1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * Copyright (C) 2004, 2008 Oracle. All rights reserved.
9 * Lots of code in this file is copy from linux/fs/ext3/xattr.c.
10 * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
12 * This program is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU General Public
14 * License version 2 as published by the Free Software Foundation.
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.
22 #include <linux/capability.h>
24 #include <linux/types.h>
25 #include <linux/slab.h>
26 #include <linux/highmem.h>
27 #include <linux/pagemap.h>
28 #include <linux/uio.h>
29 #include <linux/sched.h>
30 #include <linux/splice.h>
31 #include <linux/mount.h>
32 #include <linux/writeback.h>
33 #include <linux/falloc.h>
34 #include <linux/sort.h>
35 #include <linux/init.h>
36 #include <linux/module.h>
37 #include <linux/string.h>
39 #define MLOG_MASK_PREFIX ML_XATTR
40 #include <cluster/masklog.h>
53 #include "buffer_head_io.h"
58 struct ocfs2_xattr_def_value_root {
59 struct ocfs2_xattr_value_root xv;
60 struct ocfs2_extent_rec er;
63 struct ocfs2_xattr_bucket {
64 struct buffer_head *bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
65 struct ocfs2_xattr_header *xh;
68 #define OCFS2_XATTR_ROOT_SIZE (sizeof(struct ocfs2_xattr_def_value_root))
69 #define OCFS2_XATTR_INLINE_SIZE 80
71 static struct ocfs2_xattr_def_value_root def_xv = {
72 .xv.xr_list.l_count = cpu_to_le16(1),
75 struct xattr_handler *ocfs2_xattr_handlers[] = {
76 &ocfs2_xattr_user_handler,
77 &ocfs2_xattr_trusted_handler,
81 static struct xattr_handler *ocfs2_xattr_handler_map[] = {
82 [OCFS2_XATTR_INDEX_USER] = &ocfs2_xattr_user_handler,
83 [OCFS2_XATTR_INDEX_TRUSTED] = &ocfs2_xattr_trusted_handler,
86 struct ocfs2_xattr_info {
93 struct ocfs2_xattr_search {
94 struct buffer_head *inode_bh;
96 * xattr_bh point to the block buffer head which has extended attribute
97 * when extended attribute in inode, xattr_bh is equal to inode_bh.
99 struct buffer_head *xattr_bh;
100 struct ocfs2_xattr_header *header;
101 struct ocfs2_xattr_bucket bucket;
104 struct ocfs2_xattr_entry *here;
108 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
109 struct ocfs2_xattr_header *xh,
114 static int ocfs2_xattr_index_block_find(struct inode *inode,
115 struct buffer_head *root_bh,
118 struct ocfs2_xattr_search *xs);
120 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
121 struct ocfs2_xattr_tree_root *xt,
125 static int ocfs2_xattr_create_index_block(struct inode *inode,
126 struct ocfs2_xattr_search *xs);
128 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
129 struct ocfs2_xattr_info *xi,
130 struct ocfs2_xattr_search *xs);
132 static int ocfs2_delete_xattr_index_block(struct inode *inode,
133 struct buffer_head *xb_bh);
135 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
137 return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
140 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
142 return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
145 static inline u16 ocfs2_xattr_max_xe_in_bucket(struct super_block *sb)
147 u16 len = sb->s_blocksize -
148 offsetof(struct ocfs2_xattr_header, xh_entries);
150 return len / sizeof(struct ocfs2_xattr_entry);
153 static inline const char *ocfs2_xattr_prefix(int name_index)
155 struct xattr_handler *handler = NULL;
157 if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
158 handler = ocfs2_xattr_handler_map[name_index];
160 return handler ? handler->prefix : NULL;
163 static u32 ocfs2_xattr_name_hash(struct inode *inode,
167 /* Get hash value of uuid from super block */
168 u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
171 /* hash extended attribute name */
172 for (i = 0; i < name_len; i++) {
173 hash = (hash << OCFS2_HASH_SHIFT) ^
174 (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
182 * ocfs2_xattr_hash_entry()
184 * Compute the hash of an extended attribute.
186 static void ocfs2_xattr_hash_entry(struct inode *inode,
187 struct ocfs2_xattr_header *header,
188 struct ocfs2_xattr_entry *entry)
191 char *name = (char *)header + le16_to_cpu(entry->xe_name_offset);
193 hash = ocfs2_xattr_name_hash(inode, name, entry->xe_name_len);
194 entry->xe_name_hash = cpu_to_le32(hash);
199 static int ocfs2_xattr_extend_allocation(struct inode *inode,
201 struct buffer_head *xattr_bh,
202 struct ocfs2_xattr_value_root *xv)
205 int restart_func = 0;
207 handle_t *handle = NULL;
208 struct ocfs2_alloc_context *data_ac = NULL;
209 struct ocfs2_alloc_context *meta_ac = NULL;
210 enum ocfs2_alloc_restarted why;
211 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
212 u32 prev_clusters, logical_start = le32_to_cpu(xv->xr_clusters);
213 struct ocfs2_extent_tree et;
215 mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
217 ocfs2_init_xattr_value_extent_tree(&et, inode, xattr_bh, xv);
221 status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
228 credits = ocfs2_calc_extend_credits(osb->sb, et.et_root_el,
230 handle = ocfs2_start_trans(osb, credits);
231 if (IS_ERR(handle)) {
232 status = PTR_ERR(handle);
238 restarted_transaction:
239 status = ocfs2_journal_access(handle, inode, xattr_bh,
240 OCFS2_JOURNAL_ACCESS_WRITE);
246 prev_clusters = le32_to_cpu(xv->xr_clusters);
247 status = ocfs2_add_clusters_in_btree(osb,
257 if ((status < 0) && (status != -EAGAIN)) {
258 if (status != -ENOSPC)
263 status = ocfs2_journal_dirty(handle, xattr_bh);
269 clusters_to_add -= le32_to_cpu(xv->xr_clusters) - prev_clusters;
271 if (why != RESTART_NONE && clusters_to_add) {
272 if (why == RESTART_META) {
273 mlog(0, "restarting function.\n");
276 BUG_ON(why != RESTART_TRANS);
278 mlog(0, "restarting transaction.\n");
279 /* TODO: This can be more intelligent. */
280 credits = ocfs2_calc_extend_credits(osb->sb,
283 status = ocfs2_extend_trans(handle, credits);
285 /* handle still has to be committed at
291 goto restarted_transaction;
297 ocfs2_commit_trans(osb, handle);
301 ocfs2_free_alloc_context(data_ac);
305 ocfs2_free_alloc_context(meta_ac);
308 if ((!status) && restart_func) {
316 static int __ocfs2_remove_xattr_range(struct inode *inode,
317 struct buffer_head *root_bh,
318 struct ocfs2_xattr_value_root *xv,
319 u32 cpos, u32 phys_cpos, u32 len,
320 struct ocfs2_cached_dealloc_ctxt *dealloc)
323 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
324 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
325 struct inode *tl_inode = osb->osb_tl_inode;
327 struct ocfs2_alloc_context *meta_ac = NULL;
328 struct ocfs2_extent_tree et;
330 ocfs2_init_xattr_value_extent_tree(&et, inode, root_bh, xv);
332 ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
338 mutex_lock(&tl_inode->i_mutex);
340 if (ocfs2_truncate_log_needs_flush(osb)) {
341 ret = __ocfs2_flush_truncate_log(osb);
348 handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
349 if (IS_ERR(handle)) {
350 ret = PTR_ERR(handle);
355 ret = ocfs2_journal_access(handle, inode, root_bh,
356 OCFS2_JOURNAL_ACCESS_WRITE);
362 ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
369 le32_add_cpu(&xv->xr_clusters, -len);
371 ret = ocfs2_journal_dirty(handle, root_bh);
377 ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len);
382 ocfs2_commit_trans(osb, handle);
384 mutex_unlock(&tl_inode->i_mutex);
387 ocfs2_free_alloc_context(meta_ac);
392 static int ocfs2_xattr_shrink_size(struct inode *inode,
395 struct buffer_head *root_bh,
396 struct ocfs2_xattr_value_root *xv)
399 u32 trunc_len, cpos, phys_cpos, alloc_size;
401 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
402 struct ocfs2_cached_dealloc_ctxt dealloc;
404 ocfs2_init_dealloc_ctxt(&dealloc);
406 if (old_clusters <= new_clusters)
410 trunc_len = old_clusters - new_clusters;
412 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
413 &alloc_size, &xv->xr_list);
419 if (alloc_size > trunc_len)
420 alloc_size = trunc_len;
422 ret = __ocfs2_remove_xattr_range(inode, root_bh, xv, cpos,
423 phys_cpos, alloc_size,
430 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
431 ocfs2_remove_xattr_clusters_from_cache(inode, block,
434 trunc_len -= alloc_size;
438 ocfs2_schedule_truncate_log_flush(osb, 1);
439 ocfs2_run_deallocs(osb, &dealloc);
444 static int ocfs2_xattr_value_truncate(struct inode *inode,
445 struct buffer_head *root_bh,
446 struct ocfs2_xattr_value_root *xv,
450 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
451 u32 old_clusters = le32_to_cpu(xv->xr_clusters);
453 if (new_clusters == old_clusters)
456 if (new_clusters > old_clusters)
457 ret = ocfs2_xattr_extend_allocation(inode,
458 new_clusters - old_clusters,
461 ret = ocfs2_xattr_shrink_size(inode,
462 old_clusters, new_clusters,
468 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
469 size_t *result, const char *prefix,
470 const char *name, int name_len)
472 char *p = buffer + *result;
473 int prefix_len = strlen(prefix);
474 int total_len = prefix_len + name_len + 1;
476 *result += total_len;
478 /* we are just looking for how big our buffer needs to be */
485 memcpy(p, prefix, prefix_len);
486 memcpy(p + prefix_len, name, name_len);
487 p[prefix_len + name_len] = '\0';
492 static int ocfs2_xattr_list_entries(struct inode *inode,
493 struct ocfs2_xattr_header *header,
494 char *buffer, size_t buffer_size)
498 const char *prefix, *name;
500 for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
501 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
502 type = ocfs2_xattr_get_type(entry);
503 prefix = ocfs2_xattr_prefix(type);
506 name = (const char *)header +
507 le16_to_cpu(entry->xe_name_offset);
509 ret = ocfs2_xattr_list_entry(buffer, buffer_size,
510 &result, prefix, name,
520 static int ocfs2_xattr_ibody_list(struct inode *inode,
521 struct ocfs2_dinode *di,
525 struct ocfs2_xattr_header *header = NULL;
526 struct ocfs2_inode_info *oi = OCFS2_I(inode);
529 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
532 header = (struct ocfs2_xattr_header *)
533 ((void *)di + inode->i_sb->s_blocksize -
534 le16_to_cpu(di->i_xattr_inline_size));
536 ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
541 static int ocfs2_xattr_block_list(struct inode *inode,
542 struct ocfs2_dinode *di,
546 struct buffer_head *blk_bh = NULL;
547 struct ocfs2_xattr_block *xb;
550 if (!di->i_xattr_loc)
553 ret = ocfs2_read_block(inode, le64_to_cpu(di->i_xattr_loc), &blk_bh);
558 /*Verify the signature of xattr block*/
559 if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
560 strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
565 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
567 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
568 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
569 ret = ocfs2_xattr_list_entries(inode, header,
570 buffer, buffer_size);
572 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
573 ret = ocfs2_xattr_tree_list_index_block(inode, xt,
574 buffer, buffer_size);
582 ssize_t ocfs2_listxattr(struct dentry *dentry,
586 int ret = 0, i_ret = 0, b_ret = 0;
587 struct buffer_head *di_bh = NULL;
588 struct ocfs2_dinode *di = NULL;
589 struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
591 if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
594 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
597 ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
603 di = (struct ocfs2_dinode *)di_bh->b_data;
605 down_read(&oi->ip_xattr_sem);
606 i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
614 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
619 up_read(&oi->ip_xattr_sem);
620 ocfs2_inode_unlock(dentry->d_inode, 0);
624 return i_ret + b_ret;
627 static int ocfs2_xattr_find_entry(int name_index,
629 struct ocfs2_xattr_search *xs)
631 struct ocfs2_xattr_entry *entry;
638 name_len = strlen(name);
640 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
641 cmp = name_index - ocfs2_xattr_get_type(entry);
643 cmp = name_len - entry->xe_name_len;
645 cmp = memcmp(name, (xs->base +
646 le16_to_cpu(entry->xe_name_offset)),
654 return cmp ? -ENODATA : 0;
657 static int ocfs2_xattr_get_value_outside(struct inode *inode,
658 struct ocfs2_xattr_value_root *xv,
662 u32 cpos, p_cluster, num_clusters, bpc, clusters;
665 size_t cplen, blocksize;
666 struct buffer_head *bh = NULL;
667 struct ocfs2_extent_list *el;
670 clusters = le32_to_cpu(xv->xr_clusters);
671 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
672 blocksize = inode->i_sb->s_blocksize;
675 while (cpos < clusters) {
676 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
683 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
684 /* Copy ocfs2_xattr_value */
685 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
686 ret = ocfs2_read_block(inode, blkno, &bh);
692 cplen = len >= blocksize ? blocksize : len;
693 memcpy(buffer, bh->b_data, cplen);
702 cpos += num_clusters;
708 static int ocfs2_xattr_ibody_get(struct inode *inode,
713 struct ocfs2_xattr_search *xs)
715 struct ocfs2_inode_info *oi = OCFS2_I(inode);
716 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
717 struct ocfs2_xattr_value_root *xv;
721 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
724 xs->end = (void *)di + inode->i_sb->s_blocksize;
725 xs->header = (struct ocfs2_xattr_header *)
726 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
727 xs->base = (void *)xs->header;
728 xs->here = xs->header->xh_entries;
730 ret = ocfs2_xattr_find_entry(name_index, name, xs);
733 size = le64_to_cpu(xs->here->xe_value_size);
735 if (size > buffer_size)
737 if (ocfs2_xattr_is_local(xs->here)) {
738 memcpy(buffer, (void *)xs->base +
739 le16_to_cpu(xs->here->xe_name_offset) +
740 OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
742 xv = (struct ocfs2_xattr_value_root *)
743 (xs->base + le16_to_cpu(
744 xs->here->xe_name_offset) +
745 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
746 ret = ocfs2_xattr_get_value_outside(inode, xv,
758 static int ocfs2_xattr_block_get(struct inode *inode,
763 struct ocfs2_xattr_search *xs)
765 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
766 struct buffer_head *blk_bh = NULL;
767 struct ocfs2_xattr_block *xb;
768 struct ocfs2_xattr_value_root *xv;
770 int ret = -ENODATA, name_offset, name_len, block_off, i;
772 if (!di->i_xattr_loc)
775 memset(&xs->bucket, 0, sizeof(xs->bucket));
777 ret = ocfs2_read_block(inode, le64_to_cpu(di->i_xattr_loc), &blk_bh);
782 /*Verify the signature of xattr block*/
783 if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
784 strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
789 xs->xattr_bh = blk_bh;
790 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
792 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
793 xs->header = &xb->xb_attrs.xb_header;
794 xs->base = (void *)xs->header;
795 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
796 xs->here = xs->header->xh_entries;
798 ret = ocfs2_xattr_find_entry(name_index, name, xs);
800 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
806 size = le64_to_cpu(xs->here->xe_value_size);
809 if (size > buffer_size)
812 name_offset = le16_to_cpu(xs->here->xe_name_offset);
813 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
814 i = xs->here - xs->header->xh_entries;
816 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
817 ret = ocfs2_xattr_bucket_get_name_value(inode,
822 xs->base = xs->bucket.bhs[block_off]->b_data;
824 if (ocfs2_xattr_is_local(xs->here)) {
825 memcpy(buffer, (void *)xs->base +
826 name_offset + name_len, size);
828 xv = (struct ocfs2_xattr_value_root *)
829 (xs->base + name_offset + name_len);
830 ret = ocfs2_xattr_get_value_outside(inode, xv,
840 for (i = 0; i < OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET; i++)
841 brelse(xs->bucket.bhs[i]);
842 memset(&xs->bucket, 0, sizeof(xs->bucket));
850 * Copy an extended attribute into the buffer provided.
851 * Buffer is NULL to compute the size of buffer required.
853 static int ocfs2_xattr_get(struct inode *inode,
860 struct ocfs2_dinode *di = NULL;
861 struct buffer_head *di_bh = NULL;
862 struct ocfs2_inode_info *oi = OCFS2_I(inode);
863 struct ocfs2_xattr_search xis = {
864 .not_found = -ENODATA,
866 struct ocfs2_xattr_search xbs = {
867 .not_found = -ENODATA,
870 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
873 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
876 ret = ocfs2_inode_lock(inode, &di_bh, 0);
881 xis.inode_bh = xbs.inode_bh = di_bh;
882 di = (struct ocfs2_dinode *)di_bh->b_data;
884 down_read(&oi->ip_xattr_sem);
885 ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
888 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
890 up_read(&oi->ip_xattr_sem);
891 ocfs2_inode_unlock(inode, 0);
898 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
899 struct ocfs2_xattr_value_root *xv,
903 int ret = 0, i, cp_len, credits;
904 u16 blocksize = inode->i_sb->s_blocksize;
905 u32 p_cluster, num_clusters;
906 u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
907 u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
909 struct buffer_head *bh = NULL;
912 BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
914 credits = clusters * bpc;
915 handle = ocfs2_start_trans(OCFS2_SB(inode->i_sb), credits);
916 if (IS_ERR(handle)) {
917 ret = PTR_ERR(handle);
922 while (cpos < clusters) {
923 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
924 &num_clusters, &xv->xr_list);
930 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
932 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
933 ret = ocfs2_read_block(inode, blkno, &bh);
939 ret = ocfs2_journal_access(handle,
942 OCFS2_JOURNAL_ACCESS_WRITE);
948 cp_len = value_len > blocksize ? blocksize : value_len;
949 memcpy(bh->b_data, value, cp_len);
952 if (cp_len < blocksize)
953 memset(bh->b_data + cp_len, 0,
956 ret = ocfs2_journal_dirty(handle, bh);
965 * XXX: do we need to empty all the following
966 * blocks in this cluster?
971 cpos += num_clusters;
974 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
981 static int ocfs2_xattr_cleanup(struct inode *inode,
982 struct ocfs2_xattr_info *xi,
983 struct ocfs2_xattr_search *xs,
986 handle_t *handle = NULL;
988 size_t name_len = strlen(xi->name);
989 void *val = xs->base + offs;
990 size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
992 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
993 OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
994 if (IS_ERR(handle)) {
995 ret = PTR_ERR(handle);
999 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1000 OCFS2_JOURNAL_ACCESS_WRITE);
1005 /* Decrease xattr count */
1006 le16_add_cpu(&xs->header->xh_count, -1);
1007 /* Remove the xattr entry and tree root which has already be set*/
1008 memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
1009 memset(val, 0, size);
1011 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1015 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1020 static int ocfs2_xattr_update_entry(struct inode *inode,
1021 struct ocfs2_xattr_info *xi,
1022 struct ocfs2_xattr_search *xs,
1025 handle_t *handle = NULL;
1028 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1029 OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1030 if (IS_ERR(handle)) {
1031 ret = PTR_ERR(handle);
1035 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1036 OCFS2_JOURNAL_ACCESS_WRITE);
1042 xs->here->xe_name_offset = cpu_to_le16(offs);
1043 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1044 if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1045 ocfs2_xattr_set_local(xs->here, 1);
1047 ocfs2_xattr_set_local(xs->here, 0);
1048 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1050 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1054 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1060 * ocfs2_xattr_set_value_outside()
1062 * Set large size value in B tree.
1064 static int ocfs2_xattr_set_value_outside(struct inode *inode,
1065 struct ocfs2_xattr_info *xi,
1066 struct ocfs2_xattr_search *xs,
1069 size_t name_len = strlen(xi->name);
1070 void *val = xs->base + offs;
1071 struct ocfs2_xattr_value_root *xv = NULL;
1072 size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1075 memset(val, 0, size);
1076 memcpy(val, xi->name, name_len);
1077 xv = (struct ocfs2_xattr_value_root *)
1078 (val + OCFS2_XATTR_SIZE(name_len));
1079 xv->xr_clusters = 0;
1080 xv->xr_last_eb_blk = 0;
1081 xv->xr_list.l_tree_depth = 0;
1082 xv->xr_list.l_count = cpu_to_le16(1);
1083 xv->xr_list.l_next_free_rec = 0;
1085 ret = ocfs2_xattr_value_truncate(inode, xs->xattr_bh, xv,
1091 ret = __ocfs2_xattr_set_value_outside(inode, xv, xi->value,
1097 ret = ocfs2_xattr_update_entry(inode, xi, xs, offs);
1105 * ocfs2_xattr_set_entry_local()
1107 * Set, replace or remove extended attribute in local.
1109 static void ocfs2_xattr_set_entry_local(struct inode *inode,
1110 struct ocfs2_xattr_info *xi,
1111 struct ocfs2_xattr_search *xs,
1112 struct ocfs2_xattr_entry *last,
1115 size_t name_len = strlen(xi->name);
1118 if (xi->value && xs->not_found) {
1119 /* Insert the new xattr entry. */
1120 le16_add_cpu(&xs->header->xh_count, 1);
1121 ocfs2_xattr_set_type(last, xi->name_index);
1122 ocfs2_xattr_set_local(last, 1);
1123 last->xe_name_len = name_len;
1129 first_val = xs->base + min_offs;
1130 offs = le16_to_cpu(xs->here->xe_name_offset);
1131 val = xs->base + offs;
1133 if (le64_to_cpu(xs->here->xe_value_size) >
1134 OCFS2_XATTR_INLINE_SIZE)
1135 size = OCFS2_XATTR_SIZE(name_len) +
1136 OCFS2_XATTR_ROOT_SIZE;
1138 size = OCFS2_XATTR_SIZE(name_len) +
1139 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1141 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1142 OCFS2_XATTR_SIZE(xi->value_len)) {
1143 /* The old and the new value have the
1144 same size. Just replace the value. */
1145 ocfs2_xattr_set_local(xs->here, 1);
1146 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1147 /* Clear value bytes. */
1148 memset(val + OCFS2_XATTR_SIZE(name_len),
1150 OCFS2_XATTR_SIZE(xi->value_len));
1151 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1156 /* Remove the old name+value. */
1157 memmove(first_val + size, first_val, val - first_val);
1158 memset(first_val, 0, size);
1159 xs->here->xe_name_hash = 0;
1160 xs->here->xe_name_offset = 0;
1161 ocfs2_xattr_set_local(xs->here, 1);
1162 xs->here->xe_value_size = 0;
1166 /* Adjust all value offsets. */
1167 last = xs->header->xh_entries;
1168 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1169 size_t o = le16_to_cpu(last->xe_name_offset);
1172 last->xe_name_offset = cpu_to_le16(o + size);
1177 /* Remove the old entry. */
1179 memmove(xs->here, xs->here + 1,
1180 (void *)last - (void *)xs->here);
1181 memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1182 le16_add_cpu(&xs->header->xh_count, -1);
1186 /* Insert the new name+value. */
1187 size_t size = OCFS2_XATTR_SIZE(name_len) +
1188 OCFS2_XATTR_SIZE(xi->value_len);
1189 void *val = xs->base + min_offs - size;
1191 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1192 memset(val, 0, size);
1193 memcpy(val, xi->name, name_len);
1194 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1197 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1198 ocfs2_xattr_set_local(xs->here, 1);
1199 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1206 * ocfs2_xattr_set_entry()
1208 * Set extended attribute entry into inode or block.
1210 * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1211 * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1212 * then set value in B tree with set_value_outside().
1214 static int ocfs2_xattr_set_entry(struct inode *inode,
1215 struct ocfs2_xattr_info *xi,
1216 struct ocfs2_xattr_search *xs,
1219 struct ocfs2_xattr_entry *last;
1220 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1221 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1222 size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1224 handle_t *handle = NULL;
1226 struct ocfs2_xattr_info xi_l = {
1227 .name_index = xi->name_index,
1230 .value_len = xi->value_len,
1233 /* Compute min_offs, last and free space. */
1234 last = xs->header->xh_entries;
1236 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1237 size_t offs = le16_to_cpu(last->xe_name_offset);
1238 if (offs < min_offs)
1243 free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
1247 if (!xs->not_found) {
1249 if (ocfs2_xattr_is_local(xs->here))
1250 size = OCFS2_XATTR_SIZE(name_len) +
1251 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1253 size = OCFS2_XATTR_SIZE(name_len) +
1254 OCFS2_XATTR_ROOT_SIZE;
1255 free += (size + sizeof(struct ocfs2_xattr_entry));
1257 /* Check free space in inode or block */
1258 if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1259 if (free < sizeof(struct ocfs2_xattr_entry) +
1260 OCFS2_XATTR_SIZE(name_len) +
1261 OCFS2_XATTR_ROOT_SIZE) {
1265 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1266 xi_l.value = (void *)&def_xv;
1267 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1268 } else if (xi->value) {
1269 if (free < sizeof(struct ocfs2_xattr_entry) +
1270 OCFS2_XATTR_SIZE(name_len) +
1271 OCFS2_XATTR_SIZE(xi->value_len)) {
1277 if (!xs->not_found) {
1278 /* For existing extended attribute */
1279 size_t size = OCFS2_XATTR_SIZE(name_len) +
1280 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1281 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1282 void *val = xs->base + offs;
1284 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1285 /* Replace existing local xattr with tree root */
1286 ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1291 } else if (!ocfs2_xattr_is_local(xs->here)) {
1292 /* For existing xattr which has value outside */
1293 struct ocfs2_xattr_value_root *xv = NULL;
1294 xv = (struct ocfs2_xattr_value_root *)(val +
1295 OCFS2_XATTR_SIZE(name_len));
1297 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1299 * If new value need set outside also,
1300 * first truncate old value to new value,
1301 * then set new value with set_value_outside().
1303 ret = ocfs2_xattr_value_truncate(inode,
1312 ret = __ocfs2_xattr_set_value_outside(inode,
1321 ret = ocfs2_xattr_update_entry(inode,
1330 * If new value need set in local,
1331 * just trucate old value to zero.
1333 ret = ocfs2_xattr_value_truncate(inode,
1343 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1344 OCFS2_INODE_UPDATE_CREDITS);
1345 if (IS_ERR(handle)) {
1346 ret = PTR_ERR(handle);
1351 ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
1352 OCFS2_JOURNAL_ACCESS_WRITE);
1358 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1359 /* set extended attribute in external block. */
1360 ret = ocfs2_extend_trans(handle,
1361 OCFS2_INODE_UPDATE_CREDITS +
1362 OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1367 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1368 OCFS2_JOURNAL_ACCESS_WRITE);
1376 * Set value in local, include set tree root in local.
1377 * This is the first step for value size >INLINE_SIZE.
1379 ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1381 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1382 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1389 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1390 (flag & OCFS2_INLINE_XATTR_FL)) {
1391 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1392 unsigned int xattrsize = osb->s_xattr_inline_size;
1395 * Adjust extent record count or inline data size
1396 * to reserve space for extended attribute.
1398 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1399 struct ocfs2_inline_data *idata = &di->id2.i_data;
1400 le16_add_cpu(&idata->id_count, -xattrsize);
1401 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1402 struct ocfs2_extent_list *el = &di->id2.i_list;
1403 le16_add_cpu(&el->l_count, -(xattrsize /
1404 sizeof(struct ocfs2_extent_rec)));
1406 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1408 /* Update xattr flag */
1409 spin_lock(&oi->ip_lock);
1410 oi->ip_dyn_features |= flag;
1411 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1412 spin_unlock(&oi->ip_lock);
1413 /* Update inode ctime */
1414 inode->i_ctime = CURRENT_TIME;
1415 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1416 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1418 ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1423 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1425 if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1427 * Set value outside in B tree.
1428 * This is the second step for value size > INLINE_SIZE.
1430 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1431 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, offs);
1437 * If set value outside failed, we have to clean
1438 * the junk tree root we have already set in local.
1440 ret2 = ocfs2_xattr_cleanup(inode, xi, xs, offs);
1450 static int ocfs2_remove_value_outside(struct inode*inode,
1451 struct buffer_head *bh,
1452 struct ocfs2_xattr_header *header)
1456 for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1457 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1459 if (!ocfs2_xattr_is_local(entry)) {
1460 struct ocfs2_xattr_value_root *xv;
1463 val = (void *)header +
1464 le16_to_cpu(entry->xe_name_offset);
1465 xv = (struct ocfs2_xattr_value_root *)
1466 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
1467 ret = ocfs2_xattr_value_truncate(inode, bh, xv, 0);
1478 static int ocfs2_xattr_ibody_remove(struct inode *inode,
1479 struct buffer_head *di_bh)
1482 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1483 struct ocfs2_xattr_header *header;
1486 header = (struct ocfs2_xattr_header *)
1487 ((void *)di + inode->i_sb->s_blocksize -
1488 le16_to_cpu(di->i_xattr_inline_size));
1490 ret = ocfs2_remove_value_outside(inode, di_bh, header);
1495 static int ocfs2_xattr_block_remove(struct inode *inode,
1496 struct buffer_head *blk_bh)
1498 struct ocfs2_xattr_block *xb;
1501 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1502 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1503 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1504 ret = ocfs2_remove_value_outside(inode, blk_bh, header);
1506 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
1511 static int ocfs2_xattr_free_block(struct inode *inode,
1514 struct inode *xb_alloc_inode;
1515 struct buffer_head *xb_alloc_bh = NULL;
1516 struct buffer_head *blk_bh = NULL;
1517 struct ocfs2_xattr_block *xb;
1518 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1524 ret = ocfs2_read_block(inode, block, &blk_bh);
1530 /*Verify the signature of xattr block*/
1531 if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
1532 strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
1537 ret = ocfs2_xattr_block_remove(inode, blk_bh);
1543 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1544 blk = le64_to_cpu(xb->xb_blkno);
1545 bit = le16_to_cpu(xb->xb_suballoc_bit);
1546 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1548 xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1549 EXTENT_ALLOC_SYSTEM_INODE,
1550 le16_to_cpu(xb->xb_suballoc_slot));
1551 if (!xb_alloc_inode) {
1556 mutex_lock(&xb_alloc_inode->i_mutex);
1558 ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1564 handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
1565 if (IS_ERR(handle)) {
1566 ret = PTR_ERR(handle);
1571 ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1576 ocfs2_commit_trans(osb, handle);
1578 ocfs2_inode_unlock(xb_alloc_inode, 1);
1579 brelse(xb_alloc_bh);
1581 mutex_unlock(&xb_alloc_inode->i_mutex);
1582 iput(xb_alloc_inode);
1589 * ocfs2_xattr_remove()
1591 * Free extended attribute resources associated with this inode.
1593 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1595 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1596 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1600 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1603 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1606 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1607 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1614 if (di->i_xattr_loc) {
1615 ret = ocfs2_xattr_free_block(inode,
1616 le64_to_cpu(di->i_xattr_loc));
1623 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1624 OCFS2_INODE_UPDATE_CREDITS);
1625 if (IS_ERR(handle)) {
1626 ret = PTR_ERR(handle);
1630 ret = ocfs2_journal_access(handle, inode, di_bh,
1631 OCFS2_JOURNAL_ACCESS_WRITE);
1637 di->i_xattr_loc = 0;
1639 spin_lock(&oi->ip_lock);
1640 oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1641 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1642 spin_unlock(&oi->ip_lock);
1644 ret = ocfs2_journal_dirty(handle, di_bh);
1648 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1653 static int ocfs2_xattr_has_space_inline(struct inode *inode,
1654 struct ocfs2_dinode *di)
1656 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1657 unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1660 if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1663 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1664 struct ocfs2_inline_data *idata = &di->id2.i_data;
1665 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1666 } else if (ocfs2_inode_is_fast_symlink(inode)) {
1667 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1668 le64_to_cpu(di->i_size);
1670 struct ocfs2_extent_list *el = &di->id2.i_list;
1671 free = (le16_to_cpu(el->l_count) -
1672 le16_to_cpu(el->l_next_free_rec)) *
1673 sizeof(struct ocfs2_extent_rec);
1675 if (free >= xattrsize)
1682 * ocfs2_xattr_ibody_find()
1684 * Find extended attribute in inode block and
1685 * fill search info into struct ocfs2_xattr_search.
1687 static int ocfs2_xattr_ibody_find(struct inode *inode,
1690 struct ocfs2_xattr_search *xs)
1692 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1693 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1697 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1700 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1701 down_read(&oi->ip_alloc_sem);
1702 has_space = ocfs2_xattr_has_space_inline(inode, di);
1703 up_read(&oi->ip_alloc_sem);
1708 xs->xattr_bh = xs->inode_bh;
1709 xs->end = (void *)di + inode->i_sb->s_blocksize;
1710 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1711 xs->header = (struct ocfs2_xattr_header *)
1712 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1714 xs->header = (struct ocfs2_xattr_header *)
1715 (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
1716 xs->base = (void *)xs->header;
1717 xs->here = xs->header->xh_entries;
1719 /* Find the named attribute. */
1720 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1721 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1722 if (ret && ret != -ENODATA)
1724 xs->not_found = ret;
1731 * ocfs2_xattr_ibody_set()
1733 * Set, replace or remove an extended attribute into inode block.
1736 static int ocfs2_xattr_ibody_set(struct inode *inode,
1737 struct ocfs2_xattr_info *xi,
1738 struct ocfs2_xattr_search *xs)
1740 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1741 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1744 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1747 down_write(&oi->ip_alloc_sem);
1748 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1749 if (!ocfs2_xattr_has_space_inline(inode, di)) {
1755 ret = ocfs2_xattr_set_entry(inode, xi, xs,
1756 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
1758 up_write(&oi->ip_alloc_sem);
1764 * ocfs2_xattr_block_find()
1766 * Find extended attribute in external block and
1767 * fill search info into struct ocfs2_xattr_search.
1769 static int ocfs2_xattr_block_find(struct inode *inode,
1772 struct ocfs2_xattr_search *xs)
1774 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1775 struct buffer_head *blk_bh = NULL;
1776 struct ocfs2_xattr_block *xb;
1779 if (!di->i_xattr_loc)
1782 ret = ocfs2_read_block(inode, le64_to_cpu(di->i_xattr_loc), &blk_bh);
1787 /*Verify the signature of xattr block*/
1788 if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
1789 strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
1794 xs->xattr_bh = blk_bh;
1795 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1797 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1798 xs->header = &xb->xb_attrs.xb_header;
1799 xs->base = (void *)xs->header;
1800 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
1801 xs->here = xs->header->xh_entries;
1803 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1805 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
1809 if (ret && ret != -ENODATA) {
1810 xs->xattr_bh = NULL;
1813 xs->not_found = ret;
1822 * When all the xattrs are deleted from index btree, the ocfs2_xattr_tree
1823 * will be erased and ocfs2_xattr_block will have its ocfs2_xattr_header
1826 static int ocfs2_restore_xattr_block(struct inode *inode,
1827 struct ocfs2_xattr_search *xs)
1831 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1832 struct ocfs2_xattr_block *xb =
1833 (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1834 struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
1835 u16 xb_flags = le16_to_cpu(xb->xb_flags);
1837 BUG_ON(!(xb_flags & OCFS2_XATTR_INDEXED) ||
1838 le16_to_cpu(el->l_next_free_rec) != 0);
1840 handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1841 if (IS_ERR(handle)) {
1842 ret = PTR_ERR(handle);
1847 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1848 OCFS2_JOURNAL_ACCESS_WRITE);
1854 memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
1855 offsetof(struct ocfs2_xattr_block, xb_attrs));
1857 xb->xb_flags = cpu_to_le16(xb_flags & ~OCFS2_XATTR_INDEXED);
1859 ocfs2_journal_dirty(handle, xs->xattr_bh);
1862 ocfs2_commit_trans(osb, handle);
1868 * ocfs2_xattr_block_set()
1870 * Set, replace or remove an extended attribute into external block.
1873 static int ocfs2_xattr_block_set(struct inode *inode,
1874 struct ocfs2_xattr_info *xi,
1875 struct ocfs2_xattr_search *xs)
1877 struct buffer_head *new_bh = NULL;
1878 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1879 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1880 struct ocfs2_alloc_context *meta_ac = NULL;
1881 handle_t *handle = NULL;
1882 struct ocfs2_xattr_block *xblk = NULL;
1883 u16 suballoc_bit_start;
1888 if (!xs->xattr_bh) {
1890 * Alloc one external block for extended attribute
1893 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
1898 handle = ocfs2_start_trans(osb,
1899 OCFS2_XATTR_BLOCK_CREATE_CREDITS);
1900 if (IS_ERR(handle)) {
1901 ret = PTR_ERR(handle);
1905 ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
1906 OCFS2_JOURNAL_ACCESS_CREATE);
1912 ret = ocfs2_claim_metadata(osb, handle, meta_ac, 1,
1913 &suballoc_bit_start, &num_got,
1920 new_bh = sb_getblk(inode->i_sb, first_blkno);
1921 ocfs2_set_new_buffer_uptodate(inode, new_bh);
1923 ret = ocfs2_journal_access(handle, inode, new_bh,
1924 OCFS2_JOURNAL_ACCESS_CREATE);
1930 /* Initialize ocfs2_xattr_block */
1931 xs->xattr_bh = new_bh;
1932 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
1933 memset(xblk, 0, inode->i_sb->s_blocksize);
1934 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
1935 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
1936 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1937 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
1938 xblk->xb_blkno = cpu_to_le64(first_blkno);
1940 xs->header = &xblk->xb_attrs.xb_header;
1941 xs->base = (void *)xs->header;
1942 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
1943 xs->here = xs->header->xh_entries;
1946 ret = ocfs2_journal_dirty(handle, new_bh);
1951 di->i_xattr_loc = cpu_to_le64(first_blkno);
1952 ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1956 ocfs2_commit_trans(osb, handle);
1959 ocfs2_free_alloc_context(meta_ac);
1963 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1965 if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
1966 /* Set extended attribute into external block */
1967 ret = ocfs2_xattr_set_entry(inode, xi, xs, OCFS2_HAS_XATTR_FL);
1968 if (!ret || ret != -ENOSPC)
1971 ret = ocfs2_xattr_create_index_block(inode, xs);
1976 ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs);
1977 if (!ret && xblk->xb_attrs.xb_root.xt_list.l_next_free_rec == 0)
1978 ret = ocfs2_restore_xattr_block(inode, xs);
1988 * Set, replace or remove an extended attribute for this inode.
1989 * value is NULL to remove an existing extended attribute, else either
1990 * create or replace an extended attribute.
1992 int ocfs2_xattr_set(struct inode *inode,
1999 struct buffer_head *di_bh = NULL;
2000 struct ocfs2_dinode *di;
2002 u16 i, blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2004 struct ocfs2_xattr_info xi = {
2005 .name_index = name_index,
2008 .value_len = value_len,
2011 struct ocfs2_xattr_search xis = {
2012 .not_found = -ENODATA,
2015 struct ocfs2_xattr_search xbs = {
2016 .not_found = -ENODATA,
2019 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2022 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2027 xis.inode_bh = xbs.inode_bh = di_bh;
2028 di = (struct ocfs2_dinode *)di_bh->b_data;
2030 down_write(&OCFS2_I(inode)->ip_xattr_sem);
2032 * Scan inode and external block to find the same name
2033 * extended attribute and collect search infomation.
2035 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2038 if (xis.not_found) {
2039 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2044 if (xis.not_found && xbs.not_found) {
2046 if (flags & XATTR_REPLACE)
2053 if (flags & XATTR_CREATE)
2058 /* Remove existing extended attribute */
2060 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2061 else if (!xbs.not_found)
2062 ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2064 /* We always try to set extended attribute into inode first*/
2065 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2066 if (!ret && !xbs.not_found) {
2068 * If succeed and that extended attribute existing in
2069 * external block, then we will remove it.
2073 ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2074 } else if (ret == -ENOSPC) {
2075 if (di->i_xattr_loc && !xbs.xattr_bh) {
2076 ret = ocfs2_xattr_block_find(inode, name_index,
2082 * If no space in inode, we will set extended attribute
2083 * into external block.
2085 ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2088 if (!xis.not_found) {
2090 * If succeed and that extended attribute
2091 * existing in inode, we will remove it.
2095 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2100 up_write(&OCFS2_I(inode)->ip_xattr_sem);
2101 ocfs2_inode_unlock(inode, 1);
2103 brelse(xbs.xattr_bh);
2104 for (i = 0; i < blk_per_bucket; i++)
2105 brelse(xbs.bucket.bhs[i]);
2111 * Find the xattr extent rec which may contains name_hash.
2112 * e_cpos will be the first name hash of the xattr rec.
2113 * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2115 static int ocfs2_xattr_get_rec(struct inode *inode,
2120 struct ocfs2_extent_list *el)
2123 struct buffer_head *eb_bh = NULL;
2124 struct ocfs2_extent_block *eb;
2125 struct ocfs2_extent_rec *rec = NULL;
2128 if (el->l_tree_depth) {
2129 ret = ocfs2_find_leaf(inode, el, name_hash, &eb_bh);
2135 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2138 if (el->l_tree_depth) {
2139 ocfs2_error(inode->i_sb,
2140 "Inode %lu has non zero tree depth in "
2141 "xattr tree block %llu\n", inode->i_ino,
2142 (unsigned long long)eb_bh->b_blocknr);
2148 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2149 rec = &el->l_recs[i];
2151 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2152 e_blkno = le64_to_cpu(rec->e_blkno);
2158 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2159 "record (%u, %u, 0) in xattr", inode->i_ino,
2160 le32_to_cpu(rec->e_cpos),
2161 ocfs2_rec_clusters(el, rec));
2166 *p_blkno = le64_to_cpu(rec->e_blkno);
2167 *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2169 *e_cpos = le32_to_cpu(rec->e_cpos);
2175 typedef int (xattr_bucket_func)(struct inode *inode,
2176 struct ocfs2_xattr_bucket *bucket,
2179 static int ocfs2_find_xe_in_bucket(struct inode *inode,
2180 struct buffer_head *header_bh,
2187 int i, ret = 0, cmp = 1, block_off, new_offset;
2188 struct ocfs2_xattr_header *xh =
2189 (struct ocfs2_xattr_header *)header_bh->b_data;
2190 size_t name_len = strlen(name);
2191 struct ocfs2_xattr_entry *xe = NULL;
2192 struct buffer_head *name_bh = NULL;
2196 * We don't use binary search in the bucket because there
2197 * may be multiple entries with the same name hash.
2199 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2200 xe = &xh->xh_entries[i];
2202 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2204 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2207 cmp = name_index - ocfs2_xattr_get_type(xe);
2209 cmp = name_len - xe->xe_name_len;
2213 ret = ocfs2_xattr_bucket_get_name_value(inode,
2223 ret = ocfs2_read_block(inode, header_bh->b_blocknr + block_off,
2229 xe_name = name_bh->b_data + new_offset;
2231 cmp = memcmp(name, xe_name, name_len);
2247 * Find the specified xattr entry in a series of buckets.
2248 * This series start from p_blkno and last for num_clusters.
2249 * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2250 * the num of the valid buckets.
2252 * Return the buffer_head this xattr should reside in. And if the xattr's
2253 * hash is in the gap of 2 buckets, return the lower bucket.
2255 static int ocfs2_xattr_bucket_find(struct inode *inode,
2262 struct ocfs2_xattr_search *xs)
2265 struct buffer_head *bh = NULL;
2266 struct buffer_head *lower_bh = NULL;
2267 struct ocfs2_xattr_header *xh = NULL;
2268 struct ocfs2_xattr_entry *xe = NULL;
2270 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2271 int low_bucket = 0, bucket, high_bucket;
2275 ret = ocfs2_read_block(inode, p_blkno, &bh);
2281 xh = (struct ocfs2_xattr_header *)bh->b_data;
2282 high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
2284 while (low_bucket <= high_bucket) {
2287 bucket = (low_bucket + high_bucket) / 2;
2289 blkno = p_blkno + bucket * blk_per_bucket;
2291 ret = ocfs2_read_block(inode, blkno, &bh);
2297 xh = (struct ocfs2_xattr_header *)bh->b_data;
2298 xe = &xh->xh_entries[0];
2299 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
2300 high_bucket = bucket - 1;
2305 * Check whether the hash of the last entry in our
2306 * bucket is larger than the search one. for an empty
2307 * bucket, the last one is also the first one.
2310 xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
2312 last_hash = le32_to_cpu(xe->xe_name_hash);
2314 /* record lower_bh which may be the insert place. */
2319 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
2320 low_bucket = bucket + 1;
2324 /* the searched xattr should reside in this bucket if exists. */
2325 ret = ocfs2_find_xe_in_bucket(inode, lower_bh,
2326 name_index, name, name_hash,
2336 * Record the bucket we have found.
2337 * When the xattr's hash value is in the gap of 2 buckets, we will
2338 * always set it to the previous bucket.
2342 * We can't find any bucket whose first name_hash is less
2343 * than the find name_hash.
2345 BUG_ON(bh->b_blocknr != p_blkno);
2349 xs->bucket.bhs[0] = lower_bh;
2350 xs->bucket.xh = (struct ocfs2_xattr_header *)
2351 xs->bucket.bhs[0]->b_data;
2354 xs->header = xs->bucket.xh;
2355 xs->base = xs->bucket.bhs[0]->b_data;
2356 xs->end = xs->base + inode->i_sb->s_blocksize;
2360 * If we have found the xattr enty, read all the blocks in
2363 ret = ocfs2_read_blocks(inode, xs->bucket.bhs[0]->b_blocknr + 1,
2364 blk_per_bucket - 1, &xs->bucket.bhs[1],
2371 xs->here = &xs->header->xh_entries[index];
2372 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
2373 (unsigned long long)xs->bucket.bhs[0]->b_blocknr, index);
2383 static int ocfs2_xattr_index_block_find(struct inode *inode,
2384 struct buffer_head *root_bh,
2387 struct ocfs2_xattr_search *xs)
2390 struct ocfs2_xattr_block *xb =
2391 (struct ocfs2_xattr_block *)root_bh->b_data;
2392 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
2393 struct ocfs2_extent_list *el = &xb_root->xt_list;
2395 u32 first_hash, num_clusters = 0;
2396 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
2398 if (le16_to_cpu(el->l_next_free_rec) == 0)
2401 mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
2402 name, name_hash, name_index);
2404 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
2411 BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
2413 mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
2414 "in the rec is %u\n", num_clusters, p_blkno, first_hash);
2416 ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
2417 p_blkno, first_hash, num_clusters, xs);
2423 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
2426 xattr_bucket_func *func,
2430 int blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2431 u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
2432 u32 num_buckets = clusters * bpc;
2433 struct ocfs2_xattr_bucket bucket;
2435 memset(&bucket, 0, sizeof(bucket));
2437 mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
2440 for (i = 0; i < num_buckets; i++, blkno += blk_per_bucket) {
2441 ret = ocfs2_read_blocks(inode, blkno, blk_per_bucket,
2448 bucket.xh = (struct ocfs2_xattr_header *)bucket.bhs[0]->b_data;
2450 * The real bucket num in this series of blocks is stored
2451 * in the 1st bucket.
2454 num_buckets = le16_to_cpu(bucket.xh->xh_num_buckets);
2456 mlog(0, "iterating xattr bucket %llu, first hash %u\n", blkno,
2457 le32_to_cpu(bucket.xh->xh_entries[0].xe_name_hash));
2459 ret = func(inode, &bucket, para);
2466 for (j = 0; j < blk_per_bucket; j++)
2467 brelse(bucket.bhs[j]);
2468 memset(&bucket, 0, sizeof(bucket));
2472 for (j = 0; j < blk_per_bucket; j++)
2473 brelse(bucket.bhs[j]);
2478 struct ocfs2_xattr_tree_list {
2484 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
2485 struct ocfs2_xattr_header *xh,
2492 if (index < 0 || index >= le16_to_cpu(xh->xh_count))
2495 name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
2497 *block_off = name_offset >> inode->i_sb->s_blocksize_bits;
2498 *new_offset = name_offset % inode->i_sb->s_blocksize;
2503 static int ocfs2_list_xattr_bucket(struct inode *inode,
2504 struct ocfs2_xattr_bucket *bucket,
2508 struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
2509 int i, block_off, new_offset;
2510 const char *prefix, *name;
2512 for (i = 0 ; i < le16_to_cpu(bucket->xh->xh_count); i++) {
2513 struct ocfs2_xattr_entry *entry = &bucket->xh->xh_entries[i];
2514 type = ocfs2_xattr_get_type(entry);
2515 prefix = ocfs2_xattr_prefix(type);
2518 ret = ocfs2_xattr_bucket_get_name_value(inode,
2526 name = (const char *)bucket->bhs[block_off]->b_data +
2528 ret = ocfs2_xattr_list_entry(xl->buffer,
2532 entry->xe_name_len);
2541 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
2542 struct ocfs2_xattr_tree_root *xt,
2546 struct ocfs2_extent_list *el = &xt->xt_list;
2548 u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
2550 struct ocfs2_xattr_tree_list xl = {
2552 .buffer_size = buffer_size,
2556 if (le16_to_cpu(el->l_next_free_rec) == 0)
2559 while (name_hash > 0) {
2560 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
2561 &e_cpos, &num_clusters, el);
2567 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
2568 ocfs2_list_xattr_bucket,
2578 name_hash = e_cpos - 1;
2586 static int cmp_xe(const void *a, const void *b)
2588 const struct ocfs2_xattr_entry *l = a, *r = b;
2589 u32 l_hash = le32_to_cpu(l->xe_name_hash);
2590 u32 r_hash = le32_to_cpu(r->xe_name_hash);
2592 if (l_hash > r_hash)
2594 if (l_hash < r_hash)
2599 static void swap_xe(void *a, void *b, int size)
2601 struct ocfs2_xattr_entry *l = a, *r = b, tmp;
2604 memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
2605 memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
2609 * When the ocfs2_xattr_block is filled up, new bucket will be created
2610 * and all the xattr entries will be moved to the new bucket.
2611 * Note: we need to sort the entries since they are not saved in order
2612 * in the ocfs2_xattr_block.
2614 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
2615 struct buffer_head *xb_bh,
2616 struct buffer_head *xh_bh,
2617 struct buffer_head *data_bh)
2619 int i, blocksize = inode->i_sb->s_blocksize;
2620 u16 offset, size, off_change;
2621 struct ocfs2_xattr_entry *xe;
2622 struct ocfs2_xattr_block *xb =
2623 (struct ocfs2_xattr_block *)xb_bh->b_data;
2624 struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
2625 struct ocfs2_xattr_header *xh =
2626 (struct ocfs2_xattr_header *)xh_bh->b_data;
2627 u16 count = le16_to_cpu(xb_xh->xh_count);
2628 char *target = xh_bh->b_data, *src = xb_bh->b_data;
2630 mlog(0, "cp xattr from block %llu to bucket %llu\n",
2631 (unsigned long long)xb_bh->b_blocknr,
2632 (unsigned long long)xh_bh->b_blocknr);
2634 memset(xh_bh->b_data, 0, blocksize);
2636 memset(data_bh->b_data, 0, blocksize);
2638 * Since the xe_name_offset is based on ocfs2_xattr_header,
2639 * there is a offset change corresponding to the change of
2640 * ocfs2_xattr_header's position.
2642 off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
2643 xe = &xb_xh->xh_entries[count - 1];
2644 offset = le16_to_cpu(xe->xe_name_offset) + off_change;
2645 size = blocksize - offset;
2647 /* copy all the names and values. */
2649 target = data_bh->b_data;
2650 memcpy(target + offset, src + offset, size);
2652 /* Init new header now. */
2653 xh->xh_count = xb_xh->xh_count;
2654 xh->xh_num_buckets = cpu_to_le16(1);
2655 xh->xh_name_value_len = cpu_to_le16(size);
2656 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
2658 /* copy all the entries. */
2659 target = xh_bh->b_data;
2660 offset = offsetof(struct ocfs2_xattr_header, xh_entries);
2661 size = count * sizeof(struct ocfs2_xattr_entry);
2662 memcpy(target + offset, (char *)xb_xh + offset, size);
2664 /* Change the xe offset for all the xe because of the move. */
2665 off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
2666 offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
2667 for (i = 0; i < count; i++)
2668 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
2670 mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
2671 offset, size, off_change);
2673 sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
2678 * After we move xattr from block to index btree, we have to
2679 * update ocfs2_xattr_search to the new xe and base.
2681 * When the entry is in xattr block, xattr_bh indicates the storage place.
2682 * While if the entry is in index b-tree, "bucket" indicates the
2683 * real place of the xattr.
2685 static int ocfs2_xattr_update_xattr_search(struct inode *inode,
2686 struct ocfs2_xattr_search *xs,
2687 struct buffer_head *old_bh,
2688 struct buffer_head *new_bh)
2691 char *buf = old_bh->b_data;
2692 struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
2693 struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
2694 int i, blocksize = inode->i_sb->s_blocksize;
2695 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2697 xs->bucket.bhs[0] = new_bh;
2699 xs->bucket.xh = (struct ocfs2_xattr_header *)xs->bucket.bhs[0]->b_data;
2700 xs->header = xs->bucket.xh;
2702 xs->base = new_bh->b_data;
2703 xs->end = xs->base + inode->i_sb->s_blocksize;
2705 if (!xs->not_found) {
2706 if (OCFS2_XATTR_BUCKET_SIZE != blocksize) {
2707 ret = ocfs2_read_blocks(inode,
2708 xs->bucket.bhs[0]->b_blocknr + 1,
2709 blk_per_bucket - 1, &xs->bucket.bhs[1],
2716 i = xs->here - old_xh->xh_entries;
2717 xs->here = &xs->header->xh_entries[i];
2724 static int ocfs2_xattr_create_index_block(struct inode *inode,
2725 struct ocfs2_xattr_search *xs)
2727 int ret, credits = OCFS2_SUBALLOC_ALLOC;
2731 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2732 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2733 struct ocfs2_alloc_context *data_ac;
2734 struct buffer_head *xh_bh = NULL, *data_bh = NULL;
2735 struct buffer_head *xb_bh = xs->xattr_bh;
2736 struct ocfs2_xattr_block *xb =
2737 (struct ocfs2_xattr_block *)xb_bh->b_data;
2738 struct ocfs2_xattr_tree_root *xr;
2739 u16 xb_flags = le16_to_cpu(xb->xb_flags);
2740 u16 bpb = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2742 mlog(0, "create xattr index block for %llu\n",
2743 (unsigned long long)xb_bh->b_blocknr);
2745 BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
2747 ret = ocfs2_reserve_clusters(osb, 1, &data_ac);
2755 * We can use this lock for now, and maybe move to a dedicated mutex
2756 * if performance becomes a problem later.
2758 down_write(&oi->ip_alloc_sem);
2761 * 3 more credits, one for xattr block update, one for the 1st block
2762 * of the new xattr bucket and one for the value/data.
2765 handle = ocfs2_start_trans(osb, credits);
2766 if (IS_ERR(handle)) {
2767 ret = PTR_ERR(handle);
2772 ret = ocfs2_journal_access(handle, inode, xb_bh,
2773 OCFS2_JOURNAL_ACCESS_WRITE);
2779 ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off, &len);
2786 * The bucket may spread in many blocks, and
2787 * we will only touch the 1st block and the last block
2788 * in the whole bucket(one for entry and one for data).
2790 blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
2792 mlog(0, "allocate 1 cluster from %llu to xattr block\n", blkno);
2794 xh_bh = sb_getblk(inode->i_sb, blkno);
2801 ocfs2_set_new_buffer_uptodate(inode, xh_bh);
2803 ret = ocfs2_journal_access(handle, inode, xh_bh,
2804 OCFS2_JOURNAL_ACCESS_CREATE);
2811 data_bh = sb_getblk(inode->i_sb, blkno + bpb - 1);
2818 ocfs2_set_new_buffer_uptodate(inode, data_bh);
2820 ret = ocfs2_journal_access(handle, inode, data_bh,
2821 OCFS2_JOURNAL_ACCESS_CREATE);
2828 ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xh_bh, data_bh);
2830 ocfs2_journal_dirty(handle, xh_bh);
2832 ocfs2_journal_dirty(handle, data_bh);
2834 ocfs2_xattr_update_xattr_search(inode, xs, xb_bh, xh_bh);
2836 /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
2837 memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
2838 offsetof(struct ocfs2_xattr_block, xb_attrs));
2840 xr = &xb->xb_attrs.xb_root;
2841 xr->xt_clusters = cpu_to_le32(1);
2842 xr->xt_last_eb_blk = 0;
2843 xr->xt_list.l_tree_depth = 0;
2844 xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
2845 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2847 xr->xt_list.l_recs[0].e_cpos = 0;
2848 xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
2849 xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
2851 xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
2853 ret = ocfs2_journal_dirty(handle, xb_bh);
2860 ocfs2_commit_trans(osb, handle);
2863 up_write(&oi->ip_alloc_sem);
2867 ocfs2_free_alloc_context(data_ac);
2875 static int cmp_xe_offset(const void *a, const void *b)
2877 const struct ocfs2_xattr_entry *l = a, *r = b;
2878 u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
2879 u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
2881 if (l_name_offset < r_name_offset)
2883 if (l_name_offset > r_name_offset)
2889 * defrag a xattr bucket if we find that the bucket has some
2890 * holes beteen name/value pairs.
2891 * We will move all the name/value pairs to the end of the bucket
2892 * so that we can spare some space for insertion.
2894 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
2895 struct ocfs2_xattr_bucket *bucket)
2898 size_t end, offset, len, value_len;
2899 struct ocfs2_xattr_header *xh;
2900 char *entries, *buf, *bucket_buf = NULL;
2901 u64 blkno = bucket->bhs[0]->b_blocknr;
2902 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2904 size_t blocksize = inode->i_sb->s_blocksize;
2906 struct buffer_head **bhs;
2907 struct ocfs2_xattr_entry *xe;
2909 bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
2914 ret = ocfs2_read_blocks(inode, blkno, blk_per_bucket, bhs, 0);
2919 * In order to make the operation more efficient and generic,
2920 * we copy all the blocks into a contiguous memory and do the
2921 * defragment there, so if anything is error, we will not touch
2924 bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
2931 for (i = 0; i < blk_per_bucket; i++, buf += blocksize)
2932 memcpy(buf, bhs[i]->b_data, blocksize);
2934 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), blk_per_bucket);
2935 if (IS_ERR(handle)) {
2936 ret = PTR_ERR(handle);
2942 for (i = 0; i < blk_per_bucket; i++) {
2943 ret = ocfs2_journal_access(handle, inode, bhs[i],
2944 OCFS2_JOURNAL_ACCESS_WRITE);
2951 xh = (struct ocfs2_xattr_header *)bucket_buf;
2952 entries = (char *)xh->xh_entries;
2953 xh_free_start = le16_to_cpu(xh->xh_free_start);
2955 mlog(0, "adjust xattr bucket in %llu, count = %u, "
2956 "xh_free_start = %u, xh_name_value_len = %u.\n",
2957 blkno, le16_to_cpu(xh->xh_count), xh_free_start,
2958 le16_to_cpu(xh->xh_name_value_len));
2961 * sort all the entries by their offset.
2962 * the largest will be the first, so that we can
2963 * move them to the end one by one.
2965 sort(entries, le16_to_cpu(xh->xh_count),
2966 sizeof(struct ocfs2_xattr_entry),
2967 cmp_xe_offset, swap_xe);
2969 /* Move all name/values to the end of the bucket. */
2970 xe = xh->xh_entries;
2971 end = OCFS2_XATTR_BUCKET_SIZE;
2972 for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
2973 offset = le16_to_cpu(xe->xe_name_offset);
2974 if (ocfs2_xattr_is_local(xe))
2975 value_len = OCFS2_XATTR_SIZE(
2976 le64_to_cpu(xe->xe_value_size));
2978 value_len = OCFS2_XATTR_ROOT_SIZE;
2979 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
2982 * We must make sure that the name/value pair
2983 * exist in the same block. So adjust end to
2984 * the previous block end if needed.
2986 if (((end - len) / blocksize !=
2987 (end - 1) / blocksize))
2988 end = end - end % blocksize;
2990 if (end > offset + len) {
2991 memmove(bucket_buf + end - len,
2992 bucket_buf + offset, len);
2993 xe->xe_name_offset = cpu_to_le16(end - len);
2996 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
2997 "bucket %llu\n", (unsigned long long)blkno);
3002 mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
3003 "bucket %llu\n", (unsigned long long)blkno);
3005 if (xh_free_start == end)
3008 memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
3009 xh->xh_free_start = cpu_to_le16(end);
3011 /* sort the entries by their name_hash. */
3012 sort(entries, le16_to_cpu(xh->xh_count),
3013 sizeof(struct ocfs2_xattr_entry),
3017 for (i = 0; i < blk_per_bucket; i++, buf += blocksize) {
3018 memcpy(bhs[i]->b_data, buf, blocksize);
3019 ocfs2_journal_dirty(handle, bhs[i]);
3023 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
3027 for (i = 0; i < blk_per_bucket; i++)
3037 * Move half nums of the xattr bucket in the previous cluster to this new
3038 * cluster. We only touch the last cluster of the previous extend record.
3040 * first_bh is the first buffer_head of a series of bucket in the same
3041 * extent rec and header_bh is the header of one bucket in this cluster.
3042 * They will be updated if we move the data header_bh contains to the new
3043 * cluster. first_hash will be set as the 1st xe's name_hash of the new cluster.
3045 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3047 struct buffer_head **first_bh,
3048 struct buffer_head **header_bh,
3054 int i, ret, credits;
3055 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3056 int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3057 int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3058 int blocksize = inode->i_sb->s_blocksize;
3059 struct buffer_head *old_bh, *new_bh, *prev_bh, *new_first_bh = NULL;
3060 struct ocfs2_xattr_header *new_xh;
3061 struct ocfs2_xattr_header *xh =
3062 (struct ocfs2_xattr_header *)((*first_bh)->b_data);
3064 BUG_ON(le16_to_cpu(xh->xh_num_buckets) < num_buckets);
3065 BUG_ON(OCFS2_XATTR_BUCKET_SIZE == osb->s_clustersize);
3067 prev_bh = *first_bh;
3069 xh = (struct ocfs2_xattr_header *)prev_bh->b_data;
3071 prev_blkno += (num_clusters - 1) * bpc + bpc / 2;
3073 mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3074 prev_blkno, new_blkno);
3077 * We need to update the 1st half of the new cluster and
3078 * 1 more for the update of the 1st bucket of the previous
3081 credits = bpc / 2 + 1;
3082 ret = ocfs2_extend_trans(handle, credits);
3088 ret = ocfs2_journal_access(handle, inode, prev_bh,
3089 OCFS2_JOURNAL_ACCESS_WRITE);
3095 for (i = 0; i < bpc / 2; i++, prev_blkno++, new_blkno++) {
3096 old_bh = new_bh = NULL;
3097 new_bh = sb_getblk(inode->i_sb, new_blkno);
3104 ocfs2_set_new_buffer_uptodate(inode, new_bh);
3106 ret = ocfs2_journal_access(handle, inode, new_bh,
3107 OCFS2_JOURNAL_ACCESS_CREATE);
3114 ret = ocfs2_read_block(inode, prev_blkno, &old_bh);
3121 memcpy(new_bh->b_data, old_bh->b_data, blocksize);
3124 new_xh = (struct ocfs2_xattr_header *)new_bh->b_data;
3125 new_xh->xh_num_buckets = cpu_to_le16(num_buckets / 2);
3128 *first_hash = le32_to_cpu(
3129 new_xh->xh_entries[0].xe_name_hash);
3130 new_first_bh = new_bh;
3131 get_bh(new_first_bh);
3134 ocfs2_journal_dirty(handle, new_bh);
3136 if (*header_bh == old_bh) {
3138 *header_bh = new_bh;
3142 *first_bh = new_first_bh;
3149 le16_add_cpu(&xh->xh_num_buckets, -(num_buckets / 2));
3151 ocfs2_journal_dirty(handle, prev_bh);
3154 brelse(new_first_bh);
3158 static int ocfs2_read_xattr_bucket(struct inode *inode,
3160 struct buffer_head **bhs,
3164 u16 i, blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3167 return ocfs2_read_blocks(inode, blkno,
3168 blk_per_bucket, bhs, 0);
3170 for (i = 0; i < blk_per_bucket; i++) {
3171 bhs[i] = sb_getblk(inode->i_sb, blkno + i);
3172 if (bhs[i] == NULL) {
3177 ocfs2_set_new_buffer_uptodate(inode, bhs[i]);
3184 * Move half num of the xattrs in old bucket(blk) to new bucket(new_blk).
3185 * first_hash will record the 1st hash of the new bucket.
3187 static int ocfs2_half_xattr_bucket(struct inode *inode,
3192 int new_bucket_head)
3195 u16 count, start, len, name_value_len, xe_len, name_offset;
3196 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3197 struct buffer_head **s_bhs, **t_bhs = NULL;
3198 struct ocfs2_xattr_header *xh;
3199 struct ocfs2_xattr_entry *xe;
3200 int blocksize = inode->i_sb->s_blocksize;
3202 mlog(0, "move half of xattrs from bucket %llu to %llu\n",
3205 s_bhs = kcalloc(blk_per_bucket, sizeof(struct buffer_head *), GFP_NOFS);
3209 ret = ocfs2_read_xattr_bucket(inode, blk, s_bhs, 0);
3215 ret = ocfs2_journal_access(handle, inode, s_bhs[0],
3216 OCFS2_JOURNAL_ACCESS_WRITE);
3222 t_bhs = kcalloc(blk_per_bucket, sizeof(struct buffer_head *), GFP_NOFS);
3228 ret = ocfs2_read_xattr_bucket(inode, new_blk, t_bhs, new_bucket_head);
3234 for (i = 0; i < blk_per_bucket; i++) {
3235 ret = ocfs2_journal_access(handle, inode, t_bhs[i],
3236 OCFS2_JOURNAL_ACCESS_CREATE);
3243 /* copy the whole bucket to the new first. */
3244 for (i = 0; i < blk_per_bucket; i++)
3245 memcpy(t_bhs[i]->b_data, s_bhs[i]->b_data, blocksize);
3247 /* update the new bucket. */
3248 xh = (struct ocfs2_xattr_header *)t_bhs[0]->b_data;
3249 count = le16_to_cpu(xh->xh_count);
3253 * Calculate the total name/value len and xh_free_start for
3254 * the old bucket first.
3256 name_offset = OCFS2_XATTR_BUCKET_SIZE;
3258 for (i = 0; i < start; i++) {
3259 xe = &xh->xh_entries[i];
3260 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3261 if (ocfs2_xattr_is_local(xe))
3263 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3265 xe_len += OCFS2_XATTR_ROOT_SIZE;
3266 name_value_len += xe_len;
3267 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3268 name_offset = le16_to_cpu(xe->xe_name_offset);
3272 * Now begin the modification to the new bucket.
3274 * In the new bucket, We just move the xattr entry to the beginning
3275 * and don't touch the name/value. So there will be some holes in the
3276 * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3279 xe = &xh->xh_entries[start];
3280 len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3281 mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3282 (int)((char *)xe - (char *)xh),
3283 (int)((char *)xh->xh_entries - (char *)xh));
3284 memmove((char *)xh->xh_entries, (char *)xe, len);
3285 xe = &xh->xh_entries[count - start];
3286 len = sizeof(struct ocfs2_xattr_entry) * start;
3287 memset((char *)xe, 0, len);
3289 le16_add_cpu(&xh->xh_count, -start);
3290 le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3292 /* Calculate xh_free_start for the new bucket. */
3293 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3294 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3295 xe = &xh->xh_entries[i];
3296 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3297 if (ocfs2_xattr_is_local(xe))
3299 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3301 xe_len += OCFS2_XATTR_ROOT_SIZE;
3302 if (le16_to_cpu(xe->xe_name_offset) <
3303 le16_to_cpu(xh->xh_free_start))
3304 xh->xh_free_start = xe->xe_name_offset;
3307 /* set xh->xh_num_buckets for the new xh. */
3308 if (new_bucket_head)
3309 xh->xh_num_buckets = cpu_to_le16(1);
3311 xh->xh_num_buckets = 0;
3313 for (i = 0; i < blk_per_bucket; i++) {
3314 ocfs2_journal_dirty(handle, t_bhs[i]);
3319 /* store the first_hash of the new bucket. */
3321 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3324 * Now only update the 1st block of the old bucket.
3325 * Please note that the entry has been sorted already above.
3327 xh = (struct ocfs2_xattr_header *)s_bhs[0]->b_data;
3328 memset(&xh->xh_entries[start], 0,
3329 sizeof(struct ocfs2_xattr_entry) * (count - start));
3330 xh->xh_count = cpu_to_le16(start);
3331 xh->xh_free_start = cpu_to_le16(name_offset);
3332 xh->xh_name_value_len = cpu_to_le16(name_value_len);
3334 ocfs2_journal_dirty(handle, s_bhs[0]);
3340 for (i = 0; i < blk_per_bucket; i++)
3346 for (i = 0; i < blk_per_bucket; i++)
3355 * Copy xattr from one bucket to another bucket.
3357 * The caller must make sure that the journal transaction
3358 * has enough space for journaling.
3360 static int ocfs2_cp_xattr_bucket(struct inode *inode,
3367 int blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3368 int blocksize = inode->i_sb->s_blocksize;
3369 struct buffer_head **s_bhs, **t_bhs = NULL;
3371 BUG_ON(s_blkno == t_blkno);
3373 mlog(0, "cp bucket %llu to %llu, target is %d\n",
3374 s_blkno, t_blkno, t_is_new);
3376 s_bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
3381 ret = ocfs2_read_xattr_bucket(inode, s_blkno, s_bhs, 0);
3385 t_bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
3392 ret = ocfs2_read_xattr_bucket(inode, t_blkno, t_bhs, t_is_new);
3396 for (i = 0; i < blk_per_bucket; i++) {
3397 ret = ocfs2_journal_access(handle, inode, t_bhs[i],
3398 OCFS2_JOURNAL_ACCESS_WRITE);
3403 for (i = 0; i < blk_per_bucket; i++) {
3404 memcpy(t_bhs[i]->b_data, s_bhs[i]->b_data, blocksize);
3405 ocfs2_journal_dirty(handle, t_bhs[i]);
3410 for (i = 0; i < blk_per_bucket; i++)
3416 for (i = 0; i < blk_per_bucket; i++)
3425 * Copy one xattr cluster from src_blk to to_blk.
3426 * The to_blk will become the first bucket header of the cluster, so its
3427 * xh_num_buckets will be initialized as the bucket num in the cluster.
3429 static int ocfs2_cp_xattr_cluster(struct inode *inode,
3431 struct buffer_head *first_bh,
3436 int i, ret, credits;
3437 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3438 int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3439 int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3440 struct buffer_head *bh = NULL;
3441 struct ocfs2_xattr_header *xh;
3442 u64 to_blk_start = to_blk;
3444 mlog(0, "cp xattrs from cluster %llu to %llu\n", src_blk, to_blk);
3447 * We need to update the new cluster and 1 more for the update of
3448 * the 1st bucket of the previous extent rec.
3451 ret = ocfs2_extend_trans(handle, credits);
3457 ret = ocfs2_journal_access(handle, inode, first_bh,
3458 OCFS2_JOURNAL_ACCESS_WRITE);
3464 for (i = 0; i < num_buckets; i++) {
3465 ret = ocfs2_cp_xattr_bucket(inode, handle,
3466 src_blk, to_blk, 1);
3472 src_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3473 to_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3476 /* update the old bucket header. */
3477 xh = (struct ocfs2_xattr_header *)first_bh->b_data;
3478 le16_add_cpu(&xh->xh_num_buckets, -num_buckets);
3480 ocfs2_journal_dirty(handle, first_bh);
3482 /* update the new bucket header. */
3483 ret = ocfs2_read_block(inode, to_blk_start, &bh);
3489 ret = ocfs2_journal_access(handle, inode, bh,
3490 OCFS2_JOURNAL_ACCESS_WRITE);
3496 xh = (struct ocfs2_xattr_header *)bh->b_data;
3497 xh->xh_num_buckets = cpu_to_le16(num_buckets);
3499 ocfs2_journal_dirty(handle, bh);
3502 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3509 * Move half of the xattrs in this cluster to the new cluster.
3510 * This function should only be called when bucket size == cluster size.
3511 * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
3513 static int ocfs2_half_xattr_cluster(struct inode *inode,
3519 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3520 int ret, credits = 2 * blk_per_bucket;
3522 BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
3524 ret = ocfs2_extend_trans(handle, credits);
3530 /* Move half of the xattr in start_blk to the next bucket. */
3531 return ocfs2_half_xattr_bucket(inode, handle, prev_blk,
3532 new_blk, first_hash, 1);
3536 * Move some xattrs from the old cluster to the new one since they are not
3537 * contiguous in ocfs2 xattr tree.
3539 * new_blk starts a new separate cluster, and we will move some xattrs from
3540 * prev_blk to it. v_start will be set as the first name hash value in this
3541 * new cluster so that it can be used as e_cpos during tree insertion and
3542 * don't collide with our original b-tree operations. first_bh and header_bh
3543 * will also be updated since they will be used in ocfs2_extend_xattr_bucket
3544 * to extend the insert bucket.
3546 * The problem is how much xattr should we move to the new one and when should
3547 * we update first_bh and header_bh?
3548 * 1. If cluster size > bucket size, that means the previous cluster has more
3549 * than 1 bucket, so just move half nums of bucket into the new cluster and
3550 * update the first_bh and header_bh if the insert bucket has been moved
3551 * to the new cluster.
3552 * 2. If cluster_size == bucket_size:
3553 * a) If the previous extent rec has more than one cluster and the insert
3554 * place isn't in the last cluster, copy the entire last cluster to the
3555 * new one. This time, we don't need to upate the first_bh and header_bh
3556 * since they will not be moved into the new cluster.
3557 * b) Otherwise, move the bottom half of the xattrs in the last cluster into
3558 * the new one. And we set the extend flag to zero if the insert place is
3559 * moved into the new allocated cluster since no extend is needed.
3561 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
3563 struct buffer_head **first_bh,
3564 struct buffer_head **header_bh,
3572 int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3574 mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
3575 prev_blk, prev_clusters, new_blk);
3577 if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1)
3578 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
3587 u64 last_blk = prev_blk + bpc * (prev_clusters - 1);
3589 if (prev_clusters > 1 && (*header_bh)->b_blocknr != last_blk)
3590 ret = ocfs2_cp_xattr_cluster(inode, handle, *first_bh,
3594 ret = ocfs2_half_xattr_cluster(inode, handle,
3598 if ((*header_bh)->b_blocknr == last_blk && extend)
3607 * Add a new cluster for xattr storage.
3609 * If the new cluster is contiguous with the previous one, it will be
3610 * appended to the same extent record, and num_clusters will be updated.
3611 * If not, we will insert a new extent for it and move some xattrs in
3612 * the last cluster into the new allocated one.
3613 * We also need to limit the maximum size of a btree leaf, otherwise we'll
3614 * lose the benefits of hashing because we'll have to search large leaves.
3615 * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
3618 * first_bh is the first block of the previous extent rec and header_bh
3619 * indicates the bucket we will insert the new xattrs. They will be updated
3620 * when the header_bh is moved into the new cluster.
3622 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
3623 struct buffer_head *root_bh,
3624 struct buffer_head **first_bh,
3625 struct buffer_head **header_bh,
3632 u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3633 u32 prev_clusters = *num_clusters;
3634 u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
3636 handle_t *handle = NULL;
3637 struct ocfs2_alloc_context *data_ac = NULL;
3638 struct ocfs2_alloc_context *meta_ac = NULL;
3639 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3640 struct ocfs2_extent_tree et;
3642 mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
3643 "previous xattr blkno = %llu\n",
3644 (unsigned long long)OCFS2_I(inode)->ip_blkno,
3645 prev_cpos, prev_blkno);
3647 ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
3649 ret = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
3650 &data_ac, &meta_ac);
3656 credits = ocfs2_calc_extend_credits(osb->sb, et.et_root_el,
3658 handle = ocfs2_start_trans(osb, credits);
3659 if (IS_ERR(handle)) {
3660 ret = PTR_ERR(handle);
3666 ret = ocfs2_journal_access(handle, inode, root_bh,
3667 OCFS2_JOURNAL_ACCESS_WRITE);
3673 ret = __ocfs2_claim_clusters(osb, handle, data_ac, 1,
3674 clusters_to_add, &bit_off, &num_bits);
3681 BUG_ON(num_bits > clusters_to_add);
3683 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
3684 mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
3685 num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
3687 if (prev_blkno + prev_clusters * bpc == block &&
3688 (prev_clusters + num_bits) << osb->s_clustersize_bits <=
3689 OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
3691 * If this cluster is contiguous with the old one and
3692 * adding this new cluster, we don't surpass the limit of
3693 * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
3694 * initialized and used like other buckets in the previous
3696 * So add it as a contiguous one. The caller will handle
3699 v_start = prev_cpos + prev_clusters;
3700 *num_clusters = prev_clusters + num_bits;
3701 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
3704 ret = ocfs2_adjust_xattr_cross_cluster(inode,
3719 if (handle->h_buffer_credits < credits) {
3721 * The journal has been restarted before, and don't
3722 * have enough space for the insertion, so extend it
3725 ret = ocfs2_extend_trans(handle, credits);
3731 mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
3732 num_bits, block, v_start);
3733 ret = ocfs2_insert_extent(osb, handle, inode, &et, v_start, block,
3734 num_bits, 0, meta_ac);
3740 ret = ocfs2_journal_dirty(handle, root_bh);
3748 ocfs2_commit_trans(osb, handle);
3750 ocfs2_free_alloc_context(data_ac);
3752 ocfs2_free_alloc_context(meta_ac);
3758 * Extend a new xattr bucket and move xattrs to the end one by one until
3759 * We meet with start_bh. Only move half of the xattrs to the bucket after it.
3761 static int ocfs2_extend_xattr_bucket(struct inode *inode,
3762 struct buffer_head *first_bh,
3763 struct buffer_head *start_bh,
3767 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3768 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3769 u64 start_blk = start_bh->b_blocknr, end_blk;
3770 u32 num_buckets = num_clusters * ocfs2_xattr_buckets_per_cluster(osb);
3772 struct ocfs2_xattr_header *first_xh =
3773 (struct ocfs2_xattr_header *)first_bh->b_data;
3774 u16 bucket = le16_to_cpu(first_xh->xh_num_buckets);
3776 mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
3777 "from %llu, len = %u\n", start_blk,
3778 (unsigned long long)first_bh->b_blocknr, num_clusters);
3780 BUG_ON(bucket >= num_buckets);
3782 end_blk = first_bh->b_blocknr + (bucket - 1) * blk_per_bucket;
3785 * We will touch all the buckets after the start_bh(include it).
3786 * Add one more bucket and modify the first_bh.
3788 credits = end_blk - start_blk + 2 * blk_per_bucket + 1;
3789 handle = ocfs2_start_trans(osb, credits);
3790 if (IS_ERR(handle)) {
3791 ret = PTR_ERR(handle);
3797 ret = ocfs2_journal_access(handle, inode, first_bh,
3798 OCFS2_JOURNAL_ACCESS_WRITE);
3804 while (end_blk != start_blk) {
3805 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
3806 end_blk + blk_per_bucket, 0);
3809 end_blk -= blk_per_bucket;
3812 /* Move half of the xattr in start_blk to the next bucket. */
3813 ret = ocfs2_half_xattr_bucket(inode, handle, start_blk,
3814 start_blk + blk_per_bucket, NULL, 0);
3816 le16_add_cpu(&first_xh->xh_num_buckets, 1);
3817 ocfs2_journal_dirty(handle, first_bh);
3820 ocfs2_commit_trans(osb, handle);
3826 * Add new xattr bucket in an extent record and adjust the buckets accordingly.
3827 * xb_bh is the ocfs2_xattr_block.
3828 * We will move all the buckets starting from header_bh to the next place. As
3829 * for this one, half num of its xattrs will be moved to the next one.
3831 * We will allocate a new cluster if current cluster is full and adjust
3832 * header_bh and first_bh if the insert place is moved to the new cluster.
3834 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
3835 struct buffer_head *xb_bh,
3836 struct buffer_head *header_bh)
3838 struct ocfs2_xattr_header *first_xh = NULL;
3839 struct buffer_head *first_bh = NULL;
3840 struct ocfs2_xattr_block *xb =
3841 (struct ocfs2_xattr_block *)xb_bh->b_data;
3842 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3843 struct ocfs2_extent_list *el = &xb_root->xt_list;
3844 struct ocfs2_xattr_header *xh =
3845 (struct ocfs2_xattr_header *)header_bh->b_data;
3846 u32 name_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3847 struct super_block *sb = inode->i_sb;
3848 struct ocfs2_super *osb = OCFS2_SB(sb);
3849 int ret, num_buckets, extend = 1;
3851 u32 e_cpos, num_clusters;
3853 mlog(0, "Add new xattr bucket starting form %llu\n",
3854 (unsigned long long)header_bh->b_blocknr);
3857 * Add refrence for header_bh here because it may be
3858 * changed in ocfs2_add_new_xattr_cluster and we need
3859 * to free it in the end.
3863 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
3870 ret = ocfs2_read_block(inode, p_blkno, &first_bh);
3876 num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
3877 first_xh = (struct ocfs2_xattr_header *)first_bh->b_data;
3879 if (num_buckets == le16_to_cpu(first_xh->xh_num_buckets)) {
3880 ret = ocfs2_add_new_xattr_cluster(inode,
3895 ret = ocfs2_extend_xattr_bucket(inode,
3907 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
3908 struct ocfs2_xattr_bucket *bucket,
3911 int block_off = offs >> inode->i_sb->s_blocksize_bits;
3913 offs = offs % inode->i_sb->s_blocksize;
3914 return bucket->bhs[block_off]->b_data + offs;
3918 * Handle the normal xattr set, including replace, delete and new.
3920 * Note: "local" indicates the real data's locality. So we can't
3921 * just its bucket locality by its length.
3923 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
3924 struct ocfs2_xattr_info *xi,
3925 struct ocfs2_xattr_search *xs,
3929 struct ocfs2_xattr_entry *last, *xe;
3930 int name_len = strlen(xi->name);
3931 struct ocfs2_xattr_header *xh = xs->header;
3932 u16 count = le16_to_cpu(xh->xh_count), start;
3933 size_t blocksize = inode->i_sb->s_blocksize;
3935 size_t offs, size, new_size;
3937 last = &xh->xh_entries[count];
3938 if (!xs->not_found) {
3940 offs = le16_to_cpu(xe->xe_name_offset);
3941 if (ocfs2_xattr_is_local(xe))
3942 size = OCFS2_XATTR_SIZE(name_len) +
3943 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3945 size = OCFS2_XATTR_SIZE(name_len) +
3946 OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
3949 * If the new value will be stored outside, xi->value has been
3950 * initalized as an empty ocfs2_xattr_value_root, and the same
3951 * goes with xi->value_len, so we can set new_size safely here.
3952 * See ocfs2_xattr_set_in_bucket.
3954 new_size = OCFS2_XATTR_SIZE(name_len) +
3955 OCFS2_XATTR_SIZE(xi->value_len);
3957 le16_add_cpu(&xh->xh_name_value_len, -size);
3959 if (new_size > size)
3960 goto set_new_name_value;
3962 /* Now replace the old value with new one. */
3964 xe->xe_value_size = cpu_to_le64(xi->value_len);
3966 xe->xe_value_size = 0;
3968 val = ocfs2_xattr_bucket_get_val(inode,
3970 memset(val + OCFS2_XATTR_SIZE(name_len), 0,
3971 size - OCFS2_XATTR_SIZE(name_len));
3972 if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
3973 memcpy(val + OCFS2_XATTR_SIZE(name_len),
3974 xi->value, xi->value_len);
3976 le16_add_cpu(&xh->xh_name_value_len, new_size);
3977 ocfs2_xattr_set_local(xe, local);
3981 * Remove the old entry if there is more than one.
3982 * We don't remove the last entry so that we can
3983 * use it to indicate the hash value of the empty
3987 le16_add_cpu(&xh->xh_count, -1);
3990 (void *)last - (void *)xe);
3992 sizeof(struct ocfs2_xattr_entry));
3995 cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4000 /* find a new entry for insert. */
4001 int low = 0, high = count - 1, tmp;
4002 struct ocfs2_xattr_entry *tmp_xe;
4004 while (low <= high && count) {
4005 tmp = (low + high) / 2;
4006 tmp_xe = &xh->xh_entries[tmp];
4008 if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
4010 else if (name_hash <
4011 le32_to_cpu(tmp_xe->xe_name_hash))
4019 xe = &xh->xh_entries[low];
4021 memmove(xe + 1, xe, (void *)last - (void *)xe);
4023 le16_add_cpu(&xh->xh_count, 1);
4024 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
4025 xe->xe_name_hash = cpu_to_le32(name_hash);
4026 xe->xe_name_len = name_len;
4027 ocfs2_xattr_set_type(xe, xi->name_index);
4031 /* Insert the new name+value. */
4032 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4035 * We must make sure that the name/value pair
4036 * exists in the same block.
4038 offs = le16_to_cpu(xh->xh_free_start);
4039 start = offs - size;
4041 if (start >> inode->i_sb->s_blocksize_bits !=
4042 (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4043 offs = offs - offs % blocksize;
4044 xh->xh_free_start = cpu_to_le16(offs);
4047 val = ocfs2_xattr_bucket_get_val(inode,
4048 &xs->bucket, offs - size);
4049 xe->xe_name_offset = cpu_to_le16(offs - size);
4051 memset(val, 0, size);
4052 memcpy(val, xi->name, name_len);
4053 memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4055 xe->xe_value_size = cpu_to_le64(xi->value_len);
4056 ocfs2_xattr_set_local(xe, local);
4058 le16_add_cpu(&xh->xh_free_start, -size);
4059 le16_add_cpu(&xh->xh_name_value_len, size);
4064 static int ocfs2_xattr_bucket_handle_journal(struct inode *inode,
4066 struct ocfs2_xattr_search *xs,
4067 struct buffer_head **bhs,
4070 int ret = 0, off, block_off;
4071 struct ocfs2_xattr_entry *xe = xs->here;
4074 * First calculate all the blocks we should journal_access
4075 * and journal_dirty. The first block should always be touched.
4077 ret = ocfs2_journal_dirty(handle, bhs[0]);
4081 /* calc the data. */
4082 off = le16_to_cpu(xe->xe_name_offset);
4083 block_off = off >> inode->i_sb->s_blocksize_bits;
4084 ret = ocfs2_journal_dirty(handle, bhs[block_off]);
4092 * Set the xattr entry in the specified bucket.
4093 * The bucket is indicated by xs->bucket and it should have the enough
4094 * space for the xattr insertion.
4096 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4097 struct ocfs2_xattr_info *xi,
4098 struct ocfs2_xattr_search *xs,
4103 handle_t *handle = NULL;
4104 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4105 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4107 mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4108 (unsigned long)xi->value_len, xi->name_index,
4109 (unsigned long long)xs->bucket.bhs[0]->b_blocknr);
4111 if (!xs->bucket.bhs[1]) {
4112 ret = ocfs2_read_blocks(inode,
4113 xs->bucket.bhs[0]->b_blocknr + 1,
4114 blk_per_bucket - 1, &xs->bucket.bhs[1],
4122 handle = ocfs2_start_trans(osb, blk_per_bucket);
4123 if (IS_ERR(handle)) {
4124 ret = PTR_ERR(handle);
4130 for (i = 0; i < blk_per_bucket; i++) {
4131 ret = ocfs2_journal_access(handle, inode, xs->bucket.bhs[i],
4132 OCFS2_JOURNAL_ACCESS_WRITE);
4139 ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local);
4141 /*Only dirty the blocks we have touched in set xattr. */
4142 ret = ocfs2_xattr_bucket_handle_journal(inode, handle, xs,
4143 xs->bucket.bhs, blk_per_bucket);
4147 ocfs2_commit_trans(osb, handle);
4152 static int ocfs2_xattr_value_update_size(struct inode *inode,
4153 struct buffer_head *xe_bh,
4154 struct ocfs2_xattr_entry *xe,
4158 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4159 handle_t *handle = NULL;
4161 handle = ocfs2_start_trans(osb, 1);
4162 if (handle == NULL) {
4168 ret = ocfs2_journal_access(handle, inode, xe_bh,
4169 OCFS2_JOURNAL_ACCESS_WRITE);
4175 xe->xe_value_size = cpu_to_le64(new_size);
4177 ret = ocfs2_journal_dirty(handle, xe_bh);
4182 ocfs2_commit_trans(osb, handle);
4188 * Truncate the specified xe_off entry in xattr bucket.
4189 * bucket is indicated by header_bh and len is the new length.
4190 * Both the ocfs2_xattr_value_root and the entry will be updated here.
4192 * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4194 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4195 struct buffer_head *header_bh,
4201 struct buffer_head *value_bh = NULL;
4202 struct ocfs2_xattr_value_root *xv;
4203 struct ocfs2_xattr_entry *xe;
4204 struct ocfs2_xattr_header *xh =
4205 (struct ocfs2_xattr_header *)header_bh->b_data;
4206 size_t blocksize = inode->i_sb->s_blocksize;
4208 xe = &xh->xh_entries[xe_off];
4210 BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4212 offset = le16_to_cpu(xe->xe_name_offset) +
4213 OCFS2_XATTR_SIZE(xe->xe_name_len);
4215 value_blk = offset / blocksize;
4217 /* We don't allow ocfs2_xattr_value to be stored in different block. */
4218 BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4219 value_blk += header_bh->b_blocknr;
4221 ret = ocfs2_read_block(inode, value_blk, &value_bh);
4227 xv = (struct ocfs2_xattr_value_root *)
4228 (value_bh->b_data + offset % blocksize);
4230 mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4231 xe_off, (unsigned long long)header_bh->b_blocknr, len);
4232 ret = ocfs2_xattr_value_truncate(inode, value_bh, xv, len);
4238 ret = ocfs2_xattr_value_update_size(inode, header_bh, xe, len);
4249 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4250 struct ocfs2_xattr_search *xs,
4254 struct ocfs2_xattr_entry *xe = xs->here;
4255 struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4257 BUG_ON(!xs->bucket.bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4259 offset = xe - xh->xh_entries;
4260 ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket.bhs[0],
4268 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4269 struct ocfs2_xattr_search *xs,
4274 struct ocfs2_xattr_value_root *xv;
4275 struct ocfs2_xattr_entry *xe = xs->here;
4277 BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4279 offset = le16_to_cpu(xe->xe_name_offset) +
4280 OCFS2_XATTR_SIZE(xe->xe_name_len);
4282 xv = (struct ocfs2_xattr_value_root *)(xs->base + offset);
4284 return __ocfs2_xattr_set_value_outside(inode, xv, val, value_len);
4287 static int ocfs2_rm_xattr_cluster(struct inode *inode,
4288 struct buffer_head *root_bh,
4294 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4295 struct inode *tl_inode = osb->osb_tl_inode;
4297 struct ocfs2_xattr_block *xb =
4298 (struct ocfs2_xattr_block *)root_bh->b_data;
4299 struct ocfs2_alloc_context *meta_ac = NULL;
4300 struct ocfs2_cached_dealloc_ctxt dealloc;
4301 struct ocfs2_extent_tree et;
4303 ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4305 ocfs2_init_dealloc_ctxt(&dealloc);
4307 mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4308 cpos, len, (unsigned long long)blkno);
4310 ocfs2_remove_xattr_clusters_from_cache(inode, blkno, len);
4312 ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
4318 mutex_lock(&tl_inode->i_mutex);
4320 if (ocfs2_truncate_log_needs_flush(osb)) {
4321 ret = __ocfs2_flush_truncate_log(osb);
4328 handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
4329 if (handle == NULL) {
4335 ret = ocfs2_journal_access(handle, inode, root_bh,
4336 OCFS2_JOURNAL_ACCESS_WRITE);
4342 ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
4349 le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4351 ret = ocfs2_journal_dirty(handle, root_bh);
4357 ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4362 ocfs2_commit_trans(osb, handle);
4364 ocfs2_schedule_truncate_log_flush(osb, 1);
4366 mutex_unlock(&tl_inode->i_mutex);
4369 ocfs2_free_alloc_context(meta_ac);
4371 ocfs2_run_deallocs(osb, &dealloc);
4376 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4377 struct ocfs2_xattr_search *xs)
4379 handle_t *handle = NULL;
4380 struct ocfs2_xattr_header *xh = xs->bucket.xh;
4381 struct ocfs2_xattr_entry *last = &xh->xh_entries[
4382 le16_to_cpu(xh->xh_count) - 1];
4385 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), 1);
4386 if (IS_ERR(handle)) {
4387 ret = PTR_ERR(handle);
4392 ret = ocfs2_journal_access(handle, inode, xs->bucket.bhs[0],
4393 OCFS2_JOURNAL_ACCESS_WRITE);
4399 /* Remove the old entry. */
4400 memmove(xs->here, xs->here + 1,
4401 (void *)last - (void *)xs->here);
4402 memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4403 le16_add_cpu(&xh->xh_count, -1);
4405 ret = ocfs2_journal_dirty(handle, xs->bucket.bhs[0]);
4409 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
4413 * Set the xattr name/value in the bucket specified in xs.
4415 * As the new value in xi may be stored in the bucket or in an outside cluster,
4416 * we divide the whole process into 3 steps:
4417 * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4418 * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4419 * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4420 * 4. If the clusters for the new outside value can't be allocated, we need
4421 * to free the xattr we allocated in set.
4423 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4424 struct ocfs2_xattr_info *xi,
4425 struct ocfs2_xattr_search *xs)
4429 char *val = (char *)xi->value;
4430 struct ocfs2_xattr_entry *xe = xs->here;
4431 u32 name_hash = ocfs2_xattr_name_hash(inode, xi->name,
4434 if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
4436 * We need to truncate the xattr storage first.
4438 * If both the old and new value are stored to
4439 * outside block, we only need to truncate
4440 * the storage and then set the value outside.
4442 * If the new value should be stored within block,
4443 * we should free all the outside block first and
4444 * the modification to the xattr block will be done
4445 * by following steps.
4447 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4448 value_len = xi->value_len;
4452 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4458 goto set_value_outside;
4461 value_len = xi->value_len;
4462 /* So we have to handle the inside block change now. */
4463 if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4465 * If the new value will be stored outside of block,
4466 * initalize a new empty value root and insert it first.
4469 xi->value = &def_xv;
4470 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
4473 ret = ocfs2_xattr_set_entry_in_bucket(inode, xi, xs, name_hash, local);
4479 if (value_len <= OCFS2_XATTR_INLINE_SIZE)
4482 /* allocate the space now for the outside block storage. */
4483 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4488 if (xs->not_found) {
4490 * We can't allocate enough clusters for outside
4491 * storage and we have allocated xattr already,
4492 * so need to remove it.
4494 ocfs2_xattr_bucket_remove_xs(inode, xs);
4500 ret = ocfs2_xattr_bucket_set_value_outside(inode, xs, val, value_len);
4505 /* check whether the xattr bucket is filled up with the same hash value. */
4506 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
4507 struct ocfs2_xattr_bucket *bucket)
4509 struct ocfs2_xattr_header *xh = bucket->xh;
4511 if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
4512 xh->xh_entries[0].xe_name_hash) {
4513 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
4515 (unsigned long long)bucket->bhs[0]->b_blocknr,
4516 le32_to_cpu(xh->xh_entries[0].xe_name_hash));
4523 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
4524 struct ocfs2_xattr_info *xi,
4525 struct ocfs2_xattr_search *xs)
4527 struct ocfs2_xattr_header *xh;
4528 struct ocfs2_xattr_entry *xe;
4529 u16 count, header_size, xh_free_start;
4530 int i, free, max_free, need, old;
4531 size_t value_size = 0, name_len = strlen(xi->name);
4532 size_t blocksize = inode->i_sb->s_blocksize;
4533 int ret, allocation = 0;
4534 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4536 mlog_entry("Set xattr %s in xattr index block\n", xi->name);
4540 count = le16_to_cpu(xh->xh_count);
4541 xh_free_start = le16_to_cpu(xh->xh_free_start);
4542 header_size = sizeof(struct ocfs2_xattr_header) +
4543 count * sizeof(struct ocfs2_xattr_entry);
4544 max_free = OCFS2_XATTR_BUCKET_SIZE -
4545 le16_to_cpu(xh->xh_name_value_len) - header_size;
4547 mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
4548 "of %u which exceed block size\n",
4549 (unsigned long long)xs->bucket.bhs[0]->b_blocknr,
4552 if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4553 value_size = OCFS2_XATTR_ROOT_SIZE;
4555 value_size = OCFS2_XATTR_SIZE(xi->value_len);
4558 need = sizeof(struct ocfs2_xattr_entry) +
4559 OCFS2_XATTR_SIZE(name_len) + value_size;
4561 need = value_size + OCFS2_XATTR_SIZE(name_len);
4564 * We only replace the old value if the new length is smaller
4565 * than the old one. Otherwise we will allocate new space in the
4566 * bucket to store it.
4569 if (ocfs2_xattr_is_local(xe))
4570 old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4572 old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4574 if (old >= value_size)
4578 free = xh_free_start - header_size;
4580 * We need to make sure the new name/value pair
4581 * can exist in the same block.
4583 if (xh_free_start % blocksize < need)
4584 free -= xh_free_start % blocksize;
4586 mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
4587 "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
4588 " %u\n", xs->not_found,
4589 (unsigned long long)xs->bucket.bhs[0]->b_blocknr,
4590 free, need, max_free, le16_to_cpu(xh->xh_free_start),
4591 le16_to_cpu(xh->xh_name_value_len));
4593 if (free < need || count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
4594 if (need <= max_free &&
4595 count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
4597 * We can create the space by defragment. Since only the
4598 * name/value will be moved, the xe shouldn't be changed
4601 ret = ocfs2_defrag_xattr_bucket(inode, &xs->bucket);
4607 xh_free_start = le16_to_cpu(xh->xh_free_start);
4608 free = xh_free_start - header_size;
4609 if (xh_free_start % blocksize < need)
4610 free -= xh_free_start % blocksize;
4615 mlog(0, "Can't get enough space for xattr insert by "
4616 "defragment. Need %u bytes, but we have %d, so "
4617 "allocate new bucket for it.\n", need, free);
4621 * We have to add new buckets or clusters and one
4622 * allocation should leave us enough space for insert.
4627 * We do not allow for overlapping ranges between buckets. And
4628 * the maximum number of collisions we will allow for then is
4629 * one bucket's worth, so check it here whether we need to
4630 * add a new bucket for the insert.
4632 ret = ocfs2_check_xattr_bucket_collision(inode, &xs->bucket);
4638 ret = ocfs2_add_new_xattr_bucket(inode,
4646 for (i = 0; i < blk_per_bucket; i++)
4647 brelse(xs->bucket.bhs[i]);
4649 memset(&xs->bucket, 0, sizeof(xs->bucket));
4651 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
4654 if (ret && ret != -ENODATA)
4656 xs->not_found = ret;
4662 ret = ocfs2_xattr_set_in_bucket(inode, xi, xs);
4668 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
4669 struct ocfs2_xattr_bucket *bucket,
4673 struct ocfs2_xattr_header *xh = bucket->xh;
4675 struct ocfs2_xattr_entry *xe;
4677 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4678 xe = &xh->xh_entries[i];
4679 if (ocfs2_xattr_is_local(xe))
4682 ret = ocfs2_xattr_bucket_value_truncate(inode,
4694 static int ocfs2_delete_xattr_index_block(struct inode *inode,
4695 struct buffer_head *xb_bh)
4697 struct ocfs2_xattr_block *xb =
4698 (struct ocfs2_xattr_block *)xb_bh->b_data;
4699 struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4701 u32 name_hash = UINT_MAX, e_cpos, num_clusters;
4704 if (le16_to_cpu(el->l_next_free_rec) == 0)
4707 while (name_hash > 0) {
4708 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4709 &e_cpos, &num_clusters, el);
4715 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
4716 ocfs2_delete_xattr_in_bucket,
4723 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
4724 p_blkno, e_cpos, num_clusters);
4733 name_hash = e_cpos - 1;
4741 * 'trusted' attributes support
4744 #define XATTR_TRUSTED_PREFIX "trusted."
4746 static size_t ocfs2_xattr_trusted_list(struct inode *inode, char *list,
4747 size_t list_size, const char *name,
4750 const size_t prefix_len = sizeof(XATTR_TRUSTED_PREFIX) - 1;
4751 const size_t total_len = prefix_len + name_len + 1;
4753 if (list && total_len <= list_size) {
4754 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
4755 memcpy(list + prefix_len, name, name_len);
4756 list[prefix_len + name_len] = '\0';
4761 static int ocfs2_xattr_trusted_get(struct inode *inode, const char *name,
4762 void *buffer, size_t size)
4764 if (strcmp(name, "") == 0)
4766 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED, name,
4770 static int ocfs2_xattr_trusted_set(struct inode *inode, const char *name,
4771 const void *value, size_t size, int flags)
4773 if (strcmp(name, "") == 0)
4776 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED, name, value,
4780 struct xattr_handler ocfs2_xattr_trusted_handler = {
4781 .prefix = XATTR_TRUSTED_PREFIX,
4782 .list = ocfs2_xattr_trusted_list,
4783 .get = ocfs2_xattr_trusted_get,
4784 .set = ocfs2_xattr_trusted_set,
4789 * 'user' attributes support
4792 #define XATTR_USER_PREFIX "user."
4794 static size_t ocfs2_xattr_user_list(struct inode *inode, char *list,
4795 size_t list_size, const char *name,
4798 const size_t prefix_len = sizeof(XATTR_USER_PREFIX) - 1;
4799 const size_t total_len = prefix_len + name_len + 1;
4800 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4802 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
4805 if (list && total_len <= list_size) {
4806 memcpy(list, XATTR_USER_PREFIX, prefix_len);
4807 memcpy(list + prefix_len, name, name_len);
4808 list[prefix_len + name_len] = '\0';
4813 static int ocfs2_xattr_user_get(struct inode *inode, const char *name,
4814 void *buffer, size_t size)
4816 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4818 if (strcmp(name, "") == 0)
4820 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
4822 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
4826 static int ocfs2_xattr_user_set(struct inode *inode, const char *name,
4827 const void *value, size_t size, int flags)
4829 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4831 if (strcmp(name, "") == 0)
4833 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
4836 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER, name, value,
4840 struct xattr_handler ocfs2_xattr_user_handler = {
4841 .prefix = XATTR_USER_PREFIX,
4842 .list = ocfs2_xattr_user_list,
4843 .get = ocfs2_xattr_user_get,
4844 .set = ocfs2_xattr_user_set,