1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * Copyright (C) 2004, 2008 Oracle. All rights reserved.
9 * Lots of code in this file is copy from linux/fs/ext3/xattr.c.
10 * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
12 * This program is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU General Public
14 * License version 2 as published by the Free Software Foundation.
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
22 #include <linux/capability.h>
24 #include <linux/types.h>
25 #include <linux/slab.h>
26 #include <linux/highmem.h>
27 #include <linux/pagemap.h>
28 #include <linux/uio.h>
29 #include <linux/sched.h>
30 #include <linux/splice.h>
31 #include <linux/mount.h>
32 #include <linux/writeback.h>
33 #include <linux/falloc.h>
34 #include <linux/sort.h>
35 #include <linux/init.h>
36 #include <linux/module.h>
37 #include <linux/string.h>
38 #include <linux/security.h>
40 #define MLOG_MASK_PREFIX ML_XATTR
41 #include <cluster/masklog.h>
45 #include "blockcheck.h"
55 #include "buffer_head_io.h"
60 struct ocfs2_xattr_def_value_root {
61 struct ocfs2_xattr_value_root xv;
62 struct ocfs2_extent_rec er;
65 struct ocfs2_xattr_bucket {
66 /* The inode these xattrs are associated with */
67 struct inode *bu_inode;
69 /* The actual buffers that make up the bucket */
70 struct buffer_head *bu_bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
72 /* How many blocks make up one bucket for this filesystem */
76 struct ocfs2_xattr_set_ctxt {
78 struct ocfs2_alloc_context *meta_ac;
79 struct ocfs2_alloc_context *data_ac;
80 struct ocfs2_cached_dealloc_ctxt dealloc;
83 #define OCFS2_XATTR_ROOT_SIZE (sizeof(struct ocfs2_xattr_def_value_root))
84 #define OCFS2_XATTR_INLINE_SIZE 80
85 #define OCFS2_XATTR_FREE_IN_IBODY (OCFS2_MIN_XATTR_INLINE_SIZE \
86 - sizeof(struct ocfs2_xattr_header) \
88 #define OCFS2_XATTR_FREE_IN_BLOCK(ptr) ((ptr)->i_sb->s_blocksize \
89 - sizeof(struct ocfs2_xattr_block) \
90 - sizeof(struct ocfs2_xattr_header) \
93 static struct ocfs2_xattr_def_value_root def_xv = {
94 .xv.xr_list.l_count = cpu_to_le16(1),
97 struct xattr_handler *ocfs2_xattr_handlers[] = {
98 &ocfs2_xattr_user_handler,
99 #ifdef CONFIG_OCFS2_FS_POSIX_ACL
100 &ocfs2_xattr_acl_access_handler,
101 &ocfs2_xattr_acl_default_handler,
103 &ocfs2_xattr_trusted_handler,
104 &ocfs2_xattr_security_handler,
108 static struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
109 [OCFS2_XATTR_INDEX_USER] = &ocfs2_xattr_user_handler,
110 #ifdef CONFIG_OCFS2_FS_POSIX_ACL
111 [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS]
112 = &ocfs2_xattr_acl_access_handler,
113 [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT]
114 = &ocfs2_xattr_acl_default_handler,
116 [OCFS2_XATTR_INDEX_TRUSTED] = &ocfs2_xattr_trusted_handler,
117 [OCFS2_XATTR_INDEX_SECURITY] = &ocfs2_xattr_security_handler,
120 struct ocfs2_xattr_info {
127 struct ocfs2_xattr_search {
128 struct buffer_head *inode_bh;
130 * xattr_bh point to the block buffer head which has extended attribute
131 * when extended attribute in inode, xattr_bh is equal to inode_bh.
133 struct buffer_head *xattr_bh;
134 struct ocfs2_xattr_header *header;
135 struct ocfs2_xattr_bucket *bucket;
138 struct ocfs2_xattr_entry *here;
142 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
143 struct ocfs2_xattr_header *xh,
148 static int ocfs2_xattr_block_find(struct inode *inode,
151 struct ocfs2_xattr_search *xs);
152 static int ocfs2_xattr_index_block_find(struct inode *inode,
153 struct buffer_head *root_bh,
156 struct ocfs2_xattr_search *xs);
158 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
159 struct ocfs2_xattr_tree_root *xt,
163 static int ocfs2_xattr_create_index_block(struct inode *inode,
164 struct ocfs2_xattr_search *xs,
165 struct ocfs2_xattr_set_ctxt *ctxt);
167 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
168 struct ocfs2_xattr_info *xi,
169 struct ocfs2_xattr_search *xs,
170 struct ocfs2_xattr_set_ctxt *ctxt);
172 static int ocfs2_delete_xattr_index_block(struct inode *inode,
173 struct buffer_head *xb_bh);
174 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
175 u64 src_blk, u64 last_blk, u64 to_blk,
176 unsigned int start_bucket,
179 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
181 return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
184 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
186 return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
189 static inline u16 ocfs2_xattr_max_xe_in_bucket(struct super_block *sb)
191 u16 len = sb->s_blocksize -
192 offsetof(struct ocfs2_xattr_header, xh_entries);
194 return len / sizeof(struct ocfs2_xattr_entry);
197 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
198 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
199 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
201 static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
203 struct ocfs2_xattr_bucket *bucket;
204 int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
206 BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
208 bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
210 bucket->bu_inode = inode;
211 bucket->bu_blocks = blks;
217 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
221 for (i = 0; i < bucket->bu_blocks; i++) {
222 brelse(bucket->bu_bhs[i]);
223 bucket->bu_bhs[i] = NULL;
227 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
230 ocfs2_xattr_bucket_relse(bucket);
231 bucket->bu_inode = NULL;
237 * A bucket that has never been written to disk doesn't need to be
238 * read. We just need the buffer_heads. Don't call this for
239 * buckets that are already on disk. ocfs2_read_xattr_bucket() initializes
242 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
247 for (i = 0; i < bucket->bu_blocks; i++) {
248 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
250 if (!bucket->bu_bhs[i]) {
256 if (!ocfs2_buffer_uptodate(bucket->bu_inode,
258 ocfs2_set_new_buffer_uptodate(bucket->bu_inode,
263 ocfs2_xattr_bucket_relse(bucket);
267 /* Read the xattr bucket at xb_blkno */
268 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
273 rc = ocfs2_read_blocks(bucket->bu_inode, xb_blkno,
274 bucket->bu_blocks, bucket->bu_bhs, 0,
277 rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
280 &bucket_xh(bucket)->xh_check);
286 ocfs2_xattr_bucket_relse(bucket);
290 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
291 struct ocfs2_xattr_bucket *bucket,
296 for (i = 0; i < bucket->bu_blocks; i++) {
297 rc = ocfs2_journal_access(handle, bucket->bu_inode,
298 bucket->bu_bhs[i], type);
308 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
309 struct ocfs2_xattr_bucket *bucket)
313 ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
314 bucket->bu_bhs, bucket->bu_blocks,
315 &bucket_xh(bucket)->xh_check);
317 for (i = 0; i < bucket->bu_blocks; i++)
318 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
321 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
322 struct ocfs2_xattr_bucket *src)
325 int blocksize = src->bu_inode->i_sb->s_blocksize;
327 BUG_ON(dest->bu_blocks != src->bu_blocks);
328 BUG_ON(dest->bu_inode != src->bu_inode);
330 for (i = 0; i < src->bu_blocks; i++) {
331 memcpy(bucket_block(dest, i), bucket_block(src, i),
336 static int ocfs2_validate_xattr_block(struct super_block *sb,
337 struct buffer_head *bh)
340 struct ocfs2_xattr_block *xb =
341 (struct ocfs2_xattr_block *)bh->b_data;
343 mlog(0, "Validating xattr block %llu\n",
344 (unsigned long long)bh->b_blocknr);
346 BUG_ON(!buffer_uptodate(bh));
349 * If the ecc fails, we return the error but otherwise
350 * leave the filesystem running. We know any error is
351 * local to this block.
353 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
358 * Errors after here are fatal
361 if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
363 "Extended attribute block #%llu has bad "
365 (unsigned long long)bh->b_blocknr, 7,
370 if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
372 "Extended attribute block #%llu has an "
373 "invalid xb_blkno of %llu",
374 (unsigned long long)bh->b_blocknr,
375 (unsigned long long)le64_to_cpu(xb->xb_blkno));
379 if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
381 "Extended attribute block #%llu has an invalid "
382 "xb_fs_generation of #%u",
383 (unsigned long long)bh->b_blocknr,
384 le32_to_cpu(xb->xb_fs_generation));
391 static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
392 struct buffer_head **bh)
395 struct buffer_head *tmp = *bh;
397 rc = ocfs2_read_block(inode, xb_blkno, &tmp,
398 ocfs2_validate_xattr_block);
400 /* If ocfs2_read_block() got us a new bh, pass it up. */
407 static inline const char *ocfs2_xattr_prefix(int name_index)
409 struct xattr_handler *handler = NULL;
411 if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
412 handler = ocfs2_xattr_handler_map[name_index];
414 return handler ? handler->prefix : NULL;
417 static u32 ocfs2_xattr_name_hash(struct inode *inode,
421 /* Get hash value of uuid from super block */
422 u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
425 /* hash extended attribute name */
426 for (i = 0; i < name_len; i++) {
427 hash = (hash << OCFS2_HASH_SHIFT) ^
428 (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
436 * ocfs2_xattr_hash_entry()
438 * Compute the hash of an extended attribute.
440 static void ocfs2_xattr_hash_entry(struct inode *inode,
441 struct ocfs2_xattr_header *header,
442 struct ocfs2_xattr_entry *entry)
445 char *name = (char *)header + le16_to_cpu(entry->xe_name_offset);
447 hash = ocfs2_xattr_name_hash(inode, name, entry->xe_name_len);
448 entry->xe_name_hash = cpu_to_le32(hash);
453 static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
457 if (value_len <= OCFS2_XATTR_INLINE_SIZE)
458 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
460 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
461 size += sizeof(struct ocfs2_xattr_entry);
466 int ocfs2_calc_security_init(struct inode *dir,
467 struct ocfs2_security_xattr_info *si,
470 struct ocfs2_alloc_context **xattr_ac)
473 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
474 int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
478 * The max space of security xattr taken inline is
479 * 256(name) + 80(value) + 16(entry) = 352 bytes,
480 * So reserve one metadata block for it is ok.
482 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
483 s_size > OCFS2_XATTR_FREE_IN_IBODY) {
484 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
489 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
492 /* reserve clusters for xattr value which will be set in B tree*/
493 if (si->value_len > OCFS2_XATTR_INLINE_SIZE)
494 *want_clusters += ocfs2_clusters_for_bytes(dir->i_sb,
499 int ocfs2_calc_xattr_init(struct inode *dir,
500 struct buffer_head *dir_bh,
502 struct ocfs2_security_xattr_info *si,
505 struct ocfs2_alloc_context **xattr_ac)
508 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
514 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
517 if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
518 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
519 OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
522 a_size = ocfs2_xattr_entry_real_size(0, acl_len);
525 } else if (acl_len != 0 && acl_len != -ENODATA) {
531 if (!(s_size + a_size))
535 * The max space of security xattr taken inline is
536 * 256(name) + 80(value) + 16(entry) = 352 bytes,
537 * The max space of acl xattr taken inline is
538 * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
539 * when blocksize = 512, may reserve one more cluser for
540 * xattr bucket, otherwise reserve one metadata block
543 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
544 (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
545 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
550 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
553 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
554 (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
556 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
559 /* reserve clusters for xattr value which will be set in B tree*/
560 if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE)
561 *want_clusters += ocfs2_clusters_for_bytes(dir->i_sb,
563 if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
564 acl_len > OCFS2_XATTR_INLINE_SIZE) {
565 *want_clusters += ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
567 *want_clusters += ocfs2_clusters_for_bytes(dir->i_sb,
574 static int ocfs2_xattr_extend_allocation(struct inode *inode,
576 struct ocfs2_xattr_value_buf *vb,
577 struct ocfs2_xattr_set_ctxt *ctxt)
580 handle_t *handle = ctxt->handle;
581 enum ocfs2_alloc_restarted why;
582 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
583 u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
584 struct ocfs2_extent_tree et;
586 mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
588 ocfs2_init_xattr_value_extent_tree(&et, inode, vb);
590 status = vb->vb_access(handle, inode, vb->vb_bh,
591 OCFS2_JOURNAL_ACCESS_WRITE);
597 prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
598 status = ocfs2_add_clusters_in_btree(osb,
613 status = ocfs2_journal_dirty(handle, vb->vb_bh);
619 clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) - prev_clusters;
622 * We should have already allocated enough space before the transaction,
623 * so no need to restart.
