ext3: Don't update superblock write time when filesystem is read-only
[safe/jmp/linux-2.6] / fs / reiserfs / xattr_acl.c
index c312881..35d6e67 100644 (file)
@@ -1,3 +1,4 @@
+#include <linux/capability.h>
 #include <linux/fs.h>
 #include <linux/posix_acl.h>
 #include <linux/reiserfs_fs.h>
 #include <linux/reiserfs_acl.h>
 #include <asm/uaccess.h>
 
-static int reiserfs_set_acl(struct inode *inode, int type, struct posix_acl *acl);
+static int reiserfs_set_acl(struct reiserfs_transaction_handle *th,
+                           struct inode *inode, int type,
+                           struct posix_acl *acl);
 
 static int
 xattr_set_acl(struct inode *inode, int type, const void *value, size_t size)
 {
        struct posix_acl *acl;
-       int error;
-
+       int error, error2;
+       struct reiserfs_transaction_handle th;
+       size_t jcreate_blocks;
        if (!reiserfs_posixacl(inode->i_sb))
                return -EOPNOTSUPP;
-       if ((current->fsuid != inode->i_uid) && !capable(CAP_FOWNER))
+       if (!is_owner_or_cap(inode))
                return -EPERM;
 
        if (value) {
@@ -34,14 +38,27 @@ xattr_set_acl(struct inode *inode, int type, const void *value, size_t size)
        } else
                acl = NULL;
 
-       error = reiserfs_set_acl (inode, type, acl);
+       /* Pessimism: We can't assume that anything from the xattr root up
+        * has been created. */
+
+       jcreate_blocks = reiserfs_xattr_jcreate_nblocks(inode) +
+                        reiserfs_xattr_nblocks(inode, size) * 2;
 
-release_and_out:
+       reiserfs_write_lock(inode->i_sb);
+       error = journal_begin(&th, inode->i_sb, jcreate_blocks);
+       if (error == 0) {
+               error = reiserfs_set_acl(&th, inode, type, acl);
+               error2 = journal_end(&th, inode->i_sb, jcreate_blocks);
+               if (error2)
+                       error = error2;
+       }
+       reiserfs_write_unlock(inode->i_sb);
+
+      release_and_out:
        posix_acl_release(acl);
        return error;
 }
 
-
 static int
 xattr_get_acl(struct inode *inode, int type, void *buffer, size_t size)
 {
@@ -51,7 +68,7 @@ xattr_get_acl(struct inode *inode, int type, void *buffer, size_t size)
        if (!reiserfs_posixacl(inode->i_sb))
                return -EOPNOTSUPP;
 
-       acl = reiserfs_get_acl (inode, type);
+       acl = reiserfs_get_acl(inode, type);
        if (IS_ERR(acl))
                return PTR_ERR(acl);
        if (acl == NULL)
@@ -62,12 +79,10 @@ xattr_get_acl(struct inode *inode, int type, void *buffer, size_t size)
        return error;
 }
 
