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>
38 #include <linux/security.h>
40 #define MLOG_MASK_PREFIX ML_XATTR
41 #include <cluster/masklog.h>
45 #include "blockcheck.h"
55 #include "buffer_head_io.h"
60 struct ocfs2_xattr_def_value_root {
61 struct ocfs2_xattr_value_root xv;
62 struct ocfs2_extent_rec er;
65 struct ocfs2_xattr_bucket {
66 /* The inode these xattrs are associated with */
67 struct inode *bu_inode;
69 /* The actual buffers that make up the bucket */
70 struct buffer_head *bu_bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
72 /* How many blocks make up one bucket for this filesystem */
76 struct ocfs2_xattr_set_ctxt {
78 struct ocfs2_alloc_context *meta_ac;
79 struct ocfs2_alloc_context *data_ac;
80 struct ocfs2_cached_dealloc_ctxt dealloc;
83 #define OCFS2_XATTR_ROOT_SIZE (sizeof(struct ocfs2_xattr_def_value_root))
84 #define OCFS2_XATTR_INLINE_SIZE 80
85 #define OCFS2_XATTR_HEADER_GAP 4
86 #define OCFS2_XATTR_FREE_IN_IBODY (OCFS2_MIN_XATTR_INLINE_SIZE \
87 - sizeof(struct ocfs2_xattr_header) \
88 - OCFS2_XATTR_HEADER_GAP)
89 #define OCFS2_XATTR_FREE_IN_BLOCK(ptr) ((ptr)->i_sb->s_blocksize \
90 - sizeof(struct ocfs2_xattr_block) \
91 - sizeof(struct ocfs2_xattr_header) \
92 - OCFS2_XATTR_HEADER_GAP)
94 static struct ocfs2_xattr_def_value_root def_xv = {
95 .xv.xr_list.l_count = cpu_to_le16(1),
98 struct xattr_handler *ocfs2_xattr_handlers[] = {
99 &ocfs2_xattr_user_handler,
100 #ifdef CONFIG_OCFS2_FS_POSIX_ACL
101 &ocfs2_xattr_acl_access_handler,
102 &ocfs2_xattr_acl_default_handler,
104 &ocfs2_xattr_trusted_handler,
105 &ocfs2_xattr_security_handler,
109 static struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
110 [OCFS2_XATTR_INDEX_USER] = &ocfs2_xattr_user_handler,
111 #ifdef CONFIG_OCFS2_FS_POSIX_ACL
112 [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS]
113 = &ocfs2_xattr_acl_access_handler,
114 [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT]
115 = &ocfs2_xattr_acl_default_handler,
117 [OCFS2_XATTR_INDEX_TRUSTED] = &ocfs2_xattr_trusted_handler,
118 [OCFS2_XATTR_INDEX_SECURITY] = &ocfs2_xattr_security_handler,
121 struct ocfs2_xattr_info {
128 struct ocfs2_xattr_search {
129 struct buffer_head *inode_bh;
131 * xattr_bh point to the block buffer head which has extended attribute
132 * when extended attribute in inode, xattr_bh is equal to inode_bh.
134 struct buffer_head *xattr_bh;
135 struct ocfs2_xattr_header *header;
136 struct ocfs2_xattr_bucket *bucket;
139 struct ocfs2_xattr_entry *here;
143 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
144 struct ocfs2_xattr_header *xh,
149 static int ocfs2_xattr_block_find(struct inode *inode,
152 struct ocfs2_xattr_search *xs);
153 static int ocfs2_xattr_index_block_find(struct inode *inode,
154 struct buffer_head *root_bh,
157 struct ocfs2_xattr_search *xs);
159 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
160 struct ocfs2_xattr_tree_root *xt,
164 static int ocfs2_xattr_create_index_block(struct inode *inode,
165 struct ocfs2_xattr_search *xs,
166 struct ocfs2_xattr_set_ctxt *ctxt);
168 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
169 struct ocfs2_xattr_info *xi,
170 struct ocfs2_xattr_search *xs,
171 struct ocfs2_xattr_set_ctxt *ctxt);
173 static int ocfs2_delete_xattr_index_block(struct inode *inode,
174 struct buffer_head *xb_bh);
175 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
176 u64 src_blk, u64 last_blk, u64 to_blk,
177 unsigned int start_bucket,
180 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
182 return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
185 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
187 return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
190 static inline u16 ocfs2_xattr_max_xe_in_bucket(struct super_block *sb)
192 u16 len = sb->s_blocksize -
193 offsetof(struct ocfs2_xattr_header, xh_entries);
195 return len / sizeof(struct ocfs2_xattr_entry);
198 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
199 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
200 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
202 static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
204 struct ocfs2_xattr_bucket *bucket;
205 int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
207 BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
209 bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
211 bucket->bu_inode = inode;
212 bucket->bu_blocks = blks;
218 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
222 for (i = 0; i < bucket->bu_blocks; i++) {
223 brelse(bucket->bu_bhs[i]);
224 bucket->bu_bhs[i] = NULL;
228 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
231 ocfs2_xattr_bucket_relse(bucket);
232 bucket->bu_inode = NULL;
238 * A bucket that has never been written to disk doesn't need to be
239 * read. We just need the buffer_heads. Don't call this for
240 * buckets that are already on disk. ocfs2_read_xattr_bucket() initializes
243 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
248 for (i = 0; i < bucket->bu_blocks; i++) {
249 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
251 if (!bucket->bu_bhs[i]) {
257 if (!ocfs2_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
259 ocfs2_set_new_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
264 ocfs2_xattr_bucket_relse(bucket);
268 /* Read the xattr bucket at xb_blkno */
269 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
274 rc = ocfs2_read_blocks(INODE_CACHE(bucket->bu_inode), xb_blkno,
275 bucket->bu_blocks, bucket->bu_bhs, 0,
278 spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
279 rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
282 &bucket_xh(bucket)->xh_check);
283 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
289 ocfs2_xattr_bucket_relse(bucket);
293 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
294 struct ocfs2_xattr_bucket *bucket,
299 for (i = 0; i < bucket->bu_blocks; i++) {
300 rc = ocfs2_journal_access(handle,
301 INODE_CACHE(bucket->bu_inode),
302 bucket->bu_bhs[i], type);
312 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
313 struct ocfs2_xattr_bucket *bucket)
317 spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
318 ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
319 bucket->bu_bhs, bucket->bu_blocks,
320 &bucket_xh(bucket)->xh_check);
321 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
323 for (i = 0; i < bucket->bu_blocks; i++)
324 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
327 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
328 struct ocfs2_xattr_bucket *src)
331 int blocksize = src->bu_inode->i_sb->s_blocksize;
333 BUG_ON(dest->bu_blocks != src->bu_blocks);
334 BUG_ON(dest->bu_inode != src->bu_inode);
336 for (i = 0; i < src->bu_blocks; i++) {
337 memcpy(bucket_block(dest, i), bucket_block(src, i),
342 static int ocfs2_validate_xattr_block(struct super_block *sb,
343 struct buffer_head *bh)
346 struct ocfs2_xattr_block *xb =
347 (struct ocfs2_xattr_block *)bh->b_data;
349 mlog(0, "Validating xattr block %llu\n",
350 (unsigned long long)bh->b_blocknr);
352 BUG_ON(!buffer_uptodate(bh));
355 * If the ecc fails, we return the error but otherwise
356 * leave the filesystem running. We know any error is
357 * local to this block.
359 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
364 * Errors after here are fatal
367 if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
369 "Extended attribute block #%llu has bad "
371 (unsigned long long)bh->b_blocknr, 7,
376 if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
378 "Extended attribute block #%llu has an "
379 "invalid xb_blkno of %llu",
380 (unsigned long long)bh->b_blocknr,
381 (unsigned long long)le64_to_cpu(xb->xb_blkno));
385 if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
387 "Extended attribute block #%llu has an invalid "
388 "xb_fs_generation of #%u",
389 (unsigned long long)bh->b_blocknr,
390 le32_to_cpu(xb->xb_fs_generation));
397 static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
398 struct buffer_head **bh)
401 struct buffer_head *tmp = *bh;
403 rc = ocfs2_read_block(INODE_CACHE(inode), xb_blkno, &tmp,
404 ocfs2_validate_xattr_block);
406 /* If ocfs2_read_block() got us a new bh, pass it up. */
413 static inline const char *ocfs2_xattr_prefix(int name_index)
415 struct xattr_handler *handler = NULL;
417 if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
418 handler = ocfs2_xattr_handler_map[name_index];
420 return handler ? handler->prefix : NULL;
423 static u32 ocfs2_xattr_name_hash(struct inode *inode,
427 /* Get hash value of uuid from super block */
428 u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
431 /* hash extended attribute name */
432 for (i = 0; i < name_len; i++) {
433 hash = (hash << OCFS2_HASH_SHIFT) ^
434 (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
442 * ocfs2_xattr_hash_entry()
444 * Compute the hash of an extended attribute.
446 static void ocfs2_xattr_hash_entry(struct inode *inode,
447 struct ocfs2_xattr_header *header,
448 struct ocfs2_xattr_entry *entry)
451 char *name = (char *)header + le16_to_cpu(entry->xe_name_offset);
453 hash = ocfs2_xattr_name_hash(inode, name, entry->xe_name_len);
454 entry->xe_name_hash = cpu_to_le32(hash);
459 static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
463 if (value_len <= OCFS2_XATTR_INLINE_SIZE)
464 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
466 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
467 size += sizeof(struct ocfs2_xattr_entry);
472 int ocfs2_calc_security_init(struct inode *dir,
473 struct ocfs2_security_xattr_info *si,
476 struct ocfs2_alloc_context **xattr_ac)
479 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
480 int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
484 * The max space of security xattr taken inline is
485 * 256(name) + 80(value) + 16(entry) = 352 bytes,
486 * So reserve one metadata block for it is ok.
488 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
489 s_size > OCFS2_XATTR_FREE_IN_IBODY) {
490 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
495 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
498 /* reserve clusters for xattr value which will be set in B tree*/
499 if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
500 int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
503 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
505 *want_clusters += new_clusters;
510 int ocfs2_calc_xattr_init(struct inode *dir,
511 struct buffer_head *dir_bh,
513 struct ocfs2_security_xattr_info *si,
519 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
520 int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
523 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
526 if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
527 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
528 OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
531 a_size = ocfs2_xattr_entry_real_size(0, acl_len);
534 } else if (acl_len != 0 && acl_len != -ENODATA) {
540 if (!(s_size + a_size))
544 * The max space of security xattr taken inline is
545 * 256(name) + 80(value) + 16(entry) = 352 bytes,
546 * The max space of acl xattr taken inline is
547 * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
548 * when blocksize = 512, may reserve one more cluser for
549 * xattr bucket, otherwise reserve one metadata block
551 * If this is a new directory with inline data,
552 * we choose to reserve the entire inline area for
553 * directory contents and force an external xattr block.
555 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
556 (S_ISDIR(mode) && ocfs2_supports_inline_data(osb)) ||
557 (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
558 *want_meta = *want_meta + 1;
559 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
562 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
563 (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
565 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
569 * reserve credits and clusters for xattrs which has large value
570 * and have to be set outside
572 if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
573 new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
575 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
577 *want_clusters += new_clusters;
579 if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
580 acl_len > OCFS2_XATTR_INLINE_SIZE) {
581 /* for directory, it has DEFAULT and ACCESS two types of acls */
582 new_clusters = (S_ISDIR(mode) ? 2 : 1) *
583 ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
584 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
586 *want_clusters += new_clusters;
592 static int ocfs2_xattr_extend_allocation(struct inode *inode,
594 struct ocfs2_xattr_value_buf *vb,
595 struct ocfs2_xattr_set_ctxt *ctxt)
598 handle_t *handle = ctxt->handle;
599 enum ocfs2_alloc_restarted why;
600 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
601 u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
602 struct ocfs2_extent_tree et;
604 mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
606 ocfs2_init_xattr_value_extent_tree(&et, inode, vb);
608 status = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
609 OCFS2_JOURNAL_ACCESS_WRITE);
615 prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
616 status = ocfs2_add_clusters_in_btree(osb,
631 status = ocfs2_journal_dirty(handle, vb->vb_bh);
637 clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) - prev_clusters;
640 * We should have already allocated enough space before the transaction,
641 * so no need to restart.
