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);
1654 /* Update inode ctime */
1655 inode->i_ctime = CURRENT_TIME;
1656 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1657 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1659 ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1663 if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1665 * Set value outside in B tree.
1666 * This is the second step for value size > INLINE_SIZE.
1668 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1669 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, ctxt,
1676 * If set value outside failed, we have to clean
1677 * the junk tree root we have already set in local.
1679 ret2 = ocfs2_xattr_cleanup(inode, ctxt->handle,
1689 static int ocfs2_remove_value_outside(struct inode*inode,
1690 struct ocfs2_xattr_value_buf *vb,
1691 struct ocfs2_xattr_header *header)
1694 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1695 struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
1697 ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
1699 ctxt.handle = ocfs2_start_trans(osb,
1700 ocfs2_remove_extent_credits(osb->sb));
1701 if (IS_ERR(ctxt.handle)) {
1702 ret = PTR_ERR(ctxt.handle);
1707 for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1708 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1710 if (!ocfs2_xattr_is_local(entry)) {
1713 val = (void *)header +
1714 le16_to_cpu(entry->xe_name_offset);
1715 vb->vb_xv = (struct ocfs2_xattr_value_root *)
1716 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
1717 ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
1725 ocfs2_commit_trans(osb, ctxt.handle);
1726 ocfs2_schedule_truncate_log_flush(osb, 1);
1727 ocfs2_run_deallocs(osb, &ctxt.dealloc);
1732 static int ocfs2_xattr_ibody_remove(struct inode *inode,
1733 struct buffer_head *di_bh)
1736 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1737 struct ocfs2_xattr_header *header;
1739 struct ocfs2_xattr_value_buf vb = {
1741 .vb_access = ocfs2_journal_access_di,
1744 header = (struct ocfs2_xattr_header *)
1745 ((void *)di + inode->i_sb->s_blocksize -
1746 le16_to_cpu(di->i_xattr_inline_size));
1748 ret = ocfs2_remove_value_outside(inode, &vb, header);
1753 static int ocfs2_xattr_block_remove(struct inode *inode,
1754 struct buffer_head *blk_bh)
1756 struct ocfs2_xattr_block *xb;
1758 struct ocfs2_xattr_value_buf vb = {
1760 .vb_access = ocfs2_journal_access_xb,
1763 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1764 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1765 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1766 ret = ocfs2_remove_value_outside(inode, &vb, header);
1768 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
1773 static int ocfs2_xattr_free_block(struct inode *inode,
1776 struct inode *xb_alloc_inode;
1777 struct buffer_head *xb_alloc_bh = NULL;
1778 struct buffer_head *blk_bh = NULL;
1779 struct ocfs2_xattr_block *xb;
1780 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1786 ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
1792 ret = ocfs2_xattr_block_remove(inode, blk_bh);
1798 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1799 blk = le64_to_cpu(xb->xb_blkno);
1800 bit = le16_to_cpu(xb->xb_suballoc_bit);
1801 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1803 xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1804 EXTENT_ALLOC_SYSTEM_INODE,
1805 le16_to_cpu(xb->xb_suballoc_slot));
1806 if (!xb_alloc_inode) {
1811 mutex_lock(&xb_alloc_inode->i_mutex);
1813 ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1819 handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
1820 if (IS_ERR(handle)) {
1821 ret = PTR_ERR(handle);
1826 ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1831 ocfs2_commit_trans(osb, handle);
1833 ocfs2_inode_unlock(xb_alloc_inode, 1);
1834 brelse(xb_alloc_bh);
1836 mutex_unlock(&xb_alloc_inode->i_mutex);
1837 iput(xb_alloc_inode);
1844 * ocfs2_xattr_remove()
1846 * Free extended attribute resources associated with this inode.
1848 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1850 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1851 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1855 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1858 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1861 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1862 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1869 if (di->i_xattr_loc) {
1870 ret = ocfs2_xattr_free_block(inode,
1871 le64_to_cpu(di->i_xattr_loc));
1878 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1879 OCFS2_INODE_UPDATE_CREDITS);
1880 if (IS_ERR(handle)) {
1881 ret = PTR_ERR(handle);
1885 ret = ocfs2_journal_access_di(handle, inode, di_bh,
1886 OCFS2_JOURNAL_ACCESS_WRITE);
1892 di->i_xattr_loc = 0;
1894 spin_lock(&oi->ip_lock);
1895 oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1896 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1897 spin_unlock(&oi->ip_lock);
1899 ret = ocfs2_journal_dirty(handle, di_bh);
1903 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1908 static int ocfs2_xattr_has_space_inline(struct inode *inode,
1909 struct ocfs2_dinode *di)
1911 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1912 unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1915 if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1918 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1919 struct ocfs2_inline_data *idata = &di->id2.i_data;
1920 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1921 } else if (ocfs2_inode_is_fast_symlink(inode)) {
1922 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1923 le64_to_cpu(di->i_size);
1925 struct ocfs2_extent_list *el = &di->id2.i_list;
1926 free = (le16_to_cpu(el->l_count) -
1927 le16_to_cpu(el->l_next_free_rec)) *
1928 sizeof(struct ocfs2_extent_rec);
1930 if (free >= xattrsize)
1937 * ocfs2_xattr_ibody_find()
1939 * Find extended attribute in inode block and
1940 * fill search info into struct ocfs2_xattr_search.
1942 static int ocfs2_xattr_ibody_find(struct inode *inode,
1945 struct ocfs2_xattr_search *xs)
1947 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1948 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1952 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1955 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1956 down_read(&oi->ip_alloc_sem);
1957 has_space = ocfs2_xattr_has_space_inline(inode, di);
1958 up_read(&oi->ip_alloc_sem);
1963 xs->xattr_bh = xs->inode_bh;
1964 xs->end = (void *)di + inode->i_sb->s_blocksize;
1965 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1966 xs->header = (struct ocfs2_xattr_header *)
1967 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1969 xs->header = (struct ocfs2_xattr_header *)
1970 (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
1971 xs->base = (void *)xs->header;
1972 xs->here = xs->header->xh_entries;
1974 /* Find the named attribute. */
1975 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1976 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1977 if (ret && ret != -ENODATA)
1979 xs->not_found = ret;
1986 * ocfs2_xattr_ibody_set()
1988 * Set, replace or remove an extended attribute into inode block.
1991 static int ocfs2_xattr_ibody_set(struct inode *inode,
1992 struct ocfs2_xattr_info *xi,
1993 struct ocfs2_xattr_search *xs,
1994 struct ocfs2_xattr_set_ctxt *ctxt)
1996 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1997 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2000 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2003 down_write(&oi->ip_alloc_sem);
2004 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2005 if (!ocfs2_xattr_has_space_inline(inode, di)) {
2011 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2012 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
2014 up_write(&oi->ip_alloc_sem);
2020 * ocfs2_xattr_block_find()
2022 * Find extended attribute in external block and
2023 * fill search info into struct ocfs2_xattr_search.
2025 static int ocfs2_xattr_block_find(struct inode *inode,
2028 struct ocfs2_xattr_search *xs)
2030 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2031 struct buffer_head *blk_bh = NULL;
2032 struct ocfs2_xattr_block *xb;
2035 if (!di->i_xattr_loc)
2038 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2045 xs->xattr_bh = blk_bh;
2046 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2048 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2049 xs->header = &xb->xb_attrs.xb_header;
2050 xs->base = (void *)xs->header;
2051 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2052 xs->here = xs->header->xh_entries;
2054 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2056 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2060 if (ret && ret != -ENODATA) {
2061 xs->xattr_bh = NULL;
2064 xs->not_found = ret;
2073 * ocfs2_xattr_block_set()
2075 * Set, replace or remove an extended attribute into external block.
2078 static int ocfs2_xattr_block_set(struct inode *inode,
2079 struct ocfs2_xattr_info *xi,
2080 struct ocfs2_xattr_search *xs,
2081 struct ocfs2_xattr_set_ctxt *ctxt)
2083 struct buffer_head *new_bh = NULL;
2084 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2085 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2086 handle_t *handle = ctxt->handle;
2087 struct ocfs2_xattr_block *xblk = NULL;
2088 u16 suballoc_bit_start;
2093 if (!xs->xattr_bh) {
2094 ret = ocfs2_journal_access_di(handle, inode, xs->inode_bh,
2095 OCFS2_JOURNAL_ACCESS_CREATE);
2101 ret = ocfs2_claim_metadata(osb, handle, ctxt->meta_ac, 1,
2102 &suballoc_bit_start, &num_got,
2109 new_bh = sb_getblk(inode->i_sb, first_blkno);
2110 ocfs2_set_new_buffer_uptodate(inode, new_bh);
2112 ret = ocfs2_journal_access_xb(handle, inode, new_bh,
2113 OCFS2_JOURNAL_ACCESS_CREATE);
2119 /* Initialize ocfs2_xattr_block */
2120 xs->xattr_bh = new_bh;
2121 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2122 memset(xblk, 0, inode->i_sb->s_blocksize);
2123 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2124 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
2125 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2126 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
2127 xblk->xb_blkno = cpu_to_le64(first_blkno);
2129 xs->header = &xblk->xb_attrs.xb_header;
2130 xs->base = (void *)xs->header;
2131 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2132 xs->here = xs->header->xh_entries;
2134 ret = ocfs2_journal_dirty(handle, new_bh);
2139 di->i_xattr_loc = cpu_to_le64(first_blkno);
2140 ocfs2_journal_dirty(handle, xs->inode_bh);
2142 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2144 if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2145 /* Set extended attribute into external block */
2146 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2147 OCFS2_HAS_XATTR_FL);
2148 if (!ret || ret != -ENOSPC)
2151 ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2156 ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2163 /* Check whether the new xattr can be inserted into the inode. */
2164 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2165 struct ocfs2_xattr_info *xi,
2166 struct ocfs2_xattr_search *xs)
2169 struct ocfs2_xattr_entry *last;
2171 size_t min_offs = xs->end - xs->base;
2176 last = xs->header->xh_entries;
2178 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
2179 size_t offs = le16_to_cpu(last->xe_name_offset);
2180 if (offs < min_offs)
2185 free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
2189 BUG_ON(!xs->not_found);
2191 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2192 value_size = OCFS2_XATTR_ROOT_SIZE;
2194 value_size = OCFS2_XATTR_SIZE(xi->value_len);
2196 if (free >= sizeof(struct ocfs2_xattr_entry) +
2197 OCFS2_XATTR_SIZE(strlen(xi->name)) + value_size)
2203 static int ocfs2_calc_xattr_set_need(struct inode *inode,
2204 struct ocfs2_dinode *di,
2205 struct ocfs2_xattr_info *xi,
2206 struct ocfs2_xattr_search *xis,
2207 struct ocfs2_xattr_search *xbs,
2212 int ret = 0, old_in_xb = 0;
2213 int clusters_add = 0, meta_add = 0, credits = 0;
2214 struct buffer_head *bh = NULL;
2215 struct ocfs2_xattr_block *xb = NULL;
2216 struct ocfs2_xattr_entry *xe = NULL;
2217 struct ocfs2_xattr_value_root *xv = NULL;
2219 int name_offset, name_len = 0;
2220 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2225 * Calculate the clusters we need to write.
