move vfs_busy() to rwlocks and properly hide the locking api from vfs

ok tedu, pedro

13 files changed, 65 insertions, 63 deletions

diff --git a/sys/compat/common/vfs_syscalls_25.c b/sys/compat/common/vfs_syscalls_25.c
index 58cab278b23..f017af71580 100644
--- a/sys/compat/common/vfs_syscalls_25.c
+++ b/sys/compat/common/vfs_syscalls_25.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: vfs_syscalls_25.c,v 1.10 2006/04/30 14:20:07 sturm Exp $	*/
+/*	$OpenBSD: vfs_syscalls_25.c,v 1.11 2006/06/14 20:01:50 sturm Exp $	*/
 
 /*
  * Copyright (c) 1989, 1993
@@ -181,7 +181,7 @@ compat_25_sys_getfsstat(p, v, retval)
 
 	for (mp = CIRCLEQ_FIRST(&mountlist); mp != CIRCLEQ_END(&mountlist);
 	    mp = nmp) {
-		if (vfs_busy(mp, LK_NOWAIT)) {
+		if (vfs_busy(mp, VB_READ|VB_UMIGNORE)) {
 			nmp = CIRCLEQ_NEXT(mp, mnt_list);
 			continue;
 		}
diff --git a/sys/compat/freebsd/freebsd_file.c b/sys/compat/freebsd/freebsd_file.c
index 89ac1fcc0b2..46f0e69e051 100644
--- a/sys/compat/freebsd/freebsd_file.c
+++ b/sys/compat/freebsd/freebsd_file.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: freebsd_file.c,v 1.23 2006/04/30 14:20:07 sturm Exp $	*/
+/*	$OpenBSD: freebsd_file.c,v 1.24 2006/06/14 20:01:50 sturm Exp $	*/
 /*	$NetBSD: freebsd_file.c,v 1.3 1996/05/03 17:03:09 christos Exp $	*/
 
 /*
@@ -660,7 +660,7 @@ freebsd_sys_getfsstat(p, v, retval)
 
 	for (mp = CIRCLEQ_FIRST(&mountlist); mp != CIRCLEQ_END(&mountlist);
 	    mp = nmp) {
-		if (vfs_busy(mp, LK_NOWAIT)) {
+		if (vfs_busy(mp, VB_READ|VB_UMIGNORE)) {
 			nmp = CIRCLEQ_NEXT(mp, mnt_list);
 			continue;
 		}
diff --git a/sys/compat/netbsd/netbsd_file.c b/sys/compat/netbsd/netbsd_file.c
index f02653634b1..d7e3b0bd20d 100644
--- a/sys/compat/netbsd/netbsd_file.c
+++ b/sys/compat/netbsd/netbsd_file.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: netbsd_file.c,v 1.9 2006/04/30 14:20:07 sturm Exp $	*/
+/*	$OpenBSD: netbsd_file.c,v 1.10 2006/06/14 20:01:50 sturm Exp $	*/
 /*	$NetBSD: freebsd_file.c,v 1.3 1996/05/03 17:03:09 christos Exp $	*/
 
 /*
@@ -634,7 +634,7 @@ netbsd_sys_getfsstat(p, v, retval)
 
 	for (mp = CIRCLEQ_FIRST(&mountlist); mp != CIRCLEQ_END(&mountlist);
 	    mp = nmp) {
-		if (vfs_busy(mp, LK_NOWAIT)) {
+		if (vfs_busy(mp, VB_READ|VB_UMIGNORE)) {
 			nmp = CIRCLEQ_NEXT(mp, mnt_list);
 			continue;
 		}
diff --git a/sys/kern/vfs_lookup.c b/sys/kern/vfs_lookup.c
index 37b2ffe0fe8..9fa13cc7c93 100644
--- a/sys/kern/vfs_lookup.c
+++ b/sys/kern/vfs_lookup.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: vfs_lookup.c,v 1.32 2006/04/30 14:20:07 sturm Exp $	*/
+/*	$OpenBSD: vfs_lookup.c,v 1.33 2006/06/14 20:01:50 sturm Exp $	*/
 /*	$NetBSD: vfs_lookup.c,v 1.17 1996/02/09 19:00:59 christos Exp $	*/
 
