
+	ni->mi.dirty = true;
+	err = mi_write(&ni->mi, 0);
+
+	ntfs_mark_rec_free(sbi, ni->mi.rno);
+
+	return err;
+}
+
+/* ni_fname_name
+ *
+ * Return: File name attribute by its value.
+ */
+struct ATTR_FILE_NAME *ni_fname_name(struct ntfs_inode *ni,
+				     const struct cpu_str *uni,
+				     const struct MFT_REF *home_dir,
+				     struct mft_inode **mi,
+				     struct ATTR_LIST_ENTRY **le)
+{
+	struct ATTRIB *attr = NULL;
+	struct ATTR_FILE_NAME *fname;
+
+	*le = NULL;
+
+	/* Enumerate all names. */
+next:
+	attr = ni_find_attr(ni, attr, le, ATTR_NAME, NULL, 0, NULL, mi);
+	if (!attr)
+		return NULL;
+
+	fname = resident_data_ex(attr, SIZEOF_ATTRIBUTE_FILENAME);
+	if (!fname)
+		goto next;
+
+	if (home_dir && memcmp(home_dir, &fname->home, sizeof(*home_dir)))
+		goto next;
+
+	if (!uni)
+		goto next;
+
+	if (uni->len != fname->name_len)
+		goto next;
+
+	if (ntfs_cmp_names_cpu(uni, (struct le_str *)&fname->name_len, NULL,
+			       false))
+		goto next;
+
+	return fname;
+}
+
+/*
+ * ni_fname_type
+ *
+ * Return: File name attribute with given type.
+ */
+struct ATTR_FILE_NAME *ni_fname_type(struct ntfs_inode *ni, u8 name_type,
+				     struct mft_inode **mi,
+				     struct ATTR_LIST_ENTRY **le)
+{
+	struct ATTRIB *attr = NULL;
+	struct ATTR_FILE_NAME *fname;
+
+	*le = NULL;
+
+	if (FILE_NAME_POSIX == name_type)
+		return NULL;
+
+	/* Enumerate all names. */
+	for (;;) {
+		attr = ni_find_attr(ni, attr, le, ATTR_NAME, NULL, 0, NULL, mi);
+		if (!attr)
+			return NULL;
+
+		fname = resident_data_ex(attr, SIZEOF_ATTRIBUTE_FILENAME);
+		if (fname && name_type == fname->type)
+			return fname;
+	}
+}
+
+/*
+ * ni_new_attr_flags
+ *
+ * Process compressed/sparsed in special way.
+ * NOTE: You need to set ni->std_fa = new_fa
+ * after this function to keep internal structures in consistency.
+ */
+int ni_new_attr_flags(struct ntfs_inode *ni, enum FILE_ATTRIBUTE new_fa)
+{
+	struct ATTRIB *attr;
+	struct mft_inode *mi;
+	__le16 new_aflags;
+	u32 new_asize;
+
+	attr = ni_find_attr(ni, NULL, NULL, ATTR_DATA, NULL, 0, NULL, &mi);
+	if (!attr)
+		return -EINVAL;
+
+	new_aflags = attr->flags;
+
+	if (new_fa & FILE_ATTRIBUTE_SPARSE_FILE)
+		new_aflags |= ATTR_FLAG_SPARSED;
+	else
+		new_aflags &= ~ATTR_FLAG_SPARSED;
+
+	if (new_fa & FILE_ATTRIBUTE_COMPRESSED)
+		new_aflags |= ATTR_FLAG_COMPRESSED;
+	else
+		new_aflags &= ~ATTR_FLAG_COMPRESSED;
+
+	if (new_aflags == attr->flags)
+		return 0;
+
+	if ((new_aflags & (ATTR_FLAG_COMPRESSED | ATTR_FLAG_SPARSED)) ==
+	    (ATTR_FLAG_COMPRESSED | ATTR_FLAG_SPARSED)) {
+		ntfs_inode_warn(&ni->vfs_inode,
+				"file can't be sparsed and compressed");
+		return -EOPNOTSUPP;
+	}
+
+	if (!attr->non_res)
+		goto out;
+
+	if (attr->nres.data_size) {
+		ntfs_inode_warn(
+			&ni->vfs_inode,
+			"one can change sparsed/compressed only for empty files");
+		return -EOPNOTSUPP;
+	}
+
+	/* Resize nonresident empty attribute in-place only. */
+	new_asize = (new_aflags & (ATTR_FLAG_COMPRESSED | ATTR_FLAG_SPARSED))
+			    ? (SIZEOF_NONRESIDENT_EX + 8)
+			    : (SIZEOF_NONRESIDENT + 8);
+
+	if (!mi_resize_attr(mi, attr, new_asize - le32_to_cpu(attr->size)))
+		return -EOPNOTSUPP;
+
+	if (new_aflags & ATTR_FLAG_SPARSED) {
+		attr->name_off = SIZEOF_NONRESIDENT_EX_LE;
+		/* Windows uses 16 clusters per frame but supports one cluster per frame too. */
+		attr->nres.c_unit = 0;
+		ni->vfs_inode.i_mapping->a_ops = &ntfs_aops;
+	} else if (new_aflags & ATTR_FLAG_COMPRESSED) {
+		attr->name_off = SIZEOF_NONRESIDENT_EX_LE;
+		/* The only allowed: 16 clusters per frame. */
+		attr->nres.c_unit = NTFS_LZNT_CUNIT;
+		ni->vfs_inode.i_mapping->a_ops = &ntfs_aops_cmpr;
+	} else {
+		attr->name_off = SIZEOF_NONRESIDENT_LE;
+		/* Normal files. */
+		attr->nres.c_unit = 0;
+		ni->vfs_inode.i_mapping->a_ops = &ntfs_aops;
+	}
+	attr->nres.run_off = attr->name_off;
+out:
+	attr->flags = new_aflags;
+	mi->dirty = true;
+
+	return 0;
+}
+
+/*
+ * ni_parse_reparse
+ *
+ * Buffer is at least 24 bytes.
+ */
+enum REPARSE_SIGN ni_parse_reparse(struct ntfs_inode *ni, struct ATTRIB *attr,
+				   void *buffer)
+{
+	const struct REPARSE_DATA_BUFFER *rp = NULL;
+	u8 bits;
+	u16 len;
+	typeof(rp->CompressReparseBuffer) *cmpr;
+
+	static_assert(sizeof(struct REPARSE_DATA_BUFFER) <= 24);
+
+	/* Try to estimate reparse point. */
+	if (!attr->non_res) {
+		rp = resident_data_ex(attr, sizeof(struct REPARSE_DATA_BUFFER));
+	} else if (le64_to_cpu(attr->nres.data_size) >=
+		   sizeof(struct REPARSE_DATA_BUFFER)) {
+		struct runs_tree run;
+
+		run_init(&run);
+
+		if (!attr_load_runs_vcn(ni, ATTR_REPARSE, NULL, 0, &run, 0) &&
+		    !ntfs_read_run_nb(ni->mi.sbi, &run, 0, buffer,
+				      sizeof(struct REPARSE_DATA_BUFFER),
+				      NULL)) {
+			rp = buffer;
+		}
+
+		run_close(&run);
+	}
+
+	if (!rp)
+		return REPARSE_NONE;
+
+	len = le16_to_cpu(rp->ReparseDataLength);
+	switch (rp->ReparseTag) {
+	case (IO_REPARSE_TAG_MICROSOFT | IO_REPARSE_TAG_SYMBOLIC_LINK):
+		break; /* Symbolic link. */
+	case IO_REPARSE_TAG_MOUNT_POINT:
+		break; /* Mount points and junctions. */
+	case IO_REPARSE_TAG_SYMLINK:
+		break;
+	case IO_REPARSE_TAG_COMPRESS:
+		/*
+		 * WOF - Windows Overlay Filter - Used to compress files with
+		 * LZX/Xpress.
+		 *
+		 * Unlike native NTFS file compression, the Windows
+		 * Overlay Filter supports only read operations. This means
+		 * that it doesn't need to sector-align each compressed chunk,
+		 * so the compressed data can be packed more tightly together.
+		 * If you open the file for writing, the WOF just decompresses
+		 * the entire file, turning it back into a plain file.
+		 *
+		 * Ntfs3 driver decompresses the entire file only on write or
+		 * change size requests.
+		 */
+
+		cmpr = &rp->CompressReparseBuffer;
+		if (len < sizeof(*cmpr) ||
+		    cmpr->WofVersion != WOF_CURRENT_VERSION ||
+		    cmpr->WofProvider != WOF_PROVIDER_SYSTEM ||
+		    cmpr->ProviderVer != WOF_PROVIDER_CURRENT_VERSION) {
+			return REPARSE_NONE;
+		}
+
+		switch (cmpr->CompressionFormat) {
+		case WOF_COMPRESSION_XPRESS4K:
+			bits = 0xc; // 4k
+			break;
+		case WOF_COMPRESSION_XPRESS8K:
+			bits = 0xd; // 8k
+			break;
+		case WOF_COMPRESSION_XPRESS16K:
+			bits = 0xe; // 16k
+			break;
+		case WOF_COMPRESSION_LZX32K:
+			bits = 0xf; // 32k
+			break;
+		default:
+			bits = 0x10; // 64k
+			break;
+		}
+		ni_set_ext_compress_bits(ni, bits);
+		return REPARSE_COMPRESSED;
+
+	case IO_REPARSE_TAG_DEDUP:
+		ni->ni_flags |= NI_FLAG_DEDUPLICATED;
+		return REPARSE_DEDUPLICATED;
+
+	default:
+		if (rp->ReparseTag & IO_REPARSE_TAG_NAME_SURROGATE)
+			break;
+
+		return REPARSE_NONE;
+	}
+
+	/* Looks like normal symlink. */
+	return REPARSE_LINK;
+}
+
+/*
+ * ni_fiemap - Helper for file_fiemap().
+ *
+ * Assumed ni_lock.
+ * TODO: Less aggressive locks.
+ */
+int ni_fiemap(struct ntfs_inode *ni, struct fiemap_extent_info *fieinfo,
+	      __u64 vbo, __u64 len)
+{
+	int err = 0;
+	struct ntfs_sb_info *sbi = ni->mi.sbi;
+	u8 cluster_bits = sbi->cluster_bits;
+	struct runs_tree *run;
+	struct rw_semaphore *run_lock;
+	struct ATTRIB *attr;
+	CLST vcn = vbo >> cluster_bits;
+	CLST lcn, clen;
+	u64 valid = ni->i_valid;
+	u64 lbo, bytes;
+	u64 end, alloc_size;
+	size_t idx = -1;
+	u32 flags;
+	bool ok;
+
+	if (S_ISDIR(ni->vfs_inode.i_mode)) {
+		run = &ni->dir.alloc_run;
+		attr = ni_find_attr(ni, NULL, NULL, ATTR_ALLOC, I30_NAME,
+				    ARRAY_SIZE(I30_NAME), NULL, NULL);
+		run_lock = &ni->dir.run_lock;
+	} else {
+		run = &ni->file.run;
+		attr = ni_find_attr(ni, NULL, NULL, ATTR_DATA, NULL, 0, NULL,
+				    NULL);
+		if (!attr) {
+			err = -EINVAL;
+			goto out;
+		}
+		if (is_attr_compressed(attr)) {
+			/* Unfortunately cp -r incorrectly treats compressed clusters. */
+			err = -EOPNOTSUPP;
+			ntfs_inode_warn(
+				&ni->vfs_inode,
+				"fiemap is not supported for compressed file (cp -r)");
+			goto out;
+		}
+		run_lock = &ni->file.run_lock;
+	}
+
+	if (!attr || !attr->non_res) {
+		err = fiemap_fill_next_extent(
+			fieinfo, 0, 0,
+			attr ? le32_to_cpu(attr->res.data_size) : 0,
+			FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_LAST |
+				FIEMAP_EXTENT_MERGED);
+		goto out;
+	}
+
+	end = vbo + len;
+	alloc_size = le64_to_cpu(attr->nres.alloc_size);
+	if (end > alloc_size)
+		end = alloc_size;
+
+	down_read(run_lock);
+
+	while (vbo < end) {
+		if (idx == -1) {
+			ok = run_lookup_entry(run, vcn, &lcn, &clen, &idx);
+		} else {
+			CLST vcn_next = vcn;
+
+			ok = run_get_entry(run, ++idx, &vcn, &lcn, &clen) &&
+			     vcn == vcn_next;
+			if (!ok)
+				vcn = vcn_next;
+		}
+
+		if (!ok) {
+			up_read(run_lock);
+			down_write(run_lock);
+
+			err = attr_load_runs_vcn(ni, attr->type,
+						 attr_name(attr),
+						 attr->name_len, run, vcn);
+
+			up_write(run_lock);
+			down_read(run_lock);
+
+			if (err)
+				break;
+
+			ok = run_lookup_entry(run, vcn, &lcn, &clen, &idx);
+
+			if (!ok) {
+				err = -EINVAL;
+				break;
+			}
+		}
+
+		if (!clen) {
+			err = -EINVAL; // ?
+			break;
+		}
+
+		if (lcn == SPARSE_LCN) {
+			vcn += clen;
+			vbo = (u64)vcn << cluster_bits;
+			continue;
+		}
+
+		flags = FIEMAP_EXTENT_MERGED;
+		if (S_ISDIR(ni->vfs_inode.i_mode)) {
+			;
+		} else if (is_attr_compressed(attr)) {
+			CLST clst_data;
+
+			err = attr_is_frame_compressed(
+				ni, attr, vcn >> attr->nres.c_unit, &clst_data);
+			if (err)
+				break;
+			if (clst_data < NTFS_LZNT_CLUSTERS)
+				flags |= FIEMAP_EXTENT_ENCODED;
+		} else if (is_attr_encrypted(attr)) {
+			flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
+		}
+
+		vbo = (u64)vcn << cluster_bits;
+		bytes = (u64)clen << cluster_bits;
+		lbo = (u64)lcn << cluster_bits;
+
+		vcn += clen;
+
+		if (vbo + bytes >= end) {
+			bytes = end - vbo;
+			flags |= FIEMAP_EXTENT_LAST;
+		}
+
+		if (vbo + bytes <= valid) {
+			;
+		} else if (vbo >= valid) {
+			flags |= FIEMAP_EXTENT_UNWRITTEN;
+		} else {
+			/* vbo < valid && valid < vbo + bytes */
+			u64 dlen = valid - vbo;
+
+			err = fiemap_fill_next_extent(fieinfo, vbo, lbo, dlen,
+						      flags);
+			if (err < 0)
+				break;
+			if (err == 1) {
+				err = 0;
+				break;
+			}
+
+			vbo = valid;
+			bytes -= dlen;
+			if (!bytes)
+				continue;
+
+			lbo += dlen;
+			flags |= FIEMAP_EXTENT_UNWRITTEN;
+		}
+
+		err = fiemap_fill_next_extent(fieinfo, vbo, lbo, bytes, flags);
+		if (err < 0)
+			break;
+		if (err == 1) {
+			err = 0;
+			break;
+		}
+
+		vbo += bytes;
+	}
+
+	up_read(run_lock);
+
+out:
+	return err;
+}
+
+/*
+ * ni_readpage_cmpr
+ *
+ * When decompressing, we typically obtain more than one page per reference.
+ * We inject the additional pages into the page cache.
+ */
+int ni_readpage_cmpr(struct ntfs_inode *ni, struct page *page)
+{
+	int err;
+	struct ntfs_sb_info *sbi = ni->mi.sbi;
+	struct address_space *mapping = page->mapping;
+	pgoff_t index = page->index;
+	u64 frame_vbo, vbo = (u64)index << PAGE_SHIFT;
+	struct page **pages = NULL; /* Array of at most 16 pages. stack? */
+	u8 frame_bits;
+	CLST frame;
+	u32 i, idx, frame_size, pages_per_frame;
+	gfp_t gfp_mask;
+	struct page *pg;
+
+	if (vbo >= ni->vfs_inode.i_size) {
+		SetPageUptodate(page);
+		err = 0;
+		goto out;
+	}
+
+	if (ni->ni_flags & NI_FLAG_COMPRESSED_MASK) {
+		/* Xpress or LZX. */
+		frame_bits = ni_ext_compress_bits(ni);
+	} else {
+		/* LZNT compression. */
+		frame_bits = NTFS_LZNT_CUNIT + sbi->cluster_bits;
+	}
+	frame_size = 1u << frame_bits;
+	frame = vbo >> frame_bits;
+	frame_vbo = (u64)frame << frame_bits;
+	idx = (vbo - frame_vbo) >> PAGE_SHIFT;
+
+	pages_per_frame = frame_size >> PAGE_SHIFT;
+	pages = kcalloc(pages_per_frame, sizeof(struct page *), GFP_NOFS);
+	if (!pages) {
+		err = -ENOMEM;
+		goto out;
+	}
+
+	pages[idx] = page;
+	index = frame_vbo >> PAGE_SHIFT;
+	gfp_mask = mapping_gfp_mask(mapping);
+
+	for (i = 0; i < pages_per_frame; i++, index++) {
+		if (i == idx)
+			continue;
+
+		pg = find_or_create_page(mapping, index, gfp_mask);
+		if (!pg) {
+			err = -ENOMEM;
+			goto out1;
+		}
+		pages[i] = pg;
+	}
+
+	err = ni_read_frame(ni, frame_vbo, pages, pages_per_frame);
+
+out1:
+	if (err)
+		SetPageError(page);
+
+	for (i = 0; i < pages_per_frame; i++) {
+		pg = pages[i];
+		if (i == idx)
+			continue;
+		unlock_page(pg);
+		put_page(pg);
+	}
+
+out:
+	/* At this point, err contains 0 or -EIO depending on the "critical" page. */
+	kfree(pages);
+	unlock_page(page);
+
+	return err;
+}
+
+#ifdef CONFIG_NTFS3_LZX_XPRESS
+/*
+ * ni_decompress_file - Decompress LZX/Xpress compressed file.
+ *
+ * Remove ATTR_DATA::WofCompressedData.
+ * Remove ATTR_REPARSE.
+ */
+int ni_decompress_file(struct ntfs_inode *ni)
+{
+	struct ntfs_sb_info *sbi = ni->mi.sbi;
+	struct inode *inode = &ni->vfs_inode;
+	loff_t i_size = inode->i_size;
+	struct address_space *mapping = inode->i_mapping;
+	gfp_t gfp_mask = mapping_gfp_mask(mapping);
+	struct page **pages = NULL;
+	struct ATTR_LIST_ENTRY *le;
+	struct ATTRIB *attr;
+	CLST vcn, cend, lcn, clen, end;
+	pgoff_t index;
+	u64 vbo;
+	u8 frame_bits;
+	u32 i, frame_size, pages_per_frame, bytes;
+	struct mft_inode *mi;
+	int err;
+
+	/* Clusters for decompressed data. */
+	cend = bytes_to_cluster(sbi, i_size);
+
+	if (!i_size)
+		goto remove_wof;
+
+	/* Check in advance. */
+	if (cend > wnd_zeroes(&sbi->used.bitmap)) {
+		err = -ENOSPC;
+		goto out;
+	}
+
+	frame_bits = ni_ext_compress_bits(ni);
+	frame_size = 1u << frame_bits;
+	pages_per_frame = frame_size >> PAGE_SHIFT;
+	pages = kcalloc(pages_per_frame, sizeof(struct page *), GFP_NOFS);
+	if (!pages) {
+		err = -ENOMEM;
+		goto out;
+	}
+
+	/*
+	 * Step 1: Decompress data and copy to new allocated clusters.
+	 */
+	index = 0;
+	for (vbo = 0; vbo < i_size; vbo += bytes) {
+		u32 nr_pages;
+		bool new;
+
+		if (vbo + frame_size > i_size) {
+			bytes = i_size - vbo;
+			nr_pages = (bytes + PAGE_SIZE - 1) >> PAGE_SHIFT;
+		} else {
+			nr_pages = pages_per_frame;
+			bytes = frame_size;
+		}
+
+		end = bytes_to_cluster(sbi, vbo + bytes);
+
+		for (vcn = vbo >> sbi->cluster_bits; vcn < end; vcn += clen) {
+			err = attr_data_get_block(ni, vcn, cend - vcn, &lcn,
+						  &clen, &new);
+			if (err)
+				goto out;
+		}
+
+		for (i = 0; i < pages_per_frame; i++, index++) {
+			struct page *pg;
+
+			pg = find_or_create_page(mapping, index, gfp_mask);
+			if (!pg) {
+				while (i--) {
+					unlock_page(pages[i]);
+					put_page(pages[i]);
+				}
+				err = -ENOMEM;
+				goto out;
+			}
+			pages[i] = pg;
+		}
+
+		err = ni_read_frame(ni, vbo, pages, pages_per_frame);
+
+		if (!err) {
+			down_read(&ni->file.run_lock);
+			err = ntfs_bio_pages(sbi, &ni->file.run, pages,
+					     nr_pages, vbo, bytes,
+					     REQ_OP_WRITE);
+			up_read(&ni->file.run_lock);
+		}
+
+		for (i = 0; i < pages_per_frame; i++) {
+			unlock_page(pages[i]);
+			put_page(pages[i]);
+		}
+
+		if (err)
+			goto out;
+
+		cond_resched();
+	}
+
+remove_wof:
+	/*
+	 * Step 2: Deallocate attributes ATTR_DATA::WofCompressedData
+	 * and ATTR_REPARSE.
+	 */
+	attr = NULL;
+	le = NULL;
+	while ((attr = ni_enum_attr_ex(ni, attr, &le, NULL))) {
+		CLST svcn, evcn;
+		u32 asize, roff;
+
+		if (attr->type == ATTR_REPARSE) {
+			struct MFT_REF ref;
+
+			mi_get_ref(&ni->mi, &ref);
+			ntfs_remove_reparse(sbi, 0, &ref);
+		}
+
+		if (!attr->non_res)
+			continue;
+
+		if (attr->type != ATTR_REPARSE &&
+		    (attr->type != ATTR_DATA ||
+		     attr->name_len != ARRAY_SIZE(WOF_NAME) ||
+		     memcmp(attr_name(attr), WOF_NAME, sizeof(WOF_NAME))))
+			continue;
+
+		svcn = le64_to_cpu(attr->nres.svcn);
+		evcn = le64_to_cpu(attr->nres.evcn);
+
+		if (evcn + 1 <= svcn)
+			continue;
+
+		asize = le32_to_cpu(attr->size);
+		roff = le16_to_cpu(attr->nres.run_off);
+
+		/*run==1  Means unpack and deallocate. */
+		run_unpack_ex(RUN_DEALLOCATE, sbi, ni->mi.rno, svcn, evcn, svcn,
+			      Add2Ptr(attr, roff), asize - roff);
+	}
+
+	/*
+	 * Step 3: Remove attribute ATTR_DATA::WofCompressedData.
+	 */
+	err = ni_remove_attr(ni, ATTR_DATA, WOF_NAME, ARRAY_SIZE(WOF_NAME),
+			     false, NULL);
+	if (err)
+		goto out;
+
+	/*
+	 * Step 4: Remove ATTR_REPARSE.
+	 */
+	err = ni_remove_attr(ni, ATTR_REPARSE, NULL, 0, false, NULL);
+	if (err)
+		goto out;
+
+	/*
+	 * Step 5: Remove sparse flag from data attribute.
+	 */
+	attr = ni_find_attr(ni, NULL, NULL, ATTR_DATA, NULL, 0, NULL, &mi);
+	if (!attr) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	if (attr->non_res && is_attr_sparsed(attr)) {
+		/* Sparsed attribute header is 8 bytes bigger than normal. */
+		struct MFT_REC *rec = mi->mrec;
+		u32 used = le32_to_cpu(rec->used);
+		u32 asize = le32_to_cpu(attr->size);
+		u16 roff = le16_to_cpu(attr->nres.run_off);
+		char *rbuf = Add2Ptr(attr, roff);
+
+		memmove(rbuf - 8, rbuf, used - PtrOffset(rec, rbuf));
+		attr->size = cpu_to_le32(asize - 8);
+		attr->flags &= ~ATTR_FLAG_SPARSED;
+		attr->nres.run_off = cpu_to_le16(roff - 8);
+		attr->nres.c_unit = 0;
+		rec->used = cpu_to_le32(used - 8);
+		mi->dirty = true;
+		ni->std_fa &= ~(FILE_ATTRIBUTE_SPARSE_FILE |
+				FILE_ATTRIBUTE_REPARSE_POINT);
+
+		mark_inode_dirty(inode);
+	}
+
+	/* Clear cached flag. */
+	ni->ni_flags &= ~NI_FLAG_COMPRESSED_MASK;
+	if (ni->file.offs_page) {
+		put_page(ni->file.offs_page);
+		ni->file.offs_page = NULL;
+	}
+	mapping->a_ops = &ntfs_aops;
+
+out:
+	kfree(pages);
+	if (err) {
+		make_bad_inode(inode);
+		ntfs_set_state(sbi, NTFS_DIRTY_ERROR);
+	}
+
+	return err;
+}
+
+/*
+ * decompress_lzx_xpress - External compression LZX/Xpress.
+ */
+static int decompress_lzx_xpress(struct ntfs_sb_info *sbi, const char *cmpr,
+				 size_t cmpr_size, void *unc, size_t unc_size,
+				 u32 frame_size)
+{
+	int err;
+	void *ctx;
+
+	if (cmpr_size == unc_size) {
+		/* Frame not compressed. */
+		memcpy(unc, cmpr, unc_size);
+		return 0;
+	}
+
+	err = 0;
+	if (frame_size == 0x8000) {
+		mutex_lock(&sbi->compress.mtx_lzx);
+		/* LZX: Frame compressed. */
+		ctx = sbi->compress.lzx;
+		if (!ctx) {
+			/* Lazy initialize LZX decompress context. */
+			ctx = lzx_allocate_decompressor();
+			if (!ctx) {
+				err = -ENOMEM;
+				goto out1;
+			}
+
+			sbi->compress.lzx = ctx;
+		}
+
+		if (lzx_decompress(ctx, cmpr, cmpr_size, unc, unc_size)) {
+			/* Treat all errors as "invalid argument". */
+			err = -EINVAL;
+		}
+out1:
+		mutex_unlock(&sbi->compress.mtx_lzx);
+	} else {
+		/* XPRESS: Frame compressed. */
+		mutex_lock(&sbi->compress.mtx_xpress);
+		ctx = sbi->compress.xpress;
+		if (!ctx) {
+			/* Lazy initialize Xpress decompress context. */
+			ctx = xpress_allocate_decompressor();
+			if (!ctx) {
+				err = -ENOMEM;
+				goto out2;
+			}
+
+			sbi->compress.xpress = ctx;
+		}
+
+		if (xpress_decompress(ctx, cmpr, cmpr_size, unc, unc_size)) {
+			/* Treat all errors as "invalid argument". */
+			err = -EINVAL;
+		}
+out2:
+		mutex_unlock(&sbi->compress.mtx_xpress);
+	}
+	return err;
+}
+#endif
+
+/*
+ * ni_read_frame
+ *
+ * Pages - Array of locked pages.
+ */
+int ni_read_frame(struct ntfs_inode *ni, u64 frame_vbo, struct page **pages,
+		  u32 pages_per_frame)
+{
+	int err;
+	struct ntfs_sb_info *sbi = ni->mi.sbi;
+	u8 cluster_bits = sbi->cluster_bits;
+	char *frame_ondisk = NULL;
+	char *frame_mem = NULL;
+	struct page **pages_disk = NULL;
+	struct ATTR_LIST_ENTRY *le = NULL;
+	struct runs_tree *run = &ni->file.run;
+	u64 valid_size = ni->i_valid;
+	u64 vbo_disk;
+	size_t unc_size;
+	u32 frame_size, i, npages_disk, ondisk_size;
+	struct page *pg;
+	struct ATTRIB *attr;
+	CLST frame, clst_data;
+
+	/*
+	 * To simplify decompress algorithm do vmap for source
+	 * and target pages.
+	 */
+	for (i = 0; i < pages_per_frame; i++)
+		kmap(pages[i]);
+
+	frame_size = pages_per_frame << PAGE_SHIFT;
+	frame_mem = vmap(pages, pages_per_frame, VM_MAP, PAGE_KERNEL);
+	if (!frame_mem) {
+		err = -ENOMEM;
+		goto out;
+	}
+
+	attr = ni_find_attr(ni, NULL, &le, ATTR_DATA, NULL, 0, NULL, NULL);
+	if (!attr) {
+		err = -ENOENT;
+		goto out1;
+	}
+
+	if (!attr->non_res) {
+		u32 data_size = le32_to_cpu(attr->res.data_size);
+
+		memset(frame_mem, 0, frame_size);
+		if (frame_vbo < data_size) {
+			ondisk_size = data_size - frame_vbo;
+			memcpy(frame_mem, resident_data(attr) + frame_vbo,
+			       min(ondisk_size, frame_size));
+		}
+		err = 0;
+		goto out1;
+	}
+
+	if (frame_vbo >= valid_size) {
+		memset(frame_mem, 0, frame_size);
+		err = 0;
+		goto out1;
+	}
+
+	if (ni->ni_flags & NI_FLAG_COMPRESSED_MASK) {
+#ifndef CONFIG_NTFS3_LZX_XPRESS
+		err = -EOPNOTSUPP;
+		goto out1;
+#else
+		u32 frame_bits = ni_ext_compress_bits(ni);
+		u64 frame64 = frame_vbo >> frame_bits;
+		u64 frames, vbo_data;
+
+		if (frame_size != (1u << frame_bits)) {
+			err = -EINVAL;
+			goto out1;
+		}
+		switch (frame_size) {
+		case 0x1000:
+		case 0x2000:
+		case 0x4000:
+		case 0x8000:
+			break;
+		default:
+			/* Unknown compression. */
+			err = -EOPNOTSUPP;
+			goto out1;
+		}
+
+		attr = ni_find_attr(ni, attr, &le, ATTR_DATA, WOF_NAME,
+				    ARRAY_SIZE(WOF_NAME), NULL, NULL);
+		if (!attr) {
+			ntfs_inode_err(
+				&ni->vfs_inode,
+				"external compressed file should contains data attribute \"WofCompressedData\"");
+			err = -EINVAL;
+			goto out1;
+		}
+
+		if (!attr->non_res) {
+			run = NULL;
+		} else {
+			run = run_alloc();
+			if (!run) {
+				err = -ENOMEM;
+				goto out1;
+			}
+		}
+
+		frames = (ni->vfs_inode.i_size - 1) >> frame_bits;
+
+		err = attr_wof_frame_info(ni, attr, run, frame64, frames,
+					  frame_bits, &ondisk_size, &vbo_data);
+		if (err)
+			goto out2;
+
+		if (frame64 == frames) {
+			unc_size = 1 + ((ni->vfs_inode.i_size - 1) &
+					(frame_size - 1));
+			ondisk_size = attr_size(attr) - vbo_data;
+		} else {
+			unc_size = frame_size;
+		}
+
+		if (ondisk_size > frame_size) {
+			err = -EINVAL;
+			goto out2;
+		}
+
+		if (!attr->non_res) {
+			if (vbo_data + ondisk_size >
+			    le32_to_cpu(attr->res.data_size)) {
+				err = -EINVAL;
+				goto out1;
+			}
+
+			err = decompress_lzx_xpress(
+				sbi, Add2Ptr(resident_data(attr), vbo_data),
+				ondisk_size, frame_mem, unc_size, frame_size);
+			goto out1;
+		}
+		vbo_disk = vbo_data;
+		/* Load all runs to read [vbo_disk-vbo_to). */
+		err = attr_load_runs_range(ni, ATTR_DATA, WOF_NAME,
+					   ARRAY_SIZE(WOF_NAME), run, vbo_disk,
+					   vbo_data + ondisk_size);
+		if (err)
+			goto out2;
+		npages_disk = (ondisk_size + (vbo_disk & (PAGE_SIZE - 1)) +
+			       PAGE_SIZE - 1) >>
+			      PAGE_SHIFT;
+#endif
+	} else if (is_attr_compressed(attr)) {
+		/* LZNT compression. */
+		if (sbi->cluster_size > NTFS_LZNT_MAX_CLUSTER) {
+			err = -EOPNOTSUPP;
+			goto out1;
+		}
+
+		if (attr->nres.c_unit != NTFS_LZNT_CUNIT) {
+			err = -EOPNOTSUPP;
+			goto out1;
+		}
+
+		down_write(&ni->file.run_lock);
+		run_truncate_around(run, le64_to_cpu(attr->nres.svcn));
+		frame = frame_vbo >> (cluster_bits + NTFS_LZNT_CUNIT);
+		err = attr_is_frame_compressed(ni, attr, frame, &clst_data);
+		up_write(&ni->file.run_lock);
+		if (err)
+			goto out1;
+
+		if (!clst_data) {
+			memset(frame_mem, 0, frame_size);
+			goto out1;
+		}
+
+		frame_size = sbi->cluster_size << NTFS_LZNT_CUNIT;
+		ondisk_size = clst_data << cluster_bits;
+
+		if (clst_data >= NTFS_LZNT_CLUSTERS) {
+			/* Frame is not compressed. */
+			down_read(&ni->file.run_lock);
+			err = ntfs_bio_pages(sbi, run, pages, pages_per_frame,
+					     frame_vbo, ondisk_size,
+					     REQ_OP_READ);
+			up_read(&ni->file.run_lock);
+			goto out1;
+		}
+		vbo_disk = frame_vbo;
+		npages_disk = (ondisk_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
+	} else {
+		__builtin_unreachable();
+		err = -EINVAL;
+		goto out1;
+	}
+
+	pages_disk = kzalloc(npages_disk * sizeof(struct page *), GFP_NOFS);
+	if (!pages_disk) {
+		err = -ENOMEM;
+		goto out2;
+	}
+
+	for (i = 0; i < npages_disk; i++) {
+		pg = alloc_page(GFP_KERNEL);
+		if (!pg) {
+			err = -ENOMEM;
+			goto out3;
+		}
+		pages_disk[i] = pg;
+		lock_page(pg);
+		kmap(pg);
+	}
+
+	/* Read 'ondisk_size' bytes from disk. */
+	down_read(&ni->file.run_lock);
+	err = ntfs_bio_pages(sbi, run, pages_disk, npages_disk, vbo_disk,
+			     ondisk_size, REQ_OP_READ);
+	up_read(&ni->file.run_lock);
+	if (err)
+		goto out3;
+
+	/*
+	 * To simplify decompress algorithm do vmap for source and target pages.
+	 */
+	frame_ondisk = vmap(pages_disk, npages_disk, VM_MAP, PAGE_KERNEL_RO);
+	if (!frame_ondisk) {
+		err = -ENOMEM;
+		goto out3;
+	}
+
+	/* Decompress: Frame_ondisk -> frame_mem. */
+#ifdef CONFIG_NTFS3_LZX_XPRESS
+	if (run != &ni->file.run) {
+		/* LZX or XPRESS */
+		err = decompress_lzx_xpress(
+			sbi, frame_ondisk + (vbo_disk & (PAGE_SIZE - 1)),
+			ondisk_size, frame_mem, unc_size, frame_size);
+	} else
+#endif
+	{
+		/* LZNT - Native NTFS compression. */
+		unc_size = decompress_lznt(frame_ondisk, ondisk_size, frame_mem,
+					   frame_size);
+		if ((ssize_t)unc_size < 0)
+			err = unc_size;
+		else if (!unc_size || unc_size > frame_size)
+			err = -EINVAL;
+	}
+	if (!err && valid_size < frame_vbo + frame_size) {
+		size_t ok = valid_size - frame_vbo;
+
+		memset(frame_mem + ok, 0, frame_size - ok);
+	}
+
+	vunmap(frame_ondisk);
+
+out3:
+	for (i = 0; i < npages_disk; i++) {
+		pg = pages_disk[i];
+		if (pg) {
+			kunmap(pg);
+			unlock_page(pg);
+			put_page(pg);
+		}
+	}
+	kfree(pages_disk);
+
+out2:
+#ifdef CONFIG_NTFS3_LZX_XPRESS
+	if (run != &ni->file.run)
+		run_free(run);
+#endif
+out1:
+	vunmap(frame_mem);
+out:
+	for (i = 0; i < pages_per_frame; i++) {
+		pg = pages[i];
+		kunmap(pg);
+		ClearPageError(pg);
+		SetPageUptodate(pg);
+	}
+
+	return err;
+}
+
+/*
+ * ni_write_frame
+ *
+ * Pages - Array of locked pages.
+ */
+int ni_write_frame(struct ntfs_inode *ni, struct page **pages,
+		   u32 pages_per_frame)
+{
+	int err;
+	struct ntfs_sb_info *sbi = ni->mi.sbi;
+	u8 frame_bits = NTFS_LZNT_CUNIT + sbi->cluster_bits;
+	u32 frame_size = sbi->cluster_size << NTFS_LZNT_CUNIT;
+	u64 frame_vbo = (u64)pages[0]->index << PAGE_SHIFT;
+	CLST frame = frame_vbo >> frame_bits;
+	char *frame_ondisk = NULL;
+	struct page **pages_disk = NULL;
+	struct ATTR_LIST_ENTRY *le = NULL;
+	char *frame_mem;
+	struct ATTRIB *attr;
+	struct mft_inode *mi;
+	u32 i;
+	struct page *pg;
+	size_t compr_size, ondisk_size;
+	struct lznt *lznt;
+
+	attr = ni_find_attr(ni, NULL, &le, ATTR_DATA, NULL, 0, NULL, &mi);
+	if (!attr) {
+		err = -ENOENT;
+		goto out;
+	}
+
+	if (WARN_ON(!is_attr_compressed(attr))) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	if (sbi->cluster_size > NTFS_LZNT_MAX_CLUSTER) {
+		err = -EOPNOTSUPP;
+		goto out;
+	}
+
+	if (!attr->non_res) {
+		down_write(&ni->file.run_lock);
+		err = attr_make_nonresident(ni, attr, le, mi,
+					    le32_to_cpu(attr->res.data_size),
+					    &ni->file.run, &attr, pages[0]);
+		up_write(&ni->file.run_lock);
+		if (err)
+			goto out;
+	}
+
+	if (attr->nres.c_unit != NTFS_LZNT_CUNIT) {
+		err = -EOPNOTSUPP;
+		goto out;
+	}
+
+	pages_disk = kcalloc(pages_per_frame, sizeof(struct page *), GFP_NOFS);
+	if (!pages_disk) {
+		err = -ENOMEM;
+		goto out;
+	}
+
+	for (i = 0; i < pages_per_frame; i++) {
+		pg = alloc_page(GFP_KERNEL);
+		if (!pg) {
+			err = -ENOMEM;
+			goto out1;
+		}
+		pages_disk[i] = pg;
+		lock_page(pg);
+		kmap(pg);
+	}
+
+	/* To simplify compress algorithm do vmap for source and target pages. */
+	frame_ondisk = vmap(pages_disk, pages_per_frame, VM_MAP, PAGE_KERNEL);
+	if (!frame_ondisk) {
+		err = -ENOMEM;
+		goto out1;
+	}
+
+	for (i = 0; i < pages_per_frame; i++)
+		kmap(pages[i]);
+
+	/* Map in-memory frame for read-only. */
+	frame_mem = vmap(pages, pages_per_frame, VM_MAP, PAGE_KERNEL_RO);
+	if (!frame_mem) {
+		err = -ENOMEM;
+		goto out2;
+	}
+
+	mutex_lock(&sbi->compress.mtx_lznt);
+	lznt = NULL;
+	if (!sbi->compress.lznt) {
+		/*
+		 * LZNT implements two levels of compression:
+		 * 0 - Standard compression
+		 * 1 - Best compression, requires a lot of cpu
+		 * use mount option?
+		 */
+		lznt = get_lznt_ctx(0);
+		if (!lznt) {
+			mutex_unlock(&sbi->compress.mtx_lznt);
+			err = -ENOMEM;
+			goto out3;
+		}
+
+		sbi->compress.lznt = lznt;
+		lznt = NULL;
+	}
+
+	/* Compress: frame_mem -> frame_ondisk */
+	compr_size = compress_lznt(frame_mem, frame_size, frame_ondisk,
+				   frame_size, sbi->compress.lznt);
+	mutex_unlock(&sbi->compress.mtx_lznt);
+	kfree(lznt);
+
+	if (compr_size + sbi->cluster_size > frame_size) {
+		/* Frame is not compressed. */
+		compr_size = frame_size;
+		ondisk_size = frame_size;
+	} else if (compr_size) {
+		/* Frame is compressed. */
+		ondisk_size = ntfs_up_cluster(sbi, compr_size);
+		memset(frame_ondisk + compr_size, 0, ondisk_size - compr_size);
+	} else {
+		/* Frame is sparsed. */
+		ondisk_size = 0;
+	}
+
+	down_write(&ni->file.run_lock);
+	run_truncate_around(&ni->file.run, le64_to_cpu(attr->nres.svcn));
+	err = attr_allocate_frame(ni, frame, compr_size, ni->i_valid);
+	up_write(&ni->file.run_lock);
+	if (err)
+		goto out2;
+
+	if (!ondisk_size)
+		goto out2;
+
+	down_read(&ni->file.run_lock);
+	err = ntfs_bio_pages(sbi, &ni->file.run,
+			     ondisk_size < frame_size ? pages_disk : pages,
+			     pages_per_frame, frame_vbo, ondisk_size,
+			     REQ_OP_WRITE);
+	up_read(&ni->file.run_lock);
+
+out3:
+	vunmap(frame_mem);
+
+out2:
+	for (i = 0; i < pages_per_frame; i++)
+		kunmap(pages[i]);
+
+	vunmap(frame_ondisk);
+out1:
+	for (i = 0; i < pages_per_frame; i++) {
+		pg = pages_disk[i];
+		if (pg) {
+			kunmap(pg);
+			unlock_page(pg);
+			put_page(pg);
+		}
+	}
+	kfree(pages_disk);
+out:
+	return err;
+}
+
+/*
+ * ni_remove_name - Removes name 'de' from MFT and from directory.
+ * 'de2' and 'undo_step' are used to restore MFT/dir, if error occurs.
+ */
+int ni_remove_name(struct ntfs_inode *dir_ni, struct ntfs_inode *ni,
+		   struct NTFS_DE *de, struct NTFS_DE **de2, int *undo_step)
+{
+	int err;
+	struct ntfs_sb_info *sbi = ni->mi.sbi;
+	struct ATTR_FILE_NAME *de_name = (struct ATTR_FILE_NAME *)(de + 1);
+	struct ATTR_FILE_NAME *fname;
+	struct ATTR_LIST_ENTRY *le;
+	struct mft_inode *mi;
+	u16 de_key_size = le16_to_cpu(de->key_size);
+	u8 name_type;
+
+	*undo_step = 0;
+
+	/* Find name in record. */
+	mi_get_ref(&dir_ni->mi, &de_name->home);
+
+	fname = ni_fname_name(ni, (struct cpu_str *)&de_name->name_len,
+			      &de_name->home, &mi, &le);
+	if (!fname)
+		return -ENOENT;
+
+	memcpy(&de_name->dup, &fname->dup, sizeof(struct NTFS_DUP_INFO));
+	name_type = paired_name(fname->type);
+
+	/* Mark ntfs as dirty. It will be cleared at umount. */
+	ntfs_set_state(sbi, NTFS_DIRTY_DIRTY);
+
+	/* Step 1: Remove name from directory. */
+	err = indx_delete_entry(&dir_ni->dir, dir_ni, fname, de_key_size, sbi);
+	if (err)
+		return err;
+
+	/* Step 2: Remove name from MFT. */
+	ni_remove_attr_le(ni, attr_from_name(fname), mi, le);
+
+	*undo_step = 2;
+
+	/* Get paired name. */
+	fname = ni_fname_type(ni, name_type, &mi, &le);
+	if (fname) {
+		u16 de2_key_size = fname_full_size(fname);
+
+		*de2 = Add2Ptr(de, 1024);
+		(*de2)->key_size = cpu_to_le16(de2_key_size);
+
+		memcpy(*de2 + 1, fname, de2_key_size);
+
+		/* Step 3: Remove paired name from directory. */
+		err = indx_delete_entry(&dir_ni->dir, dir_ni, fname,
+					de2_key_size, sbi);
+		if (err)
+			return err;
+
+		/* Step 4: Remove paired name from MFT. */
+		ni_remove_attr_le(ni, attr_from_name(fname), mi, le);
+
+		*undo_step = 4;
+	}
+	return 0;
+}
+
+/*
+ * ni_remove_name_undo - Paired function for ni_remove_name.
+ *
+ * Return: True if ok
+ */
+bool ni_remove_name_undo(struct ntfs_inode *dir_ni, struct ntfs_inode *ni,
+			 struct NTFS_DE *de, struct NTFS_DE *de2, int undo_step)
+{
+	struct ntfs_sb_info *sbi = ni->mi.sbi;
+	struct ATTRIB *attr;
+	u16 de_key_size = de2 ? le16_to_cpu(de2->key_size) : 0;
+
+	switch (undo_step) {
+	case 4:
+		if (ni_insert_resident(ni, de_key_size, ATTR_NAME, NULL, 0,
+				       &attr, NULL, NULL)) {
+			return false;
+		}
+		memcpy(Add2Ptr(attr, SIZEOF_RESIDENT), de2 + 1, de_key_size);
+
+		mi_get_ref(&ni->mi, &de2->ref);
+		de2->size = cpu_to_le16(ALIGN(de_key_size, 8) +
+					sizeof(struct NTFS_DE));
+		de2->flags = 0;
+		de2->res = 0;
+
+		if (indx_insert_entry(&dir_ni->dir, dir_ni, de2, sbi, NULL,
+				      1)) {
+			return false;
+		}
+		fallthrough;
+
+	case 2:
+		de_key_size = le16_to_cpu(de->key_size);
+
+		if (ni_insert_resident(ni, de_key_size, ATTR_NAME, NULL, 0,
+				       &attr, NULL, NULL)) {
+			return false;
+		}
+
+		memcpy(Add2Ptr(attr, SIZEOF_RESIDENT), de + 1, de_key_size);
+		mi_get_ref(&ni->mi, &de->ref);
+
+		if (indx_insert_entry(&dir_ni->dir, dir_ni, de, sbi, NULL, 1)) {
+			return false;
+		}
+	}
+
+	return true;
+}
+
+/*
+ * ni_add_name - Add new name in MFT and in directory.
+ */
+int ni_add_name(struct ntfs_inode *dir_ni, struct ntfs_inode *ni,
+		struct NTFS_DE *de)
+{
+	int err;
+	struct ATTRIB *attr;
+	struct ATTR_LIST_ENTRY *le;
+	struct mft_inode *mi;
+	struct ATTR_FILE_NAME *de_name = (struct ATTR_FILE_NAME *)(de + 1);
+	u16 de_key_size = le16_to_cpu(de->key_size);
+
+	mi_get_ref(&ni->mi, &de->ref);
+	mi_get_ref(&dir_ni->mi, &de_name->home);
+
+	/* Insert new name in MFT. */
+	err = ni_insert_resident(ni, de_key_size, ATTR_NAME, NULL, 0, &attr,
+				 &mi, &le);
+	if (err)
+		return err;
+
+	memcpy(Add2Ptr(attr, SIZEOF_RESIDENT), de_name, de_key_size);
+
+	/* Insert new name in directory. */
+	err = indx_insert_entry(&dir_ni->dir, dir_ni, de, ni->mi.sbi, NULL, 0);
+	if (err)
+		ni_remove_attr_le(ni, attr, mi, le);
+
+	return err;
+}
+
+/*
+ * ni_rename - Remove one name and insert new name.
+ */
+int ni_rename(struct ntfs_inode *dir_ni, struct ntfs_inode *new_dir_ni,
+	      struct ntfs_inode *ni, struct NTFS_DE *de, struct NTFS_DE *new_de,
+	      bool *is_bad)
+{
+	int err;
+	struct NTFS_DE *de2 = NULL;
+	int undo = 0;
+
+	/*
+	 * There are two possible ways to rename:
+	 * 1) Add new name and remove old name.
+	 * 2) Remove old name and add new name.
+	 *
+	 * In most cases (not all!) adding new name in MFT and in directory can
+	 * allocate additional cluster(s).
+	 * Second way may result to bad inode if we can't add new name
+	 * and then can't restore (add) old name.
+	 */
+
+	/*
+	 * Way 1 - Add new + remove old.
+	 */
+	err = ni_add_name(new_dir_ni, ni, new_de);
+	if (!err) {
+		err = ni_remove_name(dir_ni, ni, de, &de2, &undo);
+		if (err && ni_remove_name(new_dir_ni, ni, new_de, &de2, &undo))
+			*is_bad = true;
+	}
+
+	/*
+	 * Way 2 - Remove old + add new.
+	 */
+	/*
+	 *	err = ni_remove_name(dir_ni, ni, de, &de2, &undo);
+	 *	if (!err) {
+	 *		err = ni_add_name(new_dir_ni, ni, new_de);
+	 *		if (err && !ni_remove_name_undo(dir_ni, ni, de, de2, undo))
+	 *			*is_bad = true;
+	 *	}
+	 */
+
+	return err;
+}
+
+/*
+ * ni_is_dirty - Return: True if 'ni' requires ni_write_inode.
+ */
+bool ni_is_dirty(struct inode *inode)
+{
+	struct ntfs_inode *ni = ntfs_i(inode);
+	struct rb_node *node;
+
+	if (ni->mi.dirty || ni->attr_list.dirty ||
+	    (ni->ni_flags & NI_FLAG_UPDATE_PARENT))
+		return true;
+
+	for (node = rb_first(&ni->mi_tree); node; node = rb_next(node)) {
+		if (rb_entry(node, struct mft_inode, node)->dirty)
+			return true;
+	}
+
+	return false;
+}
+
+/*
+ * ni_update_parent
+ *
+ * Update duplicate info of ATTR_FILE_NAME in MFT and in parent directories.
+ */
+static bool ni_update_parent(struct ntfs_inode *ni, struct NTFS_DUP_INFO *dup,
+			     int sync)
+{
+	struct ATTRIB *attr;
+	struct mft_inode *mi;
+	struct ATTR_LIST_ENTRY *le = NULL;
+	struct ntfs_sb_info *sbi = ni->mi.sbi;
+	struct super_block *sb = sbi->sb;
+	bool re_dirty = false;
+
+	if (ni->mi.mrec->flags & RECORD_FLAG_DIR) {
+		dup->fa |= FILE_ATTRIBUTE_DIRECTORY;
+		attr = NULL;
+		dup->alloc_size = 0;
+		dup->data_size = 0;
+	} else {
+		dup->fa &= ~FILE_ATTRIBUTE_DIRECTORY;
+
+		attr = ni_find_attr(ni, NULL, &le, ATTR_DATA, NULL, 0, NULL,
+				    &mi);
+		if (!attr) {
+			dup->alloc_size = dup->data_size = 0;
+		} else if (!attr->non_res) {
+			u32 data_size = le32_to_cpu(attr->res.data_size);
+
+			dup->alloc_size = cpu_to_le64(ALIGN(data_size, 8));
+			dup->data_size = cpu_to_le64(data_size);
+		} else {
+			u64 new_valid = ni->i_valid;
+			u64 data_size = le64_to_cpu(attr->nres.data_size);
+			__le64 valid_le;
+
+			dup->alloc_size = is_attr_ext(attr)
+						  ? attr->nres.total_size
+						  : attr->nres.alloc_size;
+			dup->data_size = attr->nres.data_size;
+
+			if (new_valid > data_size)
+				new_valid = data_size;
+
+			valid_le = cpu_to_le64(new_valid);
+			if (valid_le != attr->nres.valid_size) {
+				attr->nres.valid_size = valid_le;
+				mi->dirty = true;
+			}
+		}
+	}
+
+	/* TODO: Fill reparse info. */
+	dup->reparse = 0;
+	dup->ea_size = 0;
+
+	if (ni->ni_flags & NI_FLAG_EA) {
+		attr = ni_find_attr(ni, attr, &le, ATTR_EA_INFO, NULL, 0, NULL,
+				    NULL);
+		if (attr) {
+			const struct EA_INFO *info;
+
+			info = resident_data_ex(attr, sizeof(struct EA_INFO));
+			dup->ea_size = info->size_pack;
+		}
+	}
+
+	attr = NULL;
+	le = NULL;
+
+	while ((attr = ni_find_attr(ni, attr, &le, ATTR_NAME, NULL, 0, NULL,
+				    &mi))) {
+		struct inode *dir;
+		struct ATTR_FILE_NAME *fname;
+
+		fname = resident_data_ex(attr, SIZEOF_ATTRIBUTE_FILENAME);
+		if (!fname || !memcmp(&fname->dup, dup, sizeof(fname->dup)))
+			continue;
+
+		/* ntfs_iget5 may sleep. */
+		dir = ntfs_iget5(sb, &fname->home, NULL);
+		if (IS_ERR(dir)) {
+			ntfs_inode_warn(
+				&ni->vfs_inode,
+				"failed to open parent directory r=%lx to update",
+				(long)ino_get(&fname->home));
+			continue;
+		}
+
+		if (!is_bad_inode(dir)) {
+			struct ntfs_inode *dir_ni = ntfs_i(dir);
+
+			if (!ni_trylock(dir_ni)) {
+				re_dirty = true;
+			} else {
+				indx_update_dup(dir_ni, sbi, fname, dup, sync);
+				ni_unlock(dir_ni);
+				memcpy(&fname->dup, dup, sizeof(fname->dup));
+				mi->dirty = true;
+			}
+		}
+		iput(dir);
+	}
+
+	return re_dirty;
+}
+
+/*
+ * ni_write_inode - Write MFT base record and all subrecords to disk.
+ */
+int ni_write_inode(struct inode *inode, int sync, const char *hint)
+{
+	int err = 0, err2;
+	struct ntfs_inode *ni = ntfs_i(inode);
+	struct super_block *sb = inode->i_sb;
+	struct ntfs_sb_info *sbi = sb->s_fs_info;
+	bool re_dirty = false;
+	struct ATTR_STD_INFO *std;
+	struct rb_node *node, *next;
+	struct NTFS_DUP_INFO dup;
+
+	if (is_bad_inode(inode) || sb_rdonly(sb))
+		return 0;
+
+	if (!ni_trylock(ni)) {
+		/* 'ni' is under modification, skip for now. */
+		mark_inode_dirty_sync(inode);
+		return 0;
+	}
+
+	if (is_rec_inuse(ni->mi.mrec) &&
+	    !(sbi->flags & NTFS_FLAGS_LOG_REPLAYING) && inode->i_nlink) {
+		bool modified = false;
+
+		/* Update times in standard attribute. */
+		std = ni_std(ni);
+		if (!std) {
+			err = -EINVAL;
+			goto out;
+		}
+
+		/* Update the access times if they have changed. */
+		dup.m_time = kernel2nt(&inode->i_mtime);
+		if (std->m_time != dup.m_time) {
+			std->m_time = dup.m_time;
+			modified = true;
+		}
+
+		dup.c_time = kernel2nt(&inode->i_ctime);
+		if (std->c_time != dup.c_time) {
+			std->c_time = dup.c_time;
+			modified = true;
+		}
+
+		dup.a_time = kernel2nt(&inode->i_atime);
+		if (std->a_time != dup.a_time) {
+			std->a_time = dup.a_time;
+			modified = true;
+		}
+
+		dup.fa = ni->std_fa;
+		if (std->fa != dup.fa) {
+			std->fa = dup.fa;
+			modified = true;
+		}
+
+		if (modified)
+			ni->mi.dirty = true;
+
+		if (!ntfs_is_meta_file(sbi, inode->i_ino) &&
+		    (modified || (ni->ni_flags & NI_FLAG_UPDATE_PARENT))
+		    /* Avoid __wait_on_freeing_inode(inode). */
+		    && (sb->s_flags & SB_ACTIVE)) {
+			dup.cr_time = std->cr_time;
+			/* Not critical if this function fail. */
+			re_dirty = ni_update_parent(ni, &dup, sync);
+
+			if (re_dirty)
+				ni->ni_flags |= NI_FLAG_UPDATE_PARENT;
+			else
+				ni->ni_flags &= ~NI_FLAG_UPDATE_PARENT;
+		}
+
+		/* Update attribute list. */
+		if (ni->attr_list.size && ni->attr_list.dirty) {
+			if (inode->i_ino != MFT_REC_MFT || sync) {
+				err = ni_try_remove_attr_list(ni);
+				if (err)
+					goto out;
+			}
+
+			err = al_update(ni);
+			if (err)
+				goto out;
+		}
+	}
+
+	for (node = rb_first(&ni->mi_tree); node; node = next) {
+		struct mft_inode *mi = rb_entry(node, struct mft_inode, node);
+		bool is_empty;
+
+		next = rb_next(node);
+
+		if (!mi->dirty)
+			continue;
+
+		is_empty = !mi_enum_attr(mi, NULL);
+
+		if (is_empty)
+			clear_rec_inuse(mi->mrec);
+
+		err2 = mi_write(mi, sync);
+		if (!err && err2)
+			err = err2;
+
+		if (is_empty) {
+			ntfs_mark_rec_free(sbi, mi->rno);
+			rb_erase(node, &ni->mi_tree);
+			mi_put(mi);
+		}
+	}
+
+	if (ni->mi.dirty) {
+		err2 = mi_write(&ni->mi, sync);
+		if (!err && err2)
+			err = err2;
+	}
+out:
+	ni_unlock(ni);
+
+	if (err) {
+		ntfs_err(sb, "%s r=%lx failed, %d.", hint, inode->i_ino, err);
+		ntfs_set_state(sbi, NTFS_DIRTY_ERROR);
+		return err;
+	}
+
+	if (re_dirty)
+		mark_inode_dirty_sync(inode);
+
+	return 0;
+}
diff --git a/fs/ntfs3/fslog.c b/fs/ntfs3/fslog.c
new file mode 100644
index 000000000000..b5853aed0e25
--- /dev/null
+++ b/fs/ntfs3/fslog.c
@@ -0,0 +1,5217 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ *
+ * Copyright (C) 2019-2021 Paragon Software GmbH, All rights reserved.
+ *
+ */
+
+#include <linux/blkdev.h>
+#include <linux/buffer_head.h>
+#include <linux/fs.h>
+#include <linux/hash.h>
+#include <linux/nls.h>
+#include <linux/random.h>
+#include <linux/ratelimit.h>
+#include <linux/slab.h>
+
+#include "debug.h"
+#include "ntfs.h"
+#include "ntfs_fs.h"
+
+/*
+ * LOG FILE structs
+ */
+
+// clang-format off
+
+#define MaxLogFileSize     0x100000000ull
+#define DefaultLogPageSize 4096
+#define MinLogRecordPages  0x30
+
+struct RESTART_HDR {
+	struct NTFS_RECORD_HEADER rhdr; // 'RSTR'
+	__le32 sys_page_size; // 0x10: Page size of the system which initialized the log.