-
 /*
  * Convert from filesystem to in-memory representation.
  */
-static struct posix_acl *
-posix_acl_from_disk(const void *value, size_t size)
+static struct posix_acl *posix_acl_from_disk(const void *value, size_t size)
 {
        const char *end = (char *)value + size;
        int n, count;
@@ -76,8 +91,8 @@ posix_acl_from_disk(const void *value, size_t size)
        if (!value)
                return NULL;
        if (size < sizeof(reiserfs_acl_header))
-                return ERR_PTR(-EINVAL);
-       if (((reiserfs_acl_header *)value)->a_version !=
+               return ERR_PTR(-EINVAL);
+       if (((reiserfs_acl_header *) value)->a_version !=
            cpu_to_le32(REISERFS_ACL_VERSION))
                return ERR_PTR(-EINVAL);
        value = (char *)value + sizeof(reiserfs_acl_header);
@@ -89,41 +104,39 @@ posix_acl_from_disk(const void *value, size_t size)
        acl = posix_acl_alloc(count, GFP_NOFS);
        if (!acl)
                return ERR_PTR(-ENOMEM);
-       for (n=0; n < count; n++) {
-               reiserfs_acl_entry *entry =
-                       (reiserfs_acl_entry *)value;
+       for (n = 0; n < count; n++) {
+               reiserfs_acl_entry *entry = (reiserfs_acl_entry *) value;
                if ((char *)value + sizeof(reiserfs_acl_entry_short) > end)
                        goto fail;
-               acl->a_entries[n].e_tag  = le16_to_cpu(entry->e_tag);
+               acl->a_entries[n].e_tag = le16_to_cpu(entry->e_tag);
                acl->a_entries[n].e_perm = le16_to_cpu(entry->e_perm);
-               switch(acl->a_entries[n].e_tag) {
-                       case ACL_USER_OBJ:
-                       case ACL_GROUP_OBJ:
-                       case ACL_MASK:
-                       case ACL_OTHER:
-                               value = (char *)value +
-                                       sizeof(reiserfs_acl_entry_short);
-                               acl->a_entries[n].e_id = ACL_UNDEFINED_ID;
-                               break;
-
-                       case ACL_USER:
-                       case ACL_GROUP:
-                               value = (char *)value + sizeof(reiserfs_acl_entry);
-                               if ((char *)value > end)
-                                       goto fail;
-                               acl->a_entries[n].e_id =
-                                       le32_to_cpu(entry->e_id);
-                               break;
-
-                       default:
+               switch (acl->a_entries[n].e_tag) {
+               case ACL_USER_OBJ:
+               case ACL_GROUP_OBJ:
+               case ACL_MASK:
+               case ACL_OTHER:
+                       value = (char *)value +
+                           sizeof(reiserfs_acl_entry_short);
+                       acl->a_entries[n].e_id = ACL_UNDEFINED_ID;
+                       break;
+
+               case ACL_USER:
+               case ACL_GROUP:
+                       value = (char *)value + sizeof(reiserfs_acl_entry);
+                       if ((char *)value > end)
                                goto fail;
+                       acl->a_entries[n].e_id = le32_to_cpu(entry->e_id);
+                       break;
+
+               default:
+                       goto fail;
                }
        }
        if (value != end)
                goto fail;
        return acl;
 
-fail:
+      fail:
        posix_acl_release(acl);
        return ERR_PTR(-EINVAL);
 }
@@ -131,46 +144,46 @@ fail:
 /*
  * Convert from in-memory to filesystem representation.
  */
-static void *
-posix_acl_to_disk(const struct posix_acl *acl, size_t *size)
+static void *posix_acl_to_disk(const struct posix_acl *acl, size_t * size)
 {
        reiserfs_acl_header *ext_acl;
        char *e;
        int n;
 
        *size = reiserfs_acl_size(acl->a_count);
-       ext_acl = (reiserfs_acl_header *)kmalloc(sizeof(reiserfs_acl_header) +
-               acl->a_count * sizeof(reiserfs_acl_entry), GFP_NOFS);
+       ext_acl = kmalloc(sizeof(reiserfs_acl_header) +
+                                                 acl->a_count *
+                                                 sizeof(reiserfs_acl_entry),
+                                                 GFP_NOFS);
        if (!ext_acl)
                return ERR_PTR(-ENOMEM);
        ext_acl->a_version = cpu_to_le32(REISERFS_ACL_VERSION);
        e = (char *)ext_acl + sizeof(reiserfs_acl_header);
-       for (n=0; n < acl->a_count; n++) {
-               reiserfs_acl_entry *entry = (reiserfs_acl_entry *)e;
-               entry->e_tag  = cpu_to_le16(acl->a_entries[n].e_tag);
+       for (n = 0; n < acl->a_count; n++) {
+               reiserfs_acl_entry *entry = (reiserfs_acl_entry *) e;
+               entry->e_tag = cpu_to_le16(acl->a_entries[n].e_tag);
                entry->e_perm = cpu_to_le16(acl->a_entries[n].e_perm);
-               switch(acl->a_entries[n].e_tag) {
-                       case ACL_USER:
-                       case ACL_GROUP:
-                               entry->e_id =
-                                       cpu_to_le32(acl->a_entries[n].e_id);
-                               e += sizeof(reiserfs_acl_entry);
-                               break;
-
-                       case ACL_USER_OBJ:
-                       case ACL_GROUP_OBJ:
-                       case ACL_MASK:
-                       case ACL_OTHER:
-                               e += sizeof(reiserfs_acl_entry_short);
-                               break;
-
-                       default:
-                               goto fail;
+               switch (acl->a_entries[n].e_tag) {
+               case ACL_USER:
+               case ACL_GROUP:
+                       entry->e_id = cpu_to_le32(acl->a_entries[n].e_id);
+                       e += sizeof(reiserfs_acl_entry);
+                       break;
+
+               case ACL_USER_OBJ:
+               case ACL_GROUP_OBJ:
+               case ACL_MASK:
+               case ACL_OTHER:
+                       e += sizeof(reiserfs_acl_entry_short);
+                       break;
+
+               default:
+                       goto fail;
                }
        }
        return (char *)ext_acl;
 
