Introduce path_put()
[safe/jmp/linux-2.6] / fs / namei.c
index 5af7681..b0df7ea 100644 (file)
 #include <linux/namei.h>
 #include <linux/quotaops.h>
 #include <linux/pagemap.h>
-#include <linux/dnotify.h>
-#include <linux/smp_lock.h>
+#include <linux/fsnotify.h>
 #include <linux/personality.h>
 #include <linux/security.h>
 #include <linux/syscalls.h>
 #include <linux/mount.h>
 #include <linux/audit.h>
+#include <linux/capability.h>
+#include <linux/file.h>
+#include <linux/fcntl.h>
 #include <asm/namei.h>
 #include <asm/uaccess.h>
 
  */
 /*
  * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
- * implemented.  Let's see if raised priority of ->s_vfs_rename_sem gives
+ * implemented.  Let's see if raised priority of ->s_vfs_rename_mutex gives
  * any extra contention...
  */
 
+static int link_path_walk(const char *name, struct nameidata *nd);
+
 /* In order to reduce some races, while at the same time doing additional
  * checking and hopefully speeding things up, we copy filenames to the
  * kernel data space before using them..
  * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
  * PATH_MAX includes the nul terminator --RR.
  */
-static inline int do_getname(const char __user *filename, char *page)
+static int do_getname(const char __user *filename, char *page)
 {
        int retval;
        unsigned long len = PATH_MAX;
@@ -155,7 +159,7 @@ char * getname(const char __user * filename)
 #ifdef CONFIG_AUDITSYSCALL
 void putname(const char *name)
 {
-       if (unlikely(current->audit_context))
+       if (unlikely(!audit_dummy_context()))
                audit_putname(name);
        else
                __putname(name);
@@ -224,6 +228,10 @@ int generic_permission(struct inode *inode, int mask,
 int permission(struct inode *inode, int mask, struct nameidata *nd)
 {
        int retval, submask;
+       struct vfsmount *mnt = NULL;
+
+       if (nd)
+               mnt = nd->path.mnt;
 
        if (mask & MAY_WRITE) {
                umode_t mode = inode->i_mode;
@@ -242,19 +250,72 @@ int permission(struct inode *inode, int mask, struct nameidata *nd)
                        return -EACCES;
        }
 
+       if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode)) {
+               /*
+                * MAY_EXEC on regular files is denied if the fs is mounted
+                * with the "noexec" flag.
+                */
+               if (mnt && (mnt->mnt_flags & MNT_NOEXEC))
+                       return -EACCES;
+       }
 
        /* Ordinary permission routines do not understand MAY_APPEND. */
        submask = mask & ~MAY_APPEND;
-       if (inode->i_op && inode->i_op->permission)
+       if (inode->i_op && inode->i_op->permission) {
                retval = inode->i_op->permission(inode, submask, nd);
-       else
+               if (!retval) {
+                       /*
+                        * Exec permission on a regular file is denied if none
+                        * of the execute bits are set.
+                        *
+                        * This check should be done by the ->permission()
+                        * method.
+                        */
+                       if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode) &&
+                           !(inode->i_mode & S_IXUGO))
+                               return -EACCES;
+               }
+       } else {
                retval = generic_permission(inode, submask, NULL);
+       }
        if (retval)
                return retval;
 
        return security_inode_permission(inode, mask, nd);
 }
 
+/**
+ * vfs_permission  -  check for access rights to a given path
+ * @nd:                lookup result that describes the path
+ * @mask:      right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
+ *
+ * Used to check for read/write/execute permissions on a path.
+ * We use "fsuid" for this, letting us set arbitrary permissions
+ * for filesystem access without changing the "normal" uids which
+ * are used for other things.
+ */
+int vfs_permission(struct nameidata *nd, int mask)
+{
+       return permission(nd->path.dentry->d_inode, mask, nd);
+}
+
+/**
+ * file_permission  -  check for additional access rights to a given file
+ * @file:      file to check access rights for
+ * @mask:      right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
+ *
+ * Used to check for read/write/execute permissions on an already opened
+ * file.
+ *
+ * Note:
+ *     Do not use this function in new code.  All access checks should
+ *     be done using vfs_permission().
+ */
+int file_permission(struct file *file, int mask)
+{
+       return permission(file->f_path.dentry->d_inode, mask, NULL);
+}
+
 /*
  * get_write_access() gets write permission for a file.
  * put_write_access() releases this write permission.
@@ -288,7 +349,7 @@ int get_write_access(struct inode * inode)
 
 int deny_write_access(struct file * file)
 {
-       struct inode *inode = file->f_dentry->d_inode;
+       struct inode *inode = file->f_path.dentry->d_inode;
 
        spin_lock(&inode->i_lock);
        if (atomic_read(&inode->i_writecount) > 0) {
@@ -301,20 +362,53 @@ int deny_write_access(struct file * file)
        return 0;
 }
 
-void path_release(struct nameidata *nd)
+/**
+ * path_put - put a reference to a path
+ * @path: path to put the reference to
+ *
+ * Given a path decrement the reference count to the dentry and the vfsmount.
+ */
+void path_put(struct path *path)
 {
-       dput(nd->dentry);
-       mntput(nd->mnt);
+       dput(path->dentry);
+       mntput(path->mnt);
 }
+EXPORT_SYMBOL(path_put);
 
-/*
- * umount() mustn't call path_release()/mntput() as that would clear
- * mnt_expiry_mark
+/**
+ * release_open_intent - free up open intent resources
+ * @nd: pointer to nameidata
  */
-void path_release_on_umount(struct nameidata *nd)
+void release_open_intent(struct nameidata *nd)
 {
-       dput(nd->dentry);
-       _mntput(nd->mnt);
+       if (nd->intent.open.file->f_path.dentry == NULL)
+               put_filp(nd->intent.open.file);
+       else
+               fput(nd->intent.open.file);
+}
+
+static inline struct dentry *
+do_revalidate(struct dentry *dentry, struct nameidata *nd)
+{
+       int status = dentry->d_op->d_revalidate(dentry, nd);
+       if (unlikely(status <= 0)) {
+               /*
+                * The dentry failed validation.
+                * If d_revalidate returned 0 attempt to invalidate
+                * the dentry otherwise d_revalidate is asking us
+                * to return a fail status.
+                */
+               if (!status) {
+                       if (!d_invalidate(dentry)) {
+                               dput(dentry);
+                               dentry = NULL;
+                       }
+               } else {
+                       dput(dentry);
+                       dentry = ERR_PTR(status);
+               }
+       }
+       return dentry;
 }
 
 /*
@@ -331,12 +425,9 @@ static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name,
        if (!dentry)
                dentry = d_lookup(parent, name);
 
-       if (dentry && dentry->d_op && dentry->d_op->d_revalidate) {
-               if (!dentry->d_op->d_revalidate(dentry, nd) && !d_invalidate(dentry)) {
-                       dput(dentry);
-                       dentry = NULL;
-               }
-       }
+       if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
+               dentry = do_revalidate(dentry, nd);
+
        return dentry;
 }
 
@@ -350,7 +441,7 @@ static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name,
  * short-cut DAC fails, then call permission() to do more
  * complete permission check.
  */
-static inline int exec_permission_lite(struct inode *inode,
+static int exec_permission_lite(struct inode *inode,
                                       struct nameidata *nd)
 {
        umode_t mode = inode->i_mode;
@@ -393,7 +484,7 @@ static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, s
        struct dentry * result;
        struct inode *dir = parent->d_inode;
 
-       down(&dir->i_sem);
+       mutex_lock(&dir->i_mutex);
        /*
         * First re-do the cached lookup just in case it was created
         * while we waited for the directory semaphore..
@@ -419,7 +510,7 @@ static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, s
                        else
                                result = dentry;
                }
-               up(&dir->i_sem);
+               mutex_unlock(&dir->i_mutex);
                return result;
        }
 
@@ -427,12 +518,11 @@ static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, s
         * Uhhuh! Nasty case: the cache was re-populated while
         * we waited on the semaphore. Need to revalidate.
         */
-       up(&dir->i_sem);
+       mutex_unlock(&dir->i_mutex);
        if (result->d_op && result->d_op->d_revalidate) {
-               if (!result->d_op->d_revalidate(result, nd) && !d_invalidate(result)) {
-                       dput(result);
+               result = do_revalidate(result, nd);
+               if (!result)
                        result = ERR_PTR(-ENOENT);
-               }
        }
        return result;
 }
@@ -440,25 +530,27 @@ static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, s
 static int __emul_lookup_dentry(const char *, struct nameidata *);
 
 /* SMP-safe */
-static inline int
+static __always_inline int
 walk_init_root(const char *name, struct nameidata *nd)
 {
-       read_lock(&current->fs->lock);
-       if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
-               nd->mnt = mntget(current->fs->altrootmnt);
-               nd->dentry = dget(current->fs->altroot);
-               read_unlock(&current->fs->lock);
+       struct fs_struct *fs = current->fs;
+
+       read_lock(&fs->lock);
+       if (fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
+               nd->path.mnt = mntget(fs->altrootmnt);
+               nd->path.dentry = dget(fs->altroot);
+               read_unlock(&fs->lock);
                if (__emul_lookup_dentry(name,nd))
                        return 0;
-               read_lock(&current->fs->lock);
+               read_lock(&fs->lock);
        }
-       nd->mnt = mntget(current->fs->rootmnt);
-       nd->dentry = dget(current->fs->root);
-       read_unlock(&current->fs->lock);
+       nd->path.mnt = mntget(fs->rootmnt);
+       nd->path.dentry = dget(fs->root);
+       read_unlock(&fs->lock);
        return 1;
 }
 
