UBUNTU: Ubuntu-2.6.38-12.51
[linux-flexiantxendom0-natty.git] / fs / namei.c
index 8d3f15b..44890c6 100644 (file)
@@ -169,8 +169,8 @@ EXPORT_SYMBOL(putname);
 /*
  * This does basic POSIX ACL permission checking
  */
-static inline int __acl_permission_check(struct inode *inode, int mask,
-               int (*check_acl)(struct inode *inode, int mask), int rcu)
+static int acl_permission_check(struct inode *inode, int mask, unsigned int flags,
+               int (*check_acl)(struct inode *inode, int mask, unsigned int flags))
 {
        umode_t                 mode = inode->i_mode;
 
@@ -180,13 +180,9 @@ static inline int __acl_permission_check(struct inode *inode, int mask,
                mode >>= 6;
        else {
                if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
-                       if (rcu) {
-                               return -ECHILD;
-                       } else {
-                               int error = check_acl(inode, mask);
-                               if (error != -EAGAIN)
-                                       return error;
-                       }
+                       int error = check_acl(inode, mask, flags);
+                       if (error != -EAGAIN)
+                               return error;
                }
 
                if (in_group_p(inode->i_gid))
@@ -201,32 +197,31 @@ static inline int __acl_permission_check(struct inode *inode, int mask,
        return -EACCES;
 }
 
-static inline int acl_permission_check(struct inode *inode, int mask,
-               int (*check_acl)(struct inode *inode, int mask))
-{
-       return __acl_permission_check(inode, mask, check_acl, 0);
-}
-
 /**
- * generic_permission  -  check for access rights on a Posix-like filesystem
+ * generic_permission -  check for access rights on a Posix-like filesystem
  * @inode:     inode to check access rights for
  * @mask:      right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
  * @check_acl: optional callback to check for Posix ACLs
+ * @flags:     IPERM_FLAG_ flags.
  *
  * Used to check for read/write/execute permissions on a file.
  * We use "fsuid" for this, letting us set arbitrary permissions
  * for filesystem access without changing the "normal" uids which
- * are used for other things..
+ * are used for other things.
+ *
+ * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
+ * request cannot be satisfied (eg. requires blocking or too much complexity).
+ * It would then be called again in ref-walk mode.
  */
-int generic_permission(struct inode *inode, int mask,
-               int (*check_acl)(struct inode *inode, int mask))
+int generic_permission(struct inode *inode, int mask, unsigned int flags,
+       int (*check_acl)(struct inode *inode, int mask, unsigned int flags))
 {
        int ret;
 
        /*
         * Do the basic POSIX ACL permission checks.
         */
-       ret = acl_permission_check(inode, mask, check_acl);
+       ret = acl_permission_check(inode, mask, flags, check_acl);
        if (ret != -EACCES)
                return ret;
 
@@ -281,9 +276,10 @@ int inode_permission(struct inode *inode, int mask)
        }
 
        if (inode->i_op->permission)
-               retval = inode->i_op->permission(inode, mask);
+               retval = inode->i_op->permission(inode, mask, 0);
        else
-               retval = generic_permission(inode, mask, inode->i_op->check_acl);
+               retval = generic_permission(inode, mask, 0,
+                               inode->i_op->check_acl);
 
        if (retval)
                return retval;
@@ -357,6 +353,7 @@ int deny_write_access(struct file * file)
 
        return 0;
 }
+EXPORT_SYMBOL(deny_write_access);
 
 /**
  * path_get - get a reference to a path
@@ -387,7 +384,7 @@ EXPORT_SYMBOL(path_put);
 /**
  * nameidata_drop_rcu - drop this nameidata out of rcu-walk
  * @nd: nameidata pathwalk data to drop
- * @Returns: 0 on success, -ECHLID on failure
+ * Returns: 0 on success, -ECHILD on failure
  *
  * Path walking has 2 modes, rcu-walk and ref-walk (see
  * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt
@@ -448,7 +445,7 @@ static inline int nameidata_drop_rcu_maybe(struct nameidata *nd)
  * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk
  * @nd: nameidata pathwalk data to drop
  * @dentry: dentry to drop
- * @Returns: 0 on success, -ECHLID on failure
+ * Returns: 0 on success, -ECHILD on failure
  *
  * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root,
  * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on
@@ -510,7 +507,7 @@ static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata *nd, struct d
 /**
  * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk
  * @nd: nameidata pathwalk data to drop
- * @Returns: 0 on success, -ECHLID on failure
+ * Returns: 0 on success, -ECHILD on failure
  *
  * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk.
  * nd->path should be the final element of the lookup, so nd->root is discarded.
@@ -557,16 +554,25 @@ static inline int nameidata_drop_rcu_last_maybe(struct nameidata *nd)
  */
 void release_open_intent(struct nameidata *nd)
 {
-       if (nd->intent.open.file->f_path.dentry == NULL)
-               put_filp(nd->intent.open.file);
-       else
-               fput(nd->intent.open.file);
+       struct file *file = nd->intent.open.file;
+
+       if (file && !IS_ERR(file)) {
+               if (file->f_path.dentry == NULL)
+                       put_filp(file);
+               else
+                       fput(file);
+       }
 }
 