2226 * No matter whether we replace an old one or add a new one,
2227 * we need this for writing.
2229 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2230 credits += new_clusters *
2231 ocfs2_clusters_to_blocks(inode->i_sb, 1);
2233 if (xis->not_found && xbs->not_found) {
2234 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2236 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2237 clusters_add += new_clusters;
2238 credits += ocfs2_calc_extend_credits(inode->i_sb,
2246 if (!xis->not_found) {
2248 name_offset = le16_to_cpu(xe->xe_name_offset);
2249 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2251 credits += OCFS2_INODE_UPDATE_CREDITS;
2253 int i, block_off = 0;
2254 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2256 name_offset = le16_to_cpu(xe->xe_name_offset);
2257 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2258 i = xbs->here - xbs->header->xh_entries;
2261 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2262 ret = ocfs2_xattr_bucket_get_name_value(inode,
2263 bucket_xh(xbs->bucket),
2266 base = bucket_block(xbs->bucket, block_off);
2267 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2270 credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
2275 * delete a xattr doesn't need metadata and cluster allocation.
2276 * so just calculate the credits and return.
2278 * The credits for removing the value tree will be extended
2279 * by ocfs2_remove_extent itself.
2282 if (!ocfs2_xattr_is_local(xe))
2283 credits += ocfs2_remove_extent_credits(inode->i_sb);
2288 /* do cluster allocation guess first. */
2289 value_size = le64_to_cpu(xe->xe_value_size);
2293 * In xattr set, we always try to set the xe in inode first,
2294 * so if it can be inserted into inode successfully, the old
2295 * one will be removed from the xattr block, and this xattr
2296 * will be inserted into inode as a new xattr in inode.
2298 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
2299 clusters_add += new_clusters;
2300 credits += ocfs2_remove_extent_credits(inode->i_sb) +
2301 OCFS2_INODE_UPDATE_CREDITS;
2302 if (!ocfs2_xattr_is_local(xe))
2303 credits += ocfs2_calc_extend_credits(
2311 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2312 /* the new values will be stored outside. */
2313 u32 old_clusters = 0;
2315 if (!ocfs2_xattr_is_local(xe)) {
2316 old_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2318 xv = (struct ocfs2_xattr_value_root *)
2319 (base + name_offset + name_len);
2320 value_size = OCFS2_XATTR_ROOT_SIZE;
2324 if (old_clusters >= new_clusters) {
2325 credits += ocfs2_remove_extent_credits(inode->i_sb);
2328 meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
2329 clusters_add += new_clusters - old_clusters;
2330 credits += ocfs2_calc_extend_credits(inode->i_sb,
2334 if (value_size >= OCFS2_XATTR_ROOT_SIZE)
2339 * Now the new value will be stored inside. So if the new
2340 * value is smaller than the size of value root or the old
2341 * value, we don't need any allocation, otherwise we have
2342 * to guess metadata allocation.
2344 if ((ocfs2_xattr_is_local(xe) && value_size >= xi->value_len) ||
2345 (!ocfs2_xattr_is_local(xe) &&
2346 OCFS2_XATTR_ROOT_SIZE >= xi->value_len))
2351 /* calculate metadata allocation. */
2352 if (di->i_xattr_loc) {
2353 if (!xbs->xattr_bh) {
2354 ret = ocfs2_read_xattr_block(inode,
2355 le64_to_cpu(di->i_xattr_loc),
2362 xb = (struct ocfs2_xattr_block *)bh->b_data;
2364 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2366 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2367 struct ocfs2_extent_list *el =
2368 &xb->xb_attrs.xb_root.xt_list;
2369 meta_add += ocfs2_extend_meta_needed(el);
2370 credits += ocfs2_calc_extend_credits(inode->i_sb,
2375 * This cluster will be used either for new bucket or for
2377 * If the cluster size is the same as the bucket size, one
2378 * more is needed since we may need to extend the bucket
2382 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2383 if (OCFS2_XATTR_BUCKET_SIZE ==
2384 OCFS2_SB(inode->i_sb)->s_clustersize) {
2385 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2390 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
2394 *clusters_need = clusters_add;
2396 *meta_need = meta_add;
2398 *credits_need = credits;
2403 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
2404 struct ocfs2_dinode *di,
2405 struct ocfs2_xattr_info *xi,
2406 struct ocfs2_xattr_search *xis,
2407 struct ocfs2_xattr_search *xbs,
2408 struct ocfs2_xattr_set_ctxt *ctxt,
2411 int clusters_add, meta_add, ret;
2412 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2414 memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
2416 ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
2418 ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
2419 &clusters_add, &meta_add, credits);
2425 mlog(0, "Set xattr %s, reserve meta blocks = %d, clusters = %d, "
2426 "credits = %d\n", xi->name, meta_add, clusters_add, *credits);
2429 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
2438 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
2444 if (ctxt->meta_ac) {
2445 ocfs2_free_alloc_context(ctxt->meta_ac);
2446 ctxt->meta_ac = NULL;
2450 * We cannot have an error and a non null ctxt->data_ac.
2457 static int __ocfs2_xattr_set_handle(struct inode *inode,
2458 struct ocfs2_dinode *di,
2459 struct ocfs2_xattr_info *xi,
2460 struct ocfs2_xattr_search *xis,
2461 struct ocfs2_xattr_search *xbs,
2462 struct ocfs2_xattr_set_ctxt *ctxt)
2464 int ret = 0, credits, old_found;
2467 /* Remove existing extended attribute */
2468 if (!xis->not_found)
2469 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2470 else if (!xbs->not_found)
2471 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2473 /* We always try to set extended attribute into inode first*/
2474 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2475 if (!ret && !xbs->not_found) {
2477 * If succeed and that extended attribute existing in
2478 * external block, then we will remove it.
2483 old_found = xis->not_found;
2484 xis->not_found = -ENODATA;
2485 ret = ocfs2_calc_xattr_set_need(inode,
2493 xis->not_found = old_found;
2499 ret = ocfs2_extend_trans(ctxt->handle, credits +
2500 ctxt->handle->h_buffer_credits);
2505 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2506 } else if (ret == -ENOSPC) {
2507 if (di->i_xattr_loc && !xbs->xattr_bh) {
2508 ret = ocfs2_xattr_block_find(inode,
2514 old_found = xis->not_found;
2515 xis->not_found = -ENODATA;
2516 ret = ocfs2_calc_xattr_set_need(inode,
2524 xis->not_found = old_found;
2530 ret = ocfs2_extend_trans(ctxt->handle, credits +
2531 ctxt->handle->h_buffer_credits);
2538 * If no space in inode, we will set extended attribute
2539 * into external block.
2541 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2544 if (!xis->not_found) {
2546 * If succeed and that extended attribute
2547 * existing in inode, we will remove it.
2551 xbs->not_found = -ENODATA;
2552 ret = ocfs2_calc_xattr_set_need(inode,
2565 ret = ocfs2_extend_trans(ctxt->handle, credits +
2566 ctxt->handle->h_buffer_credits);
2571 ret = ocfs2_xattr_ibody_set(inode, xi,
2582 * This function only called duing creating inode
2583 * for init security/acl xattrs of the new inode.
2584 * The xattrs could be put into ibody or extent block,
2585 * xattr bucket would not be use in this case.
2586 * transanction credits also be reserved in here.
2588 int ocfs2_xattr_set_handle(handle_t *handle,
2589 struct inode *inode,
2590 struct buffer_head *di_bh,
2596 struct ocfs2_alloc_context *meta_ac,
2597 struct ocfs2_alloc_context *data_ac)
2599 struct ocfs2_dinode *di;
2602 struct ocfs2_xattr_info xi = {
2603 .name_index = name_index,
2606 .value_len = value_len,
2609 struct ocfs2_xattr_search xis = {
2610 .not_found = -ENODATA,
2613 struct ocfs2_xattr_search xbs = {
2614 .not_found = -ENODATA,
2617 struct ocfs2_xattr_set_ctxt ctxt = {
2623 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2626 xis.inode_bh = xbs.inode_bh = di_bh;
2627 di = (struct ocfs2_dinode *)di_bh->b_data;
2629 down_write(&OCFS2_I(inode)->ip_xattr_sem);
2631 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2634 if (xis.not_found) {
2635 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2640 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2643 up_write(&OCFS2_I(inode)->ip_xattr_sem);
2644 brelse(xbs.xattr_bh);
2652 * Set, replace or remove an extended attribute for this inode.
2653 * value is NULL to remove an existing extended attribute, else either
2654 * create or replace an extended attribute.
2656 int ocfs2_xattr_set(struct inode *inode,
2663 struct buffer_head *di_bh = NULL;
2664 struct ocfs2_dinode *di;
2666 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2667 struct inode *tl_inode = osb->osb_tl_inode;
2668 struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2670 struct ocfs2_xattr_info xi = {
2671 .name_index = name_index,
2674 .value_len = value_len,
2677 struct ocfs2_xattr_search xis = {
2678 .not_found = -ENODATA,
2681 struct ocfs2_xattr_search xbs = {
2682 .not_found = -ENODATA,
2685 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2689 * Only xbs will be used on indexed trees. xis doesn't need a
2692 xbs.bucket = ocfs2_xattr_bucket_new(inode);
2694 mlog_errno(-ENOMEM);
2698 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2701 goto cleanup_nolock;
2703 xis.inode_bh = xbs.inode_bh = di_bh;
2704 di = (struct ocfs2_dinode *)di_bh->b_data;
2706 down_write(&OCFS2_I(inode)->ip_xattr_sem);
2708 * Scan inode and external block to find the same name
2709 * extended attribute and collect search infomation.
