1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * Copyright (C) 2008 Oracle. All rights reserved.
9 * Lots of code in this file is taken from ext3.
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public
13 * License as published by the Free Software Foundation; either
14 * version 2 of the License, or (at your option) any later version.
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
21 * You should have received a copy of the GNU General Public
22 * License along with this program; if not, write to the
23 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
24 * Boston, MA 021110-1307, USA.
27 #include <linux/capability.h>
29 #include <linux/types.h>
30 #include <linux/slab.h>
31 #include <linux/highmem.h>
32 #include <linux/pagemap.h>
33 #include <linux/uio.h>
34 #include <linux/sched.h>
35 #include <linux/splice.h>
36 #include <linux/mount.h>
37 #include <linux/writeback.h>
38 #include <linux/falloc.h>
39 #include <linux/sort.h>
40 #include <linux/init.h>
41 #include <linux/module.h>
42 #include <linux/string.h>
44 #define MLOG_MASK_PREFIX ML_XATTR
45 #include <cluster/masklog.h>
58 #include "buffer_head_io.h"
63 struct ocfs2_xattr_def_value_root {
64 struct ocfs2_xattr_value_root xv;
65 struct ocfs2_extent_rec er;
68 struct ocfs2_xattr_bucket {
69 struct buffer_head *bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
70 struct ocfs2_xattr_header *xh;
73 #define OCFS2_XATTR_ROOT_SIZE (sizeof(struct ocfs2_xattr_def_value_root))
74 #define OCFS2_XATTR_INLINE_SIZE 80
76 static struct ocfs2_xattr_def_value_root def_xv = {
77 .xv.xr_list.l_count = cpu_to_le16(1),
80 struct xattr_handler *ocfs2_xattr_handlers[] = {
81 &ocfs2_xattr_user_handler,
82 &ocfs2_xattr_trusted_handler,
86 static struct xattr_handler *ocfs2_xattr_handler_map[] = {
87 [OCFS2_XATTR_INDEX_USER] = &ocfs2_xattr_user_handler,
88 [OCFS2_XATTR_INDEX_TRUSTED] = &ocfs2_xattr_trusted_handler,
91 struct ocfs2_xattr_info {
98 struct ocfs2_xattr_search {
99 struct buffer_head *inode_bh;
101 * xattr_bh point to the block buffer head which has extended attribute
102 * when extended attribute in inode, xattr_bh is equal to inode_bh.
104 struct buffer_head *xattr_bh;
105 struct ocfs2_xattr_header *header;
106 struct ocfs2_xattr_bucket bucket;
109 struct ocfs2_xattr_entry *here;
113 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
114 struct ocfs2_xattr_header *xh,
119 static int ocfs2_xattr_index_block_find(struct inode *inode,
120 struct buffer_head *root_bh,
123 struct ocfs2_xattr_search *xs);
125 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
126 struct ocfs2_xattr_tree_root *xt,
130 static int ocfs2_xattr_create_index_block(struct inode *inode,
131 struct ocfs2_xattr_search *xs);
133 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
134 struct ocfs2_xattr_info *xi,
135 struct ocfs2_xattr_search *xs);
137 static int ocfs2_delete_xattr_index_block(struct inode *inode,
138 struct buffer_head *xb_bh);
140 static inline const char *ocfs2_xattr_prefix(int name_index)
142 struct xattr_handler *handler = NULL;
144 if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
145 handler = ocfs2_xattr_handler_map[name_index];
147 return handler ? handler->prefix : NULL;
150 static u32 ocfs2_xattr_name_hash(struct inode *inode,
154 /* Get hash value of uuid from super block */
155 u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
158 /* hash extended attribute name */
159 for (i = 0; i < name_len; i++) {
160 hash = (hash << OCFS2_HASH_SHIFT) ^
161 (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
169 * ocfs2_xattr_hash_entry()
171 * Compute the hash of an extended attribute.
173 static void ocfs2_xattr_hash_entry(struct inode *inode,
174 struct ocfs2_xattr_header *header,
175 struct ocfs2_xattr_entry *entry)
178 char *name = (char *)header + le16_to_cpu(entry->xe_name_offset);
180 hash = ocfs2_xattr_name_hash(inode, name, entry->xe_name_len);
181 entry->xe_name_hash = cpu_to_le32(hash);
186 static int ocfs2_xattr_extend_allocation(struct inode *inode,
188 struct buffer_head *xattr_bh,
189 struct ocfs2_xattr_value_root *xv)
192 int restart_func = 0;
194 handle_t *handle = NULL;
195 struct ocfs2_alloc_context *data_ac = NULL;
196 struct ocfs2_alloc_context *meta_ac = NULL;
197 enum ocfs2_alloc_restarted why;
198 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
199 u32 prev_clusters, logical_start = le32_to_cpu(xv->xr_clusters);
200 struct ocfs2_extent_tree et;
202 mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
204 ocfs2_init_xattr_value_extent_tree(&et, inode, xattr_bh, xv);
208 status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
215 credits = ocfs2_calc_extend_credits(osb->sb, et.et_root_el,
217 handle = ocfs2_start_trans(osb, credits);
218 if (IS_ERR(handle)) {
219 status = PTR_ERR(handle);
225 restarted_transaction:
226 status = ocfs2_journal_access(handle, inode, xattr_bh,
227 OCFS2_JOURNAL_ACCESS_WRITE);
233 prev_clusters = le32_to_cpu(xv->xr_clusters);
234 status = ocfs2_add_clusters_in_btree(osb,
244 if ((status < 0) && (status != -EAGAIN)) {
245 if (status != -ENOSPC)
250 status = ocfs2_journal_dirty(handle, xattr_bh);
256 clusters_to_add -= le32_to_cpu(xv->xr_clusters) - prev_clusters;
258 if (why != RESTART_NONE && clusters_to_add) {
259 if (why == RESTART_META) {
260 mlog(0, "restarting function.\n");
263 BUG_ON(why != RESTART_TRANS);
265 mlog(0, "restarting transaction.\n");
266 /* TODO: This can be more intelligent. */
267 credits = ocfs2_calc_extend_credits(osb->sb,
270 status = ocfs2_extend_trans(handle, credits);
272 /* handle still has to be committed at
278 goto restarted_transaction;
284 ocfs2_commit_trans(osb, handle);
288 ocfs2_free_alloc_context(data_ac);
292 ocfs2_free_alloc_context(meta_ac);
295 if ((!status) && restart_func) {
303 static int __ocfs2_remove_xattr_range(struct inode *inode,
304 struct buffer_head *root_bh,
305 struct ocfs2_xattr_value_root *xv,
306 u32 cpos, u32 phys_cpos, u32 len,
307 struct ocfs2_cached_dealloc_ctxt *dealloc)
310 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
311 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
312 struct inode *tl_inode = osb->osb_tl_inode;
314 struct ocfs2_alloc_context *meta_ac = NULL;
315 struct ocfs2_extent_tree et;
317 ocfs2_init_xattr_value_extent_tree(&et, inode, root_bh, xv);
319 ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
325 mutex_lock(&tl_inode->i_mutex);
327 if (ocfs2_truncate_log_needs_flush(osb)) {
328 ret = __ocfs2_flush_truncate_log(osb);
335 handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
336 if (IS_ERR(handle)) {
337 ret = PTR_ERR(handle);
342 ret = ocfs2_journal_access(handle, inode, root_bh,
343 OCFS2_JOURNAL_ACCESS_WRITE);
349 ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
356 le32_add_cpu(&xv->xr_clusters, -len);
358 ret = ocfs2_journal_dirty(handle, root_bh);
364 ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len);
369 ocfs2_commit_trans(osb, handle);
371 mutex_unlock(&tl_inode->i_mutex);
374 ocfs2_free_alloc_context(meta_ac);
379 static int ocfs2_xattr_shrink_size(struct inode *inode,
382 struct buffer_head *root_bh,
383 struct ocfs2_xattr_value_root *xv)
386 u32 trunc_len, cpos, phys_cpos, alloc_size;
388 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
389 struct ocfs2_cached_dealloc_ctxt dealloc;
391 ocfs2_init_dealloc_ctxt(&dealloc);
393 if (old_clusters <= new_clusters)
397 trunc_len = old_clusters - new_clusters;
399 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
400 &alloc_size, &xv->xr_list);
406 if (alloc_size > trunc_len)
407 alloc_size = trunc_len;
409 ret = __ocfs2_remove_xattr_range(inode, root_bh, xv, cpos,
410 phys_cpos, alloc_size,
417 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
418 ocfs2_remove_xattr_clusters_from_cache(inode, block,
421 trunc_len -= alloc_size;
425 ocfs2_schedule_truncate_log_flush(osb, 1);
426 ocfs2_run_deallocs(osb, &dealloc);
431 static int ocfs2_xattr_value_truncate(struct inode *inode,
432 struct buffer_head *root_bh,
433 struct ocfs2_xattr_value_root *xv,
437 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
438 u32 old_clusters = le32_to_cpu(xv->xr_clusters);
440 if (new_clusters == old_clusters)
443 if (new_clusters > old_clusters)
444 ret = ocfs2_xattr_extend_allocation(inode,
445 new_clusters - old_clusters,
448 ret = ocfs2_xattr_shrink_size(inode,
449 old_clusters, new_clusters,
455 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
456 size_t *result, const char *prefix,
457 const char *name, int name_len)
459 char *p = buffer + *result;
460 int prefix_len = strlen(prefix);
461 int total_len = prefix_len + name_len + 1;
463 *result += total_len;
465 /* we are just looking for how big our buffer needs to be */
472 memcpy(p, prefix, prefix_len);
473 memcpy(p + prefix_len, name, name_len);
474 p[prefix_len + name_len] = '\0';
479 static int ocfs2_xattr_list_entries(struct inode *inode,
480 struct ocfs2_xattr_header *header,
481 char *buffer, size_t buffer_size)
485 const char *prefix, *name;
487 for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
488 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
489 type = ocfs2_xattr_get_type(entry);
490 prefix = ocfs2_xattr_prefix(type);
493 name = (const char *)header +
494 le16_to_cpu(entry->xe_name_offset);
496 ret = ocfs2_xattr_list_entry(buffer, buffer_size,
497 &result, prefix, name,
507 static int ocfs2_xattr_ibody_list(struct inode *inode,
508 struct ocfs2_dinode *di,
512 struct ocfs2_xattr_header *header = NULL;
513 struct ocfs2_inode_info *oi = OCFS2_I(inode);
516 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
519 header = (struct ocfs2_xattr_header *)
520 ((void *)di + inode->i_sb->s_blocksize -
521 le16_to_cpu(di->i_xattr_inline_size));
523 ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
528 static int ocfs2_xattr_block_list(struct inode *inode,
529 struct ocfs2_dinode *di,
533 struct buffer_head *blk_bh = NULL;
534 struct ocfs2_xattr_block *xb;
537 if (!di->i_xattr_loc)
540 ret = ocfs2_read_block(inode,
541 le64_to_cpu(di->i_xattr_loc),
542 &blk_bh, OCFS2_BH_CACHED);
547 /*Verify the signature of xattr block*/
548 if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
549 strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
554 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
556 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
557 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
558 ret = ocfs2_xattr_list_entries(inode, header,
559 buffer, buffer_size);
561 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
562 ret = ocfs2_xattr_tree_list_index_block(inode, xt,
563 buffer, buffer_size);
571 ssize_t ocfs2_listxattr(struct dentry *dentry,
575 int ret = 0, i_ret = 0, b_ret = 0;
576 struct buffer_head *di_bh = NULL;
577 struct ocfs2_dinode *di = NULL;
578 struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
580 if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
583 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
586 ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
592 di = (struct ocfs2_dinode *)di_bh->b_data;
594 down_read(&oi->ip_xattr_sem);
595 i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
603 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
608 up_read(&oi->ip_xattr_sem);
609 ocfs2_inode_unlock(dentry->d_inode, 0);
613 return i_ret + b_ret;
616 static int ocfs2_xattr_find_entry(int name_index,
618 struct ocfs2_xattr_search *xs)
620 struct ocfs2_xattr_entry *entry;
627 name_len = strlen(name);
629 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
630 cmp = name_index - ocfs2_xattr_get_type(entry);
632 cmp = name_len - entry->xe_name_len;
634 cmp = memcmp(name, (xs->base +
635 le16_to_cpu(entry->xe_name_offset)),
643 return cmp ? -ENODATA : 0;
646 static int ocfs2_xattr_get_value_outside(struct inode *inode,
647 struct ocfs2_xattr_value_root *xv,
651 u32 cpos, p_cluster, num_clusters, bpc, clusters;
654 size_t cplen, blocksize;
655 struct buffer_head *bh = NULL;
656 struct ocfs2_extent_list *el;
659 clusters = le32_to_cpu(xv->xr_clusters);
660 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
661 blocksize = inode->i_sb->s_blocksize;
664 while (cpos < clusters) {
665 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
672 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
673 /* Copy ocfs2_xattr_value */
674 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
675 ret = ocfs2_read_block(inode, blkno,
676 &bh, OCFS2_BH_CACHED);
682 cplen = len >= blocksize ? blocksize : len;
683 memcpy(buffer, bh->b_data, cplen);
692 cpos += num_clusters;
698 static int ocfs2_xattr_ibody_get(struct inode *inode,
703 struct ocfs2_xattr_search *xs)
705 struct ocfs2_inode_info *oi = OCFS2_I(inode);
706 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
707 struct ocfs2_xattr_value_root *xv;
711 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
714 xs->end = (void *)di + inode->i_sb->s_blocksize;
715 xs->header = (struct ocfs2_xattr_header *)
716 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
717 xs->base = (void *)xs->header;
718 xs->here = xs->header->xh_entries;
720 ret = ocfs2_xattr_find_entry(name_index, name, xs);
723 size = le64_to_cpu(xs->here->xe_value_size);
725 if (size > buffer_size)
727 if (ocfs2_xattr_is_local(xs->here)) {
728 memcpy(buffer, (void *)xs->base +
729 le16_to_cpu(xs->here->xe_name_offset) +
730 OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
732 xv = (struct ocfs2_xattr_value_root *)
733 (xs->base + le16_to_cpu(
734 xs->here->xe_name_offset) +
735 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
736 ret = ocfs2_xattr_get_value_outside(inode, xv,
748 static int ocfs2_xattr_block_get(struct inode *inode,
753 struct ocfs2_xattr_search *xs)
755 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
756 struct buffer_head *blk_bh = NULL;
757 struct ocfs2_xattr_block *xb;
758 struct ocfs2_xattr_value_root *xv;
760 int ret = -ENODATA, name_offset, name_len, block_off, i;
762 if (!di->i_xattr_loc)
765 memset(&xs->bucket, 0, sizeof(xs->bucket));
767 ret = ocfs2_read_block(inode,
768 le64_to_cpu(di->i_xattr_loc),
769 &blk_bh, OCFS2_BH_CACHED);
774 /*Verify the signature of xattr block*/
775 if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
776 strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
781 xs->xattr_bh = blk_bh;
782 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
784 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
785 xs->header = &xb->xb_attrs.xb_header;
786 xs->base = (void *)xs->header;
787 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
788 xs->here = xs->header->xh_entries;
790 ret = ocfs2_xattr_find_entry(name_index, name, xs);
792 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
798 size = le64_to_cpu(xs->here->xe_value_size);
801 if (size > buffer_size)
804 name_offset = le16_to_cpu(xs->here->xe_name_offset);
805 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
806 i = xs->here - xs->header->xh_entries;
808 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
809 ret = ocfs2_xattr_bucket_get_name_value(inode,
814 xs->base = xs->bucket.bhs[block_off]->b_data;
816 if (ocfs2_xattr_is_local(xs->here)) {
817 memcpy(buffer, (void *)xs->base +
818 name_offset + name_len, size);
820 xv = (struct ocfs2_xattr_value_root *)
821 (xs->base + name_offset + name_len);
822 ret = ocfs2_xattr_get_value_outside(inode, xv,
832 for (i = 0; i < OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET; i++)
833 brelse(xs->bucket.bhs[i]);
834 memset(&xs->bucket, 0, sizeof(xs->bucket));
842 * Copy an extended attribute into the buffer provided.
843 * Buffer is NULL to compute the size of buffer required.
845 int ocfs2_xattr_get(struct inode *inode,
852 struct ocfs2_dinode *di = NULL;
853 struct buffer_head *di_bh = NULL;
854 struct ocfs2_inode_info *oi = OCFS2_I(inode);
855 struct ocfs2_xattr_search xis = {
856 .not_found = -ENODATA,
858 struct ocfs2_xattr_search xbs = {
859 .not_found = -ENODATA,
862 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
865 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
868 ret = ocfs2_inode_lock(inode, &di_bh, 0);
873 xis.inode_bh = xbs.inode_bh = di_bh;
874 di = (struct ocfs2_dinode *)di_bh->b_data;
876 down_read(&oi->ip_xattr_sem);
877 ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
880 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
882 up_read(&oi->ip_xattr_sem);
883 ocfs2_inode_unlock(inode, 0);
890 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
891 struct ocfs2_xattr_value_root *xv,
895 int ret = 0, i, cp_len, credits;
896 u16 blocksize = inode->i_sb->s_blocksize;
897 u32 p_cluster, num_clusters;
898 u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
899 u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
901 struct buffer_head *bh = NULL;
904 BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
906 credits = clusters * bpc;
907 handle = ocfs2_start_trans(OCFS2_SB(inode->i_sb), credits);
908 if (IS_ERR(handle)) {
909 ret = PTR_ERR(handle);
914 while (cpos < clusters) {
915 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
916 &num_clusters, &xv->xr_list);
922 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
924 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
925 ret = ocfs2_read_block(inode, blkno,
926 &bh, OCFS2_BH_CACHED);
932 ret = ocfs2_journal_access(handle,
935 OCFS2_JOURNAL_ACCESS_WRITE);
941 cp_len = value_len > blocksize ? blocksize : value_len;
942 memcpy(bh->b_data, value, cp_len);
945 if (cp_len < blocksize)
946 memset(bh->b_data + cp_len, 0,
949 ret = ocfs2_journal_dirty(handle, bh);
958 * XXX: do we need to empty all the following
959 * blocks in this cluster?
964 cpos += num_clusters;
967 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
974 static int ocfs2_xattr_cleanup(struct inode *inode,
975 struct ocfs2_xattr_info *xi,
976 struct ocfs2_xattr_search *xs,
979 handle_t *handle = NULL;
981 size_t name_len = strlen(xi->name);
982 void *val = xs->base + offs;
983 size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
985 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
986 OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
987 if (IS_ERR(handle)) {
988 ret = PTR_ERR(handle);
992 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
993 OCFS2_JOURNAL_ACCESS_WRITE);
998 /* Decrease xattr count */
999 le16_add_cpu(&xs->header->xh_count, -1);
1000 /* Remove the xattr entry and tree root which has already be set*/
1001 memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
1002 memset(val, 0, size);
1004 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1008 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1013 static int ocfs2_xattr_update_entry(struct inode *inode,
1014 struct ocfs2_xattr_info *xi,
1015 struct ocfs2_xattr_search *xs,
1018 handle_t *handle = NULL;
1021 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1022 OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1023 if (IS_ERR(handle)) {
1024 ret = PTR_ERR(handle);
1028 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1029 OCFS2_JOURNAL_ACCESS_WRITE);
1035 xs->here->xe_name_offset = cpu_to_le16(offs);
1036 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1037 if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1038 ocfs2_xattr_set_local(xs->here, 1);
1040 ocfs2_xattr_set_local(xs->here, 0);
1041 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1043 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1047 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1053 * ocfs2_xattr_set_value_outside()
1055 * Set large size value in B tree.
1057 static int ocfs2_xattr_set_value_outside(struct inode *inode,
1058 struct ocfs2_xattr_info *xi,
1059 struct ocfs2_xattr_search *xs,
1062 size_t name_len = strlen(xi->name);
1063 void *val = xs->base + offs;
1064 struct ocfs2_xattr_value_root *xv = NULL;
1065 size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1068 memset(val, 0, size);
1069 memcpy(val, xi->name, name_len);
1070 xv = (struct ocfs2_xattr_value_root *)
1071 (val + OCFS2_XATTR_SIZE(name_len));
1072 xv->xr_clusters = 0;
1073 xv->xr_last_eb_blk = 0;
1074 xv->xr_list.l_tree_depth = 0;
1075 xv->xr_list.l_count = cpu_to_le16(1);
1076 xv->xr_list.l_next_free_rec = 0;
1078 ret = ocfs2_xattr_value_truncate(inode, xs->xattr_bh, xv,
1084 ret = __ocfs2_xattr_set_value_outside(inode, xv, xi->value,
1090 ret = ocfs2_xattr_update_entry(inode, xi, xs, offs);
1098 * ocfs2_xattr_set_entry_local()
1100 * Set, replace or remove extended attribute in local.
1102 static void ocfs2_xattr_set_entry_local(struct inode *inode,
1103 struct ocfs2_xattr_info *xi,
1104 struct ocfs2_xattr_search *xs,
1105 struct ocfs2_xattr_entry *last,
1108 size_t name_len = strlen(xi->name);
1111 if (xi->value && xs->not_found) {
1112 /* Insert the new xattr entry. */
1113 le16_add_cpu(&xs->header->xh_count, 1);
1114 ocfs2_xattr_set_type(last, xi->name_index);
1115 ocfs2_xattr_set_local(last, 1);
1116 last->xe_name_len = name_len;
1122 first_val = xs->base + min_offs;
1123 offs = le16_to_cpu(xs->here->xe_name_offset);
1124 val = xs->base + offs;
1126 if (le64_to_cpu(xs->here->xe_value_size) >
1127 OCFS2_XATTR_INLINE_SIZE)
1128 size = OCFS2_XATTR_SIZE(name_len) +
1129 OCFS2_XATTR_ROOT_SIZE;
1131 size = OCFS2_XATTR_SIZE(name_len) +
1132 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1134 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1135 OCFS2_XATTR_SIZE(xi->value_len)) {
1136 /* The old and the new value have the
1137 same size. Just replace the value. */
1138 ocfs2_xattr_set_local(xs->here, 1);
1139 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1140 /* Clear value bytes. */
1141 memset(val + OCFS2_XATTR_SIZE(name_len),
1143 OCFS2_XATTR_SIZE(xi->value_len));
1144 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1149 /* Remove the old name+value. */
1150 memmove(first_val + size, first_val, val - first_val);
1151 memset(first_val, 0, size);
1152 xs->here->xe_name_hash = 0;
1153 xs->here->xe_name_offset = 0;
1154 ocfs2_xattr_set_local(xs->here, 1);
1155 xs->here->xe_value_size = 0;
1159 /* Adjust all value offsets. */
1160 last = xs->header->xh_entries;
1161 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1162 size_t o = le16_to_cpu(last->xe_name_offset);
1165 last->xe_name_offset = cpu_to_le16(o + size);
1170 /* Remove the old entry. */
1172 memmove(xs->here, xs->here + 1,
1173 (void *)last - (void *)xs->here);
1174 memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1175 le16_add_cpu(&xs->header->xh_count, -1);
1179 /* Insert the new name+value. */
1180 size_t size = OCFS2_XATTR_SIZE(name_len) +
1181 OCFS2_XATTR_SIZE(xi->value_len);
1182 void *val = xs->base + min_offs - size;
1184 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1185 memset(val, 0, size);
1186 memcpy(val, xi->name, name_len);
1187 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1190 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1191 ocfs2_xattr_set_local(xs->here, 1);
1192 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1199 * ocfs2_xattr_set_entry()
1201 * Set extended attribute entry into inode or block.
1203 * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1204 * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1205 * then set value in B tree with set_value_outside().
1207 static int ocfs2_xattr_set_entry(struct inode *inode,
1208 struct ocfs2_xattr_info *xi,
1209 struct ocfs2_xattr_search *xs,
1212 struct ocfs2_xattr_entry *last;
1213 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1214 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1215 size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1217 handle_t *handle = NULL;
1219 struct ocfs2_xattr_info xi_l = {
1220 .name_index = xi->name_index,
1223 .value_len = xi->value_len,
1226 /* Compute min_offs, last and free space. */
1227 last = xs->header->xh_entries;
1229 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1230 size_t offs = le16_to_cpu(last->xe_name_offset);
1231 if (offs < min_offs)
1236 free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
1240 if (!xs->not_found) {
1242 if (ocfs2_xattr_is_local(xs->here))
1243 size = OCFS2_XATTR_SIZE(name_len) +
1244 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1246 size = OCFS2_XATTR_SIZE(name_len) +
1247 OCFS2_XATTR_ROOT_SIZE;
1248 free += (size + sizeof(struct ocfs2_xattr_entry));
1250 /* Check free space in inode or block */
1251 if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1252 if (free < sizeof(struct ocfs2_xattr_entry) +
1253 OCFS2_XATTR_SIZE(name_len) +
1254 OCFS2_XATTR_ROOT_SIZE) {
1258 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1259 xi_l.value = (void *)&def_xv;
1260 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1261 } else if (xi->value) {
1262 if (free < sizeof(struct ocfs2_xattr_entry) +
1263 OCFS2_XATTR_SIZE(name_len) +
1264 OCFS2_XATTR_SIZE(xi->value_len)) {
1270 if (!xs->not_found) {
1271 /* For existing extended attribute */
1272 size_t size = OCFS2_XATTR_SIZE(name_len) +
1273 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1274 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1275 void *val = xs->base + offs;
1277 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1278 /* Replace existing local xattr with tree root */
1279 ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1284 } else if (!ocfs2_xattr_is_local(xs->here)) {
1285 /* For existing xattr which has value outside */
1286 struct ocfs2_xattr_value_root *xv = NULL;
1287 xv = (struct ocfs2_xattr_value_root *)(val +
1288 OCFS2_XATTR_SIZE(name_len));
1290 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1292 * If new value need set outside also,
1293 * first truncate old value to new value,
1294 * then set new value with set_value_outside().
