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_FREE_IN_IBODY (OCFS2_MIN_XATTR_INLINE_SIZE \
86 - sizeof(struct ocfs2_xattr_header) \
88 #define OCFS2_XATTR_FREE_IN_BLOCK(ptr) ((ptr)->i_sb->s_blocksize \
89 - sizeof(struct ocfs2_xattr_block) \
90 - sizeof(struct ocfs2_xattr_header) \
93 static struct ocfs2_xattr_def_value_root def_xv = {
94 .xv.xr_list.l_count = cpu_to_le16(1),
97 struct xattr_handler *ocfs2_xattr_handlers[] = {
98 &ocfs2_xattr_user_handler,
99 #ifdef CONFIG_OCFS2_FS_POSIX_ACL
100 &ocfs2_xattr_acl_access_handler,
101 &ocfs2_xattr_acl_default_handler,
103 &ocfs2_xattr_trusted_handler,
104 &ocfs2_xattr_security_handler,
108 static struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
109 [OCFS2_XATTR_INDEX_USER] = &ocfs2_xattr_user_handler,
110 #ifdef CONFIG_OCFS2_FS_POSIX_ACL
111 [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS]
112 = &ocfs2_xattr_acl_access_handler,
113 [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT]
114 = &ocfs2_xattr_acl_default_handler,
116 [OCFS2_XATTR_INDEX_TRUSTED] = &ocfs2_xattr_trusted_handler,
117 [OCFS2_XATTR_INDEX_SECURITY] = &ocfs2_xattr_security_handler,
120 struct ocfs2_xattr_info {
127 struct ocfs2_xattr_search {
128 struct buffer_head *inode_bh;
130 * xattr_bh point to the block buffer head which has extended attribute
131 * when extended attribute in inode, xattr_bh is equal to inode_bh.
133 struct buffer_head *xattr_bh;
134 struct ocfs2_xattr_header *header;
135 struct ocfs2_xattr_bucket *bucket;
138 struct ocfs2_xattr_entry *here;
142 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
143 struct ocfs2_xattr_header *xh,
148 static int ocfs2_xattr_block_find(struct inode *inode,
151 struct ocfs2_xattr_search *xs);
152 static int ocfs2_xattr_index_block_find(struct inode *inode,
153 struct buffer_head *root_bh,
156 struct ocfs2_xattr_search *xs);
158 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
159 struct ocfs2_xattr_tree_root *xt,
163 static int ocfs2_xattr_create_index_block(struct inode *inode,
164 struct ocfs2_xattr_search *xs,
165 struct ocfs2_xattr_set_ctxt *ctxt);
167 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
168 struct ocfs2_xattr_info *xi,
169 struct ocfs2_xattr_search *xs,
170 struct ocfs2_xattr_set_ctxt *ctxt);
172 static int ocfs2_delete_xattr_index_block(struct inode *inode,
173 struct buffer_head *xb_bh);
174 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
175 u64 src_blk, u64 last_blk, u64 to_blk,
176 unsigned int start_bucket,
179 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
181 return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
184 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
186 return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
189 static inline u16 ocfs2_xattr_max_xe_in_bucket(struct super_block *sb)
191 u16 len = sb->s_blocksize -
192 offsetof(struct ocfs2_xattr_header, xh_entries);
194 return len / sizeof(struct ocfs2_xattr_entry);
197 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
198 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
199 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
201 static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
203 struct ocfs2_xattr_bucket *bucket;
204 int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
206 BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
208 bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
210 bucket->bu_inode = inode;
211 bucket->bu_blocks = blks;
217 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
221 for (i = 0; i < bucket->bu_blocks; i++) {
222 brelse(bucket->bu_bhs[i]);
223 bucket->bu_bhs[i] = NULL;
227 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
230 ocfs2_xattr_bucket_relse(bucket);
231 bucket->bu_inode = NULL;
237 * A bucket that has never been written to disk doesn't need to be
238 * read. We just need the buffer_heads. Don't call this for
239 * buckets that are already on disk. ocfs2_read_xattr_bucket() initializes
242 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
247 for (i = 0; i < bucket->bu_blocks; i++) {
248 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
250 if (!bucket->bu_bhs[i]) {
256 if (!ocfs2_buffer_uptodate(bucket->bu_inode,
258 ocfs2_set_new_buffer_uptodate(bucket->bu_inode,
263 ocfs2_xattr_bucket_relse(bucket);
267 /* Read the xattr bucket at xb_blkno */
268 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
273 rc = ocfs2_read_blocks(bucket->bu_inode, xb_blkno,
274 bucket->bu_blocks, bucket->bu_bhs, 0,
277 rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
280 &bucket_xh(bucket)->xh_check);
286 ocfs2_xattr_bucket_relse(bucket);
290 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
291 struct ocfs2_xattr_bucket *bucket,
296 for (i = 0; i < bucket->bu_blocks; i++) {
297 rc = ocfs2_journal_access(handle, bucket->bu_inode,
298 bucket->bu_bhs[i], type);
308 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
309 struct ocfs2_xattr_bucket *bucket)
313 ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
314 bucket->bu_bhs, bucket->bu_blocks,
315 &bucket_xh(bucket)->xh_check);
317 for (i = 0; i < bucket->bu_blocks; i++)
318 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
321 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
322 struct ocfs2_xattr_bucket *src)
325 int blocksize = src->bu_inode->i_sb->s_blocksize;
327 BUG_ON(dest->bu_blocks != src->bu_blocks);
328 BUG_ON(dest->bu_inode != src->bu_inode);
330 for (i = 0; i < src->bu_blocks; i++) {
331 memcpy(bucket_block(dest, i), bucket_block(src, i),
336 static int ocfs2_validate_xattr_block(struct super_block *sb,
337 struct buffer_head *bh)
340 struct ocfs2_xattr_block *xb =
341 (struct ocfs2_xattr_block *)bh->b_data;
343 mlog(0, "Validating xattr block %llu\n",
344 (unsigned long long)bh->b_blocknr);
346 BUG_ON(!buffer_uptodate(bh));
349 * If the ecc fails, we return the error but otherwise
350 * leave the filesystem running. We know any error is
351 * local to this block.
353 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
358 * Errors after here are fatal
361 if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
363 "Extended attribute block #%llu has bad "
365 (unsigned long long)bh->b_blocknr, 7,
370 if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
372 "Extended attribute block #%llu has an "
373 "invalid xb_blkno of %llu",
374 (unsigned long long)bh->b_blocknr,
375 (unsigned long long)le64_to_cpu(xb->xb_blkno));
379 if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
381 "Extended attribute block #%llu has an invalid "
382 "xb_fs_generation of #%u",
383 (unsigned long long)bh->b_blocknr,
384 le32_to_cpu(xb->xb_fs_generation));
391 static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
392 struct buffer_head **bh)
395 struct buffer_head *tmp = *bh;
397 rc = ocfs2_read_block(inode, xb_blkno, &tmp,
398 ocfs2_validate_xattr_block);
400 /* If ocfs2_read_block() got us a new bh, pass it up. */
407 static inline const char *ocfs2_xattr_prefix(int name_index)
409 struct xattr_handler *handler = NULL;
411 if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
412 handler = ocfs2_xattr_handler_map[name_index];
414 return handler ? handler->prefix : NULL;
417 static u32 ocfs2_xattr_name_hash(struct inode *inode,
421 /* Get hash value of uuid from super block */
422 u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
425 /* hash extended attribute name */
426 for (i = 0; i < name_len; i++) {
427 hash = (hash << OCFS2_HASH_SHIFT) ^
428 (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
436 * ocfs2_xattr_hash_entry()
438 * Compute the hash of an extended attribute.
440 static void ocfs2_xattr_hash_entry(struct inode *inode,
441 struct ocfs2_xattr_header *header,
442 struct ocfs2_xattr_entry *entry)
445 char *name = (char *)header + le16_to_cpu(entry->xe_name_offset);
447 hash = ocfs2_xattr_name_hash(inode, name, entry->xe_name_len);
448 entry->xe_name_hash = cpu_to_le32(hash);
453 static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
457 if (value_len <= OCFS2_XATTR_INLINE_SIZE)
458 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
460 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
461 size += sizeof(struct ocfs2_xattr_entry);
466 int ocfs2_calc_security_init(struct inode *dir,
467 struct ocfs2_security_xattr_info *si,
470 struct ocfs2_alloc_context **xattr_ac)
473 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
474 int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
478 * The max space of security xattr taken inline is
479 * 256(name) + 80(value) + 16(entry) = 352 bytes,
480 * So reserve one metadata block for it is ok.
482 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
483 s_size > OCFS2_XATTR_FREE_IN_IBODY) {
484 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
489 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
492 /* reserve clusters for xattr value which will be set in B tree*/
493 if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
494 int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
497 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
499 *want_clusters += new_clusters;
504 int ocfs2_calc_xattr_init(struct inode *dir,
505 struct buffer_head *dir_bh,
507 struct ocfs2_security_xattr_info *si,
510 struct ocfs2_alloc_context **xattr_ac)
513 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
514 int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
517 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
520 if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
521 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
522 OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
525 a_size = ocfs2_xattr_entry_real_size(0, acl_len);
528 } else if (acl_len != 0 && acl_len != -ENODATA) {
534 if (!(s_size + a_size))
538 * The max space of security xattr taken inline is
539 * 256(name) + 80(value) + 16(entry) = 352 bytes,
540 * The max space of acl xattr taken inline is
541 * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
542 * when blocksize = 512, may reserve one more cluser for
543 * xattr bucket, otherwise reserve one metadata block
546 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
547 (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
548 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
553 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
556 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
557 (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
559 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
563 * reserve credits and clusters for xattrs which has large value
564 * and have to be set outside
566 if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
567 new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
569 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
571 *want_clusters += new_clusters;
573 if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
574 acl_len > OCFS2_XATTR_INLINE_SIZE) {
575 /* for directory, it has DEFAULT and ACCESS two types of acls */
576 new_clusters = (S_ISDIR(mode) ? 2 : 1) *
577 ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
578 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
580 *want_clusters += new_clusters;
586 static int ocfs2_xattr_extend_allocation(struct inode *inode,
588 struct ocfs2_xattr_value_buf *vb,
589 struct ocfs2_xattr_set_ctxt *ctxt)
592 handle_t *handle = ctxt->handle;
593 enum ocfs2_alloc_restarted why;
594 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
595 u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
596 struct ocfs2_extent_tree et;
598 mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
600 ocfs2_init_xattr_value_extent_tree(&et, inode, vb);
602 status = vb->vb_access(handle, inode, vb->vb_bh,
603 OCFS2_JOURNAL_ACCESS_WRITE);
609 prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
610 status = ocfs2_add_clusters_in_btree(osb,
625 status = ocfs2_journal_dirty(handle, vb->vb_bh);
631 clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) - prev_clusters;
634 * We should have already allocated enough space before the transaction,
635 * so no need to restart.