625 BUG_ON(why != RESTART_NONE || clusters_to_add);
632 static int __ocfs2_remove_xattr_range(struct inode *inode,
633 struct ocfs2_xattr_value_buf *vb,
634 u32 cpos, u32 phys_cpos, u32 len,
635 struct ocfs2_xattr_set_ctxt *ctxt)
638 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
639 handle_t *handle = ctxt->handle;
640 struct ocfs2_extent_tree et;
642 ocfs2_init_xattr_value_extent_tree(&et, inode, vb);
644 ret = vb->vb_access(handle, inode, vb->vb_bh,
645 OCFS2_JOURNAL_ACCESS_WRITE);
651 ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, ctxt->meta_ac,
658 le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
660 ret = ocfs2_journal_dirty(handle, vb->vb_bh);
666 ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc, phys_blkno, len);
674 static int ocfs2_xattr_shrink_size(struct inode *inode,
677 struct ocfs2_xattr_value_buf *vb,
678 struct ocfs2_xattr_set_ctxt *ctxt)
681 u32 trunc_len, cpos, phys_cpos, alloc_size;
684 if (old_clusters <= new_clusters)
688 trunc_len = old_clusters - new_clusters;
690 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
692 &vb->vb_xv->xr_list);
698 if (alloc_size > trunc_len)
699 alloc_size = trunc_len;
701 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
702 phys_cpos, alloc_size,
709 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
710 ocfs2_remove_xattr_clusters_from_cache(inode, block,
713 trunc_len -= alloc_size;
720 static int ocfs2_xattr_value_truncate(struct inode *inode,
721 struct ocfs2_xattr_value_buf *vb,
723 struct ocfs2_xattr_set_ctxt *ctxt)
726 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
727 u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
729 if (new_clusters == old_clusters)
732 if (new_clusters > old_clusters)
733 ret = ocfs2_xattr_extend_allocation(inode,
734 new_clusters - old_clusters,
737 ret = ocfs2_xattr_shrink_size(inode,
738 old_clusters, new_clusters,
744 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
745 size_t *result, const char *prefix,
746 const char *name, int name_len)
748 char *p = buffer + *result;
749 int prefix_len = strlen(prefix);
750 int total_len = prefix_len + name_len + 1;
752 *result += total_len;
754 /* we are just looking for how big our buffer needs to be */
761 memcpy(p, prefix, prefix_len);
762 memcpy(p + prefix_len, name, name_len);
763 p[prefix_len + name_len] = '\0';
768 static int ocfs2_xattr_list_entries(struct inode *inode,
769 struct ocfs2_xattr_header *header,
770 char *buffer, size_t buffer_size)
774 const char *prefix, *name;
776 for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
777 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
778 type = ocfs2_xattr_get_type(entry);
779 prefix = ocfs2_xattr_prefix(type);
782 name = (const char *)header +
783 le16_to_cpu(entry->xe_name_offset);
785 ret = ocfs2_xattr_list_entry(buffer, buffer_size,
786 &result, prefix, name,
796 static int ocfs2_xattr_ibody_list(struct inode *inode,
797 struct ocfs2_dinode *di,
801 struct ocfs2_xattr_header *header = NULL;
802 struct ocfs2_inode_info *oi = OCFS2_I(inode);
805 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
808 header = (struct ocfs2_xattr_header *)
809 ((void *)di + inode->i_sb->s_blocksize -
810 le16_to_cpu(di->i_xattr_inline_size));
812 ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
817 static int ocfs2_xattr_block_list(struct inode *inode,
818 struct ocfs2_dinode *di,
822 struct buffer_head *blk_bh = NULL;
823 struct ocfs2_xattr_block *xb;
826 if (!di->i_xattr_loc)
829 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
836 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
837 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
838 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
839 ret = ocfs2_xattr_list_entries(inode, header,
840 buffer, buffer_size);
842 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
843 ret = ocfs2_xattr_tree_list_index_block(inode, xt,
844 buffer, buffer_size);
852 ssize_t ocfs2_listxattr(struct dentry *dentry,
856 int ret = 0, i_ret = 0, b_ret = 0;
857 struct buffer_head *di_bh = NULL;
858 struct ocfs2_dinode *di = NULL;
859 struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
861 if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
864 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
867 ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
873 di = (struct ocfs2_dinode *)di_bh->b_data;
875 down_read(&oi->ip_xattr_sem);
876 i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
884 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
889 up_read(&oi->ip_xattr_sem);
890 ocfs2_inode_unlock(dentry->d_inode, 0);
894 return i_ret + b_ret;
897 static int ocfs2_xattr_find_entry(int name_index,
899 struct ocfs2_xattr_search *xs)
901 struct ocfs2_xattr_entry *entry;
908 name_len = strlen(name);
910 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
911 cmp = name_index - ocfs2_xattr_get_type(entry);
913 cmp = name_len - entry->xe_name_len;
915 cmp = memcmp(name, (xs->base +
916 le16_to_cpu(entry->xe_name_offset)),
924 return cmp ? -ENODATA : 0;
927 static int ocfs2_xattr_get_value_outside(struct inode *inode,
928 struct ocfs2_xattr_value_root *xv,
932 u32 cpos, p_cluster, num_clusters, bpc, clusters;
935 size_t cplen, blocksize;
936 struct buffer_head *bh = NULL;
937 struct ocfs2_extent_list *el;
940 clusters = le32_to_cpu(xv->xr_clusters);
941 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
942 blocksize = inode->i_sb->s_blocksize;
945 while (cpos < clusters) {
946 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
953 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
954 /* Copy ocfs2_xattr_value */
955 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
956 ret = ocfs2_read_block(inode, blkno, &bh, NULL);
962 cplen = len >= blocksize ? blocksize : len;
963 memcpy(buffer, bh->b_data, cplen);
972 cpos += num_clusters;
978 static int ocfs2_xattr_ibody_get(struct inode *inode,
983 struct ocfs2_xattr_search *xs)
985 struct ocfs2_inode_info *oi = OCFS2_I(inode);
986 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
987 struct ocfs2_xattr_value_root *xv;
991 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
994 xs->end = (void *)di + inode->i_sb->s_blocksize;
995 xs->header = (struct ocfs2_xattr_header *)
996 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
997 xs->base = (void *)xs->header;
998 xs->here = xs->header->xh_entries;
1000 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1003 size = le64_to_cpu(xs->here->xe_value_size);
1005 if (size > buffer_size)
1007 if (ocfs2_xattr_is_local(xs->here)) {
1008 memcpy(buffer, (void *)xs->base +
1009 le16_to_cpu(xs->here->xe_name_offset) +
1010 OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1012 xv = (struct ocfs2_xattr_value_root *)
1013 (xs->base + le16_to_cpu(
1014 xs->here->xe_name_offset) +
1015 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1016 ret = ocfs2_xattr_get_value_outside(inode, xv,
1028 static int ocfs2_xattr_block_get(struct inode *inode,
1033 struct ocfs2_xattr_search *xs)
1035 struct ocfs2_xattr_block *xb;
1036 struct ocfs2_xattr_value_root *xv;
1038 int ret = -ENODATA, name_offset, name_len, block_off, i;
1040 xs->bucket = ocfs2_xattr_bucket_new(inode);
1047 ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1053 if (xs->not_found) {
1058 xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1059 size = le64_to_cpu(xs->here->xe_value_size);
1062 if (size > buffer_size)
1065 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1066 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1067 i = xs->here - xs->header->xh_entries;
1069 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1070 ret = ocfs2_xattr_bucket_get_name_value(inode,
1071 bucket_xh(xs->bucket),
1075 xs->base = bucket_block(xs->bucket, block_off);
1077 if (ocfs2_xattr_is_local(xs->here)) {
1078 memcpy(buffer, (void *)xs->base +
1079 name_offset + name_len, size);
1081 xv = (struct ocfs2_xattr_value_root *)
1082 (xs->base + name_offset + name_len);
1083 ret = ocfs2_xattr_get_value_outside(inode, xv,
1093 ocfs2_xattr_bucket_free(xs->bucket);
1095 brelse(xs->xattr_bh);
1096 xs->xattr_bh = NULL;
1100 int ocfs2_xattr_get_nolock(struct inode *inode,
1101 struct buffer_head *di_bh,
1108 struct ocfs2_dinode *di = NULL;
1109 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1110 struct ocfs2_xattr_search xis = {
1111 .not_found = -ENODATA,
1113 struct ocfs2_xattr_search xbs = {
1114 .not_found = -ENODATA,
1117 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1120 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1123 xis.inode_bh = xbs.inode_bh = di_bh;
1124 di = (struct ocfs2_dinode *)di_bh->b_data;
1126 down_read(&oi->ip_xattr_sem);
1127 ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1129 if (ret == -ENODATA && di->i_xattr_loc)
1130 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1132 up_read(&oi->ip_xattr_sem);
1137 /* ocfs2_xattr_get()
1139 * Copy an extended attribute into the buffer provided.
1140 * Buffer is NULL to compute the size of buffer required.
1142 static int ocfs2_xattr_get(struct inode *inode,
1149 struct buffer_head *di_bh = NULL;
1151 ret = ocfs2_inode_lock(inode, &di_bh, 0);
1156 ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1157 name, buffer, buffer_size);
1159 ocfs2_inode_unlock(inode, 0);
1166 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1168 struct ocfs2_xattr_value_root *xv,
1172 int ret = 0, i, cp_len;
1173 u16 blocksize = inode->i_sb->s_blocksize;
1174 u32 p_cluster, num_clusters;
1175 u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1176 u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1178 struct buffer_head *bh = NULL;
1180 BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1182 while (cpos < clusters) {
1183 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1184 &num_clusters, &xv->xr_list);
1190 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1192 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1193 ret = ocfs2_read_block(inode, blkno, &bh, NULL);
1199 ret = ocfs2_journal_access(handle,
1202 OCFS2_JOURNAL_ACCESS_WRITE);
1208 cp_len = value_len > blocksize ? blocksize : value_len;
1209 memcpy(bh->b_data, value, cp_len);
1210 value_len -= cp_len;
1212 if (cp_len < blocksize)
1213 memset(bh->b_data + cp_len, 0,
1214 blocksize - cp_len);
1216 ret = ocfs2_journal_dirty(handle, bh);
1225 * XXX: do we need to empty all the following
1226 * blocks in this cluster?
1231 cpos += num_clusters;
1239 static int ocfs2_xattr_cleanup(struct inode *inode,
1241 struct ocfs2_xattr_info *xi,
1242 struct ocfs2_xattr_search *xs,
1243 struct ocfs2_xattr_value_buf *vb,
1247 size_t name_len = strlen(xi->name);
1248 void *val = xs->base + offs;
1249 size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1251 ret = vb->vb_access(handle, inode, vb->vb_bh,
1252 OCFS2_JOURNAL_ACCESS_WRITE);
1257 /* Decrease xattr count */
1258 le16_add_cpu(&xs->header->xh_count, -1);
1259 /* Remove the xattr entry and tree root which has already be set*/
1260 memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
1261 memset(val, 0, size);
1263 ret = ocfs2_journal_dirty(handle, vb->vb_bh);
1270 static int ocfs2_xattr_update_entry(struct inode *inode,
1272 struct ocfs2_xattr_info *xi,
1273 struct ocfs2_xattr_search *xs,
1274 struct ocfs2_xattr_value_buf *vb,
1279 ret = vb->vb_access(handle, inode, vb->vb_bh,
1280 OCFS2_JOURNAL_ACCESS_WRITE);
1286 xs->here->xe_name_offset = cpu_to_le16(offs);
1287 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1288 if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1289 ocfs2_xattr_set_local(xs->here, 1);
1291 ocfs2_xattr_set_local(xs->here, 0);
1292 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1294 ret = ocfs2_journal_dirty(handle, vb->vb_bh);
1302 * ocfs2_xattr_set_value_outside()
1304 * Set large size value in B tree.
1306 static int ocfs2_xattr_set_value_outside(struct inode *inode,
1307 struct ocfs2_xattr_info *xi,
1308 struct ocfs2_xattr_search *xs,
1309 struct ocfs2_xattr_set_ctxt *ctxt,
1310 struct ocfs2_xattr_value_buf *vb,
1313 size_t name_len = strlen(xi->name);
1314 void *val = xs->base + offs;
1315 struct ocfs2_xattr_value_root *xv = NULL;
1316 size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1319 memset(val, 0, size);
1320 memcpy(val, xi->name, name_len);
1321 xv = (struct ocfs2_xattr_value_root *)
1322 (val + OCFS2_XATTR_SIZE(name_len));
1323 xv->xr_clusters = 0;
1324 xv->xr_last_eb_blk = 0;
1325 xv->xr_list.l_tree_depth = 0;
1326 xv->xr_list.l_count = cpu_to_le16(1);
1327 xv->xr_list.l_next_free_rec = 0;
1330 ret = ocfs2_xattr_value_truncate(inode, vb, xi->value_len, ctxt);
1335 ret = ocfs2_xattr_update_entry(inode, ctxt->handle, xi, xs, vb, offs);
1340 ret = __ocfs2_xattr_set_value_outside(inode, ctxt->handle, vb->vb_xv,
1341 xi->value, xi->value_len);
1349 * ocfs2_xattr_set_entry_local()
1351 * Set, replace or remove extended attribute in local.
1353 static void ocfs2_xattr_set_entry_local(struct inode *inode,
1354 struct ocfs2_xattr_info *xi,
1355 struct ocfs2_xattr_search *xs,
1356 struct ocfs2_xattr_entry *last,
1359 size_t name_len = strlen(xi->name);
1362 if (xi->value && xs->not_found) {
1363 /* Insert the new xattr entry. */
1364 le16_add_cpu(&xs->header->xh_count, 1);
1365 ocfs2_xattr_set_type(last, xi->name_index);
1366 ocfs2_xattr_set_local(last, 1);
1367 last->xe_name_len = name_len;
1373 first_val = xs->base + min_offs;
1374 offs = le16_to_cpu(xs->here->xe_name_offset);
1375 val = xs->base + offs;
1377 if (le64_to_cpu(xs->here->xe_value_size) >
1378 OCFS2_XATTR_INLINE_SIZE)
1379 size = OCFS2_XATTR_SIZE(name_len) +
1380 OCFS2_XATTR_ROOT_SIZE;
1382 size = OCFS2_XATTR_SIZE(name_len) +
1383 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1385 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1386 OCFS2_XATTR_SIZE(xi->value_len)) {
1387 /* The old and the new value have the
1388 same size. Just replace the value. */
1389 ocfs2_xattr_set_local(xs->here, 1);
1390 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1391 /* Clear value bytes. */
1392 memset(val + OCFS2_XATTR_SIZE(name_len),
1394 OCFS2_XATTR_SIZE(xi->value_len));
1395 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1400 /* Remove the old name+value. */
1401 memmove(first_val + size, first_val, val - first_val);
1402 memset(first_val, 0, size);
1403 xs->here->xe_name_hash = 0;
1404 xs->here->xe_name_offset = 0;
1405 ocfs2_xattr_set_local(xs->here, 1);
1406 xs->here->xe_value_size = 0;
1410 /* Adjust all value offsets. */
1411 last = xs->header->xh_entries;
1412 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1413 size_t o = le16_to_cpu(last->xe_name_offset);
1416 last->xe_name_offset = cpu_to_le16(o + size);
1421 /* Remove the old entry. */
1423 memmove(xs->here, xs->here + 1,
1424 (void *)last - (void *)xs->here);
1425 memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1426 le16_add_cpu(&xs->header->xh_count, -1);
1430 /* Insert the new name+value. */
1431 size_t size = OCFS2_XATTR_SIZE(name_len) +
1432 OCFS2_XATTR_SIZE(xi->value_len);
1433 void *val = xs->base + min_offs - size;
1435 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1436 memset(val, 0, size);
1437 memcpy(val, xi->name, name_len);
1438 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1441 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1442 ocfs2_xattr_set_local(xs->here, 1);
1443 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1450 * ocfs2_xattr_set_entry()
1452 * Set extended attribute entry into inode or block.
1454 * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1455 * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1456 * then set value in B tree with set_value_outside().