2711 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2714 if (xis.not_found) {
2715 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2720 if (xis.not_found && xbs.not_found) {
2722 if (flags & XATTR_REPLACE)
2729 if (flags & XATTR_CREATE)
2734 mutex_lock(&tl_inode->i_mutex);
2736 if (ocfs2_truncate_log_needs_flush(osb)) {
2737 ret = __ocfs2_flush_truncate_log(osb);
2739 mutex_unlock(&tl_inode->i_mutex);
2744 mutex_unlock(&tl_inode->i_mutex);
2746 ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
2747 &xbs, &ctxt, &credits);
2753 ctxt.handle = ocfs2_start_trans(osb, credits);
2754 if (IS_ERR(ctxt.handle)) {
2755 ret = PTR_ERR(ctxt.handle);
2760 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2762 ocfs2_commit_trans(osb, ctxt.handle);
2765 ocfs2_free_alloc_context(ctxt.data_ac);
2767 ocfs2_free_alloc_context(ctxt.meta_ac);
2768 if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
2769 ocfs2_schedule_truncate_log_flush(osb, 1);
2770 ocfs2_run_deallocs(osb, &ctxt.dealloc);
2772 up_write(&OCFS2_I(inode)->ip_xattr_sem);
2773 ocfs2_inode_unlock(inode, 1);
2776 brelse(xbs.xattr_bh);
2777 ocfs2_xattr_bucket_free(xbs.bucket);
2783 * Find the xattr extent rec which may contains name_hash.
2784 * e_cpos will be the first name hash of the xattr rec.
2785 * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2787 static int ocfs2_xattr_get_rec(struct inode *inode,
2792 struct ocfs2_extent_list *el)
2795 struct buffer_head *eb_bh = NULL;
2796 struct ocfs2_extent_block *eb;
2797 struct ocfs2_extent_rec *rec = NULL;
2800 if (el->l_tree_depth) {
2801 ret = ocfs2_find_leaf(inode, el, name_hash, &eb_bh);
2807 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2810 if (el->l_tree_depth) {
2811 ocfs2_error(inode->i_sb,
2812 "Inode %lu has non zero tree depth in "
2813 "xattr tree block %llu\n", inode->i_ino,
2814 (unsigned long long)eb_bh->b_blocknr);
2820 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2821 rec = &el->l_recs[i];
2823 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2824 e_blkno = le64_to_cpu(rec->e_blkno);
2830 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2831 "record (%u, %u, 0) in xattr", inode->i_ino,
2832 le32_to_cpu(rec->e_cpos),
2833 ocfs2_rec_clusters(el, rec));
2838 *p_blkno = le64_to_cpu(rec->e_blkno);
2839 *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2841 *e_cpos = le32_to_cpu(rec->e_cpos);
2847 typedef int (xattr_bucket_func)(struct inode *inode,
2848 struct ocfs2_xattr_bucket *bucket,
2851 static int ocfs2_find_xe_in_bucket(struct inode *inode,
2852 struct ocfs2_xattr_bucket *bucket,
2859 int i, ret = 0, cmp = 1, block_off, new_offset;
2860 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
2861 size_t name_len = strlen(name);
2862 struct ocfs2_xattr_entry *xe = NULL;
2866 * We don't use binary search in the bucket because there
2867 * may be multiple entries with the same name hash.
2869 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2870 xe = &xh->xh_entries[i];
2872 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2874 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2877 cmp = name_index - ocfs2_xattr_get_type(xe);
2879 cmp = name_len - xe->xe_name_len;
2883 ret = ocfs2_xattr_bucket_get_name_value(inode,
2894 xe_name = bucket_block(bucket, block_off) + new_offset;
2895 if (!memcmp(name, xe_name, name_len)) {
2907 * Find the specified xattr entry in a series of buckets.
2908 * This series start from p_blkno and last for num_clusters.
2909 * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2910 * the num of the valid buckets.
2912 * Return the buffer_head this xattr should reside in. And if the xattr's
2913 * hash is in the gap of 2 buckets, return the lower bucket.
2915 static int ocfs2_xattr_bucket_find(struct inode *inode,
2922 struct ocfs2_xattr_search *xs)
2925 struct ocfs2_xattr_header *xh = NULL;
2926 struct ocfs2_xattr_entry *xe = NULL;
2928 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2929 int low_bucket = 0, bucket, high_bucket;
2930 struct ocfs2_xattr_bucket *search;
2932 u64 blkno, lower_blkno = 0;
2934 search = ocfs2_xattr_bucket_new(inode);
2941 ret = ocfs2_read_xattr_bucket(search, p_blkno);
2947 xh = bucket_xh(search);
2948 high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
2949 while (low_bucket <= high_bucket) {
2950 ocfs2_xattr_bucket_relse(search);
2952 bucket = (low_bucket + high_bucket) / 2;
2953 blkno = p_blkno + bucket * blk_per_bucket;
2954 ret = ocfs2_read_xattr_bucket(search, blkno);
2960 xh = bucket_xh(search);
2961 xe = &xh->xh_entries[0];
2962 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
2963 high_bucket = bucket - 1;
2968 * Check whether the hash of the last entry in our
2969 * bucket is larger than the search one. for an empty
2970 * bucket, the last one is also the first one.
2973 xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
2975 last_hash = le32_to_cpu(xe->xe_name_hash);
2977 /* record lower_blkno which may be the insert place. */
2978 lower_blkno = blkno;
2980 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
2981 low_bucket = bucket + 1;
2985 /* the searched xattr should reside in this bucket if exists. */
2986 ret = ocfs2_find_xe_in_bucket(inode, search,
2987 name_index, name, name_hash,
2997 * Record the bucket we have found.
2998 * When the xattr's hash value is in the gap of 2 buckets, we will
2999 * always set it to the previous bucket.
3002 lower_blkno = p_blkno;
3004 /* This should be in cache - we just read it during the search */
3005 ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3011 xs->header = bucket_xh(xs->bucket);
3012 xs->base = bucket_block(xs->bucket, 0);
3013 xs->end = xs->base + inode->i_sb->s_blocksize;
3016 xs->here = &xs->header->xh_entries[index];
3017 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
3018 (unsigned long long)bucket_blkno(xs->bucket), index);
3023 ocfs2_xattr_bucket_free(search);
3027 static int ocfs2_xattr_index_block_find(struct inode *inode,
3028 struct buffer_head *root_bh,
3031 struct ocfs2_xattr_search *xs)
3034 struct ocfs2_xattr_block *xb =
3035 (struct ocfs2_xattr_block *)root_bh->b_data;
3036 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3037 struct ocfs2_extent_list *el = &xb_root->xt_list;
3039 u32 first_hash, num_clusters = 0;
3040 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3042 if (le16_to_cpu(el->l_next_free_rec) == 0)
3045 mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
3046 name, name_hash, name_index);
3048 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3055 BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3057 mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
3058 "in the rec is %u\n", num_clusters, (unsigned long long)p_blkno,
3061 ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3062 p_blkno, first_hash, num_clusters, xs);
3068 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3071 xattr_bucket_func *func,
3075 u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3076 u32 num_buckets = clusters * bpc;
3077 struct ocfs2_xattr_bucket *bucket;
3079 bucket = ocfs2_xattr_bucket_new(inode);
3081 mlog_errno(-ENOMEM);
3085 mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
3086 clusters, (unsigned long long)blkno);
3088 for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3089 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3096 * The real bucket num in this series of blocks is stored
3097 * in the 1st bucket.
3100 num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3102 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
3103 (unsigned long long)blkno,
3104 le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3106 ret = func(inode, bucket, para);
3109 /* Fall through to bucket_relse() */
3112 ocfs2_xattr_bucket_relse(bucket);
3117 ocfs2_xattr_bucket_free(bucket);
3121 struct ocfs2_xattr_tree_list {
3127 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
3128 struct ocfs2_xattr_header *xh,
3135 if (index < 0 || index >= le16_to_cpu(xh->xh_count))
3138 name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
3140 *block_off = name_offset >> inode->i_sb->s_blocksize_bits;
3141 *new_offset = name_offset % inode->i_sb->s_blocksize;
3146 static int ocfs2_list_xattr_bucket(struct inode *inode,
3147 struct ocfs2_xattr_bucket *bucket,
3151 struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
3152 int i, block_off, new_offset;
3153 const char *prefix, *name;
3155 for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
3156 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
3157 type = ocfs2_xattr_get_type(entry);
3158 prefix = ocfs2_xattr_prefix(type);
3161 ret = ocfs2_xattr_bucket_get_name_value(inode,
3169 name = (const char *)bucket_block(bucket, block_off) +
3171 ret = ocfs2_xattr_list_entry(xl->buffer,
3175 entry->xe_name_len);
3184 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
3185 struct ocfs2_xattr_tree_root *xt,
3189 struct ocfs2_extent_list *el = &xt->xt_list;
3191 u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
3193 struct ocfs2_xattr_tree_list xl = {
3195 .buffer_size = buffer_size,
3199 if (le16_to_cpu(el->l_next_free_rec) == 0)
3202 while (name_hash > 0) {
3203 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
3204 &e_cpos, &num_clusters, el);
3210 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
3211 ocfs2_list_xattr_bucket,
3221 name_hash = e_cpos - 1;
3229 static int cmp_xe(const void *a, const void *b)
3231 const struct ocfs2_xattr_entry *l = a, *r = b;
3232 u32 l_hash = le32_to_cpu(l->xe_name_hash);
3233 u32 r_hash = le32_to_cpu(r->xe_name_hash);
3235 if (l_hash > r_hash)
3237 if (l_hash < r_hash)
3242 static void swap_xe(void *a, void *b, int size)
3244 struct ocfs2_xattr_entry *l = a, *r = b, tmp;
3247 memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
3248 memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
3252 * When the ocfs2_xattr_block is filled up, new bucket will be created
3253 * and all the xattr entries will be moved to the new bucket.
3254 * The header goes at the start of the bucket, and the names+values are
3255 * filled from the end. This is why *target starts as the last buffer.
3256 * Note: we need to sort the entries since they are not saved in order
3257 * in the ocfs2_xattr_block.