1296 ret = ocfs2_xattr_value_truncate(inode,
1305 ret = __ocfs2_xattr_set_value_outside(inode,
1314 ret = ocfs2_xattr_update_entry(inode,
1323 * If new value need set in local,
1324 * just trucate old value to zero.
1326 ret = ocfs2_xattr_value_truncate(inode,
1336 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1337 OCFS2_INODE_UPDATE_CREDITS);
1338 if (IS_ERR(handle)) {
1339 ret = PTR_ERR(handle);
1344 ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
1345 OCFS2_JOURNAL_ACCESS_WRITE);
1351 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1352 /* set extended attribute in external block. */
1353 ret = ocfs2_extend_trans(handle,
1354 OCFS2_INODE_UPDATE_CREDITS +
1355 OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1360 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1361 OCFS2_JOURNAL_ACCESS_WRITE);
1369 * Set value in local, include set tree root in local.
1370 * This is the first step for value size >INLINE_SIZE.
1372 ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1374 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1375 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1382 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1383 (flag & OCFS2_INLINE_XATTR_FL)) {
1384 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1385 unsigned int xattrsize = osb->s_xattr_inline_size;
1388 * Adjust extent record count or inline data size
1389 * to reserve space for extended attribute.
1391 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1392 struct ocfs2_inline_data *idata = &di->id2.i_data;
1393 le16_add_cpu(&idata->id_count, -xattrsize);
1394 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1395 struct ocfs2_extent_list *el = &di->id2.i_list;
1396 le16_add_cpu(&el->l_count, -(xattrsize /
1397 sizeof(struct ocfs2_extent_rec)));
1399 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1401 /* Update xattr flag */
1402 spin_lock(&oi->ip_lock);
1403 oi->ip_dyn_features |= flag;
1404 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1405 spin_unlock(&oi->ip_lock);
1406 /* Update inode ctime */
1407 inode->i_ctime = CURRENT_TIME;
1408 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1409 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1411 ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1416 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1418 if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1420 * Set value outside in B tree.
1421 * This is the second step for value size > INLINE_SIZE.
1423 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1424 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, offs);
1430 * If set value outside failed, we have to clean
1431 * the junk tree root we have already set in local.
1433 ret2 = ocfs2_xattr_cleanup(inode, xi, xs, offs);
1443 static int ocfs2_remove_value_outside(struct inode*inode,
1444 struct buffer_head *bh,
1445 struct ocfs2_xattr_header *header)
1449 for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1450 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1452 if (!ocfs2_xattr_is_local(entry)) {
1453 struct ocfs2_xattr_value_root *xv;
1456 val = (void *)header +
1457 le16_to_cpu(entry->xe_name_offset);
1458 xv = (struct ocfs2_xattr_value_root *)
1459 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
1460 ret = ocfs2_xattr_value_truncate(inode, bh, xv, 0);
1471 static int ocfs2_xattr_ibody_remove(struct inode *inode,
1472 struct buffer_head *di_bh)
1475 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1476 struct ocfs2_xattr_header *header;
1479 header = (struct ocfs2_xattr_header *)
1480 ((void *)di + inode->i_sb->s_blocksize -
1481 le16_to_cpu(di->i_xattr_inline_size));
1483 ret = ocfs2_remove_value_outside(inode, di_bh, header);
1488 static int ocfs2_xattr_block_remove(struct inode *inode,
1489 struct buffer_head *blk_bh)
1491 struct ocfs2_xattr_block *xb;
1494 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1495 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1496 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1497 ret = ocfs2_remove_value_outside(inode, blk_bh, header);
1499 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
1504 static int ocfs2_xattr_free_block(struct inode *inode,
1507 struct inode *xb_alloc_inode;
1508 struct buffer_head *xb_alloc_bh = NULL;
1509 struct buffer_head *blk_bh = NULL;
1510 struct ocfs2_xattr_block *xb;
1511 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1517 ret = ocfs2_read_block(inode, block, &blk_bh,
1524 /*Verify the signature of xattr block*/
1525 if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
1526 strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
1531 ret = ocfs2_xattr_block_remove(inode, blk_bh);
1537 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1538 blk = le64_to_cpu(xb->xb_blkno);
1539 bit = le16_to_cpu(xb->xb_suballoc_bit);
1540 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1542 xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1543 EXTENT_ALLOC_SYSTEM_INODE,
1544 le16_to_cpu(xb->xb_suballoc_slot));
1545 if (!xb_alloc_inode) {
1550 mutex_lock(&xb_alloc_inode->i_mutex);
1552 ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1558 handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
1559 if (IS_ERR(handle)) {
1560 ret = PTR_ERR(handle);
1565 ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1570 ocfs2_commit_trans(osb, handle);
1572 ocfs2_inode_unlock(xb_alloc_inode, 1);
1573 brelse(xb_alloc_bh);
1575 mutex_unlock(&xb_alloc_inode->i_mutex);
1576 iput(xb_alloc_inode);
1583 * ocfs2_xattr_remove()
1585 * Free extended attribute resources associated with this inode.
1587 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1589 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1590 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1594 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1597 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1600 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1601 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1608 if (di->i_xattr_loc) {
1609 ret = ocfs2_xattr_free_block(inode,
1610 le64_to_cpu(di->i_xattr_loc));
1617 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1618 OCFS2_INODE_UPDATE_CREDITS);
1619 if (IS_ERR(handle)) {
1620 ret = PTR_ERR(handle);
1624 ret = ocfs2_journal_access(handle, inode, di_bh,
1625 OCFS2_JOURNAL_ACCESS_WRITE);
1631 di->i_xattr_loc = 0;
1633 spin_lock(&oi->ip_lock);
1634 oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1635 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1636 spin_unlock(&oi->ip_lock);
1638 ret = ocfs2_journal_dirty(handle, di_bh);
1642 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1647 static int ocfs2_xattr_has_space_inline(struct inode *inode,
1648 struct ocfs2_dinode *di)
1650 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1651 unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1654 if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1657 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1658 struct ocfs2_inline_data *idata = &di->id2.i_data;
1659 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1660 } else if (ocfs2_inode_is_fast_symlink(inode)) {
1661 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1662 le64_to_cpu(di->i_size);
1664 struct ocfs2_extent_list *el = &di->id2.i_list;
1665 free = (le16_to_cpu(el->l_count) -
1666 le16_to_cpu(el->l_next_free_rec)) *
1667 sizeof(struct ocfs2_extent_rec);
1669 if (free >= xattrsize)
1676 * ocfs2_xattr_ibody_find()
1678 * Find extended attribute in inode block and
1679 * fill search info into struct ocfs2_xattr_search.
1681 static int ocfs2_xattr_ibody_find(struct inode *inode,
1684 struct ocfs2_xattr_search *xs)
1686 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1687 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1691 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1694 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1695 down_read(&oi->ip_alloc_sem);
1696 has_space = ocfs2_xattr_has_space_inline(inode, di);
1697 up_read(&oi->ip_alloc_sem);
1702 xs->xattr_bh = xs->inode_bh;
1703 xs->end = (void *)di + inode->i_sb->s_blocksize;
1704 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1705 xs->header = (struct ocfs2_xattr_header *)
1706 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1708 xs->header = (struct ocfs2_xattr_header *)
1709 (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
1710 xs->base = (void *)xs->header;
1711 xs->here = xs->header->xh_entries;
1713 /* Find the named attribute. */
1714 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1715 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1716 if (ret && ret != -ENODATA)
1718 xs->not_found = ret;
1725 * ocfs2_xattr_ibody_set()
1727 * Set, replace or remove an extended attribute into inode block.
1730 static int ocfs2_xattr_ibody_set(struct inode *inode,
1731 struct ocfs2_xattr_info *xi,
1732 struct ocfs2_xattr_search *xs)
1734 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1735 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1738 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1741 down_write(&oi->ip_alloc_sem);
1742 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1743 if (!ocfs2_xattr_has_space_inline(inode, di)) {
1749 ret = ocfs2_xattr_set_entry(inode, xi, xs,
1750 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
1752 up_write(&oi->ip_alloc_sem);
1758 * ocfs2_xattr_block_find()
1760 * Find extended attribute in external block and
1761 * fill search info into struct ocfs2_xattr_search.
1763 static int ocfs2_xattr_block_find(struct inode *inode,
1766 struct ocfs2_xattr_search *xs)
1768 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1769 struct buffer_head *blk_bh = NULL;
1770 struct ocfs2_xattr_block *xb;
1773 if (!di->i_xattr_loc)
1776 ret = ocfs2_read_block(inode,
1777 le64_to_cpu(di->i_xattr_loc),
1778 &blk_bh, OCFS2_BH_CACHED);
1783 /*Verify the signature of xattr block*/
1784 if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
1785 strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
1790 xs->xattr_bh = blk_bh;
1791 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1793 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1794 xs->header = &xb->xb_attrs.xb_header;
1795 xs->base = (void *)xs->header;
1796 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
1797 xs->here = xs->header->xh_entries;
1799 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1801 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
1805 if (ret && ret != -ENODATA) {
1806 xs->xattr_bh = NULL;
1809 xs->not_found = ret;
1818 * When all the xattrs are deleted from index btree, the ocfs2_xattr_tree
1819 * will be erased and ocfs2_xattr_block will have its ocfs2_xattr_header
1822 static int ocfs2_restore_xattr_block(struct inode *inode,
1823 struct ocfs2_xattr_search *xs)
1827 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1828 struct ocfs2_xattr_block *xb =
1829 (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1830 struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
1831 u16 xb_flags = le16_to_cpu(xb->xb_flags);
1833 BUG_ON(!(xb_flags & OCFS2_XATTR_INDEXED) ||
1834 le16_to_cpu(el->l_next_free_rec) != 0);
1836 handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1837 if (IS_ERR(handle)) {
1838 ret = PTR_ERR(handle);
1843 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1844 OCFS2_JOURNAL_ACCESS_WRITE);
1850 memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
1851 offsetof(struct ocfs2_xattr_block, xb_attrs));
1853 xb->xb_flags = cpu_to_le16(xb_flags & ~OCFS2_XATTR_INDEXED);
1855 ocfs2_journal_dirty(handle, xs->xattr_bh);
1858 ocfs2_commit_trans(osb, handle);
1864 * ocfs2_xattr_block_set()
1866 * Set, replace or remove an extended attribute into external block.
1869 static int ocfs2_xattr_block_set(struct inode *inode,
1870 struct ocfs2_xattr_info *xi,
1871 struct ocfs2_xattr_search *xs)
1873 struct buffer_head *new_bh = NULL;
1874 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1875 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1876 struct ocfs2_alloc_context *meta_ac = NULL;
1877 handle_t *handle = NULL;
1878 struct ocfs2_xattr_block *xblk = NULL;
1879 u16 suballoc_bit_start;
1884 if (!xs->xattr_bh) {
1886 * Alloc one external block for extended attribute
1889 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
1894 handle = ocfs2_start_trans(osb,
1895 OCFS2_XATTR_BLOCK_CREATE_CREDITS);
1896 if (IS_ERR(handle)) {
1897 ret = PTR_ERR(handle);
1901 ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
1902 OCFS2_JOURNAL_ACCESS_CREATE);
1908 ret = ocfs2_claim_metadata(osb, handle, meta_ac, 1,
1909 &suballoc_bit_start, &num_got,
1916 new_bh = sb_getblk(inode->i_sb, first_blkno);
1917 ocfs2_set_new_buffer_uptodate(inode, new_bh);
1919 ret = ocfs2_journal_access(handle, inode, new_bh,
1920 OCFS2_JOURNAL_ACCESS_CREATE);
1926 /* Initialize ocfs2_xattr_block */
1927 xs->xattr_bh = new_bh;
1928 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
1929 memset(xblk, 0, inode->i_sb->s_blocksize);
1930 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
1931 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
1932 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1933 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
1934 xblk->xb_blkno = cpu_to_le64(first_blkno);
1936 xs->header = &xblk->xb_attrs.xb_header;
1937 xs->base = (void *)xs->header;
1938 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
1939 xs->here = xs->header->xh_entries;
1942 ret = ocfs2_journal_dirty(handle, new_bh);
1947 di->i_xattr_loc = cpu_to_le64(first_blkno);
1948 ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1952 ocfs2_commit_trans(osb, handle);
1955 ocfs2_free_alloc_context(meta_ac);
1959 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1961 if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
1962 /* Set extended attribute into external block */
1963 ret = ocfs2_xattr_set_entry(inode, xi, xs, OCFS2_HAS_XATTR_FL);
1964 if (!ret || ret != -ENOSPC)
1967 ret = ocfs2_xattr_create_index_block(inode, xs);
1972 ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs);
1973 if (!ret && xblk->xb_attrs.xb_root.xt_list.l_next_free_rec == 0)
1974 ret = ocfs2_restore_xattr_block(inode, xs);
1984 * Set, replace or remove an extended attribute for this inode.
1985 * value is NULL to remove an existing extended attribute, else either
1986 * create or replace an extended attribute.
1988 int ocfs2_xattr_set(struct inode *inode,
1995 struct buffer_head *di_bh = NULL;
1996 struct ocfs2_dinode *di;
1998 u16 i, blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2000 struct ocfs2_xattr_info xi = {
2001 .name_index = name_index,
2004 .value_len = value_len,
2007 struct ocfs2_xattr_search xis = {
2008 .not_found = -ENODATA,
2011 struct ocfs2_xattr_search xbs = {
2012 .not_found = -ENODATA,
2015 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2018 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2023 xis.inode_bh = xbs.inode_bh = di_bh;
2024 di = (struct ocfs2_dinode *)di_bh->b_data;
2026 down_write(&OCFS2_I(inode)->ip_xattr_sem);
2028 * Scan inode and external block to find the same name
2029 * extended attribute and collect search infomation.
2031 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2034 if (xis.not_found) {
2035 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2040 if (xis.not_found && xbs.not_found) {
2042 if (flags & XATTR_REPLACE)
2049 if (flags & XATTR_CREATE)
2054 /* Remove existing extended attribute */
2056 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2057 else if (!xbs.not_found)
2058 ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2060 /* We always try to set extended attribute into inode first*/
2061 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2062 if (!ret && !xbs.not_found) {
2064 * If succeed and that extended attribute existing in
2065 * external block, then we will remove it.
2069 ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2070 } else if (ret == -ENOSPC) {
2071 if (di->i_xattr_loc && !xbs.xattr_bh) {
2072 ret = ocfs2_xattr_block_find(inode, name_index,
2078 * If no space in inode, we will set extended attribute
2079 * into external block.
2081 ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2084 if (!xis.not_found) {
2086 * If succeed and that extended attribute
2087 * existing in inode, we will remove it.
2091 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2096 up_write(&OCFS2_I(inode)->ip_xattr_sem);
2097 ocfs2_inode_unlock(inode, 1);
2099 brelse(xbs.xattr_bh);
2100 for (i = 0; i < blk_per_bucket; i++)
2101 brelse(xbs.bucket.bhs[i]);
2107 * Find the xattr extent rec which may contains name_hash.
2108 * e_cpos will be the first name hash of the xattr rec.
2109 * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2111 static int ocfs2_xattr_get_rec(struct inode *inode,
2116 struct ocfs2_extent_list *el)
2119 struct buffer_head *eb_bh = NULL;
2120 struct ocfs2_extent_block *eb;
2121 struct ocfs2_extent_rec *rec = NULL;
2124 if (el->l_tree_depth) {
2125 ret = ocfs2_find_leaf(inode, el, name_hash, &eb_bh);
2131 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2134 if (el->l_tree_depth) {
2135 ocfs2_error(inode->i_sb,
2136 "Inode %lu has non zero tree depth in "
2137 "xattr tree block %llu\n", inode->i_ino,
2138 (unsigned long long)eb_bh->b_blocknr);
2144 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2145 rec = &el->l_recs[i];
2147 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2148 e_blkno = le64_to_cpu(rec->e_blkno);
2154 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2155 "record (%u, %u, 0) in xattr", inode->i_ino,
2156 le32_to_cpu(rec->e_cpos),
2157 ocfs2_rec_clusters(el, rec));
2162 *p_blkno = le64_to_cpu(rec->e_blkno);
2163 *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2165 *e_cpos = le32_to_cpu(rec->e_cpos);
2171 typedef int (xattr_bucket_func)(struct inode *inode,
2172 struct ocfs2_xattr_bucket *bucket,
2175 static int ocfs2_find_xe_in_bucket(struct inode *inode,
2176 struct buffer_head *header_bh,
2183 int i, ret = 0, cmp = 1, block_off, new_offset;
2184 struct ocfs2_xattr_header *xh =
2185 (struct ocfs2_xattr_header *)header_bh->b_data;
2186 size_t name_len = strlen(name);
2187 struct ocfs2_xattr_entry *xe = NULL;
2188 struct buffer_head *name_bh = NULL;
2192 * We don't use binary search in the bucket because there
2193 * may be multiple entries with the same name hash.
2195 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2196 xe = &xh->xh_entries[i];
2198 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2200 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2203 cmp = name_index - ocfs2_xattr_get_type(xe);
2205 cmp = name_len - xe->xe_name_len;
2209 ret = ocfs2_xattr_bucket_get_name_value(inode,
2219 ret = ocfs2_read_block(inode,
2220 header_bh->b_blocknr + block_off,
2221 &name_bh, OCFS2_BH_CACHED);
2226 xe_name = name_bh->b_data + new_offset;
2228 cmp = memcmp(name, xe_name, name_len);
2244 * Find the specified xattr entry in a series of buckets.
2245 * This series start from p_blkno and last for num_clusters.
2246 * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2247 * the num of the valid buckets.
2249 * Return the buffer_head this xattr should reside in. And if the xattr's
2250 * hash is in the gap of 2 buckets, return the lower bucket.
2252 static int ocfs2_xattr_bucket_find(struct inode *inode,
2259 struct ocfs2_xattr_search *xs)
2262 struct buffer_head *bh = NULL;
2263 struct buffer_head *lower_bh = NULL;
2264 struct ocfs2_xattr_header *xh = NULL;
2265 struct ocfs2_xattr_entry *xe = NULL;
2267 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2268 int low_bucket = 0, bucket, high_bucket;
2272 ret = ocfs2_read_block(inode, p_blkno, &bh, OCFS2_BH_CACHED);
2278 xh = (struct ocfs2_xattr_header *)bh->b_data;
2279 high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
2281 while (low_bucket <= high_bucket) {
2284 bucket = (low_bucket + high_bucket) / 2;
2286 blkno = p_blkno + bucket * blk_per_bucket;
2288 ret = ocfs2_read_block(inode, blkno, &bh, OCFS2_BH_CACHED);
2294 xh = (struct ocfs2_xattr_header *)bh->b_data;
2295 xe = &xh->xh_entries[0];
2296 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
2297 high_bucket = bucket - 1;
2302 * Check whether the hash of the last entry in our
2303 * bucket is larger than the search one. for an empty
2304 * bucket, the last one is also the first one.
2307 xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
2309 last_hash = le32_to_cpu(xe->xe_name_hash);
2311 /* record lower_bh which may be the insert place. */
2316 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
2317 low_bucket = bucket + 1;
2321 /* the searched xattr should reside in this bucket if exists. */
2322 ret = ocfs2_find_xe_in_bucket(inode, lower_bh,
2323 name_index, name, name_hash,
2333 * Record the bucket we have found.
2334 * When the xattr's hash value is in the gap of 2 buckets, we will
2335 * always set it to the previous bucket.
2339 * We can't find any bucket whose first name_hash is less
2340 * than the find name_hash.
2342 BUG_ON(bh->b_blocknr != p_blkno);
2346 xs->bucket.bhs[0] = lower_bh;
2347 xs->bucket.xh = (struct ocfs2_xattr_header *)
2348 xs->bucket.bhs[0]->b_data;
2351 xs->header = xs->bucket.xh;
2352 xs->base = xs->bucket.bhs[0]->b_data;
2353 xs->end = xs->base + inode->i_sb->s_blocksize;
2357 * If we have found the xattr enty, read all the blocks in
2360 ret = ocfs2_read_blocks(inode, xs->bucket.bhs[0]->b_blocknr + 1,
2361 blk_per_bucket - 1, &xs->bucket.bhs[1],
2368 xs->here = &xs->header->xh_entries[index];
2369 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
2370 (unsigned long long)xs->bucket.bhs[0]->b_blocknr, index);
2380 static int ocfs2_xattr_index_block_find(struct inode *inode,
2381 struct buffer_head *root_bh,
2384 struct ocfs2_xattr_search *xs)
2387 struct ocfs2_xattr_block *xb =
2388 (struct ocfs2_xattr_block *)root_bh->b_data;
2389 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
2390 struct ocfs2_extent_list *el = &xb_root->xt_list;
2392 u32 first_hash, num_clusters = 0;
2393 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
2395 if (le16_to_cpu(el->l_next_free_rec) == 0)
2398 mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
2399 name, name_hash, name_index);
2401 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
2408 BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
2410 mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
2411 "in the rec is %u\n", num_clusters, p_blkno, first_hash);
2413 ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
2414 p_blkno, first_hash, num_clusters, xs);
2420 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
2423 xattr_bucket_func *func,
2427 int blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2428 u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
2429 u32 num_buckets = clusters * bpc;
2430 struct ocfs2_xattr_bucket bucket;
2432 memset(&bucket, 0, sizeof(bucket));
2434 mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
2437 for (i = 0; i < num_buckets; i++, blkno += blk_per_bucket) {
2438 ret = ocfs2_read_blocks(inode, blkno, blk_per_bucket,
2439 bucket.bhs, OCFS2_BH_CACHED);
2445 bucket.xh = (struct ocfs2_xattr_header *)bucket.bhs[0]->b_data;
2447 * The real bucket num in this series of blocks is stored
2448 * in the 1st bucket.
2451 num_buckets = le16_to_cpu(bucket.xh->xh_num_buckets);
2453 mlog(0, "iterating xattr bucket %llu, first hash %u\n", blkno,
2454 le32_to_cpu(bucket.xh->xh_entries[0].xe_name_hash));
2456 ret = func(inode, &bucket, para);
2463 for (j = 0; j < blk_per_bucket; j++)
2464 brelse(bucket.bhs[j]);
2465 memset(&bucket, 0, sizeof(bucket));
2469 for (j = 0; j < blk_per_bucket; j++)
2470 brelse(bucket.bhs[j]);
2475 struct ocfs2_xattr_tree_list {
2481 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
2482 struct ocfs2_xattr_header *xh,
2489 if (index < 0 || index >= le16_to_cpu(xh->xh_count))
2492 name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
2494 *block_off = name_offset >> inode->i_sb->s_blocksize_bits;
2495 *new_offset = name_offset % inode->i_sb->s_blocksize;
2500 static int ocfs2_list_xattr_bucket(struct inode *inode,
2501 struct ocfs2_xattr_bucket *bucket,
2505 struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
2506 int i, block_off, new_offset;
2507 const char *prefix, *name;
2509 for (i = 0 ; i < le16_to_cpu(bucket->xh->xh_count); i++) {
2510 struct ocfs2_xattr_entry *entry = &bucket->xh->xh_entries[i];
2511 type = ocfs2_xattr_get_type(entry);
2512 prefix = ocfs2_xattr_prefix(type);
2515 ret = ocfs2_xattr_bucket_get_name_value(inode,
2523 name = (const char *)bucket->bhs[block_off]->b_data +
2525 ret = ocfs2_xattr_list_entry(xl->buffer,
2529 entry->xe_name_len);
2538 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
2539 struct ocfs2_xattr_tree_root *xt,
2543 struct ocfs2_extent_list *el = &xt->xt_list;
2545 u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
2547 struct ocfs2_xattr_tree_list xl = {
2549 .buffer_size = buffer_size,
2553 if (le16_to_cpu(el->l_next_free_rec) == 0)
2556 while (name_hash > 0) {
2557 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
2558 &e_cpos, &num_clusters, el);
2564 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
2565 ocfs2_list_xattr_bucket,
2575 name_hash = e_cpos - 1;
2583 static int cmp_xe(const void *a, const void *b)
2585 const struct ocfs2_xattr_entry *l = a, *r = b;
2586 u32 l_hash = le32_to_cpu(l->xe_name_hash);
2587 u32 r_hash = le32_to_cpu(r->xe_name_hash);
2589 if (l_hash > r_hash)
2591 if (l_hash < r_hash)
2596 static void swap_xe(void *a, void *b, int size)
2598 struct ocfs2_xattr_entry *l = a, *r = b, tmp;
2601 memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
2602 memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
2606 * When the ocfs2_xattr_block is filled up, new bucket will be created
2607 * and all the xattr entries will be moved to the new bucket.
2608 * Note: we need to sort the entries since they are not saved in order
2609 * in the ocfs2_xattr_block.
2611 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
2612 struct buffer_head *xb_bh,
2613 struct buffer_head *xh_bh,
2614 struct buffer_head *data_bh)
2616 int i, blocksize = inode->i_sb->s_blocksize;
2617 u16 offset, size, off_change;
2618 struct ocfs2_xattr_entry *xe;
2619 struct ocfs2_xattr_block *xb =
2620 (struct ocfs2_xattr_block *)xb_bh->b_data;
2621 struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
2622 struct ocfs2_xattr_header *xh =
2623 (struct ocfs2_xattr_header *)xh_bh->b_data;
2624 u16 count = le16_to_cpu(xb_xh->xh_count);
2625 char *target = xh_bh->b_data, *src = xb_bh->b_data;
2627 mlog(0, "cp xattr from block %llu to bucket %llu\n",
2628 (unsigned long long)xb_bh->b_blocknr,
2629 (unsigned long long)xh_bh->b_blocknr);
2631 memset(xh_bh->b_data, 0, blocksize);
2633 memset(data_bh->b_data, 0, blocksize);
2635 * Since the xe_name_offset is based on ocfs2_xattr_header,
2636 * there is a offset change corresponding to the change of
2637 * ocfs2_xattr_header's position.
2639 off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
2640 xe = &xb_xh->xh_entries[count - 1];
2641 offset = le16_to_cpu(xe->xe_name_offset) + off_change;
2642 size = blocksize - offset;
2644 /* copy all the names and values. */
2646 target = data_bh->b_data;
2647 memcpy(target + offset, src + offset, size);
2649 /* Init new header now. */
2650 xh->xh_count = xb_xh->xh_count;
2651 xh->xh_num_buckets = cpu_to_le16(1);
2652 xh->xh_name_value_len = cpu_to_le16(size);
2653 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
2655 /* copy all the entries. */
2656 target = xh_bh->b_data;
2657 offset = offsetof(struct ocfs2_xattr_header, xh_entries);
2658 size = count * sizeof(struct ocfs2_xattr_entry);
2659 memcpy(target + offset, (char *)xb_xh + offset, size);
2661 /* Change the xe offset for all the xe because of the move. */
2662 off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
2663 offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
2664 for (i = 0; i < count; i++)
2665 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
2667 mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
2668 offset, size, off_change);
2670 sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
2675 * After we move xattr from block to index btree, we have to
2676 * update ocfs2_xattr_search to the new xe and base.
2678 * When the entry is in xattr block, xattr_bh indicates the storage place.
2679 * While if the entry is in index b-tree, "bucket" indicates the
2680 * real place of the xattr.
2682 static int ocfs2_xattr_update_xattr_search(struct inode *inode,
2683 struct ocfs2_xattr_search *xs,
2684 struct buffer_head *old_bh,
2685 struct buffer_head *new_bh)
2688 char *buf = old_bh->b_data;
2689 struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
2690 struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
2691 int i, blocksize = inode->i_sb->s_blocksize;
2692 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2694 xs->bucket.bhs[0] = new_bh;
2696 xs->bucket.xh = (struct ocfs2_xattr_header *)xs->bucket.bhs[0]->b_data;
2697 xs->header = xs->bucket.xh;
2699 xs->base = new_bh->b_data;
2700 xs->end = xs->base + inode->i_sb->s_blocksize;
2702 if (!xs->not_found) {
2703 if (OCFS2_XATTR_BUCKET_SIZE != blocksize) {
2704 ret = ocfs2_read_blocks(inode,
2705 xs->bucket.bhs[0]->b_blocknr + 1,
2706 blk_per_bucket - 1, &xs->bucket.bhs[1],
2713 i = xs->here - old_xh->xh_entries;
2714 xs->here = &xs->header->xh_entries[i];
2721 static int ocfs2_xattr_create_index_block(struct inode *inode,
2722 struct ocfs2_xattr_search *xs)
2724 int ret, credits = OCFS2_SUBALLOC_ALLOC;
2728 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2729 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2730 struct ocfs2_alloc_context *data_ac;
2731 struct buffer_head *xh_bh = NULL, *data_bh = NULL;
2732 struct buffer_head *xb_bh = xs->xattr_bh;
2733 struct ocfs2_xattr_block *xb =
2734 (struct ocfs2_xattr_block *)xb_bh->b_data;
2735 struct ocfs2_xattr_tree_root *xr;
2736 u16 xb_flags = le16_to_cpu(xb->xb_flags);
2737 u16 bpb = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2739 mlog(0, "create xattr index block for %llu\n",
2740 (unsigned long long)xb_bh->b_blocknr);
2742 BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
2744 ret = ocfs2_reserve_clusters(osb, 1, &data_ac);
2752 * We can use this lock for now, and maybe move to a dedicated mutex
2753 * if performance becomes a problem later.
2755 down_write(&oi->ip_alloc_sem);
2758 * 3 more credits, one for xattr block update, one for the 1st block
2759 * of the new xattr bucket and one for the value/data.
2762 handle = ocfs2_start_trans(osb, credits);
2763 if (IS_ERR(handle)) {
2764 ret = PTR_ERR(handle);
2769 ret = ocfs2_journal_access(handle, inode, xb_bh,
2770 OCFS2_JOURNAL_ACCESS_WRITE);
2776 ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off, &len);
2783 * The bucket may spread in many blocks, and
2784 * we will only touch the 1st block and the last block
2785 * in the whole bucket(one for entry and one for data).
2787 blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
2789 mlog(0, "allocate 1 cluster from %llu to xattr block\n", blkno);
2791 xh_bh = sb_getblk(inode->i_sb, blkno);
2798 ocfs2_set_new_buffer_uptodate(inode, xh_bh);
2800 ret = ocfs2_journal_access(handle, inode, xh_bh,
2801 OCFS2_JOURNAL_ACCESS_CREATE);
2808 data_bh = sb_getblk(inode->i_sb, blkno + bpb - 1);
2815 ocfs2_set_new_buffer_uptodate(inode, data_bh);
2817 ret = ocfs2_journal_access(handle, inode, data_bh,
2818 OCFS2_JOURNAL_ACCESS_CREATE);
2825 ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xh_bh, data_bh);
2827 ocfs2_journal_dirty(handle, xh_bh);
2829 ocfs2_journal_dirty(handle, data_bh);
2831 ocfs2_xattr_update_xattr_search(inode, xs, xb_bh, xh_bh);
2833 /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
2834 memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
2835 offsetof(struct ocfs2_xattr_block, xb_attrs));
2837 xr = &xb->xb_attrs.xb_root;
2838 xr->xt_clusters = cpu_to_le32(1);
2839 xr->xt_last_eb_blk = 0;
2840 xr->xt_list.l_tree_depth = 0;
2841 xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
2842 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2844 xr->xt_list.l_recs[0].e_cpos = 0;
2845 xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
2846 xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
2848 xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
2850 ret = ocfs2_journal_dirty(handle, xb_bh);
2857 ocfs2_commit_trans(osb, handle);
2860 up_write(&oi->ip_alloc_sem);
2864 ocfs2_free_alloc_context(data_ac);
2872 static int cmp_xe_offset(const void *a, const void *b)
2874 const struct ocfs2_xattr_entry *l = a, *r = b;
2875 u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
2876 u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
2878 if (l_name_offset < r_name_offset)
2880 if (l_name_offset > r_name_offset)
2886 * defrag a xattr bucket if we find that the bucket has some
2887 * holes beteen name/value pairs.
2888 * We will move all the name/value pairs to the end of the bucket
2889 * so that we can spare some space for insertion.
2891 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
2892 struct ocfs2_xattr_bucket *bucket)
2895 size_t end, offset, len, value_len;
2896 struct ocfs2_xattr_header *xh;
2897 char *entries, *buf, *bucket_buf = NULL;
2898 u64 blkno = bucket->bhs[0]->b_blocknr;
2899 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2901 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2902 size_t blocksize = inode->i_sb->s_blocksize;
2904 struct buffer_head **bhs;
2905 struct ocfs2_xattr_entry *xe;
2907 bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
2912 ret = ocfs2_read_blocks(inode, blkno, blk_per_bucket, bhs,
2918 * In order to make the operation more efficient and generic,
2919 * we copy all the blocks into a contiguous memory and do the
2920 * defragment there, so if anything is error, we will not touch
2923 bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
2930 for (i = 0; i < blk_per_bucket; i++, buf += blocksize)
2931 memcpy(buf, bhs[i]->b_data, blocksize);
2933 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), blk_per_bucket);
2934 if (IS_ERR(handle)) {
2935 ret = PTR_ERR(handle);
2941 for (i = 0; i < blk_per_bucket; i++) {
2942 ret = ocfs2_journal_access(handle, inode, bhs[i],
2943 OCFS2_JOURNAL_ACCESS_WRITE);
2950 xh = (struct ocfs2_xattr_header *)bucket_buf;
2951 entries = (char *)xh->xh_entries;
2952 xh_free_start = le16_to_cpu(xh->xh_free_start);
2954 mlog(0, "adjust xattr bucket in %llu, count = %u, "
2955 "xh_free_start = %u, xh_name_value_len = %u.\n",
2956 blkno, le16_to_cpu(xh->xh_count), xh_free_start,
2957 le16_to_cpu(xh->xh_name_value_len));
2960 * sort all the entries by their offset.
2961 * the largest will be the first, so that we can
2962 * move them to the end one by one.
2964 sort(entries, le16_to_cpu(xh->xh_count),
2965 sizeof(struct ocfs2_xattr_entry),
2966 cmp_xe_offset, swap_xe);
2968 /* Move all name/values to the end of the bucket. */
2969 xe = xh->xh_entries;
2970 end = OCFS2_XATTR_BUCKET_SIZE;
2971 for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
2972 offset = le16_to_cpu(xe->xe_name_offset);
2973 if (ocfs2_xattr_is_local(xe))
2974 value_len = OCFS2_XATTR_SIZE(
2975 le64_to_cpu(xe->xe_value_size));
2977 value_len = OCFS2_XATTR_ROOT_SIZE;
2978 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
2981 * We must make sure that the name/value pair
2982 * exist in the same block. So adjust end to
2983 * the previous block end if needed.
2985 if (((end - len) / blocksize !=
2986 (end - 1) / blocksize))
2987 end = end - end % blocksize;
2989 if (end > offset + len) {
2990 memmove(bucket_buf + end - len,
2991 bucket_buf + offset, len);
2992 xe->xe_name_offset = cpu_to_le16(end - len);
2995 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
2996 "bucket %llu\n", (unsigned long long)blkno);
3001 mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
3002 "bucket %llu\n", (unsigned long long)blkno);
3004 if (xh_free_start == end)
3007 memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
3008 xh->xh_free_start = cpu_to_le16(end);
3010 /* sort the entries by their name_hash. */
3011 sort(entries, le16_to_cpu(xh->xh_count),
3012 sizeof(struct ocfs2_xattr_entry),
3016 for (i = 0; i < blk_per_bucket; i++, buf += blocksize) {
3017 memcpy(bhs[i]->b_data, buf, blocksize);
3018 ocfs2_journal_dirty(handle, bhs[i]);
3022 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
3026 for (i = 0; i < blk_per_bucket; i++)
3036 * Move half nums of the xattr bucket in the previous cluster to this new
3037 * cluster. We only touch the last cluster of the previous extend record.
3039 * first_bh is the first buffer_head of a series of bucket in the same
3040 * extent rec and header_bh is the header of one bucket in this cluster.
3041 * They will be updated if we move the data header_bh contains to the new
3042 * cluster. first_hash will be set as the 1st xe's name_hash of the new cluster.
3044 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3046 struct buffer_head **first_bh,
3047 struct buffer_head **header_bh,
3053 int i, ret, credits;
3054 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3055 int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3056 int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3057 int blocksize = inode->i_sb->s_blocksize;
3058 struct buffer_head *old_bh, *new_bh, *prev_bh, *new_first_bh = NULL;
3059 struct ocfs2_xattr_header *new_xh;
3060 struct ocfs2_xattr_header *xh =
3061 (struct ocfs2_xattr_header *)((*first_bh)->b_data);
3063 BUG_ON(le16_to_cpu(xh->xh_num_buckets) < num_buckets);
3064 BUG_ON(OCFS2_XATTR_BUCKET_SIZE == osb->s_clustersize);
3066 prev_bh = *first_bh;
3068 xh = (struct ocfs2_xattr_header *)prev_bh->b_data;
3070 prev_blkno += (num_clusters - 1) * bpc + bpc / 2;
3072 mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3073 prev_blkno, new_blkno);
3076 * We need to update the 1st half of the new cluster and
3077 * 1 more for the update of the 1st bucket of the previous
3080 credits = bpc / 2 + 1;
3081 ret = ocfs2_extend_trans(handle, credits);
3087 ret = ocfs2_journal_access(handle, inode, prev_bh,
3088 OCFS2_JOURNAL_ACCESS_WRITE);
3094 for (i = 0; i < bpc / 2; i++, prev_blkno++, new_blkno++) {
3095 old_bh = new_bh = NULL;
3096 new_bh = sb_getblk(inode->i_sb, new_blkno);
3103 ocfs2_set_new_buffer_uptodate(inode, new_bh);
3105 ret = ocfs2_journal_access(handle, inode, new_bh,
3106 OCFS2_JOURNAL_ACCESS_CREATE);
3113 ret = ocfs2_read_block(inode, prev_blkno,
3114 &old_bh, OCFS2_BH_CACHED);
3121 memcpy(new_bh->b_data, old_bh->b_data, blocksize);
3124 new_xh = (struct ocfs2_xattr_header *)new_bh->b_data;
3125 new_xh->xh_num_buckets = cpu_to_le16(num_buckets / 2);
3128 *first_hash = le32_to_cpu(
3129 new_xh->xh_entries[0].xe_name_hash);
3130 new_first_bh = new_bh;
3131 get_bh(new_first_bh);
3134 ocfs2_journal_dirty(handle, new_bh);
3136 if (*header_bh == old_bh) {
3138 *header_bh = new_bh;
3142 *first_bh = new_first_bh;
3149 le16_add_cpu(&xh->xh_num_buckets, -(num_buckets / 2));
3151 ocfs2_journal_dirty(handle, prev_bh);
3154 brelse(new_first_bh);
3158 static int ocfs2_read_xattr_bucket(struct inode *inode,
3160 struct buffer_head **bhs,
3164 u16 i, blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3167 return ocfs2_read_blocks(inode, blkno,
3168 blk_per_bucket, bhs,
3171 for (i = 0; i < blk_per_bucket; i++) {
3172 bhs[i] = sb_getblk(inode->i_sb, blkno + i);
3173 if (bhs[i] == NULL) {
3178 ocfs2_set_new_buffer_uptodate(inode, bhs[i]);
3185 * Move half num of the xattrs in old bucket(blk) to new bucket(new_blk).
3186 * first_hash will record the 1st hash of the new bucket.
3188 static int ocfs2_half_xattr_bucket(struct inode *inode,
3193 int new_bucket_head)
3196 u16 count, start, len, name_value_len, xe_len, name_offset;
3197 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3198 struct buffer_head **s_bhs, **t_bhs = NULL;
3199 struct ocfs2_xattr_header *xh;
3200 struct ocfs2_xattr_entry *xe;
3201 int blocksize = inode->i_sb->s_blocksize;
3203 mlog(0, "move half of xattrs from bucket %llu to %llu\n",
3206 s_bhs = kcalloc(blk_per_bucket, sizeof(struct buffer_head *), GFP_NOFS);
3210 ret = ocfs2_read_xattr_bucket(inode, blk, s_bhs, 0);
3216 ret = ocfs2_journal_access(handle, inode, s_bhs[0],
3217 OCFS2_JOURNAL_ACCESS_WRITE);
3223 t_bhs = kcalloc(blk_per_bucket, sizeof(struct buffer_head *), GFP_NOFS);
3229 ret = ocfs2_read_xattr_bucket(inode, new_blk, t_bhs, new_bucket_head);
3235 for (i = 0; i < blk_per_bucket; i++) {
3236 ret = ocfs2_journal_access(handle, inode, t_bhs[i],
3237 OCFS2_JOURNAL_ACCESS_CREATE);
3244 /* copy the whole bucket to the new first. */
3245 for (i = 0; i < blk_per_bucket; i++)
3246 memcpy(t_bhs[i]->b_data, s_bhs[i]->b_data, blocksize);
3248 /* update the new bucket. */
3249 xh = (struct ocfs2_xattr_header *)t_bhs[0]->b_data;
3250 count = le16_to_cpu(xh->xh_count);
3254 * Calculate the total name/value len and xh_free_start for
3255 * the old bucket first.
3257 name_offset = OCFS2_XATTR_BUCKET_SIZE;
3259 for (i = 0; i < start; i++) {
3260 xe = &xh->xh_entries[i];
3261 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3262 if (ocfs2_xattr_is_local(xe))
3264 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3266 xe_len += OCFS2_XATTR_ROOT_SIZE;
3267 name_value_len += xe_len;
3268 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3269 name_offset = le16_to_cpu(xe->xe_name_offset);
3273 * Now begin the modification to the new bucket.
3275 * In the new bucket, We just move the xattr entry to the beginning
3276 * and don't touch the name/value. So there will be some holes in the
3277 * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3280 xe = &xh->xh_entries[start];
3281 len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3282 mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3283 (int)((char *)xe - (char *)xh),
3284 (int)((char *)xh->xh_entries - (char *)xh));
3285 memmove((char *)xh->xh_entries, (char *)xe, len);
3286 xe = &xh->xh_entries[count - start];
3287 len = sizeof(struct ocfs2_xattr_entry) * start;
3288 memset((char *)xe, 0, len);
3290 le16_add_cpu(&xh->xh_count, -start);
3291 le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3293 /* Calculate xh_free_start for the new bucket. */
3294 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3295 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3296 xe = &xh->xh_entries[i];
3297 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3298 if (ocfs2_xattr_is_local(xe))
3300 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3302 xe_len += OCFS2_XATTR_ROOT_SIZE;
3303 if (le16_to_cpu(xe->xe_name_offset) <
3304 le16_to_cpu(xh->xh_free_start))
3305 xh->xh_free_start = xe->xe_name_offset;
3308 /* set xh->xh_num_buckets for the new xh. */
3309 if (new_bucket_head)
3310 xh->xh_num_buckets = cpu_to_le16(1);
3312 xh->xh_num_buckets = 0;
3314 for (i = 0; i < blk_per_bucket; i++) {
3315 ocfs2_journal_dirty(handle, t_bhs[i]);
3320 /* store the first_hash of the new bucket. */
3322 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3325 * Now only update the 1st block of the old bucket.
3326 * Please note that the entry has been sorted already above.
3328 xh = (struct ocfs2_xattr_header *)s_bhs[0]->b_data;
3329 memset(&xh->xh_entries[start], 0,
3330 sizeof(struct ocfs2_xattr_entry) * (count - start));
3331 xh->xh_count = cpu_to_le16(start);
3332 xh->xh_free_start = cpu_to_le16(name_offset);
3333 xh->xh_name_value_len = cpu_to_le16(name_value_len);
3335 ocfs2_journal_dirty(handle, s_bhs[0]);
3341 for (i = 0; i < blk_per_bucket; i++)
3347 for (i = 0; i < blk_per_bucket; i++)
3356 * Copy xattr from one bucket to another bucket.
3358 * The caller must make sure that the journal transaction
3359 * has enough space for journaling.
3361 static int ocfs2_cp_xattr_bucket(struct inode *inode,
3368 int blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3369 int blocksize = inode->i_sb->s_blocksize;
3370 struct buffer_head **s_bhs, **t_bhs = NULL;
3372 BUG_ON(s_blkno == t_blkno);
3374 mlog(0, "cp bucket %llu to %llu, target is %d\n",
3375 s_blkno, t_blkno, t_is_new);
3377 s_bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
3382 ret = ocfs2_read_xattr_bucket(inode, s_blkno, s_bhs, 0);
3386 t_bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
3393 ret = ocfs2_read_xattr_bucket(inode, t_blkno, t_bhs, t_is_new);
3397 for (i = 0; i < blk_per_bucket; i++) {
3398 ret = ocfs2_journal_access(handle, inode, t_bhs[i],
3399 OCFS2_JOURNAL_ACCESS_WRITE);
3404 for (i = 0; i < blk_per_bucket; i++) {
3405 memcpy(t_bhs[i]->b_data, s_bhs[i]->b_data, blocksize);
3406 ocfs2_journal_dirty(handle, t_bhs[i]);
3411 for (i = 0; i < blk_per_bucket; i++)
3417 for (i = 0; i < blk_per_bucket; i++)
3426 * Copy one xattr cluster from src_blk to to_blk.
3427 * The to_blk will become the first bucket header of the cluster, so its
3428 * xh_num_buckets will be initialized as the bucket num in the cluster.
3430 static int ocfs2_cp_xattr_cluster(struct inode *inode,
3432 struct buffer_head *first_bh,
3437 int i, ret, credits;
3438 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3439 int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3440 int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3441 struct buffer_head *bh = NULL;
3442 struct ocfs2_xattr_header *xh;
3443 u64 to_blk_start = to_blk;
3445 mlog(0, "cp xattrs from cluster %llu to %llu\n", src_blk, to_blk);
3448 * We need to update the new cluster and 1 more for the update of
3449 * the 1st bucket of the previous extent rec.
3452 ret = ocfs2_extend_trans(handle, credits);
3458 ret = ocfs2_journal_access(handle, inode, first_bh,
3459 OCFS2_JOURNAL_ACCESS_WRITE);
3465 for (i = 0; i < num_buckets; i++) {
3466 ret = ocfs2_cp_xattr_bucket(inode, handle,
3467 src_blk, to_blk, 1);
3473 src_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3474 to_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3477 /* update the old bucket header. */
3478 xh = (struct ocfs2_xattr_header *)first_bh->b_data;
3479 le16_add_cpu(&xh->xh_num_buckets, -num_buckets);
3481 ocfs2_journal_dirty(handle, first_bh);
3483 /* update the new bucket header. */
3484 ret = ocfs2_read_block(inode, to_blk_start, &bh, OCFS2_BH_CACHED);
3490 ret = ocfs2_journal_access(handle, inode, bh,
3491 OCFS2_JOURNAL_ACCESS_WRITE);
3497 xh = (struct ocfs2_xattr_header *)bh->b_data;
3498 xh->xh_num_buckets = cpu_to_le16(num_buckets);
3500 ocfs2_journal_dirty(handle, bh);
3503 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3510 * Move half of the xattrs in this cluster to the new cluster.
3511 * This function should only be called when bucket size == cluster size.
3512 * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
3514 static int ocfs2_half_xattr_cluster(struct inode *inode,
3520 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3521 int ret, credits = 2 * blk_per_bucket;
3523 BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
3525 ret = ocfs2_extend_trans(handle, credits);
3531 /* Move half of the xattr in start_blk to the next bucket. */
3532 return ocfs2_half_xattr_bucket(inode, handle, prev_blk,
3533 new_blk, first_hash, 1);
3537 * Move some xattrs from the old cluster to the new one since they are not
3538 * contiguous in ocfs2 xattr tree.
3540 * new_blk starts a new separate cluster, and we will move some xattrs from
3541 * prev_blk to it. v_start will be set as the first name hash value in this
3542 * new cluster so that it can be used as e_cpos during tree insertion and
3543 * don't collide with our original b-tree operations. first_bh and header_bh
3544 * will also be updated since they will be used in ocfs2_extend_xattr_bucket
3545 * to extend the insert bucket.
3547 * The problem is how much xattr should we move to the new one and when should
3548 * we update first_bh and header_bh?
3549 * 1. If cluster size > bucket size, that means the previous cluster has more
3550 * than 1 bucket, so just move half nums of bucket into the new cluster and
3551 * update the first_bh and header_bh if the insert bucket has been moved
3552 * to the new cluster.
3553 * 2. If cluster_size == bucket_size:
3554 * a) If the previous extent rec has more than one cluster and the insert
3555 * place isn't in the last cluster, copy the entire last cluster to the
3556 * new one. This time, we don't need to upate the first_bh and header_bh
3557 * since they will not be moved into the new cluster.
3558 * b) Otherwise, move the bottom half of the xattrs in the last cluster into
3559 * the new one. And we set the extend flag to zero if the insert place is
3560 * moved into the new allocated cluster since no extend is needed.
3562 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
3564 struct buffer_head **first_bh,
3565 struct buffer_head **header_bh,
3573 int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3575 mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
3576 prev_blk, prev_clusters, new_blk);
3578 if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1)
3579 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
3588 u64 last_blk = prev_blk + bpc * (prev_clusters - 1);
3590 if (prev_clusters > 1 && (*header_bh)->b_blocknr != last_blk)
3591 ret = ocfs2_cp_xattr_cluster(inode, handle, *first_bh,
3595 ret = ocfs2_half_xattr_cluster(inode, handle,
3599 if ((*header_bh)->b_blocknr == last_blk && extend)
3608 * Add a new cluster for xattr storage.
3610 * If the new cluster is contiguous with the previous one, it will be
3611 * appended to the same extent record, and num_clusters will be updated.
3612 * If not, we will insert a new extent for it and move some xattrs in
3613 * the last cluster into the new allocated one.
3614 * We also need to limit the maximum size of a btree leaf, otherwise we'll
3615 * lose the benefits of hashing because we'll have to search large leaves.
3616 * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
3619 * first_bh is the first block of the previous extent rec and header_bh
3620 * indicates the bucket we will insert the new xattrs. They will be updated
3621 * when the header_bh is moved into the new cluster.
3623 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
3624 struct buffer_head *root_bh,
3625 struct buffer_head **first_bh,
3626 struct buffer_head **header_bh,
3633 u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3634 u32 prev_clusters = *num_clusters;
3635 u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
3637 handle_t *handle = NULL;
3638 struct ocfs2_alloc_context *data_ac = NULL;
3639 struct ocfs2_alloc_context *meta_ac = NULL;
3640 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3641 struct ocfs2_extent_tree et;
3643 mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
3644 "previous xattr blkno = %llu\n",
3645 (unsigned long long)OCFS2_I(inode)->ip_blkno,
3646 prev_cpos, prev_blkno);
3648 ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
3650 ret = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
3651 &data_ac, &meta_ac);
3657 credits = ocfs2_calc_extend_credits(osb->sb, et.et_root_el,
3659 handle = ocfs2_start_trans(osb, credits);
3660 if (IS_ERR(handle)) {
3661 ret = PTR_ERR(handle);
3667 ret = ocfs2_journal_access(handle, inode, root_bh,
3668 OCFS2_JOURNAL_ACCESS_WRITE);
3674 ret = __ocfs2_claim_clusters(osb, handle, data_ac, 1,
3675 clusters_to_add, &bit_off, &num_bits);
3682 BUG_ON(num_bits > clusters_to_add);
3684 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
3685 mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
3686 num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
3688 if (prev_blkno + prev_clusters * bpc == block &&
3689 (prev_clusters + num_bits) << osb->s_clustersize_bits <=
3690 OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
3692 * If this cluster is contiguous with the old one and
3693 * adding this new cluster, we don't surpass the limit of
3694 * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
3695 * initialized and used like other buckets in the previous
3697 * So add it as a contiguous one. The caller will handle
3700 v_start = prev_cpos + prev_clusters;
3701 *num_clusters = prev_clusters + num_bits;
3702 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
3705 ret = ocfs2_adjust_xattr_cross_cluster(inode,
3720 if (handle->h_buffer_credits < credits) {
3722 * The journal has been restarted before, and don't
3723 * have enough space for the insertion, so extend it
3726 ret = ocfs2_extend_trans(handle, credits);
3732 mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
3733 num_bits, block, v_start);
3734 ret = ocfs2_insert_extent(osb, handle, inode, &et, v_start, block,
3735 num_bits, 0, meta_ac);
3741 ret = ocfs2_journal_dirty(handle, root_bh);
3749 ocfs2_commit_trans(osb, handle);
3751 ocfs2_free_alloc_context(data_ac);
3753 ocfs2_free_alloc_context(meta_ac);
3759 * Extend a new xattr bucket and move xattrs to the end one by one until
3760 * We meet with start_bh. Only move half of the xattrs to the bucket after it.
