From: Peter Staubach <staubach@redhat.com>
Date: Fri, 1 Feb 2008 16:14:09 -0500
Subject: [nfs] reduce number of wire RPC ops, increase perf
Message-id: 47A38BA1.8070704@redhat.com
O-Subject: [RHEL-5.2 PATCH] Increase in RHEL3 vs. RHEL5 NFS operations (Part II)
Bugzilla: 321111

Hi.

Attached is a patch to address bz321111, "Increase in RHEL3 vs.
RHEL5 NFS operations (Part II)".  This patch was constructed
by backporting 37 patches from upstream.  These patches result
in greatly reducing the number of over the wire RPC operations
that the NFS client must send in order to function.

During the development and debugging of this patch, two bugs
in the vanilla RHEL-5 NFS were discovered and addressed.  The
changes for those bugs are included in these changes.  One
was also a bug in upstream NFS and has been fixed there by
Trond, the NFS client upstream maintainer.

This patch was tested using the Connectathon testsuite against
several NFS servers, including RHEL-4, RHEL-5, and Solaris 10.
It was also run on a desktop system by SteveD.  (Thanx, Steve!)

    Thanx...

       ps

diff --git a/fs/nfs/delegation.c b/fs/nfs/delegation.c
index 8a7eb8e..0a443fa 100644
--- a/fs/nfs/delegation.c
+++ b/fs/nfs/delegation.c
@@ -155,15 +155,50 @@ int nfs_inode_set_delegation(struct inode *inode, struct rpc_cred *cred, struct
 	return status;
 }
 
-static int nfs_do_return_delegation(struct inode *inode, struct nfs_delegation *delegation)
+static int nfs_do_return_delegation(struct inode *inode, struct nfs_delegation *delegation, int issync)
 {
 	int res = 0;
 
-	res = nfs4_proc_delegreturn(inode, delegation->cred, &delegation->stateid);
+	res = nfs4_proc_delegreturn(inode, delegation->cred, &delegation->stateid, issync);
 	nfs_free_delegation(delegation);
 	return res;
 }
 
+static struct nfs_delegation *nfs_detach_delegation(struct inode *inode)
+{
+	struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
+	struct nfs_inode *nfsi = NFS_I(inode);
+	struct nfs_delegation *delegation;
+
+	spin_lock(&clp->cl_lock);
+	delegation = nfsi->delegation;
+	if (delegation != NULL) {
+		list_del_init(&delegation->super_list);
+		nfsi->delegation = NULL;
+		nfsi->delegation_state = 0;
+	}
+	spin_unlock(&clp->cl_lock);
+	return delegation;
+}
+
+/*
+ * This function returns the delegation without reclaiming opens
+ * or protecting against delegation reclaims.
+ * It is therefore really only safe to be called from
+ * nfs4_clear_inode()
+ */
+void nfs_inode_return_delegation_noreclaim(struct inode *inode)
+{
+	struct nfs_delegation *delegation;
+
+	if (NFS_I(inode)->delegation == NULL)
+		return;
+	delegation = nfs_detach_delegation(inode);
+	if (delegation == NULL)
+		return;
+	nfs_do_return_delegation(inode, delegation, 0);
+}
+
 /* Sync all data to disk upon delegation return */
 static void nfs_msync_inode(struct inode *inode)
 {
@@ -186,21 +221,14 @@ int __nfs_inode_return_delegation(struct inode *inode)
 	down_read(&clp->cl_sem);
 	/* Guard against new delegated open calls */
 	down_write(&nfsi->rwsem);
-	spin_lock(&clp->cl_lock);
-	delegation = nfsi->delegation;
-	if (delegation != NULL) {
-		list_del_init(&delegation->super_list);
-		nfsi->delegation = NULL;
-		nfsi->delegation_state = 0;
-	}
-	spin_unlock(&clp->cl_lock);
+	delegation = nfs_detach_delegation(inode);
 	nfs_delegation_claim_opens(inode);
 	up_write(&nfsi->rwsem);
 	up_read(&clp->cl_sem);
 	nfs_msync_inode(inode);
 
 	if (delegation != NULL)
-		res = nfs_do_return_delegation(inode, delegation);
+		res = nfs_do_return_delegation(inode, delegation, 1);
 	return res;
 }
 
@@ -341,7 +369,7 @@ static int recall_thread(void *data)
 	nfs_msync_inode(inode);
 
 	if (delegation != NULL)
-		nfs_do_return_delegation(inode, delegation);
+		nfs_do_return_delegation(inode, delegation, 1);
 	iput(inode);
 	module_put_and_exit(0);
 }
diff --git a/fs/nfs/delegation.h b/fs/nfs/delegation.h
index f6e42fb..cef3cdd 100644
--- a/fs/nfs/delegation.h
+++ b/fs/nfs/delegation.h
@@ -28,6 +28,7 @@ int nfs_inode_set_delegation(struct inode *inode, struct rpc_cred *cred, struct
 void nfs_inode_reclaim_delegation(struct inode *inode, struct rpc_cred *cred, struct nfs_openres *res);
 int __nfs_inode_return_delegation(struct inode *inode);
 int nfs_async_inode_return_delegation(struct inode *inode, const nfs4_stateid *stateid);
+void nfs_inode_return_delegation_noreclaim(struct inode *inode);
 
 struct inode *nfs_delegation_find_inode(struct nfs_client *clp, const struct nfs_fh *fhandle);
 void nfs_return_all_delegations(struct super_block *sb);
@@ -38,7 +39,7 @@ void nfs_delegation_mark_reclaim(struct nfs_client *clp);
 void nfs_delegation_reap_unclaimed(struct nfs_client *clp);
 
 /* NFSv4 delegation-related procedures */
-int nfs4_proc_delegreturn(struct inode *inode, struct rpc_cred *cred, const nfs4_stateid *stateid);
+int nfs4_proc_delegreturn(struct inode *inode, struct rpc_cred *cred, const nfs4_stateid *stateid, int issync);
 int nfs4_open_delegation_recall(struct nfs_open_context *ctx, struct nfs4_state *state);
 int nfs4_lock_delegation_recall(struct nfs4_state *state, struct file_lock *fl);
 int nfs4_copy_delegation_stateid(nfs4_stateid *dst, struct inode *inode);
diff --git a/fs/nfs/dir.c b/fs/nfs/dir.c
index 7f254fc..b0b3d8a 100644
--- a/fs/nfs/dir.c
+++ b/fs/nfs/dir.c
@@ -195,9 +195,6 @@ int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page)
 		goto error;
 	}
 	SetPageUptodate(page);
-	spin_lock(&inode->i_lock);
-	NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
-	spin_unlock(&inode->i_lock);
 	/* Ensure consistent page alignment of the data.
 	 * Note: assumes we have exclusive access to this mapping either
 	 *	 through inode->i_mutex or some other mechanism.
@@ -207,9 +204,7 @@ int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page)
 	unlock_page(page);
 	return 0;
  error:
-	SetPageError(page);
 	unlock_page(page);
-	nfs_zap_caches(inode);
 	desc->error = error;
 	return -EIO;
 }
@@ -479,9 +474,6 @@ int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
 						page,
 						NFS_SERVER(inode)->dtsize,
 						desc->plus);
-	spin_lock(&inode->i_lock);
-	NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
-	spin_unlock(&inode->i_lock);
 	desc->page = page;
 	desc->ptr = kmap(page);		/* matching kunmap in nfs_do_filldir */
 	if (desc->error >= 0) {
@@ -641,10 +633,14 @@ static inline int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
 {
 	if (IS_ROOT(dentry))
 		return 1;
-	if ((NFS_I(dir)->cache_validity & NFS_INO_INVALID_ATTR) != 0
-			|| nfs_attribute_timeout(dir))
+	if (!nfs_verify_change_attribute(dir, dentry->d_time))
 		return 0;
-	return nfs_verify_change_attribute(dir, (unsigned long)dentry->d_fsdata);
+	/* Revalidate nfsi->cache_change_attribute before we declare a match */
+	if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
+		return 0;
+	if (!nfs_verify_change_attribute(dir, dentry->d_time))
+		return 0;
+	return 1;
 }
 
