dentries: Extract common code to remove dentry from lru
[safe/jmp/linux-2.6] / fs / namei.c
index 5e2d98d..941c8e8 100644 (file)
@@ -30,7 +30,6 @@
 #include <linux/capability.h>
 #include <linux/file.h>
 #include <linux/fcntl.h>
-#include <linux/namei.h>
 #include <asm/namei.h>
 #include <asm/uaccess.h>
 
  * 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..
@@ -226,10 +227,14 @@ int generic_permission(struct inode *inode, int mask,
 
 int permission(struct inode *inode, int mask, struct nameidata *nd)
 {
-       umode_t mode = inode->i_mode;
        int retval, submask;
+       struct vfsmount *mnt = NULL;
+
+       if (nd)
+               mnt = nd->path.mnt;
 
        if (mask & MAY_WRITE) {
+               umode_t mode = inode->i_mode;
 
                /*
                 * Nobody gets write access to a read-only fs.
@@ -245,22 +250,34 @@ int permission(struct inode *inode, int mask, struct nameidata *nd)
                        return -EACCES;
        }
 
-
-       /*
-        * MAY_EXEC on regular files requires special handling: We override
-        * filesystem execute permissions if the mode bits aren't set or
-        * the fs is mounted with the "noexec" flag.
-        */
-       if ((mask & MAY_EXEC) && S_ISREG(mode) && (!(mode & S_IXUGO) ||
-                       (nd && nd->mnt && (nd->mnt->mnt_flags & MNT_NOEXEC))))
-               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;
 
@@ -279,7 +296,7 @@ int permission(struct inode *inode, int mask, struct nameidata *nd)
  */
 int vfs_permission(struct nameidata *nd, int mask)
 {
-       return permission(nd->dentry->d_inode, mask, nd);
+       return permission(nd->path.dentry->d_inode, mask, nd);
 }
 
 /**
@@ -345,21 +362,31 @@ int deny_write_access(struct file * file)
        return 0;
 }
 
-void path_release(struct nameidata *nd)
+/**
+ * path_get - get a reference to a path
+ * @path: path to get the reference to
+ *
+ * Given a path increment the reference count to the dentry and the vfsmount.
+ */
+void path_get(struct path *path)
 {
-       dput(nd->dentry);
-       mntput(nd->mnt);
+       mntget(path->mnt);
+       dget(path->dentry);
 }
+EXPORT_SYMBOL(path_get);
 
-/*
- * umount() mustn't call path_release()/mntput() as that would clear
- * mnt_expiry_mark
+/**
+ * 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_release_on_umount(struct nameidata *nd)
+void path_put(struct path *path)
 {
-       dput(nd->dentry);
-       mntput_no_expire(nd->mnt);
+       dput(path->dentry);
+       mntput(path->mnt);
 }
+EXPORT_SYMBOL(path_put);
 
 /**
  * release_open_intent - free up open intent resources
@@ -522,16 +549,16 @@ walk_init_root(const char *name, struct nameidata *nd)
        struct fs_struct *fs = current->fs;
 
        read_lock(&fs->lock);
-       if (fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
-               nd->mnt = mntget(fs->altrootmnt);
-               nd->dentry = dget(fs->altroot);
+       if (fs->altroot.dentry && !(nd->flags & LOOKUP_NOALT)) {
+               nd->path = fs->altroot;
+               path_get(&fs->altroot);
                read_unlock(&fs->lock);
                if (__emul_lookup_dentry(name,nd))
                        return 0;
                read_lock(&fs->lock);
        }
-       nd->mnt = mntget(fs->rootmnt);
-       nd->dentry = dget(fs->root);
+       nd->path = fs->root;
+       path_get(&fs->root);
        read_unlock(&fs->lock);
        return 1;
 }
@@ -544,7 +571,7 @@ static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *l
                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;
@@ -560,31 +587,31 @@ 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);
 }
 
-static inline void dput_path(struct path *path, struct nameidata *nd)
+static void path_put_conditional(struct path *path, struct nameidata *nd)
 {
        dput(path->dentry);
-       if (path->mnt != nd->mnt)
+       if (path->mnt != nd->path.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;
+       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)
@@ -596,7 +623,7 @@ static __always_inline int __do_follow_link(struct path *path, struct nameidata
        touch_atime(path->mnt, dentry);
        nd_set_link(nd, NULL);
 
