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>
41 #define MLOG_MASK_PREFIX ML_XATTR
42 #include <cluster/masklog.h>
55 #include "buffer_head_io.h"
60 struct ocfs2_xattr_def_value_root {
61 struct ocfs2_xattr_value_root xv;
62 struct ocfs2_extent_rec er;
65 struct ocfs2_xattr_bucket {
66 struct buffer_head *bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
67 struct ocfs2_xattr_header *xh;
70 #define OCFS2_XATTR_ROOT_SIZE (sizeof(struct ocfs2_xattr_def_value_root))
71 #define OCFS2_XATTR_INLINE_SIZE 80
73 static struct ocfs2_xattr_def_value_root def_xv = {
74 .xv.xr_list.l_count = cpu_to_le16(1),
77 struct xattr_handler *ocfs2_xattr_handlers[] = {
78 &ocfs2_xattr_user_handler,
79 &ocfs2_xattr_trusted_handler,
83 static struct xattr_handler *ocfs2_xattr_handler_map[] = {
84 [OCFS2_XATTR_INDEX_USER] = &ocfs2_xattr_user_handler,
85 [OCFS2_XATTR_INDEX_TRUSTED] = &ocfs2_xattr_trusted_handler,
88 struct ocfs2_xattr_info {
95 struct ocfs2_xattr_search {
96 struct buffer_head *inode_bh;
98 * xattr_bh point to the block buffer head which has extended attribute
99 * when extended attribute in inode, xattr_bh is equal to inode_bh.
101 struct buffer_head *xattr_bh;
102 struct ocfs2_xattr_header *header;
103 struct ocfs2_xattr_bucket bucket;
106 struct ocfs2_xattr_entry *here;
110 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
111 struct ocfs2_xattr_header *xh,
116 static int ocfs2_xattr_index_block_find(struct inode *inode,
117 struct buffer_head *root_bh,
120 struct ocfs2_xattr_search *xs);
122 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
123 struct ocfs2_xattr_tree_root *xt,
127 static int ocfs2_xattr_create_index_block(struct inode *inode,
128 struct ocfs2_xattr_search *xs);
130 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
131 struct ocfs2_xattr_info *xi,
132 struct ocfs2_xattr_search *xs);
134 static int ocfs2_delete_xattr_index_block(struct inode *inode,
135 struct buffer_head *xb_bh);
137 static inline struct xattr_handler *ocfs2_xattr_handler(int name_index)
139 struct xattr_handler *handler = NULL;
141 if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
142 handler = ocfs2_xattr_handler_map[name_index];
147 static inline u32 ocfs2_xattr_name_hash(struct inode *inode,
153 /* Get hash value of uuid from super block */
154 u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
157 /* hash extended attribute prefix */
158 for (i = 0; i < prefix_len; i++) {
159 hash = (hash << OCFS2_HASH_SHIFT) ^
160 (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
163 /* hash extended attribute name */
164 for (i = 0; i < name_len; i++) {
165 hash = (hash << OCFS2_HASH_SHIFT) ^
166 (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
174 * ocfs2_xattr_hash_entry()
176 * Compute the hash of an extended attribute.
178 static void ocfs2_xattr_hash_entry(struct inode *inode,
179 struct ocfs2_xattr_header *header,
180 struct ocfs2_xattr_entry *entry)
183 struct xattr_handler *handler =
184 ocfs2_xattr_handler(ocfs2_xattr_get_type(entry));
185 char *prefix = handler->prefix;
186 char *name = (char *)header + le16_to_cpu(entry->xe_name_offset);
187 int prefix_len = strlen(handler->prefix);
189 hash = ocfs2_xattr_name_hash(inode, prefix, prefix_len, name,
191 entry->xe_name_hash = cpu_to_le32(hash);
196 static int ocfs2_xattr_extend_allocation(struct inode *inode,
198 struct buffer_head *xattr_bh,
199 struct ocfs2_xattr_value_root *xv)
202 int restart_func = 0;
204 handle_t *handle = NULL;
205 struct ocfs2_alloc_context *data_ac = NULL;
206 struct ocfs2_alloc_context *meta_ac = NULL;
207 enum ocfs2_alloc_restarted why;
208 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
209 struct ocfs2_extent_list *root_el = &xv->xr_list;
210 u32 prev_clusters, logical_start = le32_to_cpu(xv->xr_clusters);
212 mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
216 status = ocfs2_lock_allocators(inode, xattr_bh, root_el,
217 clusters_to_add, 0, &data_ac,
218 &meta_ac, OCFS2_XATTR_VALUE_EXTENT, xv);
224 credits = ocfs2_calc_extend_credits(osb->sb, root_el, clusters_to_add);
225 handle = ocfs2_start_trans(osb, credits);
226 if (IS_ERR(handle)) {
227 status = PTR_ERR(handle);
233 restarted_transaction:
234 status = ocfs2_journal_access(handle, inode, xattr_bh,
235 OCFS2_JOURNAL_ACCESS_WRITE);
241 prev_clusters = le32_to_cpu(xv->xr_clusters);
242 status = ocfs2_add_clusters_in_btree(osb,
253 OCFS2_XATTR_VALUE_EXTENT,
255 if ((status < 0) && (status != -EAGAIN)) {
256 if (status != -ENOSPC)
261 status = ocfs2_journal_dirty(handle, xattr_bh);
267 clusters_to_add -= le32_to_cpu(xv->xr_clusters) - prev_clusters;
269 if (why != RESTART_NONE && clusters_to_add) {
270 if (why == RESTART_META) {
271 mlog(0, "restarting function.\n");
274 BUG_ON(why != RESTART_TRANS);
276 mlog(0, "restarting transaction.\n");
277 /* TODO: This can be more intelligent. */
278 credits = ocfs2_calc_extend_credits(osb->sb,
281 status = ocfs2_extend_trans(handle, credits);
283 /* handle still has to be committed at
289 goto restarted_transaction;
295 ocfs2_commit_trans(osb, handle);
299 ocfs2_free_alloc_context(data_ac);
303 ocfs2_free_alloc_context(meta_ac);
306 if ((!status) && restart_func) {
314 static int __ocfs2_remove_xattr_range(struct inode *inode,
315 struct buffer_head *root_bh,
316 struct ocfs2_xattr_value_root *xv,
317 u32 cpos, u32 phys_cpos, u32 len,
318 struct ocfs2_cached_dealloc_ctxt *dealloc)
321 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
322 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
323 struct inode *tl_inode = osb->osb_tl_inode;
325 struct ocfs2_alloc_context *meta_ac = NULL;
327 ret = ocfs2_lock_allocators(inode, root_bh, &xv->xr_list,
328 0, 1, NULL, &meta_ac,
329 OCFS2_XATTR_VALUE_EXTENT, xv);
335 mutex_lock(&tl_inode->i_mutex);
337 if (ocfs2_truncate_log_needs_flush(osb)) {
338 ret = __ocfs2_flush_truncate_log(osb);
345 handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
346 if (IS_ERR(handle)) {
347 ret = PTR_ERR(handle);
352 ret = ocfs2_journal_access(handle, inode, root_bh,
353 OCFS2_JOURNAL_ACCESS_WRITE);
359 ret = ocfs2_remove_extent(inode, root_bh, cpos, len, handle, meta_ac,
360 dealloc, OCFS2_XATTR_VALUE_EXTENT, xv);
366 le32_add_cpu(&xv->xr_clusters, -len);
368 ret = ocfs2_journal_dirty(handle, root_bh);
374 ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len);
379 ocfs2_commit_trans(osb, handle);
381 mutex_unlock(&tl_inode->i_mutex);
384 ocfs2_free_alloc_context(meta_ac);
389 static int ocfs2_xattr_shrink_size(struct inode *inode,
392 struct buffer_head *root_bh,
393 struct ocfs2_xattr_value_root *xv)
396 u32 trunc_len, cpos, phys_cpos, alloc_size;
398 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
399 struct ocfs2_cached_dealloc_ctxt dealloc;
401 ocfs2_init_dealloc_ctxt(&dealloc);
403 if (old_clusters <= new_clusters)
407 trunc_len = old_clusters - new_clusters;
409 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
410 &alloc_size, &xv->xr_list);
416 if (alloc_size > trunc_len)
417 alloc_size = trunc_len;
419 ret = __ocfs2_remove_xattr_range(inode, root_bh, xv, cpos,
420 phys_cpos, alloc_size,
427 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
428 ocfs2_remove_xattr_clusters_from_cache(inode, block,
431 trunc_len -= alloc_size;
435 ocfs2_schedule_truncate_log_flush(osb, 1);
436 ocfs2_run_deallocs(osb, &dealloc);
441 static int ocfs2_xattr_value_truncate(struct inode *inode,
442 struct buffer_head *root_bh,
443 struct ocfs2_xattr_value_root *xv,
447 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
448 u32 old_clusters = le32_to_cpu(xv->xr_clusters);
450 if (new_clusters == old_clusters)
453 if (new_clusters > old_clusters)
454 ret = ocfs2_xattr_extend_allocation(inode,
455 new_clusters - old_clusters,
458 ret = ocfs2_xattr_shrink_size(inode,
459 old_clusters, new_clusters,
465 static int ocfs2_xattr_list_entries(struct inode *inode,
466 struct ocfs2_xattr_header *header,
467 char *buffer, size_t buffer_size)
469 size_t rest = buffer_size;
472 for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
473 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
474 struct xattr_handler *handler =
475 ocfs2_xattr_handler(ocfs2_xattr_get_type(entry));
478 size_t size = handler->list(inode, buffer, rest,
480 le16_to_cpu(entry->xe_name_offset)),
491 return buffer_size - rest;
494 static int ocfs2_xattr_ibody_list(struct inode *inode,
495 struct ocfs2_dinode *di,
499 struct ocfs2_xattr_header *header = NULL;
500 struct ocfs2_inode_info *oi = OCFS2_I(inode);
503 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
506 header = (struct ocfs2_xattr_header *)
507 ((void *)di + inode->i_sb->s_blocksize -
508 le16_to_cpu(di->i_xattr_inline_size));
510 ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
515 static int ocfs2_xattr_block_list(struct inode *inode,
516 struct ocfs2_dinode *di,
520 struct buffer_head *blk_bh = NULL;
521 struct ocfs2_xattr_block *xb;
524 if (!di->i_xattr_loc)
527 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
528 le64_to_cpu(di->i_xattr_loc),
529 &blk_bh, OCFS2_BH_CACHED, inode);
534 /*Verify the signature of xattr block*/
535 if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
536 strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
541 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
543 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
544 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
545 ret = ocfs2_xattr_list_entries(inode, header,
546 buffer, buffer_size);
548 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
549 ret = ocfs2_xattr_tree_list_index_block(inode, xt,
550 buffer, buffer_size);
558 ssize_t ocfs2_listxattr(struct dentry *dentry,
562 int ret = 0, i_ret = 0, b_ret = 0;
563 struct buffer_head *di_bh = NULL;
564 struct ocfs2_dinode *di = NULL;
565 struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
567 if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
570 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
573 ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
579 di = (struct ocfs2_dinode *)di_bh->b_data;
581 down_read(&oi->ip_xattr_sem);
582 i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
590 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
595 up_read(&oi->ip_xattr_sem);
596 ocfs2_inode_unlock(dentry->d_inode, 0);
600 return i_ret + b_ret;
603 static int ocfs2_xattr_find_entry(int name_index,
605 struct ocfs2_xattr_search *xs)
607 struct ocfs2_xattr_entry *entry;
614 name_len = strlen(name);
616 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
617 cmp = name_index - ocfs2_xattr_get_type(entry);
619 cmp = name_len - entry->xe_name_len;
621 cmp = memcmp(name, (xs->base +
622 le16_to_cpu(entry->xe_name_offset)),
630 return cmp ? -ENODATA : 0;
633 static int ocfs2_xattr_get_value_outside(struct inode *inode,
634 struct ocfs2_xattr_value_root *xv,
638 u32 cpos, p_cluster, num_clusters, bpc, clusters;
641 size_t cplen, blocksize;
642 struct buffer_head *bh = NULL;
643 struct ocfs2_extent_list *el;
646 clusters = le32_to_cpu(xv->xr_clusters);
647 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
648 blocksize = inode->i_sb->s_blocksize;
651 while (cpos < clusters) {
652 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
659 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
660 /* Copy ocfs2_xattr_value */
661 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
662 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), blkno,
663 &bh, OCFS2_BH_CACHED, inode);
669 cplen = len >= blocksize ? blocksize : len;
670 memcpy(buffer, bh->b_data, cplen);
679 cpos += num_clusters;
685 static int ocfs2_xattr_ibody_get(struct inode *inode,
690 struct ocfs2_xattr_search *xs)
692 struct ocfs2_inode_info *oi = OCFS2_I(inode);
693 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
694 struct ocfs2_xattr_value_root *xv;
698 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
701 xs->end = (void *)di + inode->i_sb->s_blocksize;
702 xs->header = (struct ocfs2_xattr_header *)
703 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
704 xs->base = (void *)xs->header;
705 xs->here = xs->header->xh_entries;
707 ret = ocfs2_xattr_find_entry(name_index, name, xs);
710 size = le64_to_cpu(xs->here->xe_value_size);
712 if (size > buffer_size)
714 if (ocfs2_xattr_is_local(xs->here)) {
715 memcpy(buffer, (void *)xs->base +
716 le16_to_cpu(xs->here->xe_name_offset) +
717 OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
719 xv = (struct ocfs2_xattr_value_root *)
720 (xs->base + le16_to_cpu(
721 xs->here->xe_name_offset) +
722 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
723 ret = ocfs2_xattr_get_value_outside(inode, xv,
735 static int ocfs2_xattr_block_get(struct inode *inode,
740 struct ocfs2_xattr_search *xs)
742 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
743 struct buffer_head *blk_bh = NULL;
744 struct ocfs2_xattr_block *xb;
745 struct ocfs2_xattr_value_root *xv;
747 int ret = -ENODATA, name_offset, name_len, block_off, i;
749 if (!di->i_xattr_loc)
752 memset(&xs->bucket, 0, sizeof(xs->bucket));
754 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
755 le64_to_cpu(di->i_xattr_loc),
756 &blk_bh, OCFS2_BH_CACHED, inode);
761 /*Verify the signature of xattr block*/
762 if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
763 strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
768 xs->xattr_bh = blk_bh;
769 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
771 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
772 xs->header = &xb->xb_attrs.xb_header;
773 xs->base = (void *)xs->header;
774 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
775 xs->here = xs->header->xh_entries;
777 ret = ocfs2_xattr_find_entry(name_index, name, xs);
779 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
785 size = le64_to_cpu(xs->here->xe_value_size);
788 if (size > buffer_size)
791 name_offset = le16_to_cpu(xs->here->xe_name_offset);
792 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
793 i = xs->here - xs->header->xh_entries;
795 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
796 ret = ocfs2_xattr_bucket_get_name_value(inode,
801 xs->base = xs->bucket.bhs[block_off]->b_data;
803 if (ocfs2_xattr_is_local(xs->here)) {
804 memcpy(buffer, (void *)xs->base +
805 name_offset + name_len, size);
807 xv = (struct ocfs2_xattr_value_root *)
808 (xs->base + name_offset + name_len);
809 ret = ocfs2_xattr_get_value_outside(inode, xv,
819 for (i = 0; i < OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET; i++)
820 brelse(xs->bucket.bhs[i]);
821 memset(&xs->bucket, 0, sizeof(xs->bucket));
829 * Copy an extended attribute into the buffer provided.