 /*
@@ -662,6 +658,20 @@ static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigne
 }
 
 /*
+ * Use intent information to check whether or not we're going to do
+ * an O_EXCL create using this path component.
+ */
+static inline
+int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
+{
+	if (NFS_PROTO(dir)->version == 2)
+		return 0;
+	if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0)
+		return 0;
+	return (nd->intent.open.flags & O_EXCL) != 0;
+}
+
+/*
  * Inode and filehandle revalidation for lookups.
  *
  * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
@@ -726,7 +736,6 @@ static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
 	int error;
 	struct nfs_fh fhandle;
 	struct nfs_fattr fattr;
-	unsigned long verifier;
 
 	parent = dget_parent(dentry);
 	lock_kernel();
@@ -747,12 +756,8 @@ static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
 		goto out_bad;
 	}
 
-	/* Revalidate parent directory attribute cache */
-	if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
-		goto out_zap_parent;
-
 	/* Force a full look up iff the parent directory has changed */
-	if (nfs_check_verifier(dir, dentry)) {
+	if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) {
 		if (nfs_lookup_verify_inode(inode, nd))
 			goto out_zap_parent;
 		goto out_valid;
@@ -761,7 +766,6 @@ static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
 	if (NFS_STALE(inode))
 		goto out_bad;
 
-	verifier = nfs_save_change_attribute(dir);
 	error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
 	if (error)
 		goto out_bad;
@@ -770,7 +774,7 @@ static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
 	if ((error = nfs_refresh_inode(inode, &fattr)) != 0)
 		goto out_bad;
 
-	nfs_set_verifier(dentry, verifier);
+	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
  out_valid:
 	unlock_kernel();
 	dput(parent);
@@ -781,7 +785,7 @@ static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
 out_zap_parent:
 	nfs_zap_caches(dir);
  out_bad:
-	NFS_CACHEINV(dir);
+	nfs_mark_for_revalidate(dir);
 	if (inode && S_ISDIR(inode->i_mode)) {
 		/* Purge readdir caches. */
 		nfs_zap_caches(inode);
@@ -827,7 +831,6 @@ static int nfs_dentry_delete(struct dentry *dentry)
  */
 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
 {
-	nfs_inode_return_delegation(inode);
 	if (S_ISDIR(inode->i_mode))
 		/* drop any readdir cache as it could easily be old */
 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
@@ -838,7 +841,6 @@ static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
 		nfs_complete_unlink(dentry, inode);
 		unlock_kernel();
 	}
-	/* When creating a negative dentry, we want to renew d_time */
 	iput(inode);
 }
 
@@ -848,20 +850,6 @@ struct dentry_operations nfs_dentry_operations = {
 	.d_iput		= nfs_dentry_iput,
 };
 
-/*
- * Use intent information to check whether or not we're going to do
- * an O_EXCL create using this path component.
- */
-static inline
-int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
-{
-	if (NFS_PROTO(dir)->version == 2)
-		return 0;
-	if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0)
-		return 0;
-	return (nd->intent.open.flags & O_EXCL) != 0;
-}
-
 static inline int nfs_reval_fsid(struct inode *dir, const struct nfs_fattr *fattr)
 {
 	struct nfs_server *server = NFS_SERVER(dir);
@@ -984,28 +972,16 @@ static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry
 	}
 	dentry->d_op = NFS_PROTO(dir)->dentry_ops;
 
-	/* Let vfs_create() deal with O_EXCL */
+	/* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
+	 * the dentry. */
 	if (nd->intent.open.flags & O_EXCL) {
-		d_add(dentry, NULL);
+		d_instantiate(dentry, NULL);
 		goto out;
 	}
 
 	/* Open the file on the server */
 	lock_kernel();
-	/* Revalidate parent directory attribute cache */
-	error = nfs_revalidate_inode(NFS_SERVER(dir), dir);
-	if (error < 0) {
-		res = ERR_PTR(error);
-		unlock_kernel();
-		goto out;
-	}
-
-	if (nd->intent.open.flags & O_CREAT) {
-		nfs_begin_data_update(dir);
-		res = nfs4_atomic_open(dir, dentry, nd);
-		nfs_end_data_update(dir);
-	} else
-		res = nfs4_atomic_open(dir, dentry, nd);
+	res = nfs4_atomic_open(dir, dentry, nd);
 	unlock_kernel();
 	if (IS_ERR(res)) {
 		error = PTR_ERR(res);
@@ -1038,7 +1014,6 @@ static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
 	struct dentry *parent = NULL;
 	struct inode *inode = dentry->d_inode;
 	struct inode *dir;
-	unsigned long verifier;
 	int openflags, ret = 0;
 
 	parent = dget_parent(dentry);
@@ -1048,8 +1023,12 @@ static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
 	/* We can't create new files in nfs_open_revalidate(), so we
 	 * optimize away revalidation of negative dentries.
 	 */
-	if (inode == NULL)
+	if (inode == NULL) {
+		if (!nfs_neg_need_reval(dir, dentry, nd))
+			ret = 1;
 		goto out;
+	}
+
 	/* NFS only supports OPEN on regular files */
 	if (!S_ISREG(inode->i_mode))
 		goto no_open;
@@ -1066,12 +1045,11 @@ static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
 	 * change attribute *before* we do the RPC call.
 	 */
 	lock_kernel();
-	verifier = nfs_save_change_attribute(dir);
 	ret = nfs4_open_revalidate(dir, dentry, openflags, nd);
 	unlock_kernel();
 out:
 	dput(parent);
-	if (!ret)
+	if (ret == 1)
 		d_drop(dentry);
 	return ret;
 no_open:
@@ -1093,6 +1071,7 @@ static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc)
 		.len = entry->len,
 	};
 	struct inode *inode;
+	unsigned long verf = nfs_save_change_attribute(dir);
 
 	switch (name.len) {
 		case 2:
@@ -1103,6 +1082,14 @@ static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc)
 			if (name.name[0] == '.')
 				return dget(parent);
 	}
+
+	spin_lock(&dir->i_lock);
+	if (NFS_I(dir)->cache_validity & NFS_INO_INVALID_DATA) {
+		spin_unlock(&dir->i_lock);
+		return NULL;
+	}
+	spin_unlock(&dir->i_lock);
+
 	name.hash = full_name_hash(name.name, name.len);
 	dentry = d_lookup(parent, &name);
 	if (dentry != NULL) {
@@ -1142,7 +1129,7 @@ static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc)
 	}
 
 out_renew:
-	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
+	nfs_set_verifier(dentry, verf);
 	return dentry;
 }
 
