1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * Copyright (C) 2004, 2008 Oracle. All rights reserved.
9 * Lots of code in this file is copy from linux/fs/ext3/xattr.c.
10 * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
12 * This program is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU General Public
14 * License version 2 as published by the Free Software Foundation.
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
22 #include <linux/capability.h>
24 #include <linux/types.h>
25 #include <linux/slab.h>
26 #include <linux/highmem.h>
27 #include <linux/pagemap.h>
28 #include <linux/uio.h>
29 #include <linux/sched.h>
30 #include <linux/splice.h>
31 #include <linux/mount.h>
32 #include <linux/writeback.h>
33 #include <linux/falloc.h>
34 #include <linux/sort.h>
35 #include <linux/init.h>
36 #include <linux/module.h>
37 #include <linux/string.h>
38 #include <linux/security.h>
40 #define MLOG_MASK_PREFIX ML_XATTR
41 #include <cluster/masklog.h>
45 #include "blockcheck.h"
55 #include "buffer_head_io.h"
58 #include "refcounttree.h"
61 struct ocfs2_xattr_def_value_root {
62 struct ocfs2_xattr_value_root xv;
63 struct ocfs2_extent_rec er;
66 struct ocfs2_xattr_bucket {
67 /* The inode these xattrs are associated with */
68 struct inode *bu_inode;
70 /* The actual buffers that make up the bucket */
71 struct buffer_head *bu_bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
73 /* How many blocks make up one bucket for this filesystem */
77 struct ocfs2_xattr_set_ctxt {
79 struct ocfs2_alloc_context *meta_ac;
80 struct ocfs2_alloc_context *data_ac;
81 struct ocfs2_cached_dealloc_ctxt dealloc;
84 #define OCFS2_XATTR_ROOT_SIZE (sizeof(struct ocfs2_xattr_def_value_root))
85 #define OCFS2_XATTR_INLINE_SIZE 80
86 #define OCFS2_XATTR_HEADER_GAP 4
87 #define OCFS2_XATTR_FREE_IN_IBODY (OCFS2_MIN_XATTR_INLINE_SIZE \
88 - sizeof(struct ocfs2_xattr_header) \
89 - OCFS2_XATTR_HEADER_GAP)
90 #define OCFS2_XATTR_FREE_IN_BLOCK(ptr) ((ptr)->i_sb->s_blocksize \
91 - sizeof(struct ocfs2_xattr_block) \
92 - sizeof(struct ocfs2_xattr_header) \
93 - OCFS2_XATTR_HEADER_GAP)
95 static struct ocfs2_xattr_def_value_root def_xv = {
96 .xv.xr_list.l_count = cpu_to_le16(1),
99 struct xattr_handler *ocfs2_xattr_handlers[] = {
100 &ocfs2_xattr_user_handler,
101 &ocfs2_xattr_acl_access_handler,
102 &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 [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS]
111 = &ocfs2_xattr_acl_access_handler,
112 [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT]
113 = &ocfs2_xattr_acl_default_handler,
114 [OCFS2_XATTR_INDEX_TRUSTED] = &ocfs2_xattr_trusted_handler,
115 [OCFS2_XATTR_INDEX_SECURITY] = &ocfs2_xattr_security_handler,
118 struct ocfs2_xattr_info {
125 struct ocfs2_xattr_search {
126 struct buffer_head *inode_bh;
128 * xattr_bh point to the block buffer head which has extended attribute
129 * when extended attribute in inode, xattr_bh is equal to inode_bh.
131 struct buffer_head *xattr_bh;
132 struct ocfs2_xattr_header *header;
133 struct ocfs2_xattr_bucket *bucket;
136 struct ocfs2_xattr_entry *here;
140 /* Operations on struct ocfs2_xa_entry */
142 struct ocfs2_xa_loc_operations {
144 * Return a pointer to the appropriate buffer in loc->xl_storage
145 * at the given offset from loc->xl_header.
147 void *(*xlo_offset_pointer)(struct ocfs2_xa_loc *loc, int offset);
150 * Remove the name+value at this location. Do whatever is
151 * appropriate with the remaining name+value pairs.
153 void (*xlo_wipe_namevalue)(struct ocfs2_xa_loc *loc);
157 * Describes an xattr entry location. This is a memory structure
158 * tracking the on-disk structure.
160 struct ocfs2_xa_loc {
161 /* The ocfs2_xattr_header inside the on-disk storage. Not NULL. */
162 struct ocfs2_xattr_header *xl_header;
164 /* Bytes from xl_header to the end of the storage */
168 * The ocfs2_xattr_entry this location describes. If this is
169 * NULL, this location describes the on-disk structure where it
172 struct ocfs2_xattr_entry *xl_entry;
175 * Internal housekeeping
178 /* Buffer(s) containing this entry */
181 /* Operations on the storage backing this location */
182 const struct ocfs2_xa_loc_operations *xl_ops;
185 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
186 struct ocfs2_xattr_header *xh,
191 static int ocfs2_xattr_block_find(struct inode *inode,
194 struct ocfs2_xattr_search *xs);
195 static int ocfs2_xattr_index_block_find(struct inode *inode,
196 struct buffer_head *root_bh,
199 struct ocfs2_xattr_search *xs);
201 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
202 struct buffer_head *blk_bh,
206 static int ocfs2_xattr_create_index_block(struct inode *inode,
207 struct ocfs2_xattr_search *xs,
208 struct ocfs2_xattr_set_ctxt *ctxt);
210 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
211 struct ocfs2_xattr_info *xi,
212 struct ocfs2_xattr_search *xs,
213 struct ocfs2_xattr_set_ctxt *ctxt);
215 typedef int (xattr_tree_rec_func)(struct inode *inode,
216 struct buffer_head *root_bh,
217 u64 blkno, u32 cpos, u32 len, void *para);
218 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
219 struct buffer_head *root_bh,
220 xattr_tree_rec_func *rec_func,
222 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
223 struct ocfs2_xattr_bucket *bucket,
225 static int ocfs2_rm_xattr_cluster(struct inode *inode,
226 struct buffer_head *root_bh,
232 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
233 u64 src_blk, u64 last_blk, u64 to_blk,
234 unsigned int start_bucket,
236 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
237 struct ocfs2_dinode *di,
238 struct ocfs2_xattr_info *xi,
239 struct ocfs2_xattr_search *xis,
240 struct ocfs2_xattr_search *xbs,
241 struct ocfs2_refcount_tree **ref_tree,
244 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
245 struct ocfs2_xattr_bucket *bucket,
247 struct ocfs2_xattr_value_root **xv,
248 struct buffer_head **bh);
250 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
252 return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
255 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
257 return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
260 static inline u16 ocfs2_xattr_max_xe_in_bucket(struct super_block *sb)
262 u16 len = sb->s_blocksize -
263 offsetof(struct ocfs2_xattr_header, xh_entries);
265 return len / sizeof(struct ocfs2_xattr_entry);
268 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
269 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
270 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
272 static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
274 struct ocfs2_xattr_bucket *bucket;
275 int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
277 BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
279 bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
281 bucket->bu_inode = inode;
282 bucket->bu_blocks = blks;
288 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
292 for (i = 0; i < bucket->bu_blocks; i++) {
293 brelse(bucket->bu_bhs[i]);
294 bucket->bu_bhs[i] = NULL;
298 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
301 ocfs2_xattr_bucket_relse(bucket);
302 bucket->bu_inode = NULL;
308 * A bucket that has never been written to disk doesn't need to be
309 * read. We just need the buffer_heads. Don't call this for
310 * buckets that are already on disk. ocfs2_read_xattr_bucket() initializes
313 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
318 for (i = 0; i < bucket->bu_blocks; i++) {
319 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
321 if (!bucket->bu_bhs[i]) {
327 if (!ocfs2_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
329 ocfs2_set_new_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
334 ocfs2_xattr_bucket_relse(bucket);
338 /* Read the xattr bucket at xb_blkno */
339 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
344 rc = ocfs2_read_blocks(INODE_CACHE(bucket->bu_inode), xb_blkno,
345 bucket->bu_blocks, bucket->bu_bhs, 0,
348 spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
349 rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
352 &bucket_xh(bucket)->xh_check);
353 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
359 ocfs2_xattr_bucket_relse(bucket);
363 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
364 struct ocfs2_xattr_bucket *bucket,
369 for (i = 0; i < bucket->bu_blocks; i++) {
370 rc = ocfs2_journal_access(handle,
371 INODE_CACHE(bucket->bu_inode),
372 bucket->bu_bhs[i], type);
382 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
383 struct ocfs2_xattr_bucket *bucket)
387 spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
388 ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
389 bucket->bu_bhs, bucket->bu_blocks,
390 &bucket_xh(bucket)->xh_check);
391 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
393 for (i = 0; i < bucket->bu_blocks; i++)
394 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
397 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
398 struct ocfs2_xattr_bucket *src)
401 int blocksize = src->bu_inode->i_sb->s_blocksize;
403 BUG_ON(dest->bu_blocks != src->bu_blocks);
404 BUG_ON(dest->bu_inode != src->bu_inode);
406 for (i = 0; i < src->bu_blocks; i++) {
407 memcpy(bucket_block(dest, i), bucket_block(src, i),
412 static int ocfs2_validate_xattr_block(struct super_block *sb,
413 struct buffer_head *bh)
416 struct ocfs2_xattr_block *xb =
417 (struct ocfs2_xattr_block *)bh->b_data;
419 mlog(0, "Validating xattr block %llu\n",
420 (unsigned long long)bh->b_blocknr);
422 BUG_ON(!buffer_uptodate(bh));
425 * If the ecc fails, we return the error but otherwise
426 * leave the filesystem running. We know any error is
427 * local to this block.
429 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
434 * Errors after here are fatal
437 if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
439 "Extended attribute block #%llu has bad "
441 (unsigned long long)bh->b_blocknr, 7,
446 if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
448 "Extended attribute block #%llu has an "
449 "invalid xb_blkno of %llu",
450 (unsigned long long)bh->b_blocknr,
451 (unsigned long long)le64_to_cpu(xb->xb_blkno));
455 if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
457 "Extended attribute block #%llu has an invalid "
458 "xb_fs_generation of #%u",
459 (unsigned long long)bh->b_blocknr,
460 le32_to_cpu(xb->xb_fs_generation));
467 static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
468 struct buffer_head **bh)
471 struct buffer_head *tmp = *bh;
473 rc = ocfs2_read_block(INODE_CACHE(inode), xb_blkno, &tmp,
474 ocfs2_validate_xattr_block);
476 /* If ocfs2_read_block() got us a new bh, pass it up. */
483 static inline const char *ocfs2_xattr_prefix(int name_index)
485 struct xattr_handler *handler = NULL;
487 if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
488 handler = ocfs2_xattr_handler_map[name_index];
490 return handler ? handler->prefix : NULL;
493 static u32 ocfs2_xattr_name_hash(struct inode *inode,
497 /* Get hash value of uuid from super block */
498 u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
501 /* hash extended attribute name */
502 for (i = 0; i < name_len; i++) {
503 hash = (hash << OCFS2_HASH_SHIFT) ^
504 (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
512 * ocfs2_xattr_hash_entry()
514 * Compute the hash of an extended attribute.
516 static void ocfs2_xattr_hash_entry(struct inode *inode,
517 struct ocfs2_xattr_header *header,
518 struct ocfs2_xattr_entry *entry)
521 char *name = (char *)header + le16_to_cpu(entry->xe_name_offset);
523 hash = ocfs2_xattr_name_hash(inode, name, entry->xe_name_len);
524 entry->xe_name_hash = cpu_to_le32(hash);
529 static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
533 if (value_len <= OCFS2_XATTR_INLINE_SIZE)
534 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
536 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
537 size += sizeof(struct ocfs2_xattr_entry);
542 int ocfs2_calc_security_init(struct inode *dir,
543 struct ocfs2_security_xattr_info *si,
546 struct ocfs2_alloc_context **xattr_ac)
549 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
550 int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
554 * The max space of security xattr taken inline is
555 * 256(name) + 80(value) + 16(entry) = 352 bytes,
556 * So reserve one metadata block for it is ok.
558 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
559 s_size > OCFS2_XATTR_FREE_IN_IBODY) {
560 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
565 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
568 /* reserve clusters for xattr value which will be set in B tree*/
569 if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
570 int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
573 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
575 *want_clusters += new_clusters;
580 int ocfs2_calc_xattr_init(struct inode *dir,
581 struct buffer_head *dir_bh,
583 struct ocfs2_security_xattr_info *si,
589 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
590 int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
593 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
596 if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
597 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
598 OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
601 a_size = ocfs2_xattr_entry_real_size(0, acl_len);
604 } else if (acl_len != 0 && acl_len != -ENODATA) {
610 if (!(s_size + a_size))
614 * The max space of security xattr taken inline is
615 * 256(name) + 80(value) + 16(entry) = 352 bytes,
616 * The max space of acl xattr taken inline is
617 * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
618 * when blocksize = 512, may reserve one more cluser for
619 * xattr bucket, otherwise reserve one metadata block
621 * If this is a new directory with inline data,
622 * we choose to reserve the entire inline area for
623 * directory contents and force an external xattr block.
625 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
626 (S_ISDIR(mode) && ocfs2_supports_inline_data(osb)) ||
627 (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
628 *want_meta = *want_meta + 1;
629 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
632 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
633 (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
635 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
639 * reserve credits and clusters for xattrs which has large value
640 * and have to be set outside
642 if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
643 new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
645 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
647 *want_clusters += new_clusters;
649 if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
650 acl_len > OCFS2_XATTR_INLINE_SIZE) {
651 /* for directory, it has DEFAULT and ACCESS two types of acls */
652 new_clusters = (S_ISDIR(mode) ? 2 : 1) *
653 ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
654 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
656 *want_clusters += new_clusters;
662 static int ocfs2_xattr_extend_allocation(struct inode *inode,
664 struct ocfs2_xattr_value_buf *vb,
665 struct ocfs2_xattr_set_ctxt *ctxt)
668 handle_t *handle = ctxt->handle;
669 enum ocfs2_alloc_restarted why;
670 u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
671 struct ocfs2_extent_tree et;
673 mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
675 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
677 status = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
678 OCFS2_JOURNAL_ACCESS_WRITE);
684 prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
685 status = ocfs2_add_clusters_in_btree(handle,
698 status = ocfs2_journal_dirty(handle, vb->vb_bh);
704 clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) - prev_clusters;
707 * We should have already allocated enough space before the transaction,
708 * so no need to restart.