-fail:
+      fail:
        kfree(ext_acl);
        return ERR_PTR(-EINVAL);
 }
@@ -178,62 +191,56 @@ fail:
 /*
  * Inode operation get_posix_acl().
  *
- * inode->i_sem: down
+ * inode->i_mutex: down
  * BKL held [before 2.5.x]
  */
-struct posix_acl *
-reiserfs_get_acl(struct inode *inode, int type)
+struct posix_acl *reiserfs_get_acl(struct inode *inode, int type)
 {
        char *name, *value;
-       struct posix_acl *acl, **p_acl;
-       size_t size;
+       struct posix_acl *acl;
+       int size;
        int retval;
-        struct reiserfs_inode_info *reiserfs_i = REISERFS_I(inode);
-
-        switch (type) {
-            case ACL_TYPE_ACCESS:
-                name = POSIX_ACL_XATTR_ACCESS;
-                p_acl = &reiserfs_i->i_acl_access;
-                break;
-            case ACL_TYPE_DEFAULT:
-                name = POSIX_ACL_XATTR_DEFAULT;
-                p_acl = &reiserfs_i->i_acl_default;
-                break;
-            default:
-                return ERR_PTR (-EINVAL);
-        }
-
-        if (IS_ERR (*p_acl)) {
-            if (PTR_ERR (*p_acl) == -ENODATA)
-                return NULL;
-        } else if (*p_acl != NULL)
-            return posix_acl_dup (*p_acl);
-
-        size = reiserfs_xattr_get (inode, name, NULL, 0);
-        if ((int)size < 0) {
-            if (size == -ENODATA || size == -ENOSYS) {
-               *p_acl = ERR_PTR (-ENODATA);
-               return NULL;
-            }
-            return ERR_PTR (size);
-        }
 
-        value = kmalloc (size, GFP_NOFS);
-        if (!value)
-            return ERR_PTR (-ENOMEM);
+       acl = get_cached_acl(inode, type);
+       if (acl != ACL_NOT_CACHED)
+               return acl;
+
+       switch (type) {
+       case ACL_TYPE_ACCESS:
+               name = POSIX_ACL_XATTR_ACCESS;
+               break;
+       case ACL_TYPE_DEFAULT:
+               name = POSIX_ACL_XATTR_DEFAULT;
+               break;
+       default:
+               BUG();
+       }
+
+       size = reiserfs_xattr_get(inode, name, NULL, 0);
+       if (size < 0) {
+               if (size == -ENODATA || size == -ENOSYS) {
+                       set_cached_acl(inode, type, NULL);
+                       return NULL;
+               }
+               return ERR_PTR(size);
+       }
+
+       value = kmalloc(size, GFP_NOFS);
+       if (!value)
+               return ERR_PTR(-ENOMEM);
 
        retval = reiserfs_xattr_get(inode, name, value, size);
        if (retval == -ENODATA || retval == -ENOSYS) {
                /* This shouldn't actually happen as it should have
                   been caught above.. but just in case */
                acl = NULL;
-               *p_acl = ERR_PTR (-ENODATA);
-        } else if (retval < 0) {
+       } else if (retval < 0) {
                acl = ERR_PTR(retval);
        } else {
                acl = posix_acl_from_disk(value, retval);
-               *p_acl = posix_acl_dup (acl);
-        }
+       }
+       if (!IS_ERR(acl))
+               set_cached_acl(inode, type, acl);
 