637 BUG_ON(why != RESTART_NONE || clusters_to_add);
644 static int __ocfs2_remove_xattr_range(struct inode *inode,
645 struct ocfs2_xattr_value_buf *vb,
646 u32 cpos, u32 phys_cpos, u32 len,
647 struct ocfs2_xattr_set_ctxt *ctxt)
650 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
651 handle_t *handle = ctxt->handle;
652 struct ocfs2_extent_tree et;
654 ocfs2_init_xattr_value_extent_tree(&et, inode, vb);
656 ret = vb->vb_access(handle, inode, vb->vb_bh,
657 OCFS2_JOURNAL_ACCESS_WRITE);
663 ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, ctxt->meta_ac,
670 le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
672 ret = ocfs2_journal_dirty(handle, vb->vb_bh);
678 ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc, phys_blkno, len);
686 static int ocfs2_xattr_shrink_size(struct inode *inode,
689 struct ocfs2_xattr_value_buf *vb,
690 struct ocfs2_xattr_set_ctxt *ctxt)
693 u32 trunc_len, cpos, phys_cpos, alloc_size;
696 if (old_clusters <= new_clusters)
700 trunc_len = old_clusters - new_clusters;
702 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
704 &vb->vb_xv->xr_list);
710 if (alloc_size > trunc_len)
711 alloc_size = trunc_len;
713 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
714 phys_cpos, alloc_size,
721 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
722 ocfs2_remove_xattr_clusters_from_cache(inode, block,
725 trunc_len -= alloc_size;
732 static int ocfs2_xattr_value_truncate(struct inode *inode,
733 struct ocfs2_xattr_value_buf *vb,
735 struct ocfs2_xattr_set_ctxt *ctxt)
738 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
739 u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
741 if (new_clusters == old_clusters)
744 if (new_clusters > old_clusters)
745 ret = ocfs2_xattr_extend_allocation(inode,
746 new_clusters - old_clusters,
749 ret = ocfs2_xattr_shrink_size(inode,
750 old_clusters, new_clusters,
756 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
757 size_t *result, const char *prefix,
758 const char *name, int name_len)
760 char *p = buffer + *result;
761 int prefix_len = strlen(prefix);
762 int total_len = prefix_len + name_len + 1;
764 *result += total_len;
766 /* we are just looking for how big our buffer needs to be */
773 memcpy(p, prefix, prefix_len);
774 memcpy(p + prefix_len, name, name_len);
775 p[prefix_len + name_len] = '\0';
780 static int ocfs2_xattr_list_entries(struct inode *inode,
781 struct ocfs2_xattr_header *header,
782 char *buffer, size_t buffer_size)
786 const char *prefix, *name;
788 for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
789 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
790 type = ocfs2_xattr_get_type(entry);
791 prefix = ocfs2_xattr_prefix(type);
794 name = (const char *)header +
795 le16_to_cpu(entry->xe_name_offset);
797 ret = ocfs2_xattr_list_entry(buffer, buffer_size,
798 &result, prefix, name,
808 static int ocfs2_xattr_ibody_list(struct inode *inode,
809 struct ocfs2_dinode *di,
813 struct ocfs2_xattr_header *header = NULL;
814 struct ocfs2_inode_info *oi = OCFS2_I(inode);
817 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
820 header = (struct ocfs2_xattr_header *)
821 ((void *)di + inode->i_sb->s_blocksize -
822 le16_to_cpu(di->i_xattr_inline_size));
824 ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
829 static int ocfs2_xattr_block_list(struct inode *inode,
830 struct ocfs2_dinode *di,
834 struct buffer_head *blk_bh = NULL;
835 struct ocfs2_xattr_block *xb;
838 if (!di->i_xattr_loc)
841 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
848 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
849 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
850 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
851 ret = ocfs2_xattr_list_entries(inode, header,
852 buffer, buffer_size);
854 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
855 ret = ocfs2_xattr_tree_list_index_block(inode, xt,
856 buffer, buffer_size);
864 ssize_t ocfs2_listxattr(struct dentry *dentry,
868 int ret = 0, i_ret = 0, b_ret = 0;
869 struct buffer_head *di_bh = NULL;
870 struct ocfs2_dinode *di = NULL;
871 struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
873 if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
876 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
879 ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
885 di = (struct ocfs2_dinode *)di_bh->b_data;
887 down_read(&oi->ip_xattr_sem);
888 i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
896 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
901 up_read(&oi->ip_xattr_sem);
902 ocfs2_inode_unlock(dentry->d_inode, 0);
906 return i_ret + b_ret;
909 static int ocfs2_xattr_find_entry(int name_index,
911 struct ocfs2_xattr_search *xs)
913 struct ocfs2_xattr_entry *entry;
920 name_len = strlen(name);
922 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
923 cmp = name_index - ocfs2_xattr_get_type(entry);
925 cmp = name_len - entry->xe_name_len;
927 cmp = memcmp(name, (xs->base +
928 le16_to_cpu(entry->xe_name_offset)),
936 return cmp ? -ENODATA : 0;
939 static int ocfs2_xattr_get_value_outside(struct inode *inode,
940 struct ocfs2_xattr_value_root *xv,
944 u32 cpos, p_cluster, num_clusters, bpc, clusters;
947 size_t cplen, blocksize;
948 struct buffer_head *bh = NULL;
949 struct ocfs2_extent_list *el;
952 clusters = le32_to_cpu(xv->xr_clusters);
953 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
954 blocksize = inode->i_sb->s_blocksize;
957 while (cpos < clusters) {
958 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
965 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
966 /* Copy ocfs2_xattr_value */
967 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
968 ret = ocfs2_read_block(inode, blkno, &bh, NULL);
974 cplen = len >= blocksize ? blocksize : len;
975 memcpy(buffer, bh->b_data, cplen);
984 cpos += num_clusters;
990 static int ocfs2_xattr_ibody_get(struct inode *inode,
995 struct ocfs2_xattr_search *xs)
997 struct ocfs2_inode_info *oi = OCFS2_I(inode);
998 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
999 struct ocfs2_xattr_value_root *xv;
1003 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1006 xs->end = (void *)di + inode->i_sb->s_blocksize;
1007 xs->header = (struct ocfs2_xattr_header *)
1008 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1009 xs->base = (void *)xs->header;
1010 xs->here = xs->header->xh_entries;
1012 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1015 size = le64_to_cpu(xs->here->xe_value_size);
1017 if (size > buffer_size)
1019 if (ocfs2_xattr_is_local(xs->here)) {
1020 memcpy(buffer, (void *)xs->base +
1021 le16_to_cpu(xs->here->xe_name_offset) +
1022 OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1024 xv = (struct ocfs2_xattr_value_root *)
1025 (xs->base + le16_to_cpu(
1026 xs->here->xe_name_offset) +
1027 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1028 ret = ocfs2_xattr_get_value_outside(inode, xv,
1040 static int ocfs2_xattr_block_get(struct inode *inode,
1045 struct ocfs2_xattr_search *xs)
1047 struct ocfs2_xattr_block *xb;
1048 struct ocfs2_xattr_value_root *xv;
1050 int ret = -ENODATA, name_offset, name_len, block_off, i;
1052 xs->bucket = ocfs2_xattr_bucket_new(inode);
1059 ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1065 if (xs->not_found) {
1070 xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1071 size = le64_to_cpu(xs->here->xe_value_size);
1074 if (size > buffer_size)
1077 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1078 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1079 i = xs->here - xs->header->xh_entries;
1081 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1082 ret = ocfs2_xattr_bucket_get_name_value(inode,
1083 bucket_xh(xs->bucket),
1087 xs->base = bucket_block(xs->bucket, block_off);
1089 if (ocfs2_xattr_is_local(xs->here)) {
1090 memcpy(buffer, (void *)xs->base +
1091 name_offset + name_len, size);
1093 xv = (struct ocfs2_xattr_value_root *)
1094 (xs->base + name_offset + name_len);
1095 ret = ocfs2_xattr_get_value_outside(inode, xv,
1105 ocfs2_xattr_bucket_free(xs->bucket);
1107 brelse(xs->xattr_bh);
1108 xs->xattr_bh = NULL;
1112 int ocfs2_xattr_get_nolock(struct inode *inode,
1113 struct buffer_head *di_bh,
1120 struct ocfs2_dinode *di = NULL;
1121 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1122 struct ocfs2_xattr_search xis = {
1123 .not_found = -ENODATA,
1125 struct ocfs2_xattr_search xbs = {
1126 .not_found = -ENODATA,
1129 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1132 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1135 xis.inode_bh = xbs.inode_bh = di_bh;
1136 di = (struct ocfs2_dinode *)di_bh->b_data;
1138 down_read(&oi->ip_xattr_sem);
1139 ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1141 if (ret == -ENODATA && di->i_xattr_loc)
1142 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1144 up_read(&oi->ip_xattr_sem);
1149 /* ocfs2_xattr_get()
1151 * Copy an extended attribute into the buffer provided.
1152 * Buffer is NULL to compute the size of buffer required.
1154 static int ocfs2_xattr_get(struct inode *inode,
1161 struct buffer_head *di_bh = NULL;
1163 ret = ocfs2_inode_lock(inode, &di_bh, 0);
1168 ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1169 name, buffer, buffer_size);
1171 ocfs2_inode_unlock(inode, 0);
1178 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1180 struct ocfs2_xattr_value_root *xv,
1184 int ret = 0, i, cp_len;
1185 u16 blocksize = inode->i_sb->s_blocksize;
1186 u32 p_cluster, num_clusters;
1187 u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1188 u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1190 struct buffer_head *bh = NULL;
1192 BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1194 while (cpos < clusters) {
1195 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1196 &num_clusters, &xv->xr_list);
1202 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1204 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1205 ret = ocfs2_read_block(inode, blkno, &bh, NULL);
1211 ret = ocfs2_journal_access(handle,
1214 OCFS2_JOURNAL_ACCESS_WRITE);
1220 cp_len = value_len > blocksize ? blocksize : value_len;
1221 memcpy(bh->b_data, value, cp_len);
1222 value_len -= cp_len;
1224 if (cp_len < blocksize)
1225 memset(bh->b_data + cp_len, 0,
1226 blocksize - cp_len);
1228 ret = ocfs2_journal_dirty(handle, bh);
1237 * XXX: do we need to empty all the following
1238 * blocks in this cluster?
1243 cpos += num_clusters;
1251 static int ocfs2_xattr_cleanup(struct inode *inode,
1253 struct ocfs2_xattr_info *xi,
1254 struct ocfs2_xattr_search *xs,
1255 struct ocfs2_xattr_value_buf *vb,
1259 size_t name_len = strlen(xi->name);
1260 void *val = xs->base + offs;
1261 size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1263 ret = vb->vb_access(handle, inode, vb->vb_bh,
1264 OCFS2_JOURNAL_ACCESS_WRITE);
1269 /* Decrease xattr count */
1270 le16_add_cpu(&xs->header->xh_count, -1);
1271 /* Remove the xattr entry and tree root which has already be set*/
1272 memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
1273 memset(val, 0, size);
1275 ret = ocfs2_journal_dirty(handle, vb->vb_bh);
1282 static int ocfs2_xattr_update_entry(struct inode *inode,
1284 struct ocfs2_xattr_info *xi,
1285 struct ocfs2_xattr_search *xs,
1286 struct ocfs2_xattr_value_buf *vb,
1291 ret = vb->vb_access(handle, inode, vb->vb_bh,
1292 OCFS2_JOURNAL_ACCESS_WRITE);
1298 xs->here->xe_name_offset = cpu_to_le16(offs);
1299 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1300 if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1301 ocfs2_xattr_set_local(xs->here, 1);
1303 ocfs2_xattr_set_local(xs->here, 0);
1304 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1306 ret = ocfs2_journal_dirty(handle, vb->vb_bh);
1314 * ocfs2_xattr_set_value_outside()
1316 * Set large size value in B tree.
1318 static int ocfs2_xattr_set_value_outside(struct inode *inode,
1319 struct ocfs2_xattr_info *xi,
1320 struct ocfs2_xattr_search *xs,
1321 struct ocfs2_xattr_set_ctxt *ctxt,
1322 struct ocfs2_xattr_value_buf *vb,
1325 size_t name_len = strlen(xi->name);
1326 void *val = xs->base + offs;
1327 struct ocfs2_xattr_value_root *xv = NULL;
1328 size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1331 memset(val, 0, size);
1332 memcpy(val, xi->name, name_len);
1333 xv = (struct ocfs2_xattr_value_root *)
1334 (val + OCFS2_XATTR_SIZE(name_len));
1335 xv->xr_clusters = 0;
1336 xv->xr_last_eb_blk = 0;
1337 xv->xr_list.l_tree_depth = 0;
1338 xv->xr_list.l_count = cpu_to_le16(1);
1339 xv->xr_list.l_next_free_rec = 0;
1342 ret = ocfs2_xattr_value_truncate(inode, vb, xi->value_len, ctxt);
1347 ret = ocfs2_xattr_update_entry(inode, ctxt->handle, xi, xs, vb, offs);
1352 ret = __ocfs2_xattr_set_value_outside(inode, ctxt->handle, vb->vb_xv,
1353 xi->value, xi->value_len);
1361 * ocfs2_xattr_set_entry_local()
1363 * Set, replace or remove extended attribute in local.
1365 static void ocfs2_xattr_set_entry_local(struct inode *inode,
1366 struct ocfs2_xattr_info *xi,
1367 struct ocfs2_xattr_search *xs,
1368 struct ocfs2_xattr_entry *last,
1371 size_t name_len = strlen(xi->name);
1374 if (xi->value && xs->not_found) {
1375 /* Insert the new xattr entry. */
1376 le16_add_cpu(&xs->header->xh_count, 1);
1377 ocfs2_xattr_set_type(last, xi->name_index);
1378 ocfs2_xattr_set_local(last, 1);
1379 last->xe_name_len = name_len;
1385 first_val = xs->base + min_offs;
1386 offs = le16_to_cpu(xs->here->xe_name_offset);
1387 val = xs->base + offs;
1389 if (le64_to_cpu(xs->here->xe_value_size) >
1390 OCFS2_XATTR_INLINE_SIZE)
1391 size = OCFS2_XATTR_SIZE(name_len) +
1392 OCFS2_XATTR_ROOT_SIZE;
1394 size = OCFS2_XATTR_SIZE(name_len) +
1395 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1397 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1398 OCFS2_XATTR_SIZE(xi->value_len)) {
1399 /* The old and the new value have the
1400 same size. Just replace the value. */
1401 ocfs2_xattr_set_local(xs->here, 1);
1402 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1403 /* Clear value bytes. */
1404 memset(val + OCFS2_XATTR_SIZE(name_len),
1406 OCFS2_XATTR_SIZE(xi->value_len));
1407 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1412 /* Remove the old name+value. */
1413 memmove(first_val + size, first_val, val - first_val);
1414 memset(first_val, 0, size);
1415 xs->here->xe_name_hash = 0;
1416 xs->here->xe_name_offset = 0;
1417 ocfs2_xattr_set_local(xs->here, 1);
1418 xs->here->xe_value_size = 0;
1422 /* Adjust all value offsets. */
1423 last = xs->header->xh_entries;
1424 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1425 size_t o = le16_to_cpu(last->xe_name_offset);
1428 last->xe_name_offset = cpu_to_le16(o + size);
1433 /* Remove the old entry. */
1435 memmove(xs->here, xs->here + 1,
1436 (void *)last - (void *)xs->here);
1437 memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1438 le16_add_cpu(&xs->header->xh_count, -1);
1442 /* Insert the new name+value. */
1443 size_t size = OCFS2_XATTR_SIZE(name_len) +
1444 OCFS2_XATTR_SIZE(xi->value_len);
1445 void *val = xs->base + min_offs - size;
1447 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1448 memset(val, 0, size);
1449 memcpy(val, xi->name, name_len);
1450 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1453 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1454 ocfs2_xattr_set_local(xs->here, 1);
1455 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1462 * ocfs2_xattr_set_entry()
1464 * Set extended attribute entry into inode or block.
1466 * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1467 * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1468 * then set value in B tree with set_value_outside().
1470 static int ocfs2_xattr_set_entry(struct inode *inode,
1471 struct ocfs2_xattr_info *xi,
1472 struct ocfs2_xattr_search *xs,
1473 struct ocfs2_xattr_set_ctxt *ctxt,
1476 struct ocfs2_xattr_entry *last;
1477 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1478 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1479 size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1481 handle_t *handle = ctxt->handle;
1483 struct ocfs2_xattr_info xi_l = {
1484 .name_index = xi->name_index,
1487 .value_len = xi->value_len,
1489 struct ocfs2_xattr_value_buf vb = {
1490 .vb_bh = xs->xattr_bh,
1491 .vb_access = ocfs2_journal_access_di,
1494 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1495 BUG_ON(xs->xattr_bh == xs->inode_bh);
1496 vb.vb_access = ocfs2_journal_access_xb;
1498 BUG_ON(xs->xattr_bh != xs->inode_bh);
1500 /* Compute min_offs, last and free space. */
1501 last = xs->header->xh_entries;
1503 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1504 size_t offs = le16_to_cpu(last->xe_name_offset);
1505 if (offs < min_offs)
1510 free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
1514 if (!xs->not_found) {
1516 if (ocfs2_xattr_is_local(xs->here))
1517 size = OCFS2_XATTR_SIZE(name_len) +
1518 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1520 size = OCFS2_XATTR_SIZE(name_len) +
1521 OCFS2_XATTR_ROOT_SIZE;
1522 free += (size + sizeof(struct ocfs2_xattr_entry));
1524 /* Check free space in inode or block */
1525 if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1526 if (free < sizeof(struct ocfs2_xattr_entry) +
1527 OCFS2_XATTR_SIZE(name_len) +
1528 OCFS2_XATTR_ROOT_SIZE) {
1532 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1533 xi_l.value = (void *)&def_xv;
1534 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1535 } else if (xi->value) {
1536 if (free < sizeof(struct ocfs2_xattr_entry) +
1537 OCFS2_XATTR_SIZE(name_len) +
1538 OCFS2_XATTR_SIZE(xi->value_len)) {
1544 if (!xs->not_found) {
1545 /* For existing extended attribute */
1546 size_t size = OCFS2_XATTR_SIZE(name_len) +
1547 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1548 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1549 void *val = xs->base + offs;
1551 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1552 /* Replace existing local xattr with tree root */
1553 ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1558 } else if (!ocfs2_xattr_is_local(xs->here)) {
1559 /* For existing xattr which has value outside */
1560 vb.vb_xv = (struct ocfs2_xattr_value_root *)
1561 (val + OCFS2_XATTR_SIZE(name_len));
1563 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1565 * If new value need set outside also,
1566 * first truncate old value to new value,
1567 * then set new value with set_value_outside().
1569 ret = ocfs2_xattr_value_truncate(inode,
1578 ret = ocfs2_xattr_update_entry(inode,
1589 ret = __ocfs2_xattr_set_value_outside(inode,
1599 * If new value need set in local,
1600 * just trucate old value to zero.
1602 ret = ocfs2_xattr_value_truncate(inode,
1612 ret = ocfs2_journal_access_di(handle, inode, xs->inode_bh,
1613 OCFS2_JOURNAL_ACCESS_WRITE);
1619 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1620 ret = vb.vb_access(handle, inode, vb.vb_bh,
1621 OCFS2_JOURNAL_ACCESS_WRITE);
1629 * Set value in local, include set tree root in local.
1630 * This is the first step for value size >INLINE_SIZE.
1632 ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1634 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1635 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1642 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1643 (flag & OCFS2_INLINE_XATTR_FL)) {
1644 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1645 unsigned int xattrsize = osb->s_xattr_inline_size;
1648 * Adjust extent record count or inline data size
1649 * to reserve space for extended attribute.
1651 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1652 struct ocfs2_inline_data *idata = &di->id2.i_data;
1653 le16_add_cpu(&idata->id_count, -xattrsize);
1654 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1655 struct ocfs2_extent_list *el = &di->id2.i_list;
1656 le16_add_cpu(&el->l_count, -(xattrsize /
1657 sizeof(struct ocfs2_extent_rec)));
1659 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1661 /* Update xattr flag */
1662 spin_lock(&oi->ip_lock);
1663 oi->ip_dyn_features |= flag;
1664 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1665 spin_unlock(&oi->ip_lock);
1667 ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1671 if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1673 * Set value outside in B tree.
1674 * This is the second step for value size > INLINE_SIZE.
1676 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1677 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, ctxt,
1684 * If set value outside failed, we have to clean
1685 * the junk tree root we have already set in local.
1687 ret2 = ocfs2_xattr_cleanup(inode, ctxt->handle,
1697 static int ocfs2_remove_value_outside(struct inode*inode,
1698 struct ocfs2_xattr_value_buf *vb,
1699 struct ocfs2_xattr_header *header)
1702 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1703 struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
1705 ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
1707 ctxt.handle = ocfs2_start_trans(osb,
1708 ocfs2_remove_extent_credits(osb->sb));
1709 if (IS_ERR(ctxt.handle)) {
1710 ret = PTR_ERR(ctxt.handle);
1715 for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1716 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1718 if (!ocfs2_xattr_is_local(entry)) {
1721 val = (void *)header +
1722 le16_to_cpu(entry->xe_name_offset);
1723 vb->vb_xv = (struct ocfs2_xattr_value_root *)
1724 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
1725 ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
1733 ocfs2_commit_trans(osb, ctxt.handle);
1734 ocfs2_schedule_truncate_log_flush(osb, 1);
1735 ocfs2_run_deallocs(osb, &ctxt.dealloc);
1740 static int ocfs2_xattr_ibody_remove(struct inode *inode,
1741 struct buffer_head *di_bh)
1744 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1745 struct ocfs2_xattr_header *header;
1747 struct ocfs2_xattr_value_buf vb = {
1749 .vb_access = ocfs2_journal_access_di,
1752 header = (struct ocfs2_xattr_header *)
1753 ((void *)di + inode->i_sb->s_blocksize -
1754 le16_to_cpu(di->i_xattr_inline_size));
1756 ret = ocfs2_remove_value_outside(inode, &vb, header);
1761 static int ocfs2_xattr_block_remove(struct inode *inode,
1762 struct buffer_head *blk_bh)
1764 struct ocfs2_xattr_block *xb;
1766 struct ocfs2_xattr_value_buf vb = {
1768 .vb_access = ocfs2_journal_access_xb,
1771 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1772 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1773 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1774 ret = ocfs2_remove_value_outside(inode, &vb, header);
1776 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
1781 static int ocfs2_xattr_free_block(struct inode *inode,
1784 struct inode *xb_alloc_inode;
1785 struct buffer_head *xb_alloc_bh = NULL;
1786 struct buffer_head *blk_bh = NULL;
1787 struct ocfs2_xattr_block *xb;
1788 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1794 ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
1800 ret = ocfs2_xattr_block_remove(inode, blk_bh);
1806 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1807 blk = le64_to_cpu(xb->xb_blkno);
1808 bit = le16_to_cpu(xb->xb_suballoc_bit);
1809 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1811 xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1812 EXTENT_ALLOC_SYSTEM_INODE,
1813 le16_to_cpu(xb->xb_suballoc_slot));
1814 if (!xb_alloc_inode) {
1819 mutex_lock(&xb_alloc_inode->i_mutex);
1821 ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1827 handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
1828 if (IS_ERR(handle)) {
1829 ret = PTR_ERR(handle);
1834 ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1839 ocfs2_commit_trans(osb, handle);
1841 ocfs2_inode_unlock(xb_alloc_inode, 1);
1842 brelse(xb_alloc_bh);
1844 mutex_unlock(&xb_alloc_inode->i_mutex);
1845 iput(xb_alloc_inode);
1852 * ocfs2_xattr_remove()
1854 * Free extended attribute resources associated with this inode.
1856 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1858 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1859 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1863 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1866 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1869 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1870 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1877 if (di->i_xattr_loc) {
1878 ret = ocfs2_xattr_free_block(inode,
1879 le64_to_cpu(di->i_xattr_loc));
1886 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1887 OCFS2_INODE_UPDATE_CREDITS);
1888 if (IS_ERR(handle)) {
1889 ret = PTR_ERR(handle);
1893 ret = ocfs2_journal_access_di(handle, inode, di_bh,
1894 OCFS2_JOURNAL_ACCESS_WRITE);
1900 di->i_xattr_loc = 0;
1902 spin_lock(&oi->ip_lock);
1903 oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1904 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1905 spin_unlock(&oi->ip_lock);
1907 ret = ocfs2_journal_dirty(handle, di_bh);
1911 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1916 static int ocfs2_xattr_has_space_inline(struct inode *inode,
1917 struct ocfs2_dinode *di)
1919 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1920 unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1923 if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1926 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1927 struct ocfs2_inline_data *idata = &di->id2.i_data;
1928 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1929 } else if (ocfs2_inode_is_fast_symlink(inode)) {
1930 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1931 le64_to_cpu(di->i_size);
1933 struct ocfs2_extent_list *el = &di->id2.i_list;
1934 free = (le16_to_cpu(el->l_count) -
1935 le16_to_cpu(el->l_next_free_rec)) *
1936 sizeof(struct ocfs2_extent_rec);
1938 if (free >= xattrsize)
1945 * ocfs2_xattr_ibody_find()
1947 * Find extended attribute in inode block and
1948 * fill search info into struct ocfs2_xattr_search.
1950 static int ocfs2_xattr_ibody_find(struct inode *inode,
1953 struct ocfs2_xattr_search *xs)
1955 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1956 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1960 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1963 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1964 down_read(&oi->ip_alloc_sem);
1965 has_space = ocfs2_xattr_has_space_inline(inode, di);
1966 up_read(&oi->ip_alloc_sem);
1971 xs->xattr_bh = xs->inode_bh;
1972 xs->end = (void *)di + inode->i_sb->s_blocksize;
1973 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1974 xs->header = (struct ocfs2_xattr_header *)
1975 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1977 xs->header = (struct ocfs2_xattr_header *)
1978 (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
1979 xs->base = (void *)xs->header;
1980 xs->here = xs->header->xh_entries;
1982 /* Find the named attribute. */
1983 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1984 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1985 if (ret && ret != -ENODATA)
1987 xs->not_found = ret;
1994 * ocfs2_xattr_ibody_set()
1996 * Set, replace or remove an extended attribute into inode block.
1999 static int ocfs2_xattr_ibody_set(struct inode *inode,
2000 struct ocfs2_xattr_info *xi,
2001 struct ocfs2_xattr_search *xs,
2002 struct ocfs2_xattr_set_ctxt *ctxt)
2004 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2005 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2008 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2011 down_write(&oi->ip_alloc_sem);
2012 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2013 if (!ocfs2_xattr_has_space_inline(inode, di)) {
2019 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2020 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
2022 up_write(&oi->ip_alloc_sem);
2028 * ocfs2_xattr_block_find()
2030 * Find extended attribute in external block and
2031 * fill search info into struct ocfs2_xattr_search.
2033 static int ocfs2_xattr_block_find(struct inode *inode,
2036 struct ocfs2_xattr_search *xs)
2038 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2039 struct buffer_head *blk_bh = NULL;
2040 struct ocfs2_xattr_block *xb;
2043 if (!di->i_xattr_loc)
2046 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2053 xs->xattr_bh = blk_bh;
2054 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2056 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2057 xs->header = &xb->xb_attrs.xb_header;
2058 xs->base = (void *)xs->header;
2059 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2060 xs->here = xs->header->xh_entries;
2062 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2064 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2068 if (ret && ret != -ENODATA) {
2069 xs->xattr_bh = NULL;
2072 xs->not_found = ret;
2081 * ocfs2_xattr_block_set()
2083 * Set, replace or remove an extended attribute into external block.
2086 static int ocfs2_xattr_block_set(struct inode *inode,
2087 struct ocfs2_xattr_info *xi,
2088 struct ocfs2_xattr_search *xs,
2089 struct ocfs2_xattr_set_ctxt *ctxt)
2091 struct buffer_head *new_bh = NULL;
2092 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2093 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2094 handle_t *handle = ctxt->handle;
2095 struct ocfs2_xattr_block *xblk = NULL;
2096 u16 suballoc_bit_start;
2101 if (!xs->xattr_bh) {
2102 ret = ocfs2_journal_access_di(handle, inode, xs->inode_bh,
2103 OCFS2_JOURNAL_ACCESS_CREATE);
2109 ret = ocfs2_claim_metadata(osb, handle, ctxt->meta_ac, 1,
2110 &suballoc_bit_start, &num_got,
2117 new_bh = sb_getblk(inode->i_sb, first_blkno);
2118 ocfs2_set_new_buffer_uptodate(inode, new_bh);
2120 ret = ocfs2_journal_access_xb(handle, inode, new_bh,
2121 OCFS2_JOURNAL_ACCESS_CREATE);
2127 /* Initialize ocfs2_xattr_block */
2128 xs->xattr_bh = new_bh;
2129 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2130 memset(xblk, 0, inode->i_sb->s_blocksize);
2131 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2132 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
2133 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2134 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
2135 xblk->xb_blkno = cpu_to_le64(first_blkno);
2137 xs->header = &xblk->xb_attrs.xb_header;
2138 xs->base = (void *)xs->header;
2139 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2140 xs->here = xs->header->xh_entries;
2142 ret = ocfs2_journal_dirty(handle, new_bh);
2147 di->i_xattr_loc = cpu_to_le64(first_blkno);
2148 ocfs2_journal_dirty(handle, xs->inode_bh);
2150 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2152 if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2153 /* Set extended attribute into external block */
2154 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2155 OCFS2_HAS_XATTR_FL);
2156 if (!ret || ret != -ENOSPC)
2159 ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2164 ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2171 /* Check whether the new xattr can be inserted into the inode. */
2172 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2173 struct ocfs2_xattr_info *xi,
2174 struct ocfs2_xattr_search *xs)
2177 struct ocfs2_xattr_entry *last;
2179 size_t min_offs = xs->end - xs->base;
2184 last = xs->header->xh_entries;
2186 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
2187 size_t offs = le16_to_cpu(last->xe_name_offset);
2188 if (offs < min_offs)
2193 free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
2197 BUG_ON(!xs->not_found);
2199 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2200 value_size = OCFS2_XATTR_ROOT_SIZE;
2202 value_size = OCFS2_XATTR_SIZE(xi->value_len);
2204 if (free >= sizeof(struct ocfs2_xattr_entry) +
2205 OCFS2_XATTR_SIZE(strlen(xi->name)) + value_size)
2211 static int ocfs2_calc_xattr_set_need(struct inode *inode,
2212 struct ocfs2_dinode *di,
2213 struct ocfs2_xattr_info *xi,
2214 struct ocfs2_xattr_search *xis,
2215 struct ocfs2_xattr_search *xbs,
2220 int ret = 0, old_in_xb = 0;
2221 int clusters_add = 0, meta_add = 0, credits = 0;
2222 struct buffer_head *bh = NULL;
2223 struct ocfs2_xattr_block *xb = NULL;
2224 struct ocfs2_xattr_entry *xe = NULL;
2225 struct ocfs2_xattr_value_root *xv = NULL;
2227 int name_offset, name_len = 0;
2228 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2233 * Calculate the clusters we need to write.
2234 * No matter whether we replace an old one or add a new one,
2235 * we need this for writing.