-static inline int __vfs_follow_link(struct nameidata *nd, const char *link)
+static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
 {
        int res = 0;
        char *name;
@@ -466,7 +558,7 @@ static inline int __vfs_follow_link(struct nameidata *nd, const char *link)
                goto fail;
 
        if (*link == '/') {
-               path_release(nd);
+               path_put(&nd->path);
                if (!walk_init_root(link, nd))
                        /* weird __emul_prefix() stuff did it */
                        goto out;
@@ -482,39 +574,56 @@ out:
         */
        name = __getname();
        if (unlikely(!name)) {
-               path_release(nd);
+               path_put(&nd->path);
                return -ENOMEM;
        }
        strcpy(name, nd->last.name);
        nd->last.name = name;
        return 0;
 fail:
-       path_release(nd);
+       path_put(&nd->path);
        return PTR_ERR(link);
 }
 
-struct path {
-       struct vfsmount *mnt;
-       struct dentry *dentry;
-};
+static void path_put_conditional(struct path *path, struct nameidata *nd)
+{
+       dput(path->dentry);
+       if (path->mnt != nd->path.mnt)
+               mntput(path->mnt);
+}
 
-static inline int __do_follow_link(struct path *path, struct nameidata *nd)
+static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
+{
+       dput(nd->path.dentry);
+       if (nd->path.mnt != path->mnt)
+               mntput(nd->path.mnt);
+       nd->path.mnt = path->mnt;
+       nd->path.dentry = path->dentry;
+}
+
+static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
 {
        int error;
+       void *cookie;
        struct dentry *dentry = path->dentry;
 
-       touch_atime(nd->mnt, dentry);
+       touch_atime(path->mnt, dentry);
        nd_set_link(nd, NULL);
 
-       if (path->mnt == nd->mnt)
-               mntget(path->mnt);
-       error = dentry->d_inode->i_op->follow_link(dentry, nd);
-       if (!error) {
+       if (path->mnt != nd->path.mnt) {
+               path_to_nameidata(path, nd);
+               dget(dentry);
+       }
+       mntget(path->mnt);
+       cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
+       error = PTR_ERR(cookie);
+       if (!IS_ERR(cookie)) {
                char *s = nd_get_link(nd);
+               error = 0;
                if (s)
                        error = __vfs_follow_link(nd, s);
                if (dentry->d_inode->i_op->put_link)
-                       dentry->d_inode->i_op->put_link(dentry, nd);
+                       dentry->d_inode->i_op->put_link(dentry, nd, cookie);
        }
        dput(dentry);
        mntput(path->mnt);
@@ -549,10 +658,8 @@ static inline int do_follow_link(struct path *path, struct nameidata *nd)
        nd->depth--;
        return err;
 loop:
-       dput(path->dentry);
-       if (path->mnt != nd->mnt)
-               mntput(path->mnt);
-       path_release(nd);
+       path_put_conditional(path, nd);
+       path_put(&nd->path);
        return err;
 }
 
@@ -627,41 +734,43 @@ int follow_down(struct vfsmount **mnt, struct dentry **dentry)
        return 0;
 }
 
-static inline void follow_dotdot(struct nameidata *nd)
+static __always_inline void follow_dotdot(struct nameidata *nd)
 {
+       struct fs_struct *fs = current->fs;
+
        while(1) {
                struct vfsmount *parent;
-               struct dentry *old = nd->dentry;
+               struct dentry *old = nd->path.dentry;
 
-                read_lock(&current->fs->lock);
-               if (nd->dentry == current->fs->root &&
-                   nd->mnt == current->fs->rootmnt) {
-                        read_unlock(&current->fs->lock);
+                read_lock(&fs->lock);
+               if (nd->path.dentry == fs->root &&
+                   nd->path.mnt == fs->rootmnt) {
+                        read_unlock(&fs->lock);
                        break;
                }
-                read_unlock(&current->fs->lock);
+                read_unlock(&fs->lock);
                spin_lock(&dcache_lock);
-               if (nd->dentry != nd->mnt->mnt_root) {
-                       nd->dentry = dget(nd->dentry->d_parent);
+               if (nd->path.dentry != nd->path.mnt->mnt_root) {
+                       nd->path.dentry = dget(nd->path.dentry->d_parent);
                        spin_unlock(&dcache_lock);
                        dput(old);
                        break;
                }
                spin_unlock(&dcache_lock);
                spin_lock(&vfsmount_lock);
-               parent = nd->mnt->mnt_parent;
-               if (parent == nd->mnt) {
+               parent = nd->path.mnt->mnt_parent;
+               if (parent == nd->path.mnt) {
                        spin_unlock(&vfsmount_lock);
                        break;
                }
                mntget(parent);
-               nd->dentry = dget(nd->mnt->mnt_mountpoint);
+               nd->path.dentry = dget(nd->path.mnt->mnt_mountpoint);
                spin_unlock(&vfsmount_lock);
                dput(old);
-               mntput(nd->mnt);
-               nd->mnt = parent;
+               mntput(nd->path.mnt);
+               nd->path.mnt = parent;
        }
-       follow_mount(&nd->mnt, &nd->dentry);
+       follow_mount(&nd->path.mnt, &nd->path.dentry);
 }
 
 /*
@@ -672,8 +781,8 @@ static inline void follow_dotdot(struct nameidata *nd)
 static int do_lookup(struct nameidata *nd, struct qstr *name,
                     struct path *path)
 {
-       struct vfsmount *mnt = nd->mnt;
-       struct dentry *dentry = __d_lookup(nd->dentry, name);
+       struct vfsmount *mnt = nd->path.mnt;
+       struct dentry *dentry = __d_lookup(nd->path.dentry, name);
 
        if (!dentry)
                goto need_lookup;
@@ -686,18 +795,18 @@ done:
        return 0;
 
 need_lookup:
-       dentry = real_lookup(nd->dentry, name, nd);
+       dentry = real_lookup(nd->path.dentry, name, nd);
        if (IS_ERR(dentry))
                goto fail;
        goto done;
 
 need_revalidate:
-       if (dentry->d_op->d_revalidate(dentry, nd))
-               goto done;
-       if (d_invalidate(dentry))
-               goto done;
-       dput(dentry);
-       goto need_lookup;
+       dentry = do_revalidate(dentry, nd);
+       if (!dentry)
+               goto need_lookup;
+       if (IS_ERR(dentry))
+               goto fail;
+       goto done;
 
 fail:
        return PTR_ERR(dentry);
@@ -711,7 +820,7 @@ fail:
  * Returns 0 and nd will have valid dentry and mnt on success.
  * Returns error and drops reference to input namei data on failure.
  */
-static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
+static int __link_path_walk(const char *name, struct nameidata *nd)
 {
        struct path next;
        struct inode *inode;
@@ -723,9 +832,9 @@ static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
        if (!*name)
                goto return_reval;
 
-       inode = nd->dentry->d_inode;
+       inode = nd->path.dentry->d_inode;
        if (nd->depth)
-               lookup_flags = LOOKUP_FOLLOW;
+               lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
 
        /* At this point we know we have a real path component. */
        for(;;) {
@@ -733,10 +842,10 @@ static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
                struct qstr this;
                unsigned int c;
 
+               nd->flags |= LOOKUP_CONTINUE;
                err = exec_permission_lite(inode, nd);
-               if (err == -EAGAIN) { 
-                       err = permission(inode, MAY_EXEC, nd);
-               }
+               if (err == -EAGAIN)
+                       err = vfs_permission(nd, MAY_EXEC);
                if (err)
                        break;
 
@@ -771,7 +880,7 @@ static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
                                if (this.name[1] != '.')
                                        break;
                                follow_dotdot(nd);
-                               inode = nd->dentry->d_inode;
+                               inode = nd->path.dentry->d_inode;
                                /* fallthrough */
                        case 1:
                                continue;
@@ -780,12 +889,12 @@ static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
                 * See if the low-level filesystem might want
                 * to use its own hash..
                 */
-               if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
-                       err = nd->dentry->d_op->d_hash(nd->dentry, &this);
+               if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
+                       err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
+                                                           &this);
                        if (err < 0)
                                break;
                }
-               nd->flags |= LOOKUP_CONTINUE;
                /* This does the actual lookups.. */
                err = do_lookup(nd, &this, &next);
                if (err)
@@ -804,19 +913,14 @@ static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
                        if (err)
                                goto return_err;
                        err = -ENOENT;
-                       inode = nd->dentry->d_inode;
+                       inode = nd->path.dentry->d_inode;
                        if (!inode)
                                break;
                        err = -ENOTDIR; 
                        if (!inode->i_op)
                                break;
-               } else {
-                       dput(nd->dentry);
-                       if (nd->mnt != next.mnt)
-                               mntput(nd->mnt);
-                       nd->mnt = next.mnt;
-                       nd->dentry = next.dentry;
-               }
+               } else
+                       path_to_nameidata(&next, nd);
                err = -ENOTDIR; 
                if (!inode->i_op->lookup)
                        break;
@@ -826,7 +930,8 @@ static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
 last_with_slashes:
                lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 last_component:
-               nd->flags &= ~LOOKUP_CONTINUE;
+               /* Clear LOOKUP_CONTINUE iff it was previously unset */
+               nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
                if (lookup_flags & LOOKUP_PARENT)
                        goto lookup_parent;
                if (this.name[0] == '.') switch (this.len) {
@@ -836,13 +941,14 @@ last_component:
                                if (this.name[1] != '.')
                                        break;
                                follow_dotdot(nd);
-                               inode = nd->dentry->d_inode;
+                               inode = nd->path.dentry->d_inode;
                                /* fallthrough */
                        case 1:
                                goto return_reval;
                }
-               if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
-                       err = nd->dentry->d_op->d_hash(nd->dentry, &this);
+               if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) {
+                       err = nd->path.dentry->d_op->d_hash(nd->path.dentry,
+                                                           &this);
                        if (err < 0)
                                break;
                }
@@ -855,14 +961,9 @@ last_component:
                        err = do_follow_link(&next, nd);
                        if (err)
                                goto return_err;
-                       inode = nd->dentry->d_inode;
-               } else {
-                       dput(nd->dentry);
-                       if (nd->mnt != next.mnt)
-                               mntput(nd->mnt);
-                       nd->mnt = next.mnt;
-                       nd->dentry = next.dentry;
-               }
+                       inode = nd->path.dentry->d_inode;
+               } else
+                       path_to_nameidata(&next, nd);
                err = -ENOENT;
                if (!inode)
                        break;
@@ -888,22 +989,21 @@ return_reval:
                 * We bypassed the ordinary revalidation routines.
                 * We may need to check the cached dentry for staleness.
                 */
-               if (nd->dentry && nd->dentry->d_sb &&
-                   (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
+               if (nd->path.dentry && nd->path.dentry->d_sb &&
+                   (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
                        err = -ESTALE;
                        /* Note: we do not d_invalidate() */
-                       if (!nd->dentry->d_op->d_revalidate(nd->dentry, nd))
+                       if (!nd->path.dentry->d_op->d_revalidate(
+                                       nd->path.dentry, nd))
                                break;
                }
 return_base:
                return 0;
 out_dput:
-               dput(next.dentry);
-               if (nd->mnt != next.mnt)
-                       mntput(next.mnt);
+               path_put_conditional(&next, nd);
                break;
        }
-       path_release(nd);
+       path_put(&nd->path);
 return_err:
        return err;
 }
@@ -915,31 +1015,31 @@ return_err:
  * Retry the whole path once, forcing real lookup requests
  * instead of relying on the dcache.
  */
-int fastcall link_path_walk(const char *name, struct nameidata *nd)
+static int link_path_walk(const char *name, struct nameidata *nd)
 {
        struct nameidata save = *nd;
        int result;
 