-static inline struct dentry *
+static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
+{
+       return dentry->d_op->d_revalidate(dentry, nd);
+}
+
+static struct dentry *
 do_revalidate(struct dentry *dentry, struct nameidata *nd)
 {
-       int status = dentry->d_op->d_revalidate(dentry, nd);
+       int status = d_revalidate(dentry, nd);
        if (unlikely(status <= 0)) {
                /*
                 * The dentry failed validation.
@@ -574,19 +580,51 @@ do_revalidate(struct dentry *dentry, struct nameidata *nd)
                 * the dentry otherwise d_revalidate is asking us
                 * to return a fail status.
                 */
-               if (!status) {
-                       if (!d_invalidate(dentry)) {
-                               dput(dentry);
-                               dentry = NULL;
-                       }
-               } else {
+               if (status < 0) {
                        dput(dentry);
                        dentry = ERR_PTR(status);
+               } else if (!d_invalidate(dentry)) {
+                       dput(dentry);
+                       dentry = NULL;
                }
        }
        return dentry;
 }
 
+static inline struct dentry *
+do_revalidate_rcu(struct dentry *dentry, struct nameidata *nd)
+{
+       int status = d_revalidate(dentry, nd);
+       if (likely(status > 0))
+               return dentry;
+       if (status == -ECHILD) {
+               if (nameidata_dentry_drop_rcu(nd, dentry))
+                       return ERR_PTR(-ECHILD);
+               return do_revalidate(dentry, nd);
+       }
+       if (status < 0)
+               return ERR_PTR(status);
+       /* Don't d_invalidate in rcu-walk mode */
+       if (nameidata_dentry_drop_rcu(nd, dentry))
+               return ERR_PTR(-ECHILD);
+       if (!d_invalidate(dentry)) {
+               dput(dentry);
+               dentry = NULL;
+       }
+       return dentry;
+}
+
+static inline int need_reval_dot(struct dentry *dentry)
+{
+       if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
+               return 0;
+
+       if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
+               return 0;
+
+       return 1;
+}
+
 /*
  * force_reval_path - force revalidation of a dentry
  *
@@ -610,13 +648,12 @@ force_reval_path(struct path *path, struct nameidata *nd)
 
        /*
         * only check on filesystems where it's possible for the dentry to
-        * become stale. It's assumed that if this flag is set then the
-        * d_revalidate op will also be defined.
+        * become stale.
         */
-       if (!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT))
+       if (!need_reval_dot(dentry))
                return 0;
 
-       status = dentry->d_op->d_revalidate(dentry, nd);
+       status = d_revalidate(dentry, nd);
        if (status > 0)
                return 0;
 
@@ -636,22 +673,19 @@ force_reval_path(struct path *path, struct nameidata *nd)
  * short-cut DAC fails, then call ->permission() to do more
  * complete permission check.
  */
-static inline int __exec_permission(struct inode *inode, int rcu)
+static inline int exec_permission(struct inode *inode, unsigned int flags)
 {
        int ret;
 
        if (inode->i_op->permission) {
-               if (rcu)
-                       return -ECHILD;
-               ret = inode->i_op->permission(inode, MAY_EXEC);
-               if (!ret)
-                       goto ok;
-               return ret;
+               ret = inode->i_op->permission(inode, MAY_EXEC, flags);
+       } else {
+               ret = acl_permission_check(inode, MAY_EXEC, flags,
+                               inode->i_op->check_acl);
        }
-       ret = __acl_permission_check(inode, MAY_EXEC, inode->i_op->check_acl, rcu);
-       if (!ret)
+       if (likely(!ret))
                goto ok;
-       if (rcu && ret == -ECHILD)
+       if (ret == -ECHILD)
                return ret;
 
        if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH))
@@ -659,17 +693,7 @@ static inline int __exec_permission(struct inode *inode, int rcu)
 
        return ret;
 ok:
-       return security_inode_exec_permission(inode, rcu);
-}
-
-static int exec_permission(struct inode *inode)
-{
-       return __exec_permission(inode, 0);
-}
-
-static int exec_permission_rcu(struct inode *inode)
-{
-       return __exec_permission(inode, 1);
+       return security_inode_exec_permission(inode, flags);
 }
 
 static __always_inline void set_root(struct nameidata *nd)
@@ -684,9 +708,13 @@ static __always_inline void set_root_rcu(struct nameidata *nd)
 {
        if (!nd->root.mnt) {
                struct fs_struct *fs = current->fs;
-               spin_lock(&fs->lock);
-               nd->root = fs->root;
-               spin_unlock(&fs->lock);
+               unsigned seq;
+
+               do {
+                       seq = read_seqcount_begin(&fs->seq);
+                       nd->root = fs->root;
+                       nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
+               } while (read_seqcount_retry(&fs->seq, seq));
        }
 }
 