+	__le32 page_size;     // 0x14: Log page size used for this log file.
+	__le16 ra_off;        // 0x18:
+	__le16 minor_ver;     // 0x1A:
+	__le16 major_ver;     // 0x1C:
+	__le16 fixups[];
+};
+
+#define LFS_NO_CLIENT 0xffff
+#define LFS_NO_CLIENT_LE cpu_to_le16(0xffff)
+
+struct CLIENT_REC {
+	__le64 oldest_lsn;
+	__le64 restart_lsn; // 0x08:
+	__le16 prev_client; // 0x10:
+	__le16 next_client; // 0x12:
+	__le16 seq_num;     // 0x14:
+	u8 align[6];        // 0x16:
+	__le32 name_bytes;  // 0x1C: In bytes.
+	__le16 name[32];    // 0x20: Name of client.
+};
+
+static_assert(sizeof(struct CLIENT_REC) == 0x60);
+
+/* Two copies of these will exist at the beginning of the log file */
+struct RESTART_AREA {
+	__le64 current_lsn;    // 0x00: Current logical end of log file.
+	__le16 log_clients;    // 0x08: Maximum number of clients.
+	__le16 client_idx[2];  // 0x0A: Free/use index into the client record arrays.
+	__le16 flags;          // 0x0E: See RESTART_SINGLE_PAGE_IO.
+	__le32 seq_num_bits;   // 0x10: The number of bits in sequence number.
+	__le16 ra_len;         // 0x14:
+	__le16 client_off;     // 0x16:
+	__le64 l_size;         // 0x18: Usable log file size.
+	__le32 last_lsn_data_len; // 0x20:
+	__le16 rec_hdr_len;    // 0x24: Log page data offset.
+	__le16 data_off;       // 0x26: Log page data length.
+	__le32 open_log_count; // 0x28:
+	__le32 align[5];       // 0x2C:
+	struct CLIENT_REC clients[]; // 0x40:
+};
+
+struct LOG_REC_HDR {
+	__le16 redo_op;      // 0x00:  NTFS_LOG_OPERATION
+	__le16 undo_op;      // 0x02:  NTFS_LOG_OPERATION
+	__le16 redo_off;     // 0x04:  Offset to Redo record.
+	__le16 redo_len;     // 0x06:  Redo length.
+	__le16 undo_off;     // 0x08:  Offset to Undo record.
+	__le16 undo_len;     // 0x0A:  Undo length.
+	__le16 target_attr;  // 0x0C:
+	__le16 lcns_follow;  // 0x0E:
+	__le16 record_off;   // 0x10:
+	__le16 attr_off;     // 0x12:
+	__le16 cluster_off;  // 0x14:
+	__le16 reserved;     // 0x16:
+	__le64 target_vcn;   // 0x18:
+	__le64 page_lcns[];  // 0x20:
+};
+
+static_assert(sizeof(struct LOG_REC_HDR) == 0x20);
+
+#define RESTART_ENTRY_ALLOCATED    0xFFFFFFFF
+#define RESTART_ENTRY_ALLOCATED_LE cpu_to_le32(0xFFFFFFFF)
+
+struct RESTART_TABLE {
+	__le16 size;       // 0x00: In bytes
+	__le16 used;       // 0x02: Entries
+	__le16 total;      // 0x04: Entries
+	__le16 res[3];     // 0x06:
+	__le32 free_goal;  // 0x0C:
+	__le32 first_free; // 0x10:
+	__le32 last_free;  // 0x14:
+
+};
+
+static_assert(sizeof(struct RESTART_TABLE) == 0x18);
+
+struct ATTR_NAME_ENTRY {
+	__le16 off; // Offset in the Open attribute Table.
+	__le16 name_bytes;
+	__le16 name[];
+};
+
+struct OPEN_ATTR_ENRTY {
+	__le32 next;            // 0x00: RESTART_ENTRY_ALLOCATED if allocated
+	__le32 bytes_per_index; // 0x04:
+	enum ATTR_TYPE type;    // 0x08:
+	u8 is_dirty_pages;      // 0x0C:
+	u8 is_attr_name;        // 0x0B: Faked field to manage 'ptr'
+	u8 name_len;            // 0x0C: Faked field to manage 'ptr'
+	u8 res;
+	struct MFT_REF ref;     // 0x10: File Reference of file containing attribute
+	__le64 open_record_lsn; // 0x18:
+	void *ptr;              // 0x20:
+};
+
+/* 32 bit version of 'struct OPEN_ATTR_ENRTY' */
+struct OPEN_ATTR_ENRTY_32 {
+	__le32 next;            // 0x00: RESTART_ENTRY_ALLOCATED if allocated
+	__le32 ptr;             // 0x04:
+	struct MFT_REF ref;     // 0x08:
+	__le64 open_record_lsn; // 0x10:
+	u8 is_dirty_pages;      // 0x18:
+	u8 is_attr_name;        // 0x19:
+	u8 res1[2];
+	enum ATTR_TYPE type;    // 0x1C:
+	u8 name_len;            // 0x20: In wchar
+	u8 res2[3];
+	__le32 AttributeName;   // 0x24:
+	__le32 bytes_per_index; // 0x28:
+};
+
+#define SIZEOF_OPENATTRIBUTEENTRY0 0x2c
+// static_assert( 0x2C == sizeof(struct OPEN_ATTR_ENRTY_32) );
+static_assert(sizeof(struct OPEN_ATTR_ENRTY) < SIZEOF_OPENATTRIBUTEENTRY0);
+
+/*
+ * One entry exists in the Dirty Pages Table for each page which is dirty at
+ * the time the Restart Area is written.
+ */
+struct DIR_PAGE_ENTRY {
+	__le32 next;         // 0x00: RESTART_ENTRY_ALLOCATED if allocated
+	__le32 target_attr;  // 0x04: Index into the Open attribute Table
+	__le32 transfer_len; // 0x08:
+	__le32 lcns_follow;  // 0x0C:
+	__le64 vcn;          // 0x10: Vcn of dirty page
+	__le64 oldest_lsn;   // 0x18:
+	__le64 page_lcns[];  // 0x20:
+};
+
+static_assert(sizeof(struct DIR_PAGE_ENTRY) == 0x20);
+
+/* 32 bit version of 'struct DIR_PAGE_ENTRY' */
+struct DIR_PAGE_ENTRY_32 {
+	__le32 next;		// 0x00: RESTART_ENTRY_ALLOCATED if allocated
+	__le32 target_attr;	// 0x04: Index into the Open attribute Table
+	__le32 transfer_len;	// 0x08:
+	__le32 lcns_follow;	// 0x0C:
+	__le32 reserved;	// 0x10:
+	__le32 vcn_low;		// 0x14: Vcn of dirty page
+	__le32 vcn_hi;		// 0x18: Vcn of dirty page
+	__le32 oldest_lsn_low;	// 0x1C:
+	__le32 oldest_lsn_hi;	// 0x1C:
+	__le32 page_lcns_low;	// 0x24:
+	__le32 page_lcns_hi;	// 0x24:
+};
+
+static_assert(offsetof(struct DIR_PAGE_ENTRY_32, vcn_low) == 0x14);
+static_assert(sizeof(struct DIR_PAGE_ENTRY_32) == 0x2c);
+
+enum transact_state {
+	TransactionUninitialized = 0,
+	TransactionActive,
+	TransactionPrepared,
+	TransactionCommitted
+};
+
+struct TRANSACTION_ENTRY {
+	__le32 next;          // 0x00: RESTART_ENTRY_ALLOCATED if allocated
+	u8 transact_state;    // 0x04:
+	u8 reserved[3];       // 0x05:
+	__le64 first_lsn;     // 0x08:
+	__le64 prev_lsn;      // 0x10:
+	__le64 undo_next_lsn; // 0x18:
+	__le32 undo_records;  // 0x20: Number of undo log records pending abort
+	__le32 undo_len;      // 0x24: Total undo size
+};
+
+static_assert(sizeof(struct TRANSACTION_ENTRY) == 0x28);
+
+struct NTFS_RESTART {
+	__le32 major_ver;             // 0x00:
+	__le32 minor_ver;             // 0x04:
+	__le64 check_point_start;     // 0x08:
+	__le64 open_attr_table_lsn;   // 0x10:
+	__le64 attr_names_lsn;        // 0x18:
+	__le64 dirty_pages_table_lsn; // 0x20:
+	__le64 transact_table_lsn;    // 0x28:
+	__le32 open_attr_len;         // 0x30: In bytes
+	__le32 attr_names_len;        // 0x34: In bytes
+	__le32 dirty_pages_len;       // 0x38: In bytes
+	__le32 transact_table_len;    // 0x3C: In bytes
+};
+
+static_assert(sizeof(struct NTFS_RESTART) == 0x40);
+
+struct NEW_ATTRIBUTE_SIZES {
+	__le64 alloc_size;
+	__le64 valid_size;
+	__le64 data_size;
+	__le64 total_size;
+};
+
+struct BITMAP_RANGE {
+	__le32 bitmap_off;
+	__le32 bits;
+};
+
+struct LCN_RANGE {
+	__le64 lcn;
+	__le64 len;
+};
+
+/* The following type defines the different log record types. */
+#define LfsClientRecord  cpu_to_le32(1)
+#define LfsClientRestart cpu_to_le32(2)
+
+/* This is used to uniquely identify a client for a particular log file. */
+struct CLIENT_ID {
+	__le16 seq_num;
+	__le16 client_idx;
+};
+
+/* This is the header that begins every Log Record in the log file. */
+struct LFS_RECORD_HDR {
+	__le64 this_lsn;		// 0x00:
+	__le64 client_prev_lsn;		// 0x08:
+	__le64 client_undo_next_lsn;	// 0x10:
+	__le32 client_data_len;		// 0x18:
+	struct CLIENT_ID client;	// 0x1C: Owner of this log record.
+	__le32 record_type;		// 0x20: LfsClientRecord or LfsClientRestart.
+	__le32 transact_id;		// 0x24:
+	__le16 flags;			// 0x28: LOG_RECORD_MULTI_PAGE
+	u8 align[6];			// 0x2A:
+};
+
+#define LOG_RECORD_MULTI_PAGE cpu_to_le16(1)
+
+static_assert(sizeof(struct LFS_RECORD_HDR) == 0x30);
+
+struct LFS_RECORD {
+	__le16 next_record_off;	// 0x00: Offset of the free space in the page,
+	u8 align[6];		// 0x02:
+	__le64 last_end_lsn;	// 0x08: lsn for the last log record which ends on the page,
+};
+
+static_assert(sizeof(struct LFS_RECORD) == 0x10);
+
+struct RECORD_PAGE_HDR {
+	struct NTFS_RECORD_HEADER rhdr;	// 'RCRD'
+	__le32 rflags;			// 0x10: See LOG_PAGE_LOG_RECORD_END
+	__le16 page_count;		// 0x14:
+	__le16 page_pos;		// 0x16:
+	struct LFS_RECORD record_hdr;	// 0x18:
+	__le16 fixups[10];		// 0x28:
+	__le32 file_off;		// 0x3c: Used when major version >= 2
+};
+
+// clang-format on
+
+// Page contains the end of a log record.
+#define LOG_PAGE_LOG_RECORD_END cpu_to_le32(0x00000001)
+
+static inline bool is_log_record_end(const struct RECORD_PAGE_HDR *hdr)
+{
+	return hdr->rflags & LOG_PAGE_LOG_RECORD_END;
+}
+
+static_assert(offsetof(struct RECORD_PAGE_HDR, file_off) == 0x3c);
+
+/*
+ * END of NTFS LOG structures
+ */
+
+/* Define some tuning parameters to keep the restart tables a reasonable size. */
+#define INITIAL_NUMBER_TRANSACTIONS 5
+
+enum NTFS_LOG_OPERATION {
+
+	Noop = 0x00,
+	CompensationLogRecord = 0x01,
+	InitializeFileRecordSegment = 0x02,
+	DeallocateFileRecordSegment = 0x03,
+	WriteEndOfFileRecordSegment = 0x04,
+	CreateAttribute = 0x05,
+	DeleteAttribute = 0x06,
+	UpdateResidentValue = 0x07,
+	UpdateNonresidentValue = 0x08,
+	UpdateMappingPairs = 0x09,
+	DeleteDirtyClusters = 0x0A,
+	SetNewAttributeSizes = 0x0B,
+	AddIndexEntryRoot = 0x0C,
+	DeleteIndexEntryRoot = 0x0D,
+	AddIndexEntryAllocation = 0x0E,
+	DeleteIndexEntryAllocation = 0x0F,
+	WriteEndOfIndexBuffer = 0x10,
+	SetIndexEntryVcnRoot = 0x11,
+	SetIndexEntryVcnAllocation = 0x12,
+	UpdateFileNameRoot = 0x13,
+	UpdateFileNameAllocation = 0x14,
+	SetBitsInNonresidentBitMap = 0x15,
+	ClearBitsInNonresidentBitMap = 0x16,
+	HotFix = 0x17,
+	EndTopLevelAction = 0x18,
+	PrepareTransaction = 0x19,
+	CommitTransaction = 0x1A,
+	ForgetTransaction = 0x1B,
+	OpenNonresidentAttribute = 0x1C,
+	OpenAttributeTableDump = 0x1D,
+	AttributeNamesDump = 0x1E,
+	DirtyPageTableDump = 0x1F,
+	TransactionTableDump = 0x20,
+	UpdateRecordDataRoot = 0x21,
+	UpdateRecordDataAllocation = 0x22,
+
+	UpdateRelativeDataInIndex =
+		0x23, // NtOfsRestartUpdateRelativeDataInIndex
+	UpdateRelativeDataInIndex2 = 0x24,
+	ZeroEndOfFileRecord = 0x25,
+};
+
+/*
+ * Array for log records which require a target attribute.
+ * A true indicates that the corresponding restart operation
+ * requires a target attribute.
+ */
+static const u8 AttributeRequired[] = {
+	0xFC, 0xFB, 0xFF, 0x10, 0x06,
+};
+
+static inline bool is_target_required(u16 op)
+{
+	bool ret = op <= UpdateRecordDataAllocation &&
+		   (AttributeRequired[op >> 3] >> (op & 7) & 1);
+	return ret;
+}
+
+static inline bool can_skip_action(enum NTFS_LOG_OPERATION op)
+{
+	switch (op) {
+	case Noop:
+	case DeleteDirtyClusters:
+	case HotFix:
+	case EndTopLevelAction:
+	case PrepareTransaction:
+	case CommitTransaction:
+	case ForgetTransaction:
+	case CompensationLogRecord:
+	case OpenNonresidentAttribute:
+	case OpenAttributeTableDump:
+	case AttributeNamesDump:
+	case DirtyPageTableDump:
+	case TransactionTableDump:
+		return true;
+	default:
+		return false;
+	}
+}
+
+enum { lcb_ctx_undo_next, lcb_ctx_prev, lcb_ctx_next };
+
+/* Bytes per restart table. */
+static inline u32 bytes_per_rt(const struct RESTART_TABLE *rt)
+{
+	return le16_to_cpu(rt->used) * le16_to_cpu(rt->size) +
+	       sizeof(struct RESTART_TABLE);
+}
+
+/* Log record length. */
+static inline u32 lrh_length(const struct LOG_REC_HDR *lr)
+{
+	u16 t16 = le16_to_cpu(lr->lcns_follow);
+
+	return struct_size(lr, page_lcns, max_t(u16, 1, t16));
+}
+
+struct lcb {
+	struct LFS_RECORD_HDR *lrh; // Log record header of the current lsn.
+	struct LOG_REC_HDR *log_rec;
+	u32 ctx_mode; // lcb_ctx_undo_next/lcb_ctx_prev/lcb_ctx_next
+	struct CLIENT_ID client;
+	bool alloc; // If true the we should deallocate 'log_rec'.
+};
+
+static void lcb_put(struct lcb *lcb)
+{
+	if (lcb->alloc)
+		kfree(lcb->log_rec);
+	kfree(lcb->lrh);
+	kfree(lcb);
+}
+
+/* Find the oldest lsn from active clients. */
+static inline void oldest_client_lsn(const struct CLIENT_REC *ca,
+				     __le16 next_client, u64 *oldest_lsn)
+{
+	while (next_client != LFS_NO_CLIENT_LE) {
+		const struct CLIENT_REC *cr = ca + le16_to_cpu(next_client);
+		u64 lsn = le64_to_cpu(cr->oldest_lsn);
+
+		/* Ignore this block if it's oldest lsn is 0. */
+		if (lsn && lsn < *oldest_lsn)
+			*oldest_lsn = lsn;
+
+		next_client = cr->next_client;
+	}
+}
+
+static inline bool is_rst_page_hdr_valid(u32 file_off,
+					 const struct RESTART_HDR *rhdr)
+{
+	u32 sys_page = le32_to_cpu(rhdr->sys_page_size);
+	u32 page_size = le32_to_cpu(rhdr->page_size);
+	u32 end_usa;
+	u16 ro;
+
+	if (sys_page < SECTOR_SIZE || page_size < SECTOR_SIZE ||
+	    sys_page & (sys_page - 1) || page_size & (page_size - 1)) {
+		return false;
+	}
+
+	/* Check that if the file offset isn't 0, it is the system page size. */
+	if (file_off && file_off != sys_page)
+		return false;
+
+	/* Check support version 1.1+. */
+	if (le16_to_cpu(rhdr->major_ver) <= 1 && !rhdr->minor_ver)
+		return false;
+
+	if (le16_to_cpu(rhdr->major_ver) > 2)
+		return false;
+
+	ro = le16_to_cpu(rhdr->ra_off);
+	if (!IS_ALIGNED(ro, 8) || ro > sys_page)
+		return false;
+
+	end_usa = ((sys_page >> SECTOR_SHIFT) + 1) * sizeof(short);
+	end_usa += le16_to_cpu(rhdr->rhdr.fix_off);
+
+	if (ro < end_usa)
+		return false;
+
+	return true;
+}
+
+static inline bool is_rst_area_valid(const struct RESTART_HDR *rhdr)
+{
+	const struct RESTART_AREA *ra;
+	u16 cl, fl, ul;
+	u32 off, l_size, file_dat_bits, file_size_round;
+	u16 ro = le16_to_cpu(rhdr->ra_off);
+	u32 sys_page = le32_to_cpu(rhdr->sys_page_size);
+
+	if (ro + offsetof(struct RESTART_AREA, l_size) >
+	    SECTOR_SIZE - sizeof(short))
+		return false;
+
+	ra = Add2Ptr(rhdr, ro);
+	cl = le16_to_cpu(ra->log_clients);
+
+	if (cl > 1)
+		return false;
+
+	off = le16_to_cpu(ra->client_off);
+
+	if (!IS_ALIGNED(off, 8) || ro + off > SECTOR_SIZE - sizeof(short))
+		return false;
+
+	off += cl * sizeof(struct CLIENT_REC);
+
+	if (off > sys_page)
+		return false;
+
+	/*
+	 * Check the restart length field and whether the entire
+	 * restart area is contained that length.
+	 */
+	if (le16_to_cpu(rhdr->ra_off) + le16_to_cpu(ra->ra_len) > sys_page ||
+	    off > le16_to_cpu(ra->ra_len)) {
+		return false;
+	}
+
+	/*
+	 * As a final check make sure that the use list and the free list
+	 * are either empty or point to a valid client.
+	 */
+	fl = le16_to_cpu(ra->client_idx[0]);
+	ul = le16_to_cpu(ra->client_idx[1]);
+	if ((fl != LFS_NO_CLIENT && fl >= cl) ||
+	    (ul != LFS_NO_CLIENT && ul >= cl))
+		return false;
+
+	/* Make sure the sequence number bits match the log file size. */
+	l_size = le64_to_cpu(ra->l_size);
+
+	file_dat_bits = sizeof(u64) * 8 - le32_to_cpu(ra->seq_num_bits);
+	file_size_round = 1u << (file_dat_bits + 3);
+	if (file_size_round != l_size &&
+	    (file_size_round < l_size || (file_size_round / 2) > l_size)) {
+		return false;
+	}
+
+	/* The log page data offset and record header length must be quad-aligned. */
+	if (!IS_ALIGNED(le16_to_cpu(ra->data_off), 8) ||
+	    !IS_ALIGNED(le16_to_cpu(ra->rec_hdr_len), 8))
+		return false;
+
+	return true;
+}
+
+static inline bool is_client_area_valid(const struct RESTART_HDR *rhdr,
+					bool usa_error)
+{
+	u16 ro = le16_to_cpu(rhdr->ra_off);
+	const struct RESTART_AREA *ra = Add2Ptr(rhdr, ro);
+	u16 ra_len = le16_to_cpu(ra->ra_len);
+	const struct CLIENT_REC *ca;
+	u32 i;
+
+	if (usa_error && ra_len + ro > SECTOR_SIZE - sizeof(short))
+		return false;
+
+	/* Find the start of the client array. */
+	ca = Add2Ptr(ra, le16_to_cpu(ra->client_off));
+
+	/*
+	 * Start with the free list.
+	 * Check that all the clients are valid and that there isn't a cycle.
+	 * Do the in-use list on the second pass.
+	 */
+	for (i = 0; i < 2; i++) {
+		u16 client_idx = le16_to_cpu(ra->client_idx[i]);
+		bool first_client = true;
+		u16 clients = le16_to_cpu(ra->log_clients);
+
+		while (client_idx != LFS_NO_CLIENT) {
+			const struct CLIENT_REC *cr;
+
+			if (!clients ||
+			    client_idx >= le16_to_cpu(ra->log_clients))
+				return false;
+
+			clients -= 1;
+			cr = ca + client_idx;
+
+			client_idx = le16_to_cpu(cr->next_client);
+
+			if (first_client) {
+				first_client = false;
+				if (cr->prev_client != LFS_NO_CLIENT_LE)
+					return false;
+			}
+		}
+	}
+
+	return true;
+}
+
+/*
+ * remove_client
+ *
+ * Remove a client record from a client record list an restart area.
+ */
+static inline void remove_client(struct CLIENT_REC *ca,
+				 const struct CLIENT_REC *cr, __le16 *head)
+{
+	if (cr->prev_client == LFS_NO_CLIENT_LE)
+		*head = cr->next_client;
+	else
+		ca[le16_to_cpu(cr->prev_client)].next_client = cr->next_client;
+
+	if (cr->next_client != LFS_NO_CLIENT_LE)
+		ca[le16_to_cpu(cr->next_client)].prev_client = cr->prev_client;
+}
+
+/*
+ * add_client - Add a client record to the start of a list.
+ */
+static inline void add_client(struct CLIENT_REC *ca, u16 index, __le16 *head)
+{
+	struct CLIENT_REC *cr = ca + index;
+
+	cr->prev_client = LFS_NO_CLIENT_LE;
+	cr->next_client = *head;
+
+	if (*head != LFS_NO_CLIENT_LE)
+		ca[le16_to_cpu(*head)].prev_client = cpu_to_le16(index);
+
+	*head = cpu_to_le16(index);
+}
+
+static inline void *enum_rstbl(struct RESTART_TABLE *t, void *c)
+{
+	__le32 *e;
+	u32 bprt;
+	u16 rsize = t ? le16_to_cpu(t->size) : 0;
+
+	if (!c) {
+		if (!t || !t->total)
+			return NULL;
+		e = Add2Ptr(t, sizeof(struct RESTART_TABLE));
+	} else {
+		e = Add2Ptr(c, rsize);
+	}
+
+	/* Loop until we hit the first one allocated, or the end of the list. */
+	for (bprt = bytes_per_rt(t); PtrOffset(t, e) < bprt;
+	     e = Add2Ptr(e, rsize)) {
+		if (*e == RESTART_ENTRY_ALLOCATED_LE)
+			return e;
+	}
+	return NULL;
+}
+
+/*
+ * find_dp - Search for a @vcn in Dirty Page Table.
+ */
+static inline struct DIR_PAGE_ENTRY *find_dp(struct RESTART_TABLE *dptbl,
+					     u32 target_attr, u64 vcn)
+{
+	__le32 ta = cpu_to_le32(target_attr);
+	struct DIR_PAGE_ENTRY *dp = NULL;
+
+	while ((dp = enum_rstbl(dptbl, dp))) {
+		u64 dp_vcn = le64_to_cpu(dp->vcn);
+
+		if (dp->target_attr == ta && vcn >= dp_vcn &&
+		    vcn < dp_vcn + le32_to_cpu(dp->lcns_follow)) {
+			return dp;
+		}
+	}
+	return NULL;
+}
+
+static inline u32 norm_file_page(u32 page_size, u32 *l_size, bool use_default)
+{
+	if (use_default)
+		page_size = DefaultLogPageSize;
+
+	/* Round the file size down to a system page boundary. */
+	*l_size &= ~(page_size - 1);
+
+	/* File should contain at least 2 restart pages and MinLogRecordPages pages. */
+	if (*l_size < (MinLogRecordPages + 2) * page_size)
+		return 0;
+
+	return page_size;
+}
+
+static bool check_log_rec(const struct LOG_REC_HDR *lr, u32 bytes, u32 tr,
+			  u32 bytes_per_attr_entry)
+{
+	u16 t16;
+
+	if (bytes < sizeof(struct LOG_REC_HDR))
+		return false;
+	if (!tr)
+		return false;
+
+	if ((tr - sizeof(struct RESTART_TABLE)) %
+	    sizeof(struct TRANSACTION_ENTRY))
+		return false;
+
+	if (le16_to_cpu(lr->redo_off) & 7)
+		return false;
+
+	if (le16_to_cpu(lr->undo_off) & 7)
+		return false;
+
+	if (lr->target_attr)
+		goto check_lcns;
+
+	if (is_target_required(le16_to_cpu(lr->redo_op)))
+		return false;
+
+	if (is_target_required(le16_to_cpu(lr->undo_op)))
+		return false;
+
+check_lcns:
+	if (!lr->lcns_follow)
+		goto check_length;
+
+	t16 = le16_to_cpu(lr->target_attr);
+	if ((t16 - sizeof(struct RESTART_TABLE)) % bytes_per_attr_entry)
+		return false;
+
+check_length:
+	if (bytes < lrh_length(lr))
+		return false;
+
+	return true;
+}
+
+static bool check_rstbl(const struct RESTART_TABLE *rt, size_t bytes)
+{
+	u32 ts;
+	u32 i, off;
+	u16 rsize = le16_to_cpu(rt->size);
+	u16 ne = le16_to_cpu(rt->used);
+	u32 ff = le32_to_cpu(rt->first_free);
+	u32 lf = le32_to_cpu(rt->last_free);
+
+	ts = rsize * ne + sizeof(struct RESTART_TABLE);
+
+	if (!rsize || rsize > bytes ||
+	    rsize + sizeof(struct RESTART_TABLE) > bytes || bytes < ts ||
+	    le16_to_cpu(rt->total) > ne || ff > ts || lf > ts ||
+	    (ff && ff < sizeof(struct RESTART_TABLE)) ||
+	    (lf && lf < sizeof(struct RESTART_TABLE))) {
+		return false;
+	}
+
+	/*
+	 * Verify each entry is either allocated or points
+	 * to a valid offset the table.
+	 */
+	for (i = 0; i < ne; i++) {
+		off = le32_to_cpu(*(__le32 *)Add2Ptr(
+			rt, i * rsize + sizeof(struct RESTART_TABLE)));
+
+		if (off != RESTART_ENTRY_ALLOCATED && off &&
+		    (off < sizeof(struct RESTART_TABLE) ||
+		     ((off - sizeof(struct RESTART_TABLE)) % rsize))) {
+			return false;
+		}
+	}
+
+	/*
+	 * Walk through the list headed by the first entry to make
+	 * sure none of the entries are currently being used.
+	 */
+	for (off = ff; off;) {
+		if (off == RESTART_ENTRY_ALLOCATED)
+			return false;
+
+		off = le32_to_cpu(*(__le32 *)Add2Ptr(rt, off));
+	}
+
+	return true;
+}
+
+/*
+ * free_rsttbl_idx - Free a previously allocated index a Restart Table.
+ */
+static inline void free_rsttbl_idx(struct RESTART_TABLE *rt, u32 off)
+{
+	__le32 *e;
+	u32 lf = le32_to_cpu(rt->last_free);
+	__le32 off_le = cpu_to_le32(off);
+
+	e = Add2Ptr(rt, off);
+
+	if (off < le32_to_cpu(rt->free_goal)) {
+		*e = rt->first_free;
+		rt->first_free = off_le;
+		if (!lf)
+			rt->last_free = off_le;
+	} else {
+		if (lf)
+			*(__le32 *)Add2Ptr(rt, lf) = off_le;
+		else
+			rt->first_free = off_le;
+
+		rt->last_free = off_le;
+		*e = 0;
+	}
+
+	le16_sub_cpu(&rt->total, 1);
+}
+
+static inline struct RESTART_TABLE *init_rsttbl(u16 esize, u16 used)
+{
+	__le32 *e, *last_free;
+	u32 off;
+	u32 bytes = esize * used + sizeof(struct RESTART_TABLE);
+	u32 lf = sizeof(struct RESTART_TABLE) + (used - 1) * esize;
+	struct RESTART_TABLE *t = kzalloc(bytes, GFP_NOFS);
+
+	if (!t)
+		return NULL;
+
+	t->size = cpu_to_le16(esize);
+	t->used = cpu_to_le16(used);
+	t->free_goal = cpu_to_le32(~0u);
+	t->first_free = cpu_to_le32(sizeof(struct RESTART_TABLE));
+	t->last_free = cpu_to_le32(lf);
+
+	e = (__le32 *)(t + 1);
+	last_free = Add2Ptr(t, lf);
+
+	for (off = sizeof(struct RESTART_TABLE) + esize; e < last_free;
+	     e = Add2Ptr(e, esize), off += esize) {
+		*e = cpu_to_le32(off);
+	}
+	return t;
+}
+
+static inline struct RESTART_TABLE *extend_rsttbl(struct RESTART_TABLE *tbl,
+						  u32 add, u32 free_goal)
+{
+	u16 esize = le16_to_cpu(tbl->size);
+	__le32 osize = cpu_to_le32(bytes_per_rt(tbl));
+	u32 used = le16_to_cpu(tbl->used);
+	struct RESTART_TABLE *rt;
+
+	rt = init_rsttbl(esize, used + add);
+	if (!rt)
+		return NULL;
+
+	memcpy(rt + 1, tbl + 1, esize * used);
+
+	rt->free_goal = free_goal == ~0u
+				? cpu_to_le32(~0u)
+				: cpu_to_le32(sizeof(struct RESTART_TABLE) +
+					      free_goal * esize);
+
+	if (tbl->first_free) {
+		rt->first_free = tbl->first_free;
+		*(__le32 *)Add2Ptr(rt, le32_to_cpu(tbl->last_free)) = osize;
+	} else {
+		rt->first_free = osize;
+	}
+
+	rt->total = tbl->total;
+
+	kfree(tbl);
+	return rt;
+}
+
+/*
+ * alloc_rsttbl_idx
+ *
+ * Allocate an index from within a previously initialized Restart Table.
+ */
+static inline void *alloc_rsttbl_idx(struct RESTART_TABLE **tbl)
+{
+	u32 off;
+	__le32 *e;
+	struct RESTART_TABLE *t = *tbl;
+
+	if (!t->first_free) {
+		*tbl = t = extend_rsttbl(t, 16, ~0u);
+		if (!t)
+			return NULL;
+	}
+
+	off = le32_to_cpu(t->first_free);
+
+	/* Dequeue this entry and zero it. */
+	e = Add2Ptr(t, off);
+
+	t->first_free = *e;
+
+	memset(e, 0, le16_to_cpu(t->size));
+
+	*e = RESTART_ENTRY_ALLOCATED_LE;
+
+	/* If list is going empty, then we fix the last_free as well. */
+	if (!t->first_free)
+		t->last_free = 0;
+
+	le16_add_cpu(&t->total, 1);
+
+	return Add2Ptr(t, off);
+}
+
+/*
+ * alloc_rsttbl_from_idx
+ *
+ * Allocate a specific index from within a previously initialized Restart Table.
+ */
+static inline void *alloc_rsttbl_from_idx(struct RESTART_TABLE **tbl, u32 vbo)
+{
+	u32 off;
+	__le32 *e;
+	struct RESTART_TABLE *rt = *tbl;
+	u32 bytes = bytes_per_rt(rt);
+	u16 esize = le16_to_cpu(rt->size);
+
+	/* If the entry is not the table, we will have to extend the table. */
+	if (vbo >= bytes) {
+		/*
+		 * Extend the size by computing the number of entries between
+		 * the existing size and the desired index and adding 1 to that.
+		 */
+		u32 bytes2idx = vbo - bytes;
+
+		/*
+		 * There should always be an integral number of entries
+		 * being added. Now extend the table.
+		 */
+		*tbl = rt = extend_rsttbl(rt, bytes2idx / esize + 1, bytes);
+		if (!rt)
+			return NULL;
+	}
+
+	/* See if the entry is already allocated, and just return if it is. */
+	e = Add2Ptr(rt, vbo);
+
+	if (*e == RESTART_ENTRY_ALLOCATED_LE)
+		return e;
+
+	/*
+	 * Walk through the table, looking for the entry we're
+	 * interested and the previous entry.
+	 */
+	off = le32_to_cpu(rt->first_free);
+	e = Add2Ptr(rt, off);
+
+	if (off == vbo) {
+		/* this is a match */
+		rt->first_free = *e;
+		goto skip_looking;
+	}
+
+	/*
+	 * Need to walk through the list looking for the predecessor
+	 * of our entry.
+	 */
+	for (;;) {
+		/* Remember the entry just found */
+		u32 last_off = off;
+		__le32 *last_e = e;
+
+		/* Should never run of entries. */
+
+		/* Lookup up the next entry the list. */
+		off = le32_to_cpu(*last_e);
+		e = Add2Ptr(rt, off);
+
+		/* If this is our match we are done. */
+		if (off == vbo) {
+			*last_e = *e;
+
+			/*
+			 * If this was the last entry, we update that
+			 * table as well.
+			 */
+			if (le32_to_cpu(rt->last_free) == off)
+				rt->last_free = cpu_to_le32(last_off);
+			break;
+		}
+	}
+
+skip_looking:
+	/* If the list is now empty, we fix the last_free as well. */
+	if (!rt->first_free)
+		rt->last_free = 0;
+
+	/* Zero this entry. */
+	memset(e, 0, esize);
+	*e = RESTART_ENTRY_ALLOCATED_LE;
+
+	le16_add_cpu(&rt->total, 1);
+
+	return e;
+}
+
+#define RESTART_SINGLE_PAGE_IO cpu_to_le16(0x0001)
+
+#define NTFSLOG_WRAPPED 0x00000001
+#define NTFSLOG_MULTIPLE_PAGE_IO 0x00000002
+#define NTFSLOG_NO_LAST_LSN 0x00000004
+#define NTFSLOG_REUSE_TAIL 0x00000010
+#define NTFSLOG_NO_OLDEST_LSN 0x00000020
+
+/* Helper struct to work with NTFS $LogFile. */
+struct ntfs_log {
+	struct ntfs_inode *ni;
+
+	u32 l_size;
+	u32 sys_page_size;
+	u32 sys_page_mask;
+	u32 page_size;
+	u32 page_mask; // page_size - 1
+	u8 page_bits;
+	struct RECORD_PAGE_HDR *one_page_buf;
+
+	struct RESTART_TABLE *open_attr_tbl;
+	u32 transaction_id;
+	u32 clst_per_page;
+
+	u32 first_page;
+	u32 next_page;
+	u32 ra_off;
+	u32 data_off;
+	u32 restart_size;
+	u32 data_size;
+	u16 record_header_len;
+	u64 seq_num;
+	u32 seq_num_bits;
+	u32 file_data_bits;
+	u32 seq_num_mask; /* (1 << file_data_bits) - 1 */
+
+	struct RESTART_AREA *ra; /* In-memory image of the next restart area. */
+	u32 ra_size; /* The usable size of the restart area. */
+
+	/*
+	 * If true, then the in-memory restart area is to be written
+	 * to the first position on the disk.
+	 */
+	bool init_ra;
+	bool set_dirty; /* True if we need to set dirty flag. */
+
+	u64 oldest_lsn;
+
+	u32 oldest_lsn_off;
+	u64 last_lsn;
+
+	u32 total_avail;
+	u32 total_avail_pages;
+	u32 total_undo_commit;
+	u32 max_current_avail;
+	u32 current_avail;
+	u32 reserved;
+
+	short major_ver;
+	short minor_ver;
+
+	u32 l_flags; /* See NTFSLOG_XXX */
+	u32 current_openlog_count; /* On-disk value for open_log_count. */
+
+	struct CLIENT_ID client_id;
+	u32 client_undo_commit;
+};
+
+static inline u32 lsn_to_vbo(struct ntfs_log *log, const u64 lsn)
+{
+	u32 vbo = (lsn << log->seq_num_bits) >> (log->seq_num_bits - 3);
+
+	return vbo;
+}
+
+/* Compute the offset in the log file of the next log page. */
+static inline u32 next_page_off(struct ntfs_log *log, u32 off)
+{
+	off = (off & ~log->sys_page_mask) + log->page_size;
+	return off >= log->l_size ? log->first_page : off;
+}
+
+static inline u32 lsn_to_page_off(struct ntfs_log *log, u64 lsn)
+{
+	return (((u32)lsn) << 3) & log->page_mask;
+}
+
+static inline u64 vbo_to_lsn(struct ntfs_log *log, u32 off, u64 Seq)
+{
+	return (off >> 3) + (Seq << log->file_data_bits);
+}
+
+static inline bool is_lsn_in_file(struct ntfs_log *log, u64 lsn)
+{
+	return lsn >= log->oldest_lsn &&
+	       lsn <= le64_to_cpu(log->ra->current_lsn);
+}
+
+static inline u32 hdr_file_off(struct ntfs_log *log,
+			       struct RECORD_PAGE_HDR *hdr)
+{
+	if (log->major_ver < 2)
+		return le64_to_cpu(hdr->rhdr.lsn);
+
+	return le32_to_cpu(hdr->file_off);
+}
+
+static inline u64 base_lsn(struct ntfs_log *log,
+			   const struct RECORD_PAGE_HDR *hdr, u64 lsn)
+{
+	u64 h_lsn = le64_to_cpu(hdr->rhdr.lsn);
+	u64 ret = (((h_lsn >> log->file_data_bits) +
+		    (lsn < (lsn_to_vbo(log, h_lsn) & ~log->page_mask) ? 1 : 0))
+		   << log->file_data_bits) +
+		  ((((is_log_record_end(hdr) &&
+		      h_lsn <= le64_to_cpu(hdr->record_hdr.last_end_lsn))
+			     ? le16_to_cpu(hdr->record_hdr.next_record_off)
+			     : log->page_size) +
+		    lsn) >>
+		   3);
+
+	return ret;
+}
+
+static inline bool verify_client_lsn(struct ntfs_log *log,
+				     const struct CLIENT_REC *client, u64 lsn)
+{
+	return lsn >= le64_to_cpu(client->oldest_lsn) &&
+	       lsn <= le64_to_cpu(log->ra->current_lsn) && lsn;
+}
+
+struct restart_info {
+	u64 last_lsn;
+	struct RESTART_HDR *r_page;
+	u32 vbo;
+	bool chkdsk_was_run;
+	bool valid_page;
+	bool initialized;
+	bool restart;
+};
+
+static int read_log_page(struct ntfs_log *log, u32 vbo,
+			 struct RECORD_PAGE_HDR **buffer, bool *usa_error)
+{
+	int err = 0;
+	u32 page_idx = vbo >> log->page_bits;
+	u32 page_off = vbo & log->page_mask;
+	u32 bytes = log->page_size - page_off;
+	void *to_free = NULL;
+	u32 page_vbo = page_idx << log->page_bits;
+	struct RECORD_PAGE_HDR *page_buf;
+	struct ntfs_inode *ni = log->ni;
+	bool bBAAD;
+
+	if (vbo >= log->l_size)
+		return -EINVAL;
+
+	if (!*buffer) {
+		to_free = kmalloc(bytes, GFP_NOFS);
+		if (!to_free)
+			return -ENOMEM;
+		*buffer = to_free;
+	}
+
+	page_buf = page_off ? log->one_page_buf : *buffer;
+
+	err = ntfs_read_run_nb(ni->mi.sbi, &ni->file.run, page_vbo, page_buf,
+			       log->page_size, NULL);
+	if (err)
+		goto out;
+
+	if (page_buf->rhdr.sign != NTFS_FFFF_SIGNATURE)
+		ntfs_fix_post_read(&page_buf->rhdr, PAGE_SIZE, false);
+
+	if (page_buf != *buffer)
+		memcpy(*buffer, Add2Ptr(page_buf, page_off), bytes);
+
+	bBAAD = page_buf->rhdr.sign == NTFS_BAAD_SIGNATURE;
+
+	if (usa_error)
+		*usa_error = bBAAD;
+	/* Check that the update sequence array for this page is valid */
+	/* If we don't allow errors, raise an error status */
+	else if (bBAAD)
+		err = -EINVAL;
+
+out:
+	if (err && to_free) {
+		kfree(to_free);
+		*buffer = NULL;
+	}
+
+	return err;
+}
+
+/*
+ * log_read_rst
+ *
+ * It walks through 512 blocks of the file looking for a valid
+ * restart page header. It will stop the first time we find a
+ * valid page header.
+ */
+static int log_read_rst(struct ntfs_log *log, u32 l_size, bool first,
+			struct restart_info *info)
+{
+	u32 skip, vbo;
+	struct RESTART_HDR *r_page = kmalloc(DefaultLogPageSize, GFP_NOFS);
+
+	if (!r_page)
+		return -ENOMEM;
+
+	memset(info, 0, sizeof(struct restart_info));
+
+	/* Determine which restart area we are looking for. */
+	if (first) {
+		vbo = 0;
+		skip = 512;
+	} else {
+		vbo = 512;
+		skip = 0;
+	}
+
+	/* Loop continuously until we succeed. */
+	for (; vbo < l_size; vbo = 2 * vbo + skip, skip = 0) {
+		bool usa_error;
+		u32 sys_page_size;
+		bool brst, bchk;
+		struct RESTART_AREA *ra;
+
+		/* Read a page header at the current offset. */
+		if (read_log_page(log, vbo, (struct RECORD_PAGE_HDR **)&r_page,
+				  &usa_error)) {
+			/* Ignore any errors. */
+			continue;
+		}
+
+		/* Exit if the signature is a log record page. */
+		if (r_page->rhdr.sign == NTFS_RCRD_SIGNATURE) {
+			info->initialized = true;
+			break;
+		}
+
+		brst = r_page->rhdr.sign == NTFS_RSTR_SIGNATURE;
+		bchk = r_page->rhdr.sign == NTFS_CHKD_SIGNATURE;
+
+		if (!bchk && !brst) {
+			if (r_page->rhdr.sign != NTFS_FFFF_SIGNATURE) {
+				/*
+				 * Remember if the signature does not
+				 * indicate uninitialized file.
+				 */
+				info->initialized = true;
+			}
+			continue;
+		}
+
+		ra = NULL;
+		info->valid_page = false;
+		info->initialized = true;
+		info->vbo = vbo;
+
+		/* Let's check the restart area if this is a valid page. */
+		if (!is_rst_page_hdr_valid(vbo, r_page))
+			goto check_result;
+		ra = Add2Ptr(r_page, le16_to_cpu(r_page->ra_off));
+
+		if (!is_rst_area_valid(r_page))
+			goto check_result;
+
+		/*
+		 * We have a valid restart page header and restart area.
+		 * If chkdsk was run or we have no clients then we have
+		 * no more checking to do.
+		 */
+		if (bchk || ra->client_idx[1] == LFS_NO_CLIENT_LE) {
+			info->valid_page = true;
+			goto check_result;
+		}
+
+		/* Read the entire restart area. */
+		sys_page_size = le32_to_cpu(r_page->sys_page_size);
+		if (DefaultLogPageSize != sys_page_size) {
+			kfree(r_page);
+			r_page = kzalloc(sys_page_size, GFP_NOFS);
+			if (!r_page)
+				return -ENOMEM;
+
+			if (read_log_page(log, vbo,
+					  (struct RECORD_PAGE_HDR **)&r_page,
+					  &usa_error)) {
+				/* Ignore any errors. */
+				kfree(r_page);
+				r_page = NULL;
+				continue;
+			}
+		}
+
+		if (is_client_area_valid(r_page, usa_error)) {
+			info->valid_page = true;
+			ra = Add2Ptr(r_page, le16_to_cpu(r_page->ra_off));
+		}
+
+check_result:
+		/*
+		 * If chkdsk was run then update the caller's
+		 * values and return.
+		 */
+		if (r_page->rhdr.sign == NTFS_CHKD_SIGNATURE) {
+			info->chkdsk_was_run = true;
+			info->last_lsn = le64_to_cpu(r_page->rhdr.lsn);
+			info->restart = true;
+			info->r_page = r_page;
+			return 0;
+		}
+
+		/*
+		 * If we have a valid page then copy the values
+		 * we need from it.
+		 */
+		if (info->valid_page) {
+			info->last_lsn = le64_to_cpu(ra->current_lsn);
+			info->restart = true;
+			info->r_page = r_page;
+			return 0;
+		}
+	}
+
+	kfree(r_page);
+
+	return 0;
+}
+
+/*
+ * Ilog_init_pg_hdr - Init @log from restart page header.
+ */
+static void log_init_pg_hdr(struct ntfs_log *log, u32 sys_page_size,
+			    u32 page_size, u16 major_ver, u16 minor_ver)
+{
+	log->sys_page_size = sys_page_size;
+	log->sys_page_mask = sys_page_size - 1;
+	log->page_size = page_size;
+	log->page_mask = page_size - 1;
+	log->page_bits = blksize_bits(page_size);
+
+	log->clst_per_page = log->page_size >> log->ni->mi.sbi->cluster_bits;
+	if (!log->clst_per_page)
+		log->clst_per_page = 1;
+
+	log->first_page = major_ver >= 2
+				  ? 0x22 * page_size
+				  : ((sys_page_size << 1) + (page_size << 1));
+	log->major_ver = major_ver;
+	log->minor_ver = minor_ver;
+}
+
+/*
+ * log_create - Init @log in cases when we don't have a restart area to use.
+ */
+static void log_create(struct ntfs_log *log, u32 l_size, const u64 last_lsn,
+		       u32 open_log_count, bool wrapped, bool use_multi_page)
+{
+	log->l_size = l_size;
+	/* All file offsets must be quadword aligned. */
+	log->file_data_bits = blksize_bits(l_size) - 3;
+	log->seq_num_mask = (8 << log->file_data_bits) - 1;
+	log->seq_num_bits = sizeof(u64) * 8 - log->file_data_bits;
+	log->seq_num = (last_lsn >> log->file_data_bits) + 2;
+	log->next_page = log->first_page;
+	log->oldest_lsn = log->seq_num << log->file_data_bits;
+	log->oldest_lsn_off = 0;
+	log->last_lsn = log->oldest_lsn;
+
+	log->l_flags |= NTFSLOG_NO_LAST_LSN | NTFSLOG_NO_OLDEST_LSN;
+
+	/* Set the correct flags for the I/O and indicate if we have wrapped. */
+	if (wrapped)
+		log->l_flags |= NTFSLOG_WRAPPED;
+
+	if (use_multi_page)
+		log->l_flags |= NTFSLOG_MULTIPLE_PAGE_IO;
+
+	/* Compute the log page values. */
+	log->data_off = ALIGN(
+		offsetof(struct RECORD_PAGE_HDR, fixups) +
+			sizeof(short) * ((log->page_size >> SECTOR_SHIFT) + 1),
+		8);
+	log->data_size = log->page_size - log->data_off;
+	log->record_header_len = sizeof(struct LFS_RECORD_HDR);
+
+	/* Remember the different page sizes for reservation. */
+	log->reserved = log->data_size - log->record_header_len;
+
+	/* Compute the restart page values. */
+	log->ra_off = ALIGN(
+		offsetof(struct RESTART_HDR, fixups) +
+			sizeof(short) *
+				((log->sys_page_size >> SECTOR_SHIFT) + 1),
+		8);
+	log->restart_size = log->sys_page_size - log->ra_off;
+	log->ra_size = struct_size(log->ra, clients, 1);
+	log->current_openlog_count = open_log_count;
+
+	/*
+	 * The total available log file space is the number of
+	 * log file pages times the space available on each page.
+	 */
+	log->total_avail_pages = log->l_size - log->first_page;
+	log->total_avail = log->total_avail_pages >> log->page_bits;
+
+	/*
+	 * We assume that we can't use the end of the page less than
+	 * the file record size.
+	 * Then we won't need to reserve more than the caller asks for.
+	 */
+	log->max_current_avail = log->total_avail * log->reserved;
+	log->total_avail = log->total_avail * log->data_size;
+	log->current_avail = log->max_current_avail;
+}
+
+/*
+ * log_create_ra - Fill a restart area from the values stored in @log.
+ */
+static struct RESTART_AREA *log_create_ra(struct ntfs_log *log)
+{
+	struct CLIENT_REC *cr;
+	struct RESTART_AREA *ra = kzalloc(log->restart_size, GFP_NOFS);
+
+	if (!ra)
+		return NULL;
+
+	ra->current_lsn = cpu_to_le64(log->last_lsn);
+	ra->log_clients = cpu_to_le16(1);
+	ra->client_idx[1] = LFS_NO_CLIENT_LE;
+	if (log->l_flags & NTFSLOG_MULTIPLE_PAGE_IO)
+		ra->flags = RESTART_SINGLE_PAGE_IO;
+	ra->seq_num_bits = cpu_to_le32(log->seq_num_bits);
+	ra->ra_len = cpu_to_le16(log->ra_size);
+	ra->client_off = cpu_to_le16(offsetof(struct RESTART_AREA, clients));
+	ra->l_size = cpu_to_le64(log->l_size);
+	ra->rec_hdr_len = cpu_to_le16(log->record_header_len);
+	ra->data_off = cpu_to_le16(log->data_off);
+	ra->open_log_count = cpu_to_le32(log->current_openlog_count + 1);
+
+	cr = ra->clients;
+
+	cr->prev_client = LFS_NO_CLIENT_LE;
+	cr->next_client = LFS_NO_CLIENT_LE;
+
+	return ra;
+}
+
+static u32 final_log_off(struct ntfs_log *log, u64 lsn, u32 data_len)
+{
+	u32 base_vbo = lsn << 3;
+	u32 final_log_off = (base_vbo & log->seq_num_mask) & ~log->page_mask;
+	u32 page_off = base_vbo & log->page_mask;
+	u32 tail = log->page_size - page_off;
+
+	page_off -= 1;
+
+	/* Add the length of the header. */
+	data_len += log->record_header_len;
+
+	/*
+	 * If this lsn is contained this log page we are done.
+	 * Otherwise we need to walk through several log pages.
+	 */
+	if (data_len > tail) {
+		data_len -= tail;
+		tail = log->data_size;
+		page_off = log->data_off - 1;
+
+		for (;;) {
+			final_log_off = next_page_off(log, final_log_off);
+
+			/*
+			 * We are done if the remaining bytes
+			 * fit on this page.
+			 */
+			if (data_len <= tail)
+				break;
+			data_len -= tail;
+		}
+	}
+
+	/*
+	 * We add the remaining bytes to our starting position on this page
+	 * and then add that value to the file offset of this log page.
+	 */
+	return final_log_off + data_len + page_off;
+}
+
+static int next_log_lsn(struct ntfs_log *log, const struct LFS_RECORD_HDR *rh,
+			u64 *lsn)
+{
+	int err;
+	u64 this_lsn = le64_to_cpu(rh->this_lsn);
+	u32 vbo = lsn_to_vbo(log, this_lsn);
+	u32 end =
+		final_log_off(log, this_lsn, le32_to_cpu(rh->client_data_len));
+	u32 hdr_off = end & ~log->sys_page_mask;
+	u64 seq = this_lsn >> log->file_data_bits;
+	struct RECORD_PAGE_HDR *page = NULL;
+
+	/* Remember if we wrapped. */
+	if (end <= vbo)
+		seq += 1;
+
+	/* Log page header for this page. */
+	err = read_log_page(log, hdr_off, &page, NULL);
+	if (err)
+		return err;
+
+	/*
+	 * If the lsn we were given was not the last lsn on this page,
+	 * then the starting offset for the next lsn is on a quad word
+	 * boundary following the last file offset for the current lsn.
+	 * Otherwise the file offset is the start of the data on the next page.
+	 */
+	if (this_lsn == le64_to_cpu(page->rhdr.lsn)) {
+		/* If we wrapped, we need to increment the sequence number. */
+		hdr_off = next_page_off(log, hdr_off);
+		if (hdr_off == log->first_page)
+			seq += 1;
+
+		vbo = hdr_off + log->data_off;
+	} else {
+		vbo = ALIGN(end, 8);
+	}
+
+	/* Compute the lsn based on the file offset and the sequence count. */
+	*lsn = vbo_to_lsn(log, vbo, seq);
+
+	/*
+	 * If this lsn is within the legal range for the file, we return true.
+	 * Otherwise false indicates that there are no more lsn's.
+	 */
+	if (!is_lsn_in_file(log, *lsn))
+		*lsn = 0;
+
+	kfree(page);
+
+	return 0;
+}
+
+/*
+ * current_log_avail - Calculate the number of bytes available for log records.
+ */
+static u32 current_log_avail(struct ntfs_log *log)
+{
+	u32 oldest_off, next_free_off, free_bytes;
+
+	if (log->l_flags & NTFSLOG_NO_LAST_LSN) {
+		/* The entire file is available. */
+		return log->max_current_avail;
+	}
+
+	/*
+	 * If there is a last lsn the restart area then we know that we will
+	 * have to compute the free range.
+	 * If there is no oldest lsn then start at the first page of the file.
+	 */
+	oldest_off = (log->l_flags & NTFSLOG_NO_OLDEST_LSN)
+			     ? log->first_page
+			     : (log->oldest_lsn_off & ~log->sys_page_mask);
+
+	/*
+	 * We will use the next log page offset to compute the next free page.
+	 * If we are going to reuse this page go to the next page.
+	 * If we are at the first page then use the end of the file.
+	 */
+	next_free_off = (log->l_flags & NTFSLOG_REUSE_TAIL)
+				? log->next_page + log->page_size
+				: log->next_page == log->first_page
+					  ? log->l_size
+					  : log->next_page;
+
+	/* If the two offsets are the same then there is no available space. */
+	if (oldest_off == next_free_off)
+		return 0;
+	/*
+	 * If the free offset follows the oldest offset then subtract
+	 * this range from the total available pages.
+	 */
+	free_bytes =
+		oldest_off < next_free_off
+			? log->total_avail_pages - (next_free_off - oldest_off)
+			: oldest_off - next_free_off;
+
+	free_bytes >>= log->page_bits;
+	return free_bytes * log->reserved;
+}
+
+static bool check_subseq_log_page(struct ntfs_log *log,
+				  const struct RECORD_PAGE_HDR *rp, u32 vbo,
+				  u64 seq)
+{
+	u64 lsn_seq;
+	const struct NTFS_RECORD_HEADER *rhdr = &rp->rhdr;
+	u64 lsn = le64_to_cpu(rhdr->lsn);
+
+	if (rhdr->sign == NTFS_FFFF_SIGNATURE || !rhdr->sign)
+		return false;
+
+	/*
+	 * If the last lsn on the page occurs was written after the page
+	 * that caused the original error then we have a fatal error.
+	 */
+	lsn_seq = lsn >> log->file_data_bits;
+
+	/*
+	 * If the sequence number for the lsn the page is equal or greater
+	 * than lsn we expect, then this is a subsequent write.
+	 */
+	return lsn_seq >= seq ||
+	       (lsn_seq == seq - 1 && log->first_page == vbo &&
+		vbo != (lsn_to_vbo(log, lsn) & ~log->page_mask));
+}
+
+/*
+ * last_log_lsn
+ *
+ * Walks through the log pages for a file, searching for the
+ * last log page written to the file.