@@ -1152,32 +1139,40 @@ out_renew:
 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
 				struct nfs_fattr *fattr)
 {
+	struct dentry *parent = dget_parent(dentry);
+	struct inode *dir = parent->d_inode;
 	struct inode *inode;
 	int error = -EACCES;
 
+	d_drop(dentry);
+
 	/* We may have been initialized further down */
 	if (dentry->d_inode)
-		return 0;
+		goto out;
 	if (fhandle->size == 0) {
-		struct inode *dir = dentry->d_parent->d_inode;
 		error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
 		if (error)
-			return error;
+			goto out_error;
 	}
+	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
 	if (!(fattr->valid & NFS_ATTR_FATTR)) {
 		struct nfs_server *server = NFS_SB(dentry->d_sb);
 		error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
 		if (error < 0)
-			return error;
+			goto out_error;
 	}
 	inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
 	error = PTR_ERR(inode);
 	if (IS_ERR(inode))
-		return error;
-	d_instantiate(dentry, inode);
-	if (d_unhashed(dentry))
-		d_rehash(dentry);
+		goto out_error;
+	d_add(dentry, inode);
+out:
+	dput(parent);
 	return 0;
+out_error:
+	nfs_mark_for_revalidate(dir);
+	dput(parent);
+	return error;
 }
 
 /*
@@ -1203,12 +1198,9 @@ static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
 		open_flags = nd->intent.open.flags;
 
 	lock_kernel();
-	nfs_begin_data_update(dir);
 	error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd);
-	nfs_end_data_update(dir);
 	if (error != 0)
 		goto out_err;
-	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
 	unlock_kernel();
 	return 0;
 out_err:
@@ -1236,12 +1228,9 @@ nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
 	attr.ia_valid = ATTR_MODE;
 
 	lock_kernel();
-	nfs_begin_data_update(dir);
 	status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
-	nfs_end_data_update(dir);
 	if (status != 0)
 		goto out_err;
-	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
 	unlock_kernel();
 	return 0;
 out_err:
@@ -1265,12 +1254,9 @@ static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 	attr.ia_mode = mode | S_IFDIR;
 
 	lock_kernel();
-	nfs_begin_data_update(dir);
 	error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
-	nfs_end_data_update(dir);
 	if (error != 0)
 		goto out_err;
-	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
 	unlock_kernel();
 	return 0;
 out_err:
@@ -1287,12 +1273,10 @@ static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
 			dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
 
 	lock_kernel();
-	nfs_begin_data_update(dir);
 	error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
 	/* Ensure the VFS deletes this inode */
 	if (error == 0 && dentry->d_inode != NULL)
 		dentry->d_inode->i_nlink = 0;
-	nfs_end_data_update(dir);
 	unlock_kernel();
 
 	return error;
@@ -1355,17 +1339,13 @@ dentry->d_parent->d_name.name, dentry->d_name.name);
 
 	qsilly.name = silly;
 	qsilly.len  = strlen(silly);
-	nfs_begin_data_update(dir);
 	if (dentry->d_inode) {
-		nfs_begin_data_update(dentry->d_inode);
 		error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
 				dir, &qsilly);
 		nfs_mark_for_revalidate(dentry->d_inode);
-		nfs_end_data_update(dentry->d_inode);
 	} else
 		error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
 				dir, &qsilly);
-	nfs_end_data_update(dir);
 	if (!error) {
 		nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
 		d_move(dentry, sdentry);
@@ -1399,19 +1379,15 @@ static int nfs_safe_remove(struct dentry *dentry)
 		goto out;
 	}
 
-	nfs_begin_data_update(dir);
 	if (inode != NULL) {
 		nfs_inode_return_delegation(inode);
-		nfs_begin_data_update(inode);
 		error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
 		/* The VFS may want to delete this inode */
 		if (error == 0)
 			inode->i_nlink--;
 		nfs_mark_for_revalidate(inode);
-		nfs_end_data_update(inode);
 	} else
 		error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
-	nfs_end_data_update(dir);
 out:
 	return error;
 }
@@ -1502,9 +1478,7 @@ static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *sym
 		memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
 	kunmap_atomic(kaddr, KM_USER0);
 
-	nfs_begin_data_update(dir);
 	error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
-	nfs_end_data_update(dir);
 	if (error != 0) {
 		dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
 			dir->i_sb->s_id, dir->i_ino,
@@ -1544,15 +1518,12 @@ nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
 		dentry->d_parent->d_name.name, dentry->d_name.name);
 
 	lock_kernel();
-	nfs_begin_data_update(dir);
-	nfs_begin_data_update(inode);
+	d_drop(dentry);
 	error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
 	if (error == 0) {
 		atomic_inc(&inode->i_count);
-		d_instantiate(dentry, inode);
+		d_add(dentry, inode);
 	}
-	nfs_end_data_update(inode);
-	nfs_end_data_update(dir);
 	unlock_kernel();
 	return error;
 }
@@ -1661,15 +1632,9 @@ go_ahead:
 		d_delete(new_dentry);
 	}
 
-	nfs_begin_data_update(old_dir);
-	nfs_begin_data_update(new_dir);
-	nfs_begin_data_update(old_inode);
 	error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
 					   new_dir, &new_dentry->d_name);
 	nfs_mark_for_revalidate(old_inode);
-	nfs_end_data_update(old_inode);
-	nfs_end_data_update(new_dir);
-	nfs_end_data_update(old_dir);
 out:
 	if (rehash)
 		d_rehash(rehash);
diff --git a/fs/nfs/direct.c b/fs/nfs/direct.c
index 6456d25..3ac2e7f 100644
--- a/fs/nfs/direct.c
+++ b/fs/nfs/direct.c
@@ -508,7 +508,6 @@ static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode
 			nfs_direct_write_reschedule(dreq);
 			break;
 		default:
-			nfs_end_data_update(inode);
 			if (dreq->commit_data != NULL)
 				nfs_commit_free(dreq->commit_data);
 			nfs_direct_free_writedata(dreq);
@@ -530,7 +529,6 @@ static inline void nfs_alloc_commit_data(struct nfs_direct_req *dreq)
 
 static void nfs_direct_write_complete(struct nfs_direct_req *dreq, struct inode *inode)
 {
-	nfs_end_data_update(inode);
 	nfs_direct_free_writedata(dreq);
 	nfs_direct_complete(dreq);
 }
@@ -716,8 +714,6 @@ static ssize_t nfs_direct_write(struct kiocb *iocb, unsigned long user_addr, siz
 
 	nfs_add_stats(inode, NFSIOS_DIRECTWRITTENBYTES, count);
 
-	nfs_begin_data_update(inode);
-
 	rpc_clnt_sigmask(clnt, &oldset);
 	result = nfs_direct_write_schedule(dreq, user_addr, count, pos, sync);
 	if (!result)
@@ -839,10 +835,6 @@ ssize_t nfs_file_direct_write(struct kiocb *iocb, const char __user *buf, size_t
 	retval = nfs_direct_write(iocb, (unsigned long) buf, count, pos);
 