        kfree(value);
        return acl;
@@ -242,330 +249,309 @@ reiserfs_get_acl(struct inode *inode, int type)
 /*
  * Inode operation set_posix_acl().
  *
- * inode->i_sem: down
+ * inode->i_mutex: down
  * BKL held [before 2.5.x]
  */
 static int
-reiserfs_set_acl(struct inode *inode, int type, struct posix_acl *acl)
+reiserfs_set_acl(struct reiserfs_transaction_handle *th, struct inode *inode,
+                int type, struct posix_acl *acl)
 {
-        char *name;
+       char *name;
        void *value = NULL;
-       struct posix_acl **p_acl;
-       size_t size;
+       size_t size = 0;
        int error;
-        struct reiserfs_inode_info *reiserfs_i = REISERFS_I(inode);
 
        if (S_ISLNK(inode->i_mode))
                return -EOPNOTSUPP;
 
-        switch (type) {
-            case ACL_TYPE_ACCESS:
-                name = POSIX_ACL_XATTR_ACCESS;
-                p_acl = &reiserfs_i->i_acl_access;
-                if (acl) {
-                    mode_t mode = inode->i_mode;
-                    error = posix_acl_equiv_mode (acl, &mode);
-                    if (error < 0)
-                        return error;
-                    else {
-                        inode->i_mode = mode;
-                        if (error == 0)
-                            acl = NULL;
-                    }
-                }
-                break;
-            case ACL_TYPE_DEFAULT:
-                name = POSIX_ACL_XATTR_DEFAULT;
-                p_acl = &reiserfs_i->i_acl_default;
-                if (!S_ISDIR (inode->i_mode))
-                    return acl ? -EACCES : 0;
-                break;
-            default:
-                return -EINVAL;
-        }
-
-       if (acl) {
-            value = posix_acl_to_disk(acl, &size);
-            if (IS_ERR(value))
-                return (int)PTR_ERR(value);
-            error = reiserfs_xattr_set(inode, name, value, size, 0);
-       } else {
-            error = reiserfs_xattr_del (inode, name);
-            if (error == -ENODATA) {
-                /* This may seem odd here, but it means that the ACL was set
-                 * with a value representable with mode bits. If there was
-                 * an ACL before, reiserfs_xattr_del already dirtied the inode.
-                 */
-                mark_inode_dirty (inode);
-                error = 0;
-            }
-        }
-
-       if (value)
-               kfree(value);
-
-        if (!error) {
-            /* Release the old one */
-            if (!IS_ERR (*p_acl) && *p_acl)
-                posix_acl_release (*p_acl);
-
-            if (acl == NULL)
-                *p_acl = ERR_PTR (-ENODATA);
-            else
-                *p_acl = posix_acl_dup (acl);
-        }
+       switch (type) {
+       case ACL_TYPE_ACCESS:
+               name = POSIX_ACL_XATTR_ACCESS;
+               if (acl) {
+                       mode_t mode = inode->i_mode;
+                       error = posix_acl_equiv_mode(acl, &mode);
+                       if (error < 0)
+                               return error;
+                       else {
+                               inode->i_mode = mode;
+                               if (error == 0)
+                                       acl = NULL;
+                       }
+               }
+               break;
+       case ACL_TYPE_DEFAULT:
+               name = POSIX_ACL_XATTR_DEFAULT;
+               if (!S_ISDIR(inode->i_mode))
+                       return acl ? -EACCES : 0;
+               break;
+       default:
+               return -EINVAL;
+       }
+
+       if (acl) {
+               value = posix_acl_to_disk(acl, &size);
+               if (IS_ERR(value))
+                       return (int)PTR_ERR(value);
+       }
+
+       error = reiserfs_xattr_set_handle(th, inode, name, value, size, 0);
+
+       /*
+        * Ensure that the inode gets dirtied if we're only using
+        * the mode bits and an old ACL didn't exist. We don't need
+        * to check if the inode is hashed here since we won't get
+        * called by reiserfs_inherit_default_acl().
+        */
+       if (error == -ENODATA) {
+               error = 0;
+               if (type == ACL_TYPE_ACCESS) {
+                       inode->i_ctime = CURRENT_TIME_SEC;
+                       mark_inode_dirty(inode);
+               }
+       }
+
+       kfree(value);
+
+       if (!error)
+               set_cached_acl(inode, type, acl);
 
