NFSv4: Eliminate nfsv4 open race...

 Make NFSv4 return the fully initialized file pointer with the
 stateid that it created in the lookup w/intent.

 Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>

fs/nfs/dir.c

@@ -914,7 +914,6 @@ static int is_atomic_open(struct inode *dir, struct nameidata *nd)
 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
 {
 	struct dentry *res = NULL;
-	struct inode *inode = NULL;
 	int error;
 
 	/* Check that we are indeed trying to open this file */
@@ -928,8 +927,10 @@ 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 */
-	if (nd->intent.open.flags & O_EXCL)
-		goto no_entry;
+	if (nd->intent.open.flags & O_EXCL) {
+		d_add(dentry, NULL);
+		goto out;
+	}
 
 	/* Open the file on the server */
 	lock_kernel();
@@ -943,18 +944,18 @@ static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry
 
 	if (nd->intent.open.flags & O_CREAT) {
 		nfs_begin_data_update(dir);
-		inode = nfs4_atomic_open(dir, dentry, nd);
+		res = nfs4_atomic_open(dir, dentry, nd);
 		nfs_end_data_update(dir);
 	} else
-		inode = nfs4_atomic_open(dir, dentry, nd);
+		res = nfs4_atomic_open(dir, dentry, nd);
 	unlock_kernel();
-	if (IS_ERR(inode)) {
-		error = PTR_ERR(inode);
+	if (IS_ERR(res)) {
+		error = PTR_ERR(res);
 		switch (error) {
 			/* Make a negative dentry */
 			case -ENOENT:
-				inode = NULL;
-				break;
+				res = NULL;
+				goto out;
 			/* This turned out not to be a regular file */
 			case -ELOOP:
 				if (!(nd->intent.open.flags & O_NOFOLLOW))
@@ -962,13 +963,9 @@ static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry
 			/* case -EISDIR: */
 			/* case -EINVAL: */
 			default:
-				res = ERR_PTR(error);
 				goto out;
 		}
-	}
-no_entry:
-	res = d_add_unique(dentry, inode);
-	if (res != NULL)
+	} else if (res != NULL)
 		dentry = res;
 	nfs_renew_times(dentry);
 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
@@ -1012,7 +1009,7 @@ static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
 	 */
 	lock_kernel();
 	verifier = nfs_save_change_attribute(dir);
-	ret = nfs4_open_revalidate(dir, dentry, openflags);
+	ret = nfs4_open_revalidate(dir, dentry, openflags, nd);
 	if (!ret)
 		nfs_set_verifier(dentry, verifier);
 	unlock_kernel();
@@ -1135,7 +1132,7 @@ static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
 
 	lock_kernel();
 	nfs_begin_data_update(dir);
-	error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
+	error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd);
 	nfs_end_data_update(dir);
 	if (error != 0)
 		goto out_err;

fs/nfs/nfs3proc.c

@@ -299,7 +299,7 @@ static int nfs3_proc_commit(struct nfs_write_data *cdata)
  */
 static int
 nfs3_proc_create(struct inode *dir, struct dentry *dentry, struct iattr *sattr,
-		 int flags)
+		 int flags, struct nameidata *nd)
 {
 	struct nfs_fh		fhandle;
 	struct nfs_fattr	fattr;

fs/nfs/nfs4_fs.h

@@ -215,8 +215,8 @@ extern int nfs4_proc_setclientid_confirm(struct nfs4_client *);
 extern int nfs4_proc_async_renew(struct nfs4_client *);
 extern int nfs4_proc_renew(struct nfs4_client *);
 extern int nfs4_do_close(struct inode *inode, struct nfs4_state *state, mode_t mode);
-extern struct inode *nfs4_atomic_open(struct inode *, struct dentry *, struct nameidata *);
-extern int nfs4_open_revalidate(struct inode *, struct dentry *, int);
+extern struct dentry *nfs4_atomic_open(struct inode *, struct dentry *, struct nameidata *);
+extern int nfs4_open_revalidate(struct inode *, struct dentry *, int, struct nameidata *);
 
 extern struct nfs4_state_recovery_ops nfs4_reboot_recovery_ops;
 extern struct nfs4_state_recovery_ops nfs4_network_partition_recovery_ops;

fs/nfs/nfs4proc.c

@@ -47,6 +47,7 @@
 #include <linux/nfs_page.h>
 #include <linux/smp_lock.h>
 #include <linux/namei.h>
+#include <linux/mount.h>
 