 	/*
-	 * XXX: nfs_end_data_update() already ensures this file's
-	 *      cached data is subsequently invalidated.  Do we really
-	 *      need to call invalidate_inode_pages2() again here?
-	 *
 	 *      For aio writes, this invalidation will almost certainly
 	 *      occur before the writes complete.  Kind of racey.
 	 */
diff --git a/fs/nfs/inode.c b/fs/nfs/inode.c
index 47e664b..1e9ce09 100644
--- a/fs/nfs/inode.c
+++ b/fs/nfs/inode.c
@@ -99,7 +99,6 @@ void nfs_clear_inode(struct inode *inode)
 	 */
 	BUG_ON(nfs_have_writebacks(inode));
 	BUG_ON (!list_empty(&NFS_I(inode)->open_files));
-	BUG_ON(atomic_read(&NFS_I(inode)->data_updates) != 0);
 	nfs_zap_acl_cache(inode);
 	nfs_access_zap_cache(inode);
 
@@ -165,6 +164,13 @@ static void nfs_zap_acl_cache(struct inode *inode)
 	spin_unlock(&inode->i_lock);
 }
 
+void nfs_invalidate_atime(struct inode *inode)
+{
+	spin_lock(&inode->i_lock);
+	NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
+	spin_unlock(&inode->i_lock);
+}
+
 /*
  * Invalidate, but do not unhash, the inode.
  * NB: must be called with inode->i_lock held!
@@ -350,7 +356,6 @@ nfs_setattr(struct dentry *dentry, struct iattr *attr)
 		return 0;
 
 	lock_kernel();
-	nfs_begin_data_update(inode);
 	/* Write all dirty data */
 	filemap_write_and_wait(inode->i_mapping);
 	nfs_wb_all(inode);
@@ -362,7 +367,6 @@ nfs_setattr(struct dentry *dentry, struct iattr *attr)
 	error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
 	if (error == 0)
 		nfs_refresh_inode(inode, &fattr);
-	nfs_end_data_update(inode);
 	unlock_kernel();
 	return error;
 }
@@ -616,16 +620,10 @@ __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
 	status = nfs_wait_on_inode(inode);
 	if (status < 0)
 		goto out;
-	if (NFS_STALE(inode)) {
-		status = -ESTALE;
-		/* Do we trust the cached ESTALE? */
-		if (NFS_ATTRTIMEO(inode) != 0) {
-			if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME)) {
-				/* no */
-			} else
-				goto out;
-		}
-	}
+
+	status = -ESTALE;
+	if (NFS_STALE(inode))
+		goto out;
 
 	status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
 	if (status != 0) {
@@ -714,11 +712,8 @@ int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
 
 		spin_lock(&inode->i_lock);
 		nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
-		if (S_ISDIR(inode->i_mode)) {
+		if (S_ISDIR(inode->i_mode))
 			memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
-			/* This ensures we revalidate child dentries */
-			nfsi->cache_change_attribute = jiffies;
-		}
 		spin_unlock(&inode->i_lock);
 
 		nfs_fscache_renew_cookie(inode);
@@ -730,57 +725,27 @@ int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
 	return ret;
 }
 
-/**
- * nfs_begin_data_update
- * @inode - pointer to inode
- * Declare that a set of operations will update file data on the server
- */
-void nfs_begin_data_update(struct inode *inode)
-{
-	atomic_inc(&NFS_I(inode)->data_updates);
-}
-
-/**
- * nfs_end_data_update
- * @inode - pointer to inode
- * Declare end of the operations that will update file data
- * This will mark the inode as immediately needing revalidation
- * of its attribute cache.
- */
-void nfs_end_data_update(struct inode *inode)
+static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
 {
 	struct nfs_inode *nfsi = NFS_I(inode);
 
-	if (!nfs_have_delegation(inode, FMODE_READ)) {
-		/* Directories and symlinks: invalidate page cache */
-		if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) {
-			spin_lock(&inode->i_lock);
+	if ((fattr->valid & NFS_ATTR_WCC_V4) != 0 &&
+			nfsi->change_attr == fattr->pre_change_attr) {
+		nfsi->change_attr = fattr->change_attr;
+		if (S_ISDIR(inode->i_mode))
 			nfsi->cache_validity |= NFS_INO_INVALID_DATA;
-			spin_unlock(&inode->i_lock);
-		}
 	}
-	nfsi->cache_change_attribute = jiffies;
-	atomic_dec(&nfsi->data_updates);
-}
-
-static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
-{
-	struct nfs_inode *nfsi = NFS_I(inode);
-
 	/* If we have atomic WCC data, we may update some attributes */
 	if ((fattr->valid & NFS_ATTR_WCC) != 0) {
-		if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
+		if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
 			memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
-			nfsi->cache_change_attribute = jiffies;
-		}
 		if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
 			memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
-			nfsi->cache_change_attribute = jiffies;
+			if (S_ISDIR(inode->i_mode))
+				nfsi->cache_validity |= NFS_INO_INVALID_DATA;
 		}
-		if (inode->i_size == fattr->pre_size && nfsi->npages == 0) {
+		if (inode->i_size == fattr->pre_size && nfsi->npages == 0)
 			inode->i_size = fattr->size;
-			nfsi->cache_change_attribute = jiffies;
-		}
 	}
 }
 
@@ -797,7 +762,6 @@ static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fat
 {
 	struct nfs_inode *nfsi = NFS_I(inode);
 	loff_t cur_size, new_isize;
-	int data_unstable;
 
 
 	/* Has the inode gone and changed behind our back? */
@@ -806,9 +770,6 @@ static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fat
 		return -EIO;
 	}
 
-	/* Are we in the process of updating data on the server? */
-	data_unstable = nfs_caches_unstable(inode);
-
 	/* Do atomic weak cache consistency updates */
 	nfs_wcc_update_inode(inode, fattr);
 
@@ -880,17 +841,50 @@ int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
 {
 	struct nfs_inode *nfsi = NFS_I(inode);
-	int status = 0;
 
-	spin_lock(&inode->i_lock);
-	if (unlikely((fattr->valid & NFS_ATTR_FATTR) == 0)) {
-		nfsi->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
-		goto out;
+	if (fattr->valid & NFS_ATTR_FATTR) {
+		if (S_ISDIR(inode->i_mode)) {
+			spin_lock(&inode->i_lock);
+			nfsi->cache_validity |= NFS_INO_INVALID_DATA;
+			spin_unlock(&inode->i_lock);
+		}
+		return nfs_update_inode(inode, fattr);
 	}
-	status = nfs_update_inode(inode, fattr);
-out:
+
+	spin_lock(&inode->i_lock);
+	nfsi->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
+	if (S_ISDIR(inode->i_mode))
+		nfsi->cache_validity |= NFS_INO_INVALID_DATA;
 	spin_unlock(&inode->i_lock);
-	return status;
+	return 0;
+}
+
+/**
+ * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
+ * @inode - pointer to inode
+ * @fattr - updated attributes
+ *
+ * After an operation that has changed the inode metadata, mark the
+ * attribute cache as being invalid, then try to update it. Fake up
+ * weak cache consistency data, if none exist.
+ *
+ * This function is mainly designed to be used by the ->write_done() functions.
+ */
+int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
+{
+	if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
+			(fattr->valid & NFS_ATTR_WCC_V4) == 0) {
+		fattr->pre_change_attr = NFS_I(inode)->change_attr;
+		fattr->valid |= NFS_ATTR_WCC_V4;
+	}
+	if ((fattr->valid & NFS_ATTR_FATTR) != 0 &&
+			(fattr->valid & NFS_ATTR_WCC) == 0) {
+		memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
+		memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
+		fattr->pre_size = inode->i_size;
+		fattr->valid |= NFS_ATTR_WCC;
+	}
+	return nfs_post_op_update_inode(inode, fattr);
 }
 
