ocfs2: add IO error check in ocfs2_get_sector()
[safe/jmp/linux-2.6] / fs / ocfs2 / xattr.c
index 7a90892..4ddd788 100644 (file)
 #include <linux/init.h>
 #include <linux/module.h>
 #include <linux/string.h>
+#include <linux/security.h>
 
 #define MLOG_MASK_PREFIX ML_XATTR
 #include <cluster/masklog.h>
 
 #include "ocfs2.h"
 #include "alloc.h"
+#include "blockcheck.h"
 #include "dlmglue.h"
 #include "file.h"
 #include "symlink.h"
@@ -80,6 +82,14 @@ struct ocfs2_xattr_set_ctxt {
 
 #define OCFS2_XATTR_ROOT_SIZE  (sizeof(struct ocfs2_xattr_def_value_root))
 #define OCFS2_XATTR_INLINE_SIZE        80
+#define OCFS2_XATTR_HEADER_GAP 4
+#define OCFS2_XATTR_FREE_IN_IBODY      (OCFS2_MIN_XATTR_INLINE_SIZE \
+                                        - sizeof(struct ocfs2_xattr_header) \
+                                        - OCFS2_XATTR_HEADER_GAP)
+#define OCFS2_XATTR_FREE_IN_BLOCK(ptr) ((ptr)->i_sb->s_blocksize \
+                                        - sizeof(struct ocfs2_xattr_block) \
+                                        - sizeof(struct ocfs2_xattr_header) \
+                                        - OCFS2_XATTR_HEADER_GAP)
 
 static struct ocfs2_xattr_def_value_root def_xv = {
        .xv.xr_list.l_count = cpu_to_le16(1),
@@ -87,13 +97,25 @@ static struct ocfs2_xattr_def_value_root def_xv = {
 
 struct xattr_handler *ocfs2_xattr_handlers[] = {
        &ocfs2_xattr_user_handler,
+#ifdef CONFIG_OCFS2_FS_POSIX_ACL
+       &ocfs2_xattr_acl_access_handler,
+       &ocfs2_xattr_acl_default_handler,
+#endif
        &ocfs2_xattr_trusted_handler,
+       &ocfs2_xattr_security_handler,
        NULL
 };
 
 static struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
        [OCFS2_XATTR_INDEX_USER]        = &ocfs2_xattr_user_handler,
+#ifdef CONFIG_OCFS2_FS_POSIX_ACL
+       [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS]
+                                       = &ocfs2_xattr_acl_access_handler,
+       [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT]
+                                       = &ocfs2_xattr_acl_default_handler,
+#endif
        [OCFS2_XATTR_INDEX_TRUSTED]     = &ocfs2_xattr_trusted_handler,
+       [OCFS2_XATTR_INDEX_SECURITY]    = &ocfs2_xattr_security_handler,
 };
 
 struct ocfs2_xattr_info {
@@ -150,6 +172,10 @@ static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
 
 static int ocfs2_delete_xattr_index_block(struct inode *inode,
                                          struct buffer_head *xb_bh);
+static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
+                                 u64 src_blk, u64 last_blk, u64 to_blk,
+                                 unsigned int start_bucket,
+                                 u32 *first_hash);
 
 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
 {
@@ -246,7 +272,19 @@ static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
        int rc;
 
        rc = ocfs2_read_blocks(bucket->bu_inode, xb_blkno,
-                              bucket->bu_blocks, bucket->bu_bhs, 0);
+                              bucket->bu_blocks, bucket->bu_bhs, 0,
+                              NULL);
+       if (!rc) {
+               spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
+               rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
+                                                bucket->bu_bhs,
+                                                bucket->bu_blocks,
+                                                &bucket_xh(bucket)->xh_check);
+               spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
+               if (rc)
+                       mlog_errno(rc);
+       }
+
        if (rc)
                ocfs2_xattr_bucket_relse(bucket);
        return rc;
@@ -275,6 +313,12 @@ static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
 {
        int i;
 
+       spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
+       ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
+                                  bucket->bu_bhs, bucket->bu_blocks,
+                                  &bucket_xh(bucket)->xh_check);
+       spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
+
        for (i = 0; i < bucket->bu_blocks; i++)
                ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
 }
@@ -294,6 +338,77 @@ static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
        }
 }
 
+static int ocfs2_validate_xattr_block(struct super_block *sb,
+                                     struct buffer_head *bh)
+{
+       int rc;
+       struct ocfs2_xattr_block *xb =
+               (struct ocfs2_xattr_block *)bh->b_data;
+
+       mlog(0, "Validating xattr block %llu\n",
+            (unsigned long long)bh->b_blocknr);
+
+       BUG_ON(!buffer_uptodate(bh));
+
+       /*
+        * If the ecc fails, we return the error but otherwise
+        * leave the filesystem running.  We know any error is
+        * local to this block.
+        */
+       rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
+       if (rc)
+               return rc;
+
+       /*
+        * Errors after here are fatal
+        */
+
+       if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
+               ocfs2_error(sb,
+                           "Extended attribute block #%llu has bad "
+                           "signature %.*s",
+                           (unsigned long long)bh->b_blocknr, 7,
+                           xb->xb_signature);
+               return -EINVAL;
+       }
+
+       if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
+               ocfs2_error(sb,
+                           "Extended attribute block #%llu has an "
+                           "invalid xb_blkno of %llu",
+                           (unsigned long long)bh->b_blocknr,
+                           (unsigned long long)le64_to_cpu(xb->xb_blkno));
+               return -EINVAL;
+       }
+
+       if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
+               ocfs2_error(sb,
+                           "Extended attribute block #%llu has an invalid "
+                           "xb_fs_generation of #%u",
+                           (unsigned long long)bh->b_blocknr,
+                           le32_to_cpu(xb->xb_fs_generation));
+               return -EINVAL;
+       }
+
+       return 0;
+}
+
+static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
+                                 struct buffer_head **bh)
+{
+       int rc;
+       struct buffer_head *tmp = *bh;
+
+       rc = ocfs2_read_block(inode, xb_blkno, &tmp,
+                             ocfs2_validate_xattr_block);
+
+       /* If ocfs2_read_block() got us a new bh, pass it up. */
+       if (!rc && !*bh)
+               *bh = tmp;
+
+       return rc;
+}
+
 static inline const char *ocfs2_xattr_prefix(int name_index)
 {
        struct xattr_handler *handler = NULL;
@@ -340,31 +455,163 @@ static void ocfs2_xattr_hash_entry(struct inode *inode,
        return;
 }
 
+static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
+{
+       int size = 0;
+
+       if (value_len <= OCFS2_XATTR_INLINE_SIZE)
+               size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
+       else
+               size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
+       size += sizeof(struct ocfs2_xattr_entry);
+
+       return size;
+}
+
+int ocfs2_calc_security_init(struct inode *dir,
+                            struct ocfs2_security_xattr_info *si,
+                            int *want_clusters,
+                            int *xattr_credits,
+                            struct ocfs2_alloc_context **xattr_ac)
+{
+       int ret = 0;
+       struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
+       int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
+                                                si->value_len);
+
+       /*
+        * The max space of security xattr taken inline is
+        * 256(name) + 80(value) + 16(entry) = 352 bytes,
+        * So reserve one metadata block for it is ok.
+        */
+       if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
+           s_size > OCFS2_XATTR_FREE_IN_IBODY) {
+               ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
+               if (ret) {
+                       mlog_errno(ret);
+                       return ret;
+               }
+               *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
+       }
+
+       /* reserve clusters for xattr value which will be set in B tree*/
+       if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
+               int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
+                                                           si->value_len);
+
+               *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
+                                                          new_clusters);
+               *want_clusters += new_clusters;
+       }
+       return ret;
+}
+
+int ocfs2_calc_xattr_init(struct inode *dir,
+                         struct buffer_head *dir_bh,
+                         int mode,
+                         struct ocfs2_security_xattr_info *si,
+                         int *want_clusters,
+                         int *xattr_credits,
+                         struct ocfs2_alloc_context **xattr_ac)
+{
+       int ret = 0;
+       struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
+       int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
+
+       if (si->enable)
+               s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
+                                                    si->value_len);
+
+       if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
+               acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
+                                       OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
+                                       "", NULL, 0);
+               if (acl_len > 0) {
+                       a_size = ocfs2_xattr_entry_real_size(0, acl_len);
+                       if (S_ISDIR(mode))
+                               a_size <<= 1;
+               } else if (acl_len != 0 && acl_len != -ENODATA) {
+                       mlog_errno(ret);
+                       return ret;
+               }
+       }
+
+       if (!(s_size + a_size))
+               return ret;
+
+       /*
+        * The max space of security xattr taken inline is
+        * 256(name) + 80(value) + 16(entry) = 352 bytes,
+        * The max space of acl xattr taken inline is
+        * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
+        * when blocksize = 512, may reserve one more cluser for
+        * xattr bucket, otherwise reserve one metadata block
+        * for them is ok.
+        */
+       if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
+           (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
+               ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
+               if (ret) {
+                       mlog_errno(ret);
+                       return ret;
+               }
+               *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
+       }
+
+       if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
+           (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
+               *want_clusters += 1;
+               *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
+       }
+
+       /*
+        * reserve credits and clusters for xattrs which has large value
+        * and have to be set outside
+        */
+       if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
+               new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
+                                                       si->value_len);
+               *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
+                                                          new_clusters);
+               *want_clusters += new_clusters;
+       }
+       if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
+           acl_len > OCFS2_XATTR_INLINE_SIZE) {
+               /* for directory, it has DEFAULT and ACCESS two types of acls */
+               new_clusters = (S_ISDIR(mode) ? 2 : 1) *
+                               ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
+               *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
+                                                          new_clusters);
+               *want_clusters += new_clusters;
+       }
+
+       return ret;
+}
+
 static int ocfs2_xattr_extend_allocation(struct inode *inode,
                                         u32 clusters_to_add,
-                                        struct buffer_head *xattr_bh,
-                                        struct ocfs2_xattr_value_root *xv,
+                                        struct ocfs2_xattr_value_buf *vb,
                                         struct ocfs2_xattr_set_ctxt *ctxt)
 {
        int status = 0;
        handle_t *handle = ctxt->handle;
        enum ocfs2_alloc_restarted why;
        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
-       u32 prev_clusters, logical_start = le32_to_cpu(xv->xr_clusters);
+       u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
        struct ocfs2_extent_tree et;
 
        mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
 
-       ocfs2_init_xattr_value_extent_tree(&et, inode, xattr_bh, xv);
+       ocfs2_init_xattr_value_extent_tree(&et, inode, vb);
 