643 BUG_ON(why != RESTART_NONE || clusters_to_add);
650 static int __ocfs2_remove_xattr_range(struct inode *inode,
651 struct ocfs2_xattr_value_buf *vb,
652 u32 cpos, u32 phys_cpos, u32 len,
653 struct ocfs2_xattr_set_ctxt *ctxt)
656 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
657 handle_t *handle = ctxt->handle;
658 struct ocfs2_extent_tree et;
660 ocfs2_init_xattr_value_extent_tree(&et, inode, vb);
662 ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
663 OCFS2_JOURNAL_ACCESS_WRITE);
669 ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, ctxt->meta_ac,
676 le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
678 ret = ocfs2_journal_dirty(handle, vb->vb_bh);
684 ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc, phys_blkno, len);
692 static int ocfs2_xattr_shrink_size(struct inode *inode,
695 struct ocfs2_xattr_value_buf *vb,
696 struct ocfs2_xattr_set_ctxt *ctxt)
699 u32 trunc_len, cpos, phys_cpos, alloc_size;
702 if (old_clusters <= new_clusters)
706 trunc_len = old_clusters - new_clusters;
708 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
710 &vb->vb_xv->xr_list);
716 if (alloc_size > trunc_len)
717 alloc_size = trunc_len;
719 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
720 phys_cpos, alloc_size,
727 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
728 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode),
731 trunc_len -= alloc_size;
738 static int ocfs2_xattr_value_truncate(struct inode *inode,
739 struct ocfs2_xattr_value_buf *vb,
741 struct ocfs2_xattr_set_ctxt *ctxt)
744 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
745 u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
747 if (new_clusters == old_clusters)
750 if (new_clusters > old_clusters)
751 ret = ocfs2_xattr_extend_allocation(inode,
752 new_clusters - old_clusters,
755 ret = ocfs2_xattr_shrink_size(inode,
756 old_clusters, new_clusters,
762 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
763 size_t *result, const char *prefix,
764 const char *name, int name_len)
766 char *p = buffer + *result;
767 int prefix_len = strlen(prefix);
768 int total_len = prefix_len + name_len + 1;
770 *result += total_len;
772 /* we are just looking for how big our buffer needs to be */
779 memcpy(p, prefix, prefix_len);
780 memcpy(p + prefix_len, name, name_len);
781 p[prefix_len + name_len] = '\0';
786 static int ocfs2_xattr_list_entries(struct inode *inode,
787 struct ocfs2_xattr_header *header,
788 char *buffer, size_t buffer_size)
792 const char *prefix, *name;
794 for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
795 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
796 type = ocfs2_xattr_get_type(entry);
797 prefix = ocfs2_xattr_prefix(type);
800 name = (const char *)header +
801 le16_to_cpu(entry->xe_name_offset);
803 ret = ocfs2_xattr_list_entry(buffer, buffer_size,
804 &result, prefix, name,
814 static int ocfs2_xattr_ibody_list(struct inode *inode,
815 struct ocfs2_dinode *di,
819 struct ocfs2_xattr_header *header = NULL;
820 struct ocfs2_inode_info *oi = OCFS2_I(inode);
823 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
826 header = (struct ocfs2_xattr_header *)
827 ((void *)di + inode->i_sb->s_blocksize -
828 le16_to_cpu(di->i_xattr_inline_size));
830 ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
835 static int ocfs2_xattr_block_list(struct inode *inode,
836 struct ocfs2_dinode *di,
840 struct buffer_head *blk_bh = NULL;
841 struct ocfs2_xattr_block *xb;
844 if (!di->i_xattr_loc)
847 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
854 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
855 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
856 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
857 ret = ocfs2_xattr_list_entries(inode, header,
858 buffer, buffer_size);
860 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
861 ret = ocfs2_xattr_tree_list_index_block(inode, xt,
862 buffer, buffer_size);
870 ssize_t ocfs2_listxattr(struct dentry *dentry,
874 int ret = 0, i_ret = 0, b_ret = 0;
875 struct buffer_head *di_bh = NULL;
876 struct ocfs2_dinode *di = NULL;
877 struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
879 if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
882 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
885 ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
891 di = (struct ocfs2_dinode *)di_bh->b_data;
893 down_read(&oi->ip_xattr_sem);
894 i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
902 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
907 up_read(&oi->ip_xattr_sem);
908 ocfs2_inode_unlock(dentry->d_inode, 0);
912 return i_ret + b_ret;
915 static int ocfs2_xattr_find_entry(int name_index,
917 struct ocfs2_xattr_search *xs)
919 struct ocfs2_xattr_entry *entry;
926 name_len = strlen(name);
928 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
929 cmp = name_index - ocfs2_xattr_get_type(entry);
931 cmp = name_len - entry->xe_name_len;
933 cmp = memcmp(name, (xs->base +
934 le16_to_cpu(entry->xe_name_offset)),
942 return cmp ? -ENODATA : 0;
945 static int ocfs2_xattr_get_value_outside(struct inode *inode,
946 struct ocfs2_xattr_value_root *xv,
950 u32 cpos, p_cluster, num_clusters, bpc, clusters;
953 size_t cplen, blocksize;
954 struct buffer_head *bh = NULL;
955 struct ocfs2_extent_list *el;
958 clusters = le32_to_cpu(xv->xr_clusters);
959 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
960 blocksize = inode->i_sb->s_blocksize;
963 while (cpos < clusters) {
964 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
971 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
972 /* Copy ocfs2_xattr_value */
973 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
974 ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
981 cplen = len >= blocksize ? blocksize : len;
982 memcpy(buffer, bh->b_data, cplen);
991 cpos += num_clusters;
997 static int ocfs2_xattr_ibody_get(struct inode *inode,
1002 struct ocfs2_xattr_search *xs)
1004 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1005 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1006 struct ocfs2_xattr_value_root *xv;
1010 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1013 xs->end = (void *)di + inode->i_sb->s_blocksize;
1014 xs->header = (struct ocfs2_xattr_header *)
1015 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1016 xs->base = (void *)xs->header;
1017 xs->here = xs->header->xh_entries;
1019 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1022 size = le64_to_cpu(xs->here->xe_value_size);
1024 if (size > buffer_size)
1026 if (ocfs2_xattr_is_local(xs->here)) {
1027 memcpy(buffer, (void *)xs->base +
1028 le16_to_cpu(xs->here->xe_name_offset) +
1029 OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1031 xv = (struct ocfs2_xattr_value_root *)
1032 (xs->base + le16_to_cpu(
1033 xs->here->xe_name_offset) +
1034 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1035 ret = ocfs2_xattr_get_value_outside(inode, xv,
1047 static int ocfs2_xattr_block_get(struct inode *inode,
1052 struct ocfs2_xattr_search *xs)
1054 struct ocfs2_xattr_block *xb;
1055 struct ocfs2_xattr_value_root *xv;
1057 int ret = -ENODATA, name_offset, name_len, i;
1058 int uninitialized_var(block_off);
1060 xs->bucket = ocfs2_xattr_bucket_new(inode);
1067 ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1073 if (xs->not_found) {
1078 xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1079 size = le64_to_cpu(xs->here->xe_value_size);
1082 if (size > buffer_size)
1085 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1086 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1087 i = xs->here - xs->header->xh_entries;
1089 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1090 ret = ocfs2_xattr_bucket_get_name_value(inode,
1091 bucket_xh(xs->bucket),
1095 xs->base = bucket_block(xs->bucket, block_off);
1097 if (ocfs2_xattr_is_local(xs->here)) {
1098 memcpy(buffer, (void *)xs->base +
1099 name_offset + name_len, size);
1101 xv = (struct ocfs2_xattr_value_root *)
1102 (xs->base + name_offset + name_len);
1103 ret = ocfs2_xattr_get_value_outside(inode, xv,
1113 ocfs2_xattr_bucket_free(xs->bucket);
1115 brelse(xs->xattr_bh);
1116 xs->xattr_bh = NULL;
1120 int ocfs2_xattr_get_nolock(struct inode *inode,
1121 struct buffer_head *di_bh,
1128 struct ocfs2_dinode *di = NULL;
1129 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1130 struct ocfs2_xattr_search xis = {
1131 .not_found = -ENODATA,
1133 struct ocfs2_xattr_search xbs = {
1134 .not_found = -ENODATA,
1137 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1140 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1143 xis.inode_bh = xbs.inode_bh = di_bh;
1144 di = (struct ocfs2_dinode *)di_bh->b_data;
1146 down_read(&oi->ip_xattr_sem);
1147 ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1149 if (ret == -ENODATA && di->i_xattr_loc)
1150 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1152 up_read(&oi->ip_xattr_sem);
1157 /* ocfs2_xattr_get()
1159 * Copy an extended attribute into the buffer provided.
1160 * Buffer is NULL to compute the size of buffer required.
1162 static int ocfs2_xattr_get(struct inode *inode,
1169 struct buffer_head *di_bh = NULL;
1171 ret = ocfs2_inode_lock(inode, &di_bh, 0);
1176 ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1177 name, buffer, buffer_size);
1179 ocfs2_inode_unlock(inode, 0);
1186 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1188 struct ocfs2_xattr_value_root *xv,
1192 int ret = 0, i, cp_len;
1193 u16 blocksize = inode->i_sb->s_blocksize;
1194 u32 p_cluster, num_clusters;
1195 u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1196 u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1198 struct buffer_head *bh = NULL;
1200 BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1202 while (cpos < clusters) {
1203 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1204 &num_clusters, &xv->xr_list);
1210 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1212 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1213 ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1220 ret = ocfs2_journal_access(handle,
1223 OCFS2_JOURNAL_ACCESS_WRITE);
1229 cp_len = value_len > blocksize ? blocksize : value_len;
1230 memcpy(bh->b_data, value, cp_len);
1231 value_len -= cp_len;
1233 if (cp_len < blocksize)
1234 memset(bh->b_data + cp_len, 0,
1235 blocksize - cp_len);
1237 ret = ocfs2_journal_dirty(handle, bh);
1246 * XXX: do we need to empty all the following
1247 * blocks in this cluster?
1252 cpos += num_clusters;
1260 static int ocfs2_xattr_cleanup(struct inode *inode,
1262 struct ocfs2_xattr_info *xi,
1263 struct ocfs2_xattr_search *xs,
1264 struct ocfs2_xattr_value_buf *vb,
1268 size_t name_len = strlen(xi->name);
1269 void *val = xs->base + offs;
1270 size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1272 ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
1273 OCFS2_JOURNAL_ACCESS_WRITE);
1278 /* Decrease xattr count */
1279 le16_add_cpu(&xs->header->xh_count, -1);
1280 /* Remove the xattr entry and tree root which has already be set*/
1281 memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
1282 memset(val, 0, size);
1284 ret = ocfs2_journal_dirty(handle, vb->vb_bh);
1291 static int ocfs2_xattr_update_entry(struct inode *inode,
1293 struct ocfs2_xattr_info *xi,
1294 struct ocfs2_xattr_search *xs,
1295 struct ocfs2_xattr_value_buf *vb,
1300 ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
1301 OCFS2_JOURNAL_ACCESS_WRITE);
1307 xs->here->xe_name_offset = cpu_to_le16(offs);
1308 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1309 if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1310 ocfs2_xattr_set_local(xs->here, 1);
1312 ocfs2_xattr_set_local(xs->here, 0);
1313 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1315 ret = ocfs2_journal_dirty(handle, vb->vb_bh);
1323 * ocfs2_xattr_set_value_outside()
1325 * Set large size value in B tree.
1327 static int ocfs2_xattr_set_value_outside(struct inode *inode,
1328 struct ocfs2_xattr_info *xi,
1329 struct ocfs2_xattr_search *xs,
1330 struct ocfs2_xattr_set_ctxt *ctxt,
1331 struct ocfs2_xattr_value_buf *vb,
1334 size_t name_len = strlen(xi->name);
1335 void *val = xs->base + offs;
1336 struct ocfs2_xattr_value_root *xv = NULL;
1337 size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1340 memset(val, 0, size);
1341 memcpy(val, xi->name, name_len);
1342 xv = (struct ocfs2_xattr_value_root *)
1343 (val + OCFS2_XATTR_SIZE(name_len));
1344 xv->xr_clusters = 0;
1345 xv->xr_last_eb_blk = 0;
1346 xv->xr_list.l_tree_depth = 0;
1347 xv->xr_list.l_count = cpu_to_le16(1);
1348 xv->xr_list.l_next_free_rec = 0;
1351 ret = ocfs2_xattr_value_truncate(inode, vb, xi->value_len, ctxt);
1356 ret = ocfs2_xattr_update_entry(inode, ctxt->handle, xi, xs, vb, offs);
1361 ret = __ocfs2_xattr_set_value_outside(inode, ctxt->handle, vb->vb_xv,
1362 xi->value, xi->value_len);
1370 * ocfs2_xattr_set_entry_local()
1372 * Set, replace or remove extended attribute in local.
1374 static void ocfs2_xattr_set_entry_local(struct inode *inode,
1375 struct ocfs2_xattr_info *xi,
1376 struct ocfs2_xattr_search *xs,
1377 struct ocfs2_xattr_entry *last,
1380 size_t name_len = strlen(xi->name);
1383 if (xi->value && xs->not_found) {
1384 /* Insert the new xattr entry. */
1385 le16_add_cpu(&xs->header->xh_count, 1);
1386 ocfs2_xattr_set_type(last, xi->name_index);
1387 ocfs2_xattr_set_local(last, 1);
1388 last->xe_name_len = name_len;
1394 first_val = xs->base + min_offs;
1395 offs = le16_to_cpu(xs->here->xe_name_offset);
1396 val = xs->base + offs;
1398 if (le64_to_cpu(xs->here->xe_value_size) >
1399 OCFS2_XATTR_INLINE_SIZE)
1400 size = OCFS2_XATTR_SIZE(name_len) +
1401 OCFS2_XATTR_ROOT_SIZE;
1403 size = OCFS2_XATTR_SIZE(name_len) +
1404 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1406 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1407 OCFS2_XATTR_SIZE(xi->value_len)) {
1408 /* The old and the new value have the
1409 same size. Just replace the value. */
1410 ocfs2_xattr_set_local(xs->here, 1);
1411 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1412 /* Clear value bytes. */
1413 memset(val + OCFS2_XATTR_SIZE(name_len),
1415 OCFS2_XATTR_SIZE(xi->value_len));
1416 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1421 /* Remove the old name+value. */
1422 memmove(first_val + size, first_val, val - first_val);
1423 memset(first_val, 0, size);
1424 xs->here->xe_name_hash = 0;
1425 xs->here->xe_name_offset = 0;
1426 ocfs2_xattr_set_local(xs->here, 1);
1427 xs->here->xe_value_size = 0;
1431 /* Adjust all value offsets. */
1432 last = xs->header->xh_entries;
1433 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1434 size_t o = le16_to_cpu(last->xe_name_offset);
1437 last->xe_name_offset = cpu_to_le16(o + size);
1442 /* Remove the old entry. */
1444 memmove(xs->here, xs->here + 1,
1445 (void *)last - (void *)xs->here);
1446 memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1447 le16_add_cpu(&xs->header->xh_count, -1);
1451 /* Insert the new name+value. */
1452 size_t size = OCFS2_XATTR_SIZE(name_len) +
1453 OCFS2_XATTR_SIZE(xi->value_len);
1454 void *val = xs->base + min_offs - size;
1456 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1457 memset(val, 0, size);
1458 memcpy(val, xi->name, name_len);
1459 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1462 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1463 ocfs2_xattr_set_local(xs->here, 1);
1464 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1471 * ocfs2_xattr_set_entry()
1473 * Set extended attribute entry into inode or block.
1475 * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1476 * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1477 * then set value in B tree with set_value_outside().
