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 u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
601 struct ocfs2_extent_tree et;
603 mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
605 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
607 status = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
608 OCFS2_JOURNAL_ACCESS_WRITE);
614 prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
615 status = ocfs2_add_clusters_in_btree(handle,
628 status = ocfs2_journal_dirty(handle, vb->vb_bh);
634 clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) - prev_clusters;
637 * We should have already allocated enough space before the transaction,
638 * so no need to restart.
640 BUG_ON(why != RESTART_NONE || clusters_to_add);
647 static int __ocfs2_remove_xattr_range(struct inode *inode,
648 struct ocfs2_xattr_value_buf *vb,
649 u32 cpos, u32 phys_cpos, u32 len,
650 struct ocfs2_xattr_set_ctxt *ctxt)
653 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
654 handle_t *handle = ctxt->handle;
655 struct ocfs2_extent_tree et;
657 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
659 ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
660 OCFS2_JOURNAL_ACCESS_WRITE);
666 ret = ocfs2_remove_extent(handle, &et, cpos, len, ctxt->meta_ac,
673 le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
675 ret = ocfs2_journal_dirty(handle, vb->vb_bh);
681 ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc, phys_blkno, len);
689 static int ocfs2_xattr_shrink_size(struct inode *inode,
692 struct ocfs2_xattr_value_buf *vb,
693 struct ocfs2_xattr_set_ctxt *ctxt)
696 u32 trunc_len, cpos, phys_cpos, alloc_size;
699 if (old_clusters <= new_clusters)
703 trunc_len = old_clusters - new_clusters;
705 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
707 &vb->vb_xv->xr_list, NULL);
713 if (alloc_size > trunc_len)
714 alloc_size = trunc_len;
716 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
717 phys_cpos, alloc_size,
724 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
725 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode),
728 trunc_len -= alloc_size;
735 static int ocfs2_xattr_value_truncate(struct inode *inode,
736 struct ocfs2_xattr_value_buf *vb,
738 struct ocfs2_xattr_set_ctxt *ctxt)
741 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
742 u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
744 if (new_clusters == old_clusters)
747 if (new_clusters > old_clusters)
748 ret = ocfs2_xattr_extend_allocation(inode,
749 new_clusters - old_clusters,
752 ret = ocfs2_xattr_shrink_size(inode,
753 old_clusters, new_clusters,
759 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
760 size_t *result, const char *prefix,
761 const char *name, int name_len)
763 char *p = buffer + *result;
764 int prefix_len = strlen(prefix);
765 int total_len = prefix_len + name_len + 1;
767 *result += total_len;
769 /* we are just looking for how big our buffer needs to be */
776 memcpy(p, prefix, prefix_len);
777 memcpy(p + prefix_len, name, name_len);
778 p[prefix_len + name_len] = '\0';
783 static int ocfs2_xattr_list_entries(struct inode *inode,
784 struct ocfs2_xattr_header *header,
785 char *buffer, size_t buffer_size)
789 const char *prefix, *name;
791 for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
792 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
793 type = ocfs2_xattr_get_type(entry);
794 prefix = ocfs2_xattr_prefix(type);
797 name = (const char *)header +
798 le16_to_cpu(entry->xe_name_offset);
800 ret = ocfs2_xattr_list_entry(buffer, buffer_size,
801 &result, prefix, name,
811 static int ocfs2_xattr_ibody_list(struct inode *inode,
812 struct ocfs2_dinode *di,
816 struct ocfs2_xattr_header *header = NULL;
817 struct ocfs2_inode_info *oi = OCFS2_I(inode);
820 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
823 header = (struct ocfs2_xattr_header *)
824 ((void *)di + inode->i_sb->s_blocksize -
825 le16_to_cpu(di->i_xattr_inline_size));
827 ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
832 static int ocfs2_xattr_block_list(struct inode *inode,
833 struct ocfs2_dinode *di,
837 struct buffer_head *blk_bh = NULL;
838 struct ocfs2_xattr_block *xb;
841 if (!di->i_xattr_loc)
844 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
851 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
852 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
853 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
854 ret = ocfs2_xattr_list_entries(inode, header,
855 buffer, buffer_size);
857 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
858 ret = ocfs2_xattr_tree_list_index_block(inode, xt,
859 buffer, buffer_size);
867 ssize_t ocfs2_listxattr(struct dentry *dentry,
871 int ret = 0, i_ret = 0, b_ret = 0;
872 struct buffer_head *di_bh = NULL;
873 struct ocfs2_dinode *di = NULL;
874 struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
876 if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
879 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
882 ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
888 di = (struct ocfs2_dinode *)di_bh->b_data;
890 down_read(&oi->ip_xattr_sem);
891 i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
899 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
904 up_read(&oi->ip_xattr_sem);
905 ocfs2_inode_unlock(dentry->d_inode, 0);
909 return i_ret + b_ret;
912 static int ocfs2_xattr_find_entry(int name_index,
914 struct ocfs2_xattr_search *xs)
916 struct ocfs2_xattr_entry *entry;
923 name_len = strlen(name);
925 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
926 cmp = name_index - ocfs2_xattr_get_type(entry);
928 cmp = name_len - entry->xe_name_len;
930 cmp = memcmp(name, (xs->base +
931 le16_to_cpu(entry->xe_name_offset)),
939 return cmp ? -ENODATA : 0;
942 static int ocfs2_xattr_get_value_outside(struct inode *inode,
943 struct ocfs2_xattr_value_root *xv,
947 u32 cpos, p_cluster, num_clusters, bpc, clusters;
950 size_t cplen, blocksize;
951 struct buffer_head *bh = NULL;
952 struct ocfs2_extent_list *el;
955 clusters = le32_to_cpu(xv->xr_clusters);
956 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
957 blocksize = inode->i_sb->s_blocksize;
960 while (cpos < clusters) {
961 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
962 &num_clusters, el, NULL);
968 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
969 /* Copy ocfs2_xattr_value */
970 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
971 ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
978 cplen = len >= blocksize ? blocksize : len;
979 memcpy(buffer, bh->b_data, cplen);
988 cpos += num_clusters;
994 static int ocfs2_xattr_ibody_get(struct inode *inode,
999 struct ocfs2_xattr_search *xs)
1001 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1002 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1003 struct ocfs2_xattr_value_root *xv;
1007 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1010 xs->end = (void *)di + inode->i_sb->s_blocksize;
1011 xs->header = (struct ocfs2_xattr_header *)
1012 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1013 xs->base = (void *)xs->header;
1014 xs->here = xs->header->xh_entries;
1016 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1019 size = le64_to_cpu(xs->here->xe_value_size);
1021 if (size > buffer_size)
1023 if (ocfs2_xattr_is_local(xs->here)) {
1024 memcpy(buffer, (void *)xs->base +
1025 le16_to_cpu(xs->here->xe_name_offset) +
1026 OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1028 xv = (struct ocfs2_xattr_value_root *)
1029 (xs->base + le16_to_cpu(
1030 xs->here->xe_name_offset) +
1031 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1032 ret = ocfs2_xattr_get_value_outside(inode, xv,
1044 static int ocfs2_xattr_block_get(struct inode *inode,
1049 struct ocfs2_xattr_search *xs)
1051 struct ocfs2_xattr_block *xb;
1052 struct ocfs2_xattr_value_root *xv;
1054 int ret = -ENODATA, name_offset, name_len, i;
1055 int uninitialized_var(block_off);
1057 xs->bucket = ocfs2_xattr_bucket_new(inode);
1064 ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1070 if (xs->not_found) {
1075 xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1076 size = le64_to_cpu(xs->here->xe_value_size);
1079 if (size > buffer_size)
1082 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1083 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1084 i = xs->here - xs->header->xh_entries;
1086 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1087 ret = ocfs2_xattr_bucket_get_name_value(inode,
1088 bucket_xh(xs->bucket),
1092 xs->base = bucket_block(xs->bucket, block_off);
1094 if (ocfs2_xattr_is_local(xs->here)) {
1095 memcpy(buffer, (void *)xs->base +
1096 name_offset + name_len, size);
1098 xv = (struct ocfs2_xattr_value_root *)
1099 (xs->base + name_offset + name_len);
1100 ret = ocfs2_xattr_get_value_outside(inode, xv,
1110 ocfs2_xattr_bucket_free(xs->bucket);
1112 brelse(xs->xattr_bh);
1113 xs->xattr_bh = NULL;
1117 int ocfs2_xattr_get_nolock(struct inode *inode,
1118 struct buffer_head *di_bh,
1125 struct ocfs2_dinode *di = NULL;
1126 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1127 struct ocfs2_xattr_search xis = {
1128 .not_found = -ENODATA,
1130 struct ocfs2_xattr_search xbs = {
1131 .not_found = -ENODATA,
1134 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1137 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1140 xis.inode_bh = xbs.inode_bh = di_bh;
1141 di = (struct ocfs2_dinode *)di_bh->b_data;
1143 down_read(&oi->ip_xattr_sem);
1144 ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1146 if (ret == -ENODATA && di->i_xattr_loc)
1147 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1149 up_read(&oi->ip_xattr_sem);
1154 /* ocfs2_xattr_get()
1156 * Copy an extended attribute into the buffer provided.
1157 * Buffer is NULL to compute the size of buffer required.
1159 static int ocfs2_xattr_get(struct inode *inode,
1166 struct buffer_head *di_bh = NULL;
1168 ret = ocfs2_inode_lock(inode, &di_bh, 0);
1173 ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1174 name, buffer, buffer_size);
1176 ocfs2_inode_unlock(inode, 0);
1183 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1185 struct ocfs2_xattr_value_root *xv,
1189 int ret = 0, i, cp_len;
1190 u16 blocksize = inode->i_sb->s_blocksize;
1191 u32 p_cluster, num_clusters;
1192 u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1193 u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1195 struct buffer_head *bh = NULL;
1197 BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1199 while (cpos < clusters) {
1200 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1201 &num_clusters, &xv->xr_list,
1208 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1210 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1211 ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1218 ret = ocfs2_journal_access(handle,
1221 OCFS2_JOURNAL_ACCESS_WRITE);
1227 cp_len = value_len > blocksize ? blocksize : value_len;
1228 memcpy(bh->b_data, value, cp_len);
1229 value_len -= cp_len;
1231 if (cp_len < blocksize)
1232 memset(bh->b_data + cp_len, 0,
1233 blocksize - cp_len);
1235 ret = ocfs2_journal_dirty(handle, bh);
1244 * XXX: do we need to empty all the following
1245 * blocks in this cluster?
1250 cpos += num_clusters;
1258 static int ocfs2_xattr_cleanup(struct inode *inode,
1260 struct ocfs2_xattr_info *xi,
1261 struct ocfs2_xattr_search *xs,
1262 struct ocfs2_xattr_value_buf *vb,
1266 size_t name_len = strlen(xi->name);
1267 void *val = xs->base + offs;
1268 size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1270 ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
1271 OCFS2_JOURNAL_ACCESS_WRITE);
1276 /* Decrease xattr count */
1277 le16_add_cpu(&xs->header->xh_count, -1);
1278 /* Remove the xattr entry and tree root which has already be set*/
1279 memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
1280 memset(val, 0, size);
1282 ret = ocfs2_journal_dirty(handle, vb->vb_bh);
1289 static int ocfs2_xattr_update_entry(struct inode *inode,
1291 struct ocfs2_xattr_info *xi,
1292 struct ocfs2_xattr_search *xs,
1293 struct ocfs2_xattr_value_buf *vb,
1298 ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
1299 OCFS2_JOURNAL_ACCESS_WRITE);
1305 xs->here->xe_name_offset = cpu_to_le16(offs);
1306 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1307 if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1308 ocfs2_xattr_set_local(xs->here, 1);
1310 ocfs2_xattr_set_local(xs->here, 0);
1311 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1313 ret = ocfs2_journal_dirty(handle, vb->vb_bh);
1321 * ocfs2_xattr_set_value_outside()
1323 * Set large size value in B tree.
1325 static int ocfs2_xattr_set_value_outside(struct inode *inode,
1326 struct ocfs2_xattr_info *xi,
1327 struct ocfs2_xattr_search *xs,
1328 struct ocfs2_xattr_set_ctxt *ctxt,
1329 struct ocfs2_xattr_value_buf *vb,
1332 size_t name_len = strlen(xi->name);
1333 void *val = xs->base + offs;
1334 struct ocfs2_xattr_value_root *xv = NULL;
1335 size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1338 memset(val, 0, size);
1339 memcpy(val, xi->name, name_len);
1340 xv = (struct ocfs2_xattr_value_root *)
1341 (val + OCFS2_XATTR_SIZE(name_len));
1342 xv->xr_clusters = 0;
1343 xv->xr_last_eb_blk = 0;
1344 xv->xr_list.l_tree_depth = 0;
1345 xv->xr_list.l_count = cpu_to_le16(1);
1346 xv->xr_list.l_next_free_rec = 0;
1349 ret = ocfs2_xattr_value_truncate(inode, vb, xi->value_len, ctxt);
1354 ret = ocfs2_xattr_update_entry(inode, ctxt->handle, xi, xs, vb, offs);
1359 ret = __ocfs2_xattr_set_value_outside(inode, ctxt->handle, vb->vb_xv,
1360 xi->value, xi->value_len);
1368 * ocfs2_xattr_set_entry_local()
1370 * Set, replace or remove extended attribute in local.