-       if (path->mnt != nd->mnt) {
+       if (path->mnt != nd->path.mnt) {
                path_to_nameidata(path, nd);
                dget(dentry);
        }
@@ -611,8 +638,7 @@ static __always_inline int __do_follow_link(struct path *path, struct nameidata
                if (dentry->d_inode->i_op->put_link)
                        dentry->d_inode->i_op->put_link(dentry, nd, cookie);
        }
-       dput(dentry);
-       mntput(path->mnt);
+       path_put(path);
 
        return error;
 }
@@ -644,8 +670,8 @@ static inline int do_follow_link(struct path *path, struct nameidata *nd)
        nd->depth--;
        return err;
 loop:
-       dput_path(path, nd);
-       path_release(nd);
+       path_put_conditional(path, nd);
+       path_put(&nd->path);
        return err;
 }
 
@@ -726,37 +752,37 @@ static __always_inline void follow_dotdot(struct nameidata *nd)
 
        while(1) {
                struct vfsmount *parent;
-               struct dentry *old = nd->dentry;
+               struct dentry *old = nd->path.dentry;
 
                 read_lock(&fs->lock);
-               if (nd->dentry == fs->root &&
-                   nd->mnt == fs->rootmnt) {
+               if (nd->path.dentry == fs->root.dentry &&
+                   nd->path.mnt == fs->root.mnt) {
                         read_unlock(&fs->lock);
                        break;
                }
                 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);
 }
 
 /*
@@ -767,8 +793,8 @@ static __always_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;
@@ -781,7 +807,7 @@ 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;
@@ -806,7 +832,7 @@ 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;
@@ -818,7 +844,7 @@ 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 | (nd->flags & LOOKUP_CONTINUE);
 
@@ -866,7 +892,7 @@ 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;
@@ -875,8 +901,9 @@ 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;
                }
@@ -898,7 +925,7 @@ 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; 
@@ -926,13 +953,14 @@ 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;
                }
@@ -945,7 +973,7 @@ last_component:
                        err = do_follow_link(&next, nd);
                        if (err)
                                goto return_err;
-                       inode = nd->dentry->d_inode;
+                       inode = nd->path.dentry->d_inode;
                } else
                        path_to_nameidata(&next, nd);
                err = -ENOENT;
@@ -973,20 +1001,21 @@ 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_path(&next, nd);
+               path_put_conditional(&next, nd);
                break;
        }
-       path_release(nd);
+       path_put(&nd->path);
 return_err:
        return err;
 }
@@ -998,31 +1027,30 @@ 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);
+       path_put(&save.path);
 
        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);
@@ -1037,9 +1065,9 @@ 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 path old_path = nd->path;
                struct qstr last = nd->last;
                int last_type = nd->last_type;
                struct fs_struct *fs = current->fs;
@@ -1050,19 +1078,17 @@ static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
                 */
                nd->last_type = LAST_ROOT;
                read_lock(&fs->lock);
-               nd->mnt = mntget(fs->rootmnt);
-               nd->dentry = dget(fs->root);
+               nd->path = fs->root;
+               path_get(&fs->root);
                read_unlock(&fs->lock);
                if (path_walk(name, nd) == 0) {
-                       if (nd->dentry->d_inode) {
-                               dput(old_dentry);
-                               mntput(old_mnt);
+                       if (nd->path.dentry->d_inode) {
+                               path_put(&old_path);
                                return 1;
                        }
-                       path_release(nd);
+                       path_put(&nd->path);
                }
-               nd->dentry = old_dentry;
-               nd->mnt = old_mnt;
+               nd->path = old_path;
                nd->last = last;
                nd->last_type = last_type;
        }
@@ -1073,33 +1099,26 @@ void set_fs_altroot(void)
 {
        char *emul = __emul_prefix();
        struct nameidata nd;
-       struct vfsmount *mnt = NULL, *oldmnt;
-       struct dentry *dentry = NULL, *olddentry;
+       struct path path = {}, old_path;
        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;
-       }
+       if (!err)
+               path = nd.path;
 set_it:
        write_lock(&fs->lock);
-       oldmnt = fs->altrootmnt;
-       olddentry = fs->altroot;
-       fs->altrootmnt = mnt;
-       fs->altroot = dentry;
+       old_path = fs->altroot;
+       fs->altroot = path;
        write_unlock(&fs->lock);
-       if (olddentry) {
-               dput(olddentry);
-               mntput(oldmnt);
-       }
+       if (old_path.dentry)
+               path_put(&old_path);
 }
 