3259 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
3260 struct buffer_head *xb_bh,
3261 struct ocfs2_xattr_bucket *bucket)
3263 int i, blocksize = inode->i_sb->s_blocksize;
3264 int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3265 u16 offset, size, off_change;
3266 struct ocfs2_xattr_entry *xe;
3267 struct ocfs2_xattr_block *xb =
3268 (struct ocfs2_xattr_block *)xb_bh->b_data;
3269 struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
3270 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3271 u16 count = le16_to_cpu(xb_xh->xh_count);
3272 char *src = xb_bh->b_data;
3273 char *target = bucket_block(bucket, blks - 1);
3275 mlog(0, "cp xattr from block %llu to bucket %llu\n",
3276 (unsigned long long)xb_bh->b_blocknr,
3277 (unsigned long long)bucket_blkno(bucket));
3279 for (i = 0; i < blks; i++)
3280 memset(bucket_block(bucket, i), 0, blocksize);
3283 * Since the xe_name_offset is based on ocfs2_xattr_header,
3284 * there is a offset change corresponding to the change of
3285 * ocfs2_xattr_header's position.
3287 off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3288 xe = &xb_xh->xh_entries[count - 1];
3289 offset = le16_to_cpu(xe->xe_name_offset) + off_change;
3290 size = blocksize - offset;
3292 /* copy all the names and values. */
3293 memcpy(target + offset, src + offset, size);
3295 /* Init new header now. */
3296 xh->xh_count = xb_xh->xh_count;
3297 xh->xh_num_buckets = cpu_to_le16(1);
3298 xh->xh_name_value_len = cpu_to_le16(size);
3299 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
3301 /* copy all the entries. */
3302 target = bucket_block(bucket, 0);
3303 offset = offsetof(struct ocfs2_xattr_header, xh_entries);
3304 size = count * sizeof(struct ocfs2_xattr_entry);
3305 memcpy(target + offset, (char *)xb_xh + offset, size);
3307 /* Change the xe offset for all the xe because of the move. */
3308 off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
3309 offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3310 for (i = 0; i < count; i++)
3311 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
3313 mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
3314 offset, size, off_change);
3316 sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
3321 * After we move xattr from block to index btree, we have to
3322 * update ocfs2_xattr_search to the new xe and base.
3324 * When the entry is in xattr block, xattr_bh indicates the storage place.
3325 * While if the entry is in index b-tree, "bucket" indicates the
3326 * real place of the xattr.
3328 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
3329 struct ocfs2_xattr_search *xs,
3330 struct buffer_head *old_bh)
3332 char *buf = old_bh->b_data;
3333 struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
3334 struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
3337 xs->header = bucket_xh(xs->bucket);
3338 xs->base = bucket_block(xs->bucket, 0);
3339 xs->end = xs->base + inode->i_sb->s_blocksize;
3344 i = xs->here - old_xh->xh_entries;
3345 xs->here = &xs->header->xh_entries[i];
3348 static int ocfs2_xattr_create_index_block(struct inode *inode,
3349 struct ocfs2_xattr_search *xs,
3350 struct ocfs2_xattr_set_ctxt *ctxt)
3355 handle_t *handle = ctxt->handle;
3356 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3357 struct ocfs2_inode_info *oi = OCFS2_I(inode);
3358 struct buffer_head *xb_bh = xs->xattr_bh;
3359 struct ocfs2_xattr_block *xb =
3360 (struct ocfs2_xattr_block *)xb_bh->b_data;
3361 struct ocfs2_xattr_tree_root *xr;
3362 u16 xb_flags = le16_to_cpu(xb->xb_flags);
3364 mlog(0, "create xattr index block for %llu\n",
3365 (unsigned long long)xb_bh->b_blocknr);
3367 BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
3368 BUG_ON(!xs->bucket);
3372 * We can use this lock for now, and maybe move to a dedicated mutex
3373 * if performance becomes a problem later.
3375 down_write(&oi->ip_alloc_sem);
3377 ret = ocfs2_journal_access_xb(handle, inode, xb_bh,
3378 OCFS2_JOURNAL_ACCESS_WRITE);
3384 ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac,
3385 1, 1, &bit_off, &len);
3392 * The bucket may spread in many blocks, and
3393 * we will only touch the 1st block and the last block
3394 * in the whole bucket(one for entry and one for data).
3396 blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
3398 mlog(0, "allocate 1 cluster from %llu to xattr block\n",
3399 (unsigned long long)blkno);
3401 ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
3407 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
3408 OCFS2_JOURNAL_ACCESS_CREATE);
3414 ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
3415 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
3417 ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
3419 /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
3420 memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
3421 offsetof(struct ocfs2_xattr_block, xb_attrs));
3423 xr = &xb->xb_attrs.xb_root;
3424 xr->xt_clusters = cpu_to_le32(1);
3425 xr->xt_last_eb_blk = 0;
3426 xr->xt_list.l_tree_depth = 0;
3427 xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
3428 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
3430 xr->xt_list.l_recs[0].e_cpos = 0;
3431 xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
3432 xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
3434 xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
3436 ocfs2_journal_dirty(handle, xb_bh);
3439 up_write(&oi->ip_alloc_sem);
3444 static int cmp_xe_offset(const void *a, const void *b)
3446 const struct ocfs2_xattr_entry *l = a, *r = b;
3447 u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
3448 u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
3450 if (l_name_offset < r_name_offset)
3452 if (l_name_offset > r_name_offset)
3458 * defrag a xattr bucket if we find that the bucket has some
3459 * holes beteen name/value pairs.
3460 * We will move all the name/value pairs to the end of the bucket
3461 * so that we can spare some space for insertion.
3463 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
3465 struct ocfs2_xattr_bucket *bucket)
3468 size_t end, offset, len, value_len;
3469 struct ocfs2_xattr_header *xh;
3470 char *entries, *buf, *bucket_buf = NULL;
3471 u64 blkno = bucket_blkno(bucket);
3473 size_t blocksize = inode->i_sb->s_blocksize;
3474 struct ocfs2_xattr_entry *xe;
3477 * In order to make the operation more efficient and generic,
3478 * we copy all the blocks into a contiguous memory and do the
3479 * defragment there, so if anything is error, we will not touch
3482 bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
3489 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3490 memcpy(buf, bucket_block(bucket, i), blocksize);
3492 ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
3493 OCFS2_JOURNAL_ACCESS_WRITE);
3499 xh = (struct ocfs2_xattr_header *)bucket_buf;
3500 entries = (char *)xh->xh_entries;
3501 xh_free_start = le16_to_cpu(xh->xh_free_start);
3503 mlog(0, "adjust xattr bucket in %llu, count = %u, "
3504 "xh_free_start = %u, xh_name_value_len = %u.\n",
3505 (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
3506 xh_free_start, le16_to_cpu(xh->xh_name_value_len));
3509 * sort all the entries by their offset.
3510 * the largest will be the first, so that we can
3511 * move them to the end one by one.
3513 sort(entries, le16_to_cpu(xh->xh_count),
3514 sizeof(struct ocfs2_xattr_entry),
3515 cmp_xe_offset, swap_xe);
3517 /* Move all name/values to the end of the bucket. */
3518 xe = xh->xh_entries;
3519 end = OCFS2_XATTR_BUCKET_SIZE;
3520 for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
3521 offset = le16_to_cpu(xe->xe_name_offset);
3522 if (ocfs2_xattr_is_local(xe))
3523 value_len = OCFS2_XATTR_SIZE(
3524 le64_to_cpu(xe->xe_value_size));
3526 value_len = OCFS2_XATTR_ROOT_SIZE;
3527 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
3530 * We must make sure that the name/value pair
3531 * exist in the same block. So adjust end to
3532 * the previous block end if needed.
3534 if (((end - len) / blocksize !=
3535 (end - 1) / blocksize))
3536 end = end - end % blocksize;
3538 if (end > offset + len) {
3539 memmove(bucket_buf + end - len,
3540 bucket_buf + offset, len);
3541 xe->xe_name_offset = cpu_to_le16(end - len);
3544 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
3545 "bucket %llu\n", (unsigned long long)blkno);
3550 mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
3551 "bucket %llu\n", (unsigned long long)blkno);
3553 if (xh_free_start == end)
3556 memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
3557 xh->xh_free_start = cpu_to_le16(end);
3559 /* sort the entries by their name_hash. */
3560 sort(entries, le16_to_cpu(xh->xh_count),
3561 sizeof(struct ocfs2_xattr_entry),
3565 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3566 memcpy(bucket_block(bucket, i), buf, blocksize);
3567 ocfs2_xattr_bucket_journal_dirty(handle, bucket);
3575 * prev_blkno points to the start of an existing extent. new_blkno
3576 * points to a newly allocated extent. Because we know each of our
3577 * clusters contains more than bucket, we can easily split one cluster
3578 * at a bucket boundary. So we take the last cluster of the existing
3579 * extent and split it down the middle. We move the last half of the
3580 * buckets in the last cluster of the existing extent over to the new
3583 * first_bh is the buffer at prev_blkno so we can update the existing
3584 * extent's bucket count. header_bh is the bucket were we were hoping
3585 * to insert our xattr. If the bucket move places the target in the new
3586 * extent, we'll update first_bh and header_bh after modifying the old
3589 * first_hash will be set as the 1st xe's name_hash in the new extent.
3591 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3593 struct ocfs2_xattr_bucket *first,
3594 struct ocfs2_xattr_bucket *target,
3600 struct super_block *sb = inode->i_sb;
3601 int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
3602 int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
3603 int to_move = num_buckets / 2;
3605 u64 last_cluster_blkno = bucket_blkno(first) +
3606 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
3608 BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
3609 BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
3611 mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3612 (unsigned long long)last_cluster_blkno, (unsigned long long)new_blkno);
3614 ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
3615 last_cluster_blkno, new_blkno,
3616 to_move, first_hash);
3622 /* This is the first bucket that got moved */
3623 src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
3626 * If the target bucket was part of the moved buckets, we need to
3627 * update first and target.
3629 if (bucket_blkno(target) >= src_blkno) {
3630 /* Find the block for the new target bucket */
3631 src_blkno = new_blkno +
3632 (bucket_blkno(target) - src_blkno);
3634 ocfs2_xattr_bucket_relse(first);
3635 ocfs2_xattr_bucket_relse(target);
3638 * These shouldn't fail - the buffers are in the
3639 * journal from ocfs2_cp_xattr_bucket().