830 * Buffer is NULL to compute the size of buffer required.
832 int ocfs2_xattr_get(struct inode *inode,
839 struct ocfs2_dinode *di = NULL;
840 struct buffer_head *di_bh = NULL;
841 struct ocfs2_inode_info *oi = OCFS2_I(inode);
842 struct ocfs2_xattr_search xis = {
843 .not_found = -ENODATA,
845 struct ocfs2_xattr_search xbs = {
846 .not_found = -ENODATA,
849 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
852 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
855 ret = ocfs2_inode_lock(inode, &di_bh, 0);
860 xis.inode_bh = xbs.inode_bh = di_bh;
861 di = (struct ocfs2_dinode *)di_bh->b_data;
863 down_read(&oi->ip_xattr_sem);
864 ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
867 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
869 up_read(&oi->ip_xattr_sem);
870 ocfs2_inode_unlock(inode, 0);
877 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
878 struct ocfs2_xattr_value_root *xv,
882 int ret = 0, i, cp_len, credits;
883 u16 blocksize = inode->i_sb->s_blocksize;
884 u32 p_cluster, num_clusters;
885 u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
886 u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
888 struct buffer_head *bh = NULL;
891 BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
893 credits = clusters * bpc;
894 handle = ocfs2_start_trans(OCFS2_SB(inode->i_sb), credits);
895 if (IS_ERR(handle)) {
896 ret = PTR_ERR(handle);
901 while (cpos < clusters) {
902 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
903 &num_clusters, &xv->xr_list);
909 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
911 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
912 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), blkno,
913 &bh, OCFS2_BH_CACHED, inode);
919 ret = ocfs2_journal_access(handle,
922 OCFS2_JOURNAL_ACCESS_WRITE);
928 cp_len = value_len > blocksize ? blocksize : value_len;
929 memcpy(bh->b_data, value, cp_len);
932 if (cp_len < blocksize)
933 memset(bh->b_data + cp_len, 0,
936 ret = ocfs2_journal_dirty(handle, bh);
945 * XXX: do we need to empty all the following
946 * blocks in this cluster?
951 cpos += num_clusters;
954 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
961 static int ocfs2_xattr_cleanup(struct inode *inode,
962 struct ocfs2_xattr_info *xi,
963 struct ocfs2_xattr_search *xs,
966 handle_t *handle = NULL;
968 size_t name_len = strlen(xi->name);
969 void *val = xs->base + offs;
970 size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
972 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
973 OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
974 if (IS_ERR(handle)) {
975 ret = PTR_ERR(handle);
979 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
980 OCFS2_JOURNAL_ACCESS_WRITE);
985 /* Decrease xattr count */
986 le16_add_cpu(&xs->header->xh_count, -1);
987 /* Remove the xattr entry and tree root which has already be set*/
988 memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
989 memset(val, 0, size);
991 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
995 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1000 static int ocfs2_xattr_update_entry(struct inode *inode,
1001 struct ocfs2_xattr_info *xi,
1002 struct ocfs2_xattr_search *xs,
1005 handle_t *handle = NULL;
1008 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1009 OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1010 if (IS_ERR(handle)) {
1011 ret = PTR_ERR(handle);
1015 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1016 OCFS2_JOURNAL_ACCESS_WRITE);
1022 xs->here->xe_name_offset = cpu_to_le16(offs);
1023 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1024 if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1025 ocfs2_xattr_set_local(xs->here, 1);
1027 ocfs2_xattr_set_local(xs->here, 0);
1028 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1030 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1034 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1040 * ocfs2_xattr_set_value_outside()
1042 * Set large size value in B tree.
1044 static int ocfs2_xattr_set_value_outside(struct inode *inode,
1045 struct ocfs2_xattr_info *xi,
1046 struct ocfs2_xattr_search *xs,
1049 size_t name_len = strlen(xi->name);
1050 void *val = xs->base + offs;
1051 struct ocfs2_xattr_value_root *xv = NULL;
1052 size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1055 memset(val, 0, size);
1056 memcpy(val, xi->name, name_len);
1057 xv = (struct ocfs2_xattr_value_root *)
1058 (val + OCFS2_XATTR_SIZE(name_len));
1059 xv->xr_clusters = 0;
1060 xv->xr_last_eb_blk = 0;
1061 xv->xr_list.l_tree_depth = 0;
1062 xv->xr_list.l_count = cpu_to_le16(1);
1063 xv->xr_list.l_next_free_rec = 0;
1065 ret = ocfs2_xattr_value_truncate(inode, xs->xattr_bh, xv,
1071 ret = __ocfs2_xattr_set_value_outside(inode, xv, xi->value,
1077 ret = ocfs2_xattr_update_entry(inode, xi, xs, offs);
1085 * ocfs2_xattr_set_entry_local()
1087 * Set, replace or remove extended attribute in local.
1089 static void ocfs2_xattr_set_entry_local(struct inode *inode,
1090 struct ocfs2_xattr_info *xi,
1091 struct ocfs2_xattr_search *xs,
1092 struct ocfs2_xattr_entry *last,
1095 size_t name_len = strlen(xi->name);
1098 if (xi->value && xs->not_found) {
1099 /* Insert the new xattr entry. */
1100 le16_add_cpu(&xs->header->xh_count, 1);
1101 ocfs2_xattr_set_type(last, xi->name_index);
1102 ocfs2_xattr_set_local(last, 1);
1103 last->xe_name_len = name_len;
1109 first_val = xs->base + min_offs;
1110 offs = le16_to_cpu(xs->here->xe_name_offset);
1111 val = xs->base + offs;
1113 if (le64_to_cpu(xs->here->xe_value_size) >
1114 OCFS2_XATTR_INLINE_SIZE)
1115 size = OCFS2_XATTR_SIZE(name_len) +
1116 OCFS2_XATTR_ROOT_SIZE;
1118 size = OCFS2_XATTR_SIZE(name_len) +
1119 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1121 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1122 OCFS2_XATTR_SIZE(xi->value_len)) {
1123 /* The old and the new value have the
1124 same size. Just replace the value. */
1125 ocfs2_xattr_set_local(xs->here, 1);
1126 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1127 /* Clear value bytes. */
1128 memset(val + OCFS2_XATTR_SIZE(name_len),
1130 OCFS2_XATTR_SIZE(xi->value_len));
1131 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1136 /* Remove the old name+value. */
1137 memmove(first_val + size, first_val, val - first_val);
1138 memset(first_val, 0, size);
1139 xs->here->xe_name_hash = 0;
1140 xs->here->xe_name_offset = 0;
1141 ocfs2_xattr_set_local(xs->here, 1);
1142 xs->here->xe_value_size = 0;
1146 /* Adjust all value offsets. */
1147 last = xs->header->xh_entries;
1148 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1149 size_t o = le16_to_cpu(last->xe_name_offset);
1152 last->xe_name_offset = cpu_to_le16(o + size);
1157 /* Remove the old entry. */
1159 memmove(xs->here, xs->here + 1,
1160 (void *)last - (void *)xs->here);
1161 memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1162 le16_add_cpu(&xs->header->xh_count, -1);
1166 /* Insert the new name+value. */
1167 size_t size = OCFS2_XATTR_SIZE(name_len) +
1168 OCFS2_XATTR_SIZE(xi->value_len);
1169 void *val = xs->base + min_offs - size;
1171 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1172 memset(val, 0, size);
1173 memcpy(val, xi->name, name_len);
1174 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1177 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1178 ocfs2_xattr_set_local(xs->here, 1);
1179 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1186 * ocfs2_xattr_set_entry()
1188 * Set extended attribute entry into inode or block.
1190 * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1191 * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1192 * then set value in B tree with set_value_outside().
1194 static int ocfs2_xattr_set_entry(struct inode *inode,
1195 struct ocfs2_xattr_info *xi,
1196 struct ocfs2_xattr_search *xs,
1199 struct ocfs2_xattr_entry *last;
1200 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1201 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1202 size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1204 handle_t *handle = NULL;
1206 struct ocfs2_xattr_info xi_l = {
1207 .name_index = xi->name_index,
1210 .value_len = xi->value_len,
1213 /* Compute min_offs, last and free space. */
1214 last = xs->header->xh_entries;
1216 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1217 size_t offs = le16_to_cpu(last->xe_name_offset);
1218 if (offs < min_offs)
1223 free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
1227 if (!xs->not_found) {
1229 if (ocfs2_xattr_is_local(xs->here))
1230 size = OCFS2_XATTR_SIZE(name_len) +
1231 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1233 size = OCFS2_XATTR_SIZE(name_len) +
1234 OCFS2_XATTR_ROOT_SIZE;
1235 free += (size + sizeof(struct ocfs2_xattr_entry));
1237 /* Check free space in inode or block */
1238 if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1239 if (free < sizeof(struct ocfs2_xattr_entry) +
1240 OCFS2_XATTR_SIZE(name_len) +
1241 OCFS2_XATTR_ROOT_SIZE) {
1245 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1246 xi_l.value = (void *)&def_xv;
1247 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1248 } else if (xi->value) {
1249 if (free < sizeof(struct ocfs2_xattr_entry) +
1250 OCFS2_XATTR_SIZE(name_len) +
1251 OCFS2_XATTR_SIZE(xi->value_len)) {
1257 if (!xs->not_found) {
1258 /* For existing extended attribute */
1259 size_t size = OCFS2_XATTR_SIZE(name_len) +
1260 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1261 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1262 void *val = xs->base + offs;
1264 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1265 /* Replace existing local xattr with tree root */
1266 ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1271 } else if (!ocfs2_xattr_is_local(xs->here)) {
1272 /* For existing xattr which has value outside */
1273 struct ocfs2_xattr_value_root *xv = NULL;
1274 xv = (struct ocfs2_xattr_value_root *)(val +
1275 OCFS2_XATTR_SIZE(name_len));
1277 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1279 * If new value need set outside also,
1280 * first truncate old value to new value,
1281 * then set new value with set_value_outside().
1283 ret = ocfs2_xattr_value_truncate(inode,
1292 ret = __ocfs2_xattr_set_value_outside(inode,
1301 ret = ocfs2_xattr_update_entry(inode,
1310 * If new value need set in local,
1311 * just trucate old value to zero.
1313 ret = ocfs2_xattr_value_truncate(inode,
1323 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1324 OCFS2_INODE_UPDATE_CREDITS);
1325 if (IS_ERR(handle)) {
1326 ret = PTR_ERR(handle);
1331 ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
1332 OCFS2_JOURNAL_ACCESS_WRITE);
1338 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1339 /*set extended attribue in external blcok*/
1340 ret = ocfs2_extend_trans(handle,
1341 OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1346 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1347 OCFS2_JOURNAL_ACCESS_WRITE);
1355 * Set value in local, include set tree root in local.
1356 * This is the first step for value size >INLINE_SIZE.
1358 ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1360 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1361 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1368 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1369 (flag & OCFS2_INLINE_XATTR_FL)) {
1370 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1371 unsigned int xattrsize = osb->s_xattr_inline_size;
1374 * Adjust extent record count or inline data size
1375 * to reserve space for extended attribute.