1458 static int ocfs2_xattr_set_entry(struct inode *inode,
1459 struct ocfs2_xattr_info *xi,
1460 struct ocfs2_xattr_search *xs,
1461 struct ocfs2_xattr_set_ctxt *ctxt,
1464 struct ocfs2_xattr_entry *last;
1465 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1466 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1467 size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1469 handle_t *handle = ctxt->handle;
1471 struct ocfs2_xattr_info xi_l = {
1472 .name_index = xi->name_index,
1475 .value_len = xi->value_len,
1477 struct ocfs2_xattr_value_buf vb = {
1478 .vb_bh = xs->xattr_bh,
1479 .vb_access = ocfs2_journal_access_di,
1482 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1483 BUG_ON(xs->xattr_bh == xs->inode_bh);
1484 vb.vb_access = ocfs2_journal_access_xb;
1486 BUG_ON(xs->xattr_bh != xs->inode_bh);
1488 /* Compute min_offs, last and free space. */
1489 last = xs->header->xh_entries;
1491 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1492 size_t offs = le16_to_cpu(last->xe_name_offset);
1493 if (offs < min_offs)
1498 free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
1502 if (!xs->not_found) {
1504 if (ocfs2_xattr_is_local(xs->here))
1505 size = OCFS2_XATTR_SIZE(name_len) +
1506 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1508 size = OCFS2_XATTR_SIZE(name_len) +
1509 OCFS2_XATTR_ROOT_SIZE;
1510 free += (size + sizeof(struct ocfs2_xattr_entry));
1512 /* Check free space in inode or block */
1513 if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1514 if (free < sizeof(struct ocfs2_xattr_entry) +
1515 OCFS2_XATTR_SIZE(name_len) +
1516 OCFS2_XATTR_ROOT_SIZE) {
1520 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1521 xi_l.value = (void *)&def_xv;
1522 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1523 } else if (xi->value) {
1524 if (free < sizeof(struct ocfs2_xattr_entry) +
1525 OCFS2_XATTR_SIZE(name_len) +
1526 OCFS2_XATTR_SIZE(xi->value_len)) {
1532 if (!xs->not_found) {
1533 /* For existing extended attribute */
1534 size_t size = OCFS2_XATTR_SIZE(name_len) +
1535 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1536 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1537 void *val = xs->base + offs;
1539 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1540 /* Replace existing local xattr with tree root */
1541 ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1546 } else if (!ocfs2_xattr_is_local(xs->here)) {
1547 /* For existing xattr which has value outside */
1548 vb.vb_xv = (struct ocfs2_xattr_value_root *)
1549 (val + OCFS2_XATTR_SIZE(name_len));
1551 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1553 * If new value need set outside also,
1554 * first truncate old value to new value,
1555 * then set new value with set_value_outside().
1557 ret = ocfs2_xattr_value_truncate(inode,
1566 ret = ocfs2_xattr_update_entry(inode,
1577 ret = __ocfs2_xattr_set_value_outside(inode,
1587 * If new value need set in local,
1588 * just trucate old value to zero.
1590 ret = ocfs2_xattr_value_truncate(inode,
1600 ret = ocfs2_journal_access_di(handle, inode, xs->inode_bh,
1601 OCFS2_JOURNAL_ACCESS_WRITE);
1607 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1608 ret = vb.vb_access(handle, inode, vb.vb_bh,
1609 OCFS2_JOURNAL_ACCESS_WRITE);
1617 * Set value in local, include set tree root in local.
1618 * This is the first step for value size >INLINE_SIZE.
1620 ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1622 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1623 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1630 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1631 (flag & OCFS2_INLINE_XATTR_FL)) {
1632 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1633 unsigned int xattrsize = osb->s_xattr_inline_size;
1636 * Adjust extent record count or inline data size
1637 * to reserve space for extended attribute.
1639 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1640 struct ocfs2_inline_data *idata = &di->id2.i_data;
1641 le16_add_cpu(&idata->id_count, -xattrsize);
1642 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1643 struct ocfs2_extent_list *el = &di->id2.i_list;
1644 le16_add_cpu(&el->l_count, -(xattrsize /
1645 sizeof(struct ocfs2_extent_rec)));
1647 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1649 /* Update xattr flag */
1650 spin_lock(&oi->ip_lock);
1651 oi->ip_dyn_features |= flag;
1652 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1653 spin_unlock(&oi->ip_lock);
1655 ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1659 if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1661 * Set value outside in B tree.
1662 * This is the second step for value size > INLINE_SIZE.
1664 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1665 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, ctxt,
1672 * If set value outside failed, we have to clean
1673 * the junk tree root we have already set in local.
1675 ret2 = ocfs2_xattr_cleanup(inode, ctxt->handle,
1685 static int ocfs2_remove_value_outside(struct inode*inode,
1686 struct ocfs2_xattr_value_buf *vb,
1687 struct ocfs2_xattr_header *header)
1690 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1691 struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
1693 ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
1695 ctxt.handle = ocfs2_start_trans(osb,
1696 ocfs2_remove_extent_credits(osb->sb));
1697 if (IS_ERR(ctxt.handle)) {
1698 ret = PTR_ERR(ctxt.handle);
1703 for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1704 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1706 if (!ocfs2_xattr_is_local(entry)) {
1709 val = (void *)header +
1710 le16_to_cpu(entry->xe_name_offset);
1711 vb->vb_xv = (struct ocfs2_xattr_value_root *)
1712 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
1713 ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
1721 ocfs2_commit_trans(osb, ctxt.handle);
1722 ocfs2_schedule_truncate_log_flush(osb, 1);
1723 ocfs2_run_deallocs(osb, &ctxt.dealloc);
1728 static int ocfs2_xattr_ibody_remove(struct inode *inode,
1729 struct buffer_head *di_bh)
1732 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1733 struct ocfs2_xattr_header *header;
1735 struct ocfs2_xattr_value_buf vb = {
1737 .vb_access = ocfs2_journal_access_di,
1740 header = (struct ocfs2_xattr_header *)
1741 ((void *)di + inode->i_sb->s_blocksize -
1742 le16_to_cpu(di->i_xattr_inline_size));
1744 ret = ocfs2_remove_value_outside(inode, &vb, header);
1749 static int ocfs2_xattr_block_remove(struct inode *inode,
1750 struct buffer_head *blk_bh)
1752 struct ocfs2_xattr_block *xb;
1754 struct ocfs2_xattr_value_buf vb = {
1756 .vb_access = ocfs2_journal_access_xb,
1759 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1760 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1761 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1762 ret = ocfs2_remove_value_outside(inode, &vb, header);
1764 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
1769 static int ocfs2_xattr_free_block(struct inode *inode,
1772 struct inode *xb_alloc_inode;
1773 struct buffer_head *xb_alloc_bh = NULL;
1774 struct buffer_head *blk_bh = NULL;
1775 struct ocfs2_xattr_block *xb;
1776 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1782 ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
1788 ret = ocfs2_xattr_block_remove(inode, blk_bh);
1794 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1795 blk = le64_to_cpu(xb->xb_blkno);
1796 bit = le16_to_cpu(xb->xb_suballoc_bit);
1797 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1799 xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1800 EXTENT_ALLOC_SYSTEM_INODE,
1801 le16_to_cpu(xb->xb_suballoc_slot));
1802 if (!xb_alloc_inode) {
1807 mutex_lock(&xb_alloc_inode->i_mutex);
1809 ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1815 handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
1816 if (IS_ERR(handle)) {
1817 ret = PTR_ERR(handle);
1822 ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1827 ocfs2_commit_trans(osb, handle);
1829 ocfs2_inode_unlock(xb_alloc_inode, 1);
1830 brelse(xb_alloc_bh);
1832 mutex_unlock(&xb_alloc_inode->i_mutex);
1833 iput(xb_alloc_inode);
1840 * ocfs2_xattr_remove()
1842 * Free extended attribute resources associated with this inode.
1844 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1846 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1847 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1851 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1854 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1857 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1858 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1865 if (di->i_xattr_loc) {
1866 ret = ocfs2_xattr_free_block(inode,
1867 le64_to_cpu(di->i_xattr_loc));
1874 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1875 OCFS2_INODE_UPDATE_CREDITS);
1876 if (IS_ERR(handle)) {
1877 ret = PTR_ERR(handle);
1881 ret = ocfs2_journal_access_di(handle, inode, di_bh,
1882 OCFS2_JOURNAL_ACCESS_WRITE);
1888 di->i_xattr_loc = 0;
1890 spin_lock(&oi->ip_lock);
1891 oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1892 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1893 spin_unlock(&oi->ip_lock);
1895 ret = ocfs2_journal_dirty(handle, di_bh);
1899 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1904 static int ocfs2_xattr_has_space_inline(struct inode *inode,
1905 struct ocfs2_dinode *di)
1907 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1908 unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1911 if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1914 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1915 struct ocfs2_inline_data *idata = &di->id2.i_data;
1916 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1917 } else if (ocfs2_inode_is_fast_symlink(inode)) {
1918 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1919 le64_to_cpu(di->i_size);
1921 struct ocfs2_extent_list *el = &di->id2.i_list;
1922 free = (le16_to_cpu(el->l_count) -
1923 le16_to_cpu(el->l_next_free_rec)) *
1924 sizeof(struct ocfs2_extent_rec);
1926 if (free >= xattrsize)
1933 * ocfs2_xattr_ibody_find()
1935 * Find extended attribute in inode block and
1936 * fill search info into struct ocfs2_xattr_search.
1938 static int ocfs2_xattr_ibody_find(struct inode *inode,
1941 struct ocfs2_xattr_search *xs)
1943 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1944 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1948 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1951 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1952 down_read(&oi->ip_alloc_sem);
1953 has_space = ocfs2_xattr_has_space_inline(inode, di);
1954 up_read(&oi->ip_alloc_sem);
1959 xs->xattr_bh = xs->inode_bh;
1960 xs->end = (void *)di + inode->i_sb->s_blocksize;
1961 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1962 xs->header = (struct ocfs2_xattr_header *)
1963 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1965 xs->header = (struct ocfs2_xattr_header *)
1966 (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
1967 xs->base = (void *)xs->header;
1968 xs->here = xs->header->xh_entries;
1970 /* Find the named attribute. */
1971 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1972 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1973 if (ret && ret != -ENODATA)
1975 xs->not_found = ret;
1982 * ocfs2_xattr_ibody_set()
1984 * Set, replace or remove an extended attribute into inode block.
1987 static int ocfs2_xattr_ibody_set(struct inode *inode,
1988 struct ocfs2_xattr_info *xi,
1989 struct ocfs2_xattr_search *xs,
1990 struct ocfs2_xattr_set_ctxt *ctxt)
1992 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1993 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1996 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1999 down_write(&oi->ip_alloc_sem);
2000 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2001 if (!ocfs2_xattr_has_space_inline(inode, di)) {
2007 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2008 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
2010 up_write(&oi->ip_alloc_sem);
2016 * ocfs2_xattr_block_find()
2018 * Find extended attribute in external block and
2019 * fill search info into struct ocfs2_xattr_search.
2021 static int ocfs2_xattr_block_find(struct inode *inode,
2024 struct ocfs2_xattr_search *xs)
2026 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2027 struct buffer_head *blk_bh = NULL;
2028 struct ocfs2_xattr_block *xb;
2031 if (!di->i_xattr_loc)
2034 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2041 xs->xattr_bh = blk_bh;
2042 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2044 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2045 xs->header = &xb->xb_attrs.xb_header;
2046 xs->base = (void *)xs->header;
2047 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2048 xs->here = xs->header->xh_entries;
2050 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2052 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2056 if (ret && ret != -ENODATA) {
2057 xs->xattr_bh = NULL;
2060 xs->not_found = ret;
2069 * ocfs2_xattr_block_set()
2071 * Set, replace or remove an extended attribute into external block.
2074 static int ocfs2_xattr_block_set(struct inode *inode,
2075 struct ocfs2_xattr_info *xi,
2076 struct ocfs2_xattr_search *xs,
2077 struct ocfs2_xattr_set_ctxt *ctxt)
2079 struct buffer_head *new_bh = NULL;
2080 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2081 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2082 handle_t *handle = ctxt->handle;
2083 struct ocfs2_xattr_block *xblk = NULL;
2084 u16 suballoc_bit_start;
2089 if (!xs->xattr_bh) {
2090 ret = ocfs2_journal_access_di(handle, inode, xs->inode_bh,
2091 OCFS2_JOURNAL_ACCESS_CREATE);
2097 ret = ocfs2_claim_metadata(osb, handle, ctxt->meta_ac, 1,
2098 &suballoc_bit_start, &num_got,
2105 new_bh = sb_getblk(inode->i_sb, first_blkno);
2106 ocfs2_set_new_buffer_uptodate(inode, new_bh);
2108 ret = ocfs2_journal_access_xb(handle, inode, new_bh,
2109 OCFS2_JOURNAL_ACCESS_CREATE);
2115 /* Initialize ocfs2_xattr_block */
2116 xs->xattr_bh = new_bh;
2117 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2118 memset(xblk, 0, inode->i_sb->s_blocksize);
2119 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2120 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
2121 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2122 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
2123 xblk->xb_blkno = cpu_to_le64(first_blkno);
2125 xs->header = &xblk->xb_attrs.xb_header;
2126 xs->base = (void *)xs->header;
2127 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2128 xs->here = xs->header->xh_entries;
2130 ret = ocfs2_journal_dirty(handle, new_bh);
2135 di->i_xattr_loc = cpu_to_le64(first_blkno);
2136 ocfs2_journal_dirty(handle, xs->inode_bh);
2138 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2140 if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2141 /* Set extended attribute into external block */
2142 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2143 OCFS2_HAS_XATTR_FL);
2144 if (!ret || ret != -ENOSPC)
2147 ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2152 ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2159 /* Check whether the new xattr can be inserted into the inode. */
2160 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2161 struct ocfs2_xattr_info *xi,
2162 struct ocfs2_xattr_search *xs)
2165 struct ocfs2_xattr_entry *last;
2167 size_t min_offs = xs->end - xs->base;
2172 last = xs->header->xh_entries;
2174 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
2175 size_t offs = le16_to_cpu(last->xe_name_offset);
2176 if (offs < min_offs)
2181 free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
2185 BUG_ON(!xs->not_found);
2187 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2188 value_size = OCFS2_XATTR_ROOT_SIZE;
2190 value_size = OCFS2_XATTR_SIZE(xi->value_len);
2192 if (free >= sizeof(struct ocfs2_xattr_entry) +
2193 OCFS2_XATTR_SIZE(strlen(xi->name)) + value_size)
2199 static int ocfs2_calc_xattr_set_need(struct inode *inode,
2200 struct ocfs2_dinode *di,
2201 struct ocfs2_xattr_info *xi,
2202 struct ocfs2_xattr_search *xis,
2203 struct ocfs2_xattr_search *xbs,
2208 int ret = 0, old_in_xb = 0;
2209 int clusters_add = 0, meta_add = 0, credits = 0;
2210 struct buffer_head *bh = NULL;
2211 struct ocfs2_xattr_block *xb = NULL;
2212 struct ocfs2_xattr_entry *xe = NULL;
2213 struct ocfs2_xattr_value_root *xv = NULL;
2215 int name_offset, name_len = 0;
2216 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2221 * Calculate the clusters we need to write.