 /*
@@ -911,7 +905,6 @@ static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
 	struct nfs_inode *nfsi = NFS_I(inode);
 	loff_t cur_isize, new_isize;
 	unsigned int	invalid = 0;
-	int data_stable;
 
 	dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
 			__FUNCTION__, inode->i_sb->s_id, inode->i_ino,
@@ -938,53 +931,50 @@ static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
 	nfsi->read_cache_jiffies = fattr->time_start;
 	nfsi->last_updated = jiffies;
 
-	/* Are we racing with known updates of the metadata on the server? */
-	data_stable = nfs_verify_change_attribute(inode, fattr->time_start);
-	if (data_stable)
-		nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_ATIME);
+	nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ATIME
+			| NFS_INO_REVAL_PAGECACHE);
 
 	/* Do atomic weak cache consistency updates */
 	nfs_wcc_update_inode(inode, fattr);
 
+	/* More cache consistency checks */
+	if (!(fattr->valid & NFS_ATTR_FATTR_V4) != 0) {
+		/* NFSv2/v3: Check if the mtime agrees */
+		if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
+			dprintk("NFS: mtime change on server for file %s/%ld\n",
+					inode->i_sb->s_id, inode->i_ino);
+			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
+			nfsi->cache_change_attribute = jiffies;
+		}
+		/* If ctime has changed we should definitely clear access+acl caches */
+		if (!timespec_equal(&inode->i_ctime, &fattr->ctime))
+			invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
+	} else if (nfsi->change_attr != fattr->change_attr) {
+		dprintk("NFS: change_attr change on server for file %s/%ld\n",
+				inode->i_sb->s_id, inode->i_ino);
+		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
+		nfsi->cache_change_attribute = jiffies;
+	}
+
 	/* Check if our cached file size is stale */
  	new_isize = nfs_size_to_loff_t(fattr->size);
 	cur_isize = i_size_read(inode);
 	if (new_isize != cur_isize) {
-		/* Do we perhaps have any outstanding writes? */
-		if (nfsi->npages == 0) {
-			/* No, but did we race with nfs_end_data_update()? */
-			if (data_stable) {
-				inode->i_size = new_isize;
-				invalid |= NFS_INO_INVALID_DATA;
-				nfs_fscache_set_size(inode);
-			}
-			invalid |= NFS_INO_INVALID_ATTR;
-		} else if (new_isize > cur_isize) {
+		/* Do we perhaps have any outstanding writes, or has
+		 * the file grown beyond our last write? */
+		if (nfsi->npages == 0 || new_isize > cur_isize) {
 			inode->i_size = new_isize;
 			invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
 			nfs_fscache_set_size(inode);
 		}
-		nfsi->cache_change_attribute = jiffies;
 		dprintk("NFS: isize change on server for file %s/%ld\n",
 				inode->i_sb->s_id, inode->i_ino);
 	}
 
-	/* Check if the mtime agrees */
-	if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
-		memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
-		dprintk("NFS: mtime change on server for file %s/%ld\n",
-				inode->i_sb->s_id, inode->i_ino);
-		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
-		nfsi->cache_change_attribute = jiffies;
-	}
-
-	/* If ctime has changed we should definitely clear access+acl caches */
-	if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
-		invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
-		memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
-		nfsi->cache_change_attribute = jiffies;
-	}
+	memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
+	memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
 	memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
+	nfsi->change_attr = fattr->change_attr;
 
 	if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
 	    inode->i_uid != fattr->uid ||
@@ -1005,15 +995,6 @@ static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
  		inode->i_blocks = fattr->du.nfs2.blocks;
  	}
 
-	if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
-			nfsi->change_attr != fattr->change_attr) {
-		dprintk("NFS: change_attr change on server for file %s/%ld\n",
-				inode->i_sb->s_id, inode->i_ino);
-		nfsi->change_attr = fattr->change_attr;
-		invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
-		nfsi->cache_change_attribute = jiffies;
-	}
-
 	/* Update attrtimeo value if we're out of the unstable period */
 	if (invalid & NFS_INO_INVALID_ATTR) {
 		nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
@@ -1024,12 +1005,11 @@ static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
 			nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
 		nfsi->attrtimeo_timestamp = jiffies;
 	}
+	invalid &= ~NFS_INO_INVALID_ATTR;
 	/* Don't invalidate the data if we were to blame */
 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
 				|| S_ISLNK(inode->i_mode)))
 		invalid &= ~NFS_INO_INVALID_DATA;
-	if (data_stable)
-		invalid &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME|NFS_INO_REVAL_PAGECACHE);
 	if (!nfs_have_delegation(inode, FMODE_READ))
 		nfsi->cache_validity |= invalid;
 
@@ -1070,7 +1050,7 @@ static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
 void nfs4_clear_inode(struct inode *inode)
 {
 	/* If we are holding a delegation, return it! */
-	nfs_inode_return_delegation(inode);
+	nfs_inode_return_delegation_noreclaim(inode);
 	/* First call standard NFS clear_inode() code */
 	nfs_clear_inode(inode);
 }
@@ -1124,7 +1104,6 @@ static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
 		INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
 		INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
 		INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
-		atomic_set(&nfsi->data_updates, 0);
 		nfsi->ndirty = 0;
 		nfsi->ncommit = 0;
 		nfsi->npages = 0;
diff --git a/fs/nfs/nfs3acl.c b/fs/nfs/nfs3acl.c
index 7322da4..9b73625 100644
--- a/fs/nfs/nfs3acl.c
+++ b/fs/nfs/nfs3acl.c
@@ -317,13 +317,11 @@ static int nfs3_proc_setacls(struct inode *inode, struct posix_acl *acl,
 	}
 
 	dprintk("NFS call setacl\n");
-	nfs_begin_data_update(inode);
 	msg.rpc_proc = &server->client_acl->cl_procinfo[ACLPROC3_SETACL];
 	status = rpc_call_sync(server->client_acl, &msg, 0);
 	spin_lock(&inode->i_lock);
 	NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS;
 	spin_unlock(&inode->i_lock);
-	nfs_end_data_update(inode);
 	dprintk("NFS reply setacl: %d\n", status);
 
 	/* pages may have been allocated at the xdr layer. */
diff --git a/fs/nfs/nfs3proc.c b/fs/nfs/nfs3proc.c
index f95bc52..86e1529 100644
--- a/fs/nfs/nfs3proc.c
+++ b/fs/nfs/nfs3proc.c
@@ -167,6 +167,7 @@ nfs3_proc_lookup(struct inode *dir, struct qstr *name,
 	nfs_fattr_init(&dir_attr);
 	nfs_fattr_init(fattr);
 	status = rpc_call_sync(NFS_CLIENT(dir), &msg, 0);
+	nfs_refresh_inode(dir, &dir_attr);
 	if (status >= 0 && !(fattr->valid & NFS_ATTR_FATTR)) {
 		msg.rpc_proc = &nfs3_procedures[NFS3PROC_GETATTR];
 		msg.rpc_argp = fhandle;
@@ -174,8 +175,6 @@ nfs3_proc_lookup(struct inode *dir, struct qstr *name,
 		status = rpc_call_sync(NFS_CLIENT(dir), &msg, 0);
 	}
 	dprintk("NFS reply lookup: %d\n", status);
-	if (status >= 0)
-		status = nfs_refresh_inode(dir, &dir_attr);
 	return status;
 }
 
@@ -270,8 +269,8 @@ static int nfs3_proc_read(struct nfs_read_data *rdata)
 			(long long) rdata->args.offset);
 	nfs_fattr_init(fattr);
 	status = rpc_call_sync(NFS_CLIENT(inode), &msg, flags);
-	if (status >= 0)
-		nfs_refresh_inode(inode, fattr);
+	nfs_invalidate_atime(inode);
+	nfs_refresh_inode(inode, fattr);
 	dprintk("NFS reply read: %d\n", status);
 	return status;
 }
@@ -294,7 +293,7 @@ static int nfs3_proc_write(struct nfs_write_data *wdata)
 	nfs_fattr_init(fattr);
 	status = rpc_call_sync(NFS_CLIENT(inode), &msg, rpcflags);
 	if (status >= 0)
-		nfs_post_op_update_inode(inode, fattr);
+		nfs_post_op_update_inode_force_wcc(inode, fattr);
 	dprintk("NFS reply write: %d\n", status);
 	return status < 0? status : wdata->res.count;
 }
@@ -680,6 +679,9 @@ nfs3_proc_readdir(struct dentry *dentry, struct rpc_cred *cred,
 
 	nfs_fattr_init(&dir_attr);
 	status = rpc_call_sync(NFS_CLIENT(dir), &msg, 0);
+
+	nfs_invalidate_atime(dir);
+
 	nfs_refresh_inode(dir, &dir_attr);
 	dprintk("NFS reply readdir: %d\n", status);
 	unlock_kernel();
@@ -798,9 +800,9 @@ static int nfs3_read_done(struct rpc_task *task, struct nfs_read_data *data)
 {
 	if (nfs3_async_handle_jukebox(task, data->inode))
 		return -EAGAIN;
-	/* Call back common NFS readpage processing */
-	if (task->tk_status >= 0)
-		nfs_refresh_inode(data->inode, &data->fattr);
+
+	nfs_invalidate_atime(data->inode);
+	nfs_refresh_inode(data->inode, &data->fattr);
 	return 0;
 }
 
@@ -820,8 +822,7 @@ static int nfs3_write_done(struct rpc_task *task, struct nfs_write_data *data)
 {
 	if (nfs3_async_handle_jukebox(task, data->inode))
 		return -EAGAIN;
-	if (task->tk_status >= 0)
-		nfs_post_op_update_inode(data->inode, data->res.fattr);
+	nfs_refresh_inode(data->inode, data->res.fattr);
 	return 0;
 }
 
@@ -850,7 +851,7 @@ static int nfs3_commit_done(struct rpc_task *task, struct nfs_write_data *data)
 	if (nfs3_async_handle_jukebox(task, data->inode))
 		return -EAGAIN;
 	if (task->tk_status >= 0)
-		nfs_post_op_update_inode(data->inode, data->res.fattr);
+		nfs_post_op_update_inode_force_wcc(data->inode, data->res.fattr);
 	return 0;
 }
 
diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index 3d47bc6..0daf4e7 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -208,8 +208,9 @@ static void update_changeattr(struct inode *dir, struct nfs4_change_info *cinfo)
 
 	spin_lock(&dir->i_lock);
 	nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_DATA;
-	if (cinfo->before == nfsi->change_attr && cinfo->atomic)
-		nfsi->change_attr = cinfo->after;
+	if (!cinfo->atomic || cinfo->before != nfsi->change_attr)
+		nfsi->cache_change_attribute = jiffies;
+	nfsi->change_attr = cinfo->after;
 	spin_unlock(&dir->i_lock);
 }
 
@@ -1317,7 +1318,7 @@ nfs4_atomic_open(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
 	}
 	res = d_add_unique(dentry, igrab(state->inode));
 	if (res != NULL)
-		dentry = res;
+		path.dentry = res;
 	nfs_set_verifier(path.dentry, nfs_save_change_attribute(dir));
 	nfs_unblock_sillyrename(parent);
 	nfs4_intent_set_file(nd, &path, state);
@@ -1796,6 +1797,7 @@ static int _nfs4_proc_read(struct nfs_read_data *rdata)
 
 	nfs_fattr_init(fattr);
 	status = rpc_call_sync(server->client, &msg, flags);
+	nfs_invalidate_atime(inode);
 	if (!status)
 		renew_lease(server, timestamp);
 	dprintk("NFS reply read: %d\n", status);
@@ -1840,7 +1842,7 @@ static int _nfs4_proc_write(struct nfs_write_data *wdata)
 	if (status < 0)
 		return status;
 	renew_lease(server, wdata->timestamp);
-	nfs_post_op_update_inode(inode, fattr);
+	nfs_post_op_update_inode_force_wcc(inode, fattr);
 	return wdata->res.count;
 }
 
@@ -1931,11 +1933,12 @@ nfs4_proc_create(struct inode *dir, struct dentry *dentry, struct iattr *sattr,
 	}
 	state = nfs4_do_open(dir, &path, flags, sattr, cred);
 	put_rpccred(cred);
+	d_drop(dentry);
 	if (IS_ERR(state)) {
 		status = PTR_ERR(state);
 		goto out;
 	}
-	d_instantiate(dentry, igrab(state->inode));
+	d_add(dentry, igrab(state->inode));
 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
 	if (flags & O_EXCL) {
 		struct nfs_fattr fattr;
@@ -2256,6 +2259,9 @@ static int _nfs4_proc_readdir(struct dentry *dentry, struct rpc_cred *cred,
 	status = rpc_call_sync(NFS_CLIENT(dir), &msg, 0);
 	if (status == 0)
 		memcpy(NFS_COOKIEVERF(dir), res.verifier.data, NFS4_VERIFIER_SIZE);
+
+	nfs_invalidate_atime(dir);
+
 	unlock_kernel();
 	dprintk("%s: returns %d\n", __FUNCTION__, status);
 	return status;
@@ -2453,6 +2459,8 @@ static int nfs4_read_done(struct rpc_task *task, struct nfs_read_data *data)
 		rpc_restart_call(task);
 		return -EAGAIN;
 	}
+
+	nfs_invalidate_atime(data->inode);
 	if (task->tk_status > 0)
 		renew_lease(server, data->timestamp);
 	return 0;
@@ -2482,7 +2490,7 @@ static int nfs4_write_done(struct rpc_task *task, struct nfs_write_data *data)
 	}
 	if (task->tk_status >= 0) {
 		renew_lease(NFS_SERVER(inode), data->timestamp);
-		nfs_post_op_update_inode(inode, data->res.fattr);
+		nfs_post_op_update_inode_force_wcc(inode, data->res.fattr);
 	}
 	return 0;
 }
@@ -2524,8 +2532,7 @@ static int nfs4_commit_done(struct rpc_task *task, struct nfs_write_data *data)
 		rpc_restart_call(task);
 		return -EAGAIN;
 	}
-	if (task->tk_status >= 0)
-		nfs_post_op_update_inode(inode, data->res.fattr);
+	nfs_refresh_inode(inode, data->res.fattr);
 	return 0;
 }
 