 /*
@@ -513,7 +513,7 @@ dirloop:
 	 */
 	while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
 	    (cnp->cn_flags & NOCROSSMOUNT) == 0) {
-		if (vfs_busy(mp, 0))
+		if (vfs_busy(mp, VB_READ|VB_UMWAIT))
 			continue;
 		VOP_UNLOCK(dp, 0, p);
 		error = VFS_ROOT(mp, &tdp);
diff --git a/sys/kern/vfs_subr.c b/sys/kern/vfs_subr.c
index 3170b903426..88458c67924 100644
--- a/sys/kern/vfs_subr.c
+++ b/sys/kern/vfs_subr.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: vfs_subr.c,v 1.127 2006/06/02 20:25:09 pedro Exp $	*/
+/*	$OpenBSD: vfs_subr.c,v 1.128 2006/06/14 20:01:50 sturm Exp $	*/
 /*	$NetBSD: vfs_subr.c,v 1.53 1996/04/22 01:39:13 christos Exp $	*/
 
 /*
@@ -145,36 +145,31 @@ vntblinit(void)
  * Mark a mount point as busy. Used to synchronize access and to delay
  * unmounting.
  *
- * historical behavior:
- *  - LK_NOWAIT means that we should just ignore the mount point if it's
- *     being unmounted.
- *  - no flags means that we should sleep on the mountpoint and then
- *     fail.
+ * Default behaviour is to attempt getting a READ lock and in case of an
+ * ongoing unmount, to wait for it to finish and then return failure.
  */
 int
 vfs_busy(struct mount *mp, int flags)
 {
-	int lkflags;
+	int rwflags = 0;
 
-	switch (flags) {
-	case LK_NOWAIT:
-		lkflags = LK_SHARED|LK_NOWAIT;
-		break;
-	case 0:
-		lkflags = LK_SHARED;
-		break;
-	default:
-		lkflags = flags;
-	}
+	/* new mountpoints need their lock initialised */
+	if (mp->mnt_lock.rwl_name == NULL)
+		rw_init(&mp->mnt_lock, "vfslock");
 
-	/*
-	 * Always sleepfail. We will only sleep for an exclusive lock
-	 * and the exclusive lock will only be acquired when unmounting.
-	 */
-	lkflags |= LK_SLEEPFAIL;
+	if (flags & VB_WRITE)
+		rwflags |= RW_WRITE;
+	else
+		rwflags |= RW_READ;
+
+	if (flags & VB_UMWAIT)
+		rwflags |= RW_SLEEPFAIL;
+	else
+		rwflags |= RW_NOSLEEP;
+
+	if (rw_enter(&mp->mnt_lock, rwflags))
+		return (EBUSY);
 
-	if (lockmgr(&mp->mnt_lock, lkflags, NULL))
-		return (ENOENT);
 	return (0);
 }
 
@@ -184,13 +179,16 @@ vfs_busy(struct mount *mp, int flags)
 void
 vfs_unbusy(struct mount *mp)
 {
-	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL);
+	rw_exit(&mp->mnt_lock);
 }
 
 int
 vfs_isbusy(struct mount *mp) 
 {
-	return (lockstatus(&mp->mnt_lock));
+	if (RWLOCK_OWNER(&mp->mnt_lock) > 0)
+		return (1);
+	else
+		return (0);
 }
 