2222 * No matter whether we replace an old one or add a new one,
2223 * we need this for writing.
2225 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2226 credits += new_clusters *
2227 ocfs2_clusters_to_blocks(inode->i_sb, 1);
2229 if (xis->not_found && xbs->not_found) {
2230 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2232 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2233 clusters_add += new_clusters;
2234 credits += ocfs2_calc_extend_credits(inode->i_sb,
2242 if (!xis->not_found) {
2244 name_offset = le16_to_cpu(xe->xe_name_offset);
2245 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2247 credits += OCFS2_INODE_UPDATE_CREDITS;
2249 int i, block_off = 0;
2250 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2252 name_offset = le16_to_cpu(xe->xe_name_offset);
2253 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2254 i = xbs->here - xbs->header->xh_entries;
2257 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2258 ret = ocfs2_xattr_bucket_get_name_value(inode,
2259 bucket_xh(xbs->bucket),
2262 base = bucket_block(xbs->bucket, block_off);
2263 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2266 credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
2271 * delete a xattr doesn't need metadata and cluster allocation.
2272 * so just calculate the credits and return.
2274 * The credits for removing the value tree will be extended
2275 * by ocfs2_remove_extent itself.
2278 if (!ocfs2_xattr_is_local(xe))
2279 credits += ocfs2_remove_extent_credits(inode->i_sb);
2284 /* do cluster allocation guess first. */
2285 value_size = le64_to_cpu(xe->xe_value_size);
2289 * In xattr set, we always try to set the xe in inode first,
2290 * so if it can be inserted into inode successfully, the old
2291 * one will be removed from the xattr block, and this xattr
2292 * will be inserted into inode as a new xattr in inode.
2294 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
2295 clusters_add += new_clusters;
2296 credits += ocfs2_remove_extent_credits(inode->i_sb) +
2297 OCFS2_INODE_UPDATE_CREDITS;
2298 if (!ocfs2_xattr_is_local(xe))
2299 credits += ocfs2_calc_extend_credits(
2307 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2308 /* the new values will be stored outside. */
2309 u32 old_clusters = 0;
2311 if (!ocfs2_xattr_is_local(xe)) {
2312 old_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2314 xv = (struct ocfs2_xattr_value_root *)
2315 (base + name_offset + name_len);
2316 value_size = OCFS2_XATTR_ROOT_SIZE;
2320 if (old_clusters >= new_clusters) {
2321 credits += ocfs2_remove_extent_credits(inode->i_sb);
2324 meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
2325 clusters_add += new_clusters - old_clusters;
2326 credits += ocfs2_calc_extend_credits(inode->i_sb,
2330 if (value_size >= OCFS2_XATTR_ROOT_SIZE)
2335 * Now the new value will be stored inside. So if the new
2336 * value is smaller than the size of value root or the old
2337 * value, we don't need any allocation, otherwise we have
2338 * to guess metadata allocation.
2340 if ((ocfs2_xattr_is_local(xe) && value_size >= xi->value_len) ||
2341 (!ocfs2_xattr_is_local(xe) &&
2342 OCFS2_XATTR_ROOT_SIZE >= xi->value_len))
2347 /* calculate metadata allocation. */
2348 if (di->i_xattr_loc) {
2349 if (!xbs->xattr_bh) {
2350 ret = ocfs2_read_xattr_block(inode,
2351 le64_to_cpu(di->i_xattr_loc),
2358 xb = (struct ocfs2_xattr_block *)bh->b_data;
2360 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2363 * If there is already an xattr tree, good, we can calculate
2364 * like other b-trees. Otherwise we may have the chance of
2365 * create a tree, the credit calculation is borrowed from
2366 * ocfs2_calc_extend_credits with root_el = NULL. And the
2367 * new tree will be cluster based, so no meta is needed.
2369 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2370 struct ocfs2_extent_list *el =
2371 &xb->xb_attrs.xb_root.xt_list;
2372 meta_add += ocfs2_extend_meta_needed(el);
2373 credits += ocfs2_calc_extend_credits(inode->i_sb,
2376 credits += OCFS2_SUBALLOC_ALLOC + 1;
2379 * This cluster will be used either for new bucket or for
2381 * If the cluster size is the same as the bucket size, one
2382 * more is needed since we may need to extend the bucket
2386 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2387 if (OCFS2_XATTR_BUCKET_SIZE ==
2388 OCFS2_SB(inode->i_sb)->s_clustersize) {
2389 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2394 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
2398 *clusters_need = clusters_add;
2400 *meta_need = meta_add;
2402 *credits_need = credits;
2407 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
2408 struct ocfs2_dinode *di,
2409 struct ocfs2_xattr_info *xi,
2410 struct ocfs2_xattr_search *xis,
2411 struct ocfs2_xattr_search *xbs,
2412 struct ocfs2_xattr_set_ctxt *ctxt,
2415 int clusters_add, meta_add, ret;
2416 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2418 memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
2420 ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
2422 ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
2423 &clusters_add, &meta_add, credits);
2429 mlog(0, "Set xattr %s, reserve meta blocks = %d, clusters = %d, "
2430 "credits = %d\n", xi->name, meta_add, clusters_add, *credits);
2433 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
2442 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
2448 if (ctxt->meta_ac) {
2449 ocfs2_free_alloc_context(ctxt->meta_ac);
2450 ctxt->meta_ac = NULL;
2454 * We cannot have an error and a non null ctxt->data_ac.
2461 static int __ocfs2_xattr_set_handle(struct inode *inode,
2462 struct ocfs2_dinode *di,
2463 struct ocfs2_xattr_info *xi,
2464 struct ocfs2_xattr_search *xis,
2465 struct ocfs2_xattr_search *xbs,
2466 struct ocfs2_xattr_set_ctxt *ctxt)
2468 int ret = 0, credits, old_found;
2471 /* Remove existing extended attribute */
2472 if (!xis->not_found)
2473 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2474 else if (!xbs->not_found)
2475 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2477 /* We always try to set extended attribute into inode first*/
2478 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2479 if (!ret && !xbs->not_found) {
2481 * If succeed and that extended attribute existing in
2482 * external block, then we will remove it.
2487 old_found = xis->not_found;
2488 xis->not_found = -ENODATA;
2489 ret = ocfs2_calc_xattr_set_need(inode,
2497 xis->not_found = old_found;
2503 ret = ocfs2_extend_trans(ctxt->handle, credits +
2504 ctxt->handle->h_buffer_credits);
2509 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2510 } else if (ret == -ENOSPC) {
2511 if (di->i_xattr_loc && !xbs->xattr_bh) {
2512 ret = ocfs2_xattr_block_find(inode,
2518 old_found = xis->not_found;
2519 xis->not_found = -ENODATA;
2520 ret = ocfs2_calc_xattr_set_need(inode,
2528 xis->not_found = old_found;
2534 ret = ocfs2_extend_trans(ctxt->handle, credits +
2535 ctxt->handle->h_buffer_credits);
2542 * If no space in inode, we will set extended attribute
2543 * into external block.
2545 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2548 if (!xis->not_found) {
2550 * If succeed and that extended attribute
2551 * existing in inode, we will remove it.
2555 xbs->not_found = -ENODATA;
2556 ret = ocfs2_calc_xattr_set_need(inode,
2569 ret = ocfs2_extend_trans(ctxt->handle, credits +
2570 ctxt->handle->h_buffer_credits);
2575 ret = ocfs2_xattr_ibody_set(inode, xi,
2582 /* Update inode ctime. */
2583 ret = ocfs2_journal_access(ctxt->handle, inode, xis->inode_bh,
2584 OCFS2_JOURNAL_ACCESS_WRITE);
2590 inode->i_ctime = CURRENT_TIME;
2591 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
2592 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
2593 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
2600 * This function only called duing creating inode
2601 * for init security/acl xattrs of the new inode.
2602 * The xattrs could be put into ibody or extent block,
2603 * xattr bucket would not be use in this case.
2604 * transanction credits also be reserved in here.
2606 int ocfs2_xattr_set_handle(handle_t *handle,
2607 struct inode *inode,
2608 struct buffer_head *di_bh,
2614 struct ocfs2_alloc_context *meta_ac,
2615 struct ocfs2_alloc_context *data_ac)
2617 struct ocfs2_dinode *di;
2620 struct ocfs2_xattr_info xi = {
2621 .name_index = name_index,
2624 .value_len = value_len,
2627 struct ocfs2_xattr_search xis = {
2628 .not_found = -ENODATA,
2631 struct ocfs2_xattr_search xbs = {
2632 .not_found = -ENODATA,
2635 struct ocfs2_xattr_set_ctxt ctxt = {
2641 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2644 xis.inode_bh = xbs.inode_bh = di_bh;
2645 di = (struct ocfs2_dinode *)di_bh->b_data;
2647 down_write(&OCFS2_I(inode)->ip_xattr_sem);
2649 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2652 if (xis.not_found) {
2653 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2658 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2661 up_write(&OCFS2_I(inode)->ip_xattr_sem);
2662 brelse(xbs.xattr_bh);
2670 * Set, replace or remove an extended attribute for this inode.
2671 * value is NULL to remove an existing extended attribute, else either
2672 * create or replace an extended attribute.
2674 int ocfs2_xattr_set(struct inode *inode,
2681 struct buffer_head *di_bh = NULL;
2682 struct ocfs2_dinode *di;
2684 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2685 struct inode *tl_inode = osb->osb_tl_inode;
2686 struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
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 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2707 * Only xbs will be used on indexed trees. xis doesn't need a
2710 xbs.bucket = ocfs2_xattr_bucket_new(inode);
2712 mlog_errno(-ENOMEM);
2716 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2719 goto cleanup_nolock;
2721 xis.inode_bh = xbs.inode_bh = di_bh;
2722 di = (struct ocfs2_dinode *)di_bh->b_data;
2724 down_write(&OCFS2_I(inode)->ip_xattr_sem);
2726 * Scan inode and external block to find the same name
2727 * extended attribute and collect search infomation.
2729 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2732 if (xis.not_found) {
2733 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2738 if (xis.not_found && xbs.not_found) {
2740 if (flags & XATTR_REPLACE)
2747 if (flags & XATTR_CREATE)
2752 mutex_lock(&tl_inode->i_mutex);
2754 if (ocfs2_truncate_log_needs_flush(osb)) {
2755 ret = __ocfs2_flush_truncate_log(osb);
2757 mutex_unlock(&tl_inode->i_mutex);
2762 mutex_unlock(&tl_inode->i_mutex);
2764 ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
2765 &xbs, &ctxt, &credits);
2771 /* we need to update inode's ctime field, so add credit for it. */
2772 credits += OCFS2_INODE_UPDATE_CREDITS;
2773 ctxt.handle = ocfs2_start_trans(osb, credits);
2774 if (IS_ERR(ctxt.handle)) {
2775 ret = PTR_ERR(ctxt.handle);
2780 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2782 ocfs2_commit_trans(osb, ctxt.handle);
2785 ocfs2_free_alloc_context(ctxt.data_ac);
2787 ocfs2_free_alloc_context(ctxt.meta_ac);
2788 if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
2789 ocfs2_schedule_truncate_log_flush(osb, 1);
2790 ocfs2_run_deallocs(osb, &ctxt.dealloc);
2792 up_write(&OCFS2_I(inode)->ip_xattr_sem);
2793 ocfs2_inode_unlock(inode, 1);
2796 brelse(xbs.xattr_bh);
2797 ocfs2_xattr_bucket_free(xbs.bucket);
2803 * Find the xattr extent rec which may contains name_hash.
2804 * e_cpos will be the first name hash of the xattr rec.
2805 * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2807 static int ocfs2_xattr_get_rec(struct inode *inode,
2812 struct ocfs2_extent_list *el)
2815 struct buffer_head *eb_bh = NULL;
2816 struct ocfs2_extent_block *eb;
2817 struct ocfs2_extent_rec *rec = NULL;
2820 if (el->l_tree_depth) {
2821 ret = ocfs2_find_leaf(inode, el, name_hash, &eb_bh);
2827 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2830 if (el->l_tree_depth) {
2831 ocfs2_error(inode->i_sb,
2832 "Inode %lu has non zero tree depth in "
2833 "xattr tree block %llu\n", inode->i_ino,
2834 (unsigned long long)eb_bh->b_blocknr);
2840 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2841 rec = &el->l_recs[i];
2843 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2844 e_blkno = le64_to_cpu(rec->e_blkno);
2850 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2851 "record (%u, %u, 0) in xattr", inode->i_ino,
2852 le32_to_cpu(rec->e_cpos),
2853 ocfs2_rec_clusters(el, rec));
2858 *p_blkno = le64_to_cpu(rec->e_blkno);
2859 *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2861 *e_cpos = le32_to_cpu(rec->e_cpos);
2867 typedef int (xattr_bucket_func)(struct inode *inode,
2868 struct ocfs2_xattr_bucket *bucket,
2871 static int ocfs2_find_xe_in_bucket(struct inode *inode,
2872 struct ocfs2_xattr_bucket *bucket,
2879 int i, ret = 0, cmp = 1, block_off, new_offset;
2880 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
2881 size_t name_len = strlen(name);
2882 struct ocfs2_xattr_entry *xe = NULL;
2886 * We don't use binary search in the bucket because there
2887 * may be multiple entries with the same name hash.
2889 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2890 xe = &xh->xh_entries[i];
2892 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2894 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2897 cmp = name_index - ocfs2_xattr_get_type(xe);
2899 cmp = name_len - xe->xe_name_len;
2903 ret = ocfs2_xattr_bucket_get_name_value(inode,
2914 xe_name = bucket_block(bucket, block_off) + new_offset;
2915 if (!memcmp(name, xe_name, name_len)) {
2927 * Find the specified xattr entry in a series of buckets.