1377 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1378 struct ocfs2_inline_data *idata = &di->id2.i_data;
1379 le16_add_cpu(&idata->id_count, -xattrsize);
1380 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1381 struct ocfs2_extent_list *el = &di->id2.i_list;
1382 le16_add_cpu(&el->l_count, -(xattrsize /
1383 sizeof(struct ocfs2_extent_rec)));
1385 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1387 /* Update xattr flag */
1388 spin_lock(&oi->ip_lock);
1389 oi->ip_dyn_features |= flag;
1390 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1391 spin_unlock(&oi->ip_lock);
1392 /* Update inode ctime */
1393 inode->i_ctime = CURRENT_TIME;
1394 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1395 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1397 ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1402 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1404 if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1406 * Set value outside in B tree.
1407 * This is the second step for value size > INLINE_SIZE.
1409 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1410 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, offs);
1416 * If set value outside failed, we have to clean
1417 * the junk tree root we have already set in local.
1419 ret2 = ocfs2_xattr_cleanup(inode, xi, xs, offs);
1429 static int ocfs2_xattr_free_block(handle_t *handle,
1430 struct ocfs2_super *osb,
1431 struct ocfs2_xattr_block *xb)
1433 struct inode *xb_alloc_inode;
1434 struct buffer_head *xb_alloc_bh = NULL;
1435 u64 blk = le64_to_cpu(xb->xb_blkno);
1436 u16 bit = le16_to_cpu(xb->xb_suballoc_bit);
1437 u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1440 xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1441 EXTENT_ALLOC_SYSTEM_INODE,
1442 le16_to_cpu(xb->xb_suballoc_slot));
1443 if (!xb_alloc_inode) {
1448 mutex_lock(&xb_alloc_inode->i_mutex);
1450 ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1455 ret = ocfs2_extend_trans(handle, OCFS2_SUBALLOC_FREE);
1460 ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1465 ocfs2_inode_unlock(xb_alloc_inode, 1);
1466 brelse(xb_alloc_bh);
1468 mutex_unlock(&xb_alloc_inode->i_mutex);
1469 iput(xb_alloc_inode);
1474 static int ocfs2_remove_value_outside(struct inode*inode,
1475 struct buffer_head *bh,
1476 struct ocfs2_xattr_header *header)
1480 for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1481 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1483 if (!ocfs2_xattr_is_local(entry)) {
1484 struct ocfs2_xattr_value_root *xv;
1487 val = (void *)header +
1488 le16_to_cpu(entry->xe_name_offset);
1489 xv = (struct ocfs2_xattr_value_root *)
1490 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
1491 ret = ocfs2_xattr_value_truncate(inode, bh, xv, 0);
1502 static int ocfs2_xattr_ibody_remove(struct inode *inode,
1503 struct buffer_head *di_bh)
1506 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1507 struct ocfs2_xattr_header *header;
1510 header = (struct ocfs2_xattr_header *)
1511 ((void *)di + inode->i_sb->s_blocksize -
1512 le16_to_cpu(di->i_xattr_inline_size));
1514 ret = ocfs2_remove_value_outside(inode, di_bh, header);
1519 static int ocfs2_xattr_block_remove(struct inode *inode,
1520 struct buffer_head *blk_bh)
1522 struct ocfs2_xattr_block *xb;
1525 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1526 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1527 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1528 ret = ocfs2_remove_value_outside(inode, blk_bh, header);
1530 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
1536 * ocfs2_xattr_remove()
1538 * Free extended attribute resources associated with this inode.
1540 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1542 struct ocfs2_xattr_block *xb;
1543 struct buffer_head *blk_bh = NULL;
1544 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1545 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1546 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1550 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1553 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1556 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1557 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1563 if (di->i_xattr_loc) {
1564 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
1565 le64_to_cpu(di->i_xattr_loc),
1566 &blk_bh, OCFS2_BH_CACHED, inode);
1571 /*Verify the signature of xattr block*/
1572 if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
1573 strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
1578 ret = ocfs2_xattr_block_remove(inode, blk_bh);
1585 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1586 OCFS2_INODE_UPDATE_CREDITS);
1587 if (IS_ERR(handle)) {
1588 ret = PTR_ERR(handle);
1592 ret = ocfs2_journal_access(handle, inode, di_bh,
1593 OCFS2_JOURNAL_ACCESS_WRITE);
1599 if (di->i_xattr_loc) {
1600 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1601 ocfs2_xattr_free_block(handle, osb, xb);
1602 di->i_xattr_loc = cpu_to_le64(0);
1605 spin_lock(&oi->ip_lock);
1606 oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1607 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1608 spin_unlock(&oi->ip_lock);
1610 ret = ocfs2_journal_dirty(handle, di_bh);
1614 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1621 static int ocfs2_xattr_has_space_inline(struct inode *inode,
1622 struct ocfs2_dinode *di)
1624 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1625 unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1628 if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1631 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1632 struct ocfs2_inline_data *idata = &di->id2.i_data;
1633 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1634 } else if (ocfs2_inode_is_fast_symlink(inode)) {
1635 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1636 le64_to_cpu(di->i_size);
1638 struct ocfs2_extent_list *el = &di->id2.i_list;
1639 free = (le16_to_cpu(el->l_count) -
1640 le16_to_cpu(el->l_next_free_rec)) *
1641 sizeof(struct ocfs2_extent_rec);
1643 if (free >= xattrsize)
1650 * ocfs2_xattr_ibody_find()
1652 * Find extended attribute in inode block and
1653 * fill search info into struct ocfs2_xattr_search.
1655 static int ocfs2_xattr_ibody_find(struct inode *inode,
1658 struct ocfs2_xattr_search *xs)
1660 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1661 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1665 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1668 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1669 down_read(&oi->ip_alloc_sem);
1670 has_space = ocfs2_xattr_has_space_inline(inode, di);
1671 up_read(&oi->ip_alloc_sem);
1676 xs->xattr_bh = xs->inode_bh;
1677 xs->end = (void *)di + inode->i_sb->s_blocksize;
1678 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1679 xs->header = (struct ocfs2_xattr_header *)
1680 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1682 xs->header = (struct ocfs2_xattr_header *)
1683 (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
1684 xs->base = (void *)xs->header;
1685 xs->here = xs->header->xh_entries;
1687 /* Find the named attribute. */
1688 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1689 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1690 if (ret && ret != -ENODATA)
1692 xs->not_found = ret;
1699 * ocfs2_xattr_ibody_set()
1701 * Set, replace or remove an extended attribute into inode block.
1704 static int ocfs2_xattr_ibody_set(struct inode *inode,
1705 struct ocfs2_xattr_info *xi,
1706 struct ocfs2_xattr_search *xs)
1708 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1709 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1712 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1715 down_write(&oi->ip_alloc_sem);
1716 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1717 if (!ocfs2_xattr_has_space_inline(inode, di)) {
1723 ret = ocfs2_xattr_set_entry(inode, xi, xs,
1724 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
1726 up_write(&oi->ip_alloc_sem);
1732 * ocfs2_xattr_block_find()
1734 * Find extended attribute in external block and
1735 * fill search info into struct ocfs2_xattr_search.
1737 static int ocfs2_xattr_block_find(struct inode *inode,
1740 struct ocfs2_xattr_search *xs)
1742 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1743 struct buffer_head *blk_bh = NULL;
1744 struct ocfs2_xattr_block *xb;
1747 if (!di->i_xattr_loc)
1750 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
1751 le64_to_cpu(di->i_xattr_loc),
1752 &blk_bh, OCFS2_BH_CACHED, inode);
1757 /*Verify the signature of xattr block*/
1758 if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
1759 strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
1764 xs->xattr_bh = blk_bh;
1765 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1767 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1768 xs->header = &xb->xb_attrs.xb_header;
1769 xs->base = (void *)xs->header;
1770 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
1771 xs->here = xs->header->xh_entries;
1773 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1775 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
1779 if (ret && ret != -ENODATA) {
1780 xs->xattr_bh = NULL;
1783 xs->not_found = ret;
1792 * When all the xattrs are deleted from index btree, the ocfs2_xattr_tree
1793 * will be erased and ocfs2_xattr_block will have its ocfs2_xattr_header
1796 static int ocfs2_restore_xattr_block(struct inode *inode,
1797 struct ocfs2_xattr_search *xs)
1801 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1802 struct ocfs2_xattr_block *xb =
1803 (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1804 struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
1805 u16 xb_flags = le16_to_cpu(xb->xb_flags);
1807 BUG_ON(!(xb_flags & OCFS2_XATTR_INDEXED) ||
1808 le16_to_cpu(el->l_next_free_rec) != 0);
1810 handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1811 if (IS_ERR(handle)) {
1812 ret = PTR_ERR(handle);
1817 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1818 OCFS2_JOURNAL_ACCESS_WRITE);
1824 memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
1825 offsetof(struct ocfs2_xattr_block, xb_attrs));
1827 xb->xb_flags = cpu_to_le16(xb_flags & ~OCFS2_XATTR_INDEXED);
1829 ocfs2_journal_dirty(handle, xs->xattr_bh);
1832 ocfs2_commit_trans(osb, handle);
1838 * ocfs2_xattr_block_set()
1840 * Set, replace or remove an extended attribute into external block.
1843 static int ocfs2_xattr_block_set(struct inode *inode,
1844 struct ocfs2_xattr_info *xi,
1845 struct ocfs2_xattr_search *xs)
1847 struct buffer_head *new_bh = NULL;
1848 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1849 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1850 struct ocfs2_alloc_context *meta_ac = NULL;
1851 handle_t *handle = NULL;
1852 struct ocfs2_xattr_block *xblk = NULL;
1853 u16 suballoc_bit_start;
1858 if (!xs->xattr_bh) {
1860 * Alloc one external block for extended attribute
1863 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
1868 handle = ocfs2_start_trans(osb,
1869 OCFS2_XATTR_BLOCK_CREATE_CREDITS);
1870 if (IS_ERR(handle)) {
1871 ret = PTR_ERR(handle);
1875 ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
1876 OCFS2_JOURNAL_ACCESS_CREATE);
1882 ret = ocfs2_claim_metadata(osb, handle, meta_ac, 1,
1883 &suballoc_bit_start, &num_got,
1890 new_bh = sb_getblk(inode->i_sb, first_blkno);
1891 ocfs2_set_new_buffer_uptodate(inode, new_bh);
1893 ret = ocfs2_journal_access(handle, inode, new_bh,
1894 OCFS2_JOURNAL_ACCESS_CREATE);
1900 /* Initialize ocfs2_xattr_block */
1901 xs->xattr_bh = new_bh;
1902 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
1903 memset(xblk, 0, inode->i_sb->s_blocksize);
1904 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
1905 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
1906 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1907 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
1908 xblk->xb_blkno = cpu_to_le64(first_blkno);
1910 xs->header = &xblk->xb_attrs.xb_header;
1911 xs->base = (void *)xs->header;
1912 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
1913 xs->here = xs->header->xh_entries;
1916 ret = ocfs2_journal_dirty(handle, new_bh);
1921 di->i_xattr_loc = cpu_to_le64(first_blkno);
1922 ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1926 ocfs2_commit_trans(osb, handle);
1929 ocfs2_free_alloc_context(meta_ac);
1933 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1935 if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
1936 /* Set extended attribute into external block */
1937 ret = ocfs2_xattr_set_entry(inode, xi, xs, OCFS2_HAS_XATTR_FL);
1938 if (!ret || ret != -ENOSPC)
1941 ret = ocfs2_xattr_create_index_block(inode, xs);
1946 ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs);
1947 if (!ret && xblk->xb_attrs.xb_root.xt_list.l_next_free_rec == 0)
1948 ret = ocfs2_restore_xattr_block(inode, xs);
1958 * Set, replace or remove an extended attribute for this inode.
1959 * value is NULL to remove an existing extended attribute, else either
1960 * create or replace an extended attribute.
1962 int ocfs2_xattr_set(struct inode *inode,
1969 struct buffer_head *di_bh = NULL;
1970 struct ocfs2_dinode *di;
1972 u16 i, blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
1974 struct ocfs2_xattr_info xi = {
1975 .name_index = name_index,
1978 .value_len = value_len,
1981 struct ocfs2_xattr_search xis = {
1982 .not_found = -ENODATA,
1985 struct ocfs2_xattr_search xbs = {
1986 .not_found = -ENODATA,
1989 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1992 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1997 xis.inode_bh = xbs.inode_bh = di_bh;
1998 di = (struct ocfs2_dinode *)di_bh->b_data;
2000 down_write(&OCFS2_I(inode)->ip_xattr_sem);
2002 * Scan inode and external block to find the same name
2003 * extended attribute and collect search infomation.
2005 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2008 if (xis.not_found) {
2009 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2014 if (xis.not_found && xbs.not_found) {
2016 if (flags & XATTR_REPLACE)
2023 if (flags & XATTR_CREATE)
2028 /* Remove existing extended attribute */
2030 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2031 else if (!xbs.not_found)
2032 ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2034 /* We always try to set extended attribute into inode first*/
2035 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2036 if (!ret && !xbs.not_found) {
2038 * If succeed and that extended attribute existing in
2039 * external block, then we will remove it.