710 BUG_ON(why != RESTART_NONE || clusters_to_add);
717 static int __ocfs2_remove_xattr_range(struct inode *inode,
718 struct ocfs2_xattr_value_buf *vb,
719 u32 cpos, u32 phys_cpos, u32 len,
720 unsigned int ext_flags,
721 struct ocfs2_xattr_set_ctxt *ctxt)
724 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
725 handle_t *handle = ctxt->handle;
726 struct ocfs2_extent_tree et;
728 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
730 ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
731 OCFS2_JOURNAL_ACCESS_WRITE);
737 ret = ocfs2_remove_extent(handle, &et, cpos, len, ctxt->meta_ac,
744 le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
746 ret = ocfs2_journal_dirty(handle, vb->vb_bh);
752 if (ext_flags & OCFS2_EXT_REFCOUNTED)
753 ret = ocfs2_decrease_refcount(inode, handle,
754 ocfs2_blocks_to_clusters(inode->i_sb,
756 len, ctxt->meta_ac, &ctxt->dealloc, 1);
758 ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc,
767 static int ocfs2_xattr_shrink_size(struct inode *inode,
770 struct ocfs2_xattr_value_buf *vb,
771 struct ocfs2_xattr_set_ctxt *ctxt)
774 unsigned int ext_flags;
775 u32 trunc_len, cpos, phys_cpos, alloc_size;
778 if (old_clusters <= new_clusters)
782 trunc_len = old_clusters - new_clusters;
784 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
786 &vb->vb_xv->xr_list, &ext_flags);
792 if (alloc_size > trunc_len)
793 alloc_size = trunc_len;
795 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
796 phys_cpos, alloc_size,
803 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
804 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode),
807 trunc_len -= alloc_size;
814 static int ocfs2_xattr_value_truncate(struct inode *inode,
815 struct ocfs2_xattr_value_buf *vb,
817 struct ocfs2_xattr_set_ctxt *ctxt)
820 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
821 u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
823 if (new_clusters == old_clusters)
826 if (new_clusters > old_clusters)
827 ret = ocfs2_xattr_extend_allocation(inode,
828 new_clusters - old_clusters,
831 ret = ocfs2_xattr_shrink_size(inode,
832 old_clusters, new_clusters,
838 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
839 size_t *result, const char *prefix,
840 const char *name, int name_len)
842 char *p = buffer + *result;
843 int prefix_len = strlen(prefix);
844 int total_len = prefix_len + name_len + 1;
846 *result += total_len;
848 /* we are just looking for how big our buffer needs to be */
855 memcpy(p, prefix, prefix_len);
856 memcpy(p + prefix_len, name, name_len);
857 p[prefix_len + name_len] = '\0';
862 static int ocfs2_xattr_list_entries(struct inode *inode,
863 struct ocfs2_xattr_header *header,
864 char *buffer, size_t buffer_size)
868 const char *prefix, *name;
870 for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
871 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
872 type = ocfs2_xattr_get_type(entry);
873 prefix = ocfs2_xattr_prefix(type);
876 name = (const char *)header +
877 le16_to_cpu(entry->xe_name_offset);
879 ret = ocfs2_xattr_list_entry(buffer, buffer_size,
880 &result, prefix, name,
890 int ocfs2_has_inline_xattr_value_outside(struct inode *inode,
891 struct ocfs2_dinode *di)
893 struct ocfs2_xattr_header *xh;
896 xh = (struct ocfs2_xattr_header *)
897 ((void *)di + inode->i_sb->s_blocksize -
898 le16_to_cpu(di->i_xattr_inline_size));
900 for (i = 0; i < le16_to_cpu(xh->xh_count); i++)
901 if (!ocfs2_xattr_is_local(&xh->xh_entries[i]))
907 static int ocfs2_xattr_ibody_list(struct inode *inode,
908 struct ocfs2_dinode *di,
912 struct ocfs2_xattr_header *header = NULL;
913 struct ocfs2_inode_info *oi = OCFS2_I(inode);
916 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
919 header = (struct ocfs2_xattr_header *)
920 ((void *)di + inode->i_sb->s_blocksize -
921 le16_to_cpu(di->i_xattr_inline_size));
923 ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
928 static int ocfs2_xattr_block_list(struct inode *inode,
929 struct ocfs2_dinode *di,
933 struct buffer_head *blk_bh = NULL;
934 struct ocfs2_xattr_block *xb;
937 if (!di->i_xattr_loc)
940 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
947 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
948 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
949 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
950 ret = ocfs2_xattr_list_entries(inode, header,
951 buffer, buffer_size);
953 ret = ocfs2_xattr_tree_list_index_block(inode, blk_bh,
954 buffer, buffer_size);
961 ssize_t ocfs2_listxattr(struct dentry *dentry,
965 int ret = 0, i_ret = 0, b_ret = 0;
966 struct buffer_head *di_bh = NULL;
967 struct ocfs2_dinode *di = NULL;
968 struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
970 if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
973 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
976 ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
982 di = (struct ocfs2_dinode *)di_bh->b_data;
984 down_read(&oi->ip_xattr_sem);
985 i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
993 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
998 up_read(&oi->ip_xattr_sem);
999 ocfs2_inode_unlock(dentry->d_inode, 0);
1003 return i_ret + b_ret;
1006 static int ocfs2_xattr_find_entry(int name_index,
1008 struct ocfs2_xattr_search *xs)
1010 struct ocfs2_xattr_entry *entry;
1017 name_len = strlen(name);
1019 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
1020 cmp = name_index - ocfs2_xattr_get_type(entry);
1022 cmp = name_len - entry->xe_name_len;
1024 cmp = memcmp(name, (xs->base +
1025 le16_to_cpu(entry->xe_name_offset)),
1033 return cmp ? -ENODATA : 0;
1036 static int ocfs2_xattr_get_value_outside(struct inode *inode,
1037 struct ocfs2_xattr_value_root *xv,
1041 u32 cpos, p_cluster, num_clusters, bpc, clusters;
1044 size_t cplen, blocksize;
1045 struct buffer_head *bh = NULL;
1046 struct ocfs2_extent_list *el;
1049 clusters = le32_to_cpu(xv->xr_clusters);
1050 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1051 blocksize = inode->i_sb->s_blocksize;
1054 while (cpos < clusters) {
1055 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1056 &num_clusters, el, NULL);
1062 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1063 /* Copy ocfs2_xattr_value */
1064 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1065 ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1072 cplen = len >= blocksize ? blocksize : len;
1073 memcpy(buffer, bh->b_data, cplen);
1082 cpos += num_clusters;
1088 static int ocfs2_xattr_ibody_get(struct inode *inode,
1093 struct ocfs2_xattr_search *xs)
1095 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1096 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1097 struct ocfs2_xattr_value_root *xv;
1101 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1104 xs->end = (void *)di + inode->i_sb->s_blocksize;
1105 xs->header = (struct ocfs2_xattr_header *)
1106 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1107 xs->base = (void *)xs->header;
1108 xs->here = xs->header->xh_entries;
1110 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1113 size = le64_to_cpu(xs->here->xe_value_size);
1115 if (size > buffer_size)
1117 if (ocfs2_xattr_is_local(xs->here)) {
1118 memcpy(buffer, (void *)xs->base +
1119 le16_to_cpu(xs->here->xe_name_offset) +
1120 OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1122 xv = (struct ocfs2_xattr_value_root *)
1123 (xs->base + le16_to_cpu(
1124 xs->here->xe_name_offset) +
1125 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1126 ret = ocfs2_xattr_get_value_outside(inode, xv,
1138 static int ocfs2_xattr_block_get(struct inode *inode,
1143 struct ocfs2_xattr_search *xs)
1145 struct ocfs2_xattr_block *xb;
1146 struct ocfs2_xattr_value_root *xv;
1148 int ret = -ENODATA, name_offset, name_len, i;
1149 int uninitialized_var(block_off);
1151 xs->bucket = ocfs2_xattr_bucket_new(inode);
1158 ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1164 if (xs->not_found) {
1169 xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1170 size = le64_to_cpu(xs->here->xe_value_size);
1173 if (size > buffer_size)
1176 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1177 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1178 i = xs->here - xs->header->xh_entries;
1180 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1181 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
1182 bucket_xh(xs->bucket),
1186 xs->base = bucket_block(xs->bucket, block_off);
1188 if (ocfs2_xattr_is_local(xs->here)) {
1189 memcpy(buffer, (void *)xs->base +
1190 name_offset + name_len, size);
1192 xv = (struct ocfs2_xattr_value_root *)
1193 (xs->base + name_offset + name_len);
1194 ret = ocfs2_xattr_get_value_outside(inode, xv,
1204 ocfs2_xattr_bucket_free(xs->bucket);
1206 brelse(xs->xattr_bh);
1207 xs->xattr_bh = NULL;
1211 int ocfs2_xattr_get_nolock(struct inode *inode,
1212 struct buffer_head *di_bh,
1219 struct ocfs2_dinode *di = NULL;
1220 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1221 struct ocfs2_xattr_search xis = {
1222 .not_found = -ENODATA,
1224 struct ocfs2_xattr_search xbs = {
1225 .not_found = -ENODATA,
1228 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1231 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1234 xis.inode_bh = xbs.inode_bh = di_bh;
1235 di = (struct ocfs2_dinode *)di_bh->b_data;
1237 down_read(&oi->ip_xattr_sem);
1238 ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1240 if (ret == -ENODATA && di->i_xattr_loc)
1241 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1243 up_read(&oi->ip_xattr_sem);
1248 /* ocfs2_xattr_get()
1250 * Copy an extended attribute into the buffer provided.
1251 * Buffer is NULL to compute the size of buffer required.
1253 static int ocfs2_xattr_get(struct inode *inode,
1260 struct buffer_head *di_bh = NULL;
1262 ret = ocfs2_inode_lock(inode, &di_bh, 0);
1267 ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1268 name, buffer, buffer_size);
1270 ocfs2_inode_unlock(inode, 0);
1277 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1279 struct ocfs2_xattr_value_buf *vb,
1283 int ret = 0, i, cp_len;
1284 u16 blocksize = inode->i_sb->s_blocksize;
1285 u32 p_cluster, num_clusters;
1286 u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1287 u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1289 struct buffer_head *bh = NULL;
1290 unsigned int ext_flags;
1291 struct ocfs2_xattr_value_root *xv = vb->vb_xv;
1293 BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1295 while (cpos < clusters) {
1296 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1297 &num_clusters, &xv->xr_list,
1304 BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1306 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1308 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1309 ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1316 ret = ocfs2_journal_access(handle,
1319 OCFS2_JOURNAL_ACCESS_WRITE);
1325 cp_len = value_len > blocksize ? blocksize : value_len;
1326 memcpy(bh->b_data, value, cp_len);
1327 value_len -= cp_len;
1329 if (cp_len < blocksize)
1330 memset(bh->b_data + cp_len, 0,
1331 blocksize - cp_len);
1333 ret = ocfs2_journal_dirty(handle, bh);
1342 * XXX: do we need to empty all the following
1343 * blocks in this cluster?
1348 cpos += num_clusters;
1356 static int ocfs2_xattr_cleanup(struct inode *inode,
1358 struct ocfs2_xattr_info *xi,
1359 struct ocfs2_xattr_search *xs,
1360 struct ocfs2_xattr_value_buf *vb,
1364 size_t name_len = strlen(xi->name);
1365 void *val = xs->base + offs;
1366 size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1368 ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
1369 OCFS2_JOURNAL_ACCESS_WRITE);
1374 /* Decrease xattr count */
1375 le16_add_cpu(&xs->header->xh_count, -1);
1376 /* Remove the xattr entry and tree root which has already be set*/
1377 memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
1378 memset(val, 0, size);
1380 ret = ocfs2_journal_dirty(handle, vb->vb_bh);
1387 static int ocfs2_xattr_update_entry(struct inode *inode,
1389 struct ocfs2_xattr_info *xi,
1390 struct ocfs2_xattr_search *xs,
1391 struct ocfs2_xattr_value_buf *vb,
1396 ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
1397 OCFS2_JOURNAL_ACCESS_WRITE);
1403 xs->here->xe_name_offset = cpu_to_le16(offs);
1404 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1405 if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1406 ocfs2_xattr_set_local(xs->here, 1);
1408 ocfs2_xattr_set_local(xs->here, 0);
1409 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1411 ret = ocfs2_journal_dirty(handle, vb->vb_bh);
1419 * ocfs2_xattr_set_value_outside()
1421 * Set large size value in B tree.
1423 static int ocfs2_xattr_set_value_outside(struct inode *inode,
1424 struct ocfs2_xattr_info *xi,
1425 struct ocfs2_xattr_search *xs,
1426 struct ocfs2_xattr_set_ctxt *ctxt,
1427 struct ocfs2_xattr_value_buf *vb,
1430 size_t name_len = strlen(xi->name);
1431 void *val = xs->base + offs;
1432 struct ocfs2_xattr_value_root *xv = NULL;
1433 size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1436 memset(val, 0, size);
1437 memcpy(val, xi->name, name_len);
1438 xv = (struct ocfs2_xattr_value_root *)
1439 (val + OCFS2_XATTR_SIZE(name_len));
1440 xv->xr_clusters = 0;
1441 xv->xr_last_eb_blk = 0;
1442 xv->xr_list.l_tree_depth = 0;
1443 xv->xr_list.l_count = cpu_to_le16(1);
1444 xv->xr_list.l_next_free_rec = 0;
1447 ret = ocfs2_xattr_value_truncate(inode, vb, xi->value_len, ctxt);
1452 ret = ocfs2_xattr_update_entry(inode, ctxt->handle, xi, xs, vb, offs);
1457 ret = __ocfs2_xattr_set_value_outside(inode, ctxt->handle, vb,
1458 xi->value, xi->value_len);
1466 * Wipe the name+value pair and allow the storage to reclaim it. This
1467 * must be followed by either removal of the entry or a call to
1468 * ocfs2_xa_add_namevalue().
1470 static void ocfs2_xa_wipe_namevalue(struct ocfs2_xa_loc *loc)
1472 loc->xl_ops->xlo_wipe_namevalue(loc);
1475 static void *ocfs2_xa_block_offset_pointer(struct ocfs2_xa_loc *loc,
1478 BUG_ON(offset >= loc->xl_size);
1479 return (char *)loc->xl_header + offset;
1483 * Block storage for xattrs keeps the name+value pairs compacted. When
1484 * we remove one, we have to shift any that preceded it towards the end.
1486 static void ocfs2_xa_block_wipe_namevalue(struct ocfs2_xa_loc *loc)
1489 int namevalue_offset, first_namevalue_offset, namevalue_size;
1490 struct ocfs2_xattr_entry *entry = loc->xl_entry;
1491 struct ocfs2_xattr_header *xh = loc->xl_header;
1492 u64 value_size = le64_to_cpu(entry->xe_value_size);
1493 int count = le16_to_cpu(xh->xh_count);
1495 namevalue_offset = le16_to_cpu(entry->xe_name_offset);
1496 namevalue_size = OCFS2_XATTR_SIZE(entry->xe_name_len);
1497 if (value_size > OCFS2_XATTR_INLINE_SIZE)
1498 namevalue_size += OCFS2_XATTR_ROOT_SIZE;
1500 namevalue_size += OCFS2_XATTR_SIZE(value_size);
1502 for (i = 0, first_namevalue_offset = loc->xl_size;
1504 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1505 if (offset < first_namevalue_offset)
1506 first_namevalue_offset = offset;
1509 /* Shift the name+value pairs */
1510 memmove((char *)xh + first_namevalue_offset + namevalue_size,
1511 (char *)xh + first_namevalue_offset,
1512 namevalue_offset - first_namevalue_offset);
1513 memset((char *)xh + first_namevalue_offset, 0, namevalue_size);
1515 /* Now tell xh->xh_entries about it */
1516 for (i = 0; i < count; i++) {
1517 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1518 if (offset < namevalue_offset)
1519 le16_add_cpu(&xh->xh_entries[i].xe_name_offset,
1524 * Note that we don't update xh_free_start or xh_name_value_len
1525 * because they're not used in block-stored xattrs.
1530 * Operations for xattrs stored in blocks. This includes inline inode
1531 * storage and unindexed ocfs2_xattr_blocks.
1533 static const struct ocfs2_xa_loc_operations ocfs2_xa_block_loc_ops = {
1534 .xlo_offset_pointer = ocfs2_xa_block_offset_pointer,
1535 .xlo_wipe_namevalue = ocfs2_xa_block_wipe_namevalue,
1538 static void *ocfs2_xa_bucket_offset_pointer(struct ocfs2_xa_loc *loc,
1541 struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1542 int block, block_offset;
1544 BUG_ON(offset >= OCFS2_XATTR_BUCKET_SIZE);
1546 /* The header is at the front of the bucket */
1547 block = offset >> bucket->bu_inode->i_sb->s_blocksize_bits;
1548 block_offset = offset % bucket->bu_inode->i_sb->s_blocksize;
1550 return bucket_block(bucket, block) + block_offset;
1553 static void ocfs2_xa_bucket_wipe_namevalue(struct ocfs2_xa_loc *loc)
1556 struct ocfs2_xattr_entry *entry = loc->xl_entry;
1557 u64 value_size = le64_to_cpu(entry->xe_value_size);
1559 namevalue_size = OCFS2_XATTR_SIZE(entry->xe_name_len);
1560 if (value_size > OCFS2_XATTR_INLINE_SIZE)
1561 namevalue_size += OCFS2_XATTR_ROOT_SIZE;
1563 namevalue_size += OCFS2_XATTR_SIZE(value_size);
1565 le16_add_cpu(&loc->xl_header->xh_name_value_len, -namevalue_size);
1568 /* Operations for xattrs stored in buckets. */
1569 static const struct ocfs2_xa_loc_operations ocfs2_xa_bucket_loc_ops = {
1570 .xlo_offset_pointer = ocfs2_xa_bucket_offset_pointer,
1571 .xlo_wipe_namevalue = ocfs2_xa_bucket_wipe_namevalue,
1574 static void ocfs2_xa_remove_entry(struct ocfs2_xa_loc *loc)
1576 ocfs2_xa_wipe_namevalue(loc);
1579 static void ocfs2_init_dinode_xa_loc(struct ocfs2_xa_loc *loc,
1580 struct inode *inode,
1581 struct buffer_head *bh,
1582 struct ocfs2_xattr_entry *entry)
1584 struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1586 loc->xl_ops = &ocfs2_xa_block_loc_ops;
1587 loc->xl_storage = bh;
1588 loc->xl_entry = entry;
1590 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1591 loc->xl_size = le16_to_cpu(di->i_xattr_inline_size);
1594 loc->xl_size = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1597 (struct ocfs2_xattr_header *)(bh->b_data + bh->b_size -
1601 static void ocfs2_init_xattr_block_xa_loc(struct ocfs2_xa_loc *loc,
1602 struct buffer_head *bh,
1603 struct ocfs2_xattr_entry *entry)
1605 struct ocfs2_xattr_block *xb =
1606 (struct ocfs2_xattr_block *)bh->b_data;
1608 BUG_ON(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED);
1610 loc->xl_ops = &ocfs2_xa_block_loc_ops;
1611 loc->xl_storage = bh;
1612 loc->xl_header = &(xb->xb_attrs.xb_header);
1613 loc->xl_entry = entry;
1614 loc->xl_size = bh->b_size - offsetof(struct ocfs2_xattr_block,
1615 xb_attrs.xb_header);
1618 static void ocfs2_init_xattr_bucket_xa_loc(struct ocfs2_xa_loc *loc,
1619 struct ocfs2_xattr_bucket *bucket,
1620 struct ocfs2_xattr_entry *entry)
1622 loc->xl_ops = &ocfs2_xa_bucket_loc_ops;
1623 loc->xl_storage = bucket;
1624 loc->xl_header = bucket_xh(bucket);
1625 loc->xl_entry = entry;
1626 loc->xl_size = OCFS2_XATTR_BUCKET_SIZE;
1630 * ocfs2_xattr_set_entry_local()
1632 * Set, replace or remove extended attribute in local.
1634 static void ocfs2_xattr_set_entry_local(struct inode *inode,
1635 struct ocfs2_xattr_info *xi,
1636 struct ocfs2_xattr_search *xs,
1637 struct ocfs2_xattr_entry *last,
1640 size_t name_len = strlen(xi->name);
1642 struct ocfs2_xa_loc loc;
1644 if (xs->xattr_bh == xs->inode_bh)
1645 ocfs2_init_dinode_xa_loc(&loc, inode, xs->inode_bh,
1646 xs->not_found ? NULL : xs->here);
1648 ocfs2_init_xattr_block_xa_loc(&loc, xs->xattr_bh,
1649 xs->not_found ? NULL : xs->here);
1650 if (xi->value && xs->not_found) {
1651 /* Insert the new xattr entry. */
1652 le16_add_cpu(&xs->header->xh_count, 1);
1653 ocfs2_xattr_set_type(last, xi->name_index);
1654 ocfs2_xattr_set_local(last, 1);
1655 last->xe_name_len = name_len;
1661 first_val = xs->base + min_offs;
1662 offs = le16_to_cpu(xs->here->xe_name_offset);
1663 val = xs->base + offs;
1665 if (le64_to_cpu(xs->here->xe_value_size) >
1666 OCFS2_XATTR_INLINE_SIZE)
1667 size = OCFS2_XATTR_SIZE(name_len) +
1668 OCFS2_XATTR_ROOT_SIZE;
1670 size = OCFS2_XATTR_SIZE(name_len) +
1671 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1673 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1674 OCFS2_XATTR_SIZE(xi->value_len)) {
1675 /* The old and the new value have the
1676 same size. Just replace the value. */
1677 ocfs2_xattr_set_local(xs->here, 1);
1678 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1679 /* Clear value bytes. */
1680 memset(val + OCFS2_XATTR_SIZE(name_len),
1682 OCFS2_XATTR_SIZE(xi->value_len));
1683 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1689 /* Remove the old name+value. */
1690 ocfs2_xa_wipe_namevalue(&loc);
1691 xs->here->xe_name_hash = 0;
1692 xs->here->xe_name_offset = 0;
1693 ocfs2_xattr_set_local(xs->here, 1);
1694 xs->here->xe_value_size = 0;
1699 /* Remove the old entry. */
1700 i = le16_to_cpu(xs->header->xh_count) - 1;
1701 last = &xs->header->xh_entries[i];
1702 xs->header->xh_count = cpu_to_le16(i);
1704 memmove(xs->here, xs->here + 1,
1705 (void *)last - (void *)xs->here);
1706 memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1710 /* Insert the new name+value. */
1711 size_t size = OCFS2_XATTR_SIZE(name_len) +
1712 OCFS2_XATTR_SIZE(xi->value_len);
1713 void *val = xs->base + min_offs - size;
1715 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1716 memset(val, 0, size);
1717 memcpy(val, xi->name, name_len);
1718 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1721 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1722 ocfs2_xattr_set_local(xs->here, 1);
1723 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1730 * ocfs2_xattr_set_entry()
1732 * Set extended attribute entry into inode or block.
1734 * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1735 * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1736 * then set value in B tree with set_value_outside().