2928 * This series start from p_blkno and last for num_clusters.
2929 * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2930 * the num of the valid buckets.
2932 * Return the buffer_head this xattr should reside in. And if the xattr's
2933 * hash is in the gap of 2 buckets, return the lower bucket.
2935 static int ocfs2_xattr_bucket_find(struct inode *inode,
2942 struct ocfs2_xattr_search *xs)
2945 struct ocfs2_xattr_header *xh = NULL;
2946 struct ocfs2_xattr_entry *xe = NULL;
2948 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2949 int low_bucket = 0, bucket, high_bucket;
2950 struct ocfs2_xattr_bucket *search;
2952 u64 blkno, lower_blkno = 0;
2954 search = ocfs2_xattr_bucket_new(inode);
2961 ret = ocfs2_read_xattr_bucket(search, p_blkno);
2967 xh = bucket_xh(search);
2968 high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
2969 while (low_bucket <= high_bucket) {
2970 ocfs2_xattr_bucket_relse(search);
2972 bucket = (low_bucket + high_bucket) / 2;
2973 blkno = p_blkno + bucket * blk_per_bucket;
2974 ret = ocfs2_read_xattr_bucket(search, blkno);
2980 xh = bucket_xh(search);
2981 xe = &xh->xh_entries[0];
2982 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
2983 high_bucket = bucket - 1;
2988 * Check whether the hash of the last entry in our
2989 * bucket is larger than the search one. for an empty
2990 * bucket, the last one is also the first one.
2993 xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
2995 last_hash = le32_to_cpu(xe->xe_name_hash);
2997 /* record lower_blkno which may be the insert place. */
2998 lower_blkno = blkno;
3000 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3001 low_bucket = bucket + 1;
3005 /* the searched xattr should reside in this bucket if exists. */
3006 ret = ocfs2_find_xe_in_bucket(inode, search,
3007 name_index, name, name_hash,
3017 * Record the bucket we have found.
3018 * When the xattr's hash value is in the gap of 2 buckets, we will
3019 * always set it to the previous bucket.
3022 lower_blkno = p_blkno;
3024 /* This should be in cache - we just read it during the search */
3025 ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3031 xs->header = bucket_xh(xs->bucket);
3032 xs->base = bucket_block(xs->bucket, 0);
3033 xs->end = xs->base + inode->i_sb->s_blocksize;
3036 xs->here = &xs->header->xh_entries[index];
3037 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
3038 (unsigned long long)bucket_blkno(xs->bucket), index);
3043 ocfs2_xattr_bucket_free(search);
3047 static int ocfs2_xattr_index_block_find(struct inode *inode,
3048 struct buffer_head *root_bh,
3051 struct ocfs2_xattr_search *xs)
3054 struct ocfs2_xattr_block *xb =
3055 (struct ocfs2_xattr_block *)root_bh->b_data;
3056 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3057 struct ocfs2_extent_list *el = &xb_root->xt_list;
3059 u32 first_hash, num_clusters = 0;
3060 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3062 if (le16_to_cpu(el->l_next_free_rec) == 0)
3065 mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
3066 name, name_hash, name_index);
3068 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3075 BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3077 mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
3078 "in the rec is %u\n", num_clusters, (unsigned long long)p_blkno,
3081 ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3082 p_blkno, first_hash, num_clusters, xs);
3088 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3091 xattr_bucket_func *func,
3095 u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3096 u32 num_buckets = clusters * bpc;
3097 struct ocfs2_xattr_bucket *bucket;
3099 bucket = ocfs2_xattr_bucket_new(inode);
3101 mlog_errno(-ENOMEM);
3105 mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
3106 clusters, (unsigned long long)blkno);
3108 for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3109 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3116 * The real bucket num in this series of blocks is stored
3117 * in the 1st bucket.
3120 num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3122 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
3123 (unsigned long long)blkno,
3124 le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3126 ret = func(inode, bucket, para);
3129 /* Fall through to bucket_relse() */
3132 ocfs2_xattr_bucket_relse(bucket);
3137 ocfs2_xattr_bucket_free(bucket);
3141 struct ocfs2_xattr_tree_list {
3147 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
3148 struct ocfs2_xattr_header *xh,
3155 if (index < 0 || index >= le16_to_cpu(xh->xh_count))
3158 name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
3160 *block_off = name_offset >> inode->i_sb->s_blocksize_bits;
3161 *new_offset = name_offset % inode->i_sb->s_blocksize;
3166 static int ocfs2_list_xattr_bucket(struct inode *inode,
3167 struct ocfs2_xattr_bucket *bucket,
3171 struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
3172 int i, block_off, new_offset;
3173 const char *prefix, *name;
3175 for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
3176 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
3177 type = ocfs2_xattr_get_type(entry);
3178 prefix = ocfs2_xattr_prefix(type);
3181 ret = ocfs2_xattr_bucket_get_name_value(inode,
3189 name = (const char *)bucket_block(bucket, block_off) +
3191 ret = ocfs2_xattr_list_entry(xl->buffer,
3195 entry->xe_name_len);
3204 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
3205 struct ocfs2_xattr_tree_root *xt,
3209 struct ocfs2_extent_list *el = &xt->xt_list;
3211 u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
3213 struct ocfs2_xattr_tree_list xl = {
3215 .buffer_size = buffer_size,
3219 if (le16_to_cpu(el->l_next_free_rec) == 0)
3222 while (name_hash > 0) {
3223 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
3224 &e_cpos, &num_clusters, el);
3230 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
3231 ocfs2_list_xattr_bucket,
3241 name_hash = e_cpos - 1;
3249 static int cmp_xe(const void *a, const void *b)
3251 const struct ocfs2_xattr_entry *l = a, *r = b;
3252 u32 l_hash = le32_to_cpu(l->xe_name_hash);
3253 u32 r_hash = le32_to_cpu(r->xe_name_hash);
3255 if (l_hash > r_hash)
3257 if (l_hash < r_hash)
3262 static void swap_xe(void *a, void *b, int size)
3264 struct ocfs2_xattr_entry *l = a, *r = b, tmp;
3267 memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
3268 memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
3272 * When the ocfs2_xattr_block is filled up, new bucket will be created
3273 * and all the xattr entries will be moved to the new bucket.
3274 * The header goes at the start of the bucket, and the names+values are
3275 * filled from the end. This is why *target starts as the last buffer.
3276 * Note: we need to sort the entries since they are not saved in order
3277 * in the ocfs2_xattr_block.
3279 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
3280 struct buffer_head *xb_bh,
3281 struct ocfs2_xattr_bucket *bucket)
3283 int i, blocksize = inode->i_sb->s_blocksize;
3284 int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3285 u16 offset, size, off_change;
3286 struct ocfs2_xattr_entry *xe;
3287 struct ocfs2_xattr_block *xb =
3288 (struct ocfs2_xattr_block *)xb_bh->b_data;
3289 struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
3290 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3291 u16 count = le16_to_cpu(xb_xh->xh_count);
3292 char *src = xb_bh->b_data;
3293 char *target = bucket_block(bucket, blks - 1);
3295 mlog(0, "cp xattr from block %llu to bucket %llu\n",
3296 (unsigned long long)xb_bh->b_blocknr,
3297 (unsigned long long)bucket_blkno(bucket));
3299 for (i = 0; i < blks; i++)
3300 memset(bucket_block(bucket, i), 0, blocksize);
3303 * Since the xe_name_offset is based on ocfs2_xattr_header,
3304 * there is a offset change corresponding to the change of
3305 * ocfs2_xattr_header's position.
3307 off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3308 xe = &xb_xh->xh_entries[count - 1];
3309 offset = le16_to_cpu(xe->xe_name_offset) + off_change;
3310 size = blocksize - offset;
3312 /* copy all the names and values. */
3313 memcpy(target + offset, src + offset, size);
3315 /* Init new header now. */
3316 xh->xh_count = xb_xh->xh_count;
3317 xh->xh_num_buckets = cpu_to_le16(1);
3318 xh->xh_name_value_len = cpu_to_le16(size);
3319 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
3321 /* copy all the entries. */
3322 target = bucket_block(bucket, 0);
3323 offset = offsetof(struct ocfs2_xattr_header, xh_entries);
3324 size = count * sizeof(struct ocfs2_xattr_entry);
3325 memcpy(target + offset, (char *)xb_xh + offset, size);
3327 /* Change the xe offset for all the xe because of the move. */
3328 off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
3329 offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3330 for (i = 0; i < count; i++)
3331 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
3333 mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
3334 offset, size, off_change);
3336 sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
3341 * After we move xattr from block to index btree, we have to
3342 * update ocfs2_xattr_search to the new xe and base.
3344 * When the entry is in xattr block, xattr_bh indicates the storage place.
3345 * While if the entry is in index b-tree, "bucket" indicates the
3346 * real place of the xattr.
3348 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
3349 struct ocfs2_xattr_search *xs,
3350 struct buffer_head *old_bh)
3352 char *buf = old_bh->b_data;
3353 struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
3354 struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
3357 xs->header = bucket_xh(xs->bucket);
3358 xs->base = bucket_block(xs->bucket, 0);
3359 xs->end = xs->base + inode->i_sb->s_blocksize;
3364 i = xs->here - old_xh->xh_entries;
3365 xs->here = &xs->header->xh_entries[i];
3368 static int ocfs2_xattr_create_index_block(struct inode *inode,
3369 struct ocfs2_xattr_search *xs,
3370 struct ocfs2_xattr_set_ctxt *ctxt)
3375 handle_t *handle = ctxt->handle;
3376 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3377 struct ocfs2_inode_info *oi = OCFS2_I(inode);
3378 struct buffer_head *xb_bh = xs->xattr_bh;
3379 struct ocfs2_xattr_block *xb =
3380 (struct ocfs2_xattr_block *)xb_bh->b_data;
3381 struct ocfs2_xattr_tree_root *xr;
3382 u16 xb_flags = le16_to_cpu(xb->xb_flags);
3384 mlog(0, "create xattr index block for %llu\n",
3385 (unsigned long long)xb_bh->b_blocknr);
3387 BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
3388 BUG_ON(!xs->bucket);
3392 * We can use this lock for now, and maybe move to a dedicated mutex
3393 * if performance becomes a problem later.
3395 down_write(&oi->ip_alloc_sem);
3397 ret = ocfs2_journal_access_xb(handle, inode, xb_bh,
3398 OCFS2_JOURNAL_ACCESS_WRITE);
3404 ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac,
3405 1, 1, &bit_off, &len);
3412 * The bucket may spread in many blocks, and
3413 * we will only touch the 1st block and the last block
3414 * in the whole bucket(one for entry and one for data).
3416 blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
3418 mlog(0, "allocate 1 cluster from %llu to xattr block\n",
3419 (unsigned long long)blkno);
3421 ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
3427 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
3428 OCFS2_JOURNAL_ACCESS_CREATE);
3434 ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
3435 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
3437 ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
3439 /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
3440 memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
3441 offsetof(struct ocfs2_xattr_block, xb_attrs));
3443 xr = &xb->xb_attrs.xb_root;
3444 xr->xt_clusters = cpu_to_le32(1);
3445 xr->xt_last_eb_blk = 0;
3446 xr->xt_list.l_tree_depth = 0;
3447 xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
3448 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
3450 xr->xt_list.l_recs[0].e_cpos = 0;
3451 xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
3452 xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
3454 xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
3456 ocfs2_journal_dirty(handle, xb_bh);
3459 up_write(&oi->ip_alloc_sem);
3464 static int cmp_xe_offset(const void *a, const void *b)
3466 const struct ocfs2_xattr_entry *l = a, *r = b;
3467 u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
3468 u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
3470 if (l_name_offset < r_name_offset)
3472 if (l_name_offset > r_name_offset)
3478 * defrag a xattr bucket if we find that the bucket has some
3479 * holes beteen name/value pairs.
3480 * We will move all the name/value pairs to the end of the bucket
3481 * so that we can spare some space for insertion.
3483 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
3485 struct ocfs2_xattr_bucket *bucket)
3488 size_t end, offset, len, value_len;
3489 struct ocfs2_xattr_header *xh;
3490 char *entries, *buf, *bucket_buf = NULL;
3491 u64 blkno = bucket_blkno(bucket);
3493 size_t blocksize = inode->i_sb->s_blocksize;
3494 struct ocfs2_xattr_entry *xe;
3497 * In order to make the operation more efficient and generic,
3498 * we copy all the blocks into a contiguous memory and do the
3499 * defragment there, so if anything is error, we will not touch
3502 bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
3509 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3510 memcpy(buf, bucket_block(bucket, i), blocksize);
3512 ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
3513 OCFS2_JOURNAL_ACCESS_WRITE);
3519 xh = (struct ocfs2_xattr_header *)bucket_buf;
3520 entries = (char *)xh->xh_entries;
3521 xh_free_start = le16_to_cpu(xh->xh_free_start);
3523 mlog(0, "adjust xattr bucket in %llu, count = %u, "
3524 "xh_free_start = %u, xh_name_value_len = %u.\n",
3525 (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
3526 xh_free_start, le16_to_cpu(xh->xh_name_value_len));
3529 * sort all the entries by their offset.
3530 * the largest will be the first, so that we can
3531 * move them to the end one by one.
3533 sort(entries, le16_to_cpu(xh->xh_count),
3534 sizeof(struct ocfs2_xattr_entry),
3535 cmp_xe_offset, swap_xe);
3537 /* Move all name/values to the end of the bucket. */
3538 xe = xh->xh_entries;
3539 end = OCFS2_XATTR_BUCKET_SIZE;
3540 for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
3541 offset = le16_to_cpu(xe->xe_name_offset);
3542 if (ocfs2_xattr_is_local(xe))
3543 value_len = OCFS2_XATTR_SIZE(
3544 le64_to_cpu(xe->xe_value_size));
3546 value_len = OCFS2_XATTR_ROOT_SIZE;
3547 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
3550 * We must make sure that the name/value pair
3551 * exist in the same block. So adjust end to
3552 * the previous block end if needed.