3762 static int ocfs2_extend_xattr_bucket(struct inode *inode,
3763 struct buffer_head *first_bh,
3764 struct buffer_head *start_bh,
3768 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3769 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3770 u64 start_blk = start_bh->b_blocknr, end_blk;
3771 u32 num_buckets = num_clusters * ocfs2_xattr_buckets_per_cluster(osb);
3773 struct ocfs2_xattr_header *first_xh =
3774 (struct ocfs2_xattr_header *)first_bh->b_data;
3775 u16 bucket = le16_to_cpu(first_xh->xh_num_buckets);
3777 mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
3778 "from %llu, len = %u\n", start_blk,
3779 (unsigned long long)first_bh->b_blocknr, num_clusters);
3781 BUG_ON(bucket >= num_buckets);
3783 end_blk = first_bh->b_blocknr + (bucket - 1) * blk_per_bucket;
3786 * We will touch all the buckets after the start_bh(include it).
3787 * Add one more bucket and modify the first_bh.
3789 credits = end_blk - start_blk + 2 * blk_per_bucket + 1;
3790 handle = ocfs2_start_trans(osb, credits);
3791 if (IS_ERR(handle)) {
3792 ret = PTR_ERR(handle);
3798 ret = ocfs2_journal_access(handle, inode, first_bh,
3799 OCFS2_JOURNAL_ACCESS_WRITE);
3805 while (end_blk != start_blk) {
3806 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
3807 end_blk + blk_per_bucket, 0);
3810 end_blk -= blk_per_bucket;
3813 /* Move half of the xattr in start_blk to the next bucket. */
3814 ret = ocfs2_half_xattr_bucket(inode, handle, start_blk,
3815 start_blk + blk_per_bucket, NULL, 0);
3817 le16_add_cpu(&first_xh->xh_num_buckets, 1);
3818 ocfs2_journal_dirty(handle, first_bh);
3821 ocfs2_commit_trans(osb, handle);
3827 * Add new xattr bucket in an extent record and adjust the buckets accordingly.
3828 * xb_bh is the ocfs2_xattr_block.
3829 * We will move all the buckets starting from header_bh to the next place. As
3830 * for this one, half num of its xattrs will be moved to the next one.
3832 * We will allocate a new cluster if current cluster is full and adjust
3833 * header_bh and first_bh if the insert place is moved to the new cluster.
3835 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
3836 struct buffer_head *xb_bh,
3837 struct buffer_head *header_bh)
3839 struct ocfs2_xattr_header *first_xh = NULL;
3840 struct buffer_head *first_bh = NULL;
3841 struct ocfs2_xattr_block *xb =
3842 (struct ocfs2_xattr_block *)xb_bh->b_data;
3843 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3844 struct ocfs2_extent_list *el = &xb_root->xt_list;
3845 struct ocfs2_xattr_header *xh =
3846 (struct ocfs2_xattr_header *)header_bh->b_data;
3847 u32 name_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3848 struct super_block *sb = inode->i_sb;
3849 struct ocfs2_super *osb = OCFS2_SB(sb);
3850 int ret, num_buckets, extend = 1;
3852 u32 e_cpos, num_clusters;
3854 mlog(0, "Add new xattr bucket starting form %llu\n",
3855 (unsigned long long)header_bh->b_blocknr);
3858 * Add refrence for header_bh here because it may be
3859 * changed in ocfs2_add_new_xattr_cluster and we need
3860 * to free it in the end.
3864 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
3871 ret = ocfs2_read_block(inode, p_blkno,
3872 &first_bh, OCFS2_BH_CACHED);
3878 num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
3879 first_xh = (struct ocfs2_xattr_header *)first_bh->b_data;
3881 if (num_buckets == le16_to_cpu(first_xh->xh_num_buckets)) {
3882 ret = ocfs2_add_new_xattr_cluster(inode,
3897 ret = ocfs2_extend_xattr_bucket(inode,
3909 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
3910 struct ocfs2_xattr_bucket *bucket,
3913 int block_off = offs >> inode->i_sb->s_blocksize_bits;
3915 offs = offs % inode->i_sb->s_blocksize;
3916 return bucket->bhs[block_off]->b_data + offs;
3920 * Handle the normal xattr set, including replace, delete and new.
3922 * Note: "local" indicates the real data's locality. So we can't
3923 * just its bucket locality by its length.
3925 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
3926 struct ocfs2_xattr_info *xi,
3927 struct ocfs2_xattr_search *xs,
3931 struct ocfs2_xattr_entry *last, *xe;
3932 int name_len = strlen(xi->name);
3933 struct ocfs2_xattr_header *xh = xs->header;
3934 u16 count = le16_to_cpu(xh->xh_count), start;
3935 size_t blocksize = inode->i_sb->s_blocksize;
3937 size_t offs, size, new_size;
3939 last = &xh->xh_entries[count];
3940 if (!xs->not_found) {
3942 offs = le16_to_cpu(xe->xe_name_offset);
3943 if (ocfs2_xattr_is_local(xe))
3944 size = OCFS2_XATTR_SIZE(name_len) +
3945 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3947 size = OCFS2_XATTR_SIZE(name_len) +
3948 OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
3951 * If the new value will be stored outside, xi->value has been
3952 * initalized as an empty ocfs2_xattr_value_root, and the same
3953 * goes with xi->value_len, so we can set new_size safely here.
3954 * See ocfs2_xattr_set_in_bucket.
3956 new_size = OCFS2_XATTR_SIZE(name_len) +
3957 OCFS2_XATTR_SIZE(xi->value_len);
3959 le16_add_cpu(&xh->xh_name_value_len, -size);
3961 if (new_size > size)
3962 goto set_new_name_value;
3964 /* Now replace the old value with new one. */
3966 xe->xe_value_size = cpu_to_le64(xi->value_len);
3968 xe->xe_value_size = 0;
3970 val = ocfs2_xattr_bucket_get_val(inode,
3972 memset(val + OCFS2_XATTR_SIZE(name_len), 0,
3973 size - OCFS2_XATTR_SIZE(name_len));
3974 if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
3975 memcpy(val + OCFS2_XATTR_SIZE(name_len),
3976 xi->value, xi->value_len);
3978 le16_add_cpu(&xh->xh_name_value_len, new_size);
3979 ocfs2_xattr_set_local(xe, local);
3983 * Remove the old entry if there is more than one.
3984 * We don't remove the last entry so that we can
3985 * use it to indicate the hash value of the empty
3989 le16_add_cpu(&xh->xh_count, -1);
3992 (void *)last - (void *)xe);
3994 sizeof(struct ocfs2_xattr_entry));
3997 cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4002 /* find a new entry for insert. */
4003 int low = 0, high = count - 1, tmp;
4004 struct ocfs2_xattr_entry *tmp_xe;
4006 while (low <= high && count) {
4007 tmp = (low + high) / 2;
4008 tmp_xe = &xh->xh_entries[tmp];
4010 if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
4012 else if (name_hash <
4013 le32_to_cpu(tmp_xe->xe_name_hash))
4021 xe = &xh->xh_entries[low];
4023 memmove(xe + 1, xe, (void *)last - (void *)xe);
4025 le16_add_cpu(&xh->xh_count, 1);
4026 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
4027 xe->xe_name_hash = cpu_to_le32(name_hash);
4028 xe->xe_name_len = name_len;
4029 ocfs2_xattr_set_type(xe, xi->name_index);
4033 /* Insert the new name+value. */
4034 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4037 * We must make sure that the name/value pair
4038 * exists in the same block.
4040 offs = le16_to_cpu(xh->xh_free_start);
4041 start = offs - size;
4043 if (start >> inode->i_sb->s_blocksize_bits !=
4044 (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4045 offs = offs - offs % blocksize;
4046 xh->xh_free_start = cpu_to_le16(offs);
4049 val = ocfs2_xattr_bucket_get_val(inode,
4050 &xs->bucket, offs - size);
4051 xe->xe_name_offset = cpu_to_le16(offs - size);
4053 memset(val, 0, size);
4054 memcpy(val, xi->name, name_len);
4055 memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4057 xe->xe_value_size = cpu_to_le64(xi->value_len);
4058 ocfs2_xattr_set_local(xe, local);
4060 le16_add_cpu(&xh->xh_free_start, -size);
4061 le16_add_cpu(&xh->xh_name_value_len, size);
4066 static int ocfs2_xattr_bucket_handle_journal(struct inode *inode,
4068 struct ocfs2_xattr_search *xs,
4069 struct buffer_head **bhs,
4072 int ret = 0, off, block_off;
4073 struct ocfs2_xattr_entry *xe = xs->here;
4076 * First calculate all the blocks we should journal_access
4077 * and journal_dirty. The first block should always be touched.
4079 ret = ocfs2_journal_dirty(handle, bhs[0]);
4083 /* calc the data. */
4084 off = le16_to_cpu(xe->xe_name_offset);
4085 block_off = off >> inode->i_sb->s_blocksize_bits;
4086 ret = ocfs2_journal_dirty(handle, bhs[block_off]);
4094 * Set the xattr entry in the specified bucket.
4095 * The bucket is indicated by xs->bucket and it should have the enough
4096 * space for the xattr insertion.
4098 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4099 struct ocfs2_xattr_info *xi,
4100 struct ocfs2_xattr_search *xs,
4105 handle_t *handle = NULL;
4106 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4107 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4109 mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4110 (unsigned long)xi->value_len, xi->name_index,
4111 (unsigned long long)xs->bucket.bhs[0]->b_blocknr);
4113 if (!xs->bucket.bhs[1]) {
4114 ret = ocfs2_read_blocks(inode,
4115 xs->bucket.bhs[0]->b_blocknr + 1,
4116 blk_per_bucket - 1, &xs->bucket.bhs[1],
4124 handle = ocfs2_start_trans(osb, blk_per_bucket);
4125 if (IS_ERR(handle)) {
4126 ret = PTR_ERR(handle);
4132 for (i = 0; i < blk_per_bucket; i++) {
4133 ret = ocfs2_journal_access(handle, inode, xs->bucket.bhs[i],
4134 OCFS2_JOURNAL_ACCESS_WRITE);
4141 ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local);
4143 /*Only dirty the blocks we have touched in set xattr. */
4144 ret = ocfs2_xattr_bucket_handle_journal(inode, handle, xs,
4145 xs->bucket.bhs, blk_per_bucket);
4149 ocfs2_commit_trans(osb, handle);
4154 static int ocfs2_xattr_value_update_size(struct inode *inode,
4155 struct buffer_head *xe_bh,
4156 struct ocfs2_xattr_entry *xe,
4160 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4161 handle_t *handle = NULL;
4163 handle = ocfs2_start_trans(osb, 1);
4164 if (handle == NULL) {
4170 ret = ocfs2_journal_access(handle, inode, xe_bh,
4171 OCFS2_JOURNAL_ACCESS_WRITE);
4177 xe->xe_value_size = cpu_to_le64(new_size);
4179 ret = ocfs2_journal_dirty(handle, xe_bh);
4184 ocfs2_commit_trans(osb, handle);
4190 * Truncate the specified xe_off entry in xattr bucket.
4191 * bucket is indicated by header_bh and len is the new length.
4192 * Both the ocfs2_xattr_value_root and the entry will be updated here.