 #include "nfs4_fs.h"
 #include "delegation.h"
@@ -947,12 +948,26 @@ out:
 	return status;
 }
 
-struct inode *
+static void nfs4_intent_set_file(struct nameidata *nd, struct dentry *dentry, struct nfs4_state *state)
+{
+	struct file *filp;
+
+	filp = lookup_instantiate_filp(nd, dentry, NULL);
+	if (!IS_ERR(filp)) {
+		struct nfs_open_context *ctx;
+		ctx = (struct nfs_open_context *)filp->private_data;
+		ctx->state = state;
+	} else
+		nfs4_close_state(state, nd->intent.open.flags);
+}
+
+struct dentry *
 nfs4_atomic_open(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
 {
 	struct iattr attr;
 	struct rpc_cred *cred;
 	struct nfs4_state *state;
+	struct dentry *res;
 
 	if (nd->flags & LOOKUP_CREATE) {
 		attr.ia_mode = nd->intent.open.create_mode;
@@ -966,16 +981,23 @@ nfs4_atomic_open(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
 
 	cred = rpcauth_lookupcred(NFS_SERVER(dir)->client->cl_auth, 0);
 	if (IS_ERR(cred))
-		return (struct inode *)cred;
+		return (struct dentry *)cred;
 	state = nfs4_do_open(dir, dentry, nd->intent.open.flags, &attr, cred);
 	put_rpccred(cred);
-	if (IS_ERR(state))
-		return (struct inode *)state;
-	return state->inode;
+	if (IS_ERR(state)) {
+		if (PTR_ERR(state) == -ENOENT)
+			d_add(dentry, NULL);
+		return (struct dentry *)state;
+	}
+	res = d_add_unique(dentry, state->inode);
+	if (res != NULL)
+		dentry = res;
+	nfs4_intent_set_file(nd, dentry, state);
+	return res;
 }
 
 int
-nfs4_open_revalidate(struct inode *dir, struct dentry *dentry, int openflags)
+nfs4_open_revalidate(struct inode *dir, struct dentry *dentry, int openflags, struct nameidata *nd)
 {
 	struct rpc_cred *cred;
 	struct nfs4_state *state;
@@ -988,18 +1010,30 @@ nfs4_open_revalidate(struct inode *dir, struct dentry *dentry, int openflags)
 	if (IS_ERR(state))
 		state = nfs4_do_open(dir, dentry, openflags, NULL, cred);
 	put_rpccred(cred);
-	if (state == ERR_PTR(-ENOENT) && dentry->d_inode == 0)
-		return 1;
-	if (IS_ERR(state))
-		return 0;
+	if (IS_ERR(state)) {
+		switch (PTR_ERR(state)) {
+			case -EPERM:
+			case -EACCES:
+			case -EDQUOT:
+			case -ENOSPC:
+			case -EROFS:
+				lookup_instantiate_filp(nd, (struct dentry *)state, NULL);
+				return 1;
+			case -ENOENT:
+				if (dentry->d_inode == NULL)
+					return 1;
+		}
+		goto out_drop;
+	}
 	inode = state->inode;
+	iput(inode);
 	if (inode == dentry->d_inode) {
-		iput(inode);
+		nfs4_intent_set_file(nd, dentry, state);
 		return 1;
 	}
-	d_drop(dentry);
 	nfs4_close_state(state, openflags);
-	iput(inode);
+out_drop:
+	d_drop(dentry);
 	return 0;
 }
 
@@ -1500,7 +1534,7 @@ static int nfs4_proc_commit(struct nfs_write_data *cdata)
 
 static int
 nfs4_proc_create(struct inode *dir, struct dentry *dentry, struct iattr *sattr,
-                 int flags)
+                 int flags, struct nameidata *nd)
 {
 	struct nfs4_state *state;
 	struct rpc_cred *cred;
@@ -1522,13 +1556,13 @@ nfs4_proc_create(struct inode *dir, struct dentry *dentry, struct iattr *sattr,
 		struct nfs_fattr fattr;
 		status = nfs4_do_setattr(NFS_SERVER(dir), &fattr,
 		                     NFS_FH(state->inode), sattr, state);
-		if (status == 0) {
+		if (status == 0)
 			nfs_setattr_update_inode(state->inode, sattr);
-			goto out;
-		}
-	} else if (flags != 0)
-		goto out;
-	nfs4_close_state(state, flags);
+	}
+	if (status == 0 && nd != NULL && (nd->flags & LOOKUP_OPEN))
+		nfs4_intent_set_file(nd, dentry, state);
+	else
+		nfs4_close_state(state, flags);
 out:
 	return status;
 }
@@ -2175,65 +2209,6 @@ nfs4_proc_renew(struct nfs4_client *clp)
 	return 0;
 }
 