@@ -3063,12 +3070,12 @@ static const struct rpc_call_ops nfs4_delegreturn_ops = {
 	.rpc_release = nfs4_delegreturn_release,
 };
 
-static int _nfs4_proc_delegreturn(struct inode *inode, struct rpc_cred *cred, const nfs4_stateid *stateid)
+static int _nfs4_proc_delegreturn(struct inode *inode, struct rpc_cred *cred, const nfs4_stateid *stateid, int issync)
 {
 	struct nfs4_delegreturndata *data;
 	struct nfs_server *server = NFS_SERVER(inode);
 	struct rpc_task *task;
-	int status;
+	int status = 0;
 
 	data = kmalloc(sizeof(*data), GFP_KERNEL);
 	if (data == NULL)
@@ -3087,23 +3094,27 @@ static int _nfs4_proc_delegreturn(struct inode *inode, struct rpc_cred *cred, co
 	task = rpc_run_task(NFS_CLIENT(inode), RPC_TASK_ASYNC, &nfs4_delegreturn_ops, data);
 	if (IS_ERR(task))
 		return PTR_ERR(task);
+	if (!issync)
+		goto out;
 	status = nfs4_wait_for_completion_rpc_task(task);
-	if (status == 0) {
-		status = data->rpc_status;
-		if (status == 0)
-			nfs_post_op_update_inode(inode, &data->fattr);
-	}
+	if (status != 0)
+		goto out;
+	status = data->rpc_status;
+	if (status != 0)
+		goto out;
+	nfs_refresh_inode(inode, &data->fattr);
+out:
 	rpc_put_task(task);
 	return status;
 }
 
-int nfs4_proc_delegreturn(struct inode *inode, struct rpc_cred *cred, const nfs4_stateid *stateid)
+int nfs4_proc_delegreturn(struct inode *inode, struct rpc_cred *cred, const nfs4_stateid *stateid, int issync)
 {
 	struct nfs_server *server = NFS_SERVER(inode);
 	struct nfs4_exception exception = { };
 	int err;
 	do {
-		err = _nfs4_proc_delegreturn(inode, cred, stateid);
+		err = _nfs4_proc_delegreturn(inode, cred, stateid, issync);
 		switch (err) {
 			case -NFS4ERR_STALE_STATEID:
 			case -NFS4ERR_EXPIRED:
diff --git a/fs/nfs/proc.c b/fs/nfs/proc.c
index 3fcfe89..63cc5c6 100644
--- a/fs/nfs/proc.c
+++ b/fs/nfs/proc.c
@@ -203,6 +203,7 @@ static int nfs_proc_read(struct nfs_read_data *rdata)
 			(long long) rdata->args.offset);
 	nfs_fattr_init(fattr);
 	status = rpc_call_sync(NFS_CLIENT(inode), &msg, flags);
+	nfs_invalidate_atime(inode);
 	if (status >= 0) {
 		nfs_refresh_inode(inode, fattr);
 		/* Emulate the eof flag, which isn't normally needed in NFSv2
@@ -233,7 +234,7 @@ static int nfs_proc_write(struct nfs_write_data *wdata)
 	nfs_fattr_init(fattr);
 	status = rpc_call_sync(NFS_CLIENT(inode), &msg, flags);
 	if (status >= 0) {
-		nfs_post_op_update_inode(inode, fattr);
+		nfs_post_op_update_inode_force_wcc(inode, fattr);
 		wdata->res.count = wdata->args.count;
 		wdata->verf.committed = NFS_FILE_SYNC;
 	}
@@ -267,6 +268,7 @@ nfs_proc_create(struct inode *dir, struct dentry *dentry, struct iattr *sattr,
 	nfs_fattr_init(&fattr);
 	dprintk("NFS call  create %s\n", dentry->d_name.name);
 	status = rpc_call_sync(NFS_CLIENT(dir), &msg, 0);
+	nfs_mark_for_revalidate(dir);
 	if (status == 0)
 		status = nfs_instantiate(dentry, &fhandle, &fattr);
 	dprintk("NFS reply create: %d\n", status);
@@ -534,6 +536,8 @@ nfs_proc_readdir(struct dentry *dentry, struct rpc_cred *cred,
 	dprintk("NFS call  readdir %d\n", (unsigned int)cookie);
 	status = rpc_call_sync(NFS_CLIENT(dir), &msg, 0);
 
+	nfs_invalidate_atime(dir);
+
 	dprintk("NFS reply readdir: %d\n", status);
 	unlock_kernel();
 	return status;
@@ -609,6 +613,7 @@ nfs_proc_pathconf(struct nfs_server *server, struct nfs_fh *fhandle,
 
 static int nfs_read_done(struct rpc_task *task, struct nfs_read_data *data)
 {
+	nfs_invalidate_atime(data->inode);
 	if (task->tk_status >= 0) {
 		nfs_refresh_inode(data->inode, data->res.fattr);
 		/* Emulate the eof flag, which isn't normally needed in NFSv2
@@ -635,7 +640,7 @@ static void nfs_proc_read_setup(struct nfs_read_data *data)
 static int nfs_write_done(struct rpc_task *task, struct nfs_write_data *data)
 {
 	if (task->tk_status >= 0)
-		nfs_post_op_update_inode(data->inode, data->res.fattr);
+		nfs_post_op_update_inode_force_wcc(data->inode, data->res.fattr);
 	return 0;
 }
 
diff --git a/fs/nfs/read.c b/fs/nfs/read.c
index eacb590..4dde1a0 100644
--- a/fs/nfs/read.c
+++ b/fs/nfs/read.c
@@ -208,9 +208,6 @@ static int nfs_readpage_sync(struct nfs_open_context *ctx, struct inode *inode,
 		if (rdata->res.eof != 0 || result == 0)
 			break;
 	} while (count);
-	spin_lock(&inode->i_lock);
-	NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
-	spin_unlock(&inode->i_lock);
 
 	if (rdata->res.eof || rdata->res.count == rdata->args.count) {
 		SetPageUptodate(page);
@@ -494,9 +491,6 @@ int nfs_readpage_result(struct rpc_task *task, struct nfs_read_data *data)
 		set_bit(NFS_INO_STALE, &NFS_FLAGS(data->inode));
 		nfs_mark_for_revalidate(data->inode);
 	}
-	spin_lock(&data->inode->i_lock);
-	NFS_I(data->inode)->cache_validity |= NFS_INO_INVALID_ATIME;
-	spin_unlock(&data->inode->i_lock);
 	return 0;
 }
 
diff --git a/fs/nfs/unlink.c b/fs/nfs/unlink.c
index 698402b..29e2a40 100644
--- a/fs/nfs/unlink.c
+++ b/fs/nfs/unlink.c
@@ -87,7 +87,6 @@ static void nfs_async_unlink_init(struct rpc_task *task, void *calldata)
 		.rpc_cred = data->cred,
 	};
 
-	nfs_begin_data_update(dir);
 	NFS_PROTO(dir)->unlink_setup(&msg, dir);
 	rpc_call_setup(task, &msg, 0);
 }
@@ -105,8 +104,6 @@ static void nfs_async_unlink_done(struct rpc_task *task, void *calldata)
 
 	if (!NFS_PROTO(dir)->unlink_done(task, dir))
 		rpc_restart_call(task);
-	else
-		nfs_end_data_update(dir);
 }
 
 /**
diff --git a/fs/nfs/write.c b/fs/nfs/write.c
index af40a60..f8c8222 100644
--- a/fs/nfs/write.c
+++ b/fs/nfs/write.c
@@ -239,7 +239,6 @@ static int nfs_writepage_sync(struct nfs_open_context *ctx, struct inode *inode,
 		count, (long long)(page_offset(page) + offset));
 
 	set_page_writeback(page);
-	nfs_begin_data_update(inode);
 	do {
 		if (count < wsize)
 			wdata->args.count = count;
@@ -271,7 +270,6 @@ static int nfs_writepage_sync(struct nfs_open_context *ctx, struct inode *inode,
 		ClearPageError(page);
 
 io_error:
-	nfs_end_data_update(inode);
 	end_page_writeback(page);
 	nfs_writedata_release(wdata);
 	return written ? written : result;
@@ -432,7 +430,6 @@ static int nfs_inode_add_request(struct inode *inode, struct nfs_page *req)
 		return error;
 	if (!nfsi->npages) {
 		igrab(inode);
-		nfs_begin_data_update(inode);
 		if (nfs_have_delegation(inode, FMODE_WRITE))
 			nfsi->change_attr++;
 	}
@@ -458,7 +455,6 @@ static void nfs_inode_remove_request(struct nfs_page *req)
 	nfsi->npages--;
 	if (!nfsi->npages) {
 		spin_unlock(&nfsi->req_lock);
-		nfs_end_data_update(inode);
 		iput(inode);
 	} else
 		spin_unlock(&nfsi->req_lock);
diff --git a/include/linux/nfs_fs.h b/include/linux/nfs_fs.h
index f5d5094..004e5b5 100644
--- a/include/linux/nfs_fs.h
+++ b/include/linux/nfs_fs.h
@@ -250,11 +250,6 @@ static inline struct nfs_inode *NFS_I(struct inode *inode)
 
 #define NFS_FILEID(inode)		(NFS_I(inode)->fileid)
 
-static inline int nfs_caches_unstable(struct inode *inode)
-{
-	return atomic_read(&NFS_I(inode)->data_updates) != 0;
-}
-
 static inline void nfs_mark_for_revalidate(struct inode *inode)
 {
 	struct nfs_inode *nfsi = NFS_I(inode);
@@ -266,12 +261,6 @@ static inline void nfs_mark_for_revalidate(struct inode *inode)
 	spin_unlock(&inode->i_lock);
 }
 
-static inline void NFS_CACHEINV(struct inode *inode)
-{
-	if (!nfs_caches_unstable(inode))
-		nfs_mark_for_revalidate(inode);
-}
-
 static inline int nfs_server_capable(struct inode *inode, int cap)
 {
 	return NFS_SERVER(inode)->caps & cap;
@@ -289,26 +278,26 @@ static inline void nfs_set_verifier(struct dentry * dentry, unsigned long verf)
 
 /**
  * nfs_save_change_attribute - Returns the inode attribute change cookie
- * @inode - pointer to inode
+ * @dir - pointer to parent directory inode
  * The "change attribute" is updated every time we finish an operation
  * that will result in a metadata change on the server.
  */
-static inline long nfs_save_change_attribute(struct inode *inode)
+static inline unsigned long nfs_save_change_attribute(struct inode *dir)
 {
-	return NFS_I(inode)->cache_change_attribute;
+	return NFS_I(dir)->cache_change_attribute;
 }
 
 /**
- * nfs_verify_change_attribute - Detects NFS inode cache updates
- * @inode - pointer to inode
+ * nfs_verify_change_attribute - Detects NFS remote directory changes
+ * @dir - pointer to parent directory inode
  * @chattr - previously saved change attribute
- * Return "false" if metadata has been updated (or is in the process of
- * being updated) since the change attribute was saved.
+ * Return "false" if the verifiers doesn't match the change attribute.
+ * This would usually indicate that the directory contents have changed on
+ * the server, and that any dentries need revalidating.
  */
-static inline int nfs_verify_change_attribute(struct inode *inode, unsigned long chattr)
+static inline int nfs_verify_change_attribute(struct inode *dir, unsigned long chattr)
 {
-	return !nfs_caches_unstable(inode)
-		&& time_after_eq(chattr, NFS_I(inode)->cache_change_attribute);
+	return chattr == NFS_I(dir)->cache_change_attribute;
 }
 
 /*
@@ -316,10 +305,12 @@ static inline int nfs_verify_change_attribute(struct inode *inode, unsigned long
  */
 extern int nfs_sync_mapping(struct address_space *mapping);
 extern void nfs_zap_caches(struct inode *);
+extern void nfs_invalidate_atime(struct inode *);
 extern struct inode *nfs_fhget(struct super_block *, struct nfs_fh *,
 				struct nfs_fattr *);
 extern int nfs_refresh_inode(struct inode *, struct nfs_fattr *);
 extern int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr);
+extern int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr);
 extern int nfs_getattr(struct vfsmount *, struct dentry *, struct kstat *);
 extern int nfs_permission(struct inode *, int, struct nameidata *);
 extern int nfs_access_get_cached(struct inode *, struct rpc_cred *, struct nfs_access_entry *);
@@ -333,10 +324,6 @@ extern int __nfs_revalidate_inode(struct nfs_server *, struct inode *);
 extern int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping);
 extern int nfs_setattr(struct dentry *, struct iattr *);
 extern void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr);
-extern void nfs_begin_attr_update(struct inode *);
-extern void nfs_end_attr_update(struct inode *);
-extern void nfs_begin_data_update(struct inode *);
-extern void nfs_end_data_update(struct inode *);
 extern struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx);
 extern void put_nfs_open_context(struct nfs_open_context *ctx);
 extern struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, int mode);
diff --git a/include/linux/nfs_xdr.h b/include/linux/nfs_xdr.h
index 6e7e60a..eee9209 100644
--- a/include/linux/nfs_xdr.h
+++ b/include/linux/nfs_xdr.h
@@ -63,7 +63,8 @@ struct nfs_fattr {
 #define NFS_ATTR_FATTR		0x0002		/* post-op attributes */
 #define NFS_ATTR_FATTR_V3	0x0004		/* NFSv3 attributes */
 #define NFS_ATTR_FATTR_V4	0x0008		/* NFSv4 change attribute */
-#define NFS_ATTR_FATTR_V4_REFERRAL	0x0010		/* NFSv4 referral */
+#define NFS_ATTR_WCC_V4		0x0010		/* pre-op change attribute */
+#define NFS_ATTR_FATTR_V4_REFERRAL	0x0020		/* NFSv4 referral */
 
 /*
  * Info on the file system