+ */
+static int last_log_lsn(struct ntfs_log *log)
+{
+	int err;
+	bool usa_error = false;
+	bool replace_page = false;
+	bool reuse_page = log->l_flags & NTFSLOG_REUSE_TAIL;
+	bool wrapped_file, wrapped;
+
+	u32 page_cnt = 1, page_pos = 1;
+	u32 page_off = 0, page_off1 = 0, saved_off = 0;
+	u32 final_off, second_off, final_off_prev = 0, second_off_prev = 0;
+	u32 first_file_off = 0, second_file_off = 0;
+	u32 part_io_count = 0;
+	u32 tails = 0;
+	u32 this_off, curpage_off, nextpage_off, remain_pages;
+
+	u64 expected_seq, seq_base = 0, lsn_base = 0;
+	u64 best_lsn, best_lsn1, best_lsn2;
+	u64 lsn_cur, lsn1, lsn2;
+	u64 last_ok_lsn = reuse_page ? log->last_lsn : 0;
+
+	u16 cur_pos, best_page_pos;
+
+	struct RECORD_PAGE_HDR *page = NULL;
+	struct RECORD_PAGE_HDR *tst_page = NULL;
+	struct RECORD_PAGE_HDR *first_tail = NULL;
+	struct RECORD_PAGE_HDR *second_tail = NULL;
+	struct RECORD_PAGE_HDR *tail_page = NULL;
+	struct RECORD_PAGE_HDR *second_tail_prev = NULL;
+	struct RECORD_PAGE_HDR *first_tail_prev = NULL;
+	struct RECORD_PAGE_HDR *page_bufs = NULL;
+	struct RECORD_PAGE_HDR *best_page;
+
+	if (log->major_ver >= 2) {
+		final_off = 0x02 * log->page_size;
+		second_off = 0x12 * log->page_size;
+
+		// 0x10 == 0x12 - 0x2
+		page_bufs = kmalloc(log->page_size * 0x10, GFP_NOFS);
+		if (!page_bufs)
+			return -ENOMEM;
+	} else {
+		second_off = log->first_page - log->page_size;
+		final_off = second_off - log->page_size;
+	}
+
+next_tail:
+	/* Read second tail page (at pos 3/0x12000). */
+	if (read_log_page(log, second_off, &second_tail, &usa_error) ||
+	    usa_error || second_tail->rhdr.sign != NTFS_RCRD_SIGNATURE) {
+		kfree(second_tail);
+		second_tail = NULL;
+		second_file_off = 0;
+		lsn2 = 0;
+	} else {
+		second_file_off = hdr_file_off(log, second_tail);
+		lsn2 = le64_to_cpu(second_tail->record_hdr.last_end_lsn);
+	}
+
+	/* Read first tail page (at pos 2/0x2000). */
+	if (read_log_page(log, final_off, &first_tail, &usa_error) ||
+	    usa_error || first_tail->rhdr.sign != NTFS_RCRD_SIGNATURE) {
+		kfree(first_tail);
+		first_tail = NULL;
+		first_file_off = 0;
+		lsn1 = 0;
+	} else {
+		first_file_off = hdr_file_off(log, first_tail);
+		lsn1 = le64_to_cpu(first_tail->record_hdr.last_end_lsn);
+	}
+
+	if (log->major_ver < 2) {
+		int best_page;
+
+		first_tail_prev = first_tail;
+		final_off_prev = first_file_off;
+		second_tail_prev = second_tail;
+		second_off_prev = second_file_off;
+		tails = 1;
+
+		if (!first_tail && !second_tail)
+			goto tail_read;
+
+		if (first_tail && second_tail)
+			best_page = lsn1 < lsn2 ? 1 : 0;
+		else if (first_tail)
+			best_page = 0;
+		else
+			best_page = 1;
+
+		page_off = best_page ? second_file_off : first_file_off;
+		seq_base = (best_page ? lsn2 : lsn1) >> log->file_data_bits;
+		goto tail_read;
+	}
+
+	best_lsn1 = first_tail ? base_lsn(log, first_tail, first_file_off) : 0;
+	best_lsn2 =
+		second_tail ? base_lsn(log, second_tail, second_file_off) : 0;
+
+	if (first_tail && second_tail) {
+		if (best_lsn1 > best_lsn2) {
+			best_lsn = best_lsn1;
+			best_page = first_tail;
+			this_off = first_file_off;
+		} else {
+			best_lsn = best_lsn2;
+			best_page = second_tail;
+			this_off = second_file_off;
+		}
+	} else if (first_tail) {
+		best_lsn = best_lsn1;
+		best_page = first_tail;
+		this_off = first_file_off;
+	} else if (second_tail) {
+		best_lsn = best_lsn2;
+		best_page = second_tail;
+		this_off = second_file_off;
+	} else {
+		goto tail_read;
+	}
+
+	best_page_pos = le16_to_cpu(best_page->page_pos);
+
+	if (!tails) {
+		if (best_page_pos == page_pos) {
+			seq_base = best_lsn >> log->file_data_bits;
+			saved_off = page_off = le32_to_cpu(best_page->file_off);
+			lsn_base = best_lsn;
+
+			memmove(page_bufs, best_page, log->page_size);
+
+			page_cnt = le16_to_cpu(best_page->page_count);
+			if (page_cnt > 1)
+				page_pos += 1;
+
+			tails = 1;
+		}
+	} else if (seq_base == (best_lsn >> log->file_data_bits) &&
+		   saved_off + log->page_size == this_off &&
+		   lsn_base < best_lsn &&
+		   (page_pos != page_cnt || best_page_pos == page_pos ||
+		    best_page_pos == 1) &&
+		   (page_pos >= page_cnt || best_page_pos == page_pos)) {
+		u16 bppc = le16_to_cpu(best_page->page_count);
+
+		saved_off += log->page_size;
+		lsn_base = best_lsn;
+
+		memmove(Add2Ptr(page_bufs, tails * log->page_size), best_page,
+			log->page_size);
+
+		tails += 1;
+
+		if (best_page_pos != bppc) {
+			page_cnt = bppc;
+			page_pos = best_page_pos;
+
+			if (page_cnt > 1)
+				page_pos += 1;
+		} else {
+			page_pos = page_cnt = 1;
+		}
+	} else {
+		kfree(first_tail);
+		kfree(second_tail);
+		goto tail_read;
+	}
+
+	kfree(first_tail_prev);
+	first_tail_prev = first_tail;
+	final_off_prev = first_file_off;
+	first_tail = NULL;
+
+	kfree(second_tail_prev);
+	second_tail_prev = second_tail;
+	second_off_prev = second_file_off;
+	second_tail = NULL;
+
+	final_off += log->page_size;
+	second_off += log->page_size;
+
+	if (tails < 0x10)
+		goto next_tail;
+tail_read:
+	first_tail = first_tail_prev;
+	final_off = final_off_prev;
+
+	second_tail = second_tail_prev;
+	second_off = second_off_prev;
+
+	page_cnt = page_pos = 1;
+
+	curpage_off = seq_base == log->seq_num ? min(log->next_page, page_off)
+					       : log->next_page;
+
+	wrapped_file =
+		curpage_off == log->first_page &&
+		!(log->l_flags & (NTFSLOG_NO_LAST_LSN | NTFSLOG_REUSE_TAIL));
+
+	expected_seq = wrapped_file ? (log->seq_num + 1) : log->seq_num;
+
+	nextpage_off = curpage_off;
+
+next_page:
+	tail_page = NULL;
+	/* Read the next log page. */
+	err = read_log_page(log, curpage_off, &page, &usa_error);
+
+	/* Compute the next log page offset the file. */
+	nextpage_off = next_page_off(log, curpage_off);
+	wrapped = nextpage_off == log->first_page;
+
+	if (tails > 1) {
+		struct RECORD_PAGE_HDR *cur_page =
+			Add2Ptr(page_bufs, curpage_off - page_off);
+
+		if (curpage_off == saved_off) {
+			tail_page = cur_page;
+			goto use_tail_page;
+		}
+
+		if (page_off > curpage_off || curpage_off >= saved_off)
+			goto use_tail_page;
+
+		if (page_off1)
+			goto use_cur_page;
+
+		if (!err && !usa_error &&
+		    page->rhdr.sign == NTFS_RCRD_SIGNATURE &&
+		    cur_page->rhdr.lsn == page->rhdr.lsn &&
+		    cur_page->record_hdr.next_record_off ==
+			    page->record_hdr.next_record_off &&
+		    ((page_pos == page_cnt &&
+		      le16_to_cpu(page->page_pos) == 1) ||
+		     (page_pos != page_cnt &&
+		      le16_to_cpu(page->page_pos) == page_pos + 1 &&
+		      le16_to_cpu(page->page_count) == page_cnt))) {
+			cur_page = NULL;
+			goto use_tail_page;
+		}
+
+		page_off1 = page_off;
+
+use_cur_page:
+
+		lsn_cur = le64_to_cpu(cur_page->rhdr.lsn);
+
+		if (last_ok_lsn !=
+			    le64_to_cpu(cur_page->record_hdr.last_end_lsn) &&
+		    ((lsn_cur >> log->file_data_bits) +
+		     ((curpage_off <
+		       (lsn_to_vbo(log, lsn_cur) & ~log->page_mask))
+			      ? 1
+			      : 0)) != expected_seq) {
+			goto check_tail;
+		}
+
+		if (!is_log_record_end(cur_page)) {
+			tail_page = NULL;
+			last_ok_lsn = lsn_cur;
+			goto next_page_1;
+		}
+
+		log->seq_num = expected_seq;
+		log->l_flags &= ~NTFSLOG_NO_LAST_LSN;
+		log->last_lsn = le64_to_cpu(cur_page->record_hdr.last_end_lsn);
+		log->ra->current_lsn = cur_page->record_hdr.last_end_lsn;
+
+		if (log->record_header_len <=
+		    log->page_size -
+			    le16_to_cpu(cur_page->record_hdr.next_record_off)) {
+			log->l_flags |= NTFSLOG_REUSE_TAIL;
+			log->next_page = curpage_off;
+		} else {
+			log->l_flags &= ~NTFSLOG_REUSE_TAIL;
+			log->next_page = nextpage_off;
+		}
+
+		if (wrapped_file)
+			log->l_flags |= NTFSLOG_WRAPPED;
+
+		last_ok_lsn = le64_to_cpu(cur_page->record_hdr.last_end_lsn);
+		goto next_page_1;
+	}
+
+	/*
+	 * If we are at the expected first page of a transfer check to see
+	 * if either tail copy is at this offset.
+	 * If this page is the last page of a transfer, check if we wrote
+	 * a subsequent tail copy.
+	 */
+	if (page_cnt == page_pos || page_cnt == page_pos + 1) {
+		/*
+		 * Check if the offset matches either the first or second
+		 * tail copy. It is possible it will match both.
+		 */
+		if (curpage_off == final_off)
+			tail_page = first_tail;
+
+		/*
+		 * If we already matched on the first page then
+		 * check the ending lsn's.
+		 */
+		if (curpage_off == second_off) {
+			if (!tail_page ||
+			    (second_tail &&
+			     le64_to_cpu(second_tail->record_hdr.last_end_lsn) >
+				     le64_to_cpu(first_tail->record_hdr
+							 .last_end_lsn))) {
+				tail_page = second_tail;
+			}
+		}
+	}
+
+use_tail_page:
+	if (tail_page) {
+		/* We have a candidate for a tail copy. */
+		lsn_cur = le64_to_cpu(tail_page->record_hdr.last_end_lsn);
+
+		if (last_ok_lsn < lsn_cur) {
+			/*
+			 * If the sequence number is not expected,
+			 * then don't use the tail copy.
+			 */
+			if (expected_seq != (lsn_cur >> log->file_data_bits))
+				tail_page = NULL;
+		} else if (last_ok_lsn > lsn_cur) {
+			/*
+			 * If the last lsn is greater than the one on
+			 * this page then forget this tail.
+			 */
+			tail_page = NULL;
+		}
+	}
+
+	/*
+	 *If we have an error on the current page,
+	 * we will break of this loop.
+	 */
+	if (err || usa_error)
+		goto check_tail;
+
+	/*
+	 * Done if the last lsn on this page doesn't match the previous known
+	 * last lsn or the sequence number is not expected.
+	 */
+	lsn_cur = le64_to_cpu(page->rhdr.lsn);
+	if (last_ok_lsn != lsn_cur &&
+	    expected_seq != (lsn_cur >> log->file_data_bits)) {
+		goto check_tail;
+	}
+
+	/*
+	 * Check that the page position and page count values are correct.
+	 * If this is the first page of a transfer the position must be 1
+	 * and the count will be unknown.
+	 */
+	if (page_cnt == page_pos) {
+		if (page->page_pos != cpu_to_le16(1) &&
+		    (!reuse_page || page->page_pos != page->page_count)) {
+			/*
+			 * If the current page is the first page we are
+			 * looking at and we are reusing this page then
+			 * it can be either the first or last page of a
+			 * transfer. Otherwise it can only be the first.
+			 */
+			goto check_tail;
+		}
+	} else if (le16_to_cpu(page->page_count) != page_cnt ||
+		   le16_to_cpu(page->page_pos) != page_pos + 1) {
+		/*
+		 * The page position better be 1 more than the last page
+		 * position and the page count better match.
+		 */
+		goto check_tail;
+	}
+
+	/*
+	 * We have a valid page the file and may have a valid page
+	 * the tail copy area.
+	 * If the tail page was written after the page the file then
+	 * break of the loop.
+	 */
+	if (tail_page &&
+	    le64_to_cpu(tail_page->record_hdr.last_end_lsn) > lsn_cur) {
+		/* Remember if we will replace the page. */
+		replace_page = true;
+		goto check_tail;
+	}
+
+	tail_page = NULL;
+
+	if (is_log_record_end(page)) {
+		/*
+		 * Since we have read this page we know the sequence number
+		 * is the same as our expected value.
+		 */
+		log->seq_num = expected_seq;
+		log->last_lsn = le64_to_cpu(page->record_hdr.last_end_lsn);
+		log->ra->current_lsn = page->record_hdr.last_end_lsn;
+		log->l_flags &= ~NTFSLOG_NO_LAST_LSN;
+
+		/*
+		 * If there is room on this page for another header then
+		 * remember we want to reuse the page.
+		 */
+		if (log->record_header_len <=
+		    log->page_size -
+			    le16_to_cpu(page->record_hdr.next_record_off)) {
+			log->l_flags |= NTFSLOG_REUSE_TAIL;
+			log->next_page = curpage_off;
+		} else {
+			log->l_flags &= ~NTFSLOG_REUSE_TAIL;
+			log->next_page = nextpage_off;
+		}
+
+		/* Remember if we wrapped the log file. */
+		if (wrapped_file)
+			log->l_flags |= NTFSLOG_WRAPPED;
+	}
+
+	/*
+	 * Remember the last page count and position.
+	 * Also remember the last known lsn.
+	 */
+	page_cnt = le16_to_cpu(page->page_count);
+	page_pos = le16_to_cpu(page->page_pos);
+	last_ok_lsn = le64_to_cpu(page->rhdr.lsn);
+
+next_page_1:
+
+	if (wrapped) {
+		expected_seq += 1;
+		wrapped_file = 1;
+	}
+
+	curpage_off = nextpage_off;
+	kfree(page);
+	page = NULL;
+	reuse_page = 0;
+	goto next_page;
+
+check_tail:
+	if (tail_page) {
+		log->seq_num = expected_seq;
+		log->last_lsn = le64_to_cpu(tail_page->record_hdr.last_end_lsn);
+		log->ra->current_lsn = tail_page->record_hdr.last_end_lsn;
+		log->l_flags &= ~NTFSLOG_NO_LAST_LSN;
+
+		if (log->page_size -
+			    le16_to_cpu(
+				    tail_page->record_hdr.next_record_off) >=
+		    log->record_header_len) {
+			log->l_flags |= NTFSLOG_REUSE_TAIL;
+			log->next_page = curpage_off;
+		} else {
+			log->l_flags &= ~NTFSLOG_REUSE_TAIL;
+			log->next_page = nextpage_off;
+		}
+
+		if (wrapped)
+			log->l_flags |= NTFSLOG_WRAPPED;
+	}
+
+	/* Remember that the partial IO will start at the next page. */
+	second_off = nextpage_off;
+
+	/*
+	 * If the next page is the first page of the file then update
+	 * the sequence number for log records which begon the next page.
+	 */
+	if (wrapped)
+		expected_seq += 1;
+
+	/*
+	 * If we have a tail copy or are performing single page I/O we can
+	 * immediately look at the next page.
+	 */
+	if (replace_page || (log->ra->flags & RESTART_SINGLE_PAGE_IO)) {
+		page_cnt = 2;
+		page_pos = 1;
+		goto check_valid;
+	}
+
+	if (page_pos != page_cnt)
+		goto check_valid;
+	/*
+	 * If the next page causes us to wrap to the beginning of the log
+	 * file then we know which page to check next.
+	 */
+	if (wrapped) {
+		page_cnt = 2;
+		page_pos = 1;
+		goto check_valid;
+	}
+
+	cur_pos = 2;
+
+next_test_page:
+	kfree(tst_page);
+	tst_page = NULL;
+
+	/* Walk through the file, reading log pages. */
+	err = read_log_page(log, nextpage_off, &tst_page, &usa_error);
+
+	/*
+	 * If we get a USA error then assume that we correctly found
+	 * the end of the original transfer.
+	 */
+	if (usa_error)
+		goto file_is_valid;
+
+	/*
+	 * If we were able to read the page, we examine it to see if it
+	 * is the same or different Io block.
+	 */
+	if (err)
+		goto next_test_page_1;
+
+	if (le16_to_cpu(tst_page->page_pos) == cur_pos &&
+	    check_subseq_log_page(log, tst_page, nextpage_off, expected_seq)) {
+		page_cnt = le16_to_cpu(tst_page->page_count) + 1;
+		page_pos = le16_to_cpu(tst_page->page_pos);
+		goto check_valid;
+	} else {
+		goto file_is_valid;
+	}
+
+next_test_page_1:
+
+	nextpage_off = next_page_off(log, curpage_off);
+	wrapped = nextpage_off == log->first_page;
+
+	if (wrapped) {
+		expected_seq += 1;
+		page_cnt = 2;
+		page_pos = 1;
+	}
+
+	cur_pos += 1;
+	part_io_count += 1;
+	if (!wrapped)
+		goto next_test_page;
+
+check_valid:
+	/* Skip over the remaining pages this transfer. */
+	remain_pages = page_cnt - page_pos - 1;
+	part_io_count += remain_pages;
+
+	while (remain_pages--) {
+		nextpage_off = next_page_off(log, curpage_off);
+		wrapped = nextpage_off == log->first_page;
+
+		if (wrapped)
+			expected_seq += 1;
+	}
+
+	/* Call our routine to check this log page. */
+	kfree(tst_page);
+	tst_page = NULL;
+
+	err = read_log_page(log, nextpage_off, &tst_page, &usa_error);
+	if (!err && !usa_error &&
+	    check_subseq_log_page(log, tst_page, nextpage_off, expected_seq)) {
+		err = -EINVAL;
+		goto out;
+	}
+
+file_is_valid:
+
+	/* We have a valid file. */
+	if (page_off1 || tail_page) {
+		struct RECORD_PAGE_HDR *tmp_page;
+
+		if (sb_rdonly(log->ni->mi.sbi->sb)) {
+			err = -EROFS;
+			goto out;
+		}
+
+		if (page_off1) {
+			tmp_page = Add2Ptr(page_bufs, page_off1 - page_off);
+			tails -= (page_off1 - page_off) / log->page_size;
+			if (!tail_page)
+				tails -= 1;
+		} else {
+			tmp_page = tail_page;
+			tails = 1;
+		}
+
+		while (tails--) {
+			u64 off = hdr_file_off(log, tmp_page);
+
+			if (!page) {
+				page = kmalloc(log->page_size, GFP_NOFS);
+				if (!page)
+					return -ENOMEM;
+			}
+
+			/*
+			 * Correct page and copy the data from this page
+			 * into it and flush it to disk.
+			 */
+			memcpy(page, tmp_page, log->page_size);
+
+			/* Fill last flushed lsn value flush the page. */
+			if (log->major_ver < 2)
+				page->rhdr.lsn = page->record_hdr.last_end_lsn;
+			else
+				page->file_off = 0;
+
+			page->page_pos = page->page_count = cpu_to_le16(1);
+
+			ntfs_fix_pre_write(&page->rhdr, log->page_size);
+
+			err = ntfs_sb_write_run(log->ni->mi.sbi,
+						&log->ni->file.run, off, page,
+						log->page_size);
+
+			if (err)
+				goto out;
+
+			if (part_io_count && second_off == off) {
+				second_off += log->page_size;
+				part_io_count -= 1;
+			}
+
+			tmp_page = Add2Ptr(tmp_page, log->page_size);
+		}
+	}
+
+	if (part_io_count) {
+		if (sb_rdonly(log->ni->mi.sbi->sb)) {
+			err = -EROFS;
+			goto out;
+		}
+	}
+
+out:
+	kfree(second_tail);
+	kfree(first_tail);
+	kfree(page);
+	kfree(tst_page);
+	kfree(page_bufs);
+
+	return err;
+}
+
+/*
+ * read_log_rec_buf - Copy a log record from the file to a buffer.
+ *
+ * The log record may span several log pages and may even wrap the file.
+ */
+static int read_log_rec_buf(struct ntfs_log *log,
+			    const struct LFS_RECORD_HDR *rh, void *buffer)
+{
+	int err;
+	struct RECORD_PAGE_HDR *ph = NULL;
+	u64 lsn = le64_to_cpu(rh->this_lsn);
+	u32 vbo = lsn_to_vbo(log, lsn) & ~log->page_mask;
+	u32 off = lsn_to_page_off(log, lsn) + log->record_header_len;
+	u32 data_len = le32_to_cpu(rh->client_data_len);
+
+	/*
+	 * While there are more bytes to transfer,
+	 * we continue to attempt to perform the read.
+	 */
+	for (;;) {
+		bool usa_error;
+		u32 tail = log->page_size - off;
+
+		if (tail >= data_len)
+			tail = data_len;
+
+		data_len -= tail;
+
+		err = read_log_page(log, vbo, &ph, &usa_error);
+		if (err)
+			goto out;
+
+		/*
+		 * The last lsn on this page better be greater or equal
+		 * to the lsn we are copying.
+		 */
+		if (lsn > le64_to_cpu(ph->rhdr.lsn)) {
+			err = -EINVAL;
+			goto out;
+		}
+
+		memcpy(buffer, Add2Ptr(ph, off), tail);
+
+		/* If there are no more bytes to transfer, we exit the loop. */
+		if (!data_len) {
+			if (!is_log_record_end(ph) ||
+			    lsn > le64_to_cpu(ph->record_hdr.last_end_lsn)) {
+				err = -EINVAL;
+				goto out;
+			}
+			break;
+		}
+
+		if (ph->rhdr.lsn == ph->record_hdr.last_end_lsn ||
+		    lsn > le64_to_cpu(ph->rhdr.lsn)) {
+			err = -EINVAL;
+			goto out;
+		}
+
+		vbo = next_page_off(log, vbo);
+		off = log->data_off;
+
+		/*
+		 * Adjust our pointer the user's buffer to transfer
+		 * the next block to.
+		 */
+		buffer = Add2Ptr(buffer, tail);
+	}
+
+out:
+	kfree(ph);
+	return err;
+}
+
+static int read_rst_area(struct ntfs_log *log, struct NTFS_RESTART **rst_,
+			 u64 *lsn)
+{
+	int err;
+	struct LFS_RECORD_HDR *rh = NULL;
+	const struct CLIENT_REC *cr =
+		Add2Ptr(log->ra, le16_to_cpu(log->ra->client_off));
+	u64 lsnr, lsnc = le64_to_cpu(cr->restart_lsn);
+	u32 len;
+	struct NTFS_RESTART *rst;
+
+	*lsn = 0;
+	*rst_ = NULL;
+
+	/* If the client doesn't have a restart area, go ahead and exit now. */
+	if (!lsnc)
+		return 0;
+
+	err = read_log_page(log, lsn_to_vbo(log, lsnc),
+			    (struct RECORD_PAGE_HDR **)&rh, NULL);
+	if (err)
+		return err;
+
+	rst = NULL;
+	lsnr = le64_to_cpu(rh->this_lsn);
+
+	if (lsnc != lsnr) {
+		/* If the lsn values don't match, then the disk is corrupt. */
+		err = -EINVAL;
+		goto out;
+	}
+
+	*lsn = lsnr;
+	len = le32_to_cpu(rh->client_data_len);
+
+	if (!len) {
+		err = 0;
+		goto out;
+	}
+
+	if (len < sizeof(struct NTFS_RESTART)) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	rst = kmalloc(len, GFP_NOFS);
+	if (!rst) {
+		err = -ENOMEM;
+		goto out;
+	}
+
+	/* Copy the data into the 'rst' buffer. */
+	err = read_log_rec_buf(log, rh, rst);
+	if (err)
+		goto out;
+
+	*rst_ = rst;
+	rst = NULL;
+
+out:
+	kfree(rh);
+	kfree(rst);
+
+	return err;
+}
+
+static int find_log_rec(struct ntfs_log *log, u64 lsn, struct lcb *lcb)
+{
+	int err;
+	struct LFS_RECORD_HDR *rh = lcb->lrh;
+	u32 rec_len, len;
+
+	/* Read the record header for this lsn. */
+	if (!rh) {
+		err = read_log_page(log, lsn_to_vbo(log, lsn),
+				    (struct RECORD_PAGE_HDR **)&rh, NULL);
+
+		lcb->lrh = rh;
+		if (err)
+			return err;
+	}
+
+	/*
+	 * If the lsn the log record doesn't match the desired
+	 * lsn then the disk is corrupt.
+	 */
+	if (lsn != le64_to_cpu(rh->this_lsn))
+		return -EINVAL;
+
+	len = le32_to_cpu(rh->client_data_len);
+
+	/*
+	 * Check that the length field isn't greater than the total
+	 * available space the log file.
+	 */
+	rec_len = len + log->record_header_len;
+	if (rec_len >= log->total_avail)
+		return -EINVAL;
+
+	/*
+	 * If the entire log record is on this log page,
+	 * put a pointer to the log record the context block.
+	 */
+	if (rh->flags & LOG_RECORD_MULTI_PAGE) {
+		void *lr = kmalloc(len, GFP_NOFS);
+
+		if (!lr)
+			return -ENOMEM;
+
+		lcb->log_rec = lr;
+		lcb->alloc = true;
+
+		/* Copy the data into the buffer returned. */
+		err = read_log_rec_buf(log, rh, lr);
+		if (err)
+			return err;
+	} else {
+		/* If beyond the end of the current page -> an error. */
+		u32 page_off = lsn_to_page_off(log, lsn);
+
+		if (page_off + len + log->record_header_len > log->page_size)
+			return -EINVAL;
+
+		lcb->log_rec = Add2Ptr(rh, sizeof(struct LFS_RECORD_HDR));
+		lcb->alloc = false;
+	}
+
+	return 0;
+}
+
+/*
+ * read_log_rec_lcb - Init the query operation.
+ */
+static int read_log_rec_lcb(struct ntfs_log *log, u64 lsn, u32 ctx_mode,
+			    struct lcb **lcb_)
+{
+	int err;
+	const struct CLIENT_REC *cr;
+	struct lcb *lcb;
+
+	switch (ctx_mode) {
+	case lcb_ctx_undo_next:
+	case lcb_ctx_prev:
+	case lcb_ctx_next:
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	/* Check that the given lsn is the legal range for this client. */
+	cr = Add2Ptr(log->ra, le16_to_cpu(log->ra->client_off));
+
+	if (!verify_client_lsn(log, cr, lsn))
+		return -EINVAL;
+
+	lcb = kzalloc(sizeof(struct lcb), GFP_NOFS);
+	if (!lcb)
+		return -ENOMEM;
+	lcb->client = log->client_id;
+	lcb->ctx_mode = ctx_mode;
+
+	/* Find the log record indicated by the given lsn. */
+	err = find_log_rec(log, lsn, lcb);
+	if (err)
+		goto out;
+
+	*lcb_ = lcb;
+	return 0;
+
+out:
+	lcb_put(lcb);
+	*lcb_ = NULL;
+	return err;
+}
+
+/*
+ * find_client_next_lsn
+ *
+ * Attempt to find the next lsn to return to a client based on the context mode.
+ */
+static int find_client_next_lsn(struct ntfs_log *log, struct lcb *lcb, u64 *lsn)
+{
+	int err;
+	u64 next_lsn;
+	struct LFS_RECORD_HDR *hdr;
+
+	hdr = lcb->lrh;
+	*lsn = 0;
+
+	if (lcb_ctx_next != lcb->ctx_mode)
+		goto check_undo_next;
+
+	/* Loop as long as another lsn can be found. */
+	for (;;) {
+		u64 current_lsn;
+
+		err = next_log_lsn(log, hdr, &current_lsn);
+		if (err)
+			goto out;
+
+		if (!current_lsn)
+			break;
+
+		if (hdr != lcb->lrh)
+			kfree(hdr);
+
+		hdr = NULL;
+		err = read_log_page(log, lsn_to_vbo(log, current_lsn),
+				    (struct RECORD_PAGE_HDR **)&hdr, NULL);
+		if (err)
+			goto out;
+
+		if (memcmp(&hdr->client, &lcb->client,
+			   sizeof(struct CLIENT_ID))) {
+			/*err = -EINVAL; */
+		} else if (LfsClientRecord == hdr->record_type) {
+			kfree(lcb->lrh);
+			lcb->lrh = hdr;
+			*lsn = current_lsn;
+			return 0;
+		}
+	}
+
+out:
+	if (hdr != lcb->lrh)
+		kfree(hdr);
+	return err;
+
+check_undo_next:
+	if (lcb_ctx_undo_next == lcb->ctx_mode)
+		next_lsn = le64_to_cpu(hdr->client_undo_next_lsn);
+	else if (lcb_ctx_prev == lcb->ctx_mode)
+		next_lsn = le64_to_cpu(hdr->client_prev_lsn);
+	else
+		return 0;
+
+	if (!next_lsn)
+		return 0;
+
+	if (!verify_client_lsn(
+		    log, Add2Ptr(log->ra, le16_to_cpu(log->ra->client_off)),
+		    next_lsn))
+		return 0;
+
+	hdr = NULL;
+	err = read_log_page(log, lsn_to_vbo(log, next_lsn),
+			    (struct RECORD_PAGE_HDR **)&hdr, NULL);
+	if (err)
+		return err;
+	kfree(lcb->lrh);
+	lcb->lrh = hdr;
+
+	*lsn = next_lsn;
+
+	return 0;
+}
+
+static int read_next_log_rec(struct ntfs_log *log, struct lcb *lcb, u64 *lsn)
+{
+	int err;
+
+	err = find_client_next_lsn(log, lcb, lsn);
+	if (err)
+		return err;
+
+	if (!*lsn)
+		return 0;
+
+	if (lcb->alloc)
+		kfree(lcb->log_rec);
+
+	lcb->log_rec = NULL;
+	lcb->alloc = false;
+	kfree(lcb->lrh);
+	lcb->lrh = NULL;
+
+	return find_log_rec(log, *lsn, lcb);
+}
+
+static inline bool check_index_header(const struct INDEX_HDR *hdr, size_t bytes)
+{
+	__le16 mask;
+	u32 min_de, de_off, used, total;
+	const struct NTFS_DE *e;
+
+	if (hdr_has_subnode(hdr)) {
+		min_de = sizeof(struct NTFS_DE) + sizeof(u64);
+		mask = NTFS_IE_HAS_SUBNODES;
+	} else {
+		min_de = sizeof(struct NTFS_DE);
+		mask = 0;
+	}
+
+	de_off = le32_to_cpu(hdr->de_off);
+	used = le32_to_cpu(hdr->used);
+	total = le32_to_cpu(hdr->total);
+
+	if (de_off > bytes - min_de || used > bytes || total > bytes ||
+	    de_off + min_de > used || used > total) {
+		return false;
+	}
+
+	e = Add2Ptr(hdr, de_off);
+	for (;;) {
+		u16 esize = le16_to_cpu(e->size);
+		struct NTFS_DE *next = Add2Ptr(e, esize);
+
+		if (esize < min_de || PtrOffset(hdr, next) > used ||
+		    (e->flags & NTFS_IE_HAS_SUBNODES) != mask) {
+			return false;
+		}
+
+		if (de_is_last(e))
+			break;
+
+		e = next;
+	}
+
+	return true;
+}
+
+static inline bool check_index_buffer(const struct INDEX_BUFFER *ib, u32 bytes)
+{
+	u16 fo;
+	const struct NTFS_RECORD_HEADER *r = &ib->rhdr;
+
+	if (r->sign != NTFS_INDX_SIGNATURE)
+		return false;
+
+	fo = (SECTOR_SIZE - ((bytes >> SECTOR_SHIFT) + 1) * sizeof(short));
+
+	if (le16_to_cpu(r->fix_off) > fo)
+		return false;
+
+	if ((le16_to_cpu(r->fix_num) - 1) * SECTOR_SIZE != bytes)
+		return false;
+
+	return check_index_header(&ib->ihdr,
+				  bytes - offsetof(struct INDEX_BUFFER, ihdr));
+}
+
+static inline bool check_index_root(const struct ATTRIB *attr,
+				    struct ntfs_sb_info *sbi)
+{
+	bool ret;
+	const struct INDEX_ROOT *root = resident_data(attr);
+	u8 index_bits = le32_to_cpu(root->index_block_size) >= sbi->cluster_size
+				? sbi->cluster_bits
+				: SECTOR_SHIFT;
+	u8 block_clst = root->index_block_clst;
+
+	if (le32_to_cpu(attr->res.data_size) < sizeof(struct INDEX_ROOT) ||
+	    (root->type != ATTR_NAME && root->type != ATTR_ZERO) ||
+	    (root->type == ATTR_NAME &&
+	     root->rule != NTFS_COLLATION_TYPE_FILENAME) ||
+	    (le32_to_cpu(root->index_block_size) !=
+	     (block_clst << index_bits)) ||
+	    (block_clst != 1 && block_clst != 2 && block_clst != 4 &&
+	     block_clst != 8 && block_clst != 0x10 && block_clst != 0x20 &&
+	     block_clst != 0x40 && block_clst != 0x80)) {
+		return false;
+	}
+
+	ret = check_index_header(&root->ihdr,
+				 le32_to_cpu(attr->res.data_size) -
+					 offsetof(struct INDEX_ROOT, ihdr));
+	return ret;
+}
+
+static inline bool check_attr(const struct MFT_REC *rec,
+			      const struct ATTRIB *attr,
+			      struct ntfs_sb_info *sbi)
+{
+	u32 asize = le32_to_cpu(attr->size);
+	u32 rsize = 0;
+	u64 dsize, svcn, evcn;
+	u16 run_off;
+
+	/* Check the fixed part of the attribute record header. */
+	if (asize >= sbi->record_size ||
+	    asize + PtrOffset(rec, attr) >= sbi->record_size ||
+	    (attr->name_len &&
+	     le16_to_cpu(attr->name_off) + attr->name_len * sizeof(short) >
+		     asize)) {
+		return false;
+	}
+
+	/* Check the attribute fields. */
+	switch (attr->non_res) {
+	case 0:
+		rsize = le32_to_cpu(attr->res.data_size);
+		if (rsize >= asize ||
+		    le16_to_cpu(attr->res.data_off) + rsize > asize) {
+			return false;
+		}
+		break;
+
+	case 1:
+		dsize = le64_to_cpu(attr->nres.data_size);
+		svcn = le64_to_cpu(attr->nres.svcn);
+		evcn = le64_to_cpu(attr->nres.evcn);
+		run_off = le16_to_cpu(attr->nres.run_off);
+
+		if (svcn > evcn + 1 || run_off >= asize ||
+		    le64_to_cpu(attr->nres.valid_size) > dsize ||
+		    dsize > le64_to_cpu(attr->nres.alloc_size)) {
+			return false;
+		}
+
+		if (run_unpack(NULL, sbi, 0, svcn, evcn, svcn,
+			       Add2Ptr(attr, run_off), asize - run_off) < 0) {
+			return false;
+		}
+
+		return true;
+
+	default:
+		return false;
+	}
+
+	switch (attr->type) {
+	case ATTR_NAME:
+		if (fname_full_size(Add2Ptr(
+			    attr, le16_to_cpu(attr->res.data_off))) > asize) {
+			return false;
+		}
+		break;
+
+	case ATTR_ROOT:
+		return check_index_root(attr, sbi);
+
+	case ATTR_STD:
+		if (rsize < sizeof(struct ATTR_STD_INFO5) &&
+		    rsize != sizeof(struct ATTR_STD_INFO)) {
+			return false;
+		}
+		break;
+
+	case ATTR_LIST:
+	case ATTR_ID:
+	case ATTR_SECURE:
+	case ATTR_LABEL:
+	case ATTR_VOL_INFO:
+	case ATTR_DATA:
+	case ATTR_ALLOC:
+	case ATTR_BITMAP:
+	case ATTR_REPARSE:
+	case ATTR_EA_INFO:
+	case ATTR_EA:
+	case ATTR_PROPERTYSET:
+	case ATTR_LOGGED_UTILITY_STREAM:
+		break;
+
+	default:
+		return false;
+	}
+
+	return true;
+}
+
+static inline bool check_file_record(const struct MFT_REC *rec,
+				     const struct MFT_REC *rec2,
+				     struct ntfs_sb_info *sbi)
+{
+	const struct ATTRIB *attr;
+	u16 fo = le16_to_cpu(rec->rhdr.fix_off);
+	u16 fn = le16_to_cpu(rec->rhdr.fix_num);
+	u16 ao = le16_to_cpu(rec->attr_off);
+	u32 rs = sbi->record_size;
+
+	/* Check the file record header for consistency. */
+	if (rec->rhdr.sign != NTFS_FILE_SIGNATURE ||
+	    fo > (SECTOR_SIZE - ((rs >> SECTOR_SHIFT) + 1) * sizeof(short)) ||
+	    (fn - 1) * SECTOR_SIZE != rs || ao < MFTRECORD_FIXUP_OFFSET_1 ||
+	    ao > sbi->record_size - SIZEOF_RESIDENT || !is_rec_inuse(rec) ||
+	    le32_to_cpu(rec->total) != rs) {
+		return false;
+	}
+
+	/* Loop to check all of the attributes. */
+	for (attr = Add2Ptr(rec, ao); attr->type != ATTR_END;
+	     attr = Add2Ptr(attr, le32_to_cpu(attr->size))) {
+		if (check_attr(rec, attr, sbi))
+			continue;
+		return false;
+	}
+
+	return true;
+}
+
+static inline int check_lsn(const struct NTFS_RECORD_HEADER *hdr,
+			    const u64 *rlsn)
+{
+	u64 lsn;
+
+	if (!rlsn)
+		return true;
+
+	lsn = le64_to_cpu(hdr->lsn);
+
+	if (hdr->sign == NTFS_HOLE_SIGNATURE)
+		return false;
+
+	if (*rlsn > lsn)
+		return true;
+
+	return false;
+}
+
+static inline bool check_if_attr(const struct MFT_REC *rec,
+				 const struct LOG_REC_HDR *lrh)
+{
+	u16 ro = le16_to_cpu(lrh->record_off);
+	u16 o = le16_to_cpu(rec->attr_off);
+	const struct ATTRIB *attr = Add2Ptr(rec, o);
+
+	while (o < ro) {
+		u32 asize;
+
+		if (attr->type == ATTR_END)
+			break;
+
+		asize = le32_to_cpu(attr->size);
+		if (!asize)
+			break;
+
+		o += asize;
+		attr = Add2Ptr(attr, asize);
+	}
+
+	return o == ro;
+}
+
+static inline bool check_if_index_root(const struct MFT_REC *rec,
+				       const struct LOG_REC_HDR *lrh)
+{
+	u16 ro = le16_to_cpu(lrh->record_off);
+	u16 o = le16_to_cpu(rec->attr_off);
+	const struct ATTRIB *attr = Add2Ptr(rec, o);
+
+	while (o < ro) {
+		u32 asize;
+
+		if (attr->type == ATTR_END)
+			break;
+
+		asize = le32_to_cpu(attr->size);
+		if (!asize)
+			break;
+
+		o += asize;
+		attr = Add2Ptr(attr, asize);
+	}
+
+	return o == ro && attr->type == ATTR_ROOT;
+}
+
+static inline bool check_if_root_index(const struct ATTRIB *attr,
+				       const struct INDEX_HDR *hdr,
+				       const struct LOG_REC_HDR *lrh)
+{
+	u16 ao = le16_to_cpu(lrh->attr_off);
+	u32 de_off = le32_to_cpu(hdr->de_off);
+	u32 o = PtrOffset(attr, hdr) + de_off;
+	const struct NTFS_DE *e = Add2Ptr(hdr, de_off);
+	u32 asize = le32_to_cpu(attr->size);
+
+	while (o < ao) {
+		u16 esize;
+
+		if (o >= asize)
+			break;
+
+		esize = le16_to_cpu(e->size);
+		if (!esize)
+			break;
+
+		o += esize;
+		e = Add2Ptr(e, esize);
+	}
+
+	return o == ao;
+}
+
+static inline bool check_if_alloc_index(const struct INDEX_HDR *hdr,
+					u32 attr_off)
+{
+	u32 de_off = le32_to_cpu(hdr->de_off);
+	u32 o = offsetof(struct INDEX_BUFFER, ihdr) + de_off;
+	const struct NTFS_DE *e = Add2Ptr(hdr, de_off);
+	u32 used = le32_to_cpu(hdr->used);
+
+	while (o < attr_off) {
+		u16 esize;
+
+		if (de_off >= used)
+			break;
+
+		esize = le16_to_cpu(e->size);
+		if (!esize)
+			break;
+
+		o += esize;
+		de_off += esize;
+		e = Add2Ptr(e, esize);
+	}
+
+	return o == attr_off;
+}
+
+static inline void change_attr_size(struct MFT_REC *rec, struct ATTRIB *attr,
+				    u32 nsize)
+{
+	u32 asize = le32_to_cpu(attr->size);
+	int dsize = nsize - asize;
+	u8 *next = Add2Ptr(attr, asize);
+	u32 used = le32_to_cpu(rec->used);
+
+	memmove(Add2Ptr(attr, nsize), next, used - PtrOffset(rec, next));
+
+	rec->used = cpu_to_le32(used + dsize);
+	attr->size = cpu_to_le32(nsize);
+}
+
+struct OpenAttr {
+	struct ATTRIB *attr;
+	struct runs_tree *run1;
+	struct runs_tree run0;
+	struct ntfs_inode *ni;
+	// CLST rno;
+};
+
+/*
+ * cmp_type_and_name
+ *
+ * Return: 0 if 'attr' has the same type and name.
+ */
+static inline int cmp_type_and_name(const struct ATTRIB *a1,
+				    const struct ATTRIB *a2)
+{
+	return a1->type != a2->type || a1->name_len != a2->name_len ||
+	       (a1->name_len && memcmp(attr_name(a1), attr_name(a2),
+				       a1->name_len * sizeof(short)));
+}
+
+static struct OpenAttr *find_loaded_attr(struct ntfs_log *log,
+					 const struct ATTRIB *attr, CLST rno)
+{
+	struct OPEN_ATTR_ENRTY *oe = NULL;
+
+	while ((oe = enum_rstbl(log->open_attr_tbl, oe))) {
+		struct OpenAttr *op_attr;
+
+		if (ino_get(&oe->ref) != rno)
+			continue;
+
+		op_attr = (struct OpenAttr *)oe->ptr;
+		if (!cmp_type_and_name(op_attr->attr, attr))
+			return op_attr;
+	}
+	return NULL;
+}
+
+static struct ATTRIB *attr_create_nonres_log(struct ntfs_sb_info *sbi,
+					     enum ATTR_TYPE type, u64 size,
+					     const u16 *name, size_t name_len,
+					     __le16 flags)
+{
+	struct ATTRIB *attr;
+	u32 name_size = ALIGN(name_len * sizeof(short), 8);
+	bool is_ext = flags & (ATTR_FLAG_COMPRESSED | ATTR_FLAG_SPARSED);
+	u32 asize = name_size +
+		    (is_ext ? SIZEOF_NONRESIDENT_EX : SIZEOF_NONRESIDENT);
+
+	attr = kzalloc(asize, GFP_NOFS);
+	if (!attr)
+		return NULL;
+
+	attr->type = type;
+	attr->size = cpu_to_le32(asize);
+	attr->flags = flags;
+	attr->non_res = 1;
+	attr->name_len = name_len;
+
+	attr->nres.evcn = cpu_to_le64((u64)bytes_to_cluster(sbi, size) - 1);
+	attr->nres.alloc_size = cpu_to_le64(ntfs_up_cluster(sbi, size));
+	attr->nres.data_size = cpu_to_le64(size);
+	attr->nres.valid_size = attr->nres.data_size;
+	if (is_ext) {
+		attr->name_off = SIZEOF_NONRESIDENT_EX_LE;
+		if (is_attr_compressed(attr))
+			attr->nres.c_unit = COMPRESSION_UNIT;
+
+		attr->nres.run_off =
+			cpu_to_le16(SIZEOF_NONRESIDENT_EX + name_size);
+		memcpy(Add2Ptr(attr, SIZEOF_NONRESIDENT_EX), name,
+		       name_len * sizeof(short));
+	} else {
+		attr->name_off = SIZEOF_NONRESIDENT_LE;
+		attr->nres.run_off =
+			cpu_to_le16(SIZEOF_NONRESIDENT + name_size);
+		memcpy(Add2Ptr(attr, SIZEOF_NONRESIDENT), name,
+		       name_len * sizeof(short));
+	}
+
+	return attr;
+}
+
+/*
+ * do_action - Common routine for the Redo and Undo Passes.
+ * @rlsn: If it is NULL then undo.
+ */
+static int do_action(struct ntfs_log *log, struct OPEN_ATTR_ENRTY *oe,
+		     const struct LOG_REC_HDR *lrh, u32 op, void *data,
+		     u32 dlen, u32 rec_len, const u64 *rlsn)
+{
+	int err = 0;
+	struct ntfs_sb_info *sbi = log->ni->mi.sbi;
+	struct inode *inode = NULL, *inode_parent;
+	struct mft_inode *mi = NULL, *mi2_child = NULL;
+	CLST rno = 0, rno_base = 0;
+	struct INDEX_BUFFER *ib = NULL;
+	struct MFT_REC *rec = NULL;
+	struct ATTRIB *attr = NULL, *attr2;
+	struct INDEX_HDR *hdr;
+	struct INDEX_ROOT *root;
+	struct NTFS_DE *e, *e1, *e2;
+	struct NEW_ATTRIBUTE_SIZES *new_sz;
+	struct ATTR_FILE_NAME *fname;
+	struct OpenAttr *oa, *oa2;
+	u32 nsize, t32, asize, used, esize, bmp_off, bmp_bits;
+	u16 id, id2;
+	u32 record_size = sbi->record_size;
+	u64 t64;
+	u16 roff = le16_to_cpu(lrh->record_off);
+	u16 aoff = le16_to_cpu(lrh->attr_off);
+	u64 lco = 0;
+	u64 cbo = (u64)le16_to_cpu(lrh->cluster_off) << SECTOR_SHIFT;
+	u64 tvo = le64_to_cpu(lrh->target_vcn) << sbi->cluster_bits;
+	u64 vbo = cbo + tvo;
+	void *buffer_le = NULL;
+	u32 bytes = 0;
+	bool a_dirty = false;
+	u16 data_off;
+
+	oa = oe->ptr;
+
+	/* Big switch to prepare. */
+	switch (op) {
+	/* ============================================================
+	 * Process MFT records, as described by the current log record.
+	 * ============================================================
+	 */
+	case InitializeFileRecordSegment:
+	case DeallocateFileRecordSegment:
+	case WriteEndOfFileRecordSegment:
+	case CreateAttribute:
+	case DeleteAttribute:
+	case UpdateResidentValue:
+	case UpdateMappingPairs:
+	case SetNewAttributeSizes:
+	case AddIndexEntryRoot:
+	case DeleteIndexEntryRoot:
+	case SetIndexEntryVcnRoot:
+	case UpdateFileNameRoot:
+	case UpdateRecordDataRoot:
+	case ZeroEndOfFileRecord:
+		rno = vbo >> sbi->record_bits;
+		inode = ilookup(sbi->sb, rno);
+		if (inode) {
+			mi = &ntfs_i(inode)->mi;
+		} else if (op == InitializeFileRecordSegment) {
+			mi = kzalloc(sizeof(struct mft_inode), GFP_NOFS);
+			if (!mi)
+				return -ENOMEM;
+			err = mi_format_new(mi, sbi, rno, 0, false);
+			if (err)
+				goto out;
+		} else {
+			/* Read from disk. */
+			err = mi_get(sbi, rno, &mi);
+			if (err)
+				return err;
+		}
+		rec = mi->mrec;
+
+		if (op == DeallocateFileRecordSegment)
+			goto skip_load_parent;
+
+		if (InitializeFileRecordSegment != op) {
+			if (rec->rhdr.sign == NTFS_BAAD_SIGNATURE)
+				goto dirty_vol;
+			if (!check_lsn(&rec->rhdr, rlsn))
+				goto out;
+			if (!check_file_record(rec, NULL, sbi))
+				goto dirty_vol;
+			attr = Add2Ptr(rec, roff);
+		}
+
+		if (is_rec_base(rec) || InitializeFileRecordSegment == op) {
+			rno_base = rno;
+			goto skip_load_parent;
+		}
+
+		rno_base = ino_get(&rec->parent_ref);
+		inode_parent = ntfs_iget5(sbi->sb, &rec->parent_ref, NULL);
+		if (IS_ERR(inode_parent))
+			goto skip_load_parent;
+
+		if (is_bad_inode(inode_parent)) {
+			iput(inode_parent);
+			goto skip_load_parent;
+		}
+
+		if (ni_load_mi_ex(ntfs_i(inode_parent), rno, &mi2_child)) {
+			iput(inode_parent);
+		} else {
+			if (mi2_child->mrec != mi->mrec)
+				memcpy(mi2_child->mrec, mi->mrec,
+				       sbi->record_size);
+
+			if (inode)
+				iput(inode);
+			else if (mi)
+				mi_put(mi);
+
+			inode = inode_parent;
+			mi = mi2_child;
+			rec = mi2_child->mrec;
+			attr = Add2Ptr(rec, roff);
+		}
+
+skip_load_parent:
+		inode_parent = NULL;
+		break;
+
+	/*
+	 * Process attributes, as described by the current log record.
+	 */
+	case UpdateNonresidentValue:
+	case AddIndexEntryAllocation:
+	case DeleteIndexEntryAllocation:
+	case WriteEndOfIndexBuffer:
+	case SetIndexEntryVcnAllocation:
+	case UpdateFileNameAllocation:
+	case SetBitsInNonresidentBitMap:
+	case ClearBitsInNonresidentBitMap:
+	case UpdateRecordDataAllocation:
+		attr = oa->attr;
+		bytes = UpdateNonresidentValue == op ? dlen : 0;
+		lco = (u64)le16_to_cpu(lrh->lcns_follow) << sbi->cluster_bits;
+
+		if (attr->type == ATTR_ALLOC) {
+			t32 = le32_to_cpu(oe->bytes_per_index);
+			if (bytes < t32)
+				bytes = t32;
+		}
+
+		if (!bytes)
+			bytes = lco - cbo;
+
+		bytes += roff;
+		if (attr->type == ATTR_ALLOC)
+			bytes = (bytes + 511) & ~511; // align
+
+		buffer_le = kmalloc(bytes, GFP_NOFS);
+		if (!buffer_le)
+			return -ENOMEM;
+
+		err = ntfs_read_run_nb(sbi, oa->run1, vbo, buffer_le, bytes,
+				       NULL);
+		if (err)
+			goto out;
+
+		if (attr->type == ATTR_ALLOC && *(int *)buffer_le)
+			ntfs_fix_post_read(buffer_le, bytes, false);
+		break;
+
+	default:
+		WARN_ON(1);
+	}
+
+	/* Big switch to do operation. */
+	switch (op) {
+	case InitializeFileRecordSegment:
+		if (roff + dlen > record_size)
+			goto dirty_vol;
+
+		memcpy(Add2Ptr(rec, roff), data, dlen);
+		mi->dirty = true;
+		break;
+
+	case DeallocateFileRecordSegment:
+		clear_rec_inuse(rec);
+		le16_add_cpu(&rec->seq, 1);
+		mi->dirty = true;
+		break;
+
+	case WriteEndOfFileRecordSegment:
+		attr2 = (struct ATTRIB *)data;
+		if (!check_if_attr(rec, lrh) || roff + dlen > record_size)
+			goto dirty_vol;
+
+		memmove(attr, attr2, dlen);
+		rec->used = cpu_to_le32(ALIGN(roff + dlen, 8));
+
+		mi->dirty = true;
+		break;
+
+	case CreateAttribute:
+		attr2 = (struct ATTRIB *)data;
+		asize = le32_to_cpu(attr2->size);
+		used = le32_to_cpu(rec->used);
+
+		if (!check_if_attr(rec, lrh) || dlen < SIZEOF_RESIDENT ||
+		    !IS_ALIGNED(asize, 8) ||
+		    Add2Ptr(attr2, asize) > Add2Ptr(lrh, rec_len) ||
+		    dlen > record_size - used) {
+			goto dirty_vol;
+		}
+
+		memmove(Add2Ptr(attr, asize), attr, used - roff);
+		memcpy(attr, attr2, asize);
+
+		rec->used = cpu_to_le32(used + asize);
+		id = le16_to_cpu(rec->next_attr_id);
+		id2 = le16_to_cpu(attr2->id);
+		if (id <= id2)
+			rec->next_attr_id = cpu_to_le16(id2 + 1);
+		if (is_attr_indexed(attr))
+			le16_add_cpu(&rec->hard_links, 1);
+
+		oa2 = find_loaded_attr(log, attr, rno_base);
+		if (oa2) {
+			void *p2 = kmemdup(attr, le32_to_cpu(attr->size),
+					   GFP_NOFS);
+			if (p2) {
+				// run_close(oa2->run1);
+				kfree(oa2->attr);
+				oa2->attr = p2;
+			}
+		}
+
+		mi->dirty = true;
+		break;
+
+	case DeleteAttribute:
+		asize = le32_to_cpu(attr->size);
+		used = le32_to_cpu(rec->used);
+
+		if (!check_if_attr(rec, lrh))
+			goto dirty_vol;
+
+		rec->used = cpu_to_le32(used - asize);
+		if (is_attr_indexed(attr))
+			le16_add_cpu(&rec->hard_links, -1);
+
+		memmove(attr, Add2Ptr(attr, asize), used - asize - roff);
+
+		mi->dirty = true;
+		break;
+
+	case UpdateResidentValue:
+		nsize = aoff + dlen;
+
+		if (!check_if_attr(rec, lrh))
+			goto dirty_vol;
+
+		asize = le32_to_cpu(attr->size);
+		used = le32_to_cpu(rec->used);
+
+		if (lrh->redo_len == lrh->undo_len) {
+			if (nsize > asize)
+				goto dirty_vol;
+			goto move_data;
+		}
+
+		if (nsize > asize && nsize - asize > record_size - used)
+			goto dirty_vol;
+
+		nsize = ALIGN(nsize, 8);
+		data_off = le16_to_cpu(attr->res.data_off);
+
+		if (nsize < asize) {
+			memmove(Add2Ptr(attr, aoff), data, dlen);
+			data = NULL; // To skip below memmove().