2043 ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2044 } else if (ret == -ENOSPC) {
2045 if (di->i_xattr_loc && !xbs.xattr_bh) {
2046 ret = ocfs2_xattr_block_find(inode, name_index,
2052 * If no space in inode, we will set extended attribute
2053 * into external block.
2055 ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2058 if (!xis.not_found) {
2060 * If succeed and that extended attribute
2061 * existing in inode, we will remove it.
2065 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2070 up_write(&OCFS2_I(inode)->ip_xattr_sem);
2071 ocfs2_inode_unlock(inode, 1);
2073 brelse(xbs.xattr_bh);
2074 for (i = 0; i < blk_per_bucket; i++)
2075 brelse(xbs.bucket.bhs[i]);
2080 static inline u32 ocfs2_xattr_hash_by_name(struct inode *inode,
2082 const char *suffix_name)
2084 struct xattr_handler *handler = ocfs2_xattr_handler(name_index);
2085 char *prefix = handler->prefix;
2086 int prefix_len = strlen(handler->prefix);
2088 return ocfs2_xattr_name_hash(inode, prefix, prefix_len,
2089 (char *)suffix_name, strlen(suffix_name));
2093 * Find the xattr extent rec which may contains name_hash.
2094 * e_cpos will be the first name hash of the xattr rec.
2095 * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2097 static int ocfs2_xattr_get_rec(struct inode *inode,
2102 struct ocfs2_extent_list *el)
2105 struct buffer_head *eb_bh = NULL;
2106 struct ocfs2_extent_block *eb;
2107 struct ocfs2_extent_rec *rec = NULL;
2110 if (el->l_tree_depth) {
2111 ret = ocfs2_find_leaf(inode, el, name_hash, &eb_bh);
2117 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2120 if (el->l_tree_depth) {
2121 ocfs2_error(inode->i_sb,
2122 "Inode %lu has non zero tree depth in "
2123 "xattr tree block %llu\n", inode->i_ino,
2124 (unsigned long long)eb_bh->b_blocknr);
2130 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2131 rec = &el->l_recs[i];
2133 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2134 e_blkno = le64_to_cpu(rec->e_blkno);
2140 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2141 "record (%u, %u, 0) in xattr", inode->i_ino,
2142 le32_to_cpu(rec->e_cpos),
2143 ocfs2_rec_clusters(el, rec));
2148 *p_blkno = le64_to_cpu(rec->e_blkno);
2149 *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2151 *e_cpos = le32_to_cpu(rec->e_cpos);
2157 typedef int (xattr_bucket_func)(struct inode *inode,
2158 struct ocfs2_xattr_bucket *bucket,
2161 static int ocfs2_find_xe_in_bucket(struct inode *inode,
2162 struct buffer_head *header_bh,
2169 int i, ret = 0, cmp = 1, block_off, new_offset;
2170 struct ocfs2_xattr_header *xh =
2171 (struct ocfs2_xattr_header *)header_bh->b_data;
2172 size_t name_len = strlen(name);
2173 struct ocfs2_xattr_entry *xe = NULL;
2174 struct buffer_head *name_bh = NULL;
2178 * We don't use binary search in the bucket because there
2179 * may be multiple entries with the same name hash.
2181 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2182 xe = &xh->xh_entries[i];
2184 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2186 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2189 cmp = name_index - ocfs2_xattr_get_type(xe);
2191 cmp = name_len - xe->xe_name_len;
2195 ret = ocfs2_xattr_bucket_get_name_value(inode,
2205 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
2206 header_bh->b_blocknr + block_off,
2207 &name_bh, OCFS2_BH_CACHED, inode);
2212 xe_name = name_bh->b_data + new_offset;
2214 cmp = memcmp(name, xe_name, name_len);
2230 * Find the specified xattr entry in a series of buckets.
2231 * This series start from p_blkno and last for num_clusters.
2232 * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2233 * the num of the valid buckets.
2235 * Return the buffer_head this xattr should reside in. And if the xattr's
2236 * hash is in the gap of 2 buckets, return the lower bucket.
2238 static int ocfs2_xattr_bucket_find(struct inode *inode,
2245 struct ocfs2_xattr_search *xs)
2248 struct buffer_head *bh = NULL;
2249 struct buffer_head *lower_bh = NULL;
2250 struct ocfs2_xattr_header *xh = NULL;
2251 struct ocfs2_xattr_entry *xe = NULL;
2253 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2254 int low_bucket = 0, bucket, high_bucket;
2258 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno,
2259 &bh, OCFS2_BH_CACHED, inode);
2265 xh = (struct ocfs2_xattr_header *)bh->b_data;
2266 high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
2268 while (low_bucket <= high_bucket) {
2271 bucket = (low_bucket + high_bucket) / 2;
2273 blkno = p_blkno + bucket * blk_per_bucket;
2275 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), blkno,
2276 &bh, OCFS2_BH_CACHED, inode);
2282 xh = (struct ocfs2_xattr_header *)bh->b_data;
2283 xe = &xh->xh_entries[0];
2284 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
2285 high_bucket = bucket - 1;
2290 * Check whether the hash of the last entry in our
2291 * bucket is larger than the search one.
2293 xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
2294 last_hash = le32_to_cpu(xe->xe_name_hash);
2296 /* record lower_bh which may be the insert place. */
2301 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
2302 low_bucket = bucket + 1;
2306 /* the searched xattr should reside in this bucket if exists. */
2307 ret = ocfs2_find_xe_in_bucket(inode, lower_bh,
2308 name_index, name, name_hash,
2318 * Record the bucket we have found.
2319 * When the xattr's hash value is in the gap of 2 buckets, we will
2320 * always set it to the previous bucket.
2324 * We can't find any bucket whose first name_hash is less
2325 * than the find name_hash.
2327 BUG_ON(bh->b_blocknr != p_blkno);
2331 xs->bucket.bhs[0] = lower_bh;
2332 xs->bucket.xh = (struct ocfs2_xattr_header *)
2333 xs->bucket.bhs[0]->b_data;
2336 xs->header = xs->bucket.xh;
2337 xs->base = xs->bucket.bhs[0]->b_data;
2338 xs->end = xs->base + inode->i_sb->s_blocksize;
2342 * If we have found the xattr enty, read all the blocks in
2345 ret = ocfs2_read_blocks(OCFS2_SB(inode->i_sb),
2346 xs->bucket.bhs[0]->b_blocknr + 1,
2347 blk_per_bucket - 1, &xs->bucket.bhs[1],
2348 OCFS2_BH_CACHED, inode);
2354 xs->here = &xs->header->xh_entries[index];
2355 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
2356 (unsigned long long)xs->bucket.bhs[0]->b_blocknr, index);
2366 static int ocfs2_xattr_index_block_find(struct inode *inode,
2367 struct buffer_head *root_bh,
2370 struct ocfs2_xattr_search *xs)
2373 struct ocfs2_xattr_block *xb =
2374 (struct ocfs2_xattr_block *)root_bh->b_data;
2375 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
2376 struct ocfs2_extent_list *el = &xb_root->xt_list;
2378 u32 first_hash, num_clusters = 0;
2379 u32 name_hash = ocfs2_xattr_hash_by_name(inode, name_index, name);
2381 if (le16_to_cpu(el->l_next_free_rec) == 0)
2384 mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
2385 name, name_hash, name_index);
2387 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
2394 BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
2396 mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
2397 "in the rec is %u\n", num_clusters, p_blkno, first_hash);
2399 ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
2400 p_blkno, first_hash, num_clusters, xs);
2406 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
2409 xattr_bucket_func *func,
2413 int blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2414 u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
2415 u32 num_buckets = clusters * bpc;
2416 struct ocfs2_xattr_bucket bucket;
2418 memset(&bucket, 0, sizeof(bucket));
2420 mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
2423 for (i = 0; i < num_buckets; i++, blkno += blk_per_bucket) {
2424 ret = ocfs2_read_blocks(OCFS2_SB(inode->i_sb),
2425 blkno, blk_per_bucket,
2426 bucket.bhs, OCFS2_BH_CACHED, inode);
2432 bucket.xh = (struct ocfs2_xattr_header *)bucket.bhs[0]->b_data;
2434 * The real bucket num in this series of blocks is stored
2435 * in the 1st bucket.
2438 num_buckets = le16_to_cpu(bucket.xh->xh_num_buckets);
2440 mlog(0, "iterating xattr bucket %llu\n", blkno);
2442 ret = func(inode, &bucket, para);
2449 for (j = 0; j < blk_per_bucket; j++)
2450 brelse(bucket.bhs[j]);
2451 memset(&bucket, 0, sizeof(bucket));
2455 for (j = 0; j < blk_per_bucket; j++)
2456 brelse(bucket.bhs[j]);
2461 struct ocfs2_xattr_tree_list {
2466 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
2467 struct ocfs2_xattr_header *xh,
2474 if (index < 0 || index >= le16_to_cpu(xh->xh_count))
2477 name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
2479 *block_off = name_offset >> inode->i_sb->s_blocksize_bits;
2480 *new_offset = name_offset % inode->i_sb->s_blocksize;
2485 static int ocfs2_list_xattr_bucket(struct inode *inode,
2486 struct ocfs2_xattr_bucket *bucket,
2490 struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
2492 int i, block_off, new_offset;
2494 for (i = 0 ; i < le16_to_cpu(bucket->xh->xh_count); i++) {
2495 struct ocfs2_xattr_entry *entry = &bucket->xh->xh_entries[i];
2496 struct xattr_handler *handler =
2497 ocfs2_xattr_handler(ocfs2_xattr_get_type(entry));
2500 ret = ocfs2_xattr_bucket_get_name_value(inode,
2507 size = handler->list(inode, xl->buffer, xl->buffer_size,
2508 bucket->bhs[block_off]->b_data +
2510 entry->xe_name_len);
2512 if (size > xl->buffer_size)
2516 xl->buffer_size -= size;
2523 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
2524 struct ocfs2_xattr_tree_root *xt,
2528 struct ocfs2_extent_list *el = &xt->xt_list;
2530 u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
2532 struct ocfs2_xattr_tree_list xl = {
2534 .buffer_size = buffer_size,
2537 if (le16_to_cpu(el->l_next_free_rec) == 0)
2540 while (name_hash > 0) {
2541 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
2542 &e_cpos, &num_clusters, el);
2548 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
2549 ocfs2_list_xattr_bucket,
2559 name_hash = e_cpos - 1;
2562 ret = buffer_size - xl.buffer_size;
2567 static int cmp_xe(const void *a, const void *b)
2569 const struct ocfs2_xattr_entry *l = a, *r = b;
2570 u32 l_hash = le32_to_cpu(l->xe_name_hash);
2571 u32 r_hash = le32_to_cpu(r->xe_name_hash);
2573 if (l_hash > r_hash)
2575 if (l_hash < r_hash)
2580 static void swap_xe(void *a, void *b, int size)
2582 struct ocfs2_xattr_entry *l = a, *r = b, tmp;
2585 memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
2586 memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
2590 * When the ocfs2_xattr_block is filled up, new bucket will be created
2591 * and all the xattr entries will be moved to the new bucket.
2592 * Note: we need to sort the entries since they are not saved in order
2593 * in the ocfs2_xattr_block.
2595 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
2596 struct buffer_head *xb_bh,
2597 struct buffer_head *xh_bh,
2598 struct buffer_head *data_bh)
2600 int i, blocksize = inode->i_sb->s_blocksize;
2601 u16 offset, size, off_change;
2602 struct ocfs2_xattr_entry *xe;
2603 struct ocfs2_xattr_block *xb =
2604 (struct ocfs2_xattr_block *)xb_bh->b_data;
2605 struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
2606 struct ocfs2_xattr_header *xh =
2607 (struct ocfs2_xattr_header *)xh_bh->b_data;
2608 u16 count = le16_to_cpu(xb_xh->xh_count);
2609 char *target = xh_bh->b_data, *src = xb_bh->b_data;
2611 mlog(0, "cp xattr from block %llu to bucket %llu\n",
2612 (unsigned long long)xb_bh->b_blocknr,
2613 (unsigned long long)xh_bh->b_blocknr);
2615 memset(xh_bh->b_data, 0, blocksize);
2617 memset(data_bh->b_data, 0, blocksize);
2619 * Since the xe_name_offset is based on ocfs2_xattr_header,
2620 * there is a offset change corresponding to the change of
2621 * ocfs2_xattr_header's position.
2623 off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
2624 xe = &xb_xh->xh_entries[count - 1];
2625 offset = le16_to_cpu(xe->xe_name_offset) + off_change;
2626 size = blocksize - offset;
2628 /* copy all the names and values. */
2630 target = data_bh->b_data;
2631 memcpy(target + offset, src + offset, size);
2633 /* Init new header now. */
2634 xh->xh_count = xb_xh->xh_count;
2635 xh->xh_num_buckets = cpu_to_le16(1);
2636 xh->xh_name_value_len = cpu_to_le16(size);
2637 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
2639 /* copy all the entries. */
2640 target = xh_bh->b_data;
2641 offset = offsetof(struct ocfs2_xattr_header, xh_entries);
2642 size = count * sizeof(struct ocfs2_xattr_entry);
2643 memcpy(target + offset, (char *)xb_xh + offset, size);
2645 /* Change the xe offset for all the xe because of the move. */
2646 off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
2647 offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
2648 for (i = 0; i < count; i++)
2649 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
2651 mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
2652 offset, size, off_change);
2654 sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
2659 * After we move xattr from block to index btree, we have to
2660 * update ocfs2_xattr_search to the new xe and base.