1372 static void ocfs2_xattr_set_entry_local(struct inode *inode,
1373 struct ocfs2_xattr_info *xi,
1374 struct ocfs2_xattr_search *xs,
1375 struct ocfs2_xattr_entry *last,
1378 size_t name_len = strlen(xi->name);
1381 if (xi->value && xs->not_found) {
1382 /* Insert the new xattr entry. */
1383 le16_add_cpu(&xs->header->xh_count, 1);
1384 ocfs2_xattr_set_type(last, xi->name_index);
1385 ocfs2_xattr_set_local(last, 1);
1386 last->xe_name_len = name_len;
1392 first_val = xs->base + min_offs;
1393 offs = le16_to_cpu(xs->here->xe_name_offset);
1394 val = xs->base + offs;
1396 if (le64_to_cpu(xs->here->xe_value_size) >
1397 OCFS2_XATTR_INLINE_SIZE)
1398 size = OCFS2_XATTR_SIZE(name_len) +
1399 OCFS2_XATTR_ROOT_SIZE;
1401 size = OCFS2_XATTR_SIZE(name_len) +
1402 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1404 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1405 OCFS2_XATTR_SIZE(xi->value_len)) {
1406 /* The old and the new value have the
1407 same size. Just replace the value. */
1408 ocfs2_xattr_set_local(xs->here, 1);
1409 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1410 /* Clear value bytes. */
1411 memset(val + OCFS2_XATTR_SIZE(name_len),
1413 OCFS2_XATTR_SIZE(xi->value_len));
1414 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1419 /* Remove the old name+value. */
1420 memmove(first_val + size, first_val, val - first_val);
1421 memset(first_val, 0, size);
1422 xs->here->xe_name_hash = 0;
1423 xs->here->xe_name_offset = 0;
1424 ocfs2_xattr_set_local(xs->here, 1);
1425 xs->here->xe_value_size = 0;
1429 /* Adjust all value offsets. */
1430 last = xs->header->xh_entries;
1431 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1432 size_t o = le16_to_cpu(last->xe_name_offset);
1435 last->xe_name_offset = cpu_to_le16(o + size);
1440 /* Remove the old entry. */
1442 memmove(xs->here, xs->here + 1,
1443 (void *)last - (void *)xs->here);
1444 memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1445 le16_add_cpu(&xs->header->xh_count, -1);
1449 /* Insert the new name+value. */
1450 size_t size = OCFS2_XATTR_SIZE(name_len) +
1451 OCFS2_XATTR_SIZE(xi->value_len);
1452 void *val = xs->base + min_offs - size;
1454 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1455 memset(val, 0, size);
1456 memcpy(val, xi->name, name_len);
1457 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1460 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1461 ocfs2_xattr_set_local(xs->here, 1);
1462 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1469 * ocfs2_xattr_set_entry()
1471 * Set extended attribute entry into inode or block.
1473 * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1474 * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1475 * then set value in B tree with set_value_outside().
1477 static int ocfs2_xattr_set_entry(struct inode *inode,
1478 struct ocfs2_xattr_info *xi,
1479 struct ocfs2_xattr_search *xs,
1480 struct ocfs2_xattr_set_ctxt *ctxt,
1483 struct ocfs2_xattr_entry *last;
1484 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1485 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1486 size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1488 handle_t *handle = ctxt->handle;
1490 struct ocfs2_xattr_info xi_l = {
1491 .name_index = xi->name_index,
1494 .value_len = xi->value_len,
1496 struct ocfs2_xattr_value_buf vb = {
1497 .vb_bh = xs->xattr_bh,
1498 .vb_access = ocfs2_journal_access_di,
1501 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1502 BUG_ON(xs->xattr_bh == xs->inode_bh);
1503 vb.vb_access = ocfs2_journal_access_xb;
1505 BUG_ON(xs->xattr_bh != xs->inode_bh);
1507 /* Compute min_offs, last and free space. */
1508 last = xs->header->xh_entries;
1510 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1511 size_t offs = le16_to_cpu(last->xe_name_offset);
1512 if (offs < min_offs)
1517 free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
1521 if (!xs->not_found) {
1523 if (ocfs2_xattr_is_local(xs->here))
1524 size = OCFS2_XATTR_SIZE(name_len) +
1525 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1527 size = OCFS2_XATTR_SIZE(name_len) +
1528 OCFS2_XATTR_ROOT_SIZE;
1529 free += (size + sizeof(struct ocfs2_xattr_entry));
1531 /* Check free space in inode or block */
1532 if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1533 if (free < sizeof(struct ocfs2_xattr_entry) +
1534 OCFS2_XATTR_SIZE(name_len) +
1535 OCFS2_XATTR_ROOT_SIZE) {
1539 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1540 xi_l.value = (void *)&def_xv;
1541 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1542 } else if (xi->value) {
1543 if (free < sizeof(struct ocfs2_xattr_entry) +
1544 OCFS2_XATTR_SIZE(name_len) +
1545 OCFS2_XATTR_SIZE(xi->value_len)) {
1551 if (!xs->not_found) {
1552 /* For existing extended attribute */
1553 size_t size = OCFS2_XATTR_SIZE(name_len) +
1554 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1555 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1556 void *val = xs->base + offs;
1558 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1559 /* Replace existing local xattr with tree root */
1560 ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1565 } else if (!ocfs2_xattr_is_local(xs->here)) {
1566 /* For existing xattr which has value outside */
1567 vb.vb_xv = (struct ocfs2_xattr_value_root *)
1568 (val + OCFS2_XATTR_SIZE(name_len));
1570 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1572 * If new value need set outside also,
1573 * first truncate old value to new value,
1574 * then set new value with set_value_outside().
1576 ret = ocfs2_xattr_value_truncate(inode,
1585 ret = ocfs2_xattr_update_entry(inode,
1596 ret = __ocfs2_xattr_set_value_outside(inode,
1606 * If new value need set in local,
1607 * just trucate old value to zero.
1609 ret = ocfs2_xattr_value_truncate(inode,
1619 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), xs->inode_bh,
1620 OCFS2_JOURNAL_ACCESS_WRITE);
1626 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1627 ret = vb.vb_access(handle, INODE_CACHE(inode), vb.vb_bh,
1628 OCFS2_JOURNAL_ACCESS_WRITE);
1636 * Set value in local, include set tree root in local.
1637 * This is the first step for value size >INLINE_SIZE.
1639 ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1641 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1642 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1649 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1650 (flag & OCFS2_INLINE_XATTR_FL)) {
1651 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1652 unsigned int xattrsize = osb->s_xattr_inline_size;
1655 * Adjust extent record count or inline data size
1656 * to reserve space for extended attribute.
1658 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1659 struct ocfs2_inline_data *idata = &di->id2.i_data;
1660 le16_add_cpu(&idata->id_count, -xattrsize);
1661 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1662 struct ocfs2_extent_list *el = &di->id2.i_list;
1663 le16_add_cpu(&el->l_count, -(xattrsize /
1664 sizeof(struct ocfs2_extent_rec)));
1666 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1668 /* Update xattr flag */
1669 spin_lock(&oi->ip_lock);
1670 oi->ip_dyn_features |= flag;
1671 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1672 spin_unlock(&oi->ip_lock);
1674 ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1678 if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1680 * Set value outside in B tree.
1681 * This is the second step for value size > INLINE_SIZE.
1683 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1684 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, ctxt,
1691 * If set value outside failed, we have to clean
1692 * the junk tree root we have already set in local.
1694 ret2 = ocfs2_xattr_cleanup(inode, ctxt->handle,
1704 static int ocfs2_remove_value_outside(struct inode*inode,
1705 struct ocfs2_xattr_value_buf *vb,
1706 struct ocfs2_xattr_header *header)
1709 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1710 struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
1712 ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
1714 ctxt.handle = ocfs2_start_trans(osb,
1715 ocfs2_remove_extent_credits(osb->sb));
1716 if (IS_ERR(ctxt.handle)) {
1717 ret = PTR_ERR(ctxt.handle);
1722 for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1723 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1725 if (!ocfs2_xattr_is_local(entry)) {
1728 val = (void *)header +
1729 le16_to_cpu(entry->xe_name_offset);
1730 vb->vb_xv = (struct ocfs2_xattr_value_root *)
1731 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
1732 ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
1740 ocfs2_commit_trans(osb, ctxt.handle);
1741 ocfs2_schedule_truncate_log_flush(osb, 1);
1742 ocfs2_run_deallocs(osb, &ctxt.dealloc);
1747 static int ocfs2_xattr_ibody_remove(struct inode *inode,
1748 struct buffer_head *di_bh)
1751 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1752 struct ocfs2_xattr_header *header;
1754 struct ocfs2_xattr_value_buf vb = {
1756 .vb_access = ocfs2_journal_access_di,
1759 header = (struct ocfs2_xattr_header *)
1760 ((void *)di + inode->i_sb->s_blocksize -
1761 le16_to_cpu(di->i_xattr_inline_size));
1763 ret = ocfs2_remove_value_outside(inode, &vb, header);
1768 static int ocfs2_xattr_block_remove(struct inode *inode,
1769 struct buffer_head *blk_bh)
1771 struct ocfs2_xattr_block *xb;
1773 struct ocfs2_xattr_value_buf vb = {
1775 .vb_access = ocfs2_journal_access_xb,
1778 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1779 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1780 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1781 ret = ocfs2_remove_value_outside(inode, &vb, header);
1783 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
1788 static int ocfs2_xattr_free_block(struct inode *inode,
1791 struct inode *xb_alloc_inode;
1792 struct buffer_head *xb_alloc_bh = NULL;
1793 struct buffer_head *blk_bh = NULL;
1794 struct ocfs2_xattr_block *xb;
1795 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1801 ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
1807 ret = ocfs2_xattr_block_remove(inode, blk_bh);
1813 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1814 blk = le64_to_cpu(xb->xb_blkno);
1815 bit = le16_to_cpu(xb->xb_suballoc_bit);
1816 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1818 xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1819 EXTENT_ALLOC_SYSTEM_INODE,
1820 le16_to_cpu(xb->xb_suballoc_slot));
1821 if (!xb_alloc_inode) {
1826 mutex_lock(&xb_alloc_inode->i_mutex);
1828 ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1834 handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
1835 if (IS_ERR(handle)) {
1836 ret = PTR_ERR(handle);
1841 ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1846 ocfs2_commit_trans(osb, handle);
1848 ocfs2_inode_unlock(xb_alloc_inode, 1);
1849 brelse(xb_alloc_bh);
1851 mutex_unlock(&xb_alloc_inode->i_mutex);
1852 iput(xb_alloc_inode);
1859 * ocfs2_xattr_remove()
1861 * Free extended attribute resources associated with this inode.
1863 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1865 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1866 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1870 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1873 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1876 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1877 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1884 if (di->i_xattr_loc) {
1885 ret = ocfs2_xattr_free_block(inode,
1886 le64_to_cpu(di->i_xattr_loc));
1893 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1894 OCFS2_INODE_UPDATE_CREDITS);
1895 if (IS_ERR(handle)) {
1896 ret = PTR_ERR(handle);
1900 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
1901 OCFS2_JOURNAL_ACCESS_WRITE);
1907 di->i_xattr_loc = 0;
1909 spin_lock(&oi->ip_lock);
1910 oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1911 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1912 spin_unlock(&oi->ip_lock);
1914 ret = ocfs2_journal_dirty(handle, di_bh);
1918 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1923 static int ocfs2_xattr_has_space_inline(struct inode *inode,
1924 struct ocfs2_dinode *di)
1926 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1927 unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1930 if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1933 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1934 struct ocfs2_inline_data *idata = &di->id2.i_data;
1935 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1936 } else if (ocfs2_inode_is_fast_symlink(inode)) {
1937 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1938 le64_to_cpu(di->i_size);
1940 struct ocfs2_extent_list *el = &di->id2.i_list;
1941 free = (le16_to_cpu(el->l_count) -
1942 le16_to_cpu(el->l_next_free_rec)) *
1943 sizeof(struct ocfs2_extent_rec);
1945 if (free >= xattrsize)
1952 * ocfs2_xattr_ibody_find()
1954 * Find extended attribute in inode block and
1955 * fill search info into struct ocfs2_xattr_search.