1738 static int ocfs2_xattr_set_entry(struct inode *inode,
1739 struct ocfs2_xattr_info *xi,
1740 struct ocfs2_xattr_search *xs,
1741 struct ocfs2_xattr_set_ctxt *ctxt,
1744 struct ocfs2_xattr_entry *last;
1745 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1746 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1747 size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1749 handle_t *handle = ctxt->handle;
1751 struct ocfs2_xattr_info xi_l = {
1752 .name_index = xi->name_index,
1755 .value_len = xi->value_len,
1757 struct ocfs2_xattr_value_buf vb = {
1758 .vb_bh = xs->xattr_bh,
1759 .vb_access = ocfs2_journal_access_di,
1762 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1763 BUG_ON(xs->xattr_bh == xs->inode_bh);
1764 vb.vb_access = ocfs2_journal_access_xb;
1766 BUG_ON(xs->xattr_bh != xs->inode_bh);
1768 /* Compute min_offs, last and free space. */
1769 last = xs->header->xh_entries;
1771 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1772 size_t offs = le16_to_cpu(last->xe_name_offset);
1773 if (offs < min_offs)
1778 free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
1782 if (!xs->not_found) {
1784 if (ocfs2_xattr_is_local(xs->here))
1785 size = OCFS2_XATTR_SIZE(name_len) +
1786 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1788 size = OCFS2_XATTR_SIZE(name_len) +
1789 OCFS2_XATTR_ROOT_SIZE;
1790 free += (size + sizeof(struct ocfs2_xattr_entry));
1792 /* Check free space in inode or block */
1793 if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1794 if (free < sizeof(struct ocfs2_xattr_entry) +
1795 OCFS2_XATTR_SIZE(name_len) +
1796 OCFS2_XATTR_ROOT_SIZE) {
1800 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1801 xi_l.value = (void *)&def_xv;
1802 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1803 } else if (xi->value) {
1804 if (free < sizeof(struct ocfs2_xattr_entry) +
1805 OCFS2_XATTR_SIZE(name_len) +
1806 OCFS2_XATTR_SIZE(xi->value_len)) {
1812 if (!xs->not_found) {
1813 /* For existing extended attribute */
1814 size_t size = OCFS2_XATTR_SIZE(name_len) +
1815 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1816 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1817 void *val = xs->base + offs;
1819 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1820 /* Replace existing local xattr with tree root */
1821 ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1826 } else if (!ocfs2_xattr_is_local(xs->here)) {
1827 /* For existing xattr which has value outside */
1828 vb.vb_xv = (struct ocfs2_xattr_value_root *)
1829 (val + OCFS2_XATTR_SIZE(name_len));
1831 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1833 * If new value need set outside also,
1834 * first truncate old value to new value,
1835 * then set new value with set_value_outside().
1837 ret = ocfs2_xattr_value_truncate(inode,
1846 ret = ocfs2_xattr_update_entry(inode,
1857 ret = __ocfs2_xattr_set_value_outside(inode,
1867 * If new value need set in local,
1868 * just trucate old value to zero.
1870 ret = ocfs2_xattr_value_truncate(inode,
1880 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), xs->inode_bh,
1881 OCFS2_JOURNAL_ACCESS_WRITE);
1887 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1888 ret = vb.vb_access(handle, INODE_CACHE(inode), vb.vb_bh,
1889 OCFS2_JOURNAL_ACCESS_WRITE);
1897 * Set value in local, include set tree root in local.
1898 * This is the first step for value size >INLINE_SIZE.
1900 ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1902 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1903 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1910 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1911 (flag & OCFS2_INLINE_XATTR_FL)) {
1912 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1913 unsigned int xattrsize = osb->s_xattr_inline_size;
1916 * Adjust extent record count or inline data size
1917 * to reserve space for extended attribute.
1919 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1920 struct ocfs2_inline_data *idata = &di->id2.i_data;
1921 le16_add_cpu(&idata->id_count, -xattrsize);
1922 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1923 struct ocfs2_extent_list *el = &di->id2.i_list;
1924 le16_add_cpu(&el->l_count, -(xattrsize /
1925 sizeof(struct ocfs2_extent_rec)));
1927 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1929 /* Update xattr flag */
1930 spin_lock(&oi->ip_lock);
1931 oi->ip_dyn_features |= flag;
1932 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1933 spin_unlock(&oi->ip_lock);
1935 ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1939 if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1941 * Set value outside in B tree.
1942 * This is the second step for value size > INLINE_SIZE.
1944 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1945 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, ctxt,
1952 * If set value outside failed, we have to clean
1953 * the junk tree root we have already set in local.
1955 ret2 = ocfs2_xattr_cleanup(inode, ctxt->handle,
1966 * In xattr remove, if it is stored outside and refcounted, we may have
1967 * the chance to split the refcount tree. So need the allocators.
1969 static int ocfs2_lock_xattr_remove_allocators(struct inode *inode,
1970 struct ocfs2_xattr_value_root *xv,
1971 struct ocfs2_caching_info *ref_ci,
1972 struct buffer_head *ref_root_bh,
1973 struct ocfs2_alloc_context **meta_ac,
1976 int ret, meta_add = 0;
1977 u32 p_cluster, num_clusters;
1978 unsigned int ext_flags;
1981 ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
1990 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
1993 ret = ocfs2_refcounted_xattr_delete_need(inode, ref_ci,
1995 &meta_add, ref_credits);
2001 ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
2010 static int ocfs2_remove_value_outside(struct inode*inode,
2011 struct ocfs2_xattr_value_buf *vb,
2012 struct ocfs2_xattr_header *header,
2013 struct ocfs2_caching_info *ref_ci,
2014 struct buffer_head *ref_root_bh)
2016 int ret = 0, i, ref_credits;
2017 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2018 struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2021 ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
2023 for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
2024 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
2026 if (ocfs2_xattr_is_local(entry))
2029 val = (void *)header +
2030 le16_to_cpu(entry->xe_name_offset);
2031 vb->vb_xv = (struct ocfs2_xattr_value_root *)
2032 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
2034 ret = ocfs2_lock_xattr_remove_allocators(inode, vb->vb_xv,
2035 ref_ci, ref_root_bh,
2039 ctxt.handle = ocfs2_start_trans(osb, ref_credits +
2040 ocfs2_remove_extent_credits(osb->sb));
2041 if (IS_ERR(ctxt.handle)) {
2042 ret = PTR_ERR(ctxt.handle);
2047 ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
2053 ocfs2_commit_trans(osb, ctxt.handle);
2055 ocfs2_free_alloc_context(ctxt.meta_ac);
2056 ctxt.meta_ac = NULL;
2061 ocfs2_free_alloc_context(ctxt.meta_ac);
2062 ocfs2_schedule_truncate_log_flush(osb, 1);
2063 ocfs2_run_deallocs(osb, &ctxt.dealloc);
2067 static int ocfs2_xattr_ibody_remove(struct inode *inode,
2068 struct buffer_head *di_bh,
2069 struct ocfs2_caching_info *ref_ci,
2070 struct buffer_head *ref_root_bh)
2073 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2074 struct ocfs2_xattr_header *header;
2076 struct ocfs2_xattr_value_buf vb = {
2078 .vb_access = ocfs2_journal_access_di,
2081 header = (struct ocfs2_xattr_header *)
2082 ((void *)di + inode->i_sb->s_blocksize -
2083 le16_to_cpu(di->i_xattr_inline_size));
2085 ret = ocfs2_remove_value_outside(inode, &vb, header,
2086 ref_ci, ref_root_bh);
2091 struct ocfs2_rm_xattr_bucket_para {
2092 struct ocfs2_caching_info *ref_ci;
2093 struct buffer_head *ref_root_bh;
2096 static int ocfs2_xattr_block_remove(struct inode *inode,
2097 struct buffer_head *blk_bh,
2098 struct ocfs2_caching_info *ref_ci,
2099 struct buffer_head *ref_root_bh)
2101 struct ocfs2_xattr_block *xb;
2103 struct ocfs2_xattr_value_buf vb = {
2105 .vb_access = ocfs2_journal_access_xb,
2107 struct ocfs2_rm_xattr_bucket_para args = {
2109 .ref_root_bh = ref_root_bh,
2112 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2113 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2114 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
2115 ret = ocfs2_remove_value_outside(inode, &vb, header,
2116 ref_ci, ref_root_bh);
2118 ret = ocfs2_iterate_xattr_index_block(inode,
2120 ocfs2_rm_xattr_cluster,
2126 static int ocfs2_xattr_free_block(struct inode *inode,
2128 struct ocfs2_caching_info *ref_ci,
2129 struct buffer_head *ref_root_bh)
2131 struct inode *xb_alloc_inode;
2132 struct buffer_head *xb_alloc_bh = NULL;
2133 struct buffer_head *blk_bh = NULL;
2134 struct ocfs2_xattr_block *xb;
2135 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2141 ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
2147 ret = ocfs2_xattr_block_remove(inode, blk_bh, ref_ci, ref_root_bh);
2153 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2154 blk = le64_to_cpu(xb->xb_blkno);
2155 bit = le16_to_cpu(xb->xb_suballoc_bit);
2156 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2158 xb_alloc_inode = ocfs2_get_system_file_inode(osb,
2159 EXTENT_ALLOC_SYSTEM_INODE,
2160 le16_to_cpu(xb->xb_suballoc_slot));
2161 if (!xb_alloc_inode) {
2166 mutex_lock(&xb_alloc_inode->i_mutex);
2168 ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
2174 handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
2175 if (IS_ERR(handle)) {
2176 ret = PTR_ERR(handle);
2181 ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
2186 ocfs2_commit_trans(osb, handle);
2188 ocfs2_inode_unlock(xb_alloc_inode, 1);
2189 brelse(xb_alloc_bh);
2191 mutex_unlock(&xb_alloc_inode->i_mutex);
2192 iput(xb_alloc_inode);
2199 * ocfs2_xattr_remove()
2201 * Free extended attribute resources associated with this inode.
2203 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
2205 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2206 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2207 struct ocfs2_refcount_tree *ref_tree = NULL;
2208 struct buffer_head *ref_root_bh = NULL;
2209 struct ocfs2_caching_info *ref_ci = NULL;
2213 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2216 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
2219 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
2220 ret = ocfs2_lock_refcount_tree(OCFS2_SB(inode->i_sb),
2221 le64_to_cpu(di->i_refcount_loc),
2222 1, &ref_tree, &ref_root_bh);
2227 ref_ci = &ref_tree->rf_ci;
2231 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2232 ret = ocfs2_xattr_ibody_remove(inode, di_bh,
2233 ref_ci, ref_root_bh);
2240 if (di->i_xattr_loc) {
2241 ret = ocfs2_xattr_free_block(inode,
2242 le64_to_cpu(di->i_xattr_loc),
2243 ref_ci, ref_root_bh);
2250 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
2251 OCFS2_INODE_UPDATE_CREDITS);
2252 if (IS_ERR(handle)) {
2253 ret = PTR_ERR(handle);
2257 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2258 OCFS2_JOURNAL_ACCESS_WRITE);
2264 di->i_xattr_loc = 0;
2266 spin_lock(&oi->ip_lock);
2267 oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
2268 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2269 spin_unlock(&oi->ip_lock);
2271 ret = ocfs2_journal_dirty(handle, di_bh);
2275 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
2278 ocfs2_unlock_refcount_tree(OCFS2_SB(inode->i_sb), ref_tree, 1);
2279 brelse(ref_root_bh);
2283 static int ocfs2_xattr_has_space_inline(struct inode *inode,
2284 struct ocfs2_dinode *di)
2286 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2287 unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
2290 if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
2293 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2294 struct ocfs2_inline_data *idata = &di->id2.i_data;
2295 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
2296 } else if (ocfs2_inode_is_fast_symlink(inode)) {
2297 free = ocfs2_fast_symlink_chars(inode->i_sb) -
2298 le64_to_cpu(di->i_size);
2300 struct ocfs2_extent_list *el = &di->id2.i_list;
2301 free = (le16_to_cpu(el->l_count) -
2302 le16_to_cpu(el->l_next_free_rec)) *
2303 sizeof(struct ocfs2_extent_rec);
2305 if (free >= xattrsize)
2312 * ocfs2_xattr_ibody_find()
2314 * Find extended attribute in inode block and
2315 * fill search info into struct ocfs2_xattr_search.
2317 static int ocfs2_xattr_ibody_find(struct inode *inode,
2320 struct ocfs2_xattr_search *xs)
2322 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2323 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2327 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2330 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2331 down_read(&oi->ip_alloc_sem);
2332 has_space = ocfs2_xattr_has_space_inline(inode, di);
2333 up_read(&oi->ip_alloc_sem);
2338 xs->xattr_bh = xs->inode_bh;
2339 xs->end = (void *)di + inode->i_sb->s_blocksize;
2340 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
2341 xs->header = (struct ocfs2_xattr_header *)
2342 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
2344 xs->header = (struct ocfs2_xattr_header *)
2345 (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
2346 xs->base = (void *)xs->header;
2347 xs->here = xs->header->xh_entries;
2349 /* Find the named attribute. */
2350 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2351 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2352 if (ret && ret != -ENODATA)
2354 xs->not_found = ret;
2361 * ocfs2_xattr_ibody_set()
2363 * Set, replace or remove an extended attribute into inode block.
2366 static int ocfs2_xattr_ibody_set(struct inode *inode,
2367 struct ocfs2_xattr_info *xi,
2368 struct ocfs2_xattr_search *xs,
2369 struct ocfs2_xattr_set_ctxt *ctxt)
2371 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2372 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2375 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2378 down_write(&oi->ip_alloc_sem);
2379 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2380 if (!ocfs2_xattr_has_space_inline(inode, di)) {
2386 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2387 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
2389 up_write(&oi->ip_alloc_sem);
2395 * ocfs2_xattr_block_find()
2397 * Find extended attribute in external block and
2398 * fill search info into struct ocfs2_xattr_search.
2400 static int ocfs2_xattr_block_find(struct inode *inode,
2403 struct ocfs2_xattr_search *xs)
2405 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2406 struct buffer_head *blk_bh = NULL;
2407 struct ocfs2_xattr_block *xb;
2410 if (!di->i_xattr_loc)
2413 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2420 xs->xattr_bh = blk_bh;
2421 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2423 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2424 xs->header = &xb->xb_attrs.xb_header;
2425 xs->base = (void *)xs->header;
2426 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2427 xs->here = xs->header->xh_entries;
2429 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2431 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2435 if (ret && ret != -ENODATA) {
2436 xs->xattr_bh = NULL;
2439 xs->not_found = ret;
2447 static int ocfs2_create_xattr_block(handle_t *handle,
2448 struct inode *inode,
2449 struct buffer_head *inode_bh,
2450 struct ocfs2_alloc_context *meta_ac,
2451 struct buffer_head **ret_bh,
2455 u16 suballoc_bit_start;
2458 struct ocfs2_dinode *di = (struct ocfs2_dinode *)inode_bh->b_data;
2459 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2460 struct buffer_head *new_bh = NULL;
2461 struct ocfs2_xattr_block *xblk;
2463 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), inode_bh,
2464 OCFS2_JOURNAL_ACCESS_CREATE);
2470 ret = ocfs2_claim_metadata(osb, handle, meta_ac, 1,
2471 &suballoc_bit_start, &num_got,
2478 new_bh = sb_getblk(inode->i_sb, first_blkno);
2479 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2481 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode),
2483 OCFS2_JOURNAL_ACCESS_CREATE);
2489 /* Initialize ocfs2_xattr_block */
2490 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2491 memset(xblk, 0, inode->i_sb->s_blocksize);
2492 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2493 xblk->xb_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
2494 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2495 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
2496 xblk->xb_blkno = cpu_to_le64(first_blkno);
2499 struct ocfs2_xattr_tree_root *xr = &xblk->xb_attrs.xb_root;
2500 xr->xt_clusters = cpu_to_le32(1);
2501 xr->xt_last_eb_blk = 0;
2502 xr->xt_list.l_tree_depth = 0;
2503 xr->xt_list.l_count = cpu_to_le16(
2504 ocfs2_xattr_recs_per_xb(inode->i_sb));
2505 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2506 xblk->xb_flags = cpu_to_le16(OCFS2_XATTR_INDEXED);
2509 ret = ocfs2_journal_dirty(handle, new_bh);
2514 di->i_xattr_loc = cpu_to_le64(first_blkno);
2515 ocfs2_journal_dirty(handle, inode_bh);
2526 * ocfs2_xattr_block_set()
2528 * Set, replace or remove an extended attribute into external block.
2531 static int ocfs2_xattr_block_set(struct inode *inode,
2532 struct ocfs2_xattr_info *xi,
2533 struct ocfs2_xattr_search *xs,
2534 struct ocfs2_xattr_set_ctxt *ctxt)
2536 struct buffer_head *new_bh = NULL;
2537 handle_t *handle = ctxt->handle;
2538 struct ocfs2_xattr_block *xblk = NULL;
2541 if (!xs->xattr_bh) {
2542 ret = ocfs2_create_xattr_block(handle, inode, xs->inode_bh,
2543 ctxt->meta_ac, &new_bh, 0);
2549 xs->xattr_bh = new_bh;
2550 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2551 xs->header = &xblk->xb_attrs.xb_header;
2552 xs->base = (void *)xs->header;
2553 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2554 xs->here = xs->header->xh_entries;
2556 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2558 if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2559 /* Set extended attribute into external block */
2560 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2561 OCFS2_HAS_XATTR_FL);
2562 if (!ret || ret != -ENOSPC)
2565 ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2570 ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2577 /* Check whether the new xattr can be inserted into the inode. */
2578 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2579 struct ocfs2_xattr_info *xi,
2580 struct ocfs2_xattr_search *xs)
2583 struct ocfs2_xattr_entry *last;
2585 size_t min_offs = xs->end - xs->base;
2590 last = xs->header->xh_entries;
2592 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
2593 size_t offs = le16_to_cpu(last->xe_name_offset);
2594 if (offs < min_offs)
2599 free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
2603 BUG_ON(!xs->not_found);
2605 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2606 value_size = OCFS2_XATTR_ROOT_SIZE;
2608 value_size = OCFS2_XATTR_SIZE(xi->value_len);
2610 if (free >= sizeof(struct ocfs2_xattr_entry) +
2611 OCFS2_XATTR_SIZE(strlen(xi->name)) + value_size)
2617 static int ocfs2_calc_xattr_set_need(struct inode *inode,
2618 struct ocfs2_dinode *di,
2619 struct ocfs2_xattr_info *xi,
2620 struct ocfs2_xattr_search *xis,
2621 struct ocfs2_xattr_search *xbs,
2626 int ret = 0, old_in_xb = 0;
2627 int clusters_add = 0, meta_add = 0, credits = 0;
2628 struct buffer_head *bh = NULL;
2629 struct ocfs2_xattr_block *xb = NULL;
2630 struct ocfs2_xattr_entry *xe = NULL;
2631 struct ocfs2_xattr_value_root *xv = NULL;
2633 int name_offset, name_len = 0;
2634 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2639 * Calculate the clusters we need to write.
2640 * No matter whether we replace an old one or add a new one,
2641 * we need this for writing.
2643 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2644 credits += new_clusters *
2645 ocfs2_clusters_to_blocks(inode->i_sb, 1);
2647 if (xis->not_found && xbs->not_found) {
2648 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2650 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2651 clusters_add += new_clusters;
2652 credits += ocfs2_calc_extend_credits(inode->i_sb,
2660 if (!xis->not_found) {
2662 name_offset = le16_to_cpu(xe->xe_name_offset);
2663 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2665 credits += OCFS2_INODE_UPDATE_CREDITS;
2667 int i, block_off = 0;
2668 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2670 name_offset = le16_to_cpu(xe->xe_name_offset);
2671 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2672 i = xbs->here - xbs->header->xh_entries;
2675 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2676 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
2677 bucket_xh(xbs->bucket),
2680 base = bucket_block(xbs->bucket, block_off);
2681 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2684 credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
2689 * delete a xattr doesn't need metadata and cluster allocation.