 /*
@@ -212,8 +210,7 @@ vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp)
 		return (ENODEV);
 	mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK);
 	bzero((char *)mp, (u_long)sizeof(struct mount));
-	lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0);
-	(void)vfs_busy(mp, LK_NOWAIT);
+	(void) vfs_busy(mp, VB_READ|VB_UMIGNORE);
 	LIST_INIT(&mp->mnt_vnodelist);
 	mp->mnt_vfc = vfsp;
 	mp->mnt_op = vfsp->vfc_vfsops;
@@ -1165,7 +1162,7 @@ vgonel(struct vnode *vp, struct proc *p)
 		 */
 		mp = vp->v_specmountpoint;
 		if (mp != NULL) {
-			if (!vfs_busy(mp, LK_EXCLUSIVE)) {
+			if (!vfs_busy(mp, VB_WRITE|VB_UMWAIT)) {
 				flags = MNT_FORCE | MNT_DOOMED;
 				dounmount(mp, flags, p, NULL);
 			}
@@ -1332,7 +1329,7 @@ printlockedvnodes(void)
 
 	for (mp = CIRCLEQ_FIRST(&mountlist); mp != CIRCLEQ_END(&mountlist);
 	    mp = nmp) {
-		if (vfs_busy(mp, LK_NOWAIT)) {
+		if (vfs_busy(mp, VB_READ|VB_UMIGNORE)) {
 			nmp = CIRCLEQ_NEXT(mp, mnt_list);
 			continue;
 		}
@@ -1419,7 +1416,7 @@ sysctl_vnode(char *where, size_t *sizep, struct proc *p)
 
 	for (mp = CIRCLEQ_FIRST(&mountlist); mp != CIRCLEQ_END(&mountlist);
 	    mp = nmp) {
-		if (vfs_busy(mp, LK_NOWAIT)) {
+		if (vfs_busy(mp, VB_READ|VB_UMIGNORE)) {
 			nmp = CIRCLEQ_NEXT(mp, mnt_list);
 			continue;
 		}
@@ -1720,7 +1717,7 @@ vfs_unmountall(void)
 	for (mp = CIRCLEQ_LAST(&mountlist); mp != CIRCLEQ_END(&mountlist);
 	    mp = nmp) {
 		nmp = CIRCLEQ_PREV(mp, mnt_list);
-		if ((vfs_busy(mp, LK_EXCLUSIVE|LK_NOWAIT)) != 0)
+		if ((vfs_busy(mp, VB_WRITE|VB_UMIGNORE)) != 0)
 			continue;
 		if ((error = dounmount(mp, MNT_FORCE, curproc, NULL)) != 0) {
 			printf("unmount of %s failed with error %d\n",
diff --git a/sys/kern/vfs_sync.c b/sys/kern/vfs_sync.c
index 6068ff5473e..8694baa5c39 100644
--- a/sys/kern/vfs_sync.c
+++ b/sys/kern/vfs_sync.c
@@ -1,4 +1,4 @@
-/*       $OpenBSD: vfs_sync.c,v 1.39 2006/04/30 14:20:07 sturm Exp $  */
+/*       $OpenBSD: vfs_sync.c,v 1.40 2006/06/14 20:01:50 sturm Exp $  */
 
 /*
  *  Portions of this code are:
@@ -353,7 +353,7 @@ sync_fsync(void *v)
 	 * Walk the list of vnodes pushing all that are dirty and
 	 * not already on the sync list.
 	 */
-	if (vfs_busy(mp, LK_NOWAIT) == 0) {
+	if (vfs_busy(mp, VB_READ|VB_UMIGNORE) == 0) {
 		asyncflag = mp->mnt_flag & MNT_ASYNC;
 		mp->mnt_flag &= ~MNT_ASYNC;
 		VFS_SYNC(mp, MNT_LAZY, ap->a_cred, ap->a_p);
diff --git a/sys/kern/vfs_syscalls.c b/sys/kern/vfs_syscalls.c
index a1fd7f46d4f..fc4ae8e5ef3 100644
--- a/sys/kern/vfs_syscalls.c
+++ b/sys/kern/vfs_syscalls.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: vfs_syscalls.c,v 1.135 2006/05/27 17:37:42 sturm Exp $	*/
+/*	$OpenBSD: vfs_syscalls.c,v 1.136 2006/06/14 20:01:50 sturm Exp $	*/
 /*	$NetBSD: vfs_syscalls.c,v 1.71 1996/04/23 10:29:02 mycroft Exp $	*/
 
 /*
@@ -150,7 +150,7 @@ sys_mount(struct proc *p, void *v, register_t *retval)
 			if (flag & MNT_NOEXEC)
 				SCARG(uap, flags) |= MNT_NOEXEC;
 		}
-		if ((error = vfs_busy(mp, LK_NOWAIT)) != 0) {
+		if ((error = vfs_busy(mp, VB_READ|VB_UMIGNORE)) != 0) {
 			vput(vp);
 			return (error);
 		}
@@ -236,8 +236,7 @@ sys_mount(struct proc *p, void *v, register_t *retval)
 	mp = (struct mount *)malloc((u_long)sizeof(struct mount),
 		M_MOUNT, M_WAITOK);
 	bzero((char *)mp, (u_long)sizeof(struct mount));
-	lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0);
-	(void) vfs_busy(mp, LK_NOWAIT);
+	(void) vfs_busy(mp, VB_READ|VB_UMIGNORE);
 	mp->mnt_op = vfsp->vfc_vfsops;
 	mp->mnt_vfc = vfsp;
 	mp->mnt_flag |= (vfsp->vfc_flags & MNT_VISFLAGMASK);
@@ -403,7 +402,7 @@ sys_unmount(struct proc *p, void *v, register_t *retval)
 	}
 	vput(vp);
 
-	if (vfs_busy(mp, LK_EXCLUSIVE))
+	if (vfs_busy(mp, VB_WRITE|VB_UMWAIT))
 		return (EBUSY);
 
 	return (dounmount(mp, SCARG(uap, flags), p, vp));
@@ -434,7 +433,7 @@ dounmount(struct mount *mp, int flags, struct proc *p, struct vnode *olddp)
  	if (error && error != EIO && !(flags & MNT_DOOMED)) {
  		if ((mp->mnt_flag & MNT_RDONLY) == 0 && hadsyncer)
  			(void) vfs_allocate_syncvnode(mp);
-		lockmgr(&mp->mnt_lock, LK_RELEASE | LK_INTERLOCK, NULL);
+		vfs_unbusy(mp);
 		return (error);
 	}
 
@@ -449,7 +448,7 @@ dounmount(struct mount *mp, int flags, struct proc *p, struct vnode *olddp)
 	if (!LIST_EMPTY(&mp->mnt_vnodelist))
 		panic("unmount: dangling vnode");
 
-	lockmgr(&mp->mnt_lock, LK_RELEASE | LK_INTERLOCK, NULL);
+	vfs_unbusy(mp);
 	free(mp, M_MOUNT);
 
 	return (0);
@@ -472,7 +471,7 @@ sys_sync(struct proc *p, void *v, register_t *retval)
 
 	for (mp = CIRCLEQ_LAST(&mountlist); mp != CIRCLEQ_END(&mountlist);
 	    mp = nmp) {
-		if (vfs_busy(mp, LK_NOWAIT)) {
+		if (vfs_busy(mp, VB_READ|VB_UMIGNORE)) {
 			nmp = CIRCLEQ_PREV(mp, mnt_list);
 			continue;
 		}
@@ -627,7 +626,7 @@ sys_getfsstat(struct proc *p, void *v, register_t *retval)
 
 	for (mp = CIRCLEQ_FIRST(&mountlist); mp != CIRCLEQ_END(&mountlist);
 	    mp = nmp) {
-		if (vfs_busy(mp, LK_NOWAIT)) {
+		if (vfs_busy(mp, VB_READ|VB_UMIGNORE)) {
 			nmp = CIRCLEQ_NEXT(mp, mnt_list);
 			continue;
 		}
@@ -703,7 +702,7 @@ sys_fchdir(struct proc *p, void *v, register_t *retval)
 		error = VOP_ACCESS(vp, VEXEC, p->p_ucred, p);
 
 	while (!error && (mp = vp->v_mountedhere) != NULL) {
-		if (vfs_busy(mp, 0))
+		if (vfs_busy(mp, VB_READ|VB_UMWAIT))
 			continue;
 		error = VFS_ROOT(mp, &tdp);
 		vfs_unbusy(mp);
diff --git a/sys/nfs/nfs_syscalls.c b/sys/nfs/nfs_syscalls.c
index 9ac032a6323..e6156abe717 100644
--- a/sys/nfs/nfs_syscalls.c
+++ b/sys/nfs/nfs_syscalls.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: nfs_syscalls.c,v 1.46 2006/04/30 14:20:07 sturm Exp $	*/
+/*	$OpenBSD: nfs_syscalls.c,v 1.47 2006/06/14 20:01:50 sturm Exp $	*/
 /*	$NetBSD: nfs_syscalls.c,v 1.19 1996/02/18 11:53:52 fvdl Exp $	*/
 
 /*
@@ -177,7 +177,7 @@ nfs_clientd(struct nfsmount *nmp, struct ucred *cred, struct nfsd_cargs *ncd,
 		    error = tsleep((caddr_t)&nmp->nm_authstr, PSOCK | PCATCH,
 			"nqnfstimr", hz / 3);
 		    if (error == EINTR || error == ERESTART) {
-			if (vfs_busy(nmp->nm_mountp, LK_EXCLUSIVE) == 0)
+			if (vfs_busy(nmp->nm_mountp, VB_WRITE|VB_UMWAIT) == 0)
 			    dounmount(nmp->nm_mountp, MNT_FORCE, p, NULL);
 		    }
 	    }
diff --git a/sys/sys/mount.h b/sys/sys/mount.h
index d6912d908f6..153b4e20dee 100644
--- a/sys/sys/mount.h
+++ b/sys/sys/mount.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: mount.h,v 1.71 2006/05/28 03:55:56 pedro Exp $	*/
+/*	$OpenBSD: mount.h,v 1.72 2006/06/14 20:01:50 sturm Exp $	*/
 /*	$NetBSD: mount.h,v 1.48 1996/02/18 11:55:47 fvdl Exp $	*/
 
 /*
@@ -40,7 +40,7 @@
 #include <sys/ucred.h>
 #endif
 #include <sys/queue.h>
-#include <sys/lock.h>
+#include <sys/rwlock.h>
 
 typedef struct { int32_t val[2]; } fsid_t;	/* file system id type */
 
@@ -372,7 +372,7 @@ struct mount {
 	struct vnode	*mnt_vnodecovered;	/* vnode we mounted on */
 	struct vnode    *mnt_syncer;            /* syncer vnode */
 	struct vnodelst	mnt_vnodelist;		/* list of vnodes this mount */
-	struct lock     mnt_lock;               /* mount structure lock */
+	struct rwlock   mnt_lock;               /* mount structure lock */
 	int		mnt_flag;		/* flags */
 	int		mnt_maxsymlinklen;	/* max size of short symlink */
 	struct statfs	mnt_stat;		/* cache of filesystem stats */
@@ -564,6 +564,11 @@ struct netexport {
  * exported vnode operations
  */
 int	vfs_busy(struct mount *, int);
+#define VB_READ		0x01
+#define VB_WRITE	0x02
+#define VB_UMIGNORE	0x04	/* ignore unmount in progress */
+#define VB_UMWAIT	0x08	/* wait for unmount to finish */
+
 int     vfs_isbusy(struct mount *);
 int     vfs_mount_foreach_vnode(struct mount *, int (*func)(struct vnode *,
 				    void *), void *);
diff --git a/sys/ufs/mfs/mfs_vfsops.c b/sys/ufs/mfs/mfs_vfsops.c
index 1b826228a1e..6c20f2834a6 100644
--- a/sys/ufs/mfs/mfs_vfsops.c
+++ b/sys/ufs/mfs/mfs_vfsops.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: mfs_vfsops.c,v 1.32 2006/04/30 14:20:08 sturm Exp $	*/
+/*	$OpenBSD: mfs_vfsops.c,v 1.33 2006/06/14 20:01:50 sturm Exp $	*/
 /*	$NetBSD: mfs_vfsops.c,v 1.10 1996/02/09 22:31:28 christos Exp $	*/
 