        /* make sure the stuff we saved doesn't go away */
-       dget(save.dentry);
-       mntget(save.mnt);
+       dget(save.path.dentry);
+       mntget(save.path.mnt);
 
        result = __link_path_walk(name, nd);
        if (result == -ESTALE) {
                *nd = save;
-               dget(nd->dentry);
-               mntget(nd->mnt);
+               dget(nd->path.dentry);
+               mntget(nd->path.mnt);
                nd->flags |= LOOKUP_REVAL;
                result = __link_path_walk(name, nd);
        }
 
-       dput(save.dentry);
-       mntput(save.mnt);
+       dput(save.path.dentry);
+       mntput(save.path.mnt);
 
        return result;
 }
 
-int fastcall path_walk(const char * name, struct nameidata *nd)
+static int path_walk(const char *name, struct nameidata *nd)
 {
        current->total_link_count = 0;
        return link_path_walk(name, nd);
@@ -954,30 +1054,33 @@ static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
        if (path_walk(name, nd))
                return 0;               /* something went wrong... */
 
-       if (!nd->dentry->d_inode || S_ISDIR(nd->dentry->d_inode->i_mode)) {
-               struct dentry *old_dentry = nd->dentry;
-               struct vfsmount *old_mnt = nd->mnt;
+       if (!nd->path.dentry->d_inode ||
+           S_ISDIR(nd->path.dentry->d_inode->i_mode)) {
+               struct dentry *old_dentry = nd->path.dentry;
+               struct vfsmount *old_mnt = nd->path.mnt;
                struct qstr last = nd->last;
                int last_type = nd->last_type;
+               struct fs_struct *fs = current->fs;
+
                /*
-                * NAME was not found in alternate root or it's a directory.  Try to find
-                * it in the normal root:
+                * NAME was not found in alternate root or it's a directory.
+                * Try to find it in the normal root:
                 */
                nd->last_type = LAST_ROOT;
-               read_lock(&current->fs->lock);
-               nd->mnt = mntget(current->fs->rootmnt);
-               nd->dentry = dget(current->fs->root);
-               read_unlock(&current->fs->lock);
+               read_lock(&fs->lock);
+               nd->path.mnt = mntget(fs->rootmnt);
+               nd->path.dentry = dget(fs->root);
+               read_unlock(&fs->lock);
                if (path_walk(name, nd) == 0) {
-                       if (nd->dentry->d_inode) {
+                       if (nd->path.dentry->d_inode) {
                                dput(old_dentry);
                                mntput(old_mnt);
                                return 1;
                        }
-                       path_release(nd);
+                       path_put(&nd->path);
                }
-               nd->dentry = old_dentry;
-               nd->mnt = old_mnt;
+               nd->path.dentry = old_dentry;
+               nd->path.mnt = old_mnt;
                nd->last = last;
                nd->last_type = last_type;
        }
@@ -991,21 +1094,22 @@ void set_fs_altroot(void)
        struct vfsmount *mnt = NULL, *oldmnt;
        struct dentry *dentry = NULL, *olddentry;
        int err;
+       struct fs_struct *fs = current->fs;
 
        if (!emul)
                goto set_it;
        err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
        if (!err) {
-               mnt = nd.mnt;
-               dentry = nd.dentry;
+               mnt = nd.path.mnt;
+               dentry = nd.path.dentry;
        }
 set_it:
-       write_lock(&current->fs->lock);
-       oldmnt = current->fs->altrootmnt;
-       olddentry = current->fs->altroot;
-       current->fs->altrootmnt = mnt;
-       current->fs->altroot = dentry;
-       write_unlock(&current->fs->lock);
+       write_lock(&fs->lock);
+       oldmnt = fs->altrootmnt;
+       olddentry = fs->altroot;
+       fs->altrootmnt = mnt;
+       fs->altroot = dentry;
+       write_unlock(&fs->lock);
        if (olddentry) {
                dput(olddentry);
                mntput(oldmnt);
@@ -1013,56 +1117,186 @@ set_it:
 }
 
 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
-int fastcall path_lookup(const char *name, unsigned int flags, struct nameidata *nd)
+static int do_path_lookup(int dfd, const char *name,
+                               unsigned int flags, struct nameidata *nd)
 {
        int retval = 0;
+       int fput_needed;
+       struct file *file;
+       struct fs_struct *fs = current->fs;
 
        nd->last_type = LAST_ROOT; /* if there are only slashes... */
        nd->flags = flags;
        nd->depth = 0;
 
-       read_lock(&current->fs->lock);
        if (*name=='/') {
-               if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
-                       nd->mnt = mntget(current->fs->altrootmnt);
-                       nd->dentry = dget(current->fs->altroot);
-                       read_unlock(&current->fs->lock);
+               read_lock(&fs->lock);
+               if (fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
+                       nd->path.mnt = mntget(fs->altrootmnt);
+                       nd->path.dentry = dget(fs->altroot);
+                       read_unlock(&fs->lock);
                        if (__emul_lookup_dentry(name,nd))
                                goto out; /* found in altroot */
-                       read_lock(&current->fs->lock);
+                       read_lock(&fs->lock);
                }
-               nd->mnt = mntget(current->fs->rootmnt);
-               nd->dentry = dget(current->fs->root);
+               nd->path.mnt = mntget(fs->rootmnt);
+               nd->path.dentry = dget(fs->root);
+               read_unlock(&fs->lock);
+       } else if (dfd == AT_FDCWD) {
+               read_lock(&fs->lock);
+               nd->path.mnt = mntget(fs->pwdmnt);
+               nd->path.dentry = dget(fs->pwd);
+               read_unlock(&fs->lock);
        } else {
-               nd->mnt = mntget(current->fs->pwdmnt);
-               nd->dentry = dget(current->fs->pwd);
+               struct dentry *dentry;
+
+               file = fget_light(dfd, &fput_needed);
+               retval = -EBADF;
+               if (!file)
+                       goto out_fail;
+
+               dentry = file->f_path.dentry;
+
+               retval = -ENOTDIR;
+               if (!S_ISDIR(dentry->d_inode->i_mode))
+                       goto fput_fail;
+
+               retval = file_permission(file, MAY_EXEC);
+               if (retval)
+                       goto fput_fail;
+
+               nd->path.mnt = mntget(file->f_path.mnt);
+               nd->path.dentry = dget(dentry);
+
+               fput_light(file, fput_needed);
        }
-       read_unlock(&current->fs->lock);
-       current->total_link_count = 0;
-       retval = link_path_walk(name, nd);
+
+       retval = path_walk(name, nd);
 out:
-       if (unlikely(current->audit_context
-                    && nd && nd->dentry && nd->dentry->d_inode))
-               audit_inode(name, nd->dentry->d_inode);
+       if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
+                               nd->path.dentry->d_inode))
+               audit_inode(name, nd->path.dentry);
+out_fail:
        return retval;
+
+fput_fail:
+       fput_light(file, fput_needed);
+       goto out_fail;
 }
 
-/*
- * Restricted form of lookup. Doesn't follow links, single-component only,
- * needs parent already locked. Doesn't follow mounts.
- * SMP-safe.
+int path_lookup(const char *name, unsigned int flags,
+                       struct nameidata *nd)
+{
+       return do_path_lookup(AT_FDCWD, name, flags, nd);
+}
+
+/**
+ * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
+ * @dentry:  pointer to dentry of the base directory
+ * @mnt: pointer to vfs mount of the base directory
+ * @name: pointer to file name
+ * @flags: lookup flags
+ * @nd: pointer to nameidata
  */
-static struct dentry * __lookup_hash(struct qstr *name, struct dentry * base, struct nameidata *nd)
+int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
+                   const char *name, unsigned int flags,
+                   struct nameidata *nd)
 {
-       struct dentry * dentry;
+       int retval;
+
+       /* same as do_path_lookup */
+       nd->last_type = LAST_ROOT;
+       nd->flags = flags;
+       nd->depth = 0;
+
+       nd->path.mnt = mntget(mnt);
+       nd->path.dentry = dget(dentry);
+
+       retval = path_walk(name, nd);
+       if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry &&
+                               nd->path.dentry->d_inode))
+               audit_inode(name, nd->path.dentry);
+
+       return retval;
+
+}
+
+static int __path_lookup_intent_open(int dfd, const char *name,
+               unsigned int lookup_flags, struct nameidata *nd,
+               int open_flags, int create_mode)
+{
+       struct file *filp = get_empty_filp();
+       int err;
+
+       if (filp == NULL)
+               return -ENFILE;
+       nd->intent.open.file = filp;
+       nd->intent.open.flags = open_flags;
+       nd->intent.open.create_mode = create_mode;
+       err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
+       if (IS_ERR(nd->intent.open.file)) {
+               if (err == 0) {
+                       err = PTR_ERR(nd->intent.open.file);
+                       path_put(&nd->path);
+               }
+       } else if (err != 0)
+               release_open_intent(nd);
+       return err;
+}
+
+/**
+ * path_lookup_open - lookup a file path with open intent
+ * @dfd: the directory to use as base, or AT_FDCWD
+ * @name: pointer to file name
+ * @lookup_flags: lookup intent flags
+ * @nd: pointer to nameidata
+ * @open_flags: open intent flags
+ */
+int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
+               struct nameidata *nd, int open_flags)
+{
+       return __path_lookup_intent_open(dfd, name, lookup_flags, nd,
+                       open_flags, 0);
+}
+
+/**
+ * path_lookup_create - lookup a file path with open + create intent
+ * @dfd: the directory to use as base, or AT_FDCWD
+ * @name: pointer to file name
+ * @lookup_flags: lookup intent flags
+ * @nd: pointer to nameidata
+ * @open_flags: open intent flags
+ * @create_mode: create intent flags
+ */
+static int path_lookup_create(int dfd, const char *name,
+                             unsigned int lookup_flags, struct nameidata *nd,
+                             int open_flags, int create_mode)
+{
+       return __path_lookup_intent_open(dfd, name, lookup_flags|LOOKUP_CREATE,
+                       nd, open_flags, create_mode);
+}
+
+int __user_path_lookup_open(const char __user *name, unsigned int lookup_flags,
+               struct nameidata *nd, int open_flags)
+{
+       char *tmp = getname(name);
+       int err = PTR_ERR(tmp);
+
+       if (!IS_ERR(tmp)) {
+               err = __path_lookup_intent_open(AT_FDCWD, tmp, lookup_flags, nd, open_flags, 0);
+               putname(tmp);
+       }
+       return err;
+}
+
+static struct dentry *__lookup_hash(struct qstr *name,
+               struct dentry *base, struct nameidata *nd)
+{
+       struct dentry *dentry;
        struct inode *inode;
        int err;
 
        inode = base->d_inode;
-       err = permission(inode, MAY_EXEC, nd);
-       dentry = ERR_PTR(err);
-       if (err)
-               goto out;
 
        /*
         * See if the low-level filesystem might want
@@ -1091,60 +1325,109 @@ out:
        return dentry;
 }
 
-struct dentry * lookup_hash(struct qstr *name, struct dentry * base)
+/*
+ * Restricted form of lookup. Doesn't follow links, single-component only,
+ * needs parent already locked. Doesn't follow mounts.
+ * SMP-safe.
+ */
+static struct dentry *lookup_hash(struct nameidata *nd)
 {
-       return __lookup_hash(name, base, NULL);
+       int err;
+
+       err = permission(nd->path.dentry->d_inode, MAY_EXEC, nd);
+       if (err)
+               return ERR_PTR(err);
+       return __lookup_hash(&nd->last, nd->path.dentry, nd);
 }
 
-/* SMP-safe */
-struct dentry * lookup_one_len(const char * name, struct dentry * base, int len)
+static int __lookup_one_len(const char *name, struct qstr *this,
+               struct dentry *base, int len)
 {
        unsigned long hash;
-       struct qstr this;
        unsigned int c;
 
-       this.name = name;
-       this.len = len;
+       this->name = name;
+       this->len = len;
        if (!len)
-               goto access;
+               return -EACCES;
 
        hash = init_name_hash();
        while (len--) {
                c = *(const unsigned char *)name++;
                if (c == '/' || c == '\0')
-                       goto access;
+                       return -EACCES;
                hash = partial_name_hash(c, hash);
        }
-       this.hash = end_name_hash(hash);
+       this->hash = end_name_hash(hash);
+       return 0;
+}
 
-       return lookup_hash(&this, base);
-access:
-       return ERR_PTR(-EACCES);
+/**
+ * lookup_one_len:  filesystem helper to lookup single pathname component
+ * @name:      pathname component to lookup
+ * @base:      base directory to lookup from
+ * @len:       maximum length @len should be interpreted to
+ *
+ * Note that this routine is purely a helper for filesystem useage and should
+ * not be called by generic code.  Also note that by using this function to
+ * nameidata argument is passed to the filesystem methods and a filesystem
+ * using this helper needs to be prepared for that.
+ */
+struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
+{
+       int err;
+       struct qstr this;
+
+       err = __lookup_one_len(name, &this, base, len);
+       if (err)
+               return ERR_PTR(err);
+
+       err = permission(base->d_inode, MAY_EXEC, NULL);
+       if (err)
+               return ERR_PTR(err);
+       return __lookup_hash(&this, base, NULL);
 }
 
-/*
- *     namei()
+/**
+ * lookup_one_noperm - bad hack for sysfs
+ * @name:      pathname component to lookup
+ * @base:      base directory to lookup from
  *
- * is used by most simple commands to get the inode of a specified name.
- * Open, link etc use their own routines, but this is enough for things
- * like 'chmod' etc.
+ * This is a variant of lookup_one_len that doesn't perform any permission
+ * checks.   It's a horrible hack to work around the braindead sysfs
+ * architecture and should not be used anywhere else.
  *
- * namei exists in two versions: namei/lnamei. The only difference is
- * that namei follows links, while lnamei does not.
- * SMP-safe
+ * DON'T USE THIS FUNCTION EVER, thanks.
  */
-int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
+struct dentry *lookup_one_noperm(const char *name, struct dentry *base)
+{
+       int err;
+       struct qstr this;
+
+       err = __lookup_one_len(name, &this, base, strlen(name));
+       if (err)
+               return ERR_PTR(err);
+       return __lookup_hash(&this, base, NULL);
+}
+
+int __user_walk_fd(int dfd, const char __user *name, unsigned flags,
+                           struct nameidata *nd)
 {
        char *tmp = getname(name);
        int err = PTR_ERR(tmp);
 
        if (!IS_ERR(tmp)) {
-               err = path_lookup(tmp, flags, nd);
+               err = do_path_lookup(dfd, tmp, flags, nd);
                putname(tmp);
        }
        return err;
 }
 
+int __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
+{
+       return __user_walk_fd(AT_FDCWD, name, flags, nd);
+}
+
 /*
  * It's inline, so penalty for filesystems that don't use sticky bit is
  * minimal.
@@ -1179,7 +1462,7 @@ static inline int check_sticky(struct inode *dir, struct inode *inode)
  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
  *     nfs_async_unlink().
  */
-static inline int may_delete(struct inode *dir,struct dentry *victim,int isdir)
+static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
 {
        int error;
 
@@ -1187,6 +1470,7 @@ static inline int may_delete(struct inode *dir,struct dentry *victim,int isdir)
                return -ENOENT;
 
        BUG_ON(victim->d_parent->d_inode != dir);
+       audit_inode_child(victim->d_name.name, victim, dir);
 
        error = permission(dir,MAY_WRITE | MAY_EXEC, NULL);
        if (error)
@@ -1229,9 +1513,6 @@ static inline int may_create(struct inode *dir, struct dentry *child,
 }
 
 /* 
- * Special case: O_CREAT|O_EXCL implies O_NOFOLLOW for security
- * reasons.
- *
  * O_DIRECTORY translates into forcing a directory lookup.
  */
 static inline int lookup_flags(unsigned int f)
@@ -1241,9 +1522,6 @@ static inline int lookup_flags(unsigned int f)
        if (f & O_NOFOLLOW)
                retval &= ~LOOKUP_FOLLOW;
        
-       if ((f & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
-               retval &= ~LOOKUP_FOLLOW;
-       
        if (f & O_DIRECTORY)
                retval |= LOOKUP_DIRECTORY;
 
@@ -1258,39 +1536,39 @@ struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
        struct dentry *p;
 
        if (p1 == p2) {
-               down(&p1->d_inode->i_sem);
+               mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
                return NULL;
        }
 
-       down(&p1->d_inode->i_sb->s_vfs_rename_sem);
+       mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
 
        for (p = p1; p->d_parent != p; p = p->d_parent) {
                if (p->d_parent == p2) {
-                       down(&p2->d_inode->i_sem);
-                       down(&p1->d_inode->i_sem);
+                       mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
+                       mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
                        return p;
                }
        }
 
        for (p = p2; p->d_parent != p; p = p->d_parent) {
                if (p->d_parent == p1) {
-                       down(&p1->d_inode->i_sem);
-                       down(&p2->d_inode->i_sem);
+                       mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
+                       mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
                        return p;
                }
        }
 
-       down(&p1->d_inode->i_sem);
-       down(&p2->d_inode->i_sem);
+       mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
+       mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
        return NULL;
 }
 
 void unlock_rename(struct dentry *p1, struct dentry *p2)
 {
-       up(&p1->d_inode->i_sem);
+       mutex_unlock(&p1->d_inode->i_mutex);
        if (p1 != p2) {
-               up(&p2->d_inode->i_sem);
-               up(&p1->d_inode->i_sb->s_vfs_rename_sem);
+               mutex_unlock(&p2->d_inode->i_mutex);
+               mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
        }
 }
 
@@ -1311,16 +1589,14 @@ int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
                return error;
        DQUOT_INIT(dir);
        error = dir->i_op->create(dir, dentry, mode, nd);
-       if (!error) {
-               inode_dir_notify(dir, DN_CREATE);
-               security_inode_post_create(dir, dentry, mode);
-       }
+       if (!error)
+               fsnotify_create(dir, dentry);
        return error;
 }
 
 int may_open(struct nameidata *nd, int acc_mode, int flag)
 {
-       struct dentry *dentry = nd->dentry;
+       struct dentry *dentry = nd->path.dentry;
        struct inode *inode = dentry->d_inode;
        int error;
 
@@ -1330,13 +1606,9 @@ int may_open(struct nameidata *nd, int acc_mode, int flag)
        if (S_ISLNK(inode->i_mode))
                return -ELOOP;
        
-       if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
+       if (S_ISDIR(inode->i_mode) && (acc_mode & MAY_WRITE))
                return -EISDIR;
 
-       error = permission(inode, acc_mode, nd);
-       if (error)
-               return error;
-
        /*
         * FIFO's, sockets and device files are special: they don't
         * actually live on the filesystem itself, and as such you
@@ -1345,12 +1617,16 @@ int may_open(struct nameidata *nd, int acc_mode, int flag)
        if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
                flag &= ~O_TRUNC;
        } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
-               if (nd->mnt->mnt_flags & MNT_NODEV)
+               if (nd->path.mnt->mnt_flags & MNT_NODEV)
                        return -EACCES;
 
                flag &= ~O_TRUNC;
-       } else if (IS_RDONLY(inode) && (flag & FMODE_WRITE))
+       } else if (IS_RDONLY(inode) && (acc_mode & MAY_WRITE))
                return -EROFS;
+
+       error = vfs_permission(nd, acc_mode);
+       if (error)
+               return error;
        /*
         * An append-only file must be opened in append mode for writing.
         */
@@ -1363,7 +1639,7 @@ int may_open(struct nameidata *nd, int acc_mode, int flag)
 
        /* O_NOATIME can only be set by the owner or superuser */
        if (flag & O_NOATIME)
-               if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
+               if (!is_owner_or_cap(inode))
                        return -EPERM;
 
        /*
@@ -1384,8 +1660,10 @@ int may_open(struct nameidata *nd, int acc_mode, int flag)
                error = locks_verify_locked(inode);
                if (!error) {
                        DQUOT_INIT(inode);
-                       
-                       error = do_truncate(dentry, 0);
+
+                       error = do_truncate(dentry, 0,
+                                           ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
+                                           NULL);
                }
                put_write_access(inode);
                if (error)
@@ -1397,6 +1675,24 @@ int may_open(struct nameidata *nd, int acc_mode, int flag)
        return 0;
 }
 
+static int open_namei_create(struct nameidata *nd, struct path *path,
+                               int flag, int mode)
+{
+       int error;
+       struct dentry *dir = nd->path.dentry;
+
+       if (!IS_POSIXACL(dir->d_inode))
+               mode &= ~current->fs->umask;
+       error = vfs_create(dir->d_inode, path->dentry, mode, nd);
+       mutex_unlock(&dir->d_inode->i_mutex);
+       dput(nd->path.dentry);
+       nd->path.dentry = path->dentry;
+       if (error)
+               return error;
+       /* Don't check for write permission, don't truncate */
+       return may_open(nd, 0, flag & ~O_TRUNC);
+}
+
 /*
  *     open_namei()
  *
@@ -1411,29 +1707,31 @@ int may_open(struct nameidata *nd, int acc_mode, int flag)
  * for symlinks (where the permissions are checked later).
  * SMP-safe
  */
-int open_namei(const char * pathname, int flag, int mode, struct nameidata *nd)
+int open_namei(int dfd, const char *pathname, int flag,
+               int mode, struct nameidata *nd)
 {
-       int acc_mode, error = 0;
+       int acc_mode, error;
        struct path path;
        struct dentry *dir;
        int count = 0;
 
        acc_mode = ACC_MODE(flag);
 
+       /* O_TRUNC implies we need access checks for write permissions */
+       if (flag & O_TRUNC)
+               acc_mode |= MAY_WRITE;
+
        /* Allow the LSM permission hook to distinguish append 
           access from general write access. */
        if (flag & O_APPEND)
                acc_mode |= MAY_APPEND;
 
-       /* Fill in the open() intent data */
-       nd->intent.open.flags = flag;
-       nd->intent.open.create_mode = mode;
-
        /*
         * The simplest case - just a plain lookup.
         */
        if (!(flag & O_CREAT)) {
-               error = path_lookup(pathname, lookup_flags(flag)|LOOKUP_OPEN, nd);
+               error = path_lookup_open(dfd, pathname, lookup_flags(flag),
+                                        nd, flag);
                if (error)
                        return error;
                goto ok;
@@ -1442,7 +1740,7 @@ int open_namei(const char * pathname, int flag, int mode, struct nameidata *nd)
        /*
         * Create - we need to know the parent.
         */
-       error = path_lookup(pathname, LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE, nd);
+       error = path_lookup_create(dfd,pathname,LOOKUP_PARENT,nd,flag,mode);
        if (error)
                return error;
 
@@ -1455,39 +1753,38 @@ int open_namei(const char * pathname, int flag, int mode, struct nameidata *nd)
        if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
                goto exit;
 
-       dir = nd->dentry;
+       dir = nd->path.dentry;
        nd->flags &= ~LOOKUP_PARENT;
-       down(&dir->d_inode->i_sem);
-       path.dentry = __lookup_hash(&nd->last, nd->dentry, nd);
-       path.mnt = nd->mnt;
+       mutex_lock(&dir->d_inode->i_mutex);
+       path.dentry = lookup_hash(nd);
+       path.mnt = nd->path.mnt;
 
 do_last:
        error = PTR_ERR(path.dentry);
        if (IS_ERR(path.dentry)) {
-               up(&dir->d_inode->i_sem);
+               mutex_unlock(&dir->d_inode->i_mutex);
                goto exit;
        }
 
+       if (IS_ERR(nd->intent.open.file)) {
+               mutex_unlock(&dir->d_inode->i_mutex);
+               error = PTR_ERR(nd->intent.open.file);
+               goto exit_dput;
+       }
+
        /* Negative dentry, just create the file */
        if (!path.dentry->d_inode) {
-               if (!IS_POSIXACL(dir->d_inode))
-                       mode &= ~current->fs->umask;
-               error = vfs_create(dir->d_inode, path.dentry, mode, nd);
-               up(&dir->d_inode->i_sem);
-               dput(nd->dentry);
-               nd->dentry = path.dentry;
+               error = open_namei_create(nd, &path, flag, mode);
                if (error)
                        goto exit;
-               /* Don't check for write permission, don't truncate */
-               acc_mode = 0;
-               flag &= ~O_TRUNC;
-               goto ok;
+               return 0;
        }
 
        /*
         * It already exists.
         */
-       up(&dir->d_inode->i_sem);
+       mutex_unlock(&dir->d_inode->i_mutex);
+       audit_inode(pathname, path.dentry);
 
        error = -EEXIST;
        if (flag & O_EXCL)
@@ -1498,17 +1795,14 @@ do_last:
                if (flag & O_NOFOLLOW)
                        goto exit_dput;
        }
+
        error = -ENOENT;
        if (!path.dentry->d_inode)
                goto exit_dput;
        if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
                goto do_link;
 
-       dput(nd->dentry);
-       nd->dentry = path.dentry;
-       if (nd->mnt != path.mnt)
-               mntput(nd->mnt);
-       nd->mnt = path.mnt;
+       path_to_nameidata(&path, nd);
        error = -EISDIR;
        if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
                goto exit;
@@ -1519,11 +1813,11 @@ ok:
        return 0;
 
 exit_dput:
-       dput(path.dentry);
-       if (nd->mnt != path.mnt)
-               mntput(path.mnt);
+       path_put_conditional(&path, nd);
 exit:
-       path_release(nd);
+       if (!IS_ERR(nd->intent.open.file))
+               release_open_intent(nd);
+       path_put(&nd->path);
        return error;
 
 do_link:
@@ -1545,8 +1839,14 @@ do_link:
        if (error)
                goto exit_dput;
        error = __do_follow_link(&path, nd);
-       if (error)
+       if (error) {
+               /* Does someone understand code flow here? Or it is only
+                * me so stupid? Anathema to whoever designed this non-sense
+                * with "intent.open".
+                */
+               release_open_intent(nd);
                return error;
+       }
        nd->flags &= ~LOOKUP_PARENT;
        if (nd->last_type == LAST_BIND)
                goto ok;
@@ -1554,19 +1854,19 @@ do_link:
        if (nd->last_type != LAST_NORM)
                goto exit;
        if (nd->last.name[nd->last.len]) {
-               putname(nd->last.name);
+               __putname(nd->last.name);
                goto exit;
        }
        error = -ELOOP;
        if (count++==32) {
-               putname(nd->last.name);
+               __putname(nd->last.name);
                goto exit;
        }
-       dir = nd->dentry;
-       down(&dir->d_inode->i_sem);
-       path.dentry = __lookup_hash(&nd->last, nd->dentry, nd);
-       path.mnt = nd->mnt;
-       putname(nd->last.name);
+       dir = nd->path.dentry;
+       mutex_lock(&dir->d_inode->i_mutex);
+       path.dentry = lookup_hash(nd);
+       path.mnt = nd->path.mnt;
+       __putname(nd->last.name);
        goto do_last;
 }
 
@@ -1577,19 +1877,37 @@ do_link:
  *
  * Simple function to lookup and return a dentry and create it
  * if it doesn't exist.  Is SMP-safe.
+ *
+ * Returns with nd->path.dentry->d_inode->i_mutex locked.
  */
 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
 {
-       struct dentry *dentry;
+       struct dentry *dentry = ERR_PTR(-EEXIST);
 
-       down(&nd->dentry->d_inode->i_sem);
-       dentry = ERR_PTR(-EEXIST);
+       mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
+       /*
+        * Yucky last component or no last component at all?
+        * (foo/., foo/.., /////)
+        */
        if (nd->last_type != LAST_NORM)
                goto fail;
        nd->flags &= ~LOOKUP_PARENT;
-       dentry = lookup_hash(&nd->last, nd->dentry);
+       nd->flags |= LOOKUP_CREATE;
+       nd->intent.open.flags = O_EXCL;
+
+       /*
+        * Do the final lookup.
+        */
+       dentry = lookup_hash(nd);
        if (IS_ERR(dentry))
                goto fail;
+
+       /*
+        * Special case - lookup gave negative, but... we had foo/bar/
+        * From the vfs_mknod() POV we just have a negative dentry -
+        * all is fine. Let's be bastards - you had / on the end, you've
+        * been asking for (non-existent) directory. -ENOENT for you.
+        */
        if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
                goto enoent;
        return dentry;
@@ -1620,14 +1938,13 @@ int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
 
        DQUOT_INIT(dir);
        error = dir->i_op->mknod(dir, dentry, mode, dev);
-       if (!error) {
-               inode_dir_notify(dir, DN_CREATE);
-               security_inode_post_mknod(dir, dentry, mode, dev);
-       }
+       if (!error)
+               fsnotify_create(dir, dentry);
        return error;
 }
 
-asmlinkage long sys_mknod(const char __user * filename, int mode, unsigned dev)
+asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
+                               unsigned dev)
 {
        int error = 0;
        char * tmp;
@@ -1640,25 +1957,25 @@ asmlinkage long sys_mknod(const char __user * filename, int mode, unsigned dev)
        if (IS_ERR(tmp))
                return PTR_ERR(tmp);
 
-       error = path_lookup(tmp, LOOKUP_PARENT, &nd);
+       error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
        if (error)
                goto out;
        dentry = lookup_create(&nd, 0);
        error = PTR_ERR(dentry);
 
-       if (!IS_POSIXACL(nd.dentry->d_inode))
+       if (!IS_POSIXACL(nd.path.dentry->d_inode))
                mode &= ~current->fs->umask;
        if (!IS_ERR(dentry)) {
                switch (mode & S_IFMT) {
                case 0: case S_IFREG:
-                       error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
+                       error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd);
                        break;
                case S_IFCHR: case S_IFBLK:
-                       error = vfs_mknod(nd.dentry->d_inode,dentry,mode,
+                       error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,
                                        new_decode_dev(dev));
                        break;
                case S_IFIFO: case S_IFSOCK:
-                       error = vfs_mknod(nd.dentry->d_inode,dentry,mode,0);
+                       error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0);
                        break;
                case S_IFDIR:
                        error = -EPERM;
@@ -1668,14 +1985,19 @@ asmlinkage long sys_mknod(const char __user * filename, int mode, unsigned dev)
                }
                dput(dentry);
        }
-       up(&nd.dentry->d_inode->i_sem);
-       path_release(&nd);
+       mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
+       path_put(&nd.path);
 out:
        putname(tmp);
 
        return error;
 }
 
+asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev)
+{
+       return sys_mknodat(AT_FDCWD, filename, mode, dev);
+}
+
 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 {
        int error = may_create(dir, dentry, NULL);
@@ -1693,44 +2015,49 @@ int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 
        DQUOT_INIT(dir);
        error = dir->i_op->mkdir(dir, dentry, mode);
-       if (!error) {
-               inode_dir_notify(dir, DN_CREATE);
-               security_inode_post_mkdir(dir,dentry, mode);
-       }
+       if (!error)
+               fsnotify_mkdir(dir, dentry);
        return error;
 }
 
-asmlinkage long sys_mkdir(const char __user * pathname, int mode)
+asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
 {
        int error = 0;
        char * tmp;
+       struct dentry *dentry;
+       struct nameidata nd;
 
        tmp = getname(pathname);
        error = PTR_ERR(tmp);
-       if (!IS_ERR(tmp)) {
-               struct dentry *dentry;
-               struct nameidata nd;
+       if (IS_ERR(tmp))
+               goto out_err;
 
-               error = path_lookup(tmp, LOOKUP_PARENT, &nd);
-               if (error)
-                       goto out;
-               dentry = lookup_create(&nd, 1);
-               error = PTR_ERR(dentry);
-               if (!IS_ERR(dentry)) {
-                       if (!IS_POSIXACL(nd.dentry->d_inode))
-                               mode &= ~current->fs->umask;
-                       error = vfs_mkdir(nd.dentry->d_inode, dentry, mode);
-                       dput(dentry);
-               }
-               up(&nd.dentry->d_inode->i_sem);
-               path_release(&nd);
-out:
-               putname(tmp);
-       }
+       error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
+       if (error)
+               goto out;
+       dentry = lookup_create(&nd, 1);
+       error = PTR_ERR(dentry);
+       if (IS_ERR(dentry))
+               goto out_unlock;
 
+       if (!IS_POSIXACL(nd.path.dentry->d_inode))
+               mode &= ~current->fs->umask;
+       error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
+       dput(dentry);
+out_unlock:
+       mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
+       path_put(&nd.path);
+out:
+       putname(tmp);
+out_err:
        return error;
 }
 
+asmlinkage long sys_mkdir(const char __user *pathname, int mode)
+{
+       return sys_mkdirat(AT_FDCWD, pathname, mode);
+}
+
 /*
  * We try to drop the dentry early: we should have
  * a usage count of 2 if we're the only user of this
@@ -1749,8 +2076,7 @@ out:
 void dentry_unhash(struct dentry *dentry)
 {
        dget(dentry);
-       if (atomic_read(&dentry->d_count))
-               shrink_dcache_parent(dentry);
+       shrink_dcache_parent(dentry);
        spin_lock(&dcache_lock);
        spin_lock(&dentry->d_lock);
        if (atomic_read(&dentry->d_count) == 2)
@@ -1771,7 +2097,7 @@ int vfs_rmdir(struct inode *dir, struct dentry *dentry)
 
        DQUOT_INIT(dir);
 
-       down(&dentry->d_inode->i_sem);
+       mutex_lock(&dentry->d_inode->i_mutex);
        dentry_unhash(dentry);
        if (d_mountpoint(dentry))
                error = -EBUSY;
@@ -1783,9 +2109,8 @@ int vfs_rmdir(struct inode *dir, struct dentry *dentry)
                                dentry->d_inode->i_flags |= S_DEAD;
                }
        }
-       up(&dentry->d_inode->i_sem);
+       mutex_unlock(&dentry->d_inode->i_mutex);
        if (!error) {
-               inode_dir_notify(dir, DN_DELETE);
                d_delete(dentry);
        }
        dput(dentry);
@@ -1793,7 +2118,7 @@ int vfs_rmdir(struct inode *dir, struct dentry *dentry)
        return error;
 }
 
-asmlinkage long sys_rmdir(const char __user * pathname)
+static long do_rmdir(int dfd, const char __user *pathname)
 {
        int error = 0;
        char * name;
@@ -1804,7 +2129,7 @@ asmlinkage long sys_rmdir(const char __user * pathname)
        if(IS_ERR(name))
                return PTR_ERR(name);
 
-       error = path_lookup(name, LOOKUP_PARENT, &nd);
+       error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
        if (error)
                goto exit;
 
@@ -1819,21 +2144,27 @@ asmlinkage long sys_rmdir(const char __user * pathname)
                        error = -EBUSY;
                        goto exit1;
        }
-       down(&nd.dentry->d_inode->i_sem);
-       dentry = lookup_hash(&nd.last, nd.dentry);
+       mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
+       dentry = lookup_hash(&nd);
        error = PTR_ERR(dentry);
-       if (!IS_ERR(dentry)) {
-               error = vfs_rmdir(nd.dentry->d_inode, dentry);
-               dput(dentry);
-       }
-       up(&nd.dentry->d_inode->i_sem);
+       if (IS_ERR(dentry))
+               goto exit2;
+       error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
+       dput(dentry);
+exit2:
+       mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
 exit1:
-       path_release(&nd);
+       path_put(&nd.path);
 exit:
        putname(name);
        return error;
 }
 
+asmlinkage long sys_rmdir(const char __user *pathname)
+{
+       return do_rmdir(AT_FDCWD, pathname);
+}
+
 int vfs_unlink(struct inode *dir, struct dentry *dentry)
 {
        int error = may_delete(dir, dentry, 0);
@@ -1846,7 +2177,7 @@ int vfs_unlink(struct inode *dir, struct dentry *dentry)
 
        DQUOT_INIT(dir);
 
-       down(&dentry->d_inode->i_sem);
+       mutex_lock(&dentry->d_inode->i_mutex);
        if (d_mountpoint(dentry))
                error = -EBUSY;
        else {
@@ -1854,23 +2185,24 @@ int vfs_unlink(struct inode *dir, struct dentry *dentry)
                if (!error)
                        error = dir->i_op->unlink(dir, dentry);
        }
-       up(&dentry->d_inode->i_sem);
+       mutex_unlock(&dentry->d_inode->i_mutex);
 
        /* We don't d_delete() NFS sillyrenamed files--they still exist. */
        if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
+               fsnotify_link_count(dentry->d_inode);
                d_delete(dentry);
-               inode_dir_notify(dir, DN_DELETE);
        }
+
        return error;
 }
 
 /*
  * Make sure that the actual truncation of the file will occur outside its
- * directory's i_sem.  Truncate can take a long time if there is a lot of
+ * directory's i_mutex.  Truncate can take a long time if there is a lot of
  * writeout happening, and we don't want to prevent access to the directory
  * while waiting on the I/O.
  */
-asmlinkage long sys_unlink(const char __user * pathname)
+static long do_unlinkat(int dfd, const char __user *pathname)
 {
        int error = 0;
        char * name;
@@ -1882,14 +2214,14 @@ asmlinkage long sys_unlink(const char __user * pathname)
        if(IS_ERR(name))
                return PTR_ERR(name);
 
-       error = path_lookup(name, LOOKUP_PARENT, &nd);
+       error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
        if (error)
                goto exit;
        error = -EISDIR;
        if (nd.last_type != LAST_NORM)
                goto exit1;
-       down(&nd.dentry->d_inode->i_sem);
-       dentry = lookup_hash(&nd.last, nd.dentry);
+       mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
+       dentry = lookup_hash(&nd);
        error = PTR_ERR(dentry);
        if (!IS_ERR(dentry)) {
                /* Why not before? Because we want correct error value */
@@ -1898,15 +2230,15 @@ asmlinkage long sys_unlink(const char __user * pathname)
                inode = dentry->d_inode;
                if (inode)
                        atomic_inc(&inode->i_count);
-               error = vfs_unlink(nd.dentry->d_inode, dentry);
+               error = vfs_unlink(nd.path.dentry->d_inode, dentry);
        exit2:
                dput(dentry);
        }
-       up(&nd.dentry->d_inode->i_sem);
+       mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
        if (inode)
                iput(inode);    /* truncate the inode here */
 exit1:
-       path_release(&nd);
+       path_put(&nd.path);
 exit:
        putname(name);
        return error;
@@ -1917,6 +2249,22 @@ slashes:
        goto exit2;
 }
 
+asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag)
+{
+       if ((flag & ~AT_REMOVEDIR) != 0)
+               return -EINVAL;
+
+       if (flag & AT_REMOVEDIR)
+               return do_rmdir(dfd, pathname);
+
+       return do_unlinkat(dfd, pathname);
+}
+
+asmlinkage long sys_unlink(const char __user *pathname)
+{
+       return do_unlinkat(AT_FDCWD, pathname);
+}
+
 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname, int mode)
 {
        int error = may_create(dir, dentry, NULL);
@@ -1933,46 +2281,53 @@ int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname, i
 
        DQUOT_INIT(dir);
        error = dir->i_op->symlink(dir, dentry, oldname);
-       if (!error) {
-               inode_dir_notify(dir, DN_CREATE);
-               security_inode_post_symlink(dir, dentry, oldname);
-       }
+       if (!error)
+               fsnotify_create(dir, dentry);
        return error;
 }
 
-asmlinkage long sys_symlink(const char __user * oldname, const char __user * newname)
+asmlinkage long sys_symlinkat(const char __user *oldname,
+                             int newdfd, const char __user *newname)
 {
        int error = 0;
        char * from;
        char * to;
+       struct dentry *dentry;
+       struct nameidata nd;
 
        from = getname(oldname);
        if(IS_ERR(from))
                return PTR_ERR(from);
        to = getname(newname);
        error = PTR_ERR(to);
-       if (!IS_ERR(to)) {
-               struct dentry *dentry;
-               struct nameidata nd;
+       if (IS_ERR(to))
+               goto out_putname;
 
-               error = path_lookup(to, LOOKUP_PARENT, &nd);
-               if (error)
-                       goto out;
-               dentry = lookup_create(&nd, 0);
-               error = PTR_ERR(dentry);
-               if (!IS_ERR(dentry)) {
-                       error = vfs_symlink(nd.dentry->d_inode, dentry, from, S_IALLUGO);
-                       dput(dentry);
-               }
-               up(&nd.dentry->d_inode->i_sem);
-               path_release(&nd);
+       error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
+       if (error)
+               goto out;
+       dentry = lookup_create(&nd, 0);
+       error = PTR_ERR(dentry);
+       if (IS_ERR(dentry))
+               goto out_unlock;
+
+       error = vfs_symlink(nd.path.dentry->d_inode, dentry, from, S_IALLUGO);
+       dput(dentry);
+out_unlock:
+       mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
+       path_put(&nd.path);
 out:
-               putname(to);
-       }
+       putname(to);
+out_putname:
        putname(from);
        return error;
 }
 
+asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname)
+{
+       return sys_symlinkat(oldname, AT_FDCWD, newname);
+}
+
 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
 {
        struct inode *inode = old_dentry->d_inode;
@@ -2002,14 +2357,12 @@ int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_de
        if (error)
                return error;
 
-       down(&old_dentry->d_inode->i_sem);
+       mutex_lock(&old_dentry->d_inode->i_mutex);
        DQUOT_INIT(dir);
        error = dir->i_op->link(old_dentry, dir, new_dentry);
-       up(&old_dentry->d_inode->i_sem);
-       if (!error) {
-               inode_dir_notify(dir, DN_CREATE);
-               security_inode_post_link(old_dentry, dir, new_dentry);
-       }
+       mutex_unlock(&old_dentry->d_inode->i_mutex);
+       if (!error)
+               fsnotify_link(dir, old_dentry->d_inode, new_dentry);
        return error;
 }
 
@@ -2022,43 +2375,56 @@ int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_de
  * with linux 2.0, and to avoid hard-linking to directories
  * and other special files.  --ADM
  */
-asmlinkage long sys_link(const char __user * oldname, const char __user * newname)
+asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
+                          int newdfd, const char __user *newname,
+                          int flags)
 {
        struct dentry *new_dentry;
        struct nameidata nd, old_nd;
        int error;
        char * to;
 
+       if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
+               return -EINVAL;
+
        to = getname(newname);
        if (IS_ERR(to))
                return PTR_ERR(to);
 
-       error = __user_walk(oldname, 0, &old_nd);
+       error = __user_walk_fd(olddfd, oldname,
+                              flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
+                              &old_nd);
        if (error)
                goto exit;
-       error = path_lookup(to, LOOKUP_PARENT, &nd);
+       error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
        if (error)
                goto out;
        error = -EXDEV;
-       if (old_nd.mnt != nd.mnt)
+       if (old_nd.path.mnt != nd.path.mnt)
                goto out_release;
        new_dentry = lookup_create(&nd, 0);
        error = PTR_ERR(new_dentry);
-       if (!IS_ERR(new_dentry)) {
-               error = vfs_link(old_nd.dentry, nd.dentry->d_inode, new_dentry);
-               dput(new_dentry);
-       }
-       up(&nd.dentry->d_inode->i_sem);
+       if (IS_ERR(new_dentry))
+               goto out_unlock;
+       error = vfs_link(old_nd.path.dentry, nd.path.dentry->d_inode, new_dentry);
+       dput(new_dentry);
+out_unlock:
+       mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
 out_release:
-       path_release(&nd);
+       path_put(&nd.path);
 out:
-       path_release(&old_nd);
+       path_put(&old_nd.path);
 exit:
        putname(to);
 
        return error;
 }
 
+asmlinkage long sys_link(const char __user *oldname, const char __user *newname)
+{
+       return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
+}
+
 /*
  * The worst of all namespace operations - renaming directory. "Perverted"
  * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
@@ -2066,17 +2432,17 @@ exit:
  *     a) we can get into loop creation. Check is done in is_subdir().
  *     b) race potential - two innocent renames can create a loop together.
  *        That's where 4.4 screws up. Current fix: serialization on
- *        sb->s_vfs_rename_sem. We might be more accurate, but that's another
+ *        sb->s_vfs_rename_mutex. We might be more accurate, but that's another
  *        story.
  *     c) we have to lock _three_ objects - parents and victim (if it exists).
- *        And that - after we got ->i_sem on parents (until then we don't know
+ *        And that - after we got ->i_mutex on parents (until then we don't know
  *        whether the target exists).  Solution: try to be smart with locking
  *        order for inodes.  We rely on the fact that tree topology may change
- *        only under ->s_vfs_rename_sem _and_ that parent of the object we
+ *        only under ->s_vfs_rename_mutex _and_ that parent of the object we
  *        move will be locked.  Thus we can rank directories by the tree
  *        (ancestors first) and rank all non-directories after them.
  *        That works since everybody except rename does "lock parent, lookup,
- *        lock child" and rename is under ->s_vfs_rename_sem.
+ *        lock child" and rename is under ->s_vfs_rename_mutex.
  *        HOWEVER, it relies on the assumption that any object with ->lookup()
  *        has no more than 1 dentry.  If "hybrid" objects will ever appear,
  *        we'd better make sure that there's no link(2) for them.
@@ -2086,9 +2452,9 @@ exit:
  *        stuff into VFS), but the former is not going away. Solution: the same
  *        trick as in rmdir().
  *     e) conversion from fhandle to dentry may come in the wrong moment - when
- *        we are removing the target. Solution: we will have to grab ->i_sem
+ *        we are removing the target. Solution: we will have to grab ->i_mutex
  *        in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
- *        ->i_sem on parents, which works but leads to some truely excessive
+ *        ->i_mutex on parents, which works but leads to some truely excessive
  *        locking].
  */
 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
@@ -2113,7 +2479,7 @@ static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
 
        target = new_dentry->d_inode;
        if (target) {
-               down(&target->i_sem);
+               mutex_lock(&target->i_mutex);
                dentry_unhash(new_dentry);
        }
        if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
@@ -2123,16 +2489,14 @@ static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
        if (target) {
                if (!error)
                        target->i_flags |= S_DEAD;
-               up(&target->i_sem);
+               mutex_unlock(&target->i_mutex);
                if (d_unhashed(new_dentry))
                        d_rehash(new_dentry);
                dput(new_dentry);
        }
-       if (!error) {
-               d_move(old_dentry,new_dentry);
-               security_inode_post_rename(old_dir, old_dentry,
-                                          new_dir, new_dentry);
-       }
+       if (!error)
+               if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
+                       d_move(old_dentry,new_dentry);
        return error;
 }
 