@@ -719,7 +747,8 @@ static void path_put_conditional(struct path *path, struct nameidata *nd)
                mntput(path->mnt);
 }
 
-static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
+static inline void path_to_nameidata(const struct path *path,
+                                       struct nameidata *nd)
 {
        if (!(nd->flags & LOOKUP_RCU)) {
                dput(nd->path.dentry);
@@ -731,20 +760,18 @@ static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
 }
 
 static __always_inline int
-__do_follow_link(struct path *path, struct nameidata *nd, void **p)
+__do_follow_link(const struct path *link, struct nameidata *nd, void **p)
 {
        int error;
-       struct dentry *dentry = path->dentry;
+       struct dentry *dentry = link->dentry;
+
+       BUG_ON(nd->flags & LOOKUP_RCU);
 
-       touch_atime(path->mnt, dentry);
+       touch_atime(link->mnt, dentry);
        nd_set_link(nd, NULL);
 
-       if (path->mnt != nd->path.mnt) {
-               path_to_nameidata(path, nd);
-               nd->inode = nd->path.dentry->d_inode;
-               dget(dentry);
-       }
-       mntget(path->mnt);
+       if (link->mnt == nd->path.mnt)
+               mntget(link->mnt);
 
        nd->last_type = LAST_BIND;
        *p = dentry->d_inode->i_op->follow_link(dentry, nd);
@@ -770,10 +797,16 @@ __do_follow_link(struct path *path, struct nameidata *nd, void **p)
  * Without that kind of total limit, nasty chains of consecutive
  * symlinks can cause almost arbitrarily long lookups. 
  */
-static inline int do_follow_link(struct path *path, struct nameidata *nd)
+static inline int do_follow_link(struct inode *inode, struct path *path, struct nameidata *nd)
 {
        void *cookie;
        int err = -ELOOP;
+
+       /* We drop rcu-walk here */
+       if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry))
+               return -ECHILD;
+       BUG_ON(inode != path->dentry->d_inode);
+
        if (current->link_count >= MAX_NESTED_LINKS)
                goto loop;
        if (current->total_link_count >= 40)
@@ -835,54 +868,148 @@ int follow_up(struct path *path)
 }
 
 /*
- * serialization is taken care of in namespace.c
+ * Perform an automount
+ * - return -EISDIR to tell follow_managed() to stop and return the path we
+ *   were called with.
  */
-static void __follow_mount_rcu(struct nameidata *nd, struct path *path,
-                               struct inode **inode)
+static int follow_automount(struct path *path, unsigned flags,
+                           bool *need_mntput)
 {
-       while (d_mountpoint(path->dentry)) {
-               struct vfsmount *mounted;
-               mounted = __lookup_mnt(path->mnt, path->dentry, 1);
-               if (!mounted)
-                       return;
-               path->mnt = mounted;
-               path->dentry = mounted->mnt_root;
-               nd->seq = read_seqcount_begin(&path->dentry->d_seq);
-               *inode = path->dentry->d_inode;
+       struct vfsmount *mnt;
+       int err;
+
+       if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
+               return -EREMOTE;
+
+       /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
+        * and this is the terminal part of the path.
+        */
+       if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_CONTINUE))
+               return -EISDIR; /* we actually want to stop here */
+
+       /* We want to mount if someone is trying to open/create a file of any
+        * type under the mountpoint, wants to traverse through the mountpoint
+        * or wants to open the mounted directory.
+        *
+        * We don't want to mount if someone's just doing a stat and they've
+        * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
+        * appended a '/' to the name.
+        */
+       if (!(flags & LOOKUP_FOLLOW) &&
+           !(flags & (LOOKUP_CONTINUE | LOOKUP_DIRECTORY |
+                      LOOKUP_OPEN | LOOKUP_CREATE)))
+               return -EISDIR;
+
+       current->total_link_count++;
+       if (current->total_link_count >= 40)
+               return -ELOOP;
+
+       mnt = path->dentry->d_op->d_automount(path);
+       if (IS_ERR(mnt)) {
+               /*
+                * The filesystem is allowed to return -EISDIR here to indicate
+                * it doesn't want to automount.  For instance, autofs would do
+                * this so that its userspace daemon can mount on this dentry.
+                *
+                * However, we can only permit this if it's a terminal point in
+                * the path being looked up; if it wasn't then the remainder of
+                * the path is inaccessible and we should say so.
+                */
+               if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_CONTINUE))
+                       return -EREMOTE;
+               return PTR_ERR(mnt);
        }
-}
 
-static int __follow_mount(struct path *path)
-{
-       int res = 0;
-       while (d_mountpoint(path->dentry)) {
-               struct vfsmount *mounted = lookup_mnt(path);
-               if (!mounted)
-                       break;
+       if (!mnt) /* mount collision */
+               return 0;
+
+       err = finish_automount(mnt, path);
+
+       switch (err) {
+       case -EBUSY:
+               /* Someone else made a mount here whilst we were busy */
+               return 0;
+       case 0:
                dput(path->dentry);
-               if (res)
+               if (*need_mntput)
                        mntput(path->mnt);
-               path->mnt = mounted;
-               path->dentry = dget(mounted->mnt_root);
-               res = 1;
+               path->mnt = mnt;
+               path->dentry = dget(mnt->mnt_root);
+               *need_mntput = true;
+               return 0;
+       default:
+               return err;
        }
-       return res;
+
 }
 
-static void follow_mount(struct path *path)
+/*
+ * Handle a dentry that is managed in some way.
+ * - Flagged for transit management (autofs)
+ * - Flagged as mountpoint
+ * - Flagged as automount point
+ *
+ * This may only be called in refwalk mode.
+ *
+ * Serialization is taken care of in namespace.c
+ */
+static int follow_managed(struct path *path, unsigned flags)
 {
-       while (d_mountpoint(path->dentry)) {
-               struct vfsmount *mounted = lookup_mnt(path);
-               if (!mounted)
-                       break;
-               dput(path->dentry);
-               mntput(path->mnt);
-               path->mnt = mounted;
-               path->dentry = dget(mounted->mnt_root);
+       unsigned managed;
+       bool need_mntput = false;
+       int ret;
+
+       /* Given that we're not holding a lock here, we retain the value in a
+        * local variable for each dentry as we look at it so that we don't see
+        * the components of that value change under us */
+       while (managed = ACCESS_ONCE(path->dentry->d_flags),
+              managed &= DCACHE_MANAGED_DENTRY,
+              unlikely(managed != 0)) {
+               /* Allow the filesystem to manage the transit without i_mutex
+                * being held. */
+               if (managed & DCACHE_MANAGE_TRANSIT) {
+                       BUG_ON(!path->dentry->d_op);
+                       BUG_ON(!path->dentry->d_op->d_manage);
+                       ret = path->dentry->d_op->d_manage(path->dentry,
+                                                          false, false);
+                       if (ret < 0)
+                               return ret == -EISDIR ? 0 : ret;
+               }
+
+               /* Transit to a mounted filesystem. */
+               if (managed & DCACHE_MOUNTED) {
+                       struct vfsmount *mounted = lookup_mnt(path);
+                       if (mounted) {
+                               dput(path->dentry);
+                               if (need_mntput)
+                                       mntput(path->mnt);
+                               path->mnt = mounted;
+                               path->dentry = dget(mounted->mnt_root);
+                               need_mntput = true;
+                               continue;
+                       }
+
+                       /* Something is mounted on this dentry in another
+                        * namespace and/or whatever was mounted there in this
+                        * namespace got unmounted before we managed to get the
+                        * vfsmount_lock */
+               }
+
+               /* Handle an automount point */
+               if (managed & DCACHE_NEED_AUTOMOUNT) {
+                       ret = follow_automount(path, flags, &need_mntput);
+                       if (ret < 0)
+                               return ret == -EISDIR ? 0 : ret;
+                       continue;
+               }
+
+               /* We didn't change the current path point */
+               break;
        }
+       return 0;
 }
 
-int follow_down(struct path *path)
+int follow_down_one(struct path *path)
 {
        struct vfsmount *mounted;
 
@@ -897,13 +1024,41 @@ int follow_down(struct path *path)
        return 0;
 }
 
+/*
+ * Skip to top of mountpoint pile in rcuwalk mode.  We abort the rcu-walk if we
+ * meet a managed dentry and we're not walking to "..".  True is returned to
+ * continue, false to abort.
+ */
+static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
+                              struct inode **inode, bool reverse_transit)
+{
+       while (d_mountpoint(path->dentry)) {
+               struct vfsmount *mounted;
+               if (unlikely(path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) &&
+                   !reverse_transit &&
+                   path->dentry->d_op->d_manage(path->dentry, false, true) < 0)
+                       return false;
+               mounted = __lookup_mnt(path->mnt, path->dentry, 1);
+               if (!mounted)
+                       break;
+               path->mnt = mounted;
+               path->dentry = mounted->mnt_root;
+               nd->seq = read_seqcount_begin(&path->dentry->d_seq);
+               *inode = path->dentry->d_inode;
+       }
+
+       if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
+               return reverse_transit;
+       return true;
+}
+
 static int follow_dotdot_rcu(struct nameidata *nd)
 {
        struct inode *inode = nd->inode;
 
        set_root_rcu(nd);
 