3641 ret = ocfs2_read_xattr_bucket(first, new_blkno);
3646 ret = ocfs2_read_xattr_bucket(target, src_blkno);
3657 * Find the suitable pos when we divide a bucket into 2.
3658 * We have to make sure the xattrs with the same hash value exist
3659 * in the same bucket.
3661 * If this ocfs2_xattr_header covers more than one hash value, find a
3662 * place where the hash value changes. Try to find the most even split.
3663 * The most common case is that all entries have different hash values,
3664 * and the first check we make will find a place to split.
3666 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
3668 struct ocfs2_xattr_entry *entries = xh->xh_entries;
3669 int count = le16_to_cpu(xh->xh_count);
3670 int delta, middle = count / 2;
3673 * We start at the middle. Each step gets farther away in both
3674 * directions. We therefore hit the change in hash value
3675 * nearest to the middle. Note that this loop does not execute for
3678 for (delta = 0; delta < middle; delta++) {
3679 /* Let's check delta earlier than middle */
3680 if (cmp_xe(&entries[middle - delta - 1],
3681 &entries[middle - delta]))
3682 return middle - delta;
3684 /* For even counts, don't walk off the end */
3685 if ((middle + delta + 1) == count)
3688 /* Now try delta past middle */
3689 if (cmp_xe(&entries[middle + delta],
3690 &entries[middle + delta + 1]))
3691 return middle + delta + 1;
3694 /* Every entry had the same hash */
3699 * Move some xattrs in old bucket(blk) to new bucket(new_blk).
3700 * first_hash will record the 1st hash of the new bucket.
3702 * Normally half of the xattrs will be moved. But we have to make
3703 * sure that the xattrs with the same hash value are stored in the
3704 * same bucket. If all the xattrs in this bucket have the same hash
3705 * value, the new bucket will be initialized as an empty one and the
3706 * first_hash will be initialized as (hash_value+1).
3708 static int ocfs2_divide_xattr_bucket(struct inode *inode,
3713 int new_bucket_head)
3716 int count, start, len, name_value_len = 0, xe_len, name_offset = 0;
3717 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3718 struct ocfs2_xattr_header *xh;
3719 struct ocfs2_xattr_entry *xe;
3720 int blocksize = inode->i_sb->s_blocksize;
3722 mlog(0, "move some of xattrs from bucket %llu to %llu\n",
3723 (unsigned long long)blk, (unsigned long long)new_blk);
3725 s_bucket = ocfs2_xattr_bucket_new(inode);
3726 t_bucket = ocfs2_xattr_bucket_new(inode);
3727 if (!s_bucket || !t_bucket) {
3733 ret = ocfs2_read_xattr_bucket(s_bucket, blk);
3739 ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
3740 OCFS2_JOURNAL_ACCESS_WRITE);
3747 * Even if !new_bucket_head, we're overwriting t_bucket. Thus,
3748 * there's no need to read it.
3750 ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
3757 * Hey, if we're overwriting t_bucket, what difference does
3758 * ACCESS_CREATE vs ACCESS_WRITE make? See the comment in the
3759 * same part of ocfs2_cp_xattr_bucket().
3761 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3763 OCFS2_JOURNAL_ACCESS_CREATE :
3764 OCFS2_JOURNAL_ACCESS_WRITE);
3770 xh = bucket_xh(s_bucket);
3771 count = le16_to_cpu(xh->xh_count);
3772 start = ocfs2_xattr_find_divide_pos(xh);
3774 if (start == count) {
3775 xe = &xh->xh_entries[start-1];
3778 * initialized a new empty bucket here.
3779 * The hash value is set as one larger than
3780 * that of the last entry in the previous bucket.
3782 for (i = 0; i < t_bucket->bu_blocks; i++)
3783 memset(bucket_block(t_bucket, i), 0, blocksize);
3785 xh = bucket_xh(t_bucket);
3786 xh->xh_free_start = cpu_to_le16(blocksize);
3787 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
3788 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
3790 goto set_num_buckets;
3793 /* copy the whole bucket to the new first. */
3794 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
3796 /* update the new bucket. */
3797 xh = bucket_xh(t_bucket);
3800 * Calculate the total name/value len and xh_free_start for
3801 * the old bucket first.
3803 name_offset = OCFS2_XATTR_BUCKET_SIZE;
3805 for (i = 0; i < start; i++) {
3806 xe = &xh->xh_entries[i];
3807 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3808 if (ocfs2_xattr_is_local(xe))
3810 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3812 xe_len += OCFS2_XATTR_ROOT_SIZE;
3813 name_value_len += xe_len;
3814 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3815 name_offset = le16_to_cpu(xe->xe_name_offset);
3819 * Now begin the modification to the new bucket.
3821 * In the new bucket, We just move the xattr entry to the beginning
3822 * and don't touch the name/value. So there will be some holes in the
3823 * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3826 xe = &xh->xh_entries[start];
3827 len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3828 mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3829 (int)((char *)xe - (char *)xh),
3830 (int)((char *)xh->xh_entries - (char *)xh));
3831 memmove((char *)xh->xh_entries, (char *)xe, len);
3832 xe = &xh->xh_entries[count - start];
3833 len = sizeof(struct ocfs2_xattr_entry) * start;
3834 memset((char *)xe, 0, len);
3836 le16_add_cpu(&xh->xh_count, -start);
3837 le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3839 /* Calculate xh_free_start for the new bucket. */
3840 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3841 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3842 xe = &xh->xh_entries[i];
3843 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3844 if (ocfs2_xattr_is_local(xe))
3846 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3848 xe_len += OCFS2_XATTR_ROOT_SIZE;
3849 if (le16_to_cpu(xe->xe_name_offset) <
3850 le16_to_cpu(xh->xh_free_start))
3851 xh->xh_free_start = xe->xe_name_offset;
3855 /* set xh->xh_num_buckets for the new xh. */
3856 if (new_bucket_head)
3857 xh->xh_num_buckets = cpu_to_le16(1);
3859 xh->xh_num_buckets = 0;
3861 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
3863 /* store the first_hash of the new bucket. */
3865 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3868 * Now only update the 1st block of the old bucket. If we
3869 * just added a new empty bucket, there is no need to modify
3875 xh = bucket_xh(s_bucket);
3876 memset(&xh->xh_entries[start], 0,
3877 sizeof(struct ocfs2_xattr_entry) * (count - start));
3878 xh->xh_count = cpu_to_le16(start);
3879 xh->xh_free_start = cpu_to_le16(name_offset);
3880 xh->xh_name_value_len = cpu_to_le16(name_value_len);
3882 ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
3885 ocfs2_xattr_bucket_free(s_bucket);
3886 ocfs2_xattr_bucket_free(t_bucket);
3892 * Copy xattr from one bucket to another bucket.
3894 * The caller must make sure that the journal transaction
3895 * has enough space for journaling.
3897 static int ocfs2_cp_xattr_bucket(struct inode *inode,
3904 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3906 BUG_ON(s_blkno == t_blkno);
3908 mlog(0, "cp bucket %llu to %llu, target is %d\n",
3909 (unsigned long long)s_blkno, (unsigned long long)t_blkno,
3912 s_bucket = ocfs2_xattr_bucket_new(inode);
3913 t_bucket = ocfs2_xattr_bucket_new(inode);
3914 if (!s_bucket || !t_bucket) {
3920 ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
3925 * Even if !t_is_new, we're overwriting t_bucket. Thus,
3926 * there's no need to read it.
3928 ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
3933 * Hey, if we're overwriting t_bucket, what difference does
3934 * ACCESS_CREATE vs ACCESS_WRITE make? Well, if we allocated a new
3935 * cluster to fill, we came here from
3936 * ocfs2_mv_xattr_buckets(), and it is really new -
3937 * ACCESS_CREATE is required. But we also might have moved data
3938 * out of t_bucket before extending back into it.
3939 * ocfs2_add_new_xattr_bucket() can do this - its call to
3940 * ocfs2_add_new_xattr_cluster() may have created a new extent
3941 * and copied out the end of the old extent. Then it re-extends
3942 * the old extent back to create space for new xattrs. That's
3943 * how we get here, and the bucket isn't really new.
3945 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3947 OCFS2_JOURNAL_ACCESS_CREATE :
3948 OCFS2_JOURNAL_ACCESS_WRITE);
3952 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
3953 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
3956 ocfs2_xattr_bucket_free(t_bucket);
3957 ocfs2_xattr_bucket_free(s_bucket);
3963 * src_blk points to the start of an existing extent. last_blk points to
3964 * last cluster in that extent. to_blk points to a newly allocated
3965 * extent. We copy the buckets from the cluster at last_blk to the new
3966 * extent. If start_bucket is non-zero, we skip that many buckets before
3967 * we start copying. The new extent's xh_num_buckets gets set to the
3968 * number of buckets we copied. The old extent's xh_num_buckets shrinks
3969 * by the same amount.
3971 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
3972 u64 src_blk, u64 last_blk, u64 to_blk,
3973 unsigned int start_bucket,
3976 int i, ret, credits;
3977 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3978 int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3979 int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3980 struct ocfs2_xattr_bucket *old_first, *new_first;
3982 mlog(0, "mv xattrs from cluster %llu to %llu\n",
3983 (unsigned long long)last_blk, (unsigned long long)to_blk);
3985 BUG_ON(start_bucket >= num_buckets);
3987 num_buckets -= start_bucket;
3988 last_blk += (start_bucket * blks_per_bucket);
3991 /* The first bucket of the original extent */
3992 old_first = ocfs2_xattr_bucket_new(inode);
3993 /* The first bucket of the new extent */
3994 new_first = ocfs2_xattr_bucket_new(inode);
3995 if (!old_first || !new_first) {
4001 ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4008 * We need to update the first bucket of the old extent and all
4009 * the buckets going to the new extent.
4011 credits = ((num_buckets + 1) * blks_per_bucket) +
4012 handle->h_buffer_credits;
4013 ret = ocfs2_extend_trans(handle, credits);
4019 ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4020 OCFS2_JOURNAL_ACCESS_WRITE);
4026 for (i = 0; i < num_buckets; i++) {
4027 ret = ocfs2_cp_xattr_bucket(inode, handle,
4028 last_blk + (i * blks_per_bucket),
4029 to_blk + (i * blks_per_bucket),
4038 * Get the new bucket ready before we dirty anything
4039 * (This actually shouldn't fail, because we already dirtied
4040 * it once in ocfs2_cp_xattr_bucket()).