-       status = ocfs2_journal_access(handle, inode, xattr_bh,
-                                     OCFS2_JOURNAL_ACCESS_WRITE);
+       status = vb->vb_access(handle, inode, vb->vb_bh,
+                             OCFS2_JOURNAL_ACCESS_WRITE);
        if (status < 0) {
                mlog_errno(status);
                goto leave;
        }
 
-       prev_clusters = le32_to_cpu(xv->xr_clusters);
+       prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
        status = ocfs2_add_clusters_in_btree(osb,
                                             inode,
                                             &logical_start,
@@ -380,13 +627,13 @@ static int ocfs2_xattr_extend_allocation(struct inode *inode,
                goto leave;
        }
 
-       status = ocfs2_journal_dirty(handle, xattr_bh);
+       status = ocfs2_journal_dirty(handle, vb->vb_bh);
        if (status < 0) {
                mlog_errno(status);
                goto leave;
        }
 
-       clusters_to_add -= le32_to_cpu(xv->xr_clusters) - prev_clusters;
+       clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) - prev_clusters;
 
        /*
         * We should have already allocated enough space before the transaction,
@@ -400,8 +647,7 @@ leave:
 }
 
 static int __ocfs2_remove_xattr_range(struct inode *inode,
-                                     struct buffer_head *root_bh,
-                                     struct ocfs2_xattr_value_root *xv,
+                                     struct ocfs2_xattr_value_buf *vb,
                                      u32 cpos, u32 phys_cpos, u32 len,
                                      struct ocfs2_xattr_set_ctxt *ctxt)
 {
@@ -410,10 +656,10 @@ static int __ocfs2_remove_xattr_range(struct inode *inode,
        handle_t *handle = ctxt->handle;
        struct ocfs2_extent_tree et;
 
-       ocfs2_init_xattr_value_extent_tree(&et, inode, root_bh, xv);
+       ocfs2_init_xattr_value_extent_tree(&et, inode, vb);
 
-       ret = ocfs2_journal_access(handle, inode, root_bh,
-                                  OCFS2_JOURNAL_ACCESS_WRITE);
+       ret = vb->vb_access(handle, inode, vb->vb_bh,
+                           OCFS2_JOURNAL_ACCESS_WRITE);
        if (ret) {
                mlog_errno(ret);
                goto out;
@@ -426,9 +672,9 @@ static int __ocfs2_remove_xattr_range(struct inode *inode,
                goto out;
        }
 
-       le32_add_cpu(&xv->xr_clusters, -len);
+       le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
 
-       ret = ocfs2_journal_dirty(handle, root_bh);
+       ret = ocfs2_journal_dirty(handle, vb->vb_bh);
        if (ret) {
                mlog_errno(ret);
                goto out;
@@ -445,8 +691,7 @@ out:
 static int ocfs2_xattr_shrink_size(struct inode *inode,
                                   u32 old_clusters,
                                   u32 new_clusters,
-                                  struct buffer_head *root_bh,
-                                  struct ocfs2_xattr_value_root *xv,
+                                  struct ocfs2_xattr_value_buf *vb,
                                   struct ocfs2_xattr_set_ctxt *ctxt)
 {
        int ret = 0;
@@ -460,7 +705,8 @@ static int ocfs2_xattr_shrink_size(struct inode *inode,
        trunc_len = old_clusters - new_clusters;
        while (trunc_len) {
                ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
-                                              &alloc_size, &xv->xr_list);
+                                              &alloc_size,
+                                              &vb->vb_xv->xr_list);
                if (ret) {
                        mlog_errno(ret);
                        goto out;
@@ -469,7 +715,7 @@ static int ocfs2_xattr_shrink_size(struct inode *inode,
                if (alloc_size > trunc_len)
                        alloc_size = trunc_len;
 
-               ret = __ocfs2_remove_xattr_range(inode, root_bh, xv, cpos,
+               ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
                                                 phys_cpos, alloc_size,
                                                 ctxt);
                if (ret) {
@@ -489,14 +735,13 @@ out:
 }
 
 static int ocfs2_xattr_value_truncate(struct inode *inode,
-                                     struct buffer_head *root_bh,
-                                     struct ocfs2_xattr_value_root *xv,
+                                     struct ocfs2_xattr_value_buf *vb,
                                      int len,
                                      struct ocfs2_xattr_set_ctxt *ctxt)
 {
        int ret;
        u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
-       u32 old_clusters = le32_to_cpu(xv->xr_clusters);
+       u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 
        if (new_clusters == old_clusters)
                return 0;
@@ -504,11 +749,11 @@ static int ocfs2_xattr_value_truncate(struct inode *inode,
        if (new_clusters > old_clusters)
                ret = ocfs2_xattr_extend_allocation(inode,
                                                    new_clusters - old_clusters,
-                                                   root_bh, xv, ctxt);
+                                                   vb, ctxt);
        else
                ret = ocfs2_xattr_shrink_size(inode,
                                              old_clusters, new_clusters,
-                                             root_bh, xv, ctxt);
+                                             vb, ctxt);
 
        return ret;
 }
@@ -598,18 +843,14 @@ static int ocfs2_xattr_block_list(struct inode *inode,
        if (!di->i_xattr_loc)
                return ret;
 
-       ret = ocfs2_read_block(inode, le64_to_cpu(di->i_xattr_loc), &blk_bh);
+       ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
+                                    &blk_bh);
        if (ret < 0) {
                mlog_errno(ret);
                return ret;
        }
 
        xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
-       if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
-               ret = -EIO;
-               goto cleanup;
-       }
-
        if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
                struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
                ret = ocfs2_xattr_list_entries(inode, header,
@@ -619,7 +860,7 @@ static int ocfs2_xattr_block_list(struct inode *inode,
                ret = ocfs2_xattr_tree_list_index_block(inode, xt,
                                                   buffer, buffer_size);
        }
-cleanup:
+
        brelse(blk_bh);
 
        return ret;
@@ -729,7 +970,7 @@ static int ocfs2_xattr_get_value_outside(struct inode *inode,
                blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
                /* Copy ocfs2_xattr_value */
                for (i = 0; i < num_clusters * bpc; i++, blkno++) {
-                       ret = ocfs2_read_block(inode, blkno, &bh);
+                       ret = ocfs2_read_block(inode, blkno, &bh, NULL);
                        if (ret) {
                                mlog_errno(ret);
                                goto out;
@@ -873,12 +1114,8 @@ cleanup:
        return ret;
 }
 
-/* ocfs2_xattr_get()
- *
- * Copy an extended attribute into the buffer provided.
- * Buffer is NULL to compute the size of buffer required.
- */
-static int ocfs2_xattr_get(struct inode *inode,
+int ocfs2_xattr_get_nolock(struct inode *inode,
+                          struct buffer_head *di_bh,
                           int name_index,
                           const char *name,
                           void *buffer,
@@ -886,7 +1123,6 @@ static int ocfs2_xattr_get(struct inode *inode,
 {
        int ret;
        struct ocfs2_dinode *di = NULL;
-       struct buffer_head *di_bh = NULL;
        struct ocfs2_inode_info *oi = OCFS2_I(inode);
        struct ocfs2_xattr_search xis = {
                .not_found = -ENODATA,
@@ -901,11 +1137,6 @@ static int ocfs2_xattr_get(struct inode *inode,
        if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
                ret = -ENODATA;
 
-       ret = ocfs2_inode_lock(inode, &di_bh, 0);
-       if (ret < 0) {
-               mlog_errno(ret);
-               return ret;
-       }
        xis.inode_bh = xbs.inode_bh = di_bh;
        di = (struct ocfs2_dinode *)di_bh->b_data;
 
@@ -916,6 +1147,32 @@ static int ocfs2_xattr_get(struct inode *inode,
                ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
                                            buffer_size, &xbs);
        up_read(&oi->ip_xattr_sem);
+
+       return ret;
+}
+
+/* ocfs2_xattr_get()
+ *
+ * Copy an extended attribute into the buffer provided.
+ * Buffer is NULL to compute the size of buffer required.
+ */
+static int ocfs2_xattr_get(struct inode *inode,
+                          int name_index,
+                          const char *name,
+                          void *buffer,
+                          size_t buffer_size)
+{
+       int ret;
+       struct buffer_head *di_bh = NULL;
+
+       ret = ocfs2_inode_lock(inode, &di_bh, 0);
+       if (ret < 0) {
+               mlog_errno(ret);
+               return ret;
+       }
+       ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
+                                    name, buffer, buffer_size);
+
        ocfs2_inode_unlock(inode, 0);
 
        brelse(di_bh);
@@ -929,7 +1186,7 @@ static int __ocfs2_xattr_set_value_outside(struct inode *inode,
                                           const void *value,
                                           int value_len)
 {
-       int ret = 0, i, cp_len, credits;
+       int ret = 0, i, cp_len;
        u16 blocksize = inode->i_sb->s_blocksize;
        u32 p_cluster, num_clusters;
        u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
@@ -939,18 +1196,6 @@ static int __ocfs2_xattr_set_value_outside(struct inode *inode,
 
        BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
 
-       /*
-        * In __ocfs2_xattr_set_value_outside has already been dirtied,
-        * so we don't need to worry about whether ocfs2_extend_trans
-        * will create a new transactio for us or not.
-        */
-       credits = clusters * bpc;
-       ret = ocfs2_extend_trans(handle, credits);
-       if (ret) {
-               mlog_errno(ret);
-               goto out;
-       }
-
        while (cpos < clusters) {
                ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
                                               &num_clusters, &xv->xr_list);
@@ -962,7 +1207,7 @@ static int __ocfs2_xattr_set_value_outside(struct inode *inode,
                blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
 
                for (i = 0; i < num_clusters * bpc; i++, blkno++) {
-                       ret = ocfs2_read_block(inode, blkno, &bh);
+                       ret = ocfs2_read_block(inode, blkno, &bh, NULL);
                        if (ret) {
                                mlog_errno(ret);
                                goto out;
@@ -1012,6 +1257,7 @@ static int ocfs2_xattr_cleanup(struct inode *inode,
                               handle_t *handle,
                               struct ocfs2_xattr_info *xi,
                               struct ocfs2_xattr_search *xs,
+                              struct ocfs2_xattr_value_buf *vb,
                               size_t offs)
 {
        int ret = 0;
@@ -1019,8 +1265,8 @@ static int ocfs2_xattr_cleanup(struct inode *inode,
        void *val = xs->base + offs;
        size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
 
-       ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
-                                  OCFS2_JOURNAL_ACCESS_WRITE);
+       ret = vb->vb_access(handle, inode, vb->vb_bh,
+                           OCFS2_JOURNAL_ACCESS_WRITE);
        if (ret) {
                mlog_errno(ret);
                goto out;
@@ -1031,7 +1277,7 @@ static int ocfs2_xattr_cleanup(struct inode *inode,
        memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
        memset(val, 0, size);
 
-       ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
+       ret = ocfs2_journal_dirty(handle, vb->vb_bh);
        if (ret < 0)
                mlog_errno(ret);
 out:
@@ -1042,12 +1288,13 @@ static int ocfs2_xattr_update_entry(struct inode *inode,
                                    handle_t *handle,
                                    struct ocfs2_xattr_info *xi,
                                    struct ocfs2_xattr_search *xs,
+                                   struct ocfs2_xattr_value_buf *vb,
                                    size_t offs)
 {
        int ret;
 
-       ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
-                                  OCFS2_JOURNAL_ACCESS_WRITE);
+       ret = vb->vb_access(handle, inode, vb->vb_bh,
+                           OCFS2_JOURNAL_ACCESS_WRITE);
        if (ret) {
                mlog_errno(ret);
                goto out;
@@ -1061,7 +1308,7 @@ static int ocfs2_xattr_update_entry(struct inode *inode,
                ocfs2_xattr_set_local(xs->here, 0);
        ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
 
-       ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
+       ret = ocfs2_journal_dirty(handle, vb->vb_bh);
        if (ret < 0)
                mlog_errno(ret);
 out:
@@ -1077,6 +1324,7 @@ static int ocfs2_xattr_set_value_outside(struct inode *inode,
                                         struct ocfs2_xattr_info *xi,
                                         struct ocfs2_xattr_search *xs,
                                         struct ocfs2_xattr_set_ctxt *ctxt,
+                                        struct ocfs2_xattr_value_buf *vb,
                                         size_t offs)
 {
        size_t name_len = strlen(xi->name);
@@ -1094,19 +1342,19 @@ static int ocfs2_xattr_set_value_outside(struct inode *inode,
        xv->xr_list.l_tree_depth = 0;
        xv->xr_list.l_count = cpu_to_le16(1);
        xv->xr_list.l_next_free_rec = 0;
+       vb->vb_xv = xv;
 
-       ret = ocfs2_xattr_value_truncate(inode, xs->xattr_bh, xv,
-                                        xi->value_len, ctxt);
+       ret = ocfs2_xattr_value_truncate(inode, vb, xi->value_len, ctxt);
        if (ret < 0) {
                mlog_errno(ret);
                return ret;
        }
-       ret = ocfs2_xattr_update_entry(inode, ctxt->handle, xi, xs, offs);
+       ret = ocfs2_xattr_update_entry(inode, ctxt->handle, xi, xs, vb, offs);
        if (ret < 0) {
                mlog_errno(ret);
                return ret;
        }
-       ret = __ocfs2_xattr_set_value_outside(inode, ctxt->handle, xv,
+       ret = __ocfs2_xattr_set_value_outside(inode, ctxt->handle, vb->vb_xv,
                                              xi->value, xi->value_len);
        if (ret < 0)
                mlog_errno(ret);
@@ -1243,6 +1491,16 @@ static int ocfs2_xattr_set_entry(struct inode *inode,
                .value = xi->value,
                .value_len = xi->value_len,
        };
+       struct ocfs2_xattr_value_buf vb = {
+               .vb_bh = xs->xattr_bh,
+               .vb_access = ocfs2_journal_access_di,
+       };
+
+       if (!(flag & OCFS2_INLINE_XATTR_FL)) {
+               BUG_ON(xs->xattr_bh == xs->inode_bh);
+               vb.vb_access = ocfs2_journal_access_xb;
+       } else
+               BUG_ON(xs->xattr_bh != xs->inode_bh);
 
        /* Compute min_offs, last and free space. */
        last = xs->header->xh_entries;
@@ -1254,7 +1512,7 @@ static int ocfs2_xattr_set_entry(struct inode *inode,
                last += 1;
        }
 
-       free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
+       free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
        if (free < 0)
                return -EIO;
 
@@ -1298,15 +1556,14 @@ static int ocfs2_xattr_set_entry(struct inode *inode,
                if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
                        /* Replace existing local xattr with tree root */
                        ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
-                                                           ctxt, offs);
+                                                           ctxt, &vb, offs);
                        if (ret < 0)
                                mlog_errno(ret);
                        goto out;
                } else if (!ocfs2_xattr_is_local(xs->here)) {
                        /* For existing xattr which has value outside */
-                       struct ocfs2_xattr_value_root *xv = NULL;
-                       xv = (struct ocfs2_xattr_value_root *)(val +
-                               OCFS2_XATTR_SIZE(name_len));
+                       vb.vb_xv = (struct ocfs2_xattr_value_root *)
+                               (val + OCFS2_XATTR_SIZE(name_len));
 
                        if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
                                /*
@@ -1315,8 +1572,7 @@ static int ocfs2_xattr_set_entry(struct inode *inode,
                                 * then set new value with set_value_outside().
                                 */
                                ret = ocfs2_xattr_value_truncate(inode,
-                                                                xs->xattr_bh,
-                                                                xv,
+                                                                &vb,
                                                                 xi->value_len,
                                                                 ctxt);
                                if (ret < 0) {
@@ -1328,6 +1584,7 @@ static int ocfs2_xattr_set_entry(struct inode *inode,
                                                               handle,
                                                               xi,
                                                               xs,
+                                                              &vb,
                                                               offs);
                                if (ret < 0) {
                                        mlog_errno(ret);
@@ -1336,7 +1593,7 @@ static int ocfs2_xattr_set_entry(struct inode *inode,
 
                                ret = __ocfs2_xattr_set_value_outside(inode,
                                                                handle,
-                                                               xv,
+                                                               vb.vb_xv,
                                                                xi->value,
                                                                xi->value_len);
                                if (ret < 0)
@@ -1348,8 +1605,7 @@ static int ocfs2_xattr_set_entry(struct inode *inode,
                                 * just trucate old value to zero.
                                 */
                                 ret = ocfs2_xattr_value_truncate(inode,
-                                                                 xs->xattr_bh,
-                                                                 xv,
+                                                                 &vb,
                                                                  0,
                                                                  ctxt);
                                if (ret < 0)
@@ -1358,16 +1614,16 @@ static int ocfs2_xattr_set_entry(struct inode *inode,
                }
        }
 
-       ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
-                                  OCFS2_JOURNAL_ACCESS_WRITE);
+       ret = ocfs2_journal_access_di(handle, inode, xs->inode_bh,
+                                     OCFS2_JOURNAL_ACCESS_WRITE);
        if (ret) {
                mlog_errno(ret);
                goto out;
        }
 
        if (!(flag & OCFS2_INLINE_XATTR_FL)) {
-               ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
-                                          OCFS2_JOURNAL_ACCESS_WRITE);
+               ret = vb.vb_access(handle, inode, vb.vb_bh,
+                                  OCFS2_JOURNAL_ACCESS_WRITE);
                if (ret) {
                        mlog_errno(ret);
                        goto out;
@@ -1412,10 +1668,6 @@ static int ocfs2_xattr_set_entry(struct inode *inode,
        oi->ip_dyn_features |= flag;
        di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
        spin_unlock(&oi->ip_lock);
-       /* Update inode ctime */
-       inode->i_ctime = CURRENT_TIME;
-       di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
-       di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 
        ret = ocfs2_journal_dirty(handle, xs->inode_bh);
        if (ret < 0)
@@ -1427,7 +1679,8 @@ static int ocfs2_xattr_set_entry(struct inode *inode,
                 * This is the second step for value size > INLINE_SIZE.
                 */
                size_t offs = le16_to_cpu(xs->here->xe_name_offset);
-               ret = ocfs2_xattr_set_value_outside(inode, xi, xs, ctxt, offs);
+               ret = ocfs2_xattr_set_value_outside(inode, xi, xs, ctxt,
+                                                   &vb, offs);
                if (ret < 0) {
                        int ret2;
 
@@ -1437,7 +1690,7 @@ static int ocfs2_xattr_set_entry(struct inode *inode,
                         * the junk tree root we have already set in local.
                         */
                        ret2 = ocfs2_xattr_cleanup(inode, ctxt->handle,
-                                                  xi, xs, offs);
+                                                  xi, xs, &vb, offs);
                        if (ret2 < 0)
                                mlog_errno(ret2);
                }
@@ -1447,7 +1700,7 @@ out:
 }
 
 static int ocfs2_remove_value_outside(struct inode*inode,
-                                     struct buffer_head *bh,
+                                     struct ocfs2_xattr_value_buf *vb,
                                      struct ocfs2_xattr_header *header)
 {
        int ret = 0, i;
@@ -1456,7 +1709,8 @@ static int ocfs2_remove_value_outside(struct inode*inode,
 
        ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
 
-       ctxt.handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
+       ctxt.handle = ocfs2_start_trans(osb,
+                                       ocfs2_remove_extent_credits(osb->sb));
        if (IS_ERR(ctxt.handle)) {
                ret = PTR_ERR(ctxt.handle);
                mlog_errno(ret);
@@ -1467,15 +1721,13 @@ static int ocfs2_remove_value_outside(struct inode*inode,
                struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
 
                if (!ocfs2_xattr_is_local(entry)) {
-                       struct ocfs2_xattr_value_root *xv;
                        void *val;
 
                        val = (void *)header +
                                le16_to_cpu(entry->xe_name_offset);
-                       xv = (struct ocfs2_xattr_value_root *)
+                       vb->vb_xv = (struct ocfs2_xattr_value_root *)
                                (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
-                       ret = ocfs2_xattr_value_truncate(inode, bh, xv,
-                                                        0, &ctxt);
+                       ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
                        if (ret < 0) {
                                mlog_errno(ret);
                                break;
@@ -1497,12 +1749,16 @@ static int ocfs2_xattr_ibody_remove(struct inode *inode,
        struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
        struct ocfs2_xattr_header *header;
        int ret;
+       struct ocfs2_xattr_value_buf vb = {
+               .vb_bh = di_bh,
+               .vb_access = ocfs2_journal_access_di,
+       };
 
        header = (struct ocfs2_xattr_header *)
                 ((void *)di + inode->i_sb->s_blocksize -
                 le16_to_cpu(di->i_xattr_inline_size));
 
-       ret = ocfs2_remove_value_outside(inode, di_bh, header);
+       ret = ocfs2_remove_value_outside(inode, &vb, header);
 
        return ret;
 }
@@ -1512,11 +1768,15 @@ static int ocfs2_xattr_block_remove(struct inode *inode,
 {
        struct ocfs2_xattr_block *xb;
        int ret = 0;
+       struct ocfs2_xattr_value_buf vb = {
+               .vb_bh = blk_bh,
+               .vb_access = ocfs2_journal_access_xb,
+       };
 
        xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
        if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
                struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
-               ret = ocfs2_remove_value_outside(inode, blk_bh, header);
+               ret = ocfs2_remove_value_outside(inode, &vb, header);
        } else
                ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
 
@@ -1536,24 +1796,19 @@ static int ocfs2_xattr_free_block(struct inode *inode,
        u64 blk, bg_blkno;
        u16 bit;
 
-       ret = ocfs2_read_block(inode, block, &blk_bh);
+       ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
        if (ret < 0) {
                mlog_errno(ret);
                goto out;
        }
 
-       xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
-       if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
-               ret = -EIO;
-               goto out;
-       }
-
        ret = ocfs2_xattr_block_remove(inode, blk_bh);
        if (ret < 0) {
                mlog_errno(ret);
                goto out;
        }
 
+       xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
        blk = le64_to_cpu(xb->xb_blkno);
        bit = le16_to_cpu(xb->xb_suballoc_bit);
        bg_blkno = ocfs2_which_suballoc_group(blk, bit);
@@ -1640,8 +1895,8 @@ int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
                mlog_errno(ret);
                goto out;
        }
-       ret = ocfs2_journal_access(handle, inode, di_bh,
-                                  OCFS2_JOURNAL_ACCESS_WRITE);
+       ret = ocfs2_journal_access_di(handle, inode, di_bh,
+                                     OCFS2_JOURNAL_ACCESS_WRITE);
        if (ret) {
                mlog_errno(ret);
                goto out_commit;
@@ -1793,19 +2048,15 @@ static int ocfs2_xattr_block_find(struct inode *inode,
        if (!di->i_xattr_loc)
                return ret;
 
-       ret = ocfs2_read_block(inode, le64_to_cpu(di->i_xattr_loc), &blk_bh);
+       ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
+                                    &blk_bh);
        if (ret < 0) {
                mlog_errno(ret);
                return ret;
        }
 
-       xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
-       if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
-               ret = -EIO;
-               goto cleanup;
-       }
-
        xs->xattr_bh = blk_bh;
+       xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
 
        if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
                xs->header = &xb->xb_attrs.xb_header;
@@ -1853,8 +2104,8 @@ static int ocfs2_xattr_block_set(struct inode *inode,
        int ret;
 
        if (!xs->xattr_bh) {
-               ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
-                                          OCFS2_JOURNAL_ACCESS_CREATE);
+               ret = ocfs2_journal_access_di(handle, inode, xs->inode_bh,
+                                             OCFS2_JOURNAL_ACCESS_CREATE);
                if (ret < 0) {
                        mlog_errno(ret);
                        goto end;
@@ -1871,8 +2122,8 @@ static int ocfs2_xattr_block_set(struct inode *inode,
                new_bh = sb_getblk(inode->i_sb, first_blkno);
                ocfs2_set_new_buffer_uptodate(inode, new_bh);
 
-               ret = ocfs2_journal_access(handle, inode, new_bh,
-                                          OCFS2_JOURNAL_ACCESS_CREATE);
+               ret = ocfs2_journal_access_xb(handle, inode, new_bh,
+                                             OCFS2_JOURNAL_ACCESS_CREATE);
                if (ret < 0) {
                        mlog_errno(ret);
                        goto end;
@@ -1944,7 +2195,7 @@ static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
                last += 1;
        }
 
-       free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
+       free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
        if (free < 0)
                return 0;
 
@@ -1983,6 +2234,15 @@ static int ocfs2_calc_xattr_set_need(struct inode *inode,
                                                    xi->value_len);
        u64 value_size;
 
+       /*
+        * Calculate the clusters we need to write.
+        * No matter whether we replace an old one or add a new one,
+        * we need this for writing.
+        */
+       if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
+               credits += new_clusters *
+                          ocfs2_clusters_to_blocks(inode->i_sb, 1);
+
        if (xis->not_found && xbs->not_found) {
                credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
 
@@ -2003,7 +2263,7 @@ static int ocfs2_calc_xattr_set_need(struct inode *inode,
                base = xis->base;
                credits += OCFS2_INODE_UPDATE_CREDITS;
        } else {
-               int i, block_off;
+               int i, block_off = 0;
                xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
                xe = xbs->here;
                name_offset = le16_to_cpu(xe->xe_name_offset);
@@ -2033,7 +2293,7 @@ static int ocfs2_calc_xattr_set_need(struct inode *inode,
         */
        if (!xi->value) {
                if (!ocfs2_xattr_is_local(xe))
-                       credits += OCFS2_REMOVE_EXTENT_CREDITS;
+                       credits += ocfs2_remove_extent_credits(inode->i_sb);
 
                goto out;
        }
@@ -2050,7 +2310,7 @@ static int ocfs2_calc_xattr_set_need(struct inode *inode,
                 */
                if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
                        clusters_add += new_clusters;
-                       credits += OCFS2_REMOVE_EXTENT_CREDITS +
+                       credits += ocfs2_remove_extent_credits(inode->i_sb) +
                                    OCFS2_INODE_UPDATE_CREDITS;
                        if (!ocfs2_xattr_is_local(xe))
                                credits += ocfs2_calc_extend_credits(
@@ -2070,11 +2330,12 @@ static int ocfs2_calc_xattr_set_need(struct inode *inode,
                                                                 value_size);
                        xv = (struct ocfs2_xattr_value_root *)
                             (base + name_offset + name_len);
+                       value_size = OCFS2_XATTR_ROOT_SIZE;
                } else
                        xv = &def_xv.xv;
 
                if (old_clusters >= new_clusters) {
-                       credits += OCFS2_REMOVE_EXTENT_CREDITS;
+                       credits += ocfs2_remove_extent_credits(inode->i_sb);
                        goto out;
                } else {
                        meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
@@ -2083,7 +2344,8 @@ static int ocfs2_calc_xattr_set_need(struct inode *inode,
                                                             &xv->xr_list,
                                                             new_clusters -
                                                             old_clusters);
-                       goto out;
+                       if (value_size >= OCFS2_XATTR_ROOT_SIZE)
+                               goto out;
                }
        } else {
                /*
@@ -2102,9 +2364,9 @@ meta_guess:
        /* calculate metadata allocation. */
        if (di->i_xattr_loc) {
                if (!xbs->xattr_bh) {
-                       ret = ocfs2_read_block(inode,
-                                              le64_to_cpu(di->i_xattr_loc),
-                                              &bh);
+                       ret = ocfs2_read_xattr_block(inode,
+                                                    le64_to_cpu(di->i_xattr_loc),
+                                                    &bh);
                        if (ret) {
                                mlog_errno(ret);
                                goto out;
@@ -2114,13 +2376,21 @@ meta_guess:
                } else
                        xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
 
+               /*
+                * If there is already an xattr tree, good, we can calculate
+                * like other b-trees. Otherwise we may have the chance of
+                * create a tree, the credit calculation is borrowed from
+                * ocfs2_calc_extend_credits with root_el = NULL. And the
+                * new tree will be cluster based, so no meta is needed.
+                */
                if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
                        struct ocfs2_extent_list *el =
                                 &xb->xb_attrs.xb_root.xt_list;
                        meta_add += ocfs2_extend_meta_needed(el);
                        credits += ocfs2_calc_extend_credits(inode->i_sb,
                                                             el, 1);
-               }
+               } else
+                       credits += OCFS2_SUBALLOC_ALLOC + 1;
 
                /*
                 * This cluster will be used either for new bucket or for
@@ -2212,7 +2482,7 @@ static int __ocfs2_xattr_set_handle(struct inode *inode,
                                    struct ocfs2_xattr_search *xbs,
                                    struct ocfs2_xattr_set_ctxt *ctxt)
 {
-       int ret = 0, credits;
+       int ret = 0, credits, old_found;
 
        if (!xi->value) {
                /* Remove existing extended attribute */
@@ -2231,6 +2501,7 @@ static int __ocfs2_xattr_set_handle(struct inode *inode,
                        xi->value = NULL;
                        xi->value_len = 0;
 
+                       old_found = xis->not_found;
                        xis->not_found = -ENODATA;
                        ret = ocfs2_calc_xattr_set_need(inode,
                                                        di,
@@ -2240,6 +2511,7 @@ static int __ocfs2_xattr_set_handle(struct inode *inode,
                                                        NULL,
                                                        NULL,
                                                        &credits);
+                       xis->not_found = old_found;
                        if (ret) {
                                mlog_errno(ret);
                                goto out;
@@ -2260,6 +2532,7 @@ static int __ocfs2_xattr_set_handle(struct inode *inode,
                                if (ret)
                                        goto out;
 
+                               old_found = xis->not_found;
                                xis->not_found = -ENODATA;
                                ret = ocfs2_calc_xattr_set_need(inode,
                                                                di,
@@ -2269,6 +2542,7 @@ static int __ocfs2_xattr_set_handle(struct inode *inode,
                                                                NULL,
                                                                NULL,
                                                                &credits);
+                               xis->not_found = old_found;
                                if (ret) {
                                        mlog_errno(ret);
                                        goto out;
@@ -2321,11 +2595,106 @@ static int __ocfs2_xattr_set_handle(struct inode *inode,
                }
        }
 
+       if (!ret) {
+               /* Update inode ctime. */
+               ret = ocfs2_journal_access_di(ctxt->handle, inode,
+                                             xis->inode_bh,
+                                             OCFS2_JOURNAL_ACCESS_WRITE);
+               if (ret) {
+                       mlog_errno(ret);
+                       goto out;
+               }
+
+               inode->i_ctime = CURRENT_TIME;
+               di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
+               di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
+               ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
+       }
 out:
        return ret;
 }
 
 /*
+ * This function only called duing creating inode
+ * for init security/acl xattrs of the new inode.
+ * All transanction credits have been reserved in mknod.
+ */
+int ocfs2_xattr_set_handle(handle_t *handle,
+                          struct inode *inode,
+                          struct buffer_head *di_bh,
+                          int name_index,
+                          const char *name,
+                          const void *value,
+                          size_t value_len,
+                          int flags,
+                          struct ocfs2_alloc_context *meta_ac,
+                          struct ocfs2_alloc_context *data_ac)
+{
+       struct ocfs2_dinode *di;
+       int ret;
+
+       struct ocfs2_xattr_info xi = {
+               .name_index = name_index,
+               .name = name,
+               .value = value,
+               .value_len = value_len,
+       };
+
+       struct ocfs2_xattr_search xis = {
+               .not_found = -ENODATA,
+       };
+
+       struct ocfs2_xattr_search xbs = {
+               .not_found = -ENODATA,
+       };
+
+       struct ocfs2_xattr_set_ctxt ctxt = {
+               .handle = handle,
+               .meta_ac = meta_ac,
+               .data_ac = data_ac,
+       };
+
+       if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
+               return -EOPNOTSUPP;
+
+       /*
+        * In extreme situation, may need xattr bucket when
+        * block size is too small. And we have already reserved
+        * the credits for bucket in mknod.
+        */
+       if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
+               xbs.bucket = ocfs2_xattr_bucket_new(inode);
+               if (!xbs.bucket) {
+                       mlog_errno(-ENOMEM);
+                       return -ENOMEM;
+               }
+       }
+
+       xis.inode_bh = xbs.inode_bh = di_bh;
+       di = (struct ocfs2_dinode *)di_bh->b_data;
+
+       down_write(&OCFS2_I(inode)->ip_xattr_sem);
+
+       ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
+       if (ret)
+               goto cleanup;
+       if (xis.not_found) {
+               ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
+               if (ret)
+                       goto cleanup;
+       }
+
+       ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
+
+cleanup:
+       up_write(&OCFS2_I(inode)->ip_xattr_sem);
+       brelse(xbs.xattr_bh);
+       ocfs2_xattr_bucket_free(xbs.bucket);
+
+       return ret;
+}
+
+/*
  * ocfs2_xattr_set()
  *
  * Set, replace or remove an extended attribute for this inode.
@@ -2429,6 +2798,8 @@ int ocfs2_xattr_set(struct inode *inode,
                goto cleanup;
        }
 
+       /* we need to update inode's ctime field, so add credit for it. */
+       credits += OCFS2_INODE_UPDATE_CREDITS;
        ctxt.handle = ocfs2_start_trans(osb, credits);
        if (IS_ERR(ctxt.handle)) {
                ret = PTR_ERR(ctxt.handle);
@@ -2569,6 +2940,7 @@ static int ocfs2_find_xe_in_bucket(struct inode *inode,
                        break;
                }
 
+
                xe_name = bucket_block(bucket, block_off) + new_offset;
                if (!memcmp(name, xe_name, name_len)) {
                        *xe_index = i;
@@ -3052,8 +3424,8 @@ static int ocfs2_xattr_create_index_block(struct inode *inode,
         */
        down_write(&oi->ip_alloc_sem);
 
-       ret = ocfs2_journal_access(handle, inode, xb_bh,
-                                  OCFS2_JOURNAL_ACCESS_WRITE);
+       ret = ocfs2_journal_access_xb(handle, inode, xb_bh,
+                                     OCFS2_JOURNAL_ACCESS_WRITE);
        if (ret) {
                mlog_errno(ret);
                goto out;
@@ -3250,124 +3622,84 @@ out:
 }
 
 /*
- * Move half nums of the xattr bucket in the previous cluster to this new
- * cluster. We only touch the last cluster of the previous extend record.
+ * prev_blkno points to the start of an existing extent.  new_blkno
+ * points to a newly allocated extent.  Because we know each of our
+ * clusters contains more than bucket, we can easily split one cluster
+ * at a bucket boundary.  So we take the last cluster of the existing
+ * extent and split it down the middle.  We move the last half of the
+ * buckets in the last cluster of the existing extent over to the new
+ * extent.
+ *
+ * first_bh is the buffer at prev_blkno so we can update the existing
+ * extent's bucket count.  header_bh is the bucket were we were hoping
+ * to insert our xattr.  If the bucket move places the target in the new
+ * extent, we'll update first_bh and header_bh after modifying the old
+ * extent.
  *
- * first_bh is the first buffer_head of a series of bucket in the same
- * extent rec and header_bh is the header of one bucket in this cluster.
- * They will be updated if we move the data header_bh contains to the new
- * cluster. first_hash will be set as the 1st xe's name_hash of the new cluster.
+ * first_hash will be set as the 1st xe's name_hash in the new extent.
  */
 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
                                               handle_t *handle,
-                                              struct buffer_head **first_bh,
-                                              struct buffer_head **header_bh,
+                                              struct ocfs2_xattr_bucket *first,
+                                              struct ocfs2_xattr_bucket *target,
                                               u64 new_blkno,
-                                              u64 prev_blkno,
                                               u32 num_clusters,
                                               u32 *first_hash)
 {
-       int i, ret, credits;
-       struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
-       int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
-       int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
-       int blocksize = inode->i_sb->s_blocksize;
-       struct buffer_head *old_bh, *new_bh, *prev_bh, *new_first_bh = NULL;
-       struct ocfs2_xattr_header *new_xh;
-       struct ocfs2_xattr_header *xh =
-                       (struct ocfs2_xattr_header *)((*first_bh)->b_data);
-
-       BUG_ON(le16_to_cpu(xh->xh_num_buckets) < num_buckets);
-       BUG_ON(OCFS2_XATTR_BUCKET_SIZE == osb->s_clustersize);
-
-       prev_bh = *first_bh;
-       get_bh(prev_bh);
-       xh = (struct ocfs2_xattr_header *)prev_bh->b_data;
+       int ret;
+       struct super_block *sb = inode->i_sb;
+       int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
+       int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
+       int to_move = num_buckets / 2;
+       u64 src_blkno;
+       u64 last_cluster_blkno = bucket_blkno(first) +
+               ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
 
-       prev_blkno += (num_clusters - 1) * bpc + bpc / 2;
+       BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
+       BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
 
        mlog(0, "move half of xattrs in cluster %llu to %llu\n",
-            (unsigned long long)prev_blkno, (unsigned long long)new_blkno);
+            (unsigned long long)last_cluster_blkno, (unsigned long long)new_blkno);
 
-       /*
-        * We need to update the 1st half of the new cluster and
-        * 1 more for the update of the 1st bucket of the previous
-        * extent record.
-        */
-       credits = bpc / 2 + 1 + handle->h_buffer_credits;
-       ret = ocfs2_extend_trans(handle, credits);
+       ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
+                                    last_cluster_blkno, new_blkno,
+                                    to_move, first_hash);
        if (ret) {
                mlog_errno(ret);
                goto out;
        }
 
-       ret = ocfs2_journal_access(handle, inode, prev_bh,
-                                  OCFS2_JOURNAL_ACCESS_WRITE);
-       if (ret) {
-               mlog_errno(ret);
-               goto out;
-       }
+       /* This is the first bucket that got moved */
+       src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
 
-       for (i = 0; i < bpc / 2; i++, prev_blkno++, new_blkno++) {
-               old_bh = new_bh = NULL;
-               new_bh = sb_getblk(inode->i_sb, new_blkno);
-               if (!new_bh) {
-                       ret = -EIO;
-                       mlog_errno(ret);
-                       goto out;
-               }
+       /*
+        * If the target bucket was part of the moved buckets, we need to
+        * update first and target.
+        */
+       if (bucket_blkno(target) >= src_blkno) {
+               /* Find the block for the new target bucket */
+               src_blkno = new_blkno +
+                       (bucket_blkno(target) - src_blkno);
 
-               ocfs2_set_new_buffer_uptodate(inode, new_bh);
+               ocfs2_xattr_bucket_relse(first);
+               ocfs2_xattr_bucket_relse(target);
 
-               ret = ocfs2_journal_access(handle, inode, new_bh,
-                                          OCFS2_JOURNAL_ACCESS_CREATE);
-               if (ret < 0) {
+               /*
+                * These shouldn't fail - the buffers are in the
+                * journal from ocfs2_cp_xattr_bucket().
+                */
+               ret = ocfs2_read_xattr_bucket(first, new_blkno);
+               if (ret) {
                        mlog_errno(ret);
-                       brelse(new_bh);
                        goto out;
                }
-
-               ret = ocfs2_read_block(inode, prev_blkno, &old_bh);
-               if (ret < 0) {
+               ret = ocfs2_read_xattr_bucket(target, src_blkno);
+               if (ret)
                        mlog_errno(ret);
-                       brelse(new_bh);
-                       goto out;
-               }
-
-               memcpy(new_bh->b_data, old_bh->b_data, blocksize);
-
-               if (i == 0) {
-                       new_xh = (struct ocfs2_xattr_header *)new_bh->b_data;
-                       new_xh->xh_num_buckets = cpu_to_le16(num_buckets / 2);
-
-                       if (first_hash)
-                               *first_hash = le32_to_cpu(
-                                       new_xh->xh_entries[0].xe_name_hash);
-                       new_first_bh = new_bh;
-                       get_bh(new_first_bh);
-               }
 
-               ocfs2_journal_dirty(handle, new_bh);
-
-               if (*header_bh == old_bh) {
-                       brelse(*header_bh);
-                       *header_bh = new_bh;
-                       get_bh(*header_bh);
-
-                       brelse(*first_bh);
-                       *first_bh = new_first_bh;
-                       get_bh(*first_bh);
-               }
-               brelse(new_bh);
-               brelse(old_bh);
        }
 
-       le16_add_cpu(&xh->xh_num_buckets, -(num_buckets / 2));
-
-       ocfs2_journal_dirty(handle, prev_bh);
 out:
-       brelse(prev_bh);
-       brelse(new_first_bh);
        return ret;
 }
 
@@ -3471,6 +3803,11 @@ static int ocfs2_divide_xattr_bucket(struct inode *inode,
                goto out;
        }
 
+       /*
+        * Hey, if we're overwriting t_bucket, what difference does
+        * ACCESS_CREATE vs ACCESS_WRITE make?  See the comment in the
+        * same part of ocfs2_cp_xattr_bucket().
+        */
        ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
                                                new_bucket_head ?
                                                OCFS2_JOURNAL_ACCESS_CREATE :
@@ -3629,7 +3966,7 @@ static int ocfs2_cp_xattr_bucket(struct inode *inode,
                mlog_errno(ret);
                goto out;
        }
-  
+
        ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
        if (ret)
                goto out;
@@ -3642,6 +3979,19 @@ static int ocfs2_cp_xattr_bucket(struct inode *inode,
        if (ret)
                goto out;
 
+       /*
+        * Hey, if we're overwriting t_bucket, what difference does
+        * ACCESS_CREATE vs ACCESS_WRITE make?  Well, if we allocated a new
+        * cluster to fill, we came here from
+        * ocfs2_mv_xattr_buckets(), and it is really new -
+        * ACCESS_CREATE is required.  But we also might have moved data
+        * out of t_bucket before extending back into it.
+        * ocfs2_add_new_xattr_bucket() can do this - its call to
+        * ocfs2_add_new_xattr_cluster() may have created a new extent
+        * and copied out the end of the old extent.  Then it re-extends
+        * the old extent back to create space for new xattrs.  That's
+        * how we get here, and the bucket isn't really new.
+        */
        ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
                                                t_is_new ?
                                                OCFS2_JOURNAL_ACCESS_CREATE :
@@ -3660,41 +4010,64 @@ out:
 }
 
 /*
- * Copy one xattr cluster from src_blk to to_blk.
- * The to_blk will become the first bucket header of the cluster, so its
- * xh_num_buckets will be initialized as the bucket num in the cluster.
+ * src_blk points to the start of an existing extent.  last_blk points to
+ * last cluster in that extent.  to_blk points to a newly allocated
+ * extent.  We copy the buckets from the cluster at last_blk to the new
+ * extent.  If start_bucket is non-zero, we skip that many buckets before
+ * we start copying.  The new extent's xh_num_buckets gets set to the
+ * number of buckets we copied.  The old extent's xh_num_buckets shrinks
+ * by the same amount.
  */
-static int ocfs2_cp_xattr_cluster(struct inode *inode,
-                                 handle_t *handle,
-                                 struct buffer_head *first_bh,
-                                 u64 src_blk,
-                                 u64 to_blk,
+static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
+                                 u64 src_blk, u64 last_blk, u64 to_blk,
+                                 unsigned int start_bucket,
                                  u32 *first_hash)
 {
        int i, ret, credits;
        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
-       int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
+       int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
        int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
-       struct buffer_head *bh = NULL;
-       struct ocfs2_xattr_header *xh;
-       u64 to_blk_start = to_blk;
+       struct ocfs2_xattr_bucket *old_first, *new_first;
+
+       mlog(0, "mv xattrs from cluster %llu to %llu\n",
+            (unsigned long long)last_blk, (unsigned long long)to_blk);
+
+       BUG_ON(start_bucket >= num_buckets);
+       if (start_bucket) {
+               num_buckets -= start_bucket;
+               last_blk += (start_bucket * blks_per_bucket);
+       }
 
-       mlog(0, "cp xattrs from cluster %llu to %llu\n",
-            (unsigned long long)src_blk, (unsigned long long)to_blk);
+       /* The first bucket of the original extent */
+       old_first = ocfs2_xattr_bucket_new(inode);
+       /* The first bucket of the new extent */
+       new_first = ocfs2_xattr_bucket_new(inode);
+       if (!old_first || !new_first) {
+               ret = -ENOMEM;
+               mlog_errno(ret);
+               goto out;
+       }
+
+       ret = ocfs2_read_xattr_bucket(old_first, src_blk);
+       if (ret) {
+               mlog_errno(ret);
+               goto out;
+       }
 
        /*
-        * We need to update the new cluster and 1 more for the update of
-        * the 1st bucket of the previous extent rec.
+        * We need to update the first bucket of the old extent and all
+        * the buckets going to the new extent.
         */
-       credits = bpc + 1 + handle->h_buffer_credits;
+       credits = ((num_buckets + 1) * blks_per_bucket) +
+               handle->h_buffer_credits;
        ret = ocfs2_extend_trans(handle, credits);
        if (ret) {
                mlog_errno(ret);
                goto out;
        }
 
-       ret = ocfs2_journal_access(handle, inode, first_bh,
-                                  OCFS2_JOURNAL_ACCESS_WRITE);
+       ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
+                                               OCFS2_JOURNAL_ACCESS_WRITE);
        if (ret) {
                mlog_errno(ret);
                goto out;
@@ -3702,45 +4075,45 @@ static int ocfs2_cp_xattr_cluster(struct inode *inode,
 
        for (i = 0; i < num_buckets; i++) {
                ret = ocfs2_cp_xattr_bucket(inode, handle,
-                                           src_blk, to_blk, 1);
+                                           last_blk + (i * blks_per_bucket),
+                                           to_blk + (i * blks_per_bucket),
+                                           1);
                if (ret) {
                        mlog_errno(ret);
                        goto out;
                }
-
-               src_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
-               to_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
        }
 
-       /* update the old bucket header. */
-       xh = (struct ocfs2_xattr_header *)first_bh->b_data;
-       le16_add_cpu(&xh->xh_num_buckets, -num_buckets);
-
-       ocfs2_journal_dirty(handle, first_bh);
-
-       /* update the new bucket header. */
-       ret = ocfs2_read_block(inode, to_blk_start, &bh);
-       if (ret < 0) {
+       /*
+        * Get the new bucket ready before we dirty anything
+        * (This actually shouldn't fail, because we already dirtied
+        * it once in ocfs2_cp_xattr_bucket()).
+        */
+       ret = ocfs2_read_xattr_bucket(new_first, to_blk);
+       if (ret) {
                mlog_errno(ret);
                goto out;
        }
-
-       ret = ocfs2_journal_access(handle, inode, bh,
-                                  OCFS2_JOURNAL_ACCESS_WRITE);
+       ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
+                                               OCFS2_JOURNAL_ACCESS_WRITE);
        if (ret) {
                mlog_errno(ret);
                goto out;
        }
 
-       xh = (struct ocfs2_xattr_header *)bh->b_data;
-       xh->xh_num_buckets = cpu_to_le16(num_buckets);
+       /* Now update the headers */
+       le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
+       ocfs2_xattr_bucket_journal_dirty(handle, old_first);
 
-       ocfs2_journal_dirty(handle, bh);
+       bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
+       ocfs2_xattr_bucket_journal_dirty(handle, new_first);
 
        if (first_hash)
-               *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
+               *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
+
 out:
-       brelse(bh);
+       ocfs2_xattr_bucket_free(new_first);
+       ocfs2_xattr_bucket_free(old_first);
        return ret;
 }
 
@@ -3799,43 +4172,49 @@ static int ocfs2_divide_xattr_cluster(struct inode *inode,
  */
 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
                                            handle_t *handle,
-                                           struct buffer_head **first_bh,
-                                           struct buffer_head **header_bh,
+                                           struct ocfs2_xattr_bucket *first,
+                                           struct ocfs2_xattr_bucket *target,
                                            u64 new_blk,
-                                           u64 prev_blk,
                                            u32 prev_clusters,
                                            u32 *v_start,
                                            int *extend)
 {
-       int ret = 0;
-       int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
+       int ret;
 
        mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
-            (unsigned long long)prev_blk, prev_clusters,
+            (unsigned long long)bucket_blkno(first), prev_clusters,
             (unsigned long long)new_blk);
 
-       if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1)
+       if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
                ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
                                                          handle,
-                                                         first_bh,
-                                                         header_bh,
+                                                         first, target,
                                                          new_blk,
-                                                         prev_blk,
                                                          prev_clusters,
                                                          v_start);
-       else {
-               u64 last_blk = prev_blk + bpc * (prev_clusters - 1);
-
-               if (prev_clusters > 1 && (*header_bh)->b_blocknr != last_blk)
-                       ret = ocfs2_cp_xattr_cluster(inode, handle, *first_bh,
-                                                    last_blk, new_blk,
+               if (ret)
+                       mlog_errno(ret);
+       } else {
+               /* The start of the last cluster in the first extent */
+               u64 last_blk = bucket_blkno(first) +
+                       ((prev_clusters - 1) *
+                        ocfs2_clusters_to_blocks(inode->i_sb, 1));
+
+               if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
+                       ret = ocfs2_mv_xattr_buckets(inode, handle,
+                                                    bucket_blkno(first),
+                                                    last_blk, new_blk, 0,
                                                     v_start);
-               else {
+                       if (ret)
+                               mlog_errno(ret);
+               } else {
                        ret = ocfs2_divide_xattr_cluster(inode, handle,
                                                         last_blk, new_blk,
                                                         v_start);
+                       if (ret)
+                               mlog_errno(ret);
 
-                       if ((*header_bh)->b_blocknr == last_blk && extend)
+                       if ((bucket_blkno(target) == last_blk) && extend)
                                *extend = 0;
                }
        }
@@ -3861,11 +4240,10 @@ static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
  */
 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
                                       struct buffer_head *root_bh,
-                                      struct buffer_head **first_bh,
-                                      struct buffer_head **header_bh,
+                                      struct ocfs2_xattr_bucket *first,
+                                      struct ocfs2_xattr_bucket *target,
                                       u32 *num_clusters,
                                       u32 prev_cpos,
-                                      u64 prev_blkno,
                                       int *extend,
                                       struct ocfs2_xattr_set_ctxt *ctxt)
 {
@@ -3881,12 +4259,12 @@ static int ocfs2_add_new_xattr_cluster(struct inode *inode,
        mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
             "previous xattr blkno = %llu\n",
             (unsigned long long)OCFS2_I(inode)->ip_blkno,
-            prev_cpos, (unsigned long long)prev_blkno);
+            prev_cpos, (unsigned long long)bucket_blkno(first));
 
        ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
 
-       ret = ocfs2_journal_access(handle, inode, root_bh,
-                                  OCFS2_JOURNAL_ACCESS_WRITE);
+       ret = ocfs2_journal_access_xb(handle, inode, root_bh,
+                                     OCFS2_JOURNAL_ACCESS_WRITE);
        if (ret < 0) {
                mlog_errno(ret);
                goto leave;
@@ -3906,7 +4284,7 @@ static int ocfs2_add_new_xattr_cluster(struct inode *inode,
        mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
             num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
 
-       if (prev_blkno + prev_clusters * bpc == block &&
+       if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
            (prev_clusters + num_bits) << osb->s_clustersize_bits <=
             OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
                /*
@@ -3925,10 +4303,9 @@ static int ocfs2_add_new_xattr_cluster(struct inode *inode,
        } else {
                ret = ocfs2_adjust_xattr_cross_cluster(inode,
                                                       handle,
-                                                      first_bh,
-                                                      header_bh,
+                                                      first,
+                                                      target,
                                                       block,
-                                                      prev_blkno,
                                                       prev_clusters,
                                                       &v_start,
                                                       extend);
@@ -3956,37 +4333,45 @@ leave:
 }
 
 /*
- * Extend a new xattr bucket and move xattrs to the end one by one until
- * We meet with start_bh. Only move half of the xattrs to the bucket after it.
+ * We are given an extent.  'first' is the bucket at the very front of
+ * the extent.  The extent has space for an additional bucket past
+ * bucket_xh(first)->xh_num_buckets.  'target_blkno' is the block number
+ * of the target bucket.  We wish to shift every bucket past the target
+ * down one, filling in that additional space.  When we get back to the
+ * target, we split the target between itself and the now-empty bucket
+ * at target+1 (aka, target_blkno + blks_per_bucket).
  */
 static int ocfs2_extend_xattr_bucket(struct inode *inode,
                                     handle_t *handle,
-                                    struct buffer_head *first_bh,
-                                    struct buffer_head *start_bh,
+                                    struct ocfs2_xattr_bucket *first,
+                                    u64 target_blk,
                                     u32 num_clusters)
 {
        int ret, credits;
        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
        u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
-       u64 start_blk = start_bh->b_blocknr, end_blk;
-       u32 num_buckets = num_clusters * ocfs2_xattr_buckets_per_cluster(osb);
-       struct ocfs2_xattr_header *first_xh =
-                               (struct ocfs2_xattr_header *)first_bh->b_data;
-       u16 bucket = le16_to_cpu(first_xh->xh_num_buckets);
+       u64 end_blk;
+       u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
 
        mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
-            "from %llu, len = %u\n", (unsigned long long)start_blk,
-            (unsigned long long)first_bh->b_blocknr, num_clusters);
+            "from %llu, len = %u\n", (unsigned long long)target_blk,
+            (unsigned long long)bucket_blkno(first), num_clusters);
 
-       BUG_ON(bucket >= num_buckets);
+       /* The extent must have room for an additional bucket */
+       BUG_ON(new_bucket >=
+              (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
 
-       end_blk = first_bh->b_blocknr + (bucket - 1) * blk_per_bucket;
+       /* end_blk points to the last existing bucket */
+       end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
 
        /*
-        * We will touch all the buckets after the start_bh(include it).
-        * Then we add one more bucket.
+        * end_blk is the start of the last existing bucket.
+        * Thus, (end_blk - target_blk) covers the target bucket and
+        * every bucket after it up to, but not including, the last
+        * existing bucket.  Then we add the last existing bucket, the
+        * new bucket, and the first bucket (3 * blk_per_bucket).
         */
-       credits = end_blk - start_blk + 3 * blk_per_bucket + 1 +
+       credits = (end_blk - target_blk) + (3 * blk_per_bucket) +
                  handle->h_buffer_credits;
        ret = ocfs2_extend_trans(handle, credits);
        if (ret) {
@@ -3994,14 +4379,14 @@ static int ocfs2_extend_xattr_bucket(struct inode *inode,
                goto out;
        }
 
-       ret = ocfs2_journal_access(handle, inode, first_bh,
-                                  OCFS2_JOURNAL_ACCESS_WRITE);
+       ret = ocfs2_xattr_bucket_journal_access(handle, first,
+                                               OCFS2_JOURNAL_ACCESS_WRITE);
        if (ret) {
                mlog_errno(ret);
                goto out;
        }
 
-       while (end_blk != start_blk) {
+       while (end_blk != target_blk) {
                ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
                                            end_blk + blk_per_bucket, 0);
                if (ret)
@@ -4009,55 +4394,58 @@ static int ocfs2_extend_xattr_bucket(struct inode *inode,
                end_blk -= blk_per_bucket;
        }
 
-       /* Move half of the xattr in start_blk to the next bucket. */
-       ret = ocfs2_divide_xattr_bucket(inode, handle, start_blk,
-                                       start_blk + blk_per_bucket, NULL, 0);
+       /* Move half of the xattr in target_blkno to the next bucket. */
+       ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
+                                       target_blk + blk_per_bucket, NULL, 0);
 
-       le16_add_cpu(&first_xh->xh_num_buckets, 1);
-       ocfs2_journal_dirty(handle, first_bh);
+       le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
+       ocfs2_xattr_bucket_journal_dirty(handle, first);
 
 out:
        return ret;
 }
 
 /*
- * Add new xattr bucket in an extent record and adjust the buckets accordingly.
- * xb_bh is the ocfs2_xattr_block.
- * We will move all the buckets starting from header_bh to the next place. As
- * for this one, half num of its xattrs will be moved to the next one.
+ * Add new xattr bucket in an extent record and adjust the buckets
+ * accordingly.  xb_bh is the ocfs2_xattr_block, and target is the
+ * bucket we want to insert into.
  *
- * We will allocate a new cluster if current cluster is full and adjust
- * header_bh and first_bh if the insert place is moved to the new cluster.
+ * In the easy case, we will move all the buckets after target down by
+ * one. Half of target's xattrs will be moved to the next bucket.
+ *
+ * If current cluster is full, we'll allocate a new one.  This may not
+ * be contiguous.  The underlying calls will make sure that there is
+ * space for the insert, shifting buckets around if necessary.
+ * 'target' may be moved by those calls.
  */
 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
                                      struct buffer_head *xb_bh,
-                                     struct buffer_head *header_bh,
+                                     struct ocfs2_xattr_bucket *target,
                                      struct ocfs2_xattr_set_ctxt *ctxt)
 {
-       struct ocfs2_xattr_header *first_xh = NULL;
-       struct buffer_head *first_bh = NULL;
        struct ocfs2_xattr_block *xb =
                        (struct ocfs2_xattr_block *)xb_bh->b_data;
        struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
        struct ocfs2_extent_list *el = &xb_root->xt_list;
-       struct ocfs2_xattr_header *xh =
-                       (struct ocfs2_xattr_header *)header_bh->b_data;
-       u32 name_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
-       struct super_block *sb = inode->i_sb;
-       struct ocfs2_super *osb = OCFS2_SB(sb);
+       u32 name_hash =
+               le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
+       struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
        int ret, num_buckets, extend = 1;
        u64 p_blkno;
        u32 e_cpos, num_clusters;
+       /* The bucket at the front of the extent */
+       struct ocfs2_xattr_bucket *first;
 
-       mlog(0, "Add new xattr bucket starting form %llu\n",
-            (unsigned long long)header_bh->b_blocknr);
+       mlog(0, "Add new xattr bucket starting from %llu\n",
+            (unsigned long long)bucket_blkno(target));
 
-       /*
-        * Add refrence for header_bh here because it may be
-        * changed in ocfs2_add_new_xattr_cluster and we need
-        * to free it in the end.
-        */
-       get_bh(header_bh);
+       /* The first bucket of the original extent */
+       first = ocfs2_xattr_bucket_new(inode);
+       if (!first) {
+               ret = -ENOMEM;
+               mlog_errno(ret);
+               goto out;
+       }
 
        ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
                                  &num_clusters, el);
@@ -4066,23 +4454,24 @@ static int ocfs2_add_new_xattr_bucket(struct inode *inode,
                goto out;
        }
 
-       ret = ocfs2_read_block(inode, p_blkno, &first_bh);
+       ret = ocfs2_read_xattr_bucket(first, p_blkno);
        if (ret) {
                mlog_errno(ret);
                goto out;
        }
 
        num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
-       first_xh = (struct ocfs2_xattr_header *)first_bh->b_data;
-
-       if (num_buckets == le16_to_cpu(first_xh->xh_num_buckets)) {
+       if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
+               /*
+                * This can move first+target if the target bucket moves
+                * to the new extent.
+                */
                ret = ocfs2_add_new_xattr_cluster(inode,
                                                  xb_bh,
-                                                 &first_bh,
-                                                 &header_bh,
+                                                 first,
+                                                 target,
                                                  &num_clusters,
                                                  e_cpos,
-                                                 p_blkno,
                                                  &extend,
                                                  ctxt);
                if (ret) {
@@ -4091,17 +4480,19 @@ static int ocfs2_add_new_xattr_bucket(struct inode *inode,
                }
        }
 
-       if (extend)
+       if (extend) {
                ret = ocfs2_extend_xattr_bucket(inode,
                                                ctxt->handle,
-                                               first_bh,
-                                               header_bh,
+                                               first,
+                                               bucket_blkno(target),
                                                num_clusters);
-       if (ret)
-               mlog_errno(ret);
+               if (ret)
+                       mlog_errno(ret);
+       }
+
 out:
-       brelse(first_bh);
-       brelse(header_bh);
+       ocfs2_xattr_bucket_free(first);
+
        return ret;
 }
 
@@ -4304,31 +4695,6 @@ out:
        return ret;
 }
 
-static int ocfs2_xattr_value_update_size(struct inode *inode,
-                                        handle_t *handle,
-                                        struct buffer_head *xe_bh,
-                                        struct ocfs2_xattr_entry *xe,
-                                        u64 new_size)
-{
-       int ret;
-
-       ret = ocfs2_journal_access(handle, inode, xe_bh,
-                                  OCFS2_JOURNAL_ACCESS_WRITE);
-       if (ret < 0) {
-               mlog_errno(ret);
-               goto out;
-       }
-
-       xe->xe_value_size = cpu_to_le64(new_size);
-
-       ret = ocfs2_journal_dirty(handle, xe_bh);
-       if (ret < 0)
-               mlog_errno(ret);
-
-out:
-       return ret;
-}
-
 /*
  * Truncate the specified xe_off entry in xattr bucket.
  * bucket is indicated by header_bh and len is the new length.
@@ -4337,19 +4703,19 @@ out:
  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
  */
 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
-                                            struct buffer_head *header_bh,
+                                            struct ocfs2_xattr_bucket *bucket,
                                             int xe_off,
                                             int len,
                                             struct ocfs2_xattr_set_ctxt *ctxt)
 {
        int ret, offset;
        u64 value_blk;
-       struct buffer_head *value_bh = NULL;
-       struct ocfs2_xattr_value_root *xv;
        struct ocfs2_xattr_entry *xe;
-       struct ocfs2_xattr_header *xh =
-                       (struct ocfs2_xattr_header *)header_bh->b_data;
+       struct ocfs2_xattr_header *xh = bucket_xh(bucket);
        size_t blocksize = inode->i_sb->s_blocksize;
+       struct ocfs2_xattr_value_buf vb = {
+               .vb_access = ocfs2_journal_access,
+       };
 
        xe = &xh->xh_entries[xe_off];
 
@@ -4362,34 +4728,40 @@ static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
 
        /* We don't allow ocfs2_xattr_value to be stored in different block. */
        BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
-       value_blk += header_bh->b_blocknr;
 
-       ret = ocfs2_read_block(inode, value_blk, &value_bh);
-       if (ret) {
-               mlog_errno(ret);
-               goto out;
-       }
+       vb.vb_bh = bucket->bu_bhs[value_blk];
+       BUG_ON(!vb.vb_bh);
 
-       xv = (struct ocfs2_xattr_value_root *)
-               (value_bh->b_data + offset % blocksize);
+       vb.vb_xv = (struct ocfs2_xattr_value_root *)
+               (vb.vb_bh->b_data + offset % blocksize);
 
+       /*
+        * From here on out we have to dirty the bucket.  The generic
+        * value calls only modify one of the bucket's bhs, but we need
+        * to send the bucket at once.  So if they error, they *could* have
+        * modified something.  We have to assume they did, and dirty
+        * the whole bucket.  This leaves us in a consistent state.
+        */
        mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
-            xe_off, (unsigned long long)header_bh->b_blocknr, len);
-       ret = ocfs2_xattr_value_truncate(inode, value_bh, xv, len, ctxt);
+            xe_off, (unsigned long long)bucket_blkno(bucket), len);
+       ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
        if (ret) {
                mlog_errno(ret);
                goto out;
        }
 
-       ret = ocfs2_xattr_value_update_size(inode, ctxt->handle,
-                                           header_bh, xe, len);
+       ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
+                                               OCFS2_JOURNAL_ACCESS_WRITE);
        if (ret) {
                mlog_errno(ret);
                goto out;
        }
 
+       xe->xe_value_size = cpu_to_le64(len);
+
+       ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
+
 out:
-       brelse(value_bh);
        return ret;
 }
 
@@ -4405,7 +4777,7 @@ static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
        BUG_ON(!xs->bucket->bu_bhs[0] || !xe || ocfs2_xattr_is_local(xe));
 
        offset = xe - xh->xh_entries;
-       ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket->bu_bhs[0],
+       ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket,
                                                offset, len, ctxt);
        if (ret)
                mlog_errno(ret);
@@ -4475,15 +4847,15 @@ static int ocfs2_rm_xattr_cluster(struct inode *inode,
                }
        }
 
-       handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
+       handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
        if (IS_ERR(handle)) {
                ret = -ENOMEM;
                mlog_errno(ret);
                goto out;
        }
 
-       ret = ocfs2_journal_access(handle, inode, root_bh,
-                                  OCFS2_JOURNAL_ACCESS_WRITE);
+       ret = ocfs2_journal_access_xb(handle, inode, root_bh,
+                                     OCFS2_JOURNAL_ACCESS_WRITE);
        if (ret) {
                mlog_errno(ret);
                goto out_commit;
@@ -4694,8 +5066,8 @@ try_again:
        xh_free_start = le16_to_cpu(xh->xh_free_start);
        header_size = sizeof(struct ocfs2_xattr_header) +
                        count * sizeof(struct ocfs2_xattr_entry);
-       max_free = OCFS2_XATTR_BUCKET_SIZE -
-               le16_to_cpu(xh->xh_name_value_len) - header_size;
+       max_free = OCFS2_XATTR_BUCKET_SIZE - header_size -
+               le16_to_cpu(xh->xh_name_value_len) - OCFS2_XATTR_HEADER_GAP;
 
        mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
                        "of %u which exceed block size\n",
@@ -4728,7 +5100,7 @@ try_again:
                        need = 0;
        }
 
-       free = xh_free_start - header_size;
+       free = xh_free_start - header_size - OCFS2_XATTR_HEADER_GAP;
        /*
         * We need to make sure the new name/value pair
         * can exist in the same block.
@@ -4761,7 +5133,8 @@ try_again:
                        }
 
                        xh_free_start = le16_to_cpu(xh->xh_free_start);
-                       free = xh_free_start - header_size;
+                       free = xh_free_start - header_size
+                               - OCFS2_XATTR_HEADER_GAP;
                        if (xh_free_start % blocksize < need)
                                free -= xh_free_start % blocksize;
 
@@ -4795,15 +5168,21 @@ try_again:
 
                ret = ocfs2_add_new_xattr_bucket(inode,
                                                 xs->xattr_bh,
-                                                xs->bucket->bu_bhs[0],
+                                                xs->bucket,
                                                 ctxt);
                if (ret) {
                        mlog_errno(ret);
                        goto out;
                }
 
+               /*
+                * ocfs2_add_new_xattr_bucket() will have updated
+                * xs->bucket if it moved, but it will not have updated
+                * any of the other search fields.  Thus, we drop it and
+                * re-search.  Everything should be cached, so it'll be
+                * quick.
+                */
                ocfs2_xattr_bucket_relse(xs->bucket);
-
                ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
                                                   xi->name_index,
                                                   xi->name, xs);
@@ -4831,34 +5210,36 @@ static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
        struct ocfs2_xattr_entry *xe;
        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
        struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
+       int credits = ocfs2_remove_extent_credits(osb->sb) +
+               ocfs2_blocks_per_xattr_bucket(inode->i_sb);
 
-       ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
 
-       ctxt.handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
-       if (IS_ERR(ctxt.handle)) {
-               ret = PTR_ERR(ctxt.handle);
-               mlog_errno(ret);
-               goto out;
-       }
+       ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
 
        for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
                xe = &xh->xh_entries[i];
                if (ocfs2_xattr_is_local(xe))
                        continue;
 
-               ret = ocfs2_xattr_bucket_value_truncate(inode,
-                                                       bucket->bu_bhs[0],
+               ctxt.handle = ocfs2_start_trans(osb, credits);
+               if (IS_ERR(ctxt.handle)) {
+                       ret = PTR_ERR(ctxt.handle);
+                       mlog_errno(ret);
+                       break;
+               }
+
+               ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
                                                        i, 0, &ctxt);
+
+               ocfs2_commit_trans(osb, ctxt.handle);
                if (ret) {
                        mlog_errno(ret);
                        break;
                }
        }
 
-       ret = ocfs2_commit_trans(osb, ctxt.handle);
        ocfs2_schedule_truncate_log_flush(osb, 1);
        ocfs2_run_deallocs(osb, &ctxt.dealloc);
-out:
        return ret;
 }
 
@@ -4909,6 +5290,74 @@ out:
 }
 
 /*
+ * 'security' attributes support
+ */
+static size_t ocfs2_xattr_security_list(struct inode *inode, char *list,
+                                       size_t list_size, const char *name,
+                                       size_t name_len)
+{
+       const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
+       const size_t total_len = prefix_len + name_len + 1;
+
+       if (list && total_len <= list_size) {
+               memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
+               memcpy(list + prefix_len, name, name_len);
+               list[prefix_len + name_len] = '\0';
+       }
+       return total_len;
+}
+
+static int ocfs2_xattr_security_get(struct inode *inode, const char *name,
+                                   void *buffer, size_t size)
+{
+       if (strcmp(name, "") == 0)
+               return -EINVAL;
+       return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_SECURITY, name,
+                              buffer, size);
+}
+
+static int ocfs2_xattr_security_set(struct inode *inode, const char *name,
+                                   const void *value, size_t size, int flags)
+{
+       if (strcmp(name, "") == 0)
+               return -EINVAL;
+
+       return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY, name, value,
+                              size, flags);
+}
+
+int ocfs2_init_security_get(struct inode *inode,
+                           struct inode *dir,
+                           struct ocfs2_security_xattr_info *si)
+{
+       /* check whether ocfs2 support feature xattr */
+       if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
+               return -EOPNOTSUPP;
+       return security_inode_init_security(inode, dir, &si->name, &si->value,
+                                           &si->value_len);
+}
+
+int ocfs2_init_security_set(handle_t *handle,
+                           struct inode *inode,
+                           struct buffer_head *di_bh,
+                           struct ocfs2_security_xattr_info *si,
+                           struct ocfs2_alloc_context *xattr_ac,
+                           struct ocfs2_alloc_context *data_ac)
+{
+       return ocfs2_xattr_set_handle(handle, inode, di_bh,
+                                    OCFS2_XATTR_INDEX_SECURITY,
+                                    si->name, si->value, si->value_len, 0,
+                                    xattr_ac, data_ac);
+}
+
+struct xattr_handler ocfs2_xattr_security_handler = {
+       .prefix = XATTR_SECURITY_PREFIX,
+       .list   = ocfs2_xattr_security_list,
+       .get    = ocfs2_xattr_security_get,
+       .set    = ocfs2_xattr_security_set,
+};
+
+/*
  * 'trusted' attributes support
  */
 static size_t ocfs2_xattr_trusted_list(struct inode *inode, char *list,