-       while(1) {
+       while (1) {
                if (nd->path.dentry == nd->root.dentry &&
                    nd->path.mnt == nd->root.mnt) {
                        break;
@@ -926,12 +1081,80 @@ static int follow_dotdot_rcu(struct nameidata *nd)
                nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
                inode = nd->path.dentry->d_inode;
        }
-       __follow_mount_rcu(nd, &nd->path, &inode);
+       __follow_mount_rcu(nd, &nd->path, &inode, true);
        nd->inode = inode;
 
        return 0;
 }
 
+/*
+ * Follow down to the covering mount currently visible to userspace.  At each
+ * point, the filesystem owning that dentry may be queried as to whether the
+ * caller is permitted to proceed or not.
+ *
+ * Care must be taken as namespace_sem may be held (indicated by mounting_here
+ * being true).
+ */
+int follow_down(struct path *path, bool mounting_here)
+{
+       unsigned managed;
+       int ret;
+
+       while (managed = ACCESS_ONCE(path->dentry->d_flags),
+              unlikely(managed & DCACHE_MANAGED_DENTRY)) {
+               /* Allow the filesystem to manage the transit without i_mutex
+                * being held.
+                *
+                * We indicate to the filesystem if someone is trying to mount
+                * something here.  This gives autofs the chance to deny anyone
+                * other than its daemon the right to mount on its
+                * superstructure.
+                *
+                * The filesystem may sleep at this point.
+                */
+               if (managed & DCACHE_MANAGE_TRANSIT) {
+                       BUG_ON(!path->dentry->d_op);
+                       BUG_ON(!path->dentry->d_op->d_manage);
+                       ret = path->dentry->d_op->d_manage(
+                               path->dentry, mounting_here, false);
+                       if (ret < 0)
+                               return ret == -EISDIR ? 0 : ret;
+               }
+
+               /* Transit to a mounted filesystem. */
+               if (managed & DCACHE_MOUNTED) {
+                       struct vfsmount *mounted = lookup_mnt(path);
+                       if (!mounted)
+                               break;
+                       dput(path->dentry);
+                       mntput(path->mnt);
+                       path->mnt = mounted;
+                       path->dentry = dget(mounted->mnt_root);
+                       continue;
+               }
+
+               /* Don't handle automount points here */
+               break;
+       }
+       return 0;
+}
+
+/*
+ * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
+ */
+static void follow_mount(struct path *path)
+{
+       while (d_mountpoint(path->dentry)) {
+               struct vfsmount *mounted = lookup_mnt(path);
+               if (!mounted)
+                       break;
+               dput(path->dentry);
+               mntput(path->mnt);
+               path->mnt = mounted;
+               path->dentry = dget(mounted->mnt_root);
+       }
+}
+
 static void follow_dotdot(struct nameidata *nd)
 {
        set_root(nd);
@@ -996,12 +1219,14 @@ static int do_lookup(struct nameidata *nd, struct qstr *name,
        struct vfsmount *mnt = nd->path.mnt;
        struct dentry *dentry, *parent = nd->path.dentry;
        struct inode *dir;
+       int err;
+
        /*
         * See if the low-level filesystem might want
         * to use its own hash..
         */
-       if (parent->d_op && parent->d_op->d_hash) {
-               int err = parent->d_op->d_hash(parent, nd->inode, name);
+       if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
+               err = parent->d_op->d_hash(parent, nd->inode, name);
                if (err < 0)
                        return err;
        }
@@ -1026,28 +1251,43 @@ static int do_lookup(struct nameidata *nd, struct qstr *name,
                        return -ECHILD;
 
                nd->seq = seq;
-               if (dentry->d_op && dentry->d_op->d_revalidate) {
-                       /* We commonly drop rcu-walk here */
-                       if (nameidata_dentry_drop_rcu(nd, dentry))
-                               return -ECHILD;
-                       goto need_revalidate;
+               if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
+                       dentry = do_revalidate_rcu(dentry, nd);
+                       if (!dentry)
+                               goto need_lookup;
+                       if (IS_ERR(dentry))
+                               goto fail;
+                       if (!(nd->flags & LOOKUP_RCU))
+                               goto done;
                }
                path->mnt = mnt;
                path->dentry = dentry;
-               __follow_mount_rcu(nd, path, inode);
-       } else {
-               dentry = __d_lookup(parent, name);
+               if (likely(__follow_mount_rcu(nd, path, inode, false)))
+                       return 0;
+               if (nameidata_drop_rcu(nd))
+                       return -ECHILD;
+               /* fallthru */
+       }
+       dentry = __d_lookup(parent, name);
+       if (!dentry)
+               goto need_lookup;
+found:
+       if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
+               dentry = do_revalidate(dentry, nd);
                if (!dentry)
                        goto need_lookup;
-found:
-               if (dentry->d_op && dentry->d_op->d_revalidate)
-                       goto need_revalidate;
-done:
-               path->mnt = mnt;
-               path->dentry = dentry;
-               __follow_mount(path);
-               *inode = path->dentry->d_inode;
+               if (IS_ERR(dentry))
+                       goto fail;
        }
+done:
+       path->mnt = mnt;
+       path->dentry = dentry;
+       err = follow_managed(path, nd->flags);
+       if (unlikely(err < 0)) {
+               path_put_conditional(path, nd);
+               return err;
+       }
+       *inode = path->dentry->d_inode;
        return 0;
 
 need_lookup:
@@ -1080,30 +1320,11 @@ need_lookup:
        mutex_unlock(&dir->i_mutex);
        goto found;
 