+		}
+
+		memmove(Add2Ptr(attr, nsize), Add2Ptr(attr, asize),
+			used - le16_to_cpu(lrh->record_off) - asize);
+
+		rec->used = cpu_to_le32(used + nsize - asize);
+		attr->size = cpu_to_le32(nsize);
+		attr->res.data_size = cpu_to_le32(aoff + dlen - data_off);
+
+move_data:
+		if (data)
+			memmove(Add2Ptr(attr, aoff), data, dlen);
+
+		oa2 = find_loaded_attr(log, attr, rno_base);
+		if (oa2) {
+			void *p2 = kmemdup(attr, le32_to_cpu(attr->size),
+					   GFP_NOFS);
+			if (p2) {
+				// run_close(&oa2->run0);
+				oa2->run1 = &oa2->run0;
+				kfree(oa2->attr);
+				oa2->attr = p2;
+			}
+		}
+
+		mi->dirty = true;
+		break;
+
+	case UpdateMappingPairs:
+		nsize = aoff + dlen;
+		asize = le32_to_cpu(attr->size);
+		used = le32_to_cpu(rec->used);
+
+		if (!check_if_attr(rec, lrh) || !attr->non_res ||
+		    aoff < le16_to_cpu(attr->nres.run_off) || aoff > asize ||
+		    (nsize > asize && nsize - asize > record_size - used)) {
+			goto dirty_vol;
+		}
+
+		nsize = ALIGN(nsize, 8);
+
+		memmove(Add2Ptr(attr, nsize), Add2Ptr(attr, asize),
+			used - le16_to_cpu(lrh->record_off) - asize);
+		rec->used = cpu_to_le32(used + nsize - asize);
+		attr->size = cpu_to_le32(nsize);
+		memmove(Add2Ptr(attr, aoff), data, dlen);
+
+		if (run_get_highest_vcn(le64_to_cpu(attr->nres.svcn),
+					attr_run(attr), &t64)) {
+			goto dirty_vol;
+		}
+
+		attr->nres.evcn = cpu_to_le64(t64);
+		oa2 = find_loaded_attr(log, attr, rno_base);
+		if (oa2 && oa2->attr->non_res)
+			oa2->attr->nres.evcn = attr->nres.evcn;
+
+		mi->dirty = true;
+		break;
+
+	case SetNewAttributeSizes:
+		new_sz = data;
+		if (!check_if_attr(rec, lrh) || !attr->non_res)
+			goto dirty_vol;
+
+		attr->nres.alloc_size = new_sz->alloc_size;
+		attr->nres.data_size = new_sz->data_size;
+		attr->nres.valid_size = new_sz->valid_size;
+
+		if (dlen >= sizeof(struct NEW_ATTRIBUTE_SIZES))
+			attr->nres.total_size = new_sz->total_size;
+
+		oa2 = find_loaded_attr(log, attr, rno_base);
+		if (oa2) {
+			void *p2 = kmemdup(attr, le32_to_cpu(attr->size),
+					   GFP_NOFS);
+			if (p2) {
+				kfree(oa2->attr);
+				oa2->attr = p2;
+			}
+		}
+		mi->dirty = true;
+		break;
+
+	case AddIndexEntryRoot:
+		e = (struct NTFS_DE *)data;
+		esize = le16_to_cpu(e->size);
+		root = resident_data(attr);
+		hdr = &root->ihdr;
+		used = le32_to_cpu(hdr->used);
+
+		if (!check_if_index_root(rec, lrh) ||
+		    !check_if_root_index(attr, hdr, lrh) ||
+		    Add2Ptr(data, esize) > Add2Ptr(lrh, rec_len) ||
+		    esize > le32_to_cpu(rec->total) - le32_to_cpu(rec->used)) {
+			goto dirty_vol;
+		}
+
+		e1 = Add2Ptr(attr, le16_to_cpu(lrh->attr_off));
+
+		change_attr_size(rec, attr, le32_to_cpu(attr->size) + esize);
+
+		memmove(Add2Ptr(e1, esize), e1,
+			PtrOffset(e1, Add2Ptr(hdr, used)));
+		memmove(e1, e, esize);
+
+		le32_add_cpu(&attr->res.data_size, esize);
+		hdr->used = cpu_to_le32(used + esize);
+		le32_add_cpu(&hdr->total, esize);
+
+		mi->dirty = true;
+		break;
+
+	case DeleteIndexEntryRoot:
+		root = resident_data(attr);
+		hdr = &root->ihdr;
+		used = le32_to_cpu(hdr->used);
+
+		if (!check_if_index_root(rec, lrh) ||
+		    !check_if_root_index(attr, hdr, lrh)) {
+			goto dirty_vol;
+		}
+
+		e1 = Add2Ptr(attr, le16_to_cpu(lrh->attr_off));
+		esize = le16_to_cpu(e1->size);
+		e2 = Add2Ptr(e1, esize);
+
+		memmove(e1, e2, PtrOffset(e2, Add2Ptr(hdr, used)));
+
+		le32_sub_cpu(&attr->res.data_size, esize);
+		hdr->used = cpu_to_le32(used - esize);
+		le32_sub_cpu(&hdr->total, esize);
+
+		change_attr_size(rec, attr, le32_to_cpu(attr->size) - esize);
+
+		mi->dirty = true;
+		break;
+
+	case SetIndexEntryVcnRoot:
+		root = resident_data(attr);
+		hdr = &root->ihdr;
+
+		if (!check_if_index_root(rec, lrh) ||
+		    !check_if_root_index(attr, hdr, lrh)) {
+			goto dirty_vol;
+		}
+
+		e = Add2Ptr(attr, le16_to_cpu(lrh->attr_off));
+
+		de_set_vbn_le(e, *(__le64 *)data);
+		mi->dirty = true;
+		break;
+
+	case UpdateFileNameRoot:
+		root = resident_data(attr);
+		hdr = &root->ihdr;
+
+		if (!check_if_index_root(rec, lrh) ||
+		    !check_if_root_index(attr, hdr, lrh)) {
+			goto dirty_vol;
+		}
+
+		e = Add2Ptr(attr, le16_to_cpu(lrh->attr_off));
+		fname = (struct ATTR_FILE_NAME *)(e + 1);
+		memmove(&fname->dup, data, sizeof(fname->dup)); //
+		mi->dirty = true;
+		break;
+
+	case UpdateRecordDataRoot:
+		root = resident_data(attr);
+		hdr = &root->ihdr;
+
+		if (!check_if_index_root(rec, lrh) ||
+		    !check_if_root_index(attr, hdr, lrh)) {
+			goto dirty_vol;
+		}
+
+		e = Add2Ptr(attr, le16_to_cpu(lrh->attr_off));
+
+		memmove(Add2Ptr(e, le16_to_cpu(e->view.data_off)), data, dlen);
+
+		mi->dirty = true;
+		break;
+
+	case ZeroEndOfFileRecord:
+		if (roff + dlen > record_size)
+			goto dirty_vol;
+
+		memset(attr, 0, dlen);
+		mi->dirty = true;
+		break;
+
+	case UpdateNonresidentValue:
+		if (lco < cbo + roff + dlen)
+			goto dirty_vol;
+
+		memcpy(Add2Ptr(buffer_le, roff), data, dlen);
+
+		a_dirty = true;
+		if (attr->type == ATTR_ALLOC)
+			ntfs_fix_pre_write(buffer_le, bytes);
+		break;
+
+	case AddIndexEntryAllocation:
+		ib = Add2Ptr(buffer_le, roff);
+		hdr = &ib->ihdr;
+		e = data;
+		esize = le16_to_cpu(e->size);
+		e1 = Add2Ptr(ib, aoff);
+
+		if (is_baad(&ib->rhdr))
+			goto dirty_vol;
+		if (!check_lsn(&ib->rhdr, rlsn))
+			goto out;
+
+		used = le32_to_cpu(hdr->used);
+
+		if (!check_index_buffer(ib, bytes) ||
+		    !check_if_alloc_index(hdr, aoff) ||
+		    Add2Ptr(e, esize) > Add2Ptr(lrh, rec_len) ||
+		    used + esize > le32_to_cpu(hdr->total)) {
+			goto dirty_vol;
+		}
+
+		memmove(Add2Ptr(e1, esize), e1,
+			PtrOffset(e1, Add2Ptr(hdr, used)));
+		memcpy(e1, e, esize);
+
+		hdr->used = cpu_to_le32(used + esize);
+
+		a_dirty = true;
+
+		ntfs_fix_pre_write(&ib->rhdr, bytes);
+		break;
+
+	case DeleteIndexEntryAllocation:
+		ib = Add2Ptr(buffer_le, roff);
+		hdr = &ib->ihdr;
+		e = Add2Ptr(ib, aoff);
+		esize = le16_to_cpu(e->size);
+
+		if (is_baad(&ib->rhdr))
+			goto dirty_vol;
+		if (!check_lsn(&ib->rhdr, rlsn))
+			goto out;
+
+		if (!check_index_buffer(ib, bytes) ||
+		    !check_if_alloc_index(hdr, aoff)) {
+			goto dirty_vol;
+		}
+
+		e1 = Add2Ptr(e, esize);
+		nsize = esize;
+		used = le32_to_cpu(hdr->used);
+
+		memmove(e, e1, PtrOffset(e1, Add2Ptr(hdr, used)));
+
+		hdr->used = cpu_to_le32(used - nsize);
+
+		a_dirty = true;
+
+		ntfs_fix_pre_write(&ib->rhdr, bytes);
+		break;
+
+	case WriteEndOfIndexBuffer:
+		ib = Add2Ptr(buffer_le, roff);
+		hdr = &ib->ihdr;
+		e = Add2Ptr(ib, aoff);
+
+		if (is_baad(&ib->rhdr))
+			goto dirty_vol;
+		if (!check_lsn(&ib->rhdr, rlsn))
+			goto out;
+		if (!check_index_buffer(ib, bytes) ||
+		    !check_if_alloc_index(hdr, aoff) ||
+		    aoff + dlen > offsetof(struct INDEX_BUFFER, ihdr) +
+					  le32_to_cpu(hdr->total)) {
+			goto dirty_vol;
+		}
+
+		hdr->used = cpu_to_le32(dlen + PtrOffset(hdr, e));
+		memmove(e, data, dlen);
+
+		a_dirty = true;
+		ntfs_fix_pre_write(&ib->rhdr, bytes);
+		break;
+
+	case SetIndexEntryVcnAllocation:
+		ib = Add2Ptr(buffer_le, roff);
+		hdr = &ib->ihdr;
+		e = Add2Ptr(ib, aoff);
+
+		if (is_baad(&ib->rhdr))
+			goto dirty_vol;
+
+		if (!check_lsn(&ib->rhdr, rlsn))
+			goto out;
+		if (!check_index_buffer(ib, bytes) ||
+		    !check_if_alloc_index(hdr, aoff)) {
+			goto dirty_vol;
+		}
+
+		de_set_vbn_le(e, *(__le64 *)data);
+
+		a_dirty = true;
+		ntfs_fix_pre_write(&ib->rhdr, bytes);
+		break;
+
+	case UpdateFileNameAllocation:
+		ib = Add2Ptr(buffer_le, roff);
+		hdr = &ib->ihdr;
+		e = Add2Ptr(ib, aoff);
+
+		if (is_baad(&ib->rhdr))
+			goto dirty_vol;
+
+		if (!check_lsn(&ib->rhdr, rlsn))
+			goto out;
+		if (!check_index_buffer(ib, bytes) ||
+		    !check_if_alloc_index(hdr, aoff)) {
+			goto dirty_vol;
+		}
+
+		fname = (struct ATTR_FILE_NAME *)(e + 1);
+		memmove(&fname->dup, data, sizeof(fname->dup));
+
+		a_dirty = true;
+		ntfs_fix_pre_write(&ib->rhdr, bytes);
+		break;
+
+	case SetBitsInNonresidentBitMap:
+		bmp_off =
+			le32_to_cpu(((struct BITMAP_RANGE *)data)->bitmap_off);
+		bmp_bits = le32_to_cpu(((struct BITMAP_RANGE *)data)->bits);
+
+		if (cbo + (bmp_off + 7) / 8 > lco ||
+		    cbo + ((bmp_off + bmp_bits + 7) / 8) > lco) {
+			goto dirty_vol;
+		}
+
+		__bitmap_set(Add2Ptr(buffer_le, roff), bmp_off, bmp_bits);
+		a_dirty = true;
+		break;
+
+	case ClearBitsInNonresidentBitMap:
+		bmp_off =
+			le32_to_cpu(((struct BITMAP_RANGE *)data)->bitmap_off);
+		bmp_bits = le32_to_cpu(((struct BITMAP_RANGE *)data)->bits);
+
+		if (cbo + (bmp_off + 7) / 8 > lco ||
+		    cbo + ((bmp_off + bmp_bits + 7) / 8) > lco) {
+			goto dirty_vol;
+		}
+
+		__bitmap_clear(Add2Ptr(buffer_le, roff), bmp_off, bmp_bits);
+		a_dirty = true;
+		break;
+
+	case UpdateRecordDataAllocation:
+		ib = Add2Ptr(buffer_le, roff);
+		hdr = &ib->ihdr;
+		e = Add2Ptr(ib, aoff);
+
+		if (is_baad(&ib->rhdr))
+			goto dirty_vol;
+
+		if (!check_lsn(&ib->rhdr, rlsn))
+			goto out;
+		if (!check_index_buffer(ib, bytes) ||
+		    !check_if_alloc_index(hdr, aoff)) {
+			goto dirty_vol;
+		}
+
+		memmove(Add2Ptr(e, le16_to_cpu(e->view.data_off)), data, dlen);
+
+		a_dirty = true;
+		ntfs_fix_pre_write(&ib->rhdr, bytes);
+		break;
+
+	default:
+		WARN_ON(1);
+	}
+
+	if (rlsn) {
+		__le64 t64 = cpu_to_le64(*rlsn);
+
+		if (rec)
+			rec->rhdr.lsn = t64;
+		if (ib)
+			ib->rhdr.lsn = t64;
+	}
+
+	if (mi && mi->dirty) {
+		err = mi_write(mi, 0);
+		if (err)
+			goto out;
+	}
+
+	if (a_dirty) {
+		attr = oa->attr;
+		err = ntfs_sb_write_run(sbi, oa->run1, vbo, buffer_le, bytes);
+		if (err)
+			goto out;
+	}
+
+out:
+
+	if (inode)
+		iput(inode);
+	else if (mi != mi2_child)
+		mi_put(mi);
+
+	kfree(buffer_le);
+
+	return err;
+
+dirty_vol:
+	log->set_dirty = true;
+	goto out;
+}
+
+/*
+ * log_replay - Replays log and empties it.
+ *
+ * This function is called during mount operation.
+ * It replays log and empties it.
+ * Initialized is set false if logfile contains '-1'.
+ */
+int log_replay(struct ntfs_inode *ni, bool *initialized)
+{
+	int err;
+	struct ntfs_sb_info *sbi = ni->mi.sbi;
+	struct ntfs_log *log;
+
+	struct restart_info rst_info, rst_info2;
+	u64 rec_lsn, ra_lsn, checkpt_lsn = 0, rlsn = 0;
+	struct ATTR_NAME_ENTRY *attr_names = NULL;
+	struct ATTR_NAME_ENTRY *ane;
+	struct RESTART_TABLE *dptbl = NULL;
+	struct RESTART_TABLE *trtbl = NULL;
+	const struct RESTART_TABLE *rt;
+	struct RESTART_TABLE *oatbl = NULL;
+	struct inode *inode;
+	struct OpenAttr *oa;
+	struct ntfs_inode *ni_oe;
+	struct ATTRIB *attr = NULL;
+	u64 size, vcn, undo_next_lsn;
+	CLST rno, lcn, lcn0, len0, clen;
+	void *data;
+	struct NTFS_RESTART *rst = NULL;
+	struct lcb *lcb = NULL;
+	struct OPEN_ATTR_ENRTY *oe;
+	struct TRANSACTION_ENTRY *tr;
+	struct DIR_PAGE_ENTRY *dp;
+	u32 i, bytes_per_attr_entry;
+	u32 l_size = ni->vfs_inode.i_size;
+	u32 orig_file_size = l_size;
+	u32 page_size, vbo, tail, off, dlen;
+	u32 saved_len, rec_len, transact_id;
+	bool use_second_page;
+	struct RESTART_AREA *ra2, *ra = NULL;
+	struct CLIENT_REC *ca, *cr;
+	__le16 client;
+	struct RESTART_HDR *rh;
+	const struct LFS_RECORD_HDR *frh;
+	const struct LOG_REC_HDR *lrh;
+	bool is_mapped;
+	bool is_ro = sb_rdonly(sbi->sb);
+	u64 t64;
+	u16 t16;
+	u32 t32;
+
+	/* Get the size of page. NOTE: To replay we can use default page. */
+#if PAGE_SIZE >= DefaultLogPageSize && PAGE_SIZE <= DefaultLogPageSize * 2
+	page_size = norm_file_page(PAGE_SIZE, &l_size, true);
+#else
+	page_size = norm_file_page(PAGE_SIZE, &l_size, false);
+#endif
+	if (!page_size)
+		return -EINVAL;
+
+	log = kzalloc(sizeof(struct ntfs_log), GFP_NOFS);
+	if (!log)
+		return -ENOMEM;
+
+	log->ni = ni;
+	log->l_size = l_size;
+	log->one_page_buf = kmalloc(page_size, GFP_NOFS);
+
+	if (!log->one_page_buf) {
+		err = -ENOMEM;
+		goto out;
+	}
+
+	log->page_size = page_size;
+	log->page_mask = page_size - 1;
+	log->page_bits = blksize_bits(page_size);
+
+	/* Look for a restart area on the disk. */
+	err = log_read_rst(log, l_size, true, &rst_info);
+	if (err)
+		goto out;
+
+	/* remember 'initialized' */
+	*initialized = rst_info.initialized;
+
+	if (!rst_info.restart) {
+		if (rst_info.initialized) {
+			/* No restart area but the file is not initialized. */
+			err = -EINVAL;
+			goto out;
+		}
+
+		log_init_pg_hdr(log, page_size, page_size, 1, 1);
+		log_create(log, l_size, 0, get_random_int(), false, false);
+
+		log->ra = ra;
+
+		ra = log_create_ra(log);
+		if (!ra) {
+			err = -ENOMEM;
+			goto out;
+		}
+		log->ra = ra;
+		log->init_ra = true;
+
+		goto process_log;
+	}
+
+	/*
+	 * If the restart offset above wasn't zero then we won't
+	 * look for a second restart.
+	 */
+	if (rst_info.vbo)
+		goto check_restart_area;
+
+	err = log_read_rst(log, l_size, false, &rst_info2);
+
+	/* Determine which restart area to use. */
+	if (!rst_info2.restart || rst_info2.last_lsn <= rst_info.last_lsn)
+		goto use_first_page;
+
+	use_second_page = true;
+
+	if (rst_info.chkdsk_was_run && page_size != rst_info.vbo) {
+		struct RECORD_PAGE_HDR *sp = NULL;
+		bool usa_error;
+
+		if (!read_log_page(log, page_size, &sp, &usa_error) &&
+		    sp->rhdr.sign == NTFS_CHKD_SIGNATURE) {
+			use_second_page = false;
+		}
+		kfree(sp);
+	}
+
+	if (use_second_page) {
+		kfree(rst_info.r_page);
+		memcpy(&rst_info, &rst_info2, sizeof(struct restart_info));
+		rst_info2.r_page = NULL;
+	}
+
+use_first_page:
+	kfree(rst_info2.r_page);
+
+check_restart_area:
+	/*
+	 * If the restart area is at offset 0, we want
+	 * to write the second restart area first.
+	 */
+	log->init_ra = !!rst_info.vbo;
+
+	/* If we have a valid page then grab a pointer to the restart area. */
+	ra2 = rst_info.valid_page
+		      ? Add2Ptr(rst_info.r_page,
+				le16_to_cpu(rst_info.r_page->ra_off))
+		      : NULL;
+
+	if (rst_info.chkdsk_was_run ||
+	    (ra2 && ra2->client_idx[1] == LFS_NO_CLIENT_LE)) {
+		bool wrapped = false;
+		bool use_multi_page = false;
+		u32 open_log_count;
+
+		/* Do some checks based on whether we have a valid log page. */
+		if (!rst_info.valid_page) {
+			open_log_count = get_random_int();
+			goto init_log_instance;
+		}
+		open_log_count = le32_to_cpu(ra2->open_log_count);
+
+		/*
+		 * If the restart page size isn't changing then we want to
+		 * check how much work we need to do.
+		 */
+		if (page_size != le32_to_cpu(rst_info.r_page->sys_page_size))
+			goto init_log_instance;
+
+init_log_instance:
+		log_init_pg_hdr(log, page_size, page_size, 1, 1);
+
+		log_create(log, l_size, rst_info.last_lsn, open_log_count,
+			   wrapped, use_multi_page);
+
+		ra = log_create_ra(log);
+		if (!ra) {
+			err = -ENOMEM;
+			goto out;
+		}
+		log->ra = ra;
+
+		/* Put the restart areas and initialize
+		 * the log file as required.
+		 */
+		goto process_log;
+	}
+
+	if (!ra2) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	/*
+	 * If the log page or the system page sizes have changed, we can't
+	 * use the log file. We must use the system page size instead of the
+	 * default size if there is not a clean shutdown.
+	 */
+	t32 = le32_to_cpu(rst_info.r_page->sys_page_size);
+	if (page_size != t32) {
+		l_size = orig_file_size;
+		page_size =
+			norm_file_page(t32, &l_size, t32 == DefaultLogPageSize);
+	}
+
+	if (page_size != t32 ||
+	    page_size != le32_to_cpu(rst_info.r_page->page_size)) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	/* If the file size has shrunk then we won't mount it. */
+	if (l_size < le64_to_cpu(ra2->l_size)) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	log_init_pg_hdr(log, page_size, page_size,
+			le16_to_cpu(rst_info.r_page->major_ver),
+			le16_to_cpu(rst_info.r_page->minor_ver));
+
+	log->l_size = le64_to_cpu(ra2->l_size);
+	log->seq_num_bits = le32_to_cpu(ra2->seq_num_bits);
+	log->file_data_bits = sizeof(u64) * 8 - log->seq_num_bits;
+	log->seq_num_mask = (8 << log->file_data_bits) - 1;
+	log->last_lsn = le64_to_cpu(ra2->current_lsn);
+	log->seq_num = log->last_lsn >> log->file_data_bits;
+	log->ra_off = le16_to_cpu(rst_info.r_page->ra_off);
+	log->restart_size = log->sys_page_size - log->ra_off;
+	log->record_header_len = le16_to_cpu(ra2->rec_hdr_len);
+	log->ra_size = le16_to_cpu(ra2->ra_len);
+	log->data_off = le16_to_cpu(ra2->data_off);
+	log->data_size = log->page_size - log->data_off;
+	log->reserved = log->data_size - log->record_header_len;
+
+	vbo = lsn_to_vbo(log, log->last_lsn);
+
+	if (vbo < log->first_page) {
+		/* This is a pseudo lsn. */
+		log->l_flags |= NTFSLOG_NO_LAST_LSN;
+		log->next_page = log->first_page;
+		goto find_oldest;
+	}
+
+	/* Find the end of this log record. */
+	off = final_log_off(log, log->last_lsn,
+			    le32_to_cpu(ra2->last_lsn_data_len));
+
+	/* If we wrapped the file then increment the sequence number. */
+	if (off <= vbo) {
+		log->seq_num += 1;
+		log->l_flags |= NTFSLOG_WRAPPED;
+	}
+
+	/* Now compute the next log page to use. */
+	vbo &= ~log->sys_page_mask;
+	tail = log->page_size - (off & log->page_mask) - 1;
+
+	/*
+	 *If we can fit another log record on the page,
+	 * move back a page the log file.
+	 */
+	if (tail >= log->record_header_len) {
+		log->l_flags |= NTFSLOG_REUSE_TAIL;
+		log->next_page = vbo;
+	} else {
+		log->next_page = next_page_off(log, vbo);
+	}
+
+find_oldest:
+	/*
+	 * Find the oldest client lsn. Use the last
+	 * flushed lsn as a starting point.
+	 */
+	log->oldest_lsn = log->last_lsn;
+	oldest_client_lsn(Add2Ptr(ra2, le16_to_cpu(ra2->client_off)),
+			  ra2->client_idx[1], &log->oldest_lsn);
+	log->oldest_lsn_off = lsn_to_vbo(log, log->oldest_lsn);
+
+	if (log->oldest_lsn_off < log->first_page)
+		log->l_flags |= NTFSLOG_NO_OLDEST_LSN;
+
+	if (!(ra2->flags & RESTART_SINGLE_PAGE_IO))
+		log->l_flags |= NTFSLOG_WRAPPED | NTFSLOG_MULTIPLE_PAGE_IO;
+
+	log->current_openlog_count = le32_to_cpu(ra2->open_log_count);
+	log->total_avail_pages = log->l_size - log->first_page;
+	log->total_avail = log->total_avail_pages >> log->page_bits;
+	log->max_current_avail = log->total_avail * log->reserved;
+	log->total_avail = log->total_avail * log->data_size;
+
+	log->current_avail = current_log_avail(log);
+
+	ra = kzalloc(log->restart_size, GFP_NOFS);
+	if (!ra) {
+		err = -ENOMEM;
+		goto out;
+	}
+	log->ra = ra;
+
+	t16 = le16_to_cpu(ra2->client_off);
+	if (t16 == offsetof(struct RESTART_AREA, clients)) {
+		memcpy(ra, ra2, log->ra_size);
+	} else {
+		memcpy(ra, ra2, offsetof(struct RESTART_AREA, clients));
+		memcpy(ra->clients, Add2Ptr(ra2, t16),
+		       le16_to_cpu(ra2->ra_len) - t16);
+
+		log->current_openlog_count = get_random_int();
+		ra->open_log_count = cpu_to_le32(log->current_openlog_count);
+		log->ra_size = offsetof(struct RESTART_AREA, clients) +
+			       sizeof(struct CLIENT_REC);
+		ra->client_off =
+			cpu_to_le16(offsetof(struct RESTART_AREA, clients));
+		ra->ra_len = cpu_to_le16(log->ra_size);
+	}
+
+	le32_add_cpu(&ra->open_log_count, 1);
+
+	/* Now we need to walk through looking for the last lsn. */
+	err = last_log_lsn(log);
+	if (err)
+		goto out;
+
+	log->current_avail = current_log_avail(log);
+
+	/* Remember which restart area to write first. */
+	log->init_ra = rst_info.vbo;
+
+process_log:
+	/* 1.0, 1.1, 2.0 log->major_ver/minor_ver - short values. */
+	switch ((log->major_ver << 16) + log->minor_ver) {
+	case 0x10000:
+	case 0x10001:
+	case 0x20000:
+		break;
+	default:
+		ntfs_warn(sbi->sb, "\x24LogFile version %d.%d is not supported",
+			  log->major_ver, log->minor_ver);
+		err = -EOPNOTSUPP;
+		log->set_dirty = true;
+		goto out;
+	}
+
+	/* One client "NTFS" per logfile. */
+	ca = Add2Ptr(ra, le16_to_cpu(ra->client_off));
+
+	for (client = ra->client_idx[1];; client = cr->next_client) {
+		if (client == LFS_NO_CLIENT_LE) {
+			/* Insert "NTFS" client LogFile. */
+			client = ra->client_idx[0];
+			if (client == LFS_NO_CLIENT_LE)
+				return -EINVAL;
+
+			t16 = le16_to_cpu(client);
+			cr = ca + t16;
+
+			remove_client(ca, cr, &ra->client_idx[0]);
+
+			cr->restart_lsn = 0;
+			cr->oldest_lsn = cpu_to_le64(log->oldest_lsn);
+			cr->name_bytes = cpu_to_le32(8);
+			cr->name[0] = cpu_to_le16('N');
+			cr->name[1] = cpu_to_le16('T');
+			cr->name[2] = cpu_to_le16('F');
+			cr->name[3] = cpu_to_le16('S');
+
+			add_client(ca, t16, &ra->client_idx[1]);
+			break;
+		}
+
+		cr = ca + le16_to_cpu(client);
+
+		if (cpu_to_le32(8) == cr->name_bytes &&
+		    cpu_to_le16('N') == cr->name[0] &&
+		    cpu_to_le16('T') == cr->name[1] &&
+		    cpu_to_le16('F') == cr->name[2] &&
+		    cpu_to_le16('S') == cr->name[3])
+			break;
+	}
+
+	/* Update the client handle with the client block information. */
+	log->client_id.seq_num = cr->seq_num;
+	log->client_id.client_idx = client;
+
+	err = read_rst_area(log, &rst, &ra_lsn);
+	if (err)
+		goto out;
+
+	if (!rst)
+		goto out;
+
+	bytes_per_attr_entry = !rst->major_ver ? 0x2C : 0x28;
+
+	checkpt_lsn = le64_to_cpu(rst->check_point_start);
+	if (!checkpt_lsn)
+		checkpt_lsn = ra_lsn;
+
+	/* Allocate and Read the Transaction Table. */
+	if (!rst->transact_table_len)
+		goto check_dirty_page_table;
+
+	t64 = le64_to_cpu(rst->transact_table_lsn);
+	err = read_log_rec_lcb(log, t64, lcb_ctx_prev, &lcb);
+	if (err)
+		goto out;
+
+	lrh = lcb->log_rec;
+	frh = lcb->lrh;
+	rec_len = le32_to_cpu(frh->client_data_len);
+
+	if (!check_log_rec(lrh, rec_len, le32_to_cpu(frh->transact_id),
+			   bytes_per_attr_entry)) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	t16 = le16_to_cpu(lrh->redo_off);
+
+	rt = Add2Ptr(lrh, t16);
+	t32 = rec_len - t16;
+
+	/* Now check that this is a valid restart table. */
+	if (!check_rstbl(rt, t32)) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	trtbl = kmemdup(rt, t32, GFP_NOFS);
+	if (!trtbl) {
+		err = -ENOMEM;
+		goto out;
+	}
+
+	lcb_put(lcb);
+	lcb = NULL;
+
+check_dirty_page_table:
+	/* The next record back should be the Dirty Pages Table. */
+	if (!rst->dirty_pages_len)
+		goto check_attribute_names;
+
+	t64 = le64_to_cpu(rst->dirty_pages_table_lsn);
+	err = read_log_rec_lcb(log, t64, lcb_ctx_prev, &lcb);
+	if (err)
+		goto out;
+
+	lrh = lcb->log_rec;
+	frh = lcb->lrh;
+	rec_len = le32_to_cpu(frh->client_data_len);
+
+	if (!check_log_rec(lrh, rec_len, le32_to_cpu(frh->transact_id),
+			   bytes_per_attr_entry)) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	t16 = le16_to_cpu(lrh->redo_off);
+
+	rt = Add2Ptr(lrh, t16);
+	t32 = rec_len - t16;
+
+	/* Now check that this is a valid restart table. */
+	if (!check_rstbl(rt, t32)) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	dptbl = kmemdup(rt, t32, GFP_NOFS);
+	if (!dptbl) {
+		err = -ENOMEM;
+		goto out;
+	}
+
+	/* Convert Ra version '0' into version '1'. */
+	if (rst->major_ver)
+		goto end_conv_1;
+
+	dp = NULL;
+	while ((dp = enum_rstbl(dptbl, dp))) {
+		struct DIR_PAGE_ENTRY_32 *dp0 = (struct DIR_PAGE_ENTRY_32 *)dp;
+		// NOTE: Danger. Check for of boundary.
+		memmove(&dp->vcn, &dp0->vcn_low,
+			2 * sizeof(u64) +
+				le32_to_cpu(dp->lcns_follow) * sizeof(u64));
+	}
+
+end_conv_1:
+	lcb_put(lcb);
+	lcb = NULL;
+
+	/*
+	 * Go through the table and remove the duplicates,
+	 * remembering the oldest lsn values.
+	 */
+	if (sbi->cluster_size <= log->page_size)
+		goto trace_dp_table;
+
+	dp = NULL;
+	while ((dp = enum_rstbl(dptbl, dp))) {
+		struct DIR_PAGE_ENTRY *next = dp;
+
+		while ((next = enum_rstbl(dptbl, next))) {
+			if (next->target_attr == dp->target_attr &&
+			    next->vcn == dp->vcn) {
+				if (le64_to_cpu(next->oldest_lsn) <
+				    le64_to_cpu(dp->oldest_lsn)) {
+					dp->oldest_lsn = next->oldest_lsn;
+				}
+
+				free_rsttbl_idx(dptbl, PtrOffset(dptbl, next));
+			}
+		}
+	}
+trace_dp_table:
+check_attribute_names:
+	/* The next record should be the Attribute Names. */
+	if (!rst->attr_names_len)
+		goto check_attr_table;
+
+	t64 = le64_to_cpu(rst->attr_names_lsn);
+	err = read_log_rec_lcb(log, t64, lcb_ctx_prev, &lcb);
+	if (err)
+		goto out;
+
+	lrh = lcb->log_rec;
+	frh = lcb->lrh;
+	rec_len = le32_to_cpu(frh->client_data_len);
+
+	if (!check_log_rec(lrh, rec_len, le32_to_cpu(frh->transact_id),
+			   bytes_per_attr_entry)) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	t32 = lrh_length(lrh);
+	rec_len -= t32;
+
+	attr_names = kmemdup(Add2Ptr(lrh, t32), rec_len, GFP_NOFS);
+
+	lcb_put(lcb);
+	lcb = NULL;
+
+check_attr_table:
+	/* The next record should be the attribute Table. */
+	if (!rst->open_attr_len)
+		goto check_attribute_names2;
+
+	t64 = le64_to_cpu(rst->open_attr_table_lsn);
+	err = read_log_rec_lcb(log, t64, lcb_ctx_prev, &lcb);
+	if (err)
+		goto out;
+
+	lrh = lcb->log_rec;
+	frh = lcb->lrh;
+	rec_len = le32_to_cpu(frh->client_data_len);
+
+	if (!check_log_rec(lrh, rec_len, le32_to_cpu(frh->transact_id),
+			   bytes_per_attr_entry)) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	t16 = le16_to_cpu(lrh->redo_off);
+
+	rt = Add2Ptr(lrh, t16);
+	t32 = rec_len - t16;
+
+	if (!check_rstbl(rt, t32)) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	oatbl = kmemdup(rt, t32, GFP_NOFS);
+	if (!oatbl) {
+		err = -ENOMEM;
+		goto out;
+	}
+
+	log->open_attr_tbl = oatbl;
+
+	/* Clear all of the Attr pointers. */
+	oe = NULL;
+	while ((oe = enum_rstbl(oatbl, oe))) {
+		if (!rst->major_ver) {
+			struct OPEN_ATTR_ENRTY_32 oe0;
+
+			/* Really 'oe' points to OPEN_ATTR_ENRTY_32. */
+			memcpy(&oe0, oe, SIZEOF_OPENATTRIBUTEENTRY0);
+
+			oe->bytes_per_index = oe0.bytes_per_index;
+			oe->type = oe0.type;
+			oe->is_dirty_pages = oe0.is_dirty_pages;
+			oe->name_len = 0;
+			oe->ref = oe0.ref;
+			oe->open_record_lsn = oe0.open_record_lsn;
+		}
+
+		oe->is_attr_name = 0;
+		oe->ptr = NULL;
+	}
+
+	lcb_put(lcb);
+	lcb = NULL;
+
+check_attribute_names2:
+	if (!rst->attr_names_len)
+		goto trace_attribute_table;
+
+	ane = attr_names;
+	if (!oatbl)
+		goto trace_attribute_table;
+	while (ane->off) {
+		/* TODO: Clear table on exit! */
+		oe = Add2Ptr(oatbl, le16_to_cpu(ane->off));
+		t16 = le16_to_cpu(ane->name_bytes);
+		oe->name_len = t16 / sizeof(short);
+		oe->ptr = ane->name;
+		oe->is_attr_name = 2;
+		ane = Add2Ptr(ane, sizeof(struct ATTR_NAME_ENTRY) + t16);
+	}
+
+trace_attribute_table:
+	/*
+	 * If the checkpt_lsn is zero, then this is a freshly
+	 * formatted disk and we have no work to do.
+	 */
+	if (!checkpt_lsn) {
+		err = 0;
+		goto out;
+	}
+
+	if (!oatbl) {
+		oatbl = init_rsttbl(bytes_per_attr_entry, 8);
+		if (!oatbl) {
+			err = -ENOMEM;
+			goto out;
+		}
+	}
+
+	log->open_attr_tbl = oatbl;
+
+	/* Start the analysis pass from the Checkpoint lsn. */
+	rec_lsn = checkpt_lsn;
+
+	/* Read the first lsn. */
+	err = read_log_rec_lcb(log, checkpt_lsn, lcb_ctx_next, &lcb);
+	if (err)
+		goto out;
+
+	/* Loop to read all subsequent records to the end of the log file. */
+next_log_record_analyze:
+	err = read_next_log_rec(log, lcb, &rec_lsn);
+	if (err)
+		goto out;
+
+	if (!rec_lsn)
+		goto end_log_records_enumerate;
+
+	frh = lcb->lrh;
+	transact_id = le32_to_cpu(frh->transact_id);
+	rec_len = le32_to_cpu(frh->client_data_len);
+	lrh = lcb->log_rec;
+
+	if (!check_log_rec(lrh, rec_len, transact_id, bytes_per_attr_entry)) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	/*
+	 * The first lsn after the previous lsn remembered
+	 * the checkpoint is the first candidate for the rlsn.
+	 */
+	if (!rlsn)
+		rlsn = rec_lsn;
+
+	if (LfsClientRecord != frh->record_type)
+		goto next_log_record_analyze;
+
+	/*
+	 * Now update the Transaction Table for this transaction. If there
+	 * is no entry present or it is unallocated we allocate the entry.
+	 */
+	if (!trtbl) {
+		trtbl = init_rsttbl(sizeof(struct TRANSACTION_ENTRY),
+				    INITIAL_NUMBER_TRANSACTIONS);
+		if (!trtbl) {
+			err = -ENOMEM;
+			goto out;
+		}
+	}
+
+	tr = Add2Ptr(trtbl, transact_id);
+
+	if (transact_id >= bytes_per_rt(trtbl) ||
+	    tr->next != RESTART_ENTRY_ALLOCATED_LE) {
+		tr = alloc_rsttbl_from_idx(&trtbl, transact_id);
+		if (!tr) {
+			err = -ENOMEM;
+			goto out;
+		}
+		tr->transact_state = TransactionActive;
+		tr->first_lsn = cpu_to_le64(rec_lsn);
+	}
+
+	tr->prev_lsn = tr->undo_next_lsn = cpu_to_le64(rec_lsn);
+
+	/*
+	 * If this is a compensation log record, then change
+	 * the undo_next_lsn to be the undo_next_lsn of this record.
+	 */
+	if (lrh->undo_op == cpu_to_le16(CompensationLogRecord))
+		tr->undo_next_lsn = frh->client_undo_next_lsn;
+
+	/* Dispatch to handle log record depending on type. */
+	switch (le16_to_cpu(lrh->redo_op)) {
+	case InitializeFileRecordSegment:
+	case DeallocateFileRecordSegment:
+	case WriteEndOfFileRecordSegment:
+	case CreateAttribute:
+	case DeleteAttribute:
+	case UpdateResidentValue:
+	case UpdateNonresidentValue:
+	case UpdateMappingPairs:
+	case SetNewAttributeSizes:
+	case AddIndexEntryRoot:
+	case DeleteIndexEntryRoot:
+	case AddIndexEntryAllocation:
+	case DeleteIndexEntryAllocation:
+	case WriteEndOfIndexBuffer:
+	case SetIndexEntryVcnRoot:
+	case SetIndexEntryVcnAllocation:
+	case UpdateFileNameRoot:
+	case UpdateFileNameAllocation:
+	case SetBitsInNonresidentBitMap:
+	case ClearBitsInNonresidentBitMap:
+	case UpdateRecordDataRoot:
+	case UpdateRecordDataAllocation:
+	case ZeroEndOfFileRecord:
+		t16 = le16_to_cpu(lrh->target_attr);
+		t64 = le64_to_cpu(lrh->target_vcn);
+		dp = find_dp(dptbl, t16, t64);
+
+		if (dp)
+			goto copy_lcns;
+
+		/*
+		 * Calculate the number of clusters per page the system
+		 * which wrote the checkpoint, possibly creating the table.
+		 */
+		if (dptbl) {
+			t32 = (le16_to_cpu(dptbl->size) -
+			       sizeof(struct DIR_PAGE_ENTRY)) /
+			      sizeof(u64);
+		} else {
+			t32 = log->clst_per_page;
+			kfree(dptbl);
+			dptbl = init_rsttbl(struct_size(dp, page_lcns, t32),
+					    32);
+			if (!dptbl) {
+				err = -ENOMEM;
+				goto out;
+			}
+		}
+
+		dp = alloc_rsttbl_idx(&dptbl);
+		if (!dp) {
+			err = -ENOMEM;
+			goto out;
+		}
+		dp->target_attr = cpu_to_le32(t16);
+		dp->transfer_len = cpu_to_le32(t32 << sbi->cluster_bits);
+		dp->lcns_follow = cpu_to_le32(t32);
+		dp->vcn = cpu_to_le64(t64 & ~((u64)t32 - 1));
+		dp->oldest_lsn = cpu_to_le64(rec_lsn);
+
+copy_lcns:
+		/*
+		 * Copy the Lcns from the log record into the Dirty Page Entry.
+		 * TODO: For different page size support, must somehow make
+		 * whole routine a loop, case Lcns do not fit below.
+		 */
+		t16 = le16_to_cpu(lrh->lcns_follow);
+		for (i = 0; i < t16; i++) {
+			size_t j = (size_t)(le64_to_cpu(lrh->target_vcn) -
+					    le64_to_cpu(dp->vcn));
+			dp->page_lcns[j + i] = lrh->page_lcns[i];
+		}
+
+		goto next_log_record_analyze;
+
+	case DeleteDirtyClusters: {
+		u32 range_count =
+			le16_to_cpu(lrh->redo_len) / sizeof(struct LCN_RANGE);
+		const struct LCN_RANGE *r =
+			Add2Ptr(lrh, le16_to_cpu(lrh->redo_off));
+
+		/* Loop through all of the Lcn ranges this log record. */
+		for (i = 0; i < range_count; i++, r++) {
+			u64 lcn0 = le64_to_cpu(r->lcn);
+			u64 lcn_e = lcn0 + le64_to_cpu(r->len) - 1;
+
+			dp = NULL;
+			while ((dp = enum_rstbl(dptbl, dp))) {
+				u32 j;
+
+				t32 = le32_to_cpu(dp->lcns_follow);
+				for (j = 0; j < t32; j++) {
+					t64 = le64_to_cpu(dp->page_lcns[j]);
+					if (t64 >= lcn0 && t64 <= lcn_e)
+						dp->page_lcns[j] = 0;
+				}
+			}
+		}
+		goto next_log_record_analyze;
+		;
+	}
+
+	case OpenNonresidentAttribute:
+		t16 = le16_to_cpu(lrh->target_attr);
+		if (t16 >= bytes_per_rt(oatbl)) {
+			/*
+			 * Compute how big the table needs to be.
+			 * Add 10 extra entries for some cushion.
+			 */
+			u32 new_e = t16 / le16_to_cpu(oatbl->size);
+
+			new_e += 10 - le16_to_cpu(oatbl->used);
+
+			oatbl = extend_rsttbl(oatbl, new_e, ~0u);
+			log->open_attr_tbl = oatbl;
+			if (!oatbl) {
+				err = -ENOMEM;
+				goto out;
+			}
+		}
+
+		/* Point to the entry being opened. */
+		oe = alloc_rsttbl_from_idx(&oatbl, t16);
+		log->open_attr_tbl = oatbl;
+		if (!oe) {
+			err = -ENOMEM;
+			goto out;
+		}
+
+		/* Initialize this entry from the log record. */
+		t16 = le16_to_cpu(lrh->redo_off);
+		if (!rst->major_ver) {
+			/* Convert version '0' into version '1'. */
+			struct OPEN_ATTR_ENRTY_32 *oe0 = Add2Ptr(lrh, t16);
+
+			oe->bytes_per_index = oe0->bytes_per_index;
+			oe->type = oe0->type;
+			oe->is_dirty_pages = oe0->is_dirty_pages;
+			oe->name_len = 0; //oe0.name_len;
+			oe->ref = oe0->ref;
+			oe->open_record_lsn = oe0->open_record_lsn;
+		} else {
+			memcpy(oe, Add2Ptr(lrh, t16), bytes_per_attr_entry);
+		}
+
+		t16 = le16_to_cpu(lrh->undo_len);
+		if (t16) {
+			oe->ptr = kmalloc(t16, GFP_NOFS);
+			if (!oe->ptr) {
+				err = -ENOMEM;
+				goto out;
+			}
+			oe->name_len = t16 / sizeof(short);
+			memcpy(oe->ptr,
+			       Add2Ptr(lrh, le16_to_cpu(lrh->undo_off)), t16);
+			oe->is_attr_name = 1;
+		} else {
+			oe->ptr = NULL;
+			oe->is_attr_name = 0;
+		}
+
+		goto next_log_record_analyze;
+
+	case HotFix:
+		t16 = le16_to_cpu(lrh->target_attr);
+		t64 = le64_to_cpu(lrh->target_vcn);
+		dp = find_dp(dptbl, t16, t64);
+		if (dp) {
+			size_t j = le64_to_cpu(lrh->target_vcn) -
+				   le64_to_cpu(dp->vcn);
+			if (dp->page_lcns[j])
+				dp->page_lcns[j] = lrh->page_lcns[0];
+		}
+		goto next_log_record_analyze;
+
+	case EndTopLevelAction:
+		tr = Add2Ptr(trtbl, transact_id);
+		tr->prev_lsn = cpu_to_le64(rec_lsn);
+		tr->undo_next_lsn = frh->client_undo_next_lsn;
+		goto next_log_record_analyze;
+
+	case PrepareTransaction:
+		tr = Add2Ptr(trtbl, transact_id);
+		tr->transact_state = TransactionPrepared;
+		goto next_log_record_analyze;
+
+	case CommitTransaction:
+		tr = Add2Ptr(trtbl, transact_id);
+		tr->transact_state = TransactionCommitted;
+		goto next_log_record_analyze;
+
+	case ForgetTransaction:
+		free_rsttbl_idx(trtbl, transact_id);
+		goto next_log_record_analyze;
+
+	case Noop:
+	case OpenAttributeTableDump:
+	case AttributeNamesDump:
+	case DirtyPageTableDump:
+	case TransactionTableDump:
+		/* The following cases require no action the Analysis Pass. */
+		goto next_log_record_analyze;
+
+	default:
+		/*
+		 * All codes will be explicitly handled.
+		 * If we see a code we do not expect, then we are trouble.
+		 */
+		goto next_log_record_analyze;
+	}
+
+end_log_records_enumerate:
+	lcb_put(lcb);
+	lcb = NULL;
+
+	/*
+	 * Scan the Dirty Page Table and Transaction Table for
+	 * the lowest lsn, and return it as the Redo lsn.
+	 */
+	dp = NULL;
+	while ((dp = enum_rstbl(dptbl, dp))) {
+		t64 = le64_to_cpu(dp->oldest_lsn);
+		if (t64 && t64 < rlsn)
+			rlsn = t64;
+	}
+
+	tr = NULL;
+	while ((tr = enum_rstbl(trtbl, tr))) {
+		t64 = le64_to_cpu(tr->first_lsn);
+		if (t64 && t64 < rlsn)
+			rlsn = t64;
+	}
+
+	/*
+	 * Only proceed if the Dirty Page Table or Transaction
+	 * table are not empty.
+	 */
+	if ((!dptbl || !dptbl->total) && (!trtbl || !trtbl->total))
+		goto end_reply;
+
+	sbi->flags |= NTFS_FLAGS_NEED_REPLAY;
+	if (is_ro)
+		goto out;
+
+	/* Reopen all of the attributes with dirty pages. */
+	oe = NULL;
+next_open_attribute:
+
+	oe = enum_rstbl(oatbl, oe);
+	if (!oe) {
+		err = 0;
+		dp = NULL;
+		goto next_dirty_page;
+	}
+
+	oa = kzalloc(sizeof(struct OpenAttr), GFP_NOFS);
+	if (!oa) {
+		err = -ENOMEM;
+		goto out;
+	}
+
+	inode = ntfs_iget5(sbi->sb, &oe->ref, NULL);
+	if (IS_ERR(inode))
+		goto fake_attr;
+
+	if (is_bad_inode(inode)) {
+		iput(inode);
+fake_attr:
+		if (oa->ni) {
+			iput(&oa->ni->vfs_inode);
+			oa->ni = NULL;
+		}
+
+		attr = attr_create_nonres_log(sbi, oe->type, 0, oe->ptr,
+					      oe->name_len, 0);
+		if (!attr) {
+			kfree(oa);
+			err = -ENOMEM;
+			goto out;
+		}
+		oa->attr = attr;
+		oa->run1 = &oa->run0;
+		goto final_oe;
+	}
+
+	ni_oe = ntfs_i(inode);
+	oa->ni = ni_oe;
+
+	attr = ni_find_attr(ni_oe, NULL, NULL, oe->type, oe->ptr, oe->name_len,
+			    NULL, NULL);
+
+	if (!attr)
+		goto fake_attr;
+
+	t32 = le32_to_cpu(attr->size);
+	oa->attr = kmemdup(attr, t32, GFP_NOFS);
+	if (!oa->attr)
+		goto fake_attr;
+
+	if (!S_ISDIR(inode->i_mode)) {
+		if (attr->type == ATTR_DATA && !attr->name_len) {
+			oa->run1 = &ni_oe->file.run;
+			goto final_oe;
+		}
+	} else {
+		if (attr->type == ATTR_ALLOC &&
+		    attr->name_len == ARRAY_SIZE(I30_NAME) &&
+		    !memcmp(attr_name(attr), I30_NAME, sizeof(I30_NAME))) {
+			oa->run1 = &ni_oe->dir.alloc_run;
+			goto final_oe;
+		}
+	}
+
+	if (attr->non_res) {
+		u16 roff = le16_to_cpu(attr->nres.run_off);
+		CLST svcn = le64_to_cpu(attr->nres.svcn);
+
+		err = run_unpack(&oa->run0, sbi, inode->i_ino, svcn,
+				 le64_to_cpu(attr->nres.evcn), svcn,
+				 Add2Ptr(attr, roff), t32 - roff);
+		if (err < 0) {
+			kfree(oa->attr);
+			oa->attr = NULL;
+			goto fake_attr;
+		}
+		err = 0;
+	}
+	oa->run1 = &oa->run0;
+	attr = oa->attr;
+
+final_oe:
+	if (oe->is_attr_name == 1)
+		kfree(oe->ptr);
+	oe->is_attr_name = 0;
+	oe->ptr = oa;
+	oe->name_len = attr->name_len;
+
+	goto next_open_attribute;
+
+	/*
+	 * Now loop through the dirty page table to extract all of the Vcn/Lcn.
+	 * Mapping that we have, and insert it into the appropriate run.
+	 */
+next_dirty_page:
+	dp = enum_rstbl(dptbl, dp);
+	if (!dp)
+		goto do_redo_1;
+
+	oe = Add2Ptr(oatbl, le32_to_cpu(dp->target_attr));
+
+	if (oe->next != RESTART_ENTRY_ALLOCATED_LE)
+		goto next_dirty_page;
+
+	oa = oe->ptr;
+	if (!oa)
+		goto next_dirty_page;
+
+	i = -1;
+next_dirty_page_vcn:
+	i += 1;
+	if (i >= le32_to_cpu(dp->lcns_follow))
+		goto next_dirty_page;
+
+	vcn = le64_to_cpu(dp->vcn) + i;
+	size = (vcn + 1) << sbi->cluster_bits;
+
+	if (!dp->page_lcns[i])
+		goto next_dirty_page_vcn;
+
+	rno = ino_get(&oe->ref);
+	if (rno <= MFT_REC_MIRR &&
+	    size < (MFT_REC_VOL + 1) * sbi->record_size &&
+	    oe->type == ATTR_DATA) {
+		goto next_dirty_page_vcn;
+	}
+
+	lcn = le64_to_cpu(dp->page_lcns[i]);
+
+	if ((!run_lookup_entry(oa->run1, vcn, &lcn0, &len0, NULL) ||
+	     lcn0 != lcn) &&
+	    !run_add_entry(oa->run1, vcn, lcn, 1, false)) {
+		err = -ENOMEM;
+		goto out;
+	}
+	attr = oa->attr;
+	t64 = le64_to_cpu(attr->nres.alloc_size);
+	if (size > t64) {
+		attr->nres.valid_size = attr->nres.data_size =
+			attr->nres.alloc_size = cpu_to_le64(size);
+	}
+	goto next_dirty_page_vcn;
+
+do_redo_1:
+	/*
+	 * Perform the Redo Pass, to restore all of the dirty pages to the same
+	 * contents that they had immediately before the crash. If the dirty
+	 * page table is empty, then we can skip the entire Redo Pass.
+	 */
+	if (!dptbl || !dptbl->total)
+		goto do_undo_action;
+
+	rec_lsn = rlsn;
+
+	/*
+	 * Read the record at the Redo lsn, before falling
+	 * into common code to handle each record.
+	 */
+	err = read_log_rec_lcb(log, rlsn, lcb_ctx_next, &lcb);
+	if (err)
+		goto out;
+
+	/*
+	 * Now loop to read all of our log records forwards, until
+	 * we hit the end of the file, cleaning up at the end.
+	 */
+do_action_next:
+	frh = lcb->lrh;
+
+	if (LfsClientRecord != frh->record_type)
+		goto read_next_log_do_action;
+
+	transact_id = le32_to_cpu(frh->transact_id);
+	rec_len = le32_to_cpu(frh->client_data_len);
+	lrh = lcb->log_rec;
+
+	if (!check_log_rec(lrh, rec_len, transact_id, bytes_per_attr_entry)) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	/* Ignore log records that do not update pages. */
+	if (lrh->lcns_follow)
+		goto find_dirty_page;
+
+	goto read_next_log_do_action;
+
+find_dirty_page:
+	t16 = le16_to_cpu(lrh->target_attr);
+	t64 = le64_to_cpu(lrh->target_vcn);
+	dp = find_dp(dptbl, t16, t64);
+
+	if (!dp)
+		goto read_next_log_do_action;
+
+	if (rec_lsn < le64_to_cpu(dp->oldest_lsn))
+		goto read_next_log_do_action;
+
+	t16 = le16_to_cpu(lrh->target_attr);
+	if (t16 >= bytes_per_rt(oatbl)) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	oe = Add2Ptr(oatbl, t16);
+
+	if (oe->next != RESTART_ENTRY_ALLOCATED_LE) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	oa = oe->ptr;
+
+	if (!oa) {
+		err = -EINVAL;
+		goto out;
+	}
+	attr = oa->attr;
+
+	vcn = le64_to_cpu(lrh->target_vcn);
+
+	if (!run_lookup_entry(oa->run1, vcn, &lcn, NULL, NULL) ||
+	    lcn == SPARSE_LCN) {
+		goto read_next_log_do_action;
+	}
+
+	/* Point to the Redo data and get its length. */
+	data = Add2Ptr(lrh, le16_to_cpu(lrh->redo_off));
+	dlen = le16_to_cpu(lrh->redo_len);
+
+	/* Shorten length by any Lcns which were deleted. */
+	saved_len = dlen;
+
+	for (i = le16_to_cpu(lrh->lcns_follow); i; i--) {
+		size_t j;
+		u32 alen, voff;
+
+		voff = le16_to_cpu(lrh->record_off) +
+		       le16_to_cpu(lrh->attr_off);
+		voff += le16_to_cpu(lrh->cluster_off) << SECTOR_SHIFT;
+
+		/* If the Vcn question is allocated, we can just get out. */
+		j = le64_to_cpu(lrh->target_vcn) - le64_to_cpu(dp->vcn);
+		if (dp->page_lcns[j + i - 1])
+			break;
+
+		if (!saved_len)
+			saved_len = 1;
+
+		/*
+		 * Calculate the allocated space left relative to the
+		 * log record Vcn, after removing this unallocated Vcn.
+		 */
+		alen = (i - 1) << sbi->cluster_bits;
+
+		/*
+		 * If the update described this log record goes beyond
+		 * the allocated space, then we will have to reduce the length.
+		 */
+		if (voff >= alen)
+			dlen = 0;
+		else if (voff + dlen > alen)
+			dlen = alen - voff;
+	}
+
+	/*
+	 * If the resulting dlen from above is now zero,
+	 * we can skip this log record.
+	 */
+	if (!dlen && saved_len)
+		goto read_next_log_do_action;
+
+	t16 = le16_to_cpu(lrh->redo_op);
+	if (can_skip_action(t16))
+		goto read_next_log_do_action;
+
+	/* Apply the Redo operation a common routine. */
+	err = do_action(log, oe, lrh, t16, data, dlen, rec_len, &rec_lsn);
+	if (err)
+		goto out;
+
+	/* Keep reading and looping back until end of file. */
+read_next_log_do_action:
+	err = read_next_log_rec(log, lcb, &rec_lsn);
+	if (!err && rec_lsn)
+		goto do_action_next;
+
+	lcb_put(lcb);
+	lcb = NULL;
+
+do_undo_action:
+	/* Scan Transaction Table. */
+	tr = NULL;
+transaction_table_next:
+	tr = enum_rstbl(trtbl, tr);
+	if (!tr)
+		goto undo_action_done;
+
+	if (TransactionActive != tr->transact_state || !tr->undo_next_lsn) {
+		free_rsttbl_idx(trtbl, PtrOffset(trtbl, tr));
+		goto transaction_table_next;
+	}
+
+	log->transaction_id = PtrOffset(trtbl, tr);
+	undo_next_lsn = le64_to_cpu(tr->undo_next_lsn);
+
+	/*
+	 * We only have to do anything if the transaction has
+	 * something its undo_next_lsn field.
+	 */
+	if (!undo_next_lsn)
+		goto commit_undo;
+
+	/* Read the first record to be undone by this transaction. */
+	err = read_log_rec_lcb(log, undo_next_lsn, lcb_ctx_undo_next, &lcb);
+	if (err)
+		goto out;
+
+	/*
+	 * Now loop to read all of our log records forwards,
+	 * until we hit the end of the file, cleaning up at the end.
+	 */
+undo_action_next:
+
+	lrh = lcb->log_rec;
+	frh = lcb->lrh;
+	transact_id = le32_to_cpu(frh->transact_id);
+	rec_len = le32_to_cpu(frh->client_data_len);
+
+	if (!check_log_rec(lrh, rec_len, transact_id, bytes_per_attr_entry)) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	if (lrh->undo_op == cpu_to_le16(Noop))
+		goto read_next_log_undo_action;
+
+	oe = Add2Ptr(oatbl, le16_to_cpu(lrh->target_attr));
+	oa = oe->ptr;
+
+	t16 = le16_to_cpu(lrh->lcns_follow);
+	if (!t16)
+		goto add_allocated_vcns;
+
+	is_mapped = run_lookup_entry(oa->run1, le64_to_cpu(lrh->target_vcn),
+				     &lcn, &clen, NULL);
+
+	/*
+	 * If the mapping isn't already the table or the  mapping
+	 * corresponds to a hole the mapping, we need to make sure
+	 * there is no partial page already memory.
+	 */
+	if (is_mapped && lcn != SPARSE_LCN && clen >= t16)
+		goto add_allocated_vcns;
+
+	vcn = le64_to_cpu(lrh->target_vcn);
+	vcn &= ~(log->clst_per_page - 1);
+
+add_allocated_vcns:
+	for (i = 0, vcn = le64_to_cpu(lrh->target_vcn),
+	    size = (vcn + 1) << sbi->cluster_bits;
+	     i < t16; i++, vcn += 1, size += sbi->cluster_size) {
+		attr = oa->attr;
+		if (!attr->non_res) {
+			if (size > le32_to_cpu(attr->res.data_size))
+				attr->res.data_size = cpu_to_le32(size);
+		} else {
+			if (size > le64_to_cpu(attr->nres.data_size))
+				attr->nres.valid_size = attr->nres.data_size =
+					attr->nres.alloc_size =
+						cpu_to_le64(size);
+		}
+	}
+
+	t16 = le16_to_cpu(lrh->undo_op);
+	if (can_skip_action(t16))
+		goto read_next_log_undo_action;
+
+	/* Point to the Redo data and get its length. */
+	data = Add2Ptr(lrh, le16_to_cpu(lrh->undo_off));
+	dlen = le16_to_cpu(lrh->undo_len);
+
+	/* It is time to apply the undo action. */
+	err = do_action(log, oe, lrh, t16, data, dlen, rec_len, NULL);
+
+read_next_log_undo_action:
+	/*
+	 * Keep reading and looping back until we have read the
+	 * last record for this transaction.
+	 */
+	err = read_next_log_rec(log, lcb, &rec_lsn);
+	if (err)
+		goto out;
+
+	if (rec_lsn)
+		goto undo_action_next;
+
+	lcb_put(lcb);
+	lcb = NULL;
+
+commit_undo:
+	free_rsttbl_idx(trtbl, log->transaction_id);
+
+	log->transaction_id = 0;
+
+	goto transaction_table_next;
+
+undo_action_done:
+
+	ntfs_update_mftmirr(sbi, 0);
+
+	sbi->flags &= ~NTFS_FLAGS_NEED_REPLAY;
+
+end_reply:
+
+	err = 0;
+	if (is_ro)
+		goto out;
+
+	rh = kzalloc(log->page_size, GFP_NOFS);
+	if (!rh) {
+		err = -ENOMEM;
+		goto out;
+	}
+
+	rh->rhdr.sign = NTFS_RSTR_SIGNATURE;
+	rh->rhdr.fix_off = cpu_to_le16(offsetof(struct RESTART_HDR, fixups));
+	t16 = (log->page_size >> SECTOR_SHIFT) + 1;
+	rh->rhdr.fix_num = cpu_to_le16(t16);
+	rh->sys_page_size = cpu_to_le32(log->page_size);
+	rh->page_size = cpu_to_le32(log->page_size);
+
+	t16 = ALIGN(offsetof(struct RESTART_HDR, fixups) + sizeof(short) * t16,
+		    8);
+	rh->ra_off = cpu_to_le16(t16);
+	rh->minor_ver = cpu_to_le16(1); // 0x1A:
+	rh->major_ver = cpu_to_le16(1); // 0x1C:
+
+	ra2 = Add2Ptr(rh, t16);
+	memcpy(ra2, ra, sizeof(struct RESTART_AREA));
+
+	ra2->client_idx[0] = 0;
+	ra2->client_idx[1] = LFS_NO_CLIENT_LE;
+	ra2->flags = cpu_to_le16(2);
+
+	le32_add_cpu(&ra2->open_log_count, 1);
+
+	ntfs_fix_pre_write(&rh->rhdr, log->page_size);
+
+	err = ntfs_sb_write_run(sbi, &ni->file.run, 0, rh, log->page_size);
+	if (!err)
+		err = ntfs_sb_write_run(sbi, &log->ni->file.run, log->page_size,
+					rh, log->page_size);
+
+	kfree(rh);
+	if (err)
+		goto out;
+
+out:
+	kfree(rst);
+	if (lcb)
+		lcb_put(lcb);
+
+	/*
+	 * Scan the Open Attribute Table to close all of
+	 * the open attributes.
+	 */
+	oe = NULL;
+	while ((oe = enum_rstbl(oatbl, oe))) {
+		rno = ino_get(&oe->ref);
+
+		if (oe->is_attr_name == 1) {
+			kfree(oe->ptr);
+			oe->ptr = NULL;
+			continue;
+		}
+
+		if (oe->is_attr_name)
+			continue;
+
+		oa = oe->ptr;
+		if (!oa)
+			continue;
+
+		run_close(&oa->run0);
+		kfree(oa->attr);
+		if (oa->ni)
+			iput(&oa->ni->vfs_inode);
+		kfree(oa);
+	}
+
+	kfree(trtbl);
+	kfree(oatbl);
+	kfree(dptbl);
+	kfree(attr_names);
+	kfree(rst_info.r_page);
+
+	kfree(ra);
+	kfree(log->one_page_buf);
+
+	if (err)
+		sbi->flags |= NTFS_FLAGS_NEED_REPLAY;
+
+	if (err == -EROFS)
+		err = 0;
+	else if (log->set_dirty)
+		ntfs_set_state(sbi, NTFS_DIRTY_ERROR);
+
+	kfree(log);
+
+	return err;
+}
diff --git a/fs/ntfs3/fsntfs.c b/fs/ntfs3/fsntfs.c
new file mode 100644
index 000000000000..91e3743e1442
--- /dev/null
+++ b/fs/ntfs3/fsntfs.c
@@ -0,0 +1,2509 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ *
+ * Copyright (C) 2019-2021 Paragon Software GmbH, All rights reserved.
+ *
+ */
+
+#include <linux/blkdev.h>
+#include <linux/buffer_head.h>
+#include <linux/fs.h>
+#include <linux/nls.h>
+
+#include "debug.h"
+#include "ntfs.h"
+#include "ntfs_fs.h"
+
+// clang-format off
+const struct cpu_str NAME_MFT = {
+	4, 0, { '$', 'M', 'F', 'T' },
+};
+const struct cpu_str NAME_MIRROR = {
+	8, 0, { '$', 'M', 'F', 'T', 'M', 'i', 'r', 'r' },
+};
+const struct cpu_str NAME_LOGFILE = {
+	8, 0, { '$', 'L', 'o', 'g', 'F', 'i', 'l', 'e' },
+};
+const struct cpu_str NAME_VOLUME = {
+	7, 0, { '$', 'V', 'o', 'l', 'u', 'm', 'e' },
+};
+const struct cpu_str NAME_ATTRDEF = {
+	8, 0, { '$', 'A', 't', 't', 'r', 'D', 'e', 'f' },
+};
+const struct cpu_str NAME_ROOT = {
+	1, 0, { '.' },
+};
+const struct cpu_str NAME_BITMAP = {
+	7, 0, { '$', 'B', 'i', 't', 'm', 'a', 'p' },
+};
+const struct cpu_str NAME_BOOT = {
+	5, 0, { '$', 'B', 'o', 'o', 't' },
+};
+const struct cpu_str NAME_BADCLUS = {
+	8, 0, { '$', 'B', 'a', 'd', 'C', 'l', 'u', 's' },
+};
+const struct cpu_str NAME_QUOTA = {
+	6, 0, { '$', 'Q', 'u', 'o', 't', 'a' },
+};
+const struct cpu_str NAME_SECURE = {
+	7, 0, { '$', 'S', 'e', 'c', 'u', 'r', 'e' },
+};
+const struct cpu_str NAME_UPCASE = {
+	7, 0, { '$', 'U', 'p', 'C', 'a', 's', 'e' },
+};
+const struct cpu_str NAME_EXTEND = {
+	7, 0, { '$', 'E', 'x', 't', 'e', 'n', 'd' },
+};
+const struct cpu_str NAME_OBJID = {
+	6, 0, { '$', 'O', 'b', 'j', 'I', 'd' },
+};
+const struct cpu_str NAME_REPARSE = {
+	8, 0, { '$', 'R', 'e', 'p', 'a', 'r', 's', 'e' },
+};
+const struct cpu_str NAME_USNJRNL = {
+	8, 0, { '$', 'U', 's', 'n', 'J', 'r', 'n', 'l' },
+};
+const __le16 BAD_NAME[4] = {
+	cpu_to_le16('$'), cpu_to_le16('B'), cpu_to_le16('a'), cpu_to_le16('d'),
+};
+const __le16 I30_NAME[4] = {
+	cpu_to_le16('$'), cpu_to_le16('I'), cpu_to_le16('3'), cpu_to_le16('0'),
+};
+const __le16 SII_NAME[4] = {
+	cpu_to_le16('$'), cpu_to_le16('S'), cpu_to_le16('I'), cpu_to_le16('I'),
+};
+const __le16 SDH_NAME[4] = {
+	cpu_to_le16('$'), cpu_to_le16('S'), cpu_to_le16('D'), cpu_to_le16('H'),
+};
+const __le16 SDS_NAME[4] = {
+	cpu_to_le16('$'), cpu_to_le16('S'), cpu_to_le16('D'), cpu_to_le16('S'),
+};
+const __le16 SO_NAME[2] = {
+	cpu_to_le16('$'), cpu_to_le16('O'),
+};
+const __le16 SQ_NAME[2] = {
+	cpu_to_le16('$'), cpu_to_le16('Q'),
+};
+const __le16 SR_NAME[2] = {
+	cpu_to_le16('$'), cpu_to_le16('R'),
+};
+
+#ifdef CONFIG_NTFS3_LZX_XPRESS
+const __le16 WOF_NAME[17] = {
+	cpu_to_le16('W'), cpu_to_le16('o'), cpu_to_le16('f'), cpu_to_le16('C'),
+	cpu_to_le16('o'), cpu_to_le16('m'), cpu_to_le16('p'), cpu_to_le16('r'),
+	cpu_to_le16('e'), cpu_to_le16('s'), cpu_to_le16('s'), cpu_to_le16('e'),
+	cpu_to_le16('d'), cpu_to_le16('D'), cpu_to_le16('a'), cpu_to_le16('t'),
+	cpu_to_le16('a'),
+};
+#endif
+
+// clang-format on
+
+/*
+ * ntfs_fix_pre_write - Insert fixups into @rhdr before writing to disk.
+ */
+bool ntfs_fix_pre_write(struct NTFS_RECORD_HEADER *rhdr, size_t bytes)
+{
+	u16 *fixup, *ptr;
+	u16 sample;
+	u16 fo = le16_to_cpu(rhdr->fix_off);
+	u16 fn = le16_to_cpu(rhdr->fix_num);
+
+	if ((fo & 1) || fo + fn * sizeof(short) > SECTOR_SIZE || !fn-- ||
+	    fn * SECTOR_SIZE > bytes) {
+		return false;
+	}
+
+	/* Get fixup pointer. */
+	fixup = Add2Ptr(rhdr, fo);
+
+	if (*fixup >= 0x7FFF)
+		*fixup = 1;
+	else
+		*fixup += 1;
+
+	sample = *fixup;
+
+	ptr = Add2Ptr(rhdr, SECTOR_SIZE - sizeof(short));
+
+	while (fn--) {
+		*++fixup = *ptr;
+		*ptr = sample;
+		ptr += SECTOR_SIZE / sizeof(short);
+	}
+	return true;
+}
+
+/*
+ * ntfs_fix_post_read - Remove fixups after reading from disk.
+ *
+ * Return: < 0 if error, 0 if ok, 1 if need to update fixups.
+ */
+int ntfs_fix_post_read(struct NTFS_RECORD_HEADER *rhdr, size_t bytes,
+		       bool simple)
+{
+	int ret;
+	u16 *fixup, *ptr;
+	u16 sample, fo, fn;
+
+	fo = le16_to_cpu(rhdr->fix_off);
+	fn = simple ? ((bytes >> SECTOR_SHIFT) + 1)
+		    : le16_to_cpu(rhdr->fix_num);
+
+	/* Check errors. */
+	if ((fo & 1) || fo + fn * sizeof(short) > SECTOR_SIZE || !fn-- ||
+	    fn * SECTOR_SIZE > bytes) {
+		return -EINVAL; /* Native chkntfs returns ok! */
+	}
+
+	/* Get fixup pointer. */
+	fixup = Add2Ptr(rhdr, fo);
+	sample = *fixup;
+	ptr = Add2Ptr(rhdr, SECTOR_SIZE - sizeof(short));
+	ret = 0;
+
+	while (fn--) {
+		/* Test current word. */
+		if (*ptr != sample) {
+			/* Fixup does not match! Is it serious error? */
+			ret = -E_NTFS_FIXUP;
+		}
+
+		/* Replace fixup. */
+		*ptr = *++fixup;
+		ptr += SECTOR_SIZE / sizeof(short);
+	}
+
+	return ret;
+}
+
+/*
+ * ntfs_extend_init - Load $Extend file.
+ */
+int ntfs_extend_init(struct ntfs_sb_info *sbi)
+{
+	int err;
+	struct super_block *sb = sbi->sb;
+	struct inode *inode, *inode2;
+	struct MFT_REF ref;
+
+	if (sbi->volume.major_ver < 3) {
+		ntfs_notice(sb, "Skip $Extend 'cause NTFS version");
+		return 0;
+	}
+
+	ref.low = cpu_to_le32(MFT_REC_EXTEND);
+	ref.high = 0;
+	ref.seq = cpu_to_le16(MFT_REC_EXTEND);
+	inode = ntfs_iget5(sb, &ref, &NAME_EXTEND);
+	if (IS_ERR(inode)) {
+		err = PTR_ERR(inode);
+		ntfs_err(sb, "Failed to load $Extend.");
+		inode = NULL;
+		goto out;
+	}
+
+	/* If ntfs_iget5() reads from disk it never returns bad inode. */
+	if (!S_ISDIR(inode->i_mode)) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	/* Try to find $ObjId */
+	inode2 = dir_search_u(inode, &NAME_OBJID, NULL);
+	if (inode2 && !IS_ERR(inode2)) {
+		if (is_bad_inode(inode2)) {
+			iput(inode2);
+		} else {
+			sbi->objid.ni = ntfs_i(inode2);
+			sbi->objid_no = inode2->i_ino;
+		}
+	}
+
+	/* Try to find $Quota */
+	inode2 = dir_search_u(inode, &NAME_QUOTA, NULL);
+	if (inode2 && !IS_ERR(inode2)) {
+		sbi->quota_no = inode2->i_ino;
+		iput(inode2);
+	}
+
+	/* Try to find $Reparse */
+	inode2 = dir_search_u(inode, &NAME_REPARSE, NULL);
+	if (inode2 && !IS_ERR(inode2)) {
+		sbi->reparse.ni = ntfs_i(inode2);
+		sbi->reparse_no = inode2->i_ino;
+	}
+
+	/* Try to find $UsnJrnl */
+	inode2 = dir_search_u(inode, &NAME_USNJRNL, NULL);
+	if (inode2 && !IS_ERR(inode2)) {
+		sbi->usn_jrnl_no = inode2->i_ino;
+		iput(inode2);
+	}
+
+	err = 0;
+out:
+	iput(inode);
+	return err;
+}
+
+int ntfs_loadlog_and_replay(struct ntfs_inode *ni, struct ntfs_sb_info *sbi)
+{
+	int err = 0;
+	struct super_block *sb = sbi->sb;
+	bool initialized = false;
+	struct MFT_REF ref;
+	struct inode *inode;
+
+	/* Check for 4GB. */
+	if (ni->vfs_inode.i_size >= 0x100000000ull) {
+		ntfs_err(sb, "\x24LogFile is too big");
+		err = -EINVAL;
+		goto out;
+	}
+
+	sbi->flags |= NTFS_FLAGS_LOG_REPLAYING;
+
+	ref.low = cpu_to_le32(MFT_REC_MFT);
+	ref.high = 0;
+	ref.seq = cpu_to_le16(1);
+
+	inode = ntfs_iget5(sb, &ref, NULL);
+
+	if (IS_ERR(inode))
+		inode = NULL;
+
+	if (!inode) {
+		/* Try to use MFT copy. */
+		u64 t64 = sbi->mft.lbo;
+
+		sbi->mft.lbo = sbi->mft.lbo2;
+		inode = ntfs_iget5(sb, &ref, NULL);
+		sbi->mft.lbo = t64;
+		if (IS_ERR(inode))
+			inode = NULL;
+	}
+
+	if (!inode) {
+		err = -EINVAL;
+		ntfs_err(sb, "Failed to load $MFT.");
+		goto out;
+	}
+
+	sbi->mft.ni = ntfs_i(inode);
+
+	/* LogFile should not contains attribute list. */
+	err = ni_load_all_mi(sbi->mft.ni);
+	if (!err)
+		err = log_replay(ni, &initialized);
+
+	iput(inode);
+	sbi->mft.ni = NULL;
+
+	sync_blockdev(sb->s_bdev);
+	invalidate_bdev(sb->s_bdev);
+
+	if (sbi->flags & NTFS_FLAGS_NEED_REPLAY) {
+		err = 0;
+		goto out;
+	}
+
+	if (sb_rdonly(sb) || !initialized)
+		goto out;
+
+	/* Fill LogFile by '-1' if it is initialized. */
+	err = ntfs_bio_fill_1(sbi, &ni->file.run);
+
+out:
+	sbi->flags &= ~NTFS_FLAGS_LOG_REPLAYING;
+
+	return err;
+}
+
+/*
+ * ntfs_query_def
+ *
+ * Return: Current ATTR_DEF_ENTRY for given attribute type.
+ */
+const struct ATTR_DEF_ENTRY *ntfs_query_def(struct ntfs_sb_info *sbi,
+					    enum ATTR_TYPE type)
+{
+	int type_in = le32_to_cpu(type);
+	size_t min_idx = 0;
+	size_t max_idx = sbi->def_entries - 1;
+
+	while (min_idx <= max_idx) {
+		size_t i = min_idx + ((max_idx - min_idx) >> 1);
+		const struct ATTR_DEF_ENTRY *entry = sbi->def_table + i;
+		int diff = le32_to_cpu(entry->type) - type_in;
+
+		if (!diff)
+			return entry;
+		if (diff < 0)
+			min_idx = i + 1;
+		else if (i)
+			max_idx = i - 1;
+		else
+			return NULL;
+	}
+	return NULL;
+}
+
+/*
+ * ntfs_look_for_free_space - Look for a free space in bitmap.
+ */
+int ntfs_look_for_free_space(struct ntfs_sb_info *sbi, CLST lcn, CLST len,
+			     CLST *new_lcn, CLST *new_len,
+			     enum ALLOCATE_OPT opt)
+{
+	int err;
+	CLST alen = 0;
+	struct super_block *sb = sbi->sb;
+	size_t alcn, zlen, zeroes, zlcn, zlen2, ztrim, new_zlen;
+	struct wnd_bitmap *wnd = &sbi->used.bitmap;
+
+	down_write_nested(&wnd->rw_lock, BITMAP_MUTEX_CLUSTERS);
+	if (opt & ALLOCATE_MFT) {
+		zlen = wnd_zone_len(wnd);
+
+		if (!zlen) {
+			err = ntfs_refresh_zone(sbi);
+			if (err)
+				goto out;
+			zlen = wnd_zone_len(wnd);
+		}
+
+		if (!zlen) {
+			ntfs_err(sbi->sb, "no free space to extend mft");
+			goto out;
+		}
+
+		lcn = wnd_zone_bit(wnd);
+		alen = zlen > len ? len : zlen;
+
+		wnd_zone_set(wnd, lcn + alen, zlen - alen);
+
+		err = wnd_set_used(wnd, lcn, alen);
+		if (err) {
+			up_write(&wnd->rw_lock);
+			return err;
+		}
+		alcn = lcn;
+		goto out;
+	}
+	/*
+	 * 'Cause cluster 0 is always used this value means that we should use
+	 * cached value of 'next_free_lcn' to improve performance.
+	 */
+	if (!lcn)
+		lcn = sbi->used.next_free_lcn;
+
+	if (lcn >= wnd->nbits)
+		lcn = 0;
+
+	alen = wnd_find(wnd, len, lcn, BITMAP_FIND_MARK_AS_USED, &alcn);
+	if (alen)
+		goto out;
+
+	/* Try to use clusters from MftZone. */
+	zlen = wnd_zone_len(wnd);
+	zeroes = wnd_zeroes(wnd);
+
+	/* Check too big request */
+	if (len > zeroes + zlen || zlen <= NTFS_MIN_MFT_ZONE)
+		goto out;
+
+	/* How many clusters to cat from zone. */
+	zlcn = wnd_zone_bit(wnd);
+	zlen2 = zlen >> 1;
+	ztrim = len > zlen ? zlen : (len > zlen2 ? len : zlen2);
+	new_zlen = zlen - ztrim;
+
+	if (new_zlen < NTFS_MIN_MFT_ZONE) {
+		new_zlen = NTFS_MIN_MFT_ZONE;
+		if (new_zlen > zlen)
+			new_zlen = zlen;
+	}
+
+	wnd_zone_set(wnd, zlcn, new_zlen);
+
+	/* Allocate continues clusters. */
+	alen = wnd_find(wnd, len, 0,
+			BITMAP_FIND_MARK_AS_USED | BITMAP_FIND_FULL, &alcn);
+
+out:
+	if (alen) {
+		err = 0;
+		*new_len = alen;
+		*new_lcn = alcn;
+
+		ntfs_unmap_meta(sb, alcn, alen);
+
+		/* Set hint for next requests. */
+		if (!(opt & ALLOCATE_MFT))
+			sbi->used.next_free_lcn = alcn + alen;
+	} else {
+		err = -ENOSPC;
+	}
+
+	up_write(&wnd->rw_lock);
+	return err;
+}
+
+/*
+ * ntfs_extend_mft - Allocate additional MFT records.
+ *
+ * sbi->mft.bitmap is locked for write.
+ *
+ * NOTE: recursive:
+ *	ntfs_look_free_mft ->
+ *	ntfs_extend_mft ->
+ *	attr_set_size ->
+ *	ni_insert_nonresident ->
+ *	ni_insert_attr ->
+ *	ni_ins_attr_ext ->
+ *	ntfs_look_free_mft ->
+ *	ntfs_extend_mft
+ *
+ * To avoid recursive always allocate space for two new MFT records
+ * see attrib.c: "at least two MFT to avoid recursive loop".
+ */
+static int ntfs_extend_mft(struct ntfs_sb_info *sbi)
+{
+	int err;
+	struct ntfs_inode *ni = sbi->mft.ni;
+	size_t new_mft_total;
+	u64 new_mft_bytes, new_bitmap_bytes;
+	struct ATTRIB *attr;
+	struct wnd_bitmap *wnd = &sbi->mft.bitmap;
+
+	new_mft_total = (wnd->nbits + MFT_INCREASE_CHUNK + 127) & (CLST)~127;
+	new_mft_bytes = (u64)new_mft_total << sbi->record_bits;
+
+	/* Step 1: Resize $MFT::DATA. */
+	down_write(&ni->file.run_lock);
+	err = attr_set_size(ni, ATTR_DATA, NULL, 0, &ni->file.run,
+			    new_mft_bytes, NULL, false, &attr);
+
+	if (err) {
+		up_write(&ni->file.run_lock);
+		goto out;
+	}
+
+	attr->nres.valid_size = attr->nres.data_size;
+	new_mft_total = le64_to_cpu(attr->nres.alloc_size) >> sbi->record_bits;
+	ni->mi.dirty = true;
+
+	/* Step 2: Resize $MFT::BITMAP. */
+	new_bitmap_bytes = bitmap_size(new_mft_total);
+
+	err = attr_set_size(ni, ATTR_BITMAP, NULL, 0, &sbi->mft.bitmap.run,
+			    new_bitmap_bytes, &new_bitmap_bytes, true, NULL);
+
+	/* Refresh MFT Zone if necessary. */
+	down_write_nested(&sbi->used.bitmap.rw_lock, BITMAP_MUTEX_CLUSTERS);
+
+	ntfs_refresh_zone(sbi);
+
+	up_write(&sbi->used.bitmap.rw_lock);
+	up_write(&ni->file.run_lock);
+
+	if (err)
+		goto out;
+
+	err = wnd_extend(wnd, new_mft_total);
+
+	if (err)
+		goto out;
+
+	ntfs_clear_mft_tail(sbi, sbi->mft.used, new_mft_total);
+
+	err = _ni_write_inode(&ni->vfs_inode, 0);
+out:
+	return err;
+}
+
+/*
+ * ntfs_look_free_mft - Look for a free MFT record.
+ */
+int ntfs_look_free_mft(struct ntfs_sb_info *sbi, CLST *rno, bool mft,
+		       struct ntfs_inode *ni, struct mft_inode **mi)
+{
+	int err = 0;
+	size_t zbit, zlen, from, to, fr;
+	size_t mft_total;
+	struct MFT_REF ref;
+	struct super_block *sb = sbi->sb;
+	struct wnd_bitmap *wnd = &sbi->mft.bitmap;
+	u32 ir;
+
+	static_assert(sizeof(sbi->mft.reserved_bitmap) * 8 >=
+		      MFT_REC_FREE - MFT_REC_RESERVED);
+
+	if (!mft)
+		down_write_nested(&wnd->rw_lock, BITMAP_MUTEX_MFT);
+
+	zlen = wnd_zone_len(wnd);
+
+	/* Always reserve space for MFT. */
+	if (zlen) {
+		if (mft) {
+			zbit = wnd_zone_bit(wnd);
+			*rno = zbit;
+			wnd_zone_set(wnd, zbit + 1, zlen - 1);
+		}
+		goto found;
+	}
+
+	/* No MFT zone. Find the nearest to '0' free MFT. */
+	if (!wnd_find(wnd, 1, MFT_REC_FREE, 0, &zbit)) {
+		/* Resize MFT */
+		mft_total = wnd->nbits;
+
+		err = ntfs_extend_mft(sbi);
+		if (!err) {
+			zbit = mft_total;
+			goto reserve_mft;
+		}
+
+		if (!mft || MFT_REC_FREE == sbi->mft.next_reserved)
+			goto out;
+
+		err = 0;
+
+		/*
+		 * Look for free record reserved area [11-16) ==
+		 * [MFT_REC_RESERVED, MFT_REC_FREE ) MFT bitmap always
+		 * marks it as used.
+		 */
+		if (!sbi->mft.reserved_bitmap) {
+			/* Once per session create internal bitmap for 5 bits. */
+			sbi->mft.reserved_bitmap = 0xFF;
+
+			ref.high = 0;
+			for (ir = MFT_REC_RESERVED; ir < MFT_REC_FREE; ir++) {
+				struct inode *i;
+				struct ntfs_inode *ni;
+				struct MFT_REC *mrec;
+
+				ref.low = cpu_to_le32(ir);
+				ref.seq = cpu_to_le16(ir);
+
+				i = ntfs_iget5(sb, &ref, NULL);
+				if (IS_ERR(i)) {
+next:
+					ntfs_notice(
+						sb,
+						"Invalid reserved record %x",
+						ref.low);
+					continue;
+				}
+				if (is_bad_inode(i)) {
+					iput(i);
+					goto next;
+				}
+
+				ni = ntfs_i(i);
+
+				mrec = ni->mi.mrec;
+
+				if (!is_rec_base(mrec))
+					goto next;
+
+				if (mrec->hard_links)
+					goto next;
+
+				if (!ni_std(ni))
+					goto next;
+
+				if (ni_find_attr(ni, NULL, NULL, ATTR_NAME,
+						 NULL, 0, NULL, NULL))
+					goto next;
+
+				__clear_bit(ir - MFT_REC_RESERVED,
+					    &sbi->mft.reserved_bitmap);
+			}
+		}
+
+		/* Scan 5 bits for zero. Bit 0 == MFT_REC_RESERVED */
+		zbit = find_next_zero_bit(&sbi->mft.reserved_bitmap,
+					  MFT_REC_FREE, MFT_REC_RESERVED);
+		if (zbit >= MFT_REC_FREE) {
+			sbi->mft.next_reserved = MFT_REC_FREE;
+			goto out;
+		}
+
+		zlen = 1;
+		sbi->mft.next_reserved = zbit;
+	} else {
+reserve_mft:
+		zlen = zbit == MFT_REC_FREE ? (MFT_REC_USER - MFT_REC_FREE) : 4;
+		if (zbit + zlen > wnd->nbits)
+			zlen = wnd->nbits - zbit;
+
+		while (zlen > 1 && !wnd_is_free(wnd, zbit, zlen))
+			zlen -= 1;
+
+		/* [zbit, zbit + zlen) will be used for MFT itself. */
+		from = sbi->mft.used;
+		if (from < zbit)
+			from = zbit;
+		to = zbit + zlen;
+		if (from < to) {
+			ntfs_clear_mft_tail(sbi, from, to);
+			sbi->mft.used = to;
+		}
+	}
+
+	if (mft) {
+		*rno = zbit;
+		zbit += 1;
+		zlen -= 1;
+	}
+
+	wnd_zone_set(wnd, zbit, zlen);
+
+found:
+	if (!mft) {
+		/* The request to get record for general purpose. */
+		if (sbi->mft.next_free < MFT_REC_USER)
+			sbi->mft.next_free = MFT_REC_USER;
+
+		for (;;) {
+			if (sbi->mft.next_free >= sbi->mft.bitmap.nbits) {
+			} else if (!wnd_find(wnd, 1, MFT_REC_USER, 0, &fr)) {
+				sbi->mft.next_free = sbi->mft.bitmap.nbits;
+			} else {
+				*rno = fr;
+				sbi->mft.next_free = *rno + 1;
+				break;
+			}
+
+			err = ntfs_extend_mft(sbi);
+			if (err)
+				goto out;
+		}
+	}
+
+	if (ni && !ni_add_subrecord(ni, *rno, mi)) {
+		err = -ENOMEM;
+		goto out;
+	}
+
+	/* We have found a record that are not reserved for next MFT. */
+	if (*rno >= MFT_REC_FREE)
+		wnd_set_used(wnd, *rno, 1);
+	else if (*rno >= MFT_REC_RESERVED && sbi->mft.reserved_bitmap_inited)
+		__set_bit(*rno - MFT_REC_RESERVED, &sbi->mft.reserved_bitmap);
+
+out:
+	if (!mft)
+		up_write(&wnd->rw_lock);
+
+	return err;
+}
+
+/*
+ * ntfs_mark_rec_free - Mark record as free.
+ */
+void ntfs_mark_rec_free(struct ntfs_sb_info *sbi, CLST rno)
+{
+	struct wnd_bitmap *wnd = &sbi->mft.bitmap;
+
+	down_write_nested(&wnd->rw_lock, BITMAP_MUTEX_MFT);
+	if (rno >= wnd->nbits)
+		goto out;
+
+	if (rno >= MFT_REC_FREE) {
+		if (!wnd_is_used(wnd, rno, 1))
+			ntfs_set_state(sbi, NTFS_DIRTY_ERROR);
+		else
+			wnd_set_free(wnd, rno, 1);
+	} else if (rno >= MFT_REC_RESERVED && sbi->mft.reserved_bitmap_inited) {
+		__clear_bit(rno - MFT_REC_RESERVED, &sbi->mft.reserved_bitmap);
+	}
+
+	if (rno < wnd_zone_bit(wnd))
+		wnd_zone_set(wnd, rno, 1);
+	else if (rno < sbi->mft.next_free && rno >= MFT_REC_USER)
+		sbi->mft.next_free = rno;
+
+out:
+	up_write(&wnd->rw_lock);
+}
+
+/*
+ * ntfs_clear_mft_tail - Format empty records [from, to).
+ *
+ * sbi->mft.bitmap is locked for write.
+ */
+int ntfs_clear_mft_tail(struct ntfs_sb_info *sbi, size_t from, size_t to)
+{
+	int err;
+	u32 rs;
+	u64 vbo;
+	struct runs_tree *run;
+	struct ntfs_inode *ni;
+
+	if (from >= to)
+		return 0;
+
+	rs = sbi->record_size;
+	ni = sbi->mft.ni;
+	run = &ni->file.run;
+
+	down_read(&ni->file.run_lock);
+	vbo = (u64)from * rs;
+	for (; from < to; from++, vbo += rs) {
+		struct ntfs_buffers nb;
+
+		err = ntfs_get_bh(sbi, run, vbo, rs, &nb);
+		if (err)
+			goto out;
+
+		err = ntfs_write_bh(sbi, &sbi->new_rec->rhdr, &nb, 0);
+		nb_put(&nb);
+		if (err)
+			goto out;
+	}
+
+out:
+	sbi->mft.used = from;
+	up_read(&ni->file.run_lock);
+	return err;
+}
+
+/*
+ * ntfs_refresh_zone - Refresh MFT zone.
+ *
+ * sbi->used.bitmap is locked for rw.
+ * sbi->mft.bitmap is locked for write.
+ * sbi->mft.ni->file.run_lock for write.
+ */
+int ntfs_refresh_zone(struct ntfs_sb_info *sbi)
+{
+	CLST zone_limit, zone_max, lcn, vcn, len;
+	size_t lcn_s, zlen;
+	struct wnd_bitmap *wnd = &sbi->used.bitmap;
+	struct ntfs_inode *ni = sbi->mft.ni;
+
+	/* Do not change anything unless we have non empty MFT zone. */
+	if (wnd_zone_len(wnd))
+		return 0;
+
+	/*
+	 * Compute the MFT zone at two steps.
+	 * It would be nice if we are able to allocate 1/8 of
+	 * total clusters for MFT but not more then 512 MB.
+	 */
+	zone_limit = (512 * 1024 * 1024) >> sbi->cluster_bits;
+	zone_max = wnd->nbits >> 3;
+	if (zone_max > zone_limit)
+		zone_max = zone_limit;
+
+	vcn = bytes_to_cluster(sbi,
+			       (u64)sbi->mft.bitmap.nbits << sbi->record_bits);
+
+	if (!run_lookup_entry(&ni->file.run, vcn - 1, &lcn, &len, NULL))
+		lcn = SPARSE_LCN;
+
+	/* We should always find Last Lcn for MFT. */
+	if (lcn == SPARSE_LCN)
+		return -EINVAL;
+
+	lcn_s = lcn + 1;
+
+	/* Try to allocate clusters after last MFT run. */
+	zlen = wnd_find(wnd, zone_max, lcn_s, 0, &lcn_s);
+	if (!zlen) {
+		ntfs_notice(sbi->sb, "MftZone: unavailable");
+		return 0;
+	}
+
+	/* Truncate too large zone. */
+	wnd_zone_set(wnd, lcn_s, zlen);
+
+	return 0;
+}
+
+/*
+ * ntfs_update_mftmirr - Update $MFTMirr data.
+ */
+int ntfs_update_mftmirr(struct ntfs_sb_info *sbi, int wait)
+{
+	int err;
+	struct super_block *sb = sbi->sb;
+	u32 blocksize = sb->s_blocksize;
+	sector_t block1, block2;
+	u32 bytes;
+
+	if (!(sbi->flags & NTFS_FLAGS_MFTMIRR))
+		return 0;
+
+	err = 0;
+	bytes = sbi->mft.recs_mirr << sbi->record_bits;
+	block1 = sbi->mft.lbo >> sb->s_blocksize_bits;
+	block2 = sbi->mft.lbo2 >> sb->s_blocksize_bits;
+
+	for (; bytes >= blocksize; bytes -= blocksize) {
+		struct buffer_head *bh1, *bh2;
+
+		bh1 = sb_bread(sb, block1++);
+		if (!bh1) {
+			err = -EIO;
+			goto out;
+		}
+
+		bh2 = sb_getblk(sb, block2++);
+		if (!bh2) {
+			put_bh(bh1);
+			err = -EIO;
+			goto out;
+		}
+
+		if (buffer_locked(bh2))
+			__wait_on_buffer(bh2);
+
+		lock_buffer(bh2);
+		memcpy(bh2->b_data, bh1->b_data, blocksize);
+		set_buffer_uptodate(bh2);
+		mark_buffer_dirty(bh2);
+		unlock_buffer(bh2);
+
+		put_bh(bh1);
+		bh1 = NULL;
+
+		if (wait)
+			err = sync_dirty_buffer(bh2);
+
+		put_bh(bh2);
+		if (err)
+			goto out;
+	}
+
+	sbi->flags &= ~NTFS_FLAGS_MFTMIRR;
+
+out:
+	return err;
+}
+
+/*
+ * ntfs_set_state
+ *
+ * Mount: ntfs_set_state(NTFS_DIRTY_DIRTY)
+ * Umount: ntfs_set_state(NTFS_DIRTY_CLEAR)
+ * NTFS error: ntfs_set_state(NTFS_DIRTY_ERROR)
+ */
+int ntfs_set_state(struct ntfs_sb_info *sbi, enum NTFS_DIRTY_FLAGS dirty)
+{
+	int err;
+	struct ATTRIB *attr;
+	struct VOLUME_INFO *info;
+	struct mft_inode *mi;
+	struct ntfs_inode *ni;
+
+	/*
+	 * Do not change state if fs was real_dirty.
+	 * Do not change state if fs already dirty(clear).
+	 * Do not change any thing if mounted read only.
+	 */
+	if (sbi->volume.real_dirty || sb_rdonly(sbi->sb))
+		return 0;
+
+	/* Check cached value. */
+	if ((dirty == NTFS_DIRTY_CLEAR ? 0 : VOLUME_FLAG_DIRTY) ==
+	    (sbi->volume.flags & VOLUME_FLAG_DIRTY))
+		return 0;
+
+	ni = sbi->volume.ni;
+	if (!ni)
+		return -EINVAL;
+
+	mutex_lock_nested(&ni->ni_lock, NTFS_INODE_MUTEX_DIRTY);
+
+	attr = ni_find_attr(ni, NULL, NULL, ATTR_VOL_INFO, NULL, 0, NULL, &mi);
+	if (!attr) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	info = resident_data_ex(attr, SIZEOF_ATTRIBUTE_VOLUME_INFO);
+	if (!info) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	switch (dirty) {
+	case NTFS_DIRTY_ERROR:
+		ntfs_notice(sbi->sb, "Mark volume as dirty due to NTFS errors");
+		sbi->volume.real_dirty = true;
+		fallthrough;
+	case NTFS_DIRTY_DIRTY:
+		info->flags |= VOLUME_FLAG_DIRTY;
+		break;
+	case NTFS_DIRTY_CLEAR:
+		info->flags &= ~VOLUME_FLAG_DIRTY;
+		break;
+	}
+	/* Cache current volume flags. */
+	sbi->volume.flags = info->flags;
+	mi->dirty = true;
+	err = 0;
+
+out:
+	ni_unlock(ni);
+	if (err)
+		return err;
+
+	mark_inode_dirty(&ni->vfs_inode);
+	/* verify(!ntfs_update_mftmirr()); */
+
+	/*
+	 * If we used wait=1, sync_inode_metadata waits for the io for the
+	 * inode to finish. It hangs when media is removed.
+	 * So wait=0 is sent down to sync_inode_metadata
+	 * and filemap_fdatawrite is used for the data blocks.
+	 */
+	err = sync_inode_metadata(&ni->vfs_inode, 0);
+	if (!err)
+		err = filemap_fdatawrite(ni->vfs_inode.i_mapping);
+
+	return err;
+}
+
+/*
+ * security_hash - Calculates a hash of security descriptor.
+ */
+static inline __le32 security_hash(const void *sd, size_t bytes)
+{
+	u32 hash = 0;
+	const __le32 *ptr = sd;
+
+	bytes >>= 2;
+	while (bytes--)
+		hash = ((hash >> 0x1D) | (hash << 3)) + le32_to_cpu(*ptr++);
+	return cpu_to_le32(hash);
+}
+
+int ntfs_sb_read(struct super_block *sb, u64 lbo, size_t bytes, void *buffer)
+{
+	struct block_device *bdev = sb->s_bdev;
+	u32 blocksize = sb->s_blocksize;
+	u64 block = lbo >> sb->s_blocksize_bits;
+	u32 off = lbo & (blocksize - 1);
+	u32 op = blocksize - off;
+
+	for (; bytes; block += 1, off = 0, op = blocksize) {
+		struct buffer_head *bh = __bread(bdev, block, blocksize);
+
+		if (!bh)
+			return -EIO;
+
+		if (op > bytes)
+			op = bytes;
+
+		memcpy(buffer, bh->b_data + off, op);
+
+		put_bh(bh);
+
+		bytes -= op;
+		buffer = Add2Ptr(buffer, op);
+	}
+
+	return 0;
+}
+
+int ntfs_sb_write(struct super_block *sb, u64 lbo, size_t bytes,
+		  const void *buf, int wait)
+{
+	u32 blocksize = sb->s_blocksize;
+	struct block_device *bdev = sb->s_bdev;
+	sector_t block = lbo >> sb->s_blocksize_bits;
+	u32 off = lbo & (blocksize - 1);
+	u32 op = blocksize - off;
+	struct buffer_head *bh;
+
+	if (!wait && (sb->s_flags & SB_SYNCHRONOUS))
+		wait = 1;
+
+	for (; bytes; block += 1, off = 0, op = blocksize) {
+		if (op > bytes)
+			op = bytes;
+
+		if (op < blocksize) {
+			bh = __bread(bdev, block, blocksize);
+			if (!bh) {
+				ntfs_err(sb, "failed to read block %llx",
+					 (u64)block);
+				return -EIO;
+			}
+		} else {
+			bh = __getblk(bdev, block, blocksize);
+			if (!bh)
+				return -ENOMEM;
+		}
+
+		if (buffer_locked(bh))
+			__wait_on_buffer(bh);
+
+		lock_buffer(bh);
+		if (buf) {
+			memcpy(bh->b_data + off, buf, op);
+			buf = Add2Ptr(buf, op);
+		} else {
+			memset(bh->b_data + off, -1, op);
+		}
+
+		set_buffer_uptodate(bh);
+		mark_buffer_dirty(bh);
+		unlock_buffer(bh);
+
+		if (wait) {
+			int err = sync_dirty_buffer(bh);
+
+			if (err) {
+				ntfs_err(
+					sb,
+					"failed to sync buffer at block %llx, error %d",
+					(u64)block, err);
+				put_bh(bh);
+				return err;
+			}
+		}
+
+		put_bh(bh);
+
+		bytes -= op;
+	}
+	return 0;
+}
+
+int ntfs_sb_write_run(struct ntfs_sb_info *sbi, const struct runs_tree *run,
+		      u64 vbo, const void *buf, size_t bytes)
+{
+	struct super_block *sb = sbi->sb;
+	u8 cluster_bits = sbi->cluster_bits;
+	u32 off = vbo & sbi->cluster_mask;
+	CLST lcn, clen, vcn = vbo >> cluster_bits, vcn_next;
+	u64 lbo, len;
+	size_t idx;
+
+	if (!run_lookup_entry(run, vcn, &lcn, &clen, &idx))
+		return -ENOENT;
+
+	if (lcn == SPARSE_LCN)
+		return -EINVAL;
+
+	lbo = ((u64)lcn << cluster_bits) + off;
+	len = ((u64)clen << cluster_bits) - off;
+
+	for (;;) {
+		u32 op = len < bytes ? len : bytes;
+		int err = ntfs_sb_write(sb, lbo, op, buf, 0);
+
+		if (err)
+			return err;
+
+		bytes -= op;
+		if (!bytes)
+			break;
+
+		vcn_next = vcn + clen;
+		if (!run_get_entry(run, ++idx, &vcn, &lcn, &clen) ||
+		    vcn != vcn_next)
+			return -ENOENT;
+
+		if (lcn == SPARSE_LCN)
+			return -EINVAL;
+
+		if (buf)
+			buf = Add2Ptr(buf, op);
+
+		lbo = ((u64)lcn << cluster_bits);
+		len = ((u64)clen << cluster_bits);
+	}
+
+	return 0;
+}
+
+struct buffer_head *ntfs_bread_run(struct ntfs_sb_info *sbi,
+				   const struct runs_tree *run, u64 vbo)
+{
+	struct super_block *sb = sbi->sb;
+	u8 cluster_bits = sbi->cluster_bits;
+	CLST lcn;
+	u64 lbo;
+
+	if (!run_lookup_entry(run, vbo >> cluster_bits, &lcn, NULL, NULL))
+		return ERR_PTR(-ENOENT);
+
+	lbo = ((u64)lcn << cluster_bits) + (vbo & sbi->cluster_mask);
+
+	return ntfs_bread(sb, lbo >> sb->s_blocksize_bits);
+}
+
+int ntfs_read_run_nb(struct ntfs_sb_info *sbi, const struct runs_tree *run,
+		     u64 vbo, void *buf, u32 bytes, struct ntfs_buffers *nb)
+{
+	int err;
+	struct super_block *sb = sbi->sb;
+	u32 blocksize = sb->s_blocksize;
+	u8 cluster_bits = sbi->cluster_bits;
+	u32 off = vbo & sbi->cluster_mask;
+	u32 nbh = 0;
+	CLST vcn_next, vcn = vbo >> cluster_bits;
+	CLST lcn, clen;
+	u64 lbo, len;
+	size_t idx;
+	struct buffer_head *bh;
+
+	if (!run) {
+		/* First reading of $Volume + $MFTMirr + $LogFile goes here. */
+		if (vbo > MFT_REC_VOL * sbi->record_size) {
+			err = -ENOENT;
+			goto out;
+		}
+
+		/* Use absolute boot's 'MFTCluster' to read record. */
+		lbo = vbo + sbi->mft.lbo;
+		len = sbi->record_size;
+	} else if (!run_lookup_entry(run, vcn, &lcn, &clen, &idx)) {
+		err = -ENOENT;
+		goto out;
+	} else {
+		if (lcn == SPARSE_LCN) {
+			err = -EINVAL;
+			goto out;
+		}
+
+		lbo = ((u64)lcn << cluster_bits) + off;
+		len = ((u64)clen << cluster_bits) - off;
+	}
+
+	off = lbo & (blocksize - 1);
+	if (nb) {
+		nb->off = off;
+		nb->bytes = bytes;
+	}
+
+	for (;;) {
+		u32 len32 = len >= bytes ? bytes : len;
+		sector_t block = lbo >> sb->s_blocksize_bits;
+
+		do {
+			u32 op = blocksize - off;
+
+			if (op > len32)
+				op = len32;
+
+			bh = ntfs_bread(sb, block);
+			if (!bh) {
+				err = -EIO;
+				goto out;
+			}
+
+			if (buf) {
+				memcpy(buf, bh->b_data + off, op);
+				buf = Add2Ptr(buf, op);
+			}
+
+			if (!nb) {
+				put_bh(bh);
+			} else if (nbh >= ARRAY_SIZE(nb->bh)) {
+				err = -EINVAL;
+				goto out;
+			} else {
+				nb->bh[nbh++] = bh;
+				nb->nbufs = nbh;
+			}
+
+			bytes -= op;
+			if (!bytes)
+				return 0;
+			len32 -= op;
+			block += 1;
+			off = 0;
+
+		} while (len32);
+
+		vcn_next = vcn + clen;
+		if (!run_get_entry(run, ++idx, &vcn, &lcn, &clen) ||
+		    vcn != vcn_next) {
+			err = -ENOENT;
+			goto out;
+		}
+
+		if (lcn == SPARSE_LCN) {
+			err = -EINVAL;
+			goto out;
+		}
+
+		lbo = ((u64)lcn << cluster_bits);
+		len = ((u64)clen << cluster_bits);
+	}
+
+out:
+	if (!nbh)
+		return err;
+
+	while (nbh) {
+		put_bh(nb->bh[--nbh]);
+		nb->bh[nbh] = NULL;
+	}
+
+	nb->nbufs = 0;
+	return err;
+}
+
+/*
+ * ntfs_read_bh
+ *
+ * Return: < 0 if error, 0 if ok, -E_NTFS_FIXUP if need to update fixups.
+ */
+int ntfs_read_bh(struct ntfs_sb_info *sbi, const struct runs_tree *run, u64 vbo,
+		 struct NTFS_RECORD_HEADER *rhdr, u32 bytes,
+		 struct ntfs_buffers *nb)
+{
+	int err = ntfs_read_run_nb(sbi, run, vbo, rhdr, bytes, nb);
+
+	if (err)
+		return err;
+	return ntfs_fix_post_read(rhdr, nb->bytes, true);
+}
+
+int ntfs_get_bh(struct ntfs_sb_info *sbi, const struct runs_tree *run, u64 vbo,
+		u32 bytes, struct ntfs_buffers *nb)
+{
+	int err = 0;
+	struct super_block *sb = sbi->sb;
+	u32 blocksize = sb->s_blocksize;
+	u8 cluster_bits = sbi->cluster_bits;
+	CLST vcn_next, vcn = vbo >> cluster_bits;
+	u32 off;
+	u32 nbh = 0;
+	CLST lcn, clen;
+	u64 lbo, len;
+	size_t idx;
+
+	nb->bytes = bytes;
+
+	if (!run_lookup_entry(run, vcn, &lcn, &clen, &idx)) {
+		err = -ENOENT;
+		goto out;
+	}
+
+	off = vbo & sbi->cluster_mask;
+	lbo = ((u64)lcn << cluster_bits) + off;
+	len = ((u64)clen << cluster_bits) - off;
+
+	nb->off = off = lbo & (blocksize - 1);
+
+	for (;;) {
+		u32 len32 = len < bytes ? len : bytes;
+		sector_t block = lbo >> sb->s_blocksize_bits;
+
+		do {
+			u32 op;
+			struct buffer_head *bh;
+
+			if (nbh >= ARRAY_SIZE(nb->bh)) {
+				err = -EINVAL;
+				goto out;
+			}
+
+			op = blocksize - off;
+			if (op > len32)
+				op = len32;
+
+			if (op == blocksize) {
+				bh = sb_getblk(sb, block);
+				if (!bh) {
+					err = -ENOMEM;
+					goto out;
+				}
+				if (buffer_locked(bh))
+					__wait_on_buffer(bh);
+				set_buffer_uptodate(bh);
+			} else {
+				bh = ntfs_bread(sb, block);
+				if (!bh) {
+					err = -EIO;
+					goto out;
+				}
+			}
+
+			nb->bh[nbh++] = bh;
+			bytes -= op;
+			if (!bytes) {
+				nb->nbufs = nbh;
+				return 0;
+			}
+
+			block += 1;
+			len32 -= op;
+			off = 0;
+		} while (len32);
+
+		vcn_next = vcn + clen;
+		if (!run_get_entry(run, ++idx, &vcn, &lcn, &clen) ||
+		    vcn != vcn_next) {
+			err = -ENOENT;
+			goto out;
+		}
+
+		lbo = ((u64)lcn << cluster_bits);
+		len = ((u64)clen << cluster_bits);
+	}
+
+out:
+	while (nbh) {
+		put_bh(nb->bh[--nbh]);
+		nb->bh[nbh] = NULL;
+	}
+
+	nb->nbufs = 0;
+
+	return err;
+}
+
+int ntfs_write_bh(struct ntfs_sb_info *sbi, struct NTFS_RECORD_HEADER *rhdr,
+		  struct ntfs_buffers *nb, int sync)
+{
+	int err = 0;
+	struct super_block *sb = sbi->sb;
+	u32 block_size = sb->s_blocksize;
+	u32 bytes = nb->bytes;
+	u32 off = nb->off;
+	u16 fo = le16_to_cpu(rhdr->fix_off);
+	u16 fn = le16_to_cpu(rhdr->fix_num);
+	u32 idx;
+	__le16 *fixup;
+	__le16 sample;
+
+	if ((fo & 1) || fo + fn * sizeof(short) > SECTOR_SIZE || !fn-- ||
+	    fn * SECTOR_SIZE > bytes) {
+		return -EINVAL;
+	}
+
+	for (idx = 0; bytes && idx < nb->nbufs; idx += 1, off = 0) {
+		u32 op = block_size - off;
+		char *bh_data;
+		struct buffer_head *bh = nb->bh[idx];
+		__le16 *ptr, *end_data;
+
+		if (op > bytes)
+			op = bytes;
+
+		if (buffer_locked(bh))
+			__wait_on_buffer(bh);
+
+		lock_buffer(nb->bh[idx]);
+
+		bh_data = bh->b_data + off;
+		end_data = Add2Ptr(bh_data, op);
+		memcpy(bh_data, rhdr, op);
+
+		if (!idx) {
+			u16 t16;
+
+			fixup = Add2Ptr(bh_data, fo);
+			sample = *fixup;
+			t16 = le16_to_cpu(sample);
+			if (t16 >= 0x7FFF) {
+				sample = *fixup = cpu_to_le16(1);
+			} else {
+				sample = cpu_to_le16(t16 + 1);
+				*fixup = sample;
+			}
+
+			*(__le16 *)Add2Ptr(rhdr, fo) = sample;
+		}
+
+		ptr = Add2Ptr(bh_data, SECTOR_SIZE - sizeof(short));
+
+		do {
+			*++fixup = *ptr;
+			*ptr = sample;
+			ptr += SECTOR_SIZE / sizeof(short);
+		} while (ptr < end_data);
+
+		set_buffer_uptodate(bh);
+		mark_buffer_dirty(bh);
+		unlock_buffer(bh);
+
+		if (sync) {
+			int err2 = sync_dirty_buffer(bh);
+
+			if (!err && err2)
+				err = err2;
+		}
+
+		bytes -= op;
+		rhdr = Add2Ptr(rhdr, op);
+	}
+
+	return err;
+}
+
+static inline struct bio *ntfs_alloc_bio(u32 nr_vecs)
+{
+	struct bio *bio = bio_alloc(GFP_NOFS | __GFP_HIGH, nr_vecs);
+
+	if (!bio && (current->flags & PF_MEMALLOC)) {
+		while (!bio && (nr_vecs /= 2))
+			bio = bio_alloc(GFP_NOFS | __GFP_HIGH, nr_vecs);
+	}
+	return bio;
+}
+
+/*
+ * ntfs_bio_pages - Read/write pages from/to disk.
+ */
+int ntfs_bio_pages(struct ntfs_sb_info *sbi, const struct runs_tree *run,
+		   struct page **pages, u32 nr_pages, u64 vbo, u32 bytes,
+		   u32 op)
+{
+	int err = 0;
+	struct bio *new, *bio = NULL;
+	struct super_block *sb = sbi->sb;
+	struct block_device *bdev = sb->s_bdev;
+	struct page *page;
+	u8 cluster_bits = sbi->cluster_bits;
+	CLST lcn, clen, vcn, vcn_next;
+	u32 add, off, page_idx;
+	u64 lbo, len;
+	size_t run_idx;
+	struct blk_plug plug;
+
+	if (!bytes)
+		return 0;
+
+	blk_start_plug(&plug);
+
+	/* Align vbo and bytes to be 512 bytes aligned. */
+	lbo = (vbo + bytes + 511) & ~511ull;
+	vbo = vbo & ~511ull;
+	bytes = lbo - vbo;
+
+	vcn = vbo >> cluster_bits;
+	if (!run_lookup_entry(run, vcn, &lcn, &clen, &run_idx)) {
+		err = -ENOENT;
+		goto out;
+	}
+	off = vbo & sbi->cluster_mask;
+	page_idx = 0;
+	page = pages[0];
+
+	for (;;) {
+		lbo = ((u64)lcn << cluster_bits) + off;
+		len = ((u64)clen << cluster_bits) - off;
+new_bio:
+		new = ntfs_alloc_bio(nr_pages - page_idx);
+		if (!new) {
+			err = -ENOMEM;
+			goto out;
+		}
+		if (bio) {
+			bio_chain(bio, new);
+			submit_bio(bio);
+		}
+		bio = new;
+		bio_set_dev(bio, bdev);
+		bio->bi_iter.bi_sector = lbo >> 9;
+		bio->bi_opf = op;
+
+		while (len) {
+			off = vbo & (PAGE_SIZE - 1);
+			add = off + len > PAGE_SIZE ? (PAGE_SIZE - off) : len;
+
+			if (bio_add_page(bio, page, add, off) < add)
+				goto new_bio;
+
+			if (bytes <= add)
+				goto out;
+			bytes -= add;
+			vbo += add;
+
+			if (add + off == PAGE_SIZE) {
+				page_idx += 1;
+				if (WARN_ON(page_idx >= nr_pages)) {
+					err = -EINVAL;
+					goto out;
+				}
+				page = pages[page_idx];
+			}
+
+			if (len <= add)
+				break;
+			len -= add;
+			lbo += add;
+		}
+
+		vcn_next = vcn + clen;
+		if (!run_get_entry(run, ++run_idx, &vcn, &lcn, &clen) ||
+		    vcn != vcn_next) {
+			err = -ENOENT;
+			goto out;
+		}
+		off = 0;
+	}
+out:
+	if (bio) {
+		if (!err)
+			err = submit_bio_wait(bio);
+		bio_put(bio);
+	}
+	blk_finish_plug(&plug);
+
+	return err;
+}
+
+/*
+ * ntfs_bio_fill_1 - Helper for ntfs_loadlog_and_replay().
+ *
+ * Fill on-disk logfile range by (-1)
+ * this means empty logfile.
+ */
+int ntfs_bio_fill_1(struct ntfs_sb_info *sbi, const struct runs_tree *run)
+{
+	int err = 0;
+	struct super_block *sb = sbi->sb;
+	struct block_device *bdev = sb->s_bdev;
+	u8 cluster_bits = sbi->cluster_bits;
+	struct bio *new, *bio = NULL;
+	CLST lcn, clen;
+	u64 lbo, len;
+	size_t run_idx;
+	struct page *fill;
+	void *kaddr;
+	struct blk_plug plug;
+
+	fill = alloc_page(GFP_KERNEL);
+	if (!fill)
+		return -ENOMEM;
+
+	kaddr = kmap_atomic(fill);
+	memset(kaddr, -1, PAGE_SIZE);
+	kunmap_atomic(kaddr);
+	flush_dcache_page(fill);
+	lock_page(fill);
+
+	if (!run_lookup_entry(run, 0, &lcn, &clen, &run_idx)) {
+		err = -ENOENT;
+		goto out;
+	}
+
+	/*
+	 * TODO: Try blkdev_issue_write_same.
+	 */
+	blk_start_plug(&plug);
+	do {
+		lbo = (u64)lcn << cluster_bits;
+		len = (u64)clen << cluster_bits;
+new_bio:
+		new = ntfs_alloc_bio(BIO_MAX_VECS);
+		if (!new) {
+			err = -ENOMEM;
+			break;
+		}
+		if (bio) {
+			bio_chain(bio, new);
+			submit_bio(bio);
+		}
+		bio = new;
+		bio_set_dev(bio, bdev);
+		bio->bi_opf = REQ_OP_WRITE;
+		bio->bi_iter.bi_sector = lbo >> 9;
+
+		for (;;) {
+			u32 add = len > PAGE_SIZE ? PAGE_SIZE : len;
+
+			if (bio_add_page(bio, fill, add, 0) < add)
+				goto new_bio;
+
+			lbo += add;
+			if (len <= add)
+				break;
+			len -= add;
+		}
+	} while (run_get_entry(run, ++run_idx, NULL, &lcn, &clen));
+
+	if (bio) {
+		if (!err)
+			err = submit_bio_wait(bio);
+		bio_put(bio);
+	}
+	blk_finish_plug(&plug);
+out:
+	unlock_page(fill);
+	put_page(fill);
+
+	return err;
+}
+
+int ntfs_vbo_to_lbo(struct ntfs_sb_info *sbi, const struct runs_tree *run,
+		    u64 vbo, u64 *lbo, u64 *bytes)
+{
+	u32 off;
+	CLST lcn, len;
+	u8 cluster_bits = sbi->cluster_bits;
+
+	if (!run_lookup_entry(run, vbo >> cluster_bits, &lcn, &len, NULL))
+		return -ENOENT;
+
+	off = vbo & sbi->cluster_mask;
+	*lbo = lcn == SPARSE_LCN ? -1 : (((u64)lcn << cluster_bits) + off);
+	*bytes = ((u64)len << cluster_bits) - off;
+
+	return 0;
+}
+
+struct ntfs_inode *ntfs_new_inode(struct ntfs_sb_info *sbi, CLST rno, bool dir)
+{
+	int err = 0;
+	struct super_block *sb = sbi->sb;
+	struct inode *inode = new_inode(sb);
+	struct ntfs_inode *ni;
+
+	if (!inode)
+		return ERR_PTR(-ENOMEM);
+
+	ni = ntfs_i(inode);
+
+	err = mi_format_new(&ni->mi, sbi, rno, dir ? RECORD_FLAG_DIR : 0,
+			    false);
+	if (err)
+		goto out;
+
+	inode->i_ino = rno;
+	if (insert_inode_locked(inode) < 0) {
+		err = -EIO;
+		goto out;
+	}
+
+out:
+	if (err) {
+		iput(inode);
+		ni = ERR_PTR(err);
+	}
+	return ni;
+}
+
+/*
+ * O:BAG:BAD:(A;OICI;FA;;;WD)
+ * Owner S-1-5-32-544 (Administrators)
+ * Group S-1-5-32-544 (Administrators)
+ * ACE: allow S-1-1-0 (Everyone) with FILE_ALL_ACCESS
+ */
+const u8 s_default_security[] __aligned(8) = {
+	0x01, 0x00, 0x04, 0x80, 0x30, 0x00, 0x00, 0x00, 0x40, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0x02, 0x00, 0x1C, 0x00,
+	0x01, 0x00, 0x00, 0x00, 0x00, 0x03, 0x14, 0x00, 0xFF, 0x01, 0x1F, 0x00,
+	0x01, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x00,
+	0x01, 0x02, 0x00, 0x00, 0x00, 0x00, 0x00, 0x05, 0x20, 0x00, 0x00, 0x00,
+	0x20, 0x02, 0x00, 0x00, 0x01, 0x02, 0x00, 0x00, 0x00, 0x00, 0x00, 0x05,
+	0x20, 0x00, 0x00, 0x00, 0x20, 0x02, 0x00, 0x00,
+};
+
+static_assert(sizeof(s_default_security) == 0x50);
+
+static inline u32 sid_length(const struct SID *sid)
+{
+	return struct_size(sid, SubAuthority, sid->SubAuthorityCount);
+}
+
+/*
+ * is_acl_valid
+ *
+ * Thanks Mark Harmstone for idea.
+ */
+static bool is_acl_valid(const struct ACL *acl, u32 len)
+{
+	const struct ACE_HEADER *ace;
+	u32 i;
+	u16 ace_count, ace_size;
+
+	if (acl->AclRevision != ACL_REVISION &&
+	    acl->AclRevision != ACL_REVISION_DS) {
+		/*
+		 * This value should be ACL_REVISION, unless the ACL contains an
+		 * object-specific ACE, in which case this value must be ACL_REVISION_DS.
+		 * All ACEs in an ACL must be at the same revision level.
+		 */
+		return false;
+	}
+
+	if (acl->Sbz1)
+		return false;
+
+	if (le16_to_cpu(acl->AclSize) > len)
+		return false;
+
+	if (acl->Sbz2)
+		return false;
+
+	len -= sizeof(struct ACL);
+	ace = (struct ACE_HEADER *)&acl[1];
+	ace_count = le16_to_cpu(acl->AceCount);
+
+	for (i = 0; i < ace_count; i++) {
+		if (len < sizeof(struct ACE_HEADER))
+			return false;
+
+		ace_size = le16_to_cpu(ace->AceSize);
+		if (len < ace_size)
+			return false;
+
+		len -= ace_size;
+		ace = Add2Ptr(ace, ace_size);
+	}
+
+	return true;
+}
+
+bool is_sd_valid(const struct SECURITY_DESCRIPTOR_RELATIVE *sd, u32 len)
+{
+	u32 sd_owner, sd_group, sd_sacl, sd_dacl;
+
+	if (len < sizeof(struct SECURITY_DESCRIPTOR_RELATIVE))
+		return false;
+
+	if (sd->Revision != 1)
+		return false;
+
+	if (sd->Sbz1)
+		return false;
+
+	if (!(sd->Control & SE_SELF_RELATIVE))
+		return false;
+
+	sd_owner = le32_to_cpu(sd->Owner);
+	if (sd_owner) {
+		const struct SID *owner = Add2Ptr(sd, sd_owner);
+
+		if (sd_owner + offsetof(struct SID, SubAuthority) > len)
+			return false;
+
+		if (owner->Revision != 1)
+			return false;
+
+		if (sd_owner + sid_length(owner) > len)
+			return false;
+	}
+
+	sd_group = le32_to_cpu(sd->Group);
+	if (sd_group) {
+		const struct SID *group = Add2Ptr(sd, sd_group);
+
+		if (sd_group + offsetof(struct SID, SubAuthority) > len)
+			return false;
+
+		if (group->Revision != 1)
+			return false;
+
+		if (sd_group + sid_length(group) > len)
+			return false;
+	}
+
+	sd_sacl = le32_to_cpu(sd->Sacl);
+	if (sd_sacl) {
+		const struct ACL *sacl = Add2Ptr(sd, sd_sacl);
+
+		if (sd_sacl + sizeof(struct ACL) > len)
+			return false;
+
+		if (!is_acl_valid(sacl, len - sd_sacl))
+			return false;
+	}
+
+	sd_dacl = le32_to_cpu(sd->Dacl);
+	if (sd_dacl) {
+		const struct ACL *dacl = Add2Ptr(sd, sd_dacl);
+
+		if (sd_dacl + sizeof(struct ACL) > len)
+			return false;
+
+		if (!is_acl_valid(dacl, len - sd_dacl))
+			return false;
+	}
+
+	return true;
+}
+
+/*
+ * ntfs_security_init - Load and parse $Secure.
+ */
+int ntfs_security_init(struct ntfs_sb_info *sbi)
+{
+	int err;
+	struct super_block *sb = sbi->sb;
+	struct inode *inode;
+	struct ntfs_inode *ni;
+	struct MFT_REF ref;
+	struct ATTRIB *attr;
+	struct ATTR_LIST_ENTRY *le;
+	u64 sds_size;
+	size_t off;
+	struct NTFS_DE *ne;
+	struct NTFS_DE_SII *sii_e;
+	struct ntfs_fnd *fnd_sii = NULL;
+	const struct INDEX_ROOT *root_sii;
+	const struct INDEX_ROOT *root_sdh;
+	struct ntfs_index *indx_sdh = &sbi->security.index_sdh;
+	struct ntfs_index *indx_sii = &sbi->security.index_sii;
+
+	ref.low = cpu_to_le32(MFT_REC_SECURE);
+	ref.high = 0;
+	ref.seq = cpu_to_le16(MFT_REC_SECURE);
+
+	inode = ntfs_iget5(sb, &ref, &NAME_SECURE);
+	if (IS_ERR(inode)) {
+		err = PTR_ERR(inode);
+		ntfs_err(sb, "Failed to load $Secure.");
+		inode = NULL;
+		goto out;
+	}
+
+	ni = ntfs_i(inode);
+
+	le = NULL;
+
+	attr = ni_find_attr(ni, NULL, &le, ATTR_ROOT, SDH_NAME,
+			    ARRAY_SIZE(SDH_NAME), NULL, NULL);
+	if (!attr) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	root_sdh = resident_data(attr);
+	if (root_sdh->type != ATTR_ZERO ||
+	    root_sdh->rule != NTFS_COLLATION_TYPE_SECURITY_HASH) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	err = indx_init(indx_sdh, sbi, attr, INDEX_MUTEX_SDH);
+	if (err)
+		goto out;
+
+	attr = ni_find_attr(ni, attr, &le, ATTR_ROOT, SII_NAME,
+			    ARRAY_SIZE(SII_NAME), NULL, NULL);
+	if (!attr) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	root_sii = resident_data(attr);
+	if (root_sii->type != ATTR_ZERO ||
+	    root_sii->rule != NTFS_COLLATION_TYPE_UINT) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	err = indx_init(indx_sii, sbi, attr, INDEX_MUTEX_SII);
+	if (err)
+		goto out;
+
+	fnd_sii = fnd_get();
+	if (!fnd_sii) {
+		err = -ENOMEM;
+		goto out;
+	}
+
+	sds_size = inode->i_size;
+
+	/* Find the last valid Id. */
+	sbi->security.next_id = SECURITY_ID_FIRST;
+	/* Always write new security at the end of bucket. */
+	sbi->security.next_off =
+		ALIGN(sds_size - SecurityDescriptorsBlockSize, 16);
+
+	off = 0;
+	ne = NULL;
+
+	for (;;) {
+		u32 next_id;
+
+		err = indx_find_raw(indx_sii, ni, root_sii, &ne, &off, fnd_sii);
+		if (err || !ne)
+			break;
+
+		sii_e = (struct NTFS_DE_SII *)ne;
+		if (le16_to_cpu(ne->view.data_size) < SIZEOF_SECURITY_HDR)
+			continue;
+
+		next_id = le32_to_cpu(sii_e->sec_id) + 1;
+		if (next_id >= sbi->security.next_id)
+			sbi->security.next_id = next_id;
+	}
+
+	sbi->security.ni = ni;
+	inode = NULL;
+out:
+	iput(inode);
+	fnd_put(fnd_sii);
+
+	return err;
+}
+
+/*
+ * ntfs_get_security_by_id - Read security descriptor by id.
+ */
+int ntfs_get_security_by_id(struct ntfs_sb_info *sbi, __le32 security_id,
+			    struct SECURITY_DESCRIPTOR_RELATIVE **sd,
+			    size_t *size)
+{
+	int err;
+	int diff;
+	struct ntfs_inode *ni = sbi->security.ni;
+	struct ntfs_index *indx = &sbi->security.index_sii;
+	void *p = NULL;
+	struct NTFS_DE_SII *sii_e;
+	struct ntfs_fnd *fnd_sii;
+	struct SECURITY_HDR d_security;
+	const struct INDEX_ROOT *root_sii;
+	u32 t32;
+
+	*sd = NULL;
+
+	mutex_lock_nested(&ni->ni_lock, NTFS_INODE_MUTEX_SECURITY);
+
+	fnd_sii = fnd_get();
+	if (!fnd_sii) {
+		err = -ENOMEM;
+		goto out;
+	}
+
+	root_sii = indx_get_root(indx, ni, NULL, NULL);
+	if (!root_sii) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	/* Try to find this SECURITY descriptor in SII indexes. */
+	err = indx_find(indx, ni, root_sii, &security_id, sizeof(security_id),
+			NULL, &diff, (struct NTFS_DE **)&sii_e, fnd_sii);
+	if (err)
+		goto out;
+
+	if (diff)
+		goto out;
+
+	t32 = le32_to_cpu(sii_e->sec_hdr.size);
+	if (t32 < SIZEOF_SECURITY_HDR) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	if (t32 > SIZEOF_SECURITY_HDR + 0x10000) {
+		/* Looks like too big security. 0x10000 - is arbitrary big number. */
+		err = -EFBIG;
+		goto out;
+	}
+
+	*size = t32 - SIZEOF_SECURITY_HDR;
+
+	p = kmalloc(*size, GFP_NOFS);
+	if (!p) {
+		err = -ENOMEM;
+		goto out;
+	}
+
+	err = ntfs_read_run_nb(sbi, &ni->file.run,
+			       le64_to_cpu(sii_e->sec_hdr.off), &d_security,
+			       sizeof(d_security), NULL);
+	if (err)
+		goto out;
+
+	if (memcmp(&d_security, &sii_e->sec_hdr, SIZEOF_SECURITY_HDR)) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	err = ntfs_read_run_nb(sbi, &ni->file.run,
+			       le64_to_cpu(sii_e->sec_hdr.off) +
+				       SIZEOF_SECURITY_HDR,
+			       p, *size, NULL);
+	if (err)
+		goto out;
+
+	*sd = p;
+	p = NULL;
+
+out:
+	kfree(p);
+	fnd_put(fnd_sii);
+	ni_unlock(ni);
+
+	return err;
+}
+
+/*
+ * ntfs_insert_security - Insert security descriptor into $Secure::SDS.
+ *
+ * SECURITY Descriptor Stream data is organized into chunks of 256K bytes
+ * and it contains a mirror copy of each security descriptor.  When writing
+ * to a security descriptor at location X, another copy will be written at
+ * location (X+256K).
+ * When writing a security descriptor that will cross the 256K boundary,
+ * the pointer will be advanced by 256K to skip
+ * over the mirror portion.
+ */
+int ntfs_insert_security(struct ntfs_sb_info *sbi,
+			 const struct SECURITY_DESCRIPTOR_RELATIVE *sd,
+			 u32 size_sd, __le32 *security_id, bool *inserted)
+{
+	int err, diff;
+	struct ntfs_inode *ni = sbi->security.ni;
+	struct ntfs_index *indx_sdh = &sbi->security.index_sdh;
+	struct ntfs_index *indx_sii = &sbi->security.index_sii;
+	struct NTFS_DE_SDH *e;
+	struct NTFS_DE_SDH sdh_e;
+	struct NTFS_DE_SII sii_e;
+	struct SECURITY_HDR *d_security;
+	u32 new_sec_size = size_sd + SIZEOF_SECURITY_HDR;
+	u32 aligned_sec_size = ALIGN(new_sec_size, 16);
+	struct SECURITY_KEY hash_key;
+	struct ntfs_fnd *fnd_sdh = NULL;
+	const struct INDEX_ROOT *root_sdh;
+	const struct INDEX_ROOT *root_sii;
+	u64 mirr_off, new_sds_size;
+	u32 next, left;
+
+	static_assert((1 << Log2OfSecurityDescriptorsBlockSize) ==
+		      SecurityDescriptorsBlockSize);
+
+	hash_key.hash = security_hash(sd, size_sd);
+	hash_key.sec_id = SECURITY_ID_INVALID;
+
+	if (inserted)
+		*inserted = false;
+	*security_id = SECURITY_ID_INVALID;
+
+	/* Allocate a temporal buffer. */
+	d_security = kzalloc(aligned_sec_size, GFP_NOFS);
+	if (!d_security)
+		return -ENOMEM;
+
+	mutex_lock_nested(&ni->ni_lock, NTFS_INODE_MUTEX_SECURITY);
+
+	fnd_sdh = fnd_get();
+	if (!fnd_sdh) {
+		err = -ENOMEM;
+		goto out;
+	}
+
+	root_sdh = indx_get_root(indx_sdh, ni, NULL, NULL);
+	if (!root_sdh) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	root_sii = indx_get_root(indx_sii, ni, NULL, NULL);
+	if (!root_sii) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	/*
+	 * Check if such security already exists.
+	 * Use "SDH" and hash -> to get the offset in "SDS".
+	 */
+	err = indx_find(indx_sdh, ni, root_sdh, &hash_key, sizeof(hash_key),
+			&d_security->key.sec_id, &diff, (struct NTFS_DE **)&e,
+			fnd_sdh);
+	if (err)
+		goto out;
+
+	while (e) {
+		if (le32_to_cpu(e->sec_hdr.size) == new_sec_size) {
+			err = ntfs_read_run_nb(sbi, &ni->file.run,
+					       le64_to_cpu(e->sec_hdr.off),
+					       d_security, new_sec_size, NULL);
+			if (err)
+				goto out;
+
+			if (le32_to_cpu(d_security->size) == new_sec_size &&
+			    d_security->key.hash == hash_key.hash &&
+			    !memcmp(d_security + 1, sd, size_sd)) {
+				*security_id = d_security->key.sec_id;
+				/* Such security already exists. */
+				err = 0;
+				goto out;
+			}
+		}
+
+		err = indx_find_sort(indx_sdh, ni, root_sdh,
+				     (struct NTFS_DE **)&e, fnd_sdh);
+		if (err)
+			goto out;
+
+		if (!e || e->key.hash != hash_key.hash)
+			break;
+	}
+
+	/* Zero unused space. */
+	next = sbi->security.next_off & (SecurityDescriptorsBlockSize - 1);
+	left = SecurityDescriptorsBlockSize - next;
+
+	/* Zero gap until SecurityDescriptorsBlockSize. */
+	if (left < new_sec_size) {
+		/* Zero "left" bytes from sbi->security.next_off. */
+		sbi->security.next_off += SecurityDescriptorsBlockSize + left;
+	}
+
+	/* Zero tail of previous security. */
+	//used = ni->vfs_inode.i_size & (SecurityDescriptorsBlockSize - 1);
+
+	/*
+	 * Example:
+	 * 0x40438 == ni->vfs_inode.i_size
+	 * 0x00440 == sbi->security.next_off
+	 * need to zero [0x438-0x440)
+	 * if (next > used) {
+	 *  u32 tozero = next - used;
+	 *  zero "tozero" bytes from sbi->security.next_off - tozero
+	 */
+
+	/* Format new security descriptor. */
+	d_security->key.hash = hash_key.hash;
+	d_security->key.sec_id = cpu_to_le32(sbi->security.next_id);
+	d_security->off = cpu_to_le64(sbi->security.next_off);
+	d_security->size = cpu_to_le32(new_sec_size);
+	memcpy(d_security + 1, sd, size_sd);
+
+	/* Write main SDS bucket. */
+	err = ntfs_sb_write_run(sbi, &ni->file.run, sbi->security.next_off,
+				d_security, aligned_sec_size);
+
+	if (err)
+		goto out;
+
+	mirr_off = sbi->security.next_off + SecurityDescriptorsBlockSize;
+	new_sds_size = mirr_off + aligned_sec_size;
+
+	if (new_sds_size > ni->vfs_inode.i_size) {
+		err = attr_set_size(ni, ATTR_DATA, SDS_NAME,
+				    ARRAY_SIZE(SDS_NAME), &ni->file.run,
+				    new_sds_size, &new_sds_size, false, NULL);
+		if (err)
+			goto out;
+	}
+
+	/* Write copy SDS bucket. */
+	err = ntfs_sb_write_run(sbi, &ni->file.run, mirr_off, d_security,
+				aligned_sec_size);
+	if (err)
+		goto out;
+
+	/* Fill SII entry. */
+	sii_e.de.view.data_off =
+		cpu_to_le16(offsetof(struct NTFS_DE_SII, sec_hdr));
+	sii_e.de.view.data_size = cpu_to_le16(SIZEOF_SECURITY_HDR);
+	sii_e.de.view.res = 0;
+	sii_e.de.size = cpu_to_le16(SIZEOF_SII_DIRENTRY);
+	sii_e.de.key_size = cpu_to_le16(sizeof(d_security->key.sec_id));
+	sii_e.de.flags = 0;
+	sii_e.de.res = 0;
+	sii_e.sec_id = d_security->key.sec_id;
+	memcpy(&sii_e.sec_hdr, d_security, SIZEOF_SECURITY_HDR);
+
+	err = indx_insert_entry(indx_sii, ni, &sii_e.de, NULL, NULL, 0);
+	if (err)
+		goto out;
+
+	/* Fill SDH entry. */
+	sdh_e.de.view.data_off =
+		cpu_to_le16(offsetof(struct NTFS_DE_SDH, sec_hdr));
+	sdh_e.de.view.data_size = cpu_to_le16(SIZEOF_SECURITY_HDR);
+	sdh_e.de.view.res = 0;
+	sdh_e.de.size = cpu_to_le16(SIZEOF_SDH_DIRENTRY);
+	sdh_e.de.key_size = cpu_to_le16(sizeof(sdh_e.key));
+	sdh_e.de.flags = 0;
+	sdh_e.de.res = 0;
+	sdh_e.key.hash = d_security->key.hash;
+	sdh_e.key.sec_id = d_security->key.sec_id;
+	memcpy(&sdh_e.sec_hdr, d_security, SIZEOF_SECURITY_HDR);
+	sdh_e.magic[0] = cpu_to_le16('I');
+	sdh_e.magic[1] = cpu_to_le16('I');
+
+	fnd_clear(fnd_sdh);
+	err = indx_insert_entry(indx_sdh, ni, &sdh_e.de, (void *)(size_t)1,
+				fnd_sdh, 0);
+	if (err)
+		goto out;
+
+	*security_id = d_security->key.sec_id;
+	if (inserted)
+		*inserted = true;
+
+	/* Update Id and offset for next descriptor. */
+	sbi->security.next_id += 1;
+	sbi->security.next_off += aligned_sec_size;
+
+out:
+	fnd_put(fnd_sdh);
+	mark_inode_dirty(&ni->vfs_inode);
+	ni_unlock(ni);
+	kfree(d_security);
+
+	return err;
+}
+
+/*
+ * ntfs_reparse_init - Load and parse $Extend/$Reparse.
+ */
+int ntfs_reparse_init(struct ntfs_sb_info *sbi)
+{
+	int err;
+	struct ntfs_inode *ni = sbi->reparse.ni;
+	struct ntfs_index *indx = &sbi->reparse.index_r;
+	struct ATTRIB *attr;
+	struct ATTR_LIST_ENTRY *le;
+	const struct INDEX_ROOT *root_r;
+
+	if (!ni)
+		return 0;
+
+	le = NULL;
+	attr = ni_find_attr(ni, NULL, &le, ATTR_ROOT, SR_NAME,
+			    ARRAY_SIZE(SR_NAME), NULL, NULL);
+	if (!attr) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	root_r = resident_data(attr);
+	if (root_r->type != ATTR_ZERO ||
+	    root_r->rule != NTFS_COLLATION_TYPE_UINTS) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	err = indx_init(indx, sbi, attr, INDEX_MUTEX_SR);
+	if (err)
+		goto out;
+
+out:
+	return err;
+}
+
+/*
+ * ntfs_objid_init - Load and parse $Extend/$ObjId.
+ */
+int ntfs_objid_init(struct ntfs_sb_info *sbi)
+{
+	int err;
+	struct ntfs_inode *ni = sbi->objid.ni;
+	struct ntfs_index *indx = &sbi->objid.index_o;
+	struct ATTRIB *attr;
+	struct ATTR_LIST_ENTRY *le;
+	const struct INDEX_ROOT *root;
+
+	if (!ni)
+		return 0;
+
+	le = NULL;
+	attr = ni_find_attr(ni, NULL, &le, ATTR_ROOT, SO_NAME,
+			    ARRAY_SIZE(SO_NAME), NULL, NULL);
+	if (!attr) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	root = resident_data(attr);
+	if (root->type != ATTR_ZERO ||
+	    root->rule != NTFS_COLLATION_TYPE_UINTS) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	err = indx_init(indx, sbi, attr, INDEX_MUTEX_SO);
+	if (err)
+		goto out;
+
+out:
+	return err;
+}
+
+int ntfs_objid_remove(struct ntfs_sb_info *sbi, struct GUID *guid)
+{
+	int err;
+	struct ntfs_inode *ni = sbi->objid.ni;
+	struct ntfs_index *indx = &sbi->objid.index_o;
+
+	if (!ni)
+		return -EINVAL;
+
+	mutex_lock_nested(&ni->ni_lock, NTFS_INODE_MUTEX_OBJID);
+
+	err = indx_delete_entry(indx, ni, guid, sizeof(*guid), NULL);
+
+	mark_inode_dirty(&ni->vfs_inode);
+	ni_unlock(ni);
+
+	return err;
+}
+
+int ntfs_insert_reparse(struct ntfs_sb_info *sbi, __le32 rtag,
+			const struct MFT_REF *ref)
+{
+	int err;
+	struct ntfs_inode *ni = sbi->reparse.ni;
+	struct ntfs_index *indx = &sbi->reparse.index_r;
+	struct NTFS_DE_R re;
+
+	if (!ni)
+		return -EINVAL;
+
+	memset(&re, 0, sizeof(re));
+
+	re.de.view.data_off = cpu_to_le16(offsetof(struct NTFS_DE_R, zero));
+	re.de.size = cpu_to_le16(sizeof(struct NTFS_DE_R));
+	re.de.key_size = cpu_to_le16(sizeof(re.key));
+
+	re.key.ReparseTag = rtag;
+	memcpy(&re.key.ref, ref, sizeof(*ref));
+
+	mutex_lock_nested(&ni->ni_lock, NTFS_INODE_MUTEX_REPARSE);
+
+	err = indx_insert_entry(indx, ni, &re.de, NULL, NULL, 0);
+
+	mark_inode_dirty(&ni->vfs_inode);
+	ni_unlock(ni);
+
+	return err;
+}
+
+int ntfs_remove_reparse(struct ntfs_sb_info *sbi, __le32 rtag,
+			const struct MFT_REF *ref)
+{
+	int err, diff;
+	struct ntfs_inode *ni = sbi->reparse.ni;
+	struct ntfs_index *indx = &sbi->reparse.index_r;
+	struct ntfs_fnd *fnd = NULL;
+	struct REPARSE_KEY rkey;
+	struct NTFS_DE_R *re;
+	struct INDEX_ROOT *root_r;
+
+	if (!ni)
+		return -EINVAL;
+
+	rkey.ReparseTag = rtag;
+	rkey.ref = *ref;
+
+	mutex_lock_nested(&ni->ni_lock, NTFS_INODE_MUTEX_REPARSE);
+
+	if (rtag) {
+		err = indx_delete_entry(indx, ni, &rkey, sizeof(rkey), NULL);
+		goto out1;
+	}
+
+	fnd = fnd_get();
+	if (!fnd) {
+		err = -ENOMEM;
+		goto out1;
+	}
+
+	root_r = indx_get_root(indx, ni, NULL, NULL);
+	if (!root_r) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	/* 1 - forces to ignore rkey.ReparseTag when comparing keys. */
+	err = indx_find(indx, ni, root_r, &rkey, sizeof(rkey), (void *)1, &diff,
+			(struct NTFS_DE **)&re, fnd);
+	if (err)
+		goto out;
+
+	if (memcmp(&re->key.ref, ref, sizeof(*ref))) {
+		/* Impossible. Looks like volume corrupt? */
+		goto out;
+	}
+
+	memcpy(&rkey, &re->key, sizeof(rkey));
+
+	fnd_put(fnd);
+	fnd = NULL;
+
+	err = indx_delete_entry(indx, ni, &rkey, sizeof(rkey), NULL);
+	if (err)
+		goto out;
+
+out:
+	fnd_put(fnd);
+
+out1:
+	mark_inode_dirty(&ni->vfs_inode);
+	ni_unlock(ni);
+
+	return err;
+}
+
+static inline void ntfs_unmap_and_discard(struct ntfs_sb_info *sbi, CLST lcn,
+					  CLST len)
+{
+	ntfs_unmap_meta(sbi->sb, lcn, len);
+	ntfs_discard(sbi, lcn, len);
+}
+
+void mark_as_free_ex(struct ntfs_sb_info *sbi, CLST lcn, CLST len, bool trim)
+{
+	CLST end, i;
+	struct wnd_bitmap *wnd = &sbi->used.bitmap;
+
+	down_write_nested(&wnd->rw_lock, BITMAP_MUTEX_CLUSTERS);
+	if (!wnd_is_used(wnd, lcn, len)) {
+		ntfs_set_state(sbi, NTFS_DIRTY_ERROR);
+
+		end = lcn + len;
+		len = 0;
+		for (i = lcn; i < end; i++) {
+			if (wnd_is_used(wnd, i, 1)) {
+				if (!len)
+					lcn = i;
+				len += 1;
+				continue;
+			}
+
+			if (!len)
+				continue;
+
+			if (trim)
+				ntfs_unmap_and_discard(sbi, lcn, len);
+
+			wnd_set_free(wnd, lcn, len);
+			len = 0;
+		}
+
+		if (!len)
+			goto out;
+	}
+
+	if (trim)
+		ntfs_unmap_and_discard(sbi, lcn, len);
+	wnd_set_free(wnd, lcn, len);
+
+out:
+	up_write(&wnd->rw_lock);
+}
+
+/*
+ * run_deallocate - Deallocate clusters.
+ */
+int run_deallocate(struct ntfs_sb_info *sbi, struct runs_tree *run, bool trim)
+{
+	CLST lcn, len;
+	size_t idx = 0;
+
+	while (run_get_entry(run, idx++, NULL, &lcn, &len)) {
+		if (lcn == SPARSE_LCN)
+			continue;
+
+		mark_as_free_ex(sbi, lcn, len, trim);
+	}
+
+	return 0;
+}
diff --git a/fs/ntfs3/index.c b/fs/ntfs3/index.c
new file mode 100644
index 000000000000..0daca9adc54c
--- /dev/null
+++ b/fs/ntfs3/index.c
@@ -0,0 +1,2650 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ *
+ * Copyright (C) 2019-2021 Paragon Software GmbH, All rights reserved.
+ *
+ */
+
+#include <linux/blkdev.h>
+#include <linux/buffer_head.h>
+#include <linux/fs.h>
+#include <linux/nls.h>
+
+#include "debug.h"
+#include "ntfs.h"
+#include "ntfs_fs.h"
+
+static const struct INDEX_NAMES {
+	const __le16 *name;
+	u8 name_len;
+} s_index_names[INDEX_MUTEX_TOTAL] = {
+	{ I30_NAME, ARRAY_SIZE(I30_NAME) }, { SII_NAME, ARRAY_SIZE(SII_NAME) },
+	{ SDH_NAME, ARRAY_SIZE(SDH_NAME) }, { SO_NAME, ARRAY_SIZE(SO_NAME) },
+	{ SQ_NAME, ARRAY_SIZE(SQ_NAME) },   { SR_NAME, ARRAY_SIZE(SR_NAME) },
+};
+
+/*
+ * cmp_fnames - Compare two names in index.
+ *
+ * if l1 != 0
+ *   Both names are little endian on-disk ATTR_FILE_NAME structs.
+ * else
+ *   key1 - cpu_str, key2 - ATTR_FILE_NAME
+ */
+static int cmp_fnames(const void *key1, size_t l1, const void *key2, size_t l2,
+		      const void *data)
+{
+	const struct ATTR_FILE_NAME *f2 = key2;
+	const struct ntfs_sb_info *sbi = data;
+	const struct ATTR_FILE_NAME *f1;
+	u16 fsize2;
+	bool both_case;
+
+	if (l2 <= offsetof(struct ATTR_FILE_NAME, name))
+		return -1;
+
+	fsize2 = fname_full_size(f2);
+	if (l2 < fsize2)
+		return -1;
+
+	both_case = f2->type != FILE_NAME_DOS /*&& !sbi->options.nocase*/;
+	if (!l1) {
+		const struct le_str *s2 = (struct le_str *)&f2->name_len;
+
+		/*
+		 * If names are equal (case insensitive)
+		 * try to compare it case sensitive.
+		 */
+		return ntfs_cmp_names_cpu(key1, s2, sbi->upcase, both_case);
+	}
+
+	f1 = key1;
+	return ntfs_cmp_names(f1->name, f1->name_len, f2->name, f2->name_len,
+			      sbi->upcase, both_case);
+}
+
+/*
+ * cmp_uint - $SII of $Secure and $Q of Quota
+ */
+static int cmp_uint(const void *key1, size_t l1, const void *key2, size_t l2,
+		    const void *data)
+{
+	const u32 *k1 = key1;
+	const u32 *k2 = key2;
+
+	if (l2 < sizeof(u32))
+		return -1;
+
+	if (*k1 < *k2)
+		return -1;
+	if (*k1 > *k2)
+		return 1;
+	return 0;
+}
+
+/*
+ * cmp_sdh - $SDH of $Secure
+ */
+static int cmp_sdh(const void *key1, size_t l1, const void *key2, size_t l2,
+		   const void *data)
+{
+	const struct SECURITY_KEY *k1 = key1;
+	const struct SECURITY_KEY *k2 = key2;
+	u32 t1, t2;
+
+	if (l2 < sizeof(struct SECURITY_KEY))
+		return -1;
+
+	t1 = le32_to_cpu(k1->hash);
+	t2 = le32_to_cpu(k2->hash);
+
+	/* First value is a hash value itself. */
+	if (t1 < t2)
+		return -1;
+	if (t1 > t2)
+		return 1;
+
+	/* Second value is security Id. */
+	if (data) {
+		t1 = le32_to_cpu(k1->sec_id);
+		t2 = le32_to_cpu(k2->sec_id);
+		if (t1 < t2)
+			return -1;
+		if (t1 > t2)
+			return 1;
+	}
+
+	return 0;
+}
+
+/*
+ * cmp_uints - $O of ObjId and "$R" for Reparse.
+ */
+static int cmp_uints(const void *key1, size_t l1, const void *key2, size_t l2,
+		     const void *data)
+{
+	const __le32 *k1 = key1;
+	const __le32 *k2 = key2;
+	size_t count;
+
+	if ((size_t)data == 1) {
+		/*
+		 * ni_delete_all -> ntfs_remove_reparse ->
+		 * delete all with this reference.
+		 * k1, k2 - pointers to REPARSE_KEY
+		 */
+
+		k1 += 1; // Skip REPARSE_KEY.ReparseTag
+		k2 += 1; // Skip REPARSE_KEY.ReparseTag
+		if (l2 <= sizeof(int))
+			return -1;
+		l2 -= sizeof(int);
+		if (l1 <= sizeof(int))
+			return 1;
+		l1 -= sizeof(int);
+	}
+
+	if (l2 < sizeof(int))
+		return -1;
+
+	for (count = min(l1, l2) >> 2; count > 0; --count, ++k1, ++k2) {
+		u32 t1 = le32_to_cpu(*k1);
+		u32 t2 = le32_to_cpu(*k2);
+
+		if (t1 > t2)
+			return 1;
+		if (t1 < t2)
+			return -1;
+	}
+
+	if (l1 > l2)
+		return 1;
+	if (l1 < l2)
+		return -1;
+
+	return 0;
+}
+
+static inline NTFS_CMP_FUNC get_cmp_func(const struct INDEX_ROOT *root)
+{
+	switch (root->type) {
+	case ATTR_NAME:
+		if (root->rule == NTFS_COLLATION_TYPE_FILENAME)
+			return &cmp_fnames;
+		break;
+	case ATTR_ZERO:
+		switch (root->rule) {
+		case NTFS_COLLATION_TYPE_UINT:
+			return &cmp_uint;
+		case NTFS_COLLATION_TYPE_SECURITY_HASH:
+			return &cmp_sdh;
+		case NTFS_COLLATION_TYPE_UINTS:
+			return &cmp_uints;
+		default:
+			break;
+		}
+		break;
+	default:
+		break;
+	}
+
+	return NULL;
+}
+
+struct bmp_buf {
+	struct ATTRIB *b;
+	struct mft_inode *mi;
+	struct buffer_head *bh;
+	ulong *buf;
+	size_t bit;
+	u32 nbits;
+	u64 new_valid;
+};
+
+static int bmp_buf_get(struct ntfs_index *indx, struct ntfs_inode *ni,
+		       size_t bit, struct bmp_buf *bbuf)
+{
+	struct ATTRIB *b;
+	size_t data_size, valid_size, vbo, off = bit >> 3;
+	struct ntfs_sb_info *sbi = ni->mi.sbi;
+	CLST vcn = off >> sbi->cluster_bits;
+	struct ATTR_LIST_ENTRY *le = NULL;
+	struct buffer_head *bh;
+	struct super_block *sb;
+	u32 blocksize;
+	const struct INDEX_NAMES *in = &s_index_names[indx->type];
+
+	bbuf->bh = NULL;
+
+	b = ni_find_attr(ni, NULL, &le, ATTR_BITMAP, in->name, in->name_len,
+			 &vcn, &bbuf->mi);
+	bbuf->b = b;
+	if (!b)
+		return -EINVAL;
+
+	if (!b->non_res) {
+		data_size = le32_to_cpu(b->res.data_size);
+
+		if (off >= data_size)
+			return -EINVAL;
+
+		bbuf->buf = (ulong *)resident_data(b);
+		bbuf->bit = 0;
+		bbuf->nbits = data_size * 8;
+
+		return 0;
+	}
+
+	data_size = le64_to_cpu(b->nres.data_size);
+	if (WARN_ON(off >= data_size)) {
+		/* Looks like filesystem error. */
+		return -EINVAL;
+	}
+
+	valid_size = le64_to_cpu(b->nres.valid_size);
+
+	bh = ntfs_bread_run(sbi, &indx->bitmap_run, off);
+	if (!bh)
+		return -EIO;
+
+	if (IS_ERR(bh))
+		return PTR_ERR(bh);
+
+	bbuf->bh = bh;
+
+	if (buffer_locked(bh))
+		__wait_on_buffer(bh);
+
+	lock_buffer(bh);
+
+	sb = sbi->sb;
+	blocksize = sb->s_blocksize;
+
+	vbo = off & ~(size_t)sbi->block_mask;
+
+	bbuf->new_valid = vbo + blocksize;
+	if (bbuf->new_valid <= valid_size)
+		bbuf->new_valid = 0;
+	else if (bbuf->new_valid > data_size)
+		bbuf->new_valid = data_size;
+
+	if (vbo >= valid_size) {
+		memset(bh->b_data, 0, blocksize);
+	} else if (vbo + blocksize > valid_size) {
+		u32 voff = valid_size & sbi->block_mask;
+
+		memset(bh->b_data + voff, 0, blocksize - voff);
+	}
+
+	bbuf->buf = (ulong *)bh->b_data;
+	bbuf->bit = 8 * (off & ~(size_t)sbi->block_mask);
+	bbuf->nbits = 8 * blocksize;
+
+	return 0;
+}
+
+static void bmp_buf_put(struct bmp_buf *bbuf, bool dirty)
+{
+	struct buffer_head *bh = bbuf->bh;
+	struct ATTRIB *b = bbuf->b;
+
+	if (!bh) {
+		if (b && !b->non_res && dirty)
+			bbuf->mi->dirty = true;
+		return;
+	}
+
+	if (!dirty)
+		goto out;
+
+	if (bbuf->new_valid) {
+		b->nres.valid_size = cpu_to_le64(bbuf->new_valid);
+		bbuf->mi->dirty = true;
+	}
+
+	set_buffer_uptodate(bh);
+	mark_buffer_dirty(bh);
+
+out:
+	unlock_buffer(bh);
+	put_bh(bh);
+}
+
+/*
+ * indx_mark_used - Mark the bit @bit as used.
+ */
+static int indx_mark_used(struct ntfs_index *indx, struct ntfs_inode *ni,
+			  size_t bit)
+{
+	int err;
+	struct bmp_buf bbuf;
+
+	err = bmp_buf_get(indx, ni, bit, &bbuf);
+	if (err)
+		return err;
+
+	__set_bit(bit - bbuf.bit, bbuf.buf);
+
+	bmp_buf_put(&bbuf, true);
+
+	return 0;
+}
+
+/*
+ * indx_mark_free - Mark the bit @bit as free.
+ */
+static int indx_mark_free(struct ntfs_index *indx, struct ntfs_inode *ni,
+			  size_t bit)
+{
+	int err;
+	struct bmp_buf bbuf;
+
+	err = bmp_buf_get(indx, ni, bit, &bbuf);
+	if (err)
+		return err;
+
+	__clear_bit(bit - bbuf.bit, bbuf.buf);
+
+	bmp_buf_put(&bbuf, true);
+
+	return 0;
+}
+
+/*
+ * scan_nres_bitmap
+ *
+ * If ntfs_readdir calls this function (indx_used_bit -> scan_nres_bitmap),
+ * inode is shared locked and no ni_lock.
+ * Use rw_semaphore for read/write access to bitmap_run.
+ */
+static int scan_nres_bitmap(struct ntfs_inode *ni, struct ATTRIB *bitmap,
+			    struct ntfs_index *indx, size_t from,
+			    bool (*fn)(const ulong *buf, u32 bit, u32 bits,
+				       size_t *ret),
+			    size_t *ret)
+{
+	struct ntfs_sb_info *sbi = ni->mi.sbi;
+	struct super_block *sb = sbi->sb;
+	struct runs_tree *run = &indx->bitmap_run;
+	struct rw_semaphore *lock = &indx->run_lock;
+	u32 nbits = sb->s_blocksize * 8;
+	u32 blocksize = sb->s_blocksize;
+	u64 valid_size = le64_to_cpu(bitmap->nres.valid_size);
+	u64 data_size = le64_to_cpu(bitmap->nres.data_size);
+	sector_t eblock = bytes_to_block(sb, data_size);
+	size_t vbo = from >> 3;
+	sector_t blk = (vbo & sbi->cluster_mask) >> sb->s_blocksize_bits;
+	sector_t vblock = vbo >> sb->s_blocksize_bits;
+	sector_t blen, block;
+	CLST lcn, clen, vcn, vcn_next;
+	size_t idx;
+	struct buffer_head *bh;
+	bool ok;
+
+	*ret = MINUS_ONE_T;
+
+	if (vblock >= eblock)
+		return 0;
+
+	from &= nbits - 1;
+	vcn = vbo >> sbi->cluster_bits;
+
+	down_read(lock);
+	ok = run_lookup_entry(run, vcn, &lcn, &clen, &idx);
+	up_read(lock);
+
+next_run:
+	if (!ok) {
+		int err;
+		const struct INDEX_NAMES *name = &s_index_names[indx->type];
+
+		down_write(lock);
+		err = attr_load_runs_vcn(ni, ATTR_BITMAP, name->name,
+					 name->name_len, run, vcn);
+		up_write(lock);
+		if (err)
+			return err;
+		down_read(lock);
+		ok = run_lookup_entry(run, vcn, &lcn, &clen, &idx);
+		up_read(lock);
+		if (!ok)
+			return -EINVAL;
+	}
+
+	blen = (sector_t)clen * sbi->blocks_per_cluster;
+	block = (sector_t)lcn * sbi->blocks_per_cluster;
+
+	for (; blk < blen; blk++, from = 0) {
+		bh = ntfs_bread(sb, block + blk);
+		if (!bh)
+			return -EIO;
+
+		vbo = (u64)vblock << sb->s_blocksize_bits;
+		if (vbo >= valid_size) {
+			memset(bh->b_data, 0, blocksize);
+		} else if (vbo + blocksize > valid_size) {
+			u32 voff = valid_size & sbi->block_mask;
+
+			memset(bh->b_data + voff, 0, blocksize - voff);
+		}
+
+		if (vbo + blocksize > data_size)
+			nbits = 8 * (data_size - vbo);
+
+		ok = nbits > from ? (*fn)((ulong *)bh->b_data, from, nbits, ret)
+				  : false;
+		put_bh(bh);
+
+		if (ok) {
+			*ret += 8 * vbo;
+			return 0;
+		}
+
+		if (++vblock >= eblock) {
+			*ret = MINUS_ONE_T;
+			return 0;
+		}
+	}
+	blk = 0;
+	vcn_next = vcn + clen;
+	down_read(lock);
+	ok = run_get_entry(run, ++idx, &vcn, &lcn, &clen) && vcn == vcn_next;
+	if (!ok)
+		vcn = vcn_next;
+	up_read(lock);
+	goto next_run;
+}
+
+static bool scan_for_free(const ulong *buf, u32 bit, u32 bits, size_t *ret)
+{
+	size_t pos = find_next_zero_bit(buf, bits, bit);
+
+	if (pos >= bits)
+		return false;
+	*ret = pos;
+	return true;
+}
+
+/*
+ * indx_find_free - Look for free bit.
+ *
+ * Return: -1 if no free bits.
+ */
+static int indx_find_free(struct ntfs_index *indx, struct ntfs_inode *ni,
+			  size_t *bit, struct ATTRIB **bitmap)
+{
+	struct ATTRIB *b;
+	struct ATTR_LIST_ENTRY *le = NULL;
+	const struct INDEX_NAMES *in = &s_index_names[indx->type];
+	int err;
+
+	b = ni_find_attr(ni, NULL, &le, ATTR_BITMAP, in->name, in->name_len,
+			 NULL, NULL);
+
+	if (!b)
+		return -ENOENT;
+
+	*bitmap = b;
+	*bit = MINUS_ONE_T;
+
+	if (!b->non_res) {
+		u32 nbits = 8 * le32_to_cpu(b->res.data_size);
+		size_t pos = find_next_zero_bit(resident_data(b), nbits, 0);
+
+		if (pos < nbits)
+			*bit = pos;
+	} else {
+		err = scan_nres_bitmap(ni, b, indx, 0, &scan_for_free, bit);
+
+		if (err)
+			return err;
+	}
+
+	return 0;
+}
+
+static bool scan_for_used(const ulong *buf, u32 bit, u32 bits, size_t *ret)
+{
+	size_t pos = find_next_bit(buf, bits, bit);
+
+	if (pos >= bits)
+		return false;
+	*ret = pos;
+	return true;
+}
+
+/*
+ * indx_used_bit - Look for used bit.
+ *
+ * Return: MINUS_ONE_T if no used bits.
+ */
+int indx_used_bit(struct ntfs_index *indx, struct ntfs_inode *ni, size_t *bit)
+{
+	struct ATTRIB *b;
+	struct ATTR_LIST_ENTRY *le = NULL;
+	size_t from = *bit;
+	const struct INDEX_NAMES *in = &s_index_names[indx->type];
+	int err;
+
+	b = ni_find_attr(ni, NULL, &le, ATTR_BITMAP, in->name, in->name_len,
+			 NULL, NULL);
+
+	if (!b)
+		return -ENOENT;
+
+	*bit = MINUS_ONE_T;
+
+	if (!b->non_res) {
+		u32 nbits = le32_to_cpu(b->res.data_size) * 8;
+		size_t pos = find_next_bit(resident_data(b), nbits, from);
+
+		if (pos < nbits)
+			*bit = pos;
+	} else {
+		err = scan_nres_bitmap(ni, b, indx, from, &scan_for_used, bit);
+		if (err)
+			return err;
+	}
+
+	return 0;
+}
+
+/*
+ * hdr_find_split
+ *
+ * Find a point at which the index allocation buffer would like to be split.
+ * NOTE: This function should never return 'END' entry NULL returns on error.
+ */
+static const struct NTFS_DE *hdr_find_split(const struct INDEX_HDR *hdr)
+{
+	size_t o;
+	const struct NTFS_DE *e = hdr_first_de(hdr);
+	u32 used_2 = le32_to_cpu(hdr->used) >> 1;
+	u16 esize;
+
+	if (!e || de_is_last(e))
+		return NULL;
+
+	esize = le16_to_cpu(e->size);
+	for (o = le32_to_cpu(hdr->de_off) + esize; o < used_2; o += esize) {
+		const struct NTFS_DE *p = e;
+
+		e = Add2Ptr(hdr, o);
+
+		/* We must not return END entry. */
+		if (de_is_last(e))
+			return p;
+
+		esize = le16_to_cpu(e->size);
+	}
+
+	return e;
+}
+
+/*
+ * hdr_insert_head - Insert some entries at the beginning of the buffer.
+ *
+ * It is used to insert entries into a newly-created buffer.
+ */
+static const struct NTFS_DE *hdr_insert_head(struct INDEX_HDR *hdr,
+					     const void *ins, u32 ins_bytes)
+{
+	u32 to_move;
+	struct NTFS_DE *e = hdr_first_de(hdr);
+	u32 used = le32_to_cpu(hdr->used);
+
+	if (!e)
+		return NULL;
+
+	/* Now we just make room for the inserted entries and jam it in. */
+	to_move = used - le32_to_cpu(hdr->de_off);
+	memmove(Add2Ptr(e, ins_bytes), e, to_move);
+	memcpy(e, ins, ins_bytes);
+	hdr->used = cpu_to_le32(used + ins_bytes);
+
+	return e;
+}
+
+void fnd_clear(struct ntfs_fnd *fnd)
+{
+	int i;
+
+	for (i = 0; i < fnd->level; i++) {
+		struct indx_node *n = fnd->nodes[i];
+
+		if (!n)
+			continue;
+
+		put_indx_node(n);
+		fnd->nodes[i] = NULL;
+	}
+	fnd->level = 0;
+	fnd->root_de = NULL;
+}
+
+static int fnd_push(struct ntfs_fnd *fnd, struct indx_node *n,
+		    struct NTFS_DE *e)
+{
+	int i;
+
+	i = fnd->level;
+	if (i < 0 || i >= ARRAY_SIZE(fnd->nodes))
+		return -EINVAL;
+	fnd->nodes[i] = n;
+	fnd->de[i] = e;
+	fnd->level += 1;
+	return 0;
+}
+
+static struct indx_node *fnd_pop(struct ntfs_fnd *fnd)
+{
+	struct indx_node *n;
+	int i = fnd->level;
+
+	i -= 1;
+	n = fnd->nodes[i];
+	fnd->nodes[i] = NULL;
+	fnd->level = i;
+
+	return n;
+}
+
+static bool fnd_is_empty(struct ntfs_fnd *fnd)
+{
+	if (!fnd->level)
+		return !fnd->root_de;
+
+	return !fnd->de[fnd->level - 1];
+}
+
+/*
+ * hdr_find_e - Locate an entry the index buffer.
+ *
+ * If no matching entry is found, it returns the first entry which is greater
+ * than the desired entry If the search key is greater than all the entries the
+ * buffer, it returns the 'end' entry. This function does a binary search of the
+ * current index buffer, for the first entry that is <= to the search value.
+ *
+ * Return: NULL if error.
+ */
+static struct NTFS_DE *hdr_find_e(const struct ntfs_index *indx,
+				  const struct INDEX_HDR *hdr, const void *key,
+				  size_t key_len, const void *ctx, int *diff)
+{
+	struct NTFS_DE *e;
+	NTFS_CMP_FUNC cmp = indx->cmp;
+	u32 e_size, e_key_len;
+	u32 end = le32_to_cpu(hdr->used);
+	u32 off = le32_to_cpu(hdr->de_off);
+
+#ifdef NTFS3_INDEX_BINARY_SEARCH
+	int max_idx = 0, fnd, min_idx;
+	int nslots = 64;
+	u16 *offs;
+
+	if (end > 0x10000)
+		goto next;
+
+	offs = kmalloc(sizeof(u16) * nslots, GFP_NOFS);
+	if (!offs)
+		goto next;
+
+	/* Use binary search algorithm. */
+next1:
+	if (off + sizeof(struct NTFS_DE) > end) {
+		e = NULL;
+		goto out1;
+	}
+	e = Add2Ptr(hdr, off);
+	e_size = le16_to_cpu(e->size);
+
+	if (e_size < sizeof(struct NTFS_DE) || off + e_size > end) {
+		e = NULL;
+		goto out1;
+	}
+
+	if (max_idx >= nslots) {
+		u16 *ptr;
+		int new_slots = ALIGN(2 * nslots, 8);
+
+		ptr = kmalloc(sizeof(u16) * new_slots, GFP_NOFS);
+		if (ptr)
+			memcpy(ptr, offs, sizeof(u16) * max_idx);
+		kfree(offs);
+		offs = ptr;
+		nslots = new_slots;
+		if (!ptr)
+			goto next;
+	}
+
+	/* Store entry table. */
+	offs[max_idx] = off;
+
+	if (!de_is_last(e)) {
+		off += e_size;
+		max_idx += 1;
+		goto next1;
+	}
+
+	/*
+	 * Table of pointers is created.
+	 * Use binary search to find entry that is <= to the search value.
+	 */
+	fnd = -1;
+	min_idx = 0;
+
+	while (min_idx <= max_idx) {
+		int mid_idx = min_idx + ((max_idx - min_idx) >> 1);
+		int diff2;
+
+		e = Add2Ptr(hdr, offs[mid_idx]);
+
+		e_key_len = le16_to_cpu(e->key_size);
+
+		diff2 = (*cmp)(key, key_len, e + 1, e_key_len, ctx);
+
+		if (!diff2) {
+			*diff = 0;
+			goto out1;
+		}
+
+		if (diff2 < 0) {
+			max_idx = mid_idx - 1;
+			fnd = mid_idx;
+			if (!fnd)
+				break;
+		} else {
+			min_idx = mid_idx + 1;
+		}
+	}
+
+	if (fnd == -1) {
+		e = NULL;
+		goto out1;
+	}
+
+	*diff = -1;
+	e = Add2Ptr(hdr, offs[fnd]);
+
+out1:
+	kfree(offs);
+
+	return e;
+#endif
+
+next:
+	/*
+	 * Entries index are sorted.
+	 * Enumerate all entries until we find entry
+	 * that is <= to the search value.
+	 */
+	if (off + sizeof(struct NTFS_DE) > end)
+		return NULL;
+
+	e = Add2Ptr(hdr, off);
+	e_size = le16_to_cpu(e->size);
+
+	if (e_size < sizeof(struct NTFS_DE) || off + e_size > end)
+		return NULL;
+
+	off += e_size;
+
+	e_key_len = le16_to_cpu(e->key_size);
+
+	*diff = (*cmp)(key, key_len, e + 1, e_key_len, ctx);
+	if (!*diff)
+		return e;
+
+	if (*diff <= 0)
+		return e;
+
+	if (de_is_last(e)) {
+		*diff = 1;
+		return e;
+	}
+	goto next;
+}
+
+/*
+ * hdr_insert_de - Insert an index entry into the buffer.
+ *
+ * 'before' should be a pointer previously returned from hdr_find_e.
+ */
+static struct NTFS_DE *hdr_insert_de(const struct ntfs_index *indx,
+				     struct INDEX_HDR *hdr,
+				     const struct NTFS_DE *de,
+				     struct NTFS_DE *before, const void *ctx)
+{
+	int diff;
+	size_t off = PtrOffset(hdr, before);
+	u32 used = le32_to_cpu(hdr->used);
+	u32 total = le32_to_cpu(hdr->total);
+	u16 de_size = le16_to_cpu(de->size);
+
+	/* First, check to see if there's enough room. */
+	if (used + de_size > total)
+		return NULL;
+
+	/* We know there's enough space, so we know we'll succeed. */
+	if (before) {
+		/* Check that before is inside Index. */
+		if (off >= used || off < le32_to_cpu(hdr->de_off) ||
+		    off + le16_to_cpu(before->size) > total) {
+			return NULL;
+		}
+		goto ok;
+	}
+	/* No insert point is applied. Get it manually. */
+	before = hdr_find_e(indx, hdr, de + 1, le16_to_cpu(de->key_size), ctx,
+			    &diff);
+	if (!before)
+		return NULL;
+	off = PtrOffset(hdr, before);
+
+ok:
+	/* Now we just make room for the entry and jam it in. */
+	memmove(Add2Ptr(before, de_size), before, used - off);
+
+	hdr->used = cpu_to_le32(used + de_size);
+	memcpy(before, de, de_size);
+
+	return before;
+}
+
+/*
+ * hdr_delete_de - Remove an entry from the index buffer.
+ */
+static inline struct NTFS_DE *hdr_delete_de(struct INDEX_HDR *hdr,
+					    struct NTFS_DE *re)
+{
+	u32 used = le32_to_cpu(hdr->used);
+	u16 esize = le16_to_cpu(re->size);
+	u32 off = PtrOffset(hdr, re);
+	int bytes = used - (off + esize);
+
+	if (off >= used || esize < sizeof(struct NTFS_DE) ||
+	    bytes < sizeof(struct NTFS_DE))
+		return NULL;
+
+	hdr->used = cpu_to_le32(used - esize);
+	memmove(re, Add2Ptr(re, esize), bytes);
+
+	return re;
+}
+
+void indx_clear(struct ntfs_index *indx)
+{
+	run_close(&indx->alloc_run);
+	run_close(&indx->bitmap_run);
+}
+
+int indx_init(struct ntfs_index *indx, struct ntfs_sb_info *sbi,
+	      const struct ATTRIB *attr, enum index_mutex_classed type)
+{
+	u32 t32;
+	const struct INDEX_ROOT *root = resident_data(attr);
+
+	/* Check root fields. */
+	if (!root->index_block_clst)
+		return -EINVAL;
+
+	indx->type = type;
+	indx->idx2vbn_bits = __ffs(root->index_block_clst);
+
+	t32 = le32_to_cpu(root->index_block_size);
+	indx->index_bits = blksize_bits(t32);
+
+	/* Check index record size. */
+	if (t32 < sbi->cluster_size) {
+		/* Index record is smaller than a cluster, use 512 blocks. */
+		if (t32 != root->index_block_clst * SECTOR_SIZE)
+			return -EINVAL;
+
+		/* Check alignment to a cluster. */
+		if ((sbi->cluster_size >> SECTOR_SHIFT) &
+		    (root->index_block_clst - 1)) {
+			return -EINVAL;
+		}
+
+		indx->vbn2vbo_bits = SECTOR_SHIFT;
+	} else {
+		/* Index record must be a multiple of cluster size. */
+		if (t32 != root->index_block_clst << sbi->cluster_bits)
+			return -EINVAL;
+
+		indx->vbn2vbo_bits = sbi->cluster_bits;
+	}
+
+	init_rwsem(&indx->run_lock);
+
+	indx->cmp = get_cmp_func(root);
+	return indx->cmp ? 0 : -EINVAL;
+}
+
+static struct indx_node *indx_new(struct ntfs_index *indx,
+				  struct ntfs_inode *ni, CLST vbn,
+				  const __le64 *sub_vbn)
+{
+	int err;
+	struct NTFS_DE *e;
+	struct indx_node *r;
+	struct INDEX_HDR *hdr;
+	struct INDEX_BUFFER *index;
+	u64 vbo = (u64)vbn << indx->vbn2vbo_bits;
+	u32 bytes = 1u << indx->index_bits;
+	u16 fn;
+	u32 eo;
+
+	r = kzalloc(sizeof(struct indx_node), GFP_NOFS);
+	if (!r)
+		return ERR_PTR(-ENOMEM);
+
+	index = kzalloc(bytes, GFP_NOFS);
+	if (!index) {
+		kfree(r);
+		return ERR_PTR(-ENOMEM);
+	}
+
+	err = ntfs_get_bh(ni->mi.sbi, &indx->alloc_run, vbo, bytes, &r->nb);
+
+	if (err) {
+		kfree(index);
+		kfree(r);
+		return ERR_PTR(err);
+	}
+
+	/* Create header. */
+	index->rhdr.sign = NTFS_INDX_SIGNATURE;
+	index->rhdr.fix_off = cpu_to_le16(sizeof(struct INDEX_BUFFER)); // 0x28
+	fn = (bytes >> SECTOR_SHIFT) + 1; // 9
+	index->rhdr.fix_num = cpu_to_le16(fn);
+	index->vbn = cpu_to_le64(vbn);
+	hdr = &index->ihdr;
+	eo = ALIGN(sizeof(struct INDEX_BUFFER) + fn * sizeof(short), 8);
+	hdr->de_off = cpu_to_le32(eo);
+
+	e = Add2Ptr(hdr, eo);
+
+	if (sub_vbn) {
+		e->flags = NTFS_IE_LAST | NTFS_IE_HAS_SUBNODES;
+		e->size = cpu_to_le16(sizeof(struct NTFS_DE) + sizeof(u64));
+		hdr->used =
+			cpu_to_le32(eo + sizeof(struct NTFS_DE) + sizeof(u64));
+		de_set_vbn_le(e, *sub_vbn);
+		hdr->flags = 1;
+	} else {
+		e->size = cpu_to_le16(sizeof(struct NTFS_DE));
+		hdr->used = cpu_to_le32(eo + sizeof(struct NTFS_DE));
+		e->flags = NTFS_IE_LAST;
+	}
+
+	hdr->total = cpu_to_le32(bytes - offsetof(struct INDEX_BUFFER, ihdr));
+
+	r->index = index;
+	return r;
+}
+
+struct INDEX_ROOT *indx_get_root(struct ntfs_index *indx, struct ntfs_inode *ni,
+				 struct ATTRIB **attr, struct mft_inode **mi)
+{
+	struct ATTR_LIST_ENTRY *le = NULL;
+	struct ATTRIB *a;
+	const struct INDEX_NAMES *in = &s_index_names[indx->type];
+
+	a = ni_find_attr(ni, NULL, &le, ATTR_ROOT, in->name, in->name_len, NULL,
+			 mi);
+	if (!a)
+		return NULL;
+
+	if (attr)
+		*attr = a;
+
+	return resident_data_ex(a, sizeof(struct INDEX_ROOT));
+}
+
+static int indx_write(struct ntfs_index *indx, struct ntfs_inode *ni,
+		      struct indx_node *node, int sync)
+{
+	struct INDEX_BUFFER *ib = node->index;
+
+	return ntfs_write_bh(ni->mi.sbi, &ib->rhdr, &node->nb, sync);
+}
+
+/*
+ * indx_read
+ *
+ * If ntfs_readdir calls this function
+ * inode is shared locked and no ni_lock.
+ * Use rw_semaphore for read/write access to alloc_run.
+ */
+int indx_read(struct ntfs_index *indx, struct ntfs_inode *ni, CLST vbn,
+	      struct indx_node **node)
+{
+	int err;
+	struct INDEX_BUFFER *ib;
+	struct runs_tree *run = &indx->alloc_run;
+	struct rw_semaphore *lock = &indx->run_lock;
+	u64 vbo = (u64)vbn << indx->vbn2vbo_bits;
+	u32 bytes = 1u << indx->index_bits;
+	struct indx_node *in = *node;
+	const struct INDEX_NAMES *name;
+
+	if (!in) {
+		in = kzalloc(sizeof(struct indx_node), GFP_NOFS);
+		if (!in)
+			return -ENOMEM;
+	} else {
+		nb_put(&in->nb);
+	}
+
+	ib = in->index;
+	if (!ib) {
+		ib = kmalloc(bytes, GFP_NOFS);
+		if (!ib) {
+			err = -ENOMEM;
+			goto out;
+		}
+	}
+
+	down_read(lock);
+	err = ntfs_read_bh(ni->mi.sbi, run, vbo, &ib->rhdr, bytes, &in->nb);
+	up_read(lock);
+	if (!err)
+		goto ok;
+
+	if (err == -E_NTFS_FIXUP)
+		goto ok;
+
+	if (err != -ENOENT)
+		goto out;
+
+	name = &s_index_names[indx->type];
+	down_write(lock);
+	err = attr_load_runs_range(ni, ATTR_ALLOC, name->name, name->name_len,
+				   run, vbo, vbo + bytes);
+	up_write(lock);
+	if (err)
+		goto out;
+
+	down_read(lock);
+	err = ntfs_read_bh(ni->mi.sbi, run, vbo, &ib->rhdr, bytes, &in->nb);
+	up_read(lock);
+	if (err == -E_NTFS_FIXUP)
+		goto ok;
+
+	if (err)
+		goto out;
+
+ok:
+	if (err == -E_NTFS_FIXUP) {
+		ntfs_write_bh(ni->mi.sbi, &ib->rhdr, &in->nb, 0);
+		err = 0;
+	}
+
+	in->index = ib;
+	*node = in;
+
+out:
+	if (ib != in->index)
+		kfree(ib);
+
+	if (*node != in) {
+		nb_put(&in->nb);
+		kfree(in);
+	}
+
+	return err;
+}
+
+/*
+ * indx_find - Scan NTFS directory for given entry.
+ */
+int indx_find(struct ntfs_index *indx, struct ntfs_inode *ni,
+	      const struct INDEX_ROOT *root, const void *key, size_t key_len,
+	      const void *ctx, int *diff, struct NTFS_DE **entry,
+	      struct ntfs_fnd *fnd)
+{
+	int err;
+	struct NTFS_DE *e;
+	const struct INDEX_HDR *hdr;
+	struct indx_node *node;
+
+	if (!root)
+		root = indx_get_root(&ni->dir, ni, NULL, NULL);
+
+	if (!root) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	hdr = &root->ihdr;
+
+	/* Check cache. */
+	e = fnd->level ? fnd->de[fnd->level - 1] : fnd->root_de;
+	if (e && !de_is_last(e) &&
+	    !(*indx->cmp)(key, key_len, e + 1, le16_to_cpu(e->key_size), ctx)) {
+		*entry = e;
+		*diff = 0;
+		return 0;
+	}
+
+	/* Soft finder reset. */
+	fnd_clear(fnd);
+
+	/* Lookup entry that is <= to the search value. */
+	e = hdr_find_e(indx, hdr, key, key_len, ctx, diff);
+	if (!e)
+		return -EINVAL;
+
+	if (fnd)
+		fnd->root_de = e;
+
+	err = 0;
+
+	for (;;) {
+		node = NULL;
+		if (*diff >= 0 || !de_has_vcn_ex(e)) {
+			*entry = e;
+			goto out;
+		}
+
+		/* Read next level. */
+		err = indx_read(indx, ni, de_get_vbn(e), &node);
+		if (err)
+			goto out;
+
+		/* Lookup entry that is <= to the search value. */
+		e = hdr_find_e(indx, &node->index->ihdr, key, key_len, ctx,
+			       diff);
+		if (!e) {
+			err = -EINVAL;
+			put_indx_node(node);
+			goto out;
+		}
+
+		fnd_push(fnd, node, e);
+	}
+
+out:
+	return err;
+}
+
+int indx_find_sort(struct ntfs_index *indx, struct ntfs_inode *ni,
+		   const struct INDEX_ROOT *root, struct NTFS_DE **entry,
+		   struct ntfs_fnd *fnd)
+{
+	int err;
+	struct indx_node *n = NULL;
+	struct NTFS_DE *e;
+	size_t iter = 0;
+	int level = fnd->level;
+
+	if (!*entry) {
+		/* Start find. */
+		e = hdr_first_de(&root->ihdr);
+		if (!e)
+			return 0;
+		fnd_clear(fnd);
+		fnd->root_de = e;
+	} else if (!level) {
+		if (de_is_last(fnd->root_de)) {
+			*entry = NULL;
+			return 0;
+		}
+
+		e = hdr_next_de(&root->ihdr, fnd->root_de);
+		if (!e)
+			return -EINVAL;
+		fnd->root_de = e;
+	} else {
+		n = fnd->nodes[level - 1];
+		e = fnd->de[level - 1];
+
+		if (de_is_last(e))
+			goto pop_level;
+
+		e = hdr_next_de(&n->index->ihdr, e);
+		if (!e)
+			return -EINVAL;
+
+		fnd->de[level - 1] = e;
+	}
+
+	/* Just to avoid tree cycle. */
+next_iter:
+	if (iter++ >= 1000)
+		return -EINVAL;
+
+	while (de_has_vcn_ex(e)) {
+		if (le16_to_cpu(e->size) <
+		    sizeof(struct NTFS_DE) + sizeof(u64)) {
+			if (n) {
+				fnd_pop(fnd);
+				kfree(n);
+			}
+			return -EINVAL;
+		}
+
+		/* Read next level. */
+		err = indx_read(indx, ni, de_get_vbn(e), &n);
+		if (err)
+			return err;
+
+		/* Try next level. */
+		e = hdr_first_de(&n->index->ihdr);
+		if (!e) {
+			kfree(n);
+			return -EINVAL;
+		}
+
+		fnd_push(fnd, n, e);
+	}
+
+	if (le16_to_cpu(e->size) > sizeof(struct NTFS_DE)) {
+		*entry = e;
+		return 0;
+	}
+
+pop_level:
+	for (;;) {
+		if (!de_is_last(e))
+			goto next_iter;
+
+		/* Pop one level. */
+		if (n) {
+			fnd_pop(fnd);
+			kfree(n);
+		}
+
+		level = fnd->level;
+
+		if (level) {
+			n = fnd->nodes[level - 1];
+			e = fnd->de[level - 1];
+		} else if (fnd->root_de) {
+			n = NULL;
+			e = fnd->root_de;
+			fnd->root_de = NULL;
+		} else {
+			*entry = NULL;
+			return 0;
+		}
+
+		if (le16_to_cpu(e->size) > sizeof(struct NTFS_DE)) {
+			*entry = e;
+			if (!fnd->root_de)
+				fnd->root_de = e;
+			return 0;
+		}
+	}
+}
+
+int indx_find_raw(struct ntfs_index *indx, struct ntfs_inode *ni,
+		  const struct INDEX_ROOT *root, struct NTFS_DE **entry,
+		  size_t *off, struct ntfs_fnd *fnd)
+{
+	int err;
+	struct indx_node *n = NULL;
+	struct NTFS_DE *e = NULL;
+	struct NTFS_DE *e2;
+	size_t bit;
+	CLST next_used_vbn;
+	CLST next_vbn;
+	u32 record_size = ni->mi.sbi->record_size;
+
+	/* Use non sorted algorithm. */
+	if (!*entry) {
+		/* This is the first call. */
+		e = hdr_first_de(&root->ihdr);
+		if (!e)
+			return 0;
+		fnd_clear(fnd);
+		fnd->root_de = e;
+
+		/* The first call with setup of initial element. */
+		if (*off >= record_size) {
+			next_vbn = (((*off - record_size) >> indx->index_bits))
+				   << indx->idx2vbn_bits;
+			/* Jump inside cycle 'for'. */
+			goto next;
+		}
+
+		/* Start enumeration from root. */
+		*off = 0;
+	} else if (!fnd->root_de)
+		return -EINVAL;
+
+	for (;;) {
+		/* Check if current entry can be used. */
+		if (e && le16_to_cpu(e->size) > sizeof(struct NTFS_DE))
+			goto ok;
+
+		if (!fnd->level) {
+			/* Continue to enumerate root. */
+			if (!de_is_last(fnd->root_de)) {
+				e = hdr_next_de(&root->ihdr, fnd->root_de);
+				if (!e)
+					return -EINVAL;
+				fnd->root_de = e;
+				continue;
+			}
+
+			/* Start to enumerate indexes from 0. */
+			next_vbn = 0;
+		} else {
+			/* Continue to enumerate indexes. */
+			e2 = fnd->de[fnd->level - 1];
+
+			n = fnd->nodes[fnd->level - 1];
+
+			if (!de_is_last(e2)) {
+				e = hdr_next_de(&n->index->ihdr, e2);
+				if (!e)
+					return -EINVAL;
+				fnd->de[fnd->level - 1] = e;
+				continue;
+			}
+
+			/* Continue with next index. */
+			next_vbn = le64_to_cpu(n->index->vbn) +
+				   root->index_block_clst;
+		}
+
+next:
+		/* Release current index. */
+		if (n) {
+			fnd_pop(fnd);
+			put_indx_node(n);
+			n = NULL;
+		}
+
+		/* Skip all free indexes. */
+		bit = next_vbn >> indx->idx2vbn_bits;
+		err = indx_used_bit(indx, ni, &bit);
+		if (err == -ENOENT || bit == MINUS_ONE_T) {
+			/* No used indexes. */
+			*entry = NULL;
+			return 0;
+		}
+
+		next_used_vbn = bit << indx->idx2vbn_bits;
+
+		/* Read buffer into memory. */
+		err = indx_read(indx, ni, next_used_vbn, &n);
+		if (err)
+			return err;
+
+		e = hdr_first_de(&n->index->ihdr);
+		fnd_push(fnd, n, e);
+		if (!e)
+			return -EINVAL;
+	}
+
+ok:
+	/* Return offset to restore enumerator if necessary. */
+	if (!n) {
+		/* 'e' points in root, */
+		*off = PtrOffset(&root->ihdr, e);
+	} else {
+		/* 'e' points in index, */
+		*off = (le64_to_cpu(n->index->vbn) << indx->vbn2vbo_bits) +
+		       record_size + PtrOffset(&n->index->ihdr, e);
+	}
+
+	*entry = e;
+	return 0;
+}
+
+/*
+ * indx_create_allocate - Create "Allocation + Bitmap" attributes.
+ */
+static int indx_create_allocate(struct ntfs_index *indx, struct ntfs_inode *ni,
+				CLST *vbn)
+{
+	int err = -ENOMEM;
+	struct ntfs_sb_info *sbi = ni->mi.sbi;
+	struct ATTRIB *bitmap;
+	struct ATTRIB *alloc;
+	u32 data_size = 1u << indx->index_bits;
+	u32 alloc_size = ntfs_up_cluster(sbi, data_size);
+	CLST len = alloc_size >> sbi->cluster_bits;
+	const struct INDEX_NAMES *in = &s_index_names[indx->type];
+	CLST alen;
+	struct runs_tree run;
+
+	run_init(&run);
+
+	err = attr_allocate_clusters(sbi, &run, 0, 0, len, NULL, 0, &alen, 0,
+				     NULL);
+	if (err)
+		goto out;
+
+	err = ni_insert_nonresident(ni, ATTR_ALLOC, in->name, in->name_len,
+				    &run, 0, len, 0, &alloc, NULL);
+	if (err)
+		goto out1;
+
+	alloc->nres.valid_size = alloc->nres.data_size = cpu_to_le64(data_size);
+
+	err = ni_insert_resident(ni, bitmap_size(1), ATTR_BITMAP, in->name,
+				 in->name_len, &bitmap, NULL, NULL);
+	if (err)
+		goto out2;
+
+	if (in->name == I30_NAME) {
+		ni->vfs_inode.i_size = data_size;
+		inode_set_bytes(&ni->vfs_inode, alloc_size);
+	}
+
+	memcpy(&indx->alloc_run, &run, sizeof(run));
+
+	*vbn = 0;
+
+	return 0;
+
+out2:
+	mi_remove_attr(NULL, &ni->mi, alloc);
+
+out1:
+	run_deallocate(sbi, &run, false);
+
+out:
+	return err;
+}
+
+/*
+ * indx_add_allocate - Add clusters to index.
+ */
+static int indx_add_allocate(struct ntfs_index *indx, struct ntfs_inode *ni,
+			     CLST *vbn)
+{
+	int err;
+	size_t bit;
+	u64 data_size;
+	u64 bmp_size, bmp_size_v;
+	struct ATTRIB *bmp, *alloc;
+	struct mft_inode *mi;
+	const struct INDEX_NAMES *in = &s_index_names[indx->type];
+
+	err = indx_find_free(indx, ni, &bit, &bmp);
+	if (err)
+		goto out1;
+
+	if (bit != MINUS_ONE_T) {
+		bmp = NULL;
+	} else {
+		if (bmp->non_res) {
+			bmp_size = le64_to_cpu(bmp->nres.data_size);
+			bmp_size_v = le64_to_cpu(bmp->nres.valid_size);
+		} else {
+			bmp_size = bmp_size_v = le32_to_cpu(bmp->res.data_size);
+		}
+
+		bit = bmp_size << 3;
+	}
+
+	data_size = (u64)(bit + 1) << indx->index_bits;
+
+	if (bmp) {
+		/* Increase bitmap. */
+		err = attr_set_size(ni, ATTR_BITMAP, in->name, in->name_len,
+				    &indx->bitmap_run, bitmap_size(bit + 1),
+				    NULL, true, NULL);
+		if (err)
+			goto out1;
+	}
+
+	alloc = ni_find_attr(ni, NULL, NULL, ATTR_ALLOC, in->name, in->name_len,
+			     NULL, &mi);
+	if (!alloc) {
+		err = -EINVAL;
+		if (bmp)
+			goto out2;
+		goto out1;
+	}
+
+	/* Increase allocation. */
+	err = attr_set_size(ni, ATTR_ALLOC, in->name, in->name_len,
+			    &indx->alloc_run, data_size, &data_size, true,
+			    NULL);
+	if (err) {
+		if (bmp)
+			goto out2;
+		goto out1;
+	}
+
+	*vbn = bit << indx->idx2vbn_bits;
+
+	return 0;
+
+out2:
+	/* Ops. No space? */
+	attr_set_size(ni, ATTR_BITMAP, in->name, in->name_len,
+		      &indx->bitmap_run, bmp_size, &bmp_size_v, false, NULL);
+
+out1:
+	return err;
+}
+
+/*
+ * indx_insert_into_root - Attempt to insert an entry into the index root.
+ *
+ * @undo - True if we undoing previous remove.
+ * If necessary, it will twiddle the index b-tree.
+ */
+static int indx_insert_into_root(struct ntfs_index *indx, struct ntfs_inode *ni,
+				 const struct NTFS_DE *new_de,
+				 struct NTFS_DE *root_de, const void *ctx,
+				 struct ntfs_fnd *fnd, bool undo)
+{
+	int err = 0;
+	struct NTFS_DE *e, *e0, *re;
+	struct mft_inode *mi;
+	struct ATTRIB *attr;
+	struct INDEX_HDR *hdr;
+	struct indx_node *n;
+	CLST new_vbn;
+	__le64 *sub_vbn, t_vbn;
+	u16 new_de_size;
+	u32 hdr_used, hdr_total, asize, to_move;
+	u32 root_size, new_root_size;
+	struct ntfs_sb_info *sbi;
+	int ds_root;
+	struct INDEX_ROOT *root, *a_root;
+
+	/* Get the record this root placed in. */
+	root = indx_get_root(indx, ni, &attr, &mi);
+	if (!root)
+		return -EINVAL;
+
+	/*
+	 * Try easy case:
+	 * hdr_insert_de will succeed if there's
+	 * room the root for the new entry.
+	 */
+	hdr = &root->ihdr;
+	sbi = ni->mi.sbi;
+	new_de_size = le16_to_cpu(new_de->size);
+	hdr_used = le32_to_cpu(hdr->used);
+	hdr_total = le32_to_cpu(hdr->total);
+	asize = le32_to_cpu(attr->size);
+	root_size = le32_to_cpu(attr->res.data_size);
+
+	ds_root = new_de_size + hdr_used - hdr_total;
+
+	/* If 'undo' is set then reduce requirements. */
+	if ((undo || asize + ds_root < sbi->max_bytes_per_attr) &&
+	    mi_resize_attr(mi, attr, ds_root)) {
+		hdr->total = cpu_to_le32(hdr_total + ds_root);
+		e = hdr_insert_de(indx, hdr, new_de, root_de, ctx);
+		WARN_ON(!e);
+		fnd_clear(fnd);
+		fnd->root_de = e;
+
+		return 0;
+	}
+
+	/* Make a copy of root attribute to restore if error. */
+	a_root = kmemdup(attr, asize, GFP_NOFS);
+	if (!a_root)
+		return -ENOMEM;
+
+	/*
+	 * Copy all the non-end entries from
+	 * the index root to the new buffer.
+	 */
+	to_move = 0;
+	e0 = hdr_first_de(hdr);
+
+	/* Calculate the size to copy. */
+	for (e = e0;; e = hdr_next_de(hdr, e)) {
+		if (!e) {
+			err = -EINVAL;
+			goto out_free_root;
+		}
+
+		if (de_is_last(e))
+			break;
+		to_move += le16_to_cpu(e->size);
+	}
+
+	if (!to_move) {
+		re = NULL;
+	} else {
+		re = kmemdup(e0, to_move, GFP_NOFS);
+		if (!re) {
+			err = -ENOMEM;
+			goto out_free_root;
+		}
+	}
+
+	sub_vbn = NULL;
+	if (de_has_vcn(e)) {
+		t_vbn = de_get_vbn_le(e);
+		sub_vbn = &t_vbn;
+	}
+
+	new_root_size = sizeof(struct INDEX_ROOT) + sizeof(struct NTFS_DE) +
+			sizeof(u64);
+	ds_root = new_root_size - root_size;
+
+	if (ds_root > 0 && asize + ds_root > sbi->max_bytes_per_attr) {
+		/* Make root external. */
+		err = -EOPNOTSUPP;
+		goto out_free_re;
+	}
+
+	if (ds_root)
+		mi_resize_attr(mi, attr, ds_root);
+
+	/* Fill first entry (vcn will be set later). */
+	e = (struct NTFS_DE *)(root + 1);
+	memset(e, 0, sizeof(struct NTFS_DE));
+	e->size = cpu_to_le16(sizeof(struct NTFS_DE) + sizeof(u64));
+	e->flags = NTFS_IE_HAS_SUBNODES | NTFS_IE_LAST;
+
+	hdr->flags = 1;
+	hdr->used = hdr->total =
+		cpu_to_le32(new_root_size - offsetof(struct INDEX_ROOT, ihdr));
+
+	fnd->root_de = hdr_first_de(hdr);
+	mi->dirty = true;
+
+	/* Create alloc and bitmap attributes (if not). */
+	err = run_is_empty(&indx->alloc_run)
+		      ? indx_create_allocate(indx, ni, &new_vbn)
+		      : indx_add_allocate(indx, ni, &new_vbn);
+
+	/* Layout of record may be changed, so rescan root. */
+	root = indx_get_root(indx, ni, &attr, &mi);
+	if (!root) {
+		/* Bug? */
+		ntfs_set_state(sbi, NTFS_DIRTY_ERROR);
+		err = -EINVAL;
+		goto out_free_re;
+	}
+
+	if (err) {
+		/* Restore root. */
+		if (mi_resize_attr(mi, attr, -ds_root))
+			memcpy(attr, a_root, asize);
+		else {
+			/* Bug? */
+			ntfs_set_state(sbi, NTFS_DIRTY_ERROR);
+		}
+		goto out_free_re;
+	}
+
+	e = (struct NTFS_DE *)(root + 1);
+	*(__le64 *)(e + 1) = cpu_to_le64(new_vbn);
+	mi->dirty = true;
+
+	/* Now we can create/format the new buffer and copy the entries into. */
+	n = indx_new(indx, ni, new_vbn, sub_vbn);
+	if (IS_ERR(n)) {
+		err = PTR_ERR(n);
+		goto out_free_re;
+	}
+
+	hdr = &n->index->ihdr;
+	hdr_used = le32_to_cpu(hdr->used);
+	hdr_total = le32_to_cpu(hdr->total);
+
+	/* Copy root entries into new buffer. */
+	hdr_insert_head(hdr, re, to_move);
+
+	/* Update bitmap attribute. */
+	indx_mark_used(indx, ni, new_vbn >> indx->idx2vbn_bits);
+
+	/* Check if we can insert new entry new index buffer. */
+	if (hdr_used + new_de_size > hdr_total) {
+		/*
+		 * This occurs if MFT record is the same or bigger than index
+		 * buffer. Move all root new index and have no space to add
+		 * new entry classic case when MFT record is 1K and index
+		 * buffer 4K the problem should not occurs.
+		 */
+		kfree(re);
+		indx_write(indx, ni, n, 0);
+
+		put_indx_node(n);
+		fnd_clear(fnd);
+		err = indx_insert_entry(indx, ni, new_de, ctx, fnd, undo);
+		goto out_free_root;
+	}
+
+	/*
+	 * Now root is a parent for new index buffer.
+	 * Insert NewEntry a new buffer.
+	 */
+	e = hdr_insert_de(indx, hdr, new_de, NULL, ctx);
+	if (!e) {
+		err = -EINVAL;
+		goto out_put_n;
+	}
+	fnd_push(fnd, n, e);
+
+	/* Just write updates index into disk. */
+	indx_write(indx, ni, n, 0);
+
+	n = NULL;
+
+out_put_n:
+	put_indx_node(n);
+out_free_re:
+	kfree(re);
+out_free_root:
+	kfree(a_root);
+	return err;
+}
+
+/*
+ * indx_insert_into_buffer
+ *
+ * Attempt to insert an entry into an Index Allocation Buffer.
+ * If necessary, it will split the buffer.
+ */
+static int
+indx_insert_into_buffer(struct ntfs_index *indx, struct ntfs_inode *ni,
+			struct INDEX_ROOT *root, const struct NTFS_DE *new_de,
+			const void *ctx, int level, struct ntfs_fnd *fnd)
+{
+	int err;
+	const struct NTFS_DE *sp;
+	struct NTFS_DE *e, *de_t, *up_e = NULL;
+	struct indx_node *n2 = NULL;
+	struct indx_node *n1 = fnd->nodes[level];
+	struct INDEX_HDR *hdr1 = &n1->index->ihdr;
+	struct INDEX_HDR *hdr2;
+	u32 to_copy, used;
+	CLST new_vbn;
+	__le64 t_vbn, *sub_vbn;
+	u16 sp_size;
+
+	/* Try the most easy case. */
+	e = fnd->level - 1 == level ? fnd->de[level] : NULL;
+	e = hdr_insert_de(indx, hdr1, new_de, e, ctx);
+	fnd->de[level] = e;
+	if (e) {
+		/* Just write updated index into disk. */
+		indx_write(indx, ni, n1, 0);
+		return 0;
+	}
+
+	/*
+	 * No space to insert into buffer. Split it.
+	 * To split we:
+	 *  - Save split point ('cause index buffers will be changed)
+	 * - Allocate NewBuffer and copy all entries <= sp into new buffer
+	 * - Remove all entries (sp including) from TargetBuffer
+	 * - Insert NewEntry into left or right buffer (depending on sp <=>
+	 *     NewEntry)
+	 * - Insert sp into parent buffer (or root)
+	 * - Make sp a parent for new buffer
+	 */
+	sp = hdr_find_split(hdr1);
+	if (!sp)
+		return -EINVAL;
+
+	sp_size = le16_to_cpu(sp->size);
+	up_e = kmalloc(sp_size + sizeof(u64), GFP_NOFS);
+	if (!up_e)
+		return -ENOMEM;
+	memcpy(up_e, sp, sp_size);
+
+	if (!hdr1->flags) {
+		up_e->flags |= NTFS_IE_HAS_SUBNODES;
+		up_e->size = cpu_to_le16(sp_size + sizeof(u64));
+		sub_vbn = NULL;
+	} else {
+		t_vbn = de_get_vbn_le(up_e);
+		sub_vbn = &t_vbn;
+	}
+
+	/* Allocate on disk a new index allocation buffer. */
+	err = indx_add_allocate(indx, ni, &new_vbn);
+	if (err)
+		goto out;
+
+	/* Allocate and format memory a new index buffer. */
+	n2 = indx_new(indx, ni, new_vbn, sub_vbn);
+	if (IS_ERR(n2)) {
+		err = PTR_ERR(n2);
+		goto out;
+	}
+
+	hdr2 = &n2->index->ihdr;
+
+	/* Make sp a parent for new buffer. */
+	de_set_vbn(up_e, new_vbn);
+
+	/* Copy all the entries <= sp into the new buffer. */
+	de_t = hdr_first_de(hdr1);
+	to_copy = PtrOffset(de_t, sp);
+	hdr_insert_head(hdr2, de_t, to_copy);
+
+	/* Remove all entries (sp including) from hdr1. */
+	used = le32_to_cpu(hdr1->used) - to_copy - sp_size;
+	memmove(de_t, Add2Ptr(sp, sp_size), used - le32_to_cpu(hdr1->de_off));
+	hdr1->used = cpu_to_le32(used);
+
+	/*
+	 * Insert new entry into left or right buffer
+	 * (depending on sp <=> new_de).
+	 */
+	hdr_insert_de(indx,
+		      (*indx->cmp)(new_de + 1, le16_to_cpu(new_de->key_size),
+				   up_e + 1, le16_to_cpu(up_e->key_size),
+				   ctx) < 0
+			      ? hdr2
+			      : hdr1,
+		      new_de, NULL, ctx);
+
+	indx_mark_used(indx, ni, new_vbn >> indx->idx2vbn_bits);
+
+	indx_write(indx, ni, n1, 0);
+	indx_write(indx, ni, n2, 0);
+
+	put_indx_node(n2);
+
+	/*
+	 * We've finished splitting everybody, so we are ready to
+	 * insert the promoted entry into the parent.
+	 */
+	if (!level) {
+		/* Insert in root. */
+		err = indx_insert_into_root(indx, ni, up_e, NULL, ctx, fnd, 0);
+		if (err)
+			goto out;
+	} else {
+		/*
+		 * The target buffer's parent is another index buffer.
+		 * TODO: Remove recursion.
+		 */
+		err = indx_insert_into_buffer(indx, ni, root, up_e, ctx,
+					      level - 1, fnd);
+		if (err)
+			goto out;
+	}
+
+out:
+	kfree(up_e);
+
+	return err;
+}
+
+/*
+ * indx_insert_entry - Insert new entry into index.
+ *
+ * @undo - True if we undoing previous remove.
+ */
+int indx_insert_entry(struct ntfs_index *indx, struct ntfs_inode *ni,
+		      const struct NTFS_DE *new_de, const void *ctx,
+		      struct ntfs_fnd *fnd, bool undo)
+{
+	int err;
+	int diff;
+	struct NTFS_DE *e;
+	struct ntfs_fnd *fnd_a = NULL;
+	struct INDEX_ROOT *root;
+
+	if (!fnd) {
+		fnd_a = fnd_get();
+		if (!fnd_a) {
+			err = -ENOMEM;
+			goto out1;
+		}
+		fnd = fnd_a;
+	}
+
+	root = indx_get_root(indx, ni, NULL, NULL);
+	if (!root) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	if (fnd_is_empty(fnd)) {
+		/*
+		 * Find the spot the tree where we want to
+		 * insert the new entry.
+		 */
+		err = indx_find(indx, ni, root, new_de + 1,
+				le16_to_cpu(new_de->key_size), ctx, &diff, &e,
+				fnd);
+		if (err)
+			goto out;
+
+		if (!diff) {
+			err = -EEXIST;
+			goto out;
+		}
+	}
+
+	if (!fnd->level) {
+		/*
+		 * The root is also a leaf, so we'll insert the
+		 * new entry into it.
+		 */
+		err = indx_insert_into_root(indx, ni, new_de, fnd->root_de, ctx,
+					    fnd, undo);
+		if (err)
+			goto out;
+	} else {
+		/*
+		 * Found a leaf buffer, so we'll insert the new entry into it.
+		 */
+		err = indx_insert_into_buffer(indx, ni, root, new_de, ctx,
+					      fnd->level - 1, fnd);
+		if (err)
+			goto out;
+	}
+
+out:
+	fnd_put(fnd_a);
+out1:
+	return err;
+}
+
+/*
+ * indx_find_buffer - Locate a buffer from the tree.
+ */
+static struct indx_node *indx_find_buffer(struct ntfs_index *indx,
+					  struct ntfs_inode *ni,
+					  const struct INDEX_ROOT *root,
+					  __le64 vbn, struct indx_node *n)
+{
+	int err;
+	const struct NTFS_DE *e;
+	struct indx_node *r;
+	const struct INDEX_HDR *hdr = n ? &n->index->ihdr : &root->ihdr;
+
+	/* Step 1: Scan one level. */
+	for (e = hdr_first_de(hdr);; e = hdr_next_de(hdr, e)) {
+		if (!e)
+			return ERR_PTR(-EINVAL);
+
+		if (de_has_vcn(e) && vbn == de_get_vbn_le(e))
+			return n;
+
+		if (de_is_last(e))
+			break;
+	}
+
+	/* Step2: Do recursion. */
+	e = Add2Ptr(hdr, le32_to_cpu(hdr->de_off));
+	for (;;) {
+		if (de_has_vcn_ex(e)) {
+			err = indx_read(indx, ni, de_get_vbn(e), &n);
+			if (err)
+				return ERR_PTR(err);
+
+			r = indx_find_buffer(indx, ni, root, vbn, n);
+			if (r)
+				return r;
+		}
+
+		if (de_is_last(e))
+			break;
+
+		e = Add2Ptr(e, le16_to_cpu(e->size));
+	}
+
+	return NULL;
+}
+
+/*
+ * indx_shrink - Deallocate unused tail indexes.
+ */
+static int indx_shrink(struct ntfs_index *indx, struct ntfs_inode *ni,
+		       size_t bit)
+{
+	int err = 0;
+	u64 bpb, new_data;
+	size_t nbits;
+	struct ATTRIB *b;
+	struct ATTR_LIST_ENTRY *le = NULL;
+	const struct INDEX_NAMES *in = &s_index_names[indx->type];
+
+	b = ni_find_attr(ni, NULL, &le, ATTR_BITMAP, in->name, in->name_len,
+			 NULL, NULL);
+
+	if (!b)
+		return -ENOENT;
+
+	if (!b->non_res) {
+		unsigned long pos;
+		const unsigned long *bm = resident_data(b);
+
+		nbits = (size_t)le32_to_cpu(b->res.data_size) * 8;
+
+		if (bit >= nbits)
+			return 0;
+
+		pos = find_next_bit(bm, nbits, bit);
+		if (pos < nbits)
+			return 0;
+	} else {
+		size_t used = MINUS_ONE_T;
+
+		nbits = le64_to_cpu(b->nres.data_size) * 8;
+
+		if (bit >= nbits)
+			return 0;
+
+		err = scan_nres_bitmap(ni, b, indx, bit, &scan_for_used, &used);
+		if (err)
+			return err;
+
+		if (used != MINUS_ONE_T)
+			return 0;
+	}
+
+	new_data = (u64)bit << indx->index_bits;
+
+	err = attr_set_size(ni, ATTR_ALLOC, in->name, in->name_len,
+			    &indx->alloc_run, new_data, &new_data, false, NULL);
+	if (err)
+		return err;
+
+	bpb = bitmap_size(bit);
+	if (bpb * 8 == nbits)
+		return 0;
+
+	err = attr_set_size(ni, ATTR_BITMAP, in->name, in->name_len,
+			    &indx->bitmap_run, bpb, &bpb, false, NULL);
+
+	return err;
+}
+
+static int indx_free_children(struct ntfs_index *indx, struct ntfs_inode *ni,
+			      const struct NTFS_DE *e, bool trim)
+{
+	int err;
+	struct indx_node *n;
+	struct INDEX_HDR *hdr;
+	CLST vbn = de_get_vbn(e);
+	size_t i;
+
+	err = indx_read(indx, ni, vbn, &n);
+	if (err)
+		return err;
+
+	hdr = &n->index->ihdr;
+	/* First, recurse into the children, if any. */
+	if (hdr_has_subnode(hdr)) {
+		for (e = hdr_first_de(hdr); e; e = hdr_next_de(hdr, e)) {
+			indx_free_children(indx, ni, e, false);
+			if (de_is_last(e))
+				break;
+		}
+	}
+
+	put_indx_node(n);
+
+	i = vbn >> indx->idx2vbn_bits;
+	/*
+	 * We've gotten rid of the children; add this buffer to the free list.
+	 */
+	indx_mark_free(indx, ni, i);
+
+	if (!trim)
+		return 0;
+
+	/*
+	 * If there are no used indexes after current free index
+	 * then we can truncate allocation and bitmap.
+	 * Use bitmap to estimate the case.
+	 */
+	indx_shrink(indx, ni, i + 1);
+	return 0;
+}
+
+/*
+ * indx_get_entry_to_replace
+ *
+ * Find a replacement entry for a deleted entry.
+ * Always returns a node entry:
+ * NTFS_IE_HAS_SUBNODES is set the flags and the size includes the sub_vcn.
+ */
+static int indx_get_entry_to_replace(struct ntfs_index *indx,
+				     struct ntfs_inode *ni,
+				     const struct NTFS_DE *de_next,
+				     struct NTFS_DE **de_to_replace,
+				     struct ntfs_fnd *fnd)
+{
+	int err;
+	int level = -1;
+	CLST vbn;
+	struct NTFS_DE *e, *te, *re;
+	struct indx_node *n;
+	struct INDEX_BUFFER *ib;
+
+	*de_to_replace = NULL;
+
+	/* Find first leaf entry down from de_next. */
+	vbn = de_get_vbn(de_next);
+	for (;;) {
+		n = NULL;
+		err = indx_read(indx, ni, vbn, &n);
+		if (err)
+			goto out;
+
+		e = hdr_first_de(&n->index->ihdr);
+		fnd_push(fnd, n, e);
+
+		if (!de_is_last(e)) {
+			/*
+			 * This buffer is non-empty, so its first entry
+			 * could be used as the replacement entry.
+			 */
+			level = fnd->level - 1;
+		}
+
+		if (!de_has_vcn(e))
+			break;
+
+		/* This buffer is a node. Continue to go down. */
+		vbn = de_get_vbn(e);
+	}
+
+	if (level == -1)
+		goto out;
+
+	n = fnd->nodes[level];
+	te = hdr_first_de(&n->index->ihdr);
+	/* Copy the candidate entry into the replacement entry buffer. */
+	re = kmalloc(le16_to_cpu(te->size) + sizeof(u64), GFP_NOFS);
+	if (!re) {
+		err = -ENOMEM;
+		goto out;
+	}
+
+	*de_to_replace = re;
+	memcpy(re, te, le16_to_cpu(te->size));
+
+	if (!de_has_vcn(re)) {
+		/*
+		 * The replacement entry we found doesn't have a sub_vcn.
+		 * increase its size to hold one.
+		 */
+		le16_add_cpu(&re->size, sizeof(u64));
+		re->flags |= NTFS_IE_HAS_SUBNODES;
+	} else {
+		/*
+		 * The replacement entry we found was a node entry, which
+		 * means that all its child buffers are empty. Return them
+		 * to the free pool.
+		 */
+		indx_free_children(indx, ni, te, true);
+	}
+
+	/*
+	 * Expunge the replacement entry from its former location,
+	 * and then write that buffer.
+	 */
+	ib = n->index;
+	e = hdr_delete_de(&ib->ihdr, te);
+
+	fnd->de[level] = e;
+	indx_write(indx, ni, n, 0);
+
+	/* Check to see if this action created an empty leaf. */
+	if (ib_is_leaf(ib) && ib_is_empty(ib))
+		return 0;
+
+out:
+	fnd_clear(fnd);
+	return err;
+}
+
+/*
+ * indx_delete_entry - Delete an entry from the index.
+ */
+int indx_delete_entry(struct ntfs_index *indx, struct ntfs_inode *ni,
+		      const void *key, u32 key_len, const void *ctx)
+{
+	int err, diff;
+	struct INDEX_ROOT *root;
+	struct INDEX_HDR *hdr;
+	struct ntfs_fnd *fnd, *fnd2;
+	struct INDEX_BUFFER *ib;
+	struct NTFS_DE *e, *re, *next, *prev, *me;
+	struct indx_node *n, *n2d = NULL;
+	__le64 sub_vbn;
+	int level, level2;
+	struct ATTRIB *attr;
+	struct mft_inode *mi;
+	u32 e_size, root_size, new_root_size;
+	size_t trim_bit;
+	const struct INDEX_NAMES *in;
+
+	fnd = fnd_get();
+	if (!fnd) {
+		err = -ENOMEM;
+		goto out2;
+	}
+
+	fnd2 = fnd_get();
+	if (!fnd2) {
+		err = -ENOMEM;
+		goto out1;
+	}
+
+	root = indx_get_root(indx, ni, &attr, &mi);
+	if (!root) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	/* Locate the entry to remove. */
+	err = indx_find(indx, ni, root, key, key_len, ctx, &diff, &e, fnd);
+	if (err)
+		goto out;
+
+	if (!e || diff) {
+		err = -ENOENT;
+		goto out;
+	}
+
+	level = fnd->level;
+
+	if (level) {
+		n = fnd->nodes[level - 1];
+		e = fnd->de[level - 1];
+		ib = n->index;
+		hdr = &ib->ihdr;
+	} else {
+		hdr = &root->ihdr;
+		e = fnd->root_de;
+		n = NULL;
+	}
+
+	e_size = le16_to_cpu(e->size);
+
+	if (!de_has_vcn_ex(e)) {
+		/* The entry to delete is a leaf, so we can just rip it out. */
+		hdr_delete_de(hdr, e);
+
+		if (!level) {
+			hdr->total = hdr->used;
+
+			/* Shrink resident root attribute. */
+			mi_resize_attr(mi, attr, 0 - e_size);
+			goto out;
+		}
+
+		indx_write(indx, ni, n, 0);
+
+		/*
+		 * Check to see if removing that entry made
+		 * the leaf empty.
+		 */
+		if (ib_is_leaf(ib) && ib_is_empty(ib)) {
+			fnd_pop(fnd);
+			fnd_push(fnd2, n, e);
+		}
+	} else {
+		/*
+		 * The entry we wish to delete is a node buffer, so we
+		 * have to find a replacement for it.
+		 */
+		next = de_get_next(e);
+
+		err = indx_get_entry_to_replace(indx, ni, next, &re, fnd2);
+		if (err)
+			goto out;
+
+		if (re) {
+			de_set_vbn_le(re, de_get_vbn_le(e));
+			hdr_delete_de(hdr, e);
+
+			err = level ? indx_insert_into_buffer(indx, ni, root,
+							      re, ctx,
+							      fnd->level - 1,
+							      fnd)
+				    : indx_insert_into_root(indx, ni, re, e,
+							    ctx, fnd, 0);
+			kfree(re);
+
+			if (err)
+				goto out;
+		} else {
+			/*
+			 * There is no replacement for the current entry.
+			 * This means that the subtree rooted at its node
+			 * is empty, and can be deleted, which turn means
+			 * that the node can just inherit the deleted
+			 * entry sub_vcn.
+			 */
+			indx_free_children(indx, ni, next, true);
+
+			de_set_vbn_le(next, de_get_vbn_le(e));
+			hdr_delete_de(hdr, e);
+			if (level) {
+				indx_write(indx, ni, n, 0);
+			} else {
+				hdr->total = hdr->used;
+
+				/* Shrink resident root attribute. */
+				mi_resize_attr(mi, attr, 0 - e_size);
+			}
+		}
+	}
+
+	/* Delete a branch of tree. */
+	if (!fnd2 || !fnd2->level)
+		goto out;
+
+	/* Reinit root 'cause it can be changed. */
+	root = indx_get_root(indx, ni, &attr, &mi);
+	if (!root) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	n2d = NULL;
+	sub_vbn = fnd2->nodes[0]->index->vbn;
+	level2 = 0;
+	level = fnd->level;
+
+	hdr = level ? &fnd->nodes[level - 1]->index->ihdr : &root->ihdr;
+
+	/* Scan current level. */
+	for (e = hdr_first_de(hdr);; e = hdr_next_de(hdr, e)) {
+		if (!e) {
+			err = -EINVAL;
+			goto out;
+		}
+
+		if (de_has_vcn(e) && sub_vbn == de_get_vbn_le(e))
+			break;
+
+		if (de_is_last(e)) {
+			e = NULL;
+			break;
+		}
+	}
+
+	if (!e) {
+		/* Do slow search from root. */
+		struct indx_node *in;
+
+		fnd_clear(fnd);
+
+		in = indx_find_buffer(indx, ni, root, sub_vbn, NULL);
+		if (IS_ERR(in)) {
+			err = PTR_ERR(in);
+			goto out;
+		}
+
+		if (in)
+			fnd_push(fnd, in, NULL);
+	}
+
+	/* Merge fnd2 -> fnd. */
+	for (level = 0; level < fnd2->level; level++) {
+		fnd_push(fnd, fnd2->nodes[level], fnd2->de[level]);
+		fnd2->nodes[level] = NULL;
+	}
+	fnd2->level = 0;
+
+	hdr = NULL;
+	for (level = fnd->level; level; level--) {
+		struct indx_node *in = fnd->nodes[level - 1];
+
+		ib = in->index;
+		if (ib_is_empty(ib)) {
+			sub_vbn = ib->vbn;
+		} else {
+			hdr = &ib->ihdr;
+			n2d = in;
+			level2 = level;
+			break;
+		}
+	}
+
+	if (!hdr)
+		hdr = &root->ihdr;
+
+	e = hdr_first_de(hdr);
+	if (!e) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	if (hdr != &root->ihdr || !de_is_last(e)) {
+		prev = NULL;
+		while (!de_is_last(e)) {
+			if (de_has_vcn(e) && sub_vbn == de_get_vbn_le(e))
+				break;
+			prev = e;
+			e = hdr_next_de(hdr, e);
+			if (!e) {
+				err = -EINVAL;
+				goto out;
+			}
+		}
+
+		if (sub_vbn != de_get_vbn_le(e)) {
+			/*
+			 * Didn't find the parent entry, although this buffer
+			 * is the parent trail. Something is corrupt.
+			 */
+			err = -EINVAL;
+			goto out;
+		}
+
+		if (de_is_last(e)) {
+			/*
+			 * Since we can't remove the end entry, we'll remove
+			 * its predecessor instead. This means we have to
+			 * transfer the predecessor's sub_vcn to the end entry.
+			 * Note: This index block is not empty, so the
+			 * predecessor must exist.
+			 */
+			if (!prev) {
+				err = -EINVAL;
+				goto out;
+			}
+
+			if (de_has_vcn(prev)) {
+				de_set_vbn_le(e, de_get_vbn_le(prev));
+			} else if (de_has_vcn(e)) {
+				le16_sub_cpu(&e->size, sizeof(u64));
+				e->flags &= ~NTFS_IE_HAS_SUBNODES;
+				le32_sub_cpu(&hdr->used, sizeof(u64));
+			}
+			e = prev;
+		}
+
+		/*
+		 * Copy the current entry into a temporary buffer (stripping
+		 * off its down-pointer, if any) and delete it from the current
+		 * buffer or root, as appropriate.
+		 */
+		e_size = le16_to_cpu(e->size);
+		me = kmemdup(e, e_size, GFP_NOFS);
+		if (!me) {
+			err = -ENOMEM;
+			goto out;
+		}
+
+		if (de_has_vcn(me)) {
+			me->flags &= ~NTFS_IE_HAS_SUBNODES;
+			le16_sub_cpu(&me->size, sizeof(u64));
+		}
+
+		hdr_delete_de(hdr, e);
+
+		if (hdr == &root->ihdr) {
+			level = 0;
+			hdr->total = hdr->used;
+
+			/* Shrink resident root attribute. */
+			mi_resize_attr(mi, attr, 0 - e_size);
+		} else {
+			indx_write(indx, ni, n2d, 0);
+			level = level2;
+		}
+
+		/* Mark unused buffers as free. */
+		trim_bit = -1;
+		for (; level < fnd->level; level++) {
+			ib = fnd->nodes[level]->index;
+			if (ib_is_empty(ib)) {
+				size_t k = le64_to_cpu(ib->vbn) >>
+					   indx->idx2vbn_bits;
+
+				indx_mark_free(indx, ni, k);
+				if (k < trim_bit)
+					trim_bit = k;
+			}
+		}
+
+		fnd_clear(fnd);
+		/*fnd->root_de = NULL;*/
+
+		/*
+		 * Re-insert the entry into the tree.
+		 * Find the spot the tree where we want to insert the new entry.
+		 */
+		err = indx_insert_entry(indx, ni, me, ctx, fnd, 0);
+		kfree(me);
+		if (err)
+			goto out;
+
+		if (trim_bit != -1)
+			indx_shrink(indx, ni, trim_bit);
+	} else {
+		/*
+		 * This tree needs to be collapsed down to an empty root.
+		 * Recreate the index root as an empty leaf and free all
+		 * the bits the index allocation bitmap.
+		 */
+		fnd_clear(fnd);
+		fnd_clear(fnd2);
+
+		in = &s_index_names[indx->type];
+
+		err = attr_set_size(ni, ATTR_ALLOC, in->name, in->name_len,
+				    &indx->alloc_run, 0, NULL, false, NULL);
+		err = ni_remove_attr(ni, ATTR_ALLOC, in->name, in->name_len,
+				     false, NULL);
+		run_close(&indx->alloc_run);
+
+		err = attr_set_size(ni, ATTR_BITMAP, in->name, in->name_len,
+				    &indx->bitmap_run, 0, NULL, false, NULL);
+		err = ni_remove_attr(ni, ATTR_BITMAP, in->name, in->name_len,
+				     false, NULL);
+		run_close(&indx->bitmap_run);
+
+		root = indx_get_root(indx, ni, &attr, &mi);
+		if (!root) {
+			err = -EINVAL;
+			goto out;
+		}
+
+		root_size = le32_to_cpu(attr->res.data_size);
+		new_root_size =
+			sizeof(struct INDEX_ROOT) + sizeof(struct NTFS_DE);
+
+		if (new_root_size != root_size &&
+		    !mi_resize_attr(mi, attr, new_root_size - root_size)) {
+			err = -EINVAL;
+			goto out;
+		}
+
+		/* Fill first entry. */
+		e = (struct NTFS_DE *)(root + 1);
+		e->ref.low = 0;
+		e->ref.high = 0;
+		e->ref.seq = 0;
+		e->size = cpu_to_le16(sizeof(struct NTFS_DE));
+		e->flags = NTFS_IE_LAST; // 0x02
+		e->key_size = 0;
+		e->res = 0;
+
+		hdr = &root->ihdr;
+		hdr->flags = 0;
+		hdr->used = hdr->total = cpu_to_le32(
+			new_root_size - offsetof(struct INDEX_ROOT, ihdr));
+		mi->dirty = true;
+	}
+
+out:
+	fnd_put(fnd2);
+out1:
+	fnd_put(fnd);
+out2:
+	return err;
+}
+
+/*
+ * Update duplicated information in directory entry
+ * 'dup' - info from MFT record
+ */
+int indx_update_dup(struct ntfs_inode *ni, struct ntfs_sb_info *sbi,
+		    const struct ATTR_FILE_NAME *fname,
+		    const struct NTFS_DUP_INFO *dup, int sync)
+{
+	int err, diff;
+	struct NTFS_DE *e = NULL;
+	struct ATTR_FILE_NAME *e_fname;
+	struct ntfs_fnd *fnd;
+	struct INDEX_ROOT *root;
+	struct mft_inode *mi;
+	struct ntfs_index *indx = &ni->dir;
+
+	fnd = fnd_get();
+	if (!fnd)
+		return -ENOMEM;
+
+	root = indx_get_root(indx, ni, NULL, &mi);
+	if (!root) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	/* Find entry in directory. */
+	err = indx_find(indx, ni, root, fname, fname_full_size(fname), sbi,
+			&diff, &e, fnd);
+	if (err)
+		goto out;
+
+	if (!e) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	if (diff) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	e_fname = (struct ATTR_FILE_NAME *)(e + 1);
+
+	if (!memcmp(&e_fname->dup, dup, sizeof(*dup))) {
+		/*
+		 * Nothing to update in index! Try to avoid this call.
+		 */
+		goto out;
+	}
+
+	memcpy(&e_fname->dup, dup, sizeof(*dup));
+
+	if (fnd->level) {
+		/* Directory entry in index. */
+		err = indx_write(indx, ni, fnd->nodes[fnd->level - 1], sync);
+	} else {
+		/* Directory entry in directory MFT record. */
+		mi->dirty = true;
+		if (sync)
+			err = mi_write(mi, 1);
+		else
+			mark_inode_dirty(&ni->vfs_inode);
+	}
+
+out:
+	fnd_put(fnd);
+	return err;
+}
diff --git a/fs/ntfs3/inode.c b/fs/ntfs3/inode.c
new file mode 100644
index 000000000000..db2a5a4c38e4
--- /dev/null
+++ b/fs/ntfs3/inode.c
@@ -0,0 +1,1957 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ *
+ * Copyright (C) 2019-2021 Paragon Software GmbH, All rights reserved.
+ *
+ */
+
+#include <linux/blkdev.h>
+#include <linux/buffer_head.h>
+#include <linux/fs.h>
+#include <linux/iversion.h>
+#include <linux/mpage.h>
+#include <linux/namei.h>
+#include <linux/nls.h>
+#include <linux/uio.h>
+#include <linux/writeback.h>
+
+#include "debug.h"
+#include "ntfs.h"
+#include "ntfs_fs.h"
+
+/*
+ * ntfs_read_mft - Read record and parses MFT.
+ */
+static struct inode *ntfs_read_mft(struct inode *inode,
+				   const struct cpu_str *name,
+				   const struct MFT_REF *ref)
+{
+	int err = 0;
+	struct ntfs_inode *ni = ntfs_i(inode);
+	struct super_block *sb = inode->i_sb;
+	struct ntfs_sb_info *sbi = sb->s_fs_info;
+	mode_t mode = 0;
+	struct ATTR_STD_INFO5 *std5 = NULL;
+	struct ATTR_LIST_ENTRY *le;
+	struct ATTRIB *attr;
+	bool is_match = false;
+	bool is_root = false;
+	bool is_dir;
+	unsigned long ino = inode->i_ino;
+	u32 rp_fa = 0, asize, t32;
+	u16 roff, rsize, names = 0;
+	const struct ATTR_FILE_NAME *fname = NULL;
+	const struct INDEX_ROOT *root;
+	struct REPARSE_DATA_BUFFER rp; // 0x18 bytes
+	u64 t64;
+	struct MFT_REC *rec;
+	struct runs_tree *run;
+
+	inode->i_op = NULL;
+	/* Setup 'uid' and 'gid' */
+	inode->i_uid = sbi->options.fs_uid;
+	inode->i_gid = sbi->options.fs_gid;
+
+	err = mi_init(&ni->mi, sbi, ino);
+	if (err)
+		goto out;
+
+	if (!sbi->mft.ni && ino == MFT_REC_MFT && !sb->s_root) {
+		t64 = sbi->mft.lbo >> sbi->cluster_bits;
+		t32 = bytes_to_cluster(sbi, MFT_REC_VOL * sbi->record_size);
+		sbi->mft.ni = ni;
+		init_rwsem(&ni->file.run_lock);
+
+		if (!run_add_entry(&ni->file.run, 0, t64, t32, true)) {
+			err = -ENOMEM;
+			goto out;
+		}
+	}
+
+	err = mi_read(&ni->mi, ino == MFT_REC_MFT);
+
+	if (err)
+		goto out;
+
+	rec = ni->mi.mrec;
+
+	if (sbi->flags & NTFS_FLAGS_LOG_REPLAYING) {
+		;
+	} else if (ref->seq != rec->seq) {
+		err = -EINVAL;
+		ntfs_err(sb, "MFT: r=%lx, expect seq=%x instead of %x!", ino,
+			 le16_to_cpu(ref->seq), le16_to_cpu(rec->seq));
+		goto out;
+	} else if (!is_rec_inuse(rec)) {
+		err = -EINVAL;
+		ntfs_err(sb, "Inode r=%x is not in use!", (u32)ino);
+		goto out;
+	}
+
+	if (le32_to_cpu(rec->total) != sbi->record_size) {
+		/* Bad inode? */
+		err = -EINVAL;
+		goto out;
+	}
+
+	if (!is_rec_base(rec))
+		goto Ok;
+
+	/* Record should contain $I30 root. */
+	is_dir = rec->flags & RECORD_FLAG_DIR;
+
+	inode->i_generation = le16_to_cpu(rec->seq);
+
+	/* Enumerate all struct Attributes MFT. */
+	le = NULL;
+	attr = NULL;
+
+	/*
+	 * To reduce tab pressure use goto instead of
+	 * while( (attr = ni_enum_attr_ex(ni, attr, &le, NULL) ))
+	 */
+next_attr:
+	run = NULL;
+	err = -EINVAL;
+	attr = ni_enum_attr_ex(ni, attr, &le, NULL);
+	if (!attr)
+		goto end_enum;
+
+	if (le && le->vcn) {
+		/* This is non primary attribute segment. Ignore if not MFT. */
+		if (ino != MFT_REC_MFT || attr->type != ATTR_DATA)
+			goto next_attr;
+
+		run = &ni->file.run;
+		asize = le32_to_cpu(attr->size);
+		goto attr_unpack_run;
+	}
+
+	roff = attr->non_res ? 0 : le16_to_cpu(attr->res.data_off);
+	rsize = attr->non_res ? 0 : le32_to_cpu(attr->res.data_size);
+	asize = le32_to_cpu(attr->size);
+
+	switch (attr->type) {
+	case ATTR_STD:
+		if (attr->non_res ||
+		    asize < sizeof(struct ATTR_STD_INFO) + roff ||
+		    rsize < sizeof(struct ATTR_STD_INFO))
+			goto out;
+
+		if (std5)
+			goto next_attr;
+
+		std5 = Add2Ptr(attr, roff);
+
+#ifdef STATX_BTIME
+		nt2kernel(std5->cr_time, &ni->i_crtime);
+#endif
+		nt2kernel(std5->a_time, &inode->i_atime);
+		nt2kernel(std5->c_time, &inode->i_ctime);
+		nt2kernel(std5->m_time, &inode->i_mtime);
+
+		ni->std_fa = std5->fa;
+
+		if (asize >= sizeof(struct ATTR_STD_INFO5) + roff &&
+		    rsize >= sizeof(struct ATTR_STD_INFO5))
+			ni->std_security_id = std5->security_id;
+		goto next_attr;
+
+	case ATTR_LIST:
+		if (attr->name_len || le || ino == MFT_REC_LOG)
+			goto out;
+
+		err = ntfs_load_attr_list(ni, attr);
+		if (err)
+			goto out;
+
+		le = NULL;
+		attr = NULL;
+		goto next_attr;
+
+	case ATTR_NAME:
+		if (attr->non_res || asize < SIZEOF_ATTRIBUTE_FILENAME + roff ||
+		    rsize < SIZEOF_ATTRIBUTE_FILENAME)
+			goto out;
+
+		fname = Add2Ptr(attr, roff);
+		if (fname->type == FILE_NAME_DOS)
+			goto next_attr;
+
+		names += 1;
+		if (name && name->len == fname->name_len &&
+		    !ntfs_cmp_names_cpu(name, (struct le_str *)&fname->name_len,
+					NULL, false))
+			is_match = true;
+
+		goto next_attr;
+
+	case ATTR_DATA:
+		if (is_dir) {
+			/* Ignore data attribute in dir record. */
+			goto next_attr;
+		}
+
+		if (ino == MFT_REC_BADCLUST && !attr->non_res)
+			goto next_attr;
+
+		if (attr->name_len &&
+		    ((ino != MFT_REC_BADCLUST || !attr->non_res ||
+		      attr->name_len != ARRAY_SIZE(BAD_NAME) ||
+		      memcmp(attr_name(attr), BAD_NAME, sizeof(BAD_NAME))) &&
+		     (ino != MFT_REC_SECURE || !attr->non_res ||
+		      attr->name_len != ARRAY_SIZE(SDS_NAME) ||
+		      memcmp(attr_name(attr), SDS_NAME, sizeof(SDS_NAME))))) {
+			/* File contains stream attribute. Ignore it. */
+			goto next_attr;
+		}
+
+		if (is_attr_sparsed(attr))
+			ni->std_fa |= FILE_ATTRIBUTE_SPARSE_FILE;
+		else
+			ni->std_fa &= ~FILE_ATTRIBUTE_SPARSE_FILE;
+
+		if (is_attr_compressed(attr))
+			ni->std_fa |= FILE_ATTRIBUTE_COMPRESSED;
+		else
+			ni->std_fa &= ~FILE_ATTRIBUTE_COMPRESSED;
+
+		if (is_attr_encrypted(attr))
+			ni->std_fa |= FILE_ATTRIBUTE_ENCRYPTED;
+		else
+			ni->std_fa &= ~FILE_ATTRIBUTE_ENCRYPTED;
+
+		if (!attr->non_res) {
+			ni->i_valid = inode->i_size = rsize;
+			inode_set_bytes(inode, rsize);
+			t32 = asize;
+		} else {
+			t32 = le16_to_cpu(attr->nres.run_off);
+		}
+
+		mode = S_IFREG | (0777 & sbi->options.fs_fmask_inv);
+
+		if (!attr->non_res) {
+			ni->ni_flags |= NI_FLAG_RESIDENT;
+			goto next_attr;
+		}
+
+		inode_set_bytes(inode, attr_ondisk_size(attr));
+
+		ni->i_valid = le64_to_cpu(attr->nres.valid_size);
+		inode->i_size = le64_to_cpu(attr->nres.data_size);
+		if (!attr->nres.alloc_size)
+			goto next_attr;
+
+		run = ino == MFT_REC_BITMAP ? &sbi->used.bitmap.run
+					    : &ni->file.run;
+		break;
+
+	case ATTR_ROOT:
+		if (attr->non_res)
+			goto out;
+
+		root = Add2Ptr(attr, roff);
+		is_root = true;
+
+		if (attr->name_len != ARRAY_SIZE(I30_NAME) ||
+		    memcmp(attr_name(attr), I30_NAME, sizeof(I30_NAME)))
+			goto next_attr;
+
+		if (root->type != ATTR_NAME ||
+		    root->rule != NTFS_COLLATION_TYPE_FILENAME)
+			goto out;
+
+		if (!is_dir)
+			goto next_attr;
+
+		ni->ni_flags |= NI_FLAG_DIR;
+
+		err = indx_init(&ni->dir, sbi, attr, INDEX_MUTEX_I30);
+		if (err)
+			goto out;
+
+		mode = sb->s_root
+			       ? (S_IFDIR | (0777 & sbi->options.fs_dmask_inv))
+			       : (S_IFDIR | 0777);
+		goto next_attr;
+
+	case ATTR_ALLOC:
+		if (!is_root || attr->name_len != ARRAY_SIZE(I30_NAME) ||
+		    memcmp(attr_name(attr), I30_NAME, sizeof(I30_NAME)))
+			goto next_attr;
+
+		inode->i_size = le64_to_cpu(attr->nres.data_size);
+		ni->i_valid = le64_to_cpu(attr->nres.valid_size);
+		inode_set_bytes(inode, le64_to_cpu(attr->nres.alloc_size));
+
+		run = &ni->dir.alloc_run;
+		break;
+
+	case ATTR_BITMAP:
+		if (ino == MFT_REC_MFT) {
+			if (!attr->non_res)
+				goto out;
+#ifndef CONFIG_NTFS3_64BIT_CLUSTER
+			/* 0x20000000 = 2^32 / 8 */
+			if (le64_to_cpu(attr->nres.alloc_size) >= 0x20000000)
+				goto out;
+#endif
+			run = &sbi->mft.bitmap.run;
+			break;
+		} else if (is_dir && attr->name_len == ARRAY_SIZE(I30_NAME) &&
+			   !memcmp(attr_name(attr), I30_NAME,
+				   sizeof(I30_NAME)) &&
+			   attr->non_res) {
+			run = &ni->dir.bitmap_run;
+			break;
+		}
+		goto next_attr;
+
+	case ATTR_REPARSE:
+		if (attr->name_len)
+			goto next_attr;
+
+		rp_fa = ni_parse_reparse(ni, attr, &rp);
+		switch (rp_fa) {
+		case REPARSE_LINK:
+			if (!attr->non_res) {
+				inode->i_size = rsize;
+				inode_set_bytes(inode, rsize);
+				t32 = asize;
+			} else {
+				inode->i_size =
+					le64_to_cpu(attr->nres.data_size);
+				t32 = le16_to_cpu(attr->nres.run_off);
+			}
+
+			/* Looks like normal symlink. */
+			ni->i_valid = inode->i_size;
+
+			/* Clear directory bit. */
+			if (ni->ni_flags & NI_FLAG_DIR) {
+				indx_clear(&ni->dir);
+				memset(&ni->dir, 0, sizeof(ni->dir));
+				ni->ni_flags &= ~NI_FLAG_DIR;
+			} else {
+				run_close(&ni->file.run);
+			}
+			mode = S_IFLNK | 0777;
+			is_dir = false;
+			if (attr->non_res) {
+				run = &ni->file.run;
+				goto attr_unpack_run; // Double break.
+			}
+			break;
+
+		case REPARSE_COMPRESSED:
+			break;
+
+		case REPARSE_DEDUPLICATED:
+			break;
+		}
+		goto next_attr;
+
+	case ATTR_EA_INFO:
+		if (!attr->name_len &&
+		    resident_data_ex(attr, sizeof(struct EA_INFO))) {
+			ni->ni_flags |= NI_FLAG_EA;
+			/*
+			 * ntfs_get_wsl_perm updates inode->i_uid, inode->i_gid, inode->i_mode
+			 */
+			inode->i_mode = mode;
+			ntfs_get_wsl_perm(inode);
+			mode = inode->i_mode;
+		}
+		goto next_attr;
+
+	default:
+		goto next_attr;
+	}
+
+attr_unpack_run:
+	roff = le16_to_cpu(attr->nres.run_off);
+
+	t64 = le64_to_cpu(attr->nres.svcn);
+	err = run_unpack_ex(run, sbi, ino, t64, le64_to_cpu(attr->nres.evcn),
+			    t64, Add2Ptr(attr, roff), asize - roff);
+	if (err < 0)
+		goto out;
+	err = 0;
+	goto next_attr;
+
+end_enum:
+
+	if (!std5)
+		goto out;
+
+	if (!is_match && name) {
+		/* Reuse rec as buffer for ascii name. */
+		err = -ENOENT;
+		goto out;
+	}
+
+	if (std5->fa & FILE_ATTRIBUTE_READONLY)
+		mode &= ~0222;
+
+	if (!names) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	if (names != le16_to_cpu(rec->hard_links)) {
+		/* Correct minor error on the fly. Do not mark inode as dirty. */
+		rec->hard_links = cpu_to_le16(names);
+		ni->mi.dirty = true;
+	}
+
+	set_nlink(inode, names);
+
+	if (S_ISDIR(mode)) {
+		ni->std_fa |= FILE_ATTRIBUTE_DIRECTORY;
+
+		/*
+		 * Dot and dot-dot should be included in count but was not
+		 * included in enumeration.
+		 * Usually a hard links to directories are disabled.
+		 */
+		inode->i_op = &ntfs_dir_inode_operations;
+		inode->i_fop = &ntfs_dir_operations;
+		ni->i_valid = 0;
+	} else if (S_ISLNK(mode)) {
+		ni->std_fa &= ~FILE_ATTRIBUTE_DIRECTORY;
+		inode->i_op = &ntfs_link_inode_operations;
+		inode->i_fop = NULL;
+		inode_nohighmem(inode); // ??
+	} else if (S_ISREG(mode)) {
+		ni->std_fa &= ~FILE_ATTRIBUTE_DIRECTORY;
+		inode->i_op = &ntfs_file_inode_operations;
+		inode->i_fop = &ntfs_file_operations;
+		inode->i_mapping->a_ops =
+			is_compressed(ni) ? &ntfs_aops_cmpr : &ntfs_aops;
+		if (ino != MFT_REC_MFT)
+			init_rwsem(&ni->file.run_lock);
+	} else if (S_ISCHR(mode) || S_ISBLK(mode) || S_ISFIFO(mode) ||
+		   S_ISSOCK(mode)) {
+		inode->i_op = &ntfs_special_inode_operations;
+		init_special_inode(inode, mode, inode->i_rdev);
+	} else if (fname && fname->home.low == cpu_to_le32(MFT_REC_EXTEND) &&
+		   fname->home.seq == cpu_to_le16(MFT_REC_EXTEND)) {
+		/* Records in $Extend are not a files or general directories. */
+	} else {
+		err = -EINVAL;
+		goto out;
+	}
+
+	if ((sbi->options.sys_immutable &&
+	     (std5->fa & FILE_ATTRIBUTE_SYSTEM)) &&
+	    !S_ISFIFO(mode) && !S_ISSOCK(mode) && !S_ISLNK(mode)) {
+		inode->i_flags |= S_IMMUTABLE;
+	} else {
+		inode->i_flags &= ~S_IMMUTABLE;
+	}
+
+	inode->i_mode = mode;
+	if (!(ni->ni_flags & NI_FLAG_EA)) {
+		/* If no xattr then no security (stored in xattr). */
+		inode->i_flags |= S_NOSEC;
+	}
+
+Ok:
+	if (ino == MFT_REC_MFT && !sb->s_root)
+		sbi->mft.ni = NULL;
+
+	unlock_new_inode(inode);
+
+	return inode;
+
+out:
+	if (ino == MFT_REC_MFT && !sb->s_root)
+		sbi->mft.ni = NULL;
+
+	iget_failed(inode);
+	return ERR_PTR(err);
+}
+
+/*
+ * ntfs_test_inode
+ *
+ * Return: 1 if match.
+ */
+static int ntfs_test_inode(struct inode *inode, void *data)
+{
+	struct MFT_REF *ref = data;
+
+	return ino_get(ref) == inode->i_ino;
+}
+
+static int ntfs_set_inode(struct inode *inode, void *data)
+{
+	const struct MFT_REF *ref = data;
+
+	inode->i_ino = ino_get(ref);
+	return 0;
+}
+
+struct inode *ntfs_iget5(struct super_block *sb, const struct MFT_REF *ref,
+			 const struct cpu_str *name)
+{
+	struct inode *inode;
+
+	inode = iget5_locked(sb, ino_get(ref), ntfs_test_inode, ntfs_set_inode,
+			     (void *)ref);
+	if (unlikely(!inode))
+		return ERR_PTR(-ENOMEM);
+
+	/* If this is a freshly allocated inode, need to read it now. */
+	if (inode->i_state & I_NEW)
+		inode = ntfs_read_mft(inode, name, ref);
+	else if (ref->seq != ntfs_i(inode)->mi.mrec->seq) {
+		/* Inode overlaps? */
+		make_bad_inode(inode);
+	}
+
+	return inode;
+}
+
+enum get_block_ctx {
+	GET_BLOCK_GENERAL = 0,
+	GET_BLOCK_WRITE_BEGIN = 1,
+	GET_BLOCK_DIRECT_IO_R = 2,
+	GET_BLOCK_DIRECT_IO_W = 3,
+	GET_BLOCK_BMAP = 4,
+};
+
+static noinline int ntfs_get_block_vbo(struct inode *inode, u64 vbo,
+				       struct buffer_head *bh, int create,
+				       enum get_block_ctx ctx)
+{
+	struct super_block *sb = inode->i_sb;
+	struct ntfs_sb_info *sbi = sb->s_fs_info;
+	struct ntfs_inode *ni = ntfs_i(inode);
+	struct page *page = bh->b_page;
+	u8 cluster_bits = sbi->cluster_bits;
+	u32 block_size = sb->s_blocksize;
+	u64 bytes, lbo, valid;
+	u32 off;
+	int err;
+	CLST vcn, lcn, len;
+	bool new;
+
+	/* Clear previous state. */
+	clear_buffer_new(bh);
+	clear_buffer_uptodate(bh);
+
+	/* Direct write uses 'create=0'. */
+	if (!create && vbo >= ni->i_valid) {
+		/* Out of valid. */
+		return 0;
+	}
+
+	if (vbo >= inode->i_size) {
+		/* Out of size. */
+		return 0;
+	}
+
+	if (is_resident(ni)) {
+		ni_lock(ni);
+		err = attr_data_read_resident(ni, page);
+		ni_unlock(ni);
+
+		if (!err)
+			set_buffer_uptodate(bh);
+		bh->b_size = block_size;
+		return err;
+	}
+
+	vcn = vbo >> cluster_bits;
+	off = vbo & sbi->cluster_mask;
+	new = false;
+
+	err = attr_data_get_block(ni, vcn, 1, &lcn, &len, create ? &new : NULL);
+	if (err)
+		goto out;
+
+	if (!len)
+		return 0;
+
+	bytes = ((u64)len << cluster_bits) - off;
+
+	if (lcn == SPARSE_LCN) {
+		if (!create) {
+			if (bh->b_size > bytes)
+				bh->b_size = bytes;
+			return 0;
+		}
+		WARN_ON(1);
+	}
+
+	if (new) {
+		set_buffer_new(bh);
+		if ((len << cluster_bits) > block_size)
+			ntfs_sparse_cluster(inode, page, vcn, len);
+	}
+
+	lbo = ((u64)lcn << cluster_bits) + off;
+
+	set_buffer_mapped(bh);
+	bh->b_bdev = sb->s_bdev;
+	bh->b_blocknr = lbo >> sb->s_blocksize_bits;
+
+	valid = ni->i_valid;
+
+	if (ctx == GET_BLOCK_DIRECT_IO_W) {
+		/* ntfs_direct_IO will update ni->i_valid. */
+		if (vbo >= valid)
+			set_buffer_new(bh);
+	} else if (create) {
+		/* Normal write. */
+		if (bytes > bh->b_size)
+			bytes = bh->b_size;
+
+		if (vbo >= valid)
+			set_buffer_new(bh);
+
+		if (vbo + bytes > valid) {
+			ni->i_valid = vbo + bytes;
+			mark_inode_dirty(inode);
+		}
+	} else if (vbo >= valid) {
+		/* Read out of valid data. */
+		/* Should never be here 'cause already checked. */
+		clear_buffer_mapped(bh);
+	} else if (vbo + bytes <= valid) {
+		/* Normal read. */
+	} else if (vbo + block_size <= valid) {
+		/* Normal short read. */
+		bytes = block_size;
+	} else {
+		/*
+		 * Read across valid size: vbo < valid && valid < vbo + block_size
+		 */
+		bytes = block_size;
+
+		if (page) {
+			u32 voff = valid - vbo;
+
+			bh->b_size = block_size;
+			off = vbo & (PAGE_SIZE - 1);
+			set_bh_page(bh, page, off);
+			ll_rw_block(REQ_OP_READ, 0, 1, &bh);
+			wait_on_buffer(bh);
+			if (!buffer_uptodate(bh)) {
+				err = -EIO;
+				goto out;
+			}
+			zero_user_segment(page, off + voff, off + block_size);
+		}
+	}
+
+	if (bh->b_size > bytes)
+		bh->b_size = bytes;
+
+#ifndef __LP64__
+	if (ctx == GET_BLOCK_DIRECT_IO_W || ctx == GET_BLOCK_DIRECT_IO_R) {
+		static_assert(sizeof(size_t) < sizeof(loff_t));
+		if (bytes > 0x40000000u)
+			bh->b_size = 0x40000000u;
+	}
+#endif
+
+	return 0;
+
+out:
+	return err;
+}
+
+int ntfs_get_block(struct inode *inode, sector_t vbn,
+		   struct buffer_head *bh_result, int create)
+{
+	return ntfs_get_block_vbo(inode, (u64)vbn << inode->i_blkbits,
+				  bh_result, create, GET_BLOCK_GENERAL);
+}
+
+static int ntfs_get_block_bmap(struct inode *inode, sector_t vsn,
+			       struct buffer_head *bh_result, int create)
+{
+	return ntfs_get_block_vbo(inode,
+				  (u64)vsn << inode->i_sb->s_blocksize_bits,
+				  bh_result, create, GET_BLOCK_BMAP);
+}
+
+static sector_t ntfs_bmap(struct address_space *mapping, sector_t block)
+{
+	return generic_block_bmap(mapping, block, ntfs_get_block_bmap);
+}
+
+static int ntfs_readpage(struct file *file, struct page *page)
+{
+	int err;
+	struct address_space *mapping = page->mapping;
+	struct inode *inode = mapping->host;
+	struct ntfs_inode *ni = ntfs_i(inode);
+
+	if (is_resident(ni)) {
+		ni_lock(ni);
+		err = attr_data_read_resident(ni, page);
+		ni_unlock(ni);
+		if (err != E_NTFS_NONRESIDENT) {
+			unlock_page(page);
+			return err;
+		}
+	}
+
+	if (is_compressed(ni)) {
+		ni_lock(ni);
+		err = ni_readpage_cmpr(ni, page);
+		ni_unlock(ni);
+		return err;
+	}
+
+	/* Normal + sparse files. */
+	return mpage_readpage(page, ntfs_get_block);
+}
+
+static void ntfs_readahead(struct readahead_control *rac)
+{
+	struct address_space *mapping = rac->mapping;
+	struct inode *inode = mapping->host;
+	struct ntfs_inode *ni = ntfs_i(inode);
+	u64 valid;
+	loff_t pos;
+
+	if (is_resident(ni)) {
+		/* No readahead for resident. */
+		return;
+	}
+
+	if (is_compressed(ni)) {
+		/* No readahead for compressed. */
+		return;
+	}
+
+	valid = ni->i_valid;
+	pos = readahead_pos(rac);
+
+	if (valid < i_size_read(inode) && pos <= valid &&
+	    valid < pos + readahead_length(rac)) {
+		/* Range cross 'valid'. Read it page by page. */
+		return;
+	}
+
+	mpage_readahead(rac, ntfs_get_block);
+}
+
+static int ntfs_get_block_direct_IO_R(struct inode *inode, sector_t iblock,
+				      struct buffer_head *bh_result, int create)
+{
+	return ntfs_get_block_vbo(inode, (u64)iblock << inode->i_blkbits,
+				  bh_result, create, GET_BLOCK_DIRECT_IO_R);
+}
+
+static int ntfs_get_block_direct_IO_W(struct inode *inode, sector_t iblock,
+				      struct buffer_head *bh_result, int create)
+{
+	return ntfs_get_block_vbo(inode, (u64)iblock << inode->i_blkbits,
+				  bh_result, create, GET_BLOCK_DIRECT_IO_W);
+}
+
+static ssize_t ntfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
+{
+	struct file *file = iocb->ki_filp;
+	struct address_space *mapping = file->f_mapping;
+	struct inode *inode = mapping->host;
+	struct ntfs_inode *ni = ntfs_i(inode);
+	loff_t vbo = iocb->ki_pos;
+	loff_t end;
+	int wr = iov_iter_rw(iter) & WRITE;
+	loff_t valid;
+	ssize_t ret;
+
+	if (is_resident(ni)) {
+		/* Switch to buffered write. */
+		ret = 0;
+		goto out;
+	}
+
+	ret = blockdev_direct_IO(iocb, inode, iter,
+				 wr ? ntfs_get_block_direct_IO_W
+				    : ntfs_get_block_direct_IO_R);
+
+	if (ret <= 0)
+		goto out;
+
+	end = vbo + ret;
+	valid = ni->i_valid;
+	if (wr) {
+		if (end > valid && !S_ISBLK(inode->i_mode)) {
+			ni->i_valid = end;
+			mark_inode_dirty(inode);
+		}
+	} else if (vbo < valid && valid < end) {
+		/* Fix page. */
+		iov_iter_revert(iter, end - valid);
+		iov_iter_zero(end - valid, iter);
+	}
+
+out:
+	return ret;
+}
+
+int ntfs_set_size(struct inode *inode, u64 new_size)
+{
+	struct super_block *sb = inode->i_sb;
+	struct ntfs_sb_info *sbi = sb->s_fs_info;
+	struct ntfs_inode *ni = ntfs_i(inode);
+	int err;
+
+	/* Check for maximum file size. */
+	if (is_sparsed(ni) || is_compressed(ni)) {
+		if (new_size > sbi->maxbytes_sparse) {
+			err = -EFBIG;
+			goto out;
+		}
+	} else if (new_size > sbi->maxbytes) {
+		err = -EFBIG;
+		goto out;
+	}
+
+	ni_lock(ni);
+	down_write(&ni->file.run_lock);
+
+	err = attr_set_size(ni, ATTR_DATA, NULL, 0, &ni->file.run, new_size,
+			    &ni->i_valid, true, NULL);
+
+	up_write(&ni->file.run_lock);
+	ni_unlock(ni);
+
+	mark_inode_dirty(inode);
+
+out:
+	return err;
+}
+
+static int ntfs_writepage(struct page *page, struct writeback_control *wbc)
+{
+	struct address_space *mapping = page->mapping;
+	struct inode *inode = mapping->host;
+	struct ntfs_inode *ni = ntfs_i(inode);
+	int err;
+
+	if (is_resident(ni)) {
+		ni_lock(ni);
+		err = attr_data_write_resident(ni, page);
+		ni_unlock(ni);
+		if (err != E_NTFS_NONRESIDENT) {
+			unlock_page(page);
+			return err;
+		}
+	}
+
+	return block_write_full_page(page, ntfs_get_block, wbc);
+}
+
+static int ntfs_writepages(struct address_space *mapping,
+			   struct writeback_control *wbc)
+{
+	struct inode *inode = mapping->host;
+	struct ntfs_inode *ni = ntfs_i(inode);
+	/* Redirect call to 'ntfs_writepage' for resident files. */
+	get_block_t *get_block = is_resident(ni) ? NULL : &ntfs_get_block;
+
+	return mpage_writepages(mapping, wbc, get_block);
+}
+
+static int ntfs_get_block_write_begin(struct inode *inode, sector_t vbn,
+				      struct buffer_head *bh_result, int create)
+{
+	return ntfs_get_block_vbo(inode, (u64)vbn << inode->i_blkbits,
+				  bh_result, create, GET_BLOCK_WRITE_BEGIN);
+}
+
+static int ntfs_write_begin(struct file *file, struct address_space *mapping,
+			    loff_t pos, u32 len, u32 flags, struct page **pagep,
+			    void **fsdata)
+{
+	int err;
+	struct inode *inode = mapping->host;
+	struct ntfs_inode *ni = ntfs_i(inode);
+
+	*pagep = NULL;
+	if (is_resident(ni)) {
+		struct page *page = grab_cache_page_write_begin(
+			mapping, pos >> PAGE_SHIFT, flags);
+
+		if (!page) {
+			err = -ENOMEM;
+			goto out;
+		}
+
+		ni_lock(ni);
+		err = attr_data_read_resident(ni, page);
+		ni_unlock(ni);
+
+		if (!err) {
+			*pagep = page;
+			goto out;
+		}
+		unlock_page(page);
+		put_page(page);
+
+		if (err != E_NTFS_NONRESIDENT)
+			goto out;
+	}
+
+	err = block_write_begin(mapping, pos, len, flags, pagep,
+				ntfs_get_block_write_begin);
+
+out:
+	return err;
+}
+
+/*
+ * ntfs_write_end - Address_space_operations::write_end.
+ */
+static int ntfs_write_end(struct file *file, struct address_space *mapping,
+			  loff_t pos, u32 len, u32 copied, struct page *page,
+			  void *fsdata)
+
+{
+	struct inode *inode = mapping->host;
+	struct ntfs_inode *ni = ntfs_i(inode);
+	u64 valid = ni->i_valid;
+	bool dirty = false;
+	int err;
+
+	if (is_resident(ni)) {
+		ni_lock(ni);
+		err = attr_data_write_resident(ni, page);
+		ni_unlock(ni);
+		if (!err) {
+			dirty = true;
+			/* Clear any buffers in page. */
+			if (page_has_buffers(page)) {
+				struct buffer_head *head, *bh;
+
+				bh = head = page_buffers(page);
+				do {
+					clear_buffer_dirty(bh);
+					clear_buffer_mapped(bh);
+					set_buffer_uptodate(bh);
+				} while (head != (bh = bh->b_this_page));
+			}
+			SetPageUptodate(page);
+			err = copied;
+		}
+		unlock_page(page);
+		put_page(page);
+	} else {
+		err = generic_write_end(file, mapping, pos, len, copied, page,
+					fsdata);
+	}
+
+	if (err >= 0) {
+		if (!(ni->std_fa & FILE_ATTRIBUTE_ARCHIVE)) {
+			inode->i_ctime = inode->i_mtime = current_time(inode);
+			ni->std_fa |= FILE_ATTRIBUTE_ARCHIVE;
+			dirty = true;
+		}
+
+		if (valid != ni->i_valid) {
+			/* ni->i_valid is changed in ntfs_get_block_vbo. */
+			dirty = true;
+		}
+
+		if (dirty)
+			mark_inode_dirty(inode);
+	}
+
+	return err;
+}
+
+int reset_log_file(struct inode *inode)
+{
+	int err;
+	loff_t pos = 0;
+	u32 log_size = inode->i_size;
+	struct address_space *mapping = inode->i_mapping;
+
+	for (;;) {
+		u32 len;
+		void *kaddr;
+		struct page *page;
+
+		len = pos + PAGE_SIZE > log_size ? (log_size - pos) : PAGE_SIZE;
+
+		err = block_write_begin(mapping, pos, len, 0, &page,
+					ntfs_get_block_write_begin);
+		if (err)
+			goto out;
+
+		kaddr = kmap_atomic(page);
+		memset(kaddr, -1, len);
+		kunmap_atomic(kaddr);
+		flush_dcache_page(page);
+
+		err = block_write_end(NULL, mapping, pos, len, len, page, NULL);
+		if (err < 0)
+			goto out;
+		pos += len;
+
+		if (pos >= log_size)
+			break;
+		balance_dirty_pages_ratelimited(mapping);
+	}
+out:
+	mark_inode_dirty_sync(inode);
+
+	return err;
+}
+
+int ntfs3_write_inode(struct inode *inode, struct writeback_control *wbc)
+{
+	return _ni_write_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
+}
+
+int ntfs_sync_inode(struct inode *inode)
+{
+	return _ni_write_inode(inode, 1);
+}
+
+/*
+ * writeback_inode - Helper function for ntfs_flush_inodes().
+ *
+ * This writes both the inode and the file data blocks, waiting
+ * for in flight data blocks before the start of the call.  It
+ * does not wait for any io started during the call.
+ */
+static int writeback_inode(struct inode *inode)
+{
+	int ret = sync_inode_metadata(inode, 0);
+
+	if (!ret)
+		ret = filemap_fdatawrite(inode->i_mapping);
+	return ret;
+}
+
+/*
+ * ntfs_flush_inodes
+ *
+ * Write data and metadata corresponding to i1 and i2.  The io is
+ * started but we do not wait for any of it to finish.
+ *
+ * filemap_flush() is used for the block device, so if there is a dirty
+ * page for a block already in flight, we will not wait and start the
+ * io over again.
+ */
+int ntfs_flush_inodes(struct super_block *sb, struct inode *i1,
+		      struct inode *i2)
+{
+	int ret = 0;
+
+	if (i1)
+		ret = writeback_inode(i1);
+	if (!ret && i2)
+		ret = writeback_inode(i2);
+	if (!ret)
+		ret = filemap_flush(sb->s_bdev->bd_inode->i_mapping);
+	return ret;
+}
+
+int inode_write_data(struct inode *inode, const void *data, size_t bytes)
+{
+	pgoff_t idx;
+
+	/* Write non resident data. */
+	for (idx = 0; bytes; idx++) {
+		size_t op = bytes > PAGE_SIZE ? PAGE_SIZE : bytes;
+		struct page *page = ntfs_map_page(inode->i_mapping, idx);
+
+		if (IS_ERR(page))
+			return PTR_ERR(page);
+
+		lock_page(page);
+		WARN_ON(!PageUptodate(page));
+		ClearPageUptodate(page);
+
+		memcpy(page_address(page), data, op);
+
+		flush_dcache_page(page);
+		SetPageUptodate(page);
+		unlock_page(page);
+
+		ntfs_unmap_page(page);
+
+		bytes -= op;
+		data = Add2Ptr(data, PAGE_SIZE);
+	}
+	return 0;
+}
+
+/*
+ * ntfs_reparse_bytes
+ *
+ * Number of bytes for REPARSE_DATA_BUFFER(IO_REPARSE_TAG_SYMLINK)
+ * for unicode string of @uni_len length.
+ */
+static inline u32 ntfs_reparse_bytes(u32 uni_len)
+{
+	/* Header + unicode string + decorated unicode string. */
+	return sizeof(short) * (2 * uni_len + 4) +
+	       offsetof(struct REPARSE_DATA_BUFFER,
+			SymbolicLinkReparseBuffer.PathBuffer);
+}
+
+static struct REPARSE_DATA_BUFFER *
+ntfs_create_reparse_buffer(struct ntfs_sb_info *sbi, const char *symname,
+			   u32 size, u16 *nsize)
+{
+	int i, err;
+	struct REPARSE_DATA_BUFFER *rp;
+	__le16 *rp_name;
+	typeof(rp->SymbolicLinkReparseBuffer) *rs;
+
+	rp = kzalloc(ntfs_reparse_bytes(2 * size + 2), GFP_NOFS);
+	if (!rp)
+		return ERR_PTR(-ENOMEM);
+
+	rs = &rp->SymbolicLinkReparseBuffer;
+	rp_name = rs->PathBuffer;
+
+	/* Convert link name to UTF-16. */
+	err = ntfs_nls_to_utf16(sbi, symname, size,
+				(struct cpu_str *)(rp_name - 1), 2 * size,
+				UTF16_LITTLE_ENDIAN);
+	if (err < 0)
+		goto out;
+
+	/* err = the length of unicode name of symlink. */
+	*nsize = ntfs_reparse_bytes(err);
+
+	if (*nsize > sbi->reparse.max_size) {
+		err = -EFBIG;
+		goto out;
+	}
+
+	/* Translate Linux '/' into Windows '\'. */
+	for (i = 0; i < err; i++) {
+		if (rp_name[i] == cpu_to_le16('/'))
+			rp_name[i] = cpu_to_le16('\\');
+	}
+
+	rp->ReparseTag = IO_REPARSE_TAG_SYMLINK;
+	rp->ReparseDataLength =
+		cpu_to_le16(*nsize - offsetof(struct REPARSE_DATA_BUFFER,
+					      SymbolicLinkReparseBuffer));
+
+	/* PrintName + SubstituteName. */
+	rs->SubstituteNameOffset = cpu_to_le16(sizeof(short) * err);
+	rs->SubstituteNameLength = cpu_to_le16(sizeof(short) * err + 8);
+	rs->PrintNameLength = rs->SubstituteNameOffset;
+
+	/*
+	 * TODO: Use relative path if possible to allow Windows to
+	 * parse this path.
+	 * 0-absolute path 1- relative path (SYMLINK_FLAG_RELATIVE).
+	 */
+	rs->Flags = 0;
+
+	memmove(rp_name + err + 4, rp_name, sizeof(short) * err);
+
+	/* Decorate SubstituteName. */
+	rp_name += err;
+	rp_name[0] = cpu_to_le16('\\');
+	rp_name[1] = cpu_to_le16('?');
+	rp_name[2] = cpu_to_le16('?');
+	rp_name[3] = cpu_to_le16('\\');
+
+	return rp;
+out:
+	kfree(rp);
+	return ERR_PTR(err);
+}
+
+struct inode *ntfs_create_inode(struct user_namespace *mnt_userns,
+				struct inode *dir, struct dentry *dentry,
+				const struct cpu_str *uni, umode_t mode,
+				dev_t dev, const char *symname, u32 size,
+				struct ntfs_fnd *fnd)
+{
+	int err;
+	struct super_block *sb = dir->i_sb;
+	struct ntfs_sb_info *sbi = sb->s_fs_info;
+	const struct qstr *name = &dentry->d_name;
+	CLST ino = 0;
+	struct ntfs_inode *dir_ni = ntfs_i(dir);
+	struct ntfs_inode *ni = NULL;
+	struct inode *inode = NULL;
+	struct ATTRIB *attr;
+	struct ATTR_STD_INFO5 *std5;
+	struct ATTR_FILE_NAME *fname;
+	struct MFT_REC *rec;
+	u32 asize, dsize, sd_size;
+	enum FILE_ATTRIBUTE fa;
+	__le32 security_id = SECURITY_ID_INVALID;
+	CLST vcn;
+	const void *sd;
+	u16 t16, nsize = 0, aid = 0;
+	struct INDEX_ROOT *root, *dir_root;
+	struct NTFS_DE *e, *new_de = NULL;
+	struct REPARSE_DATA_BUFFER *rp = NULL;
+	bool rp_inserted = false;
+
+	dir_root = indx_get_root(&dir_ni->dir, dir_ni, NULL, NULL);
+	if (!dir_root)
+		return ERR_PTR(-EINVAL);
+
+	if (S_ISDIR(mode)) {
+		/* Use parent's directory attributes. */
+		fa = dir_ni->std_fa | FILE_ATTRIBUTE_DIRECTORY |
+		     FILE_ATTRIBUTE_ARCHIVE;
+		/*
+		 * By default child directory inherits parent attributes.
+		 * Root directory is hidden + system.
+		 * Make an exception for children in root.
+		 */
+		if (dir->i_ino == MFT_REC_ROOT)
+			fa &= ~(FILE_ATTRIBUTE_HIDDEN | FILE_ATTRIBUTE_SYSTEM);
+	} else if (S_ISLNK(mode)) {
+		/* It is good idea that link should be the same type (file/dir) as target */
+		fa = FILE_ATTRIBUTE_REPARSE_POINT;
+
+		/*
+		 * Linux: there are dir/file/symlink and so on.
+		 * NTFS: symlinks are "dir + reparse" or "file + reparse"
+		 * It is good idea to create:
+		 * dir + reparse if 'symname' points to directory
+		 * or
+		 * file + reparse if 'symname' points to file
+		 * Unfortunately kern_path hangs if symname contains 'dir'.
+		 */
+
+		/*
+		 *	struct path path;
+		 *
+		 *	if (!kern_path(symname, LOOKUP_FOLLOW, &path)){
+		 *		struct inode *target = d_inode(path.dentry);
+		 *
+		 *		if (S_ISDIR(target->i_mode))
+		 *			fa |= FILE_ATTRIBUTE_DIRECTORY;
+		 *		// if ( target->i_sb == sb ){
+		 *		//	use relative path?
+		 *		// }
+		 *		path_put(&path);
+		 *	}
+		 */
+	} else if (S_ISREG(mode)) {
+		if (sbi->options.sparse) {
+			/* Sparsed regular file, cause option 'sparse'. */
+			fa = FILE_ATTRIBUTE_SPARSE_FILE |
+			     FILE_ATTRIBUTE_ARCHIVE;
+		} else if (dir_ni->std_fa & FILE_ATTRIBUTE_COMPRESSED) {
+			/* Compressed regular file, if parent is compressed. */
+			fa = FILE_ATTRIBUTE_COMPRESSED | FILE_ATTRIBUTE_ARCHIVE;
+		} else {
+			/* Regular file, default attributes. */
+			fa = FILE_ATTRIBUTE_ARCHIVE;
+		}
+	} else {
+		fa = FILE_ATTRIBUTE_ARCHIVE;
+	}
+
+	if (!(mode & 0222))
+		fa |= FILE_ATTRIBUTE_READONLY;
+
+	/* Allocate PATH_MAX bytes. */
+	new_de = __getname();
+	if (!new_de) {
+		err = -ENOMEM;
+		goto out1;
+	}
+
+	/* Mark rw ntfs as dirty. it will be cleared at umount. */
+	ntfs_set_state(sbi, NTFS_DIRTY_DIRTY);
+
+	/* Step 1: allocate and fill new mft record. */
+	err = ntfs_look_free_mft(sbi, &ino, false, NULL, NULL);
+	if (err)
+		goto out2;
+
+	ni = ntfs_new_inode(sbi, ino, fa & FILE_ATTRIBUTE_DIRECTORY);
+	if (IS_ERR(ni)) {
+		err = PTR_ERR(ni);
+		ni = NULL;
+		goto out3;
+	}
+	inode = &ni->vfs_inode;
+	inode_init_owner(mnt_userns, inode, dir, mode);
+	mode = inode->i_mode;
+
+	inode->i_atime = inode->i_mtime = inode->i_ctime = ni->i_crtime =
+		current_time(inode);
+
+	rec = ni->mi.mrec;
+	rec->hard_links = cpu_to_le16(1);
+	attr = Add2Ptr(rec, le16_to_cpu(rec->attr_off));
+
+	/* Get default security id. */
+	sd = s_default_security;
+	sd_size = sizeof(s_default_security);
+
+	if (is_ntfs3(sbi)) {
+		security_id = dir_ni->std_security_id;
+		if (le32_to_cpu(security_id) < SECURITY_ID_FIRST) {
+			security_id = sbi->security.def_security_id;
+
+			if (security_id == SECURITY_ID_INVALID &&
+			    !ntfs_insert_security(sbi, sd, sd_size,
+						  &security_id, NULL))
+				sbi->security.def_security_id = security_id;
+		}
+	}
+
+	/* Insert standard info. */
+	std5 = Add2Ptr(attr, SIZEOF_RESIDENT);
+
+	if (security_id == SECURITY_ID_INVALID) {
+		dsize = sizeof(struct ATTR_STD_INFO);
+	} else {
+		dsize = sizeof(struct ATTR_STD_INFO5);
+		std5->security_id = security_id;
+		ni->std_security_id = security_id;
+	}
+	asize = SIZEOF_RESIDENT + dsize;
+
+	attr->type = ATTR_STD;
+	attr->size = cpu_to_le32(asize);
+	attr->id = cpu_to_le16(aid++);
+	attr->res.data_off = SIZEOF_RESIDENT_LE;
+	attr->res.data_size = cpu_to_le32(dsize);
+
+	std5->cr_time = std5->m_time = std5->c_time = std5->a_time =
+		kernel2nt(&inode->i_atime);
+
+	ni->std_fa = fa;
+	std5->fa = fa;
+
+	attr = Add2Ptr(attr, asize);
+
+	/* Insert file name. */
+	err = fill_name_de(sbi, new_de, name, uni);
+	if (err)
+		goto out4;
+
+	mi_get_ref(&ni->mi, &new_de->ref);
+
+	fname = (struct ATTR_FILE_NAME *)(new_de + 1);
+	mi_get_ref(&dir_ni->mi, &fname->home);
+	fname->dup.cr_time = fname->dup.m_time = fname->dup.c_time =
+		fname->dup.a_time = std5->cr_time;
+	fname->dup.alloc_size = fname->dup.data_size = 0;
+	fname->dup.fa = std5->fa;
+	fname->dup.ea_size = fname->dup.reparse = 0;
+
+	dsize = le16_to_cpu(new_de->key_size);
+	asize = ALIGN(SIZEOF_RESIDENT + dsize, 8);
+
+	attr->type = ATTR_NAME;
+	attr->size = cpu_to_le32(asize);
+	attr->res.data_off = SIZEOF_RESIDENT_LE;
+	attr->res.flags = RESIDENT_FLAG_INDEXED;
+	attr->id = cpu_to_le16(aid++);
+	attr->res.data_size = cpu_to_le32(dsize);
+	memcpy(Add2Ptr(attr, SIZEOF_RESIDENT), fname, dsize);
+
+	attr = Add2Ptr(attr, asize);
+
+	if (security_id == SECURITY_ID_INVALID) {
+		/* Insert security attribute. */
+		asize = SIZEOF_RESIDENT + ALIGN(sd_size, 8);
+
+		attr->type = ATTR_SECURE;
+		attr->size = cpu_to_le32(asize);
+		attr->id = cpu_to_le16(aid++);
+		attr->res.data_off = SIZEOF_RESIDENT_LE;
+		attr->res.data_size = cpu_to_le32(sd_size);
+		memcpy(Add2Ptr(attr, SIZEOF_RESIDENT), sd, sd_size);
+
+		attr = Add2Ptr(attr, asize);
+	}
+
+	attr->id = cpu_to_le16(aid++);
+	if (fa & FILE_ATTRIBUTE_DIRECTORY) {
+		/*
+		 * Regular directory or symlink to directory.
+		 * Create root attribute.
+		 */
+		dsize = sizeof(struct INDEX_ROOT) + sizeof(struct NTFS_DE);
+		asize = sizeof(I30_NAME) + SIZEOF_RESIDENT + dsize;
+
+		attr->type = ATTR_ROOT;
+		attr->size = cpu_to_le32(asize);
+
+		attr->name_len = ARRAY_SIZE(I30_NAME);
+		attr->name_off = SIZEOF_RESIDENT_LE;
+		attr->res.data_off =
+			cpu_to_le16(sizeof(I30_NAME) + SIZEOF_RESIDENT);
+		attr->res.data_size = cpu_to_le32(dsize);
+		memcpy(Add2Ptr(attr, SIZEOF_RESIDENT), I30_NAME,
+		       sizeof(I30_NAME));
+
+		root = Add2Ptr(attr, sizeof(I30_NAME) + SIZEOF_RESIDENT);
+		memcpy(root, dir_root, offsetof(struct INDEX_ROOT, ihdr));
+		root->ihdr.de_off =
+			cpu_to_le32(sizeof(struct INDEX_HDR)); // 0x10
+		root->ihdr.used = cpu_to_le32(sizeof(struct INDEX_HDR) +
+					      sizeof(struct NTFS_DE));
+		root->ihdr.total = root->ihdr.used;
+
+		e = Add2Ptr(root, sizeof(struct INDEX_ROOT));
+		e->size = cpu_to_le16(sizeof(struct NTFS_DE));
+		e->flags = NTFS_IE_LAST;
+	} else if (S_ISLNK(mode)) {
+		/*
+		 * Symlink to file.
+		 * Create empty resident data attribute.
+		 */
+		asize = SIZEOF_RESIDENT;
+
+		/* Insert empty ATTR_DATA */
+		attr->type = ATTR_DATA;
+		attr->size = cpu_to_le32(SIZEOF_RESIDENT);
+		attr->name_off = SIZEOF_RESIDENT_LE;
+		attr->res.data_off = SIZEOF_RESIDENT_LE;
+	} else if (S_ISREG(mode)) {
+		/*
+		 * Regular file. Create empty non resident data attribute.
+		 */
+		attr->type = ATTR_DATA;
+		attr->non_res = 1;
+		attr->nres.evcn = cpu_to_le64(-1ll);
+		if (fa & FILE_ATTRIBUTE_SPARSE_FILE) {
+			attr->size = cpu_to_le32(SIZEOF_NONRESIDENT_EX + 8);
+			attr->name_off = SIZEOF_NONRESIDENT_EX_LE;
+			attr->flags = ATTR_FLAG_SPARSED;
+			asize = SIZEOF_NONRESIDENT_EX + 8;
+		} else if (fa & FILE_ATTRIBUTE_COMPRESSED) {
+			attr->size = cpu_to_le32(SIZEOF_NONRESIDENT_EX + 8);
+			attr->name_off = SIZEOF_NONRESIDENT_EX_LE;
+			attr->flags = ATTR_FLAG_COMPRESSED;
+			attr->nres.c_unit = COMPRESSION_UNIT;
+			asize = SIZEOF_NONRESIDENT_EX + 8;
+		} else {
+			attr->size = cpu_to_le32(SIZEOF_NONRESIDENT + 8);
+			attr->name_off = SIZEOF_NONRESIDENT_LE;
+			asize = SIZEOF_NONRESIDENT + 8;
+		}
+		attr->nres.run_off = attr->name_off;
+	} else {
+		/*
+		 * Node. Create empty resident data attribute.
+		 */
+		attr->type = ATTR_DATA;
+		attr->size = cpu_to_le32(SIZEOF_RESIDENT);
+		attr->name_off = SIZEOF_RESIDENT_LE;
+		if (fa & FILE_ATTRIBUTE_SPARSE_FILE)
+			attr->flags = ATTR_FLAG_SPARSED;
+		else if (fa & FILE_ATTRIBUTE_COMPRESSED)
+			attr->flags = ATTR_FLAG_COMPRESSED;
+		attr->res.data_off = SIZEOF_RESIDENT_LE;
+		asize = SIZEOF_RESIDENT;
+		ni->ni_flags |= NI_FLAG_RESIDENT;
+	}
+
+	if (S_ISDIR(mode)) {
+		ni->ni_flags |= NI_FLAG_DIR;
+		err = indx_init(&ni->dir, sbi, attr, INDEX_MUTEX_I30);
+		if (err)
+			goto out4;
+	} else if (S_ISLNK(mode)) {
+		rp = ntfs_create_reparse_buffer(sbi, symname, size, &nsize);
+
+		if (IS_ERR(rp)) {
+			err = PTR_ERR(rp);
+			rp = NULL;
+			goto out4;
+		}
+
+		/*
+		 * Insert ATTR_REPARSE.
+		 */
+		attr = Add2Ptr(attr, asize);
+		attr->type = ATTR_REPARSE;
+		attr->id = cpu_to_le16(aid++);
+
+		/* Resident or non resident? */
+		asize = ALIGN(SIZEOF_RESIDENT + nsize, 8);
+		t16 = PtrOffset(rec, attr);
+
+		/* 0x78 - the size of EA + EAINFO to store WSL */
+		if (asize + t16 + 0x78 + 8 > sbi->record_size) {
+			CLST alen;
+			CLST clst = bytes_to_cluster(sbi, nsize);
+
+			/* Bytes per runs. */
+			t16 = sbi->record_size - t16 - SIZEOF_NONRESIDENT;
+
+			attr->non_res = 1;
+			attr->nres.evcn = cpu_to_le64(clst - 1);
+			attr->name_off = SIZEOF_NONRESIDENT_LE;
+			attr->nres.run_off = attr->name_off;
+			attr->nres.data_size = cpu_to_le64(nsize);
+			attr->nres.valid_size = attr->nres.data_size;
+			attr->nres.alloc_size =
+				cpu_to_le64(ntfs_up_cluster(sbi, nsize));
+
+			err = attr_allocate_clusters(sbi, &ni->file.run, 0, 0,
+						     clst, NULL, 0, &alen, 0,
+						     NULL);
+			if (err)
+				goto out5;
+
+			err = run_pack(&ni->file.run, 0, clst,
+				       Add2Ptr(attr, SIZEOF_NONRESIDENT), t16,
+				       &vcn);
+			if (err < 0)
+				goto out5;
+
+			if (vcn != clst) {
+				err = -EINVAL;
+				goto out5;
+			}
+
+			asize = SIZEOF_NONRESIDENT + ALIGN(err, 8);
+			inode->i_size = nsize;
+		} else {
+			attr->res.data_off = SIZEOF_RESIDENT_LE;
+			attr->res.data_size = cpu_to_le32(nsize);
+			memcpy(Add2Ptr(attr, SIZEOF_RESIDENT), rp, nsize);
+			inode->i_size = nsize;
+			nsize = 0;
+		}
+
+		attr->size = cpu_to_le32(asize);
+
+		err = ntfs_insert_reparse(sbi, IO_REPARSE_TAG_SYMLINK,
+					  &new_de->ref);
+		if (err)
+			goto out5;
+
+		rp_inserted = true;
+	}
+
+	attr = Add2Ptr(attr, asize);
+	attr->type = ATTR_END;
+
+	rec->used = cpu_to_le32(PtrOffset(rec, attr) + 8);
+	rec->next_attr_id = cpu_to_le16(aid);
+
+	/* Step 2: Add new name in index. */
+	err = indx_insert_entry(&dir_ni->dir, dir_ni, new_de, sbi, fnd, 0);
+	if (err)
+		goto out6;
+
+	inode->i_generation = le16_to_cpu(rec->seq);
+
+	dir->i_mtime = dir->i_ctime = inode->i_atime;
+
+	if (S_ISDIR(mode)) {
+		inode->i_op = &ntfs_dir_inode_operations;
+		inode->i_fop = &ntfs_dir_operations;
+	} else if (S_ISLNK(mode)) {
+		inode->i_op = &ntfs_link_inode_operations;
+		inode->i_fop = NULL;
+		inode->i_mapping->a_ops = &ntfs_aops;
+	} else if (S_ISREG(mode)) {
+		inode->i_op = &ntfs_file_inode_operations;
+		inode->i_fop = &ntfs_file_operations;
+		inode->i_mapping->a_ops =
+			is_compressed(ni) ? &ntfs_aops_cmpr : &ntfs_aops;
+		init_rwsem(&ni->file.run_lock);
+	} else {
+		inode->i_op = &ntfs_special_inode_operations;
+		init_special_inode(inode, mode, dev);
+	}
+
+#ifdef CONFIG_NTFS3_FS_POSIX_ACL
+	if (!S_ISLNK(mode) && (sb->s_flags & SB_POSIXACL)) {
+		err = ntfs_init_acl(mnt_userns, inode, dir);
+		if (err)
+			goto out6;
+	} else
+#endif
+	{
+		inode->i_flags |= S_NOSEC;
+	}
+
+	/* Write non resident data. */
+	if (nsize) {
+		err = ntfs_sb_write_run(sbi, &ni->file.run, 0, rp, nsize);
+		if (err)
+			goto out7;
+	}
+
+	/*
+	 * Call 'd_instantiate' after inode->i_op is set
+	 * but before finish_open.
+	 */
+	d_instantiate(dentry, inode);
+
+	ntfs_save_wsl_perm(inode);
+	mark_inode_dirty(dir);
+	mark_inode_dirty(inode);
+
+	/* Normal exit. */
+	goto out2;
+
+out7:
+
+	/* Undo 'indx_insert_entry'. */
+	indx_delete_entry(&dir_ni->dir, dir_ni, new_de + 1,
+			  le16_to_cpu(new_de->key_size), sbi);
+out6:
+	if (rp_inserted)
+		ntfs_remove_reparse(sbi, IO_REPARSE_TAG_SYMLINK, &new_de->ref);
+
+out5:
+	if (S_ISDIR(mode) || run_is_empty(&ni->file.run))
+		goto out4;
+
+	run_deallocate(sbi, &ni->file.run, false);
+
+out4:
+	clear_rec_inuse(rec);
+	clear_nlink(inode);
+	ni->mi.dirty = false;
+	discard_new_inode(inode);
+out3:
+	ntfs_mark_rec_free(sbi, ino);
+
+out2:
+	__putname(new_de);
+	kfree(rp);
+
+out1:
+	if (err)
+		return ERR_PTR(err);
+
+	unlock_new_inode(inode);
+
+	return inode;
+}
+
+int ntfs_link_inode(struct inode *inode, struct dentry *dentry)
+{
+	int err;
+	struct ntfs_inode *ni = ntfs_i(inode);
+	struct ntfs_sb_info *sbi = inode->i_sb->s_fs_info;
+	struct NTFS_DE *de;
+	struct ATTR_FILE_NAME *de_name;
+
+	/* Allocate PATH_MAX bytes. */
+	de = __getname();
+	if (!de)
+		return -ENOMEM;
+
+	/* Mark rw ntfs as dirty. It will be cleared at umount. */
+	ntfs_set_state(sbi, NTFS_DIRTY_DIRTY);
+
+	/* Construct 'de'. */
+	err = fill_name_de(sbi, de, &dentry->d_name, NULL);
+	if (err)
+		goto out;
+
+	de_name = (struct ATTR_FILE_NAME *)(de + 1);
+	/* Fill duplicate info. */
+	de_name->dup.cr_time = de_name->dup.m_time = de_name->dup.c_time =
+		de_name->dup.a_time = kernel2nt(&inode->i_ctime);
+	de_name->dup.alloc_size = de_name->dup.data_size =
+		cpu_to_le64(inode->i_size);
+	de_name->dup.fa = ni->std_fa;
+	de_name->dup.ea_size = de_name->dup.reparse = 0;
+
+	err = ni_add_name(ntfs_i(d_inode(dentry->d_parent)), ni, de);
+out:
+	__putname(de);
+	return err;
+}
+
+/*
+ * ntfs_unlink_inode
+ *
+ * inode_operations::unlink
+ * inode_operations::rmdir
+ */
+int ntfs_unlink_inode(struct inode *dir, const struct dentry *dentry)
+{
+	int err;
+	struct ntfs_sb_info *sbi = dir->i_sb->s_fs_info;
+	struct inode *inode = d_inode(dentry);
+	struct ntfs_inode *ni = ntfs_i(inode);
+	struct ntfs_inode *dir_ni = ntfs_i(dir);
+	struct NTFS_DE *de, *de2 = NULL;
+	int undo_remove;
+
+	if (ntfs_is_meta_file(sbi, ni->mi.rno))
+		return -EINVAL;
+
+	/* Allocate PATH_MAX bytes. */
+	de = __getname();
+	if (!de)
+		return -ENOMEM;
+
+	ni_lock(ni);
+
+	if (S_ISDIR(inode->i_mode) && !dir_is_empty(inode)) {
+		err = -ENOTEMPTY;
+		goto out;
+	}
+
+	err = fill_name_de(sbi, de, &dentry->d_name, NULL);
+	if (err < 0)
+		goto out;
+
+	undo_remove = 0;
+	err = ni_remove_name(dir_ni, ni, de, &de2, &undo_remove);
+
+	if (!err) {
+		drop_nlink(inode);
+		dir->i_mtime = dir->i_ctime = current_time(dir);
+		mark_inode_dirty(dir);
+		inode->i_ctime = dir->i_ctime;
+		if (inode->i_nlink)
+			mark_inode_dirty(inode);
+	} else if (!ni_remove_name_undo(dir_ni, ni, de, de2, undo_remove)) {
+		make_bad_inode(inode);
+		ntfs_inode_err(inode, "failed to undo unlink");
+		ntfs_set_state(sbi, NTFS_DIRTY_ERROR);
+	} else {
+		if (ni_is_dirty(dir))
+			mark_inode_dirty(dir);
+		if (ni_is_dirty(inode))
+			mark_inode_dirty(inode);
+	}
+
+out:
+	ni_unlock(ni);
+	__putname(de);
+	return err;
+}
+
+void ntfs_evict_inode(struct inode *inode)
+{
+	truncate_inode_pages_final(&inode->i_data);
+
+	if (inode->i_nlink)
+		_ni_write_inode(inode, inode_needs_sync(inode));
+
+	invalidate_inode_buffers(inode);
+	clear_inode(inode);
+
+	ni_clear(ntfs_i(inode));
+}
+
+static noinline int ntfs_readlink_hlp(struct inode *inode, char *buffer,
+				      int buflen)
+{
+	int i, err = 0;
+	struct ntfs_inode *ni = ntfs_i(inode);
+	struct super_block *sb = inode->i_sb;
+	struct ntfs_sb_info *sbi = sb->s_fs_info;
+	u64 i_size = inode->i_size;
+	u16 nlen = 0;
+	void *to_free = NULL;
+	struct REPARSE_DATA_BUFFER *rp;
+	struct le_str *uni;
+	struct ATTRIB *attr;
+
+	/* Reparse data present. Try to parse it. */
+	static_assert(!offsetof(struct REPARSE_DATA_BUFFER, ReparseTag));
+	static_assert(sizeof(u32) == sizeof(rp->ReparseTag));
+
+	*buffer = 0;
+
+	/* Read into temporal buffer. */
+	if (i_size > sbi->reparse.max_size || i_size <= sizeof(u32)) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	attr = ni_find_attr(ni, NULL, NULL, ATTR_REPARSE, NULL, 0, NULL, NULL);
+	if (!attr) {
+		err = -EINVAL;
+		goto out;
+	}
+
+	if (!attr->non_res) {
+		rp = resident_data_ex(attr, i_size);
+		if (!rp) {
+			err = -EINVAL;
+			goto out;
+		}
+	} else {
+		rp = kmalloc(i_size, GFP_NOFS);
+		if (!rp) {
+			err = -ENOMEM;
+			goto out;
+		}
+		to_free = rp;
+		err = ntfs_read_run_nb(sbi, &ni->file.run, 0, rp, i_size, NULL);
+		if (err)
+			goto out;
+	}
+
+	err = -EINVAL;
+
+	/* Microsoft Tag. */
+	switch (rp->ReparseTag) {
+	case IO_REPARSE_TAG_MOUNT_POINT:
+		/* Mount points and junctions. */
+		/* Can we use 'Rp->MountPointReparseBuffer.PrintNameLength'? */
+		if (i_size <= offsetof(struct REPARSE_DATA_BUFFER,
+				       MountPointReparseBuffer.PathBuffer))
+			goto out;
+		uni = Add2Ptr(rp,
+			      offsetof(struct REPARSE_DATA_BUFFER,
+				       MountPointReparseBuffer.PathBuffer) +
+				      le16_to_cpu(rp->MountPointReparseBuffer
+							  .PrintNameOffset) -
+				      2);
+		nlen = le16_to_cpu(rp->MountPointReparseBuffer.PrintNameLength);
+		break;
+
+	case IO_REPARSE_TAG_SYMLINK:
+		/* FolderSymbolicLink */
+		/* Can we use 'Rp->SymbolicLinkReparseBuffer.PrintNameLength'? */
+		if (i_size <= offsetof(struct REPARSE_DATA_BUFFER,
+				       SymbolicLinkReparseBuffer.PathBuffer))
+			goto out;
+		uni = Add2Ptr(rp,
+			      offsetof(struct REPARSE_DATA_BUFFER,
+				       SymbolicLinkReparseBuffer.PathBuffer) +
+				      le16_to_cpu(rp->SymbolicLinkReparseBuffer
+							  .PrintNameOffset) -
+				      2);
+		nlen = le16_to_cpu(
+			rp->SymbolicLinkReparseBuffer.PrintNameLength);
+		break;
+
+	case IO_REPARSE_TAG_CLOUD:
+	case IO_REPARSE_TAG_CLOUD_1:
+	case IO_REPARSE_TAG_CLOUD_2:
+	case IO_REPARSE_TAG_CLOUD_3:
+	case IO_REPARSE_TAG_CLOUD_4:
+	case IO_REPARSE_TAG_CLOUD_5:
+	case IO_REPARSE_TAG_CLOUD_6:
+	case IO_REPARSE_TAG_CLOUD_7:
+	case IO_REPARSE_TAG_CLOUD_8:
+	case IO_REPARSE_TAG_CLOUD_9:
+	case IO_REPARSE_TAG_CLOUD_A:
+	case IO_REPARSE_TAG_CLOUD_B:
+	case IO_REPARSE_TAG_CLOUD_C:
+	case IO_REPARSE_TAG_CLOUD_D:
+	case IO_REPARSE_TAG_CLOUD_E:
+	case IO_REPARSE_TAG_CLOUD_F:
+		err = sizeof("OneDrive") - 1;
+		if (err > buflen)
+			err = buflen;
+		memcpy(buffer, "OneDrive", err);
+		goto out;
+
+	default:
+		if (IsReparseTagMicrosoft(rp->ReparseTag)) {
+			/* Unknown Microsoft Tag. */
+			goto out;
+		}
+		if (!IsReparseTagNameSurrogate(rp->ReparseTag) ||
+		    i_size <= sizeof(struct REPARSE_POINT)) {
+			goto out;
+		}
+
+		/* Users tag. */
+		uni = Add2Ptr(rp, sizeof(struct REPARSE_POINT) - 2);
+		nlen = le16_to_cpu(rp->ReparseDataLength) -
+		       sizeof(struct REPARSE_POINT);
+	}
+
+	/* Convert nlen from bytes to UNICODE chars. */
+	nlen >>= 1;
+
+	/* Check that name is available. */
+	if (!nlen || &uni->name[nlen] > (__le16 *)Add2Ptr(rp, i_size))
+		goto out;
+
+	/* If name is already zero terminated then truncate it now. */
+	if (!uni->name[nlen - 1])
+		nlen -= 1;
+	uni->len = nlen;
+
+	err = ntfs_utf16_to_nls(sbi, uni, buffer, buflen);
+
+	if (err < 0)
+		goto out;
+
+	/* Translate Windows '\' into Linux '/'. */
+	for (i = 0; i < err; i++) {
+		if (buffer[i] == '\\')
+			buffer[i] = '/';
+	}
+
+	/* Always set last zero. */
+	buffer[err] = 0;
+out:
+	kfree(to_free);
+	return err;
+}
+
+static const char *ntfs_get_link(struct dentry *de, struct inode *inode,
+				 struct delayed_call *done)
+{
+	int err;
+	char *ret;
+
+	if (!de)
+		return ERR_PTR(-ECHILD);
+
+	ret = kmalloc(PAGE_SIZE, GFP_NOFS);
+	if (!ret)
+		return ERR_PTR(-ENOMEM);
+
+	err = ntfs_readlink_hlp(inode, ret, PAGE_SIZE);
+	if (err < 0) {
+		kfree(ret);
+		return ERR_PTR(err);
+	}
+
+	set_delayed_call(done, kfree_link, ret);
+
+	return ret;
+}
+
+// clang-format off
+const struct inode_operations ntfs_link_inode_operations = {
+	.get_link	= ntfs_get_link,
+	.setattr	= ntfs3_setattr,
+	.listxattr	= ntfs_listxattr,
+	.permission	= ntfs_permission,
+	.get_acl	= ntfs_get_acl,
+	.set_acl	= ntfs_set_acl,
+};
+
+const struct address_space_operations ntfs_aops = {
+	.readpage	= ntfs_readpage,
+	.readahead	= ntfs_readahead,
+	.writepage	= ntfs_writepage,
+	.writepages	= ntfs_writepages,
+	.write_begin	= ntfs_write_begin,
+	.write_end	= ntfs_write_end,
+	.direct_IO	= ntfs_direct_IO,
+	.bmap		= ntfs_bmap,
+	.set_page_dirty = __set_page_dirty_buffers,
+};
+
+const struct address_space_operations ntfs_aops_cmpr = {
+	.readpage	= ntfs_readpage,
+	.readahead	= ntfs_readahead,
+};
+// clang-format on
diff --git a/fs/ntfs3/lib/decompress_common.c b/fs/ntfs3/lib/decompress_common.c
new file mode 100644
index 000000000000..e96652240859
--- /dev/null
+++ b/fs/ntfs3/lib/decompress_common.c
@@ -0,0 +1,319 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * decompress_common.c - Code shared by the XPRESS and LZX decompressors
+ *
+ * Copyright (C) 2015 Eric Biggers