2662 * When the entry is in xattr block, xattr_bh indicates the storage place.
2663 * While if the entry is in index b-tree, "bucket" indicates the
2664 * real place of the xattr.
2666 static int ocfs2_xattr_update_xattr_search(struct inode *inode,
2667 struct ocfs2_xattr_search *xs,
2668 struct buffer_head *old_bh,
2669 struct buffer_head *new_bh)
2672 char *buf = old_bh->b_data;
2673 struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
2674 struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
2675 int i, blocksize = inode->i_sb->s_blocksize;
2676 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2678 xs->bucket.bhs[0] = new_bh;
2680 xs->bucket.xh = (struct ocfs2_xattr_header *)xs->bucket.bhs[0]->b_data;
2681 xs->header = xs->bucket.xh;
2683 xs->base = new_bh->b_data;
2684 xs->end = xs->base + inode->i_sb->s_blocksize;
2686 if (!xs->not_found) {
2687 if (OCFS2_XATTR_BUCKET_SIZE != blocksize) {
2688 ret = ocfs2_read_blocks(OCFS2_SB(inode->i_sb),
2689 xs->bucket.bhs[0]->b_blocknr + 1,
2690 blk_per_bucket - 1, &xs->bucket.bhs[1],
2691 OCFS2_BH_CACHED, inode);
2697 i = xs->here - old_xh->xh_entries;
2698 xs->here = &xs->header->xh_entries[i];
2705 static int ocfs2_xattr_create_index_block(struct inode *inode,
2706 struct ocfs2_xattr_search *xs)
2708 int ret, credits = OCFS2_SUBALLOC_ALLOC;
2712 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2713 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2714 struct ocfs2_alloc_context *data_ac;
2715 struct buffer_head *xh_bh = NULL, *data_bh = NULL;
2716 struct buffer_head *xb_bh = xs->xattr_bh;
2717 struct ocfs2_xattr_block *xb =
2718 (struct ocfs2_xattr_block *)xb_bh->b_data;
2719 struct ocfs2_xattr_tree_root *xr;
2720 u16 xb_flags = le16_to_cpu(xb->xb_flags);
2721 u16 bpb = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2723 mlog(0, "create xattr index block for %llu\n",
2724 (unsigned long long)xb_bh->b_blocknr);
2726 BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
2728 ret = ocfs2_reserve_clusters(osb, 1, &data_ac);
2736 * We can use this lock for now, and maybe move to a dedicated mutex
2737 * if performance becomes a problem later.
2739 down_write(&oi->ip_alloc_sem);
2742 * 3 more credits, one for xattr block update, one for the 1st block
2743 * of the new xattr bucket and one for the value/data.
2746 handle = ocfs2_start_trans(osb, credits);
2747 if (IS_ERR(handle)) {
2748 ret = PTR_ERR(handle);
2753 ret = ocfs2_journal_access(handle, inode, xb_bh,
2754 OCFS2_JOURNAL_ACCESS_WRITE);
2760 ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off, &len);
2767 * The bucket may spread in many blocks, and
2768 * we will only touch the 1st block and the last block
2769 * in the whole bucket(one for entry and one for data).
2771 blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
2773 mlog(0, "allocate 1 cluster from %llu to xattr block\n", blkno);
2775 xh_bh = sb_getblk(inode->i_sb, blkno);
2782 ocfs2_set_new_buffer_uptodate(inode, xh_bh);
2784 ret = ocfs2_journal_access(handle, inode, xh_bh,
2785 OCFS2_JOURNAL_ACCESS_CREATE);
2792 data_bh = sb_getblk(inode->i_sb, blkno + bpb - 1);
2799 ocfs2_set_new_buffer_uptodate(inode, data_bh);
2801 ret = ocfs2_journal_access(handle, inode, data_bh,
2802 OCFS2_JOURNAL_ACCESS_CREATE);
2809 ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xh_bh, data_bh);
2811 ocfs2_journal_dirty(handle, xh_bh);
2813 ocfs2_journal_dirty(handle, data_bh);
2815 ocfs2_xattr_update_xattr_search(inode, xs, xb_bh, xh_bh);
2817 /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
2818 memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
2819 offsetof(struct ocfs2_xattr_block, xb_attrs));
2821 xr = &xb->xb_attrs.xb_root;
2822 xr->xt_clusters = cpu_to_le32(1);
2823 xr->xt_last_eb_blk = 0;
2824 xr->xt_list.l_tree_depth = 0;
2825 xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
2826 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2828 xr->xt_list.l_recs[0].e_cpos = 0;
2829 xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
2830 xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
2832 xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
2834 ret = ocfs2_journal_dirty(handle, xb_bh);
2841 ocfs2_commit_trans(osb, handle);
2844 up_write(&oi->ip_alloc_sem);
2848 ocfs2_free_alloc_context(data_ac);
2856 static int cmp_xe_offset(const void *a, const void *b)
2858 const struct ocfs2_xattr_entry *l = a, *r = b;
2859 u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
2860 u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
2862 if (l_name_offset < r_name_offset)
2864 if (l_name_offset > r_name_offset)
2870 * defrag a xattr bucket if we find that the bucket has some
2871 * holes beteen name/value pairs.
2872 * We will move all the name/value pairs to the end of the bucket
2873 * so that we can spare some space for insertion.
2875 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
2876 struct ocfs2_xattr_bucket *bucket)
2879 size_t end, offset, len, value_len;
2880 struct ocfs2_xattr_header *xh;
2881 char *entries, *buf, *bucket_buf = NULL;
2882 u64 blkno = bucket->bhs[0]->b_blocknr;
2883 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2885 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2886 size_t blocksize = inode->i_sb->s_blocksize;
2888 struct buffer_head **bhs;
2889 struct ocfs2_xattr_entry *xe;
2891 bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
2896 ret = ocfs2_read_blocks(osb, blkno, blk_per_bucket, bhs,
2897 OCFS2_BH_CACHED, inode);
2902 * In order to make the operation more efficient and generic,
2903 * we copy all the blocks into a contiguous memory and do the
2904 * defragment there, so if anything is error, we will not touch
2907 bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
2914 for (i = 0; i < blk_per_bucket; i++, buf += blocksize)
2915 memcpy(buf, bhs[i]->b_data, blocksize);
2917 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), blk_per_bucket);
2918 if (IS_ERR(handle)) {
2919 ret = PTR_ERR(handle);
2925 for (i = 0; i < blk_per_bucket; i++) {
2926 ret = ocfs2_journal_access(handle, inode, bhs[i],
2927 OCFS2_JOURNAL_ACCESS_WRITE);
2934 xh = (struct ocfs2_xattr_header *)bucket_buf;
2935 entries = (char *)xh->xh_entries;
2936 xh_free_start = le16_to_cpu(xh->xh_free_start);
2938 mlog(0, "adjust xattr bucket in %llu, count = %u, "
2939 "xh_free_start = %u, xh_name_value_len = %u.\n",
2940 blkno, le16_to_cpu(xh->xh_count), xh_free_start,
2941 le16_to_cpu(xh->xh_name_value_len));
2944 * sort all the entries by their offset.
2945 * the largest will be the first, so that we can
2946 * move them to the end one by one.
2948 sort(entries, le16_to_cpu(xh->xh_count),
2949 sizeof(struct ocfs2_xattr_entry),
2950 cmp_xe_offset, swap_xe);
2952 /* Move all name/values to the end of the bucket. */
2953 xe = xh->xh_entries;
2954 end = OCFS2_XATTR_BUCKET_SIZE;
2955 for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
2956 offset = le16_to_cpu(xe->xe_name_offset);
2957 if (ocfs2_xattr_is_local(xe))
2958 value_len = OCFS2_XATTR_SIZE(
2959 le64_to_cpu(xe->xe_value_size));
2961 value_len = OCFS2_XATTR_ROOT_SIZE;
2962 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
2965 * We must make sure that the name/value pair
2966 * exist in the same block. So adjust end to
2967 * the previous block end if needed.
2969 if (((end - len) / blocksize !=
2970 (end - 1) / blocksize))
2971 end = end - end % blocksize;
2973 if (end > offset + len) {
2974 memmove(bucket_buf + end - len,
2975 bucket_buf + offset, len);
2976 xe->xe_name_offset = cpu_to_le16(end - len);
2979 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
2980 "bucket %llu\n", (unsigned long long)blkno);
2985 mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
2986 "bucket %llu\n", (unsigned long long)blkno);
2988 if (xh_free_start == end)
2991 memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
2992 xh->xh_free_start = cpu_to_le16(end);
2994 /* sort the entries by their name_hash. */
2995 sort(entries, le16_to_cpu(xh->xh_count),
2996 sizeof(struct ocfs2_xattr_entry),
3000 for (i = 0; i < blk_per_bucket; i++, buf += blocksize) {
3001 memcpy(bhs[i]->b_data, buf, blocksize);
3002 ocfs2_journal_dirty(handle, bhs[i]);
3006 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
3010 for (i = 0; i < blk_per_bucket; i++)
3020 * Move half nums of the xattr bucket in the previous cluster to this new
3021 * cluster. We only touch the last cluster of the previous extend record.
3023 * first_bh is the first buffer_head of a series of bucket in the same
3024 * extent rec and header_bh is the header of one bucket in this cluster.
3025 * They will be updated if we move the data header_bh contains to the new
3026 * cluster. first_hash will be set as the 1st xe's name_hash of the new cluster.
3028 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3030 struct buffer_head **first_bh,
3031 struct buffer_head **header_bh,
3037 int i, ret, credits;
3038 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3039 int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3040 int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3041 int blocksize = inode->i_sb->s_blocksize;
3042 struct buffer_head *old_bh, *new_bh, *prev_bh, *new_first_bh = NULL;
3043 struct ocfs2_xattr_header *new_xh;
3044 struct ocfs2_xattr_header *xh =
3045 (struct ocfs2_xattr_header *)((*first_bh)->b_data);
3047 BUG_ON(le16_to_cpu(xh->xh_num_buckets) < num_buckets);
3048 BUG_ON(OCFS2_XATTR_BUCKET_SIZE == osb->s_clustersize);
3050 prev_bh = *first_bh;
3052 xh = (struct ocfs2_xattr_header *)prev_bh->b_data;
3054 prev_blkno += (num_clusters - 1) * bpc + bpc / 2;
3056 mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3057 prev_blkno, new_blkno);
3060 * We need to update the 1st half of the new cluster and
3061 * 1 more for the update of the 1st bucket of the previous
3064 credits = bpc / 2 + 1;
3065 ret = ocfs2_extend_trans(handle, credits);
3071 ret = ocfs2_journal_access(handle, inode, prev_bh,
3072 OCFS2_JOURNAL_ACCESS_WRITE);
3078 for (i = 0; i < bpc / 2; i++, prev_blkno++, new_blkno++) {
3079 old_bh = new_bh = NULL;
3080 new_bh = sb_getblk(inode->i_sb, new_blkno);
3087 ocfs2_set_new_buffer_uptodate(inode, new_bh);
3089 ret = ocfs2_journal_access(handle, inode, new_bh,
3090 OCFS2_JOURNAL_ACCESS_CREATE);
3097 ret = ocfs2_read_block(osb, prev_blkno,
3098 &old_bh, OCFS2_BH_CACHED, inode);
3105 memcpy(new_bh->b_data, old_bh->b_data, blocksize);
3108 new_xh = (struct ocfs2_xattr_header *)new_bh->b_data;
3109 new_xh->xh_num_buckets = cpu_to_le16(num_buckets / 2);
3112 *first_hash = le32_to_cpu(
3113 new_xh->xh_entries[0].xe_name_hash);
3114 new_first_bh = new_bh;
3115 get_bh(new_first_bh);
3118 ocfs2_journal_dirty(handle, new_bh);
3120 if (*header_bh == old_bh) {
3122 *header_bh = new_bh;
3126 *first_bh = new_first_bh;
3133 le16_add_cpu(&xh->xh_num_buckets, -(num_buckets / 2));
3135 ocfs2_journal_dirty(handle, prev_bh);
3138 brelse(new_first_bh);
3142 static int ocfs2_read_xattr_bucket(struct inode *inode,
3144 struct buffer_head **bhs,
3148 u16 i, blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3151 return ocfs2_read_blocks(OCFS2_SB(inode->i_sb), blkno,
3152 blk_per_bucket, bhs,
3153 OCFS2_BH_CACHED, inode);
3155 for (i = 0; i < blk_per_bucket; i++) {
3156 bhs[i] = sb_getblk(inode->i_sb, blkno + i);
3157 if (bhs[i] == NULL) {
3162 ocfs2_set_new_buffer_uptodate(inode, bhs[i]);
3169 * Move half num of the xattrs in old bucket(blk) to new bucket(new_blk).
3170 * first_hash will record the 1st hash of the new bucket.