2690 * so just calculate the credits and return.
2692 * The credits for removing the value tree will be extended
2693 * by ocfs2_remove_extent itself.
2696 if (!ocfs2_xattr_is_local(xe))
2697 credits += ocfs2_remove_extent_credits(inode->i_sb);
2702 /* do cluster allocation guess first. */
2703 value_size = le64_to_cpu(xe->xe_value_size);
2707 * In xattr set, we always try to set the xe in inode first,
2708 * so if it can be inserted into inode successfully, the old
2709 * one will be removed from the xattr block, and this xattr
2710 * will be inserted into inode as a new xattr in inode.
2712 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
2713 clusters_add += new_clusters;
2714 credits += ocfs2_remove_extent_credits(inode->i_sb) +
2715 OCFS2_INODE_UPDATE_CREDITS;
2716 if (!ocfs2_xattr_is_local(xe))
2717 credits += ocfs2_calc_extend_credits(
2725 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2726 /* the new values will be stored outside. */
2727 u32 old_clusters = 0;
2729 if (!ocfs2_xattr_is_local(xe)) {
2730 old_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2732 xv = (struct ocfs2_xattr_value_root *)
2733 (base + name_offset + name_len);
2734 value_size = OCFS2_XATTR_ROOT_SIZE;
2738 if (old_clusters >= new_clusters) {
2739 credits += ocfs2_remove_extent_credits(inode->i_sb);
2742 meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
2743 clusters_add += new_clusters - old_clusters;
2744 credits += ocfs2_calc_extend_credits(inode->i_sb,
2748 if (value_size >= OCFS2_XATTR_ROOT_SIZE)
2753 * Now the new value will be stored inside. So if the new
2754 * value is smaller than the size of value root or the old
2755 * value, we don't need any allocation, otherwise we have
2756 * to guess metadata allocation.
2758 if ((ocfs2_xattr_is_local(xe) && value_size >= xi->value_len) ||
2759 (!ocfs2_xattr_is_local(xe) &&
2760 OCFS2_XATTR_ROOT_SIZE >= xi->value_len))
2765 /* calculate metadata allocation. */
2766 if (di->i_xattr_loc) {
2767 if (!xbs->xattr_bh) {
2768 ret = ocfs2_read_xattr_block(inode,
2769 le64_to_cpu(di->i_xattr_loc),
2776 xb = (struct ocfs2_xattr_block *)bh->b_data;
2778 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2781 * If there is already an xattr tree, good, we can calculate
2782 * like other b-trees. Otherwise we may have the chance of
2783 * create a tree, the credit calculation is borrowed from
2784 * ocfs2_calc_extend_credits with root_el = NULL. And the
2785 * new tree will be cluster based, so no meta is needed.
2787 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2788 struct ocfs2_extent_list *el =
2789 &xb->xb_attrs.xb_root.xt_list;
2790 meta_add += ocfs2_extend_meta_needed(el);
2791 credits += ocfs2_calc_extend_credits(inode->i_sb,
2794 credits += OCFS2_SUBALLOC_ALLOC + 1;
2797 * This cluster will be used either for new bucket or for
2799 * If the cluster size is the same as the bucket size, one
2800 * more is needed since we may need to extend the bucket
2804 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2805 if (OCFS2_XATTR_BUCKET_SIZE ==
2806 OCFS2_SB(inode->i_sb)->s_clustersize) {
2807 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2812 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
2816 *clusters_need = clusters_add;
2818 *meta_need = meta_add;
2820 *credits_need = credits;
2825 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
2826 struct ocfs2_dinode *di,
2827 struct ocfs2_xattr_info *xi,
2828 struct ocfs2_xattr_search *xis,
2829 struct ocfs2_xattr_search *xbs,
2830 struct ocfs2_xattr_set_ctxt *ctxt,
2834 int clusters_add, meta_add, ret;
2835 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2837 memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
2839 ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
2841 ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
2842 &clusters_add, &meta_add, credits);
2848 meta_add += extra_meta;
2849 mlog(0, "Set xattr %s, reserve meta blocks = %d, clusters = %d, "
2850 "credits = %d\n", xi->name, meta_add, clusters_add, *credits);
2853 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
2862 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
2868 if (ctxt->meta_ac) {
2869 ocfs2_free_alloc_context(ctxt->meta_ac);
2870 ctxt->meta_ac = NULL;
2874 * We cannot have an error and a non null ctxt->data_ac.
2881 static int __ocfs2_xattr_set_handle(struct inode *inode,
2882 struct ocfs2_dinode *di,
2883 struct ocfs2_xattr_info *xi,
2884 struct ocfs2_xattr_search *xis,
2885 struct ocfs2_xattr_search *xbs,
2886 struct ocfs2_xattr_set_ctxt *ctxt)
2888 int ret = 0, credits, old_found;
2891 /* Remove existing extended attribute */
2892 if (!xis->not_found)
2893 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2894 else if (!xbs->not_found)
2895 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2897 /* We always try to set extended attribute into inode first*/
2898 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2899 if (!ret && !xbs->not_found) {
2901 * If succeed and that extended attribute existing in
2902 * external block, then we will remove it.
2907 old_found = xis->not_found;
2908 xis->not_found = -ENODATA;
2909 ret = ocfs2_calc_xattr_set_need(inode,
2917 xis->not_found = old_found;
2923 ret = ocfs2_extend_trans(ctxt->handle, credits +
2924 ctxt->handle->h_buffer_credits);
2929 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2930 } else if (ret == -ENOSPC) {
2931 if (di->i_xattr_loc && !xbs->xattr_bh) {
2932 ret = ocfs2_xattr_block_find(inode,
2938 old_found = xis->not_found;
2939 xis->not_found = -ENODATA;
2940 ret = ocfs2_calc_xattr_set_need(inode,
2948 xis->not_found = old_found;
2954 ret = ocfs2_extend_trans(ctxt->handle, credits +
2955 ctxt->handle->h_buffer_credits);
2962 * If no space in inode, we will set extended attribute
2963 * into external block.
2965 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2968 if (!xis->not_found) {
2970 * If succeed and that extended attribute
2971 * existing in inode, we will remove it.
2975 xbs->not_found = -ENODATA;
2976 ret = ocfs2_calc_xattr_set_need(inode,
2989 ret = ocfs2_extend_trans(ctxt->handle, credits +
2990 ctxt->handle->h_buffer_credits);
2995 ret = ocfs2_xattr_ibody_set(inode, xi,
3002 /* Update inode ctime. */
3003 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
3005 OCFS2_JOURNAL_ACCESS_WRITE);
3011 inode->i_ctime = CURRENT_TIME;
3012 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
3013 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
3014 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
3021 * This function only called duing creating inode
3022 * for init security/acl xattrs of the new inode.
3023 * All transanction credits have been reserved in mknod.
3025 int ocfs2_xattr_set_handle(handle_t *handle,
3026 struct inode *inode,
3027 struct buffer_head *di_bh,
3033 struct ocfs2_alloc_context *meta_ac,
3034 struct ocfs2_alloc_context *data_ac)
3036 struct ocfs2_dinode *di;
3039 struct ocfs2_xattr_info xi = {
3040 .name_index = name_index,
3043 .value_len = value_len,
3046 struct ocfs2_xattr_search xis = {
3047 .not_found = -ENODATA,
3050 struct ocfs2_xattr_search xbs = {
3051 .not_found = -ENODATA,
3054 struct ocfs2_xattr_set_ctxt ctxt = {
3060 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3064 * In extreme situation, may need xattr bucket when
3065 * block size is too small. And we have already reserved
3066 * the credits for bucket in mknod.
3068 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
3069 xbs.bucket = ocfs2_xattr_bucket_new(inode);
3071 mlog_errno(-ENOMEM);
3076 xis.inode_bh = xbs.inode_bh = di_bh;
3077 di = (struct ocfs2_dinode *)di_bh->b_data;
3079 down_write(&OCFS2_I(inode)->ip_xattr_sem);
3081 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3084 if (xis.not_found) {
3085 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3090 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3093 up_write(&OCFS2_I(inode)->ip_xattr_sem);
3094 brelse(xbs.xattr_bh);
3095 ocfs2_xattr_bucket_free(xbs.bucket);
3103 * Set, replace or remove an extended attribute for this inode.
3104 * value is NULL to remove an existing extended attribute, else either
3105 * create or replace an extended attribute.
3107 int ocfs2_xattr_set(struct inode *inode,
3114 struct buffer_head *di_bh = NULL;
3115 struct ocfs2_dinode *di;
3116 int ret, credits, ref_meta = 0, ref_credits = 0;
3117 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3118 struct inode *tl_inode = osb->osb_tl_inode;
3119 struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
3120 struct ocfs2_refcount_tree *ref_tree = NULL;
3122 struct ocfs2_xattr_info xi = {
3123 .name_index = name_index,
3126 .value_len = value_len,
3129 struct ocfs2_xattr_search xis = {
3130 .not_found = -ENODATA,
3133 struct ocfs2_xattr_search xbs = {
3134 .not_found = -ENODATA,
3137 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3141 * Only xbs will be used on indexed trees. xis doesn't need a
3144 xbs.bucket = ocfs2_xattr_bucket_new(inode);
3146 mlog_errno(-ENOMEM);
3150 ret = ocfs2_inode_lock(inode, &di_bh, 1);
3153 goto cleanup_nolock;
3155 xis.inode_bh = xbs.inode_bh = di_bh;
3156 di = (struct ocfs2_dinode *)di_bh->b_data;
3158 down_write(&OCFS2_I(inode)->ip_xattr_sem);
3160 * Scan inode and external block to find the same name
3161 * extended attribute and collect search infomation.
3163 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3166 if (xis.not_found) {
3167 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3172 if (xis.not_found && xbs.not_found) {
3174 if (flags & XATTR_REPLACE)
3181 if (flags & XATTR_CREATE)
3185 /* Check whether the value is refcounted and do some prepartion. */
3186 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL &&
3187 (!xis.not_found || !xbs.not_found)) {
3188 ret = ocfs2_prepare_refcount_xattr(inode, di, &xi,
3189 &xis, &xbs, &ref_tree,
3190 &ref_meta, &ref_credits);
3197 mutex_lock(&tl_inode->i_mutex);
3199 if (ocfs2_truncate_log_needs_flush(osb)) {
3200 ret = __ocfs2_flush_truncate_log(osb);
3202 mutex_unlock(&tl_inode->i_mutex);
3207 mutex_unlock(&tl_inode->i_mutex);
3209 ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
3210 &xbs, &ctxt, ref_meta, &credits);
3216 /* we need to update inode's ctime field, so add credit for it. */
3217 credits += OCFS2_INODE_UPDATE_CREDITS;
3218 ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
3219 if (IS_ERR(ctxt.handle)) {
3220 ret = PTR_ERR(ctxt.handle);
3225 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3227 ocfs2_commit_trans(osb, ctxt.handle);
3230 ocfs2_free_alloc_context(ctxt.data_ac);
3232 ocfs2_free_alloc_context(ctxt.meta_ac);
3233 if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
3234 ocfs2_schedule_truncate_log_flush(osb, 1);
3235 ocfs2_run_deallocs(osb, &ctxt.dealloc);
3239 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3240 up_write(&OCFS2_I(inode)->ip_xattr_sem);
3241 if (!value && !ret) {
3242 ret = ocfs2_try_remove_refcount_tree(inode, di_bh);
3246 ocfs2_inode_unlock(inode, 1);
3249 brelse(xbs.xattr_bh);
3250 ocfs2_xattr_bucket_free(xbs.bucket);
3256 * Find the xattr extent rec which may contains name_hash.
3257 * e_cpos will be the first name hash of the xattr rec.
3258 * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
3260 static int ocfs2_xattr_get_rec(struct inode *inode,
3265 struct ocfs2_extent_list *el)
3268 struct buffer_head *eb_bh = NULL;
3269 struct ocfs2_extent_block *eb;
3270 struct ocfs2_extent_rec *rec = NULL;
3273 if (el->l_tree_depth) {
3274 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, name_hash,
3281 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
3284 if (el->l_tree_depth) {
3285 ocfs2_error(inode->i_sb,
3286 "Inode %lu has non zero tree depth in "
3287 "xattr tree block %llu\n", inode->i_ino,
3288 (unsigned long long)eb_bh->b_blocknr);
3294 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
3295 rec = &el->l_recs[i];
3297 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
3298 e_blkno = le64_to_cpu(rec->e_blkno);
3304 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
3305 "record (%u, %u, 0) in xattr", inode->i_ino,
3306 le32_to_cpu(rec->e_cpos),
3307 ocfs2_rec_clusters(el, rec));
3312 *p_blkno = le64_to_cpu(rec->e_blkno);
3313 *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
3315 *e_cpos = le32_to_cpu(rec->e_cpos);
3321 typedef int (xattr_bucket_func)(struct inode *inode,
3322 struct ocfs2_xattr_bucket *bucket,
3325 static int ocfs2_find_xe_in_bucket(struct inode *inode,
3326 struct ocfs2_xattr_bucket *bucket,
3333 int i, ret = 0, cmp = 1, block_off, new_offset;
3334 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3335 size_t name_len = strlen(name);
3336 struct ocfs2_xattr_entry *xe = NULL;
3340 * We don't use binary search in the bucket because there
3341 * may be multiple entries with the same name hash.
3343 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3344 xe = &xh->xh_entries[i];
3346 if (name_hash > le32_to_cpu(xe->xe_name_hash))
3348 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
3351 cmp = name_index - ocfs2_xattr_get_type(xe);
3353 cmp = name_len - xe->xe_name_len;
3357 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3368 xe_name = bucket_block(bucket, block_off) + new_offset;
3369 if (!memcmp(name, xe_name, name_len)) {
3381 * Find the specified xattr entry in a series of buckets.
3382 * This series start from p_blkno and last for num_clusters.
3383 * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
3384 * the num of the valid buckets.
3386 * Return the buffer_head this xattr should reside in. And if the xattr's
3387 * hash is in the gap of 2 buckets, return the lower bucket.
3389 static int ocfs2_xattr_bucket_find(struct inode *inode,
3396 struct ocfs2_xattr_search *xs)
3399 struct ocfs2_xattr_header *xh = NULL;
3400 struct ocfs2_xattr_entry *xe = NULL;
3402 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3403 int low_bucket = 0, bucket, high_bucket;
3404 struct ocfs2_xattr_bucket *search;
3406 u64 blkno, lower_blkno = 0;
3408 search = ocfs2_xattr_bucket_new(inode);
3415 ret = ocfs2_read_xattr_bucket(search, p_blkno);
3421 xh = bucket_xh(search);
3422 high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
3423 while (low_bucket <= high_bucket) {
3424 ocfs2_xattr_bucket_relse(search);
3426 bucket = (low_bucket + high_bucket) / 2;
3427 blkno = p_blkno + bucket * blk_per_bucket;
3428 ret = ocfs2_read_xattr_bucket(search, blkno);
3434 xh = bucket_xh(search);
3435 xe = &xh->xh_entries[0];
3436 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3437 high_bucket = bucket - 1;
3442 * Check whether the hash of the last entry in our
3443 * bucket is larger than the search one. for an empty
3444 * bucket, the last one is also the first one.
3447 xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3449 last_hash = le32_to_cpu(xe->xe_name_hash);
3451 /* record lower_blkno which may be the insert place. */
3452 lower_blkno = blkno;
3454 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3455 low_bucket = bucket + 1;
3459 /* the searched xattr should reside in this bucket if exists. */
3460 ret = ocfs2_find_xe_in_bucket(inode, search,
3461 name_index, name, name_hash,
3471 * Record the bucket we have found.
3472 * When the xattr's hash value is in the gap of 2 buckets, we will
3473 * always set it to the previous bucket.
3476 lower_blkno = p_blkno;
3478 /* This should be in cache - we just read it during the search */
3479 ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3485 xs->header = bucket_xh(xs->bucket);
3486 xs->base = bucket_block(xs->bucket, 0);
3487 xs->end = xs->base + inode->i_sb->s_blocksize;
3490 xs->here = &xs->header->xh_entries[index];
3491 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
3492 (unsigned long long)bucket_blkno(xs->bucket), index);
3497 ocfs2_xattr_bucket_free(search);
3501 static int ocfs2_xattr_index_block_find(struct inode *inode,
3502 struct buffer_head *root_bh,
3505 struct ocfs2_xattr_search *xs)
3508 struct ocfs2_xattr_block *xb =
3509 (struct ocfs2_xattr_block *)root_bh->b_data;
3510 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3511 struct ocfs2_extent_list *el = &xb_root->xt_list;
3513 u32 first_hash, num_clusters = 0;
3514 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3516 if (le16_to_cpu(el->l_next_free_rec) == 0)
3519 mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
3520 name, name_hash, name_index);
3522 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3529 BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3531 mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
3532 "in the rec is %u\n", num_clusters, (unsigned long long)p_blkno,
3535 ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3536 p_blkno, first_hash, num_clusters, xs);
3542 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3545 xattr_bucket_func *func,
3549 u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3550 u32 num_buckets = clusters * bpc;
3551 struct ocfs2_xattr_bucket *bucket;
3553 bucket = ocfs2_xattr_bucket_new(inode);
3555 mlog_errno(-ENOMEM);
3559 mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
3560 clusters, (unsigned long long)blkno);
3562 for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3563 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3570 * The real bucket num in this series of blocks is stored
3571 * in the 1st bucket.
