- Fix interlock handling in vn_lock(). Previously, vn_lock() could return

   with interlock held in error conditions when the caller did not specify
   LK_INTERLOCK.
 - Add several comments to vn_lock() describing the rational behind the code
   flow since it was not immediately obvious.

1 files changed, 12 insertions, 10 deletions

diff --git a/sys/kern/vfs_vnops.c b/sys/kern/vfs_vnops.c
index 875e3f0..3f00fda 100644
--- a/sys/kern/vfs_vnops.c
+++ b/sys/kern/vfs_vnops.c
@@ -843,25 +843,27 @@ debug_vn_lock(vp, flags, td, filename, line)
 			vp->v_iflag |= VI_XWANT;
 			msleep(vp, VI_MTX(vp), PINOD | PDROP,
 			    "vn_lock", 0);
-			mp_fixme("interlock not released.");
+			/*
+			 * Since we're just going to return, unlock interlock
+			 * if the caller didn't call us with it held.
+			 */
+			if ((flags & (LK_INTERLOCK|LK_RETRY)) == 0)
+				VI_UNLOCK(vp);
 			error = ENOENT;
 		} else {
-#if 0
-			/* this can now occur in normal operation */
-			if (vp->v_vxproc != NULL)
-				log(LOG_INFO, "VXLOCK interlock avoided in vn_lock\n");
-#endif
 #ifdef	DEBUG_LOCKS
 			vp->filename = filename;
 			vp->line = line;
 #endif
+			/*
+			 * lockmgr drops interlock before it will return for
+			 * any reason.  So force the code above to relock it.
+			 */
 			error = VOP_LOCK(vp,
 				    flags | LK_NOPAUSE | LK_INTERLOCK, td);
-			if (error == 0)
-				return (error);
+			flags &= ~LK_INTERLOCK;
 		}
-		flags &= ~LK_INTERLOCK;
-	} while (flags & LK_RETRY);
+	} while (flags & LK_RETRY && error != 0);
 	return (error);
 }