3554 if (((end - len) / blocksize !=
3555 (end - 1) / blocksize))
3556 end = end - end % blocksize;
3558 if (end > offset + len) {
3559 memmove(bucket_buf + end - len,
3560 bucket_buf + offset, len);
3561 xe->xe_name_offset = cpu_to_le16(end - len);
3564 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
3565 "bucket %llu\n", (unsigned long long)blkno);
3570 mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
3571 "bucket %llu\n", (unsigned long long)blkno);
3573 if (xh_free_start == end)
3576 memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
3577 xh->xh_free_start = cpu_to_le16(end);
3579 /* sort the entries by their name_hash. */
3580 sort(entries, le16_to_cpu(xh->xh_count),
3581 sizeof(struct ocfs2_xattr_entry),
3585 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3586 memcpy(bucket_block(bucket, i), buf, blocksize);
3587 ocfs2_xattr_bucket_journal_dirty(handle, bucket);
3595 * prev_blkno points to the start of an existing extent. new_blkno
3596 * points to a newly allocated extent. Because we know each of our
3597 * clusters contains more than bucket, we can easily split one cluster
3598 * at a bucket boundary. So we take the last cluster of the existing
3599 * extent and split it down the middle. We move the last half of the
3600 * buckets in the last cluster of the existing extent over to the new
3603 * first_bh is the buffer at prev_blkno so we can update the existing
3604 * extent's bucket count. header_bh is the bucket were we were hoping
3605 * to insert our xattr. If the bucket move places the target in the new
3606 * extent, we'll update first_bh and header_bh after modifying the old
3609 * first_hash will be set as the 1st xe's name_hash in the new extent.
3611 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3613 struct ocfs2_xattr_bucket *first,
3614 struct ocfs2_xattr_bucket *target,
3620 struct super_block *sb = inode->i_sb;
3621 int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
3622 int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
3623 int to_move = num_buckets / 2;
3625 u64 last_cluster_blkno = bucket_blkno(first) +
3626 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
3628 BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
3629 BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
3631 mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3632 (unsigned long long)last_cluster_blkno, (unsigned long long)new_blkno);
3634 ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
3635 last_cluster_blkno, new_blkno,
3636 to_move, first_hash);
3642 /* This is the first bucket that got moved */
3643 src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
3646 * If the target bucket was part of the moved buckets, we need to
3647 * update first and target.
3649 if (bucket_blkno(target) >= src_blkno) {
3650 /* Find the block for the new target bucket */
3651 src_blkno = new_blkno +
3652 (bucket_blkno(target) - src_blkno);
3654 ocfs2_xattr_bucket_relse(first);
3655 ocfs2_xattr_bucket_relse(target);
3658 * These shouldn't fail - the buffers are in the
3659 * journal from ocfs2_cp_xattr_bucket().
3661 ret = ocfs2_read_xattr_bucket(first, new_blkno);
3666 ret = ocfs2_read_xattr_bucket(target, src_blkno);
3677 * Find the suitable pos when we divide a bucket into 2.
3678 * We have to make sure the xattrs with the same hash value exist
3679 * in the same bucket.
3681 * If this ocfs2_xattr_header covers more than one hash value, find a
3682 * place where the hash value changes. Try to find the most even split.
3683 * The most common case is that all entries have different hash values,
3684 * and the first check we make will find a place to split.
3686 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
3688 struct ocfs2_xattr_entry *entries = xh->xh_entries;
3689 int count = le16_to_cpu(xh->xh_count);
3690 int delta, middle = count / 2;
3693 * We start at the middle. Each step gets farther away in both
3694 * directions. We therefore hit the change in hash value
3695 * nearest to the middle. Note that this loop does not execute for
3698 for (delta = 0; delta < middle; delta++) {
3699 /* Let's check delta earlier than middle */
3700 if (cmp_xe(&entries[middle - delta - 1],
3701 &entries[middle - delta]))
3702 return middle - delta;
3704 /* For even counts, don't walk off the end */
3705 if ((middle + delta + 1) == count)
3708 /* Now try delta past middle */
3709 if (cmp_xe(&entries[middle + delta],
3710 &entries[middle + delta + 1]))
3711 return middle + delta + 1;
3714 /* Every entry had the same hash */
3719 * Move some xattrs in old bucket(blk) to new bucket(new_blk).
3720 * first_hash will record the 1st hash of the new bucket.
3722 * Normally half of the xattrs will be moved. But we have to make
3723 * sure that the xattrs with the same hash value are stored in the
3724 * same bucket. If all the xattrs in this bucket have the same hash
3725 * value, the new bucket will be initialized as an empty one and the
3726 * first_hash will be initialized as (hash_value+1).
3728 static int ocfs2_divide_xattr_bucket(struct inode *inode,
3733 int new_bucket_head)
3736 int count, start, len, name_value_len = 0, xe_len, name_offset = 0;
3737 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3738 struct ocfs2_xattr_header *xh;
3739 struct ocfs2_xattr_entry *xe;
3740 int blocksize = inode->i_sb->s_blocksize;
3742 mlog(0, "move some of xattrs from bucket %llu to %llu\n",
3743 (unsigned long long)blk, (unsigned long long)new_blk);
3745 s_bucket = ocfs2_xattr_bucket_new(inode);
3746 t_bucket = ocfs2_xattr_bucket_new(inode);
3747 if (!s_bucket || !t_bucket) {
3753 ret = ocfs2_read_xattr_bucket(s_bucket, blk);
3759 ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
3760 OCFS2_JOURNAL_ACCESS_WRITE);
3767 * Even if !new_bucket_head, we're overwriting t_bucket. Thus,
3768 * there's no need to read it.
3770 ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
3777 * Hey, if we're overwriting t_bucket, what difference does
3778 * ACCESS_CREATE vs ACCESS_WRITE make? See the comment in the
3779 * same part of ocfs2_cp_xattr_bucket().
3781 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3783 OCFS2_JOURNAL_ACCESS_CREATE :
3784 OCFS2_JOURNAL_ACCESS_WRITE);
3790 xh = bucket_xh(s_bucket);
3791 count = le16_to_cpu(xh->xh_count);
3792 start = ocfs2_xattr_find_divide_pos(xh);
3794 if (start == count) {
3795 xe = &xh->xh_entries[start-1];
3798 * initialized a new empty bucket here.
3799 * The hash value is set as one larger than
3800 * that of the last entry in the previous bucket.
3802 for (i = 0; i < t_bucket->bu_blocks; i++)
3803 memset(bucket_block(t_bucket, i), 0, blocksize);
3805 xh = bucket_xh(t_bucket);
3806 xh->xh_free_start = cpu_to_le16(blocksize);
3807 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
3808 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
3810 goto set_num_buckets;
3813 /* copy the whole bucket to the new first. */
3814 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
3816 /* update the new bucket. */
3817 xh = bucket_xh(t_bucket);
3820 * Calculate the total name/value len and xh_free_start for
3821 * the old bucket first.
3823 name_offset = OCFS2_XATTR_BUCKET_SIZE;
3825 for (i = 0; i < start; i++) {
3826 xe = &xh->xh_entries[i];
3827 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3828 if (ocfs2_xattr_is_local(xe))
3830 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3832 xe_len += OCFS2_XATTR_ROOT_SIZE;
3833 name_value_len += xe_len;
3834 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3835 name_offset = le16_to_cpu(xe->xe_name_offset);
3839 * Now begin the modification to the new bucket.
3841 * In the new bucket, We just move the xattr entry to the beginning
3842 * and don't touch the name/value. So there will be some holes in the
3843 * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3846 xe = &xh->xh_entries[start];
3847 len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3848 mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3849 (int)((char *)xe - (char *)xh),
3850 (int)((char *)xh->xh_entries - (char *)xh));
3851 memmove((char *)xh->xh_entries, (char *)xe, len);
3852 xe = &xh->xh_entries[count - start];
3853 len = sizeof(struct ocfs2_xattr_entry) * start;
3854 memset((char *)xe, 0, len);
3856 le16_add_cpu(&xh->xh_count, -start);
3857 le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3859 /* Calculate xh_free_start for the new bucket. */
3860 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3861 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3862 xe = &xh->xh_entries[i];
3863 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3864 if (ocfs2_xattr_is_local(xe))
3866 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3868 xe_len += OCFS2_XATTR_ROOT_SIZE;
3869 if (le16_to_cpu(xe->xe_name_offset) <
3870 le16_to_cpu(xh->xh_free_start))
3871 xh->xh_free_start = xe->xe_name_offset;
3875 /* set xh->xh_num_buckets for the new xh. */
3876 if (new_bucket_head)
3877 xh->xh_num_buckets = cpu_to_le16(1);
3879 xh->xh_num_buckets = 0;
3881 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
3883 /* store the first_hash of the new bucket. */
3885 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3888 * Now only update the 1st block of the old bucket. If we
3889 * just added a new empty bucket, there is no need to modify
3895 xh = bucket_xh(s_bucket);
3896 memset(&xh->xh_entries[start], 0,
3897 sizeof(struct ocfs2_xattr_entry) * (count - start));
3898 xh->xh_count = cpu_to_le16(start);
3899 xh->xh_free_start = cpu_to_le16(name_offset);
3900 xh->xh_name_value_len = cpu_to_le16(name_value_len);
3902 ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
3905 ocfs2_xattr_bucket_free(s_bucket);
3906 ocfs2_xattr_bucket_free(t_bucket);
3912 * Copy xattr from one bucket to another bucket.
3914 * The caller must make sure that the journal transaction
3915 * has enough space for journaling.
3917 static int ocfs2_cp_xattr_bucket(struct inode *inode,
3924 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3926 BUG_ON(s_blkno == t_blkno);
3928 mlog(0, "cp bucket %llu to %llu, target is %d\n",
3929 (unsigned long long)s_blkno, (unsigned long long)t_blkno,
3932 s_bucket = ocfs2_xattr_bucket_new(inode);
3933 t_bucket = ocfs2_xattr_bucket_new(inode);
3934 if (!s_bucket || !t_bucket) {
3940 ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
3945 * Even if !t_is_new, we're overwriting t_bucket. Thus,
3946 * there's no need to read it.
3948 ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
3953 * Hey, if we're overwriting t_bucket, what difference does
3954 * ACCESS_CREATE vs ACCESS_WRITE make? Well, if we allocated a new
3955 * cluster to fill, we came here from
3956 * ocfs2_mv_xattr_buckets(), and it is really new -
3957 * ACCESS_CREATE is required. But we also might have moved data
3958 * out of t_bucket before extending back into it.
3959 * ocfs2_add_new_xattr_bucket() can do this - its call to
3960 * ocfs2_add_new_xattr_cluster() may have created a new extent
3961 * and copied out the end of the old extent. Then it re-extends
3962 * the old extent back to create space for new xattrs. That's
3963 * how we get here, and the bucket isn't really new.
3965 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3967 OCFS2_JOURNAL_ACCESS_CREATE :
3968 OCFS2_JOURNAL_ACCESS_WRITE);
3972 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
3973 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
3976 ocfs2_xattr_bucket_free(t_bucket);
3977 ocfs2_xattr_bucket_free(s_bucket);
3983 * src_blk points to the start of an existing extent. last_blk points to
3984 * last cluster in that extent. to_blk points to a newly allocated
3985 * extent. We copy the buckets from the cluster at last_blk to the new
3986 * extent. If start_bucket is non-zero, we skip that many buckets before
3987 * we start copying. The new extent's xh_num_buckets gets set to the
3988 * number of buckets we copied. The old extent's xh_num_buckets shrinks
3989 * by the same amount.
3991 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
3992 u64 src_blk, u64 last_blk, u64 to_blk,
3993 unsigned int start_bucket,
3996 int i, ret, credits;
3997 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3998 int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3999 int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4000 struct ocfs2_xattr_bucket *old_first, *new_first;
4002 mlog(0, "mv xattrs from cluster %llu to %llu\n",
4003 (unsigned long long)last_blk, (unsigned long long)to_blk);
4005 BUG_ON(start_bucket >= num_buckets);
4007 num_buckets -= start_bucket;
4008 last_blk += (start_bucket * blks_per_bucket);
4011 /* The first bucket of the original extent */
4012 old_first = ocfs2_xattr_bucket_new(inode);
4013 /* The first bucket of the new extent */
4014 new_first = ocfs2_xattr_bucket_new(inode);
4015 if (!old_first || !new_first) {
4021 ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4028 * We need to update the first bucket of the old extent and all
4029 * the buckets going to the new extent.
4031 credits = ((num_buckets + 1) * blks_per_bucket) +
4032 handle->h_buffer_credits;
4033 ret = ocfs2_extend_trans(handle, credits);
4039 ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4040 OCFS2_JOURNAL_ACCESS_WRITE);
4046 for (i = 0; i < num_buckets; i++) {
4047 ret = ocfs2_cp_xattr_bucket(inode, handle,
4048 last_blk + (i * blks_per_bucket),
4049 to_blk + (i * blks_per_bucket),
4058 * Get the new bucket ready before we dirty anything
4059 * (This actually shouldn't fail, because we already dirtied
4060 * it once in ocfs2_cp_xattr_bucket()).
4062 ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4067 ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4068 OCFS2_JOURNAL_ACCESS_WRITE);
4074 /* Now update the headers */
4075 le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4076 ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4078 bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4079 ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4082 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4085 ocfs2_xattr_bucket_free(new_first);
4086 ocfs2_xattr_bucket_free(old_first);
4091 * Move some xattrs in this cluster to the new cluster.
4092 * This function should only be called when bucket size == cluster size.
4093 * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4095 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4101 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4102 int ret, credits = 2 * blk_per_bucket + handle->h_buffer_credits;
4104 BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4106 ret = ocfs2_extend_trans(handle, credits);
4112 /* Move half of the xattr in start_blk to the next bucket. */
4113 return ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4114 new_blk, first_hash, 1);
4118 * Move some xattrs from the old cluster to the new one since they are not
4119 * contiguous in ocfs2 xattr tree.
4121 * new_blk starts a new separate cluster, and we will move some xattrs from
4122 * prev_blk to it. v_start will be set as the first name hash value in this
4123 * new cluster so that it can be used as e_cpos during tree insertion and
4124 * don't collide with our original b-tree operations. first_bh and header_bh
4125 * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4126 * to extend the insert bucket.
4128 * The problem is how much xattr should we move to the new one and when should
4129 * we update first_bh and header_bh?
4130 * 1. If cluster size > bucket size, that means the previous cluster has more
4131 * than 1 bucket, so just move half nums of bucket into the new cluster and
4132 * update the first_bh and header_bh if the insert bucket has been moved
4133 * to the new cluster.
4134 * 2. If cluster_size == bucket_size:
4135 * a) If the previous extent rec has more than one cluster and the insert
4136 * place isn't in the last cluster, copy the entire last cluster to the
4137 * new one. This time, we don't need to upate the first_bh and header_bh
4138 * since they will not be moved into the new cluster.