3172 static int ocfs2_half_xattr_bucket(struct inode *inode,
3177 int new_bucket_head)
3180 u16 count, start, len, name_value_len, xe_len, name_offset;
3181 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3182 struct buffer_head **s_bhs, **t_bhs = NULL;
3183 struct ocfs2_xattr_header *xh;
3184 struct ocfs2_xattr_entry *xe;
3185 int blocksize = inode->i_sb->s_blocksize;
3187 mlog(0, "move half of xattrs from bucket %llu to %llu\n",
3190 s_bhs = kcalloc(blk_per_bucket, sizeof(struct buffer_head *), GFP_NOFS);
3194 ret = ocfs2_read_xattr_bucket(inode, blk, s_bhs, 0);
3200 ret = ocfs2_journal_access(handle, inode, s_bhs[0],
3201 OCFS2_JOURNAL_ACCESS_WRITE);
3207 t_bhs = kcalloc(blk_per_bucket, sizeof(struct buffer_head *), GFP_NOFS);
3213 ret = ocfs2_read_xattr_bucket(inode, new_blk, t_bhs, new_bucket_head);
3219 for (i = 0; i < blk_per_bucket; i++) {
3220 ret = ocfs2_journal_access(handle, inode, t_bhs[i],
3221 OCFS2_JOURNAL_ACCESS_CREATE);
3228 /* copy the whole bucket to the new first. */
3229 for (i = 0; i < blk_per_bucket; i++)
3230 memcpy(t_bhs[i]->b_data, s_bhs[i]->b_data, blocksize);
3232 /* update the new bucket. */
3233 xh = (struct ocfs2_xattr_header *)t_bhs[0]->b_data;
3234 count = le16_to_cpu(xh->xh_count);
3238 * Calculate the total name/value len and xh_free_start for
3239 * the old bucket first.
3241 name_offset = OCFS2_XATTR_BUCKET_SIZE;
3243 for (i = 0; i < start; i++) {
3244 xe = &xh->xh_entries[i];
3245 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3246 if (ocfs2_xattr_is_local(xe))
3248 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3250 xe_len += OCFS2_XATTR_ROOT_SIZE;
3251 name_value_len += xe_len;
3252 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3253 name_offset = le16_to_cpu(xe->xe_name_offset);
3257 * Now begin the modification to the new bucket.
3259 * In the new bucket, We just move the xattr entry to the beginning
3260 * and don't touch the name/value. So there will be some holes in the
3261 * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3264 xe = &xh->xh_entries[start];
3265 len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3266 mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3267 (int)((char *)xe - (char *)xh),
3268 (int)((char *)xh->xh_entries - (char *)xh));
3269 memmove((char *)xh->xh_entries, (char *)xe, len);
3270 xe = &xh->xh_entries[count - start];
3271 len = sizeof(struct ocfs2_xattr_entry) * start;
3272 memset((char *)xe, 0, len);
3274 le16_add_cpu(&xh->xh_count, -start);
3275 le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3277 /* Calculate xh_free_start for the new bucket. */
3278 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3279 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3280 xe = &xh->xh_entries[i];
3281 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3282 if (ocfs2_xattr_is_local(xe))
3284 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3286 xe_len += OCFS2_XATTR_ROOT_SIZE;
3287 if (le16_to_cpu(xe->xe_name_offset) <
3288 le16_to_cpu(xh->xh_free_start))
3289 xh->xh_free_start = xe->xe_name_offset;
3292 /* set xh->xh_num_buckets for the new xh. */
3293 if (new_bucket_head)
3294 xh->xh_num_buckets = cpu_to_le16(1);
3296 xh->xh_num_buckets = 0;
3298 for (i = 0; i < blk_per_bucket; i++) {
3299 ocfs2_journal_dirty(handle, t_bhs[i]);
3304 /* store the first_hash of the new bucket. */
3306 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3309 * Now only update the 1st block of the old bucket.
3310 * Please note that the entry has been sorted already above.
3312 xh = (struct ocfs2_xattr_header *)s_bhs[0]->b_data;
3313 memset(&xh->xh_entries[start], 0,
3314 sizeof(struct ocfs2_xattr_entry) * (count - start));
3315 xh->xh_count = cpu_to_le16(start);
3316 xh->xh_free_start = cpu_to_le16(name_offset);
3317 xh->xh_name_value_len = cpu_to_le16(name_value_len);
3319 ocfs2_journal_dirty(handle, s_bhs[0]);
3325 for (i = 0; i < blk_per_bucket; i++)
3331 for (i = 0; i < blk_per_bucket; i++)
3340 * Copy xattr from one bucket to another bucket.
3342 * The caller must make sure that the journal transaction
3343 * has enough space for journaling.
3345 static int ocfs2_cp_xattr_bucket(struct inode *inode,
3352 int blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3353 int blocksize = inode->i_sb->s_blocksize;
3354 struct buffer_head **s_bhs, **t_bhs = NULL;
3356 BUG_ON(s_blkno == t_blkno);
3358 mlog(0, "cp bucket %llu to %llu, target is %d\n",
3359 s_blkno, t_blkno, t_is_new);
3361 s_bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
3366 ret = ocfs2_read_xattr_bucket(inode, s_blkno, s_bhs, 0);
3370 t_bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
3377 ret = ocfs2_read_xattr_bucket(inode, t_blkno, t_bhs, t_is_new);
3381 for (i = 0; i < blk_per_bucket; i++) {
3382 ret = ocfs2_journal_access(handle, inode, t_bhs[i],
3383 OCFS2_JOURNAL_ACCESS_WRITE);
3388 for (i = 0; i < blk_per_bucket; i++) {
3389 memcpy(t_bhs[i]->b_data, s_bhs[i]->b_data, blocksize);
3390 ocfs2_journal_dirty(handle, t_bhs[i]);
3395 for (i = 0; i < blk_per_bucket; i++)
3401 for (i = 0; i < blk_per_bucket; i++)
3410 * Copy one xattr cluster from src_blk to to_blk.
3411 * The to_blk will become the first bucket header of the cluster, so its
3412 * xh_num_buckets will be initialized as the bucket num in the cluster.
3414 static int ocfs2_cp_xattr_cluster(struct inode *inode,
3416 struct buffer_head *first_bh,
3421 int i, ret, credits;
3422 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3423 int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3424 int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3425 struct buffer_head *bh = NULL;
3426 struct ocfs2_xattr_header *xh;
3427 u64 to_blk_start = to_blk;
3429 mlog(0, "cp xattrs from cluster %llu to %llu\n", src_blk, to_blk);
3432 * We need to update the new cluster and 1 more for the update of
3433 * the 1st bucket of the previous extent rec.
3436 ret = ocfs2_extend_trans(handle, credits);
3442 ret = ocfs2_journal_access(handle, inode, first_bh,
3443 OCFS2_JOURNAL_ACCESS_WRITE);
3449 for (i = 0; i < num_buckets; i++) {
3450 ret = ocfs2_cp_xattr_bucket(inode, handle,
3451 src_blk, to_blk, 1);
3457 src_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3458 to_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3461 /* update the old bucket header. */
3462 xh = (struct ocfs2_xattr_header *)first_bh->b_data;
3463 le16_add_cpu(&xh->xh_num_buckets, -num_buckets);
3465 ocfs2_journal_dirty(handle, first_bh);
3467 /* update the new bucket header. */
3468 ret = ocfs2_read_block(osb, to_blk_start, &bh, OCFS2_BH_CACHED, inode);
3474 ret = ocfs2_journal_access(handle, inode, bh,
3475 OCFS2_JOURNAL_ACCESS_WRITE);
3481 xh = (struct ocfs2_xattr_header *)bh->b_data;
3482 xh->xh_num_buckets = cpu_to_le16(num_buckets);
3484 ocfs2_journal_dirty(handle, bh);
3487 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3494 * Move half of the xattrs in this cluster to the new cluster.
3495 * This function should only be called when bucket size == cluster size.
3496 * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
3498 static int ocfs2_half_xattr_cluster(struct inode *inode,
3504 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3505 int ret, credits = 2 * blk_per_bucket;
3507 BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
3509 ret = ocfs2_extend_trans(handle, credits);
3515 /* Move half of the xattr in start_blk to the next bucket. */
3516 return ocfs2_half_xattr_bucket(inode, handle, prev_blk,
3517 new_blk, first_hash, 1);
3521 * Move some xattrs from the old cluster to the new one since they are not
3522 * contiguous in ocfs2 xattr tree.
3524 * new_blk starts a new separate cluster, and we will move some xattrs from
3525 * prev_blk to it. v_start will be set as the first name hash value in this
3526 * new cluster so that it can be used as e_cpos during tree insertion and
3527 * don't collide with our original b-tree operations. first_bh and header_bh
3528 * will also be updated since they will be used in ocfs2_extend_xattr_bucket
3529 * to extend the insert bucket.
3531 * The problem is how much xattr should we move to the new one and when should
3532 * we update first_bh and header_bh?
3533 * 1. If cluster size > bucket size, that means the previous cluster has more
3534 * than 1 bucket, so just move half nums of bucket into the new cluster and
3535 * update the first_bh and header_bh if the insert bucket has been moved
3536 * to the new cluster.
3537 * 2. If cluster_size == bucket_size:
3538 * a) If the previous extent rec has more than one cluster and the insert
3539 * place isn't in the last cluster, copy the entire last cluster to the
3540 * new one. This time, we don't need to upate the first_bh and header_bh
3541 * since they will not be moved into the new cluster.
3542 * b) Otherwise, move the bottom half of the xattrs in the last cluster into
3543 * the new one. And we set the extend flag to zero if the insert place is
3544 * moved into the new allocated cluster since no extend is needed.
3546 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
3548 struct buffer_head **first_bh,
3549 struct buffer_head **header_bh,
3557 int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3559 mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
3560 prev_blk, prev_clusters, new_blk);
3562 if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1)
3563 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
3572 u64 last_blk = prev_blk + bpc * (prev_clusters - 1);
3574 if (prev_clusters > 1 && (*header_bh)->b_blocknr != last_blk)
3575 ret = ocfs2_cp_xattr_cluster(inode, handle, *first_bh,
3579 ret = ocfs2_half_xattr_cluster(inode, handle,
3583 if ((*header_bh)->b_blocknr == last_blk && extend)
3592 * Add a new cluster for xattr storage.
3594 * If the new cluster is contiguous with the previous one, it will be
3595 * appended to the same extent record, and num_clusters will be updated.
3596 * If not, we will insert a new extent for it and move some xattrs in
3597 * the last cluster into the new allocated one.
3598 * We also need to limit the maximum size of a btree leaf, otherwise we'll
3599 * lose the benefits of hashing because we'll have to search large leaves.
3600 * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
3603 * first_bh is the first block of the previous extent rec and header_bh
3604 * indicates the bucket we will insert the new xattrs. They will be updated
3605 * when the header_bh is moved into the new cluster.
3607 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
3608 struct buffer_head *root_bh,
3609 struct buffer_head **first_bh,
3610 struct buffer_head **header_bh,
3617 u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3618 u32 prev_clusters = *num_clusters;
3619 u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
3621 handle_t *handle = NULL;
3622 struct ocfs2_alloc_context *data_ac = NULL;
3623 struct ocfs2_alloc_context *meta_ac = NULL;
3624 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3625 struct ocfs2_xattr_block *xb =
3626 (struct ocfs2_xattr_block *)root_bh->b_data;
3627 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3628 struct ocfs2_extent_list *root_el = &xb_root->xt_list;
3629 enum ocfs2_extent_tree_type type = OCFS2_XATTR_TREE_EXTENT;
3631 mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
3632 "previous xattr blkno = %llu\n",
3633 (unsigned long long)OCFS2_I(inode)->ip_blkno,
3634 prev_cpos, prev_blkno);
3636 ret = ocfs2_lock_allocators(inode, root_bh, root_el,
3637 clusters_to_add, 0, &data_ac,
3638 &meta_ac, type, NULL);
3644 credits = ocfs2_calc_extend_credits(osb->sb, root_el, clusters_to_add);
3645 handle = ocfs2_start_trans(osb, credits);
3646 if (IS_ERR(handle)) {
3647 ret = PTR_ERR(handle);
3653 ret = ocfs2_journal_access(handle, inode, root_bh,
3654 OCFS2_JOURNAL_ACCESS_WRITE);
3660 ret = __ocfs2_claim_clusters(osb, handle, data_ac, 1,
3661 clusters_to_add, &bit_off, &num_bits);
3668 BUG_ON(num_bits > clusters_to_add);
3670 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
3671 mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
3672 num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
3674 if (prev_blkno + prev_clusters * bpc == block &&
3675 (prev_clusters + num_bits) << osb->s_clustersize_bits <=
3676 OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
3678 * If this cluster is contiguous with the old one and
3679 * adding this new cluster, we don't surpass the limit of
3680 * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
3681 * initialized and used like other buckets in the previous
3683 * So add it as a contiguous one. The caller will handle
3686 v_start = prev_cpos + prev_clusters;
3687 *num_clusters = prev_clusters + num_bits;
3688 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
3691 ret = ocfs2_adjust_xattr_cross_cluster(inode,
3706 mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
3707 num_bits, block, v_start);
3708 ret = ocfs2_xattr_tree_insert_extent(osb, handle, inode, root_bh,
3709 v_start, block, num_bits,
3716 ret = ocfs2_journal_dirty(handle, root_bh);
3724 ocfs2_commit_trans(osb, handle);
3726 ocfs2_free_alloc_context(data_ac);
3728 ocfs2_free_alloc_context(meta_ac);
3734 * Extend a new xattr bucket and move xattrs to the end one by one until
3735 * We meet with start_bh. Only move half of the xattrs to the bucket after it.