2237 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2238 credits += new_clusters *
2239 ocfs2_clusters_to_blocks(inode->i_sb, 1);
2241 if (xis->not_found && xbs->not_found) {
2242 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2244 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2245 clusters_add += new_clusters;
2246 credits += ocfs2_calc_extend_credits(inode->i_sb,
2254 if (!xis->not_found) {
2256 name_offset = le16_to_cpu(xe->xe_name_offset);
2257 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2259 credits += OCFS2_INODE_UPDATE_CREDITS;
2261 int i, block_off = 0;
2262 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2264 name_offset = le16_to_cpu(xe->xe_name_offset);
2265 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2266 i = xbs->here - xbs->header->xh_entries;
2269 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2270 ret = ocfs2_xattr_bucket_get_name_value(inode,
2271 bucket_xh(xbs->bucket),
2274 base = bucket_block(xbs->bucket, block_off);
2275 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2278 credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
2283 * delete a xattr doesn't need metadata and cluster allocation.
2284 * so just calculate the credits and return.
2286 * The credits for removing the value tree will be extended
2287 * by ocfs2_remove_extent itself.
2290 if (!ocfs2_xattr_is_local(xe))
2291 credits += ocfs2_remove_extent_credits(inode->i_sb);
2296 /* do cluster allocation guess first. */
2297 value_size = le64_to_cpu(xe->xe_value_size);
2301 * In xattr set, we always try to set the xe in inode first,
2302 * so if it can be inserted into inode successfully, the old
2303 * one will be removed from the xattr block, and this xattr
2304 * will be inserted into inode as a new xattr in inode.
2306 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
2307 clusters_add += new_clusters;
2308 credits += ocfs2_remove_extent_credits(inode->i_sb) +
2309 OCFS2_INODE_UPDATE_CREDITS;
2310 if (!ocfs2_xattr_is_local(xe))
2311 credits += ocfs2_calc_extend_credits(
2319 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2320 /* the new values will be stored outside. */
2321 u32 old_clusters = 0;
2323 if (!ocfs2_xattr_is_local(xe)) {
2324 old_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2326 xv = (struct ocfs2_xattr_value_root *)
2327 (base + name_offset + name_len);
2328 value_size = OCFS2_XATTR_ROOT_SIZE;
2332 if (old_clusters >= new_clusters) {
2333 credits += ocfs2_remove_extent_credits(inode->i_sb);
2336 meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
2337 clusters_add += new_clusters - old_clusters;
2338 credits += ocfs2_calc_extend_credits(inode->i_sb,
2342 if (value_size >= OCFS2_XATTR_ROOT_SIZE)
2347 * Now the new value will be stored inside. So if the new
2348 * value is smaller than the size of value root or the old
2349 * value, we don't need any allocation, otherwise we have
2350 * to guess metadata allocation.
2352 if ((ocfs2_xattr_is_local(xe) && value_size >= xi->value_len) ||
2353 (!ocfs2_xattr_is_local(xe) &&
2354 OCFS2_XATTR_ROOT_SIZE >= xi->value_len))
2359 /* calculate metadata allocation. */
2360 if (di->i_xattr_loc) {
2361 if (!xbs->xattr_bh) {
2362 ret = ocfs2_read_xattr_block(inode,
2363 le64_to_cpu(di->i_xattr_loc),
2370 xb = (struct ocfs2_xattr_block *)bh->b_data;
2372 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2375 * If there is already an xattr tree, good, we can calculate
2376 * like other b-trees. Otherwise we may have the chance of
2377 * create a tree, the credit calculation is borrowed from
2378 * ocfs2_calc_extend_credits with root_el = NULL. And the
2379 * new tree will be cluster based, so no meta is needed.
2381 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2382 struct ocfs2_extent_list *el =
2383 &xb->xb_attrs.xb_root.xt_list;
2384 meta_add += ocfs2_extend_meta_needed(el);
2385 credits += ocfs2_calc_extend_credits(inode->i_sb,
2388 credits += OCFS2_SUBALLOC_ALLOC + 1;
2391 * This cluster will be used either for new bucket or for
2393 * If the cluster size is the same as the bucket size, one
2394 * more is needed since we may need to extend the bucket
2398 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2399 if (OCFS2_XATTR_BUCKET_SIZE ==
2400 OCFS2_SB(inode->i_sb)->s_clustersize) {
2401 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2406 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
2410 *clusters_need = clusters_add;
2412 *meta_need = meta_add;
2414 *credits_need = credits;
2419 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
2420 struct ocfs2_dinode *di,
2421 struct ocfs2_xattr_info *xi,
2422 struct ocfs2_xattr_search *xis,
2423 struct ocfs2_xattr_search *xbs,
2424 struct ocfs2_xattr_set_ctxt *ctxt,
2427 int clusters_add, meta_add, ret;
2428 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2430 memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
2432 ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
2434 ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
2435 &clusters_add, &meta_add, credits);
2441 mlog(0, "Set xattr %s, reserve meta blocks = %d, clusters = %d, "
2442 "credits = %d\n", xi->name, meta_add, clusters_add, *credits);
2445 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
2454 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
2460 if (ctxt->meta_ac) {
2461 ocfs2_free_alloc_context(ctxt->meta_ac);
2462 ctxt->meta_ac = NULL;
2466 * We cannot have an error and a non null ctxt->data_ac.
2473 static int __ocfs2_xattr_set_handle(struct inode *inode,
2474 struct ocfs2_dinode *di,
2475 struct ocfs2_xattr_info *xi,
2476 struct ocfs2_xattr_search *xis,
2477 struct ocfs2_xattr_search *xbs,
2478 struct ocfs2_xattr_set_ctxt *ctxt)
2480 int ret = 0, credits, old_found;
2483 /* Remove existing extended attribute */
2484 if (!xis->not_found)
2485 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2486 else if (!xbs->not_found)
2487 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2489 /* We always try to set extended attribute into inode first*/
2490 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2491 if (!ret && !xbs->not_found) {
2493 * If succeed and that extended attribute existing in
2494 * external block, then we will remove it.
2499 old_found = xis->not_found;
2500 xis->not_found = -ENODATA;
2501 ret = ocfs2_calc_xattr_set_need(inode,
2509 xis->not_found = old_found;
2515 ret = ocfs2_extend_trans(ctxt->handle, credits +
2516 ctxt->handle->h_buffer_credits);
2521 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2522 } else if (ret == -ENOSPC) {
2523 if (di->i_xattr_loc && !xbs->xattr_bh) {
2524 ret = ocfs2_xattr_block_find(inode,
2530 old_found = xis->not_found;
2531 xis->not_found = -ENODATA;
2532 ret = ocfs2_calc_xattr_set_need(inode,
2540 xis->not_found = old_found;
2546 ret = ocfs2_extend_trans(ctxt->handle, credits +
2547 ctxt->handle->h_buffer_credits);
2554 * If no space in inode, we will set extended attribute
2555 * into external block.
2557 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2560 if (!xis->not_found) {
2562 * If succeed and that extended attribute
2563 * existing in inode, we will remove it.
2567 xbs->not_found = -ENODATA;
2568 ret = ocfs2_calc_xattr_set_need(inode,
2581 ret = ocfs2_extend_trans(ctxt->handle, credits +
2582 ctxt->handle->h_buffer_credits);
2587 ret = ocfs2_xattr_ibody_set(inode, xi,
2594 /* Update inode ctime. */
2595 ret = ocfs2_journal_access_di(ctxt->handle, inode,
2597 OCFS2_JOURNAL_ACCESS_WRITE);
2603 inode->i_ctime = CURRENT_TIME;
2604 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
2605 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
2606 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
2613 * This function only called duing creating inode
2614 * for init security/acl xattrs of the new inode.
2615 * All transanction credits have been reserved in mknod.
2617 int ocfs2_xattr_set_handle(handle_t *handle,
2618 struct inode *inode,
2619 struct buffer_head *di_bh,
2625 struct ocfs2_alloc_context *meta_ac,
2626 struct ocfs2_alloc_context *data_ac)
2628 struct ocfs2_dinode *di;
2631 struct ocfs2_xattr_info xi = {
2632 .name_index = name_index,
2635 .value_len = value_len,
2638 struct ocfs2_xattr_search xis = {
2639 .not_found = -ENODATA,
2642 struct ocfs2_xattr_search xbs = {
2643 .not_found = -ENODATA,
2646 struct ocfs2_xattr_set_ctxt ctxt = {
2652 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2656 * In extreme situation, may need xattr bucket when
2657 * block size is too small. And we have already reserved
2658 * the credits for bucket in mknod.
2660 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
2661 xbs.bucket = ocfs2_xattr_bucket_new(inode);
2663 mlog_errno(-ENOMEM);
2668 xis.inode_bh = xbs.inode_bh = di_bh;
2669 di = (struct ocfs2_dinode *)di_bh->b_data;
2671 down_write(&OCFS2_I(inode)->ip_xattr_sem);
2673 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2676 if (xis.not_found) {
2677 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2682 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2685 up_write(&OCFS2_I(inode)->ip_xattr_sem);
2686 brelse(xbs.xattr_bh);
2687 ocfs2_xattr_bucket_free(xbs.bucket);
2695 * Set, replace or remove an extended attribute for this inode.
2696 * value is NULL to remove an existing extended attribute, else either
2697 * create or replace an extended attribute.
2699 int ocfs2_xattr_set(struct inode *inode,
2706 struct buffer_head *di_bh = NULL;
2707 struct ocfs2_dinode *di;
2709 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2710 struct inode *tl_inode = osb->osb_tl_inode;
2711 struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2713 struct ocfs2_xattr_info xi = {
2714 .name_index = name_index,
2717 .value_len = value_len,
2720 struct ocfs2_xattr_search xis = {
2721 .not_found = -ENODATA,
2724 struct ocfs2_xattr_search xbs = {
2725 .not_found = -ENODATA,
2728 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2732 * Only xbs will be used on indexed trees. xis doesn't need a
2735 xbs.bucket = ocfs2_xattr_bucket_new(inode);
2737 mlog_errno(-ENOMEM);
2741 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2744 goto cleanup_nolock;
2746 xis.inode_bh = xbs.inode_bh = di_bh;
2747 di = (struct ocfs2_dinode *)di_bh->b_data;
2749 down_write(&OCFS2_I(inode)->ip_xattr_sem);
2751 * Scan inode and external block to find the same name
2752 * extended attribute and collect search infomation.
2754 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2757 if (xis.not_found) {
2758 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2763 if (xis.not_found && xbs.not_found) {
2765 if (flags & XATTR_REPLACE)
2772 if (flags & XATTR_CREATE)
2777 mutex_lock(&tl_inode->i_mutex);
2779 if (ocfs2_truncate_log_needs_flush(osb)) {
2780 ret = __ocfs2_flush_truncate_log(osb);
2782 mutex_unlock(&tl_inode->i_mutex);
2787 mutex_unlock(&tl_inode->i_mutex);
2789 ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
2790 &xbs, &ctxt, &credits);
2796 /* we need to update inode's ctime field, so add credit for it. */
2797 credits += OCFS2_INODE_UPDATE_CREDITS;
2798 ctxt.handle = ocfs2_start_trans(osb, credits);
2799 if (IS_ERR(ctxt.handle)) {
2800 ret = PTR_ERR(ctxt.handle);
2805 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2807 ocfs2_commit_trans(osb, ctxt.handle);
2810 ocfs2_free_alloc_context(ctxt.data_ac);
2812 ocfs2_free_alloc_context(ctxt.meta_ac);
2813 if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
2814 ocfs2_schedule_truncate_log_flush(osb, 1);
2815 ocfs2_run_deallocs(osb, &ctxt.dealloc);
2817 up_write(&OCFS2_I(inode)->ip_xattr_sem);
2818 ocfs2_inode_unlock(inode, 1);
2821 brelse(xbs.xattr_bh);
2822 ocfs2_xattr_bucket_free(xbs.bucket);
2828 * Find the xattr extent rec which may contains name_hash.
2829 * e_cpos will be the first name hash of the xattr rec.
2830 * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2832 static int ocfs2_xattr_get_rec(struct inode *inode,
2837 struct ocfs2_extent_list *el)
2840 struct buffer_head *eb_bh = NULL;
2841 struct ocfs2_extent_block *eb;
2842 struct ocfs2_extent_rec *rec = NULL;
2845 if (el->l_tree_depth) {
2846 ret = ocfs2_find_leaf(inode, el, name_hash, &eb_bh);
2852 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2855 if (el->l_tree_depth) {
2856 ocfs2_error(inode->i_sb,
2857 "Inode %lu has non zero tree depth in "
2858 "xattr tree block %llu\n", inode->i_ino,
2859 (unsigned long long)eb_bh->b_blocknr);
2865 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2866 rec = &el->l_recs[i];
2868 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2869 e_blkno = le64_to_cpu(rec->e_blkno);
2875 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2876 "record (%u, %u, 0) in xattr", inode->i_ino,
2877 le32_to_cpu(rec->e_cpos),
2878 ocfs2_rec_clusters(el, rec));
2883 *p_blkno = le64_to_cpu(rec->e_blkno);
2884 *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2886 *e_cpos = le32_to_cpu(rec->e_cpos);
2892 typedef int (xattr_bucket_func)(struct inode *inode,
2893 struct ocfs2_xattr_bucket *bucket,
2896 static int ocfs2_find_xe_in_bucket(struct inode *inode,
2897 struct ocfs2_xattr_bucket *bucket,
2904 int i, ret = 0, cmp = 1, block_off, new_offset;
2905 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
2906 size_t name_len = strlen(name);
2907 struct ocfs2_xattr_entry *xe = NULL;
2911 * We don't use binary search in the bucket because there
2912 * may be multiple entries with the same name hash.
2914 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2915 xe = &xh->xh_entries[i];
2917 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2919 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2922 cmp = name_index - ocfs2_xattr_get_type(xe);
2924 cmp = name_len - xe->xe_name_len;
2928 ret = ocfs2_xattr_bucket_get_name_value(inode,
2939 xe_name = bucket_block(bucket, block_off) + new_offset;
2940 if (!memcmp(name, xe_name, name_len)) {
2952 * Find the specified xattr entry in a series of buckets.
2953 * This series start from p_blkno and last for num_clusters.
2954 * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2955 * the num of the valid buckets.
2957 * Return the buffer_head this xattr should reside in. And if the xattr's
2958 * hash is in the gap of 2 buckets, return the lower bucket.
2960 static int ocfs2_xattr_bucket_find(struct inode *inode,
2967 struct ocfs2_xattr_search *xs)
2970 struct ocfs2_xattr_header *xh = NULL;
2971 struct ocfs2_xattr_entry *xe = NULL;
2973 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2974 int low_bucket = 0, bucket, high_bucket;
2975 struct ocfs2_xattr_bucket *search;
2977 u64 blkno, lower_blkno = 0;
2979 search = ocfs2_xattr_bucket_new(inode);
2986 ret = ocfs2_read_xattr_bucket(search, p_blkno);
2992 xh = bucket_xh(search);
2993 high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
2994 while (low_bucket <= high_bucket) {
2995 ocfs2_xattr_bucket_relse(search);
2997 bucket = (low_bucket + high_bucket) / 2;
2998 blkno = p_blkno + bucket * blk_per_bucket;
2999 ret = ocfs2_read_xattr_bucket(search, blkno);
3005 xh = bucket_xh(search);
3006 xe = &xh->xh_entries[0];
3007 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3008 high_bucket = bucket - 1;
3013 * Check whether the hash of the last entry in our
3014 * bucket is larger than the search one. for an empty
3015 * bucket, the last one is also the first one.
3018 xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3020 last_hash = le32_to_cpu(xe->xe_name_hash);
3022 /* record lower_blkno which may be the insert place. */
3023 lower_blkno = blkno;
3025 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3026 low_bucket = bucket + 1;
3030 /* the searched xattr should reside in this bucket if exists. */
3031 ret = ocfs2_find_xe_in_bucket(inode, search,
3032 name_index, name, name_hash,
3042 * Record the bucket we have found.
3043 * When the xattr's hash value is in the gap of 2 buckets, we will
3044 * always set it to the previous bucket.
3047 lower_blkno = p_blkno;
3049 /* This should be in cache - we just read it during the search */
3050 ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3056 xs->header = bucket_xh(xs->bucket);
3057 xs->base = bucket_block(xs->bucket, 0);
3058 xs->end = xs->base + inode->i_sb->s_blocksize;
3061 xs->here = &xs->header->xh_entries[index];
3062 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
3063 (unsigned long long)bucket_blkno(xs->bucket), index);
3068 ocfs2_xattr_bucket_free(search);
3072 static int ocfs2_xattr_index_block_find(struct inode *inode,
3073 struct buffer_head *root_bh,
3076 struct ocfs2_xattr_search *xs)
3079 struct ocfs2_xattr_block *xb =
3080 (struct ocfs2_xattr_block *)root_bh->b_data;
3081 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3082 struct ocfs2_extent_list *el = &xb_root->xt_list;
3084 u32 first_hash, num_clusters = 0;
3085 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3087 if (le16_to_cpu(el->l_next_free_rec) == 0)
3090 mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
3091 name, name_hash, name_index);
3093 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3100 BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3102 mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
3103 "in the rec is %u\n", num_clusters, (unsigned long long)p_blkno,
3106 ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3107 p_blkno, first_hash, num_clusters, xs);
3113 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3116 xattr_bucket_func *func,
3120 u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3121 u32 num_buckets = clusters * bpc;
3122 struct ocfs2_xattr_bucket *bucket;
3124 bucket = ocfs2_xattr_bucket_new(inode);
3126 mlog_errno(-ENOMEM);
3130 mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
3131 clusters, (unsigned long long)blkno);
3133 for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3134 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3141 * The real bucket num in this series of blocks is stored
3142 * in the 1st bucket.
3145 num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3147 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
3148 (unsigned long long)blkno,
3149 le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3151 ret = func(inode, bucket, para);
3154 /* Fall through to bucket_relse() */
3157 ocfs2_xattr_bucket_relse(bucket);
3162 ocfs2_xattr_bucket_free(bucket);
3166 struct ocfs2_xattr_tree_list {
3172 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
3173 struct ocfs2_xattr_header *xh,
3180 if (index < 0 || index >= le16_to_cpu(xh->xh_count))
3183 name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
3185 *block_off = name_offset >> inode->i_sb->s_blocksize_bits;
3186 *new_offset = name_offset % inode->i_sb->s_blocksize;
3191 static int ocfs2_list_xattr_bucket(struct inode *inode,
3192 struct ocfs2_xattr_bucket *bucket,
3196 struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
3197 int i, block_off, new_offset;
3198 const char *prefix, *name;
3200 for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
3201 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
3202 type = ocfs2_xattr_get_type(entry);
3203 prefix = ocfs2_xattr_prefix(type);
3206 ret = ocfs2_xattr_bucket_get_name_value(inode,
3214 name = (const char *)bucket_block(bucket, block_off) +
3216 ret = ocfs2_xattr_list_entry(xl->buffer,
3220 entry->xe_name_len);
3229 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
3230 struct ocfs2_xattr_tree_root *xt,
3234 struct ocfs2_extent_list *el = &xt->xt_list;
3236 u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
3238 struct ocfs2_xattr_tree_list xl = {
3240 .buffer_size = buffer_size,
3244 if (le16_to_cpu(el->l_next_free_rec) == 0)
3247 while (name_hash > 0) {
3248 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
3249 &e_cpos, &num_clusters, el);
3255 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
3256 ocfs2_list_xattr_bucket,
3266 name_hash = e_cpos - 1;
3274 static int cmp_xe(const void *a, const void *b)
3276 const struct ocfs2_xattr_entry *l = a, *r = b;
3277 u32 l_hash = le32_to_cpu(l->xe_name_hash);
3278 u32 r_hash = le32_to_cpu(r->xe_name_hash);
3280 if (l_hash > r_hash)
3282 if (l_hash < r_hash)
3287 static void swap_xe(void *a, void *b, int size)
3289 struct ocfs2_xattr_entry *l = a, *r = b, tmp;
3292 memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
3293 memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
3297 * When the ocfs2_xattr_block is filled up, new bucket will be created
3298 * and all the xattr entries will be moved to the new bucket.
3299 * The header goes at the start of the bucket, and the names+values are
3300 * filled from the end. This is why *target starts as the last buffer.
3301 * Note: we need to sort the entries since they are not saved in order
3302 * in the ocfs2_xattr_block.
3304 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
3305 struct buffer_head *xb_bh,
3306 struct ocfs2_xattr_bucket *bucket)
3308 int i, blocksize = inode->i_sb->s_blocksize;
3309 int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3310 u16 offset, size, off_change;
3311 struct ocfs2_xattr_entry *xe;
3312 struct ocfs2_xattr_block *xb =
3313 (struct ocfs2_xattr_block *)xb_bh->b_data;
3314 struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
3315 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3316 u16 count = le16_to_cpu(xb_xh->xh_count);
3317 char *src = xb_bh->b_data;
3318 char *target = bucket_block(bucket, blks - 1);
3320 mlog(0, "cp xattr from block %llu to bucket %llu\n",
3321 (unsigned long long)xb_bh->b_blocknr,
3322 (unsigned long long)bucket_blkno(bucket));
3324 for (i = 0; i < blks; i++)
3325 memset(bucket_block(bucket, i), 0, blocksize);
3328 * Since the xe_name_offset is based on ocfs2_xattr_header,
3329 * there is a offset change corresponding to the change of
3330 * ocfs2_xattr_header's position.
3332 off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3333 xe = &xb_xh->xh_entries[count - 1];
3334 offset = le16_to_cpu(xe->xe_name_offset) + off_change;
3335 size = blocksize - offset;
3337 /* copy all the names and values. */
3338 memcpy(target + offset, src + offset, size);
3340 /* Init new header now. */
3341 xh->xh_count = xb_xh->xh_count;
3342 xh->xh_num_buckets = cpu_to_le16(1);
3343 xh->xh_name_value_len = cpu_to_le16(size);
3344 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
3346 /* copy all the entries. */
3347 target = bucket_block(bucket, 0);
3348 offset = offsetof(struct ocfs2_xattr_header, xh_entries);
3349 size = count * sizeof(struct ocfs2_xattr_entry);
3350 memcpy(target + offset, (char *)xb_xh + offset, size);
3352 /* Change the xe offset for all the xe because of the move. */
3353 off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
3354 offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3355 for (i = 0; i < count; i++)
3356 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
3358 mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
3359 offset, size, off_change);
3361 sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
3366 * After we move xattr from block to index btree, we have to
3367 * update ocfs2_xattr_search to the new xe and base.
3369 * When the entry is in xattr block, xattr_bh indicates the storage place.
3370 * While if the entry is in index b-tree, "bucket" indicates the
3371 * real place of the xattr.
3373 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
3374 struct ocfs2_xattr_search *xs,
3375 struct buffer_head *old_bh)
3377 char *buf = old_bh->b_data;
3378 struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
3379 struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
3382 xs->header = bucket_xh(xs->bucket);
3383 xs->base = bucket_block(xs->bucket, 0);
3384 xs->end = xs->base + inode->i_sb->s_blocksize;
3389 i = xs->here - old_xh->xh_entries;
3390 xs->here = &xs->header->xh_entries[i];
3393 static int ocfs2_xattr_create_index_block(struct inode *inode,
3394 struct ocfs2_xattr_search *xs,
3395 struct ocfs2_xattr_set_ctxt *ctxt)
3400 handle_t *handle = ctxt->handle;
3401 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3402 struct ocfs2_inode_info *oi = OCFS2_I(inode);
3403 struct buffer_head *xb_bh = xs->xattr_bh;
3404 struct ocfs2_xattr_block *xb =
3405 (struct ocfs2_xattr_block *)xb_bh->b_data;
3406 struct ocfs2_xattr_tree_root *xr;
3407 u16 xb_flags = le16_to_cpu(xb->xb_flags);
3409 mlog(0, "create xattr index block for %llu\n",
3410 (unsigned long long)xb_bh->b_blocknr);
3412 BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
3413 BUG_ON(!xs->bucket);
3417 * We can use this lock for now, and maybe move to a dedicated mutex
3418 * if performance becomes a problem later.
3420 down_write(&oi->ip_alloc_sem);
3422 ret = ocfs2_journal_access_xb(handle, inode, xb_bh,
3423 OCFS2_JOURNAL_ACCESS_WRITE);
3429 ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac,
3430 1, 1, &bit_off, &len);
3437 * The bucket may spread in many blocks, and
3438 * we will only touch the 1st block and the last block
3439 * in the whole bucket(one for entry and one for data).
3441 blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
3443 mlog(0, "allocate 1 cluster from %llu to xattr block\n",
3444 (unsigned long long)blkno);
3446 ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
3452 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
3453 OCFS2_JOURNAL_ACCESS_CREATE);
3459 ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
3460 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
3462 ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
3464 /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
3465 memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
3466 offsetof(struct ocfs2_xattr_block, xb_attrs));
3468 xr = &xb->xb_attrs.xb_root;
3469 xr->xt_clusters = cpu_to_le32(1);
3470 xr->xt_last_eb_blk = 0;
3471 xr->xt_list.l_tree_depth = 0;
3472 xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
3473 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
3475 xr->xt_list.l_recs[0].e_cpos = 0;
3476 xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
3477 xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
3479 xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
3481 ocfs2_journal_dirty(handle, xb_bh);
3484 up_write(&oi->ip_alloc_sem);
3489 static int cmp_xe_offset(const void *a, const void *b)
3491 const struct ocfs2_xattr_entry *l = a, *r = b;
3492 u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
3493 u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
3495 if (l_name_offset < r_name_offset)
3497 if (l_name_offset > r_name_offset)
3503 * defrag a xattr bucket if we find that the bucket has some
3504 * holes beteen name/value pairs.
3505 * We will move all the name/value pairs to the end of the bucket
3506 * so that we can spare some space for insertion.
3508 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
3510 struct ocfs2_xattr_bucket *bucket)
3513 size_t end, offset, len, value_len;
3514 struct ocfs2_xattr_header *xh;
3515 char *entries, *buf, *bucket_buf = NULL;
3516 u64 blkno = bucket_blkno(bucket);
3518 size_t blocksize = inode->i_sb->s_blocksize;
3519 struct ocfs2_xattr_entry *xe;
3522 * In order to make the operation more efficient and generic,
3523 * we copy all the blocks into a contiguous memory and do the
3524 * defragment there, so if anything is error, we will not touch
3527 bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
3534 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3535 memcpy(buf, bucket_block(bucket, i), blocksize);
3537 ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
3538 OCFS2_JOURNAL_ACCESS_WRITE);
3544 xh = (struct ocfs2_xattr_header *)bucket_buf;
3545 entries = (char *)xh->xh_entries;
3546 xh_free_start = le16_to_cpu(xh->xh_free_start);
3548 mlog(0, "adjust xattr bucket in %llu, count = %u, "
3549 "xh_free_start = %u, xh_name_value_len = %u.\n",
3550 (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
3551 xh_free_start, le16_to_cpu(xh->xh_name_value_len));
3554 * sort all the entries by their offset.
3555 * the largest will be the first, so that we can
3556 * move them to the end one by one.
3558 sort(entries, le16_to_cpu(xh->xh_count),
3559 sizeof(struct ocfs2_xattr_entry),
3560 cmp_xe_offset, swap_xe);
3562 /* Move all name/values to the end of the bucket. */
3563 xe = xh->xh_entries;
3564 end = OCFS2_XATTR_BUCKET_SIZE;
3565 for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
3566 offset = le16_to_cpu(xe->xe_name_offset);
3567 if (ocfs2_xattr_is_local(xe))
3568 value_len = OCFS2_XATTR_SIZE(
3569 le64_to_cpu(xe->xe_value_size));
3571 value_len = OCFS2_XATTR_ROOT_SIZE;
3572 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
3575 * We must make sure that the name/value pair
3576 * exist in the same block. So adjust end to
3577 * the previous block end if needed.
3579 if (((end - len) / blocksize !=
3580 (end - 1) / blocksize))
3581 end = end - end % blocksize;
3583 if (end > offset + len) {
3584 memmove(bucket_buf + end - len,
3585 bucket_buf + offset, len);
3586 xe->xe_name_offset = cpu_to_le16(end - len);
3589 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
3590 "bucket %llu\n", (unsigned long long)blkno);
3595 mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
3596 "bucket %llu\n", (unsigned long long)blkno);
3598 if (xh_free_start == end)
3601 memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
3602 xh->xh_free_start = cpu_to_le16(end);
3604 /* sort the entries by their name_hash. */
3605 sort(entries, le16_to_cpu(xh->xh_count),
3606 sizeof(struct ocfs2_xattr_entry),
3610 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3611 memcpy(bucket_block(bucket, i), buf, blocksize);
3612 ocfs2_xattr_bucket_journal_dirty(handle, bucket);
3620 * prev_blkno points to the start of an existing extent. new_blkno
3621 * points to a newly allocated extent. Because we know each of our
3622 * clusters contains more than bucket, we can easily split one cluster
3623 * at a bucket boundary. So we take the last cluster of the existing
3624 * extent and split it down the middle. We move the last half of the
3625 * buckets in the last cluster of the existing extent over to the new
3628 * first_bh is the buffer at prev_blkno so we can update the existing
3629 * extent's bucket count. header_bh is the bucket were we were hoping
3630 * to insert our xattr. If the bucket move places the target in the new
3631 * extent, we'll update first_bh and header_bh after modifying the old
3634 * first_hash will be set as the 1st xe's name_hash in the new extent.
3636 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3638 struct ocfs2_xattr_bucket *first,
3639 struct ocfs2_xattr_bucket *target,
3645 struct super_block *sb = inode->i_sb;
3646 int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
3647 int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
3648 int to_move = num_buckets / 2;
3650 u64 last_cluster_blkno = bucket_blkno(first) +
3651 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
3653 BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
3654 BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
3656 mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3657 (unsigned long long)last_cluster_blkno, (unsigned long long)new_blkno);
3659 ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
3660 last_cluster_blkno, new_blkno,
3661 to_move, first_hash);
3667 /* This is the first bucket that got moved */
3668 src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
3671 * If the target bucket was part of the moved buckets, we need to
3672 * update first and target.
3674 if (bucket_blkno(target) >= src_blkno) {
3675 /* Find the block for the new target bucket */
3676 src_blkno = new_blkno +
3677 (bucket_blkno(target) - src_blkno);
3679 ocfs2_xattr_bucket_relse(first);
3680 ocfs2_xattr_bucket_relse(target);
3683 * These shouldn't fail - the buffers are in the
3684 * journal from ocfs2_cp_xattr_bucket().
3686 ret = ocfs2_read_xattr_bucket(first, new_blkno);
3691 ret = ocfs2_read_xattr_bucket(target, src_blkno);
3702 * Find the suitable pos when we divide a bucket into 2.
3703 * We have to make sure the xattrs with the same hash value exist
3704 * in the same bucket.
3706 * If this ocfs2_xattr_header covers more than one hash value, find a
3707 * place where the hash value changes. Try to find the most even split.
3708 * The most common case is that all entries have different hash values,
3709 * and the first check we make will find a place to split.
3711 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
3713 struct ocfs2_xattr_entry *entries = xh->xh_entries;
3714 int count = le16_to_cpu(xh->xh_count);
3715 int delta, middle = count / 2;
3718 * We start at the middle. Each step gets farther away in both
3719 * directions. We therefore hit the change in hash value
3720 * nearest to the middle. Note that this loop does not execute for
3723 for (delta = 0; delta < middle; delta++) {
3724 /* Let's check delta earlier than middle */
3725 if (cmp_xe(&entries[middle - delta - 1],
3726 &entries[middle - delta]))
3727 return middle - delta;
3729 /* For even counts, don't walk off the end */
3730 if ((middle + delta + 1) == count)
3733 /* Now try delta past middle */
3734 if (cmp_xe(&entries[middle + delta],
3735 &entries[middle + delta + 1]))
3736 return middle + delta + 1;
3739 /* Every entry had the same hash */
3744 * Move some xattrs in old bucket(blk) to new bucket(new_blk).
3745 * first_hash will record the 1st hash of the new bucket.
3747 * Normally half of the xattrs will be moved. But we have to make
3748 * sure that the xattrs with the same hash value are stored in the
3749 * same bucket. If all the xattrs in this bucket have the same hash
3750 * value, the new bucket will be initialized as an empty one and the
3751 * first_hash will be initialized as (hash_value+1).
3753 static int ocfs2_divide_xattr_bucket(struct inode *inode,
3758 int new_bucket_head)
3761 int count, start, len, name_value_len = 0, xe_len, name_offset = 0;
3762 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3763 struct ocfs2_xattr_header *xh;
3764 struct ocfs2_xattr_entry *xe;
3765 int blocksize = inode->i_sb->s_blocksize;
3767 mlog(0, "move some of xattrs from bucket %llu to %llu\n",
3768 (unsigned long long)blk, (unsigned long long)new_blk);
3770 s_bucket = ocfs2_xattr_bucket_new(inode);
3771 t_bucket = ocfs2_xattr_bucket_new(inode);
3772 if (!s_bucket || !t_bucket) {
3778 ret = ocfs2_read_xattr_bucket(s_bucket, blk);
3784 ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
3785 OCFS2_JOURNAL_ACCESS_WRITE);
3792 * Even if !new_bucket_head, we're overwriting t_bucket. Thus,
3793 * there's no need to read it.
3795 ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
3802 * Hey, if we're overwriting t_bucket, what difference does
3803 * ACCESS_CREATE vs ACCESS_WRITE make? See the comment in the
3804 * same part of ocfs2_cp_xattr_bucket().
3806 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3808 OCFS2_JOURNAL_ACCESS_CREATE :
3809 OCFS2_JOURNAL_ACCESS_WRITE);
3815 xh = bucket_xh(s_bucket);
3816 count = le16_to_cpu(xh->xh_count);
3817 start = ocfs2_xattr_find_divide_pos(xh);
3819 if (start == count) {
3820 xe = &xh->xh_entries[start-1];
3823 * initialized a new empty bucket here.
3824 * The hash value is set as one larger than
3825 * that of the last entry in the previous bucket.
3827 for (i = 0; i < t_bucket->bu_blocks; i++)
3828 memset(bucket_block(t_bucket, i), 0, blocksize);
3830 xh = bucket_xh(t_bucket);
3831 xh->xh_free_start = cpu_to_le16(blocksize);
3832 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
3833 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
3835 goto set_num_buckets;
3838 /* copy the whole bucket to the new first. */
3839 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
3841 /* update the new bucket. */
3842 xh = bucket_xh(t_bucket);
3845 * Calculate the total name/value len and xh_free_start for
3846 * the old bucket first.
3848 name_offset = OCFS2_XATTR_BUCKET_SIZE;
3850 for (i = 0; i < start; i++) {
3851 xe = &xh->xh_entries[i];
3852 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3853 if (ocfs2_xattr_is_local(xe))
3855 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3857 xe_len += OCFS2_XATTR_ROOT_SIZE;
3858 name_value_len += xe_len;
3859 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3860 name_offset = le16_to_cpu(xe->xe_name_offset);
3864 * Now begin the modification to the new bucket.
3866 * In the new bucket, We just move the xattr entry to the beginning
3867 * and don't touch the name/value. So there will be some holes in the
3868 * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3871 xe = &xh->xh_entries[start];
3872 len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3873 mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3874 (int)((char *)xe - (char *)xh),
3875 (int)((char *)xh->xh_entries - (char *)xh));
3876 memmove((char *)xh->xh_entries, (char *)xe, len);
3877 xe = &xh->xh_entries[count - start];
3878 len = sizeof(struct ocfs2_xattr_entry) * start;
3879 memset((char *)xe, 0, len);
3881 le16_add_cpu(&xh->xh_count, -start);
3882 le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3884 /* Calculate xh_free_start for the new bucket. */
3885 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3886 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3887 xe = &xh->xh_entries[i];
3888 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3889 if (ocfs2_xattr_is_local(xe))
3891 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3893 xe_len += OCFS2_XATTR_ROOT_SIZE;
3894 if (le16_to_cpu(xe->xe_name_offset) <
3895 le16_to_cpu(xh->xh_free_start))
3896 xh->xh_free_start = xe->xe_name_offset;
3900 /* set xh->xh_num_buckets for the new xh. */
3901 if (new_bucket_head)
3902 xh->xh_num_buckets = cpu_to_le16(1);
3904 xh->xh_num_buckets = 0;
3906 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
3908 /* store the first_hash of the new bucket. */
3910 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3913 * Now only update the 1st block of the old bucket. If we
3914 * just added a new empty bucket, there is no need to modify
3920 xh = bucket_xh(s_bucket);
3921 memset(&xh->xh_entries[start], 0,
3922 sizeof(struct ocfs2_xattr_entry) * (count - start));
3923 xh->xh_count = cpu_to_le16(start);
3924 xh->xh_free_start = cpu_to_le16(name_offset);
3925 xh->xh_name_value_len = cpu_to_le16(name_value_len);
3927 ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
3930 ocfs2_xattr_bucket_free(s_bucket);
3931 ocfs2_xattr_bucket_free(t_bucket);
3937 * Copy xattr from one bucket to another bucket.
3939 * The caller must make sure that the journal transaction
3940 * has enough space for journaling.
3942 static int ocfs2_cp_xattr_bucket(struct inode *inode,
3949 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3951 BUG_ON(s_blkno == t_blkno);
3953 mlog(0, "cp bucket %llu to %llu, target is %d\n",
3954 (unsigned long long)s_blkno, (unsigned long long)t_blkno,
3957 s_bucket = ocfs2_xattr_bucket_new(inode);
3958 t_bucket = ocfs2_xattr_bucket_new(inode);
3959 if (!s_bucket || !t_bucket) {
3965 ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
3970 * Even if !t_is_new, we're overwriting t_bucket. Thus,
3971 * there's no need to read it.
3973 ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
3978 * Hey, if we're overwriting t_bucket, what difference does
3979 * ACCESS_CREATE vs ACCESS_WRITE make? Well, if we allocated a new
3980 * cluster to fill, we came here from
3981 * ocfs2_mv_xattr_buckets(), and it is really new -
3982 * ACCESS_CREATE is required. But we also might have moved data
3983 * out of t_bucket before extending back into it.
3984 * ocfs2_add_new_xattr_bucket() can do this - its call to
3985 * ocfs2_add_new_xattr_cluster() may have created a new extent
3986 * and copied out the end of the old extent. Then it re-extends
3987 * the old extent back to create space for new xattrs. That's
3988 * how we get here, and the bucket isn't really new.
3990 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3992 OCFS2_JOURNAL_ACCESS_CREATE :
3993 OCFS2_JOURNAL_ACCESS_WRITE);
3997 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
3998 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4001 ocfs2_xattr_bucket_free(t_bucket);
4002 ocfs2_xattr_bucket_free(s_bucket);
4008 * src_blk points to the start of an existing extent. last_blk points to
4009 * last cluster in that extent. to_blk points to a newly allocated
4010 * extent. We copy the buckets from the cluster at last_blk to the new
4011 * extent. If start_bucket is non-zero, we skip that many buckets before
4012 * we start copying. The new extent's xh_num_buckets gets set to the
4013 * number of buckets we copied. The old extent's xh_num_buckets shrinks
4014 * by the same amount.
4016 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4017 u64 src_blk, u64 last_blk, u64 to_blk,
4018 unsigned int start_bucket,
4021 int i, ret, credits;
4022 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4023 int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4024 int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4025 struct ocfs2_xattr_bucket *old_first, *new_first;
4027 mlog(0, "mv xattrs from cluster %llu to %llu\n",
4028 (unsigned long long)last_blk, (unsigned long long)to_blk);
4030 BUG_ON(start_bucket >= num_buckets);
4032 num_buckets -= start_bucket;
4033 last_blk += (start_bucket * blks_per_bucket);
4036 /* The first bucket of the original extent */
4037 old_first = ocfs2_xattr_bucket_new(inode);
4038 /* The first bucket of the new extent */
4039 new_first = ocfs2_xattr_bucket_new(inode);
4040 if (!old_first || !new_first) {
4046 ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4053 * We need to update the first bucket of the old extent and all
4054 * the buckets going to the new extent.
4056 credits = ((num_buckets + 1) * blks_per_bucket) +
4057 handle->h_buffer_credits;
4058 ret = ocfs2_extend_trans(handle, credits);
4064 ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4065 OCFS2_JOURNAL_ACCESS_WRITE);
4071 for (i = 0; i < num_buckets; i++) {
4072 ret = ocfs2_cp_xattr_bucket(inode, handle,
4073 last_blk + (i * blks_per_bucket),
4074 to_blk + (i * blks_per_bucket),
4083 * Get the new bucket ready before we dirty anything
4084 * (This actually shouldn't fail, because we already dirtied
4085 * it once in ocfs2_cp_xattr_bucket()).
4087 ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4092 ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4093 OCFS2_JOURNAL_ACCESS_WRITE);
4099 /* Now update the headers */
4100 le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4101 ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4103 bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4104 ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4107 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4110 ocfs2_xattr_bucket_free(new_first);
4111 ocfs2_xattr_bucket_free(old_first);
4116 * Move some xattrs in this cluster to the new cluster.
4117 * This function should only be called when bucket size == cluster size.
4118 * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4120 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4126 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4127 int ret, credits = 2 * blk_per_bucket + handle->h_buffer_credits;
4129 BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4131 ret = ocfs2_extend_trans(handle, credits);
4137 /* Move half of the xattr in start_blk to the next bucket. */
4138 return ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4139 new_blk, first_hash, 1);
4143 * Move some xattrs from the old cluster to the new one since they are not
4144 * contiguous in ocfs2 xattr tree.
4146 * new_blk starts a new separate cluster, and we will move some xattrs from
4147 * prev_blk to it. v_start will be set as the first name hash value in this
4148 * new cluster so that it can be used as e_cpos during tree insertion and
4149 * don't collide with our original b-tree operations. first_bh and header_bh
4150 * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4151 * to extend the insert bucket.
4153 * The problem is how much xattr should we move to the new one and when should
4154 * we update first_bh and header_bh?
4155 * 1. If cluster size > bucket size, that means the previous cluster has more
4156 * than 1 bucket, so just move half nums of bucket into the new cluster and
4157 * update the first_bh and header_bh if the insert bucket has been moved
4158 * to the new cluster.
4159 * 2. If cluster_size == bucket_size:
4160 * a) If the previous extent rec has more than one cluster and the insert
4161 * place isn't in the last cluster, copy the entire last cluster to the
4162 * new one. This time, we don't need to upate the first_bh and header_bh
4163 * since they will not be moved into the new cluster.
4164 * b) Otherwise, move the bottom half of the xattrs in the last cluster into
4165 * the new one. And we set the extend flag to zero if the insert place is
4166 * moved into the new allocated cluster since no extend is needed.
4168 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
4170 struct ocfs2_xattr_bucket *first,
4171 struct ocfs2_xattr_bucket *target,
4179 mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
4180 (unsigned long long)bucket_blkno(first), prev_clusters,
4181 (unsigned long long)new_blk);
4183 if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
4184 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
4193 /* The start of the last cluster in the first extent */
4194 u64 last_blk = bucket_blkno(first) +
4195 ((prev_clusters - 1) *
4196 ocfs2_clusters_to_blocks(inode->i_sb, 1));
4198 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
4199 ret = ocfs2_mv_xattr_buckets(inode, handle,
4200 bucket_blkno(first),
4201 last_blk, new_blk, 0,
4206 ret = ocfs2_divide_xattr_cluster(inode, handle,
4212 if ((bucket_blkno(target) == last_blk) && extend)
4221 * Add a new cluster for xattr storage.
4223 * If the new cluster is contiguous with the previous one, it will be
4224 * appended to the same extent record, and num_clusters will be updated.
4225 * If not, we will insert a new extent for it and move some xattrs in
4226 * the last cluster into the new allocated one.
4227 * We also need to limit the maximum size of a btree leaf, otherwise we'll
4228 * lose the benefits of hashing because we'll have to search large leaves.
4229 * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
4232 * first_bh is the first block of the previous extent rec and header_bh
4233 * indicates the bucket we will insert the new xattrs. They will be updated
4234 * when the header_bh is moved into the new cluster.
4236 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
4237 struct buffer_head *root_bh,
4238 struct ocfs2_xattr_bucket *first,
4239 struct ocfs2_xattr_bucket *target,
4243 struct ocfs2_xattr_set_ctxt *ctxt)
4246 u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
4247 u32 prev_clusters = *num_clusters;
4248 u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
4250 handle_t *handle = ctxt->handle;
4251 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4252 struct ocfs2_extent_tree et;
4254 mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
4255 "previous xattr blkno = %llu\n",
4256 (unsigned long long)OCFS2_I(inode)->ip_blkno,
4257 prev_cpos, (unsigned long long)bucket_blkno(first));
4259 ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4261 ret = ocfs2_journal_access_xb(handle, inode, root_bh,
4262 OCFS2_JOURNAL_ACCESS_WRITE);
4268 ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac, 1,
4269 clusters_to_add, &bit_off, &num_bits);
4276 BUG_ON(num_bits > clusters_to_add);
4278 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
4279 mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
4280 num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
4282 if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
4283 (prev_clusters + num_bits) << osb->s_clustersize_bits <=
4284 OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
4286 * If this cluster is contiguous with the old one and
4287 * adding this new cluster, we don't surpass the limit of
4288 * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
4289 * initialized and used like other buckets in the previous
4291 * So add it as a contiguous one. The caller will handle
4294 v_start = prev_cpos + prev_clusters;
4295 *num_clusters = prev_clusters + num_bits;
4296 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
4299 ret = ocfs2_adjust_xattr_cross_cluster(inode,
4313 mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
4314 num_bits, (unsigned long long)block, v_start);
4315 ret = ocfs2_insert_extent(osb, handle, inode, &et, v_start, block,
4316 num_bits, 0, ctxt->meta_ac);
4322 ret = ocfs2_journal_dirty(handle, root_bh);
4331 * We are given an extent. 'first' is the bucket at the very front of
4332 * the extent. The extent has space for an additional bucket past
4333 * bucket_xh(first)->xh_num_buckets. 'target_blkno' is the block number
4334 * of the target bucket. We wish to shift every bucket past the target
4335 * down one, filling in that additional space. When we get back to the
4336 * target, we split the target between itself and the now-empty bucket
4337 * at target+1 (aka, target_blkno + blks_per_bucket).
4339 static int ocfs2_extend_xattr_bucket(struct inode *inode,
4341 struct ocfs2_xattr_bucket *first,
4346 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4347 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4349 u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
4351 mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
4352 "from %llu, len = %u\n", (unsigned long long)target_blk,
4353 (unsigned long long)bucket_blkno(first), num_clusters);
4355 /* The extent must have room for an additional bucket */
4356 BUG_ON(new_bucket >=
4357 (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
4359 /* end_blk points to the last existing bucket */
4360 end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
4363 * end_blk is the start of the last existing bucket.
4364 * Thus, (end_blk - target_blk) covers the target bucket and
4365 * every bucket after it up to, but not including, the last
4366 * existing bucket. Then we add the last existing bucket, the
4367 * new bucket, and the first bucket (3 * blk_per_bucket).
4369 credits = (end_blk - target_blk) + (3 * blk_per_bucket) +
4370 handle->h_buffer_credits;
4371 ret = ocfs2_extend_trans(handle, credits);
4377 ret = ocfs2_xattr_bucket_journal_access(handle, first,
4378 OCFS2_JOURNAL_ACCESS_WRITE);
4384 while (end_blk != target_blk) {
4385 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
4386 end_blk + blk_per_bucket, 0);
4389 end_blk -= blk_per_bucket;
4392 /* Move half of the xattr in target_blkno to the next bucket. */
4393 ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
4394 target_blk + blk_per_bucket, NULL, 0);
4396 le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
4397 ocfs2_xattr_bucket_journal_dirty(handle, first);
4404 * Add new xattr bucket in an extent record and adjust the buckets
4405 * accordingly. xb_bh is the ocfs2_xattr_block, and target is the
4406 * bucket we want to insert into.
4408 * In the easy case, we will move all the buckets after target down by
4409 * one. Half of target's xattrs will be moved to the next bucket.
4411 * If current cluster is full, we'll allocate a new one. This may not
4412 * be contiguous. The underlying calls will make sure that there is
4413 * space for the insert, shifting buckets around if necessary.
4414 * 'target' may be moved by those calls.
4416 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
4417 struct buffer_head *xb_bh,
4418 struct ocfs2_xattr_bucket *target,
4419 struct ocfs2_xattr_set_ctxt *ctxt)
4421 struct ocfs2_xattr_block *xb =
4422 (struct ocfs2_xattr_block *)xb_bh->b_data;
4423 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
4424 struct ocfs2_extent_list *el = &xb_root->xt_list;
4426 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
4427 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4428 int ret, num_buckets, extend = 1;
4430 u32 e_cpos, num_clusters;
4431 /* The bucket at the front of the extent */
4432 struct ocfs2_xattr_bucket *first;
4434 mlog(0, "Add new xattr bucket starting from %llu\n",
4435 (unsigned long long)bucket_blkno(target));
4437 /* The first bucket of the original extent */
4438 first = ocfs2_xattr_bucket_new(inode);
4445 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
4452 ret = ocfs2_read_xattr_bucket(first, p_blkno);
4458 num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
4459 if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
4461 * This can move first+target if the target bucket moves
4462 * to the new extent.
4464 ret = ocfs2_add_new_xattr_cluster(inode,
4479 ret = ocfs2_extend_xattr_bucket(inode,
4482 bucket_blkno(target),
4489 ocfs2_xattr_bucket_free(first);
4494 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
4495 struct ocfs2_xattr_bucket *bucket,
4498 int block_off = offs >> inode->i_sb->s_blocksize_bits;
4500 offs = offs % inode->i_sb->s_blocksize;
4501 return bucket_block(bucket, block_off) + offs;
4505 * Handle the normal xattr set, including replace, delete and new.
4507 * Note: "local" indicates the real data's locality. So we can't
4508 * just its bucket locality by its length.
4510 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
4511 struct ocfs2_xattr_info *xi,
4512 struct ocfs2_xattr_search *xs,
4516 struct ocfs2_xattr_entry *last, *xe;
4517 int name_len = strlen(xi->name);
4518 struct ocfs2_xattr_header *xh = xs->header;
4519 u16 count = le16_to_cpu(xh->xh_count), start;
4520 size_t blocksize = inode->i_sb->s_blocksize;
4522 size_t offs, size, new_size;
4524 last = &xh->xh_entries[count];
4525 if (!xs->not_found) {
4527 offs = le16_to_cpu(xe->xe_name_offset);
4528 if (ocfs2_xattr_is_local(xe))
4529 size = OCFS2_XATTR_SIZE(name_len) +
4530 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4532 size = OCFS2_XATTR_SIZE(name_len) +
4533 OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4536 * If the new value will be stored outside, xi->value has been
4537 * initalized as an empty ocfs2_xattr_value_root, and the same
4538 * goes with xi->value_len, so we can set new_size safely here.
4539 * See ocfs2_xattr_set_in_bucket.
4541 new_size = OCFS2_XATTR_SIZE(name_len) +
4542 OCFS2_XATTR_SIZE(xi->value_len);
4544 le16_add_cpu(&xh->xh_name_value_len, -size);
4546 if (new_size > size)
4547 goto set_new_name_value;
4549 /* Now replace the old value with new one. */
4551 xe->xe_value_size = cpu_to_le64(xi->value_len);
4553 xe->xe_value_size = 0;
4555 val = ocfs2_xattr_bucket_get_val(inode,
4557 memset(val + OCFS2_XATTR_SIZE(name_len), 0,
4558 size - OCFS2_XATTR_SIZE(name_len));
4559 if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
4560 memcpy(val + OCFS2_XATTR_SIZE(name_len),
4561 xi->value, xi->value_len);
4563 le16_add_cpu(&xh->xh_name_value_len, new_size);
4564 ocfs2_xattr_set_local(xe, local);
4568 * Remove the old entry if there is more than one.
4569 * We don't remove the last entry so that we can
4570 * use it to indicate the hash value of the empty
4574 le16_add_cpu(&xh->xh_count, -1);
4577 (void *)last - (void *)xe);
4579 sizeof(struct ocfs2_xattr_entry));
4582 cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4587 /* find a new entry for insert. */
4588 int low = 0, high = count - 1, tmp;
4589 struct ocfs2_xattr_entry *tmp_xe;
4591 while (low <= high && count) {
4592 tmp = (low + high) / 2;
4593 tmp_xe = &xh->xh_entries[tmp];
4595 if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
4597 else if (name_hash <
4598 le32_to_cpu(tmp_xe->xe_name_hash))
4606 xe = &xh->xh_entries[low];
4608 memmove(xe + 1, xe, (void *)last - (void *)xe);
4610 le16_add_cpu(&xh->xh_count, 1);
4611 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
4612 xe->xe_name_hash = cpu_to_le32(name_hash);
4613 xe->xe_name_len = name_len;
4614 ocfs2_xattr_set_type(xe, xi->name_index);
4618 /* Insert the new name+value. */
4619 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4622 * We must make sure that the name/value pair
4623 * exists in the same block.
4625 offs = le16_to_cpu(xh->xh_free_start);
4626 start = offs - size;
4628 if (start >> inode->i_sb->s_blocksize_bits !=
4629 (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4630 offs = offs - offs % blocksize;
4631 xh->xh_free_start = cpu_to_le16(offs);
4634 val = ocfs2_xattr_bucket_get_val(inode, xs->bucket, offs - size);
4635 xe->xe_name_offset = cpu_to_le16(offs - size);
4637 memset(val, 0, size);
4638 memcpy(val, xi->name, name_len);
4639 memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4641 xe->xe_value_size = cpu_to_le64(xi->value_len);
4642 ocfs2_xattr_set_local(xe, local);
4644 le16_add_cpu(&xh->xh_free_start, -size);
4645 le16_add_cpu(&xh->xh_name_value_len, size);
4651 * Set the xattr entry in the specified bucket.
4652 * The bucket is indicated by xs->bucket and it should have the enough
4653 * space for the xattr insertion.
4655 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4657 struct ocfs2_xattr_info *xi,
4658 struct ocfs2_xattr_search *xs,
4665 mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4666 (unsigned long)xi->value_len, xi->name_index,
4667 (unsigned long long)bucket_blkno(xs->bucket));
4669 if (!xs->bucket->bu_bhs[1]) {
4670 blkno = bucket_blkno(xs->bucket);
4671 ocfs2_xattr_bucket_relse(xs->bucket);
4672 ret = ocfs2_read_xattr_bucket(xs->bucket, blkno);
4679 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4680 OCFS2_JOURNAL_ACCESS_WRITE);
4686 ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local);
4687 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4694 * Truncate the specified xe_off entry in xattr bucket.
4695 * bucket is indicated by header_bh and len is the new length.
4696 * Both the ocfs2_xattr_value_root and the entry will be updated here.
4698 * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4700 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4701 struct ocfs2_xattr_bucket *bucket,
4704 struct ocfs2_xattr_set_ctxt *ctxt)
4708 struct ocfs2_xattr_entry *xe;
4709 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4710 size_t blocksize = inode->i_sb->s_blocksize;
4711 struct ocfs2_xattr_value_buf vb = {
4712 .vb_access = ocfs2_journal_access,
4715 xe = &xh->xh_entries[xe_off];
4717 BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4719 offset = le16_to_cpu(xe->xe_name_offset) +
4720 OCFS2_XATTR_SIZE(xe->xe_name_len);
4722 value_blk = offset / blocksize;
4724 /* We don't allow ocfs2_xattr_value to be stored in different block. */
4725 BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4727 vb.vb_bh = bucket->bu_bhs[value_blk];
4730 vb.vb_xv = (struct ocfs2_xattr_value_root *)
4731 (vb.vb_bh->b_data + offset % blocksize);
4734 * From here on out we have to dirty the bucket. The generic
4735 * value calls only modify one of the bucket's bhs, but we need
4736 * to send the bucket at once. So if they error, they *could* have
4737 * modified something. We have to assume they did, and dirty
4738 * the whole bucket. This leaves us in a consistent state.
4740 mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4741 xe_off, (unsigned long long)bucket_blkno(bucket), len);
4742 ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
4748 ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
4749 OCFS2_JOURNAL_ACCESS_WRITE);
4755 xe->xe_value_size = cpu_to_le64(len);
4757 ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
4763 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4764 struct ocfs2_xattr_search *xs,
4766 struct ocfs2_xattr_set_ctxt *ctxt)
4769 struct ocfs2_xattr_entry *xe = xs->here;
4770 struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4772 BUG_ON(!xs->bucket->bu_bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4774 offset = xe - xh->xh_entries;
4775 ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket,
4783 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4785 struct ocfs2_xattr_search *xs,
4790 struct ocfs2_xattr_value_root *xv;
4791 struct ocfs2_xattr_entry *xe = xs->here;
4793 BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4795 offset = le16_to_cpu(xe->xe_name_offset) +
4796 OCFS2_XATTR_SIZE(xe->xe_name_len);
4798 xv = (struct ocfs2_xattr_value_root *)(xs->base + offset);
4800 return __ocfs2_xattr_set_value_outside(inode, handle,
4801 xv, val, value_len);
4804 static int ocfs2_rm_xattr_cluster(struct inode *inode,
4805 struct buffer_head *root_bh,
4811 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4812 struct inode *tl_inode = osb->osb_tl_inode;
4814 struct ocfs2_xattr_block *xb =
4815 (struct ocfs2_xattr_block *)root_bh->b_data;
4816 struct ocfs2_alloc_context *meta_ac = NULL;
4817 struct ocfs2_cached_dealloc_ctxt dealloc;
4818 struct ocfs2_extent_tree et;
4820 ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4822 ocfs2_init_dealloc_ctxt(&dealloc);
4824 mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4825 cpos, len, (unsigned long long)blkno);
4827 ocfs2_remove_xattr_clusters_from_cache(inode, blkno, len);
4829 ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
4835 mutex_lock(&tl_inode->i_mutex);
4837 if (ocfs2_truncate_log_needs_flush(osb)) {
4838 ret = __ocfs2_flush_truncate_log(osb);
4845 handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
4846 if (IS_ERR(handle)) {
4852 ret = ocfs2_journal_access_xb(handle, inode, root_bh,
4853 OCFS2_JOURNAL_ACCESS_WRITE);
4859 ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
4866 le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4868 ret = ocfs2_journal_dirty(handle, root_bh);
4874 ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4879 ocfs2_commit_trans(osb, handle);
4881 ocfs2_schedule_truncate_log_flush(osb, 1);
4883 mutex_unlock(&tl_inode->i_mutex);
4886 ocfs2_free_alloc_context(meta_ac);
4888 ocfs2_run_deallocs(osb, &dealloc);
4893 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4895 struct ocfs2_xattr_search *xs)
4897 struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
4898 struct ocfs2_xattr_entry *last = &xh->xh_entries[
4899 le16_to_cpu(xh->xh_count) - 1];
4902 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4903 OCFS2_JOURNAL_ACCESS_WRITE);
4909 /* Remove the old entry. */
4910 memmove(xs->here, xs->here + 1,
4911 (void *)last - (void *)xs->here);
4912 memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4913 le16_add_cpu(&xh->xh_count, -1);
4915 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4919 * Set the xattr name/value in the bucket specified in xs.
4921 * As the new value in xi may be stored in the bucket or in an outside cluster,
4922 * we divide the whole process into 3 steps:
4923 * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4924 * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4925 * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4926 * 4. If the clusters for the new outside value can't be allocated, we need
4927 * to free the xattr we allocated in set.
4929 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4930 struct ocfs2_xattr_info *xi,
4931 struct ocfs2_xattr_search *xs,
4932 struct ocfs2_xattr_set_ctxt *ctxt)
4936 char *val = (char *)xi->value;
4937 struct ocfs2_xattr_entry *xe = xs->here;
4938 u32 name_hash = ocfs2_xattr_name_hash(inode, xi->name,
4941 if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
4943 * We need to truncate the xattr storage first.
4945 * If both the old and new value are stored to
4946 * outside block, we only need to truncate
4947 * the storage and then set the value outside.
4949 * If the new value should be stored within block,
4950 * we should free all the outside block first and
4951 * the modification to the xattr block will be done
4952 * by following steps.
4954 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4955 value_len = xi->value_len;
4959 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4966 goto set_value_outside;
4969 value_len = xi->value_len;
4970 /* So we have to handle the inside block change now. */
4971 if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4973 * If the new value will be stored outside of block,
4974 * initalize a new empty value root and insert it first.
4977 xi->value = &def_xv;
4978 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
4981 ret = ocfs2_xattr_set_entry_in_bucket(inode, ctxt->handle, xi, xs,
4988 if (value_len <= OCFS2_XATTR_INLINE_SIZE)
4991 /* allocate the space now for the outside block storage. */
4992 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4997 if (xs->not_found) {
4999 * We can't allocate enough clusters for outside
5000 * storage and we have allocated xattr already,
5001 * so need to remove it.
5003 ocfs2_xattr_bucket_remove_xs(inode, ctxt->handle, xs);
5009 ret = ocfs2_xattr_bucket_set_value_outside(inode, ctxt->handle,
5010 xs, val, value_len);
5016 * check whether the xattr bucket is filled up with the same hash value.
5017 * If we want to insert the xattr with the same hash, return -ENOSPC.
5018 * If we want to insert a xattr with different hash value, go ahead
5019 * and ocfs2_divide_xattr_bucket will handle this.
5021 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
5022 struct ocfs2_xattr_bucket *bucket,
5025 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5026 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5028 if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5031 if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5032 xh->xh_entries[0].xe_name_hash) {
5033 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5035 (unsigned long long)bucket_blkno(bucket),
5036 le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5043 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5044 struct ocfs2_xattr_info *xi,
5045 struct ocfs2_xattr_search *xs,
5046 struct ocfs2_xattr_set_ctxt *ctxt)
5048 struct ocfs2_xattr_header *xh;
5049 struct ocfs2_xattr_entry *xe;
5050 u16 count, header_size, xh_free_start;
5051 int free, max_free, need, old;
5052 size_t value_size = 0, name_len = strlen(xi->name);
5053 size_t blocksize = inode->i_sb->s_blocksize;
5054 int ret, allocation = 0;
5056 mlog_entry("Set xattr %s in xattr index block\n", xi->name);
5060 count = le16_to_cpu(xh->xh_count);
5061 xh_free_start = le16_to_cpu(xh->xh_free_start);
5062 header_size = sizeof(struct ocfs2_xattr_header) +
5063 count * sizeof(struct ocfs2_xattr_entry);
5064 max_free = OCFS2_XATTR_BUCKET_SIZE -
5065 le16_to_cpu(xh->xh_name_value_len) - header_size;
5067 mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
5068 "of %u which exceed block size\n",
5069 (unsigned long long)bucket_blkno(xs->bucket),
5072 if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
5073 value_size = OCFS2_XATTR_ROOT_SIZE;
5075 value_size = OCFS2_XATTR_SIZE(xi->value_len);
5078 need = sizeof(struct ocfs2_xattr_entry) +
5079 OCFS2_XATTR_SIZE(name_len) + value_size;
5081 need = value_size + OCFS2_XATTR_SIZE(name_len);
5084 * We only replace the old value if the new length is smaller
5085 * than the old one. Otherwise we will allocate new space in the
5086 * bucket to store it.
5089 if (ocfs2_xattr_is_local(xe))
5090 old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
5092 old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
5094 if (old >= value_size)
5098 free = xh_free_start - header_size;
5100 * We need to make sure the new name/value pair
5101 * can exist in the same block.
5103 if (xh_free_start % blocksize < need)
5104 free -= xh_free_start % blocksize;
5106 mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
5107 "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
5108 " %u\n", xs->not_found,
5109 (unsigned long long)bucket_blkno(xs->bucket),
5110 free, need, max_free, le16_to_cpu(xh->xh_free_start),
5111 le16_to_cpu(xh->xh_name_value_len));
5115 count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb))) {
5116 if (need <= max_free &&
5117 count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
5119 * We can create the space by defragment. Since only the
5120 * name/value will be moved, the xe shouldn't be changed
5123 ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5130 xh_free_start = le16_to_cpu(xh->xh_free_start);
5131 free = xh_free_start - header_size;
5132 if (xh_free_start % blocksize < need)
5133 free -= xh_free_start % blocksize;
5138 mlog(0, "Can't get enough space for xattr insert by "
5139 "defragment. Need %u bytes, but we have %d, so "
5140 "allocate new bucket for it.\n", need, free);
5144 * We have to add new buckets or clusters and one
5145 * allocation should leave us enough space for insert.
5150 * We do not allow for overlapping ranges between buckets. And
5151 * the maximum number of collisions we will allow for then is
5152 * one bucket's worth, so check it here whether we need to
5153 * add a new bucket for the insert.
5155 ret = ocfs2_check_xattr_bucket_collision(inode,
5163 ret = ocfs2_add_new_xattr_bucket(inode,
5173 * ocfs2_add_new_xattr_bucket() will have updated
5174 * xs->bucket if it moved, but it will not have updated
5175 * any of the other search fields. Thus, we drop it and
5176 * re-search. Everything should be cached, so it'll be
5179 ocfs2_xattr_bucket_relse(xs->bucket);
5180 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5183 if (ret && ret != -ENODATA)
5185 xs->not_found = ret;
5191 ret = ocfs2_xattr_set_in_bucket(inode, xi, xs, ctxt);
5197 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5198 struct ocfs2_xattr_bucket *bucket,
5202 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5204 struct ocfs2_xattr_entry *xe;
5205 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5206 struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5207 int credits = ocfs2_remove_extent_credits(osb->sb) +
5208 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5211 ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5213 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5214 xe = &xh->xh_entries[i];
5215 if (ocfs2_xattr_is_local(xe))
5218 ctxt.handle = ocfs2_start_trans(osb, credits);
5219 if (IS_ERR(ctxt.handle)) {
5220 ret = PTR_ERR(ctxt.handle);
5225 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5228 ocfs2_commit_trans(osb, ctxt.handle);
5235 ocfs2_schedule_truncate_log_flush(osb, 1);
5236 ocfs2_run_deallocs(osb, &ctxt.dealloc);
5240 static int ocfs2_delete_xattr_index_block(struct inode *inode,
5241 struct buffer_head *xb_bh)
5243 struct ocfs2_xattr_block *xb =
5244 (struct ocfs2_xattr_block *)xb_bh->b_data;
5245 struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
5247 u32 name_hash = UINT_MAX, e_cpos, num_clusters;
5250 if (le16_to_cpu(el->l_next_free_rec) == 0)
5253 while (name_hash > 0) {
5254 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
5255 &e_cpos, &num_clusters, el);
5261 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
5262 ocfs2_delete_xattr_in_bucket,
5269 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
5270 p_blkno, e_cpos, num_clusters);
5279 name_hash = e_cpos - 1;
5287 * 'security' attributes support
5289 static size_t ocfs2_xattr_security_list(struct inode *inode, char *list,
5290 size_t list_size, const char *name,
5293 const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
5294 const size_t total_len = prefix_len + name_len + 1;
5296 if (list && total_len <= list_size) {
5297 memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
5298 memcpy(list + prefix_len, name, name_len);
5299 list[prefix_len + name_len] = '\0';
5304 static int ocfs2_xattr_security_get(struct inode *inode, const char *name,
5305 void *buffer, size_t size)
5307 if (strcmp(name, "") == 0)
5309 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_SECURITY, name,
5313 static int ocfs2_xattr_security_set(struct inode *inode, const char *name,
5314 const void *value, size_t size, int flags)
5316 if (strcmp(name, "") == 0)
5319 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY, name, value,
5323 int ocfs2_init_security_get(struct inode *inode,
5325 struct ocfs2_security_xattr_info *si)
5327 /* check whether ocfs2 support feature xattr */
5328 if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
5330 return security_inode_init_security(inode, dir, &si->name, &si->value,
5334 int ocfs2_init_security_set(handle_t *handle,
5335 struct inode *inode,
5336 struct buffer_head *di_bh,
5337 struct ocfs2_security_xattr_info *si,
5338 struct ocfs2_alloc_context *xattr_ac,
5339 struct ocfs2_alloc_context *data_ac)
5341 return ocfs2_xattr_set_handle(handle, inode, di_bh,
5342 OCFS2_XATTR_INDEX_SECURITY,
5343 si->name, si->value, si->value_len, 0,
5347 struct xattr_handler ocfs2_xattr_security_handler = {
5348 .prefix = XATTR_SECURITY_PREFIX,
5349 .list = ocfs2_xattr_security_list,
5350 .get = ocfs2_xattr_security_get,
5351 .set = ocfs2_xattr_security_set,
5355 * 'trusted' attributes support
5357 static size_t ocfs2_xattr_trusted_list(struct inode *inode, char *list,
5358 size_t list_size, const char *name,
5361 const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
5362 const size_t total_len = prefix_len + name_len + 1;
5364 if (list && total_len <= list_size) {
5365 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
5366 memcpy(list + prefix_len, name, name_len);
5367 list[prefix_len + name_len] = '\0';
5372 static int ocfs2_xattr_trusted_get(struct inode *inode, const char *name,
5373 void *buffer, size_t size)
5375 if (strcmp(name, "") == 0)
5377 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED, name,
5381 static int ocfs2_xattr_trusted_set(struct inode *inode, const char *name,
5382 const void *value, size_t size, int flags)
5384 if (strcmp(name, "") == 0)
5387 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED, name, value,
5391 struct xattr_handler ocfs2_xattr_trusted_handler = {
5392 .prefix = XATTR_TRUSTED_PREFIX,
5393 .list = ocfs2_xattr_trusted_list,
5394 .get = ocfs2_xattr_trusted_get,
5395 .set = ocfs2_xattr_trusted_set,
5399 * 'user' attributes support
5401 static size_t ocfs2_xattr_user_list(struct inode *inode, char *list,
5402 size_t list_size, const char *name,
5405 const size_t prefix_len = XATTR_USER_PREFIX_LEN;
5406 const size_t total_len = prefix_len + name_len + 1;
5407 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5409 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5412 if (list && total_len <= list_size) {
5413 memcpy(list, XATTR_USER_PREFIX, prefix_len);
5414 memcpy(list + prefix_len, name, name_len);
5415 list[prefix_len + name_len] = '\0';
5420 static int ocfs2_xattr_user_get(struct inode *inode, const char *name,
5421 void *buffer, size_t size)
5423 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5425 if (strcmp(name, "") == 0)
5427 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5429 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
5433 static int ocfs2_xattr_user_set(struct inode *inode, const char *name,
5434 const void *value, size_t size, int flags)
5436 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5438 if (strcmp(name, "") == 0)
5440 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5443 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER, name, value,
5447 struct xattr_handler ocfs2_xattr_user_handler = {
5448 .prefix = XATTR_USER_PREFIX,
5449 .list = ocfs2_xattr_user_list,
5450 .get = ocfs2_xattr_user_get,
5451 .set = ocfs2_xattr_user_set,