4139 * b) Otherwise, move the bottom half of the xattrs in the last cluster into
4140 * the new one. And we set the extend flag to zero if the insert place is
4141 * moved into the new allocated cluster since no extend is needed.
4143 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
4145 struct ocfs2_xattr_bucket *first,
4146 struct ocfs2_xattr_bucket *target,
4154 mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
4155 (unsigned long long)bucket_blkno(first), prev_clusters,
4156 (unsigned long long)new_blk);
4158 if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
4159 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
4168 /* The start of the last cluster in the first extent */
4169 u64 last_blk = bucket_blkno(first) +
4170 ((prev_clusters - 1) *
4171 ocfs2_clusters_to_blocks(inode->i_sb, 1));
4173 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
4174 ret = ocfs2_mv_xattr_buckets(inode, handle,
4175 bucket_blkno(first),
4176 last_blk, new_blk, 0,
4181 ret = ocfs2_divide_xattr_cluster(inode, handle,
4187 if ((bucket_blkno(target) == last_blk) && extend)
4196 * Add a new cluster for xattr storage.
4198 * If the new cluster is contiguous with the previous one, it will be
4199 * appended to the same extent record, and num_clusters will be updated.
4200 * If not, we will insert a new extent for it and move some xattrs in
4201 * the last cluster into the new allocated one.
4202 * We also need to limit the maximum size of a btree leaf, otherwise we'll
4203 * lose the benefits of hashing because we'll have to search large leaves.
4204 * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
4207 * first_bh is the first block of the previous extent rec and header_bh
4208 * indicates the bucket we will insert the new xattrs. They will be updated
4209 * when the header_bh is moved into the new cluster.
4211 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
4212 struct buffer_head *root_bh,
4213 struct ocfs2_xattr_bucket *first,
4214 struct ocfs2_xattr_bucket *target,
4218 struct ocfs2_xattr_set_ctxt *ctxt)
4221 u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
4222 u32 prev_clusters = *num_clusters;
4223 u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
4225 handle_t *handle = ctxt->handle;
4226 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4227 struct ocfs2_extent_tree et;
4229 mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
4230 "previous xattr blkno = %llu\n",
4231 (unsigned long long)OCFS2_I(inode)->ip_blkno,
4232 prev_cpos, (unsigned long long)bucket_blkno(first));
4234 ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4236 ret = ocfs2_journal_access_xb(handle, inode, root_bh,
4237 OCFS2_JOURNAL_ACCESS_WRITE);
4243 ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac, 1,
4244 clusters_to_add, &bit_off, &num_bits);
4251 BUG_ON(num_bits > clusters_to_add);
4253 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
4254 mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
4255 num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
4257 if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
4258 (prev_clusters + num_bits) << osb->s_clustersize_bits <=
4259 OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
4261 * If this cluster is contiguous with the old one and
4262 * adding this new cluster, we don't surpass the limit of
4263 * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
4264 * initialized and used like other buckets in the previous
4266 * So add it as a contiguous one. The caller will handle
4269 v_start = prev_cpos + prev_clusters;
4270 *num_clusters = prev_clusters + num_bits;
4271 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
4274 ret = ocfs2_adjust_xattr_cross_cluster(inode,
4288 mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
4289 num_bits, (unsigned long long)block, v_start);
4290 ret = ocfs2_insert_extent(osb, handle, inode, &et, v_start, block,
4291 num_bits, 0, ctxt->meta_ac);
4297 ret = ocfs2_journal_dirty(handle, root_bh);
4306 * We are given an extent. 'first' is the bucket at the very front of
4307 * the extent. The extent has space for an additional bucket past
4308 * bucket_xh(first)->xh_num_buckets. 'target_blkno' is the block number
4309 * of the target bucket. We wish to shift every bucket past the target
4310 * down one, filling in that additional space. When we get back to the
4311 * target, we split the target between itself and the now-empty bucket
4312 * at target+1 (aka, target_blkno + blks_per_bucket).
4314 static int ocfs2_extend_xattr_bucket(struct inode *inode,
4316 struct ocfs2_xattr_bucket *first,
4321 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4322 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4324 u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
4326 mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
4327 "from %llu, len = %u\n", (unsigned long long)target_blk,
4328 (unsigned long long)bucket_blkno(first), num_clusters);
4330 /* The extent must have room for an additional bucket */
4331 BUG_ON(new_bucket >=
4332 (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
4334 /* end_blk points to the last existing bucket */
4335 end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
4338 * end_blk is the start of the last existing bucket.
4339 * Thus, (end_blk - target_blk) covers the target bucket and
4340 * every bucket after it up to, but not including, the last
4341 * existing bucket. Then we add the last existing bucket, the
4342 * new bucket, and the first bucket (3 * blk_per_bucket).
4344 credits = (end_blk - target_blk) + (3 * blk_per_bucket) +
4345 handle->h_buffer_credits;
4346 ret = ocfs2_extend_trans(handle, credits);
4352 ret = ocfs2_xattr_bucket_journal_access(handle, first,
4353 OCFS2_JOURNAL_ACCESS_WRITE);
4359 while (end_blk != target_blk) {
4360 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
4361 end_blk + blk_per_bucket, 0);
4364 end_blk -= blk_per_bucket;
4367 /* Move half of the xattr in target_blkno to the next bucket. */
4368 ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
4369 target_blk + blk_per_bucket, NULL, 0);
4371 le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
4372 ocfs2_xattr_bucket_journal_dirty(handle, first);
4379 * Add new xattr bucket in an extent record and adjust the buckets
4380 * accordingly. xb_bh is the ocfs2_xattr_block, and target is the
4381 * bucket we want to insert into.
4383 * In the easy case, we will move all the buckets after target down by
4384 * one. Half of target's xattrs will be moved to the next bucket.
4386 * If current cluster is full, we'll allocate a new one. This may not
4387 * be contiguous. The underlying calls will make sure that there is
4388 * space for the insert, shifting buckets around if necessary.
4389 * 'target' may be moved by those calls.
4391 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
4392 struct buffer_head *xb_bh,
4393 struct ocfs2_xattr_bucket *target,
4394 struct ocfs2_xattr_set_ctxt *ctxt)
4396 struct ocfs2_xattr_block *xb =
4397 (struct ocfs2_xattr_block *)xb_bh->b_data;
4398 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
4399 struct ocfs2_extent_list *el = &xb_root->xt_list;
4401 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
4402 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4403 int ret, num_buckets, extend = 1;
4405 u32 e_cpos, num_clusters;
4406 /* The bucket at the front of the extent */
4407 struct ocfs2_xattr_bucket *first;
4409 mlog(0, "Add new xattr bucket starting from %llu\n",
4410 (unsigned long long)bucket_blkno(target));
4412 /* The first bucket of the original extent */
4413 first = ocfs2_xattr_bucket_new(inode);
4420 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
4427 ret = ocfs2_read_xattr_bucket(first, p_blkno);
4433 num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
4434 if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
4436 * This can move first+target if the target bucket moves
4437 * to the new extent.
4439 ret = ocfs2_add_new_xattr_cluster(inode,
4454 ret = ocfs2_extend_xattr_bucket(inode,
4457 bucket_blkno(target),
4464 ocfs2_xattr_bucket_free(first);
4469 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
4470 struct ocfs2_xattr_bucket *bucket,
4473 int block_off = offs >> inode->i_sb->s_blocksize_bits;
4475 offs = offs % inode->i_sb->s_blocksize;
4476 return bucket_block(bucket, block_off) + offs;
4480 * Handle the normal xattr set, including replace, delete and new.
4482 * Note: "local" indicates the real data's locality. So we can't
4483 * just its bucket locality by its length.
4485 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
4486 struct ocfs2_xattr_info *xi,
4487 struct ocfs2_xattr_search *xs,
4491 struct ocfs2_xattr_entry *last, *xe;
4492 int name_len = strlen(xi->name);
4493 struct ocfs2_xattr_header *xh = xs->header;
4494 u16 count = le16_to_cpu(xh->xh_count), start;
4495 size_t blocksize = inode->i_sb->s_blocksize;
4497 size_t offs, size, new_size;
4499 last = &xh->xh_entries[count];
4500 if (!xs->not_found) {
4502 offs = le16_to_cpu(xe->xe_name_offset);
4503 if (ocfs2_xattr_is_local(xe))
4504 size = OCFS2_XATTR_SIZE(name_len) +
4505 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4507 size = OCFS2_XATTR_SIZE(name_len) +
4508 OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4511 * If the new value will be stored outside, xi->value has been
4512 * initalized as an empty ocfs2_xattr_value_root, and the same
4513 * goes with xi->value_len, so we can set new_size safely here.
4514 * See ocfs2_xattr_set_in_bucket.
4516 new_size = OCFS2_XATTR_SIZE(name_len) +
4517 OCFS2_XATTR_SIZE(xi->value_len);
4519 le16_add_cpu(&xh->xh_name_value_len, -size);
4521 if (new_size > size)
4522 goto set_new_name_value;
4524 /* Now replace the old value with new one. */
4526 xe->xe_value_size = cpu_to_le64(xi->value_len);
4528 xe->xe_value_size = 0;
4530 val = ocfs2_xattr_bucket_get_val(inode,
4532 memset(val + OCFS2_XATTR_SIZE(name_len), 0,
4533 size - OCFS2_XATTR_SIZE(name_len));
4534 if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
4535 memcpy(val + OCFS2_XATTR_SIZE(name_len),
4536 xi->value, xi->value_len);
4538 le16_add_cpu(&xh->xh_name_value_len, new_size);
4539 ocfs2_xattr_set_local(xe, local);
4543 * Remove the old entry if there is more than one.
4544 * We don't remove the last entry so that we can
4545 * use it to indicate the hash value of the empty
4549 le16_add_cpu(&xh->xh_count, -1);
4552 (void *)last - (void *)xe);
4554 sizeof(struct ocfs2_xattr_entry));
4557 cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4562 /* find a new entry for insert. */
4563 int low = 0, high = count - 1, tmp;
4564 struct ocfs2_xattr_entry *tmp_xe;
4566 while (low <= high && count) {
4567 tmp = (low + high) / 2;
4568 tmp_xe = &xh->xh_entries[tmp];
4570 if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
4572 else if (name_hash <
4573 le32_to_cpu(tmp_xe->xe_name_hash))
4581 xe = &xh->xh_entries[low];
4583 memmove(xe + 1, xe, (void *)last - (void *)xe);
4585 le16_add_cpu(&xh->xh_count, 1);
4586 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
4587 xe->xe_name_hash = cpu_to_le32(name_hash);
4588 xe->xe_name_len = name_len;
4589 ocfs2_xattr_set_type(xe, xi->name_index);
4593 /* Insert the new name+value. */
4594 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4597 * We must make sure that the name/value pair
4598 * exists in the same block.
4600 offs = le16_to_cpu(xh->xh_free_start);
4601 start = offs - size;
4603 if (start >> inode->i_sb->s_blocksize_bits !=
4604 (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4605 offs = offs - offs % blocksize;
4606 xh->xh_free_start = cpu_to_le16(offs);
4609 val = ocfs2_xattr_bucket_get_val(inode, xs->bucket, offs - size);
4610 xe->xe_name_offset = cpu_to_le16(offs - size);
4612 memset(val, 0, size);
4613 memcpy(val, xi->name, name_len);
4614 memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4616 xe->xe_value_size = cpu_to_le64(xi->value_len);
4617 ocfs2_xattr_set_local(xe, local);
4619 le16_add_cpu(&xh->xh_free_start, -size);
4620 le16_add_cpu(&xh->xh_name_value_len, size);
4626 * Set the xattr entry in the specified bucket.
4627 * The bucket is indicated by xs->bucket and it should have the enough
4628 * space for the xattr insertion.
4630 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4632 struct ocfs2_xattr_info *xi,
4633 struct ocfs2_xattr_search *xs,
4640 mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4641 (unsigned long)xi->value_len, xi->name_index,
4642 (unsigned long long)bucket_blkno(xs->bucket));
4644 if (!xs->bucket->bu_bhs[1]) {
4645 blkno = bucket_blkno(xs->bucket);
4646 ocfs2_xattr_bucket_relse(xs->bucket);
4647 ret = ocfs2_read_xattr_bucket(xs->bucket, blkno);
4654 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4655 OCFS2_JOURNAL_ACCESS_WRITE);
4661 ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local);
4662 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4669 * Truncate the specified xe_off entry in xattr bucket.
4670 * bucket is indicated by header_bh and len is the new length.
4671 * Both the ocfs2_xattr_value_root and the entry will be updated here.
4673 * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4675 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4676 struct ocfs2_xattr_bucket *bucket,
4679 struct ocfs2_xattr_set_ctxt *ctxt)
4683 struct ocfs2_xattr_entry *xe;
4684 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4685 size_t blocksize = inode->i_sb->s_blocksize;
4686 struct ocfs2_xattr_value_buf vb = {
4687 .vb_access = ocfs2_journal_access,
4690 xe = &xh->xh_entries[xe_off];
4692 BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4694 offset = le16_to_cpu(xe->xe_name_offset) +
4695 OCFS2_XATTR_SIZE(xe->xe_name_len);
4697 value_blk = offset / blocksize;
4699 /* We don't allow ocfs2_xattr_value to be stored in different block. */
4700 BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4702 vb.vb_bh = bucket->bu_bhs[value_blk];
4705 vb.vb_xv = (struct ocfs2_xattr_value_root *)
4706 (vb.vb_bh->b_data + offset % blocksize);
4708 ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
4709 OCFS2_JOURNAL_ACCESS_WRITE);
4716 * From here on out we have to dirty the bucket. The generic
4717 * value calls only modify one of the bucket's bhs, but we need
4718 * to send the bucket at once. So if they error, they *could* have
4719 * modified something. We have to assume they did, and dirty
4720 * the whole bucket. This leaves us in a consistent state.
