svn commit: r324704 - head/sys/kern

[Mark Johnston]

Author: markj
Date: Tue Oct 17 19:41:45 2017
New Revision: 324704
URL: https://svnweb.freebsd.org/changeset/base/324704

Log:
  Fix a racy VI_DOOMED check in MNT_VNODE_FOREACH_ALL().
  
  MNT_VNODE_FOREACH_ALL() is supposed to avoid returning doomed vnodes,
  but the VI_DOOMED check it used was done without the vnode interlock
  held, so it could race with a concurrent vgone().
  
  Submitted by:	Don Morris <don.morris at isilon.com>
  Reviewed by:	kib, mckusick
  MFC after:	1 week
  Sponsored by:	Dell EMC Isilon
  Differential Revision:	https://reviews.freebsd.org/D12704

Modified:
  head/sys/kern/vfs_subr.c

Modified: head/sys/kern/vfs_subr.c
==============================================================================
--- head/sys/kern/vfs_subr.c	Tue Oct 17 19:11:29 2017	(r324703)
+++ head/sys/kern/vfs_subr.c	Tue Oct 17 19:41:45 2017	(r324704)
@@ -5299,12 +5299,18 @@ __mnt_vnode_next_all(struct vnode **mvp, struct mount 
 		kern_yield(PRI_USER);
 	MNT_ILOCK(mp);
 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
-	vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
-	while (vp != NULL && (vp->v_type == VMARKER ||
-	    (vp->v_iflag & VI_DOOMED) != 0))
-		vp = TAILQ_NEXT(vp, v_nmntvnodes);
-
-	/* Check if we are done */
+	for (vp = TAILQ_NEXT(*mvp, v_nmntvnodes); vp != NULL;
+	    vp = TAILQ_NEXT(vp, v_nmntvnodes)) {
+		/* Allow a racy peek at VI_DOOMED to save a lock acquisition. */
+		if (vp->v_type == VMARKER || (vp->v_iflag & VI_DOOMED) != 0)
+			continue;
+		VI_LOCK(vp);
+		if ((vp->v_iflag & VI_DOOMED) != 0) {
+			VI_UNLOCK(vp);
+			continue;
+		}
+		break;
+	}
 	if (vp == NULL) {
 		__mnt_vnode_markerfree_all(mvp, mp);
 		/* MNT_IUNLOCK(mp); -- done in above function */
@@ -5313,7 +5319,6 @@ __mnt_vnode_next_all(struct vnode **mvp, struct mount 
 	}
 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
-	VI_LOCK(vp);
 	MNT_IUNLOCK(mp);
 	return (vp);
 }
@@ -5326,14 +5331,20 @@ __mnt_vnode_first_all(struct vnode **mvp, struct mount
 	*mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
 	MNT_ILOCK(mp);
 	MNT_REF(mp);
+	(*mvp)->v_mount = mp;
 	(*mvp)->v_type = VMARKER;
 
-	vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
-	while (vp != NULL && (vp->v_type == VMARKER ||
-	    (vp->v_iflag & VI_DOOMED) != 0))
-		vp = TAILQ_NEXT(vp, v_nmntvnodes);
-
-	/* Check if we are done */
+	TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
+		/* Allow a racy peek at VI_DOOMED to save a lock acquisition. */
+		if (vp->v_type == VMARKER || (vp->v_iflag & VI_DOOMED) != 0)
+			continue;
+		VI_LOCK(vp);
+		if ((vp->v_iflag & VI_DOOMED) != 0) {
+			VI_UNLOCK(vp);
+			continue;
+		}
+		break;
+	}
 	if (vp == NULL) {
 		MNT_REL(mp);
 		MNT_IUNLOCK(mp);
@@ -5341,13 +5352,10 @@ __mnt_vnode_first_all(struct vnode **mvp, struct mount
 		*mvp = NULL;
 		return (NULL);
 	}
-	(*mvp)->v_mount = mp;
 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
-	VI_LOCK(vp);
 	MNT_IUNLOCK(mp);
 	return (vp);
 }
-
 
 void
 __mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp)