Merge to Fedora kernel-2.6.17-1.2142_FC4 patched with stable patch-2.6.17.13-vs2...
[linux-2.6.git] / fs / namei.c
index 45b138d..1819223 100644 (file)
 #include <linux/namei.h>
 #include <linux/quotaops.h>
 #include <linux/pagemap.h>
-#include <linux/dnotify.h>
+#include <linux/fsnotify.h>
 #include <linux/smp_lock.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 <linux/namei.h>
+#include <linux/proc_fs.h>
+#include <linux/vserver/inode.h>
+#include <linux/vserver/debug.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...
  */
 
  * 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;
 
-       if ((unsigned long) filename >= TASK_SIZE) {
-               if (!segment_eq(get_fs(), KERNEL_DS))
+       if (!segment_eq(get_fs(), KERNEL_DS)) {
+               if ((unsigned long) filename >= TASK_SIZE)
                        return -EFAULT;
-       } else if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
-               len = TASK_SIZE - (unsigned long) filename;
+               if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
+                       len = TASK_SIZE - (unsigned long) filename;
+       }
 
-       retval = strncpy_from_user((char *)page, filename, len);
+       retval = strncpy_from_user(page, filename, len);
        if (retval > 0) {
                if (retval < len)
                        return 0;
@@ -146,45 +155,52 @@ char * getname(const char __user * filename)
                        result = ERR_PTR(retval);
                }
        }
-       if (unlikely(current->audit_context) && !IS_ERR(result) && result)
-               audit_getname(result);
+       audit_getname(result);
        return result;
 }
 
-/*
- *     vfs_permission()
+#ifdef CONFIG_AUDITSYSCALL
+void putname(const char *name)
+{
+       if (unlikely(current->audit_context))
+               audit_putname(name);
+       else
+               __putname(name);
+}
+EXPORT_SYMBOL(putname);
+#endif
+
+
+/**
+ * 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
  *
- * is used to check for read/write/execute permissions on a file.
+ * 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..
  */
-int vfs_permission(struct inode * inode, int mask)
+int generic_permission(struct inode *inode, int mask,
+               int (*check_acl)(struct inode *inode, int mask))
 {
        umode_t                 mode = inode->i_mode;
 
-       if (IS_BARRIER(inode) && !vx_check(0, VX_ADMIN|VX_WATCH))
-               return -EACCES;
-
-       if (mask & MAY_WRITE) {
-               /*
-                * Nobody gets write access to a read-only fs.
-                */
-               if (IS_RDONLY(inode) &&
-                   (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
-                       return -EROFS;
-
-               /*
-                * Nobody gets write access to an immutable file.
-                */
-               if (IS_IMMUTABLE(inode))
-                       return -EACCES;
-       }
-
        if (current->fsuid == inode->i_uid)
                mode >>= 6;
-       else if (in_group_p(inode->i_gid))
-               mode >>= 3;
+       else {
+               if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
+                       int error = check_acl(inode, mask);
+                       if (error == -EACCES)
+                               goto check_capabilities;
+                       else if (error != -EAGAIN)
+                               return error;
+               }
+
+               if (in_group_p(inode->i_gid))
+                       mode >>= 3;
+       }
 
        /*
         * If the DACs are ok we don't need any capability check.
@@ -192,6 +208,7 @@ int vfs_permission(struct inode * inode, int mask)
        if (((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask))
                return 0;
 
+ check_capabilities:
        /*
         * Read/write DACs are always overridable.
         * Executable DACs are overridable if at least one exec bit is set.
@@ -211,35 +228,99 @@ int vfs_permission(struct inode * inode, int mask)
        return -EACCES;
 }
 
-static inline int xid_permission(struct inode *inode)
+static inline int vx_barrier(struct inode *inode)
+{
+       if (IS_BARRIER(inode) && !vx_check(0, VX_ADMIN)) {
+               vxwprintk(1, "xid=%d did hit the barrier.",
+                       vx_current_xid());
+               return 1;
+       }
+       return 0;
+}
+
+static inline int xid_permission(struct inode *inode, int mask, struct nameidata *nd)
 {
+       if (vx_barrier(inode))
+               return -EACCES;
        if (inode->i_xid == 0)
                return 0;
        if (vx_check(inode->i_xid, VX_ADMIN|VX_WATCH|VX_IDENT))
                return 0;
+
+       vxwprintk(1, "xid=%d denied access to %p[#%d,%lu] »%s«.",
+               vx_current_xid(), inode, inode->i_xid, inode->i_ino,
+               vxd_cond_path(nd));
        return -EACCES;
 }
 
-int permission(struct inode * inode,int mask, struct nameidata *nd)
+int permission(struct inode *inode, int mask, struct nameidata *nd)
 {
-       int retval;
-       int submask;
+       int retval, submask;
+
+       if (mask & MAY_WRITE) {
+               umode_t mode = inode->i_mode;
+
+               /*
+                * Nobody gets write access to a read-only fs.
+                */
+               if ((IS_RDONLY(inode) || (nd && MNT_IS_RDONLY(nd->mnt))) &&
+                   (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
+                       return -EROFS;
+
+               /*
+                * Nobody gets write access to an immutable file.
+                */
+               if (IS_IMMUTABLE(inode))
+                       return -EACCES;
+       }
+
 
        /* Ordinary permission routines do not understand MAY_APPEND. */
        submask = mask & ~MAY_APPEND;
-
-       if ((retval = xid_permission(inode)))
+       if ((retval = xid_permission(inode, mask, nd)))
                return retval;
        if (inode->i_op && inode->i_op->permission)
                retval = inode->i_op->permission(inode, submask, nd);
        else
-               retval = vfs_permission(inode, submask);
+               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->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_dentry->d_inode, mask, NULL);
+}
+
 /*
  * get_write_access() gets write permission for a file.
  * put_write_access() releases this write permission.
@@ -292,6 +373,28 @@ void path_release(struct nameidata *nd)
        mntput(nd->mnt);
 }
 
+/*
+ * umount() mustn't call path_release()/mntput() as that would clear
+ * mnt_expiry_mark
+ */
+void path_release_on_umount(struct nameidata *nd)
+{
+       dput(nd->dentry);
+       mntput_no_expire(nd->mnt);
+}
+
+/**
+ * release_open_intent - free up open intent resources
+ * @nd: pointer to nameidata
+ */
+void release_open_intent(struct nameidata *nd)
+{
+       if (nd->intent.open.file->f_dentry == NULL)
+               put_filp(nd->intent.open.file);
+       else
+               fput(nd->intent.open.file);
+}
+
 /*
  * Internal lookup() using the new generic dcache.
  * SMP-safe
@@ -318,19 +421,21 @@ static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name,
 /*
  * Short-cut version of permission(), for calling by
  * path_walk(), when dcache lock is held.  Combines parts
- * of permission() and vfs_permission(), and tests ONLY for
+ * of permission() and generic_permission(), and tests ONLY for
  * MAY_EXEC permission.
  *
  * If appropriate, check DAC only.  If not appropriate, or
  * 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;
 
-       if ((inode->i_op && inode->i_op->permission))
+       if (vx_barrier(inode))
+               return -EACCES;
+       if (inode->i_op && inode->i_op->permission)
                return -EAGAIN;
 
        if (current->fsuid == inode->i_uid)
@@ -344,6 +449,9 @@ static inline int exec_permission_lite(struct inode *inode,
        if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
                goto ok;
 
+       if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
+               goto ok;
+
        if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
                goto ok;
 
@@ -365,7 +473,7 @@ 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..
@@ -391,7 +499,7 @@ 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;
        }
 
@@ -399,7 +507,7 @@ 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);
@@ -409,6 +517,113 @@ static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, s
        return result;
 }
 
+static int __emul_lookup_dentry(const char *, struct nameidata *);
+
+/* SMP-safe */
+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);
+               if (__emul_lookup_dentry(name,nd))
+                       return 0;
+               read_lock(&current->fs->lock);
+       }
+       nd->mnt = mntget(current->fs->rootmnt);
+       nd->dentry = dget(current->fs->root);
+       read_unlock(&current->fs->lock);
+       return 1;
+}
+
+static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
+{
+       int res = 0;
+       char *name;
+       if (IS_ERR(link))
+               goto fail;
+
+       if (*link == '/') {
+               path_release(nd);
+               if (!walk_init_root(link, nd))
+                       /* weird __emul_prefix() stuff did it */
+                       goto out;
+       }
+       res = link_path_walk(link, nd);
+out:
+       if (nd->depth || res || nd->last_type!=LAST_NORM)
+               return res;
+       /*
+        * If it is an iterative symlinks resolution in open_namei() we
+        * have to copy the last component. And all that crap because of
+        * bloody create() on broken symlinks. Furrfu...
+        */
+       name = __getname();
+       if (unlikely(!name)) {
+               path_release(nd);
+               return -ENOMEM;
+       }
+       strcpy(name, nd->last.name);
+       nd->last.name = name;
+       return 0;
+fail:
+       path_release(nd);
+       return PTR_ERR(link);
+}
+
+struct path {
+       struct vfsmount *mnt;
+       struct dentry *dentry;
+};
+
+static inline void dput_path(struct path *path, struct nameidata *nd)
+{
+       dput(path->dentry);
+       if (path->mnt != nd->mnt)
+               mntput(path->mnt);
+}
+
+static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
+{
+       dput(nd->dentry);
+       if (nd->mnt != path->mnt)
+               mntput(nd->mnt);
+       nd->mnt = path->mnt;
+       nd->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(path->mnt, dentry);
+       nd_set_link(nd, NULL);
+
+       if (path->mnt != nd->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, cookie);
+       }
+       dput(dentry);
+       mntput(path->mnt);
+
+       return error;
+}
+
 /*
  * This limits recursive symlink follows to 8, while
  * limiting consecutive symlinks to 40.
@@ -416,24 +631,27 @@ static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, s
  * Without that kind of total limit, nasty chains of consecutive
  * symlinks can cause almost arbitrarily long lookups. 
  */
-static inline int do_follow_link(struct dentry *dentry, struct nameidata *nd)
+static inline int do_follow_link(struct path *path, struct nameidata *nd)
 {
        int err = -ELOOP;
-       if (current->link_count >= 5)
+       if (current->link_count >= MAX_NESTED_LINKS)
                goto loop;
        if (current->total_link_count >= 40)
                goto loop;
+       BUG_ON(nd->depth >= MAX_NESTED_LINKS);
        cond_resched();
-       err = security_inode_follow_link(dentry, nd);
+       err = security_inode_follow_link(path->dentry, nd);
        if (err)
                goto loop;
        current->link_count++;
        current->total_link_count++;
-       touch_atime(nd->mnt, dentry);
-       err = dentry->d_inode->i_op->follow_link(dentry, nd);
+       nd->depth++;
+       err = __do_follow_link(path, nd);
        current->link_count--;
+       nd->depth--;
        return err;
 loop:
+       dput_path(path, nd);
        path_release(nd);
        return err;
 }
@@ -461,114 +679,142 @@ int follow_up(struct vfsmount **mnt, struct dentry **dentry)
 /* no need for dcache_lock, as serialization is taken care in
  * namespace.c
  */
-static int follow_mount(struct vfsmount **mnt, struct dentry **dentry)
+static int __follow_mount(struct path *path)
 {
        int res = 0;
+       while (d_mountpoint(path->dentry)) {
+               struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
+               if (!mounted)
+                       break;
+               dput(path->dentry);
+               if (res)
+                       mntput(path->mnt);
+               path->mnt = mounted;
+               path->dentry = dget(mounted->mnt_root);
+               res = 1;
+       }
+       return res;
+}
+
+static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
+{
        while (d_mountpoint(*dentry)) {
                struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
                if (!mounted)
                        break;
+               dput(*dentry);
                mntput(*mnt);
                *mnt = mounted;
-               dput(*dentry);
                *dentry = dget(mounted->mnt_root);
-               res = 1;
        }
-       return res;
 }
 
 /* no need for dcache_lock, as serialization is taken care in
  * namespace.c
  */
-static inline int __follow_down(struct vfsmount **mnt, struct dentry **dentry)
+int follow_down(struct vfsmount **mnt, struct dentry **dentry)
 {
        struct vfsmount *mounted;
 
        mounted = lookup_mnt(*mnt, *dentry);
        if (mounted) {
+               dput(*dentry);
                mntput(*mnt);
                *mnt = mounted;
-               dput(*dentry);
                *dentry = dget(mounted->mnt_root);
                return 1;
        }
        return 0;
 }
 
-int follow_down(struct vfsmount **mnt, struct dentry **dentry)
-{
-       return __follow_down(mnt,dentry);
-}
-static inline void follow_dotdot(struct vfsmount **mnt, struct dentry **dentry)
+static __always_inline void follow_dotdot(struct nameidata *nd)
 {
        while(1) {
                struct vfsmount *parent;
-               struct dentry *old = *dentry;
+               struct dentry *old = nd->dentry;
 
                 read_lock(&current->fs->lock);
-               if (*dentry == current->fs->root &&
-                   *mnt == current->fs->rootmnt) {
+               if (nd->dentry == current->fs->root &&
+                   nd->mnt == current->fs->rootmnt) {
                         read_unlock(&current->fs->lock);
-                       break;
+                       /* for sane '/' avoid follow_mount() */
+                       return;
                }
                 read_unlock(&current->fs->lock);
                spin_lock(&dcache_lock);
-               if (*dentry != (*mnt)->mnt_root) {
-                       *dentry = dget((*dentry)->d_parent);
+               if (nd->dentry != nd->mnt->mnt_root) {
+                       nd->dentry = dget(nd->dentry->d_parent);
                        spin_unlock(&dcache_lock);
                        dput(old);
                        break;
                }
                spin_unlock(&dcache_lock);
                spin_lock(&vfsmount_lock);
-               parent = (*mnt)->mnt_parent;
-               if (parent == *mnt) {
+               parent = nd->mnt->mnt_parent;
+               if (parent == nd->mnt) {
                        spin_unlock(&vfsmount_lock);
                        break;
                }
                mntget(parent);
-               *dentry = dget((*mnt)->mnt_mountpoint);
+               nd->dentry = dget(nd->mnt->mnt_mountpoint);
                spin_unlock(&vfsmount_lock);
                dput(old);
-               mntput(*mnt);
-               *mnt = parent;
+               mntput(nd->mnt);
+               nd->mnt = parent;
        }
-       follow_mount(mnt, dentry);
+       follow_mount(&nd->mnt, &nd->dentry);
 }
 
-struct path {
-       struct vfsmount *mnt;
-       struct dentry *dentry;
-};
-
 /*
  *  It's more convoluted than I'd like it to be, but... it's still fairly
  *  small and for now I'd prefer to have fast path as straight as possible.
  *  It _is_ time-critical.
  */
 static int do_lookup(struct nameidata *nd, struct qstr *name,
-                    struct path *path)
+                    struct path *path, int atomic)
 {
        struct vfsmount *mnt = nd->mnt;
        struct dentry *dentry = __d_lookup(nd->dentry, name);
+       struct inode *inode;
 
        if (!dentry)
                goto need_lookup;
        if (dentry->d_op && dentry->d_op->d_revalidate)
                goto need_revalidate;
+       inode = dentry->d_inode;
+       if (!inode)
+               goto done;
+       if (!vx_check(inode->i_xid, VX_WATCH|VX_ADMIN|VX_HOSTID|VX_IDENT))
+               goto hidden;
+       if (inode->i_sb->s_magic == PROC_SUPER_MAGIC) {
+               struct proc_dir_entry *de = PDE(inode);
+
+               if (de && !vx_hide_check(0, de->vx_flags))
+                       goto hidden;
+       }
 done:
        path->mnt = mnt;
        path->dentry = dentry;
+       __follow_mount(path);
        return 0;
+hidden:
+       vxwprintk(1, "xid=%d did lookup hidden %p[#%d,%lu] »%s«.",
+               vx_current_xid(), inode, inode->i_xid, inode->i_ino,
+               vxd_path(dentry, mnt));
+       dput(dentry);
+       return -ENOENT;
 
 need_lookup:
+       if (atomic)
+               return -EWOULDBLOCKIO;
        dentry = real_lookup(nd->dentry, name, nd);
        if (IS_ERR(dentry))
                goto fail;
        goto done;
 
 need_revalidate:
+       if (atomic)
+               return -EWOULDBLOCKIO;
        if (dentry->d_op->d_revalidate(dentry, nd))
                goto done;
        if (d_invalidate(dentry))
@@ -582,27 +828,29 @@ fail:
 
 /*
  * 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.
  *
- * This is the basic name resolution function, turning a pathname
- * into the final dentry.
- *
- * We expect 'base' to be positive and a directory.
+ * Returns 0 and nd will have valid dentry and mnt on success.
+ * Returns error and drops reference to input namei data on failure.
  */
-int fastcall link_path_walk(const char * name, struct nameidata *nd)
+static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
 {
        struct path next;
        struct inode *inode;
-       int err;
+       int err, atomic;
        unsigned int lookup_flags = nd->flags;
-       
+
+       atomic = (lookup_flags & LOOKUP_ATOMIC);
+
        while (*name=='/')
                name++;
        if (!*name)
                goto return_reval;
 
        inode = nd->dentry->d_inode;
-       if (current->link_count)
-               lookup_flags = LOOKUP_FOLLOW;
+       if (nd->depth)
+               lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
 
        /* At this point we know we have a real path component. */
        for(;;) {
@@ -610,10 +858,10 @@ int fastcall 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;
 
@@ -647,7 +895,7 @@ int fastcall link_path_walk(const char * name, struct nameidata *nd)
                        case 2: 
                                if (this.name[1] != '.')
                                        break;
-                               follow_dotdot(&nd->mnt, &nd->dentry);
+                               follow_dotdot(nd);
                                inode = nd->dentry->d_inode;
                                /* fallthrough */
                        case 1:
@@ -662,13 +910,10 @@ int fastcall link_path_walk(const char * name, struct nameidata *nd)
                        if (err < 0)
                                break;
                }
-               nd->flags |= LOOKUP_CONTINUE;
                /* This does the actual lookups.. */
-               err = do_lookup(nd, &this, &next);
+               err = do_lookup(nd, &this, &next, atomic);
                if (err)
                        break;
-               /* Check mountpoints.. */
-               follow_mount(&next.mnt, &next.dentry);
 
                err = -ENOENT;
                inode = next.dentry->d_inode;
@@ -679,10 +924,7 @@ int fastcall link_path_walk(const char * name, struct nameidata *nd)
                        goto out_dput;
 
                if (inode->i_op->follow_link) {
-                       mntget(next.mnt);
-                       err = do_follow_link(next.dentry, nd);
-                       dput(next.dentry);
-                       mntput(next.mnt);
+                       err = do_follow_link(&next, nd);
                        if (err)
                                goto return_err;
                        err = -ENOENT;
@@ -692,11 +934,8 @@ int fastcall link_path_walk(const char * name, struct nameidata *nd)
                        err = -ENOTDIR; 
                        if (!inode->i_op)
                                break;
-               } else {
-                       dput(nd->dentry);
-                       nd->mnt = next.mnt;
-                       nd->dentry = next.dentry;
-               }
+               } else
+                       path_to_nameidata(&next, nd);
                err = -ENOTDIR; 
                if (!inode->i_op->lookup)
                        break;
@@ -706,7 +945,8 @@ int fastcall 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) {
@@ -715,7 +955,7 @@ last_component:
                        case 2: 
                                if (this.name[1] != '.')
                                        break;
-                               follow_dotdot(&nd->mnt, &nd->dentry);
+                               follow_dotdot(nd);
                                inode = nd->dentry->d_inode;
                                /* fallthrough */
                        case 1:
@@ -726,25 +966,18 @@ last_component:
                        if (err < 0)
                                break;
                }
-               err = do_lookup(nd, &this, &next);
+               err = do_lookup(nd, &this, &next, atomic);
                if (err)
                        break;
-               follow_mount(&next.mnt, &next.dentry);
                inode = next.dentry->d_inode;
                if ((lookup_flags & LOOKUP_FOLLOW)
                    && inode && inode->i_op && inode->i_op->follow_link) {
-                       mntget(next.mnt);
-                       err = do_follow_link(next.dentry, nd);
-                       dput(next.dentry);
-                       mntput(next.mnt);
+                       err = do_follow_link(&next, nd);
                        if (err)
                                goto return_err;
                        inode = nd->dentry->d_inode;
-               } else {
-                       dput(nd->dentry);
-                       nd->mnt = next.mnt;
-                       nd->dentry = next.dentry;
-               }
+               } else
+                       path_to_nameidata(&next, nd);
                err = -ENOENT;
                if (!inode)
                        break;
@@ -780,7 +1013,7 @@ return_reval:
 return_base:
                return 0;
 out_dput:
-               dput(next.dentry);
+               dput_path(&next, nd);
                break;
        }
        path_release(nd);
@@ -788,42 +1021,78 @@ return_err:
        return err;
 }
 
+/*
+ * Wrapper to retry pathname resolution whenever the underlying
+ * file system returns an ESTALE.
+ *
+ * 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)
+{
+       struct nameidata save = *nd;
+       int result;
+
+       /* make sure the stuff we saved doesn't go away */
+       dget(save.dentry);
+       mntget(save.mnt);
+
+       result = __link_path_walk(name, nd);
+       if (result == -ESTALE) {
+               *nd = save;
+               dget(nd->dentry);
+               mntget(nd->mnt);
+               nd->flags |= LOOKUP_REVAL;
+               result = __link_path_walk(name, nd);
+       }
+
+       dput(save.dentry);
+       mntput(save.mnt);
+
+       return result;
+}
+
 int fastcall path_walk(const char * name, struct nameidata *nd)
 {
        current->total_link_count = 0;
        return link_path_walk(name, nd);
 }
 
-/* SMP-safe */
-/* returns 1 if everything is done */
+/* 
+ * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
+ * everything is done. Returns 0 and drops input nd, if lookup failed;
+ */
 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 nameidata nd_root;
+               struct dentry *old_dentry = nd->dentry;
+               struct vfsmount *old_mnt = nd->mnt;
+               struct qstr last = nd->last;
+               int last_type = nd->last_type;
                /*
                 * NAME was not found in alternate root or it's a directory.  Try to find
                 * it in the normal root:
                 */
-               nd_root.last_type = LAST_ROOT;
-               nd_root.flags = nd->flags;
-               memcpy(&nd_root.intent, &nd->intent, sizeof(nd_root.intent));
+               nd->last_type = LAST_ROOT;
                read_lock(&current->fs->lock);
-               nd_root.mnt = mntget(current->fs->rootmnt);
-               nd_root.dentry = dget(current->fs->root);
+               nd->mnt = mntget(current->fs->rootmnt);
+               nd->dentry = dget(current->fs->root);
                read_unlock(&current->fs->lock);
-               if (path_walk(name, &nd_root))
-                       return 1;
-               if (nd_root.dentry->d_inode) {
+               if (path_walk(name, nd) == 0) {
+                       if (nd->dentry->d_inode) {
+                               dput(old_dentry);
+                               mntput(old_mnt);
+                               return 1;
+                       }
                        path_release(nd);
-                       nd->dentry = nd_root.dentry;
-                       nd->mnt = nd_root.mnt;
-                       nd->last = nd_root.last;
-                       return 1;
                }
-               path_release(&nd_root);
+               nd->dentry = old_dentry;
+               nd->mnt = old_mnt;
+               nd->last = last;
+               nd->last_type = last_type;
        }
        return 1;
 }
@@ -856,58 +1125,147 @@ set_it:
        }
 }
 
-/* SMP-safe */
-static inline int
-walk_init_root(const char *name, struct nameidata *nd)
+/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
+static int fastcall do_path_lookup(int dfd, const char *name,
+                               unsigned int flags, 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);
-               if (__emul_lookup_dentry(name,nd))
-                       return 0;
-               read_lock(&current->fs->lock);
-       }
-       nd->mnt = mntget(current->fs->rootmnt);
-       nd->dentry = dget(current->fs->root);
-       read_unlock(&current->fs->lock);
-       return 1;
-}
-
-int fastcall path_lookup(const char *name, unsigned int flags, struct nameidata *nd)
-{
-       int retval;
+       int retval = 0;
+       int fput_needed;
+       struct file *file;
 
        nd->last_type = LAST_ROOT; /* if there are only slashes... */
        nd->flags = flags;
+       nd->depth = 0;
 
-       read_lock(&current->fs->lock);
        if (*name=='/') {
+               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);
                        if (__emul_lookup_dentry(name,nd))
-                               return 0;
+                               goto out; /* found in altroot */
                        read_lock(&current->fs->lock);
                }
                nd->mnt = mntget(current->fs->rootmnt);
                nd->dentry = dget(current->fs->root);
-       }
-       else{
+               read_unlock(&current->fs->lock);
+       } else if (dfd == AT_FDCWD) {
+               read_lock(&current->fs->lock);
                nd->mnt = mntget(current->fs->pwdmnt);
                nd->dentry = dget(current->fs->pwd);
+               read_unlock(&current->fs->lock);
+       } else {
+               struct dentry *dentry;
+
+               file = fget_light(dfd, &fput_needed);
+               retval = -EBADF;
+               if (!file)
+                       goto out_fail;
+
+               dentry = file->f_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->mnt = mntget(file->f_vfsmnt);
+               nd->dentry = dget(dentry);
+
+               fput_light(file, fput_needed);
        }
-       read_unlock(&current->fs->lock);
        current->total_link_count = 0;
        retval = link_path_walk(name, nd);
-       if (unlikely(current->audit_context
-                    && nd && nd->dentry && nd->dentry->d_inode))
-               audit_inode(name,
-                           nd->dentry->d_inode->i_ino,
-                           nd->dentry->d_inode->i_rdev);
+out:
+       if (likely(retval == 0)) {
+               if (unlikely(current->audit_context && nd && nd->dentry &&
+                               nd->dentry->d_inode))
+               audit_inode(name, nd->dentry->d_inode, flags);
+       }
+out_fail:
        return retval;
+
+fput_fail:
+       fput_light(file, fput_needed);
+       goto out_fail;
+}
+
+int fastcall path_lookup(const char *name, unsigned int flags,
+                       struct nameidata *nd)
+{
+       return do_path_lookup(AT_FDCWD, name, flags, nd);
+}
+
+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_release(nd);
+               }
+       } 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;
 }
 
 /*
@@ -954,9 +1312,9 @@ out:
        return dentry;
 }
 
-struct dentry * lookup_hash(struct qstr *name, struct dentry * base)
+static struct dentry *lookup_hash(struct nameidata *nd)
 {
-       return __lookup_hash(name, base, NULL);
+       return __lookup_hash(&nd->last, nd->dentry, nd);
 }
 
 /* SMP-safe */
@@ -980,7 +1338,7 @@ struct dentry * lookup_one_len(const char * name, struct dentry * base, int len)
        }
        this.hash = end_name_hash(hash);
 
-       return lookup_hash(&this, base);
+       return __lookup_hash(&this, base, NULL);
 access:
        return ERR_PTR(-EACCES);
 }
@@ -996,18 +1354,24 @@ access:
  * that namei follows links, while lnamei does not.
  * SMP-safe
  */
-int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
+int fastcall __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 fastcall __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.
@@ -1042,12 +1406,18 @@ 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, struct nameidata *nd)
 {
        int error;
-       if (!victim->d_inode || victim->d_parent->d_inode != dir)
+
+       if (!victim->d_inode)
                return -ENOENT;
-       error = permission(dir,MAY_WRITE | MAY_EXEC, NULL);
+
+       BUG_ON(victim->d_parent->d_inode != dir);
+       audit_inode_child(victim->d_name.name, victim->d_inode, dir->i_ino);
+
+       error = permission(dir,MAY_WRITE | MAY_EXEC, nd);
        if (error)
                return error;
        if (IS_APPEND(dir))
@@ -1088,9 +1458,6 @@ 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)
@@ -1100,11 +1467,10 @@ 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;
+       if (f & O_ATOMICLOOKUP)
+               retval |= LOOKUP_ATOMIC;
 
        return retval;
 }
@@ -1117,39 +1483,39 @@ struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
        struct dentry *p;
 
        if (p1 == p2) {
-               down(&p1->d_inode->i_sem);
+               mutex_lock(&p1->d_inode->i_mutex);
                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(&p2->d_inode->i_mutex);
+                       mutex_lock(&p1->d_inode->i_mutex);
                        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(&p1->d_inode->i_mutex);
+                       mutex_lock(&p2->d_inode->i_mutex);
                        return p;
                }
        }
 
-       down(&p1->d_inode->i_sem);
-       down(&p2->d_inode->i_sem);
+       mutex_lock(&p1->d_inode->i_mutex);
+       mutex_lock(&p2->d_inode->i_mutex);
        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);
        }
 }
 
@@ -1170,10 +1536,8 @@ 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;
 }
 
@@ -1192,7 +1556,7 @@ int may_open(struct nameidata *nd, int acc_mode, int flag)
        if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
                return -EISDIR;
 
-       error = permission(inode, acc_mode, nd);
+       error = vfs_permission(nd, acc_mode);
        if (error)
                return error;
 
@@ -1208,7 +1572,8 @@ int may_open(struct nameidata *nd, int acc_mode, int flag)
                        return -EACCES;
 
                flag &= ~O_TRUNC;
-       } else if (IS_RDONLY(inode) && (flag & FMODE_WRITE))
+       } else if ((IS_RDONLY(inode) || MNT_IS_RDONLY(nd->mnt))
+               && (flag & FMODE_WRITE))
                return -EROFS;
        /*
         * An append-only file must be opened in append mode for writing.
@@ -1220,6 +1585,11 @@ int may_open(struct nameidata *nd, int acc_mode, int flag)
                        return -EPERM;
        }
 
+       /* O_NOATIME can only be set by the owner or superuser */
+       if (flag & O_NOATIME)
+               if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
+                       return -EPERM;
+
        /*
         * Ensure there are no outstanding leases on the file.
         */
@@ -1239,7 +1609,7 @@ int may_open(struct nameidata *nd, int acc_mode, int flag)
                if (!error) {
                        DQUOT_INIT(inode);
                        
-                       error = do_truncate(dentry, 0);
+                       error = do_truncate(dentry, 0, ATTR_MTIME|ATTR_CTIME, NULL);
                }
                put_write_access(inode);
                if (error)
@@ -1265,29 +1635,31 @@ 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;
-       struct dentry *dentry;
+       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;
@@ -1296,7 +1668,7 @@ 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;
 
@@ -1311,24 +1683,31 @@ int open_namei(const char * pathname, int flag, int mode, struct nameidata *nd)
 
        dir = nd->dentry;
        nd->flags &= ~LOOKUP_PARENT;
-       down(&dir->d_inode->i_sem);
-       dentry = __lookup_hash(&nd->last, nd->dentry, nd);
+       mutex_lock(&dir->d_inode->i_mutex);
+       path.dentry = lookup_hash(nd);
+       path.mnt = nd->mnt;
 
 do_last:
-       error = PTR_ERR(dentry);
-       if (IS_ERR(dentry)) {
-               up(&dir->d_inode->i_sem);
+       error = PTR_ERR(path.dentry);
+       if (IS_ERR(path.dentry)) {
+               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 (!dentry->d_inode) {
+       if (!path.dentry->d_inode) {
                if (!IS_POSIXACL(dir->d_inode))
                        mode &= ~current->fs->umask;
-               error = vfs_create(dir->d_inode, dentry, mode, nd);
-               up(&dir->d_inode->i_sem);
+               error = vfs_create(dir->d_inode, path.dentry, mode, nd);
+               mutex_unlock(&dir->d_inode->i_mutex);
                dput(nd->dentry);
-               nd->dentry = dentry;
+               nd->dentry = path.dentry;
                if (error)
                        goto exit;
                /* Don't check for write permission, don't truncate */
@@ -1340,28 +1719,26 @@ do_last:
        /*
         * It already exists.
         */
-       up(&dir->d_inode->i_sem);
+       mutex_unlock(&dir->d_inode->i_mutex);
 
        error = -EEXIST;
        if (flag & O_EXCL)
                goto exit_dput;
 
-       if (d_mountpoint(dentry)) {
+       if (__follow_mount(&path)) {
                error = -ELOOP;
                if (flag & O_NOFOLLOW)
                        goto exit_dput;
-               while (__follow_down(&nd->mnt,&dentry) && d_mountpoint(dentry));
        }
        error = -ENOENT;
-       if (!dentry->d_inode)
+       if (!path.dentry->d_inode)
                goto exit_dput;
-       if (dentry->d_inode->i_op && dentry->d_inode->i_op->follow_link)
+       if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
                goto do_link;
 
-       dput(nd->dentry);
-       nd->dentry = dentry;
+       path_to_nameidata(&path, nd);
        error = -EISDIR;
-       if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode))
+       if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
                goto exit;
 ok:
        error = may_open(nd, acc_mode, flag);
@@ -1370,8 +1747,10 @@ ok:
        return 0;
 
 exit_dput:
-       dput(dentry);
+       dput_path(&path, nd);
 exit:
+       if (!IS_ERR(nd->intent.open.file))
+               release_open_intent(nd);
        path_release(nd);
        return error;
 
@@ -1390,35 +1769,32 @@ do_link:
         * are done. Procfs-like symlinks just set LAST_BIND.
         */
        nd->flags |= LOOKUP_PARENT;
-       error = security_inode_follow_link(dentry, nd);
+       error = security_inode_follow_link(path.dentry, nd);
        if (error)
                goto exit_dput;
-       touch_atime(nd->mnt, dentry);
-       error = dentry->d_inode->i_op->follow_link(dentry, nd);
-       dput(dentry);
+       error = __do_follow_link(&path, nd);
        if (error)
                return error;
        nd->flags &= ~LOOKUP_PARENT;
-       if (nd->last_type == LAST_BIND) {
-               dentry = nd->dentry;
+       if (nd->last_type == LAST_BIND)
                goto ok;
-       }
        error = -EISDIR;
        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);
-       dentry = __lookup_hash(&nd->last, nd->dentry, nd);
-       putname(nd->last.name);
+       mutex_lock(&dir->d_inode->i_mutex);
+       path.dentry = lookup_hash(nd);
+       path.mnt = nd->mnt;
+       __putname(nd->last.name);
        goto do_last;
 }
 
@@ -1429,19 +1805,35 @@ do_link:
  *
  * Simple function to lookup and return a dentry and create it
  * if it doesn't exist.  Is SMP-safe.
+ *
+ * Returns with nd->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(&nd->dentry->d_inode->i_mutex);
+       /*
+        * 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);
+
+       /*
+        * 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;
@@ -1451,10 +1843,12 @@ enoent:
 fail:
        return dentry;
 }
+EXPORT_SYMBOL_GPL(lookup_create);
 
-int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
+int vfs_mknod(struct inode *dir, struct dentry *dentry,
+       int mode, dev_t dev, struct nameidata *nd)
 {
-       int error = may_create(dir, dentry, NULL);
+       int error = may_create(dir, dentry, nd);
 
        if (error)
                return error;
@@ -1471,14 +1865,13 @@ 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;
@@ -1491,7 +1884,7 @@ 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);
@@ -1505,11 +1898,12 @@ asmlinkage long sys_mknod(const char __user * filename, int mode, unsigned dev)
                        error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
                        break;
                case S_IFCHR: case S_IFBLK:
-                       error = vfs_mknod(nd.dentry->d_inode,dentry,mode,
-                                       new_decode_dev(dev));
+                       error = vfs_mknod(nd.dentry->d_inode, dentry, mode,
+                                       new_decode_dev(dev), &nd);
                        break;
                case S_IFIFO: case S_IFSOCK:
-                       error = vfs_mknod(nd.dentry->d_inode,dentry,mode,0);
+                       error = vfs_mknod(nd.dentry->d_inode, dentry, mode,
+                                       0, &nd);
                        break;
                case S_IFDIR:
                        error = -EPERM;
@@ -1519,7 +1913,7 @@ asmlinkage long sys_mknod(const char __user * filename, int mode, unsigned dev)
                }
                dput(dentry);
        }
-       up(&nd.dentry->d_inode->i_sem);
+       mutex_unlock(&nd.dentry->d_inode->i_mutex);
        path_release(&nd);
 out:
        putname(tmp);
@@ -1527,9 +1921,15 @@ out:
        return error;
 }
 
-int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
+asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev)
 {
-       int error = may_create(dir, dentry, NULL);
+       return sys_mknodat(AT_FDCWD, filename, mode, dev);
+}
+
+int vfs_mkdir(struct inode *dir, struct dentry *dentry,
+       int mode, struct nameidata *nd)
+{
+       int error = may_create(dir, dentry, nd);
 
        if (error)
                return error;
@@ -1544,14 +1944,12 @@ 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;
@@ -1562,7 +1960,7 @@ asmlinkage long sys_mkdir(const char __user * pathname, int mode)
                struct dentry *dentry;
                struct nameidata nd;
 
-               error = path_lookup(tmp, LOOKUP_PARENT, &nd);
+               error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
                if (error)
                        goto out;
                dentry = lookup_create(&nd, 1);
@@ -1570,10 +1968,11 @@ asmlinkage long sys_mkdir(const char __user * pathname, int mode)
                if (!IS_ERR(dentry)) {
                        if (!IS_POSIXACL(nd.dentry->d_inode))
                                mode &= ~current->fs->umask;
-                       error = vfs_mkdir(nd.dentry->d_inode, dentry, mode);
+                       error = vfs_mkdir(nd.dentry->d_inode, dentry,
+                               mode, &nd);
                        dput(dentry);
                }
-               up(&nd.dentry->d_inode->i_sem);
+               mutex_unlock(&nd.dentry->d_inode->i_mutex);
                path_release(&nd);
 out:
                putname(tmp);
@@ -1582,6 +1981,11 @@ out:
        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
@@ -1597,26 +2001,23 @@ out:
  * if it cannot handle the case of removing a directory
  * that is still in use by something else..
  */
-static void d_unhash(struct dentry *dentry)
+void dentry_unhash(struct dentry *dentry)
 {
        dget(dentry);
-       spin_lock(&dcache_lock);
-       switch (atomic_read(&dentry->d_count)) {
-       default:
-               spin_unlock(&dcache_lock);
+       if (atomic_read(&dentry->d_count))
                shrink_dcache_parent(dentry);
-               spin_lock(&dcache_lock);
-               if (atomic_read(&dentry->d_count) != 2)
-                       break;
-       case 2:
+       spin_lock(&dcache_lock);
+       spin_lock(&dentry->d_lock);
+       if (atomic_read(&dentry->d_count) == 2)
                __d_drop(dentry);
-       }
+       spin_unlock(&dentry->d_lock);
        spin_unlock(&dcache_lock);
 }
 
-int vfs_rmdir(struct inode *dir, struct dentry *dentry)
+int vfs_rmdir(struct inode *dir, struct dentry *dentry,
+       struct nameidata *nd)
 {
-       int error = may_delete(dir, dentry, 1);
+       int error = may_delete(dir, dentry, 1, nd);
 
        if (error)
                return error;
@@ -1626,8 +2027,8 @@ int vfs_rmdir(struct inode *dir, struct dentry *dentry)
 
        DQUOT_INIT(dir);
 
-       down(&dentry->d_inode->i_sem);
-       d_unhash(dentry);
+       mutex_lock(&dentry->d_inode->i_mutex);
+       dentry_unhash(dentry);
        if (d_mountpoint(dentry))
                error = -EBUSY;
        else {
@@ -1638,9 +2039,8 @@ 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);
@@ -1648,7 +2048,7 @@ 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;
@@ -1659,7 +2059,7 @@ 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;
 
@@ -1674,14 +2074,14 @@ 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(&nd.dentry->d_inode->i_mutex);
+       dentry = lookup_hash(&nd);
        error = PTR_ERR(dentry);
        if (!IS_ERR(dentry)) {
-               error = vfs_rmdir(nd.dentry->d_inode, dentry);
+               error = vfs_rmdir(nd.dentry->d_inode, dentry, &nd);
                dput(dentry);
        }
-       up(&nd.dentry->d_inode->i_sem);
+       mutex_unlock(&nd.dentry->d_inode->i_mutex);
 exit1:
        path_release(&nd);
 exit:
@@ -1689,9 +2089,15 @@ exit:
        return error;
 }
 
-int vfs_unlink(struct inode *dir, struct dentry *dentry)
+asmlinkage long sys_rmdir(const char __user *pathname)
 {
-       int error = may_delete(dir, dentry, 0);
+       return do_rmdir(AT_FDCWD, pathname);
+}
+
+int vfs_unlink(struct inode *dir, struct dentry *dentry,
+       struct nameidata *nd)
+{
+       int error = may_delete(dir, dentry, 0, nd);
 
        if (error)
                return error;
@@ -1701,7 +2107,7 @@ 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 {
@@ -1709,23 +2115,23 @@ 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)) {
                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;
@@ -1737,14 +2143,14 @@ 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(&nd.dentry->d_inode->i_mutex);
+       dentry = lookup_hash(&nd);
        error = PTR_ERR(dentry);
        if (!IS_ERR(dentry)) {
                /* Why not before? Because we want correct error value */
@@ -1753,18 +2159,17 @@ 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.dentry->d_inode, dentry, &nd);
        exit2:
                dput(dentry);
        }
-       up(&nd.dentry->d_inode->i_sem);
+       mutex_unlock(&nd.dentry->d_inode->i_mutex);
+       if (inode)
+               iput(inode);    /* truncate the inode here */
 exit1:
        path_release(&nd);
 exit:
        putname(name);
-
-       if (inode)
-               iput(inode);    /* truncate the inode here */
        return error;
 
 slashes:
@@ -1773,9 +2178,26 @@ slashes:
        goto exit2;
 }
 
-int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
+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, struct nameidata *nd)
 {
-       int error = may_create(dir, dentry, NULL);
+       int error = may_create(dir, dentry, nd);
 
        if (error)
                return error;
@@ -1789,14 +2211,13 @@ int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
 
        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;
@@ -1811,16 +2232,17 @@ asmlinkage long sys_symlink(const char __user * oldname, const char __user * new
                struct dentry *dentry;
                struct nameidata nd;
 
-               error = path_lookup(to, LOOKUP_PARENT, &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)) {
-                       error = vfs_symlink(nd.dentry->d_inode, dentry, from);
+                       error = vfs_symlink(nd.dentry->d_inode, dentry,
+                               from, S_IALLUGO, &nd);
                        dput(dentry);
                }
-               up(&nd.dentry->d_inode->i_sem);
+               mutex_unlock(&nd.dentry->d_inode->i_mutex);
                path_release(&nd);
 out:
                putname(to);
@@ -1829,7 +2251,13 @@ out:
        return error;
 }
 
-int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
+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 nameidata *nd)
 {
        struct inode *inode = old_dentry->d_inode;
        int error;
@@ -1837,7 +2265,7 @@ int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_de
        if (!inode)
                return -ENOENT;
 
-       error = may_create(dir, new_dentry, NULL);
+       error = may_create(dir, new_dentry, nd);
        if (error)
                return error;
 
@@ -1858,14 +2286,12 @@ 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_create(dir, new_dentry);
        return error;
 }
 
@@ -1878,21 +2304,26 @@ 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 != 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, 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;
@@ -1901,10 +2332,11 @@ asmlinkage long sys_link(const char __user * oldname, const char __user * newnam
        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);
+               error = vfs_link(old_nd.dentry, nd.dentry->d_inode,
+                       new_dentry, &nd);
                dput(new_dentry);
        }
-       up(&nd.dentry->d_inode->i_sem);
+       mutex_unlock(&nd.dentry->d_inode->i_mutex);
 out_release:
        path_release(&nd);
 out:
@@ -1915,6 +2347,11 @@ exit:
        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...
@@ -1922,17 +2359,17 @@ 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.
@@ -1942,13 +2379,13 @@ 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].
  */
-int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
-              struct inode *new_dir, struct dentry *new_dentry)
+static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
+                         struct inode *new_dir, struct dentry *new_dentry)
 {
        int error = 0;
        struct inode *target;
@@ -1969,8 +2406,8 @@ int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
 
        target = new_dentry->d_inode;
        if (target) {
-               down(&target->i_sem);
-               d_unhash(new_dentry);
+               mutex_lock(&target->i_mutex);
+               dentry_unhash(new_dentry);
        }
        if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
                error = -EBUSY;
@@ -1979,21 +2416,18 @@ 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) {
+       if (!error)
                d_move(old_dentry,new_dentry);
-               security_inode_post_rename(old_dir, old_dentry,
-                                          new_dir, new_dentry);
-       }
        return error;
 }
 