4722 mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4723 xe_off, (unsigned long long)bucket_blkno(bucket), len);
4724 ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
4730 xe->xe_value_size = cpu_to_le64(len);
4733 ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
4739 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4740 struct ocfs2_xattr_search *xs,
4742 struct ocfs2_xattr_set_ctxt *ctxt)
4745 struct ocfs2_xattr_entry *xe = xs->here;
4746 struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4748 BUG_ON(!xs->bucket->bu_bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4750 offset = xe - xh->xh_entries;
4751 ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket,
4759 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4761 struct ocfs2_xattr_search *xs,
4766 struct ocfs2_xattr_value_root *xv;
4767 struct ocfs2_xattr_entry *xe = xs->here;
4769 BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4771 offset = le16_to_cpu(xe->xe_name_offset) +
4772 OCFS2_XATTR_SIZE(xe->xe_name_len);
4774 xv = (struct ocfs2_xattr_value_root *)(xs->base + offset);
4776 return __ocfs2_xattr_set_value_outside(inode, handle,
4777 xv, val, value_len);
4780 static int ocfs2_rm_xattr_cluster(struct inode *inode,
4781 struct buffer_head *root_bh,
4787 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4788 struct inode *tl_inode = osb->osb_tl_inode;
4790 struct ocfs2_xattr_block *xb =
4791 (struct ocfs2_xattr_block *)root_bh->b_data;
4792 struct ocfs2_alloc_context *meta_ac = NULL;
4793 struct ocfs2_cached_dealloc_ctxt dealloc;
4794 struct ocfs2_extent_tree et;
4796 ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4798 ocfs2_init_dealloc_ctxt(&dealloc);
4800 mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4801 cpos, len, (unsigned long long)blkno);
4803 ocfs2_remove_xattr_clusters_from_cache(inode, blkno, len);
4805 ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
4811 mutex_lock(&tl_inode->i_mutex);
4813 if (ocfs2_truncate_log_needs_flush(osb)) {
4814 ret = __ocfs2_flush_truncate_log(osb);
4821 handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
4822 if (IS_ERR(handle)) {
4828 ret = ocfs2_journal_access_xb(handle, inode, root_bh,
4829 OCFS2_JOURNAL_ACCESS_WRITE);
4835 ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
4842 le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4844 ret = ocfs2_journal_dirty(handle, root_bh);
4850 ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4855 ocfs2_commit_trans(osb, handle);
4857 ocfs2_schedule_truncate_log_flush(osb, 1);
4859 mutex_unlock(&tl_inode->i_mutex);
4862 ocfs2_free_alloc_context(meta_ac);
4864 ocfs2_run_deallocs(osb, &dealloc);
4869 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4871 struct ocfs2_xattr_search *xs)
4873 struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
4874 struct ocfs2_xattr_entry *last = &xh->xh_entries[
4875 le16_to_cpu(xh->xh_count) - 1];
4878 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4879 OCFS2_JOURNAL_ACCESS_WRITE);
4885 /* Remove the old entry. */
4886 memmove(xs->here, xs->here + 1,
4887 (void *)last - (void *)xs->here);
4888 memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4889 le16_add_cpu(&xh->xh_count, -1);
4891 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4895 * Set the xattr name/value in the bucket specified in xs.
4897 * As the new value in xi may be stored in the bucket or in an outside cluster,
4898 * we divide the whole process into 3 steps:
4899 * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4900 * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4901 * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4902 * 4. If the clusters for the new outside value can't be allocated, we need
4903 * to free the xattr we allocated in set.
4905 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4906 struct ocfs2_xattr_info *xi,
4907 struct ocfs2_xattr_search *xs,
4908 struct ocfs2_xattr_set_ctxt *ctxt)
4912 char *val = (char *)xi->value;
4913 struct ocfs2_xattr_entry *xe = xs->here;
4914 u32 name_hash = ocfs2_xattr_name_hash(inode, xi->name,
4917 if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
4919 * We need to truncate the xattr storage first.
4921 * If both the old and new value are stored to
4922 * outside block, we only need to truncate
4923 * the storage and then set the value outside.
4925 * If the new value should be stored within block,
4926 * we should free all the outside block first and
4927 * the modification to the xattr block will be done
4928 * by following steps.
4930 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4931 value_len = xi->value_len;
4935 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4942 goto set_value_outside;
4945 value_len = xi->value_len;
4946 /* So we have to handle the inside block change now. */
4947 if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4949 * If the new value will be stored outside of block,
4950 * initalize a new empty value root and insert it first.
4953 xi->value = &def_xv;
4954 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
4957 ret = ocfs2_xattr_set_entry_in_bucket(inode, ctxt->handle, xi, xs,
4964 if (value_len <= OCFS2_XATTR_INLINE_SIZE)
4967 /* allocate the space now for the outside block storage. */
4968 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4973 if (xs->not_found) {
4975 * We can't allocate enough clusters for outside
4976 * storage and we have allocated xattr already,
4977 * so need to remove it.
4979 ocfs2_xattr_bucket_remove_xs(inode, ctxt->handle, xs);
4985 ret = ocfs2_xattr_bucket_set_value_outside(inode, ctxt->handle,
4986 xs, val, value_len);
4992 * check whether the xattr bucket is filled up with the same hash value.
4993 * If we want to insert the xattr with the same hash, return -ENOSPC.
4994 * If we want to insert a xattr with different hash value, go ahead
4995 * and ocfs2_divide_xattr_bucket will handle this.
4997 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
4998 struct ocfs2_xattr_bucket *bucket,
5001 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5002 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5004 if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5007 if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5008 xh->xh_entries[0].xe_name_hash) {
5009 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5011 (unsigned long long)bucket_blkno(bucket),
5012 le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5019 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5020 struct ocfs2_xattr_info *xi,
5021 struct ocfs2_xattr_search *xs,
5022 struct ocfs2_xattr_set_ctxt *ctxt)
5024 struct ocfs2_xattr_header *xh;
5025 struct ocfs2_xattr_entry *xe;
5026 u16 count, header_size, xh_free_start;
5027 int free, max_free, need, old;
5028 size_t value_size = 0, name_len = strlen(xi->name);
5029 size_t blocksize = inode->i_sb->s_blocksize;
5030 int ret, allocation = 0;
5032 mlog_entry("Set xattr %s in xattr index block\n", xi->name);
5036 count = le16_to_cpu(xh->xh_count);
5037 xh_free_start = le16_to_cpu(xh->xh_free_start);
5038 header_size = sizeof(struct ocfs2_xattr_header) +
5039 count * sizeof(struct ocfs2_xattr_entry);
5040 max_free = OCFS2_XATTR_BUCKET_SIZE -
5041 le16_to_cpu(xh->xh_name_value_len) - header_size;
5043 mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
5044 "of %u which exceed block size\n",
5045 (unsigned long long)bucket_blkno(xs->bucket),
5048 if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
5049 value_size = OCFS2_XATTR_ROOT_SIZE;
5051 value_size = OCFS2_XATTR_SIZE(xi->value_len);
5054 need = sizeof(struct ocfs2_xattr_entry) +
5055 OCFS2_XATTR_SIZE(name_len) + value_size;
5057 need = value_size + OCFS2_XATTR_SIZE(name_len);
5060 * We only replace the old value if the new length is smaller
5061 * than the old one. Otherwise we will allocate new space in the
5062 * bucket to store it.
5065 if (ocfs2_xattr_is_local(xe))
5066 old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
5068 old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
5070 if (old >= value_size)
5074 free = xh_free_start - header_size;
5076 * We need to make sure the new name/value pair
5077 * can exist in the same block.
5079 if (xh_free_start % blocksize < need)
5080 free -= xh_free_start % blocksize;
5082 mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
5083 "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
5084 " %u\n", xs->not_found,
5085 (unsigned long long)bucket_blkno(xs->bucket),
5086 free, need, max_free, le16_to_cpu(xh->xh_free_start),
5087 le16_to_cpu(xh->xh_name_value_len));
5091 count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb))) {
5092 if (need <= max_free &&
5093 count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
5095 * We can create the space by defragment. Since only the
5096 * name/value will be moved, the xe shouldn't be changed
5099 ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5106 xh_free_start = le16_to_cpu(xh->xh_free_start);
5107 free = xh_free_start - header_size;
5108 if (xh_free_start % blocksize < need)
5109 free -= xh_free_start % blocksize;
5114 mlog(0, "Can't get enough space for xattr insert by "
5115 "defragment. Need %u bytes, but we have %d, so "
5116 "allocate new bucket for it.\n", need, free);
5120 * We have to add new buckets or clusters and one
5121 * allocation should leave us enough space for insert.
5126 * We do not allow for overlapping ranges between buckets. And
5127 * the maximum number of collisions we will allow for then is
5128 * one bucket's worth, so check it here whether we need to
5129 * add a new bucket for the insert.
5131 ret = ocfs2_check_xattr_bucket_collision(inode,
5139 ret = ocfs2_add_new_xattr_bucket(inode,
5149 * ocfs2_add_new_xattr_bucket() will have updated
5150 * xs->bucket if it moved, but it will not have updated
5151 * any of the other search fields. Thus, we drop it and
5152 * re-search. Everything should be cached, so it'll be
5155 ocfs2_xattr_bucket_relse(xs->bucket);
5156 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5159 if (ret && ret != -ENODATA)
5161 xs->not_found = ret;
5167 ret = ocfs2_xattr_set_in_bucket(inode, xi, xs, ctxt);
5173 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5174 struct ocfs2_xattr_bucket *bucket,
5178 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5180 struct ocfs2_xattr_entry *xe;
5181 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5182 struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5183 int credits = ocfs2_remove_extent_credits(osb->sb) +
5184 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5187 ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5189 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5190 xe = &xh->xh_entries[i];
5191 if (ocfs2_xattr_is_local(xe))
5194 ctxt.handle = ocfs2_start_trans(osb, credits);
5195 if (IS_ERR(ctxt.handle)) {
5196 ret = PTR_ERR(ctxt.handle);
5201 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5204 ocfs2_commit_trans(osb, ctxt.handle);
5211 ocfs2_schedule_truncate_log_flush(osb, 1);
5212 ocfs2_run_deallocs(osb, &ctxt.dealloc);
5216 static int ocfs2_delete_xattr_index_block(struct inode *inode,
5217 struct buffer_head *xb_bh)
5219 struct ocfs2_xattr_block *xb =
5220 (struct ocfs2_xattr_block *)xb_bh->b_data;
5221 struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
5223 u32 name_hash = UINT_MAX, e_cpos, num_clusters;
5226 if (le16_to_cpu(el->l_next_free_rec) == 0)
5229 while (name_hash > 0) {
5230 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
5231 &e_cpos, &num_clusters, el);
5237 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
5238 ocfs2_delete_xattr_in_bucket,
5245 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
5246 p_blkno, e_cpos, num_clusters);
5255 name_hash = e_cpos - 1;
5263 * 'security' attributes support
5265 static size_t ocfs2_xattr_security_list(struct inode *inode, char *list,
5266 size_t list_size, const char *name,
5269 const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
5270 const size_t total_len = prefix_len + name_len + 1;
5272 if (list && total_len <= list_size) {
5273 memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
5274 memcpy(list + prefix_len, name, name_len);
5275 list[prefix_len + name_len] = '\0';
5280 static int ocfs2_xattr_security_get(struct inode *inode, const char *name,
5281 void *buffer, size_t size)
5283 if (strcmp(name, "") == 0)
5285 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_SECURITY, name,
5289 static int ocfs2_xattr_security_set(struct inode *inode, const char *name,
5290 const void *value, size_t size, int flags)
5292 if (strcmp(name, "") == 0)
5295 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY, name, value,
5299 int ocfs2_init_security_get(struct inode *inode,
5301 struct ocfs2_security_xattr_info *si)
5303 return security_inode_init_security(inode, dir, &si->name, &si->value,
5307 int ocfs2_init_security_set(handle_t *handle,
5308 struct inode *inode,
5309 struct buffer_head *di_bh,
5310 struct ocfs2_security_xattr_info *si,
5311 struct ocfs2_alloc_context *xattr_ac,
5312 struct ocfs2_alloc_context *data_ac)
5314 return ocfs2_xattr_set_handle(handle, inode, di_bh,
5315 OCFS2_XATTR_INDEX_SECURITY,
5316 si->name, si->value, si->value_len, 0,
5320 struct xattr_handler ocfs2_xattr_security_handler = {
5321 .prefix = XATTR_SECURITY_PREFIX,
5322 .list = ocfs2_xattr_security_list,
5323 .get = ocfs2_xattr_security_get,
5324 .set = ocfs2_xattr_security_set,
5328 * 'trusted' attributes support
5330 static size_t ocfs2_xattr_trusted_list(struct inode *inode, char *list,
5331 size_t list_size, const char *name,
5334 const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
5335 const size_t total_len = prefix_len + name_len + 1;
5337 if (list && total_len <= list_size) {
5338 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
5339 memcpy(list + prefix_len, name, name_len);
5340 list[prefix_len + name_len] = '\0';
5345 static int ocfs2_xattr_trusted_get(struct inode *inode, const char *name,
5346 void *buffer, size_t size)
5348 if (strcmp(name, "") == 0)
5350 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED, name,
5354 static int ocfs2_xattr_trusted_set(struct inode *inode, const char *name,
5355 const void *value, size_t size, int flags)
5357 if (strcmp(name, "") == 0)
5360 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED, name, value,
5364 struct xattr_handler ocfs2_xattr_trusted_handler = {
5365 .prefix = XATTR_TRUSTED_PREFIX,
5366 .list = ocfs2_xattr_trusted_list,
5367 .get = ocfs2_xattr_trusted_get,
5368 .set = ocfs2_xattr_trusted_set,
5372 * 'user' attributes support
5374 static size_t ocfs2_xattr_user_list(struct inode *inode, char *list,
5375 size_t list_size, const char *name,
5378 const size_t prefix_len = XATTR_USER_PREFIX_LEN;
5379 const size_t total_len = prefix_len + name_len + 1;
5380 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5382 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5385 if (list && total_len <= list_size) {
5386 memcpy(list, XATTR_USER_PREFIX, prefix_len);
5387 memcpy(list + prefix_len, name, name_len);
5388 list[prefix_len + name_len] = '\0';
5393 static int ocfs2_xattr_user_get(struct inode *inode, const char *name,
5394 void *buffer, size_t size)
5396 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5398 if (strcmp(name, "") == 0)
5400 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5402 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
5406 static int ocfs2_xattr_user_set(struct inode *inode, const char *name,
5407 const void *value, size_t size, int flags)
5409 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5411 if (strcmp(name, "") == 0)
5413 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5416 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER, name, value,
5420 struct xattr_handler ocfs2_xattr_user_handler = {
5421 .prefix = XATTR_USER_PREFIX,
5422 .list = ocfs2_xattr_user_list,
5423 .get = ocfs2_xattr_user_get,
5424 .set = ocfs2_xattr_user_set,