-need_revalidate:
-       dentry = do_revalidate(dentry, nd);
-       if (!dentry)
-               goto need_lookup;
-       if (IS_ERR(dentry))
-               goto fail;
-       goto done;
-
 fail:
        return PTR_ERR(dentry);
 }
 
 /*
- * This is a temporary kludge to deal with "automount" symlinks; proper
- * solution is to trigger them on follow_mount(), so that do_lookup()
- * would DTRT.  To be killed before 2.6.34-final.
- */
-static inline int follow_on_final(struct inode *inode, unsigned lookup_flags)
-{
-       return inode && unlikely(inode->i_op->follow_link) &&
-               ((lookup_flags & LOOKUP_FOLLOW) || S_ISDIR(inode->i_mode));
-}
-
-/*
  * Name resolution.
  * This is the basic name resolution function, turning a pathname into
  * the final dentry. We expect 'base' to be positive and a directory.
@@ -1134,7 +1355,7 @@ static int link_path_walk(const char *name, struct nameidata *nd)
 
                nd->flags |= LOOKUP_CONTINUE;
                if (nd->flags & LOOKUP_RCU) {
-                       err = exec_permission_rcu(nd->inode);
+                       err = exec_permission(nd->inode, IPERM_FLAG_RCU);
                        if (err == -ECHILD) {
                                if (nameidata_drop_rcu(nd))
                                        return -ECHILD;
@@ -1142,7 +1363,7 @@ static int link_path_walk(const char *name, struct nameidata *nd)
                        }
                } else {
 exec_again:
-                       err = exec_permission(nd->inode);
+                       err = exec_permission(nd->inode, 0);
                }
                if (err)
                        break;
@@ -1195,11 +1416,7 @@ exec_again:
                        goto out_dput;
 
                if (inode->i_op->follow_link) {
-                       /* We commonly drop rcu-walk here */
-                       if (nameidata_dentry_drop_rcu_maybe(nd, next.dentry))
-                               return -ECHILD;
-                       BUG_ON(inode != next.dentry->d_inode);
-                       err = do_follow_link(&next, nd);
+                       err = do_follow_link(inode, &next, nd);
                        if (err)
                                goto return_err;
                        nd->inode = nd->path.dentry->d_inode;
@@ -1241,11 +1458,9 @@ last_component:
                err = do_lookup(nd, &this, &next, &inode);
                if (err)
                        break;
-               if (follow_on_final(inode, lookup_flags)) {
-                       if (nameidata_dentry_drop_rcu_maybe(nd, next.dentry))
-                               return -ECHILD;
-                       BUG_ON(inode != next.dentry->d_inode);
-                       err = do_follow_link(&next, nd);
+               if (inode && unlikely(inode->i_op->follow_link) &&
+                   (lookup_flags & LOOKUP_FOLLOW)) {
+                       err = do_follow_link(inode, &next, nd);
                        if (err)
                                goto return_err;
                        nd->inode = nd->path.dentry->d_inode;
@@ -1278,15 +1493,16 @@ return_reval:
                 * We bypassed the ordinary revalidation routines.
                 * We may need to check the cached dentry for staleness.
                 */
-               if (nd->path.dentry && nd->path.dentry->d_sb &&
-                   (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
-                       if (nameidata_drop_rcu_maybe(nd))
+               if (need_reval_dot(nd->path.dentry)) {
+                       if (nameidata_drop_rcu_last_maybe(nd))
                                return -ECHILD;
-                       err = -ESTALE;
                        /* Note: we do not d_invalidate() */
-                       if (!nd->path.dentry->d_op->d_revalidate(
-                                       nd->path.dentry, nd))
+                       err = d_revalidate(nd->path.dentry, nd);
+                       if (!err)
+                               err = -ESTALE;
+                       if (err < 0)
                                break;
+                       return 0;
                }
 return_base:
                if (nameidata_drop_rcu_last_maybe(nd))
@@ -1332,6 +1548,7 @@ static int path_walk(const char *name, struct nameidata *nd)
                /* nd->path had been dropped */
                current->total_link_count = 0;
                nd->path = save;
+               nd->inode = save.dentry->d_inode;
                path_get(&nd->path);
                nd->flags |= LOOKUP_REVAL;
                result = link_path_walk(name, nd);
@@ -1369,26 +1586,31 @@ static int path_init_rcu(int dfd, const char *name, unsigned int flags, struct n
 
        if (*name=='/') {
                struct fs_struct *fs = current->fs;
+               unsigned seq;
 
                br_read_lock(vfsmount_lock);
                rcu_read_lock();
 