-int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
-              struct inode *new_dir, struct dentry *new_dentry)
+static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
+                           struct inode *new_dir, struct dentry *new_dentry)
 {
        struct inode *target;
        int error;
@@ -2005,7 +2439,7 @@ 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
@@ -2014,10 +2448,9 @@ int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
                /* The following d_move() should become unconditional */
                if (!(old_dir->i_sb->s_type->fs_flags & FS_ODD_RENAME))
                        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;
 }
@@ -2027,18 +2460,19 @@ 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;
  
-       error = may_delete(old_dir, old_dentry, is_dir);
+       error = may_delete(old_dir, old_dentry, is_dir, NULL);
        if (error)
                return error;
 
        if (!new_dentry->d_inode)
                error = may_create(new_dir, new_dentry, NULL);
        else
-               error = may_delete(new_dir, new_dentry, is_dir);
+               error = may_delete(new_dir, new_dentry, is_dir, NULL);
        if (error)
                return error;
 
@@ -2048,22 +2482,24 @@ 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->d_inode);
        }
+       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;
@@ -2071,11 +2507,11 @@ 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;
 
@@ -2094,7 +2530,7 @@ static inline int do_rename(const char * oldname, const char * newname)
 
        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;
@@ -2114,7 +2550,10 @@ 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);
+       error = -EROFS;
+       if (MNT_IS_RDONLY(newnd.mnt))
+               goto exit4;
+       new_dentry = lookup_hash(&newnd);
        error = PTR_ERR(new_dentry);
        if (IS_ERR(new_dentry))
                goto exit4;
@@ -2139,7 +2578,8 @@ 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;
@@ -2151,13 +2591,18 @@ 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;
@@ -2175,40 +2620,25 @@ out:
        return len;
 }
 
-static inline int
-__vfs_follow_link(struct nameidata *nd, const char *link)
+/*
+ * A helper for ->readlink().  This should be used *ONLY* for symlinks that
+ * have ->follow_link() touching nd only in nd_set_link().  Using (or not
+ * using) it for any given inode is up to filesystem.
+ */
+int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
 {
-       int res = 0;
-       char *name;
-       if (IS_ERR(link))
-               goto fail;
+       struct nameidata nd;
+       void *cookie;
 
-       if (*link == '/') {
-               path_release(nd);
-               if (!walk_init_root(link, nd))
-                       /* weird __emul_prefix() stuff did it */
-                       goto out;
+       nd.depth = 0;
+       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, cookie);
+               cookie = ERR_PTR(res);
        }
-       res = link_path_walk(link, nd);
-out:
-       if (current->link_count || res || nd->last_type!=LAST_NORM)
-               return res;
-       /*
-        * If it is an iterative symlinks resolution in open_namei() we
-        * have to copy the last component. And all that crap because of
-        * bloody create() on broken symlinks. Furrfu...
-        */
-       name = __getname();
-       if (unlikely(!name)) {
-               path_release(nd);
-               return -ENOMEM;
-       }
-       strcpy(name, nd->last.name);
-       nd->last.name = name;
-       return 0;
-fail:
-       path_release(nd);
-       return PTR_ERR(link);
+       return PTR_ERR(cookie);
 }
 
 int vfs_follow_link(struct nameidata *nd, const char *link)
@@ -2251,34 +2681,52 @@ int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
        return res;
 }
 
-int page_follow_link(struct dentry *dentry, struct nameidata *nd)
+void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
 {
        struct page *page = NULL;
-       char *s = page_getlink(dentry, &page);
-       int res = __vfs_follow_link(nd, s);
+       nd_set_link(nd, page_getlink(dentry, &page));
+       return page;
+}
+
+void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
+{
+       struct page *page = cookie;
+
        if (page) {
                kunmap(page);
                page_cache_release(page);
        }
-       return res;
 }
 
-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);
+       struct page *page;
        int err = -ENOMEM;
        char *kaddr;
 
+retry:
+       page = find_or_create_page(mapping, 0, gfp_mask);
        if (!page)
                goto fail;
        err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
+       if (err == AOP_TRUNCATED_PAGE) {
+               page_cache_release(page);
+               goto retry;
+       }
        if (err)
                goto fail_map;
        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);
+       err = mapping->a_ops->commit_write(NULL, page, 0, len-1);
+       if (err == AOP_TRUNCATED_PAGE) {
+               page_cache_release(page);
+               goto retry;
+       }
+       if (err)
+               goto fail_map;
        /*
         * 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
@@ -2288,7 +2736,8 @@ int page_symlink(struct inode *inode, const char *symname, int len)
         */
        if (!PageUptodate(page)) {
                err = mapping->a_ops->readpage(NULL, page);
-               wait_on_page_locked(page);
+               if (err != AOP_TRUNCATED_PAGE)
+                       wait_on_page_locked(page);
        } else {
                unlock_page(page);
        }
@@ -2304,37 +2753,49 @@ fail:
        return err;
 }
 
+int page_symlink(struct inode *inode, const char *symname, int len)
+{
+       return __page_symlink(inode, symname, len,
+                       mapping_gfp_mask(inode->i_mapping));
+}
+
 struct inode_operations page_symlink_inode_operations = {
-       .readlink       = page_readlink,
-       .follow_link    = page_follow_link,
+       .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_create);
-EXPORT_SYMBOL(lookup_hash);
 EXPORT_SYMBOL(lookup_one_len);
-EXPORT_SYMBOL(page_follow_link);
+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(permission);
+EXPORT_SYMBOL(vfs_permission);
+EXPORT_SYMBOL(file_permission);
 EXPORT_SYMBOL(unlock_rename);
 EXPORT_SYMBOL(vfs_create);
 EXPORT_SYMBOL(vfs_follow_link);
 EXPORT_SYMBOL(vfs_link);
 EXPORT_SYMBOL(vfs_mkdir);
 EXPORT_SYMBOL(vfs_mknod);
-EXPORT_SYMBOL(vfs_permission);
+EXPORT_SYMBOL(generic_permission);
 EXPORT_SYMBOL(vfs_readlink);
 EXPORT_SYMBOL(vfs_rename);
 EXPORT_SYMBOL(vfs_rmdir);
 EXPORT_SYMBOL(vfs_symlink);
 EXPORT_SYMBOL(vfs_unlink);
+EXPORT_SYMBOL(dentry_unhash);
+EXPORT_SYMBOL(generic_readlink);