        return error;
 }
 
-/* dir->i_sem: down,
+/* dir->i_mutex: locked,
  * inode is new and not released into the wild yet */
 int
-reiserfs_inherit_default_acl (struct inode *dir, struct dentry *dentry, struct inode *inode)
+reiserfs_inherit_default_acl(struct reiserfs_transaction_handle *th,
+                            struct inode *dir, struct dentry *dentry,
+                            struct inode *inode)
 {
-    struct posix_acl *acl;
-    int err = 0;
-
-    /* ACLs only get applied to files and directories */
-    if (S_ISLNK (inode->i_mode))
-        return 0;
-
-    /* ACLs can only be used on "new" objects, so if it's an old object
-     * there is nothing to inherit from */
-    if (get_inode_sd_version (dir) == STAT_DATA_V1)
-        goto apply_umask;
-
-    /* Don't apply ACLs to objects in the .reiserfs_priv tree.. This
-     * would be useless since permissions are ignored, and a pain because
-     * it introduces locking cycles */
-    if (is_reiserfs_priv_object (dir)) {
-        reiserfs_mark_inode_private (inode);
-        goto apply_umask;
-    }
-
-    acl = reiserfs_get_acl (dir, ACL_TYPE_DEFAULT);
-    if (IS_ERR (acl)) {
-        if (PTR_ERR (acl) == -ENODATA)
-            goto apply_umask;
-        return PTR_ERR (acl);
-    }
-
-    if (acl) {
-        struct posix_acl *acl_copy;
-        mode_t mode = inode->i_mode;
-        int need_acl;
-
-        /* Copy the default ACL to the default ACL of a new directory */
-        if (S_ISDIR (inode->i_mode)) {
-            err = reiserfs_set_acl (inode, ACL_TYPE_DEFAULT, acl);
-            if (err)
-                goto cleanup;
-        }
-
-        /* Now we reconcile the new ACL and the mode,
-           potentially modifying both */
-        acl_copy = posix_acl_clone (acl, GFP_NOFS);
-        if (!acl_copy) {
-            err = -ENOMEM;
-            goto cleanup;
-        }
-
-
-        need_acl = posix_acl_create_masq (acl_copy, &mode);
-        if (need_acl >= 0) {
-            if (mode != inode->i_mode) {
-                inode->i_mode = mode;
-            }
-
-            /* If we need an ACL.. */
-            if (need_acl > 0) {
-                err = reiserfs_set_acl (inode, ACL_TYPE_ACCESS, acl_copy);
-                if (err)
-                    goto cleanup_copy;
-            }
-        }
-cleanup_copy:
-        posix_acl_release (acl_copy);
-cleanup:
-        posix_acl_release (acl);
-    } else {
-apply_umask:
-        /* no ACL, apply umask */
-        inode->i_mode &= ~current->fs->umask;
-    }
-
-    return err;
+       struct posix_acl *acl;
+       int err = 0;
+
+       /* ACLs only get applied to files and directories */
+       if (S_ISLNK(inode->i_mode))
+               return 0;
+
+       /* ACLs can only be used on "new" objects, so if it's an old object
+        * there is nothing to inherit from */
+       if (get_inode_sd_version(dir) == STAT_DATA_V1)
+               goto apply_umask;
+
+       /* Don't apply ACLs to objects in the .reiserfs_priv tree.. This
+        * would be useless since permissions are ignored, and a pain because
+        * it introduces locking cycles */
+       if (IS_PRIVATE(dir)) {
+               inode->i_flags |= S_PRIVATE;
+               goto apply_umask;
+       }
+
+       acl = reiserfs_get_acl(dir, ACL_TYPE_DEFAULT);
+       if (IS_ERR(acl))
+               return PTR_ERR(acl);
+
+       if (acl) {
+               struct posix_acl *acl_copy;
+               mode_t mode = inode->i_mode;
+               int need_acl;
+
+               /* Copy the default ACL to the default ACL of a new directory */
+               if (S_ISDIR(inode->i_mode)) {
+                       err = reiserfs_set_acl(th, inode, ACL_TYPE_DEFAULT,
+                                              acl);
+                       if (err)
+                               goto cleanup;
+               }
+
+               /* Now we reconcile the new ACL and the mode,
+                  potentially modifying both */
+               acl_copy = posix_acl_clone(acl, GFP_NOFS);
+               if (!acl_copy) {
+                       err = -ENOMEM;
+                       goto cleanup;
+               }
+
+               need_acl = posix_acl_create_masq(acl_copy, &mode);
+               if (need_acl >= 0) {
+                       if (mode != inode->i_mode) {
+                               inode->i_mode = mode;
+                       }
+
+                       /* If we need an ACL.. */
+                       if (need_acl > 0) {
+                               err = reiserfs_set_acl(th, inode,
+                                                      ACL_TYPE_ACCESS,
+                                                      acl_copy);
+                               if (err)
+                                       goto cleanup_copy;
+                       }
+               }
+             cleanup_copy:
+               posix_acl_release(acl_copy);
+             cleanup:
+               posix_acl_release(acl);
+       } else {
+             apply_umask:
+               /* no ACL, apply umask */
+               inode->i_mode &= ~current_umask();
+       }
+
+       return err;
 }
 