4042 ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4047 ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4048 OCFS2_JOURNAL_ACCESS_WRITE);
4054 /* Now update the headers */
4055 le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4056 ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4058 bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4059 ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4062 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4065 ocfs2_xattr_bucket_free(new_first);
4066 ocfs2_xattr_bucket_free(old_first);
4071 * Move some xattrs in this cluster to the new cluster.
4072 * This function should only be called when bucket size == cluster size.
4073 * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4075 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4081 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4082 int ret, credits = 2 * blk_per_bucket + handle->h_buffer_credits;
4084 BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4086 ret = ocfs2_extend_trans(handle, credits);
4092 /* Move half of the xattr in start_blk to the next bucket. */
4093 return ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4094 new_blk, first_hash, 1);
4098 * Move some xattrs from the old cluster to the new one since they are not
4099 * contiguous in ocfs2 xattr tree.
4101 * new_blk starts a new separate cluster, and we will move some xattrs from
4102 * prev_blk to it. v_start will be set as the first name hash value in this
4103 * new cluster so that it can be used as e_cpos during tree insertion and
4104 * don't collide with our original b-tree operations. first_bh and header_bh
4105 * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4106 * to extend the insert bucket.
4108 * The problem is how much xattr should we move to the new one and when should
4109 * we update first_bh and header_bh?
4110 * 1. If cluster size > bucket size, that means the previous cluster has more
4111 * than 1 bucket, so just move half nums of bucket into the new cluster and
4112 * update the first_bh and header_bh if the insert bucket has been moved
4113 * to the new cluster.
4114 * 2. If cluster_size == bucket_size:
4115 * a) If the previous extent rec has more than one cluster and the insert
4116 * place isn't in the last cluster, copy the entire last cluster to the
4117 * new one. This time, we don't need to upate the first_bh and header_bh
4118 * since they will not be moved into the new cluster.
4119 * b) Otherwise, move the bottom half of the xattrs in the last cluster into
4120 * the new one. And we set the extend flag to zero if the insert place is
4121 * moved into the new allocated cluster since no extend is needed.
4123 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
4125 struct ocfs2_xattr_bucket *first,
4126 struct ocfs2_xattr_bucket *target,
4134 mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
4135 (unsigned long long)bucket_blkno(first), prev_clusters,
4136 (unsigned long long)new_blk);
4138 if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
4139 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
4148 /* The start of the last cluster in the first extent */
4149 u64 last_blk = bucket_blkno(first) +
4150 ((prev_clusters - 1) *
4151 ocfs2_clusters_to_blocks(inode->i_sb, 1));
4153 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
4154 ret = ocfs2_mv_xattr_buckets(inode, handle,
4155 bucket_blkno(first),
4156 last_blk, new_blk, 0,
4161 ret = ocfs2_divide_xattr_cluster(inode, handle,
4167 if ((bucket_blkno(target) == last_blk) && extend)
4176 * Add a new cluster for xattr storage.
4178 * If the new cluster is contiguous with the previous one, it will be
4179 * appended to the same extent record, and num_clusters will be updated.
4180 * If not, we will insert a new extent for it and move some xattrs in
4181 * the last cluster into the new allocated one.
4182 * We also need to limit the maximum size of a btree leaf, otherwise we'll
4183 * lose the benefits of hashing because we'll have to search large leaves.
4184 * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
4187 * first_bh is the first block of the previous extent rec and header_bh
4188 * indicates the bucket we will insert the new xattrs. They will be updated
4189 * when the header_bh is moved into the new cluster.
4191 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
4192 struct buffer_head *root_bh,
4193 struct ocfs2_xattr_bucket *first,
4194 struct ocfs2_xattr_bucket *target,
4198 struct ocfs2_xattr_set_ctxt *ctxt)
4201 u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
4202 u32 prev_clusters = *num_clusters;
4203 u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
4205 handle_t *handle = ctxt->handle;
4206 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4207 struct ocfs2_extent_tree et;
4209 mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
4210 "previous xattr blkno = %llu\n",
4211 (unsigned long long)OCFS2_I(inode)->ip_blkno,
4212 prev_cpos, (unsigned long long)bucket_blkno(first));
4214 ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4216 ret = ocfs2_journal_access_xb(handle, inode, root_bh,
4217 OCFS2_JOURNAL_ACCESS_WRITE);
4223 ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac, 1,
4224 clusters_to_add, &bit_off, &num_bits);
4231 BUG_ON(num_bits > clusters_to_add);
4233 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
4234 mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
4235 num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
4237 if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
4238 (prev_clusters + num_bits) << osb->s_clustersize_bits <=
4239 OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
4241 * If this cluster is contiguous with the old one and
4242 * adding this new cluster, we don't surpass the limit of
4243 * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
4244 * initialized and used like other buckets in the previous
4246 * So add it as a contiguous one. The caller will handle
4249 v_start = prev_cpos + prev_clusters;
4250 *num_clusters = prev_clusters + num_bits;
4251 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
4254 ret = ocfs2_adjust_xattr_cross_cluster(inode,
4268 mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
4269 num_bits, (unsigned long long)block, v_start);
4270 ret = ocfs2_insert_extent(osb, handle, inode, &et, v_start, block,
4271 num_bits, 0, ctxt->meta_ac);
4277 ret = ocfs2_journal_dirty(handle, root_bh);
4286 * We are given an extent. 'first' is the bucket at the very front of
4287 * the extent. The extent has space for an additional bucket past
4288 * bucket_xh(first)->xh_num_buckets. 'target_blkno' is the block number
4289 * of the target bucket. We wish to shift every bucket past the target
4290 * down one, filling in that additional space. When we get back to the
4291 * target, we split the target between itself and the now-empty bucket
4292 * at target+1 (aka, target_blkno + blks_per_bucket).
4294 static int ocfs2_extend_xattr_bucket(struct inode *inode,
4296 struct ocfs2_xattr_bucket *first,
4301 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4302 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4304 u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
4306 mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
4307 "from %llu, len = %u\n", (unsigned long long)target_blk,
4308 (unsigned long long)bucket_blkno(first), num_clusters);
4310 /* The extent must have room for an additional bucket */
4311 BUG_ON(new_bucket >=
4312 (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
4314 /* end_blk points to the last existing bucket */
4315 end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
4318 * end_blk is the start of the last existing bucket.
4319 * Thus, (end_blk - target_blk) covers the target bucket and
4320 * every bucket after it up to, but not including, the last
4321 * existing bucket. Then we add the last existing bucket, the
4322 * new bucket, and the first bucket (3 * blk_per_bucket).
4324 credits = (end_blk - target_blk) + (3 * blk_per_bucket) +
4325 handle->h_buffer_credits;
4326 ret = ocfs2_extend_trans(handle, credits);
4332 ret = ocfs2_xattr_bucket_journal_access(handle, first,
4333 OCFS2_JOURNAL_ACCESS_WRITE);
4339 while (end_blk != target_blk) {
4340 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
4341 end_blk + blk_per_bucket, 0);
4344 end_blk -= blk_per_bucket;
4347 /* Move half of the xattr in target_blkno to the next bucket. */
4348 ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
4349 target_blk + blk_per_bucket, NULL, 0);
4351 le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
4352 ocfs2_xattr_bucket_journal_dirty(handle, first);
4359 * Add new xattr bucket in an extent record and adjust the buckets
4360 * accordingly. xb_bh is the ocfs2_xattr_block, and target is the
4361 * bucket we want to insert into.
4363 * In the easy case, we will move all the buckets after target down by
4364 * one. Half of target's xattrs will be moved to the next bucket.
4366 * If current cluster is full, we'll allocate a new one. This may not
4367 * be contiguous. The underlying calls will make sure that there is
4368 * space for the insert, shifting buckets around if necessary.
4369 * 'target' may be moved by those calls.
4371 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
4372 struct buffer_head *xb_bh,
4373 struct ocfs2_xattr_bucket *target,
4374 struct ocfs2_xattr_set_ctxt *ctxt)
4376 struct ocfs2_xattr_block *xb =
4377 (struct ocfs2_xattr_block *)xb_bh->b_data;
4378 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
4379 struct ocfs2_extent_list *el = &xb_root->xt_list;
4381 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
4382 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4383 int ret, num_buckets, extend = 1;
4385 u32 e_cpos, num_clusters;
4386 /* The bucket at the front of the extent */
4387 struct ocfs2_xattr_bucket *first;
4389 mlog(0, "Add new xattr bucket starting from %llu\n",
4390 (unsigned long long)bucket_blkno(target));
4392 /* The first bucket of the original extent */
4393 first = ocfs2_xattr_bucket_new(inode);
4400 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
4407 ret = ocfs2_read_xattr_bucket(first, p_blkno);
4413 num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
4414 if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
4416 * This can move first+target if the target bucket moves
4417 * to the new extent.
4419 ret = ocfs2_add_new_xattr_cluster(inode,
4434 ret = ocfs2_extend_xattr_bucket(inode,
4437 bucket_blkno(target),
4444 ocfs2_xattr_bucket_free(first);
4449 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
4450 struct ocfs2_xattr_bucket *bucket,
4453 int block_off = offs >> inode->i_sb->s_blocksize_bits;
4455 offs = offs % inode->i_sb->s_blocksize;
4456 return bucket_block(bucket, block_off) + offs;
4460 * Handle the normal xattr set, including replace, delete and new.
4462 * Note: "local" indicates the real data's locality. So we can't
4463 * just its bucket locality by its length.
4465 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
4466 struct ocfs2_xattr_info *xi,
4467 struct ocfs2_xattr_search *xs,
4471 struct ocfs2_xattr_entry *last, *xe;
4472 int name_len = strlen(xi->name);
4473 struct ocfs2_xattr_header *xh = xs->header;
4474 u16 count = le16_to_cpu(xh->xh_count), start;
4475 size_t blocksize = inode->i_sb->s_blocksize;
4477 size_t offs, size, new_size;
4479 last = &xh->xh_entries[count];
4480 if (!xs->not_found) {
4482 offs = le16_to_cpu(xe->xe_name_offset);
4483 if (ocfs2_xattr_is_local(xe))
4484 size = OCFS2_XATTR_SIZE(name_len) +
4485 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4487 size = OCFS2_XATTR_SIZE(name_len) +
4488 OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4491 * If the new value will be stored outside, xi->value has been
4492 * initalized as an empty ocfs2_xattr_value_root, and the same
4493 * goes with xi->value_len, so we can set new_size safely here.