3574 num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3576 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
3577 (unsigned long long)blkno,
3578 le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3580 ret = func(inode, bucket, para);
3581 if (ret && ret != -ERANGE)
3583 /* Fall through to bucket_relse() */
3586 ocfs2_xattr_bucket_relse(bucket);
3591 ocfs2_xattr_bucket_free(bucket);
3595 struct ocfs2_xattr_tree_list {
3601 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
3602 struct ocfs2_xattr_header *xh,
3609 if (index < 0 || index >= le16_to_cpu(xh->xh_count))
3612 name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
3614 *block_off = name_offset >> sb->s_blocksize_bits;
3615 *new_offset = name_offset % sb->s_blocksize;
3620 static int ocfs2_list_xattr_bucket(struct inode *inode,
3621 struct ocfs2_xattr_bucket *bucket,
3625 struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
3626 int i, block_off, new_offset;
3627 const char *prefix, *name;
3629 for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
3630 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
3631 type = ocfs2_xattr_get_type(entry);
3632 prefix = ocfs2_xattr_prefix(type);
3635 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3643 name = (const char *)bucket_block(bucket, block_off) +
3645 ret = ocfs2_xattr_list_entry(xl->buffer,
3649 entry->xe_name_len);
3658 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
3659 struct buffer_head *blk_bh,
3660 xattr_tree_rec_func *rec_func,
3663 struct ocfs2_xattr_block *xb =
3664 (struct ocfs2_xattr_block *)blk_bh->b_data;
3665 struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
3667 u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
3670 if (!el->l_next_free_rec || !rec_func)
3673 while (name_hash > 0) {
3674 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
3675 &e_cpos, &num_clusters, el);
3681 ret = rec_func(inode, blk_bh, p_blkno, e_cpos,
3682 num_clusters, para);
3692 name_hash = e_cpos - 1;
3699 static int ocfs2_list_xattr_tree_rec(struct inode *inode,
3700 struct buffer_head *root_bh,
3701 u64 blkno, u32 cpos, u32 len, void *para)
3703 return ocfs2_iterate_xattr_buckets(inode, blkno, len,
3704 ocfs2_list_xattr_bucket, para);
3707 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
3708 struct buffer_head *blk_bh,
3713 struct ocfs2_xattr_tree_list xl = {
3715 .buffer_size = buffer_size,
3719 ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
3720 ocfs2_list_xattr_tree_rec, &xl);
3731 static int cmp_xe(const void *a, const void *b)
3733 const struct ocfs2_xattr_entry *l = a, *r = b;
3734 u32 l_hash = le32_to_cpu(l->xe_name_hash);
3735 u32 r_hash = le32_to_cpu(r->xe_name_hash);
3737 if (l_hash > r_hash)
3739 if (l_hash < r_hash)
3744 static void swap_xe(void *a, void *b, int size)
3746 struct ocfs2_xattr_entry *l = a, *r = b, tmp;
3749 memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
3750 memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
3754 * When the ocfs2_xattr_block is filled up, new bucket will be created
3755 * and all the xattr entries will be moved to the new bucket.
3756 * The header goes at the start of the bucket, and the names+values are
3757 * filled from the end. This is why *target starts as the last buffer.
3758 * Note: we need to sort the entries since they are not saved in order
3759 * in the ocfs2_xattr_block.
3761 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
3762 struct buffer_head *xb_bh,
3763 struct ocfs2_xattr_bucket *bucket)
3765 int i, blocksize = inode->i_sb->s_blocksize;
3766 int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3767 u16 offset, size, off_change;
3768 struct ocfs2_xattr_entry *xe;
3769 struct ocfs2_xattr_block *xb =
3770 (struct ocfs2_xattr_block *)xb_bh->b_data;
3771 struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
3772 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3773 u16 count = le16_to_cpu(xb_xh->xh_count);
3774 char *src = xb_bh->b_data;
3775 char *target = bucket_block(bucket, blks - 1);
3777 mlog(0, "cp xattr from block %llu to bucket %llu\n",
3778 (unsigned long long)xb_bh->b_blocknr,
3779 (unsigned long long)bucket_blkno(bucket));
3781 for (i = 0; i < blks; i++)
3782 memset(bucket_block(bucket, i), 0, blocksize);
3785 * Since the xe_name_offset is based on ocfs2_xattr_header,
3786 * there is a offset change corresponding to the change of
3787 * ocfs2_xattr_header's position.
3789 off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3790 xe = &xb_xh->xh_entries[count - 1];
3791 offset = le16_to_cpu(xe->xe_name_offset) + off_change;
3792 size = blocksize - offset;
3794 /* copy all the names and values. */
3795 memcpy(target + offset, src + offset, size);
3797 /* Init new header now. */
3798 xh->xh_count = xb_xh->xh_count;
3799 xh->xh_num_buckets = cpu_to_le16(1);
3800 xh->xh_name_value_len = cpu_to_le16(size);
3801 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
3803 /* copy all the entries. */
3804 target = bucket_block(bucket, 0);
3805 offset = offsetof(struct ocfs2_xattr_header, xh_entries);
3806 size = count * sizeof(struct ocfs2_xattr_entry);
3807 memcpy(target + offset, (char *)xb_xh + offset, size);
3809 /* Change the xe offset for all the xe because of the move. */
3810 off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
3811 offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3812 for (i = 0; i < count; i++)
3813 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
3815 mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
3816 offset, size, off_change);
3818 sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
3823 * After we move xattr from block to index btree, we have to
3824 * update ocfs2_xattr_search to the new xe and base.
3826 * When the entry is in xattr block, xattr_bh indicates the storage place.
3827 * While if the entry is in index b-tree, "bucket" indicates the
3828 * real place of the xattr.
3830 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
3831 struct ocfs2_xattr_search *xs,
3832 struct buffer_head *old_bh)
3834 char *buf = old_bh->b_data;
3835 struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
3836 struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
3839 xs->header = bucket_xh(xs->bucket);
3840 xs->base = bucket_block(xs->bucket, 0);
3841 xs->end = xs->base + inode->i_sb->s_blocksize;
3846 i = xs->here - old_xh->xh_entries;
3847 xs->here = &xs->header->xh_entries[i];
3850 static int ocfs2_xattr_create_index_block(struct inode *inode,
3851 struct ocfs2_xattr_search *xs,
3852 struct ocfs2_xattr_set_ctxt *ctxt)
3857 handle_t *handle = ctxt->handle;
3858 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3859 struct ocfs2_inode_info *oi = OCFS2_I(inode);
3860 struct buffer_head *xb_bh = xs->xattr_bh;
3861 struct ocfs2_xattr_block *xb =
3862 (struct ocfs2_xattr_block *)xb_bh->b_data;
3863 struct ocfs2_xattr_tree_root *xr;
3864 u16 xb_flags = le16_to_cpu(xb->xb_flags);
3866 mlog(0, "create xattr index block for %llu\n",
3867 (unsigned long long)xb_bh->b_blocknr);
3869 BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
3870 BUG_ON(!xs->bucket);
3874 * We can use this lock for now, and maybe move to a dedicated mutex
3875 * if performance becomes a problem later.
3877 down_write(&oi->ip_alloc_sem);
3879 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), xb_bh,
3880 OCFS2_JOURNAL_ACCESS_WRITE);
3886 ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac,
3887 1, 1, &bit_off, &len);
3894 * The bucket may spread in many blocks, and
3895 * we will only touch the 1st block and the last block
3896 * in the whole bucket(one for entry and one for data).
3898 blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
3900 mlog(0, "allocate 1 cluster from %llu to xattr block\n",
3901 (unsigned long long)blkno);
3903 ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
3909 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
3910 OCFS2_JOURNAL_ACCESS_CREATE);
3916 ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
3917 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
3919 ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
3921 /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
3922 memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
3923 offsetof(struct ocfs2_xattr_block, xb_attrs));
3925 xr = &xb->xb_attrs.xb_root;
3926 xr->xt_clusters = cpu_to_le32(1);
3927 xr->xt_last_eb_blk = 0;
3928 xr->xt_list.l_tree_depth = 0;
3929 xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
3930 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
3932 xr->xt_list.l_recs[0].e_cpos = 0;
3933 xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
3934 xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
3936 xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
3938 ocfs2_journal_dirty(handle, xb_bh);
3941 up_write(&oi->ip_alloc_sem);
3946 static int cmp_xe_offset(const void *a, const void *b)
3948 const struct ocfs2_xattr_entry *l = a, *r = b;
3949 u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
3950 u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
3952 if (l_name_offset < r_name_offset)
3954 if (l_name_offset > r_name_offset)
3960 * defrag a xattr bucket if we find that the bucket has some
3961 * holes beteen name/value pairs.
3962 * We will move all the name/value pairs to the end of the bucket
3963 * so that we can spare some space for insertion.
3965 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
3967 struct ocfs2_xattr_bucket *bucket)
3970 size_t end, offset, len, value_len;
3971 struct ocfs2_xattr_header *xh;
3972 char *entries, *buf, *bucket_buf = NULL;
3973 u64 blkno = bucket_blkno(bucket);
3975 size_t blocksize = inode->i_sb->s_blocksize;
3976 struct ocfs2_xattr_entry *xe;
3979 * In order to make the operation more efficient and generic,
3980 * we copy all the blocks into a contiguous memory and do the
3981 * defragment there, so if anything is error, we will not touch
3984 bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
3991 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3992 memcpy(buf, bucket_block(bucket, i), blocksize);
3994 ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
3995 OCFS2_JOURNAL_ACCESS_WRITE);
4001 xh = (struct ocfs2_xattr_header *)bucket_buf;
4002 entries = (char *)xh->xh_entries;
4003 xh_free_start = le16_to_cpu(xh->xh_free_start);
4005 mlog(0, "adjust xattr bucket in %llu, count = %u, "
4006 "xh_free_start = %u, xh_name_value_len = %u.\n",
4007 (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
4008 xh_free_start, le16_to_cpu(xh->xh_name_value_len));
4011 * sort all the entries by their offset.
4012 * the largest will be the first, so that we can
4013 * move them to the end one by one.
4015 sort(entries, le16_to_cpu(xh->xh_count),
4016 sizeof(struct ocfs2_xattr_entry),
4017 cmp_xe_offset, swap_xe);
4019 /* Move all name/values to the end of the bucket. */
4020 xe = xh->xh_entries;
4021 end = OCFS2_XATTR_BUCKET_SIZE;
4022 for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
4023 offset = le16_to_cpu(xe->xe_name_offset);
4024 if (ocfs2_xattr_is_local(xe))
4025 value_len = OCFS2_XATTR_SIZE(
4026 le64_to_cpu(xe->xe_value_size));
4028 value_len = OCFS2_XATTR_ROOT_SIZE;
4029 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
4032 * We must make sure that the name/value pair
4033 * exist in the same block. So adjust end to
4034 * the previous block end if needed.
4036 if (((end - len) / blocksize !=
4037 (end - 1) / blocksize))
4038 end = end - end % blocksize;
4040 if (end > offset + len) {
4041 memmove(bucket_buf + end - len,
4042 bucket_buf + offset, len);
4043 xe->xe_name_offset = cpu_to_le16(end - len);
4046 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
4047 "bucket %llu\n", (unsigned long long)blkno);
4052 mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
4053 "bucket %llu\n", (unsigned long long)blkno);
4055 if (xh_free_start == end)
4058 memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
4059 xh->xh_free_start = cpu_to_le16(end);
4061 /* sort the entries by their name_hash. */
4062 sort(entries, le16_to_cpu(xh->xh_count),
4063 sizeof(struct ocfs2_xattr_entry),
4067 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4068 memcpy(bucket_block(bucket, i), buf, blocksize);
4069 ocfs2_xattr_bucket_journal_dirty(handle, bucket);
4077 * prev_blkno points to the start of an existing extent. new_blkno
4078 * points to a newly allocated extent. Because we know each of our
4079 * clusters contains more than bucket, we can easily split one cluster
4080 * at a bucket boundary. So we take the last cluster of the existing
4081 * extent and split it down the middle. We move the last half of the
4082 * buckets in the last cluster of the existing extent over to the new
4085 * first_bh is the buffer at prev_blkno so we can update the existing
4086 * extent's bucket count. header_bh is the bucket were we were hoping
4087 * to insert our xattr. If the bucket move places the target in the new
4088 * extent, we'll update first_bh and header_bh after modifying the old
4091 * first_hash will be set as the 1st xe's name_hash in the new extent.
4093 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
4095 struct ocfs2_xattr_bucket *first,
4096 struct ocfs2_xattr_bucket *target,
4102 struct super_block *sb = inode->i_sb;
4103 int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
4104 int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
4105 int to_move = num_buckets / 2;
4107 u64 last_cluster_blkno = bucket_blkno(first) +
4108 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
4110 BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
4111 BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
4113 mlog(0, "move half of xattrs in cluster %llu to %llu\n",
4114 (unsigned long long)last_cluster_blkno, (unsigned long long)new_blkno);
4116 ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
4117 last_cluster_blkno, new_blkno,
4118 to_move, first_hash);
4124 /* This is the first bucket that got moved */
4125 src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
4128 * If the target bucket was part of the moved buckets, we need to
4129 * update first and target.
4131 if (bucket_blkno(target) >= src_blkno) {
4132 /* Find the block for the new target bucket */
4133 src_blkno = new_blkno +
4134 (bucket_blkno(target) - src_blkno);
4136 ocfs2_xattr_bucket_relse(first);
4137 ocfs2_xattr_bucket_relse(target);
4140 * These shouldn't fail - the buffers are in the
4141 * journal from ocfs2_cp_xattr_bucket().
4143 ret = ocfs2_read_xattr_bucket(first, new_blkno);
4148 ret = ocfs2_read_xattr_bucket(target, src_blkno);
4159 * Find the suitable pos when we divide a bucket into 2.
4160 * We have to make sure the xattrs with the same hash value exist
4161 * in the same bucket.
4163 * If this ocfs2_xattr_header covers more than one hash value, find a
4164 * place where the hash value changes. Try to find the most even split.
4165 * The most common case is that all entries have different hash values,
4166 * and the first check we make will find a place to split.
4168 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
4170 struct ocfs2_xattr_entry *entries = xh->xh_entries;
4171 int count = le16_to_cpu(xh->xh_count);
4172 int delta, middle = count / 2;
4175 * We start at the middle. Each step gets farther away in both
4176 * directions. We therefore hit the change in hash value
4177 * nearest to the middle. Note that this loop does not execute for
4180 for (delta = 0; delta < middle; delta++) {
4181 /* Let's check delta earlier than middle */
4182 if (cmp_xe(&entries[middle - delta - 1],
4183 &entries[middle - delta]))
4184 return middle - delta;
4186 /* For even counts, don't walk off the end */
4187 if ((middle + delta + 1) == count)
4190 /* Now try delta past middle */
4191 if (cmp_xe(&entries[middle + delta],
4192 &entries[middle + delta + 1]))
4193 return middle + delta + 1;
4196 /* Every entry had the same hash */
4201 * Move some xattrs in old bucket(blk) to new bucket(new_blk).
4202 * first_hash will record the 1st hash of the new bucket.
4204 * Normally half of the xattrs will be moved. But we have to make
4205 * sure that the xattrs with the same hash value are stored in the
4206 * same bucket. If all the xattrs in this bucket have the same hash
4207 * value, the new bucket will be initialized as an empty one and the
4208 * first_hash will be initialized as (hash_value+1).
4210 static int ocfs2_divide_xattr_bucket(struct inode *inode,
4215 int new_bucket_head)
4218 int count, start, len, name_value_len = 0, xe_len, name_offset = 0;
4219 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4220 struct ocfs2_xattr_header *xh;
4221 struct ocfs2_xattr_entry *xe;
4222 int blocksize = inode->i_sb->s_blocksize;
4224 mlog(0, "move some of xattrs from bucket %llu to %llu\n",
4225 (unsigned long long)blk, (unsigned long long)new_blk);
4227 s_bucket = ocfs2_xattr_bucket_new(inode);
4228 t_bucket = ocfs2_xattr_bucket_new(inode);
4229 if (!s_bucket || !t_bucket) {
4235 ret = ocfs2_read_xattr_bucket(s_bucket, blk);
4241 ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
4242 OCFS2_JOURNAL_ACCESS_WRITE);
4249 * Even if !new_bucket_head, we're overwriting t_bucket. Thus,
4250 * there's no need to read it.
4252 ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
4259 * Hey, if we're overwriting t_bucket, what difference does
4260 * ACCESS_CREATE vs ACCESS_WRITE make? See the comment in the
4261 * same part of ocfs2_cp_xattr_bucket().
4263 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4265 OCFS2_JOURNAL_ACCESS_CREATE :
4266 OCFS2_JOURNAL_ACCESS_WRITE);
4272 xh = bucket_xh(s_bucket);
4273 count = le16_to_cpu(xh->xh_count);
4274 start = ocfs2_xattr_find_divide_pos(xh);
4276 if (start == count) {
4277 xe = &xh->xh_entries[start-1];
4280 * initialized a new empty bucket here.
4281 * The hash value is set as one larger than
4282 * that of the last entry in the previous bucket.
4284 for (i = 0; i < t_bucket->bu_blocks; i++)
4285 memset(bucket_block(t_bucket, i), 0, blocksize);
4287 xh = bucket_xh(t_bucket);
4288 xh->xh_free_start = cpu_to_le16(blocksize);
4289 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
4290 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
4292 goto set_num_buckets;
4295 /* copy the whole bucket to the new first. */
4296 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4298 /* update the new bucket. */
4299 xh = bucket_xh(t_bucket);
4302 * Calculate the total name/value len and xh_free_start for
4303 * the old bucket first.
4305 name_offset = OCFS2_XATTR_BUCKET_SIZE;
4307 for (i = 0; i < start; i++) {
4308 xe = &xh->xh_entries[i];
4309 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
4310 if (ocfs2_xattr_is_local(xe))
4312 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4314 xe_len += OCFS2_XATTR_ROOT_SIZE;
4315 name_value_len += xe_len;
4316 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
4317 name_offset = le16_to_cpu(xe->xe_name_offset);
4321 * Now begin the modification to the new bucket.
4323 * In the new bucket, We just move the xattr entry to the beginning
4324 * and don't touch the name/value. So there will be some holes in the
4325 * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
4328 xe = &xh->xh_entries[start];
4329 len = sizeof(struct ocfs2_xattr_entry) * (count - start);
4330 mlog(0, "mv xattr entry len %d from %d to %d\n", len,
4331 (int)((char *)xe - (char *)xh),
4332 (int)((char *)xh->xh_entries - (char *)xh));
4333 memmove((char *)xh->xh_entries, (char *)xe, len);
4334 xe = &xh->xh_entries[count - start];
4335 len = sizeof(struct ocfs2_xattr_entry) * start;
4336 memset((char *)xe, 0, len);
4338 le16_add_cpu(&xh->xh_count, -start);
4339 le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
4341 /* Calculate xh_free_start for the new bucket. */
4342 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4343 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4344 xe = &xh->xh_entries[i];
4345 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
4346 if (ocfs2_xattr_is_local(xe))
4348 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4350 xe_len += OCFS2_XATTR_ROOT_SIZE;
4351 if (le16_to_cpu(xe->xe_name_offset) <
4352 le16_to_cpu(xh->xh_free_start))
4353 xh->xh_free_start = xe->xe_name_offset;
4357 /* set xh->xh_num_buckets for the new xh. */
4358 if (new_bucket_head)
4359 xh->xh_num_buckets = cpu_to_le16(1);
4361 xh->xh_num_buckets = 0;
4363 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4365 /* store the first_hash of the new bucket. */
4367 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
4370 * Now only update the 1st block of the old bucket. If we
4371 * just added a new empty bucket, there is no need to modify
4377 xh = bucket_xh(s_bucket);
4378 memset(&xh->xh_entries[start], 0,
4379 sizeof(struct ocfs2_xattr_entry) * (count - start));
4380 xh->xh_count = cpu_to_le16(start);
4381 xh->xh_free_start = cpu_to_le16(name_offset);
4382 xh->xh_name_value_len = cpu_to_le16(name_value_len);
4384 ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
4387 ocfs2_xattr_bucket_free(s_bucket);
4388 ocfs2_xattr_bucket_free(t_bucket);
4394 * Copy xattr from one bucket to another bucket.
4396 * The caller must make sure that the journal transaction
4397 * has enough space for journaling.
4399 static int ocfs2_cp_xattr_bucket(struct inode *inode,
4406 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4408 BUG_ON(s_blkno == t_blkno);
4410 mlog(0, "cp bucket %llu to %llu, target is %d\n",
4411 (unsigned long long)s_blkno, (unsigned long long)t_blkno,
4414 s_bucket = ocfs2_xattr_bucket_new(inode);
4415 t_bucket = ocfs2_xattr_bucket_new(inode);
4416 if (!s_bucket || !t_bucket) {
4422 ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
4427 * Even if !t_is_new, we're overwriting t_bucket. Thus,
4428 * there's no need to read it.
4430 ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
4435 * Hey, if we're overwriting t_bucket, what difference does
4436 * ACCESS_CREATE vs ACCESS_WRITE make? Well, if we allocated a new
4437 * cluster to fill, we came here from
4438 * ocfs2_mv_xattr_buckets(), and it is really new -
4439 * ACCESS_CREATE is required. But we also might have moved data
4440 * out of t_bucket before extending back into it.
4441 * ocfs2_add_new_xattr_bucket() can do this - its call to
4442 * ocfs2_add_new_xattr_cluster() may have created a new extent
4443 * and copied out the end of the old extent. Then it re-extends
4444 * the old extent back to create space for new xattrs. That's
4445 * how we get here, and the bucket isn't really new.
4447 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4449 OCFS2_JOURNAL_ACCESS_CREATE :
4450 OCFS2_JOURNAL_ACCESS_WRITE);
4454 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4455 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4458 ocfs2_xattr_bucket_free(t_bucket);
4459 ocfs2_xattr_bucket_free(s_bucket);
4465 * src_blk points to the start of an existing extent. last_blk points to
4466 * last cluster in that extent. to_blk points to a newly allocated
4467 * extent. We copy the buckets from the cluster at last_blk to the new
4468 * extent. If start_bucket is non-zero, we skip that many buckets before
4469 * we start copying. The new extent's xh_num_buckets gets set to the
4470 * number of buckets we copied. The old extent's xh_num_buckets shrinks
4471 * by the same amount.
4473 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4474 u64 src_blk, u64 last_blk, u64 to_blk,
4475 unsigned int start_bucket,
4478 int i, ret, credits;
4479 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4480 int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4481 int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4482 struct ocfs2_xattr_bucket *old_first, *new_first;
4484 mlog(0, "mv xattrs from cluster %llu to %llu\n",
4485 (unsigned long long)last_blk, (unsigned long long)to_blk);
4487 BUG_ON(start_bucket >= num_buckets);
4489 num_buckets -= start_bucket;
4490 last_blk += (start_bucket * blks_per_bucket);
4493 /* The first bucket of the original extent */
4494 old_first = ocfs2_xattr_bucket_new(inode);
4495 /* The first bucket of the new extent */
4496 new_first = ocfs2_xattr_bucket_new(inode);
4497 if (!old_first || !new_first) {
4503 ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4510 * We need to update the first bucket of the old extent and all
4511 * the buckets going to the new extent.
4513 credits = ((num_buckets + 1) * blks_per_bucket) +
4514 handle->h_buffer_credits;
4515 ret = ocfs2_extend_trans(handle, credits);
4521 ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4522 OCFS2_JOURNAL_ACCESS_WRITE);
4528 for (i = 0; i < num_buckets; i++) {
4529 ret = ocfs2_cp_xattr_bucket(inode, handle,
4530 last_blk + (i * blks_per_bucket),
4531 to_blk + (i * blks_per_bucket),
4540 * Get the new bucket ready before we dirty anything
4541 * (This actually shouldn't fail, because we already dirtied
4542 * it once in ocfs2_cp_xattr_bucket()).
4544 ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4549 ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4550 OCFS2_JOURNAL_ACCESS_WRITE);
4556 /* Now update the headers */
4557 le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4558 ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4560 bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4561 ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4564 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4567 ocfs2_xattr_bucket_free(new_first);
4568 ocfs2_xattr_bucket_free(old_first);
4573 * Move some xattrs in this cluster to the new cluster.
4574 * This function should only be called when bucket size == cluster size.
4575 * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4577 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4583 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4584 int ret, credits = 2 * blk_per_bucket + handle->h_buffer_credits;
4586 BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4588 ret = ocfs2_extend_trans(handle, credits);
4594 /* Move half of the xattr in start_blk to the next bucket. */
4595 return ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4596 new_blk, first_hash, 1);
4600 * Move some xattrs from the old cluster to the new one since they are not
4601 * contiguous in ocfs2 xattr tree.
4603 * new_blk starts a new separate cluster, and we will move some xattrs from
4604 * prev_blk to it. v_start will be set as the first name hash value in this
4605 * new cluster so that it can be used as e_cpos during tree insertion and
4606 * don't collide with our original b-tree operations. first_bh and header_bh
4607 * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4608 * to extend the insert bucket.
4610 * The problem is how much xattr should we move to the new one and when should
4611 * we update first_bh and header_bh?
4612 * 1. If cluster size > bucket size, that means the previous cluster has more
4613 * than 1 bucket, so just move half nums of bucket into the new cluster and
4614 * update the first_bh and header_bh if the insert bucket has been moved
4615 * to the new cluster.
4616 * 2. If cluster_size == bucket_size:
4617 * a) If the previous extent rec has more than one cluster and the insert
4618 * place isn't in the last cluster, copy the entire last cluster to the
4619 * new one. This time, we don't need to upate the first_bh and header_bh
4620 * since they will not be moved into the new cluster.
4621 * b) Otherwise, move the bottom half of the xattrs in the last cluster into
4622 * the new one. And we set the extend flag to zero if the insert place is
4623 * moved into the new allocated cluster since no extend is needed.
4625 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
4627 struct ocfs2_xattr_bucket *first,
4628 struct ocfs2_xattr_bucket *target,
4636 mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
4637 (unsigned long long)bucket_blkno(first), prev_clusters,
4638 (unsigned long long)new_blk);
4640 if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
4641 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
4650 /* The start of the last cluster in the first extent */
4651 u64 last_blk = bucket_blkno(first) +
4652 ((prev_clusters - 1) *
4653 ocfs2_clusters_to_blocks(inode->i_sb, 1));
4655 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
4656 ret = ocfs2_mv_xattr_buckets(inode, handle,
4657 bucket_blkno(first),
4658 last_blk, new_blk, 0,
4663 ret = ocfs2_divide_xattr_cluster(inode, handle,
4669 if ((bucket_blkno(target) == last_blk) && extend)
4678 * Add a new cluster for xattr storage.
4680 * If the new cluster is contiguous with the previous one, it will be
4681 * appended to the same extent record, and num_clusters will be updated.
4682 * If not, we will insert a new extent for it and move some xattrs in
4683 * the last cluster into the new allocated one.
4684 * We also need to limit the maximum size of a btree leaf, otherwise we'll
4685 * lose the benefits of hashing because we'll have to search large leaves.
4686 * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
4689 * first_bh is the first block of the previous extent rec and header_bh
4690 * indicates the bucket we will insert the new xattrs. They will be updated
4691 * when the header_bh is moved into the new cluster.
4693 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
4694 struct buffer_head *root_bh,
4695 struct ocfs2_xattr_bucket *first,
4696 struct ocfs2_xattr_bucket *target,
4700 struct ocfs2_xattr_set_ctxt *ctxt)
4703 u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
4704 u32 prev_clusters = *num_clusters;
4705 u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
4707 handle_t *handle = ctxt->handle;
4708 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4709 struct ocfs2_extent_tree et;
4711 mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
4712 "previous xattr blkno = %llu\n",
4713 (unsigned long long)OCFS2_I(inode)->ip_blkno,
4714 prev_cpos, (unsigned long long)bucket_blkno(first));
4716 ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
4718 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
4719 OCFS2_JOURNAL_ACCESS_WRITE);
4725 ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac, 1,
4726 clusters_to_add, &bit_off, &num_bits);
4733 BUG_ON(num_bits > clusters_to_add);
4735 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
4736 mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
4737 num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
4739 if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
4740 (prev_clusters + num_bits) << osb->s_clustersize_bits <=
4741 OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
4743 * If this cluster is contiguous with the old one and
4744 * adding this new cluster, we don't surpass the limit of
4745 * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
4746 * initialized and used like other buckets in the previous
4748 * So add it as a contiguous one. The caller will handle
4751 v_start = prev_cpos + prev_clusters;
4752 *num_clusters = prev_clusters + num_bits;
4753 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
4756 ret = ocfs2_adjust_xattr_cross_cluster(inode,
4770 mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
4771 num_bits, (unsigned long long)block, v_start);
4772 ret = ocfs2_insert_extent(handle, &et, v_start, block,
4773 num_bits, 0, ctxt->meta_ac);
4779 ret = ocfs2_journal_dirty(handle, root_bh);
4788 * We are given an extent. 'first' is the bucket at the very front of
4789 * the extent. The extent has space for an additional bucket past
4790 * bucket_xh(first)->xh_num_buckets. 'target_blkno' is the block number
4791 * of the target bucket. We wish to shift every bucket past the target
4792 * down one, filling in that additional space. When we get back to the
4793 * target, we split the target between itself and the now-empty bucket
4794 * at target+1 (aka, target_blkno + blks_per_bucket).
4796 static int ocfs2_extend_xattr_bucket(struct inode *inode,
4798 struct ocfs2_xattr_bucket *first,
4803 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4804 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4806 u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
4808 mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
4809 "from %llu, len = %u\n", (unsigned long long)target_blk,
4810 (unsigned long long)bucket_blkno(first), num_clusters);
4812 /* The extent must have room for an additional bucket */
4813 BUG_ON(new_bucket >=
4814 (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
4816 /* end_blk points to the last existing bucket */
4817 end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
4820 * end_blk is the start of the last existing bucket.
4821 * Thus, (end_blk - target_blk) covers the target bucket and
4822 * every bucket after it up to, but not including, the last
4823 * existing bucket. Then we add the last existing bucket, the
4824 * new bucket, and the first bucket (3 * blk_per_bucket).
4826 credits = (end_blk - target_blk) + (3 * blk_per_bucket) +
4827 handle->h_buffer_credits;
4828 ret = ocfs2_extend_trans(handle, credits);
4834 ret = ocfs2_xattr_bucket_journal_access(handle, first,
4835 OCFS2_JOURNAL_ACCESS_WRITE);
4841 while (end_blk != target_blk) {
4842 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
4843 end_blk + blk_per_bucket, 0);
4846 end_blk -= blk_per_bucket;
4849 /* Move half of the xattr in target_blkno to the next bucket. */
4850 ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
4851 target_blk + blk_per_bucket, NULL, 0);
4853 le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
4854 ocfs2_xattr_bucket_journal_dirty(handle, first);
4861 * Add new xattr bucket in an extent record and adjust the buckets
4862 * accordingly. xb_bh is the ocfs2_xattr_block, and target is the
4863 * bucket we want to insert into.
4865 * In the easy case, we will move all the buckets after target down by
4866 * one. Half of target's xattrs will be moved to the next bucket.
4868 * If current cluster is full, we'll allocate a new one. This may not
4869 * be contiguous. The underlying calls will make sure that there is
4870 * space for the insert, shifting buckets around if necessary.
4871 * 'target' may be moved by those calls.
4873 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
4874 struct buffer_head *xb_bh,
4875 struct ocfs2_xattr_bucket *target,
4876 struct ocfs2_xattr_set_ctxt *ctxt)
4878 struct ocfs2_xattr_block *xb =
4879 (struct ocfs2_xattr_block *)xb_bh->b_data;
4880 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
4881 struct ocfs2_extent_list *el = &xb_root->xt_list;
4883 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
4884 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4885 int ret, num_buckets, extend = 1;
4887 u32 e_cpos, num_clusters;
4888 /* The bucket at the front of the extent */
4889 struct ocfs2_xattr_bucket *first;
4891 mlog(0, "Add new xattr bucket starting from %llu\n",
4892 (unsigned long long)bucket_blkno(target));
4894 /* The first bucket of the original extent */
4895 first = ocfs2_xattr_bucket_new(inode);
4902 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
4909 ret = ocfs2_read_xattr_bucket(first, p_blkno);
4915 num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
4916 if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
4918 * This can move first+target if the target bucket moves
4919 * to the new extent.
4921 ret = ocfs2_add_new_xattr_cluster(inode,
4936 ret = ocfs2_extend_xattr_bucket(inode,
4939 bucket_blkno(target),
4946 ocfs2_xattr_bucket_free(first);
4951 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
4952 struct ocfs2_xattr_bucket *bucket,
4955 int block_off = offs >> inode->i_sb->s_blocksize_bits;
4957 offs = offs % inode->i_sb->s_blocksize;
4958 return bucket_block(bucket, block_off) + offs;
4962 * Handle the normal xattr set, including replace, delete and new.
4964 * Note: "local" indicates the real data's locality. So we can't
4965 * just its bucket locality by its length.
4967 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
4968 struct ocfs2_xattr_info *xi,
4969 struct ocfs2_xattr_search *xs,
4973 struct ocfs2_xattr_entry *last, *xe;
4974 int name_len = strlen(xi->name);
4975 struct ocfs2_xattr_header *xh = xs->header;
4976 u16 count = le16_to_cpu(xh->xh_count), start;
4977 size_t blocksize = inode->i_sb->s_blocksize;
4979 size_t offs, size, new_size;
4980 struct ocfs2_xa_loc loc;
4982 ocfs2_init_xattr_bucket_xa_loc(&loc, xs->bucket,
4983 xs->not_found ? NULL : xs->here);
4984 last = &xh->xh_entries[count];
4985 if (!xs->not_found) {
4987 offs = le16_to_cpu(xe->xe_name_offset);
4988 if (ocfs2_xattr_is_local(xe))
4989 size = OCFS2_XATTR_SIZE(name_len) +
4990 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4992 size = OCFS2_XATTR_SIZE(name_len) +
4993 OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4996 * If the new value will be stored outside, xi->value has been
4997 * initalized as an empty ocfs2_xattr_value_root, and the same
4998 * goes with xi->value_len, so we can set new_size safely here.
4999 * See ocfs2_xattr_set_in_bucket.
5001 new_size = OCFS2_XATTR_SIZE(name_len) +
5002 OCFS2_XATTR_SIZE(xi->value_len);
5004 ocfs2_xa_wipe_namevalue(&loc);
5006 if (new_size > size)
5007 goto set_new_name_value;
5009 /* Now replace the old value with new one. */
5011 xe->xe_value_size = cpu_to_le64(xi->value_len);
5013 xe->xe_value_size = 0;
5015 val = ocfs2_xattr_bucket_get_val(inode,
5017 memset(val + OCFS2_XATTR_SIZE(name_len), 0,
5018 size - OCFS2_XATTR_SIZE(name_len));
5019 if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
5020 memcpy(val + OCFS2_XATTR_SIZE(name_len),
5021 xi->value, xi->value_len);
5023 le16_add_cpu(&xh->xh_name_value_len, new_size);
5024 ocfs2_xattr_set_local(xe, local);
5028 * Remove the old entry if there is more than one.
5029 * We don't remove the last entry so that we can
5030 * use it to indicate the hash value of the empty
5034 le16_add_cpu(&xh->xh_count, -1);
5037 (void *)last - (void *)xe);
5039 sizeof(struct ocfs2_xattr_entry));
5042 cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
5047 /* find a new entry for insert. */
5048 int low = 0, high = count - 1, tmp;
5049 struct ocfs2_xattr_entry *tmp_xe;
5051 while (low <= high && count) {
5052 tmp = (low + high) / 2;
5053 tmp_xe = &xh->xh_entries[tmp];
5055 if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
5057 else if (name_hash <
5058 le32_to_cpu(tmp_xe->xe_name_hash))
5066 xe = &xh->xh_entries[low];
5068 memmove(xe + 1, xe, (void *)last - (void *)xe);
5070 le16_add_cpu(&xh->xh_count, 1);
5071 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
5072 xe->xe_name_hash = cpu_to_le32(name_hash);
5073 xe->xe_name_len = name_len;
5074 ocfs2_xattr_set_type(xe, xi->name_index);
5078 /* Insert the new name+value. */
5079 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
5082 * We must make sure that the name/value pair
5083 * exists in the same block.
5085 offs = le16_to_cpu(xh->xh_free_start);
5086 start = offs - size;
5088 if (start >> inode->i_sb->s_blocksize_bits !=
5089 (offs - 1) >> inode->i_sb->s_blocksize_bits) {
5090 offs = offs - offs % blocksize;
5091 xh->xh_free_start = cpu_to_le16(offs);
5094 val = ocfs2_xattr_bucket_get_val(inode, xs->bucket, offs - size);
5095 xe->xe_name_offset = cpu_to_le16(offs - size);
5097 memset(val, 0, size);
5098 memcpy(val, xi->name, name_len);
5099 memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
5101 xe->xe_value_size = cpu_to_le64(xi->value_len);
5102 ocfs2_xattr_set_local(xe, local);
5104 le16_add_cpu(&xh->xh_free_start, -size);
5105 le16_add_cpu(&xh->xh_name_value_len, size);
5111 * Set the xattr entry in the specified bucket.
5112 * The bucket is indicated by xs->bucket and it should have the enough
5113 * space for the xattr insertion.
5115 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
5117 struct ocfs2_xattr_info *xi,
5118 struct ocfs2_xattr_search *xs,
5125 mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
5126 (unsigned long)xi->value_len, xi->name_index,
5127 (unsigned long long)bucket_blkno(xs->bucket));
5129 if (!xs->bucket->bu_bhs[1]) {
5130 blkno = bucket_blkno(xs->bucket);
5131 ocfs2_xattr_bucket_relse(xs->bucket);
5132 ret = ocfs2_read_xattr_bucket(xs->bucket, blkno);
5139 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
5140 OCFS2_JOURNAL_ACCESS_WRITE);
5146 ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local);
5147 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
5154 * Truncate the specified xe_off entry in xattr bucket.
5155 * bucket is indicated by header_bh and len is the new length.
5156 * Both the ocfs2_xattr_value_root and the entry will be updated here.
5158 * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
5160 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
5161 struct ocfs2_xattr_bucket *bucket,
5164 struct ocfs2_xattr_set_ctxt *ctxt)
5168 struct ocfs2_xattr_entry *xe;
5169 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5170 size_t blocksize = inode->i_sb->s_blocksize;
5171 struct ocfs2_xattr_value_buf vb = {
5172 .vb_access = ocfs2_journal_access,
5175 xe = &xh->xh_entries[xe_off];
5177 BUG_ON(!xe || ocfs2_xattr_is_local(xe));
5179 offset = le16_to_cpu(xe->xe_name_offset) +
5180 OCFS2_XATTR_SIZE(xe->xe_name_len);
5182 value_blk = offset / blocksize;
5184 /* We don't allow ocfs2_xattr_value to be stored in different block. */
5185 BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
5187 vb.vb_bh = bucket->bu_bhs[value_blk];
5190 vb.vb_xv = (struct ocfs2_xattr_value_root *)
5191 (vb.vb_bh->b_data + offset % blocksize);
5194 * From here on out we have to dirty the bucket. The generic
5195 * value calls only modify one of the bucket's bhs, but we need
5196 * to send the bucket at once. So if they error, they *could* have
5197 * modified something. We have to assume they did, and dirty
5198 * the whole bucket. This leaves us in a consistent state.
5200 mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
5201 xe_off, (unsigned long long)bucket_blkno(bucket), len);
5202 ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
5208 ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
5209 OCFS2_JOURNAL_ACCESS_WRITE);
5215 xe->xe_value_size = cpu_to_le64(len);
5217 ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
5223 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
5224 struct ocfs2_xattr_search *xs,
5226 struct ocfs2_xattr_set_ctxt *ctxt)
5229 struct ocfs2_xattr_entry *xe = xs->here;
5230 struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
5232 BUG_ON(!xs->bucket->bu_bhs[0] || !xe || ocfs2_xattr_is_local(xe));
5234 offset = xe - xh->xh_entries;
5235 ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket,
5243 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
5245 struct ocfs2_xattr_search *xs,
5249 int ret, offset, block_off;
5250 struct ocfs2_xattr_value_root *xv;
5251 struct ocfs2_xattr_entry *xe = xs->here;
5252 struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
5254 struct ocfs2_xattr_value_buf vb = {
5255 .vb_access = ocfs2_journal_access,
5258 BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
5260 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb, xh,
5261 xe - xh->xh_entries,
5269 base = bucket_block(xs->bucket, block_off);
5270 xv = (struct ocfs2_xattr_value_root *)(base + offset +
5271 OCFS2_XATTR_SIZE(xe->xe_name_len));
5274 vb.vb_bh = xs->bucket->bu_bhs[block_off];
5275 ret = __ocfs2_xattr_set_value_outside(inode, handle,
5276 &vb, val, value_len);
5283 static int ocfs2_rm_xattr_cluster(struct inode *inode,
5284 struct buffer_head *root_bh,
5291 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5292 struct inode *tl_inode = osb->osb_tl_inode;
5294 struct ocfs2_xattr_block *xb =
5295 (struct ocfs2_xattr_block *)root_bh->b_data;
5296 struct ocfs2_alloc_context *meta_ac = NULL;
5297 struct ocfs2_cached_dealloc_ctxt dealloc;
5298 struct ocfs2_extent_tree et;
5300 ret = ocfs2_iterate_xattr_buckets(inode, blkno, len,
5301 ocfs2_delete_xattr_in_bucket, para);
5307 ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5309 ocfs2_init_dealloc_ctxt(&dealloc);
5311 mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
5312 cpos, len, (unsigned long long)blkno);
5314 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode), blkno,
5317 ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
5323 mutex_lock(&tl_inode->i_mutex);
5325 if (ocfs2_truncate_log_needs_flush(osb)) {
5326 ret = __ocfs2_flush_truncate_log(osb);
5333 handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
5334 if (IS_ERR(handle)) {
5340 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5341 OCFS2_JOURNAL_ACCESS_WRITE);
5347 ret = ocfs2_remove_extent(handle, &et, cpos, len, meta_ac,
5354 le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
5356 ret = ocfs2_journal_dirty(handle, root_bh);
5362 ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
5367 ocfs2_commit_trans(osb, handle);
5369 ocfs2_schedule_truncate_log_flush(osb, 1);
5371 mutex_unlock(&tl_inode->i_mutex);
5374 ocfs2_free_alloc_context(meta_ac);
5376 ocfs2_run_deallocs(osb, &dealloc);
5381 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
5383 struct ocfs2_xattr_search *xs)
5385 struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
5386 struct ocfs2_xattr_entry *last = &xh->xh_entries[
5387 le16_to_cpu(xh->xh_count) - 1];
5390 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
5391 OCFS2_JOURNAL_ACCESS_WRITE);
5397 /* Remove the old entry. */
5398 memmove(xs->here, xs->here + 1,
5399 (void *)last - (void *)xs->here);
5400 memset(last, 0, sizeof(struct ocfs2_xattr_entry));
5401 le16_add_cpu(&xh->xh_count, -1);
5403 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
5407 * Set the xattr name/value in the bucket specified in xs.
5409 * As the new value in xi may be stored in the bucket or in an outside cluster,
5410 * we divide the whole process into 3 steps:
5411 * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
5412 * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
5413 * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
5414 * 4. If the clusters for the new outside value can't be allocated, we need
5415 * to free the xattr we allocated in set.
5417 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
5418 struct ocfs2_xattr_info *xi,
5419 struct ocfs2_xattr_search *xs,
5420 struct ocfs2_xattr_set_ctxt *ctxt)
5424 char *val = (char *)xi->value;
5425 struct ocfs2_xattr_entry *xe = xs->here;
5426 u32 name_hash = ocfs2_xattr_name_hash(inode, xi->name,
5429 if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
5431 * We need to truncate the xattr storage first.
5433 * If both the old and new value are stored to
5434 * outside block, we only need to truncate
5435 * the storage and then set the value outside.
5437 * If the new value should be stored within block,
5438 * we should free all the outside block first and
5439 * the modification to the xattr block will be done
5440 * by following steps.
5442 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
5443 value_len = xi->value_len;
5447 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
5454 goto set_value_outside;
5457 value_len = xi->value_len;
5458 /* So we have to handle the inside block change now. */
5459 if (value_len > OCFS2_XATTR_INLINE_SIZE) {
5461 * If the new value will be stored outside of block,
5462 * initalize a new empty value root and insert it first.
5465 xi->value = &def_xv;
5466 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
5469 ret = ocfs2_xattr_set_entry_in_bucket(inode, ctxt->handle, xi, xs,
5476 if (value_len <= OCFS2_XATTR_INLINE_SIZE)
5479 /* allocate the space now for the outside block storage. */
5480 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
5485 if (xs->not_found) {
5487 * We can't allocate enough clusters for outside
5488 * storage and we have allocated xattr already,
5489 * so need to remove it.
5491 ocfs2_xattr_bucket_remove_xs(inode, ctxt->handle, xs);
5497 ret = ocfs2_xattr_bucket_set_value_outside(inode, ctxt->handle,
5498 xs, val, value_len);
5504 * check whether the xattr bucket is filled up with the same hash value.
5505 * If we want to insert the xattr with the same hash, return -ENOSPC.
5506 * If we want to insert a xattr with different hash value, go ahead
5507 * and ocfs2_divide_xattr_bucket will handle this.
5509 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
5510 struct ocfs2_xattr_bucket *bucket,
5513 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5514 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5516 if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5519 if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5520 xh->xh_entries[0].xe_name_hash) {
5521 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5523 (unsigned long long)bucket_blkno(bucket),
5524 le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5531 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5532 struct ocfs2_xattr_info *xi,
5533 struct ocfs2_xattr_search *xs,
5534 struct ocfs2_xattr_set_ctxt *ctxt)
5536 struct ocfs2_xattr_header *xh;
5537 struct ocfs2_xattr_entry *xe;
5538 u16 count, header_size, xh_free_start;
5539 int free, max_free, need, old;
5540 size_t value_size = 0, name_len = strlen(xi->name);
5541 size_t blocksize = inode->i_sb->s_blocksize;
5542 int ret, allocation = 0;
5544 mlog_entry("Set xattr %s in xattr index block\n", xi->name);
5548 count = le16_to_cpu(xh->xh_count);
5549 xh_free_start = le16_to_cpu(xh->xh_free_start);
5550 header_size = sizeof(struct ocfs2_xattr_header) +
5551 count * sizeof(struct ocfs2_xattr_entry);
5552 max_free = OCFS2_XATTR_BUCKET_SIZE - header_size -
5553 le16_to_cpu(xh->xh_name_value_len) - OCFS2_XATTR_HEADER_GAP;
5555 mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
5556 "of %u which exceed block size\n",
5557 (unsigned long long)bucket_blkno(xs->bucket),
5560 if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
5561 value_size = OCFS2_XATTR_ROOT_SIZE;
5563 value_size = OCFS2_XATTR_SIZE(xi->value_len);
5566 need = sizeof(struct ocfs2_xattr_entry) +
5567 OCFS2_XATTR_SIZE(name_len) + value_size;
5569 need = value_size + OCFS2_XATTR_SIZE(name_len);
5572 * We only replace the old value if the new length is smaller
5573 * than the old one. Otherwise we will allocate new space in the
5574 * bucket to store it.
5577 if (ocfs2_xattr_is_local(xe))
5578 old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
5580 old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
5582 if (old >= value_size)
5586 free = xh_free_start - header_size - OCFS2_XATTR_HEADER_GAP;
5588 * We need to make sure the new name/value pair
5589 * can exist in the same block.
5591 if (xh_free_start % blocksize < need)
5592 free -= xh_free_start % blocksize;
5594 mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
5595 "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
5596 " %u\n", xs->not_found,
5597 (unsigned long long)bucket_blkno(xs->bucket),
5598 free, need, max_free, le16_to_cpu(xh->xh_free_start),
5599 le16_to_cpu(xh->xh_name_value_len));
5603 count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb))) {
5604 if (need <= max_free &&
5605 count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
5607 * We can create the space by defragment. Since only the
5608 * name/value will be moved, the xe shouldn't be changed
5611 ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5618 xh_free_start = le16_to_cpu(xh->xh_free_start);
5619 free = xh_free_start - header_size
5620 - OCFS2_XATTR_HEADER_GAP;
5621 if (xh_free_start % blocksize < need)
5622 free -= xh_free_start % blocksize;
5627 mlog(0, "Can't get enough space for xattr insert by "
5628 "defragment. Need %u bytes, but we have %d, so "
5629 "allocate new bucket for it.\n", need, free);
5633 * We have to add new buckets or clusters and one
5634 * allocation should leave us enough space for insert.
5639 * We do not allow for overlapping ranges between buckets. And
5640 * the maximum number of collisions we will allow for then is
5641 * one bucket's worth, so check it here whether we need to
5642 * add a new bucket for the insert.
5644 ret = ocfs2_check_xattr_bucket_collision(inode,
5652 ret = ocfs2_add_new_xattr_bucket(inode,
5662 * ocfs2_add_new_xattr_bucket() will have updated
5663 * xs->bucket if it moved, but it will not have updated
5664 * any of the other search fields. Thus, we drop it and
5665 * re-search. Everything should be cached, so it'll be
5668 ocfs2_xattr_bucket_relse(xs->bucket);
5669 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5672 if (ret && ret != -ENODATA)
5674 xs->not_found = ret;
5680 ret = ocfs2_xattr_set_in_bucket(inode, xi, xs, ctxt);
5686 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5687 struct ocfs2_xattr_bucket *bucket,
5690 int ret = 0, ref_credits;
5691 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5693 struct ocfs2_xattr_entry *xe;
5694 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5695 struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5696 int credits = ocfs2_remove_extent_credits(osb->sb) +
5697 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5698 struct ocfs2_xattr_value_root *xv;
5699 struct ocfs2_rm_xattr_bucket_para *args =
5700 (struct ocfs2_rm_xattr_bucket_para *)para;
5702 ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5704 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5705 xe = &xh->xh_entries[i];
5706 if (ocfs2_xattr_is_local(xe))
5709 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket,
5712 ret = ocfs2_lock_xattr_remove_allocators(inode, xv,
5718 ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
5719 if (IS_ERR(ctxt.handle)) {
5720 ret = PTR_ERR(ctxt.handle);
5725 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5728 ocfs2_commit_trans(osb, ctxt.handle);
5730 ocfs2_free_alloc_context(ctxt.meta_ac);
5731 ctxt.meta_ac = NULL;
5740 ocfs2_free_alloc_context(ctxt.meta_ac);
5741 ocfs2_schedule_truncate_log_flush(osb, 1);
5742 ocfs2_run_deallocs(osb, &ctxt.dealloc);
5747 * Whenever we modify a xattr value root in the bucket(e.g, CoW
5748 * or change the extent record flag), we need to recalculate
5749 * the metaecc for the whole bucket. So it is done here.
5752 * We have to give the extra credits for the caller.
5754 static int ocfs2_xattr_bucket_post_refcount(struct inode *inode,
5759 struct ocfs2_xattr_bucket *bucket =
5760 (struct ocfs2_xattr_bucket *)para;
5762 ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
5763 OCFS2_JOURNAL_ACCESS_WRITE);
5769 ocfs2_xattr_bucket_journal_dirty(handle, bucket);
5775 * Special action we need if the xattr value is refcounted.
5777 * 1. If the xattr is refcounted, lock the tree.
5778 * 2. CoW the xattr if we are setting the new value and the value
5779 * will be stored outside.
5780 * 3. In other case, decrease_refcount will work for us, so just
5781 * lock the refcount tree, calculate the meta and credits is OK.
5783 * We have to do CoW before ocfs2_init_xattr_set_ctxt since
5784 * currently CoW is a completed transaction, while this function
5785 * will also lock the allocators and let us deadlock. So we will
5786 * CoW the whole xattr value.
5788 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
5789 struct ocfs2_dinode *di,
5790 struct ocfs2_xattr_info *xi,
5791 struct ocfs2_xattr_search *xis,
5792 struct ocfs2_xattr_search *xbs,
5793 struct ocfs2_refcount_tree **ref_tree,
5798 struct ocfs2_xattr_block *xb;
5799 struct ocfs2_xattr_entry *xe;
5801 u32 p_cluster, num_clusters;
5802 unsigned int ext_flags;
5803 int name_offset, name_len;
5804 struct ocfs2_xattr_value_buf vb;
5805 struct ocfs2_xattr_bucket *bucket = NULL;
5806 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5807 struct ocfs2_post_refcount refcount;
5808 struct ocfs2_post_refcount *p = NULL;
5809 struct buffer_head *ref_root_bh = NULL;
5811 if (!xis->not_found) {
5813 name_offset = le16_to_cpu(xe->xe_name_offset);
5814 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5816 vb.vb_bh = xis->inode_bh;
5817 vb.vb_access = ocfs2_journal_access_di;
5819 int i, block_off = 0;
5820 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
5822 name_offset = le16_to_cpu(xe->xe_name_offset);
5823 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5824 i = xbs->here - xbs->header->xh_entries;
5826 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
5827 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
5828 bucket_xh(xbs->bucket),
5835 base = bucket_block(xbs->bucket, block_off);
5836 vb.vb_bh = xbs->bucket->bu_bhs[block_off];
5837 vb.vb_access = ocfs2_journal_access;
5839 if (ocfs2_meta_ecc(osb)) {
5840 /*create parameters for ocfs2_post_refcount. */
5841 bucket = xbs->bucket;
5842 refcount.credits = bucket->bu_blocks;
5843 refcount.para = bucket;
5845 ocfs2_xattr_bucket_post_refcount;
5850 vb.vb_bh = xbs->xattr_bh;
5851 vb.vb_access = ocfs2_journal_access_xb;
5855 if (ocfs2_xattr_is_local(xe))
5858 vb.vb_xv = (struct ocfs2_xattr_value_root *)
5859 (base + name_offset + name_len);
5861 ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
5862 &num_clusters, &vb.vb_xv->xr_list,
5870 * We just need to check the 1st extent record, since we always
5871 * CoW the whole xattr. So there shouldn't be a xattr with
5872 * some REFCOUNT extent recs after the 1st one.
5874 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
5877 ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
5878 1, ref_tree, &ref_root_bh);
5885 * If we are deleting the xattr or the new size will be stored inside,
5886 * cool, leave it there, the xattr truncate process will remove them
5887 * for us(it still needs the refcount tree lock and the meta, credits).
5888 * And the worse case is that every cluster truncate will split the
5889 * refcount tree, and make the original extent become 3. So we will need
5890 * 2 * cluster more extent recs at most.
5892 if (!xi->value || xi->value_len <= OCFS2_XATTR_INLINE_SIZE) {
5894 ret = ocfs2_refcounted_xattr_delete_need(inode,
5895 &(*ref_tree)->rf_ci,
5896 ref_root_bh, vb.vb_xv,
5903 ret = ocfs2_refcount_cow_xattr(inode, di, &vb,
5904 *ref_tree, ref_root_bh, 0,
5905 le32_to_cpu(vb.vb_xv->xr_clusters), p);
5910 brelse(ref_root_bh);
5915 * Add the REFCOUNTED flags for all the extent rec in ocfs2_xattr_value_root.
5916 * The physical clusters will be added to refcount tree.
5918 static int ocfs2_xattr_value_attach_refcount(struct inode *inode,
5919 struct ocfs2_xattr_value_root *xv,
5920 struct ocfs2_extent_tree *value_et,
5921 struct ocfs2_caching_info *ref_ci,
5922 struct buffer_head *ref_root_bh,
5923 struct ocfs2_cached_dealloc_ctxt *dealloc,
5924 struct ocfs2_post_refcount *refcount)
5927 u32 clusters = le32_to_cpu(xv->xr_clusters);
5928 u32 cpos, p_cluster, num_clusters;
5929 struct ocfs2_extent_list *el = &xv->xr_list;
5930 unsigned int ext_flags;
5933 while (cpos < clusters) {
5934 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
5935 &num_clusters, el, &ext_flags);
5937 cpos += num_clusters;
5938 if ((ext_flags & OCFS2_EXT_REFCOUNTED))
5943 ret = ocfs2_add_refcount_flag(inode, value_et,
5944 ref_ci, ref_root_bh,
5945 cpos - num_clusters,
5946 p_cluster, num_clusters,
5958 * Given a normal ocfs2_xattr_header, refcount all the entries which
5959 * have value stored outside.
5960 * Used for xattrs stored in inode and ocfs2_xattr_block.
5962 static int ocfs2_xattr_attach_refcount_normal(struct inode *inode,
5963 struct ocfs2_xattr_value_buf *vb,
5964 struct ocfs2_xattr_header *header,
5965 struct ocfs2_caching_info *ref_ci,
5966 struct buffer_head *ref_root_bh,
5967 struct ocfs2_cached_dealloc_ctxt *dealloc)
5970 struct ocfs2_xattr_entry *xe;
5971 struct ocfs2_xattr_value_root *xv;
5972 struct ocfs2_extent_tree et;
5975 for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
5976 xe = &header->xh_entries[i];
5978 if (ocfs2_xattr_is_local(xe))
5981 xv = (struct ocfs2_xattr_value_root *)((void *)header +
5982 le16_to_cpu(xe->xe_name_offset) +
5983 OCFS2_XATTR_SIZE(xe->xe_name_len));
5986 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
5988 ret = ocfs2_xattr_value_attach_refcount(inode, xv, &et,
5989 ref_ci, ref_root_bh,
6000 static int ocfs2_xattr_inline_attach_refcount(struct inode *inode,
6001 struct buffer_head *fe_bh,
6002 struct ocfs2_caching_info *ref_ci,
6003 struct buffer_head *ref_root_bh,
6004 struct ocfs2_cached_dealloc_ctxt *dealloc)
6006 struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
6007 struct ocfs2_xattr_header *header = (struct ocfs2_xattr_header *)
6008 (fe_bh->b_data + inode->i_sb->s_blocksize -
6009 le16_to_cpu(di->i_xattr_inline_size));
6010 struct ocfs2_xattr_value_buf vb = {
6012 .vb_access = ocfs2_journal_access_di,
6015 return ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
6016 ref_ci, ref_root_bh, dealloc);
6019 struct ocfs2_xattr_tree_value_refcount_para {
6020 struct ocfs2_caching_info *ref_ci;
6021 struct buffer_head *ref_root_bh;
6022 struct ocfs2_cached_dealloc_ctxt *dealloc;
6025 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
6026 struct ocfs2_xattr_bucket *bucket,
6028 struct ocfs2_xattr_value_root **xv,
6029 struct buffer_head **bh)
6031 int ret, block_off, name_offset;
6032 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
6033 struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
6036 ret = ocfs2_xattr_bucket_get_name_value(sb,
6046 base = bucket_block(bucket, block_off);
6048 *xv = (struct ocfs2_xattr_value_root *)(base + name_offset +
6049 OCFS2_XATTR_SIZE(xe->xe_name_len));
6052 *bh = bucket->bu_bhs[block_off];
6058 * For a given xattr bucket, refcount all the entries which
6059 * have value stored outside.