 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
-static int fastcall do_path_lookup(int dfd, const char *name,
+static int do_path_lookup(int dfd, const char *name,
                                unsigned int flags, struct nameidata *nd)
 {
        int retval = 0;
@@ -1113,21 +1132,21 @@ static int fastcall do_path_lookup(int dfd, const char *name,
 
        if (*name=='/') {
                read_lock(&fs->lock);
-               if (fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
-                       nd->mnt = mntget(fs->altrootmnt);
-                       nd->dentry = dget(fs->altroot);
+               if (fs->altroot.dentry && !(nd->flags & LOOKUP_NOALT)) {
+                       nd->path = fs->altroot;
+                       path_get(&fs->altroot);
                        read_unlock(&fs->lock);
                        if (__emul_lookup_dentry(name,nd))
                                goto out; /* found in altroot */
                        read_lock(&fs->lock);
                }
-               nd->mnt = mntget(fs->rootmnt);
-               nd->dentry = dget(fs->root);
+               nd->path = fs->root;
+               path_get(&fs->root);
                read_unlock(&fs->lock);
        } else if (dfd == AT_FDCWD) {
                read_lock(&fs->lock);
-               nd->mnt = mntget(fs->pwdmnt);
-               nd->dentry = dget(fs->pwd);
+               nd->path = fs->pwd;
+               path_get(&fs->pwd);
                read_unlock(&fs->lock);
        } else {
                struct dentry *dentry;
@@ -1147,17 +1166,17 @@ static int fastcall do_path_lookup(int dfd, const char *name,
                if (retval)
                        goto fput_fail;
 
-               nd->mnt = mntget(file->f_path.mnt);
-               nd->dentry = dget(dentry);
+               nd->path = file->f_path;
+               path_get(&file->f_path);
 
                fput_light(file, fput_needed);
        }
 
        retval = path_walk(name, nd);
 out:
-       if (unlikely(!retval && !audit_dummy_context() && 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;
 
@@ -1166,12 +1185,43 @@ fput_fail:
        goto out_fail;
 }
 
-int fastcall path_lookup(const char *name, unsigned int flags,
+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
+ */
+int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
+                   const char *name, unsigned int flags,
+                   struct nameidata *nd)
+{
+       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)
@@ -1188,7 +1238,7 @@ static int __path_lookup_intent_open(int dfd, const char *name,
        if (IS_ERR(nd->intent.open.file)) {
                if (err == 0) {
                        err = PTR_ERR(nd->intent.open.file);
-                       path_release(nd);
+                       path_put(&nd->path);
                }
        } else if (err != 0)
                release_open_intent(nd);
@@ -1240,7 +1290,8 @@ int __user_path_lookup_open(const char __user *name, unsigned int lookup_flags,
        return err;
 }
 
-static inline struct dentry *__lookup_hash_kern(struct qstr *name, struct dentry *base, struct nameidata *nd)
+static struct dentry *__lookup_hash(struct qstr *name,
+               struct dentry *base, struct nameidata *nd)
 {
        struct dentry *dentry;
        struct inode *inode;
@@ -1280,31 +1331,18 @@ out:
  * needs parent already locked. Doesn't follow mounts.
  * SMP-safe.
  */
-static inline struct dentry * __lookup_hash(struct qstr *name, struct dentry *base, struct nameidata *nd)
+static struct dentry *lookup_hash(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);
+       err = permission(nd->path.dentry->d_inode, MAY_EXEC, nd);
        if (err)
-               goto out;
-
-       dentry = __lookup_hash_kern(name, base, nd);
-out:
-       return dentry;
-}
-
-static struct dentry *lookup_hash(struct nameidata *nd)
-{
-       return __lookup_hash(&nd->last, nd->dentry, nd);
+               return ERR_PTR(err);
+       return __lookup_hash(&nd->last, nd->path.dentry, nd);
 }
 
-/* SMP-safe */
-static inline int __lookup_one_len(const char *name, struct qstr *this, struct dentry *base, int len)
+static int __lookup_one_len(const char *name, struct qstr *this,
+               struct dentry *base, int len)
 {
        unsigned long hash;
        unsigned int c;
@@ -1325,6 +1363,17 @@ static inline int __lookup_one_len(const char *name, struct qstr *this, struct d
        return 0;
 }
 
+/**
+ * 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;
@@ -1333,21 +1382,36 @@ struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
        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);
 }
 
