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"
58 #include "refcounttree.h"
60 struct ocfs2_xattr_def_value_root {
61 struct ocfs2_xattr_value_root xv;
62 struct ocfs2_extent_rec er;
65 struct ocfs2_xattr_bucket {
66 /* The inode these xattrs are associated with */
67 struct inode *bu_inode;
69 /* The actual buffers that make up the bucket */
70 struct buffer_head *bu_bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
72 /* How many blocks make up one bucket for this filesystem */
76 struct ocfs2_xattr_set_ctxt {
78 struct ocfs2_alloc_context *meta_ac;
79 struct ocfs2_alloc_context *data_ac;
80 struct ocfs2_cached_dealloc_ctxt dealloc;
83 #define OCFS2_XATTR_ROOT_SIZE (sizeof(struct ocfs2_xattr_def_value_root))
84 #define OCFS2_XATTR_INLINE_SIZE 80
85 #define OCFS2_XATTR_HEADER_GAP 4
86 #define OCFS2_XATTR_FREE_IN_IBODY (OCFS2_MIN_XATTR_INLINE_SIZE \
87 - sizeof(struct ocfs2_xattr_header) \
88 - OCFS2_XATTR_HEADER_GAP)
89 #define OCFS2_XATTR_FREE_IN_BLOCK(ptr) ((ptr)->i_sb->s_blocksize \
90 - sizeof(struct ocfs2_xattr_block) \
91 - sizeof(struct ocfs2_xattr_header) \
92 - OCFS2_XATTR_HEADER_GAP)
94 static struct ocfs2_xattr_def_value_root def_xv = {
95 .xv.xr_list.l_count = cpu_to_le16(1),
98 struct xattr_handler *ocfs2_xattr_handlers[] = {
99 &ocfs2_xattr_user_handler,
100 #ifdef CONFIG_OCFS2_FS_POSIX_ACL
101 &ocfs2_xattr_acl_access_handler,
102 &ocfs2_xattr_acl_default_handler,
104 &ocfs2_xattr_trusted_handler,
105 &ocfs2_xattr_security_handler,
109 static struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
110 [OCFS2_XATTR_INDEX_USER] = &ocfs2_xattr_user_handler,
111 #ifdef CONFIG_OCFS2_FS_POSIX_ACL
112 [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS]
113 = &ocfs2_xattr_acl_access_handler,
114 [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT]
115 = &ocfs2_xattr_acl_default_handler,
117 [OCFS2_XATTR_INDEX_TRUSTED] = &ocfs2_xattr_trusted_handler,
118 [OCFS2_XATTR_INDEX_SECURITY] = &ocfs2_xattr_security_handler,
121 struct ocfs2_xattr_info {
128 struct ocfs2_xattr_search {
129 struct buffer_head *inode_bh;
131 * xattr_bh point to the block buffer head which has extended attribute
132 * when extended attribute in inode, xattr_bh is equal to inode_bh.
134 struct buffer_head *xattr_bh;
135 struct ocfs2_xattr_header *header;
136 struct ocfs2_xattr_bucket *bucket;
139 struct ocfs2_xattr_entry *here;
143 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
144 struct ocfs2_xattr_header *xh,
149 static int ocfs2_xattr_block_find(struct inode *inode,
152 struct ocfs2_xattr_search *xs);
153 static int ocfs2_xattr_index_block_find(struct inode *inode,
154 struct buffer_head *root_bh,
157 struct ocfs2_xattr_search *xs);
159 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
160 struct ocfs2_xattr_tree_root *xt,
164 static int ocfs2_xattr_create_index_block(struct inode *inode,
165 struct ocfs2_xattr_search *xs,
166 struct ocfs2_xattr_set_ctxt *ctxt);
168 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
169 struct ocfs2_xattr_info *xi,
170 struct ocfs2_xattr_search *xs,
171 struct ocfs2_xattr_set_ctxt *ctxt);
173 static int ocfs2_delete_xattr_index_block(struct inode *inode,
174 struct buffer_head *xb_bh);
175 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
176 u64 src_blk, u64 last_blk, u64 to_blk,
177 unsigned int start_bucket,
179 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
180 struct ocfs2_dinode *di,
181 struct ocfs2_xattr_info *xi,
182 struct ocfs2_xattr_search *xis,
183 struct ocfs2_xattr_search *xbs,
184 struct ocfs2_refcount_tree **ref_tree,
188 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
190 return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
193 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
195 return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
198 static inline u16 ocfs2_xattr_max_xe_in_bucket(struct super_block *sb)
200 u16 len = sb->s_blocksize -
201 offsetof(struct ocfs2_xattr_header, xh_entries);
203 return len / sizeof(struct ocfs2_xattr_entry);
206 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
207 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
208 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
210 static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
212 struct ocfs2_xattr_bucket *bucket;
213 int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
215 BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
217 bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
219 bucket->bu_inode = inode;
220 bucket->bu_blocks = blks;
226 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
230 for (i = 0; i < bucket->bu_blocks; i++) {
231 brelse(bucket->bu_bhs[i]);
232 bucket->bu_bhs[i] = NULL;
236 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
239 ocfs2_xattr_bucket_relse(bucket);
240 bucket->bu_inode = NULL;
246 * A bucket that has never been written to disk doesn't need to be
247 * read. We just need the buffer_heads. Don't call this for
248 * buckets that are already on disk. ocfs2_read_xattr_bucket() initializes
251 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
256 for (i = 0; i < bucket->bu_blocks; i++) {
257 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
259 if (!bucket->bu_bhs[i]) {
265 if (!ocfs2_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
267 ocfs2_set_new_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
272 ocfs2_xattr_bucket_relse(bucket);
276 /* Read the xattr bucket at xb_blkno */
277 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
282 rc = ocfs2_read_blocks(INODE_CACHE(bucket->bu_inode), xb_blkno,
283 bucket->bu_blocks, bucket->bu_bhs, 0,
286 spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
287 rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
290 &bucket_xh(bucket)->xh_check);
291 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
297 ocfs2_xattr_bucket_relse(bucket);
301 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
302 struct ocfs2_xattr_bucket *bucket,
307 for (i = 0; i < bucket->bu_blocks; i++) {
308 rc = ocfs2_journal_access(handle,
309 INODE_CACHE(bucket->bu_inode),
310 bucket->bu_bhs[i], type);
320 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
321 struct ocfs2_xattr_bucket *bucket)
325 spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
326 ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
327 bucket->bu_bhs, bucket->bu_blocks,
328 &bucket_xh(bucket)->xh_check);
329 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
331 for (i = 0; i < bucket->bu_blocks; i++)
332 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
335 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
336 struct ocfs2_xattr_bucket *src)
339 int blocksize = src->bu_inode->i_sb->s_blocksize;
341 BUG_ON(dest->bu_blocks != src->bu_blocks);
342 BUG_ON(dest->bu_inode != src->bu_inode);
344 for (i = 0; i < src->bu_blocks; i++) {
345 memcpy(bucket_block(dest, i), bucket_block(src, i),
350 static int ocfs2_validate_xattr_block(struct super_block *sb,
351 struct buffer_head *bh)
354 struct ocfs2_xattr_block *xb =
355 (struct ocfs2_xattr_block *)bh->b_data;
357 mlog(0, "Validating xattr block %llu\n",
358 (unsigned long long)bh->b_blocknr);
360 BUG_ON(!buffer_uptodate(bh));
363 * If the ecc fails, we return the error but otherwise
364 * leave the filesystem running. We know any error is
365 * local to this block.
367 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
372 * Errors after here are fatal
375 if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
377 "Extended attribute block #%llu has bad "
379 (unsigned long long)bh->b_blocknr, 7,
384 if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
386 "Extended attribute block #%llu has an "
387 "invalid xb_blkno of %llu",
388 (unsigned long long)bh->b_blocknr,
389 (unsigned long long)le64_to_cpu(xb->xb_blkno));
393 if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
395 "Extended attribute block #%llu has an invalid "
396 "xb_fs_generation of #%u",
397 (unsigned long long)bh->b_blocknr,
398 le32_to_cpu(xb->xb_fs_generation));
405 static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
406 struct buffer_head **bh)
409 struct buffer_head *tmp = *bh;
411 rc = ocfs2_read_block(INODE_CACHE(inode), xb_blkno, &tmp,
412 ocfs2_validate_xattr_block);
414 /* If ocfs2_read_block() got us a new bh, pass it up. */
421 static inline const char *ocfs2_xattr_prefix(int name_index)
423 struct xattr_handler *handler = NULL;
425 if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
426 handler = ocfs2_xattr_handler_map[name_index];
428 return handler ? handler->prefix : NULL;
431 static u32 ocfs2_xattr_name_hash(struct inode *inode,
435 /* Get hash value of uuid from super block */
436 u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
439 /* hash extended attribute name */
440 for (i = 0; i < name_len; i++) {
441 hash = (hash << OCFS2_HASH_SHIFT) ^
442 (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
450 * ocfs2_xattr_hash_entry()
452 * Compute the hash of an extended attribute.
454 static void ocfs2_xattr_hash_entry(struct inode *inode,
455 struct ocfs2_xattr_header *header,
456 struct ocfs2_xattr_entry *entry)
459 char *name = (char *)header + le16_to_cpu(entry->xe_name_offset);
461 hash = ocfs2_xattr_name_hash(inode, name, entry->xe_name_len);
462 entry->xe_name_hash = cpu_to_le32(hash);
467 static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
471 if (value_len <= OCFS2_XATTR_INLINE_SIZE)
472 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
474 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
475 size += sizeof(struct ocfs2_xattr_entry);
480 int ocfs2_calc_security_init(struct inode *dir,
481 struct ocfs2_security_xattr_info *si,
484 struct ocfs2_alloc_context **xattr_ac)
487 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
488 int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
492 * The max space of security xattr taken inline is
493 * 256(name) + 80(value) + 16(entry) = 352 bytes,
494 * So reserve one metadata block for it is ok.
496 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
497 s_size > OCFS2_XATTR_FREE_IN_IBODY) {
498 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
503 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
506 /* reserve clusters for xattr value which will be set in B tree*/
507 if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
508 int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
511 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
513 *want_clusters += new_clusters;
518 int ocfs2_calc_xattr_init(struct inode *dir,
519 struct buffer_head *dir_bh,
521 struct ocfs2_security_xattr_info *si,
527 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
528 int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
531 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
534 if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
535 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
536 OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
539 a_size = ocfs2_xattr_entry_real_size(0, acl_len);
542 } else if (acl_len != 0 && acl_len != -ENODATA) {
548 if (!(s_size + a_size))
552 * The max space of security xattr taken inline is
553 * 256(name) + 80(value) + 16(entry) = 352 bytes,
554 * The max space of acl xattr taken inline is
555 * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
556 * when blocksize = 512, may reserve one more cluser for
557 * xattr bucket, otherwise reserve one metadata block
559 * If this is a new directory with inline data,
560 * we choose to reserve the entire inline area for
561 * directory contents and force an external xattr block.
563 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
564 (S_ISDIR(mode) && ocfs2_supports_inline_data(osb)) ||
565 (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
566 *want_meta = *want_meta + 1;
567 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
570 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
571 (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
573 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
577 * reserve credits and clusters for xattrs which has large value
578 * and have to be set outside
580 if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
581 new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
583 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
585 *want_clusters += new_clusters;
587 if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
588 acl_len > OCFS2_XATTR_INLINE_SIZE) {
589 /* for directory, it has DEFAULT and ACCESS two types of acls */
590 new_clusters = (S_ISDIR(mode) ? 2 : 1) *
591 ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
592 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
594 *want_clusters += new_clusters;
600 static int ocfs2_xattr_extend_allocation(struct inode *inode,
602 struct ocfs2_xattr_value_buf *vb,
603 struct ocfs2_xattr_set_ctxt *ctxt)
606 handle_t *handle = ctxt->handle;
607 enum ocfs2_alloc_restarted why;
608 u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
609 struct ocfs2_extent_tree et;
611 mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
613 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
615 status = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
616 OCFS2_JOURNAL_ACCESS_WRITE);
622 prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
623 status = ocfs2_add_clusters_in_btree(handle,
636 status = ocfs2_journal_dirty(handle, vb->vb_bh);
642 clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) - prev_clusters;
645 * We should have already allocated enough space before the transaction,
646 * so no need to restart.
648 BUG_ON(why != RESTART_NONE || clusters_to_add);
655 static int __ocfs2_remove_xattr_range(struct inode *inode,
656 struct ocfs2_xattr_value_buf *vb,
657 u32 cpos, u32 phys_cpos, u32 len,
658 unsigned int ext_flags,
659 struct ocfs2_xattr_set_ctxt *ctxt)
662 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
663 handle_t *handle = ctxt->handle;
664 struct ocfs2_extent_tree et;
666 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
668 ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
669 OCFS2_JOURNAL_ACCESS_WRITE);
675 ret = ocfs2_remove_extent(handle, &et, cpos, len, ctxt->meta_ac,
682 le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
684 ret = ocfs2_journal_dirty(handle, vb->vb_bh);
690 if (ext_flags & OCFS2_EXT_REFCOUNTED)
691 ret = ocfs2_decrease_refcount(inode, handle,
692 ocfs2_blocks_to_clusters(inode->i_sb,
694 len, ctxt->meta_ac, &ctxt->dealloc, 1);
696 ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc,
705 static int ocfs2_xattr_shrink_size(struct inode *inode,
708 struct ocfs2_xattr_value_buf *vb,
709 struct ocfs2_xattr_set_ctxt *ctxt)
712 unsigned int ext_flags;
713 u32 trunc_len, cpos, phys_cpos, alloc_size;
716 if (old_clusters <= new_clusters)
720 trunc_len = old_clusters - new_clusters;
722 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
724 &vb->vb_xv->xr_list, &ext_flags);
730 if (alloc_size > trunc_len)
731 alloc_size = trunc_len;
733 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
734 phys_cpos, alloc_size,
741 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
742 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode),
745 trunc_len -= alloc_size;
752 static int ocfs2_xattr_value_truncate(struct inode *inode,
753 struct ocfs2_xattr_value_buf *vb,
755 struct ocfs2_xattr_set_ctxt *ctxt)
758 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
759 u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
761 if (new_clusters == old_clusters)
764 if (new_clusters > old_clusters)
765 ret = ocfs2_xattr_extend_allocation(inode,
766 new_clusters - old_clusters,
769 ret = ocfs2_xattr_shrink_size(inode,
770 old_clusters, new_clusters,
776 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
777 size_t *result, const char *prefix,
778 const char *name, int name_len)
780 char *p = buffer + *result;
781 int prefix_len = strlen(prefix);
782 int total_len = prefix_len + name_len + 1;
784 *result += total_len;
786 /* we are just looking for how big our buffer needs to be */
793 memcpy(p, prefix, prefix_len);
794 memcpy(p + prefix_len, name, name_len);
795 p[prefix_len + name_len] = '\0';
800 static int ocfs2_xattr_list_entries(struct inode *inode,
801 struct ocfs2_xattr_header *header,
802 char *buffer, size_t buffer_size)
806 const char *prefix, *name;
808 for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
809 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
810 type = ocfs2_xattr_get_type(entry);
811 prefix = ocfs2_xattr_prefix(type);
814 name = (const char *)header +
815 le16_to_cpu(entry->xe_name_offset);
817 ret = ocfs2_xattr_list_entry(buffer, buffer_size,
818 &result, prefix, name,
828 static int ocfs2_xattr_ibody_list(struct inode *inode,
829 struct ocfs2_dinode *di,
833 struct ocfs2_xattr_header *header = NULL;
834 struct ocfs2_inode_info *oi = OCFS2_I(inode);
837 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
840 header = (struct ocfs2_xattr_header *)
841 ((void *)di + inode->i_sb->s_blocksize -
842 le16_to_cpu(di->i_xattr_inline_size));
844 ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
849 static int ocfs2_xattr_block_list(struct inode *inode,
850 struct ocfs2_dinode *di,
854 struct buffer_head *blk_bh = NULL;
855 struct ocfs2_xattr_block *xb;
858 if (!di->i_xattr_loc)
861 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
868 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
869 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
870 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
871 ret = ocfs2_xattr_list_entries(inode, header,
872 buffer, buffer_size);
874 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
875 ret = ocfs2_xattr_tree_list_index_block(inode, xt,
876 buffer, buffer_size);
884 ssize_t ocfs2_listxattr(struct dentry *dentry,
888 int ret = 0, i_ret = 0, b_ret = 0;
889 struct buffer_head *di_bh = NULL;
890 struct ocfs2_dinode *di = NULL;
891 struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
893 if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
896 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
899 ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
905 di = (struct ocfs2_dinode *)di_bh->b_data;
907 down_read(&oi->ip_xattr_sem);
908 i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
916 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
921 up_read(&oi->ip_xattr_sem);
922 ocfs2_inode_unlock(dentry->d_inode, 0);
926 return i_ret + b_ret;
929 static int ocfs2_xattr_find_entry(int name_index,
931 struct ocfs2_xattr_search *xs)
933 struct ocfs2_xattr_entry *entry;
940 name_len = strlen(name);
942 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
943 cmp = name_index - ocfs2_xattr_get_type(entry);
945 cmp = name_len - entry->xe_name_len;
947 cmp = memcmp(name, (xs->base +
948 le16_to_cpu(entry->xe_name_offset)),
956 return cmp ? -ENODATA : 0;
959 static int ocfs2_xattr_get_value_outside(struct inode *inode,
960 struct ocfs2_xattr_value_root *xv,
964 u32 cpos, p_cluster, num_clusters, bpc, clusters;
967 size_t cplen, blocksize;
968 struct buffer_head *bh = NULL;
969 struct ocfs2_extent_list *el;
972 clusters = le32_to_cpu(xv->xr_clusters);
973 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
974 blocksize = inode->i_sb->s_blocksize;
977 while (cpos < clusters) {
978 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
979 &num_clusters, el, NULL);
985 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
986 /* Copy ocfs2_xattr_value */
987 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
988 ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
995 cplen = len >= blocksize ? blocksize : len;
996 memcpy(buffer, bh->b_data, cplen);
1005 cpos += num_clusters;
1011 static int ocfs2_xattr_ibody_get(struct inode *inode,
1016 struct ocfs2_xattr_search *xs)
1018 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1019 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1020 struct ocfs2_xattr_value_root *xv;
1024 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1027 xs->end = (void *)di + inode->i_sb->s_blocksize;
1028 xs->header = (struct ocfs2_xattr_header *)
1029 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1030 xs->base = (void *)xs->header;
1031 xs->here = xs->header->xh_entries;
1033 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1036 size = le64_to_cpu(xs->here->xe_value_size);
1038 if (size > buffer_size)
1040 if (ocfs2_xattr_is_local(xs->here)) {
1041 memcpy(buffer, (void *)xs->base +
1042 le16_to_cpu(xs->here->xe_name_offset) +
1043 OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1045 xv = (struct ocfs2_xattr_value_root *)
1046 (xs->base + le16_to_cpu(
1047 xs->here->xe_name_offset) +
1048 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1049 ret = ocfs2_xattr_get_value_outside(inode, xv,
1061 static int ocfs2_xattr_block_get(struct inode *inode,
1066 struct ocfs2_xattr_search *xs)
1068 struct ocfs2_xattr_block *xb;
1069 struct ocfs2_xattr_value_root *xv;
1071 int ret = -ENODATA, name_offset, name_len, i;
1072 int uninitialized_var(block_off);
1074 xs->bucket = ocfs2_xattr_bucket_new(inode);
1081 ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1087 if (xs->not_found) {
1092 xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1093 size = le64_to_cpu(xs->here->xe_value_size);
1096 if (size > buffer_size)
1099 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1100 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1101 i = xs->here - xs->header->xh_entries;
1103 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1104 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
1105 bucket_xh(xs->bucket),
1109 xs->base = bucket_block(xs->bucket, block_off);
1111 if (ocfs2_xattr_is_local(xs->here)) {
1112 memcpy(buffer, (void *)xs->base +
1113 name_offset + name_len, size);
1115 xv = (struct ocfs2_xattr_value_root *)
1116 (xs->base + name_offset + name_len);
1117 ret = ocfs2_xattr_get_value_outside(inode, xv,
1127 ocfs2_xattr_bucket_free(xs->bucket);
1129 brelse(xs->xattr_bh);
1130 xs->xattr_bh = NULL;
1134 int ocfs2_xattr_get_nolock(struct inode *inode,
1135 struct buffer_head *di_bh,
1142 struct ocfs2_dinode *di = NULL;
1143 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1144 struct ocfs2_xattr_search xis = {
1145 .not_found = -ENODATA,
1147 struct ocfs2_xattr_search xbs = {
1148 .not_found = -ENODATA,
1151 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1154 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1157 xis.inode_bh = xbs.inode_bh = di_bh;
1158 di = (struct ocfs2_dinode *)di_bh->b_data;
1160 down_read(&oi->ip_xattr_sem);
1161 ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1163 if (ret == -ENODATA && di->i_xattr_loc)
1164 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1166 up_read(&oi->ip_xattr_sem);
1171 /* ocfs2_xattr_get()
1173 * Copy an extended attribute into the buffer provided.
1174 * Buffer is NULL to compute the size of buffer required.
1176 static int ocfs2_xattr_get(struct inode *inode,
1183 struct buffer_head *di_bh = NULL;
1185 ret = ocfs2_inode_lock(inode, &di_bh, 0);
1190 ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1191 name, buffer, buffer_size);
1193 ocfs2_inode_unlock(inode, 0);
1200 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1202 struct ocfs2_xattr_value_buf *vb,
1206 int ret = 0, i, cp_len;
1207 u16 blocksize = inode->i_sb->s_blocksize;
1208 u32 p_cluster, num_clusters;
1209 u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1210 u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1212 struct buffer_head *bh = NULL;
1213 unsigned int ext_flags;
1214 struct ocfs2_xattr_value_root *xv = vb->vb_xv;
1216 BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1218 while (cpos < clusters) {
1219 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1220 &num_clusters, &xv->xr_list,
1227 BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1229 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1231 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1232 ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1239 ret = ocfs2_journal_access(handle,
1242 OCFS2_JOURNAL_ACCESS_WRITE);
1248 cp_len = value_len > blocksize ? blocksize : value_len;
1249 memcpy(bh->b_data, value, cp_len);
1250 value_len -= cp_len;
1252 if (cp_len < blocksize)
1253 memset(bh->b_data + cp_len, 0,
1254 blocksize - cp_len);
1256 ret = ocfs2_journal_dirty(handle, bh);
1265 * XXX: do we need to empty all the following
1266 * blocks in this cluster?
1271 cpos += num_clusters;
1279 static int ocfs2_xattr_cleanup(struct inode *inode,
1281 struct ocfs2_xattr_info *xi,
1282 struct ocfs2_xattr_search *xs,
1283 struct ocfs2_xattr_value_buf *vb,
1287 size_t name_len = strlen(xi->name);
1288 void *val = xs->base + offs;
1289 size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1291 ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
1292 OCFS2_JOURNAL_ACCESS_WRITE);
1297 /* Decrease xattr count */
1298 le16_add_cpu(&xs->header->xh_count, -1);
1299 /* Remove the xattr entry and tree root which has already be set*/
1300 memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
1301 memset(val, 0, size);
1303 ret = ocfs2_journal_dirty(handle, vb->vb_bh);
1310 static int ocfs2_xattr_update_entry(struct inode *inode,
1312 struct ocfs2_xattr_info *xi,
1313 struct ocfs2_xattr_search *xs,
1314 struct ocfs2_xattr_value_buf *vb,
1319 ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
1320 OCFS2_JOURNAL_ACCESS_WRITE);
1326 xs->here->xe_name_offset = cpu_to_le16(offs);
1327 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1328 if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1329 ocfs2_xattr_set_local(xs->here, 1);
1331 ocfs2_xattr_set_local(xs->here, 0);
1332 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1334 ret = ocfs2_journal_dirty(handle, vb->vb_bh);
1342 * ocfs2_xattr_set_value_outside()
1344 * Set large size value in B tree.
1346 static int ocfs2_xattr_set_value_outside(struct inode *inode,
1347 struct ocfs2_xattr_info *xi,
1348 struct ocfs2_xattr_search *xs,
1349 struct ocfs2_xattr_set_ctxt *ctxt,
1350 struct ocfs2_xattr_value_buf *vb,
1353 size_t name_len = strlen(xi->name);
1354 void *val = xs->base + offs;
1355 struct ocfs2_xattr_value_root *xv = NULL;
1356 size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1359 memset(val, 0, size);
1360 memcpy(val, xi->name, name_len);
1361 xv = (struct ocfs2_xattr_value_root *)
1362 (val + OCFS2_XATTR_SIZE(name_len));
1363 xv->xr_clusters = 0;
1364 xv->xr_last_eb_blk = 0;
1365 xv->xr_list.l_tree_depth = 0;
1366 xv->xr_list.l_count = cpu_to_le16(1);
1367 xv->xr_list.l_next_free_rec = 0;
1370 ret = ocfs2_xattr_value_truncate(inode, vb, xi->value_len, ctxt);
1375 ret = ocfs2_xattr_update_entry(inode, ctxt->handle, xi, xs, vb, offs);
1380 ret = __ocfs2_xattr_set_value_outside(inode, ctxt->handle, vb,
1381 xi->value, xi->value_len);
1389 * ocfs2_xattr_set_entry_local()
1391 * Set, replace or remove extended attribute in local.
1393 static void ocfs2_xattr_set_entry_local(struct inode *inode,
1394 struct ocfs2_xattr_info *xi,
1395 struct ocfs2_xattr_search *xs,
1396 struct ocfs2_xattr_entry *last,
1399 size_t name_len = strlen(xi->name);
1402 if (xi->value && xs->not_found) {
1403 /* Insert the new xattr entry. */
1404 le16_add_cpu(&xs->header->xh_count, 1);
1405 ocfs2_xattr_set_type(last, xi->name_index);
1406 ocfs2_xattr_set_local(last, 1);
1407 last->xe_name_len = name_len;
1413 first_val = xs->base + min_offs;
1414 offs = le16_to_cpu(xs->here->xe_name_offset);
1415 val = xs->base + offs;
1417 if (le64_to_cpu(xs->here->xe_value_size) >
1418 OCFS2_XATTR_INLINE_SIZE)
1419 size = OCFS2_XATTR_SIZE(name_len) +
1420 OCFS2_XATTR_ROOT_SIZE;
1422 size = OCFS2_XATTR_SIZE(name_len) +
1423 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1425 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1426 OCFS2_XATTR_SIZE(xi->value_len)) {
1427 /* The old and the new value have the
1428 same size. Just replace the value. */
1429 ocfs2_xattr_set_local(xs->here, 1);
1430 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1431 /* Clear value bytes. */
1432 memset(val + OCFS2_XATTR_SIZE(name_len),
1434 OCFS2_XATTR_SIZE(xi->value_len));
1435 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1440 /* Remove the old name+value. */
1441 memmove(first_val + size, first_val, val - first_val);
1442 memset(first_val, 0, size);
1443 xs->here->xe_name_hash = 0;
1444 xs->here->xe_name_offset = 0;
1445 ocfs2_xattr_set_local(xs->here, 1);
1446 xs->here->xe_value_size = 0;
1450 /* Adjust all value offsets. */
1451 last = xs->header->xh_entries;
1452 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1453 size_t o = le16_to_cpu(last->xe_name_offset);
1456 last->xe_name_offset = cpu_to_le16(o + size);
1461 /* Remove the old entry. */
1463 memmove(xs->here, xs->here + 1,
1464 (void *)last - (void *)xs->here);
1465 memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1466 le16_add_cpu(&xs->header->xh_count, -1);
1470 /* Insert the new name+value. */
1471 size_t size = OCFS2_XATTR_SIZE(name_len) +
1472 OCFS2_XATTR_SIZE(xi->value_len);
1473 void *val = xs->base + min_offs - size;
1475 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1476 memset(val, 0, size);
1477 memcpy(val, xi->name, name_len);
1478 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1481 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1482 ocfs2_xattr_set_local(xs->here, 1);
1483 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1490 * ocfs2_xattr_set_entry()
1492 * Set extended attribute entry into inode or block.
1494 * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1495 * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1496 * then set value in B tree with set_value_outside().
1498 static int ocfs2_xattr_set_entry(struct inode *inode,
1499 struct ocfs2_xattr_info *xi,
1500 struct ocfs2_xattr_search *xs,
1501 struct ocfs2_xattr_set_ctxt *ctxt,
1504 struct ocfs2_xattr_entry *last;
1505 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1506 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1507 size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1509 handle_t *handle = ctxt->handle;
1511 struct ocfs2_xattr_info xi_l = {
1512 .name_index = xi->name_index,
1515 .value_len = xi->value_len,
1517 struct ocfs2_xattr_value_buf vb = {
1518 .vb_bh = xs->xattr_bh,
1519 .vb_access = ocfs2_journal_access_di,
1522 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1523 BUG_ON(xs->xattr_bh == xs->inode_bh);
1524 vb.vb_access = ocfs2_journal_access_xb;
1526 BUG_ON(xs->xattr_bh != xs->inode_bh);
1528 /* Compute min_offs, last and free space. */
1529 last = xs->header->xh_entries;
1531 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1532 size_t offs = le16_to_cpu(last->xe_name_offset);
1533 if (offs < min_offs)
1538 free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
1542 if (!xs->not_found) {
1544 if (ocfs2_xattr_is_local(xs->here))
1545 size = OCFS2_XATTR_SIZE(name_len) +
1546 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1548 size = OCFS2_XATTR_SIZE(name_len) +
1549 OCFS2_XATTR_ROOT_SIZE;
1550 free += (size + sizeof(struct ocfs2_xattr_entry));
1552 /* Check free space in inode or block */
1553 if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1554 if (free < sizeof(struct ocfs2_xattr_entry) +
1555 OCFS2_XATTR_SIZE(name_len) +
1556 OCFS2_XATTR_ROOT_SIZE) {
1560 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1561 xi_l.value = (void *)&def_xv;
1562 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1563 } else if (xi->value) {
1564 if (free < sizeof(struct ocfs2_xattr_entry) +
1565 OCFS2_XATTR_SIZE(name_len) +
1566 OCFS2_XATTR_SIZE(xi->value_len)) {
1572 if (!xs->not_found) {
1573 /* For existing extended attribute */
1574 size_t size = OCFS2_XATTR_SIZE(name_len) +
1575 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1576 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1577 void *val = xs->base + offs;
1579 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1580 /* Replace existing local xattr with tree root */
1581 ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1586 } else if (!ocfs2_xattr_is_local(xs->here)) {
1587 /* For existing xattr which has value outside */
1588 vb.vb_xv = (struct ocfs2_xattr_value_root *)
1589 (val + OCFS2_XATTR_SIZE(name_len));
1591 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1593 * If new value need set outside also,
1594 * first truncate old value to new value,
1595 * then set new value with set_value_outside().
1597 ret = ocfs2_xattr_value_truncate(inode,
1606 ret = ocfs2_xattr_update_entry(inode,
1617 ret = __ocfs2_xattr_set_value_outside(inode,
1627 * If new value need set in local,
1628 * just trucate old value to zero.
1630 ret = ocfs2_xattr_value_truncate(inode,
1640 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), xs->inode_bh,
1641 OCFS2_JOURNAL_ACCESS_WRITE);
1647 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1648 ret = vb.vb_access(handle, INODE_CACHE(inode), vb.vb_bh,
1649 OCFS2_JOURNAL_ACCESS_WRITE);
1657 * Set value in local, include set tree root in local.
1658 * This is the first step for value size >INLINE_SIZE.
1660 ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1662 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1663 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1670 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1671 (flag & OCFS2_INLINE_XATTR_FL)) {
1672 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1673 unsigned int xattrsize = osb->s_xattr_inline_size;
1676 * Adjust extent record count or inline data size
1677 * to reserve space for extended attribute.
1679 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1680 struct ocfs2_inline_data *idata = &di->id2.i_data;
1681 le16_add_cpu(&idata->id_count, -xattrsize);
1682 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1683 struct ocfs2_extent_list *el = &di->id2.i_list;
1684 le16_add_cpu(&el->l_count, -(xattrsize /
1685 sizeof(struct ocfs2_extent_rec)));
1687 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1689 /* Update xattr flag */
1690 spin_lock(&oi->ip_lock);
1691 oi->ip_dyn_features |= flag;
1692 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1693 spin_unlock(&oi->ip_lock);
1695 ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1699 if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1701 * Set value outside in B tree.
1702 * This is the second step for value size > INLINE_SIZE.
1704 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1705 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, ctxt,
1712 * If set value outside failed, we have to clean
1713 * the junk tree root we have already set in local.
1715 ret2 = ocfs2_xattr_cleanup(inode, ctxt->handle,
1725 static int ocfs2_remove_value_outside(struct inode*inode,
1726 struct ocfs2_xattr_value_buf *vb,
1727 struct ocfs2_xattr_header *header)
1730 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1731 struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
1733 ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
1735 ctxt.handle = ocfs2_start_trans(osb,
1736 ocfs2_remove_extent_credits(osb->sb));
1737 if (IS_ERR(ctxt.handle)) {
1738 ret = PTR_ERR(ctxt.handle);
1743 for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1744 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1746 if (!ocfs2_xattr_is_local(entry)) {
1749 val = (void *)header +
1750 le16_to_cpu(entry->xe_name_offset);
1751 vb->vb_xv = (struct ocfs2_xattr_value_root *)
1752 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
1753 ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
1761 ocfs2_commit_trans(osb, ctxt.handle);
1762 ocfs2_schedule_truncate_log_flush(osb, 1);
1763 ocfs2_run_deallocs(osb, &ctxt.dealloc);
1768 static int ocfs2_xattr_ibody_remove(struct inode *inode,
1769 struct buffer_head *di_bh)
1772 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1773 struct ocfs2_xattr_header *header;
1775 struct ocfs2_xattr_value_buf vb = {
1777 .vb_access = ocfs2_journal_access_di,
1780 header = (struct ocfs2_xattr_header *)
1781 ((void *)di + inode->i_sb->s_blocksize -
1782 le16_to_cpu(di->i_xattr_inline_size));
1784 ret = ocfs2_remove_value_outside(inode, &vb, header);
1789 static int ocfs2_xattr_block_remove(struct inode *inode,
1790 struct buffer_head *blk_bh)
1792 struct ocfs2_xattr_block *xb;
1794 struct ocfs2_xattr_value_buf vb = {
1796 .vb_access = ocfs2_journal_access_xb,
1799 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1800 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1801 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1802 ret = ocfs2_remove_value_outside(inode, &vb, header);
1804 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
1809 static int ocfs2_xattr_free_block(struct inode *inode,
1812 struct inode *xb_alloc_inode;
1813 struct buffer_head *xb_alloc_bh = NULL;
1814 struct buffer_head *blk_bh = NULL;
1815 struct ocfs2_xattr_block *xb;
1816 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1822 ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
1828 ret = ocfs2_xattr_block_remove(inode, blk_bh);
1834 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1835 blk = le64_to_cpu(xb->xb_blkno);
1836 bit = le16_to_cpu(xb->xb_suballoc_bit);
1837 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1839 xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1840 EXTENT_ALLOC_SYSTEM_INODE,
1841 le16_to_cpu(xb->xb_suballoc_slot));
1842 if (!xb_alloc_inode) {
1847 mutex_lock(&xb_alloc_inode->i_mutex);
1849 ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1855 handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
1856 if (IS_ERR(handle)) {
1857 ret = PTR_ERR(handle);
1862 ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1867 ocfs2_commit_trans(osb, handle);
1869 ocfs2_inode_unlock(xb_alloc_inode, 1);
1870 brelse(xb_alloc_bh);
1872 mutex_unlock(&xb_alloc_inode->i_mutex);
1873 iput(xb_alloc_inode);
1880 * ocfs2_xattr_remove()
1882 * Free extended attribute resources associated with this inode.
1884 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1886 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1887 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1891 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1894 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1897 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1898 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1905 if (di->i_xattr_loc) {
1906 ret = ocfs2_xattr_free_block(inode,
1907 le64_to_cpu(di->i_xattr_loc));
1914 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1915 OCFS2_INODE_UPDATE_CREDITS);
1916 if (IS_ERR(handle)) {
1917 ret = PTR_ERR(handle);
1921 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
1922 OCFS2_JOURNAL_ACCESS_WRITE);
1928 di->i_xattr_loc = 0;
1930 spin_lock(&oi->ip_lock);
1931 oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1932 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1933 spin_unlock(&oi->ip_lock);
1935 ret = ocfs2_journal_dirty(handle, di_bh);
1939 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1944 static int ocfs2_xattr_has_space_inline(struct inode *inode,
1945 struct ocfs2_dinode *di)
1947 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1948 unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1951 if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1954 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1955 struct ocfs2_inline_data *idata = &di->id2.i_data;
1956 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1957 } else if (ocfs2_inode_is_fast_symlink(inode)) {
1958 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1959 le64_to_cpu(di->i_size);
1961 struct ocfs2_extent_list *el = &di->id2.i_list;
1962 free = (le16_to_cpu(el->l_count) -
1963 le16_to_cpu(el->l_next_free_rec)) *
1964 sizeof(struct ocfs2_extent_rec);
1966 if (free >= xattrsize)
1973 * ocfs2_xattr_ibody_find()
1975 * Find extended attribute in inode block and
1976 * fill search info into struct ocfs2_xattr_search.
1978 static int ocfs2_xattr_ibody_find(struct inode *inode,
1981 struct ocfs2_xattr_search *xs)
1983 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1984 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1988 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1991 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1992 down_read(&oi->ip_alloc_sem);
1993 has_space = ocfs2_xattr_has_space_inline(inode, di);
1994 up_read(&oi->ip_alloc_sem);
1999 xs->xattr_bh = xs->inode_bh;
2000 xs->end = (void *)di + inode->i_sb->s_blocksize;
2001 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
2002 xs->header = (struct ocfs2_xattr_header *)
2003 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
2005 xs->header = (struct ocfs2_xattr_header *)
2006 (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
2007 xs->base = (void *)xs->header;
2008 xs->here = xs->header->xh_entries;
2010 /* Find the named attribute. */
2011 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2012 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2013 if (ret && ret != -ENODATA)
2015 xs->not_found = ret;
2022 * ocfs2_xattr_ibody_set()
2024 * Set, replace or remove an extended attribute into inode block.
2027 static int ocfs2_xattr_ibody_set(struct inode *inode,
2028 struct ocfs2_xattr_info *xi,
2029 struct ocfs2_xattr_search *xs,
2030 struct ocfs2_xattr_set_ctxt *ctxt)
2032 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2033 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2036 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2039 down_write(&oi->ip_alloc_sem);
2040 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2041 if (!ocfs2_xattr_has_space_inline(inode, di)) {
2047 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2048 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
2050 up_write(&oi->ip_alloc_sem);
2056 * ocfs2_xattr_block_find()
2058 * Find extended attribute in external block and
2059 * fill search info into struct ocfs2_xattr_search.
2061 static int ocfs2_xattr_block_find(struct inode *inode,
2064 struct ocfs2_xattr_search *xs)
2066 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2067 struct buffer_head *blk_bh = NULL;
2068 struct ocfs2_xattr_block *xb;
2071 if (!di->i_xattr_loc)
2074 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2081 xs->xattr_bh = blk_bh;
2082 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2084 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2085 xs->header = &xb->xb_attrs.xb_header;
2086 xs->base = (void *)xs->header;
2087 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2088 xs->here = xs->header->xh_entries;
2090 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2092 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2096 if (ret && ret != -ENODATA) {
2097 xs->xattr_bh = NULL;
2100 xs->not_found = ret;
2109 * ocfs2_xattr_block_set()
2111 * Set, replace or remove an extended attribute into external block.
2114 static int ocfs2_xattr_block_set(struct inode *inode,
2115 struct ocfs2_xattr_info *xi,
2116 struct ocfs2_xattr_search *xs,
2117 struct ocfs2_xattr_set_ctxt *ctxt)
2119 struct buffer_head *new_bh = NULL;
2120 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2121 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2122 handle_t *handle = ctxt->handle;
2123 struct ocfs2_xattr_block *xblk = NULL;
2124 u16 suballoc_bit_start;
2129 if (!xs->xattr_bh) {
2130 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2132 OCFS2_JOURNAL_ACCESS_CREATE);
2138 ret = ocfs2_claim_metadata(osb, handle, ctxt->meta_ac, 1,
2139 &suballoc_bit_start, &num_got,
2146 new_bh = sb_getblk(inode->i_sb, first_blkno);
2147 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2149 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode),
2151 OCFS2_JOURNAL_ACCESS_CREATE);
2157 /* Initialize ocfs2_xattr_block */
2158 xs->xattr_bh = new_bh;
2159 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2160 memset(xblk, 0, inode->i_sb->s_blocksize);
2161 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2162 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
2163 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2164 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
2165 xblk->xb_blkno = cpu_to_le64(first_blkno);
2167 xs->header = &xblk->xb_attrs.xb_header;
2168 xs->base = (void *)xs->header;
2169 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2170 xs->here = xs->header->xh_entries;
2172 ret = ocfs2_journal_dirty(handle, new_bh);
2177 di->i_xattr_loc = cpu_to_le64(first_blkno);
2178 ocfs2_journal_dirty(handle, xs->inode_bh);
2180 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2182 if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2183 /* Set extended attribute into external block */
2184 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2185 OCFS2_HAS_XATTR_FL);
2186 if (!ret || ret != -ENOSPC)
2189 ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2194 ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2201 /* Check whether the new xattr can be inserted into the inode. */
2202 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2203 struct ocfs2_xattr_info *xi,
2204 struct ocfs2_xattr_search *xs)
2207 struct ocfs2_xattr_entry *last;
2209 size_t min_offs = xs->end - xs->base;
2214 last = xs->header->xh_entries;
2216 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
2217 size_t offs = le16_to_cpu(last->xe_name_offset);
2218 if (offs < min_offs)
2223 free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
2227 BUG_ON(!xs->not_found);
2229 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2230 value_size = OCFS2_XATTR_ROOT_SIZE;
2232 value_size = OCFS2_XATTR_SIZE(xi->value_len);
2234 if (free >= sizeof(struct ocfs2_xattr_entry) +
2235 OCFS2_XATTR_SIZE(strlen(xi->name)) + value_size)
2241 static int ocfs2_calc_xattr_set_need(struct inode *inode,
2242 struct ocfs2_dinode *di,
2243 struct ocfs2_xattr_info *xi,
2244 struct ocfs2_xattr_search *xis,
2245 struct ocfs2_xattr_search *xbs,
2250 int ret = 0, old_in_xb = 0;
2251 int clusters_add = 0, meta_add = 0, credits = 0;
2252 struct buffer_head *bh = NULL;
2253 struct ocfs2_xattr_block *xb = NULL;
2254 struct ocfs2_xattr_entry *xe = NULL;
2255 struct ocfs2_xattr_value_root *xv = NULL;
2257 int name_offset, name_len = 0;
2258 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2263 * Calculate the clusters we need to write.
2264 * No matter whether we replace an old one or add a new one,
2265 * we need this for writing.
2267 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2268 credits += new_clusters *
2269 ocfs2_clusters_to_blocks(inode->i_sb, 1);
2271 if (xis->not_found && xbs->not_found) {
2272 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2274 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2275 clusters_add += new_clusters;
2276 credits += ocfs2_calc_extend_credits(inode->i_sb,
2284 if (!xis->not_found) {
2286 name_offset = le16_to_cpu(xe->xe_name_offset);
2287 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2289 credits += OCFS2_INODE_UPDATE_CREDITS;
2291 int i, block_off = 0;
2292 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2294 name_offset = le16_to_cpu(xe->xe_name_offset);
2295 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2296 i = xbs->here - xbs->header->xh_entries;
2299 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2300 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
2301 bucket_xh(xbs->bucket),
2304 base = bucket_block(xbs->bucket, block_off);
2305 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2308 credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
2313 * delete a xattr doesn't need metadata and cluster allocation.
2314 * so just calculate the credits and return.
2316 * The credits for removing the value tree will be extended
2317 * by ocfs2_remove_extent itself.
2320 if (!ocfs2_xattr_is_local(xe))
2321 credits += ocfs2_remove_extent_credits(inode->i_sb);
2326 /* do cluster allocation guess first. */
2327 value_size = le64_to_cpu(xe->xe_value_size);
2331 * In xattr set, we always try to set the xe in inode first,
2332 * so if it can be inserted into inode successfully, the old
2333 * one will be removed from the xattr block, and this xattr
2334 * will be inserted into inode as a new xattr in inode.
2336 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
2337 clusters_add += new_clusters;
2338 credits += ocfs2_remove_extent_credits(inode->i_sb) +
2339 OCFS2_INODE_UPDATE_CREDITS;
2340 if (!ocfs2_xattr_is_local(xe))
2341 credits += ocfs2_calc_extend_credits(
2349 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2350 /* the new values will be stored outside. */
2351 u32 old_clusters = 0;
2353 if (!ocfs2_xattr_is_local(xe)) {
2354 old_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2356 xv = (struct ocfs2_xattr_value_root *)
2357 (base + name_offset + name_len);
2358 value_size = OCFS2_XATTR_ROOT_SIZE;
2362 if (old_clusters >= new_clusters) {
2363 credits += ocfs2_remove_extent_credits(inode->i_sb);
2366 meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
2367 clusters_add += new_clusters - old_clusters;
2368 credits += ocfs2_calc_extend_credits(inode->i_sb,
2372 if (value_size >= OCFS2_XATTR_ROOT_SIZE)
2377 * Now the new value will be stored inside. So if the new
2378 * value is smaller than the size of value root or the old
2379 * value, we don't need any allocation, otherwise we have
2380 * to guess metadata allocation.
2382 if ((ocfs2_xattr_is_local(xe) && value_size >= xi->value_len) ||
2383 (!ocfs2_xattr_is_local(xe) &&
2384 OCFS2_XATTR_ROOT_SIZE >= xi->value_len))
2389 /* calculate metadata allocation. */
2390 if (di->i_xattr_loc) {
2391 if (!xbs->xattr_bh) {
2392 ret = ocfs2_read_xattr_block(inode,
2393 le64_to_cpu(di->i_xattr_loc),
2400 xb = (struct ocfs2_xattr_block *)bh->b_data;
2402 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2405 * If there is already an xattr tree, good, we can calculate
2406 * like other b-trees. Otherwise we may have the chance of
2407 * create a tree, the credit calculation is borrowed from
2408 * ocfs2_calc_extend_credits with root_el = NULL. And the
2409 * new tree will be cluster based, so no meta is needed.
2411 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2412 struct ocfs2_extent_list *el =
2413 &xb->xb_attrs.xb_root.xt_list;
2414 meta_add += ocfs2_extend_meta_needed(el);
2415 credits += ocfs2_calc_extend_credits(inode->i_sb,
2418 credits += OCFS2_SUBALLOC_ALLOC + 1;
2421 * This cluster will be used either for new bucket or for
2423 * If the cluster size is the same as the bucket size, one
2424 * more is needed since we may need to extend the bucket
2428 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2429 if (OCFS2_XATTR_BUCKET_SIZE ==
2430 OCFS2_SB(inode->i_sb)->s_clustersize) {
2431 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2436 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
2440 *clusters_need = clusters_add;
2442 *meta_need = meta_add;
2444 *credits_need = credits;
2449 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
2450 struct ocfs2_dinode *di,
2451 struct ocfs2_xattr_info *xi,
2452 struct ocfs2_xattr_search *xis,
2453 struct ocfs2_xattr_search *xbs,
2454 struct ocfs2_xattr_set_ctxt *ctxt,
2458 int clusters_add, meta_add, ret;
2459 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2461 memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
2463 ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
2465 ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
2466 &clusters_add, &meta_add, credits);
2472 meta_add += extra_meta;
2473 mlog(0, "Set xattr %s, reserve meta blocks = %d, clusters = %d, "
2474 "credits = %d\n", xi->name, meta_add, clusters_add, *credits);
2477 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
2486 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
2492 if (ctxt->meta_ac) {
2493 ocfs2_free_alloc_context(ctxt->meta_ac);
2494 ctxt->meta_ac = NULL;
2498 * We cannot have an error and a non null ctxt->data_ac.
2505 static int __ocfs2_xattr_set_handle(struct inode *inode,
2506 struct ocfs2_dinode *di,
2507 struct ocfs2_xattr_info *xi,
2508 struct ocfs2_xattr_search *xis,
2509 struct ocfs2_xattr_search *xbs,
2510 struct ocfs2_xattr_set_ctxt *ctxt)
2512 int ret = 0, credits, old_found;
2515 /* Remove existing extended attribute */
2516 if (!xis->not_found)
2517 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2518 else if (!xbs->not_found)
2519 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2521 /* We always try to set extended attribute into inode first*/
2522 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2523 if (!ret && !xbs->not_found) {
2525 * If succeed and that extended attribute existing in
2526 * external block, then we will remove it.
2531 old_found = xis->not_found;
2532 xis->not_found = -ENODATA;
2533 ret = ocfs2_calc_xattr_set_need(inode,
2541 xis->not_found = old_found;
2547 ret = ocfs2_extend_trans(ctxt->handle, credits +
2548 ctxt->handle->h_buffer_credits);
2553 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2554 } else if (ret == -ENOSPC) {
2555 if (di->i_xattr_loc && !xbs->xattr_bh) {
2556 ret = ocfs2_xattr_block_find(inode,
2562 old_found = xis->not_found;
2563 xis->not_found = -ENODATA;
2564 ret = ocfs2_calc_xattr_set_need(inode,
2572 xis->not_found = old_found;
2578 ret = ocfs2_extend_trans(ctxt->handle, credits +
2579 ctxt->handle->h_buffer_credits);
2586 * If no space in inode, we will set extended attribute
2587 * into external block.
2589 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2592 if (!xis->not_found) {
2594 * If succeed and that extended attribute
2595 * existing in inode, we will remove it.
2599 xbs->not_found = -ENODATA;
2600 ret = ocfs2_calc_xattr_set_need(inode,
2613 ret = ocfs2_extend_trans(ctxt->handle, credits +
2614 ctxt->handle->h_buffer_credits);
2619 ret = ocfs2_xattr_ibody_set(inode, xi,
2626 /* Update inode ctime. */
2627 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
2629 OCFS2_JOURNAL_ACCESS_WRITE);
2635 inode->i_ctime = CURRENT_TIME;
2636 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
2637 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
2638 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
2645 * This function only called duing creating inode
2646 * for init security/acl xattrs of the new inode.
2647 * All transanction credits have been reserved in mknod.
2649 int ocfs2_xattr_set_handle(handle_t *handle,
2650 struct inode *inode,
2651 struct buffer_head *di_bh,
2657 struct ocfs2_alloc_context *meta_ac,
2658 struct ocfs2_alloc_context *data_ac)
2660 struct ocfs2_dinode *di;
2663 struct ocfs2_xattr_info xi = {
2664 .name_index = name_index,
2667 .value_len = value_len,
2670 struct ocfs2_xattr_search xis = {
2671 .not_found = -ENODATA,
2674 struct ocfs2_xattr_search xbs = {
2675 .not_found = -ENODATA,
2678 struct ocfs2_xattr_set_ctxt ctxt = {
2684 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2688 * In extreme situation, may need xattr bucket when
2689 * block size is too small. And we have already reserved
2690 * the credits for bucket in mknod.
2692 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
2693 xbs.bucket = ocfs2_xattr_bucket_new(inode);
2695 mlog_errno(-ENOMEM);
2700 xis.inode_bh = xbs.inode_bh = di_bh;
2701 di = (struct ocfs2_dinode *)di_bh->b_data;
2703 down_write(&OCFS2_I(inode)->ip_xattr_sem);
2705 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2708 if (xis.not_found) {
2709 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2714 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2717 up_write(&OCFS2_I(inode)->ip_xattr_sem);
2718 brelse(xbs.xattr_bh);
2719 ocfs2_xattr_bucket_free(xbs.bucket);
2727 * Set, replace or remove an extended attribute for this inode.
2728 * value is NULL to remove an existing extended attribute, else either
2729 * create or replace an extended attribute.
2731 int ocfs2_xattr_set(struct inode *inode,
2738 struct buffer_head *di_bh = NULL;
2739 struct ocfs2_dinode *di;
2740 int ret, credits, ref_meta = 0, ref_credits = 0;
2741 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2742 struct inode *tl_inode = osb->osb_tl_inode;
2743 struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2744 struct ocfs2_refcount_tree *ref_tree = NULL;
2746 struct ocfs2_xattr_info xi = {
2747 .name_index = name_index,
2750 .value_len = value_len,
2753 struct ocfs2_xattr_search xis = {
2754 .not_found = -ENODATA,
2757 struct ocfs2_xattr_search xbs = {
2758 .not_found = -ENODATA,
2761 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2765 * Only xbs will be used on indexed trees. xis doesn't need a
2768 xbs.bucket = ocfs2_xattr_bucket_new(inode);
2770 mlog_errno(-ENOMEM);
2774 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2777 goto cleanup_nolock;
2779 xis.inode_bh = xbs.inode_bh = di_bh;
2780 di = (struct ocfs2_dinode *)di_bh->b_data;
2782 down_write(&OCFS2_I(inode)->ip_xattr_sem);
2784 * Scan inode and external block to find the same name
2785 * extended attribute and collect search infomation.
2787 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2790 if (xis.not_found) {
2791 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2796 if (xis.not_found && xbs.not_found) {
2798 if (flags & XATTR_REPLACE)
2805 if (flags & XATTR_CREATE)
2809 /* Check whether the value is refcounted and do some prepartion. */
2810 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL &&
2811 (!xis.not_found || !xbs.not_found)) {
2812 ret = ocfs2_prepare_refcount_xattr(inode, di, &xi,
2813 &xis, &xbs, &ref_tree,
2814 &ref_meta, &ref_credits);
2821 mutex_lock(&tl_inode->i_mutex);
2823 if (ocfs2_truncate_log_needs_flush(osb)) {
2824 ret = __ocfs2_flush_truncate_log(osb);
2826 mutex_unlock(&tl_inode->i_mutex);
2831 mutex_unlock(&tl_inode->i_mutex);
2833 ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
2834 &xbs, &ctxt, ref_meta, &credits);
2840 /* we need to update inode's ctime field, so add credit for it. */
2841 credits += OCFS2_INODE_UPDATE_CREDITS;
2842 ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
2843 if (IS_ERR(ctxt.handle)) {
2844 ret = PTR_ERR(ctxt.handle);
2849 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2851 ocfs2_commit_trans(osb, ctxt.handle);
2854 ocfs2_free_alloc_context(ctxt.data_ac);
2856 ocfs2_free_alloc_context(ctxt.meta_ac);
2857 if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
2858 ocfs2_schedule_truncate_log_flush(osb, 1);
2859 ocfs2_run_deallocs(osb, &ctxt.dealloc);
2862 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
2863 up_write(&OCFS2_I(inode)->ip_xattr_sem);
2864 ocfs2_inode_unlock(inode, 1);
2867 brelse(xbs.xattr_bh);
2868 ocfs2_xattr_bucket_free(xbs.bucket);
2874 * Find the xattr extent rec which may contains name_hash.
2875 * e_cpos will be the first name hash of the xattr rec.
2876 * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2878 static int ocfs2_xattr_get_rec(struct inode *inode,
2883 struct ocfs2_extent_list *el)
2886 struct buffer_head *eb_bh = NULL;
2887 struct ocfs2_extent_block *eb;
2888 struct ocfs2_extent_rec *rec = NULL;
2891 if (el->l_tree_depth) {
2892 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, name_hash,
2899 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2902 if (el->l_tree_depth) {
2903 ocfs2_error(inode->i_sb,
2904 "Inode %lu has non zero tree depth in "
2905 "xattr tree block %llu\n", inode->i_ino,
2906 (unsigned long long)eb_bh->b_blocknr);
2912 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2913 rec = &el->l_recs[i];
2915 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2916 e_blkno = le64_to_cpu(rec->e_blkno);
2922 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2923 "record (%u, %u, 0) in xattr", inode->i_ino,
2924 le32_to_cpu(rec->e_cpos),
2925 ocfs2_rec_clusters(el, rec));
2930 *p_blkno = le64_to_cpu(rec->e_blkno);
2931 *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2933 *e_cpos = le32_to_cpu(rec->e_cpos);
2939 typedef int (xattr_bucket_func)(struct inode *inode,
2940 struct ocfs2_xattr_bucket *bucket,
2943 static int ocfs2_find_xe_in_bucket(struct inode *inode,
2944 struct ocfs2_xattr_bucket *bucket,
2951 int i, ret = 0, cmp = 1, block_off, new_offset;
2952 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
2953 size_t name_len = strlen(name);
2954 struct ocfs2_xattr_entry *xe = NULL;
2958 * We don't use binary search in the bucket because there
2959 * may be multiple entries with the same name hash.
2961 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2962 xe = &xh->xh_entries[i];
2964 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2966 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2969 cmp = name_index - ocfs2_xattr_get_type(xe);
2971 cmp = name_len - xe->xe_name_len;
2975 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
2986 xe_name = bucket_block(bucket, block_off) + new_offset;
2987 if (!memcmp(name, xe_name, name_len)) {
2999 * Find the specified xattr entry in a series of buckets.
3000 * This series start from p_blkno and last for num_clusters.
3001 * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
3002 * the num of the valid buckets.
3004 * Return the buffer_head this xattr should reside in. And if the xattr's
3005 * hash is in the gap of 2 buckets, return the lower bucket.
3007 static int ocfs2_xattr_bucket_find(struct inode *inode,
3014 struct ocfs2_xattr_search *xs)
3017 struct ocfs2_xattr_header *xh = NULL;
3018 struct ocfs2_xattr_entry *xe = NULL;
3020 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3021 int low_bucket = 0, bucket, high_bucket;
3022 struct ocfs2_xattr_bucket *search;
3024 u64 blkno, lower_blkno = 0;
3026 search = ocfs2_xattr_bucket_new(inode);
3033 ret = ocfs2_read_xattr_bucket(search, p_blkno);
3039 xh = bucket_xh(search);
3040 high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
3041 while (low_bucket <= high_bucket) {
3042 ocfs2_xattr_bucket_relse(search);
3044 bucket = (low_bucket + high_bucket) / 2;
3045 blkno = p_blkno + bucket * blk_per_bucket;
3046 ret = ocfs2_read_xattr_bucket(search, blkno);
3052 xh = bucket_xh(search);
3053 xe = &xh->xh_entries[0];
3054 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3055 high_bucket = bucket - 1;
3060 * Check whether the hash of the last entry in our
3061 * bucket is larger than the search one. for an empty
3062 * bucket, the last one is also the first one.
3065 xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3067 last_hash = le32_to_cpu(xe->xe_name_hash);
3069 /* record lower_blkno which may be the insert place. */
3070 lower_blkno = blkno;
3072 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3073 low_bucket = bucket + 1;
3077 /* the searched xattr should reside in this bucket if exists. */
3078 ret = ocfs2_find_xe_in_bucket(inode, search,
3079 name_index, name, name_hash,
3089 * Record the bucket we have found.
3090 * When the xattr's hash value is in the gap of 2 buckets, we will
3091 * always set it to the previous bucket.
3094 lower_blkno = p_blkno;
3096 /* This should be in cache - we just read it during the search */
3097 ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3103 xs->header = bucket_xh(xs->bucket);
3104 xs->base = bucket_block(xs->bucket, 0);
3105 xs->end = xs->base + inode->i_sb->s_blocksize;
3108 xs->here = &xs->header->xh_entries[index];
3109 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
3110 (unsigned long long)bucket_blkno(xs->bucket), index);
3115 ocfs2_xattr_bucket_free(search);
3119 static int ocfs2_xattr_index_block_find(struct inode *inode,
3120 struct buffer_head *root_bh,
3123 struct ocfs2_xattr_search *xs)
3126 struct ocfs2_xattr_block *xb =
3127 (struct ocfs2_xattr_block *)root_bh->b_data;
3128 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3129 struct ocfs2_extent_list *el = &xb_root->xt_list;
3131 u32 first_hash, num_clusters = 0;
3132 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3134 if (le16_to_cpu(el->l_next_free_rec) == 0)
3137 mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
3138 name, name_hash, name_index);
3140 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3147 BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3149 mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
3150 "in the rec is %u\n", num_clusters, (unsigned long long)p_blkno,
3153 ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3154 p_blkno, first_hash, num_clusters, xs);
3160 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3163 xattr_bucket_func *func,
3167 u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3168 u32 num_buckets = clusters * bpc;
3169 struct ocfs2_xattr_bucket *bucket;
3171 bucket = ocfs2_xattr_bucket_new(inode);
3173 mlog_errno(-ENOMEM);
3177 mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
3178 clusters, (unsigned long long)blkno);
3180 for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3181 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3188 * The real bucket num in this series of blocks is stored
3189 * in the 1st bucket.
3192 num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3194 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
3195 (unsigned long long)blkno,
3196 le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3198 ret = func(inode, bucket, para);
3199 if (ret && ret != -ERANGE)
3201 /* Fall through to bucket_relse() */
3204 ocfs2_xattr_bucket_relse(bucket);
3209 ocfs2_xattr_bucket_free(bucket);
3213 struct ocfs2_xattr_tree_list {
3219 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
3220 struct ocfs2_xattr_header *xh,
3227 if (index < 0 || index >= le16_to_cpu(xh->xh_count))
3230 name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
3232 *block_off = name_offset >> sb->s_blocksize_bits;
3233 *new_offset = name_offset % sb->s_blocksize;
3238 static int ocfs2_list_xattr_bucket(struct inode *inode,
3239 struct ocfs2_xattr_bucket *bucket,
3243 struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
3244 int i, block_off, new_offset;
3245 const char *prefix, *name;
3247 for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
3248 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
3249 type = ocfs2_xattr_get_type(entry);
3250 prefix = ocfs2_xattr_prefix(type);
3253 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3261 name = (const char *)bucket_block(bucket, block_off) +
3263 ret = ocfs2_xattr_list_entry(xl->buffer,
3267 entry->xe_name_len);
3276 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
3277 struct ocfs2_xattr_tree_root *xt,
3281 struct ocfs2_extent_list *el = &xt->xt_list;
3283 u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
3285 struct ocfs2_xattr_tree_list xl = {
3287 .buffer_size = buffer_size,
3291 if (le16_to_cpu(el->l_next_free_rec) == 0)
3294 while (name_hash > 0) {
3295 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
3296 &e_cpos, &num_clusters, el);
3302 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
3303 ocfs2_list_xattr_bucket,
3314 name_hash = e_cpos - 1;
3322 static int cmp_xe(const void *a, const void *b)
3324 const struct ocfs2_xattr_entry *l = a, *r = b;
3325 u32 l_hash = le32_to_cpu(l->xe_name_hash);
3326 u32 r_hash = le32_to_cpu(r->xe_name_hash);
3328 if (l_hash > r_hash)
3330 if (l_hash < r_hash)
3335 static void swap_xe(void *a, void *b, int size)
3337 struct ocfs2_xattr_entry *l = a, *r = b, tmp;
3340 memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
3341 memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
3345 * When the ocfs2_xattr_block is filled up, new bucket will be created
3346 * and all the xattr entries will be moved to the new bucket.
3347 * The header goes at the start of the bucket, and the names+values are
3348 * filled from the end. This is why *target starts as the last buffer.
3349 * Note: we need to sort the entries since they are not saved in order
3350 * in the ocfs2_xattr_block.
3352 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
3353 struct buffer_head *xb_bh,
3354 struct ocfs2_xattr_bucket *bucket)
3356 int i, blocksize = inode->i_sb->s_blocksize;
3357 int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3358 u16 offset, size, off_change;
3359 struct ocfs2_xattr_entry *xe;
3360 struct ocfs2_xattr_block *xb =
3361 (struct ocfs2_xattr_block *)xb_bh->b_data;
3362 struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
3363 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3364 u16 count = le16_to_cpu(xb_xh->xh_count);
3365 char *src = xb_bh->b_data;
3366 char *target = bucket_block(bucket, blks - 1);
3368 mlog(0, "cp xattr from block %llu to bucket %llu\n",
3369 (unsigned long long)xb_bh->b_blocknr,
3370 (unsigned long long)bucket_blkno(bucket));
3372 for (i = 0; i < blks; i++)
3373 memset(bucket_block(bucket, i), 0, blocksize);
3376 * Since the xe_name_offset is based on ocfs2_xattr_header,
3377 * there is a offset change corresponding to the change of
3378 * ocfs2_xattr_header's position.
3380 off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3381 xe = &xb_xh->xh_entries[count - 1];
3382 offset = le16_to_cpu(xe->xe_name_offset) + off_change;
3383 size = blocksize - offset;
3385 /* copy all the names and values. */
3386 memcpy(target + offset, src + offset, size);
3388 /* Init new header now. */
3389 xh->xh_count = xb_xh->xh_count;
3390 xh->xh_num_buckets = cpu_to_le16(1);
3391 xh->xh_name_value_len = cpu_to_le16(size);
3392 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
3394 /* copy all the entries. */
3395 target = bucket_block(bucket, 0);
3396 offset = offsetof(struct ocfs2_xattr_header, xh_entries);
3397 size = count * sizeof(struct ocfs2_xattr_entry);
3398 memcpy(target + offset, (char *)xb_xh + offset, size);
3400 /* Change the xe offset for all the xe because of the move. */
3401 off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
3402 offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3403 for (i = 0; i < count; i++)
3404 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
3406 mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
3407 offset, size, off_change);
3409 sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
3414 * After we move xattr from block to index btree, we have to
3415 * update ocfs2_xattr_search to the new xe and base.
3417 * When the entry is in xattr block, xattr_bh indicates the storage place.
3418 * While if the entry is in index b-tree, "bucket" indicates the
3419 * real place of the xattr.
3421 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
3422 struct ocfs2_xattr_search *xs,
3423 struct buffer_head *old_bh)
3425 char *buf = old_bh->b_data;
3426 struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
3427 struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
3430 xs->header = bucket_xh(xs->bucket);
3431 xs->base = bucket_block(xs->bucket, 0);
3432 xs->end = xs->base + inode->i_sb->s_blocksize;
3437 i = xs->here - old_xh->xh_entries;
3438 xs->here = &xs->header->xh_entries[i];
3441 static int ocfs2_xattr_create_index_block(struct inode *inode,
3442 struct ocfs2_xattr_search *xs,
3443 struct ocfs2_xattr_set_ctxt *ctxt)
3448 handle_t *handle = ctxt->handle;
3449 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3450 struct ocfs2_inode_info *oi = OCFS2_I(inode);
3451 struct buffer_head *xb_bh = xs->xattr_bh;
3452 struct ocfs2_xattr_block *xb =
3453 (struct ocfs2_xattr_block *)xb_bh->b_data;
3454 struct ocfs2_xattr_tree_root *xr;
3455 u16 xb_flags = le16_to_cpu(xb->xb_flags);
3457 mlog(0, "create xattr index block for %llu\n",
3458 (unsigned long long)xb_bh->b_blocknr);
3460 BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
3461 BUG_ON(!xs->bucket);
3465 * We can use this lock for now, and maybe move to a dedicated mutex
3466 * if performance becomes a problem later.
3468 down_write(&oi->ip_alloc_sem);
3470 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), xb_bh,
3471 OCFS2_JOURNAL_ACCESS_WRITE);
3477 ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac,
3478 1, 1, &bit_off, &len);
3485 * The bucket may spread in many blocks, and
3486 * we will only touch the 1st block and the last block
3487 * in the whole bucket(one for entry and one for data).
3489 blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
3491 mlog(0, "allocate 1 cluster from %llu to xattr block\n",
3492 (unsigned long long)blkno);
3494 ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
3500 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
3501 OCFS2_JOURNAL_ACCESS_CREATE);
3507 ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
3508 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
3510 ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
3512 /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
3513 memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
3514 offsetof(struct ocfs2_xattr_block, xb_attrs));
3516 xr = &xb->xb_attrs.xb_root;
3517 xr->xt_clusters = cpu_to_le32(1);
3518 xr->xt_last_eb_blk = 0;
3519 xr->xt_list.l_tree_depth = 0;
3520 xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
3521 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
3523 xr->xt_list.l_recs[0].e_cpos = 0;
3524 xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
3525 xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
3527 xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
3529 ocfs2_journal_dirty(handle, xb_bh);
3532 up_write(&oi->ip_alloc_sem);
3537 static int cmp_xe_offset(const void *a, const void *b)
3539 const struct ocfs2_xattr_entry *l = a, *r = b;
3540 u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
3541 u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
3543 if (l_name_offset < r_name_offset)
3545 if (l_name_offset > r_name_offset)
3551 * defrag a xattr bucket if we find that the bucket has some
3552 * holes beteen name/value pairs.
3553 * We will move all the name/value pairs to the end of the bucket
3554 * so that we can spare some space for insertion.
3556 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
3558 struct ocfs2_xattr_bucket *bucket)
3561 size_t end, offset, len, value_len;
3562 struct ocfs2_xattr_header *xh;
3563 char *entries, *buf, *bucket_buf = NULL;
3564 u64 blkno = bucket_blkno(bucket);
3566 size_t blocksize = inode->i_sb->s_blocksize;
3567 struct ocfs2_xattr_entry *xe;
3570 * In order to make the operation more efficient and generic,
3571 * we copy all the blocks into a contiguous memory and do the
3572 * defragment there, so if anything is error, we will not touch
3575 bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
3582 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3583 memcpy(buf, bucket_block(bucket, i), blocksize);
3585 ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
3586 OCFS2_JOURNAL_ACCESS_WRITE);
3592 xh = (struct ocfs2_xattr_header *)bucket_buf;
3593 entries = (char *)xh->xh_entries;
3594 xh_free_start = le16_to_cpu(xh->xh_free_start);
3596 mlog(0, "adjust xattr bucket in %llu, count = %u, "
3597 "xh_free_start = %u, xh_name_value_len = %u.\n",
3598 (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
3599 xh_free_start, le16_to_cpu(xh->xh_name_value_len));
3602 * sort all the entries by their offset.
3603 * the largest will be the first, so that we can
3604 * move them to the end one by one.
3606 sort(entries, le16_to_cpu(xh->xh_count),
3607 sizeof(struct ocfs2_xattr_entry),
3608 cmp_xe_offset, swap_xe);
3610 /* Move all name/values to the end of the bucket. */
3611 xe = xh->xh_entries;
3612 end = OCFS2_XATTR_BUCKET_SIZE;
3613 for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
3614 offset = le16_to_cpu(xe->xe_name_offset);
3615 if (ocfs2_xattr_is_local(xe))
3616 value_len = OCFS2_XATTR_SIZE(
3617 le64_to_cpu(xe->xe_value_size));
3619 value_len = OCFS2_XATTR_ROOT_SIZE;
3620 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
3623 * We must make sure that the name/value pair
3624 * exist in the same block. So adjust end to
3625 * the previous block end if needed.
3627 if (((end - len) / blocksize !=
3628 (end - 1) / blocksize))
3629 end = end - end % blocksize;
3631 if (end > offset + len) {
3632 memmove(bucket_buf + end - len,
3633 bucket_buf + offset, len);
3634 xe->xe_name_offset = cpu_to_le16(end - len);
3637 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
3638 "bucket %llu\n", (unsigned long long)blkno);
3643 mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
3644 "bucket %llu\n", (unsigned long long)blkno);
3646 if (xh_free_start == end)
3649 memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
3650 xh->xh_free_start = cpu_to_le16(end);
3652 /* sort the entries by their name_hash. */
3653 sort(entries, le16_to_cpu(xh->xh_count),
3654 sizeof(struct ocfs2_xattr_entry),
3658 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3659 memcpy(bucket_block(bucket, i), buf, blocksize);
3660 ocfs2_xattr_bucket_journal_dirty(handle, bucket);
3668 * prev_blkno points to the start of an existing extent. new_blkno
3669 * points to a newly allocated extent. Because we know each of our
3670 * clusters contains more than bucket, we can easily split one cluster
3671 * at a bucket boundary. So we take the last cluster of the existing
3672 * extent and split it down the middle. We move the last half of the
3673 * buckets in the last cluster of the existing extent over to the new
3676 * first_bh is the buffer at prev_blkno so we can update the existing
3677 * extent's bucket count. header_bh is the bucket were we were hoping
3678 * to insert our xattr. If the bucket move places the target in the new
3679 * extent, we'll update first_bh and header_bh after modifying the old
3682 * first_hash will be set as the 1st xe's name_hash in the new extent.
3684 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3686 struct ocfs2_xattr_bucket *first,
3687 struct ocfs2_xattr_bucket *target,
3693 struct super_block *sb = inode->i_sb;
3694 int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
3695 int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
3696 int to_move = num_buckets / 2;
3698 u64 last_cluster_blkno = bucket_blkno(first) +
3699 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
3701 BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
3702 BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
3704 mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3705 (unsigned long long)last_cluster_blkno, (unsigned long long)new_blkno);
3707 ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
3708 last_cluster_blkno, new_blkno,
3709 to_move, first_hash);
3715 /* This is the first bucket that got moved */
3716 src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
3719 * If the target bucket was part of the moved buckets, we need to
3720 * update first and target.
3722 if (bucket_blkno(target) >= src_blkno) {
3723 /* Find the block for the new target bucket */
3724 src_blkno = new_blkno +
3725 (bucket_blkno(target) - src_blkno);
3727 ocfs2_xattr_bucket_relse(first);
3728 ocfs2_xattr_bucket_relse(target);
3731 * These shouldn't fail - the buffers are in the
3732 * journal from ocfs2_cp_xattr_bucket().
3734 ret = ocfs2_read_xattr_bucket(first, new_blkno);
3739 ret = ocfs2_read_xattr_bucket(target, src_blkno);
3750 * Find the suitable pos when we divide a bucket into 2.
3751 * We have to make sure the xattrs with the same hash value exist
3752 * in the same bucket.
3754 * If this ocfs2_xattr_header covers more than one hash value, find a
3755 * place where the hash value changes. Try to find the most even split.
3756 * The most common case is that all entries have different hash values,
3757 * and the first check we make will find a place to split.
3759 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
3761 struct ocfs2_xattr_entry *entries = xh->xh_entries;
3762 int count = le16_to_cpu(xh->xh_count);
3763 int delta, middle = count / 2;
3766 * We start at the middle. Each step gets farther away in both
3767 * directions. We therefore hit the change in hash value
3768 * nearest to the middle. Note that this loop does not execute for
3771 for (delta = 0; delta < middle; delta++) {
3772 /* Let's check delta earlier than middle */
3773 if (cmp_xe(&entries[middle - delta - 1],
3774 &entries[middle - delta]))
3775 return middle - delta;
3777 /* For even counts, don't walk off the end */
3778 if ((middle + delta + 1) == count)
3781 /* Now try delta past middle */
3782 if (cmp_xe(&entries[middle + delta],
3783 &entries[middle + delta + 1]))
3784 return middle + delta + 1;
3787 /* Every entry had the same hash */
3792 * Move some xattrs in old bucket(blk) to new bucket(new_blk).
3793 * first_hash will record the 1st hash of the new bucket.
3795 * Normally half of the xattrs will be moved. But we have to make
3796 * sure that the xattrs with the same hash value are stored in the
3797 * same bucket. If all the xattrs in this bucket have the same hash
3798 * value, the new bucket will be initialized as an empty one and the
3799 * first_hash will be initialized as (hash_value+1).
3801 static int ocfs2_divide_xattr_bucket(struct inode *inode,
3806 int new_bucket_head)
3809 int count, start, len, name_value_len = 0, xe_len, name_offset = 0;
3810 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3811 struct ocfs2_xattr_header *xh;
3812 struct ocfs2_xattr_entry *xe;
3813 int blocksize = inode->i_sb->s_blocksize;
3815 mlog(0, "move some of xattrs from bucket %llu to %llu\n",
3816 (unsigned long long)blk, (unsigned long long)new_blk);
3818 s_bucket = ocfs2_xattr_bucket_new(inode);
3819 t_bucket = ocfs2_xattr_bucket_new(inode);
3820 if (!s_bucket || !t_bucket) {
3826 ret = ocfs2_read_xattr_bucket(s_bucket, blk);
3832 ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
3833 OCFS2_JOURNAL_ACCESS_WRITE);
3840 * Even if !new_bucket_head, we're overwriting t_bucket. Thus,
3841 * there's no need to read it.
3843 ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
3850 * Hey, if we're overwriting t_bucket, what difference does
3851 * ACCESS_CREATE vs ACCESS_WRITE make? See the comment in the
3852 * same part of ocfs2_cp_xattr_bucket().
3854 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3856 OCFS2_JOURNAL_ACCESS_CREATE :
3857 OCFS2_JOURNAL_ACCESS_WRITE);
3863 xh = bucket_xh(s_bucket);
3864 count = le16_to_cpu(xh->xh_count);
3865 start = ocfs2_xattr_find_divide_pos(xh);
3867 if (start == count) {
3868 xe = &xh->xh_entries[start-1];
3871 * initialized a new empty bucket here.
3872 * The hash value is set as one larger than
3873 * that of the last entry in the previous bucket.
3875 for (i = 0; i < t_bucket->bu_blocks; i++)
3876 memset(bucket_block(t_bucket, i), 0, blocksize);
3878 xh = bucket_xh(t_bucket);
3879 xh->xh_free_start = cpu_to_le16(blocksize);
3880 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
3881 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
3883 goto set_num_buckets;
3886 /* copy the whole bucket to the new first. */
3887 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
3889 /* update the new bucket. */
3890 xh = bucket_xh(t_bucket);
3893 * Calculate the total name/value len and xh_free_start for
3894 * the old bucket first.
3896 name_offset = OCFS2_XATTR_BUCKET_SIZE;
3898 for (i = 0; i < start; i++) {
3899 xe = &xh->xh_entries[i];
3900 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3901 if (ocfs2_xattr_is_local(xe))
3903 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3905 xe_len += OCFS2_XATTR_ROOT_SIZE;
3906 name_value_len += xe_len;
3907 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3908 name_offset = le16_to_cpu(xe->xe_name_offset);
3912 * Now begin the modification to the new bucket.
3914 * In the new bucket, We just move the xattr entry to the beginning
3915 * and don't touch the name/value. So there will be some holes in the
3916 * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3919 xe = &xh->xh_entries[start];
3920 len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3921 mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3922 (int)((char *)xe - (char *)xh),
3923 (int)((char *)xh->xh_entries - (char *)xh));
3924 memmove((char *)xh->xh_entries, (char *)xe, len);
3925 xe = &xh->xh_entries[count - start];
3926 len = sizeof(struct ocfs2_xattr_entry) * start;
3927 memset((char *)xe, 0, len);
3929 le16_add_cpu(&xh->xh_count, -start);
3930 le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3932 /* Calculate xh_free_start for the new bucket. */
3933 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3934 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3935 xe = &xh->xh_entries[i];
3936 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3937 if (ocfs2_xattr_is_local(xe))
3939 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3941 xe_len += OCFS2_XATTR_ROOT_SIZE;
3942 if (le16_to_cpu(xe->xe_name_offset) <
3943 le16_to_cpu(xh->xh_free_start))
3944 xh->xh_free_start = xe->xe_name_offset;
3948 /* set xh->xh_num_buckets for the new xh. */
3949 if (new_bucket_head)
3950 xh->xh_num_buckets = cpu_to_le16(1);
3952 xh->xh_num_buckets = 0;
3954 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
3956 /* store the first_hash of the new bucket. */
3958 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3961 * Now only update the 1st block of the old bucket. If we
3962 * just added a new empty bucket, there is no need to modify
3968 xh = bucket_xh(s_bucket);
3969 memset(&xh->xh_entries[start], 0,
3970 sizeof(struct ocfs2_xattr_entry) * (count - start));
3971 xh->xh_count = cpu_to_le16(start);
3972 xh->xh_free_start = cpu_to_le16(name_offset);
3973 xh->xh_name_value_len = cpu_to_le16(name_value_len);
3975 ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
3978 ocfs2_xattr_bucket_free(s_bucket);
3979 ocfs2_xattr_bucket_free(t_bucket);
3985 * Copy xattr from one bucket to another bucket.
3987 * The caller must make sure that the journal transaction
3988 * has enough space for journaling.
3990 static int ocfs2_cp_xattr_bucket(struct inode *inode,
3997 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3999 BUG_ON(s_blkno == t_blkno);
4001 mlog(0, "cp bucket %llu to %llu, target is %d\n",
4002 (unsigned long long)s_blkno, (unsigned long long)t_blkno,
4005 s_bucket = ocfs2_xattr_bucket_new(inode);
4006 t_bucket = ocfs2_xattr_bucket_new(inode);
4007 if (!s_bucket || !t_bucket) {
4013 ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
4018 * Even if !t_is_new, we're overwriting t_bucket. Thus,
4019 * there's no need to read it.
4021 ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
4026 * Hey, if we're overwriting t_bucket, what difference does
4027 * ACCESS_CREATE vs ACCESS_WRITE make? Well, if we allocated a new
4028 * cluster to fill, we came here from
4029 * ocfs2_mv_xattr_buckets(), and it is really new -
4030 * ACCESS_CREATE is required. But we also might have moved data
4031 * out of t_bucket before extending back into it.
4032 * ocfs2_add_new_xattr_bucket() can do this - its call to
4033 * ocfs2_add_new_xattr_cluster() may have created a new extent
4034 * and copied out the end of the old extent. Then it re-extends
4035 * the old extent back to create space for new xattrs. That's
4036 * how we get here, and the bucket isn't really new.
4038 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4040 OCFS2_JOURNAL_ACCESS_CREATE :
4041 OCFS2_JOURNAL_ACCESS_WRITE);
4045 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4046 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4049 ocfs2_xattr_bucket_free(t_bucket);
4050 ocfs2_xattr_bucket_free(s_bucket);
4056 * src_blk points to the start of an existing extent. last_blk points to
4057 * last cluster in that extent. to_blk points to a newly allocated
4058 * extent. We copy the buckets from the cluster at last_blk to the new
4059 * extent. If start_bucket is non-zero, we skip that many buckets before
4060 * we start copying. The new extent's xh_num_buckets gets set to the
4061 * number of buckets we copied. The old extent's xh_num_buckets shrinks
4062 * by the same amount.
4064 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4065 u64 src_blk, u64 last_blk, u64 to_blk,
4066 unsigned int start_bucket,
4069 int i, ret, credits;
4070 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4071 int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4072 int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4073 struct ocfs2_xattr_bucket *old_first, *new_first;
4075 mlog(0, "mv xattrs from cluster %llu to %llu\n",
4076 (unsigned long long)last_blk, (unsigned long long)to_blk);
4078 BUG_ON(start_bucket >= num_buckets);
4080 num_buckets -= start_bucket;
4081 last_blk += (start_bucket * blks_per_bucket);
4084 /* The first bucket of the original extent */
4085 old_first = ocfs2_xattr_bucket_new(inode);
4086 /* The first bucket of the new extent */
4087 new_first = ocfs2_xattr_bucket_new(inode);
4088 if (!old_first || !new_first) {
4094 ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4101 * We need to update the first bucket of the old extent and all
4102 * the buckets going to the new extent.
4104 credits = ((num_buckets + 1) * blks_per_bucket) +
4105 handle->h_buffer_credits;
4106 ret = ocfs2_extend_trans(handle, credits);
4112 ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4113 OCFS2_JOURNAL_ACCESS_WRITE);
4119 for (i = 0; i < num_buckets; i++) {
4120 ret = ocfs2_cp_xattr_bucket(inode, handle,
4121 last_blk + (i * blks_per_bucket),
4122 to_blk + (i * blks_per_bucket),
4131 * Get the new bucket ready before we dirty anything
4132 * (This actually shouldn't fail, because we already dirtied
4133 * it once in ocfs2_cp_xattr_bucket()).
4135 ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4140 ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4141 OCFS2_JOURNAL_ACCESS_WRITE);
4147 /* Now update the headers */
4148 le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4149 ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4151 bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4152 ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4155 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4158 ocfs2_xattr_bucket_free(new_first);
4159 ocfs2_xattr_bucket_free(old_first);
4164 * Move some xattrs in this cluster to the new cluster.
4165 * This function should only be called when bucket size == cluster size.
4166 * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4168 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4174 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4175 int ret, credits = 2 * blk_per_bucket + handle->h_buffer_credits;
4177 BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4179 ret = ocfs2_extend_trans(handle, credits);
4185 /* Move half of the xattr in start_blk to the next bucket. */
4186 return ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4187 new_blk, first_hash, 1);
4191 * Move some xattrs from the old cluster to the new one since they are not
4192 * contiguous in ocfs2 xattr tree.
4194 * new_blk starts a new separate cluster, and we will move some xattrs from
4195 * prev_blk to it. v_start will be set as the first name hash value in this
4196 * new cluster so that it can be used as e_cpos during tree insertion and
4197 * don't collide with our original b-tree operations. first_bh and header_bh
4198 * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4199 * to extend the insert bucket.
4201 * The problem is how much xattr should we move to the new one and when should
4202 * we update first_bh and header_bh?
4203 * 1. If cluster size > bucket size, that means the previous cluster has more
4204 * than 1 bucket, so just move half nums of bucket into the new cluster and
4205 * update the first_bh and header_bh if the insert bucket has been moved
4206 * to the new cluster.
4207 * 2. If cluster_size == bucket_size:
4208 * a) If the previous extent rec has more than one cluster and the insert
4209 * place isn't in the last cluster, copy the entire last cluster to the
4210 * new one. This time, we don't need to upate the first_bh and header_bh
4211 * since they will not be moved into the new cluster.
4212 * b) Otherwise, move the bottom half of the xattrs in the last cluster into
4213 * the new one. And we set the extend flag to zero if the insert place is
4214 * moved into the new allocated cluster since no extend is needed.
4216 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
4218 struct ocfs2_xattr_bucket *first,
4219 struct ocfs2_xattr_bucket *target,
4227 mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
4228 (unsigned long long)bucket_blkno(first), prev_clusters,
4229 (unsigned long long)new_blk);
4231 if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
4232 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
4241 /* The start of the last cluster in the first extent */
4242 u64 last_blk = bucket_blkno(first) +
4243 ((prev_clusters - 1) *
4244 ocfs2_clusters_to_blocks(inode->i_sb, 1));
4246 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
4247 ret = ocfs2_mv_xattr_buckets(inode, handle,
4248 bucket_blkno(first),
4249 last_blk, new_blk, 0,
4254 ret = ocfs2_divide_xattr_cluster(inode, handle,
4260 if ((bucket_blkno(target) == last_blk) && extend)
4269 * Add a new cluster for xattr storage.
4271 * If the new cluster is contiguous with the previous one, it will be
4272 * appended to the same extent record, and num_clusters will be updated.
4273 * If not, we will insert a new extent for it and move some xattrs in
4274 * the last cluster into the new allocated one.
4275 * We also need to limit the maximum size of a btree leaf, otherwise we'll
4276 * lose the benefits of hashing because we'll have to search large leaves.
4277 * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
4280 * first_bh is the first block of the previous extent rec and header_bh
4281 * indicates the bucket we will insert the new xattrs. They will be updated
4282 * when the header_bh is moved into the new cluster.
4284 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
4285 struct buffer_head *root_bh,
4286 struct ocfs2_xattr_bucket *first,
4287 struct ocfs2_xattr_bucket *target,
4291 struct ocfs2_xattr_set_ctxt *ctxt)
4294 u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
4295 u32 prev_clusters = *num_clusters;
4296 u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
4298 handle_t *handle = ctxt->handle;
4299 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4300 struct ocfs2_extent_tree et;
4302 mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
4303 "previous xattr blkno = %llu\n",
4304 (unsigned long long)OCFS2_I(inode)->ip_blkno,
4305 prev_cpos, (unsigned long long)bucket_blkno(first));
4307 ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
4309 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
4310 OCFS2_JOURNAL_ACCESS_WRITE);
4316 ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac, 1,
4317 clusters_to_add, &bit_off, &num_bits);
4324 BUG_ON(num_bits > clusters_to_add);
4326 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
4327 mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
4328 num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
4330 if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
4331 (prev_clusters + num_bits) << osb->s_clustersize_bits <=
4332 OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
4334 * If this cluster is contiguous with the old one and
4335 * adding this new cluster, we don't surpass the limit of
4336 * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
4337 * initialized and used like other buckets in the previous
4339 * So add it as a contiguous one. The caller will handle
4342 v_start = prev_cpos + prev_clusters;
4343 *num_clusters = prev_clusters + num_bits;
4344 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
4347 ret = ocfs2_adjust_xattr_cross_cluster(inode,
4361 mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
4362 num_bits, (unsigned long long)block, v_start);
4363 ret = ocfs2_insert_extent(handle, &et, v_start, block,
4364 num_bits, 0, ctxt->meta_ac);
4370 ret = ocfs2_journal_dirty(handle, root_bh);
4379 * We are given an extent. 'first' is the bucket at the very front of
4380 * the extent. The extent has space for an additional bucket past
4381 * bucket_xh(first)->xh_num_buckets. 'target_blkno' is the block number
4382 * of the target bucket. We wish to shift every bucket past the target
4383 * down one, filling in that additional space. When we get back to the
4384 * target, we split the target between itself and the now-empty bucket
4385 * at target+1 (aka, target_blkno + blks_per_bucket).
4387 static int ocfs2_extend_xattr_bucket(struct inode *inode,
4389 struct ocfs2_xattr_bucket *first,
4394 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4395 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4397 u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
4399 mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
4400 "from %llu, len = %u\n", (unsigned long long)target_blk,
4401 (unsigned long long)bucket_blkno(first), num_clusters);
4403 /* The extent must have room for an additional bucket */
4404 BUG_ON(new_bucket >=
4405 (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
4407 /* end_blk points to the last existing bucket */
4408 end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
4411 * end_blk is the start of the last existing bucket.
4412 * Thus, (end_blk - target_blk) covers the target bucket and
4413 * every bucket after it up to, but not including, the last
4414 * existing bucket. Then we add the last existing bucket, the
4415 * new bucket, and the first bucket (3 * blk_per_bucket).
4417 credits = (end_blk - target_blk) + (3 * blk_per_bucket) +
4418 handle->h_buffer_credits;
4419 ret = ocfs2_extend_trans(handle, credits);
4425 ret = ocfs2_xattr_bucket_journal_access(handle, first,
4426 OCFS2_JOURNAL_ACCESS_WRITE);
4432 while (end_blk != target_blk) {
4433 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
4434 end_blk + blk_per_bucket, 0);
4437 end_blk -= blk_per_bucket;
4440 /* Move half of the xattr in target_blkno to the next bucket. */
4441 ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
4442 target_blk + blk_per_bucket, NULL, 0);
4444 le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
4445 ocfs2_xattr_bucket_journal_dirty(handle, first);
4452 * Add new xattr bucket in an extent record and adjust the buckets
4453 * accordingly. xb_bh is the ocfs2_xattr_block, and target is the
4454 * bucket we want to insert into.
4456 * In the easy case, we will move all the buckets after target down by
4457 * one. Half of target's xattrs will be moved to the next bucket.
4459 * If current cluster is full, we'll allocate a new one. This may not
4460 * be contiguous. The underlying calls will make sure that there is
4461 * space for the insert, shifting buckets around if necessary.
4462 * 'target' may be moved by those calls.
4464 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
4465 struct buffer_head *xb_bh,
4466 struct ocfs2_xattr_bucket *target,
4467 struct ocfs2_xattr_set_ctxt *ctxt)
4469 struct ocfs2_xattr_block *xb =
4470 (struct ocfs2_xattr_block *)xb_bh->b_data;
4471 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
4472 struct ocfs2_extent_list *el = &xb_root->xt_list;
4474 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
4475 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4476 int ret, num_buckets, extend = 1;
4478 u32 e_cpos, num_clusters;
4479 /* The bucket at the front of the extent */
4480 struct ocfs2_xattr_bucket *first;
4482 mlog(0, "Add new xattr bucket starting from %llu\n",
4483 (unsigned long long)bucket_blkno(target));
4485 /* The first bucket of the original extent */
4486 first = ocfs2_xattr_bucket_new(inode);
4493 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
4500 ret = ocfs2_read_xattr_bucket(first, p_blkno);
4506 num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
4507 if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
4509 * This can move first+target if the target bucket moves
4510 * to the new extent.
4512 ret = ocfs2_add_new_xattr_cluster(inode,
4527 ret = ocfs2_extend_xattr_bucket(inode,
4530 bucket_blkno(target),
4537 ocfs2_xattr_bucket_free(first);
4542 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
4543 struct ocfs2_xattr_bucket *bucket,
4546 int block_off = offs >> inode->i_sb->s_blocksize_bits;
4548 offs = offs % inode->i_sb->s_blocksize;
4549 return bucket_block(bucket, block_off) + offs;
4553 * Handle the normal xattr set, including replace, delete and new.
4555 * Note: "local" indicates the real data's locality. So we can't
4556 * just its bucket locality by its length.
4558 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
4559 struct ocfs2_xattr_info *xi,
4560 struct ocfs2_xattr_search *xs,
4564 struct ocfs2_xattr_entry *last, *xe;
4565 int name_len = strlen(xi->name);
4566 struct ocfs2_xattr_header *xh = xs->header;
4567 u16 count = le16_to_cpu(xh->xh_count), start;
4568 size_t blocksize = inode->i_sb->s_blocksize;
4570 size_t offs, size, new_size;
4572 last = &xh->xh_entries[count];
4573 if (!xs->not_found) {
4575 offs = le16_to_cpu(xe->xe_name_offset);
4576 if (ocfs2_xattr_is_local(xe))
4577 size = OCFS2_XATTR_SIZE(name_len) +
4578 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4580 size = OCFS2_XATTR_SIZE(name_len) +
4581 OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4584 * If the new value will be stored outside, xi->value has been
4585 * initalized as an empty ocfs2_xattr_value_root, and the same
4586 * goes with xi->value_len, so we can set new_size safely here.
4587 * See ocfs2_xattr_set_in_bucket.
4589 new_size = OCFS2_XATTR_SIZE(name_len) +
4590 OCFS2_XATTR_SIZE(xi->value_len);
4592 le16_add_cpu(&xh->xh_name_value_len, -size);
4594 if (new_size > size)
4595 goto set_new_name_value;
4597 /* Now replace the old value with new one. */
4599 xe->xe_value_size = cpu_to_le64(xi->value_len);
4601 xe->xe_value_size = 0;
4603 val = ocfs2_xattr_bucket_get_val(inode,
4605 memset(val + OCFS2_XATTR_SIZE(name_len), 0,
4606 size - OCFS2_XATTR_SIZE(name_len));
4607 if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
4608 memcpy(val + OCFS2_XATTR_SIZE(name_len),
4609 xi->value, xi->value_len);
4611 le16_add_cpu(&xh->xh_name_value_len, new_size);
4612 ocfs2_xattr_set_local(xe, local);
4616 * Remove the old entry if there is more than one.
4617 * We don't remove the last entry so that we can
4618 * use it to indicate the hash value of the empty
4622 le16_add_cpu(&xh->xh_count, -1);
4625 (void *)last - (void *)xe);
4627 sizeof(struct ocfs2_xattr_entry));
4630 cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4635 /* find a new entry for insert. */
4636 int low = 0, high = count - 1, tmp;
4637 struct ocfs2_xattr_entry *tmp_xe;
4639 while (low <= high && count) {
4640 tmp = (low + high) / 2;
4641 tmp_xe = &xh->xh_entries[tmp];
4643 if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
4645 else if (name_hash <
4646 le32_to_cpu(tmp_xe->xe_name_hash))
4654 xe = &xh->xh_entries[low];
4656 memmove(xe + 1, xe, (void *)last - (void *)xe);
4658 le16_add_cpu(&xh->xh_count, 1);
4659 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
4660 xe->xe_name_hash = cpu_to_le32(name_hash);
4661 xe->xe_name_len = name_len;
4662 ocfs2_xattr_set_type(xe, xi->name_index);
4666 /* Insert the new name+value. */
4667 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4670 * We must make sure that the name/value pair
4671 * exists in the same block.
4673 offs = le16_to_cpu(xh->xh_free_start);
4674 start = offs - size;
4676 if (start >> inode->i_sb->s_blocksize_bits !=
4677 (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4678 offs = offs - offs % blocksize;
4679 xh->xh_free_start = cpu_to_le16(offs);
4682 val = ocfs2_xattr_bucket_get_val(inode, xs->bucket, offs - size);
4683 xe->xe_name_offset = cpu_to_le16(offs - size);
4685 memset(val, 0, size);
4686 memcpy(val, xi->name, name_len);
4687 memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4689 xe->xe_value_size = cpu_to_le64(xi->value_len);
4690 ocfs2_xattr_set_local(xe, local);
4692 le16_add_cpu(&xh->xh_free_start, -size);
4693 le16_add_cpu(&xh->xh_name_value_len, size);
4699 * Set the xattr entry in the specified bucket.
4700 * The bucket is indicated by xs->bucket and it should have the enough
4701 * space for the xattr insertion.
4703 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4705 struct ocfs2_xattr_info *xi,
4706 struct ocfs2_xattr_search *xs,
4713 mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4714 (unsigned long)xi->value_len, xi->name_index,
4715 (unsigned long long)bucket_blkno(xs->bucket));
4717 if (!xs->bucket->bu_bhs[1]) {
4718 blkno = bucket_blkno(xs->bucket);
4719 ocfs2_xattr_bucket_relse(xs->bucket);
4720 ret = ocfs2_read_xattr_bucket(xs->bucket, blkno);
4727 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4728 OCFS2_JOURNAL_ACCESS_WRITE);
4734 ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local);
4735 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4742 * Truncate the specified xe_off entry in xattr bucket.
4743 * bucket is indicated by header_bh and len is the new length.
4744 * Both the ocfs2_xattr_value_root and the entry will be updated here.
4746 * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4748 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4749 struct ocfs2_xattr_bucket *bucket,
4752 struct ocfs2_xattr_set_ctxt *ctxt)
4756 struct ocfs2_xattr_entry *xe;
4757 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4758 size_t blocksize = inode->i_sb->s_blocksize;
4759 struct ocfs2_xattr_value_buf vb = {
4760 .vb_access = ocfs2_journal_access,
4763 xe = &xh->xh_entries[xe_off];
4765 BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4767 offset = le16_to_cpu(xe->xe_name_offset) +
4768 OCFS2_XATTR_SIZE(xe->xe_name_len);
4770 value_blk = offset / blocksize;
4772 /* We don't allow ocfs2_xattr_value to be stored in different block. */
4773 BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4775 vb.vb_bh = bucket->bu_bhs[value_blk];
4778 vb.vb_xv = (struct ocfs2_xattr_value_root *)
4779 (vb.vb_bh->b_data + offset % blocksize);
4782 * From here on out we have to dirty the bucket. The generic
4783 * value calls only modify one of the bucket's bhs, but we need
4784 * to send the bucket at once. So if they error, they *could* have
4785 * modified something. We have to assume they did, and dirty
4786 * the whole bucket. This leaves us in a consistent state.
4788 mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4789 xe_off, (unsigned long long)bucket_blkno(bucket), len);
4790 ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
4796 ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
4797 OCFS2_JOURNAL_ACCESS_WRITE);
4803 xe->xe_value_size = cpu_to_le64(len);
4805 ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
4811 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4812 struct ocfs2_xattr_search *xs,
4814 struct ocfs2_xattr_set_ctxt *ctxt)
4817 struct ocfs2_xattr_entry *xe = xs->here;
4818 struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4820 BUG_ON(!xs->bucket->bu_bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4822 offset = xe - xh->xh_entries;
4823 ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket,
4831 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4833 struct ocfs2_xattr_search *xs,
4837 int ret, offset, block_off;
4838 struct ocfs2_xattr_value_root *xv;
4839 struct ocfs2_xattr_entry *xe = xs->here;
4840 struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
4842 struct ocfs2_xattr_value_buf vb = {
4843 .vb_access = ocfs2_journal_access,
4846 BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4848 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb, xh,
4849 xe - xh->xh_entries,
4857 base = bucket_block(xs->bucket, block_off);
4858 xv = (struct ocfs2_xattr_value_root *)(base + offset +
4859 OCFS2_XATTR_SIZE(xe->xe_name_len));
4862 vb.vb_bh = xs->bucket->bu_bhs[block_off];
4863 ret = __ocfs2_xattr_set_value_outside(inode, handle,
4864 &vb, val, value_len);
4871 static int ocfs2_rm_xattr_cluster(struct inode *inode,
4872 struct buffer_head *root_bh,
4878 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4879 struct inode *tl_inode = osb->osb_tl_inode;
4881 struct ocfs2_xattr_block *xb =
4882 (struct ocfs2_xattr_block *)root_bh->b_data;
4883 struct ocfs2_alloc_context *meta_ac = NULL;
4884 struct ocfs2_cached_dealloc_ctxt dealloc;
4885 struct ocfs2_extent_tree et;
4887 ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
4889 ocfs2_init_dealloc_ctxt(&dealloc);
4891 mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4892 cpos, len, (unsigned long long)blkno);
4894 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode), blkno,
4897 ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
4903 mutex_lock(&tl_inode->i_mutex);
4905 if (ocfs2_truncate_log_needs_flush(osb)) {
4906 ret = __ocfs2_flush_truncate_log(osb);
4913 handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
4914 if (IS_ERR(handle)) {
4920 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
4921 OCFS2_JOURNAL_ACCESS_WRITE);
4927 ret = ocfs2_remove_extent(handle, &et, cpos, len, meta_ac,
4934 le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4936 ret = ocfs2_journal_dirty(handle, root_bh);
4942 ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4947 ocfs2_commit_trans(osb, handle);
4949 ocfs2_schedule_truncate_log_flush(osb, 1);
4951 mutex_unlock(&tl_inode->i_mutex);
4954 ocfs2_free_alloc_context(meta_ac);
4956 ocfs2_run_deallocs(osb, &dealloc);
4961 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4963 struct ocfs2_xattr_search *xs)
4965 struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
4966 struct ocfs2_xattr_entry *last = &xh->xh_entries[
4967 le16_to_cpu(xh->xh_count) - 1];
4970 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4971 OCFS2_JOURNAL_ACCESS_WRITE);
4977 /* Remove the old entry. */
4978 memmove(xs->here, xs->here + 1,
4979 (void *)last - (void *)xs->here);
4980 memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4981 le16_add_cpu(&xh->xh_count, -1);
4983 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4987 * Set the xattr name/value in the bucket specified in xs.
4989 * As the new value in xi may be stored in the bucket or in an outside cluster,
4990 * we divide the whole process into 3 steps:
4991 * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4992 * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4993 * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4994 * 4. If the clusters for the new outside value can't be allocated, we need
4995 * to free the xattr we allocated in set.
4997 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4998 struct ocfs2_xattr_info *xi,
4999 struct ocfs2_xattr_search *xs,
5000 struct ocfs2_xattr_set_ctxt *ctxt)
5004 char *val = (char *)xi->value;
5005 struct ocfs2_xattr_entry *xe = xs->here;
5006 u32 name_hash = ocfs2_xattr_name_hash(inode, xi->name,
5009 if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
5011 * We need to truncate the xattr storage first.
5013 * If both the old and new value are stored to
5014 * outside block, we only need to truncate
5015 * the storage and then set the value outside.
5017 * If the new value should be stored within block,
5018 * we should free all the outside block first and
5019 * the modification to the xattr block will be done
5020 * by following steps.
5022 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
5023 value_len = xi->value_len;
5027 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
5034 goto set_value_outside;
5037 value_len = xi->value_len;
5038 /* So we have to handle the inside block change now. */
5039 if (value_len > OCFS2_XATTR_INLINE_SIZE) {
5041 * If the new value will be stored outside of block,
5042 * initalize a new empty value root and insert it first.
5045 xi->value = &def_xv;
5046 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
5049 ret = ocfs2_xattr_set_entry_in_bucket(inode, ctxt->handle, xi, xs,
5056 if (value_len <= OCFS2_XATTR_INLINE_SIZE)
5059 /* allocate the space now for the outside block storage. */
5060 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
5065 if (xs->not_found) {
5067 * We can't allocate enough clusters for outside
5068 * storage and we have allocated xattr already,
5069 * so need to remove it.
5071 ocfs2_xattr_bucket_remove_xs(inode, ctxt->handle, xs);
5077 ret = ocfs2_xattr_bucket_set_value_outside(inode, ctxt->handle,
5078 xs, val, value_len);
5084 * check whether the xattr bucket is filled up with the same hash value.
5085 * If we want to insert the xattr with the same hash, return -ENOSPC.
5086 * If we want to insert a xattr with different hash value, go ahead
5087 * and ocfs2_divide_xattr_bucket will handle this.
5089 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
5090 struct ocfs2_xattr_bucket *bucket,
5093 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5094 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5096 if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5099 if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5100 xh->xh_entries[0].xe_name_hash) {
5101 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5103 (unsigned long long)bucket_blkno(bucket),
5104 le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5111 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5112 struct ocfs2_xattr_info *xi,
5113 struct ocfs2_xattr_search *xs,
5114 struct ocfs2_xattr_set_ctxt *ctxt)
5116 struct ocfs2_xattr_header *xh;
5117 struct ocfs2_xattr_entry *xe;
5118 u16 count, header_size, xh_free_start;
5119 int free, max_free, need, old;
5120 size_t value_size = 0, name_len = strlen(xi->name);
5121 size_t blocksize = inode->i_sb->s_blocksize;
5122 int ret, allocation = 0;
5124 mlog_entry("Set xattr %s in xattr index block\n", xi->name);
5128 count = le16_to_cpu(xh->xh_count);
5129 xh_free_start = le16_to_cpu(xh->xh_free_start);
5130 header_size = sizeof(struct ocfs2_xattr_header) +
5131 count * sizeof(struct ocfs2_xattr_entry);
5132 max_free = OCFS2_XATTR_BUCKET_SIZE - header_size -
5133 le16_to_cpu(xh->xh_name_value_len) - OCFS2_XATTR_HEADER_GAP;
5135 mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
5136 "of %u which exceed block size\n",
5137 (unsigned long long)bucket_blkno(xs->bucket),
5140 if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
5141 value_size = OCFS2_XATTR_ROOT_SIZE;
5143 value_size = OCFS2_XATTR_SIZE(xi->value_len);
5146 need = sizeof(struct ocfs2_xattr_entry) +
5147 OCFS2_XATTR_SIZE(name_len) + value_size;
5149 need = value_size + OCFS2_XATTR_SIZE(name_len);
5152 * We only replace the old value if the new length is smaller
5153 * than the old one. Otherwise we will allocate new space in the
5154 * bucket to store it.
5157 if (ocfs2_xattr_is_local(xe))
5158 old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
5160 old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
5162 if (old >= value_size)
5166 free = xh_free_start - header_size - OCFS2_XATTR_HEADER_GAP;
5168 * We need to make sure the new name/value pair
5169 * can exist in the same block.
5171 if (xh_free_start % blocksize < need)
5172 free -= xh_free_start % blocksize;
5174 mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
5175 "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
5176 " %u\n", xs->not_found,
5177 (unsigned long long)bucket_blkno(xs->bucket),
5178 free, need, max_free, le16_to_cpu(xh->xh_free_start),
5179 le16_to_cpu(xh->xh_name_value_len));
5183 count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb))) {
5184 if (need <= max_free &&
5185 count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
5187 * We can create the space by defragment. Since only the
5188 * name/value will be moved, the xe shouldn't be changed
5191 ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5198 xh_free_start = le16_to_cpu(xh->xh_free_start);
5199 free = xh_free_start - header_size
5200 - OCFS2_XATTR_HEADER_GAP;
5201 if (xh_free_start % blocksize < need)
5202 free -= xh_free_start % blocksize;
5207 mlog(0, "Can't get enough space for xattr insert by "
5208 "defragment. Need %u bytes, but we have %d, so "
5209 "allocate new bucket for it.\n", need, free);
5213 * We have to add new buckets or clusters and one
5214 * allocation should leave us enough space for insert.
5219 * We do not allow for overlapping ranges between buckets. And
5220 * the maximum number of collisions we will allow for then is
5221 * one bucket's worth, so check it here whether we need to
5222 * add a new bucket for the insert.
5224 ret = ocfs2_check_xattr_bucket_collision(inode,
5232 ret = ocfs2_add_new_xattr_bucket(inode,
5242 * ocfs2_add_new_xattr_bucket() will have updated
5243 * xs->bucket if it moved, but it will not have updated
5244 * any of the other search fields. Thus, we drop it and
5245 * re-search. Everything should be cached, so it'll be
5248 ocfs2_xattr_bucket_relse(xs->bucket);
5249 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5252 if (ret && ret != -ENODATA)
5254 xs->not_found = ret;
5260 ret = ocfs2_xattr_set_in_bucket(inode, xi, xs, ctxt);
5266 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5267 struct ocfs2_xattr_bucket *bucket,
5271 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5273 struct ocfs2_xattr_entry *xe;
5274 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5275 struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5276 int credits = ocfs2_remove_extent_credits(osb->sb) +
5277 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5280 ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5282 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5283 xe = &xh->xh_entries[i];
5284 if (ocfs2_xattr_is_local(xe))
5287 ctxt.handle = ocfs2_start_trans(osb, credits);
5288 if (IS_ERR(ctxt.handle)) {
5289 ret = PTR_ERR(ctxt.handle);
5294 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5297 ocfs2_commit_trans(osb, ctxt.handle);
5304 ocfs2_schedule_truncate_log_flush(osb, 1);
5305 ocfs2_run_deallocs(osb, &ctxt.dealloc);
5309 static int ocfs2_delete_xattr_index_block(struct inode *inode,
5310 struct buffer_head *xb_bh)
5312 struct ocfs2_xattr_block *xb =
5313 (struct ocfs2_xattr_block *)xb_bh->b_data;
5314 struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
5316 u32 name_hash = UINT_MAX, e_cpos, num_clusters;
5319 if (le16_to_cpu(el->l_next_free_rec) == 0)
5322 while (name_hash > 0) {
5323 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
5324 &e_cpos, &num_clusters, el);
5330 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
5331 ocfs2_delete_xattr_in_bucket,
5338 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
5339 p_blkno, e_cpos, num_clusters);
5348 name_hash = e_cpos - 1;
5356 * Whenever we modify a xattr value root in the bucket(e.g, CoW
5357 * or change the extent record flag), we need to recalculate
5358 * the metaecc for the whole bucket. So it is done here.
5361 * We have to give the extra credits for the caller.
5363 static int ocfs2_xattr_bucket_post_refcount(struct inode *inode,
5368 struct ocfs2_xattr_bucket *bucket =
5369 (struct ocfs2_xattr_bucket *)para;
5371 ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
5372 OCFS2_JOURNAL_ACCESS_WRITE);
5378 ocfs2_xattr_bucket_journal_dirty(handle, bucket);
5384 * Special action we need if the xattr value is refcounted.
5386 * 1. If the xattr is refcounted, lock the tree.
5387 * 2. CoW the xattr if we are setting the new value and the value
5388 * will be stored outside.
5389 * 3. In other case, decrease_refcount will work for us, so just
5390 * lock the refcount tree, calculate the meta and credits is OK.
5392 * We have to do CoW before ocfs2_init_xattr_set_ctxt since
5393 * currently CoW is a completed transaction, while this function
5394 * will also lock the allocators and let us deadlock. So we will
5395 * CoW the whole xattr value.
5397 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
5398 struct ocfs2_dinode *di,
5399 struct ocfs2_xattr_info *xi,
5400 struct ocfs2_xattr_search *xis,
5401 struct ocfs2_xattr_search *xbs,
5402 struct ocfs2_refcount_tree **ref_tree,
5407 struct ocfs2_xattr_block *xb;
5408 struct ocfs2_xattr_entry *xe;
5410 u32 p_cluster, num_clusters;
5411 unsigned int ext_flags;
5412 int name_offset, name_len;
5413 struct ocfs2_xattr_value_buf vb;
5414 struct ocfs2_xattr_bucket *bucket = NULL;
5415 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5416 struct ocfs2_post_refcount refcount;
5417 struct ocfs2_post_refcount *p = NULL;
5418 struct buffer_head *ref_root_bh = NULL;
5420 if (!xis->not_found) {
5422 name_offset = le16_to_cpu(xe->xe_name_offset);
5423 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5425 vb.vb_bh = xis->inode_bh;
5426 vb.vb_access = ocfs2_journal_access_di;
5428 int i, block_off = 0;
5429 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
5431 name_offset = le16_to_cpu(xe->xe_name_offset);
5432 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5433 i = xbs->here - xbs->header->xh_entries;
5435 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
5436 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
5437 bucket_xh(xbs->bucket),
5444 base = bucket_block(xbs->bucket, block_off);
5445 vb.vb_bh = xbs->bucket->bu_bhs[block_off];
5446 vb.vb_access = ocfs2_journal_access;
5448 if (ocfs2_meta_ecc(osb)) {
5449 /*create parameters for ocfs2_post_refcount. */
5450 bucket = xbs->bucket;
5451 refcount.credits = bucket->bu_blocks;
5452 refcount.para = bucket;
5454 ocfs2_xattr_bucket_post_refcount;
5459 vb.vb_bh = xbs->xattr_bh;
5460 vb.vb_access = ocfs2_journal_access_xb;
5464 if (ocfs2_xattr_is_local(xe))
5467 vb.vb_xv = (struct ocfs2_xattr_value_root *)
5468 (base + name_offset + name_len);
5470 ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
5471 &num_clusters, &vb.vb_xv->xr_list,
5479 * We just need to check the 1st extent record, since we always
5480 * CoW the whole xattr. So there shouldn't be a xattr with
5481 * some REFCOUNT extent recs after the 1st one.
5483 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
5486 ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
5487 1, ref_tree, &ref_root_bh);
5494 * If we are deleting the xattr or the new size will be stored inside,
5495 * cool, leave it there, the xattr truncate process will remove them
5496 * for us(it still needs the refcount tree lock and the meta, credits).
5497 * And the worse case is that every cluster truncate will split the
5498 * refcount tree, and make the original extent become 3. So we will need
5499 * 2 * cluster more extent recs at most.
5501 if (!xi->value || xi->value_len <= OCFS2_XATTR_INLINE_SIZE) {
5503 ret = ocfs2_refcounted_xattr_delete_need(inode,
5504 &(*ref_tree)->rf_ci,
5505 ref_root_bh, vb.vb_xv,
5512 ret = ocfs2_refcount_cow_xattr(inode, di, &vb,
5513 *ref_tree, ref_root_bh, 0,
5514 le32_to_cpu(vb.vb_xv->xr_clusters), p);
5519 brelse(ref_root_bh);
5524 * 'security' attributes support
5526 static size_t ocfs2_xattr_security_list(struct inode *inode, char *list,
5527 size_t list_size, const char *name,
5530 const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
5531 const size_t total_len = prefix_len + name_len + 1;
5533 if (list && total_len <= list_size) {
5534 memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
5535 memcpy(list + prefix_len, name, name_len);
5536 list[prefix_len + name_len] = '\0';
5541 static int ocfs2_xattr_security_get(struct inode *inode, const char *name,
5542 void *buffer, size_t size)
5544 if (strcmp(name, "") == 0)
5546 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_SECURITY, name,
5550 static int ocfs2_xattr_security_set(struct inode *inode, const char *name,
5551 const void *value, size_t size, int flags)
5553 if (strcmp(name, "") == 0)
5556 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY, name, value,
5560 int ocfs2_init_security_get(struct inode *inode,
5562 struct ocfs2_security_xattr_info *si)
5564 /* check whether ocfs2 support feature xattr */
5565 if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
5567 return security_inode_init_security(inode, dir, &si->name, &si->value,
5571 int ocfs2_init_security_set(handle_t *handle,
5572 struct inode *inode,
5573 struct buffer_head *di_bh,
5574 struct ocfs2_security_xattr_info *si,
5575 struct ocfs2_alloc_context *xattr_ac,
5576 struct ocfs2_alloc_context *data_ac)
5578 return ocfs2_xattr_set_handle(handle, inode, di_bh,
5579 OCFS2_XATTR_INDEX_SECURITY,
5580 si->name, si->value, si->value_len, 0,
5584 struct xattr_handler ocfs2_xattr_security_handler = {
5585 .prefix = XATTR_SECURITY_PREFIX,
5586 .list = ocfs2_xattr_security_list,
5587 .get = ocfs2_xattr_security_get,
5588 .set = ocfs2_xattr_security_set,
5592 * 'trusted' attributes support
5594 static size_t ocfs2_xattr_trusted_list(struct inode *inode, char *list,
5595 size_t list_size, const char *name,
5598 const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
5599 const size_t total_len = prefix_len + name_len + 1;
5601 if (list && total_len <= list_size) {
5602 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
5603 memcpy(list + prefix_len, name, name_len);
5604 list[prefix_len + name_len] = '\0';
5609 static int ocfs2_xattr_trusted_get(struct inode *inode, const char *name,
5610 void *buffer, size_t size)
5612 if (strcmp(name, "") == 0)
5614 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED, name,
5618 static int ocfs2_xattr_trusted_set(struct inode *inode, const char *name,
5619 const void *value, size_t size, int flags)
5621 if (strcmp(name, "") == 0)
5624 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED, name, value,
5628 struct xattr_handler ocfs2_xattr_trusted_handler = {
5629 .prefix = XATTR_TRUSTED_PREFIX,
5630 .list = ocfs2_xattr_trusted_list,
5631 .get = ocfs2_xattr_trusted_get,
5632 .set = ocfs2_xattr_trusted_set,
5636 * 'user' attributes support
5638 static size_t ocfs2_xattr_user_list(struct inode *inode, char *list,
5639 size_t list_size, const char *name,
5642 const size_t prefix_len = XATTR_USER_PREFIX_LEN;
5643 const size_t total_len = prefix_len + name_len + 1;
5644 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5646 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5649 if (list && total_len <= list_size) {
5650 memcpy(list, XATTR_USER_PREFIX, prefix_len);
5651 memcpy(list + prefix_len, name, name_len);
5652 list[prefix_len + name_len] = '\0';
5657 static int ocfs2_xattr_user_get(struct inode *inode, const char *name,
5658 void *buffer, size_t size)
5660 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5662 if (strcmp(name, "") == 0)
5664 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5666 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
5670 static int ocfs2_xattr_user_set(struct inode *inode, const char *name,
5671 const void *value, size_t size, int flags)
5673 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5675 if (strcmp(name, "") == 0)
5677 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5680 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER, name, value,
5684 struct xattr_handler ocfs2_xattr_user_handler = {
5685 .prefix = XATTR_USER_PREFIX,
5686 .list = ocfs2_xattr_user_list,
5687 .get = ocfs2_xattr_user_get,
5688 .set = ocfs2_xattr_user_set,