-/* Looks up and caches the result of the default ACL.
- * We do this so that we don't need to carry the xattr_sem into
- * reiserfs_new_inode if we don't need to */
-int
-reiserfs_cache_default_acl (struct inode *inode)
+/* This is used to cache the default acl before a new object is created.
+ * The biggest reason for this is to get an idea of how many blocks will
+ * actually be required for the create operation if we must inherit an ACL.
+ * An ACL write can add up to 3 object creations and an additional file write
+ * so we'd prefer not to reserve that many blocks in the journal if we can.
+ * It also has the advantage of not loading the ACL with a transaction open,
+ * this may seem silly, but if the owner of the directory is doing the
+ * creation, the ACL may not be loaded since the permissions wouldn't require
+ * it.
+ * We return the number of blocks required for the transaction.
+ */
+int reiserfs_cache_default_acl(struct inode *inode)
 {
-    int ret = 0;
-    if (reiserfs_posixacl (inode->i_sb) &&
-        !is_reiserfs_priv_object (inode)) {
-        struct posix_acl *acl;
-        reiserfs_read_lock_xattr_i (inode);
-        reiserfs_read_lock_xattrs (inode->i_sb);
-        acl = reiserfs_get_acl (inode, ACL_TYPE_DEFAULT);
-        reiserfs_read_unlock_xattrs (inode->i_sb);
-        reiserfs_read_unlock_xattr_i (inode);
-        ret = acl ? 1 : 0;
-        posix_acl_release (acl);
-    }
-
-    return ret;
+       struct posix_acl *acl;
+       int nblocks = 0;
+
+       if (IS_PRIVATE(inode))
+               return 0;
+
+       acl = reiserfs_get_acl(inode, ACL_TYPE_DEFAULT);
+
+       if (acl && !IS_ERR(acl)) {
+               int size = reiserfs_acl_size(acl->a_count);
+
+               /* Other xattrs can be created during inode creation. We don't
+                * want to claim too many blocks, so we check to see if we
+                * we need to create the tree to the xattrs, and then we
+                * just want two files. */
+               nblocks = reiserfs_xattr_jcreate_nblocks(inode);
+               nblocks += JOURNAL_BLOCKS_PER_OBJECT(inode->i_sb);
+
+               REISERFS_I(inode)->i_flags |= i_has_xattr_dir;
+
+               /* We need to account for writes + bitmaps for two files */
+               nblocks += reiserfs_xattr_nblocks(inode, size) * 4;
+               posix_acl_release(acl);
+       }
+
+       return nblocks;
 }
 
-int
-reiserfs_acl_chmod (struct inode *inode)
+int reiserfs_acl_chmod(struct inode *inode)
 {
-        struct posix_acl *acl, *clone;
-        int error;
+       struct posix_acl *acl, *clone;
+       int error;
 