-struct dentry *lookup_one_len_kern(const char *name, struct dentry *base, int len)
+/**
+ * lookup_one_noperm - bad hack for sysfs
+ * @name:      pathname component to lookup
+ * @base:      base directory to lookup from
+ *
+ * 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.
+ *
+ * DON'T USE THIS FUNCTION EVER, thanks.
+ */
+struct dentry *lookup_one_noperm(const char *name, struct dentry *base)
 {
        int err;
        struct qstr this;
 
-       err = __lookup_one_len(name, &this, base, len);
+       err = __lookup_one_len(name, &this, base, strlen(name));
        if (err)
                return ERR_PTR(err);
-       return __lookup_hash_kern(&this, base, NULL);
+       return __lookup_hash(&this, base, NULL);
 }
 
-int fastcall __user_walk_fd(int dfd, const char __user *name, unsigned flags,
+int __user_walk_fd(int dfd, const char __user *name, unsigned flags,
                            struct nameidata *nd)
 {
        char *tmp = getname(name);
@@ -1360,7 +1424,7 @@ int fastcall __user_walk_fd(int dfd, const char __user *name, unsigned flags,
        return err;
 }
 
-int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
+int __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
 {
        return __user_walk_fd(AT_FDCWD, name, flags, nd);
 }
@@ -1407,7 +1471,7 @@ static 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->d_inode, dir);
+       audit_inode_child(victim->d_name.name, victim, dir);
 
        error = permission(dir,MAY_WRITE | MAY_EXEC, NULL);
        if (error)
@@ -1533,7 +1597,7 @@ int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
 
 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;
 
@@ -1543,13 +1607,9 @@ 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 = vfs_permission(nd, acc_mode);
-       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
@@ -1558,12 +1618,16 @@ 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.
         */
@@ -1576,7 +1640,7 @@ 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;
 
        /*
@@ -1597,8 +1661,10 @@ 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, ATTR_MTIME|ATTR_CTIME, NULL);
+
+                       error = do_truncate(dentry, 0,
+                                           ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
+                                           NULL);
                }
                put_write_access(inode);
                if (error)
@@ -1614,14 +1680,14 @@ static int open_namei_create(struct nameidata *nd, struct path *path,
                                int flag, int mode)
 {
        int error;
-       struct dentry *dir = nd->dentry;
+       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->dentry);
-       nd->dentry = path->dentry;
+       dput(nd->path.dentry);
+       nd->path.dentry = path->dentry;
        if (error)
                return error;
        /* Don't check for write permission, don't truncate */
@@ -1688,11 +1754,11 @@ int open_namei(int dfd, const char *pathname, int flag,
        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;
        mutex_lock(&dir->d_inode->i_mutex);
        path.dentry = lookup_hash(nd);
-       path.mnt = nd->mnt;
+       path.mnt = nd->path.mnt;
 
 do_last:
        error = PTR_ERR(path.dentry);
@@ -1719,7 +1785,7 @@ do_last:
         * It already exists.
         */
        mutex_unlock(&dir->d_inode->i_mutex);
-       audit_inode(pathname, path.dentry->d_inode);
+       audit_inode(pathname, path.dentry);
 
        error = -EEXIST;
        if (flag & O_EXCL)
@@ -1748,11 +1814,11 @@ ok:
        return 0;
 
 exit_dput:
-       dput_path(&path, nd);
+       path_put_conditional(&path, nd);
 exit:
        if (!IS_ERR(nd->intent.open.file))
                release_open_intent(nd);
-       path_release(nd);
+       path_put(&nd->path);
        return error;
 
 do_link:
@@ -1797,10 +1863,10 @@ do_link:
                __putname(nd->last.name);
                goto exit;
        }
-       dir = nd->dentry;
+       dir = nd->path.dentry;
        mutex_lock(&dir->d_inode->i_mutex);
        path.dentry = lookup_hash(nd);
-       path.mnt = nd->mnt;
+       path.mnt = nd->path.mnt;
        __putname(nd->last.name);
        goto do_last;
 }
@@ -1813,13 +1879,13 @@ 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.
+ * Returns with nd->path.dentry->d_inode->i_mutex locked.
  */
 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
 {
        struct dentry *dentry = ERR_PTR(-EEXIST);
 