1957 static int ocfs2_xattr_ibody_find(struct inode *inode,
1960 struct ocfs2_xattr_search *xs)
1962 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1963 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1967 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1970 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1971 down_read(&oi->ip_alloc_sem);
1972 has_space = ocfs2_xattr_has_space_inline(inode, di);
1973 up_read(&oi->ip_alloc_sem);
1978 xs->xattr_bh = xs->inode_bh;
1979 xs->end = (void *)di + inode->i_sb->s_blocksize;
1980 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1981 xs->header = (struct ocfs2_xattr_header *)
1982 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1984 xs->header = (struct ocfs2_xattr_header *)
1985 (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
1986 xs->base = (void *)xs->header;
1987 xs->here = xs->header->xh_entries;
1989 /* Find the named attribute. */
1990 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1991 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1992 if (ret && ret != -ENODATA)
1994 xs->not_found = ret;
2001 * ocfs2_xattr_ibody_set()
2003 * Set, replace or remove an extended attribute into inode block.
2006 static int ocfs2_xattr_ibody_set(struct inode *inode,
2007 struct ocfs2_xattr_info *xi,
2008 struct ocfs2_xattr_search *xs,
2009 struct ocfs2_xattr_set_ctxt *ctxt)
2011 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2012 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2015 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2018 down_write(&oi->ip_alloc_sem);
2019 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2020 if (!ocfs2_xattr_has_space_inline(inode, di)) {
2026 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2027 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
2029 up_write(&oi->ip_alloc_sem);
2035 * ocfs2_xattr_block_find()
2037 * Find extended attribute in external block and
2038 * fill search info into struct ocfs2_xattr_search.
2040 static int ocfs2_xattr_block_find(struct inode *inode,
2043 struct ocfs2_xattr_search *xs)
2045 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2046 struct buffer_head *blk_bh = NULL;
2047 struct ocfs2_xattr_block *xb;
2050 if (!di->i_xattr_loc)
2053 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2060 xs->xattr_bh = blk_bh;
2061 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2063 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2064 xs->header = &xb->xb_attrs.xb_header;
2065 xs->base = (void *)xs->header;
2066 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2067 xs->here = xs->header->xh_entries;
2069 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2071 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2075 if (ret && ret != -ENODATA) {
2076 xs->xattr_bh = NULL;
2079 xs->not_found = ret;
2088 * ocfs2_xattr_block_set()
2090 * Set, replace or remove an extended attribute into external block.
2093 static int ocfs2_xattr_block_set(struct inode *inode,
2094 struct ocfs2_xattr_info *xi,
2095 struct ocfs2_xattr_search *xs,
2096 struct ocfs2_xattr_set_ctxt *ctxt)
2098 struct buffer_head *new_bh = NULL;
2099 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2100 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2101 handle_t *handle = ctxt->handle;
2102 struct ocfs2_xattr_block *xblk = NULL;
2103 u16 suballoc_bit_start;
2108 if (!xs->xattr_bh) {
2109 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2111 OCFS2_JOURNAL_ACCESS_CREATE);
2117 ret = ocfs2_claim_metadata(osb, handle, ctxt->meta_ac, 1,
2118 &suballoc_bit_start, &num_got,
2125 new_bh = sb_getblk(inode->i_sb, first_blkno);
2126 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2128 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode),
2130 OCFS2_JOURNAL_ACCESS_CREATE);
2136 /* Initialize ocfs2_xattr_block */
2137 xs->xattr_bh = new_bh;
2138 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2139 memset(xblk, 0, inode->i_sb->s_blocksize);
2140 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2141 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
2142 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2143 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
2144 xblk->xb_blkno = cpu_to_le64(first_blkno);
2146 xs->header = &xblk->xb_attrs.xb_header;
2147 xs->base = (void *)xs->header;
2148 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2149 xs->here = xs->header->xh_entries;
2151 ret = ocfs2_journal_dirty(handle, new_bh);
2156 di->i_xattr_loc = cpu_to_le64(first_blkno);
2157 ocfs2_journal_dirty(handle, xs->inode_bh);
2159 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2161 if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2162 /* Set extended attribute into external block */
2163 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2164 OCFS2_HAS_XATTR_FL);
2165 if (!ret || ret != -ENOSPC)
2168 ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2173 ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2180 /* Check whether the new xattr can be inserted into the inode. */
2181 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2182 struct ocfs2_xattr_info *xi,
2183 struct ocfs2_xattr_search *xs)
2186 struct ocfs2_xattr_entry *last;
2188 size_t min_offs = xs->end - xs->base;
2193 last = xs->header->xh_entries;
2195 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
2196 size_t offs = le16_to_cpu(last->xe_name_offset);
2197 if (offs < min_offs)
2202 free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
2206 BUG_ON(!xs->not_found);
2208 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2209 value_size = OCFS2_XATTR_ROOT_SIZE;
2211 value_size = OCFS2_XATTR_SIZE(xi->value_len);
2213 if (free >= sizeof(struct ocfs2_xattr_entry) +
2214 OCFS2_XATTR_SIZE(strlen(xi->name)) + value_size)
2220 static int ocfs2_calc_xattr_set_need(struct inode *inode,
2221 struct ocfs2_dinode *di,
2222 struct ocfs2_xattr_info *xi,
2223 struct ocfs2_xattr_search *xis,
2224 struct ocfs2_xattr_search *xbs,
2229 int ret = 0, old_in_xb = 0;
2230 int clusters_add = 0, meta_add = 0, credits = 0;
2231 struct buffer_head *bh = NULL;
2232 struct ocfs2_xattr_block *xb = NULL;
2233 struct ocfs2_xattr_entry *xe = NULL;
2234 struct ocfs2_xattr_value_root *xv = NULL;
2236 int name_offset, name_len = 0;
2237 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2242 * Calculate the clusters we need to write.
2243 * No matter whether we replace an old one or add a new one,
2244 * we need this for writing.
2246 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2247 credits += new_clusters *
2248 ocfs2_clusters_to_blocks(inode->i_sb, 1);
2250 if (xis->not_found && xbs->not_found) {
2251 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2253 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2254 clusters_add += new_clusters;
2255 credits += ocfs2_calc_extend_credits(inode->i_sb,
2263 if (!xis->not_found) {
2265 name_offset = le16_to_cpu(xe->xe_name_offset);
2266 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2268 credits += OCFS2_INODE_UPDATE_CREDITS;
2270 int i, block_off = 0;
2271 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2273 name_offset = le16_to_cpu(xe->xe_name_offset);
2274 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2275 i = xbs->here - xbs->header->xh_entries;
2278 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2279 ret = ocfs2_xattr_bucket_get_name_value(inode,
2280 bucket_xh(xbs->bucket),
2283 base = bucket_block(xbs->bucket, block_off);
2284 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2287 credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
2292 * delete a xattr doesn't need metadata and cluster allocation.
2293 * so just calculate the credits and return.
2295 * The credits for removing the value tree will be extended
2296 * by ocfs2_remove_extent itself.
2299 if (!ocfs2_xattr_is_local(xe))
2300 credits += ocfs2_remove_extent_credits(inode->i_sb);
2305 /* do cluster allocation guess first. */
2306 value_size = le64_to_cpu(xe->xe_value_size);
2310 * In xattr set, we always try to set the xe in inode first,
2311 * so if it can be inserted into inode successfully, the old
2312 * one will be removed from the xattr block, and this xattr
2313 * will be inserted into inode as a new xattr in inode.
2315 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
2316 clusters_add += new_clusters;
2317 credits += ocfs2_remove_extent_credits(inode->i_sb) +
2318 OCFS2_INODE_UPDATE_CREDITS;
2319 if (!ocfs2_xattr_is_local(xe))
2320 credits += ocfs2_calc_extend_credits(
2328 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2329 /* the new values will be stored outside. */
2330 u32 old_clusters = 0;
2332 if (!ocfs2_xattr_is_local(xe)) {
2333 old_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2335 xv = (struct ocfs2_xattr_value_root *)
2336 (base + name_offset + name_len);
2337 value_size = OCFS2_XATTR_ROOT_SIZE;
2341 if (old_clusters >= new_clusters) {
2342 credits += ocfs2_remove_extent_credits(inode->i_sb);
2345 meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
2346 clusters_add += new_clusters - old_clusters;
2347 credits += ocfs2_calc_extend_credits(inode->i_sb,
2351 if (value_size >= OCFS2_XATTR_ROOT_SIZE)
2356 * Now the new value will be stored inside. So if the new
2357 * value is smaller than the size of value root or the old
2358 * value, we don't need any allocation, otherwise we have
2359 * to guess metadata allocation.
2361 if ((ocfs2_xattr_is_local(xe) && value_size >= xi->value_len) ||
2362 (!ocfs2_xattr_is_local(xe) &&
2363 OCFS2_XATTR_ROOT_SIZE >= xi->value_len))
2368 /* calculate metadata allocation. */
2369 if (di->i_xattr_loc) {
2370 if (!xbs->xattr_bh) {
2371 ret = ocfs2_read_xattr_block(inode,
2372 le64_to_cpu(di->i_xattr_loc),
2379 xb = (struct ocfs2_xattr_block *)bh->b_data;
2381 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2384 * If there is already an xattr tree, good, we can calculate
2385 * like other b-trees. Otherwise we may have the chance of
2386 * create a tree, the credit calculation is borrowed from
2387 * ocfs2_calc_extend_credits with root_el = NULL. And the
2388 * new tree will be cluster based, so no meta is needed.
2390 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2391 struct ocfs2_extent_list *el =
2392 &xb->xb_attrs.xb_root.xt_list;
2393 meta_add += ocfs2_extend_meta_needed(el);
2394 credits += ocfs2_calc_extend_credits(inode->i_sb,
2397 credits += OCFS2_SUBALLOC_ALLOC + 1;
2400 * This cluster will be used either for new bucket or for
2402 * If the cluster size is the same as the bucket size, one
2403 * more is needed since we may need to extend the bucket
2407 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2408 if (OCFS2_XATTR_BUCKET_SIZE ==
2409 OCFS2_SB(inode->i_sb)->s_clustersize) {
2410 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2415 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
2419 *clusters_need = clusters_add;
2421 *meta_need = meta_add;
2423 *credits_need = credits;
2428 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
2429 struct ocfs2_dinode *di,
2430 struct ocfs2_xattr_info *xi,
2431 struct ocfs2_xattr_search *xis,
2432 struct ocfs2_xattr_search *xbs,
2433 struct ocfs2_xattr_set_ctxt *ctxt,
2436 int clusters_add, meta_add, ret;
2437 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2439 memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
2441 ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
2443 ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
2444 &clusters_add, &meta_add, credits);
2450 mlog(0, "Set xattr %s, reserve meta blocks = %d, clusters = %d, "
2451 "credits = %d\n", xi->name, meta_add, clusters_add, *credits);
2454 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
2463 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
2469 if (ctxt->meta_ac) {
2470 ocfs2_free_alloc_context(ctxt->meta_ac);
2471 ctxt->meta_ac = NULL;
2475 * We cannot have an error and a non null ctxt->data_ac.
2482 static int __ocfs2_xattr_set_handle(struct inode *inode,
2483 struct ocfs2_dinode *di,
2484 struct ocfs2_xattr_info *xi,
2485 struct ocfs2_xattr_search *xis,
2486 struct ocfs2_xattr_search *xbs,
2487 struct ocfs2_xattr_set_ctxt *ctxt)
2489 int ret = 0, credits, old_found;
2492 /* Remove existing extended attribute */
2493 if (!xis->not_found)
2494 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2495 else if (!xbs->not_found)
2496 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2498 /* We always try to set extended attribute into inode first*/
2499 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2500 if (!ret && !xbs->not_found) {
2502 * If succeed and that extended attribute existing in
2503 * external block, then we will remove it.
2508 old_found = xis->not_found;
2509 xis->not_found = -ENODATA;
2510 ret = ocfs2_calc_xattr_set_need(inode,
2518 xis->not_found = old_found;
2524 ret = ocfs2_extend_trans(ctxt->handle, credits +
2525 ctxt->handle->h_buffer_credits);
2530 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2531 } else if (ret == -ENOSPC) {
2532 if (di->i_xattr_loc && !xbs->xattr_bh) {
2533 ret = ocfs2_xattr_block_find(inode,
2539 old_found = xis->not_found;
2540 xis->not_found = -ENODATA;
2541 ret = ocfs2_calc_xattr_set_need(inode,
2549 xis->not_found = old_found;
2555 ret = ocfs2_extend_trans(ctxt->handle, credits +
2556 ctxt->handle->h_buffer_credits);
2563 * If no space in inode, we will set extended attribute
2564 * into external block.
2566 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2569 if (!xis->not_found) {
2571 * If succeed and that extended attribute
2572 * existing in inode, we will remove it.
2576 xbs->not_found = -ENODATA;
2577 ret = ocfs2_calc_xattr_set_need(inode,
2590 ret = ocfs2_extend_trans(ctxt->handle, credits +
2591 ctxt->handle->h_buffer_credits);
2596 ret = ocfs2_xattr_ibody_set(inode, xi,
2603 /* Update inode ctime. */
2604 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
2606 OCFS2_JOURNAL_ACCESS_WRITE);
2612 inode->i_ctime = CURRENT_TIME;
2613 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
2614 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
2615 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
2622 * This function only called duing creating inode
2623 * for init security/acl xattrs of the new inode.