1479 static int ocfs2_xattr_set_entry(struct inode *inode,
1480 struct ocfs2_xattr_info *xi,
1481 struct ocfs2_xattr_search *xs,
1482 struct ocfs2_xattr_set_ctxt *ctxt,
1485 struct ocfs2_xattr_entry *last;
1486 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1487 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1488 size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1490 handle_t *handle = ctxt->handle;
1492 struct ocfs2_xattr_info xi_l = {
1493 .name_index = xi->name_index,
1496 .value_len = xi->value_len,
1498 struct ocfs2_xattr_value_buf vb = {
1499 .vb_bh = xs->xattr_bh,
1500 .vb_access = ocfs2_journal_access_di,
1503 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1504 BUG_ON(xs->xattr_bh == xs->inode_bh);
1505 vb.vb_access = ocfs2_journal_access_xb;
1507 BUG_ON(xs->xattr_bh != xs->inode_bh);
1509 /* Compute min_offs, last and free space. */
1510 last = xs->header->xh_entries;
1512 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1513 size_t offs = le16_to_cpu(last->xe_name_offset);
1514 if (offs < min_offs)
1519 free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
1523 if (!xs->not_found) {
1525 if (ocfs2_xattr_is_local(xs->here))
1526 size = OCFS2_XATTR_SIZE(name_len) +
1527 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1529 size = OCFS2_XATTR_SIZE(name_len) +
1530 OCFS2_XATTR_ROOT_SIZE;
1531 free += (size + sizeof(struct ocfs2_xattr_entry));
1533 /* Check free space in inode or block */
1534 if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1535 if (free < sizeof(struct ocfs2_xattr_entry) +
1536 OCFS2_XATTR_SIZE(name_len) +
1537 OCFS2_XATTR_ROOT_SIZE) {
1541 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1542 xi_l.value = (void *)&def_xv;
1543 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1544 } else if (xi->value) {
1545 if (free < sizeof(struct ocfs2_xattr_entry) +
1546 OCFS2_XATTR_SIZE(name_len) +
1547 OCFS2_XATTR_SIZE(xi->value_len)) {
1553 if (!xs->not_found) {
1554 /* For existing extended attribute */
1555 size_t size = OCFS2_XATTR_SIZE(name_len) +
1556 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1557 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1558 void *val = xs->base + offs;
1560 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1561 /* Replace existing local xattr with tree root */
1562 ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1567 } else if (!ocfs2_xattr_is_local(xs->here)) {
1568 /* For existing xattr which has value outside */
1569 vb.vb_xv = (struct ocfs2_xattr_value_root *)
1570 (val + OCFS2_XATTR_SIZE(name_len));
1572 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1574 * If new value need set outside also,
1575 * first truncate old value to new value,
1576 * then set new value with set_value_outside().
1578 ret = ocfs2_xattr_value_truncate(inode,
1587 ret = ocfs2_xattr_update_entry(inode,
1598 ret = __ocfs2_xattr_set_value_outside(inode,
1608 * If new value need set in local,
1609 * just trucate old value to zero.
1611 ret = ocfs2_xattr_value_truncate(inode,
1621 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), xs->inode_bh,
1622 OCFS2_JOURNAL_ACCESS_WRITE);
1628 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1629 ret = vb.vb_access(handle, INODE_CACHE(inode), vb.vb_bh,
1630 OCFS2_JOURNAL_ACCESS_WRITE);
1638 * Set value in local, include set tree root in local.
1639 * This is the first step for value size >INLINE_SIZE.
1641 ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1643 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1644 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1651 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1652 (flag & OCFS2_INLINE_XATTR_FL)) {
1653 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1654 unsigned int xattrsize = osb->s_xattr_inline_size;
1657 * Adjust extent record count or inline data size
1658 * to reserve space for extended attribute.
1660 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1661 struct ocfs2_inline_data *idata = &di->id2.i_data;
1662 le16_add_cpu(&idata->id_count, -xattrsize);
1663 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1664 struct ocfs2_extent_list *el = &di->id2.i_list;
1665 le16_add_cpu(&el->l_count, -(xattrsize /
1666 sizeof(struct ocfs2_extent_rec)));
1668 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1670 /* Update xattr flag */
1671 spin_lock(&oi->ip_lock);
1672 oi->ip_dyn_features |= flag;
1673 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1674 spin_unlock(&oi->ip_lock);
1676 ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1680 if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1682 * Set value outside in B tree.
1683 * This is the second step for value size > INLINE_SIZE.
1685 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1686 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, ctxt,
1693 * If set value outside failed, we have to clean
1694 * the junk tree root we have already set in local.
1696 ret2 = ocfs2_xattr_cleanup(inode, ctxt->handle,
1706 static int ocfs2_remove_value_outside(struct inode*inode,
1707 struct ocfs2_xattr_value_buf *vb,
1708 struct ocfs2_xattr_header *header)
1711 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1712 struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
1714 ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
1716 ctxt.handle = ocfs2_start_trans(osb,
1717 ocfs2_remove_extent_credits(osb->sb));
1718 if (IS_ERR(ctxt.handle)) {
1719 ret = PTR_ERR(ctxt.handle);
1724 for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1725 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1727 if (!ocfs2_xattr_is_local(entry)) {
1730 val = (void *)header +
1731 le16_to_cpu(entry->xe_name_offset);
1732 vb->vb_xv = (struct ocfs2_xattr_value_root *)
1733 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
1734 ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
1742 ocfs2_commit_trans(osb, ctxt.handle);
1743 ocfs2_schedule_truncate_log_flush(osb, 1);
1744 ocfs2_run_deallocs(osb, &ctxt.dealloc);
1749 static int ocfs2_xattr_ibody_remove(struct inode *inode,
1750 struct buffer_head *di_bh)
1753 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1754 struct ocfs2_xattr_header *header;
1756 struct ocfs2_xattr_value_buf vb = {
1758 .vb_access = ocfs2_journal_access_di,
1761 header = (struct ocfs2_xattr_header *)
1762 ((void *)di + inode->i_sb->s_blocksize -
1763 le16_to_cpu(di->i_xattr_inline_size));
1765 ret = ocfs2_remove_value_outside(inode, &vb, header);
1770 static int ocfs2_xattr_block_remove(struct inode *inode,
1771 struct buffer_head *blk_bh)
1773 struct ocfs2_xattr_block *xb;
1775 struct ocfs2_xattr_value_buf vb = {
1777 .vb_access = ocfs2_journal_access_xb,
1780 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1781 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1782 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1783 ret = ocfs2_remove_value_outside(inode, &vb, header);
1785 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
1790 static int ocfs2_xattr_free_block(struct inode *inode,
1793 struct inode *xb_alloc_inode;
1794 struct buffer_head *xb_alloc_bh = NULL;
1795 struct buffer_head *blk_bh = NULL;
1796 struct ocfs2_xattr_block *xb;
1797 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1803 ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
1809 ret = ocfs2_xattr_block_remove(inode, blk_bh);
1815 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1816 blk = le64_to_cpu(xb->xb_blkno);
1817 bit = le16_to_cpu(xb->xb_suballoc_bit);
1818 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1820 xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1821 EXTENT_ALLOC_SYSTEM_INODE,
1822 le16_to_cpu(xb->xb_suballoc_slot));
1823 if (!xb_alloc_inode) {
1828 mutex_lock(&xb_alloc_inode->i_mutex);
1830 ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1836 handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
1837 if (IS_ERR(handle)) {
1838 ret = PTR_ERR(handle);
1843 ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1848 ocfs2_commit_trans(osb, handle);
1850 ocfs2_inode_unlock(xb_alloc_inode, 1);
1851 brelse(xb_alloc_bh);
1853 mutex_unlock(&xb_alloc_inode->i_mutex);
1854 iput(xb_alloc_inode);
1861 * ocfs2_xattr_remove()
1863 * Free extended attribute resources associated with this inode.
1865 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1867 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1868 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1872 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1875 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1878 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1879 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1886 if (di->i_xattr_loc) {
1887 ret = ocfs2_xattr_free_block(inode,
1888 le64_to_cpu(di->i_xattr_loc));
1895 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1896 OCFS2_INODE_UPDATE_CREDITS);
1897 if (IS_ERR(handle)) {
1898 ret = PTR_ERR(handle);
1902 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
1903 OCFS2_JOURNAL_ACCESS_WRITE);
1909 di->i_xattr_loc = 0;
1911 spin_lock(&oi->ip_lock);
1912 oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1913 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1914 spin_unlock(&oi->ip_lock);
1916 ret = ocfs2_journal_dirty(handle, di_bh);
1920 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1925 static int ocfs2_xattr_has_space_inline(struct inode *inode,
1926 struct ocfs2_dinode *di)
1928 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1929 unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1932 if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1935 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1936 struct ocfs2_inline_data *idata = &di->id2.i_data;
1937 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1938 } else if (ocfs2_inode_is_fast_symlink(inode)) {
1939 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1940 le64_to_cpu(di->i_size);
1942 struct ocfs2_extent_list *el = &di->id2.i_list;
1943 free = (le16_to_cpu(el->l_count) -
1944 le16_to_cpu(el->l_next_free_rec)) *
1945 sizeof(struct ocfs2_extent_rec);
1947 if (free >= xattrsize)
1954 * ocfs2_xattr_ibody_find()
1956 * Find extended attribute in inode block and
1957 * fill search info into struct ocfs2_xattr_search.
1959 static int ocfs2_xattr_ibody_find(struct inode *inode,
1962 struct ocfs2_xattr_search *xs)
1964 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1965 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1969 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1972 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1973 down_read(&oi->ip_alloc_sem);
1974 has_space = ocfs2_xattr_has_space_inline(inode, di);
1975 up_read(&oi->ip_alloc_sem);
1980 xs->xattr_bh = xs->inode_bh;
1981 xs->end = (void *)di + inode->i_sb->s_blocksize;
1982 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1983 xs->header = (struct ocfs2_xattr_header *)
1984 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1986 xs->header = (struct ocfs2_xattr_header *)
1987 (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
1988 xs->base = (void *)xs->header;
1989 xs->here = xs->header->xh_entries;
1991 /* Find the named attribute. */
1992 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1993 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1994 if (ret && ret != -ENODATA)
1996 xs->not_found = ret;
2003 * ocfs2_xattr_ibody_set()
2005 * Set, replace or remove an extended attribute into inode block.
2008 static int ocfs2_xattr_ibody_set(struct inode *inode,
2009 struct ocfs2_xattr_info *xi,
2010 struct ocfs2_xattr_search *xs,
2011 struct ocfs2_xattr_set_ctxt *ctxt)
2013 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2014 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2017 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2020 down_write(&oi->ip_alloc_sem);
2021 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2022 if (!ocfs2_xattr_has_space_inline(inode, di)) {
2028 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2029 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
2031 up_write(&oi->ip_alloc_sem);
2037 * ocfs2_xattr_block_find()
2039 * Find extended attribute in external block and
2040 * fill search info into struct ocfs2_xattr_search.
2042 static int ocfs2_xattr_block_find(struct inode *inode,
2045 struct ocfs2_xattr_search *xs)
2047 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2048 struct buffer_head *blk_bh = NULL;
2049 struct ocfs2_xattr_block *xb;
2052 if (!di->i_xattr_loc)
2055 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2062 xs->xattr_bh = blk_bh;
2063 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2065 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2066 xs->header = &xb->xb_attrs.xb_header;
2067 xs->base = (void *)xs->header;
2068 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2069 xs->here = xs->header->xh_entries;
2071 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2073 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2077 if (ret && ret != -ENODATA) {
2078 xs->xattr_bh = NULL;
2081 xs->not_found = ret;
2090 * ocfs2_xattr_block_set()
2092 * Set, replace or remove an extended attribute into external block.
2095 static int ocfs2_xattr_block_set(struct inode *inode,
2096 struct ocfs2_xattr_info *xi,
2097 struct ocfs2_xattr_search *xs,
2098 struct ocfs2_xattr_set_ctxt *ctxt)
2100 struct buffer_head *new_bh = NULL;
2101 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2102 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2103 handle_t *handle = ctxt->handle;
2104 struct ocfs2_xattr_block *xblk = NULL;
2105 u16 suballoc_bit_start;
2110 if (!xs->xattr_bh) {
2111 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2113 OCFS2_JOURNAL_ACCESS_CREATE);
2119 ret = ocfs2_claim_metadata(osb, handle, ctxt->meta_ac, 1,
2120 &suballoc_bit_start, &num_got,
2127 new_bh = sb_getblk(inode->i_sb, first_blkno);
2128 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2130 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode),
2132 OCFS2_JOURNAL_ACCESS_CREATE);
2138 /* Initialize ocfs2_xattr_block */
2139 xs->xattr_bh = new_bh;
2140 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2141 memset(xblk, 0, inode->i_sb->s_blocksize);
2142 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2143 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
2144 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2145 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
2146 xblk->xb_blkno = cpu_to_le64(first_blkno);
2148 xs->header = &xblk->xb_attrs.xb_header;
2149 xs->base = (void *)xs->header;
2150 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2151 xs->here = xs->header->xh_entries;
2153 ret = ocfs2_journal_dirty(handle, new_bh);
2158 di->i_xattr_loc = cpu_to_le64(first_blkno);
2159 ocfs2_journal_dirty(handle, xs->inode_bh);
2161 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2163 if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2164 /* Set extended attribute into external block */
2165 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2166 OCFS2_HAS_XATTR_FL);
2167 if (!ret || ret != -ENOSPC)
2170 ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2175 ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2182 /* Check whether the new xattr can be inserted into the inode. */
2183 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2184 struct ocfs2_xattr_info *xi,
2185 struct ocfs2_xattr_search *xs)
2188 struct ocfs2_xattr_entry *last;
2190 size_t min_offs = xs->end - xs->base;
2195 last = xs->header->xh_entries;
2197 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
2198 size_t offs = le16_to_cpu(last->xe_name_offset);
2199 if (offs < min_offs)
2204 free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
2208 BUG_ON(!xs->not_found);
2210 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2211 value_size = OCFS2_XATTR_ROOT_SIZE;
2213 value_size = OCFS2_XATTR_SIZE(xi->value_len);
2215 if (free >= sizeof(struct ocfs2_xattr_entry) +
2216 OCFS2_XATTR_SIZE(strlen(xi->name)) + value_size)
2222 static int ocfs2_calc_xattr_set_need(struct inode *inode,
2223 struct ocfs2_dinode *di,
2224 struct ocfs2_xattr_info *xi,
2225 struct ocfs2_xattr_search *xis,
2226 struct ocfs2_xattr_search *xbs,
2231 int ret = 0, old_in_xb = 0;
2232 int clusters_add = 0, meta_add = 0, credits = 0;
2233 struct buffer_head *bh = NULL;
2234 struct ocfs2_xattr_block *xb = NULL;
2235 struct ocfs2_xattr_entry *xe = NULL;
2236 struct ocfs2_xattr_value_root *xv = NULL;
2238 int name_offset, name_len = 0;
2239 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2244 * Calculate the clusters we need to write.
2245 * No matter whether we replace an old one or add a new one,
2246 * we need this for writing.
2248 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2249 credits += new_clusters *
2250 ocfs2_clusters_to_blocks(inode->i_sb, 1);
2252 if (xis->not_found && xbs->not_found) {
2253 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2255 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2256 clusters_add += new_clusters;
2257 credits += ocfs2_calc_extend_credits(inode->i_sb,
2265 if (!xis->not_found) {
2267 name_offset = le16_to_cpu(xe->xe_name_offset);
2268 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2270 credits += OCFS2_INODE_UPDATE_CREDITS;
2272 int i, block_off = 0;
2273 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2275 name_offset = le16_to_cpu(xe->xe_name_offset);
2276 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2277 i = xbs->here - xbs->header->xh_entries;
2280 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2281 ret = ocfs2_xattr_bucket_get_name_value(inode,
2282 bucket_xh(xbs->bucket),
2285 base = bucket_block(xbs->bucket, block_off);
2286 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2289 credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
2294 * delete a xattr doesn't need metadata and cluster allocation.
2295 * so just calculate the credits and return.
2297 * The credits for removing the value tree will be extended
2298 * by ocfs2_remove_extent itself.
2301 if (!ocfs2_xattr_is_local(xe))
2302 credits += ocfs2_remove_extent_credits(inode->i_sb);
2307 /* do cluster allocation guess first. */
2308 value_size = le64_to_cpu(xe->xe_value_size);
2312 * In xattr set, we always try to set the xe in inode first,
2313 * so if it can be inserted into inode successfully, the old
2314 * one will be removed from the xattr block, and this xattr
2315 * will be inserted into inode as a new xattr in inode.
2317 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
2318 clusters_add += new_clusters;
2319 credits += ocfs2_remove_extent_credits(inode->i_sb) +
2320 OCFS2_INODE_UPDATE_CREDITS;
2321 if (!ocfs2_xattr_is_local(xe))
2322 credits += ocfs2_calc_extend_credits(
2330 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2331 /* the new values will be stored outside. */
2332 u32 old_clusters = 0;
2334 if (!ocfs2_xattr_is_local(xe)) {
2335 old_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2337 xv = (struct ocfs2_xattr_value_root *)
2338 (base + name_offset + name_len);
2339 value_size = OCFS2_XATTR_ROOT_SIZE;
2343 if (old_clusters >= new_clusters) {
2344 credits += ocfs2_remove_extent_credits(inode->i_sb);
2347 meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
2348 clusters_add += new_clusters - old_clusters;
2349 credits += ocfs2_calc_extend_credits(inode->i_sb,
2353 if (value_size >= OCFS2_XATTR_ROOT_SIZE)
2358 * Now the new value will be stored inside. So if the new
2359 * value is smaller than the size of value root or the old
2360 * value, we don't need any allocation, otherwise we have
2361 * to guess metadata allocation.
2363 if ((ocfs2_xattr_is_local(xe) && value_size >= xi->value_len) ||
2364 (!ocfs2_xattr_is_local(xe) &&
2365 OCFS2_XATTR_ROOT_SIZE >= xi->value_len))
2370 /* calculate metadata allocation. */
2371 if (di->i_xattr_loc) {
2372 if (!xbs->xattr_bh) {
2373 ret = ocfs2_read_xattr_block(inode,
2374 le64_to_cpu(di->i_xattr_loc),
2381 xb = (struct ocfs2_xattr_block *)bh->b_data;
2383 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2386 * If there is already an xattr tree, good, we can calculate
2387 * like other b-trees. Otherwise we may have the chance of
2388 * create a tree, the credit calculation is borrowed from
2389 * ocfs2_calc_extend_credits with root_el = NULL. And the
2390 * new tree will be cluster based, so no meta is needed.
2392 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2393 struct ocfs2_extent_list *el =
2394 &xb->xb_attrs.xb_root.xt_list;
2395 meta_add += ocfs2_extend_meta_needed(el);
2396 credits += ocfs2_calc_extend_credits(inode->i_sb,
2399 credits += OCFS2_SUBALLOC_ALLOC + 1;
2402 * This cluster will be used either for new bucket or for
2404 * If the cluster size is the same as the bucket size, one
2405 * more is needed since we may need to extend the bucket
2409 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2410 if (OCFS2_XATTR_BUCKET_SIZE ==
2411 OCFS2_SB(inode->i_sb)->s_clustersize) {
2412 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2417 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
2421 *clusters_need = clusters_add;
2423 *meta_need = meta_add;
2425 *credits_need = credits;
2430 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
2431 struct ocfs2_dinode *di,
2432 struct ocfs2_xattr_info *xi,
2433 struct ocfs2_xattr_search *xis,
2434 struct ocfs2_xattr_search *xbs,
2435 struct ocfs2_xattr_set_ctxt *ctxt,
2438 int clusters_add, meta_add, ret;
2439 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2441 memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
2443 ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
2445 ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
2446 &clusters_add, &meta_add, credits);
2452 mlog(0, "Set xattr %s, reserve meta blocks = %d, clusters = %d, "
2453 "credits = %d\n", xi->name, meta_add, clusters_add, *credits);
2456 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
2465 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
2471 if (ctxt->meta_ac) {
2472 ocfs2_free_alloc_context(ctxt->meta_ac);
2473 ctxt->meta_ac = NULL;
2477 * We cannot have an error and a non null ctxt->data_ac.
2484 static int __ocfs2_xattr_set_handle(struct inode *inode,
2485 struct ocfs2_dinode *di,
2486 struct ocfs2_xattr_info *xi,
2487 struct ocfs2_xattr_search *xis,
2488 struct ocfs2_xattr_search *xbs,
2489 struct ocfs2_xattr_set_ctxt *ctxt)
2491 int ret = 0, credits, old_found;
2494 /* Remove existing extended attribute */
2495 if (!xis->not_found)
2496 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2497 else if (!xbs->not_found)
2498 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2500 /* We always try to set extended attribute into inode first*/
2501 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2502 if (!ret && !xbs->not_found) {
2504 * If succeed and that extended attribute existing in
2505 * external block, then we will remove it.
2510 old_found = xis->not_found;
2511 xis->not_found = -ENODATA;
2512 ret = ocfs2_calc_xattr_set_need(inode,
2520 xis->not_found = old_found;
2526 ret = ocfs2_extend_trans(ctxt->handle, credits +
2527 ctxt->handle->h_buffer_credits);
2532 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2533 } else if (ret == -ENOSPC) {
2534 if (di->i_xattr_loc && !xbs->xattr_bh) {
2535 ret = ocfs2_xattr_block_find(inode,
2541 old_found = xis->not_found;
2542 xis->not_found = -ENODATA;
2543 ret = ocfs2_calc_xattr_set_need(inode,
2551 xis->not_found = old_found;
2557 ret = ocfs2_extend_trans(ctxt->handle, credits +
2558 ctxt->handle->h_buffer_credits);
2565 * If no space in inode, we will set extended attribute
2566 * into external block.
2568 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2571 if (!xis->not_found) {
2573 * If succeed and that extended attribute
2574 * existing in inode, we will remove it.
2578 xbs->not_found = -ENODATA;
2579 ret = ocfs2_calc_xattr_set_need(inode,
2592 ret = ocfs2_extend_trans(ctxt->handle, credits +
2593 ctxt->handle->h_buffer_credits);
2598 ret = ocfs2_xattr_ibody_set(inode, xi,
2605 /* Update inode ctime. */
2606 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
2608 OCFS2_JOURNAL_ACCESS_WRITE);
2614 inode->i_ctime = CURRENT_TIME;
2615 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
2616 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
2617 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
2624 * This function only called duing creating inode
2625 * for init security/acl xattrs of the new inode.
2626 * All transanction credits have been reserved in mknod.
2628 int ocfs2_xattr_set_handle(handle_t *handle,
2629 struct inode *inode,
2630 struct buffer_head *di_bh,
2636 struct ocfs2_alloc_context *meta_ac,
2637 struct ocfs2_alloc_context *data_ac)
2639 struct ocfs2_dinode *di;
2642 struct ocfs2_xattr_info xi = {
2643 .name_index = name_index,
2646 .value_len = value_len,
2649 struct ocfs2_xattr_search xis = {
2650 .not_found = -ENODATA,
2653 struct ocfs2_xattr_search xbs = {
2654 .not_found = -ENODATA,
2657 struct ocfs2_xattr_set_ctxt ctxt = {
2663 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2667 * In extreme situation, may need xattr bucket when
2668 * block size is too small. And we have already reserved
2669 * the credits for bucket in mknod.
2671 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
2672 xbs.bucket = ocfs2_xattr_bucket_new(inode);
2674 mlog_errno(-ENOMEM);
2679 xis.inode_bh = xbs.inode_bh = di_bh;
2680 di = (struct ocfs2_dinode *)di_bh->b_data;
2682 down_write(&OCFS2_I(inode)->ip_xattr_sem);
2684 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2687 if (xis.not_found) {
2688 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2693 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2696 up_write(&OCFS2_I(inode)->ip_xattr_sem);
2697 brelse(xbs.xattr_bh);
2698 ocfs2_xattr_bucket_free(xbs.bucket);
2706 * Set, replace or remove an extended attribute for this inode.
2707 * value is NULL to remove an existing extended attribute, else either
2708 * create or replace an extended attribute.
2710 int ocfs2_xattr_set(struct inode *inode,
2717 struct buffer_head *di_bh = NULL;
2718 struct ocfs2_dinode *di;
2720 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2721 struct inode *tl_inode = osb->osb_tl_inode;
2722 struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2724 struct ocfs2_xattr_info xi = {
2725 .name_index = name_index,
2728 .value_len = value_len,
2731 struct ocfs2_xattr_search xis = {
2732 .not_found = -ENODATA,
2735 struct ocfs2_xattr_search xbs = {
2736 .not_found = -ENODATA,
2739 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2743 * Only xbs will be used on indexed trees. xis doesn't need a
2746 xbs.bucket = ocfs2_xattr_bucket_new(inode);
2748 mlog_errno(-ENOMEM);
2752 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2755 goto cleanup_nolock;
2757 xis.inode_bh = xbs.inode_bh = di_bh;
2758 di = (struct ocfs2_dinode *)di_bh->b_data;
2760 down_write(&OCFS2_I(inode)->ip_xattr_sem);
2762 * Scan inode and external block to find the same name
2763 * extended attribute and collect search infomation.
2765 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2768 if (xis.not_found) {
2769 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2774 if (xis.not_found && xbs.not_found) {
2776 if (flags & XATTR_REPLACE)
2783 if (flags & XATTR_CREATE)
2788 mutex_lock(&tl_inode->i_mutex);
2790 if (ocfs2_truncate_log_needs_flush(osb)) {
2791 ret = __ocfs2_flush_truncate_log(osb);
2793 mutex_unlock(&tl_inode->i_mutex);
2798 mutex_unlock(&tl_inode->i_mutex);
2800 ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
2801 &xbs, &ctxt, &credits);
2807 /* we need to update inode's ctime field, so add credit for it. */
2808 credits += OCFS2_INODE_UPDATE_CREDITS;
2809 ctxt.handle = ocfs2_start_trans(osb, credits);
2810 if (IS_ERR(ctxt.handle)) {
2811 ret = PTR_ERR(ctxt.handle);
2816 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2818 ocfs2_commit_trans(osb, ctxt.handle);
2821 ocfs2_free_alloc_context(ctxt.data_ac);
2823 ocfs2_free_alloc_context(ctxt.meta_ac);
2824 if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
2825 ocfs2_schedule_truncate_log_flush(osb, 1);
2826 ocfs2_run_deallocs(osb, &ctxt.dealloc);
2828 up_write(&OCFS2_I(inode)->ip_xattr_sem);
2829 ocfs2_inode_unlock(inode, 1);
2832 brelse(xbs.xattr_bh);
2833 ocfs2_xattr_bucket_free(xbs.bucket);
2839 * Find the xattr extent rec which may contains name_hash.
2840 * e_cpos will be the first name hash of the xattr rec.
2841 * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2843 static int ocfs2_xattr_get_rec(struct inode *inode,
2848 struct ocfs2_extent_list *el)
2851 struct buffer_head *eb_bh = NULL;
2852 struct ocfs2_extent_block *eb;
2853 struct ocfs2_extent_rec *rec = NULL;
2856 if (el->l_tree_depth) {
2857 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, name_hash,
2864 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2867 if (el->l_tree_depth) {
2868 ocfs2_error(inode->i_sb,
2869 "Inode %lu has non zero tree depth in "
2870 "xattr tree block %llu\n", inode->i_ino,
2871 (unsigned long long)eb_bh->b_blocknr);
2877 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2878 rec = &el->l_recs[i];
2880 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2881 e_blkno = le64_to_cpu(rec->e_blkno);
2887 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2888 "record (%u, %u, 0) in xattr", inode->i_ino,
2889 le32_to_cpu(rec->e_cpos),
2890 ocfs2_rec_clusters(el, rec));
2895 *p_blkno = le64_to_cpu(rec->e_blkno);
2896 *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2898 *e_cpos = le32_to_cpu(rec->e_cpos);
2904 typedef int (xattr_bucket_func)(struct inode *inode,
2905 struct ocfs2_xattr_bucket *bucket,
2908 static int ocfs2_find_xe_in_bucket(struct inode *inode,
2909 struct ocfs2_xattr_bucket *bucket,
2916 int i, ret = 0, cmp = 1, block_off, new_offset;
2917 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
2918 size_t name_len = strlen(name);
2919 struct ocfs2_xattr_entry *xe = NULL;
2923 * We don't use binary search in the bucket because there
2924 * may be multiple entries with the same name hash.
2926 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2927 xe = &xh->xh_entries[i];
2929 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2931 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2934 cmp = name_index - ocfs2_xattr_get_type(xe);
2936 cmp = name_len - xe->xe_name_len;
2940 ret = ocfs2_xattr_bucket_get_name_value(inode,
2951 xe_name = bucket_block(bucket, block_off) + new_offset;
2952 if (!memcmp(name, xe_name, name_len)) {
2964 * Find the specified xattr entry in a series of buckets.
2965 * This series start from p_blkno and last for num_clusters.
2966 * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2967 * the num of the valid buckets.
2969 * Return the buffer_head this xattr should reside in. And if the xattr's
2970 * hash is in the gap of 2 buckets, return the lower bucket.
2972 static int ocfs2_xattr_bucket_find(struct inode *inode,
2979 struct ocfs2_xattr_search *xs)
2982 struct ocfs2_xattr_header *xh = NULL;
2983 struct ocfs2_xattr_entry *xe = NULL;
2985 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2986 int low_bucket = 0, bucket, high_bucket;
2987 struct ocfs2_xattr_bucket *search;
2989 u64 blkno, lower_blkno = 0;
2991 search = ocfs2_xattr_bucket_new(inode);
2998 ret = ocfs2_read_xattr_bucket(search, p_blkno);
3004 xh = bucket_xh(search);
3005 high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
3006 while (low_bucket <= high_bucket) {
3007 ocfs2_xattr_bucket_relse(search);
3009 bucket = (low_bucket + high_bucket) / 2;
3010 blkno = p_blkno + bucket * blk_per_bucket;
3011 ret = ocfs2_read_xattr_bucket(search, blkno);
3017 xh = bucket_xh(search);
3018 xe = &xh->xh_entries[0];
3019 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3020 high_bucket = bucket - 1;
3025 * Check whether the hash of the last entry in our
3026 * bucket is larger than the search one. for an empty
3027 * bucket, the last one is also the first one.
3030 xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3032 last_hash = le32_to_cpu(xe->xe_name_hash);
3034 /* record lower_blkno which may be the insert place. */
3035 lower_blkno = blkno;
3037 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3038 low_bucket = bucket + 1;
3042 /* the searched xattr should reside in this bucket if exists. */
3043 ret = ocfs2_find_xe_in_bucket(inode, search,
3044 name_index, name, name_hash,
3054 * Record the bucket we have found.
3055 * When the xattr's hash value is in the gap of 2 buckets, we will
3056 * always set it to the previous bucket.
3059 lower_blkno = p_blkno;
3061 /* This should be in cache - we just read it during the search */
3062 ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3068 xs->header = bucket_xh(xs->bucket);
3069 xs->base = bucket_block(xs->bucket, 0);
3070 xs->end = xs->base + inode->i_sb->s_blocksize;
3073 xs->here = &xs->header->xh_entries[index];
3074 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
3075 (unsigned long long)bucket_blkno(xs->bucket), index);
3080 ocfs2_xattr_bucket_free(search);
3084 static int ocfs2_xattr_index_block_find(struct inode *inode,
3085 struct buffer_head *root_bh,
3088 struct ocfs2_xattr_search *xs)
3091 struct ocfs2_xattr_block *xb =
3092 (struct ocfs2_xattr_block *)root_bh->b_data;
3093 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3094 struct ocfs2_extent_list *el = &xb_root->xt_list;
3096 u32 first_hash, num_clusters = 0;
3097 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3099 if (le16_to_cpu(el->l_next_free_rec) == 0)
3102 mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
3103 name, name_hash, name_index);
3105 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3112 BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3114 mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
3115 "in the rec is %u\n", num_clusters, (unsigned long long)p_blkno,
3118 ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3119 p_blkno, first_hash, num_clusters, xs);
3125 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3128 xattr_bucket_func *func,
3132 u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3133 u32 num_buckets = clusters * bpc;
3134 struct ocfs2_xattr_bucket *bucket;
3136 bucket = ocfs2_xattr_bucket_new(inode);
3138 mlog_errno(-ENOMEM);
3142 mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
3143 clusters, (unsigned long long)blkno);
3145 for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3146 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3153 * The real bucket num in this series of blocks is stored
3154 * in the 1st bucket.
3157 num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3159 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
3160 (unsigned long long)blkno,
3161 le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3163 ret = func(inode, bucket, para);
3164 if (ret && ret != -ERANGE)
3166 /* Fall through to bucket_relse() */
3169 ocfs2_xattr_bucket_relse(bucket);
3174 ocfs2_xattr_bucket_free(bucket);
3178 struct ocfs2_xattr_tree_list {
3184 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
3185 struct ocfs2_xattr_header *xh,
3192 if (index < 0 || index >= le16_to_cpu(xh->xh_count))
3195 name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
3197 *block_off = name_offset >> inode->i_sb->s_blocksize_bits;
3198 *new_offset = name_offset % inode->i_sb->s_blocksize;
3203 static int ocfs2_list_xattr_bucket(struct inode *inode,
3204 struct ocfs2_xattr_bucket *bucket,
3208 struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
3209 int i, block_off, new_offset;
3210 const char *prefix, *name;
3212 for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
3213 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
3214 type = ocfs2_xattr_get_type(entry);
3215 prefix = ocfs2_xattr_prefix(type);
3218 ret = ocfs2_xattr_bucket_get_name_value(inode,
3226 name = (const char *)bucket_block(bucket, block_off) +
3228 ret = ocfs2_xattr_list_entry(xl->buffer,
3232 entry->xe_name_len);
3241 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
3242 struct ocfs2_xattr_tree_root *xt,
3246 struct ocfs2_extent_list *el = &xt->xt_list;
3248 u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
3250 struct ocfs2_xattr_tree_list xl = {
3252 .buffer_size = buffer_size,
3256 if (le16_to_cpu(el->l_next_free_rec) == 0)
3259 while (name_hash > 0) {
3260 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
3261 &e_cpos, &num_clusters, el);
3267 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
3268 ocfs2_list_xattr_bucket,
3279 name_hash = e_cpos - 1;
3287 static int cmp_xe(const void *a, const void *b)
3289 const struct ocfs2_xattr_entry *l = a, *r = b;
3290 u32 l_hash = le32_to_cpu(l->xe_name_hash);
3291 u32 r_hash = le32_to_cpu(r->xe_name_hash);
3293 if (l_hash > r_hash)
3295 if (l_hash < r_hash)
3300 static void swap_xe(void *a, void *b, int size)
3302 struct ocfs2_xattr_entry *l = a, *r = b, tmp;
3305 memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
3306 memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
3310 * When the ocfs2_xattr_block is filled up, new bucket will be created
3311 * and all the xattr entries will be moved to the new bucket.
3312 * The header goes at the start of the bucket, and the names+values are
3313 * filled from the end. This is why *target starts as the last buffer.
3314 * Note: we need to sort the entries since they are not saved in order
3315 * in the ocfs2_xattr_block.
3317 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
3318 struct buffer_head *xb_bh,
3319 struct ocfs2_xattr_bucket *bucket)
3321 int i, blocksize = inode->i_sb->s_blocksize;
3322 int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3323 u16 offset, size, off_change;
3324 struct ocfs2_xattr_entry *xe;
3325 struct ocfs2_xattr_block *xb =
3326 (struct ocfs2_xattr_block *)xb_bh->b_data;
3327 struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
3328 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3329 u16 count = le16_to_cpu(xb_xh->xh_count);
3330 char *src = xb_bh->b_data;
3331 char *target = bucket_block(bucket, blks - 1);
3333 mlog(0, "cp xattr from block %llu to bucket %llu\n",
3334 (unsigned long long)xb_bh->b_blocknr,
3335 (unsigned long long)bucket_blkno(bucket));
3337 for (i = 0; i < blks; i++)
3338 memset(bucket_block(bucket, i), 0, blocksize);
3341 * Since the xe_name_offset is based on ocfs2_xattr_header,
3342 * there is a offset change corresponding to the change of
3343 * ocfs2_xattr_header's position.
3345 off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3346 xe = &xb_xh->xh_entries[count - 1];
3347 offset = le16_to_cpu(xe->xe_name_offset) + off_change;
3348 size = blocksize - offset;
3350 /* copy all the names and values. */
3351 memcpy(target + offset, src + offset, size);
3353 /* Init new header now. */
3354 xh->xh_count = xb_xh->xh_count;
3355 xh->xh_num_buckets = cpu_to_le16(1);
3356 xh->xh_name_value_len = cpu_to_le16(size);
3357 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
3359 /* copy all the entries. */
3360 target = bucket_block(bucket, 0);
3361 offset = offsetof(struct ocfs2_xattr_header, xh_entries);
3362 size = count * sizeof(struct ocfs2_xattr_entry);
3363 memcpy(target + offset, (char *)xb_xh + offset, size);
3365 /* Change the xe offset for all the xe because of the move. */
3366 off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
3367 offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3368 for (i = 0; i < count; i++)
3369 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
3371 mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
3372 offset, size, off_change);
3374 sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
3379 * After we move xattr from block to index btree, we have to
3380 * update ocfs2_xattr_search to the new xe and base.
3382 * When the entry is in xattr block, xattr_bh indicates the storage place.
3383 * While if the entry is in index b-tree, "bucket" indicates the
3384 * real place of the xattr.
3386 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
3387 struct ocfs2_xattr_search *xs,
3388 struct buffer_head *old_bh)
3390 char *buf = old_bh->b_data;
3391 struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
3392 struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
3395 xs->header = bucket_xh(xs->bucket);
3396 xs->base = bucket_block(xs->bucket, 0);
3397 xs->end = xs->base + inode->i_sb->s_blocksize;
3402 i = xs->here - old_xh->xh_entries;
3403 xs->here = &xs->header->xh_entries[i];
3406 static int ocfs2_xattr_create_index_block(struct inode *inode,
3407 struct ocfs2_xattr_search *xs,
3408 struct ocfs2_xattr_set_ctxt *ctxt)
3413 handle_t *handle = ctxt->handle;
3414 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3415 struct ocfs2_inode_info *oi = OCFS2_I(inode);
3416 struct buffer_head *xb_bh = xs->xattr_bh;
3417 struct ocfs2_xattr_block *xb =
3418 (struct ocfs2_xattr_block *)xb_bh->b_data;
3419 struct ocfs2_xattr_tree_root *xr;
3420 u16 xb_flags = le16_to_cpu(xb->xb_flags);
3422 mlog(0, "create xattr index block for %llu\n",
3423 (unsigned long long)xb_bh->b_blocknr);
3425 BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
3426 BUG_ON(!xs->bucket);
3430 * We can use this lock for now, and maybe move to a dedicated mutex
3431 * if performance becomes a problem later.
3433 down_write(&oi->ip_alloc_sem);
3435 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), xb_bh,
3436 OCFS2_JOURNAL_ACCESS_WRITE);
3442 ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac,
3443 1, 1, &bit_off, &len);
3450 * The bucket may spread in many blocks, and
3451 * we will only touch the 1st block and the last block
3452 * in the whole bucket(one for entry and one for data).
3454 blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
3456 mlog(0, "allocate 1 cluster from %llu to xattr block\n",
3457 (unsigned long long)blkno);
3459 ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
3465 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
3466 OCFS2_JOURNAL_ACCESS_CREATE);
3472 ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
3473 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
3475 ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
3477 /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
3478 memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
3479 offsetof(struct ocfs2_xattr_block, xb_attrs));
3481 xr = &xb->xb_attrs.xb_root;
3482 xr->xt_clusters = cpu_to_le32(1);
3483 xr->xt_last_eb_blk = 0;
3484 xr->xt_list.l_tree_depth = 0;
3485 xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
3486 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
3488 xr->xt_list.l_recs[0].e_cpos = 0;
3489 xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
3490 xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
3492 xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
3494 ocfs2_journal_dirty(handle, xb_bh);
3497 up_write(&oi->ip_alloc_sem);
3502 static int cmp_xe_offset(const void *a, const void *b)
3504 const struct ocfs2_xattr_entry *l = a, *r = b;
3505 u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
3506 u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
3508 if (l_name_offset < r_name_offset)
3510 if (l_name_offset > r_name_offset)
3516 * defrag a xattr bucket if we find that the bucket has some
3517 * holes beteen name/value pairs.
3518 * We will move all the name/value pairs to the end of the bucket
3519 * so that we can spare some space for insertion.
3521 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
3523 struct ocfs2_xattr_bucket *bucket)
3526 size_t end, offset, len, value_len;
3527 struct ocfs2_xattr_header *xh;
3528 char *entries, *buf, *bucket_buf = NULL;
3529 u64 blkno = bucket_blkno(bucket);
3531 size_t blocksize = inode->i_sb->s_blocksize;
3532 struct ocfs2_xattr_entry *xe;
3535 * In order to make the operation more efficient and generic,
3536 * we copy all the blocks into a contiguous memory and do the
3537 * defragment there, so if anything is error, we will not touch
3540 bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
3547 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3548 memcpy(buf, bucket_block(bucket, i), blocksize);
3550 ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
3551 OCFS2_JOURNAL_ACCESS_WRITE);
3557 xh = (struct ocfs2_xattr_header *)bucket_buf;
3558 entries = (char *)xh->xh_entries;
3559 xh_free_start = le16_to_cpu(xh->xh_free_start);
3561 mlog(0, "adjust xattr bucket in %llu, count = %u, "
3562 "xh_free_start = %u, xh_name_value_len = %u.\n",
3563 (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
3564 xh_free_start, le16_to_cpu(xh->xh_name_value_len));
3567 * sort all the entries by their offset.
3568 * the largest will be the first, so that we can
3569 * move them to the end one by one.
3571 sort(entries, le16_to_cpu(xh->xh_count),
3572 sizeof(struct ocfs2_xattr_entry),
3573 cmp_xe_offset, swap_xe);
3575 /* Move all name/values to the end of the bucket. */
3576 xe = xh->xh_entries;
3577 end = OCFS2_XATTR_BUCKET_SIZE;
3578 for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
3579 offset = le16_to_cpu(xe->xe_name_offset);
3580 if (ocfs2_xattr_is_local(xe))
3581 value_len = OCFS2_XATTR_SIZE(
3582 le64_to_cpu(xe->xe_value_size));
3584 value_len = OCFS2_XATTR_ROOT_SIZE;
3585 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
3588 * We must make sure that the name/value pair
3589 * exist in the same block. So adjust end to
3590 * the previous block end if needed.
3592 if (((end - len) / blocksize !=
3593 (end - 1) / blocksize))
3594 end = end - end % blocksize;
3596 if (end > offset + len) {
3597 memmove(bucket_buf + end - len,
3598 bucket_buf + offset, len);
3599 xe->xe_name_offset = cpu_to_le16(end - len);
3602 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
3603 "bucket %llu\n", (unsigned long long)blkno);
3608 mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
3609 "bucket %llu\n", (unsigned long long)blkno);
3611 if (xh_free_start == end)
3614 memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
3615 xh->xh_free_start = cpu_to_le16(end);
3617 /* sort the entries by their name_hash. */
3618 sort(entries, le16_to_cpu(xh->xh_count),
3619 sizeof(struct ocfs2_xattr_entry),
3623 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3624 memcpy(bucket_block(bucket, i), buf, blocksize);
3625 ocfs2_xattr_bucket_journal_dirty(handle, bucket);
3633 * prev_blkno points to the start of an existing extent. new_blkno
3634 * points to a newly allocated extent. Because we know each of our
3635 * clusters contains more than bucket, we can easily split one cluster
3636 * at a bucket boundary. So we take the last cluster of the existing
3637 * extent and split it down the middle. We move the last half of the
3638 * buckets in the last cluster of the existing extent over to the new
3641 * first_bh is the buffer at prev_blkno so we can update the existing
3642 * extent's bucket count. header_bh is the bucket were we were hoping
3643 * to insert our xattr. If the bucket move places the target in the new
3644 * extent, we'll update first_bh and header_bh after modifying the old
3647 * first_hash will be set as the 1st xe's name_hash in the new extent.
3649 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3651 struct ocfs2_xattr_bucket *first,
3652 struct ocfs2_xattr_bucket *target,
3658 struct super_block *sb = inode->i_sb;
3659 int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
3660 int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
3661 int to_move = num_buckets / 2;
3663 u64 last_cluster_blkno = bucket_blkno(first) +
3664 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
3666 BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
3667 BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
3669 mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3670 (unsigned long long)last_cluster_blkno, (unsigned long long)new_blkno);
3672 ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
3673 last_cluster_blkno, new_blkno,
3674 to_move, first_hash);
3680 /* This is the first bucket that got moved */
3681 src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
3684 * If the target bucket was part of the moved buckets, we need to
3685 * update first and target.
3687 if (bucket_blkno(target) >= src_blkno) {
3688 /* Find the block for the new target bucket */
3689 src_blkno = new_blkno +
3690 (bucket_blkno(target) - src_blkno);
3692 ocfs2_xattr_bucket_relse(first);
3693 ocfs2_xattr_bucket_relse(target);
3696 * These shouldn't fail - the buffers are in the
3697 * journal from ocfs2_cp_xattr_bucket().
3699 ret = ocfs2_read_xattr_bucket(first, new_blkno);
3704 ret = ocfs2_read_xattr_bucket(target, src_blkno);
3715 * Find the suitable pos when we divide a bucket into 2.
3716 * We have to make sure the xattrs with the same hash value exist
3717 * in the same bucket.
3719 * If this ocfs2_xattr_header covers more than one hash value, find a
3720 * place where the hash value changes. Try to find the most even split.
3721 * The most common case is that all entries have different hash values,
3722 * and the first check we make will find a place to split.
3724 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
3726 struct ocfs2_xattr_entry *entries = xh->xh_entries;
3727 int count = le16_to_cpu(xh->xh_count);
3728 int delta, middle = count / 2;
3731 * We start at the middle. Each step gets farther away in both
3732 * directions. We therefore hit the change in hash value
3733 * nearest to the middle. Note that this loop does not execute for
3736 for (delta = 0; delta < middle; delta++) {
3737 /* Let's check delta earlier than middle */
3738 if (cmp_xe(&entries[middle - delta - 1],
3739 &entries[middle - delta]))
3740 return middle - delta;
3742 /* For even counts, don't walk off the end */
3743 if ((middle + delta + 1) == count)
3746 /* Now try delta past middle */
3747 if (cmp_xe(&entries[middle + delta],
3748 &entries[middle + delta + 1]))
3749 return middle + delta + 1;
3752 /* Every entry had the same hash */
3757 * Move some xattrs in old bucket(blk) to new bucket(new_blk).
3758 * first_hash will record the 1st hash of the new bucket.
3760 * Normally half of the xattrs will be moved. But we have to make
3761 * sure that the xattrs with the same hash value are stored in the
3762 * same bucket. If all the xattrs in this bucket have the same hash
3763 * value, the new bucket will be initialized as an empty one and the
3764 * first_hash will be initialized as (hash_value+1).
3766 static int ocfs2_divide_xattr_bucket(struct inode *inode,
3771 int new_bucket_head)
3774 int count, start, len, name_value_len = 0, xe_len, name_offset = 0;
3775 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3776 struct ocfs2_xattr_header *xh;
3777 struct ocfs2_xattr_entry *xe;
3778 int blocksize = inode->i_sb->s_blocksize;
3780 mlog(0, "move some of xattrs from bucket %llu to %llu\n",
3781 (unsigned long long)blk, (unsigned long long)new_blk);
3783 s_bucket = ocfs2_xattr_bucket_new(inode);
3784 t_bucket = ocfs2_xattr_bucket_new(inode);
3785 if (!s_bucket || !t_bucket) {
3791 ret = ocfs2_read_xattr_bucket(s_bucket, blk);
3797 ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
3798 OCFS2_JOURNAL_ACCESS_WRITE);
3805 * Even if !new_bucket_head, we're overwriting t_bucket. Thus,
3806 * there's no need to read it.
3808 ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
3815 * Hey, if we're overwriting t_bucket, what difference does
3816 * ACCESS_CREATE vs ACCESS_WRITE make? See the comment in the
3817 * same part of ocfs2_cp_xattr_bucket().
3819 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3821 OCFS2_JOURNAL_ACCESS_CREATE :
3822 OCFS2_JOURNAL_ACCESS_WRITE);
3828 xh = bucket_xh(s_bucket);
3829 count = le16_to_cpu(xh->xh_count);
3830 start = ocfs2_xattr_find_divide_pos(xh);
3832 if (start == count) {
3833 xe = &xh->xh_entries[start-1];
3836 * initialized a new empty bucket here.
3837 * The hash value is set as one larger than
3838 * that of the last entry in the previous bucket.
3840 for (i = 0; i < t_bucket->bu_blocks; i++)
3841 memset(bucket_block(t_bucket, i), 0, blocksize);
3843 xh = bucket_xh(t_bucket);
3844 xh->xh_free_start = cpu_to_le16(blocksize);
3845 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
3846 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
3848 goto set_num_buckets;
3851 /* copy the whole bucket to the new first. */
3852 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
3854 /* update the new bucket. */
3855 xh = bucket_xh(t_bucket);
3858 * Calculate the total name/value len and xh_free_start for
3859 * the old bucket first.
3861 name_offset = OCFS2_XATTR_BUCKET_SIZE;
3863 for (i = 0; i < start; i++) {
3864 xe = &xh->xh_entries[i];
3865 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3866 if (ocfs2_xattr_is_local(xe))
3868 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3870 xe_len += OCFS2_XATTR_ROOT_SIZE;
3871 name_value_len += xe_len;
3872 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3873 name_offset = le16_to_cpu(xe->xe_name_offset);
3877 * Now begin the modification to the new bucket.
3879 * In the new bucket, We just move the xattr entry to the beginning
3880 * and don't touch the name/value. So there will be some holes in the
3881 * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3884 xe = &xh->xh_entries[start];
3885 len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3886 mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3887 (int)((char *)xe - (char *)xh),
3888 (int)((char *)xh->xh_entries - (char *)xh));
3889 memmove((char *)xh->xh_entries, (char *)xe, len);
3890 xe = &xh->xh_entries[count - start];
3891 len = sizeof(struct ocfs2_xattr_entry) * start;
3892 memset((char *)xe, 0, len);
3894 le16_add_cpu(&xh->xh_count, -start);
3895 le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3897 /* Calculate xh_free_start for the new bucket. */
3898 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3899 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3900 xe = &xh->xh_entries[i];
3901 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3902 if (ocfs2_xattr_is_local(xe))
3904 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3906 xe_len += OCFS2_XATTR_ROOT_SIZE;
3907 if (le16_to_cpu(xe->xe_name_offset) <
3908 le16_to_cpu(xh->xh_free_start))
3909 xh->xh_free_start = xe->xe_name_offset;
3913 /* set xh->xh_num_buckets for the new xh. */
3914 if (new_bucket_head)
3915 xh->xh_num_buckets = cpu_to_le16(1);
3917 xh->xh_num_buckets = 0;
3919 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
3921 /* store the first_hash of the new bucket. */
3923 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3926 * Now only update the 1st block of the old bucket. If we
3927 * just added a new empty bucket, there is no need to modify
3933 xh = bucket_xh(s_bucket);
3934 memset(&xh->xh_entries[start], 0,
3935 sizeof(struct ocfs2_xattr_entry) * (count - start));
3936 xh->xh_count = cpu_to_le16(start);
3937 xh->xh_free_start = cpu_to_le16(name_offset);
3938 xh->xh_name_value_len = cpu_to_le16(name_value_len);
3940 ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
3943 ocfs2_xattr_bucket_free(s_bucket);
3944 ocfs2_xattr_bucket_free(t_bucket);
3950 * Copy xattr from one bucket to another bucket.
3952 * The caller must make sure that the journal transaction
3953 * has enough space for journaling.
3955 static int ocfs2_cp_xattr_bucket(struct inode *inode,
3962 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3964 BUG_ON(s_blkno == t_blkno);
3966 mlog(0, "cp bucket %llu to %llu, target is %d\n",
3967 (unsigned long long)s_blkno, (unsigned long long)t_blkno,
3970 s_bucket = ocfs2_xattr_bucket_new(inode);
3971 t_bucket = ocfs2_xattr_bucket_new(inode);
3972 if (!s_bucket || !t_bucket) {
3978 ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
3983 * Even if !t_is_new, we're overwriting t_bucket. Thus,
3984 * there's no need to read it.
3986 ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
3991 * Hey, if we're overwriting t_bucket, what difference does
3992 * ACCESS_CREATE vs ACCESS_WRITE make? Well, if we allocated a new
3993 * cluster to fill, we came here from
3994 * ocfs2_mv_xattr_buckets(), and it is really new -
3995 * ACCESS_CREATE is required. But we also might have moved data
3996 * out of t_bucket before extending back into it.
3997 * ocfs2_add_new_xattr_bucket() can do this - its call to
3998 * ocfs2_add_new_xattr_cluster() may have created a new extent
3999 * and copied out the end of the old extent. Then it re-extends
4000 * the old extent back to create space for new xattrs. That's
4001 * how we get here, and the bucket isn't really new.
4003 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4005 OCFS2_JOURNAL_ACCESS_CREATE :
4006 OCFS2_JOURNAL_ACCESS_WRITE);
4010 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4011 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4014 ocfs2_xattr_bucket_free(t_bucket);
4015 ocfs2_xattr_bucket_free(s_bucket);
4021 * src_blk points to the start of an existing extent. last_blk points to
4022 * last cluster in that extent. to_blk points to a newly allocated
4023 * extent. We copy the buckets from the cluster at last_blk to the new
4024 * extent. If start_bucket is non-zero, we skip that many buckets before
4025 * we start copying. The new extent's xh_num_buckets gets set to the
4026 * number of buckets we copied. The old extent's xh_num_buckets shrinks
4027 * by the same amount.
4029 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4030 u64 src_blk, u64 last_blk, u64 to_blk,
4031 unsigned int start_bucket,
4034 int i, ret, credits;
4035 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4036 int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4037 int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4038 struct ocfs2_xattr_bucket *old_first, *new_first;
4040 mlog(0, "mv xattrs from cluster %llu to %llu\n",
4041 (unsigned long long)last_blk, (unsigned long long)to_blk);
4043 BUG_ON(start_bucket >= num_buckets);
4045 num_buckets -= start_bucket;
4046 last_blk += (start_bucket * blks_per_bucket);
4049 /* The first bucket of the original extent */
4050 old_first = ocfs2_xattr_bucket_new(inode);
4051 /* The first bucket of the new extent */
4052 new_first = ocfs2_xattr_bucket_new(inode);
4053 if (!old_first || !new_first) {
4059 ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4066 * We need to update the first bucket of the old extent and all
4067 * the buckets going to the new extent.
4069 credits = ((num_buckets + 1) * blks_per_bucket) +
4070 handle->h_buffer_credits;
4071 ret = ocfs2_extend_trans(handle, credits);
4077 ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4078 OCFS2_JOURNAL_ACCESS_WRITE);
4084 for (i = 0; i < num_buckets; i++) {
4085 ret = ocfs2_cp_xattr_bucket(inode, handle,
4086 last_blk + (i * blks_per_bucket),
4087 to_blk + (i * blks_per_bucket),
4096 * Get the new bucket ready before we dirty anything
4097 * (This actually shouldn't fail, because we already dirtied
4098 * it once in ocfs2_cp_xattr_bucket()).
4100 ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4105 ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4106 OCFS2_JOURNAL_ACCESS_WRITE);
4112 /* Now update the headers */
4113 le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4114 ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4116 bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4117 ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4120 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4123 ocfs2_xattr_bucket_free(new_first);
4124 ocfs2_xattr_bucket_free(old_first);
4129 * Move some xattrs in this cluster to the new cluster.
4130 * This function should only be called when bucket size == cluster size.
4131 * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4133 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4139 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4140 int ret, credits = 2 * blk_per_bucket + handle->h_buffer_credits;
4142 BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4144 ret = ocfs2_extend_trans(handle, credits);
4150 /* Move half of the xattr in start_blk to the next bucket. */
4151 return ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4152 new_blk, first_hash, 1);
4156 * Move some xattrs from the old cluster to the new one since they are not
4157 * contiguous in ocfs2 xattr tree.
4159 * new_blk starts a new separate cluster, and we will move some xattrs from
4160 * prev_blk to it. v_start will be set as the first name hash value in this
4161 * new cluster so that it can be used as e_cpos during tree insertion and
4162 * don't collide with our original b-tree operations. first_bh and header_bh
4163 * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4164 * to extend the insert bucket.
4166 * The problem is how much xattr should we move to the new one and when should
4167 * we update first_bh and header_bh?
4168 * 1. If cluster size > bucket size, that means the previous cluster has more
4169 * than 1 bucket, so just move half nums of bucket into the new cluster and
4170 * update the first_bh and header_bh if the insert bucket has been moved
4171 * to the new cluster.
4172 * 2. If cluster_size == bucket_size:
4173 * a) If the previous extent rec has more than one cluster and the insert
4174 * place isn't in the last cluster, copy the entire last cluster to the
4175 * new one. This time, we don't need to upate the first_bh and header_bh
4176 * since they will not be moved into the new cluster.
4177 * b) Otherwise, move the bottom half of the xattrs in the last cluster into
4178 * the new one. And we set the extend flag to zero if the insert place is
4179 * moved into the new allocated cluster since no extend is needed.
4181 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
4183 struct ocfs2_xattr_bucket *first,
4184 struct ocfs2_xattr_bucket *target,
4192 mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
4193 (unsigned long long)bucket_blkno(first), prev_clusters,
4194 (unsigned long long)new_blk);
4196 if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
4197 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
4206 /* The start of the last cluster in the first extent */
4207 u64 last_blk = bucket_blkno(first) +
4208 ((prev_clusters - 1) *
4209 ocfs2_clusters_to_blocks(inode->i_sb, 1));
4211 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
4212 ret = ocfs2_mv_xattr_buckets(inode, handle,
4213 bucket_blkno(first),
4214 last_blk, new_blk, 0,
4219 ret = ocfs2_divide_xattr_cluster(inode, handle,
4225 if ((bucket_blkno(target) == last_blk) && extend)
4234 * Add a new cluster for xattr storage.
4236 * If the new cluster is contiguous with the previous one, it will be
4237 * appended to the same extent record, and num_clusters will be updated.
4238 * If not, we will insert a new extent for it and move some xattrs in
4239 * the last cluster into the new allocated one.
4240 * We also need to limit the maximum size of a btree leaf, otherwise we'll
4241 * lose the benefits of hashing because we'll have to search large leaves.
4242 * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
4245 * first_bh is the first block of the previous extent rec and header_bh
4246 * indicates the bucket we will insert the new xattrs. They will be updated
4247 * when the header_bh is moved into the new cluster.
4249 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
4250 struct buffer_head *root_bh,
4251 struct ocfs2_xattr_bucket *first,
4252 struct ocfs2_xattr_bucket *target,
4256 struct ocfs2_xattr_set_ctxt *ctxt)
4259 u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
4260 u32 prev_clusters = *num_clusters;
4261 u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
4263 handle_t *handle = ctxt->handle;
4264 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4265 struct ocfs2_extent_tree et;
4267 mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
4268 "previous xattr blkno = %llu\n",
4269 (unsigned long long)OCFS2_I(inode)->ip_blkno,
4270 prev_cpos, (unsigned long long)bucket_blkno(first));
4272 ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4274 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
4275 OCFS2_JOURNAL_ACCESS_WRITE);
4281 ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac, 1,
4282 clusters_to_add, &bit_off, &num_bits);
4289 BUG_ON(num_bits > clusters_to_add);
4291 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
4292 mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
4293 num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
4295 if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
4296 (prev_clusters + num_bits) << osb->s_clustersize_bits <=
4297 OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
4299 * If this cluster is contiguous with the old one and
4300 * adding this new cluster, we don't surpass the limit of
4301 * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
4302 * initialized and used like other buckets in the previous
4304 * So add it as a contiguous one. The caller will handle
4307 v_start = prev_cpos + prev_clusters;
4308 *num_clusters = prev_clusters + num_bits;
4309 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
4312 ret = ocfs2_adjust_xattr_cross_cluster(inode,
4326 mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
4327 num_bits, (unsigned long long)block, v_start);
4328 ret = ocfs2_insert_extent(osb, handle, inode, &et, v_start, block,
4329 num_bits, 0, ctxt->meta_ac);
4335 ret = ocfs2_journal_dirty(handle, root_bh);
4344 * We are given an extent. 'first' is the bucket at the very front of
4345 * the extent. The extent has space for an additional bucket past
4346 * bucket_xh(first)->xh_num_buckets. 'target_blkno' is the block number
4347 * of the target bucket. We wish to shift every bucket past the target
4348 * down one, filling in that additional space. When we get back to the
4349 * target, we split the target between itself and the now-empty bucket
4350 * at target+1 (aka, target_blkno + blks_per_bucket).
4352 static int ocfs2_extend_xattr_bucket(struct inode *inode,
4354 struct ocfs2_xattr_bucket *first,
4359 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4360 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4362 u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
4364 mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
4365 "from %llu, len = %u\n", (unsigned long long)target_blk,
4366 (unsigned long long)bucket_blkno(first), num_clusters);
4368 /* The extent must have room for an additional bucket */
4369 BUG_ON(new_bucket >=
4370 (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
4372 /* end_blk points to the last existing bucket */
4373 end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
4376 * end_blk is the start of the last existing bucket.
4377 * Thus, (end_blk - target_blk) covers the target bucket and
4378 * every bucket after it up to, but not including, the last
4379 * existing bucket. Then we add the last existing bucket, the
4380 * new bucket, and the first bucket (3 * blk_per_bucket).
4382 credits = (end_blk - target_blk) + (3 * blk_per_bucket) +
4383 handle->h_buffer_credits;
4384 ret = ocfs2_extend_trans(handle, credits);
4390 ret = ocfs2_xattr_bucket_journal_access(handle, first,
4391 OCFS2_JOURNAL_ACCESS_WRITE);
4397 while (end_blk != target_blk) {
4398 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
4399 end_blk + blk_per_bucket, 0);
4402 end_blk -= blk_per_bucket;
4405 /* Move half of the xattr in target_blkno to the next bucket. */
4406 ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
4407 target_blk + blk_per_bucket, NULL, 0);
4409 le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
4410 ocfs2_xattr_bucket_journal_dirty(handle, first);
4417 * Add new xattr bucket in an extent record and adjust the buckets
4418 * accordingly. xb_bh is the ocfs2_xattr_block, and target is the
4419 * bucket we want to insert into.
4421 * In the easy case, we will move all the buckets after target down by
4422 * one. Half of target's xattrs will be moved to the next bucket.
4424 * If current cluster is full, we'll allocate a new one. This may not
4425 * be contiguous. The underlying calls will make sure that there is
4426 * space for the insert, shifting buckets around if necessary.
4427 * 'target' may be moved by those calls.
4429 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
4430 struct buffer_head *xb_bh,
4431 struct ocfs2_xattr_bucket *target,
4432 struct ocfs2_xattr_set_ctxt *ctxt)
4434 struct ocfs2_xattr_block *xb =
4435 (struct ocfs2_xattr_block *)xb_bh->b_data;
4436 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
4437 struct ocfs2_extent_list *el = &xb_root->xt_list;
4439 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
4440 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4441 int ret, num_buckets, extend = 1;
4443 u32 e_cpos, num_clusters;
4444 /* The bucket at the front of the extent */
4445 struct ocfs2_xattr_bucket *first;
4447 mlog(0, "Add new xattr bucket starting from %llu\n",
4448 (unsigned long long)bucket_blkno(target));
4450 /* The first bucket of the original extent */
4451 first = ocfs2_xattr_bucket_new(inode);
4458 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
4465 ret = ocfs2_read_xattr_bucket(first, p_blkno);
4471 num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
4472 if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
4474 * This can move first+target if the target bucket moves
4475 * to the new extent.
4477 ret = ocfs2_add_new_xattr_cluster(inode,
4492 ret = ocfs2_extend_xattr_bucket(inode,
4495 bucket_blkno(target),
4502 ocfs2_xattr_bucket_free(first);
4507 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
4508 struct ocfs2_xattr_bucket *bucket,
4511 int block_off = offs >> inode->i_sb->s_blocksize_bits;
4513 offs = offs % inode->i_sb->s_blocksize;
4514 return bucket_block(bucket, block_off) + offs;
4518 * Handle the normal xattr set, including replace, delete and new.
4520 * Note: "local" indicates the real data's locality. So we can't
4521 * just its bucket locality by its length.
4523 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
4524 struct ocfs2_xattr_info *xi,
4525 struct ocfs2_xattr_search *xs,
4529 struct ocfs2_xattr_entry *last, *xe;
4530 int name_len = strlen(xi->name);
4531 struct ocfs2_xattr_header *xh = xs->header;
4532 u16 count = le16_to_cpu(xh->xh_count), start;
4533 size_t blocksize = inode->i_sb->s_blocksize;
4535 size_t offs, size, new_size;
4537 last = &xh->xh_entries[count];
4538 if (!xs->not_found) {
4540 offs = le16_to_cpu(xe->xe_name_offset);
4541 if (ocfs2_xattr_is_local(xe))
4542 size = OCFS2_XATTR_SIZE(name_len) +
4543 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4545 size = OCFS2_XATTR_SIZE(name_len) +
4546 OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4549 * If the new value will be stored outside, xi->value has been
4550 * initalized as an empty ocfs2_xattr_value_root, and the same
4551 * goes with xi->value_len, so we can set new_size safely here.
4552 * See ocfs2_xattr_set_in_bucket.
4554 new_size = OCFS2_XATTR_SIZE(name_len) +
4555 OCFS2_XATTR_SIZE(xi->value_len);
4557 le16_add_cpu(&xh->xh_name_value_len, -size);
4559 if (new_size > size)
4560 goto set_new_name_value;
4562 /* Now replace the old value with new one. */
4564 xe->xe_value_size = cpu_to_le64(xi->value_len);
4566 xe->xe_value_size = 0;
4568 val = ocfs2_xattr_bucket_get_val(inode,
4570 memset(val + OCFS2_XATTR_SIZE(name_len), 0,
4571 size - OCFS2_XATTR_SIZE(name_len));
4572 if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
4573 memcpy(val + OCFS2_XATTR_SIZE(name_len),
4574 xi->value, xi->value_len);
4576 le16_add_cpu(&xh->xh_name_value_len, new_size);
4577 ocfs2_xattr_set_local(xe, local);
4581 * Remove the old entry if there is more than one.
4582 * We don't remove the last entry so that we can
4583 * use it to indicate the hash value of the empty
4587 le16_add_cpu(&xh->xh_count, -1);
4590 (void *)last - (void *)xe);
4592 sizeof(struct ocfs2_xattr_entry));
4595 cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4600 /* find a new entry for insert. */
4601 int low = 0, high = count - 1, tmp;
4602 struct ocfs2_xattr_entry *tmp_xe;
4604 while (low <= high && count) {
4605 tmp = (low + high) / 2;
4606 tmp_xe = &xh->xh_entries[tmp];
4608 if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
4610 else if (name_hash <
4611 le32_to_cpu(tmp_xe->xe_name_hash))
4619 xe = &xh->xh_entries[low];
4621 memmove(xe + 1, xe, (void *)last - (void *)xe);
4623 le16_add_cpu(&xh->xh_count, 1);
4624 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
4625 xe->xe_name_hash = cpu_to_le32(name_hash);
4626 xe->xe_name_len = name_len;
4627 ocfs2_xattr_set_type(xe, xi->name_index);
4631 /* Insert the new name+value. */
4632 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4635 * We must make sure that the name/value pair
4636 * exists in the same block.
4638 offs = le16_to_cpu(xh->xh_free_start);
4639 start = offs - size;
4641 if (start >> inode->i_sb->s_blocksize_bits !=
4642 (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4643 offs = offs - offs % blocksize;
4644 xh->xh_free_start = cpu_to_le16(offs);
4647 val = ocfs2_xattr_bucket_get_val(inode, xs->bucket, offs - size);
4648 xe->xe_name_offset = cpu_to_le16(offs - size);
4650 memset(val, 0, size);
4651 memcpy(val, xi->name, name_len);
4652 memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4654 xe->xe_value_size = cpu_to_le64(xi->value_len);
4655 ocfs2_xattr_set_local(xe, local);
4657 le16_add_cpu(&xh->xh_free_start, -size);
4658 le16_add_cpu(&xh->xh_name_value_len, size);
4664 * Set the xattr entry in the specified bucket.
4665 * The bucket is indicated by xs->bucket and it should have the enough
4666 * space for the xattr insertion.
4668 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4670 struct ocfs2_xattr_info *xi,
4671 struct ocfs2_xattr_search *xs,
4678 mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4679 (unsigned long)xi->value_len, xi->name_index,
4680 (unsigned long long)bucket_blkno(xs->bucket));
4682 if (!xs->bucket->bu_bhs[1]) {
4683 blkno = bucket_blkno(xs->bucket);
4684 ocfs2_xattr_bucket_relse(xs->bucket);
4685 ret = ocfs2_read_xattr_bucket(xs->bucket, blkno);
4692 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4693 OCFS2_JOURNAL_ACCESS_WRITE);
4699 ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local);
4700 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4707 * Truncate the specified xe_off entry in xattr bucket.
4708 * bucket is indicated by header_bh and len is the new length.
4709 * Both the ocfs2_xattr_value_root and the entry will be updated here.
4711 * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4713 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4714 struct ocfs2_xattr_bucket *bucket,
4717 struct ocfs2_xattr_set_ctxt *ctxt)
4721 struct ocfs2_xattr_entry *xe;
4722 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4723 size_t blocksize = inode->i_sb->s_blocksize;
4724 struct ocfs2_xattr_value_buf vb = {
4725 .vb_access = ocfs2_journal_access,
4728 xe = &xh->xh_entries[xe_off];
4730 BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4732 offset = le16_to_cpu(xe->xe_name_offset) +
4733 OCFS2_XATTR_SIZE(xe->xe_name_len);
4735 value_blk = offset / blocksize;
4737 /* We don't allow ocfs2_xattr_value to be stored in different block. */
4738 BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4740 vb.vb_bh = bucket->bu_bhs[value_blk];
4743 vb.vb_xv = (struct ocfs2_xattr_value_root *)
4744 (vb.vb_bh->b_data + offset % blocksize);
4747 * From here on out we have to dirty the bucket. The generic
4748 * value calls only modify one of the bucket's bhs, but we need
4749 * to send the bucket at once. So if they error, they *could* have
4750 * modified something. We have to assume they did, and dirty
4751 * the whole bucket. This leaves us in a consistent state.
4753 mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4754 xe_off, (unsigned long long)bucket_blkno(bucket), len);
4755 ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
4761 ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
4762 OCFS2_JOURNAL_ACCESS_WRITE);
4768 xe->xe_value_size = cpu_to_le64(len);
4770 ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
4776 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4777 struct ocfs2_xattr_search *xs,
4779 struct ocfs2_xattr_set_ctxt *ctxt)
4782 struct ocfs2_xattr_entry *xe = xs->here;
4783 struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4785 BUG_ON(!xs->bucket->bu_bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4787 offset = xe - xh->xh_entries;
4788 ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket,
4796 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4798 struct ocfs2_xattr_search *xs,
4802 int ret, offset, block_off;
4803 struct ocfs2_xattr_value_root *xv;
4804 struct ocfs2_xattr_entry *xe = xs->here;
4805 struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
4808 BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4810 ret = ocfs2_xattr_bucket_get_name_value(inode, xh,
4811 xe - xh->xh_entries,
4819 base = bucket_block(xs->bucket, block_off);
4820 xv = (struct ocfs2_xattr_value_root *)(base + offset +
4821 OCFS2_XATTR_SIZE(xe->xe_name_len));
4823 ret = __ocfs2_xattr_set_value_outside(inode, handle,
4824 xv, val, value_len);
4831 static int ocfs2_rm_xattr_cluster(struct inode *inode,
4832 struct buffer_head *root_bh,
4838 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4839 struct inode *tl_inode = osb->osb_tl_inode;
4841 struct ocfs2_xattr_block *xb =
4842 (struct ocfs2_xattr_block *)root_bh->b_data;
4843 struct ocfs2_alloc_context *meta_ac = NULL;
4844 struct ocfs2_cached_dealloc_ctxt dealloc;
4845 struct ocfs2_extent_tree et;
4847 ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4849 ocfs2_init_dealloc_ctxt(&dealloc);
4851 mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4852 cpos, len, (unsigned long long)blkno);
4854 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode), blkno,
4857 ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
4863 mutex_lock(&tl_inode->i_mutex);
4865 if (ocfs2_truncate_log_needs_flush(osb)) {
4866 ret = __ocfs2_flush_truncate_log(osb);
4873 handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
4874 if (IS_ERR(handle)) {
4880 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
4881 OCFS2_JOURNAL_ACCESS_WRITE);
4887 ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
4894 le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4896 ret = ocfs2_journal_dirty(handle, root_bh);
4902 ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4907 ocfs2_commit_trans(osb, handle);
4909 ocfs2_schedule_truncate_log_flush(osb, 1);
4911 mutex_unlock(&tl_inode->i_mutex);
4914 ocfs2_free_alloc_context(meta_ac);
4916 ocfs2_run_deallocs(osb, &dealloc);
4921 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4923 struct ocfs2_xattr_search *xs)
4925 struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
4926 struct ocfs2_xattr_entry *last = &xh->xh_entries[
4927 le16_to_cpu(xh->xh_count) - 1];
4930 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4931 OCFS2_JOURNAL_ACCESS_WRITE);
4937 /* Remove the old entry. */
4938 memmove(xs->here, xs->here + 1,
4939 (void *)last - (void *)xs->here);
4940 memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4941 le16_add_cpu(&xh->xh_count, -1);
4943 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4947 * Set the xattr name/value in the bucket specified in xs.
4949 * As the new value in xi may be stored in the bucket or in an outside cluster,
4950 * we divide the whole process into 3 steps:
4951 * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4952 * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4953 * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4954 * 4. If the clusters for the new outside value can't be allocated, we need
4955 * to free the xattr we allocated in set.
4957 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4958 struct ocfs2_xattr_info *xi,
4959 struct ocfs2_xattr_search *xs,
4960 struct ocfs2_xattr_set_ctxt *ctxt)
4964 char *val = (char *)xi->value;
4965 struct ocfs2_xattr_entry *xe = xs->here;
4966 u32 name_hash = ocfs2_xattr_name_hash(inode, xi->name,
4969 if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
4971 * We need to truncate the xattr storage first.
4973 * If both the old and new value are stored to
4974 * outside block, we only need to truncate
4975 * the storage and then set the value outside.
4977 * If the new value should be stored within block,
4978 * we should free all the outside block first and
4979 * the modification to the xattr block will be done
4980 * by following steps.
4982 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4983 value_len = xi->value_len;
4987 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4994 goto set_value_outside;
4997 value_len = xi->value_len;
4998 /* So we have to handle the inside block change now. */
4999 if (value_len > OCFS2_XATTR_INLINE_SIZE) {
5001 * If the new value will be stored outside of block,
5002 * initalize a new empty value root and insert it first.
5005 xi->value = &def_xv;
5006 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
5009 ret = ocfs2_xattr_set_entry_in_bucket(inode, ctxt->handle, xi, xs,
5016 if (value_len <= OCFS2_XATTR_INLINE_SIZE)
5019 /* allocate the space now for the outside block storage. */
5020 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
5025 if (xs->not_found) {
5027 * We can't allocate enough clusters for outside
5028 * storage and we have allocated xattr already,
5029 * so need to remove it.
5031 ocfs2_xattr_bucket_remove_xs(inode, ctxt->handle, xs);
5037 ret = ocfs2_xattr_bucket_set_value_outside(inode, ctxt->handle,
5038 xs, val, value_len);
5044 * check whether the xattr bucket is filled up with the same hash value.
5045 * If we want to insert the xattr with the same hash, return -ENOSPC.
5046 * If we want to insert a xattr with different hash value, go ahead
5047 * and ocfs2_divide_xattr_bucket will handle this.
5049 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
5050 struct ocfs2_xattr_bucket *bucket,
5053 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5054 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5056 if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5059 if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5060 xh->xh_entries[0].xe_name_hash) {
5061 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5063 (unsigned long long)bucket_blkno(bucket),
5064 le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5071 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5072 struct ocfs2_xattr_info *xi,
5073 struct ocfs2_xattr_search *xs,
5074 struct ocfs2_xattr_set_ctxt *ctxt)
5076 struct ocfs2_xattr_header *xh;
5077 struct ocfs2_xattr_entry *xe;
5078 u16 count, header_size, xh_free_start;
5079 int free, max_free, need, old;
5080 size_t value_size = 0, name_len = strlen(xi->name);
5081 size_t blocksize = inode->i_sb->s_blocksize;
5082 int ret, allocation = 0;
5084 mlog_entry("Set xattr %s in xattr index block\n", xi->name);
5088 count = le16_to_cpu(xh->xh_count);
5089 xh_free_start = le16_to_cpu(xh->xh_free_start);
5090 header_size = sizeof(struct ocfs2_xattr_header) +
5091 count * sizeof(struct ocfs2_xattr_entry);
5092 max_free = OCFS2_XATTR_BUCKET_SIZE - header_size -
5093 le16_to_cpu(xh->xh_name_value_len) - OCFS2_XATTR_HEADER_GAP;
5095 mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
5096 "of %u which exceed block size\n",
5097 (unsigned long long)bucket_blkno(xs->bucket),
5100 if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
5101 value_size = OCFS2_XATTR_ROOT_SIZE;
5103 value_size = OCFS2_XATTR_SIZE(xi->value_len);
5106 need = sizeof(struct ocfs2_xattr_entry) +
5107 OCFS2_XATTR_SIZE(name_len) + value_size;
5109 need = value_size + OCFS2_XATTR_SIZE(name_len);
5112 * We only replace the old value if the new length is smaller
5113 * than the old one. Otherwise we will allocate new space in the
5114 * bucket to store it.
5117 if (ocfs2_xattr_is_local(xe))
5118 old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
5120 old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
5122 if (old >= value_size)
5126 free = xh_free_start - header_size - OCFS2_XATTR_HEADER_GAP;
5128 * We need to make sure the new name/value pair
5129 * can exist in the same block.
5131 if (xh_free_start % blocksize < need)
5132 free -= xh_free_start % blocksize;
5134 mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
5135 "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
5136 " %u\n", xs->not_found,
5137 (unsigned long long)bucket_blkno(xs->bucket),
5138 free, need, max_free, le16_to_cpu(xh->xh_free_start),
5139 le16_to_cpu(xh->xh_name_value_len));
5143 count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb))) {
5144 if (need <= max_free &&
5145 count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
5147 * We can create the space by defragment. Since only the
5148 * name/value will be moved, the xe shouldn't be changed
5151 ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5158 xh_free_start = le16_to_cpu(xh->xh_free_start);
5159 free = xh_free_start - header_size
5160 - OCFS2_XATTR_HEADER_GAP;
5161 if (xh_free_start % blocksize < need)
5162 free -= xh_free_start % blocksize;
5167 mlog(0, "Can't get enough space for xattr insert by "
5168 "defragment. Need %u bytes, but we have %d, so "
5169 "allocate new bucket for it.\n", need, free);
5173 * We have to add new buckets or clusters and one
5174 * allocation should leave us enough space for insert.
5179 * We do not allow for overlapping ranges between buckets. And
5180 * the maximum number of collisions we will allow for then is
5181 * one bucket's worth, so check it here whether we need to
5182 * add a new bucket for the insert.
5184 ret = ocfs2_check_xattr_bucket_collision(inode,
5192 ret = ocfs2_add_new_xattr_bucket(inode,
5202 * ocfs2_add_new_xattr_bucket() will have updated
5203 * xs->bucket if it moved, but it will not have updated
5204 * any of the other search fields. Thus, we drop it and
5205 * re-search. Everything should be cached, so it'll be
5208 ocfs2_xattr_bucket_relse(xs->bucket);
5209 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5212 if (ret && ret != -ENODATA)
5214 xs->not_found = ret;
5220 ret = ocfs2_xattr_set_in_bucket(inode, xi, xs, ctxt);
5226 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5227 struct ocfs2_xattr_bucket *bucket,
5231 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5233 struct ocfs2_xattr_entry *xe;
5234 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5235 struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5236 int credits = ocfs2_remove_extent_credits(osb->sb) +
5237 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5240 ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5242 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5243 xe = &xh->xh_entries[i];
5244 if (ocfs2_xattr_is_local(xe))
5247 ctxt.handle = ocfs2_start_trans(osb, credits);
5248 if (IS_ERR(ctxt.handle)) {
5249 ret = PTR_ERR(ctxt.handle);
5254 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5257 ocfs2_commit_trans(osb, ctxt.handle);
5264 ocfs2_schedule_truncate_log_flush(osb, 1);
5265 ocfs2_run_deallocs(osb, &ctxt.dealloc);
5269 static int ocfs2_delete_xattr_index_block(struct inode *inode,
5270 struct buffer_head *xb_bh)
5272 struct ocfs2_xattr_block *xb =
5273 (struct ocfs2_xattr_block *)xb_bh->b_data;
5274 struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
5276 u32 name_hash = UINT_MAX, e_cpos, num_clusters;
5279 if (le16_to_cpu(el->l_next_free_rec) == 0)
5282 while (name_hash > 0) {
5283 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
5284 &e_cpos, &num_clusters, el);
5290 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
5291 ocfs2_delete_xattr_in_bucket,
5298 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
5299 p_blkno, e_cpos, num_clusters);
5308 name_hash = e_cpos - 1;
5316 * 'security' attributes support
5318 static size_t ocfs2_xattr_security_list(struct inode *inode, char *list,
5319 size_t list_size, const char *name,
5322 const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
5323 const size_t total_len = prefix_len + name_len + 1;
5325 if (list && total_len <= list_size) {
5326 memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
5327 memcpy(list + prefix_len, name, name_len);
5328 list[prefix_len + name_len] = '\0';
5333 static int ocfs2_xattr_security_get(struct inode *inode, const char *name,
5334 void *buffer, size_t size)
5336 if (strcmp(name, "") == 0)
5338 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_SECURITY, name,
5342 static int ocfs2_xattr_security_set(struct inode *inode, const char *name,
5343 const void *value, size_t size, int flags)
5345 if (strcmp(name, "") == 0)
5348 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY, name, value,
5352 int ocfs2_init_security_get(struct inode *inode,
5354 struct ocfs2_security_xattr_info *si)
5356 /* check whether ocfs2 support feature xattr */
5357 if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
5359 return security_inode_init_security(inode, dir, &si->name, &si->value,
5363 int ocfs2_init_security_set(handle_t *handle,
5364 struct inode *inode,
5365 struct buffer_head *di_bh,
5366 struct ocfs2_security_xattr_info *si,
5367 struct ocfs2_alloc_context *xattr_ac,
5368 struct ocfs2_alloc_context *data_ac)
5370 return ocfs2_xattr_set_handle(handle, inode, di_bh,
5371 OCFS2_XATTR_INDEX_SECURITY,
5372 si->name, si->value, si->value_len, 0,
5376 struct xattr_handler ocfs2_xattr_security_handler = {
5377 .prefix = XATTR_SECURITY_PREFIX,
5378 .list = ocfs2_xattr_security_list,
5379 .get = ocfs2_xattr_security_get,
5380 .set = ocfs2_xattr_security_set,
5384 * 'trusted' attributes support
5386 static size_t ocfs2_xattr_trusted_list(struct inode *inode, char *list,
5387 size_t list_size, const char *name,
5390 const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
5391 const size_t total_len = prefix_len + name_len + 1;
5393 if (list && total_len <= list_size) {
5394 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
5395 memcpy(list + prefix_len, name, name_len);
5396 list[prefix_len + name_len] = '\0';
5401 static int ocfs2_xattr_trusted_get(struct inode *inode, const char *name,
5402 void *buffer, size_t size)
5404 if (strcmp(name, "") == 0)
5406 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED, name,
5410 static int ocfs2_xattr_trusted_set(struct inode *inode, const char *name,
5411 const void *value, size_t size, int flags)
5413 if (strcmp(name, "") == 0)
5416 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED, name, value,
5420 struct xattr_handler ocfs2_xattr_trusted_handler = {
5421 .prefix = XATTR_TRUSTED_PREFIX,
5422 .list = ocfs2_xattr_trusted_list,
5423 .get = ocfs2_xattr_trusted_get,
5424 .set = ocfs2_xattr_trusted_set,
5428 * 'user' attributes support
5430 static size_t ocfs2_xattr_user_list(struct inode *inode, char *list,
5431 size_t list_size, const char *name,
5434 const size_t prefix_len = XATTR_USER_PREFIX_LEN;
5435 const size_t total_len = prefix_len + name_len + 1;
5436 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5438 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5441 if (list && total_len <= list_size) {
5442 memcpy(list, XATTR_USER_PREFIX, prefix_len);
5443 memcpy(list + prefix_len, name, name_len);
5444 list[prefix_len + name_len] = '\0';
5449 static int ocfs2_xattr_user_get(struct inode *inode, const char *name,
5450 void *buffer, size_t size)
5452 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5454 if (strcmp(name, "") == 0)
5456 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5458 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
5462 static int ocfs2_xattr_user_set(struct inode *inode, const char *name,
5463 const void *value, size_t size, int flags)
5465 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5467 if (strcmp(name, "") == 0)
5469 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5472 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER, name, value,
5476 struct xattr_handler ocfs2_xattr_user_handler = {
5477 .prefix = XATTR_USER_PREFIX,
5478 .list = ocfs2_xattr_user_list,
5479 .get = ocfs2_xattr_user_get,
5480 .set = ocfs2_xattr_user_set,