4194 * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4196 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4197 struct buffer_head *header_bh,
4203 struct buffer_head *value_bh = NULL;
4204 struct ocfs2_xattr_value_root *xv;
4205 struct ocfs2_xattr_entry *xe;
4206 struct ocfs2_xattr_header *xh =
4207 (struct ocfs2_xattr_header *)header_bh->b_data;
4208 size_t blocksize = inode->i_sb->s_blocksize;
4210 xe = &xh->xh_entries[xe_off];
4212 BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4214 offset = le16_to_cpu(xe->xe_name_offset) +
4215 OCFS2_XATTR_SIZE(xe->xe_name_len);
4217 value_blk = offset / blocksize;
4219 /* We don't allow ocfs2_xattr_value to be stored in different block. */
4220 BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4221 value_blk += header_bh->b_blocknr;
4223 ret = ocfs2_read_block(inode, value_blk,
4224 &value_bh, OCFS2_BH_CACHED);
4230 xv = (struct ocfs2_xattr_value_root *)
4231 (value_bh->b_data + offset % blocksize);
4233 mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4234 xe_off, (unsigned long long)header_bh->b_blocknr, len);
4235 ret = ocfs2_xattr_value_truncate(inode, value_bh, xv, len);
4241 ret = ocfs2_xattr_value_update_size(inode, header_bh, xe, len);
4252 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4253 struct ocfs2_xattr_search *xs,
4257 struct ocfs2_xattr_entry *xe = xs->here;
4258 struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4260 BUG_ON(!xs->bucket.bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4262 offset = xe - xh->xh_entries;
4263 ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket.bhs[0],
4271 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4272 struct ocfs2_xattr_search *xs,
4277 struct ocfs2_xattr_value_root *xv;
4278 struct ocfs2_xattr_entry *xe = xs->here;
4280 BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4282 offset = le16_to_cpu(xe->xe_name_offset) +
4283 OCFS2_XATTR_SIZE(xe->xe_name_len);
4285 xv = (struct ocfs2_xattr_value_root *)(xs->base + offset);
4287 return __ocfs2_xattr_set_value_outside(inode, xv, val, value_len);
4290 static int ocfs2_rm_xattr_cluster(struct inode *inode,
4291 struct buffer_head *root_bh,
4297 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4298 struct inode *tl_inode = osb->osb_tl_inode;
4300 struct ocfs2_xattr_block *xb =
4301 (struct ocfs2_xattr_block *)root_bh->b_data;
4302 struct ocfs2_alloc_context *meta_ac = NULL;
4303 struct ocfs2_cached_dealloc_ctxt dealloc;
4304 struct ocfs2_extent_tree et;
4306 ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4308 ocfs2_init_dealloc_ctxt(&dealloc);
4310 mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4311 cpos, len, (unsigned long long)blkno);
4313 ocfs2_remove_xattr_clusters_from_cache(inode, blkno, len);
4315 ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
4321 mutex_lock(&tl_inode->i_mutex);
4323 if (ocfs2_truncate_log_needs_flush(osb)) {
4324 ret = __ocfs2_flush_truncate_log(osb);
4331 handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
4332 if (handle == NULL) {
4338 ret = ocfs2_journal_access(handle, inode, root_bh,
4339 OCFS2_JOURNAL_ACCESS_WRITE);
4345 ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
4352 le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4354 ret = ocfs2_journal_dirty(handle, root_bh);
4360 ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4365 ocfs2_commit_trans(osb, handle);
4367 ocfs2_schedule_truncate_log_flush(osb, 1);
4369 mutex_unlock(&tl_inode->i_mutex);
4372 ocfs2_free_alloc_context(meta_ac);
4374 ocfs2_run_deallocs(osb, &dealloc);
4379 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4380 struct ocfs2_xattr_search *xs)
4382 handle_t *handle = NULL;
4383 struct ocfs2_xattr_header *xh = xs->bucket.xh;
4384 struct ocfs2_xattr_entry *last = &xh->xh_entries[
4385 le16_to_cpu(xh->xh_count) - 1];
4388 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), 1);
4389 if (IS_ERR(handle)) {
4390 ret = PTR_ERR(handle);
4395 ret = ocfs2_journal_access(handle, inode, xs->bucket.bhs[0],
4396 OCFS2_JOURNAL_ACCESS_WRITE);
4402 /* Remove the old entry. */
4403 memmove(xs->here, xs->here + 1,
4404 (void *)last - (void *)xs->here);
4405 memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4406 le16_add_cpu(&xh->xh_count, -1);
4408 ret = ocfs2_journal_dirty(handle, xs->bucket.bhs[0]);
4412 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
4416 * Set the xattr name/value in the bucket specified in xs.
4418 * As the new value in xi may be stored in the bucket or in an outside cluster,
4419 * we divide the whole process into 3 steps:
4420 * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4421 * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4422 * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4423 * 4. If the clusters for the new outside value can't be allocated, we need
4424 * to free the xattr we allocated in set.
4426 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4427 struct ocfs2_xattr_info *xi,
4428 struct ocfs2_xattr_search *xs)
4432 char *val = (char *)xi->value;
4433 struct ocfs2_xattr_entry *xe = xs->here;
4434 u32 name_hash = ocfs2_xattr_name_hash(inode, xi->name,
4437 if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
4439 * We need to truncate the xattr storage first.
4441 * If both the old and new value are stored to
4442 * outside block, we only need to truncate
4443 * the storage and then set the value outside.
4445 * If the new value should be stored within block,
4446 * we should free all the outside block first and
4447 * the modification to the xattr block will be done
4448 * by following steps.
4450 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4451 value_len = xi->value_len;
4455 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4461 goto set_value_outside;
4464 value_len = xi->value_len;
4465 /* So we have to handle the inside block change now. */
4466 if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4468 * If the new value will be stored outside of block,
4469 * initalize a new empty value root and insert it first.
4472 xi->value = &def_xv;
4473 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
4476 ret = ocfs2_xattr_set_entry_in_bucket(inode, xi, xs, name_hash, local);
4482 if (value_len <= OCFS2_XATTR_INLINE_SIZE)
4485 /* allocate the space now for the outside block storage. */
4486 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4491 if (xs->not_found) {
4493 * We can't allocate enough clusters for outside
4494 * storage and we have allocated xattr already,
4495 * so need to remove it.
4497 ocfs2_xattr_bucket_remove_xs(inode, xs);
4503 ret = ocfs2_xattr_bucket_set_value_outside(inode, xs, val, value_len);
4508 /* check whether the xattr bucket is filled up with the same hash value. */
4509 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
4510 struct ocfs2_xattr_bucket *bucket)
4512 struct ocfs2_xattr_header *xh = bucket->xh;
4514 if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
4515 xh->xh_entries[0].xe_name_hash) {
4516 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
4518 (unsigned long long)bucket->bhs[0]->b_blocknr,
4519 le32_to_cpu(xh->xh_entries[0].xe_name_hash));
4526 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
4527 struct ocfs2_xattr_info *xi,
4528 struct ocfs2_xattr_search *xs)
4530 struct ocfs2_xattr_header *xh;
4531 struct ocfs2_xattr_entry *xe;
4532 u16 count, header_size, xh_free_start;
4533 int i, free, max_free, need, old;
4534 size_t value_size = 0, name_len = strlen(xi->name);
4535 size_t blocksize = inode->i_sb->s_blocksize;
4536 int ret, allocation = 0;
4537 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4539 mlog_entry("Set xattr %s in xattr index block\n", xi->name);
4543 count = le16_to_cpu(xh->xh_count);
4544 xh_free_start = le16_to_cpu(xh->xh_free_start);
4545 header_size = sizeof(struct ocfs2_xattr_header) +
4546 count * sizeof(struct ocfs2_xattr_entry);
4547 max_free = OCFS2_XATTR_BUCKET_SIZE -
4548 le16_to_cpu(xh->xh_name_value_len) - header_size;
4550 mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
4551 "of %u which exceed block size\n",
4552 (unsigned long long)xs->bucket.bhs[0]->b_blocknr,
4555 if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4556 value_size = OCFS2_XATTR_ROOT_SIZE;
4558 value_size = OCFS2_XATTR_SIZE(xi->value_len);
4561 need = sizeof(struct ocfs2_xattr_entry) +
4562 OCFS2_XATTR_SIZE(name_len) + value_size;
4564 need = value_size + OCFS2_XATTR_SIZE(name_len);
4567 * We only replace the old value if the new length is smaller
4568 * than the old one. Otherwise we will allocate new space in the
4569 * bucket to store it.
4572 if (ocfs2_xattr_is_local(xe))
4573 old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4575 old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4577 if (old >= value_size)
4581 free = xh_free_start - header_size;
4583 * We need to make sure the new name/value pair
4584 * can exist in the same block.
4586 if (xh_free_start % blocksize < need)
4587 free -= xh_free_start % blocksize;
4589 mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
4590 "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
4591 " %u\n", xs->not_found,
4592 (unsigned long long)xs->bucket.bhs[0]->b_blocknr,
4593 free, need, max_free, le16_to_cpu(xh->xh_free_start),
4594 le16_to_cpu(xh->xh_name_value_len));
4596 if (free < need || count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
4597 if (need <= max_free &&
4598 count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
4600 * We can create the space by defragment. Since only the
4601 * name/value will be moved, the xe shouldn't be changed
4604 ret = ocfs2_defrag_xattr_bucket(inode, &xs->bucket);
4610 xh_free_start = le16_to_cpu(xh->xh_free_start);
4611 free = xh_free_start - header_size;
4612 if (xh_free_start % blocksize < need)
4613 free -= xh_free_start % blocksize;
4618 mlog(0, "Can't get enough space for xattr insert by "
4619 "defragment. Need %u bytes, but we have %d, so "
4620 "allocate new bucket for it.\n", need, free);
4624 * We have to add new buckets or clusters and one
4625 * allocation should leave us enough space for insert.
4630 * We do not allow for overlapping ranges between buckets. And
4631 * the maximum number of collisions we will allow for then is
4632 * one bucket's worth, so check it here whether we need to
4633 * add a new bucket for the insert.
4635 ret = ocfs2_check_xattr_bucket_collision(inode, &xs->bucket);
4641 ret = ocfs2_add_new_xattr_bucket(inode,
4649 for (i = 0; i < blk_per_bucket; i++)
4650 brelse(xs->bucket.bhs[i]);
4652 memset(&xs->bucket, 0, sizeof(xs->bucket));
4654 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
4657 if (ret && ret != -ENODATA)
4659 xs->not_found = ret;
4665 ret = ocfs2_xattr_set_in_bucket(inode, xi, xs);
4671 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
4672 struct ocfs2_xattr_bucket *bucket,
4676 struct ocfs2_xattr_header *xh = bucket->xh;
4678 struct ocfs2_xattr_entry *xe;
4680 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4681 xe = &xh->xh_entries[i];
4682 if (ocfs2_xattr_is_local(xe))
4685 ret = ocfs2_xattr_bucket_value_truncate(inode,
4697 static int ocfs2_delete_xattr_index_block(struct inode *inode,
4698 struct buffer_head *xb_bh)
4700 struct ocfs2_xattr_block *xb =
4701 (struct ocfs2_xattr_block *)xb_bh->b_data;
4702 struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4704 u32 name_hash = UINT_MAX, e_cpos, num_clusters;
4707 if (le16_to_cpu(el->l_next_free_rec) == 0)
4710 while (name_hash > 0) {
4711 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4712 &e_cpos, &num_clusters, el);
4718 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
4719 ocfs2_delete_xattr_in_bucket,
4726 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
4727 p_blkno, e_cpos, num_clusters);
4736 name_hash = e_cpos - 1;
4744 * 'trusted' attributes support
4747 #define XATTR_TRUSTED_PREFIX "trusted."
4749 static size_t ocfs2_xattr_trusted_list(struct inode *inode, char *list,
4750 size_t list_size, const char *name,
4753 const size_t prefix_len = sizeof(XATTR_TRUSTED_PREFIX) - 1;
4754 const size_t total_len = prefix_len + name_len + 1;
4756 if (list && total_len <= list_size) {
4757 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
4758 memcpy(list + prefix_len, name, name_len);
4759 list[prefix_len + name_len] = '\0';
4764 static int ocfs2_xattr_trusted_get(struct inode *inode, const char *name,
4765 void *buffer, size_t size)
4767 if (strcmp(name, "") == 0)
4769 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED, name,
4773 static int ocfs2_xattr_trusted_set(struct inode *inode, const char *name,
4774 const void *value, size_t size, int flags)
4776 if (strcmp(name, "") == 0)
4779 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED, name, value,
4783 struct xattr_handler ocfs2_xattr_trusted_handler = {
4784 .prefix = XATTR_TRUSTED_PREFIX,
4785 .list = ocfs2_xattr_trusted_list,
4786 .get = ocfs2_xattr_trusted_get,
4787 .set = ocfs2_xattr_trusted_set,
4792 * 'user' attributes support
4795 #define XATTR_USER_PREFIX "user."
4797 static size_t ocfs2_xattr_user_list(struct inode *inode, char *list,
4798 size_t list_size, const char *name,
4801 const size_t prefix_len = sizeof(XATTR_USER_PREFIX) - 1;
4802 const size_t total_len = prefix_len + name_len + 1;
4803 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4805 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
4808 if (list && total_len <= list_size) {
4809 memcpy(list, XATTR_USER_PREFIX, prefix_len);
4810 memcpy(list + prefix_len, name, name_len);
4811 list[prefix_len + name_len] = '\0';
4816 static int ocfs2_xattr_user_get(struct inode *inode, const char *name,
4817 void *buffer, size_t size)
4819 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4821 if (strcmp(name, "") == 0)
4823 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
4825 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
4829 static int ocfs2_xattr_user_set(struct inode *inode, const char *name,
4830 const void *value, size_t size, int flags)
4832 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4834 if (strcmp(name, "") == 0)
4836 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
4839 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER, name, value,
4843 struct xattr_handler ocfs2_xattr_user_handler = {
4844 .prefix = XATTR_USER_PREFIX,
4845 .list = ocfs2_xattr_user_list,
4846 .get = ocfs2_xattr_user_get,
4847 .set = ocfs2_xattr_user_set,