From 03964c8e09da7a6bf48a37dae84394274f383e65 Mon Sep 17 00:00:00 2001
From: John Baldwin <jhb@FreeBSD.org>
Date: Thu, 19 Feb 2009 22:28:48 +0000
Subject: [PATCH] Enable caching of negative pathname lookups in the NFS
 client.  To avoid stale entries, we save a copy of the directory's
 modification time when the first negative cache entry was added in the
 directory's NFS node. When a negative cache entry is hit during a pathname
 lookup, the parent directory's modification time is checked.  If it has
 changed, all of the negative cache entries for that parent are purged and the
 lookup falls back to using the RPC.  This required adding a new
 cache_purge_negative() method to the name cache to purge only negative cache
 entries for a given directory.

Submitted by:	mohans, Rick Macklem, Ricardo Labiaga @ NetApp
Reviewed by:	mohans
---
 sys/kern/vfs_cache.c      | 18 ++++++++++++
 sys/nfsclient/nfs_vnops.c | 61 ++++++++++++++++++++++++++++++++++-----
 sys/nfsclient/nfsnode.h   |  1 +
 sys/sys/vnode.h           |  1 +
 4 files changed, 73 insertions(+), 8 deletions(-)

diff --git a/sys/kern/vfs_cache.c b/sys/kern/vfs_cache.c
index 96544334164..3c32808b4b7 100644
--- a/sys/kern/vfs_cache.c
+++ b/sys/kern/vfs_cache.c
@@ -662,6 +662,24 @@ cache_purge(vp)
 	CACHE_WUNLOCK();
 }
 
+/*
+ * Invalidate all negative entries for a particular directory vnode.
+ */
+void
+cache_purge_negative(vp)
+	struct vnode *vp;
+{
+	struct namecache *cp, *ncp;
+
+	CTR1(KTR_VFS, "cache_purge_negative(%p)", vp);
+	CACHE_WLOCK();
+	LIST_FOREACH_SAFE(cp, &vp->v_cache_src, nc_src, ncp) {
+		if (cp->nc_vp == NULL)
+			cache_zap(cp);
+	}
+	CACHE_WUNLOCK();
+}
+
 /*
  * Flush all entries referencing a particular filesystem.
  */
diff --git a/sys/nfsclient/nfs_vnops.c b/sys/nfsclient/nfs_vnops.c
index 90400ccedab..dbdf5cbe929 100644
--- a/sys/nfsclient/nfs_vnops.c
+++ b/sys/nfsclient/nfs_vnops.c
@@ -852,6 +852,7 @@ nfs_lookup(struct vop_lookup_args *ap)
 	struct componentname *cnp = ap->a_cnp;
 	struct vnode *dvp = ap->a_dvp;
 	struct vnode **vpp = ap->a_vpp;
+	struct vattr vattr;
 	int flags = cnp->cn_flags;
 	struct vnode *newvp;
 	struct nfsmount *nmp;
@@ -880,8 +881,12 @@ nfs_lookup(struct vop_lookup_args *ap)
 	if (error > 0 && error != ENOENT)
 		return (error);
 	if (error == -1) {
-		struct vattr vattr;
-
+		/*
+		 * We only accept a positive hit in the cache if the
+		 * change time of the file matches our cached copy.
+		 * Otherwise, we discard the cache entry and fallback
+		 * to doing a lookup RPC.
+		 */
 		newvp = *vpp;
 		if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred)
 		    && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) {
@@ -897,6 +902,24 @@ nfs_lookup(struct vop_lookup_args *ap)
 		else 
 			vrele(newvp);
 		*vpp = NULLVP;
+	} else if (error == ENOENT) {
+		/*
+		 * We only accept a negative hit in the cache if the
+		 * modification time of the parent directory matches
+		 * our cached copy.  Otherwise, we discard all of the
+		 * negative cache entries for this directory.
+		 */
+		if (VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
+		    vattr.va_mtime.tv_sec == np->n_dmtime) {
+			nfsstats.lookupcache_hits++;
+			if ((cnp->cn_nameiop == CREATE ||
+			    cnp->cn_nameiop == RENAME) &&
+			    (flags & ISLASTCN))
+				cnp->cn_flags |= SAVENAME;
+			return (ENOENT);
+		}
+		cache_purge_negative(dvp);
+		np->n_dmtime = 0;
 	}
 	error = 0;
 	newvp = NULLVP;
@@ -982,16 +1005,38 @@ nfsmout:
 			vput(newvp);
 			*vpp = NULLVP;
 		}
+
+		if (error != ENOENT)
+			goto done;
+
+		/* The requested file was not found. */
 		if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
-		    (flags & ISLASTCN) && error == ENOENT) {
+		    (flags & ISLASTCN)) {
+			/*
+			 * XXX: UFS does a full VOP_ACCESS(dvp,
+			 * VWRITE) here instead of just checking
+			 * MNT_RDONLY.
+			 */
 			if (dvp->v_mount->mnt_flag & MNT_RDONLY)
-				error = EROFS;
-			else
-				error = EJUSTRETURN;
-		}
-		if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
+				return (EROFS);
 			cnp->cn_flags |= SAVENAME;
+			return (EJUSTRETURN);
+		}
+
+		if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE) {
+			/*
+			 * Maintain n_dmtime as the modification time
+			 * of the parent directory when the oldest -ve
+			 * name cache entry for this directory was
+			 * added.
+			 */
+			if (np->n_dmtime == 0)
+				np->n_dmtime = np->n_vattr.va_mtime.tv_sec;
+			cache_enter(dvp, NULL, cnp);
+		}
+		return (ENOENT);
 	}
+done:
 	return (error);
 }
 
diff --git a/sys/nfsclient/nfsnode.h b/sys/nfsclient/nfsnode.h
index 03e5e7f1843..7781ae645c5 100644
--- a/sys/nfsclient/nfsnode.h
+++ b/sys/nfsclient/nfsnode.h
@@ -109,6 +109,7 @@ struct nfsnode {
 	time_t			n_modestamp;	/* mode cache timestamp */
 	struct timespec		n_mtime;	/* Prev modify time. */
 	time_t			n_ctime;	/* Prev create time. */
+	time_t			n_dmtime;	/* Prev dir modify time. */
 	time_t			n_expiry;	/* Lease expiry time */
 	nfsfh_t			*n_fhp;		/* NFS File Handle */
 	struct vnode		*n_vnode;	/* associated vnode */
diff --git a/sys/sys/vnode.h b/sys/sys/vnode.h
index 8e96a1d8886..2d6c0e4984f 100644
--- a/sys/sys/vnode.h
+++ b/sys/sys/vnode.h
@@ -562,6 +562,7 @@ void	cache_enter(struct vnode *dvp, struct vnode *vp,
 int	cache_lookup(struct vnode *dvp, struct vnode **vpp,
 	    struct componentname *cnp);
 void	cache_purge(struct vnode *vp);
+void	cache_purge_negative(struct vnode *vp);
 void	cache_purgevfs(struct mount *mp);
 int	change_dir(struct vnode *vp, struct thread *td);
 int	change_root(struct vnode *vp, struct thread *td);