-        if (S_ISLNK(inode->i_mode))
-                return -EOPNOTSUPP;
+       if (S_ISLNK(inode->i_mode))
+               return -EOPNOTSUPP;
 
-       if (get_inode_sd_version (inode) == STAT_DATA_V1 ||
-           !reiserfs_posixacl(inode->i_sb))
-        {
-           return 0;
+       if (get_inode_sd_version(inode) == STAT_DATA_V1 ||
+           !reiserfs_posixacl(inode->i_sb)) {
+               return 0;
        }
 
-        reiserfs_read_lock_xattrs (inode->i_sb);
-        acl = reiserfs_get_acl(inode, ACL_TYPE_ACCESS);
-        reiserfs_read_unlock_xattrs (inode->i_sb);
-        if (!acl)
-                return 0;
-        if (IS_ERR(acl))
-                return PTR_ERR(acl);
-        clone = posix_acl_clone(acl, GFP_NOFS);
-        posix_acl_release(acl);
-        if (!clone)
-                return -ENOMEM;
-        error = posix_acl_chmod_masq(clone, inode->i_mode);
-        if (!error) {
-                int lock = !has_xattr_dir (inode);
-                reiserfs_write_lock_xattr_i (inode);
-                if (lock)
-                    reiserfs_write_lock_xattrs (inode->i_sb);
-                else
-                    reiserfs_read_lock_xattrs (inode->i_sb);
-                error = reiserfs_set_acl(inode, ACL_TYPE_ACCESS, clone);
-                if (lock)
-                    reiserfs_write_unlock_xattrs (inode->i_sb);
-                else
-                    reiserfs_read_unlock_xattrs (inode->i_sb);
-                reiserfs_write_unlock_xattr_i (inode);
-        }
-        posix_acl_release(clone);
-        return error;
+       acl = reiserfs_get_acl(inode, ACL_TYPE_ACCESS);
+       if (!acl)
+               return 0;
+       if (IS_ERR(acl))
+               return PTR_ERR(acl);
+       clone = posix_acl_clone(acl, GFP_NOFS);
+       posix_acl_release(acl);
+       if (!clone)
+               return -ENOMEM;
+       error = posix_acl_chmod_masq(clone, inode->i_mode);
+       if (!error) {
+               struct reiserfs_transaction_handle th;
+               size_t size = reiserfs_xattr_nblocks(inode,
+                                            reiserfs_acl_size(clone->a_count));
+               reiserfs_write_lock(inode->i_sb);
+               error = journal_begin(&th, inode->i_sb, size * 2);
+               if (!error) {
+                       int error2;
+                       error = reiserfs_set_acl(&th, inode, ACL_TYPE_ACCESS,
+                                                clone);
+                       error2 = journal_end(&th, inode->i_sb, size * 2);
+                       if (error2)
+                               error = error2;
+               }
+               reiserfs_write_unlock(inode->i_sb);
+       }
+       posix_acl_release(clone);
+       return error;
 }
 
 static int
 posix_acl_access_get(struct inode *inode, const char *name,
-                         void *buffer, size_t size)
+                    void *buffer, size_t size)
 {
-       if (strlen(name) != sizeof(POSIX_ACL_XATTR_ACCESS)-1)
+       if (strlen(name) != sizeof(POSIX_ACL_XATTR_ACCESS) - 1)
                return -EINVAL;
        return xattr_get_acl(inode, ACL_TYPE_ACCESS, buffer, size);
 }
 
 static int
 posix_acl_access_set(struct inode *inode, const char *name,
-                         const void *value, size_t size, int flags)
+                    const void *value, size_t size, int flags)
 {
-       if (strlen(name) != sizeof(POSIX_ACL_XATTR_ACCESS)-1)
+       if (strlen(name) != sizeof(POSIX_ACL_XATTR_ACCESS) - 1)
                return -EINVAL;
        return xattr_set_acl(inode, ACL_TYPE_ACCESS, value, size);
 }
 