-               spin_lock(&fs->lock);
-               nd->root = fs->root;
-               nd->path = nd->root;
-               nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
-               spin_unlock(&fs->lock);
+               do {
+                       seq = read_seqcount_begin(&fs->seq);
+                       nd->root = fs->root;
+                       nd->path = nd->root;
+                       nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
+               } while (read_seqcount_retry(&fs->seq, seq));
 
        } else if (dfd == AT_FDCWD) {
                struct fs_struct *fs = current->fs;
+               unsigned seq;
 
                br_read_lock(vfsmount_lock);
                rcu_read_lock();
 
-               spin_lock(&fs->lock);
-               nd->path = fs->pwd;
-               nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
-               spin_unlock(&fs->lock);
+               do {
+                       seq = read_seqcount_begin(&fs->seq);
+                       nd->path = fs->pwd;
+                       nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
+               } while (read_seqcount_retry(&fs->seq, seq));
+
        } else {
                struct dentry *dentry;
 
@@ -1411,7 +1633,7 @@ static int path_init_rcu(int dfd, const char *name, unsigned int flags, struct n
                if (fput_needed)
                        nd->file = file;
 
-               nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
+               nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
                br_read_lock(vfsmount_lock);
                rcu_read_lock();
        }
@@ -1586,7 +1808,7 @@ static struct dentry *__lookup_hash(struct qstr *name,
        struct dentry *dentry;
        int err;
 
-       err = exec_permission(inode);
+       err = exec_permission(inode, 0);
        if (err)
                return ERR_PTR(err);
 
@@ -1594,7 +1816,7 @@ static struct dentry *__lookup_hash(struct qstr *name,
         * See if the low-level filesystem might want
         * to use its own hash..
         */
-       if (base->d_op && base->d_op->d_hash) {
+       if (base->d_flags & DCACHE_OP_HASH) {
                err = base->d_op->d_hash(base, inode, name);
                dentry = ERR_PTR(err);
                if (err < 0)
@@ -1608,7 +1830,7 @@ static struct dentry *__lookup_hash(struct qstr *name,
         */
        dentry = d_lookup(base, name);
 
-       if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
+       if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE))
                dentry = do_revalidate(dentry, nd);
 
        if (!dentry)
@@ -1622,12 +1844,13 @@ out:
  * needs parent already locked. Doesn't follow mounts.
  * SMP-safe.
  */
-static struct dentry *lookup_hash(struct nameidata *nd)
+struct dentry *lookup_hash(struct nameidata *nd)
 {
        return __lookup_hash(&nd->last, nd->path.dentry, nd);
 }
+EXPORT_SYMBOL(lookup_hash);
 
-static int __lookup_one_len(const char *name, struct qstr *this,
+int __lookup_one_len(const char *name, struct qstr *this,
                struct dentry *base, int len)
 {
        unsigned long hash;
@@ -1648,6 +1871,7 @@ static int __lookup_one_len(const char *name, struct qstr *this,
        this->hash = end_name_hash(hash);
        return 0;
 }
+EXPORT_SYMBOL(__lookup_one_len);
 
 /**
  * lookup_one_len - filesystem helper to lookup single pathname component
@@ -1909,8 +2133,9 @@ int may_open(struct path *path, int acc_mode, int flag)
        return break_lease(inode, flag);
 }
 
-static int handle_truncate(struct path *path)
+static int handle_truncate(struct file *filp)
 {
+       struct path *path = &filp->f_path;
        struct inode *inode = path->dentry->d_inode;
        int error = get_write_access(inode);
        if (error)
@@ -1924,7 +2149,7 @@ static int handle_truncate(struct path *path)
        if (!error) {
                error = do_truncate(path->dentry, 0,
                                    ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
-                                   NULL);
+                                   filp);
        }
        put_write_access(inode);
        return error;
@@ -2022,7 +2247,7 @@ static struct file *finish_open(struct nameidata *nd,
        }
        if (!IS_ERR(filp)) {
                if (will_truncate) {
-                       error = handle_truncate(&nd->path);
+                       error = handle_truncate(filp);
                        if (error) {
                                fput(filp);
                                filp = ERR_PTR(error);
@@ -2040,8 +2265,6 @@ static struct file *finish_open(struct nameidata *nd,
        return filp;
 
 exit:
-       if (!IS_ERR(nd->intent.open.file))
-               release_open_intent(nd);
        path_put(&nd->path);
        return ERR_PTR(error);
 }
@@ -2062,9 +2285,12 @@ static struct file *do_last(struct nameidata *nd, struct path *path,
                follow_dotdot(nd);
                dir = nd->path.dentry;
        case LAST_DOT:
-               if (nd->path.mnt->mnt_sb->s_type->fs_flags & FS_REVAL_DOT) {
-                       if (!dir->d_op->d_revalidate(dir, nd)) {
-                               error = -ESTALE;
+               if (need_reval_dot(dir)) {
+                       int status = d_revalidate(nd->path.dentry, nd);
+                       if (!status)
+                               status = -ESTALE;
+                       if (status < 0) {
+                               error = status;
                                goto exit;
                        }
                }
@@ -2136,11 +2362,9 @@ static struct file *do_last(struct nameidata *nd, struct path *path,
        if (open_flag & O_EXCL)
                goto exit_dput;
 