3737 static int ocfs2_extend_xattr_bucket(struct inode *inode,
3738 struct buffer_head *first_bh,
3739 struct buffer_head *start_bh,
3743 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3744 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3745 u64 start_blk = start_bh->b_blocknr, end_blk;
3746 u32 num_buckets = num_clusters * ocfs2_xattr_buckets_per_cluster(osb);
3748 struct ocfs2_xattr_header *first_xh =
3749 (struct ocfs2_xattr_header *)first_bh->b_data;
3750 u16 bucket = le16_to_cpu(first_xh->xh_num_buckets);
3752 mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
3753 "from %llu, len = %u\n", start_blk,
3754 (unsigned long long)first_bh->b_blocknr, num_clusters);
3756 BUG_ON(bucket >= num_buckets);
3758 end_blk = first_bh->b_blocknr + (bucket - 1) * blk_per_bucket;
3761 * We will touch all the buckets after the start_bh(include it).
3762 * Add one more bucket and modify the first_bh.
3764 credits = end_blk - start_blk + 2 * blk_per_bucket + 1;
3765 handle = ocfs2_start_trans(osb, credits);
3766 if (IS_ERR(handle)) {
3767 ret = PTR_ERR(handle);
3773 ret = ocfs2_journal_access(handle, inode, first_bh,
3774 OCFS2_JOURNAL_ACCESS_WRITE);
3780 while (end_blk != start_blk) {
3781 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
3782 end_blk + blk_per_bucket, 0);
3785 end_blk -= blk_per_bucket;
3788 /* Move half of the xattr in start_blk to the next bucket. */
3789 ret = ocfs2_half_xattr_bucket(inode, handle, start_blk,
3790 start_blk + blk_per_bucket, NULL, 0);
3792 le16_add_cpu(&first_xh->xh_num_buckets, 1);
3793 ocfs2_journal_dirty(handle, first_bh);
3796 ocfs2_commit_trans(osb, handle);
3802 * Add new xattr bucket in an extent record and adjust the buckets accordingly.
3803 * xb_bh is the ocfs2_xattr_block.
3804 * We will move all the buckets starting from header_bh to the next place. As
3805 * for this one, half num of its xattrs will be moved to the next one.
3807 * We will allocate a new cluster if current cluster is full and adjust
3808 * header_bh and first_bh if the insert place is moved to the new cluster.
3810 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
3811 struct buffer_head *xb_bh,
3812 struct buffer_head *header_bh)
3814 struct ocfs2_xattr_header *first_xh = NULL;
3815 struct buffer_head *first_bh = NULL;
3816 struct ocfs2_xattr_block *xb =
3817 (struct ocfs2_xattr_block *)xb_bh->b_data;
3818 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3819 struct ocfs2_extent_list *el = &xb_root->xt_list;
3820 struct ocfs2_xattr_header *xh =
3821 (struct ocfs2_xattr_header *)header_bh->b_data;
3822 u32 name_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3823 struct super_block *sb = inode->i_sb;
3824 struct ocfs2_super *osb = OCFS2_SB(sb);
3825 int ret, num_buckets, extend = 1;
3827 u32 e_cpos, num_clusters;
3829 mlog(0, "Add new xattr bucket starting form %llu\n",
3830 (unsigned long long)header_bh->b_blocknr);
3833 * Add refrence for header_bh here because it may be
3834 * changed in ocfs2_add_new_xattr_cluster and we need
3835 * to free it in the end.
3839 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
3846 ret = ocfs2_read_block(osb, p_blkno,
3847 &first_bh, OCFS2_BH_CACHED, inode);
3853 num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
3854 first_xh = (struct ocfs2_xattr_header *)first_bh->b_data;
3856 if (num_buckets == le16_to_cpu(first_xh->xh_num_buckets)) {
3857 ret = ocfs2_add_new_xattr_cluster(inode,
3872 ret = ocfs2_extend_xattr_bucket(inode,
3884 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
3885 struct ocfs2_xattr_bucket *bucket,
3888 int block_off = offs >> inode->i_sb->s_blocksize_bits;
3890 offs = offs % inode->i_sb->s_blocksize;
3891 return bucket->bhs[block_off]->b_data + offs;
3895 * Handle the normal xattr set, including replace, delete and new.
3896 * When the bucket is empty, "is_empty" is set and the caller can
3899 * Note: "local" indicates the real data's locality. So we can't
3900 * just its bucket locality by its length.
3902 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
3903 struct ocfs2_xattr_info *xi,
3904 struct ocfs2_xattr_search *xs,
3909 struct ocfs2_xattr_entry *last, *xe;
3910 int name_len = strlen(xi->name);
3911 struct ocfs2_xattr_header *xh = xs->header;
3912 u16 count = le16_to_cpu(xh->xh_count), start;
3913 size_t blocksize = inode->i_sb->s_blocksize;
3915 size_t offs, size, new_size;
3917 last = &xh->xh_entries[count];
3918 if (!xs->not_found) {
3920 offs = le16_to_cpu(xe->xe_name_offset);
3921 if (ocfs2_xattr_is_local(xe))
3922 size = OCFS2_XATTR_SIZE(name_len) +
3923 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3925 size = OCFS2_XATTR_SIZE(name_len) +
3926 OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
3929 * If the new value will be stored outside, xi->value has been
3930 * initalized as an empty ocfs2_xattr_value_root, and the same
3931 * goes with xi->value_len, so we can set new_size safely here.
3932 * See ocfs2_xattr_set_in_bucket.
3934 new_size = OCFS2_XATTR_SIZE(name_len) +
3935 OCFS2_XATTR_SIZE(xi->value_len);
3937 le16_add_cpu(&xh->xh_name_value_len, -size);
3939 if (new_size > size)
3940 goto set_new_name_value;
3942 /* Now replace the old value with new one. */
3944 xe->xe_value_size = cpu_to_le64(xi->value_len);
3946 xe->xe_value_size = 0;
3948 val = ocfs2_xattr_bucket_get_val(inode,
3950 memset(val + OCFS2_XATTR_SIZE(name_len), 0,
3951 size - OCFS2_XATTR_SIZE(name_len));
3952 if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
3953 memcpy(val + OCFS2_XATTR_SIZE(name_len),
3954 xi->value, xi->value_len);
3956 le16_add_cpu(&xh->xh_name_value_len, new_size);
3957 ocfs2_xattr_set_local(xe, local);
3960 /* Remove the old entry. */
3963 (void *)last - (void *)xe);
3964 memset(last, 0, sizeof(struct ocfs2_xattr_entry));
3965 le16_add_cpu(&xh->xh_count, -1);
3966 if (xh->xh_count == 0 && is_empty)
3971 /* find a new entry for insert. */
3972 int low = 0, high = count - 1, tmp;
3973 struct ocfs2_xattr_entry *tmp_xe;
3975 while (low <= high) {
3976 tmp = (low + high) / 2;
3977 tmp_xe = &xh->xh_entries[tmp];
3979 if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
3981 else if (name_hash <
3982 le32_to_cpu(tmp_xe->xe_name_hash))
3988 xe = &xh->xh_entries[low];
3990 memmove(xe + 1, xe, (void *)last - (void *)xe);
3992 le16_add_cpu(&xh->xh_count, 1);
3993 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
3994 xe->xe_name_hash = cpu_to_le32(name_hash);
3995 xe->xe_name_len = name_len;
3996 ocfs2_xattr_set_type(xe, xi->name_index);
4000 /* Insert the new name+value. */
4001 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4004 * We must make sure that the name/value pair
4005 * exists in the same block.
4007 offs = le16_to_cpu(xh->xh_free_start);
4008 start = offs - size;
4010 if (start >> inode->i_sb->s_blocksize_bits !=
4011 (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4012 offs = offs - offs % blocksize;
4013 xh->xh_free_start = cpu_to_le16(offs);
4016 val = ocfs2_xattr_bucket_get_val(inode,
4017 &xs->bucket, offs - size);
4018 xe->xe_name_offset = cpu_to_le16(offs - size);
4020 memset(val, 0, size);
4021 memcpy(val, xi->name, name_len);
4022 memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4024 xe->xe_value_size = cpu_to_le64(xi->value_len);
4025 ocfs2_xattr_set_local(xe, local);
4027 le16_add_cpu(&xh->xh_free_start, -size);
4028 le16_add_cpu(&xh->xh_name_value_len, size);
4033 static int ocfs2_xattr_bucket_handle_journal(struct inode *inode,
4035 struct ocfs2_xattr_search *xs,
4036 struct buffer_head **bhs,
4039 int ret = 0, off, block_off;
4040 struct ocfs2_xattr_entry *xe = xs->here;
4043 * First calculate all the blocks we should journal_access
4044 * and journal_dirty. The first block should always be touched.
4046 ret = ocfs2_journal_dirty(handle, bhs[0]);
4050 /* calc the data. */
4051 off = le16_to_cpu(xe->xe_name_offset);
4052 block_off = off >> inode->i_sb->s_blocksize_bits;
4053 ret = ocfs2_journal_dirty(handle, bhs[block_off]);
4061 * Set the xattr entry in the specified bucket.
4062 * The bucket is indicated by xs->bucket and it should have the enough
4063 * space for the xattr insertion.
4065 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4066 struct ocfs2_xattr_info *xi,
4067 struct ocfs2_xattr_search *xs,
4073 handle_t *handle = NULL;
4074 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4075 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4077 mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4078 (unsigned long)xi->value_len, xi->name_index,
4079 (unsigned long long)xs->bucket.bhs[0]->b_blocknr);
4081 if (!xs->bucket.bhs[1]) {
4082 ret = ocfs2_read_blocks(osb,
4083 xs->bucket.bhs[0]->b_blocknr + 1,
4084 blk_per_bucket - 1, &xs->bucket.bhs[1],
4085 OCFS2_BH_CACHED, inode);
4092 handle = ocfs2_start_trans(osb, blk_per_bucket);
4093 if (IS_ERR(handle)) {
4094 ret = PTR_ERR(handle);
4100 for (i = 0; i < blk_per_bucket; i++) {
4101 ret = ocfs2_journal_access(handle, inode, xs->bucket.bhs[i],
4102 OCFS2_JOURNAL_ACCESS_WRITE);
4109 ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash,
4110 local, bucket_empty);
4112 /*Only dirty the blocks we have touched in set xattr. */
4113 ret = ocfs2_xattr_bucket_handle_journal(inode, handle, xs,
4114 xs->bucket.bhs, blk_per_bucket);
4118 ocfs2_commit_trans(osb, handle);
4123 static int ocfs2_xattr_value_update_size(struct inode *inode,
4124 struct buffer_head *xe_bh,
4125 struct ocfs2_xattr_entry *xe,
4129 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4130 handle_t *handle = NULL;
4132 handle = ocfs2_start_trans(osb, 1);
4133 if (handle == NULL) {
4139 ret = ocfs2_journal_access(handle, inode, xe_bh,
4140 OCFS2_JOURNAL_ACCESS_WRITE);
4146 xe->xe_value_size = cpu_to_le64(new_size);
4148 ret = ocfs2_journal_dirty(handle, xe_bh);
4153 ocfs2_commit_trans(osb, handle);
4159 * Truncate the specified xe_off entry in xattr bucket.
4160 * bucket is indicated by header_bh and len is the new length.
4161 * Both the ocfs2_xattr_value_root and the entry will be updated here.
4163 * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4165 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4166 struct buffer_head *header_bh,
4172 struct buffer_head *value_bh = NULL;
4173 struct ocfs2_xattr_value_root *xv;
4174 struct ocfs2_xattr_entry *xe;
4175 struct ocfs2_xattr_header *xh =
4176 (struct ocfs2_xattr_header *)header_bh->b_data;
4177 size_t blocksize = inode->i_sb->s_blocksize;
4179 xe = &xh->xh_entries[xe_off];
4181 BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4183 offset = le16_to_cpu(xe->xe_name_offset) +
4184 OCFS2_XATTR_SIZE(xe->xe_name_len);
4186 value_blk = offset / blocksize;
4188 /* We don't allow ocfs2_xattr_value to be stored in different block. */
4189 BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4190 value_blk += header_bh->b_blocknr;
4192 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), value_blk,
4193 &value_bh, OCFS2_BH_CACHED, inode);
4199 xv = (struct ocfs2_xattr_value_root *)
4200 (value_bh->b_data + offset % blocksize);
4202 mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4203 xe_off, (unsigned long long)header_bh->b_blocknr, len);
4204 ret = ocfs2_xattr_value_truncate(inode, value_bh, xv, len);
4210 ret = ocfs2_xattr_value_update_size(inode, header_bh, xe, len);
4221 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4222 struct ocfs2_xattr_search *xs,
4226 struct ocfs2_xattr_entry *xe = xs->here;
4227 struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4229 BUG_ON(!xs->bucket.bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4231 offset = xe - xh->xh_entries;
4232 ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket.bhs[0],
4240 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4241 struct ocfs2_xattr_search *xs,
4246 struct ocfs2_xattr_value_root *xv;
4247 struct ocfs2_xattr_entry *xe = xs->here;
4249 BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4251 offset = le16_to_cpu(xe->xe_name_offset) +
4252 OCFS2_XATTR_SIZE(xe->xe_name_len);
4254 xv = (struct ocfs2_xattr_value_root *)(xs->base + offset);
4256 return __ocfs2_xattr_set_value_outside(inode, xv, val, value_len);
4260 * Remove the xattr bucket pointed by bucket_bh.
