Merge tag 'xfs-for-linus-4.1-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/dgc/linux-xfs

Pull xfs update from Dave Chinner:
 "This update contains:

   - RENAME_WHITEOUT support

   - conversion of per-cpu superblock accounting to use generic counters

   - new inode mmap lock so that we can lock page faults out of
     truncate, hole punch and other direct extent manipulation functions
     to avoid racing mmap writes from causing data corruption

   - rework of direct IO submission and completion to solve data
     corruption issue when running concurrent extending DIO writes.
     Also solves problem of running IO completion transactions in
     interrupt context during size extending AIO writes.

   - FALLOC_FL_INSERT_RANGE support for inserting holes into a file via
     direct extent manipulation to avoid needing to copy data within the
     file

   - attribute block header field overflow fix for 64k block size
     filesystems

   - Lots of changes to log messaging to be more informative and concise
     when errors occur.  Also prevent a lot of unnecessary log spamming
     due to cascading failures in error conditions.

   - lots of cleanups and bug fixes

  One thing of note is the direct IO fixes that we merged last week
  after the window opened.  Even though a little late, they fix a user
  reported data corruption and have been pretty well tested.  I figured
  there was not much point waiting another 2 weeks for -rc1 to be
  released just so I could send them to you..."

* tag 'xfs-for-linus-4.1-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/dgc/linux-xfs: (49 commits)
  xfs: using generic_file_direct_write() is unnecessary
  xfs: direct IO EOF zeroing needs to drain AIO
  xfs: DIO write completion size updates race
  xfs: DIO writes within EOF don't need an ioend
  xfs: handle DIO overwrite EOF update completion correctly
  xfs: DIO needs an ioend for writes
  xfs: move DIO mapping size calculation
  xfs: factor DIO write mapping from get_blocks
  xfs: unlock i_mutex in xfs_break_layouts
  xfs: kill unnecessary firstused overflow check on attr3 leaf removal
  xfs: use larger in-core attr firstused field and detect overflow
  xfs: pass attr geometry to attr leaf header conversion functions
  xfs: disallow ro->rw remount on norecovery mount
  xfs: xfs_shift_file_space can be static
  xfs: Add support FALLOC_FL_INSERT_RANGE for fallocate
  fs: Add support FALLOC_FL_INSERT_RANGE for fallocate
  xfs: Fix incorrect positive ENOMEM return
  xfs: xfs_mru_cache_insert() should use GFP_NOFS
  xfs: %pF is only for function pointers
  xfs: fix shadow warning in xfs_da3_root_split()
  ...

1 files changed, 32 insertions, 59 deletions

diff --git a/fs/xfs/xfs_iops.c b/fs/xfs/xfs_iops.c
index e53a90331422..2f1839e4dd1b 100644
--- a/fs/xfs/xfs_iops.c
+++ b/fs/xfs/xfs_iops.c
@@ -187,6 +187,8 @@ xfs_generic_create(
 	else
 		d_instantiate(dentry, inode);
 
+	xfs_finish_inode_setup(ip);
+
  out_free_acl:
 	if (default_acl)
 		posix_acl_release(default_acl);
@@ -195,6 +197,7 @@ xfs_generic_create(
 	return error;
 
  out_cleanup_inode:
+	xfs_finish_inode_setup(ip);
 	if (!tmpfile)
 		xfs_cleanup_inode(dir, inode, dentry);
 	iput(inode);
@@ -367,9 +370,11 @@ xfs_vn_symlink(
 		goto out_cleanup_inode;
 
 	d_instantiate(dentry, inode);
+	xfs_finish_inode_setup(cip);
 	return 0;
 
  out_cleanup_inode:
+	xfs_finish_inode_setup(cip);
 	xfs_cleanup_inode(dir, inode, dentry);
 	iput(inode);
  out:
@@ -389,7 +394,7 @@ xfs_vn_rename(
 	struct xfs_name	oname;
 	struct xfs_name	nname;
 
-	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE))
+	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
 		return -EINVAL;
 
 	/* if we are exchanging files, we need to set i_mode of both files */
@@ -766,6 +771,7 @@ xfs_setattr_size(
 		return error;
 
 	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
+	ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
 	ASSERT(S_ISREG(ip->i_d.di_mode));
 	ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
 		ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
@@ -829,55 +835,27 @@ xfs_setattr_size(
 	inode_dio_wait(inode);
 
 	/*
-	 * Do all the page cache truncate work outside the transaction context
-	 * as the "lock" order is page lock->log space reservation.  i.e.
-	 * locking pages inside the transaction can ABBA deadlock with
-	 * writeback. We have to do the VFS inode size update before we truncate
-	 * the pagecache, however, to avoid racing with page faults beyond the
-	 * new EOF they are not serialised against truncate operations except by
-	 * page locks and size updates.
+	 * We've already locked out new page faults, so now we can safely remove
+	 * pages from the page cache knowing they won't get refaulted until we
+	 * drop the XFS_MMAP_EXCL lock after the extent manipulations are
+	 * complete. The truncate_setsize() call also cleans partial EOF page
+	 * PTEs on extending truncates and hence ensures sub-page block size
+	 * filesystems are correctly handled, too.
 	 *
-	 * Hence we are in a situation where a truncate can fail with ENOMEM
-	 * from xfs_trans_reserve(), but having already truncated the in-memory
-	 * version of the file (i.e. made user visible changes). There's not
-	 * much we can do about this, except to hope that the caller sees ENOMEM
-	 * and retries the truncate operation.
+	 * We have to do all the page cache truncate work outside the
+	 * transaction context as the "lock" order is page lock->log space
+	 * reservation as defined by extent allocation in the writeback path.
+	 * Hence a truncate can fail with ENOMEM from xfs_trans_reserve(), but
+	 * having already truncated the in-memory version of the file (i.e. made
+	 * user visible changes). There's not much we can do about this, except
+	 * to hope that the caller sees ENOMEM and retries the truncate
+	 * operation.
 	 */
 	error = block_truncate_page(inode->i_mapping, newsize, xfs_get_blocks);
 	if (error)
 		return error;
 	truncate_setsize(inode, newsize);
 
-	/*
-	 * The "we can't serialise against page faults" pain gets worse.
-	 *
-	 * If the file is mapped then we have to clean the page at the old EOF
-	 * when extending the file. Extending the file can expose changes the
-	 * underlying page mapping (e.g. from beyond EOF to a hole or
-	 * unwritten), and so on the next attempt to write to that page we need
-	 * to remap it for write. i.e. we need .page_mkwrite() to be called.
-	 * Hence we need to clean the page to clean the pte and so a new write
-	 * fault will be triggered appropriately.
-	 *
-	 * If we do it before we change the inode size, then we can race with a
-	 * page fault that maps the page with exactly the same problem. If we do
-	 * it after we change the file size, then a new page fault can come in
-	 * and allocate space before we've run the rest of the truncate
-	 * transaction. That's kinda grotesque, but it's better than have data
-	 * over a hole, and so that's the lesser evil that has been chosen here.
-	 *
-	 * The real solution, however, is to have some mechanism for locking out
-	 * page faults while a truncate is in progress.
-	 */
-	if (newsize > oldsize && mapping_mapped(VFS_I(ip)->i_mapping)) {
-		error = filemap_write_and_wait_range(
-				VFS_I(ip)->i_mapping,
-				round_down(oldsize, PAGE_CACHE_SIZE),
-				round_up(oldsize, PAGE_CACHE_SIZE) - 1);
-		if (error)
-			return error;
-	}
-
 	tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
 	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
 	if (error)
@@ -975,9 +953,13 @@ xfs_vn_setattr(
 		uint		iolock = XFS_IOLOCK_EXCL;
 
 		xfs_ilock(ip, iolock);
-		error = xfs_break_layouts(dentry->d_inode, &iolock);
-		if (!error)
+		error = xfs_break_layouts(dentry->d_inode, &iolock, true);
+		if (!error) {
+			xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
+			iolock |= XFS_MMAPLOCK_EXCL;
+
 			error = xfs_setattr_size(ip, iattr);
+		}
 		xfs_iunlock(ip, iolock);
 	} else {
 		error = xfs_setattr_nonsize(ip, iattr, 0);
@@ -1228,16 +1210,12 @@ xfs_diflags_to_iflags(
 }
 
 /*
- * Initialize the Linux inode, set up the operation vectors and
- * unlock the inode.
+ * Initialize the Linux inode and set up the operation vectors.
  *
- * When reading existing inodes from disk this is called directly
- * from xfs_iget, when creating a new inode it is called from
- * xfs_ialloc after setting up the inode.
- *
- * We are always called with an uninitialised linux inode here.
- * We need to initialise the necessary fields and take a reference
- * on it.
+ * When reading existing inodes from disk this is called directly from xfs_iget,
+ * when creating a new inode it is called from xfs_ialloc after setting up the
+ * inode. These callers have different criteria for clearing XFS_INEW, so leave
+ * it up to the caller to deal with unlocking the inode appropriately.
  */
 void
 xfs_setup_inode(
@@ -1324,9 +1302,4 @@ xfs_setup_inode(
 		inode_has_no_xattr(inode);
 		cache_no_acl(inode);
 	}
-
-	xfs_iflags_clear(ip, XFS_INEW);
-	barrier();
-
-	unlock_new_inode(inode);
 }