Serialize NFS vinvalbuf operations by acquiring/upgrading to the

vnode EXCLUSIVE lock. This prevents threads from adding pages to
the vnode while an invalidation is in progress, closing potential
races. In the bioread() path, callers acquire the SHARED vnode lock
- so while an invalidate was in progress, it was possible to fault
in new pages onto the vnode causing the invalidation to take a while
or fail. We saw these races at Yahoo! with very large files+heavy
concurrent access. Forcing an upgrade to EXCLUSIVE lock before doing
the invalidation closes all these races.

Submitted by:	Mohan Srinivasan mohans at yahoo-inc dot com

sys/nfsclient/nfs_bio.c

@@ -1066,6 +1066,7 @@ nfs_vinvalbuf(struct vnode *vp, int flags, struct ucred *cred,
 	struct nfsnode *np = VTONFS(vp);
 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 	int error = 0, slpflag, slptimeo;
+ 	int old_lock = 0;
 
 	ASSERT_VOP_LOCKED(vp, "nfs_vinvalbuf");
 
@@ -1086,40 +1087,36 @@ nfs_vinvalbuf(struct vnode *vp, int flags, struct ucred *cred,
 		slpflag = 0;
 		slptimeo = 0;
 	}
-	/*
-	 * First wait for any other process doing a flush to complete.
-	 */
-	while (np->n_flag & NFLUSHINPROG) {
-		np->n_flag |= NFLUSHWANT;
-		error = tsleep(&np->n_flag, PRIBIO + 2, "nfsvinval",
-			slptimeo);
-		if (error && intrflg &&
-		    nfs_sigintr(nmp, NULL, td))
-			return (EINTR);
-	}
+
+ 	if ((old_lock = VOP_ISLOCKED(vp, td)) != LK_EXCLUSIVE) {
+ 		if (old_lock == LK_SHARED) {
+ 			/* Upgrade to exclusive lock, this might block */
+ 			vn_lock(vp, LK_UPGRADE | LK_RETRY, td);
+ 		} else {
+ 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
+ 		}
+  	}
 
 	/*
 	 * Now, flush as required.
 	 */
-	np->n_flag |= NFLUSHINPROG;
 	error = vinvalbuf(vp, flags, cred, td, slpflag, 0);
 	while (error) {
-		if (intrflg && (error = nfs_sigintr(nmp, NULL, td))) {
-			np->n_flag &= ~NFLUSHINPROG;
-			if (np->n_flag & NFLUSHWANT) {
-				np->n_flag &= ~NFLUSHWANT;
-				wakeup(&np->n_flag);
-			}
-			return (error);
-		}
+		if (intrflg && (error = nfs_sigintr(nmp, NULL, td)))
+			goto out;
 		error = vinvalbuf(vp, flags, cred, td, 0, slptimeo);
 	}
-	np->n_flag &= ~(NMODIFIED | NFLUSHINPROG);
-	if (np->n_flag & NFLUSHWANT) {
-		np->n_flag &= ~NFLUSHWANT;
-		wakeup(&np->n_flag);
-	}
-	return (0);
+	np->n_flag &= ~NMODIFIED;
+out:
+ 	if (old_lock != LK_EXCLUSIVE) {
+ 		if (old_lock == LK_SHARED) {
+ 			/* Downgrade from exclusive lock, this might block */
+ 			vn_lock(vp, LK_DOWNGRADE, td);
+ 		} else {
+ 			VOP_UNLOCK(vp, 0, td);
+ 		}
+  	}
+	return error;
 }
 
 /*

sys/nfsclient/nfs_node.c

@@ -315,7 +315,7 @@ nfs_inactive(struct vop_inactive_args *ap)
 		vrele(sp->s_dvp);
 		FREE((caddr_t)sp, M_NFSREQ);
 	}
-	np->n_flag &= (NMODIFIED | NFLUSHINPROG | NFLUSHWANT);
+	np->n_flag &= NMODIFIED;
 	VOP_UNLOCK(ap->a_vp, 0, ap->a_td);
 	return (0);
 }

sys/nfsclient/nfs_vnops.c

@@ -504,9 +504,7 @@ nfs_close(struct vop_close_args *ap)
 		    error = nfs_flush(vp, ap->a_cred, MNT_WAIT, ap->a_td, cm);
 		    /* np->n_flag &= ~NMODIFIED; */
 		} else {
-		    vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, ap->a_td);
 		    error = nfs_vinvalbuf(vp, V_SAVE, ap->a_cred, ap->a_td, 1);
-		    VOP_UNLOCK(vp, 0, ap->a_td);
 		}
 		np->n_attrstamp = 0;
 	    }

sys/nfsclient/nfsnode.h

@@ -139,8 +139,6 @@ struct nfsnode {
 /*
  * Flags for n_flag
  */
-#define	NFLUSHWANT	0x0001	/* Want wakeup from a flush in prog. */
-#define	NFLUSHINPROG	0x0002	/* Avoid multiple calls to vinvalbuf() */
 #define	NMODIFIED	0x0004	/* Might have a modified buffer in bio */
 #define	NWRITEERR	0x0008	/* Flag write errors so close will know */
 /* 0x20, 0x40, 0x80 free */