4261 * All the buckets after it in the same xattr extent rec will be
4262 * move forward one by one.
4264 static int ocfs2_rm_xattr_bucket(struct inode *inode,
4265 struct buffer_head *first_bh,
4266 struct ocfs2_xattr_bucket *bucket)
4268 int ret = 0, credits;
4269 struct ocfs2_xattr_header *xh =
4270 (struct ocfs2_xattr_header *)first_bh->b_data;
4271 u16 bucket_num = le16_to_cpu(xh->xh_num_buckets);
4272 u64 end, start = bucket->bhs[0]->b_blocknr;
4273 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4275 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4277 end = first_bh->b_blocknr + (bucket_num - 1) * blk_per_bucket;
4279 mlog(0, "rm xattr bucket %llu\n", start);
4281 * We need to update the first xattr_header and all the buckets starting
4282 * from start in this xattr rec.
4284 * XXX: Should we empty the old last bucket here?
4286 credits = 1 + end - start;
4287 handle = ocfs2_start_trans(osb, credits);
4288 if (IS_ERR(handle)) {
4289 ret = PTR_ERR(handle);
4294 ret = ocfs2_journal_access(handle, inode, first_bh,
4295 OCFS2_JOURNAL_ACCESS_WRITE);
4302 while (start < end) {
4303 ret = ocfs2_cp_xattr_bucket(inode, handle,
4304 start + blk_per_bucket,
4310 start += blk_per_bucket;
4313 /* update the first_bh. */
4314 xh->xh_num_buckets = cpu_to_le16(bucket_num - 1);
4315 ocfs2_journal_dirty(handle, first_bh);
4318 ocfs2_commit_trans(osb, handle);
4322 static int ocfs2_rm_xattr_cluster(struct inode *inode,
4323 struct buffer_head *root_bh,
4329 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4330 struct inode *tl_inode = osb->osb_tl_inode;
4332 struct ocfs2_xattr_block *xb =
4333 (struct ocfs2_xattr_block *)root_bh->b_data;
4334 struct ocfs2_extent_list *root_el = &xb->xb_attrs.xb_root.xt_list;
4335 struct ocfs2_alloc_context *meta_ac = NULL;
4336 struct ocfs2_cached_dealloc_ctxt dealloc;
4338 ocfs2_init_dealloc_ctxt(&dealloc);
4340 mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4341 cpos, len, (unsigned long long)blkno);
4343 ocfs2_remove_xattr_clusters_from_cache(inode, blkno, len);
4345 ret = ocfs2_lock_allocators(inode, root_bh, root_el,
4346 0, 1, NULL, &meta_ac,
4347 OCFS2_XATTR_TREE_EXTENT, NULL);
4353 mutex_lock(&tl_inode->i_mutex);
4355 if (ocfs2_truncate_log_needs_flush(osb)) {
4356 ret = __ocfs2_flush_truncate_log(osb);
4363 handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
4364 if (handle == NULL) {
4370 ret = ocfs2_journal_access(handle, inode, root_bh,
4371 OCFS2_JOURNAL_ACCESS_WRITE);
4377 ret = ocfs2_remove_extent(inode, root_bh, cpos, len, handle, meta_ac,
4378 &dealloc, OCFS2_XATTR_TREE_EXTENT, NULL);
4384 le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4386 ret = ocfs2_journal_dirty(handle, root_bh);
4392 ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4397 ocfs2_commit_trans(osb, handle);
4399 ocfs2_schedule_truncate_log_flush(osb, 1);
4401 mutex_unlock(&tl_inode->i_mutex);
4404 ocfs2_free_alloc_context(meta_ac);
4406 ocfs2_run_deallocs(osb, &dealloc);
4412 * Free the xattr bucket indicated by xs->bucket and if all the buckets
4413 * in the clusters is free, free the clusters also.
4415 static int ocfs2_xattr_bucket_shrink(struct inode *inode,
4416 struct ocfs2_xattr_info *xi,
4417 struct ocfs2_xattr_search *xs,
4421 u32 e_cpos, num_clusters;
4423 struct buffer_head *first_bh = NULL;
4424 struct ocfs2_xattr_header *first_xh;
4425 struct ocfs2_xattr_block *xb =
4426 (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
4428 BUG_ON(xs->header->xh_count != 0);
4430 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4431 &e_cpos, &num_clusters,
4432 &xb->xb_attrs.xb_root.xt_list);
4438 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno,
4439 &first_bh, OCFS2_BH_CACHED, inode);
4445 ret = ocfs2_rm_xattr_bucket(inode, first_bh, &xs->bucket);
4451 first_xh = (struct ocfs2_xattr_header *)first_bh->b_data;
4452 if (first_xh->xh_num_buckets == 0)
4453 ret = ocfs2_rm_xattr_cluster(inode, xs->xattr_bh,
4462 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4463 struct ocfs2_xattr_search *xs)
4465 handle_t *handle = NULL;
4466 struct ocfs2_xattr_header *xh = xs->bucket.xh;
4467 struct ocfs2_xattr_entry *last = &xh->xh_entries[
4468 le16_to_cpu(xh->xh_count) - 1];
4471 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), 1);
4472 if (IS_ERR(handle)) {
4473 ret = PTR_ERR(handle);
4478 ret = ocfs2_journal_access(handle, inode, xs->bucket.bhs[0],
4479 OCFS2_JOURNAL_ACCESS_WRITE);
4485 /* Remove the old entry. */
4486 memmove(xs->here, xs->here + 1,
4487 (void *)last - (void *)xs->here);
4488 memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4489 le16_add_cpu(&xh->xh_count, -1);
4491 ret = ocfs2_journal_dirty(handle, xs->bucket.bhs[0]);
4495 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
4499 * Set the xattr name/value in the bucket specified in xs.
4501 * As the new value in xi may be stored in the bucket or in an outside cluster,
4502 * we divide the whole process into 3 steps:
4503 * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4504 * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4505 * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4506 * 4. If the clusters for the new outside value can't be allocated, we need
4507 * to free the xattr we allocated in set.
4509 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4510 struct ocfs2_xattr_info *xi,
4511 struct ocfs2_xattr_search *xs)
4513 int ret, local = 1, bucket_empty = 0;
4515 char *val = (char *)xi->value;
4516 struct ocfs2_xattr_entry *xe = xs->here;
4517 u32 name_hash = ocfs2_xattr_hash_by_name(inode,
4518 xi->name_index, xi->name);
4520 if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
4522 * We need to truncate the xattr storage first.
4524 * If both the old and new value are stored to
4525 * outside block, we only need to truncate
4526 * the storage and then set the value outside.
4528 * If the new value should be stored within block,
4529 * we should free all the outside block first and
4530 * the modification to the xattr block will be done
4531 * by following steps.
4533 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4534 value_len = xi->value_len;
4538 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4544 goto set_value_outside;
4547 value_len = xi->value_len;
4548 /* So we have to handle the inside block change now. */
4549 if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4551 * If the new value will be stored outside of block,
4552 * initalize a new empty value root and insert it first.
4555 xi->value = &def_xv;
4556 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
4559 ret = ocfs2_xattr_set_entry_in_bucket(inode, xi, xs, name_hash,
4560 local, &bucket_empty);
4566 if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4567 /* allocate the space now for the outside block storage. */
4568 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4573 if (xs->not_found) {
4575 * We can't allocate enough clusters for outside
4576 * storage and we have allocated xattr already,
4577 * so need to remove it.
4579 ocfs2_xattr_bucket_remove_xs(inode, xs);
4585 ret = ocfs2_xattr_bucket_shrink(inode, xi,
4591 ret = ocfs2_xattr_bucket_set_value_outside(inode, xs, val, value_len);
4596 /* check whether the xattr bucket is filled up with the same hash value. */
4597 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
4598 struct ocfs2_xattr_bucket *bucket)
4600 struct ocfs2_xattr_header *xh = bucket->xh;
4602 if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
4603 xh->xh_entries[0].xe_name_hash) {
4604 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
4606 (unsigned long long)bucket->bhs[0]->b_blocknr,
4607 le32_to_cpu(xh->xh_entries[0].xe_name_hash));
4614 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
4615 struct ocfs2_xattr_info *xi,
4616 struct ocfs2_xattr_search *xs)
4618 struct ocfs2_xattr_header *xh;
4619 struct ocfs2_xattr_entry *xe;
4620 u16 count, header_size, xh_free_start;
4621 int i, free, max_free, need, old;
4622 size_t value_size = 0, name_len = strlen(xi->name);
4623 size_t blocksize = inode->i_sb->s_blocksize;
4624 int ret, allocation = 0;
4625 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4627 mlog_entry("Set xattr %s in xattr index block\n", xi->name);
4631 count = le16_to_cpu(xh->xh_count);
4632 xh_free_start = le16_to_cpu(xh->xh_free_start);
4633 header_size = sizeof(struct ocfs2_xattr_header) +
4634 count * sizeof(struct ocfs2_xattr_entry);
4635 max_free = OCFS2_XATTR_BUCKET_SIZE -
4636 le16_to_cpu(xh->xh_name_value_len) - header_size;
4638 mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
4639 "of %u which exceed block size\n",
4640 (unsigned long long)xs->bucket.bhs[0]->b_blocknr,
4643 if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4644 value_size = OCFS2_XATTR_ROOT_SIZE;
4646 value_size = OCFS2_XATTR_SIZE(xi->value_len);
4649 need = sizeof(struct ocfs2_xattr_entry) +
4650 OCFS2_XATTR_SIZE(name_len) + value_size;
4652 need = value_size + OCFS2_XATTR_SIZE(name_len);
4655 * We only replace the old value if the new length is smaller
4656 * than the old one. Otherwise we will allocate new space in the
4657 * bucket to store it.
4660 if (ocfs2_xattr_is_local(xe))
4661 old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4663 old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4665 if (old >= value_size)
4669 free = xh_free_start - header_size;
4671 * We need to make sure the new name/value pair
4672 * can exist in the same block.
4674 if (xh_free_start % blocksize < need)
4675 free -= xh_free_start % blocksize;
4677 mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
4678 "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
4679 " %u\n", xs->not_found,
4680 (unsigned long long)xs->bucket.bhs[0]->b_blocknr,
4681 free, need, max_free, le16_to_cpu(xh->xh_free_start),
4682 le16_to_cpu(xh->xh_name_value_len));
4684 if (free < need || count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
4685 if (need <= max_free &&
4686 count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
4688 * We can create the space by defragment. Since only the
4689 * name/value will be moved, the xe shouldn't be changed
4692 ret = ocfs2_defrag_xattr_bucket(inode, &xs->bucket);
4698 xh_free_start = le16_to_cpu(xh->xh_free_start);
4699 free = xh_free_start - header_size;
4700 if (xh_free_start % blocksize < need)
4701 free -= xh_free_start % blocksize;
4706 mlog(0, "Can't get enough space for xattr insert by "
4707 "defragment. Need %u bytes, but we have %d, so "
4708 "allocate new bucket for it.\n", need, free);
4712 * We have to add new buckets or clusters and one
4713 * allocation should leave us enough space for insert.
4718 * We do not allow for overlapping ranges between buckets. And
4719 * the maximum number of collisions we will allow for then is
4720 * one bucket's worth, so check it here whether we need to
4721 * add a new bucket for the insert.
4723 ret = ocfs2_check_xattr_bucket_collision(inode, &xs->bucket);
4729 ret = ocfs2_add_new_xattr_bucket(inode,
4737 for (i = 0; i < blk_per_bucket; i++)
4738 brelse(xs->bucket.bhs[i]);
4740 memset(&xs->bucket, 0, sizeof(xs->bucket));
4742 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
4745 if (ret && ret != -ENODATA)
4747 xs->not_found = ret;
4753 ret = ocfs2_xattr_set_in_bucket(inode, xi, xs);
4759 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
4760 struct ocfs2_xattr_bucket *bucket,
4764 struct ocfs2_xattr_header *xh = bucket->xh;
4766 struct ocfs2_xattr_entry *xe;
4768 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4769 xe = &xh->xh_entries[i];
4770 if (ocfs2_xattr_is_local(xe))
4773 ret = ocfs2_xattr_bucket_value_truncate(inode,
4785 static int ocfs2_delete_xattr_index_block(struct inode *inode,
4786 struct buffer_head *xb_bh)
4788 struct ocfs2_xattr_block *xb =
4789 (struct ocfs2_xattr_block *)xb_bh->b_data;
4790 struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4792 u32 name_hash = UINT_MAX, e_cpos, num_clusters;
4795 if (le16_to_cpu(el->l_next_free_rec) == 0)
4798 while (name_hash > 0) {
4799 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4800 &e_cpos, &num_clusters, el);
4806 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
4807 ocfs2_delete_xattr_in_bucket,
4814 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
4815 p_blkno, e_cpos, num_clusters);
4824 name_hash = e_cpos - 1;