@@ -2149,19 +2513,17 @@ static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
        dget(new_dentry);
        target = new_dentry->d_inode;
        if (target)
-               down(&target->i_sem);
+               mutex_lock(&target->i_mutex);
        if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
                error = -EBUSY;
        else
                error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
        if (!error) {
-               /* The following d_move() should become unconditional */
-               if (!(old_dir->i_sb->s_type->fs_flags & FS_ODD_RENAME))
+               if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
                        d_move(old_dentry, new_dentry);
-               security_inode_post_rename(old_dir, old_dentry, new_dir, new_dentry);
        }
        if (target)
-               up(&target->i_sem);
+               mutex_unlock(&target->i_mutex);
        dput(new_dentry);
        return error;
 }
@@ -2171,6 +2533,7 @@ int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
 {
        int error;
        int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
+       const char *old_name;
 
        if (old_dentry->d_inode == new_dentry->d_inode)
                return 0;
@@ -2192,22 +2555,24 @@ int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
        DQUOT_INIT(old_dir);
        DQUOT_INIT(new_dir);
 
+       old_name = fsnotify_oldname_init(old_dentry->d_name.name);
+
        if (is_dir)
                error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
        else
                error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
        if (!error) {
-               if (old_dir == new_dir)
-                       inode_dir_notify(old_dir, DN_RENAME);
-               else {
-                       inode_dir_notify(old_dir, DN_DELETE);
-                       inode_dir_notify(new_dir, DN_CREATE);
-               }
+               const char *new_name = old_dentry->d_name.name;
+               fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
+                             new_dentry->d_inode, old_dentry);
        }
+       fsnotify_oldname_free(old_name);
+
        return error;
 }
 
-static inline int do_rename(const char * oldname, const char * newname)
+static int do_rename(int olddfd, const char *oldname,
+                       int newdfd, const char *newname)
 {
        int error = 0;
        struct dentry * old_dir, * new_dir;
@@ -2215,30 +2580,30 @@ static inline int do_rename(const char * oldname, const char * newname)
        struct dentry * trap;
        struct nameidata oldnd, newnd;
 
-       error = path_lookup(oldname, LOOKUP_PARENT, &oldnd);
+       error = do_path_lookup(olddfd, oldname, LOOKUP_PARENT, &oldnd);
        if (error)
                goto exit;
 
-       error = path_lookup(newname, LOOKUP_PARENT, &newnd);
+       error = do_path_lookup(newdfd, newname, LOOKUP_PARENT, &newnd);
        if (error)
                goto exit1;
 
        error = -EXDEV;
-       if (oldnd.mnt != newnd.mnt)
+       if (oldnd.path.mnt != newnd.path.mnt)
                goto exit2;
 
-       old_dir = oldnd.dentry;
+       old_dir = oldnd.path.dentry;
        error = -EBUSY;
        if (oldnd.last_type != LAST_NORM)
                goto exit2;
 
-       new_dir = newnd.dentry;
+       new_dir = newnd.path.dentry;
        if (newnd.last_type != LAST_NORM)
                goto exit2;
 
        trap = lock_rename(new_dir, old_dir);
 
-       old_dentry = lookup_hash(&oldnd.last, old_dir);
+       old_dentry = lookup_hash(&oldnd);
        error = PTR_ERR(old_dentry);
        if (IS_ERR(old_dentry))
                goto exit3;
@@ -2258,7 +2623,7 @@ static inline int do_rename(const char * oldname, const char * newname)
        error = -EINVAL;
        if (old_dentry == trap)
                goto exit4;
-       new_dentry = lookup_hash(&newnd.last, new_dir);
+       new_dentry = lookup_hash(&newnd);
        error = PTR_ERR(new_dentry);
        if (IS_ERR(new_dentry))
                goto exit4;
@@ -2276,14 +2641,15 @@ exit4:
 exit3:
        unlock_rename(new_dir, old_dir);
 exit2:
-       path_release(&newnd);
+       path_put(&newnd.path);
 exit1:
-       path_release(&oldnd);
+       path_put(&oldnd.path);
 exit:
        return error;
 }
 
-asmlinkage long sys_rename(const char __user * oldname, const char __user * newname)
+asmlinkage long sys_renameat(int olddfd, const char __user *oldname,
+                            int newdfd, const char __user *newname)
 {
        int error;
        char * from;
@@ -2295,13 +2661,18 @@ asmlinkage long sys_rename(const char __user * oldname, const char __user * newn
        to = getname(newname);
        error = PTR_ERR(to);
        if (!IS_ERR(to)) {
-               error = do_rename(from,to);
+               error = do_rename(olddfd, from, newdfd, to);
                putname(to);
        }
        putname(from);
        return error;
 }
 
+asmlinkage long sys_rename(const char __user *oldname, const char __user *newname)
+{
+       return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
+}
+
 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
 {
        int len;
@@ -2327,15 +2698,17 @@ out:
 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
 {
        struct nameidata nd;
-       int res;
+       void *cookie;
+
        nd.depth = 0;
-       res = dentry->d_inode->i_op->follow_link(dentry, &nd);
-       if (!res) {
-               res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
+       cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
+       if (!IS_ERR(cookie)) {
+               int res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
                if (dentry->d_inode->i_op->put_link)
-                       dentry->d_inode->i_op->put_link(dentry, &nd);
+                       dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
+               cookie = ERR_PTR(res);
        }
-       return res;
+       return PTR_ERR(cookie);
 }
 
 int vfs_follow_link(struct nameidata *nd, const char *link)
@@ -2348,22 +2721,11 @@ static char *page_getlink(struct dentry * dentry, struct page **ppage)
 {
        struct page * page;
        struct address_space *mapping = dentry->d_inode->i_mapping;
-       page = read_cache_page(mapping, 0, (filler_t *)mapping->a_ops->readpage,
-                               NULL);
+       page = read_mapping_page(mapping, 0, NULL);
        if (IS_ERR(page))
-               goto sync_fail;
-       wait_on_page_locked(page);
-       if (!PageUptodate(page))
-               goto async_fail;
+               return (char*)page;
        *ppage = page;
        return kmap(page);
-
-async_fail:
-       page_cache_release(page);
-       return ERR_PTR(-EIO);
-
-sync_fail:
-       return (char*)page;
 }
 
 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
@@ -2378,90 +2740,86 @@ int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
        return res;
 }
 
-int page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
+void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
 {
-       struct page *page;
+       struct page *page = NULL;
        nd_set_link(nd, page_getlink(dentry, &page));
-       return 0;
+       return page;
 }
 
-void page_put_link(struct dentry *dentry, struct nameidata *nd)
+void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
 {
-       if (!IS_ERR(nd_get_link(nd))) {
-               struct page *page;
-               page = find_get_page(dentry->d_inode->i_mapping, 0);
-               if (!page)
-                       BUG();
+       struct page *page = cookie;
+
+       if (page) {
                kunmap(page);
                page_cache_release(page);
-               page_cache_release(page);
        }
 }
 
-int page_symlink(struct inode *inode, const char *symname, int len)
+int __page_symlink(struct inode *inode, const char *symname, int len,
+               gfp_t gfp_mask)
 {
        struct address_space *mapping = inode->i_mapping;
-       struct page *page = grab_cache_page(mapping, 0);
-       int err = -ENOMEM;
+       struct page *page;
+       void *fsdata;
+       int err;
        char *kaddr;
 
-       if (!page)
-               goto fail;
-       err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
+retry:
+       err = pagecache_write_begin(NULL, mapping, 0, len-1,
+                               AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
        if (err)
-               goto fail_map;
+               goto fail;
+
        kaddr = kmap_atomic(page, KM_USER0);
        memcpy(kaddr, symname, len-1);
        kunmap_atomic(kaddr, KM_USER0);
-       mapping->a_ops->commit_write(NULL, page, 0, len-1);
-       /*
-        * Notice that we are _not_ going to block here - end of page is
-        * unmapped, so this will only try to map the rest of page, see
-        * that it is unmapped (typically even will not look into inode -
-        * ->i_size will be enough for everything) and zero it out.
-        * OTOH it's obviously correct and should make the page up-to-date.
-        */
-       if (!PageUptodate(page)) {
-               err = mapping->a_ops->readpage(NULL, page);
-               wait_on_page_locked(page);
-       } else {
-               unlock_page(page);
-       }
-       page_cache_release(page);
+
+       err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
+                                                       page, fsdata);
        if (err < 0)
                goto fail;
+       if (err < len-1)
+               goto retry;
+
        mark_inode_dirty(inode);
        return 0;
-fail_map:
-       unlock_page(page);
-       page_cache_release(page);
 fail:
        return err;
 }
 
-struct inode_operations page_symlink_inode_operations = {
+int page_symlink(struct inode *inode, const char *symname, int len)
+{
+       return __page_symlink(inode, symname, len,
+                       mapping_gfp_mask(inode->i_mapping));
+}
+
+const struct inode_operations page_symlink_inode_operations = {
        .readlink       = generic_readlink,
        .follow_link    = page_follow_link_light,
        .put_link       = page_put_link,
 };
 
 EXPORT_SYMBOL(__user_walk);
+EXPORT_SYMBOL(__user_walk_fd);
 EXPORT_SYMBOL(follow_down);
 EXPORT_SYMBOL(follow_up);
 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
 EXPORT_SYMBOL(getname);
 EXPORT_SYMBOL(lock_rename);
-EXPORT_SYMBOL(lookup_hash);
 EXPORT_SYMBOL(lookup_one_len);
 EXPORT_SYMBOL(page_follow_link_light);
 EXPORT_SYMBOL(page_put_link);
 EXPORT_SYMBOL(page_readlink);
+EXPORT_SYMBOL(__page_symlink);
 EXPORT_SYMBOL(page_symlink);
 EXPORT_SYMBOL(page_symlink_inode_operations);
 EXPORT_SYMBOL(path_lookup);
-EXPORT_SYMBOL(path_release);
-EXPORT_SYMBOL(path_walk);
+EXPORT_SYMBOL(vfs_path_lookup);
 EXPORT_SYMBOL(permission);
+EXPORT_SYMBOL(vfs_permission);
+EXPORT_SYMBOL(file_permission);
 EXPORT_SYMBOL(unlock_rename);
 EXPORT_SYMBOL(vfs_create);
 EXPORT_SYMBOL(vfs_follow_link);