Part 1 of making shared lookups more resilient with respect to forced

unmounts.  When we upgrade a vnode lock from shared to exclusive during
a name cache lookup, fail the lookup with EBADF if the vnode is invalidated
while we are waiting for the exclusive lock.

Also, for correctness (though I'm not sure it can occur in practice),
downgrade an exclusively locked vnode if it should be share locked.

Tested by:	pho

2 files changed, 22 insertions, 9 deletions

diff --git a/sys/kern/vfs_cache.c b/sys/kern/vfs_cache.c
index ba97f38..33e8759 100644
--- a/sys/kern/vfs_cache.c
+++ b/sys/kern/vfs_cache.c
@@ -304,7 +304,9 @@ cache_zap(ncp)
  * succeeds, the vnode is returned in *vpp, and a status of -1 is
  * returned. If the lookup determines that the name does not exist
  * (negative cacheing), a status of ENOENT is returned. If the lookup
- * fails, a status of zero is returned.
+ * fails, a status of zero is returned.  If the directory vnode is
+ * recycled out from under us due to a forced unmount, a status of
+ * EBADF is returned.
  *
  * vpp is locked and ref'd on return.  If we're looking up DOTDOT, dvp is
  * unlocked.  If we're looking up . an extra ref is taken, but the lock is
@@ -428,10 +430,18 @@ success:
 		 * differently...
 		 */
 		ltype = cnp->cn_lkflags & LK_TYPE_MASK;
-		if (ltype == VOP_ISLOCKED(*vpp))
-			return (-1);
-		else if (ltype == LK_EXCLUSIVE)
-			vn_lock(*vpp, LK_UPGRADE | LK_RETRY);
+		if (ltype != VOP_ISLOCKED(*vpp)) {
+			if (ltype == LK_EXCLUSIVE) {
+				vn_lock(*vpp, LK_UPGRADE | LK_RETRY);
+				if ((*vpp)->v_iflag & VI_DOOMED) {
+					/* forced unmount */
+					vrele(*vpp);
+					*vpp = NULL;
+					return (EBADF);
+				}
+			} else
+				vn_lock(*vpp, LK_DOWNGRADE | LK_RETRY);
+		}
 		return (-1);
 	}
 	ltype = 0;	/* silence gcc warning */
@@ -671,9 +681,9 @@ vfs_cache_lookup(ap)
 	error = cache_lookup(dvp, vpp, cnp);
 	if (error == 0)
 		return (VOP_CACHEDLOOKUP(dvp, vpp, cnp));
-	if (error == ENOENT)
-		return (error);
-	return (0);
+	if (error == -1)
+		return (0);
+	return (error);
 }
 
 
diff --git a/sys/nfsclient/nfs_vnops.c b/sys/nfsclient/nfs_vnops.c
index 04b3b34..bbf6810 100644
--- a/sys/nfsclient/nfs_vnops.c
+++ b/sys/nfsclient/nfs_vnops.c
@@ -888,7 +888,10 @@ nfs_lookup(struct vop_lookup_args *ap)
 		*vpp = NULLVP;
 		return (error);
 	}
-	if ((error = cache_lookup(dvp, vpp, cnp)) && error != ENOENT) {
+	error = cache_lookup(dvp, vpp, cnp);
+	if (error > 0 && error != ENOENT)
+		return (error);
+	if (error == -1) {
 		struct vattr vattr;
 
 		newvp = *vpp;