 /*
@@ -263,7 +263,7 @@ mfs_start(struct mount *mp, int flags, struct proc *p)
 		 * EINTR/ERESTART.
 		 */
 		if (sleepreturn != 0) {
-			if (vfs_busy(mp, LK_EXCLUSIVE|LK_NOWAIT) ||
+			if (vfs_busy(mp, VB_WRITE|VB_UMIGNORE) ||
 			    dounmount(mp, 0, p, NULL))
 				CLRSIG(p, CURSIG(p));
 			sleepreturn = 0;
diff --git a/sys/ufs/ufs/inode.h b/sys/ufs/ufs/inode.h
index b7bc81d2152..b6436af8398 100644
--- a/sys/ufs/ufs/inode.h
+++ b/sys/ufs/ufs/inode.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: inode.h,v 1.32 2005/12/28 20:48:18 pedro Exp $	*/
+/*	$OpenBSD: inode.h,v 1.33 2006/06/14 20:01:50 sturm Exp $	*/
 /*	$NetBSD: inode.h,v 1.8 1995/06/15 23:22:50 cgd Exp $	*/
 
 /*
@@ -38,6 +38,7 @@
  */
 
 #include <sys/buf.h>
+#include <sys/lock.h>
 #include <ufs/ufs/dinode.h>
 #include <ufs/ufs/dir.h>
 #include <ufs/ext2fs/ext2fs_dinode.h>
diff --git a/sys/ufs/ufs/ufs_quota.c b/sys/ufs/ufs/ufs_quota.c
index 3b00eddb730..c1736332089 100644
--- a/sys/ufs/ufs/ufs_quota.c
+++ b/sys/ufs/ufs/ufs_quota.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: ufs_quota.c,v 1.23 2006/04/30 14:20:08 sturm Exp $	*/
+/*	$OpenBSD: ufs_quota.c,v 1.24 2006/06/14 20:01:50 sturm Exp $	*/
 /*	$NetBSD: ufs_quota.c,v 1.8 1996/02/09 22:36:09 christos Exp $	*/
 
 /*
@@ -1028,7 +1028,7 @@ ufs_quotactl(struct mount *mp, int cmds, uid_t uid, caddr_t arg,
 	if ((u_int)type >= MAXQUOTAS)
 		return (EINVAL);
 
-	if (vfs_busy(mp, LK_NOWAIT))
+	if (vfs_busy(mp, VB_READ|VB_UMIGNORE))
 		return (0);
  
 
diff --git a/sys/xfs/xfs_dev-common.c b/sys/xfs/xfs_dev-common.c
index 533bcae5a02..55100533536 100644
--- a/sys/xfs/xfs_dev-common.c
+++ b/sys/xfs/xfs_dev-common.c
@@ -180,7 +180,7 @@ xfs_devclose_common(dev_t dev, d_thread_t *proc)
      */
 
     if (xfs[minor(dev)].mp != NULL) {
-	if (xfs_vfs_busy(xfs[minor(dev)].mp, 0, NULL, proc)) {
+	if (xfs_vfs_busy(xfs[minor(dev)].mp, VB_READ|VB_UMWAIT, NULL, proc)) {
 	    NNPFSDEB(XDEBNODE, ("xfs_dev_close: vfs_busy() --> BUSY\n"));
 	    return EBUSY;
 	}