-       mutex_lock_nested(&nd->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
+       mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
        /*
         * Yucky last component or no last component at all?
         * (foo/., foo/.., /////)
@@ -1898,19 +1964,19 @@ asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
        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;
@@ -1920,8 +1986,8 @@ asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
                }
                dput(dentry);
        }
-       mutex_unlock(&nd.dentry->d_inode->i_mutex);
-       path_release(&nd);
+       mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
+       path_put(&nd.path);
 out:
        putname(tmp);
 
@@ -1975,13 +2041,13 @@ asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
        if (IS_ERR(dentry))
                goto out_unlock;
 
-       if (!IS_POSIXACL(nd.dentry->d_inode))
+       if (!IS_POSIXACL(nd.path.dentry->d_inode))
                mode &= ~current->fs->umask;
-       error = vfs_mkdir(nd.dentry->d_inode, dentry, mode);
+       error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode);
        dput(dentry);
 out_unlock:
-       mutex_unlock(&nd.dentry->d_inode->i_mutex);
-       path_release(&nd);
+       mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
+       path_put(&nd.path);
 out:
        putname(tmp);
 out_err:
@@ -2079,17 +2145,17 @@ static long do_rmdir(int dfd, const char __user *pathname)
                        error = -EBUSY;
                        goto exit1;
        }
-       mutex_lock_nested(&nd.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
+       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))
                goto exit2;
-       error = vfs_rmdir(nd.dentry->d_inode, dentry);
+       error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
        dput(dentry);
 exit2:
-       mutex_unlock(&nd.dentry->d_inode->i_mutex);
+       mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
 exit1:
-       path_release(&nd);
+       path_put(&nd.path);
 exit:
        putname(name);
        return error;
@@ -2124,6 +2190,7 @@ int vfs_unlink(struct inode *dir, struct dentry *dentry)
 
        /* 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);
        }
 
@@ -2154,7 +2221,7 @@ static long do_unlinkat(int dfd, const char __user *pathname)
        error = -EISDIR;
        if (nd.last_type != LAST_NORM)
                goto exit1;
-       mutex_lock_nested(&nd.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
+       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)) {
@@ -2164,15 +2231,15 @@ static long do_unlinkat(int dfd, 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);
        }
-       mutex_unlock(&nd.dentry->d_inode->i_mutex);
+       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;
@@ -2245,11 +2312,11 @@ asmlinkage long sys_symlinkat(const char __user *oldname,
        if (IS_ERR(dentry))
                goto out_unlock;
 
-       error = vfs_symlink(nd.dentry->d_inode, dentry, from, S_IALLUGO);
+       error = vfs_symlink(nd.path.dentry->d_inode, dentry, from, S_IALLUGO);
        dput(dentry);
 out_unlock:
-       mutex_unlock(&nd.dentry->d_inode->i_mutex);
-       path_release(&nd);
+       mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
+       path_put(&nd.path);
 out:
        putname(to);
 out_putname:
@@ -2296,7 +2363,7 @@ int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_de
        error = dir->i_op->link(old_dentry, dir, new_dentry);
        mutex_unlock(&old_dentry->d_inode->i_mutex);
        if (!error)
-               fsnotify_create(dir, new_dentry);
+               fsnotify_link(dir, old_dentry->d_inode, new_dentry);
        return error;
 }
 
@@ -2334,20 +2401,20 @@ asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
        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))
                goto out_unlock;
-       error = vfs_link(old_nd.dentry, nd.dentry->d_inode, new_dentry);
+       error = vfs_link(old_nd.path.dentry, nd.path.dentry->d_inode, new_dentry);
        dput(new_dentry);
 out_unlock:
-       mutex_unlock(&nd.dentry->d_inode->i_mutex);
+       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);
 
@@ -2498,7 +2565,7 @@ int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
        if (!error) {
                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);
+                             new_dentry->d_inode, old_dentry);
        }
        fsnotify_oldname_free(old_name);
 
@@ -2523,15 +2590,15 @@ static int do_rename(int olddfd, const char *oldname,
                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;
 
@@ -2575,9 +2642,9 @@ 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;
 }
@@ -2696,53 +2763,29 @@ int __page_symlink(struct inode *inode, const char *symname, int len,
 {
        struct address_space *mapping = inode->i_mapping;
        struct page *page;
+       void *fsdata;
        int err;
        char *kaddr;
 
 retry:
-       err = -ENOMEM;
-       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;
-       }
+       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);
-       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
-        * 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);
-               if (err != AOP_TRUNCATED_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;
 }
@@ -2774,8 +2817,7 @@ 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);