Add marker vnodes to ensure that all vnodes associated with the mount point are

iterated over when using MNT_VNODE_FOREACH.

Reviewed by:	truckman

3 files changed, 132 insertions, 37 deletions

diff --git a/sys/kern/vfs_default.c b/sys/kern/vfs_default.c
index d8adf7c..be8c931 100644
--- a/sys/kern/vfs_default.c
+++ b/sys/kern/vfs_default.c
@@ -545,7 +545,7 @@ vfs_stdsync(mp, waitfor, td)
 	int waitfor;
 	struct thread *td;
 {
-	struct vnode *vp, *nvp;
+	struct vnode *vp, *mvp;
 	int error, lockreq, allerror = 0;
 
 	lockreq = LK_EXCLUSIVE | LK_INTERLOCK;
@@ -556,7 +556,7 @@ vfs_stdsync(mp, waitfor, td)
 	 */
 	MNT_ILOCK(mp);
 loop:
-	MNT_VNODE_FOREACH(vp, mp, nvp) {
+	MNT_VNODE_FOREACH(vp, mp, mvp) {
 
 		VI_LOCK(vp);
 		if (vp->v_bufobj.bo_dirty.bv_cnt == 0) {
@@ -567,8 +567,10 @@ loop:
 
 		if ((error = vget(vp, lockreq, td)) != 0) {
 			MNT_ILOCK(mp);
-			if (error == ENOENT)
+			if (error == ENOENT) {
+				MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
 				goto loop;
+			}
 			continue;
 		}
 		error = VOP_FSYNC(vp, waitfor, td);
diff --git a/sys/kern/vfs_mount.c b/sys/kern/vfs_mount.c
index ee8c1dc..b7b31a8 100644
--- a/sys/kern/vfs_mount.c
+++ b/sys/kern/vfs_mount.c
@@ -90,6 +90,7 @@ SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
     "Unprivileged users may mount and unmount file systems");
 
 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
+MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
 
 /* List of mounted filesystems. */
 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
@@ -445,6 +446,28 @@ static void
 vfs_mount_destroy(struct mount *mp, struct thread *td)
 {
 
+	MNT_ILOCK(mp);
+	if (mp->mnt_holdcnt != 0) {
+		printf("Waiting for mount point to be unheld\n");
+		while (mp->mnt_holdcnt != 0) {
+			mp->mnt_holdcntwaiters++;
+			msleep(&mp->mnt_holdcnt, MNT_MTX(mp),
+			       PZERO, "mntdestroy", 0);
+			mp->mnt_holdcntwaiters--;
+		}
+		printf("mount point unheld\n");
+	}
+	if (mp->mnt_writeopcount > 0) {
+		printf("Waiting for mount point write ops\n");
+		while (mp->mnt_writeopcount > 0) {
+			mp->mnt_kern_flag |= MNTK_SUSPEND;
+			msleep(&mp->mnt_writeopcount,
+			       MNT_MTX(mp),
+			       PZERO, "mntdestroy2", 0);
+		}
+		printf("mount point write ops completed\n");
+	}
+	MNT_IUNLOCK(mp);
 	mp->mnt_vfc->vfc_refcount--;
 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist))
 		panic("unmount: dangling vnode");
@@ -453,6 +476,12 @@ vfs_mount_destroy(struct mount *mp, struct thread *td)
 	MNT_ILOCK(mp);
 	if (mp->mnt_kern_flag & MNTK_MWAIT)
 		wakeup(mp);
+	if (mp->mnt_writeopcount != 0)
+		panic("vfs_mount_destroy: nonzero writeopcount");
+	if (mp->mnt_nvnodelistsize != 0)
+		panic("vfs_mount_destroy: nonzero nvnodelistsize");
+	mp->mnt_writeopcount = -1000;
+	mp->mnt_nvnodelistsize = -1000;
 	MNT_IUNLOCK(mp);
 	mtx_destroy(&mp->mnt_mtx);
 #ifdef MAC
@@ -1658,27 +1687,96 @@ vfs_copyopt(opts, name, dest, len)
  */
 
 struct vnode *
-__mnt_vnode_next(struct vnode **nvp, struct mount *mp)
+__mnt_vnode_next(struct vnode **mvp, struct mount *mp)
 {
 	struct vnode *vp;
 
-	mtx_assert(&mp->mnt_mtx, MA_OWNED);
+	mtx_assert(MNT_MTX(mp), MA_OWNED);
 
-	vp = *nvp;
+	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
+	vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
+	while (vp != NULL && vp->v_type == VMARKER) 
+		vp = TAILQ_NEXT(vp, v_nmntvnodes);
+	
 	/* Check if we are done */
-	if (vp == NULL)
+	if (vp == NULL) {
+		__mnt_vnode_markerfree(mvp, mp);
 		return (NULL);
-	/* If our next vnode is no longer ours, start over */
-	if (vp->v_mount != mp) 
-		vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
-	/* Save pointer to next vnode in list */
-	if (vp != NULL)
-		*nvp = TAILQ_NEXT(vp, v_nmntvnodes);
-	else
-		*nvp = NULL;
+	}
+	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
+	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
 	return (vp);
 }
 
+struct vnode *
+__mnt_vnode_first(struct vnode **mvp, struct mount *mp)
+{
+	struct vnode *vp;
+
+	mtx_assert(MNT_MTX(mp), MA_OWNED);
+
+	vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
+	while (vp != NULL && vp->v_type == VMARKER) 
+		vp = TAILQ_NEXT(vp, v_nmntvnodes);
+	
+	/* Check if we are done */
+	if (vp == NULL) {
+		*mvp = NULL;
+		return (NULL);
+	}
+	mp->mnt_holdcnt++;
+	MNT_IUNLOCK(mp);
+	*mvp = (struct vnode *) malloc(sizeof(struct vnode),
+				       M_VNODE_MARKER,
+				       M_WAITOK | M_ZERO);
+	MNT_ILOCK(mp);
+	(*mvp)->v_type = VMARKER;
+
+	vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
+	while (vp != NULL && vp->v_type == VMARKER) 
+		vp = TAILQ_NEXT(vp, v_nmntvnodes);
+	
+	/* Check if we are done */
+	if (vp == NULL) {
+		MNT_IUNLOCK(mp);
+		free(*mvp, M_VNODE_MARKER);
+		MNT_ILOCK(mp);
+		*mvp = NULL;
+		mp->mnt_holdcnt--;
+		if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0)
+			wakeup(&mp->mnt_holdcnt);
+		return (NULL);
+	}
+	mp->mnt_markercnt++;
+	(*mvp)->v_mount = mp;
+	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
+	return (vp);
+}
+
+
+void
+__mnt_vnode_markerfree(struct vnode **mvp, struct mount *mp)
+{
+
+	if (*mvp == NULL)
+		return;
+	
+	mtx_assert(MNT_MTX(mp), MA_OWNED);
+
+	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
+	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
+	MNT_IUNLOCK(mp);
+	free(*mvp, M_VNODE_MARKER);
+	MNT_ILOCK(mp);
+	*mvp = NULL;
+
+	mp->mnt_markercnt--;
+	mp->mnt_holdcnt--;
+	if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0)
+		wakeup(&mp->mnt_holdcnt);
+}
+
+
 int
 __vfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td)
 {
diff --git a/sys/kern/vfs_subr.c b/sys/kern/vfs_subr.c
index a57b2ba..629e22d 100644
--- a/sys/kern/vfs_subr.c
+++ b/sys/kern/vfs_subr.c
@@ -134,9 +134,9 @@ enum vtype iftovt_tab[16] = {
 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
 };
-int vttoif_tab[9] = {
+int vttoif_tab[10] = {
 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
-	S_IFSOCK, S_IFIFO, S_IFMT,
+	S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
 };
 
 /*
@@ -566,7 +566,12 @@ vlrureclaim(struct mount *mp)
 	vn_start_write(NULL, &mp, V_WAIT);
 	MNT_ILOCK(mp);
 	count = mp->mnt_nvnodelistsize / 10 + 1;
-	while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) {
+	while (count != 0) {
+		vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
+		while (vp != NULL && vp->v_type == VMARKER)
+			vp = TAILQ_NEXT(vp, v_nmntvnodes);
+		if (vp == NULL)
+			break;
 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
 		--count;
@@ -968,6 +973,8 @@ insmntque(struct vnode *vp, struct mount *mp)
 	VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
 	MNT_ILOCK(vp->v_mount);
 	TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
+	VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
+		("neg mount point vnode list size"));
 	mp->mnt_nvnodelistsize++;
 	MNT_IUNLOCK(vp->v_mount);
 }
@@ -2241,7 +2248,7 @@ vflush(mp, rootrefs, flags, td)
 	int flags;
 	struct thread *td;
 {
-	struct vnode *vp, *nvp, *rootvp = NULL;
+	struct vnode *vp, *mvp, *rootvp = NULL;
 	struct vattr vattr;
 	int busy = 0, error;
 
@@ -2260,7 +2267,7 @@ vflush(mp, rootrefs, flags, td)
 	}
 	MNT_ILOCK(mp);
 loop:
-	MNT_VNODE_FOREACH(vp, mp, nvp) {
+	MNT_VNODE_FOREACH(vp, mp, mvp) {
 
 		VI_LOCK(vp);
 		vholdl(vp);
@@ -2269,6 +2276,7 @@ loop:
 		if (error) {
 			vdrop(vp);
 			MNT_ILOCK(mp);
+			MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
 			goto loop;
 		}
 		/*
@@ -2489,7 +2497,8 @@ count_dev(dev)
  * Print out a description of a vnode.
  */
 static char *typename[] =
-{"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
+{"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
+ "VMARKER"};
 
 void
 vn_printf(struct vnode *vp, const char *fmt, ...)
@@ -2818,20 +2827,11 @@ vfs_unmountall()
 void
 vfs_msync(struct mount *mp, int flags)
 {
-	struct vnode *vp, *nvp;
+	struct vnode *vp, *mvp;
 	struct vm_object *obj;
-	int tries;
 
-	tries = 5;
 	MNT_ILOCK(mp);
-loop:
-	TAILQ_FOREACH_SAFE(vp, &mp->mnt_nvnodelist, v_nmntvnodes, nvp) {
-		if (vp->v_mount != mp) {
-			if (--tries > 0)
-				goto loop;
-			break;
-		}
-
+	MNT_VNODE_FOREACH(vp, mp, mvp) {
 		VI_LOCK(vp);
 		if ((vp->v_iflag & VI_OBJDIRTY) &&
 		    (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
@@ -2856,11 +2856,6 @@ loop:
 				vput(vp);
 			}
 			MNT_ILOCK(mp);
-			if (TAILQ_NEXT(vp, v_nmntvnodes) != nvp) {
-				if (--tries > 0)
-					goto loop;
-				break;
-			}
 		} else
 			VI_UNLOCK(vp);
 	}