-static int
-posix_acl_access_del (struct inode *inode, const char *name)
+static size_t posix_acl_access_list(struct inode *inode, char *list,
+                                   size_t list_size, const char *name,
+                                   size_t name_len)
 {
-    struct reiserfs_inode_info *reiserfs_i = REISERFS_I(inode);
-    struct posix_acl **acl = &reiserfs_i->i_acl_access;
-    if (strlen(name) != sizeof(POSIX_ACL_XATTR_ACCESS)-1)
-       return -EINVAL;
-    if (!IS_ERR (*acl) && *acl) {
-        posix_acl_release (*acl);
-        *acl = ERR_PTR (-ENODATA);
-    }
-
-    return 0;
-}
-
-static int
-posix_acl_access_list (struct inode *inode, const char *name, int namelen, char *out)
-{
-    int len = namelen;
-    if (!reiserfs_posixacl (inode->i_sb))
-        return 0;
-    if (out)
-        memcpy (out, name, len);
-
-    return len;
+       const size_t size = sizeof(POSIX_ACL_XATTR_ACCESS);
+       if (!reiserfs_posixacl(inode->i_sb))
+               return 0;
+       if (list && size <= list_size)
+               memcpy(list, POSIX_ACL_XATTR_ACCESS, size);
+       return size;
 }
 
-struct reiserfs_xattr_handler posix_acl_access_handler = {
+struct xattr_handler reiserfs_posix_acl_access_handler = {
        .prefix = POSIX_ACL_XATTR_ACCESS,
        .get = posix_acl_access_get,
        .set = posix_acl_access_set,
-       .del = posix_acl_access_del,
        .list = posix_acl_access_list,
 };
 
 static int
-posix_acl_default_get (struct inode *inode, const char *name,
-                          void *buffer, size_t size)
+posix_acl_default_get(struct inode *inode, const char *name,
+                     void *buffer, size_t size)
 {
-       if (strlen(name) != sizeof(POSIX_ACL_XATTR_DEFAULT)-1)
+       if (strlen(name) != sizeof(POSIX_ACL_XATTR_DEFAULT) - 1)
                return -EINVAL;
        return xattr_get_acl(inode, ACL_TYPE_DEFAULT, buffer, size);
 }
 
 static int
 posix_acl_default_set(struct inode *inode, const char *name,
-                          const void *value, size_t size, int flags)
+                     const void *value, size_t size, int flags)
 {
-       if (strlen(name) != sizeof(POSIX_ACL_XATTR_DEFAULT)-1)
+       if (strlen(name) != sizeof(POSIX_ACL_XATTR_DEFAULT) - 1)
                return -EINVAL;
        return xattr_set_acl(inode, ACL_TYPE_DEFAULT, value, size);
 }
 
-static int
-posix_acl_default_del (struct inode *inode, const char *name)
-{
-    struct reiserfs_inode_info *reiserfs_i = REISERFS_I(inode);
-    struct posix_acl **acl = &reiserfs_i->i_acl_default;
-    if (strlen(name) != sizeof(POSIX_ACL_XATTR_DEFAULT)-1)
-       return -EINVAL;
-    if (!IS_ERR (*acl) && *acl) {
-        posix_acl_release (*acl);
-        *acl = ERR_PTR (-ENODATA);
-    }
-
-    return 0;
-}
-
-static int
-posix_acl_default_list (struct inode *inode, const char *name, int namelen, char *out)
+static size_t posix_acl_default_list(struct inode *inode, char *list,
+                                    size_t list_size, const char *name,
+                                    size_t name_len)
 {
-    int len = namelen;
-    if (!reiserfs_posixacl (inode->i_sb))
-        return 0;
-    if (out)
-        memcpy (out, name, len);
-
-    return len;
+       const size_t size = sizeof(POSIX_ACL_XATTR_DEFAULT);
+       if (!reiserfs_posixacl(inode->i_sb))
+               return 0;
+       if (list && size <= list_size)
+               memcpy(list, POSIX_ACL_XATTR_DEFAULT, size);
+       return size;
 }
 
-struct reiserfs_xattr_handler posix_acl_default_handler = {
+struct xattr_handler reiserfs_posix_acl_default_handler = {
        .prefix = POSIX_ACL_XATTR_DEFAULT,
        .get = posix_acl_default_get,
        .set = posix_acl_default_set,
-       .del = posix_acl_default_del,
        .list = posix_acl_default_list,
 };