-       if (__follow_mount(path)) {
-               error = -ELOOP;
-               if (open_flag & O_NOFOLLOW)
-                       goto exit_dput;
-       }
+       error = follow_managed(path, nd->flags);
+       if (error < 0)
+               goto exit_dput;
 
        error = -ENOENT;
        if (!path->dentry->d_inode)
@@ -2163,8 +2387,6 @@ exit_mutex_unlock:
 exit_dput:
        path_put_conditional(path, nd);
 exit:
-       if (!IS_ERR(nd->intent.open.file))
-               release_open_intent(nd);
        path_put(&nd->path);
        return ERR_PTR(error);
 }
@@ -2238,21 +2460,29 @@ struct file *do_filp_open(int dfd, const char *pathname,
        /* !O_CREAT, simple open */
        error = do_path_lookup(dfd, pathname, flags, &nd);
        if (unlikely(error))
-               goto out_filp;
+               goto out_filp2;
        error = -ELOOP;
        if (!(nd.flags & LOOKUP_FOLLOW)) {
                if (nd.inode->i_op->follow_link)
-                       goto out_path;
+                       goto out_path2;
        }
        error = -ENOTDIR;
        if (nd.flags & LOOKUP_DIRECTORY) {
                if (!nd.inode->i_op->lookup)
-                       goto out_path;
+                       goto out_path2;
        }
        audit_inode(pathname, nd.path.dentry);
        filp = finish_open(&nd, open_flag, acc_mode);
+out2:
+       release_open_intent(&nd);
        return filp;
 
+out_path2:
+       path_put(&nd.path);
+out_filp2:
+       filp = ERR_PTR(error);
+       goto out2;
+
 creat:
        /* OK, have to create the file. Find the parent. */
        error = path_init_rcu(dfd, pathname,
@@ -2285,11 +2515,11 @@ reval:
        nd.flags = flags;
        filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
        while (unlikely(!filp)) { /* trailing symlink */
-               struct path holder;
+               struct path link = path;
+               struct inode *linki = link.dentry->d_inode;
                void *cookie;
                error = -ELOOP;
-               /* S_ISDIR part is a temporary automount kludge */
-               if (!(nd.flags & LOOKUP_FOLLOW) && !S_ISDIR(nd.inode->i_mode))
+               if (!(nd.flags & LOOKUP_FOLLOW))
                        goto exit_dput;
                if (count++ == 32)
                        goto exit_dput;
@@ -2305,29 +2535,29 @@ reval:
                 * just set LAST_BIND.
                 */
                nd.flags |= LOOKUP_PARENT;
-               error = security_inode_follow_link(path.dentry, &nd);
+               error = security_inode_follow_link(link.dentry, &nd);
                if (error)
                        goto exit_dput;
-               error = __do_follow_link(&path, &nd, &cookie);
+               error = __do_follow_link(&link, &nd, &cookie);
                if (unlikely(error)) {
-                       if (!IS_ERR(cookie) && nd.inode->i_op->put_link)
-                               nd.inode->i_op->put_link(path.dentry, &nd, cookie);
+                       if (!IS_ERR(cookie) && linki->i_op->put_link)
+                               linki->i_op->put_link(link.dentry, &nd, cookie);
                        /* nd.path had been dropped */
-                       nd.path = path;
+                       nd.path = link;
                        goto out_path;
                }
-               holder = path;
                nd.flags &= ~LOOKUP_PARENT;
                filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname);
-               if (nd.inode->i_op->put_link)
-                       nd.inode->i_op->put_link(holder.dentry, &nd, cookie);
-               path_put(&holder);
+               if (linki->i_op->put_link)
+                       linki->i_op->put_link(link.dentry, &nd, cookie);
+               path_put(&link);
        }
 out:
        if (nd.root.mnt)
                path_put(&nd.root);
        if (filp == ERR_PTR(-ESTALE) && !(flags & LOOKUP_REVAL))
                goto reval;
+       release_open_intent(&nd);
        return filp;
 
 exit_dput:
@@ -2335,8 +2565,6 @@ exit_dput:
 out_path:
        path_put(&nd.path);
 out_filp:
-       if (!IS_ERR(nd.intent.open.file))
-               release_open_intent(&nd);
        filp = ERR_PTR(error);
        goto out;
 }
@@ -3349,6 +3577,7 @@ const struct inode_operations page_symlink_inode_operations = {
 };
 
 EXPORT_SYMBOL(user_path_at);
+EXPORT_SYMBOL(follow_down_one);
 EXPORT_SYMBOL(follow_down);
 EXPORT_SYMBOL(follow_up);
 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */