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;
2108 static int ocfs2_create_xattr_block(handle_t *handle,
2109 struct inode *inode,
2110 struct buffer_head *inode_bh,
2111 struct ocfs2_alloc_context *meta_ac,
2112 struct buffer_head **ret_bh)
2115 u16 suballoc_bit_start;
2118 struct ocfs2_dinode *di = (struct ocfs2_dinode *)inode_bh->b_data;
2119 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2120 struct buffer_head *new_bh = NULL;
2121 struct ocfs2_xattr_block *xblk;
2123 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), inode_bh,
2124 OCFS2_JOURNAL_ACCESS_CREATE);
2130 ret = ocfs2_claim_metadata(osb, handle, meta_ac, 1,
2131 &suballoc_bit_start, &num_got,
2138 new_bh = sb_getblk(inode->i_sb, first_blkno);
2139 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2141 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode),
2143 OCFS2_JOURNAL_ACCESS_CREATE);
2149 /* Initialize ocfs2_xattr_block */
2150 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2151 memset(xblk, 0, inode->i_sb->s_blocksize);
2152 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2153 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
2154 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2155 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
2156 xblk->xb_blkno = cpu_to_le64(first_blkno);
2158 ret = ocfs2_journal_dirty(handle, new_bh);
2163 di->i_xattr_loc = cpu_to_le64(first_blkno);
2164 ocfs2_journal_dirty(handle, inode_bh);
2175 * ocfs2_xattr_block_set()
2177 * Set, replace or remove an extended attribute into external block.
2180 static int ocfs2_xattr_block_set(struct inode *inode,
2181 struct ocfs2_xattr_info *xi,
2182 struct ocfs2_xattr_search *xs,
2183 struct ocfs2_xattr_set_ctxt *ctxt)
2185 struct buffer_head *new_bh = NULL;
2186 handle_t *handle = ctxt->handle;
2187 struct ocfs2_xattr_block *xblk = NULL;
2190 if (!xs->xattr_bh) {
2191 ret = ocfs2_create_xattr_block(handle, inode, xs->inode_bh,
2192 ctxt->meta_ac, &new_bh);
2198 xs->xattr_bh = new_bh;
2199 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2200 xs->header = &xblk->xb_attrs.xb_header;
2201 xs->base = (void *)xs->header;
2202 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2203 xs->here = xs->header->xh_entries;
2205 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2207 if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2208 /* Set extended attribute into external block */
2209 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2210 OCFS2_HAS_XATTR_FL);
2211 if (!ret || ret != -ENOSPC)
2214 ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2219 ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2226 /* Check whether the new xattr can be inserted into the inode. */
2227 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2228 struct ocfs2_xattr_info *xi,
2229 struct ocfs2_xattr_search *xs)
2232 struct ocfs2_xattr_entry *last;
2234 size_t min_offs = xs->end - xs->base;
2239 last = xs->header->xh_entries;
2241 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
2242 size_t offs = le16_to_cpu(last->xe_name_offset);
2243 if (offs < min_offs)
2248 free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
2252 BUG_ON(!xs->not_found);
2254 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2255 value_size = OCFS2_XATTR_ROOT_SIZE;
2257 value_size = OCFS2_XATTR_SIZE(xi->value_len);
2259 if (free >= sizeof(struct ocfs2_xattr_entry) +
2260 OCFS2_XATTR_SIZE(strlen(xi->name)) + value_size)
2266 static int ocfs2_calc_xattr_set_need(struct inode *inode,
2267 struct ocfs2_dinode *di,
2268 struct ocfs2_xattr_info *xi,
2269 struct ocfs2_xattr_search *xis,
2270 struct ocfs2_xattr_search *xbs,
2275 int ret = 0, old_in_xb = 0;
2276 int clusters_add = 0, meta_add = 0, credits = 0;
2277 struct buffer_head *bh = NULL;
2278 struct ocfs2_xattr_block *xb = NULL;
2279 struct ocfs2_xattr_entry *xe = NULL;
2280 struct ocfs2_xattr_value_root *xv = NULL;
2282 int name_offset, name_len = 0;
2283 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2288 * Calculate the clusters we need to write.
2289 * No matter whether we replace an old one or add a new one,
2290 * we need this for writing.
2292 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2293 credits += new_clusters *
2294 ocfs2_clusters_to_blocks(inode->i_sb, 1);
2296 if (xis->not_found && xbs->not_found) {
2297 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2299 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2300 clusters_add += new_clusters;
2301 credits += ocfs2_calc_extend_credits(inode->i_sb,
2309 if (!xis->not_found) {
2311 name_offset = le16_to_cpu(xe->xe_name_offset);
2312 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2314 credits += OCFS2_INODE_UPDATE_CREDITS;
2316 int i, block_off = 0;
2317 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2319 name_offset = le16_to_cpu(xe->xe_name_offset);
2320 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2321 i = xbs->here - xbs->header->xh_entries;
2324 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2325 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
2326 bucket_xh(xbs->bucket),
2329 base = bucket_block(xbs->bucket, block_off);
2330 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2333 credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
2338 * delete a xattr doesn't need metadata and cluster allocation.
2339 * so just calculate the credits and return.
2341 * The credits for removing the value tree will be extended
2342 * by ocfs2_remove_extent itself.
2345 if (!ocfs2_xattr_is_local(xe))
2346 credits += ocfs2_remove_extent_credits(inode->i_sb);
2351 /* do cluster allocation guess first. */
2352 value_size = le64_to_cpu(xe->xe_value_size);
2356 * In xattr set, we always try to set the xe in inode first,
2357 * so if it can be inserted into inode successfully, the old
2358 * one will be removed from the xattr block, and this xattr
2359 * will be inserted into inode as a new xattr in inode.
2361 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
2362 clusters_add += new_clusters;
2363 credits += ocfs2_remove_extent_credits(inode->i_sb) +
2364 OCFS2_INODE_UPDATE_CREDITS;
2365 if (!ocfs2_xattr_is_local(xe))
2366 credits += ocfs2_calc_extend_credits(
2374 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2375 /* the new values will be stored outside. */
2376 u32 old_clusters = 0;
2378 if (!ocfs2_xattr_is_local(xe)) {
2379 old_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2381 xv = (struct ocfs2_xattr_value_root *)
2382 (base + name_offset + name_len);
2383 value_size = OCFS2_XATTR_ROOT_SIZE;
2387 if (old_clusters >= new_clusters) {
2388 credits += ocfs2_remove_extent_credits(inode->i_sb);
2391 meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
2392 clusters_add += new_clusters - old_clusters;
2393 credits += ocfs2_calc_extend_credits(inode->i_sb,
2397 if (value_size >= OCFS2_XATTR_ROOT_SIZE)
2402 * Now the new value will be stored inside. So if the new
2403 * value is smaller than the size of value root or the old
2404 * value, we don't need any allocation, otherwise we have
2405 * to guess metadata allocation.
2407 if ((ocfs2_xattr_is_local(xe) && value_size >= xi->value_len) ||
2408 (!ocfs2_xattr_is_local(xe) &&
2409 OCFS2_XATTR_ROOT_SIZE >= xi->value_len))
2414 /* calculate metadata allocation. */
2415 if (di->i_xattr_loc) {
2416 if (!xbs->xattr_bh) {
2417 ret = ocfs2_read_xattr_block(inode,
2418 le64_to_cpu(di->i_xattr_loc),
2425 xb = (struct ocfs2_xattr_block *)bh->b_data;
2427 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2430 * If there is already an xattr tree, good, we can calculate
2431 * like other b-trees. Otherwise we may have the chance of
2432 * create a tree, the credit calculation is borrowed from
2433 * ocfs2_calc_extend_credits with root_el = NULL. And the
2434 * new tree will be cluster based, so no meta is needed.
2436 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2437 struct ocfs2_extent_list *el =
2438 &xb->xb_attrs.xb_root.xt_list;
2439 meta_add += ocfs2_extend_meta_needed(el);
2440 credits += ocfs2_calc_extend_credits(inode->i_sb,
2443 credits += OCFS2_SUBALLOC_ALLOC + 1;
2446 * This cluster will be used either for new bucket or for
2448 * If the cluster size is the same as the bucket size, one
2449 * more is needed since we may need to extend the bucket
2453 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2454 if (OCFS2_XATTR_BUCKET_SIZE ==
2455 OCFS2_SB(inode->i_sb)->s_clustersize) {
2456 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2461 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
2465 *clusters_need = clusters_add;
2467 *meta_need = meta_add;
2469 *credits_need = credits;
2474 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
2475 struct ocfs2_dinode *di,
2476 struct ocfs2_xattr_info *xi,
2477 struct ocfs2_xattr_search *xis,
2478 struct ocfs2_xattr_search *xbs,
2479 struct ocfs2_xattr_set_ctxt *ctxt,
2483 int clusters_add, meta_add, ret;
2484 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2486 memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
2488 ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
2490 ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
2491 &clusters_add, &meta_add, credits);
2497 meta_add += extra_meta;
2498 mlog(0, "Set xattr %s, reserve meta blocks = %d, clusters = %d, "
2499 "credits = %d\n", xi->name, meta_add, clusters_add, *credits);
2502 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
2511 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
2517 if (ctxt->meta_ac) {
2518 ocfs2_free_alloc_context(ctxt->meta_ac);
2519 ctxt->meta_ac = NULL;
2523 * We cannot have an error and a non null ctxt->data_ac.
2530 static int __ocfs2_xattr_set_handle(struct inode *inode,
2531 struct ocfs2_dinode *di,
2532 struct ocfs2_xattr_info *xi,
2533 struct ocfs2_xattr_search *xis,
2534 struct ocfs2_xattr_search *xbs,
2535 struct ocfs2_xattr_set_ctxt *ctxt)
2537 int ret = 0, credits, old_found;
2540 /* Remove existing extended attribute */
2541 if (!xis->not_found)
2542 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2543 else if (!xbs->not_found)
2544 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2546 /* We always try to set extended attribute into inode first*/
2547 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2548 if (!ret && !xbs->not_found) {
2550 * If succeed and that extended attribute existing in
2551 * external block, then we will remove it.
2556 old_found = xis->not_found;
2557 xis->not_found = -ENODATA;
2558 ret = ocfs2_calc_xattr_set_need(inode,
2566 xis->not_found = old_found;
2572 ret = ocfs2_extend_trans(ctxt->handle, credits +
2573 ctxt->handle->h_buffer_credits);
2578 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2579 } else if (ret == -ENOSPC) {
2580 if (di->i_xattr_loc && !xbs->xattr_bh) {
2581 ret = ocfs2_xattr_block_find(inode,
2587 old_found = xis->not_found;
2588 xis->not_found = -ENODATA;
2589 ret = ocfs2_calc_xattr_set_need(inode,
2597 xis->not_found = old_found;
2603 ret = ocfs2_extend_trans(ctxt->handle, credits +
2604 ctxt->handle->h_buffer_credits);
2611 * If no space in inode, we will set extended attribute
2612 * into external block.
2614 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2617 if (!xis->not_found) {
2619 * If succeed and that extended attribute
2620 * existing in inode, we will remove it.
2624 xbs->not_found = -ENODATA;
2625 ret = ocfs2_calc_xattr_set_need(inode,
2638 ret = ocfs2_extend_trans(ctxt->handle, credits +
2639 ctxt->handle->h_buffer_credits);
2644 ret = ocfs2_xattr_ibody_set(inode, xi,
2651 /* Update inode ctime. */
2652 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
2654 OCFS2_JOURNAL_ACCESS_WRITE);
2660 inode->i_ctime = CURRENT_TIME;
2661 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
2662 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
2663 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
2670 * This function only called duing creating inode
2671 * for init security/acl xattrs of the new inode.
2672 * All transanction credits have been reserved in mknod.
2674 int ocfs2_xattr_set_handle(handle_t *handle,
2675 struct inode *inode,
2676 struct buffer_head *di_bh,
2682 struct ocfs2_alloc_context *meta_ac,
2683 struct ocfs2_alloc_context *data_ac)
2685 struct ocfs2_dinode *di;
2688 struct ocfs2_xattr_info xi = {
2689 .name_index = name_index,
2692 .value_len = value_len,
2695 struct ocfs2_xattr_search xis = {
2696 .not_found = -ENODATA,
2699 struct ocfs2_xattr_search xbs = {
2700 .not_found = -ENODATA,
2703 struct ocfs2_xattr_set_ctxt ctxt = {
2709 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2713 * In extreme situation, may need xattr bucket when
2714 * block size is too small. And we have already reserved
2715 * the credits for bucket in mknod.
2717 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
2718 xbs.bucket = ocfs2_xattr_bucket_new(inode);
2720 mlog_errno(-ENOMEM);
2725 xis.inode_bh = xbs.inode_bh = di_bh;
2726 di = (struct ocfs2_dinode *)di_bh->b_data;
2728 down_write(&OCFS2_I(inode)->ip_xattr_sem);
2730 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2733 if (xis.not_found) {
2734 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2739 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2742 up_write(&OCFS2_I(inode)->ip_xattr_sem);
2743 brelse(xbs.xattr_bh);
2744 ocfs2_xattr_bucket_free(xbs.bucket);
2752 * Set, replace or remove an extended attribute for this inode.
2753 * value is NULL to remove an existing extended attribute, else either
2754 * create or replace an extended attribute.
2756 int ocfs2_xattr_set(struct inode *inode,
2763 struct buffer_head *di_bh = NULL;
2764 struct ocfs2_dinode *di;
2765 int ret, credits, ref_meta = 0, ref_credits = 0;
2766 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2767 struct inode *tl_inode = osb->osb_tl_inode;
2768 struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2769 struct ocfs2_refcount_tree *ref_tree = NULL;
2771 struct ocfs2_xattr_info xi = {
2772 .name_index = name_index,
2775 .value_len = value_len,
2778 struct ocfs2_xattr_search xis = {
2779 .not_found = -ENODATA,
2782 struct ocfs2_xattr_search xbs = {
2783 .not_found = -ENODATA,
2786 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2790 * Only xbs will be used on indexed trees. xis doesn't need a
2793 xbs.bucket = ocfs2_xattr_bucket_new(inode);
2795 mlog_errno(-ENOMEM);
2799 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2802 goto cleanup_nolock;
2804 xis.inode_bh = xbs.inode_bh = di_bh;
2805 di = (struct ocfs2_dinode *)di_bh->b_data;
2807 down_write(&OCFS2_I(inode)->ip_xattr_sem);
2809 * Scan inode and external block to find the same name
2810 * extended attribute and collect search infomation.
2812 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2815 if (xis.not_found) {
2816 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2821 if (xis.not_found && xbs.not_found) {
2823 if (flags & XATTR_REPLACE)
2830 if (flags & XATTR_CREATE)
2834 /* Check whether the value is refcounted and do some prepartion. */
2835 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL &&
2836 (!xis.not_found || !xbs.not_found)) {
2837 ret = ocfs2_prepare_refcount_xattr(inode, di, &xi,
2838 &xis, &xbs, &ref_tree,
2839 &ref_meta, &ref_credits);
2846 mutex_lock(&tl_inode->i_mutex);
2848 if (ocfs2_truncate_log_needs_flush(osb)) {
2849 ret = __ocfs2_flush_truncate_log(osb);
2851 mutex_unlock(&tl_inode->i_mutex);
2856 mutex_unlock(&tl_inode->i_mutex);
2858 ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
2859 &xbs, &ctxt, ref_meta, &credits);
2865 /* we need to update inode's ctime field, so add credit for it. */
2866 credits += OCFS2_INODE_UPDATE_CREDITS;
2867 ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
2868 if (IS_ERR(ctxt.handle)) {
2869 ret = PTR_ERR(ctxt.handle);
2874 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2876 ocfs2_commit_trans(osb, ctxt.handle);
2879 ocfs2_free_alloc_context(ctxt.data_ac);
2881 ocfs2_free_alloc_context(ctxt.meta_ac);
2882 if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
2883 ocfs2_schedule_truncate_log_flush(osb, 1);
2884 ocfs2_run_deallocs(osb, &ctxt.dealloc);
2887 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
2888 up_write(&OCFS2_I(inode)->ip_xattr_sem);
2889 ocfs2_inode_unlock(inode, 1);
2892 brelse(xbs.xattr_bh);
2893 ocfs2_xattr_bucket_free(xbs.bucket);
2899 * Find the xattr extent rec which may contains name_hash.
2900 * e_cpos will be the first name hash of the xattr rec.
2901 * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2903 static int ocfs2_xattr_get_rec(struct inode *inode,
2908 struct ocfs2_extent_list *el)
2911 struct buffer_head *eb_bh = NULL;
2912 struct ocfs2_extent_block *eb;
2913 struct ocfs2_extent_rec *rec = NULL;
2916 if (el->l_tree_depth) {
2917 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, name_hash,
2924 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2927 if (el->l_tree_depth) {
2928 ocfs2_error(inode->i_sb,
2929 "Inode %lu has non zero tree depth in "
2930 "xattr tree block %llu\n", inode->i_ino,
2931 (unsigned long long)eb_bh->b_blocknr);
2937 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2938 rec = &el->l_recs[i];
2940 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2941 e_blkno = le64_to_cpu(rec->e_blkno);
2947 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2948 "record (%u, %u, 0) in xattr", inode->i_ino,
2949 le32_to_cpu(rec->e_cpos),
2950 ocfs2_rec_clusters(el, rec));
2955 *p_blkno = le64_to_cpu(rec->e_blkno);
2956 *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2958 *e_cpos = le32_to_cpu(rec->e_cpos);
2964 typedef int (xattr_bucket_func)(struct inode *inode,
2965 struct ocfs2_xattr_bucket *bucket,
2968 static int ocfs2_find_xe_in_bucket(struct inode *inode,
2969 struct ocfs2_xattr_bucket *bucket,
2976 int i, ret = 0, cmp = 1, block_off, new_offset;
2977 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
2978 size_t name_len = strlen(name);
2979 struct ocfs2_xattr_entry *xe = NULL;
2983 * We don't use binary search in the bucket because there
2984 * may be multiple entries with the same name hash.
2986 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2987 xe = &xh->xh_entries[i];
2989 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2991 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2994 cmp = name_index - ocfs2_xattr_get_type(xe);
2996 cmp = name_len - xe->xe_name_len;
3000 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3011 xe_name = bucket_block(bucket, block_off) + new_offset;
3012 if (!memcmp(name, xe_name, name_len)) {
3024 * Find the specified xattr entry in a series of buckets.
3025 * This series start from p_blkno and last for num_clusters.
3026 * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
3027 * the num of the valid buckets.
3029 * Return the buffer_head this xattr should reside in. And if the xattr's
3030 * hash is in the gap of 2 buckets, return the lower bucket.
3032 static int ocfs2_xattr_bucket_find(struct inode *inode,
3039 struct ocfs2_xattr_search *xs)
3042 struct ocfs2_xattr_header *xh = NULL;
3043 struct ocfs2_xattr_entry *xe = NULL;
3045 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3046 int low_bucket = 0, bucket, high_bucket;
3047 struct ocfs2_xattr_bucket *search;
3049 u64 blkno, lower_blkno = 0;
3051 search = ocfs2_xattr_bucket_new(inode);
3058 ret = ocfs2_read_xattr_bucket(search, p_blkno);
3064 xh = bucket_xh(search);
3065 high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
3066 while (low_bucket <= high_bucket) {
3067 ocfs2_xattr_bucket_relse(search);
3069 bucket = (low_bucket + high_bucket) / 2;
3070 blkno = p_blkno + bucket * blk_per_bucket;
3071 ret = ocfs2_read_xattr_bucket(search, blkno);
3077 xh = bucket_xh(search);
3078 xe = &xh->xh_entries[0];
3079 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3080 high_bucket = bucket - 1;
3085 * Check whether the hash of the last entry in our
3086 * bucket is larger than the search one. for an empty
3087 * bucket, the last one is also the first one.
3090 xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3092 last_hash = le32_to_cpu(xe->xe_name_hash);
3094 /* record lower_blkno which may be the insert place. */
3095 lower_blkno = blkno;
3097 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3098 low_bucket = bucket + 1;
3102 /* the searched xattr should reside in this bucket if exists. */
3103 ret = ocfs2_find_xe_in_bucket(inode, search,
3104 name_index, name, name_hash,
3114 * Record the bucket we have found.
3115 * When the xattr's hash value is in the gap of 2 buckets, we will
3116 * always set it to the previous bucket.
3119 lower_blkno = p_blkno;
3121 /* This should be in cache - we just read it during the search */
3122 ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3128 xs->header = bucket_xh(xs->bucket);
3129 xs->base = bucket_block(xs->bucket, 0);
3130 xs->end = xs->base + inode->i_sb->s_blocksize;
3133 xs->here = &xs->header->xh_entries[index];
3134 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
3135 (unsigned long long)bucket_blkno(xs->bucket), index);
3140 ocfs2_xattr_bucket_free(search);
3144 static int ocfs2_xattr_index_block_find(struct inode *inode,
3145 struct buffer_head *root_bh,
3148 struct ocfs2_xattr_search *xs)
3151 struct ocfs2_xattr_block *xb =
3152 (struct ocfs2_xattr_block *)root_bh->b_data;
3153 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3154 struct ocfs2_extent_list *el = &xb_root->xt_list;
3156 u32 first_hash, num_clusters = 0;
3157 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3159 if (le16_to_cpu(el->l_next_free_rec) == 0)
3162 mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
3163 name, name_hash, name_index);
3165 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3172 BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3174 mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
3175 "in the rec is %u\n", num_clusters, (unsigned long long)p_blkno,
3178 ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3179 p_blkno, first_hash, num_clusters, xs);
3185 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3188 xattr_bucket_func *func,
3192 u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3193 u32 num_buckets = clusters * bpc;
3194 struct ocfs2_xattr_bucket *bucket;
3196 bucket = ocfs2_xattr_bucket_new(inode);
3198 mlog_errno(-ENOMEM);
3202 mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
3203 clusters, (unsigned long long)blkno);
3205 for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3206 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3213 * The real bucket num in this series of blocks is stored
3214 * in the 1st bucket.
3217 num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3219 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
3220 (unsigned long long)blkno,
3221 le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3223 ret = func(inode, bucket, para);
3224 if (ret && ret != -ERANGE)
3226 /* Fall through to bucket_relse() */
3229 ocfs2_xattr_bucket_relse(bucket);
3234 ocfs2_xattr_bucket_free(bucket);
3238 struct ocfs2_xattr_tree_list {
3244 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
3245 struct ocfs2_xattr_header *xh,
3252 if (index < 0 || index >= le16_to_cpu(xh->xh_count))
3255 name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
3257 *block_off = name_offset >> sb->s_blocksize_bits;
3258 *new_offset = name_offset % sb->s_blocksize;
3263 static int ocfs2_list_xattr_bucket(struct inode *inode,
3264 struct ocfs2_xattr_bucket *bucket,
3268 struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
3269 int i, block_off, new_offset;
3270 const char *prefix, *name;
3272 for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
3273 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
3274 type = ocfs2_xattr_get_type(entry);
3275 prefix = ocfs2_xattr_prefix(type);
3278 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3286 name = (const char *)bucket_block(bucket, block_off) +
3288 ret = ocfs2_xattr_list_entry(xl->buffer,
3292 entry->xe_name_len);
3301 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
3302 struct ocfs2_xattr_tree_root *xt,
3306 struct ocfs2_extent_list *el = &xt->xt_list;
3308 u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
3310 struct ocfs2_xattr_tree_list xl = {
3312 .buffer_size = buffer_size,
3316 if (le16_to_cpu(el->l_next_free_rec) == 0)
3319 while (name_hash > 0) {
3320 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
3321 &e_cpos, &num_clusters, el);
3327 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
3328 ocfs2_list_xattr_bucket,
3339 name_hash = e_cpos - 1;
3347 static int cmp_xe(const void *a, const void *b)
3349 const struct ocfs2_xattr_entry *l = a, *r = b;
3350 u32 l_hash = le32_to_cpu(l->xe_name_hash);
3351 u32 r_hash = le32_to_cpu(r->xe_name_hash);
3353 if (l_hash > r_hash)
3355 if (l_hash < r_hash)
3360 static void swap_xe(void *a, void *b, int size)
3362 struct ocfs2_xattr_entry *l = a, *r = b, tmp;
3365 memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
3366 memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
3370 * When the ocfs2_xattr_block is filled up, new bucket will be created
3371 * and all the xattr entries will be moved to the new bucket.
3372 * The header goes at the start of the bucket, and the names+values are
3373 * filled from the end. This is why *target starts as the last buffer.
3374 * Note: we need to sort the entries since they are not saved in order
3375 * in the ocfs2_xattr_block.
3377 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
3378 struct buffer_head *xb_bh,
3379 struct ocfs2_xattr_bucket *bucket)
3381 int i, blocksize = inode->i_sb->s_blocksize;
3382 int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3383 u16 offset, size, off_change;
3384 struct ocfs2_xattr_entry *xe;
3385 struct ocfs2_xattr_block *xb =
3386 (struct ocfs2_xattr_block *)xb_bh->b_data;
3387 struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
3388 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3389 u16 count = le16_to_cpu(xb_xh->xh_count);
3390 char *src = xb_bh->b_data;
3391 char *target = bucket_block(bucket, blks - 1);
3393 mlog(0, "cp xattr from block %llu to bucket %llu\n",
3394 (unsigned long long)xb_bh->b_blocknr,
3395 (unsigned long long)bucket_blkno(bucket));
3397 for (i = 0; i < blks; i++)
3398 memset(bucket_block(bucket, i), 0, blocksize);
3401 * Since the xe_name_offset is based on ocfs2_xattr_header,
3402 * there is a offset change corresponding to the change of
3403 * ocfs2_xattr_header's position.
3405 off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3406 xe = &xb_xh->xh_entries[count - 1];
3407 offset = le16_to_cpu(xe->xe_name_offset) + off_change;
3408 size = blocksize - offset;
3410 /* copy all the names and values. */
3411 memcpy(target + offset, src + offset, size);
3413 /* Init new header now. */
3414 xh->xh_count = xb_xh->xh_count;
3415 xh->xh_num_buckets = cpu_to_le16(1);
3416 xh->xh_name_value_len = cpu_to_le16(size);
3417 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
3419 /* copy all the entries. */
3420 target = bucket_block(bucket, 0);
3421 offset = offsetof(struct ocfs2_xattr_header, xh_entries);
3422 size = count * sizeof(struct ocfs2_xattr_entry);
3423 memcpy(target + offset, (char *)xb_xh + offset, size);
3425 /* Change the xe offset for all the xe because of the move. */
3426 off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
3427 offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3428 for (i = 0; i < count; i++)
3429 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
3431 mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
3432 offset, size, off_change);
3434 sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
3439 * After we move xattr from block to index btree, we have to
3440 * update ocfs2_xattr_search to the new xe and base.
3442 * When the entry is in xattr block, xattr_bh indicates the storage place.
3443 * While if the entry is in index b-tree, "bucket" indicates the
3444 * real place of the xattr.
3446 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
3447 struct ocfs2_xattr_search *xs,
3448 struct buffer_head *old_bh)
3450 char *buf = old_bh->b_data;
3451 struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
3452 struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
3455 xs->header = bucket_xh(xs->bucket);
3456 xs->base = bucket_block(xs->bucket, 0);
3457 xs->end = xs->base + inode->i_sb->s_blocksize;
3462 i = xs->here - old_xh->xh_entries;
3463 xs->here = &xs->header->xh_entries[i];
3466 static int ocfs2_xattr_create_index_block(struct inode *inode,
3467 struct ocfs2_xattr_search *xs,
3468 struct ocfs2_xattr_set_ctxt *ctxt)
3473 handle_t *handle = ctxt->handle;
3474 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3475 struct ocfs2_inode_info *oi = OCFS2_I(inode);
3476 struct buffer_head *xb_bh = xs->xattr_bh;
3477 struct ocfs2_xattr_block *xb =
3478 (struct ocfs2_xattr_block *)xb_bh->b_data;
3479 struct ocfs2_xattr_tree_root *xr;
3480 u16 xb_flags = le16_to_cpu(xb->xb_flags);
3482 mlog(0, "create xattr index block for %llu\n",
3483 (unsigned long long)xb_bh->b_blocknr);
3485 BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
3486 BUG_ON(!xs->bucket);
3490 * We can use this lock for now, and maybe move to a dedicated mutex
3491 * if performance becomes a problem later.
3493 down_write(&oi->ip_alloc_sem);
3495 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), xb_bh,
3496 OCFS2_JOURNAL_ACCESS_WRITE);
3502 ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac,
3503 1, 1, &bit_off, &len);
3510 * The bucket may spread in many blocks, and
3511 * we will only touch the 1st block and the last block
3512 * in the whole bucket(one for entry and one for data).
3514 blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
3516 mlog(0, "allocate 1 cluster from %llu to xattr block\n",
3517 (unsigned long long)blkno);
3519 ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
3525 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
3526 OCFS2_JOURNAL_ACCESS_CREATE);
3532 ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
3533 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
3535 ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
3537 /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
3538 memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
3539 offsetof(struct ocfs2_xattr_block, xb_attrs));
3541 xr = &xb->xb_attrs.xb_root;
3542 xr->xt_clusters = cpu_to_le32(1);
3543 xr->xt_last_eb_blk = 0;
3544 xr->xt_list.l_tree_depth = 0;
3545 xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
3546 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
3548 xr->xt_list.l_recs[0].e_cpos = 0;
3549 xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
3550 xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
3552 xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
3554 ocfs2_journal_dirty(handle, xb_bh);
3557 up_write(&oi->ip_alloc_sem);
3562 static int cmp_xe_offset(const void *a, const void *b)
3564 const struct ocfs2_xattr_entry *l = a, *r = b;
3565 u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
3566 u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
3568 if (l_name_offset < r_name_offset)
3570 if (l_name_offset > r_name_offset)
3576 * defrag a xattr bucket if we find that the bucket has some
3577 * holes beteen name/value pairs.
3578 * We will move all the name/value pairs to the end of the bucket
3579 * so that we can spare some space for insertion.
3581 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
3583 struct ocfs2_xattr_bucket *bucket)
3586 size_t end, offset, len, value_len;
3587 struct ocfs2_xattr_header *xh;
3588 char *entries, *buf, *bucket_buf = NULL;
3589 u64 blkno = bucket_blkno(bucket);
3591 size_t blocksize = inode->i_sb->s_blocksize;
3592 struct ocfs2_xattr_entry *xe;
3595 * In order to make the operation more efficient and generic,
3596 * we copy all the blocks into a contiguous memory and do the
3597 * defragment there, so if anything is error, we will not touch
3600 bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
3607 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3608 memcpy(buf, bucket_block(bucket, i), blocksize);
3610 ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
3611 OCFS2_JOURNAL_ACCESS_WRITE);
3617 xh = (struct ocfs2_xattr_header *)bucket_buf;
3618 entries = (char *)xh->xh_entries;
3619 xh_free_start = le16_to_cpu(xh->xh_free_start);
3621 mlog(0, "adjust xattr bucket in %llu, count = %u, "
3622 "xh_free_start = %u, xh_name_value_len = %u.\n",
3623 (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
3624 xh_free_start, le16_to_cpu(xh->xh_name_value_len));
3627 * sort all the entries by their offset.
3628 * the largest will be the first, so that we can
3629 * move them to the end one by one.
3631 sort(entries, le16_to_cpu(xh->xh_count),
3632 sizeof(struct ocfs2_xattr_entry),
3633 cmp_xe_offset, swap_xe);
3635 /* Move all name/values to the end of the bucket. */
3636 xe = xh->xh_entries;
3637 end = OCFS2_XATTR_BUCKET_SIZE;
3638 for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
3639 offset = le16_to_cpu(xe->xe_name_offset);
3640 if (ocfs2_xattr_is_local(xe))
3641 value_len = OCFS2_XATTR_SIZE(
3642 le64_to_cpu(xe->xe_value_size));
3644 value_len = OCFS2_XATTR_ROOT_SIZE;
3645 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
3648 * We must make sure that the name/value pair
3649 * exist in the same block. So adjust end to
3650 * the previous block end if needed.
3652 if (((end - len) / blocksize !=
3653 (end - 1) / blocksize))
3654 end = end - end % blocksize;
3656 if (end > offset + len) {
3657 memmove(bucket_buf + end - len,
3658 bucket_buf + offset, len);
3659 xe->xe_name_offset = cpu_to_le16(end - len);
3662 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
3663 "bucket %llu\n", (unsigned long long)blkno);
3668 mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
3669 "bucket %llu\n", (unsigned long long)blkno);
3671 if (xh_free_start == end)
3674 memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
3675 xh->xh_free_start = cpu_to_le16(end);
3677 /* sort the entries by their name_hash. */
3678 sort(entries, le16_to_cpu(xh->xh_count),
3679 sizeof(struct ocfs2_xattr_entry),
3683 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3684 memcpy(bucket_block(bucket, i), buf, blocksize);
3685 ocfs2_xattr_bucket_journal_dirty(handle, bucket);
3693 * prev_blkno points to the start of an existing extent. new_blkno
3694 * points to a newly allocated extent. Because we know each of our
3695 * clusters contains more than bucket, we can easily split one cluster
3696 * at a bucket boundary. So we take the last cluster of the existing
3697 * extent and split it down the middle. We move the last half of the
3698 * buckets in the last cluster of the existing extent over to the new
3701 * first_bh is the buffer at prev_blkno so we can update the existing
3702 * extent's bucket count. header_bh is the bucket were we were hoping
3703 * to insert our xattr. If the bucket move places the target in the new
3704 * extent, we'll update first_bh and header_bh after modifying the old
3707 * first_hash will be set as the 1st xe's name_hash in the new extent.
3709 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3711 struct ocfs2_xattr_bucket *first,
3712 struct ocfs2_xattr_bucket *target,
3718 struct super_block *sb = inode->i_sb;
3719 int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
3720 int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
3721 int to_move = num_buckets / 2;
3723 u64 last_cluster_blkno = bucket_blkno(first) +
3724 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
3726 BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
3727 BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
3729 mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3730 (unsigned long long)last_cluster_blkno, (unsigned long long)new_blkno);
3732 ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
3733 last_cluster_blkno, new_blkno,
3734 to_move, first_hash);
3740 /* This is the first bucket that got moved */
3741 src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
3744 * If the target bucket was part of the moved buckets, we need to
3745 * update first and target.
3747 if (bucket_blkno(target) >= src_blkno) {
3748 /* Find the block for the new target bucket */
3749 src_blkno = new_blkno +
3750 (bucket_blkno(target) - src_blkno);
3752 ocfs2_xattr_bucket_relse(first);
3753 ocfs2_xattr_bucket_relse(target);
3756 * These shouldn't fail - the buffers are in the
3757 * journal from ocfs2_cp_xattr_bucket().
3759 ret = ocfs2_read_xattr_bucket(first, new_blkno);
3764 ret = ocfs2_read_xattr_bucket(target, src_blkno);
3775 * Find the suitable pos when we divide a bucket into 2.
3776 * We have to make sure the xattrs with the same hash value exist
3777 * in the same bucket.
3779 * If this ocfs2_xattr_header covers more than one hash value, find a
3780 * place where the hash value changes. Try to find the most even split.
3781 * The most common case is that all entries have different hash values,
3782 * and the first check we make will find a place to split.
3784 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
3786 struct ocfs2_xattr_entry *entries = xh->xh_entries;
3787 int count = le16_to_cpu(xh->xh_count);
3788 int delta, middle = count / 2;
3791 * We start at the middle. Each step gets farther away in both
3792 * directions. We therefore hit the change in hash value
3793 * nearest to the middle. Note that this loop does not execute for
3796 for (delta = 0; delta < middle; delta++) {
3797 /* Let's check delta earlier than middle */
3798 if (cmp_xe(&entries[middle - delta - 1],
3799 &entries[middle - delta]))
3800 return middle - delta;
3802 /* For even counts, don't walk off the end */
3803 if ((middle + delta + 1) == count)
3806 /* Now try delta past middle */
3807 if (cmp_xe(&entries[middle + delta],
3808 &entries[middle + delta + 1]))
3809 return middle + delta + 1;
3812 /* Every entry had the same hash */
3817 * Move some xattrs in old bucket(blk) to new bucket(new_blk).
3818 * first_hash will record the 1st hash of the new bucket.
3820 * Normally half of the xattrs will be moved. But we have to make
3821 * sure that the xattrs with the same hash value are stored in the
3822 * same bucket. If all the xattrs in this bucket have the same hash
3823 * value, the new bucket will be initialized as an empty one and the
3824 * first_hash will be initialized as (hash_value+1).
3826 static int ocfs2_divide_xattr_bucket(struct inode *inode,
3831 int new_bucket_head)
3834 int count, start, len, name_value_len = 0, xe_len, name_offset = 0;
3835 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3836 struct ocfs2_xattr_header *xh;
3837 struct ocfs2_xattr_entry *xe;
3838 int blocksize = inode->i_sb->s_blocksize;
3840 mlog(0, "move some of xattrs from bucket %llu to %llu\n",
3841 (unsigned long long)blk, (unsigned long long)new_blk);
3843 s_bucket = ocfs2_xattr_bucket_new(inode);
3844 t_bucket = ocfs2_xattr_bucket_new(inode);
3845 if (!s_bucket || !t_bucket) {
3851 ret = ocfs2_read_xattr_bucket(s_bucket, blk);
3857 ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
3858 OCFS2_JOURNAL_ACCESS_WRITE);
3865 * Even if !new_bucket_head, we're overwriting t_bucket. Thus,
3866 * there's no need to read it.
3868 ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
3875 * Hey, if we're overwriting t_bucket, what difference does
3876 * ACCESS_CREATE vs ACCESS_WRITE make? See the comment in the
3877 * same part of ocfs2_cp_xattr_bucket().
3879 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3881 OCFS2_JOURNAL_ACCESS_CREATE :
3882 OCFS2_JOURNAL_ACCESS_WRITE);
3888 xh = bucket_xh(s_bucket);
3889 count = le16_to_cpu(xh->xh_count);
3890 start = ocfs2_xattr_find_divide_pos(xh);
3892 if (start == count) {
3893 xe = &xh->xh_entries[start-1];
3896 * initialized a new empty bucket here.
3897 * The hash value is set as one larger than
3898 * that of the last entry in the previous bucket.
3900 for (i = 0; i < t_bucket->bu_blocks; i++)
3901 memset(bucket_block(t_bucket, i), 0, blocksize);
3903 xh = bucket_xh(t_bucket);
3904 xh->xh_free_start = cpu_to_le16(blocksize);
3905 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
3906 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
3908 goto set_num_buckets;
3911 /* copy the whole bucket to the new first. */
3912 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
3914 /* update the new bucket. */
3915 xh = bucket_xh(t_bucket);
3918 * Calculate the total name/value len and xh_free_start for
3919 * the old bucket first.
3921 name_offset = OCFS2_XATTR_BUCKET_SIZE;
3923 for (i = 0; i < start; i++) {
3924 xe = &xh->xh_entries[i];
3925 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3926 if (ocfs2_xattr_is_local(xe))
3928 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3930 xe_len += OCFS2_XATTR_ROOT_SIZE;
3931 name_value_len += xe_len;
3932 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3933 name_offset = le16_to_cpu(xe->xe_name_offset);
3937 * Now begin the modification to the new bucket.
3939 * In the new bucket, We just move the xattr entry to the beginning
3940 * and don't touch the name/value. So there will be some holes in the
3941 * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3944 xe = &xh->xh_entries[start];
3945 len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3946 mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3947 (int)((char *)xe - (char *)xh),
3948 (int)((char *)xh->xh_entries - (char *)xh));
3949 memmove((char *)xh->xh_entries, (char *)xe, len);
3950 xe = &xh->xh_entries[count - start];
3951 len = sizeof(struct ocfs2_xattr_entry) * start;
3952 memset((char *)xe, 0, len);
3954 le16_add_cpu(&xh->xh_count, -start);
3955 le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3957 /* Calculate xh_free_start for the new bucket. */
3958 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3959 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3960 xe = &xh->xh_entries[i];
3961 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3962 if (ocfs2_xattr_is_local(xe))
3964 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3966 xe_len += OCFS2_XATTR_ROOT_SIZE;
3967 if (le16_to_cpu(xe->xe_name_offset) <
3968 le16_to_cpu(xh->xh_free_start))
3969 xh->xh_free_start = xe->xe_name_offset;
3973 /* set xh->xh_num_buckets for the new xh. */
3974 if (new_bucket_head)
3975 xh->xh_num_buckets = cpu_to_le16(1);
3977 xh->xh_num_buckets = 0;
3979 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
3981 /* store the first_hash of the new bucket. */
3983 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3986 * Now only update the 1st block of the old bucket. If we
3987 * just added a new empty bucket, there is no need to modify
3993 xh = bucket_xh(s_bucket);
3994 memset(&xh->xh_entries[start], 0,
3995 sizeof(struct ocfs2_xattr_entry) * (count - start));
3996 xh->xh_count = cpu_to_le16(start);
3997 xh->xh_free_start = cpu_to_le16(name_offset);
3998 xh->xh_name_value_len = cpu_to_le16(name_value_len);
4000 ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
4003 ocfs2_xattr_bucket_free(s_bucket);
4004 ocfs2_xattr_bucket_free(t_bucket);
4010 * Copy xattr from one bucket to another bucket.
4012 * The caller must make sure that the journal transaction
4013 * has enough space for journaling.
4015 static int ocfs2_cp_xattr_bucket(struct inode *inode,
4022 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4024 BUG_ON(s_blkno == t_blkno);
4026 mlog(0, "cp bucket %llu to %llu, target is %d\n",
4027 (unsigned long long)s_blkno, (unsigned long long)t_blkno,
4030 s_bucket = ocfs2_xattr_bucket_new(inode);
4031 t_bucket = ocfs2_xattr_bucket_new(inode);
4032 if (!s_bucket || !t_bucket) {
4038 ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
4043 * Even if !t_is_new, we're overwriting t_bucket. Thus,
4044 * there's no need to read it.
4046 ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
4051 * Hey, if we're overwriting t_bucket, what difference does
4052 * ACCESS_CREATE vs ACCESS_WRITE make? Well, if we allocated a new
4053 * cluster to fill, we came here from
4054 * ocfs2_mv_xattr_buckets(), and it is really new -
4055 * ACCESS_CREATE is required. But we also might have moved data
4056 * out of t_bucket before extending back into it.
4057 * ocfs2_add_new_xattr_bucket() can do this - its call to
4058 * ocfs2_add_new_xattr_cluster() may have created a new extent
4059 * and copied out the end of the old extent. Then it re-extends
4060 * the old extent back to create space for new xattrs. That's
4061 * how we get here, and the bucket isn't really new.
4063 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4065 OCFS2_JOURNAL_ACCESS_CREATE :
4066 OCFS2_JOURNAL_ACCESS_WRITE);
4070 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4071 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4074 ocfs2_xattr_bucket_free(t_bucket);
4075 ocfs2_xattr_bucket_free(s_bucket);
4081 * src_blk points to the start of an existing extent. last_blk points to
4082 * last cluster in that extent. to_blk points to a newly allocated
4083 * extent. We copy the buckets from the cluster at last_blk to the new
4084 * extent. If start_bucket is non-zero, we skip that many buckets before
4085 * we start copying. The new extent's xh_num_buckets gets set to the
4086 * number of buckets we copied. The old extent's xh_num_buckets shrinks
4087 * by the same amount.
4089 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4090 u64 src_blk, u64 last_blk, u64 to_blk,
4091 unsigned int start_bucket,
4094 int i, ret, credits;
4095 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4096 int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4097 int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4098 struct ocfs2_xattr_bucket *old_first, *new_first;
4100 mlog(0, "mv xattrs from cluster %llu to %llu\n",
4101 (unsigned long long)last_blk, (unsigned long long)to_blk);
4103 BUG_ON(start_bucket >= num_buckets);
4105 num_buckets -= start_bucket;
4106 last_blk += (start_bucket * blks_per_bucket);
4109 /* The first bucket of the original extent */
4110 old_first = ocfs2_xattr_bucket_new(inode);
4111 /* The first bucket of the new extent */
4112 new_first = ocfs2_xattr_bucket_new(inode);
4113 if (!old_first || !new_first) {
4119 ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4126 * We need to update the first bucket of the old extent and all
4127 * the buckets going to the new extent.
4129 credits = ((num_buckets + 1) * blks_per_bucket) +
4130 handle->h_buffer_credits;
4131 ret = ocfs2_extend_trans(handle, credits);
4137 ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4138 OCFS2_JOURNAL_ACCESS_WRITE);
4144 for (i = 0; i < num_buckets; i++) {
4145 ret = ocfs2_cp_xattr_bucket(inode, handle,
4146 last_blk + (i * blks_per_bucket),
4147 to_blk + (i * blks_per_bucket),
4156 * Get the new bucket ready before we dirty anything
4157 * (This actually shouldn't fail, because we already dirtied
4158 * it once in ocfs2_cp_xattr_bucket()).
4160 ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4165 ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4166 OCFS2_JOURNAL_ACCESS_WRITE);
4172 /* Now update the headers */
4173 le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4174 ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4176 bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4177 ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4180 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4183 ocfs2_xattr_bucket_free(new_first);
4184 ocfs2_xattr_bucket_free(old_first);
4189 * Move some xattrs in this cluster to the new cluster.
4190 * This function should only be called when bucket size == cluster size.
4191 * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4193 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4199 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4200 int ret, credits = 2 * blk_per_bucket + handle->h_buffer_credits;
4202 BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4204 ret = ocfs2_extend_trans(handle, credits);
4210 /* Move half of the xattr in start_blk to the next bucket. */
4211 return ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4212 new_blk, first_hash, 1);
4216 * Move some xattrs from the old cluster to the new one since they are not
4217 * contiguous in ocfs2 xattr tree.
4219 * new_blk starts a new separate cluster, and we will move some xattrs from
4220 * prev_blk to it. v_start will be set as the first name hash value in this
4221 * new cluster so that it can be used as e_cpos during tree insertion and
4222 * don't collide with our original b-tree operations. first_bh and header_bh
4223 * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4224 * to extend the insert bucket.
4226 * The problem is how much xattr should we move to the new one and when should
4227 * we update first_bh and header_bh?
4228 * 1. If cluster size > bucket size, that means the previous cluster has more
4229 * than 1 bucket, so just move half nums of bucket into the new cluster and
4230 * update the first_bh and header_bh if the insert bucket has been moved
4231 * to the new cluster.
4232 * 2. If cluster_size == bucket_size:
4233 * a) If the previous extent rec has more than one cluster and the insert
4234 * place isn't in the last cluster, copy the entire last cluster to the
4235 * new one. This time, we don't need to upate the first_bh and header_bh
4236 * since they will not be moved into the new cluster.
4237 * b) Otherwise, move the bottom half of the xattrs in the last cluster into
4238 * the new one. And we set the extend flag to zero if the insert place is
4239 * moved into the new allocated cluster since no extend is needed.
4241 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
4243 struct ocfs2_xattr_bucket *first,
4244 struct ocfs2_xattr_bucket *target,
4252 mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
4253 (unsigned long long)bucket_blkno(first), prev_clusters,
4254 (unsigned long long)new_blk);
4256 if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
4257 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
4266 /* The start of the last cluster in the first extent */
4267 u64 last_blk = bucket_blkno(first) +
4268 ((prev_clusters - 1) *
4269 ocfs2_clusters_to_blocks(inode->i_sb, 1));
4271 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
4272 ret = ocfs2_mv_xattr_buckets(inode, handle,
4273 bucket_blkno(first),
4274 last_blk, new_blk, 0,
4279 ret = ocfs2_divide_xattr_cluster(inode, handle,
4285 if ((bucket_blkno(target) == last_blk) && extend)
4294 * Add a new cluster for xattr storage.
4296 * If the new cluster is contiguous with the previous one, it will be
4297 * appended to the same extent record, and num_clusters will be updated.
4298 * If not, we will insert a new extent for it and move some xattrs in
4299 * the last cluster into the new allocated one.
4300 * We also need to limit the maximum size of a btree leaf, otherwise we'll
4301 * lose the benefits of hashing because we'll have to search large leaves.
4302 * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
4305 * first_bh is the first block of the previous extent rec and header_bh
4306 * indicates the bucket we will insert the new xattrs. They will be updated
4307 * when the header_bh is moved into the new cluster.
4309 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
4310 struct buffer_head *root_bh,
4311 struct ocfs2_xattr_bucket *first,
4312 struct ocfs2_xattr_bucket *target,
4316 struct ocfs2_xattr_set_ctxt *ctxt)
4319 u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
4320 u32 prev_clusters = *num_clusters;
4321 u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
4323 handle_t *handle = ctxt->handle;
4324 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4325 struct ocfs2_extent_tree et;
4327 mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
4328 "previous xattr blkno = %llu\n",
4329 (unsigned long long)OCFS2_I(inode)->ip_blkno,
4330 prev_cpos, (unsigned long long)bucket_blkno(first));
4332 ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
4334 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
4335 OCFS2_JOURNAL_ACCESS_WRITE);
4341 ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac, 1,
4342 clusters_to_add, &bit_off, &num_bits);
4349 BUG_ON(num_bits > clusters_to_add);
4351 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
4352 mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
4353 num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
4355 if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
4356 (prev_clusters + num_bits) << osb->s_clustersize_bits <=
4357 OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
4359 * If this cluster is contiguous with the old one and
4360 * adding this new cluster, we don't surpass the limit of
4361 * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
4362 * initialized and used like other buckets in the previous
4364 * So add it as a contiguous one. The caller will handle
4367 v_start = prev_cpos + prev_clusters;
4368 *num_clusters = prev_clusters + num_bits;
4369 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
4372 ret = ocfs2_adjust_xattr_cross_cluster(inode,
4386 mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
4387 num_bits, (unsigned long long)block, v_start);
4388 ret = ocfs2_insert_extent(handle, &et, v_start, block,
4389 num_bits, 0, ctxt->meta_ac);
4395 ret = ocfs2_journal_dirty(handle, root_bh);
4404 * We are given an extent. 'first' is the bucket at the very front of
4405 * the extent. The extent has space for an additional bucket past
4406 * bucket_xh(first)->xh_num_buckets. 'target_blkno' is the block number
4407 * of the target bucket. We wish to shift every bucket past the target
4408 * down one, filling in that additional space. When we get back to the
4409 * target, we split the target between itself and the now-empty bucket
4410 * at target+1 (aka, target_blkno + blks_per_bucket).
4412 static int ocfs2_extend_xattr_bucket(struct inode *inode,
4414 struct ocfs2_xattr_bucket *first,
4419 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4420 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4422 u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
4424 mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
4425 "from %llu, len = %u\n", (unsigned long long)target_blk,
4426 (unsigned long long)bucket_blkno(first), num_clusters);
4428 /* The extent must have room for an additional bucket */
4429 BUG_ON(new_bucket >=
4430 (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
4432 /* end_blk points to the last existing bucket */
4433 end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
4436 * end_blk is the start of the last existing bucket.
4437 * Thus, (end_blk - target_blk) covers the target bucket and
4438 * every bucket after it up to, but not including, the last
4439 * existing bucket. Then we add the last existing bucket, the
4440 * new bucket, and the first bucket (3 * blk_per_bucket).
4442 credits = (end_blk - target_blk) + (3 * blk_per_bucket) +
4443 handle->h_buffer_credits;
4444 ret = ocfs2_extend_trans(handle, credits);
4450 ret = ocfs2_xattr_bucket_journal_access(handle, first,
4451 OCFS2_JOURNAL_ACCESS_WRITE);
4457 while (end_blk != target_blk) {
4458 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
4459 end_blk + blk_per_bucket, 0);
4462 end_blk -= blk_per_bucket;
4465 /* Move half of the xattr in target_blkno to the next bucket. */
4466 ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
4467 target_blk + blk_per_bucket, NULL, 0);
4469 le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
4470 ocfs2_xattr_bucket_journal_dirty(handle, first);
4477 * Add new xattr bucket in an extent record and adjust the buckets
4478 * accordingly. xb_bh is the ocfs2_xattr_block, and target is the
4479 * bucket we want to insert into.
4481 * In the easy case, we will move all the buckets after target down by
4482 * one. Half of target's xattrs will be moved to the next bucket.
4484 * If current cluster is full, we'll allocate a new one. This may not
4485 * be contiguous. The underlying calls will make sure that there is
4486 * space for the insert, shifting buckets around if necessary.
4487 * 'target' may be moved by those calls.
4489 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
4490 struct buffer_head *xb_bh,
4491 struct ocfs2_xattr_bucket *target,
4492 struct ocfs2_xattr_set_ctxt *ctxt)
4494 struct ocfs2_xattr_block *xb =
4495 (struct ocfs2_xattr_block *)xb_bh->b_data;
4496 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
4497 struct ocfs2_extent_list *el = &xb_root->xt_list;
4499 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
4500 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4501 int ret, num_buckets, extend = 1;
4503 u32 e_cpos, num_clusters;
4504 /* The bucket at the front of the extent */
4505 struct ocfs2_xattr_bucket *first;
4507 mlog(0, "Add new xattr bucket starting from %llu\n",
4508 (unsigned long long)bucket_blkno(target));
4510 /* The first bucket of the original extent */
4511 first = ocfs2_xattr_bucket_new(inode);
4518 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
4525 ret = ocfs2_read_xattr_bucket(first, p_blkno);
4531 num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
4532 if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
4534 * This can move first+target if the target bucket moves
4535 * to the new extent.
4537 ret = ocfs2_add_new_xattr_cluster(inode,
4552 ret = ocfs2_extend_xattr_bucket(inode,
4555 bucket_blkno(target),
4562 ocfs2_xattr_bucket_free(first);
4567 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
4568 struct ocfs2_xattr_bucket *bucket,
4571 int block_off = offs >> inode->i_sb->s_blocksize_bits;
4573 offs = offs % inode->i_sb->s_blocksize;
4574 return bucket_block(bucket, block_off) + offs;
4578 * Handle the normal xattr set, including replace, delete and new.
4580 * Note: "local" indicates the real data's locality. So we can't
4581 * just its bucket locality by its length.
4583 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
4584 struct ocfs2_xattr_info *xi,
4585 struct ocfs2_xattr_search *xs,
4589 struct ocfs2_xattr_entry *last, *xe;
4590 int name_len = strlen(xi->name);
4591 struct ocfs2_xattr_header *xh = xs->header;
4592 u16 count = le16_to_cpu(xh->xh_count), start;
4593 size_t blocksize = inode->i_sb->s_blocksize;
4595 size_t offs, size, new_size;
4597 last = &xh->xh_entries[count];
4598 if (!xs->not_found) {
4600 offs = le16_to_cpu(xe->xe_name_offset);
4601 if (ocfs2_xattr_is_local(xe))
4602 size = OCFS2_XATTR_SIZE(name_len) +
4603 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4605 size = OCFS2_XATTR_SIZE(name_len) +
4606 OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4609 * If the new value will be stored outside, xi->value has been
4610 * initalized as an empty ocfs2_xattr_value_root, and the same
4611 * goes with xi->value_len, so we can set new_size safely here.
4612 * See ocfs2_xattr_set_in_bucket.
4614 new_size = OCFS2_XATTR_SIZE(name_len) +
4615 OCFS2_XATTR_SIZE(xi->value_len);
4617 le16_add_cpu(&xh->xh_name_value_len, -size);
4619 if (new_size > size)
4620 goto set_new_name_value;
4622 /* Now replace the old value with new one. */
4624 xe->xe_value_size = cpu_to_le64(xi->value_len);
4626 xe->xe_value_size = 0;
4628 val = ocfs2_xattr_bucket_get_val(inode,
4630 memset(val + OCFS2_XATTR_SIZE(name_len), 0,
4631 size - OCFS2_XATTR_SIZE(name_len));
4632 if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
4633 memcpy(val + OCFS2_XATTR_SIZE(name_len),
4634 xi->value, xi->value_len);
4636 le16_add_cpu(&xh->xh_name_value_len, new_size);
4637 ocfs2_xattr_set_local(xe, local);
4641 * Remove the old entry if there is more than one.
4642 * We don't remove the last entry so that we can
4643 * use it to indicate the hash value of the empty
4647 le16_add_cpu(&xh->xh_count, -1);
4650 (void *)last - (void *)xe);
4652 sizeof(struct ocfs2_xattr_entry));
4655 cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4660 /* find a new entry for insert. */
4661 int low = 0, high = count - 1, tmp;
4662 struct ocfs2_xattr_entry *tmp_xe;
4664 while (low <= high && count) {
4665 tmp = (low + high) / 2;
4666 tmp_xe = &xh->xh_entries[tmp];
4668 if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
4670 else if (name_hash <
4671 le32_to_cpu(tmp_xe->xe_name_hash))
4679 xe = &xh->xh_entries[low];
4681 memmove(xe + 1, xe, (void *)last - (void *)xe);
4683 le16_add_cpu(&xh->xh_count, 1);
4684 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
4685 xe->xe_name_hash = cpu_to_le32(name_hash);
4686 xe->xe_name_len = name_len;
4687 ocfs2_xattr_set_type(xe, xi->name_index);
4691 /* Insert the new name+value. */
4692 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4695 * We must make sure that the name/value pair
4696 * exists in the same block.
4698 offs = le16_to_cpu(xh->xh_free_start);
4699 start = offs - size;
4701 if (start >> inode->i_sb->s_blocksize_bits !=
4702 (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4703 offs = offs - offs % blocksize;
4704 xh->xh_free_start = cpu_to_le16(offs);
4707 val = ocfs2_xattr_bucket_get_val(inode, xs->bucket, offs - size);
4708 xe->xe_name_offset = cpu_to_le16(offs - size);
4710 memset(val, 0, size);
4711 memcpy(val, xi->name, name_len);
4712 memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4714 xe->xe_value_size = cpu_to_le64(xi->value_len);
4715 ocfs2_xattr_set_local(xe, local);
4717 le16_add_cpu(&xh->xh_free_start, -size);
4718 le16_add_cpu(&xh->xh_name_value_len, size);
4724 * Set the xattr entry in the specified bucket.
4725 * The bucket is indicated by xs->bucket and it should have the enough
4726 * space for the xattr insertion.
4728 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4730 struct ocfs2_xattr_info *xi,
4731 struct ocfs2_xattr_search *xs,
4738 mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4739 (unsigned long)xi->value_len, xi->name_index,
4740 (unsigned long long)bucket_blkno(xs->bucket));
4742 if (!xs->bucket->bu_bhs[1]) {
4743 blkno = bucket_blkno(xs->bucket);
4744 ocfs2_xattr_bucket_relse(xs->bucket);
4745 ret = ocfs2_read_xattr_bucket(xs->bucket, blkno);
4752 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4753 OCFS2_JOURNAL_ACCESS_WRITE);
4759 ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local);
4760 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4767 * Truncate the specified xe_off entry in xattr bucket.
4768 * bucket is indicated by header_bh and len is the new length.
4769 * Both the ocfs2_xattr_value_root and the entry will be updated here.
4771 * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4773 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4774 struct ocfs2_xattr_bucket *bucket,
4777 struct ocfs2_xattr_set_ctxt *ctxt)
4781 struct ocfs2_xattr_entry *xe;
4782 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4783 size_t blocksize = inode->i_sb->s_blocksize;
4784 struct ocfs2_xattr_value_buf vb = {
4785 .vb_access = ocfs2_journal_access,
4788 xe = &xh->xh_entries[xe_off];
4790 BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4792 offset = le16_to_cpu(xe->xe_name_offset) +
4793 OCFS2_XATTR_SIZE(xe->xe_name_len);
4795 value_blk = offset / blocksize;
4797 /* We don't allow ocfs2_xattr_value to be stored in different block. */
4798 BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4800 vb.vb_bh = bucket->bu_bhs[value_blk];
4803 vb.vb_xv = (struct ocfs2_xattr_value_root *)
4804 (vb.vb_bh->b_data + offset % blocksize);
4807 * From here on out we have to dirty the bucket. The generic
4808 * value calls only modify one of the bucket's bhs, but we need
4809 * to send the bucket at once. So if they error, they *could* have
4810 * modified something. We have to assume they did, and dirty
4811 * the whole bucket. This leaves us in a consistent state.
4813 mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4814 xe_off, (unsigned long long)bucket_blkno(bucket), len);
4815 ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
4821 ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
4822 OCFS2_JOURNAL_ACCESS_WRITE);
4828 xe->xe_value_size = cpu_to_le64(len);
4830 ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
4836 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4837 struct ocfs2_xattr_search *xs,
4839 struct ocfs2_xattr_set_ctxt *ctxt)
4842 struct ocfs2_xattr_entry *xe = xs->here;
4843 struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4845 BUG_ON(!xs->bucket->bu_bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4847 offset = xe - xh->xh_entries;
4848 ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket,
4856 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4858 struct ocfs2_xattr_search *xs,
4862 int ret, offset, block_off;
4863 struct ocfs2_xattr_value_root *xv;
4864 struct ocfs2_xattr_entry *xe = xs->here;
4865 struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
4867 struct ocfs2_xattr_value_buf vb = {
4868 .vb_access = ocfs2_journal_access,
4871 BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4873 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb, xh,
4874 xe - xh->xh_entries,
4882 base = bucket_block(xs->bucket, block_off);
4883 xv = (struct ocfs2_xattr_value_root *)(base + offset +
4884 OCFS2_XATTR_SIZE(xe->xe_name_len));
4887 vb.vb_bh = xs->bucket->bu_bhs[block_off];
4888 ret = __ocfs2_xattr_set_value_outside(inode, handle,
4889 &vb, val, value_len);
4896 static int ocfs2_rm_xattr_cluster(struct inode *inode,
4897 struct buffer_head *root_bh,
4903 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4904 struct inode *tl_inode = osb->osb_tl_inode;
4906 struct ocfs2_xattr_block *xb =
4907 (struct ocfs2_xattr_block *)root_bh->b_data;
4908 struct ocfs2_alloc_context *meta_ac = NULL;
4909 struct ocfs2_cached_dealloc_ctxt dealloc;
4910 struct ocfs2_extent_tree et;
4912 ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
4914 ocfs2_init_dealloc_ctxt(&dealloc);
4916 mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4917 cpos, len, (unsigned long long)blkno);
4919 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode), blkno,
4922 ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
4928 mutex_lock(&tl_inode->i_mutex);
4930 if (ocfs2_truncate_log_needs_flush(osb)) {
4931 ret = __ocfs2_flush_truncate_log(osb);
4938 handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
4939 if (IS_ERR(handle)) {
4945 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
4946 OCFS2_JOURNAL_ACCESS_WRITE);
4952 ret = ocfs2_remove_extent(handle, &et, cpos, len, meta_ac,
4959 le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4961 ret = ocfs2_journal_dirty(handle, root_bh);
4967 ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4972 ocfs2_commit_trans(osb, handle);
4974 ocfs2_schedule_truncate_log_flush(osb, 1);
4976 mutex_unlock(&tl_inode->i_mutex);
4979 ocfs2_free_alloc_context(meta_ac);
4981 ocfs2_run_deallocs(osb, &dealloc);
4986 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4988 struct ocfs2_xattr_search *xs)
4990 struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
4991 struct ocfs2_xattr_entry *last = &xh->xh_entries[
4992 le16_to_cpu(xh->xh_count) - 1];
4995 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4996 OCFS2_JOURNAL_ACCESS_WRITE);
5002 /* Remove the old entry. */
5003 memmove(xs->here, xs->here + 1,
5004 (void *)last - (void *)xs->here);
5005 memset(last, 0, sizeof(struct ocfs2_xattr_entry));
5006 le16_add_cpu(&xh->xh_count, -1);
5008 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
5012 * Set the xattr name/value in the bucket specified in xs.
5014 * As the new value in xi may be stored in the bucket or in an outside cluster,
5015 * we divide the whole process into 3 steps:
5016 * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
5017 * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
5018 * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
5019 * 4. If the clusters for the new outside value can't be allocated, we need
5020 * to free the xattr we allocated in set.
5022 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
5023 struct ocfs2_xattr_info *xi,
5024 struct ocfs2_xattr_search *xs,
5025 struct ocfs2_xattr_set_ctxt *ctxt)
5029 char *val = (char *)xi->value;
5030 struct ocfs2_xattr_entry *xe = xs->here;
5031 u32 name_hash = ocfs2_xattr_name_hash(inode, xi->name,
5034 if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
5036 * We need to truncate the xattr storage first.
5038 * If both the old and new value are stored to
5039 * outside block, we only need to truncate
5040 * the storage and then set the value outside.
5042 * If the new value should be stored within block,
5043 * we should free all the outside block first and
5044 * the modification to the xattr block will be done
5045 * by following steps.
5047 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
5048 value_len = xi->value_len;
5052 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
5059 goto set_value_outside;
5062 value_len = xi->value_len;
5063 /* So we have to handle the inside block change now. */
5064 if (value_len > OCFS2_XATTR_INLINE_SIZE) {
5066 * If the new value will be stored outside of block,
5067 * initalize a new empty value root and insert it first.
5070 xi->value = &def_xv;
5071 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
5074 ret = ocfs2_xattr_set_entry_in_bucket(inode, ctxt->handle, xi, xs,
5081 if (value_len <= OCFS2_XATTR_INLINE_SIZE)
5084 /* allocate the space now for the outside block storage. */
5085 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
5090 if (xs->not_found) {
5092 * We can't allocate enough clusters for outside
5093 * storage and we have allocated xattr already,
5094 * so need to remove it.
5096 ocfs2_xattr_bucket_remove_xs(inode, ctxt->handle, xs);
5102 ret = ocfs2_xattr_bucket_set_value_outside(inode, ctxt->handle,
5103 xs, val, value_len);
5109 * check whether the xattr bucket is filled up with the same hash value.
5110 * If we want to insert the xattr with the same hash, return -ENOSPC.
5111 * If we want to insert a xattr with different hash value, go ahead
5112 * and ocfs2_divide_xattr_bucket will handle this.
5114 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
5115 struct ocfs2_xattr_bucket *bucket,
5118 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5119 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5121 if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5124 if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5125 xh->xh_entries[0].xe_name_hash) {
5126 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5128 (unsigned long long)bucket_blkno(bucket),
5129 le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5136 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5137 struct ocfs2_xattr_info *xi,
5138 struct ocfs2_xattr_search *xs,
5139 struct ocfs2_xattr_set_ctxt *ctxt)
5141 struct ocfs2_xattr_header *xh;
5142 struct ocfs2_xattr_entry *xe;
5143 u16 count, header_size, xh_free_start;
5144 int free, max_free, need, old;
5145 size_t value_size = 0, name_len = strlen(xi->name);
5146 size_t blocksize = inode->i_sb->s_blocksize;
5147 int ret, allocation = 0;
5149 mlog_entry("Set xattr %s in xattr index block\n", xi->name);
5153 count = le16_to_cpu(xh->xh_count);
5154 xh_free_start = le16_to_cpu(xh->xh_free_start);
5155 header_size = sizeof(struct ocfs2_xattr_header) +
5156 count * sizeof(struct ocfs2_xattr_entry);
5157 max_free = OCFS2_XATTR_BUCKET_SIZE - header_size -
5158 le16_to_cpu(xh->xh_name_value_len) - OCFS2_XATTR_HEADER_GAP;
5160 mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
5161 "of %u which exceed block size\n",
5162 (unsigned long long)bucket_blkno(xs->bucket),
5165 if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
5166 value_size = OCFS2_XATTR_ROOT_SIZE;
5168 value_size = OCFS2_XATTR_SIZE(xi->value_len);
5171 need = sizeof(struct ocfs2_xattr_entry) +
5172 OCFS2_XATTR_SIZE(name_len) + value_size;
5174 need = value_size + OCFS2_XATTR_SIZE(name_len);
5177 * We only replace the old value if the new length is smaller
5178 * than the old one. Otherwise we will allocate new space in the
5179 * bucket to store it.
5182 if (ocfs2_xattr_is_local(xe))
5183 old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
5185 old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
5187 if (old >= value_size)
5191 free = xh_free_start - header_size - OCFS2_XATTR_HEADER_GAP;
5193 * We need to make sure the new name/value pair
5194 * can exist in the same block.
5196 if (xh_free_start % blocksize < need)
5197 free -= xh_free_start % blocksize;
5199 mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
5200 "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
5201 " %u\n", xs->not_found,
5202 (unsigned long long)bucket_blkno(xs->bucket),
5203 free, need, max_free, le16_to_cpu(xh->xh_free_start),
5204 le16_to_cpu(xh->xh_name_value_len));
5208 count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb))) {
5209 if (need <= max_free &&
5210 count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
5212 * We can create the space by defragment. Since only the
5213 * name/value will be moved, the xe shouldn't be changed
5216 ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5223 xh_free_start = le16_to_cpu(xh->xh_free_start);
5224 free = xh_free_start - header_size
5225 - OCFS2_XATTR_HEADER_GAP;
5226 if (xh_free_start % blocksize < need)
5227 free -= xh_free_start % blocksize;
5232 mlog(0, "Can't get enough space for xattr insert by "
5233 "defragment. Need %u bytes, but we have %d, so "
5234 "allocate new bucket for it.\n", need, free);
5238 * We have to add new buckets or clusters and one
5239 * allocation should leave us enough space for insert.
5244 * We do not allow for overlapping ranges between buckets. And
5245 * the maximum number of collisions we will allow for then is
5246 * one bucket's worth, so check it here whether we need to
5247 * add a new bucket for the insert.
5249 ret = ocfs2_check_xattr_bucket_collision(inode,
5257 ret = ocfs2_add_new_xattr_bucket(inode,
5267 * ocfs2_add_new_xattr_bucket() will have updated
5268 * xs->bucket if it moved, but it will not have updated
5269 * any of the other search fields. Thus, we drop it and
5270 * re-search. Everything should be cached, so it'll be
5273 ocfs2_xattr_bucket_relse(xs->bucket);
5274 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5277 if (ret && ret != -ENODATA)
5279 xs->not_found = ret;
5285 ret = ocfs2_xattr_set_in_bucket(inode, xi, xs, ctxt);
5291 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5292 struct ocfs2_xattr_bucket *bucket,
5296 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5298 struct ocfs2_xattr_entry *xe;
5299 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5300 struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5301 int credits = ocfs2_remove_extent_credits(osb->sb) +
5302 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5305 ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5307 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5308 xe = &xh->xh_entries[i];
5309 if (ocfs2_xattr_is_local(xe))
5312 ctxt.handle = ocfs2_start_trans(osb, credits);
5313 if (IS_ERR(ctxt.handle)) {
5314 ret = PTR_ERR(ctxt.handle);
5319 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5322 ocfs2_commit_trans(osb, ctxt.handle);
5329 ocfs2_schedule_truncate_log_flush(osb, 1);
5330 ocfs2_run_deallocs(osb, &ctxt.dealloc);
5334 static int ocfs2_delete_xattr_index_block(struct inode *inode,
5335 struct buffer_head *xb_bh)
5337 struct ocfs2_xattr_block *xb =
5338 (struct ocfs2_xattr_block *)xb_bh->b_data;
5339 struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
5341 u32 name_hash = UINT_MAX, e_cpos, num_clusters;
5344 if (le16_to_cpu(el->l_next_free_rec) == 0)
5347 while (name_hash > 0) {
5348 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
5349 &e_cpos, &num_clusters, el);
5355 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
5356 ocfs2_delete_xattr_in_bucket,
5363 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
5364 p_blkno, e_cpos, num_clusters);
5373 name_hash = e_cpos - 1;
5381 * Whenever we modify a xattr value root in the bucket(e.g, CoW
5382 * or change the extent record flag), we need to recalculate
5383 * the metaecc for the whole bucket. So it is done here.
5386 * We have to give the extra credits for the caller.
5388 static int ocfs2_xattr_bucket_post_refcount(struct inode *inode,
5393 struct ocfs2_xattr_bucket *bucket =
5394 (struct ocfs2_xattr_bucket *)para;
5396 ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
5397 OCFS2_JOURNAL_ACCESS_WRITE);
5403 ocfs2_xattr_bucket_journal_dirty(handle, bucket);
5409 * Special action we need if the xattr value is refcounted.
5411 * 1. If the xattr is refcounted, lock the tree.
5412 * 2. CoW the xattr if we are setting the new value and the value
5413 * will be stored outside.
5414 * 3. In other case, decrease_refcount will work for us, so just
5415 * lock the refcount tree, calculate the meta and credits is OK.
5417 * We have to do CoW before ocfs2_init_xattr_set_ctxt since
5418 * currently CoW is a completed transaction, while this function
5419 * will also lock the allocators and let us deadlock. So we will
5420 * CoW the whole xattr value.
5422 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
5423 struct ocfs2_dinode *di,
5424 struct ocfs2_xattr_info *xi,
5425 struct ocfs2_xattr_search *xis,
5426 struct ocfs2_xattr_search *xbs,
5427 struct ocfs2_refcount_tree **ref_tree,
5432 struct ocfs2_xattr_block *xb;
5433 struct ocfs2_xattr_entry *xe;
5435 u32 p_cluster, num_clusters;
5436 unsigned int ext_flags;
5437 int name_offset, name_len;
5438 struct ocfs2_xattr_value_buf vb;
5439 struct ocfs2_xattr_bucket *bucket = NULL;
5440 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5441 struct ocfs2_post_refcount refcount;
5442 struct ocfs2_post_refcount *p = NULL;
5443 struct buffer_head *ref_root_bh = NULL;
5445 if (!xis->not_found) {
5447 name_offset = le16_to_cpu(xe->xe_name_offset);
5448 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5450 vb.vb_bh = xis->inode_bh;
5451 vb.vb_access = ocfs2_journal_access_di;
5453 int i, block_off = 0;
5454 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
5456 name_offset = le16_to_cpu(xe->xe_name_offset);
5457 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5458 i = xbs->here - xbs->header->xh_entries;
5460 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
5461 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
5462 bucket_xh(xbs->bucket),
5469 base = bucket_block(xbs->bucket, block_off);
5470 vb.vb_bh = xbs->bucket->bu_bhs[block_off];
5471 vb.vb_access = ocfs2_journal_access;
5473 if (ocfs2_meta_ecc(osb)) {
5474 /*create parameters for ocfs2_post_refcount. */
5475 bucket = xbs->bucket;
5476 refcount.credits = bucket->bu_blocks;
5477 refcount.para = bucket;
5479 ocfs2_xattr_bucket_post_refcount;
5484 vb.vb_bh = xbs->xattr_bh;
5485 vb.vb_access = ocfs2_journal_access_xb;
5489 if (ocfs2_xattr_is_local(xe))
5492 vb.vb_xv = (struct ocfs2_xattr_value_root *)
5493 (base + name_offset + name_len);
5495 ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
5496 &num_clusters, &vb.vb_xv->xr_list,
5504 * We just need to check the 1st extent record, since we always
5505 * CoW the whole xattr. So there shouldn't be a xattr with
5506 * some REFCOUNT extent recs after the 1st one.
5508 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
5511 ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
5512 1, ref_tree, &ref_root_bh);
5519 * If we are deleting the xattr or the new size will be stored inside,
5520 * cool, leave it there, the xattr truncate process will remove them
5521 * for us(it still needs the refcount tree lock and the meta, credits).
5522 * And the worse case is that every cluster truncate will split the
5523 * refcount tree, and make the original extent become 3. So we will need
5524 * 2 * cluster more extent recs at most.
5526 if (!xi->value || xi->value_len <= OCFS2_XATTR_INLINE_SIZE) {
5528 ret = ocfs2_refcounted_xattr_delete_need(inode,
5529 &(*ref_tree)->rf_ci,
5530 ref_root_bh, vb.vb_xv,
5537 ret = ocfs2_refcount_cow_xattr(inode, di, &vb,
5538 *ref_tree, ref_root_bh, 0,
5539 le32_to_cpu(vb.vb_xv->xr_clusters), p);
5544 brelse(ref_root_bh);
5549 * 'security' attributes support
5551 static size_t ocfs2_xattr_security_list(struct inode *inode, char *list,
5552 size_t list_size, const char *name,
5555 const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
5556 const size_t total_len = prefix_len + name_len + 1;
5558 if (list && total_len <= list_size) {
5559 memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
5560 memcpy(list + prefix_len, name, name_len);
5561 list[prefix_len + name_len] = '\0';
5566 static int ocfs2_xattr_security_get(struct inode *inode, const char *name,
5567 void *buffer, size_t size)
5569 if (strcmp(name, "") == 0)
5571 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_SECURITY, name,
5575 static int ocfs2_xattr_security_set(struct inode *inode, const char *name,
5576 const void *value, size_t size, int flags)
5578 if (strcmp(name, "") == 0)
5581 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY, name, value,
5585 int ocfs2_init_security_get(struct inode *inode,
5587 struct ocfs2_security_xattr_info *si)
5589 /* check whether ocfs2 support feature xattr */
5590 if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
5592 return security_inode_init_security(inode, dir, &si->name, &si->value,
5596 int ocfs2_init_security_set(handle_t *handle,
5597 struct inode *inode,
5598 struct buffer_head *di_bh,
5599 struct ocfs2_security_xattr_info *si,
5600 struct ocfs2_alloc_context *xattr_ac,
5601 struct ocfs2_alloc_context *data_ac)
5603 return ocfs2_xattr_set_handle(handle, inode, di_bh,
5604 OCFS2_XATTR_INDEX_SECURITY,
5605 si->name, si->value, si->value_len, 0,
5609 struct xattr_handler ocfs2_xattr_security_handler = {
5610 .prefix = XATTR_SECURITY_PREFIX,
5611 .list = ocfs2_xattr_security_list,
5612 .get = ocfs2_xattr_security_get,
5613 .set = ocfs2_xattr_security_set,
5617 * 'trusted' attributes support
5619 static size_t ocfs2_xattr_trusted_list(struct inode *inode, char *list,
5620 size_t list_size, const char *name,
5623 const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
5624 const size_t total_len = prefix_len + name_len + 1;
5626 if (list && total_len <= list_size) {
5627 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
5628 memcpy(list + prefix_len, name, name_len);
5629 list[prefix_len + name_len] = '\0';
5634 static int ocfs2_xattr_trusted_get(struct inode *inode, const char *name,
5635 void *buffer, size_t size)
5637 if (strcmp(name, "") == 0)
5639 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED, name,
5643 static int ocfs2_xattr_trusted_set(struct inode *inode, const char *name,
5644 const void *value, size_t size, int flags)
5646 if (strcmp(name, "") == 0)
5649 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED, name, value,
5653 struct xattr_handler ocfs2_xattr_trusted_handler = {
5654 .prefix = XATTR_TRUSTED_PREFIX,
5655 .list = ocfs2_xattr_trusted_list,
5656 .get = ocfs2_xattr_trusted_get,
5657 .set = ocfs2_xattr_trusted_set,
5661 * 'user' attributes support
5663 static size_t ocfs2_xattr_user_list(struct inode *inode, char *list,
5664 size_t list_size, const char *name,
5667 const size_t prefix_len = XATTR_USER_PREFIX_LEN;
5668 const size_t total_len = prefix_len + name_len + 1;
5669 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5671 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5674 if (list && total_len <= list_size) {
5675 memcpy(list, XATTR_USER_PREFIX, prefix_len);
5676 memcpy(list + prefix_len, name, name_len);
5677 list[prefix_len + name_len] = '\0';
5682 static int ocfs2_xattr_user_get(struct inode *inode, const char *name,
5683 void *buffer, size_t size)
5685 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5687 if (strcmp(name, "") == 0)
5689 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5691 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
5695 static int ocfs2_xattr_user_set(struct inode *inode, const char *name,
5696 const void *value, size_t size, int flags)
5698 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5700 if (strcmp(name, "") == 0)
5702 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5705 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER, name, value,
5709 struct xattr_handler ocfs2_xattr_user_handler = {
5710 .prefix = XATTR_USER_PREFIX,
5711 .list = ocfs2_xattr_user_list,
5712 .get = ocfs2_xattr_user_get,
5713 .set = ocfs2_xattr_user_set,