-/*
- * We will need to arrange for the VFS layer to provide an atomic open.
- * Until then, this open method is prone to inefficiency and race conditions
- * due to the lookup, potential create, and open VFS calls from sys_open()
- * placed on the wire.
- */
-static int
-nfs4_proc_file_open(struct inode *inode, struct file *filp)
-{
-	struct dentry *dentry = filp->f_dentry;
-	struct nfs_open_context *ctx;
-	struct nfs4_state *state = NULL;
-	struct rpc_cred *cred;
-	int status = -ENOMEM;
-
-	dprintk("nfs4_proc_file_open: starting on (%.*s/%.*s)\n",
-	                       (int)dentry->d_parent->d_name.len,
-	                       dentry->d_parent->d_name.name,
-	                       (int)dentry->d_name.len, dentry->d_name.name);
-
-
-	/* Find our open stateid */
-	cred = rpcauth_lookupcred(NFS_SERVER(inode)->client->cl_auth, 0);
-	if (IS_ERR(cred))
-		return PTR_ERR(cred);
-	ctx = alloc_nfs_open_context(dentry, cred);
-	put_rpccred(cred);
-	if (unlikely(ctx == NULL))
-		return -ENOMEM;
-	status = -EIO; /* ERACE actually */
-	state = nfs4_find_state(inode, cred, filp->f_mode);
-	if (unlikely(state == NULL))
-		goto no_state;
-	ctx->state = state;
-	nfs4_close_state(state, filp->f_mode);
-	ctx->mode = filp->f_mode;
-	nfs_file_set_open_context(filp, ctx);
-	put_nfs_open_context(ctx);
-	if (filp->f_mode & FMODE_WRITE)
-		nfs_begin_data_update(inode);
-	return 0;
-no_state:
-	printk(KERN_WARNING "NFS: v4 raced in function %s\n", __FUNCTION__);
-	put_nfs_open_context(ctx);
-	return status;
-}
-
-/*
- * Release our state
- */
-static int
-nfs4_proc_file_release(struct inode *inode, struct file *filp)
-{
-	if (filp->f_mode & FMODE_WRITE)
-		nfs_end_data_update(inode);
-	nfs_file_clear_open_context(filp);
-	return 0;
-}
-
 static inline int nfs4_server_supports_acls(struct nfs_server *server)
 {
 	return (server->caps & NFS_CAP_ACLS)
@@ -3145,8 +3120,8 @@ struct nfs_rpc_ops	nfs_v4_clientops = {
 	.read_setup	= nfs4_proc_read_setup,
 	.write_setup	= nfs4_proc_write_setup,
 	.commit_setup	= nfs4_proc_commit_setup,
-	.file_open      = nfs4_proc_file_open,
-	.file_release   = nfs4_proc_file_release,
+	.file_open      = nfs_open,
+	.file_release   = nfs_release,
 	.lock		= nfs4_proc_lock,
 	.clear_acl_cache = nfs4_zap_acl_attr,
 };

fs/nfs/proc.c

@@ -216,7 +216,7 @@ static int nfs_proc_write(struct nfs_write_data *wdata)
 
 static int
 nfs_proc_create(struct inode *dir, struct dentry *dentry, struct iattr *sattr,
-		int flags)
+		int flags, struct nameidata *nd)
 {
 	struct nfs_fh		fhandle;
 	struct nfs_fattr	fattr;

include/linux/nfs_xdr.h

@@ -712,7 +712,7 @@ struct nfs_rpc_ops {
 	int	(*write)   (struct nfs_write_data *);
 	int	(*commit)  (struct nfs_write_data *);
 	int	(*create)  (struct inode *, struct dentry *,
-			    struct iattr *, int);
+			    struct iattr *, int, struct nameidata *);
 	int	(*remove)  (struct inode *, struct qstr *);
 	int	(*unlink_setup)  (struct rpc_message *,
 			    struct dentry *, struct qstr *);