2624 * All transanction credits have been reserved in mknod.
2626 int ocfs2_xattr_set_handle(handle_t *handle,
2627 struct inode *inode,
2628 struct buffer_head *di_bh,
2634 struct ocfs2_alloc_context *meta_ac,
2635 struct ocfs2_alloc_context *data_ac)
2637 struct ocfs2_dinode *di;
2640 struct ocfs2_xattr_info xi = {
2641 .name_index = name_index,
2644 .value_len = value_len,
2647 struct ocfs2_xattr_search xis = {
2648 .not_found = -ENODATA,
2651 struct ocfs2_xattr_search xbs = {
2652 .not_found = -ENODATA,
2655 struct ocfs2_xattr_set_ctxt ctxt = {
2661 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2665 * In extreme situation, may need xattr bucket when
2666 * block size is too small. And we have already reserved
2667 * the credits for bucket in mknod.
2669 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
2670 xbs.bucket = ocfs2_xattr_bucket_new(inode);
2672 mlog_errno(-ENOMEM);
2677 xis.inode_bh = xbs.inode_bh = di_bh;
2678 di = (struct ocfs2_dinode *)di_bh->b_data;
2680 down_write(&OCFS2_I(inode)->ip_xattr_sem);
2682 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2685 if (xis.not_found) {
2686 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2691 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2694 up_write(&OCFS2_I(inode)->ip_xattr_sem);
2695 brelse(xbs.xattr_bh);
2696 ocfs2_xattr_bucket_free(xbs.bucket);
2704 * Set, replace or remove an extended attribute for this inode.
2705 * value is NULL to remove an existing extended attribute, else either
2706 * create or replace an extended attribute.
2708 int ocfs2_xattr_set(struct inode *inode,
2715 struct buffer_head *di_bh = NULL;
2716 struct ocfs2_dinode *di;
2718 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2719 struct inode *tl_inode = osb->osb_tl_inode;
2720 struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2722 struct ocfs2_xattr_info xi = {
2723 .name_index = name_index,
2726 .value_len = value_len,
2729 struct ocfs2_xattr_search xis = {
2730 .not_found = -ENODATA,
2733 struct ocfs2_xattr_search xbs = {
2734 .not_found = -ENODATA,
2737 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2741 * Only xbs will be used on indexed trees. xis doesn't need a
2744 xbs.bucket = ocfs2_xattr_bucket_new(inode);
2746 mlog_errno(-ENOMEM);
2750 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2753 goto cleanup_nolock;
2755 xis.inode_bh = xbs.inode_bh = di_bh;
2756 di = (struct ocfs2_dinode *)di_bh->b_data;
2758 down_write(&OCFS2_I(inode)->ip_xattr_sem);
2760 * Scan inode and external block to find the same name
2761 * extended attribute and collect search infomation.
2763 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2766 if (xis.not_found) {
2767 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2772 if (xis.not_found && xbs.not_found) {
2774 if (flags & XATTR_REPLACE)
2781 if (flags & XATTR_CREATE)
2786 mutex_lock(&tl_inode->i_mutex);
2788 if (ocfs2_truncate_log_needs_flush(osb)) {
2789 ret = __ocfs2_flush_truncate_log(osb);
2791 mutex_unlock(&tl_inode->i_mutex);
2796 mutex_unlock(&tl_inode->i_mutex);
2798 ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
2799 &xbs, &ctxt, &credits);
2805 /* we need to update inode's ctime field, so add credit for it. */
2806 credits += OCFS2_INODE_UPDATE_CREDITS;
2807 ctxt.handle = ocfs2_start_trans(osb, credits);
2808 if (IS_ERR(ctxt.handle)) {
2809 ret = PTR_ERR(ctxt.handle);
2814 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2816 ocfs2_commit_trans(osb, ctxt.handle);
2819 ocfs2_free_alloc_context(ctxt.data_ac);
2821 ocfs2_free_alloc_context(ctxt.meta_ac);
2822 if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
2823 ocfs2_schedule_truncate_log_flush(osb, 1);
2824 ocfs2_run_deallocs(osb, &ctxt.dealloc);
2826 up_write(&OCFS2_I(inode)->ip_xattr_sem);
2827 ocfs2_inode_unlock(inode, 1);
2830 brelse(xbs.xattr_bh);
2831 ocfs2_xattr_bucket_free(xbs.bucket);
2837 * Find the xattr extent rec which may contains name_hash.
2838 * e_cpos will be the first name hash of the xattr rec.
2839 * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2841 static int ocfs2_xattr_get_rec(struct inode *inode,
2846 struct ocfs2_extent_list *el)
2849 struct buffer_head *eb_bh = NULL;
2850 struct ocfs2_extent_block *eb;
2851 struct ocfs2_extent_rec *rec = NULL;
2854 if (el->l_tree_depth) {
2855 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, name_hash,
2862 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2865 if (el->l_tree_depth) {
2866 ocfs2_error(inode->i_sb,
2867 "Inode %lu has non zero tree depth in "
2868 "xattr tree block %llu\n", inode->i_ino,
2869 (unsigned long long)eb_bh->b_blocknr);
2875 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2876 rec = &el->l_recs[i];
2878 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2879 e_blkno = le64_to_cpu(rec->e_blkno);
2885 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2886 "record (%u, %u, 0) in xattr", inode->i_ino,
2887 le32_to_cpu(rec->e_cpos),
2888 ocfs2_rec_clusters(el, rec));
2893 *p_blkno = le64_to_cpu(rec->e_blkno);
2894 *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2896 *e_cpos = le32_to_cpu(rec->e_cpos);
2902 typedef int (xattr_bucket_func)(struct inode *inode,
2903 struct ocfs2_xattr_bucket *bucket,
2906 static int ocfs2_find_xe_in_bucket(struct inode *inode,
2907 struct ocfs2_xattr_bucket *bucket,
2914 int i, ret = 0, cmp = 1, block_off, new_offset;
2915 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
2916 size_t name_len = strlen(name);
2917 struct ocfs2_xattr_entry *xe = NULL;
2921 * We don't use binary search in the bucket because there
2922 * may be multiple entries with the same name hash.
2924 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2925 xe = &xh->xh_entries[i];
2927 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2929 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2932 cmp = name_index - ocfs2_xattr_get_type(xe);
2934 cmp = name_len - xe->xe_name_len;
2938 ret = ocfs2_xattr_bucket_get_name_value(inode,
2949 xe_name = bucket_block(bucket, block_off) + new_offset;
2950 if (!memcmp(name, xe_name, name_len)) {
2962 * Find the specified xattr entry in a series of buckets.
2963 * This series start from p_blkno and last for num_clusters.
2964 * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2965 * the num of the valid buckets.
2967 * Return the buffer_head this xattr should reside in. And if the xattr's
2968 * hash is in the gap of 2 buckets, return the lower bucket.
2970 static int ocfs2_xattr_bucket_find(struct inode *inode,
2977 struct ocfs2_xattr_search *xs)
2980 struct ocfs2_xattr_header *xh = NULL;
2981 struct ocfs2_xattr_entry *xe = NULL;
2983 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2984 int low_bucket = 0, bucket, high_bucket;
2985 struct ocfs2_xattr_bucket *search;
2987 u64 blkno, lower_blkno = 0;
2989 search = ocfs2_xattr_bucket_new(inode);
2996 ret = ocfs2_read_xattr_bucket(search, p_blkno);
3002 xh = bucket_xh(search);
3003 high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
3004 while (low_bucket <= high_bucket) {
3005 ocfs2_xattr_bucket_relse(search);
3007 bucket = (low_bucket + high_bucket) / 2;
3008 blkno = p_blkno + bucket * blk_per_bucket;
3009 ret = ocfs2_read_xattr_bucket(search, blkno);
3015 xh = bucket_xh(search);
3016 xe = &xh->xh_entries[0];
3017 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3018 high_bucket = bucket - 1;
3023 * Check whether the hash of the last entry in our
3024 * bucket is larger than the search one. for an empty
3025 * bucket, the last one is also the first one.
3028 xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3030 last_hash = le32_to_cpu(xe->xe_name_hash);
3032 /* record lower_blkno which may be the insert place. */
3033 lower_blkno = blkno;
3035 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3036 low_bucket = bucket + 1;
3040 /* the searched xattr should reside in this bucket if exists. */
3041 ret = ocfs2_find_xe_in_bucket(inode, search,
3042 name_index, name, name_hash,
3052 * Record the bucket we have found.
3053 * When the xattr's hash value is in the gap of 2 buckets, we will
3054 * always set it to the previous bucket.
3057 lower_blkno = p_blkno;
3059 /* This should be in cache - we just read it during the search */
3060 ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3066 xs->header = bucket_xh(xs->bucket);
3067 xs->base = bucket_block(xs->bucket, 0);
3068 xs->end = xs->base + inode->i_sb->s_blocksize;
3071 xs->here = &xs->header->xh_entries[index];
3072 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
3073 (unsigned long long)bucket_blkno(xs->bucket), index);
3078 ocfs2_xattr_bucket_free(search);
3082 static int ocfs2_xattr_index_block_find(struct inode *inode,
3083 struct buffer_head *root_bh,
3086 struct ocfs2_xattr_search *xs)
3089 struct ocfs2_xattr_block *xb =
3090 (struct ocfs2_xattr_block *)root_bh->b_data;
3091 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3092 struct ocfs2_extent_list *el = &xb_root->xt_list;
3094 u32 first_hash, num_clusters = 0;
3095 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3097 if (le16_to_cpu(el->l_next_free_rec) == 0)
3100 mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
3101 name, name_hash, name_index);
3103 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3110 BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3112 mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
3113 "in the rec is %u\n", num_clusters, (unsigned long long)p_blkno,
3116 ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3117 p_blkno, first_hash, num_clusters, xs);
3123 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3126 xattr_bucket_func *func,
3130 u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3131 u32 num_buckets = clusters * bpc;
3132 struct ocfs2_xattr_bucket *bucket;
3134 bucket = ocfs2_xattr_bucket_new(inode);
3136 mlog_errno(-ENOMEM);
3140 mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
3141 clusters, (unsigned long long)blkno);
3143 for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3144 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3151 * The real bucket num in this series of blocks is stored
3152 * in the 1st bucket.
3155 num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3157 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
3158 (unsigned long long)blkno,
3159 le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3161 ret = func(inode, bucket, para);
3162 if (ret && ret != -ERANGE)
3164 /* Fall through to bucket_relse() */
3167 ocfs2_xattr_bucket_relse(bucket);
3172 ocfs2_xattr_bucket_free(bucket);
3176 struct ocfs2_xattr_tree_list {
3182 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
3183 struct ocfs2_xattr_header *xh,
3190 if (index < 0 || index >= le16_to_cpu(xh->xh_count))
3193 name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
3195 *block_off = name_offset >> inode->i_sb->s_blocksize_bits;
3196 *new_offset = name_offset % inode->i_sb->s_blocksize;
3201 static int ocfs2_list_xattr_bucket(struct inode *inode,
3202 struct ocfs2_xattr_bucket *bucket,
3206 struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
3207 int i, block_off, new_offset;
3208 const char *prefix, *name;
3210 for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
3211 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
3212 type = ocfs2_xattr_get_type(entry);
3213 prefix = ocfs2_xattr_prefix(type);
3216 ret = ocfs2_xattr_bucket_get_name_value(inode,
3224 name = (const char *)bucket_block(bucket, block_off) +
3226 ret = ocfs2_xattr_list_entry(xl->buffer,
3230 entry->xe_name_len);
3239 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
3240 struct ocfs2_xattr_tree_root *xt,
3244 struct ocfs2_extent_list *el = &xt->xt_list;
3246 u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
3248 struct ocfs2_xattr_tree_list xl = {
3250 .buffer_size = buffer_size,
3254 if (le16_to_cpu(el->l_next_free_rec) == 0)
3257 while (name_hash > 0) {
3258 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
3259 &e_cpos, &num_clusters, el);
3265 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
3266 ocfs2_list_xattr_bucket,
3277 name_hash = e_cpos - 1;
3285 static int cmp_xe(const void *a, const void *b)
3287 const struct ocfs2_xattr_entry *l = a, *r = b;
3288 u32 l_hash = le32_to_cpu(l->xe_name_hash);
3289 u32 r_hash = le32_to_cpu(r->xe_name_hash);
3291 if (l_hash > r_hash)
3293 if (l_hash < r_hash)
3298 static void swap_xe(void *a, void *b, int size)
3300 struct ocfs2_xattr_entry *l = a, *r = b, tmp;
3303 memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
3304 memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
3308 * When the ocfs2_xattr_block is filled up, new bucket will be created
3309 * and all the xattr entries will be moved to the new bucket.
3310 * The header goes at the start of the bucket, and the names+values are
3311 * filled from the end. This is why *target starts as the last buffer.
3312 * Note: we need to sort the entries since they are not saved in order
3313 * in the ocfs2_xattr_block.
3315 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
3316 struct buffer_head *xb_bh,
3317 struct ocfs2_xattr_bucket *bucket)
3319 int i, blocksize = inode->i_sb->s_blocksize;
3320 int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3321 u16 offset, size, off_change;
3322 struct ocfs2_xattr_entry *xe;
3323 struct ocfs2_xattr_block *xb =
3324 (struct ocfs2_xattr_block *)xb_bh->b_data;
3325 struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
3326 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3327 u16 count = le16_to_cpu(xb_xh->xh_count);
3328 char *src = xb_bh->b_data;
3329 char *target = bucket_block(bucket, blks - 1);
3331 mlog(0, "cp xattr from block %llu to bucket %llu\n",
3332 (unsigned long long)xb_bh->b_blocknr,
3333 (unsigned long long)bucket_blkno(bucket));
3335 for (i = 0; i < blks; i++)
3336 memset(bucket_block(bucket, i), 0, blocksize);
3339 * Since the xe_name_offset is based on ocfs2_xattr_header,
3340 * there is a offset change corresponding to the change of
3341 * ocfs2_xattr_header's position.
3343 off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3344 xe = &xb_xh->xh_entries[count - 1];
3345 offset = le16_to_cpu(xe->xe_name_offset) + off_change;
3346 size = blocksize - offset;
3348 /* copy all the names and values. */
3349 memcpy(target + offset, src + offset, size);
3351 /* Init new header now. */
3352 xh->xh_count = xb_xh->xh_count;
3353 xh->xh_num_buckets = cpu_to_le16(1);
3354 xh->xh_name_value_len = cpu_to_le16(size);
3355 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
3357 /* copy all the entries. */
3358 target = bucket_block(bucket, 0);
3359 offset = offsetof(struct ocfs2_xattr_header, xh_entries);
3360 size = count * sizeof(struct ocfs2_xattr_entry);
3361 memcpy(target + offset, (char *)xb_xh + offset, size);
3363 /* Change the xe offset for all the xe because of the move. */
3364 off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
3365 offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3366 for (i = 0; i < count; i++)
3367 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
3369 mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
3370 offset, size, off_change);
3372 sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
3377 * After we move xattr from block to index btree, we have to
3378 * update ocfs2_xattr_search to the new xe and base.
3380 * When the entry is in xattr block, xattr_bh indicates the storage place.
3381 * While if the entry is in index b-tree, "bucket" indicates the
3382 * real place of the xattr.
3384 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
3385 struct ocfs2_xattr_search *xs,
3386 struct buffer_head *old_bh)
3388 char *buf = old_bh->b_data;
3389 struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
3390 struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
3393 xs->header = bucket_xh(xs->bucket);
3394 xs->base = bucket_block(xs->bucket, 0);
3395 xs->end = xs->base + inode->i_sb->s_blocksize;
3400 i = xs->here - old_xh->xh_entries;
3401 xs->here = &xs->header->xh_entries[i];
3404 static int ocfs2_xattr_create_index_block(struct inode *inode,
3405 struct ocfs2_xattr_search *xs,
3406 struct ocfs2_xattr_set_ctxt *ctxt)
3411 handle_t *handle = ctxt->handle;
3412 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3413 struct ocfs2_inode_info *oi = OCFS2_I(inode);
3414 struct buffer_head *xb_bh = xs->xattr_bh;
3415 struct ocfs2_xattr_block *xb =
3416 (struct ocfs2_xattr_block *)xb_bh->b_data;
3417 struct ocfs2_xattr_tree_root *xr;
3418 u16 xb_flags = le16_to_cpu(xb->xb_flags);
3420 mlog(0, "create xattr index block for %llu\n",
3421 (unsigned long long)xb_bh->b_blocknr);
3423 BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
3424 BUG_ON(!xs->bucket);
3428 * We can use this lock for now, and maybe move to a dedicated mutex
3429 * if performance becomes a problem later.
3431 down_write(&oi->ip_alloc_sem);
3433 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), xb_bh,
3434 OCFS2_JOURNAL_ACCESS_WRITE);
3440 ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac,
3441 1, 1, &bit_off, &len);
3448 * The bucket may spread in many blocks, and
3449 * we will only touch the 1st block and the last block
3450 * in the whole bucket(one for entry and one for data).
3452 blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
3454 mlog(0, "allocate 1 cluster from %llu to xattr block\n",
3455 (unsigned long long)blkno);
3457 ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
3463 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
3464 OCFS2_JOURNAL_ACCESS_CREATE);
3470 ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
3471 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
3473 ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
3475 /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
3476 memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
3477 offsetof(struct ocfs2_xattr_block, xb_attrs));
3479 xr = &xb->xb_attrs.xb_root;
3480 xr->xt_clusters = cpu_to_le32(1);
3481 xr->xt_last_eb_blk = 0;
3482 xr->xt_list.l_tree_depth = 0;
3483 xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
3484 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
3486 xr->xt_list.l_recs[0].e_cpos = 0;
3487 xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
3488 xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
3490 xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
3492 ocfs2_journal_dirty(handle, xb_bh);
3495 up_write(&oi->ip_alloc_sem);
3500 static int cmp_xe_offset(const void *a, const void *b)
3502 const struct ocfs2_xattr_entry *l = a, *r = b;
3503 u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
3504 u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
3506 if (l_name_offset < r_name_offset)
3508 if (l_name_offset > r_name_offset)
3514 * defrag a xattr bucket if we find that the bucket has some
3515 * holes beteen name/value pairs.
3516 * We will move all the name/value pairs to the end of the bucket
3517 * so that we can spare some space for insertion.
3519 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
3521 struct ocfs2_xattr_bucket *bucket)
3524 size_t end, offset, len, value_len;
3525 struct ocfs2_xattr_header *xh;
3526 char *entries, *buf, *bucket_buf = NULL;
3527 u64 blkno = bucket_blkno(bucket);
3529 size_t blocksize = inode->i_sb->s_blocksize;
3530 struct ocfs2_xattr_entry *xe;
3533 * In order to make the operation more efficient and generic,
3534 * we copy all the blocks into a contiguous memory and do the
3535 * defragment there, so if anything is error, we will not touch
3538 bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
3545 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3546 memcpy(buf, bucket_block(bucket, i), blocksize);
3548 ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
3549 OCFS2_JOURNAL_ACCESS_WRITE);
3555 xh = (struct ocfs2_xattr_header *)bucket_buf;
3556 entries = (char *)xh->xh_entries;
3557 xh_free_start = le16_to_cpu(xh->xh_free_start);
3559 mlog(0, "adjust xattr bucket in %llu, count = %u, "
3560 "xh_free_start = %u, xh_name_value_len = %u.\n",
3561 (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
3562 xh_free_start, le16_to_cpu(xh->xh_name_value_len));
3565 * sort all the entries by their offset.
3566 * the largest will be the first, so that we can
3567 * move them to the end one by one.
3569 sort(entries, le16_to_cpu(xh->xh_count),
3570 sizeof(struct ocfs2_xattr_entry),
3571 cmp_xe_offset, swap_xe);
3573 /* Move all name/values to the end of the bucket. */
3574 xe = xh->xh_entries;
3575 end = OCFS2_XATTR_BUCKET_SIZE;
3576 for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
3577 offset = le16_to_cpu(xe->xe_name_offset);
3578 if (ocfs2_xattr_is_local(xe))
3579 value_len = OCFS2_XATTR_SIZE(
3580 le64_to_cpu(xe->xe_value_size));
3582 value_len = OCFS2_XATTR_ROOT_SIZE;
3583 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
3586 * We must make sure that the name/value pair
3587 * exist in the same block. So adjust end to
3588 * the previous block end if needed.
3590 if (((end - len) / blocksize !=
3591 (end - 1) / blocksize))
3592 end = end - end % blocksize;
3594 if (end > offset + len) {
3595 memmove(bucket_buf + end - len,
3596 bucket_buf + offset, len);
3597 xe->xe_name_offset = cpu_to_le16(end - len);
3600 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
3601 "bucket %llu\n", (unsigned long long)blkno);
3606 mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
3607 "bucket %llu\n", (unsigned long long)blkno);
3609 if (xh_free_start == end)
3612 memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
3613 xh->xh_free_start = cpu_to_le16(end);
3615 /* sort the entries by their name_hash. */
3616 sort(entries, le16_to_cpu(xh->xh_count),
3617 sizeof(struct ocfs2_xattr_entry),
3621 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3622 memcpy(bucket_block(bucket, i), buf, blocksize);
3623 ocfs2_xattr_bucket_journal_dirty(handle, bucket);
3631 * prev_blkno points to the start of an existing extent. new_blkno
3632 * points to a newly allocated extent. Because we know each of our
3633 * clusters contains more than bucket, we can easily split one cluster
3634 * at a bucket boundary. So we take the last cluster of the existing
3635 * extent and split it down the middle. We move the last half of the
3636 * buckets in the last cluster of the existing extent over to the new
3639 * first_bh is the buffer at prev_blkno so we can update the existing
3640 * extent's bucket count. header_bh is the bucket were we were hoping
3641 * to insert our xattr. If the bucket move places the target in the new
3642 * extent, we'll update first_bh and header_bh after modifying the old
3645 * first_hash will be set as the 1st xe's name_hash in the new extent.
3647 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3649 struct ocfs2_xattr_bucket *first,
3650 struct ocfs2_xattr_bucket *target,
3656 struct super_block *sb = inode->i_sb;
3657 int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
3658 int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
3659 int to_move = num_buckets / 2;
3661 u64 last_cluster_blkno = bucket_blkno(first) +
3662 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
3664 BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
3665 BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
3667 mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3668 (unsigned long long)last_cluster_blkno, (unsigned long long)new_blkno);
3670 ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
3671 last_cluster_blkno, new_blkno,
3672 to_move, first_hash);
3678 /* This is the first bucket that got moved */
3679 src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
3682 * If the target bucket was part of the moved buckets, we need to
3683 * update first and target.
3685 if (bucket_blkno(target) >= src_blkno) {
3686 /* Find the block for the new target bucket */
3687 src_blkno = new_blkno +
3688 (bucket_blkno(target) - src_blkno);
3690 ocfs2_xattr_bucket_relse(first);
3691 ocfs2_xattr_bucket_relse(target);
3694 * These shouldn't fail - the buffers are in the
3695 * journal from ocfs2_cp_xattr_bucket().
3697 ret = ocfs2_read_xattr_bucket(first, new_blkno);
3702 ret = ocfs2_read_xattr_bucket(target, src_blkno);
3713 * Find the suitable pos when we divide a bucket into 2.
3714 * We have to make sure the xattrs with the same hash value exist
3715 * in the same bucket.
3717 * If this ocfs2_xattr_header covers more than one hash value, find a
3718 * place where the hash value changes. Try to find the most even split.
3719 * The most common case is that all entries have different hash values,
3720 * and the first check we make will find a place to split.
3722 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
3724 struct ocfs2_xattr_entry *entries = xh->xh_entries;
3725 int count = le16_to_cpu(xh->xh_count);
3726 int delta, middle = count / 2;
3729 * We start at the middle. Each step gets farther away in both
3730 * directions. We therefore hit the change in hash value
3731 * nearest to the middle. Note that this loop does not execute for
3734 for (delta = 0; delta < middle; delta++) {
3735 /* Let's check delta earlier than middle */
3736 if (cmp_xe(&entries[middle - delta - 1],
3737 &entries[middle - delta]))
3738 return middle - delta;
3740 /* For even counts, don't walk off the end */
3741 if ((middle + delta + 1) == count)
3744 /* Now try delta past middle */
3745 if (cmp_xe(&entries[middle + delta],
3746 &entries[middle + delta + 1]))
3747 return middle + delta + 1;
3750 /* Every entry had the same hash */
3755 * Move some xattrs in old bucket(blk) to new bucket(new_blk).
3756 * first_hash will record the 1st hash of the new bucket.
3758 * Normally half of the xattrs will be moved. But we have to make
3759 * sure that the xattrs with the same hash value are stored in the
3760 * same bucket. If all the xattrs in this bucket have the same hash
3761 * value, the new bucket will be initialized as an empty one and the
3762 * first_hash will be initialized as (hash_value+1).
3764 static int ocfs2_divide_xattr_bucket(struct inode *inode,
3769 int new_bucket_head)
3772 int count, start, len, name_value_len = 0, xe_len, name_offset = 0;
3773 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3774 struct ocfs2_xattr_header *xh;
3775 struct ocfs2_xattr_entry *xe;
3776 int blocksize = inode->i_sb->s_blocksize;
3778 mlog(0, "move some of xattrs from bucket %llu to %llu\n",
3779 (unsigned long long)blk, (unsigned long long)new_blk);
3781 s_bucket = ocfs2_xattr_bucket_new(inode);
3782 t_bucket = ocfs2_xattr_bucket_new(inode);
3783 if (!s_bucket || !t_bucket) {
3789 ret = ocfs2_read_xattr_bucket(s_bucket, blk);
3795 ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
3796 OCFS2_JOURNAL_ACCESS_WRITE);
3803 * Even if !new_bucket_head, we're overwriting t_bucket. Thus,
3804 * there's no need to read it.
3806 ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
3813 * Hey, if we're overwriting t_bucket, what difference does
3814 * ACCESS_CREATE vs ACCESS_WRITE make? See the comment in the
3815 * same part of ocfs2_cp_xattr_bucket().
3817 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3819 OCFS2_JOURNAL_ACCESS_CREATE :
3820 OCFS2_JOURNAL_ACCESS_WRITE);
3826 xh = bucket_xh(s_bucket);
3827 count = le16_to_cpu(xh->xh_count);
3828 start = ocfs2_xattr_find_divide_pos(xh);
3830 if (start == count) {
3831 xe = &xh->xh_entries[start-1];
3834 * initialized a new empty bucket here.
3835 * The hash value is set as one larger than
3836 * that of the last entry in the previous bucket.
3838 for (i = 0; i < t_bucket->bu_blocks; i++)
3839 memset(bucket_block(t_bucket, i), 0, blocksize);
3841 xh = bucket_xh(t_bucket);
3842 xh->xh_free_start = cpu_to_le16(blocksize);
3843 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
3844 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
3846 goto set_num_buckets;
3849 /* copy the whole bucket to the new first. */
3850 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
3852 /* update the new bucket. */
3853 xh = bucket_xh(t_bucket);
3856 * Calculate the total name/value len and xh_free_start for
3857 * the old bucket first.
3859 name_offset = OCFS2_XATTR_BUCKET_SIZE;
3861 for (i = 0; i < start; i++) {
3862 xe = &xh->xh_entries[i];
3863 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3864 if (ocfs2_xattr_is_local(xe))
3866 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3868 xe_len += OCFS2_XATTR_ROOT_SIZE;
3869 name_value_len += xe_len;
3870 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3871 name_offset = le16_to_cpu(xe->xe_name_offset);
3875 * Now begin the modification to the new bucket.
3877 * In the new bucket, We just move the xattr entry to the beginning
3878 * and don't touch the name/value. So there will be some holes in the
3879 * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3882 xe = &xh->xh_entries[start];
3883 len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3884 mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3885 (int)((char *)xe - (char *)xh),
3886 (int)((char *)xh->xh_entries - (char *)xh));
3887 memmove((char *)xh->xh_entries, (char *)xe, len);
3888 xe = &xh->xh_entries[count - start];
3889 len = sizeof(struct ocfs2_xattr_entry) * start;
3890 memset((char *)xe, 0, len);
3892 le16_add_cpu(&xh->xh_count, -start);
3893 le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3895 /* Calculate xh_free_start for the new bucket. */
3896 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3897 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3898 xe = &xh->xh_entries[i];
3899 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3900 if (ocfs2_xattr_is_local(xe))
3902 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3904 xe_len += OCFS2_XATTR_ROOT_SIZE;
3905 if (le16_to_cpu(xe->xe_name_offset) <
3906 le16_to_cpu(xh->xh_free_start))
3907 xh->xh_free_start = xe->xe_name_offset;
3911 /* set xh->xh_num_buckets for the new xh. */
3912 if (new_bucket_head)
3913 xh->xh_num_buckets = cpu_to_le16(1);
3915 xh->xh_num_buckets = 0;
3917 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
3919 /* store the first_hash of the new bucket. */
3921 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3924 * Now only update the 1st block of the old bucket. If we
3925 * just added a new empty bucket, there is no need to modify
3931 xh = bucket_xh(s_bucket);
3932 memset(&xh->xh_entries[start], 0,
3933 sizeof(struct ocfs2_xattr_entry) * (count - start));
3934 xh->xh_count = cpu_to_le16(start);
3935 xh->xh_free_start = cpu_to_le16(name_offset);
3936 xh->xh_name_value_len = cpu_to_le16(name_value_len);
3938 ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
3941 ocfs2_xattr_bucket_free(s_bucket);
3942 ocfs2_xattr_bucket_free(t_bucket);
3948 * Copy xattr from one bucket to another bucket.
3950 * The caller must make sure that the journal transaction
3951 * has enough space for journaling.
3953 static int ocfs2_cp_xattr_bucket(struct inode *inode,
3960 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3962 BUG_ON(s_blkno == t_blkno);
3964 mlog(0, "cp bucket %llu to %llu, target is %d\n",
3965 (unsigned long long)s_blkno, (unsigned long long)t_blkno,
3968 s_bucket = ocfs2_xattr_bucket_new(inode);
3969 t_bucket = ocfs2_xattr_bucket_new(inode);
3970 if (!s_bucket || !t_bucket) {
3976 ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
3981 * Even if !t_is_new, we're overwriting t_bucket. Thus,
3982 * there's no need to read it.
3984 ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
3989 * Hey, if we're overwriting t_bucket, what difference does
3990 * ACCESS_CREATE vs ACCESS_WRITE make? Well, if we allocated a new
3991 * cluster to fill, we came here from
3992 * ocfs2_mv_xattr_buckets(), and it is really new -
3993 * ACCESS_CREATE is required. But we also might have moved data
3994 * out of t_bucket before extending back into it.
3995 * ocfs2_add_new_xattr_bucket() can do this - its call to
3996 * ocfs2_add_new_xattr_cluster() may have created a new extent
3997 * and copied out the end of the old extent. Then it re-extends
3998 * the old extent back to create space for new xattrs. That's
3999 * how we get here, and the bucket isn't really new.
4001 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4003 OCFS2_JOURNAL_ACCESS_CREATE :
4004 OCFS2_JOURNAL_ACCESS_WRITE);
4008 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4009 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4012 ocfs2_xattr_bucket_free(t_bucket);
4013 ocfs2_xattr_bucket_free(s_bucket);
4019 * src_blk points to the start of an existing extent. last_blk points to
4020 * last cluster in that extent. to_blk points to a newly allocated
4021 * extent. We copy the buckets from the cluster at last_blk to the new
4022 * extent. If start_bucket is non-zero, we skip that many buckets before
4023 * we start copying. The new extent's xh_num_buckets gets set to the
4024 * number of buckets we copied. The old extent's xh_num_buckets shrinks
4025 * by the same amount.
4027 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4028 u64 src_blk, u64 last_blk, u64 to_blk,
4029 unsigned int start_bucket,
4032 int i, ret, credits;
4033 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4034 int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4035 int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4036 struct ocfs2_xattr_bucket *old_first, *new_first;
4038 mlog(0, "mv xattrs from cluster %llu to %llu\n",
4039 (unsigned long long)last_blk, (unsigned long long)to_blk);
4041 BUG_ON(start_bucket >= num_buckets);
4043 num_buckets -= start_bucket;
4044 last_blk += (start_bucket * blks_per_bucket);
4047 /* The first bucket of the original extent */
4048 old_first = ocfs2_xattr_bucket_new(inode);
4049 /* The first bucket of the new extent */
4050 new_first = ocfs2_xattr_bucket_new(inode);
4051 if (!old_first || !new_first) {
4057 ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4064 * We need to update the first bucket of the old extent and all
4065 * the buckets going to the new extent.
4067 credits = ((num_buckets + 1) * blks_per_bucket) +
4068 handle->h_buffer_credits;
4069 ret = ocfs2_extend_trans(handle, credits);
4075 ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4076 OCFS2_JOURNAL_ACCESS_WRITE);
4082 for (i = 0; i < num_buckets; i++) {
4083 ret = ocfs2_cp_xattr_bucket(inode, handle,
4084 last_blk + (i * blks_per_bucket),
4085 to_blk + (i * blks_per_bucket),
4094 * Get the new bucket ready before we dirty anything
4095 * (This actually shouldn't fail, because we already dirtied
4096 * it once in ocfs2_cp_xattr_bucket()).
4098 ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4103 ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4104 OCFS2_JOURNAL_ACCESS_WRITE);
4110 /* Now update the headers */
4111 le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4112 ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4114 bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4115 ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4118 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4121 ocfs2_xattr_bucket_free(new_first);
4122 ocfs2_xattr_bucket_free(old_first);
4127 * Move some xattrs in this cluster to the new cluster.
4128 * This function should only be called when bucket size == cluster size.
4129 * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4131 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4137 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4138 int ret, credits = 2 * blk_per_bucket + handle->h_buffer_credits;
4140 BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4142 ret = ocfs2_extend_trans(handle, credits);
4148 /* Move half of the xattr in start_blk to the next bucket. */
4149 return ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4150 new_blk, first_hash, 1);
4154 * Move some xattrs from the old cluster to the new one since they are not
4155 * contiguous in ocfs2 xattr tree.
4157 * new_blk starts a new separate cluster, and we will move some xattrs from
4158 * prev_blk to it. v_start will be set as the first name hash value in this
4159 * new cluster so that it can be used as e_cpos during tree insertion and
4160 * don't collide with our original b-tree operations. first_bh and header_bh
4161 * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4162 * to extend the insert bucket.
4164 * The problem is how much xattr should we move to the new one and when should
4165 * we update first_bh and header_bh?
4166 * 1. If cluster size > bucket size, that means the previous cluster has more
4167 * than 1 bucket, so just move half nums of bucket into the new cluster and
4168 * update the first_bh and header_bh if the insert bucket has been moved
4169 * to the new cluster.
4170 * 2. If cluster_size == bucket_size:
4171 * a) If the previous extent rec has more than one cluster and the insert
4172 * place isn't in the last cluster, copy the entire last cluster to the
4173 * new one. This time, we don't need to upate the first_bh and header_bh
4174 * since they will not be moved into the new cluster.
4175 * b) Otherwise, move the bottom half of the xattrs in the last cluster into
4176 * the new one. And we set the extend flag to zero if the insert place is
4177 * moved into the new allocated cluster since no extend is needed.
4179 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
4181 struct ocfs2_xattr_bucket *first,
4182 struct ocfs2_xattr_bucket *target,
4190 mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
4191 (unsigned long long)bucket_blkno(first), prev_clusters,
4192 (unsigned long long)new_blk);
4194 if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
4195 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
4204 /* The start of the last cluster in the first extent */
4205 u64 last_blk = bucket_blkno(first) +
4206 ((prev_clusters - 1) *
4207 ocfs2_clusters_to_blocks(inode->i_sb, 1));
4209 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
4210 ret = ocfs2_mv_xattr_buckets(inode, handle,
4211 bucket_blkno(first),
4212 last_blk, new_blk, 0,
4217 ret = ocfs2_divide_xattr_cluster(inode, handle,
4223 if ((bucket_blkno(target) == last_blk) && extend)
4232 * Add a new cluster for xattr storage.
4234 * If the new cluster is contiguous with the previous one, it will be
4235 * appended to the same extent record, and num_clusters will be updated.
4236 * If not, we will insert a new extent for it and move some xattrs in
4237 * the last cluster into the new allocated one.
4238 * We also need to limit the maximum size of a btree leaf, otherwise we'll
4239 * lose the benefits of hashing because we'll have to search large leaves.
4240 * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
4243 * first_bh is the first block of the previous extent rec and header_bh
4244 * indicates the bucket we will insert the new xattrs. They will be updated
4245 * when the header_bh is moved into the new cluster.
4247 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
4248 struct buffer_head *root_bh,
4249 struct ocfs2_xattr_bucket *first,
4250 struct ocfs2_xattr_bucket *target,
4254 struct ocfs2_xattr_set_ctxt *ctxt)
4257 u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
4258 u32 prev_clusters = *num_clusters;
4259 u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
4261 handle_t *handle = ctxt->handle;
4262 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4263 struct ocfs2_extent_tree et;
4265 mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
4266 "previous xattr blkno = %llu\n",
4267 (unsigned long long)OCFS2_I(inode)->ip_blkno,
4268 prev_cpos, (unsigned long long)bucket_blkno(first));
4270 ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
4272 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
4273 OCFS2_JOURNAL_ACCESS_WRITE);
4279 ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac, 1,
4280 clusters_to_add, &bit_off, &num_bits);
4287 BUG_ON(num_bits > clusters_to_add);
4289 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
4290 mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
4291 num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
4293 if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
4294 (prev_clusters + num_bits) << osb->s_clustersize_bits <=
4295 OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
4297 * If this cluster is contiguous with the old one and
4298 * adding this new cluster, we don't surpass the limit of
4299 * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
4300 * initialized and used like other buckets in the previous
4302 * So add it as a contiguous one. The caller will handle
4305 v_start = prev_cpos + prev_clusters;
4306 *num_clusters = prev_clusters + num_bits;
4307 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
4310 ret = ocfs2_adjust_xattr_cross_cluster(inode,
4324 mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
4325 num_bits, (unsigned long long)block, v_start);
4326 ret = ocfs2_insert_extent(handle, &et, v_start, block,
4327 num_bits, 0, ctxt->meta_ac);
4333 ret = ocfs2_journal_dirty(handle, root_bh);
4342 * We are given an extent. 'first' is the bucket at the very front of
4343 * the extent. The extent has space for an additional bucket past
4344 * bucket_xh(first)->xh_num_buckets. 'target_blkno' is the block number
4345 * of the target bucket. We wish to shift every bucket past the target
4346 * down one, filling in that additional space. When we get back to the
4347 * target, we split the target between itself and the now-empty bucket
4348 * at target+1 (aka, target_blkno + blks_per_bucket).
4350 static int ocfs2_extend_xattr_bucket(struct inode *inode,
4352 struct ocfs2_xattr_bucket *first,
4357 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4358 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4360 u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
4362 mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
4363 "from %llu, len = %u\n", (unsigned long long)target_blk,
4364 (unsigned long long)bucket_blkno(first), num_clusters);
4366 /* The extent must have room for an additional bucket */
4367 BUG_ON(new_bucket >=
4368 (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
4370 /* end_blk points to the last existing bucket */
4371 end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
4374 * end_blk is the start of the last existing bucket.
4375 * Thus, (end_blk - target_blk) covers the target bucket and
4376 * every bucket after it up to, but not including, the last
4377 * existing bucket. Then we add the last existing bucket, the
4378 * new bucket, and the first bucket (3 * blk_per_bucket).
4380 credits = (end_blk - target_blk) + (3 * blk_per_bucket) +
4381 handle->h_buffer_credits;
4382 ret = ocfs2_extend_trans(handle, credits);
4388 ret = ocfs2_xattr_bucket_journal_access(handle, first,
4389 OCFS2_JOURNAL_ACCESS_WRITE);
4395 while (end_blk != target_blk) {
4396 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
4397 end_blk + blk_per_bucket, 0);
4400 end_blk -= blk_per_bucket;
4403 /* Move half of the xattr in target_blkno to the next bucket. */
4404 ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
4405 target_blk + blk_per_bucket, NULL, 0);
4407 le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
4408 ocfs2_xattr_bucket_journal_dirty(handle, first);
4415 * Add new xattr bucket in an extent record and adjust the buckets
4416 * accordingly. xb_bh is the ocfs2_xattr_block, and target is the
4417 * bucket we want to insert into.
4419 * In the easy case, we will move all the buckets after target down by
4420 * one. Half of target's xattrs will be moved to the next bucket.
4422 * If current cluster is full, we'll allocate a new one. This may not
4423 * be contiguous. The underlying calls will make sure that there is
4424 * space for the insert, shifting buckets around if necessary.
4425 * 'target' may be moved by those calls.
4427 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
4428 struct buffer_head *xb_bh,
4429 struct ocfs2_xattr_bucket *target,
4430 struct ocfs2_xattr_set_ctxt *ctxt)
4432 struct ocfs2_xattr_block *xb =
4433 (struct ocfs2_xattr_block *)xb_bh->b_data;
4434 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
4435 struct ocfs2_extent_list *el = &xb_root->xt_list;
4437 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
4438 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4439 int ret, num_buckets, extend = 1;
4441 u32 e_cpos, num_clusters;
4442 /* The bucket at the front of the extent */
4443 struct ocfs2_xattr_bucket *first;
4445 mlog(0, "Add new xattr bucket starting from %llu\n",
4446 (unsigned long long)bucket_blkno(target));
4448 /* The first bucket of the original extent */
4449 first = ocfs2_xattr_bucket_new(inode);
4456 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
4463 ret = ocfs2_read_xattr_bucket(first, p_blkno);
4469 num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
4470 if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
4472 * This can move first+target if the target bucket moves
4473 * to the new extent.
4475 ret = ocfs2_add_new_xattr_cluster(inode,
4490 ret = ocfs2_extend_xattr_bucket(inode,
4493 bucket_blkno(target),
4500 ocfs2_xattr_bucket_free(first);
4505 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
4506 struct ocfs2_xattr_bucket *bucket,
4509 int block_off = offs >> inode->i_sb->s_blocksize_bits;
4511 offs = offs % inode->i_sb->s_blocksize;
4512 return bucket_block(bucket, block_off) + offs;
4516 * Handle the normal xattr set, including replace, delete and new.
4518 * Note: "local" indicates the real data's locality. So we can't
4519 * just its bucket locality by its length.
4521 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
4522 struct ocfs2_xattr_info *xi,
4523 struct ocfs2_xattr_search *xs,
4527 struct ocfs2_xattr_entry *last, *xe;
4528 int name_len = strlen(xi->name);
4529 struct ocfs2_xattr_header *xh = xs->header;
4530 u16 count = le16_to_cpu(xh->xh_count), start;
4531 size_t blocksize = inode->i_sb->s_blocksize;
4533 size_t offs, size, new_size;
4535 last = &xh->xh_entries[count];
4536 if (!xs->not_found) {
4538 offs = le16_to_cpu(xe->xe_name_offset);
4539 if (ocfs2_xattr_is_local(xe))
4540 size = OCFS2_XATTR_SIZE(name_len) +
4541 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4543 size = OCFS2_XATTR_SIZE(name_len) +
4544 OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4547 * If the new value will be stored outside, xi->value has been
4548 * initalized as an empty ocfs2_xattr_value_root, and the same
4549 * goes with xi->value_len, so we can set new_size safely here.
4550 * See ocfs2_xattr_set_in_bucket.
4552 new_size = OCFS2_XATTR_SIZE(name_len) +
4553 OCFS2_XATTR_SIZE(xi->value_len);
4555 le16_add_cpu(&xh->xh_name_value_len, -size);
4557 if (new_size > size)
4558 goto set_new_name_value;
4560 /* Now replace the old value with new one. */
4562 xe->xe_value_size = cpu_to_le64(xi->value_len);
4564 xe->xe_value_size = 0;
4566 val = ocfs2_xattr_bucket_get_val(inode,
4568 memset(val + OCFS2_XATTR_SIZE(name_len), 0,
4569 size - OCFS2_XATTR_SIZE(name_len));
4570 if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
4571 memcpy(val + OCFS2_XATTR_SIZE(name_len),
4572 xi->value, xi->value_len);
4574 le16_add_cpu(&xh->xh_name_value_len, new_size);
4575 ocfs2_xattr_set_local(xe, local);
4579 * Remove the old entry if there is more than one.
4580 * We don't remove the last entry so that we can
4581 * use it to indicate the hash value of the empty
4585 le16_add_cpu(&xh->xh_count, -1);
4588 (void *)last - (void *)xe);
4590 sizeof(struct ocfs2_xattr_entry));
4593 cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4598 /* find a new entry for insert. */
4599 int low = 0, high = count - 1, tmp;
4600 struct ocfs2_xattr_entry *tmp_xe;
4602 while (low <= high && count) {
4603 tmp = (low + high) / 2;
4604 tmp_xe = &xh->xh_entries[tmp];
4606 if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
4608 else if (name_hash <
4609 le32_to_cpu(tmp_xe->xe_name_hash))
4617 xe = &xh->xh_entries[low];
4619 memmove(xe + 1, xe, (void *)last - (void *)xe);
4621 le16_add_cpu(&xh->xh_count, 1);
4622 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
4623 xe->xe_name_hash = cpu_to_le32(name_hash);
4624 xe->xe_name_len = name_len;
4625 ocfs2_xattr_set_type(xe, xi->name_index);
4629 /* Insert the new name+value. */
4630 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4633 * We must make sure that the name/value pair
4634 * exists in the same block.
4636 offs = le16_to_cpu(xh->xh_free_start);
4637 start = offs - size;
4639 if (start >> inode->i_sb->s_blocksize_bits !=
4640 (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4641 offs = offs - offs % blocksize;
4642 xh->xh_free_start = cpu_to_le16(offs);
4645 val = ocfs2_xattr_bucket_get_val(inode, xs->bucket, offs - size);
4646 xe->xe_name_offset = cpu_to_le16(offs - size);
4648 memset(val, 0, size);
4649 memcpy(val, xi->name, name_len);
4650 memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4652 xe->xe_value_size = cpu_to_le64(xi->value_len);
4653 ocfs2_xattr_set_local(xe, local);
4655 le16_add_cpu(&xh->xh_free_start, -size);
4656 le16_add_cpu(&xh->xh_name_value_len, size);
4662 * Set the xattr entry in the specified bucket.
4663 * The bucket is indicated by xs->bucket and it should have the enough
4664 * space for the xattr insertion.
4666 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4668 struct ocfs2_xattr_info *xi,
4669 struct ocfs2_xattr_search *xs,
4676 mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4677 (unsigned long)xi->value_len, xi->name_index,
4678 (unsigned long long)bucket_blkno(xs->bucket));
4680 if (!xs->bucket->bu_bhs[1]) {
4681 blkno = bucket_blkno(xs->bucket);
4682 ocfs2_xattr_bucket_relse(xs->bucket);
4683 ret = ocfs2_read_xattr_bucket(xs->bucket, blkno);
4690 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4691 OCFS2_JOURNAL_ACCESS_WRITE);
4697 ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local);
4698 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4705 * Truncate the specified xe_off entry in xattr bucket.
4706 * bucket is indicated by header_bh and len is the new length.
4707 * Both the ocfs2_xattr_value_root and the entry will be updated here.
4709 * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4711 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4712 struct ocfs2_xattr_bucket *bucket,
4715 struct ocfs2_xattr_set_ctxt *ctxt)
4719 struct ocfs2_xattr_entry *xe;
4720 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4721 size_t blocksize = inode->i_sb->s_blocksize;
4722 struct ocfs2_xattr_value_buf vb = {
4723 .vb_access = ocfs2_journal_access,
4726 xe = &xh->xh_entries[xe_off];
4728 BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4730 offset = le16_to_cpu(xe->xe_name_offset) +
4731 OCFS2_XATTR_SIZE(xe->xe_name_len);
4733 value_blk = offset / blocksize;
4735 /* We don't allow ocfs2_xattr_value to be stored in different block. */
4736 BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4738 vb.vb_bh = bucket->bu_bhs[value_blk];
4741 vb.vb_xv = (struct ocfs2_xattr_value_root *)
4742 (vb.vb_bh->b_data + offset % blocksize);
4745 * From here on out we have to dirty the bucket. The generic
4746 * value calls only modify one of the bucket's bhs, but we need
4747 * to send the bucket at once. So if they error, they *could* have
4748 * modified something. We have to assume they did, and dirty
4749 * the whole bucket. This leaves us in a consistent state.
4751 mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4752 xe_off, (unsigned long long)bucket_blkno(bucket), len);
4753 ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
4759 ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
4760 OCFS2_JOURNAL_ACCESS_WRITE);
4766 xe->xe_value_size = cpu_to_le64(len);
4768 ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
4774 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4775 struct ocfs2_xattr_search *xs,
4777 struct ocfs2_xattr_set_ctxt *ctxt)
4780 struct ocfs2_xattr_entry *xe = xs->here;
4781 struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4783 BUG_ON(!xs->bucket->bu_bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4785 offset = xe - xh->xh_entries;
4786 ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket,
4794 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4796 struct ocfs2_xattr_search *xs,
4800 int ret, offset, block_off;
4801 struct ocfs2_xattr_value_root *xv;
4802 struct ocfs2_xattr_entry *xe = xs->here;
4803 struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
4806 BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4808 ret = ocfs2_xattr_bucket_get_name_value(inode, xh,
4809 xe - xh->xh_entries,
4817 base = bucket_block(xs->bucket, block_off);
4818 xv = (struct ocfs2_xattr_value_root *)(base + offset +
4819 OCFS2_XATTR_SIZE(xe->xe_name_len));
4821 ret = __ocfs2_xattr_set_value_outside(inode, handle,
4822 xv, val, value_len);
4829 static int ocfs2_rm_xattr_cluster(struct inode *inode,
4830 struct buffer_head *root_bh,
4836 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4837 struct inode *tl_inode = osb->osb_tl_inode;
4839 struct ocfs2_xattr_block *xb =
4840 (struct ocfs2_xattr_block *)root_bh->b_data;
4841 struct ocfs2_alloc_context *meta_ac = NULL;
4842 struct ocfs2_cached_dealloc_ctxt dealloc;
4843 struct ocfs2_extent_tree et;
4845 ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
4847 ocfs2_init_dealloc_ctxt(&dealloc);
4849 mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4850 cpos, len, (unsigned long long)blkno);
4852 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode), blkno,
4855 ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
4861 mutex_lock(&tl_inode->i_mutex);
4863 if (ocfs2_truncate_log_needs_flush(osb)) {
4864 ret = __ocfs2_flush_truncate_log(osb);
4871 handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
4872 if (IS_ERR(handle)) {
4878 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
4879 OCFS2_JOURNAL_ACCESS_WRITE);
4885 ret = ocfs2_remove_extent(handle, &et, cpos, len, meta_ac,
4892 le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4894 ret = ocfs2_journal_dirty(handle, root_bh);
4900 ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4905 ocfs2_commit_trans(osb, handle);
4907 ocfs2_schedule_truncate_log_flush(osb, 1);
4909 mutex_unlock(&tl_inode->i_mutex);
4912 ocfs2_free_alloc_context(meta_ac);
4914 ocfs2_run_deallocs(osb, &dealloc);
4919 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4921 struct ocfs2_xattr_search *xs)
4923 struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
4924 struct ocfs2_xattr_entry *last = &xh->xh_entries[
4925 le16_to_cpu(xh->xh_count) - 1];
4928 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4929 OCFS2_JOURNAL_ACCESS_WRITE);
4935 /* Remove the old entry. */
4936 memmove(xs->here, xs->here + 1,
4937 (void *)last - (void *)xs->here);
4938 memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4939 le16_add_cpu(&xh->xh_count, -1);
4941 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4945 * Set the xattr name/value in the bucket specified in xs.
4947 * As the new value in xi may be stored in the bucket or in an outside cluster,
4948 * we divide the whole process into 3 steps:
4949 * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4950 * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4951 * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4952 * 4. If the clusters for the new outside value can't be allocated, we need
4953 * to free the xattr we allocated in set.
4955 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4956 struct ocfs2_xattr_info *xi,
4957 struct ocfs2_xattr_search *xs,
4958 struct ocfs2_xattr_set_ctxt *ctxt)
4962 char *val = (char *)xi->value;
4963 struct ocfs2_xattr_entry *xe = xs->here;
4964 u32 name_hash = ocfs2_xattr_name_hash(inode, xi->name,
4967 if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
4969 * We need to truncate the xattr storage first.
4971 * If both the old and new value are stored to
4972 * outside block, we only need to truncate
4973 * the storage and then set the value outside.
4975 * If the new value should be stored within block,
4976 * we should free all the outside block first and
4977 * the modification to the xattr block will be done
4978 * by following steps.
4980 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4981 value_len = xi->value_len;
4985 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4992 goto set_value_outside;
4995 value_len = xi->value_len;
4996 /* So we have to handle the inside block change now. */
4997 if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4999 * If the new value will be stored outside of block,
5000 * initalize a new empty value root and insert it first.
5003 xi->value = &def_xv;
5004 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
5007 ret = ocfs2_xattr_set_entry_in_bucket(inode, ctxt->handle, xi, xs,
5014 if (value_len <= OCFS2_XATTR_INLINE_SIZE)
5017 /* allocate the space now for the outside block storage. */
5018 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
5023 if (xs->not_found) {
5025 * We can't allocate enough clusters for outside
5026 * storage and we have allocated xattr already,
5027 * so need to remove it.
5029 ocfs2_xattr_bucket_remove_xs(inode, ctxt->handle, xs);
5035 ret = ocfs2_xattr_bucket_set_value_outside(inode, ctxt->handle,
5036 xs, val, value_len);
5042 * check whether the xattr bucket is filled up with the same hash value.
5043 * If we want to insert the xattr with the same hash, return -ENOSPC.
5044 * If we want to insert a xattr with different hash value, go ahead
5045 * and ocfs2_divide_xattr_bucket will handle this.
5047 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
5048 struct ocfs2_xattr_bucket *bucket,
5051 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5052 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5054 if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5057 if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5058 xh->xh_entries[0].xe_name_hash) {
5059 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5061 (unsigned long long)bucket_blkno(bucket),
5062 le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5069 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5070 struct ocfs2_xattr_info *xi,
5071 struct ocfs2_xattr_search *xs,
5072 struct ocfs2_xattr_set_ctxt *ctxt)
5074 struct ocfs2_xattr_header *xh;
5075 struct ocfs2_xattr_entry *xe;
5076 u16 count, header_size, xh_free_start;
5077 int free, max_free, need, old;
5078 size_t value_size = 0, name_len = strlen(xi->name);
5079 size_t blocksize = inode->i_sb->s_blocksize;
5080 int ret, allocation = 0;
5082 mlog_entry("Set xattr %s in xattr index block\n", xi->name);
5086 count = le16_to_cpu(xh->xh_count);
5087 xh_free_start = le16_to_cpu(xh->xh_free_start);
5088 header_size = sizeof(struct ocfs2_xattr_header) +
5089 count * sizeof(struct ocfs2_xattr_entry);
5090 max_free = OCFS2_XATTR_BUCKET_SIZE - header_size -
5091 le16_to_cpu(xh->xh_name_value_len) - OCFS2_XATTR_HEADER_GAP;
5093 mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
5094 "of %u which exceed block size\n",
5095 (unsigned long long)bucket_blkno(xs->bucket),
5098 if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
5099 value_size = OCFS2_XATTR_ROOT_SIZE;
5101 value_size = OCFS2_XATTR_SIZE(xi->value_len);
5104 need = sizeof(struct ocfs2_xattr_entry) +
5105 OCFS2_XATTR_SIZE(name_len) + value_size;
5107 need = value_size + OCFS2_XATTR_SIZE(name_len);
5110 * We only replace the old value if the new length is smaller
5111 * than the old one. Otherwise we will allocate new space in the
5112 * bucket to store it.
5115 if (ocfs2_xattr_is_local(xe))
5116 old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
5118 old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
5120 if (old >= value_size)
5124 free = xh_free_start - header_size - OCFS2_XATTR_HEADER_GAP;
5126 * We need to make sure the new name/value pair
5127 * can exist in the same block.
5129 if (xh_free_start % blocksize < need)
5130 free -= xh_free_start % blocksize;
5132 mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
5133 "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
5134 " %u\n", xs->not_found,
5135 (unsigned long long)bucket_blkno(xs->bucket),
5136 free, need, max_free, le16_to_cpu(xh->xh_free_start),
5137 le16_to_cpu(xh->xh_name_value_len));
5141 count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb))) {
5142 if (need <= max_free &&
5143 count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
5145 * We can create the space by defragment. Since only the
5146 * name/value will be moved, the xe shouldn't be changed
5149 ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5156 xh_free_start = le16_to_cpu(xh->xh_free_start);
5157 free = xh_free_start - header_size
5158 - OCFS2_XATTR_HEADER_GAP;
5159 if (xh_free_start % blocksize < need)
5160 free -= xh_free_start % blocksize;
5165 mlog(0, "Can't get enough space for xattr insert by "
5166 "defragment. Need %u bytes, but we have %d, so "
5167 "allocate new bucket for it.\n", need, free);
5171 * We have to add new buckets or clusters and one
5172 * allocation should leave us enough space for insert.
5177 * We do not allow for overlapping ranges between buckets. And
5178 * the maximum number of collisions we will allow for then is
5179 * one bucket's worth, so check it here whether we need to
5180 * add a new bucket for the insert.
5182 ret = ocfs2_check_xattr_bucket_collision(inode,
5190 ret = ocfs2_add_new_xattr_bucket(inode,
5200 * ocfs2_add_new_xattr_bucket() will have updated
5201 * xs->bucket if it moved, but it will not have updated
5202 * any of the other search fields. Thus, we drop it and
5203 * re-search. Everything should be cached, so it'll be
5206 ocfs2_xattr_bucket_relse(xs->bucket);
5207 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5210 if (ret && ret != -ENODATA)
5212 xs->not_found = ret;
5218 ret = ocfs2_xattr_set_in_bucket(inode, xi, xs, ctxt);
5224 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5225 struct ocfs2_xattr_bucket *bucket,
5229 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5231 struct ocfs2_xattr_entry *xe;
5232 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5233 struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5234 int credits = ocfs2_remove_extent_credits(osb->sb) +
5235 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5238 ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5240 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5241 xe = &xh->xh_entries[i];
5242 if (ocfs2_xattr_is_local(xe))
5245 ctxt.handle = ocfs2_start_trans(osb, credits);
5246 if (IS_ERR(ctxt.handle)) {
5247 ret = PTR_ERR(ctxt.handle);
5252 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5255 ocfs2_commit_trans(osb, ctxt.handle);
5262 ocfs2_schedule_truncate_log_flush(osb, 1);
5263 ocfs2_run_deallocs(osb, &ctxt.dealloc);
5267 static int ocfs2_delete_xattr_index_block(struct inode *inode,
5268 struct buffer_head *xb_bh)
5270 struct ocfs2_xattr_block *xb =
5271 (struct ocfs2_xattr_block *)xb_bh->b_data;
5272 struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
5274 u32 name_hash = UINT_MAX, e_cpos, num_clusters;
5277 if (le16_to_cpu(el->l_next_free_rec) == 0)
5280 while (name_hash > 0) {
5281 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
5282 &e_cpos, &num_clusters, el);
5288 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
5289 ocfs2_delete_xattr_in_bucket,
5296 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
5297 p_blkno, e_cpos, num_clusters);
5306 name_hash = e_cpos - 1;
5314 * 'security' attributes support
5316 static size_t ocfs2_xattr_security_list(struct inode *inode, char *list,
5317 size_t list_size, const char *name,
5320 const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
5321 const size_t total_len = prefix_len + name_len + 1;
5323 if (list && total_len <= list_size) {
5324 memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
5325 memcpy(list + prefix_len, name, name_len);
5326 list[prefix_len + name_len] = '\0';
5331 static int ocfs2_xattr_security_get(struct inode *inode, const char *name,
5332 void *buffer, size_t size)
5334 if (strcmp(name, "") == 0)
5336 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_SECURITY, name,
5340 static int ocfs2_xattr_security_set(struct inode *inode, const char *name,
5341 const void *value, size_t size, int flags)
5343 if (strcmp(name, "") == 0)
5346 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY, name, value,
5350 int ocfs2_init_security_get(struct inode *inode,
5352 struct ocfs2_security_xattr_info *si)
5354 /* check whether ocfs2 support feature xattr */
5355 if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
5357 return security_inode_init_security(inode, dir, &si->name, &si->value,
5361 int ocfs2_init_security_set(handle_t *handle,
5362 struct inode *inode,
5363 struct buffer_head *di_bh,
5364 struct ocfs2_security_xattr_info *si,
5365 struct ocfs2_alloc_context *xattr_ac,
5366 struct ocfs2_alloc_context *data_ac)
5368 return ocfs2_xattr_set_handle(handle, inode, di_bh,
5369 OCFS2_XATTR_INDEX_SECURITY,
5370 si->name, si->value, si->value_len, 0,
5374 struct xattr_handler ocfs2_xattr_security_handler = {
5375 .prefix = XATTR_SECURITY_PREFIX,
5376 .list = ocfs2_xattr_security_list,
5377 .get = ocfs2_xattr_security_get,
5378 .set = ocfs2_xattr_security_set,
5382 * 'trusted' attributes support
5384 static size_t ocfs2_xattr_trusted_list(struct inode *inode, char *list,
5385 size_t list_size, const char *name,
5388 const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
5389 const size_t total_len = prefix_len + name_len + 1;
5391 if (list && total_len <= list_size) {
5392 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
5393 memcpy(list + prefix_len, name, name_len);
5394 list[prefix_len + name_len] = '\0';
5399 static int ocfs2_xattr_trusted_get(struct inode *inode, const char *name,
5400 void *buffer, size_t size)
5402 if (strcmp(name, "") == 0)
5404 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED, name,
5408 static int ocfs2_xattr_trusted_set(struct inode *inode, const char *name,
5409 const void *value, size_t size, int flags)
5411 if (strcmp(name, "") == 0)
5414 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED, name, value,
5418 struct xattr_handler ocfs2_xattr_trusted_handler = {
5419 .prefix = XATTR_TRUSTED_PREFIX,
5420 .list = ocfs2_xattr_trusted_list,
5421 .get = ocfs2_xattr_trusted_get,
5422 .set = ocfs2_xattr_trusted_set,
5426 * 'user' attributes support
5428 static size_t ocfs2_xattr_user_list(struct inode *inode, char *list,
5429 size_t list_size, const char *name,
5432 const size_t prefix_len = XATTR_USER_PREFIX_LEN;
5433 const size_t total_len = prefix_len + name_len + 1;
5434 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5436 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5439 if (list && total_len <= list_size) {
5440 memcpy(list, XATTR_USER_PREFIX, prefix_len);
5441 memcpy(list + prefix_len, name, name_len);
5442 list[prefix_len + name_len] = '\0';
5447 static int ocfs2_xattr_user_get(struct inode *inode, const char *name,
5448 void *buffer, size_t size)
5450 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5452 if (strcmp(name, "") == 0)
5454 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5456 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
5460 static int ocfs2_xattr_user_set(struct inode *inode, const char *name,
5461 const void *value, size_t size, int flags)
5463 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5465 if (strcmp(name, "") == 0)
5467 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5470 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER, name, value,
5474 struct xattr_handler ocfs2_xattr_user_handler = {
5475 .prefix = XATTR_USER_PREFIX,
5476 .list = ocfs2_xattr_user_list,
5477 .get = ocfs2_xattr_user_get,
5478 .set = ocfs2_xattr_user_set,