4494 * See ocfs2_xattr_set_in_bucket.
4496 new_size = OCFS2_XATTR_SIZE(name_len) +
4497 OCFS2_XATTR_SIZE(xi->value_len);
4499 le16_add_cpu(&xh->xh_name_value_len, -size);
4501 if (new_size > size)
4502 goto set_new_name_value;
4504 /* Now replace the old value with new one. */
4506 xe->xe_value_size = cpu_to_le64(xi->value_len);
4508 xe->xe_value_size = 0;
4510 val = ocfs2_xattr_bucket_get_val(inode,
4512 memset(val + OCFS2_XATTR_SIZE(name_len), 0,
4513 size - OCFS2_XATTR_SIZE(name_len));
4514 if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
4515 memcpy(val + OCFS2_XATTR_SIZE(name_len),
4516 xi->value, xi->value_len);
4518 le16_add_cpu(&xh->xh_name_value_len, new_size);
4519 ocfs2_xattr_set_local(xe, local);
4523 * Remove the old entry if there is more than one.
4524 * We don't remove the last entry so that we can
4525 * use it to indicate the hash value of the empty
4529 le16_add_cpu(&xh->xh_count, -1);
4532 (void *)last - (void *)xe);
4534 sizeof(struct ocfs2_xattr_entry));
4537 cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4542 /* find a new entry for insert. */
4543 int low = 0, high = count - 1, tmp;
4544 struct ocfs2_xattr_entry *tmp_xe;
4546 while (low <= high && count) {
4547 tmp = (low + high) / 2;
4548 tmp_xe = &xh->xh_entries[tmp];
4550 if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
4552 else if (name_hash <
4553 le32_to_cpu(tmp_xe->xe_name_hash))
4561 xe = &xh->xh_entries[low];
4563 memmove(xe + 1, xe, (void *)last - (void *)xe);
4565 le16_add_cpu(&xh->xh_count, 1);
4566 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
4567 xe->xe_name_hash = cpu_to_le32(name_hash);
4568 xe->xe_name_len = name_len;
4569 ocfs2_xattr_set_type(xe, xi->name_index);
4573 /* Insert the new name+value. */
4574 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4577 * We must make sure that the name/value pair
4578 * exists in the same block.
4580 offs = le16_to_cpu(xh->xh_free_start);
4581 start = offs - size;
4583 if (start >> inode->i_sb->s_blocksize_bits !=
4584 (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4585 offs = offs - offs % blocksize;
4586 xh->xh_free_start = cpu_to_le16(offs);
4589 val = ocfs2_xattr_bucket_get_val(inode, xs->bucket, offs - size);
4590 xe->xe_name_offset = cpu_to_le16(offs - size);
4592 memset(val, 0, size);
4593 memcpy(val, xi->name, name_len);
4594 memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4596 xe->xe_value_size = cpu_to_le64(xi->value_len);
4597 ocfs2_xattr_set_local(xe, local);
4599 le16_add_cpu(&xh->xh_free_start, -size);
4600 le16_add_cpu(&xh->xh_name_value_len, size);
4606 * Set the xattr entry in the specified bucket.
4607 * The bucket is indicated by xs->bucket and it should have the enough
4608 * space for the xattr insertion.
4610 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4612 struct ocfs2_xattr_info *xi,
4613 struct ocfs2_xattr_search *xs,
4620 mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4621 (unsigned long)xi->value_len, xi->name_index,
4622 (unsigned long long)bucket_blkno(xs->bucket));
4624 if (!xs->bucket->bu_bhs[1]) {
4625 blkno = bucket_blkno(xs->bucket);
4626 ocfs2_xattr_bucket_relse(xs->bucket);
4627 ret = ocfs2_read_xattr_bucket(xs->bucket, blkno);
4634 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4635 OCFS2_JOURNAL_ACCESS_WRITE);
4641 ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local);
4642 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4649 * Truncate the specified xe_off entry in xattr bucket.
4650 * bucket is indicated by header_bh and len is the new length.
4651 * Both the ocfs2_xattr_value_root and the entry will be updated here.
4653 * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4655 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4656 struct ocfs2_xattr_bucket *bucket,
4659 struct ocfs2_xattr_set_ctxt *ctxt)
4663 struct ocfs2_xattr_entry *xe;
4664 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4665 size_t blocksize = inode->i_sb->s_blocksize;
4666 struct ocfs2_xattr_value_buf vb = {
4667 .vb_access = ocfs2_journal_access,
4670 xe = &xh->xh_entries[xe_off];
4672 BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4674 offset = le16_to_cpu(xe->xe_name_offset) +
4675 OCFS2_XATTR_SIZE(xe->xe_name_len);
4677 value_blk = offset / blocksize;
4679 /* We don't allow ocfs2_xattr_value to be stored in different block. */
4680 BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4682 vb.vb_bh = bucket->bu_bhs[value_blk];
4685 vb.vb_xv = (struct ocfs2_xattr_value_root *)
4686 (vb.vb_bh->b_data + offset % blocksize);
4688 ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
4689 OCFS2_JOURNAL_ACCESS_WRITE);
4696 * From here on out we have to dirty the bucket. The generic
4697 * value calls only modify one of the bucket's bhs, but we need
4698 * to send the bucket at once. So if they error, they *could* have
4699 * modified something. We have to assume they did, and dirty
4700 * the whole bucket. This leaves us in a consistent state.
4702 mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4703 xe_off, (unsigned long long)bucket_blkno(bucket), len);
4704 ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
4710 xe->xe_value_size = cpu_to_le64(len);
4713 ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
4719 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4720 struct ocfs2_xattr_search *xs,
4722 struct ocfs2_xattr_set_ctxt *ctxt)
4725 struct ocfs2_xattr_entry *xe = xs->here;
4726 struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4728 BUG_ON(!xs->bucket->bu_bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4730 offset = xe - xh->xh_entries;
4731 ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket,
4739 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4741 struct ocfs2_xattr_search *xs,
4746 struct ocfs2_xattr_value_root *xv;
4747 struct ocfs2_xattr_entry *xe = xs->here;
4749 BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4751 offset = le16_to_cpu(xe->xe_name_offset) +
4752 OCFS2_XATTR_SIZE(xe->xe_name_len);
4754 xv = (struct ocfs2_xattr_value_root *)(xs->base + offset);
4756 return __ocfs2_xattr_set_value_outside(inode, handle,
4757 xv, val, value_len);
4760 static int ocfs2_rm_xattr_cluster(struct inode *inode,
4761 struct buffer_head *root_bh,
4767 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4768 struct inode *tl_inode = osb->osb_tl_inode;
4770 struct ocfs2_xattr_block *xb =
4771 (struct ocfs2_xattr_block *)root_bh->b_data;
4772 struct ocfs2_alloc_context *meta_ac = NULL;
4773 struct ocfs2_cached_dealloc_ctxt dealloc;
4774 struct ocfs2_extent_tree et;
4776 ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4778 ocfs2_init_dealloc_ctxt(&dealloc);
4780 mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4781 cpos, len, (unsigned long long)blkno);
4783 ocfs2_remove_xattr_clusters_from_cache(inode, blkno, len);
4785 ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
4791 mutex_lock(&tl_inode->i_mutex);
4793 if (ocfs2_truncate_log_needs_flush(osb)) {
4794 ret = __ocfs2_flush_truncate_log(osb);
4801 handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
4802 if (IS_ERR(handle)) {
4808 ret = ocfs2_journal_access_xb(handle, inode, root_bh,
4809 OCFS2_JOURNAL_ACCESS_WRITE);
4815 ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
4822 le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4824 ret = ocfs2_journal_dirty(handle, root_bh);
4830 ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4835 ocfs2_commit_trans(osb, handle);
4837 ocfs2_schedule_truncate_log_flush(osb, 1);
4839 mutex_unlock(&tl_inode->i_mutex);
4842 ocfs2_free_alloc_context(meta_ac);
4844 ocfs2_run_deallocs(osb, &dealloc);
4849 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4851 struct ocfs2_xattr_search *xs)
4853 struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
4854 struct ocfs2_xattr_entry *last = &xh->xh_entries[
4855 le16_to_cpu(xh->xh_count) - 1];
4858 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4859 OCFS2_JOURNAL_ACCESS_WRITE);
4865 /* Remove the old entry. */
4866 memmove(xs->here, xs->here + 1,
4867 (void *)last - (void *)xs->here);
4868 memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4869 le16_add_cpu(&xh->xh_count, -1);
4871 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4875 * Set the xattr name/value in the bucket specified in xs.
4877 * As the new value in xi may be stored in the bucket or in an outside cluster,
4878 * we divide the whole process into 3 steps:
4879 * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4880 * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4881 * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4882 * 4. If the clusters for the new outside value can't be allocated, we need
4883 * to free the xattr we allocated in set.
4885 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4886 struct ocfs2_xattr_info *xi,
4887 struct ocfs2_xattr_search *xs,
4888 struct ocfs2_xattr_set_ctxt *ctxt)
4892 char *val = (char *)xi->value;
4893 struct ocfs2_xattr_entry *xe = xs->here;
4894 u32 name_hash = ocfs2_xattr_name_hash(inode, xi->name,
4897 if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
4899 * We need to truncate the xattr storage first.
4901 * If both the old and new value are stored to
4902 * outside block, we only need to truncate
4903 * the storage and then set the value outside.
4905 * If the new value should be stored within block,
4906 * we should free all the outside block first and
4907 * the modification to the xattr block will be done
4908 * by following steps.
4910 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4911 value_len = xi->value_len;
4915 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4922 goto set_value_outside;
4925 value_len = xi->value_len;
4926 /* So we have to handle the inside block change now. */
4927 if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4929 * If the new value will be stored outside of block,
4930 * initalize a new empty value root and insert it first.
4933 xi->value = &def_xv;
4934 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
4937 ret = ocfs2_xattr_set_entry_in_bucket(inode, ctxt->handle, xi, xs,
4944 if (value_len <= OCFS2_XATTR_INLINE_SIZE)
4947 /* allocate the space now for the outside block storage. */
4948 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4953 if (xs->not_found) {
4955 * We can't allocate enough clusters for outside
4956 * storage and we have allocated xattr already,
4957 * so need to remove it.
4959 ocfs2_xattr_bucket_remove_xs(inode, ctxt->handle, xs);
4965 ret = ocfs2_xattr_bucket_set_value_outside(inode, ctxt->handle,
4966 xs, val, value_len);
4972 * check whether the xattr bucket is filled up with the same hash value.
4973 * If we want to insert the xattr with the same hash, return -ENOSPC.
4974 * If we want to insert a xattr with different hash value, go ahead
4975 * and ocfs2_divide_xattr_bucket will handle this.
4977 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
4978 struct ocfs2_xattr_bucket *bucket,
4981 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4982 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
4984 if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
4987 if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
4988 xh->xh_entries[0].xe_name_hash) {
4989 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
4991 (unsigned long long)bucket_blkno(bucket),
4992 le32_to_cpu(xh->xh_entries[0].xe_name_hash));
4999 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5000 struct ocfs2_xattr_info *xi,
5001 struct ocfs2_xattr_search *xs,
5002 struct ocfs2_xattr_set_ctxt *ctxt)
5004 struct ocfs2_xattr_header *xh;
5005 struct ocfs2_xattr_entry *xe;
5006 u16 count, header_size, xh_free_start;
5007 int free, max_free, need, old;
5008 size_t value_size = 0, name_len = strlen(xi->name);
5009 size_t blocksize = inode->i_sb->s_blocksize;
5010 int ret, allocation = 0;
5012 mlog_entry("Set xattr %s in xattr index block\n", xi->name);
5016 count = le16_to_cpu(xh->xh_count);
5017 xh_free_start = le16_to_cpu(xh->xh_free_start);
5018 header_size = sizeof(struct ocfs2_xattr_header) +
5019 count * sizeof(struct ocfs2_xattr_entry);
5020 max_free = OCFS2_XATTR_BUCKET_SIZE -
5021 le16_to_cpu(xh->xh_name_value_len) - header_size;
5023 mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
5024 "of %u which exceed block size\n",
5025 (unsigned long long)bucket_blkno(xs->bucket),
5028 if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
5029 value_size = OCFS2_XATTR_ROOT_SIZE;
5031 value_size = OCFS2_XATTR_SIZE(xi->value_len);
5034 need = sizeof(struct ocfs2_xattr_entry) +
5035 OCFS2_XATTR_SIZE(name_len) + value_size;
5037 need = value_size + OCFS2_XATTR_SIZE(name_len);
5040 * We only replace the old value if the new length is smaller
5041 * than the old one. Otherwise we will allocate new space in the
5042 * bucket to store it.
5045 if (ocfs2_xattr_is_local(xe))
5046 old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
5048 old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
5050 if (old >= value_size)
5054 free = xh_free_start - header_size;
5056 * We need to make sure the new name/value pair
5057 * can exist in the same block.
5059 if (xh_free_start % blocksize < need)
5060 free -= xh_free_start % blocksize;
5062 mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
5063 "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
5064 " %u\n", xs->not_found,
5065 (unsigned long long)bucket_blkno(xs->bucket),
5066 free, need, max_free, le16_to_cpu(xh->xh_free_start),
5067 le16_to_cpu(xh->xh_name_value_len));
5071 count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb))) {
5072 if (need <= max_free &&
5073 count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
5075 * We can create the space by defragment. Since only the
5076 * name/value will be moved, the xe shouldn't be changed
5079 ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5086 xh_free_start = le16_to_cpu(xh->xh_free_start);
5087 free = xh_free_start - header_size;
5088 if (xh_free_start % blocksize < need)
5089 free -= xh_free_start % blocksize;
5094 mlog(0, "Can't get enough space for xattr insert by "
5095 "defragment. Need %u bytes, but we have %d, so "
5096 "allocate new bucket for it.\n", need, free);
5100 * We have to add new buckets or clusters and one
5101 * allocation should leave us enough space for insert.
5106 * We do not allow for overlapping ranges between buckets. And
5107 * the maximum number of collisions we will allow for then is
5108 * one bucket's worth, so check it here whether we need to
5109 * add a new bucket for the insert.
5111 ret = ocfs2_check_xattr_bucket_collision(inode,
5119 ret = ocfs2_add_new_xattr_bucket(inode,
5129 * ocfs2_add_new_xattr_bucket() will have updated
5130 * xs->bucket if it moved, but it will not have updated
5131 * any of the other search fields. Thus, we drop it and
5132 * re-search. Everything should be cached, so it'll be
5135 ocfs2_xattr_bucket_relse(xs->bucket);
5136 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5139 if (ret && ret != -ENODATA)
5141 xs->not_found = ret;
5147 ret = ocfs2_xattr_set_in_bucket(inode, xi, xs, ctxt);
5153 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5154 struct ocfs2_xattr_bucket *bucket,
5158 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5160 struct ocfs2_xattr_entry *xe;
5161 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5162 struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5163 int credits = ocfs2_remove_extent_credits(osb->sb) +
5164 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5167 ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5169 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5170 xe = &xh->xh_entries[i];
5171 if (ocfs2_xattr_is_local(xe))
5174 ctxt.handle = ocfs2_start_trans(osb, credits);
5175 if (IS_ERR(ctxt.handle)) {
5176 ret = PTR_ERR(ctxt.handle);
5181 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5184 ocfs2_commit_trans(osb, ctxt.handle);
5191 ocfs2_schedule_truncate_log_flush(osb, 1);
5192 ocfs2_run_deallocs(osb, &ctxt.dealloc);
5196 static int ocfs2_delete_xattr_index_block(struct inode *inode,
5197 struct buffer_head *xb_bh)
5199 struct ocfs2_xattr_block *xb =
5200 (struct ocfs2_xattr_block *)xb_bh->b_data;
5201 struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
5203 u32 name_hash = UINT_MAX, e_cpos, num_clusters;
5206 if (le16_to_cpu(el->l_next_free_rec) == 0)
5209 while (name_hash > 0) {
5210 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
5211 &e_cpos, &num_clusters, el);
5217 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
5218 ocfs2_delete_xattr_in_bucket,
5225 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
5226 p_blkno, e_cpos, num_clusters);
5235 name_hash = e_cpos - 1;
5243 * 'security' attributes support
5245 static size_t ocfs2_xattr_security_list(struct inode *inode, char *list,
5246 size_t list_size, const char *name,
5249 const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
5250 const size_t total_len = prefix_len + name_len + 1;
5252 if (list && total_len <= list_size) {
5253 memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
5254 memcpy(list + prefix_len, name, name_len);
5255 list[prefix_len + name_len] = '\0';
5260 static int ocfs2_xattr_security_get(struct inode *inode, const char *name,
5261 void *buffer, size_t size)
5263 if (strcmp(name, "") == 0)
5265 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_SECURITY, name,
5269 static int ocfs2_xattr_security_set(struct inode *inode, const char *name,
5270 const void *value, size_t size, int flags)
5272 if (strcmp(name, "") == 0)
5275 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY, name, value,
5279 int ocfs2_init_security_get(struct inode *inode,
5281 struct ocfs2_security_xattr_info *si)
5283 return security_inode_init_security(inode, dir, &si->name, &si->value,
5287 int ocfs2_init_security_set(handle_t *handle,
5288 struct inode *inode,
5289 struct buffer_head *di_bh,
5290 struct ocfs2_security_xattr_info *si,
5291 struct ocfs2_alloc_context *xattr_ac,
5292 struct ocfs2_alloc_context *data_ac)
5294 return ocfs2_xattr_set_handle(handle, inode, di_bh,
5295 OCFS2_XATTR_INDEX_SECURITY,
5296 si->name, si->value, si->value_len, 0,
5300 struct xattr_handler ocfs2_xattr_security_handler = {
5301 .prefix = XATTR_SECURITY_PREFIX,
5302 .list = ocfs2_xattr_security_list,
5303 .get = ocfs2_xattr_security_get,
5304 .set = ocfs2_xattr_security_set,
5308 * 'trusted' attributes support
5310 static size_t ocfs2_xattr_trusted_list(struct inode *inode, char *list,
5311 size_t list_size, const char *name,
5314 const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
5315 const size_t total_len = prefix_len + name_len + 1;
5317 if (list && total_len <= list_size) {
5318 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
5319 memcpy(list + prefix_len, name, name_len);
5320 list[prefix_len + name_len] = '\0';
5325 static int ocfs2_xattr_trusted_get(struct inode *inode, const char *name,
5326 void *buffer, size_t size)
5328 if (strcmp(name, "") == 0)
5330 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED, name,
5334 static int ocfs2_xattr_trusted_set(struct inode *inode, const char *name,
5335 const void *value, size_t size, int flags)
5337 if (strcmp(name, "") == 0)
5340 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED, name, value,
5344 struct xattr_handler ocfs2_xattr_trusted_handler = {
5345 .prefix = XATTR_TRUSTED_PREFIX,
5346 .list = ocfs2_xattr_trusted_list,
5347 .get = ocfs2_xattr_trusted_get,
5348 .set = ocfs2_xattr_trusted_set,
5352 * 'user' attributes support
5354 static size_t ocfs2_xattr_user_list(struct inode *inode, char *list,
5355 size_t list_size, const char *name,
5358 const size_t prefix_len = XATTR_USER_PREFIX_LEN;
5359 const size_t total_len = prefix_len + name_len + 1;
5360 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5362 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5365 if (list && total_len <= list_size) {
5366 memcpy(list, XATTR_USER_PREFIX, prefix_len);
5367 memcpy(list + prefix_len, name, name_len);
5368 list[prefix_len + name_len] = '\0';
5373 static int ocfs2_xattr_user_get(struct inode *inode, const char *name,
5374 void *buffer, size_t size)
5376 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5378 if (strcmp(name, "") == 0)
5380 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5382 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
5386 static int ocfs2_xattr_user_set(struct inode *inode, const char *name,
5387 const void *value, size_t size, int flags)
5389 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5391 if (strcmp(name, "") == 0)
5393 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5396 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER, name, value,
5400 struct xattr_handler ocfs2_xattr_user_handler = {
5401 .prefix = XATTR_USER_PREFIX,
5402 .list = ocfs2_xattr_user_list,
5403 .get = ocfs2_xattr_user_get,
5404 .set = ocfs2_xattr_user_set,