6061 static int ocfs2_xattr_bucket_value_refcount(struct inode *inode,
6062 struct ocfs2_xattr_bucket *bucket,
6066 struct ocfs2_extent_tree et;
6067 struct ocfs2_xattr_tree_value_refcount_para *ref =
6068 (struct ocfs2_xattr_tree_value_refcount_para *)para;
6069 struct ocfs2_xattr_header *xh =
6070 (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6071 struct ocfs2_xattr_entry *xe;
6072 struct ocfs2_xattr_value_buf vb = {
6073 .vb_access = ocfs2_journal_access,
6075 struct ocfs2_post_refcount refcount = {
6076 .credits = bucket->bu_blocks,
6078 .func = ocfs2_xattr_bucket_post_refcount,
6080 struct ocfs2_post_refcount *p = NULL;
6082 /* We only need post_refcount if we support metaecc. */
6083 if (ocfs2_meta_ecc(OCFS2_SB(inode->i_sb)))
6086 mlog(0, "refcount bucket %llu, count = %u\n",
6087 (unsigned long long)bucket_blkno(bucket),
6088 le16_to_cpu(xh->xh_count));
6089 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6090 xe = &xh->xh_entries[i];
6092 if (ocfs2_xattr_is_local(xe))
6095 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket, i,
6096 &vb.vb_xv, &vb.vb_bh);
6102 ocfs2_init_xattr_value_extent_tree(&et,
6103 INODE_CACHE(inode), &vb);
6105 ret = ocfs2_xattr_value_attach_refcount(inode, vb.vb_xv,
6119 static int ocfs2_refcount_xattr_tree_rec(struct inode *inode,
6120 struct buffer_head *root_bh,
6121 u64 blkno, u32 cpos, u32 len, void *para)
6123 return ocfs2_iterate_xattr_buckets(inode, blkno, len,
6124 ocfs2_xattr_bucket_value_refcount,
6128 static int ocfs2_xattr_block_attach_refcount(struct inode *inode,
6129 struct buffer_head *blk_bh,
6130 struct ocfs2_caching_info *ref_ci,
6131 struct buffer_head *ref_root_bh,
6132 struct ocfs2_cached_dealloc_ctxt *dealloc)
6135 struct ocfs2_xattr_block *xb =
6136 (struct ocfs2_xattr_block *)blk_bh->b_data;
6138 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
6139 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
6140 struct ocfs2_xattr_value_buf vb = {
6142 .vb_access = ocfs2_journal_access_xb,
6145 ret = ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
6146 ref_ci, ref_root_bh,
6149 struct ocfs2_xattr_tree_value_refcount_para para = {
6151 .ref_root_bh = ref_root_bh,
6155 ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
6156 ocfs2_refcount_xattr_tree_rec,
6163 int ocfs2_xattr_attach_refcount_tree(struct inode *inode,
6164 struct buffer_head *fe_bh,
6165 struct ocfs2_caching_info *ref_ci,
6166 struct buffer_head *ref_root_bh,
6167 struct ocfs2_cached_dealloc_ctxt *dealloc)
6170 struct ocfs2_inode_info *oi = OCFS2_I(inode);
6171 struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
6172 struct buffer_head *blk_bh = NULL;
6174 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
6175 ret = ocfs2_xattr_inline_attach_refcount(inode, fe_bh,
6176 ref_ci, ref_root_bh,
6184 if (!di->i_xattr_loc)
6187 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
6194 ret = ocfs2_xattr_block_attach_refcount(inode, blk_bh, ref_ci,
6195 ref_root_bh, dealloc);
6205 typedef int (should_xattr_reflinked)(struct ocfs2_xattr_entry *xe);
6207 * Store the information we need in xattr reflink.
6208 * old_bh and new_bh are inode bh for the old and new inode.
6210 struct ocfs2_xattr_reflink {
6211 struct inode *old_inode;
6212 struct inode *new_inode;
6213 struct buffer_head *old_bh;
6214 struct buffer_head *new_bh;
6215 struct ocfs2_caching_info *ref_ci;
6216 struct buffer_head *ref_root_bh;
6217 struct ocfs2_cached_dealloc_ctxt *dealloc;
6218 should_xattr_reflinked *xattr_reflinked;
6222 * Given a xattr header and xe offset,
6223 * return the proper xv and the corresponding bh.
6224 * xattr in inode, block and xattr tree have different implementaions.
6226 typedef int (get_xattr_value_root)(struct super_block *sb,
6227 struct buffer_head *bh,
6228 struct ocfs2_xattr_header *xh,
6230 struct ocfs2_xattr_value_root **xv,
6231 struct buffer_head **ret_bh,
6235 * Calculate all the xattr value root metadata stored in this xattr header and
6236 * credits we need if we create them from the scratch.
6237 * We use get_xattr_value_root so that all types of xattr container can use it.
6239 static int ocfs2_value_metas_in_xattr_header(struct super_block *sb,
6240 struct buffer_head *bh,
6241 struct ocfs2_xattr_header *xh,
6242 int *metas, int *credits,
6244 get_xattr_value_root *func,
6248 struct ocfs2_xattr_value_root *xv;
6249 struct ocfs2_xattr_entry *xe;
6251 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6252 xe = &xh->xh_entries[i];
6253 if (ocfs2_xattr_is_local(xe))
6256 ret = func(sb, bh, xh, i, &xv, NULL, para);
6262 *metas += le16_to_cpu(xv->xr_list.l_tree_depth) *
6263 le16_to_cpu(xv->xr_list.l_next_free_rec);
6265 *credits += ocfs2_calc_extend_credits(sb,
6267 le32_to_cpu(xv->xr_clusters));
6270 * If the value is a tree with depth > 1, We don't go deep
6271 * to the extent block, so just calculate a maximum record num.
6273 if (!xv->xr_list.l_tree_depth)
6274 *num_recs += le16_to_cpu(xv->xr_list.l_next_free_rec);
6276 *num_recs += ocfs2_clusters_for_bytes(sb,
6283 /* Used by xattr inode and block to return the right xv and buffer_head. */
6284 static int ocfs2_get_xattr_value_root(struct super_block *sb,
6285 struct buffer_head *bh,
6286 struct ocfs2_xattr_header *xh,
6288 struct ocfs2_xattr_value_root **xv,
6289 struct buffer_head **ret_bh,
6292 struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
6294 *xv = (struct ocfs2_xattr_value_root *)((void *)xh +
6295 le16_to_cpu(xe->xe_name_offset) +
6296 OCFS2_XATTR_SIZE(xe->xe_name_len));
6305 * Lock the meta_ac and caculate how much credits we need for reflink xattrs.
6306 * It is only used for inline xattr and xattr block.
6308 static int ocfs2_reflink_lock_xattr_allocators(struct ocfs2_super *osb,
6309 struct ocfs2_xattr_header *xh,
6310 struct buffer_head *ref_root_bh,
6312 struct ocfs2_alloc_context **meta_ac)
6314 int ret, meta_add = 0, num_recs = 0;
6315 struct ocfs2_refcount_block *rb =
6316 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
6320 ret = ocfs2_value_metas_in_xattr_header(osb->sb, NULL, xh,
6321 &meta_add, credits, &num_recs,
6322 ocfs2_get_xattr_value_root,
6330 * We need to add/modify num_recs in refcount tree, so just calculate
6331 * an approximate number we need for refcount tree change.
6332 * Sometimes we need to split the tree, and after split, half recs
6333 * will be moved to the new block, and a new block can only provide
6334 * half number of recs. So we multiple new blocks by 2.
6336 num_recs = num_recs / ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6337 meta_add += num_recs;
6338 *credits += num_recs + num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6339 if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6340 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6341 le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6345 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add, meta_ac);
6354 * Given a xattr header, reflink all the xattrs in this container.
6355 * It can be used for inode, block and bucket.
6358 * Before we call this function, the caller has memcpy the xattr in
6359 * old_xh to the new_xh.
6361 * If args.xattr_reflinked is set, call it to decide whether the xe should
6362 * be reflinked or not. If not, remove it from the new xattr header.
6364 static int ocfs2_reflink_xattr_header(handle_t *handle,
6365 struct ocfs2_xattr_reflink *args,
6366 struct buffer_head *old_bh,
6367 struct ocfs2_xattr_header *xh,
6368 struct buffer_head *new_bh,
6369 struct ocfs2_xattr_header *new_xh,
6370 struct ocfs2_xattr_value_buf *vb,
6371 struct ocfs2_alloc_context *meta_ac,
6372 get_xattr_value_root *func,
6376 struct super_block *sb = args->old_inode->i_sb;
6377 struct buffer_head *value_bh;
6378 struct ocfs2_xattr_entry *xe, *last;
6379 struct ocfs2_xattr_value_root *xv, *new_xv;
6380 struct ocfs2_extent_tree data_et;
6381 u32 clusters, cpos, p_cluster, num_clusters;
6382 unsigned int ext_flags = 0;
6384 mlog(0, "reflink xattr in container %llu, count = %u\n",
6385 (unsigned long long)old_bh->b_blocknr, le16_to_cpu(xh->xh_count));
6387 last = &new_xh->xh_entries[le16_to_cpu(new_xh->xh_count)];
6388 for (i = 0, j = 0; i < le16_to_cpu(xh->xh_count); i++, j++) {
6389 xe = &xh->xh_entries[i];
6391 if (args->xattr_reflinked && !args->xattr_reflinked(xe)) {
6392 xe = &new_xh->xh_entries[j];
6394 le16_add_cpu(&new_xh->xh_count, -1);
6395 if (new_xh->xh_count) {
6397 (void *)last - (void *)xe);
6399 sizeof(struct ocfs2_xattr_entry));
6403 * We don't want j to increase in the next round since
6404 * it is already moved ahead.
6410 if (ocfs2_xattr_is_local(xe))
6413 ret = func(sb, old_bh, xh, i, &xv, NULL, para);
6419 ret = func(sb, new_bh, new_xh, j, &new_xv, &value_bh, para);
6426 * For the xattr which has l_tree_depth = 0, all the extent
6427 * recs have already be copied to the new xh with the
6428 * propriate OCFS2_EXT_REFCOUNTED flag we just need to
6429 * increase the refount count int the refcount tree.
6431 * For the xattr which has l_tree_depth > 0, we need
6432 * to initialize it to the empty default value root,
6433 * and then insert the extents one by one.
6435 if (xv->xr_list.l_tree_depth) {
6436 memcpy(new_xv, &def_xv, sizeof(def_xv));
6438 vb->vb_bh = value_bh;
6439 ocfs2_init_xattr_value_extent_tree(&data_et,
6440 INODE_CACHE(args->new_inode), vb);
6443 clusters = le32_to_cpu(xv->xr_clusters);
6445 while (cpos < clusters) {
6446 ret = ocfs2_xattr_get_clusters(args->old_inode,
6459 if (xv->xr_list.l_tree_depth) {
6460 ret = ocfs2_insert_extent(handle,
6462 ocfs2_clusters_to_blocks(
6463 args->old_inode->i_sb,
6465 num_clusters, ext_flags,
6473 ret = ocfs2_increase_refcount(handle, args->ref_ci,
6475 p_cluster, num_clusters,
6476 meta_ac, args->dealloc);
6482 cpos += num_clusters;
6490 static int ocfs2_reflink_xattr_inline(struct ocfs2_xattr_reflink *args)
6492 int ret = 0, credits = 0;
6494 struct ocfs2_super *osb = OCFS2_SB(args->old_inode->i_sb);
6495 struct ocfs2_dinode *di = (struct ocfs2_dinode *)args->old_bh->b_data;
6496 int inline_size = le16_to_cpu(di->i_xattr_inline_size);
6497 int header_off = osb->sb->s_blocksize - inline_size;
6498 struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)
6499 (args->old_bh->b_data + header_off);
6500 struct ocfs2_xattr_header *new_xh = (struct ocfs2_xattr_header *)
6501 (args->new_bh->b_data + header_off);
6502 struct ocfs2_alloc_context *meta_ac = NULL;
6503 struct ocfs2_inode_info *new_oi;
6504 struct ocfs2_dinode *new_di;
6505 struct ocfs2_xattr_value_buf vb = {
6506 .vb_bh = args->new_bh,
6507 .vb_access = ocfs2_journal_access_di,
6510 ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6511 &credits, &meta_ac);
6517 handle = ocfs2_start_trans(osb, credits);
6518 if (IS_ERR(handle)) {
6519 ret = PTR_ERR(handle);
6524 ret = ocfs2_journal_access_di(handle, INODE_CACHE(args->new_inode),
6525 args->new_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6531 memcpy(args->new_bh->b_data + header_off,
6532 args->old_bh->b_data + header_off, inline_size);
6534 new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6535 new_di->i_xattr_inline_size = cpu_to_le16(inline_size);
6537 ret = ocfs2_reflink_xattr_header(handle, args, args->old_bh, xh,
6538 args->new_bh, new_xh, &vb, meta_ac,
6539 ocfs2_get_xattr_value_root, NULL);
6545 new_oi = OCFS2_I(args->new_inode);
6546 spin_lock(&new_oi->ip_lock);
6547 new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL | OCFS2_INLINE_XATTR_FL;
6548 new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6549 spin_unlock(&new_oi->ip_lock);
6551 ocfs2_journal_dirty(handle, args->new_bh);
6554 ocfs2_commit_trans(osb, handle);
6558 ocfs2_free_alloc_context(meta_ac);
6562 static int ocfs2_create_empty_xattr_block(struct inode *inode,
6563 struct buffer_head *fe_bh,
6564 struct buffer_head **ret_bh,
6569 struct ocfs2_alloc_context *meta_ac;
6570 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6572 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
6578 handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_CREATE_CREDITS);
6579 if (IS_ERR(handle)) {
6580 ret = PTR_ERR(handle);
6585 mlog(0, "create new xattr block for inode %llu, index = %d\n",
6586 (unsigned long long)fe_bh->b_blocknr, indexed);
6587 ret = ocfs2_create_xattr_block(handle, inode, fe_bh,
6588 meta_ac, ret_bh, indexed);
6592 ocfs2_commit_trans(osb, handle);
6594 ocfs2_free_alloc_context(meta_ac);
6598 static int ocfs2_reflink_xattr_block(struct ocfs2_xattr_reflink *args,
6599 struct buffer_head *blk_bh,
6600 struct buffer_head *new_blk_bh)
6602 int ret = 0, credits = 0;
6604 struct ocfs2_inode_info *new_oi = OCFS2_I(args->new_inode);
6605 struct ocfs2_dinode *new_di;
6606 struct ocfs2_super *osb = OCFS2_SB(args->new_inode->i_sb);
6607 int header_off = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
6608 struct ocfs2_xattr_block *xb =
6609 (struct ocfs2_xattr_block *)blk_bh->b_data;
6610 struct ocfs2_xattr_header *xh = &xb->xb_attrs.xb_header;
6611 struct ocfs2_xattr_block *new_xb =
6612 (struct ocfs2_xattr_block *)new_blk_bh->b_data;
6613 struct ocfs2_xattr_header *new_xh = &new_xb->xb_attrs.xb_header;
6614 struct ocfs2_alloc_context *meta_ac;
6615 struct ocfs2_xattr_value_buf vb = {
6616 .vb_bh = new_blk_bh,
6617 .vb_access = ocfs2_journal_access_xb,
6620 ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6621 &credits, &meta_ac);
6627 /* One more credits in case we need to add xattr flags in new inode. */
6628 handle = ocfs2_start_trans(osb, credits + 1);
6629 if (IS_ERR(handle)) {
6630 ret = PTR_ERR(handle);
6635 if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6636 ret = ocfs2_journal_access_di(handle,
6637 INODE_CACHE(args->new_inode),
6639 OCFS2_JOURNAL_ACCESS_WRITE);
6646 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(args->new_inode),
6647 new_blk_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6653 memcpy(new_blk_bh->b_data + header_off, blk_bh->b_data + header_off,
6654 osb->sb->s_blocksize - header_off);
6656 ret = ocfs2_reflink_xattr_header(handle, args, blk_bh, xh,
6657 new_blk_bh, new_xh, &vb, meta_ac,
6658 ocfs2_get_xattr_value_root, NULL);
6664 ocfs2_journal_dirty(handle, new_blk_bh);
6666 if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6667 new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6668 spin_lock(&new_oi->ip_lock);
6669 new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
6670 new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6671 spin_unlock(&new_oi->ip_lock);
6673 ocfs2_journal_dirty(handle, args->new_bh);
6677 ocfs2_commit_trans(osb, handle);
6680 ocfs2_free_alloc_context(meta_ac);
6684 struct ocfs2_reflink_xattr_tree_args {
6685 struct ocfs2_xattr_reflink *reflink;
6686 struct buffer_head *old_blk_bh;
6687 struct buffer_head *new_blk_bh;
6688 struct ocfs2_xattr_bucket *old_bucket;
6689 struct ocfs2_xattr_bucket *new_bucket;
6694 * We have to handle the case that both old bucket and new bucket
6695 * will call this function to get the right ret_bh.
6696 * So The caller must give us the right bh.
6698 static int ocfs2_get_reflink_xattr_value_root(struct super_block *sb,
6699 struct buffer_head *bh,
6700 struct ocfs2_xattr_header *xh,
6702 struct ocfs2_xattr_value_root **xv,
6703 struct buffer_head **ret_bh,
6706 struct ocfs2_reflink_xattr_tree_args *args =
6707 (struct ocfs2_reflink_xattr_tree_args *)para;
6708 struct ocfs2_xattr_bucket *bucket;
6710 if (bh == args->old_bucket->bu_bhs[0])
6711 bucket = args->old_bucket;
6713 bucket = args->new_bucket;
6715 return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6719 struct ocfs2_value_tree_metas {
6725 static int ocfs2_value_tree_metas_in_bucket(struct super_block *sb,
6726 struct buffer_head *bh,
6727 struct ocfs2_xattr_header *xh,
6729 struct ocfs2_xattr_value_root **xv,
6730 struct buffer_head **ret_bh,
6733 struct ocfs2_xattr_bucket *bucket =
6734 (struct ocfs2_xattr_bucket *)para;
6736 return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6740 static int ocfs2_calc_value_tree_metas(struct inode *inode,
6741 struct ocfs2_xattr_bucket *bucket,
6744 struct ocfs2_value_tree_metas *metas =
6745 (struct ocfs2_value_tree_metas *)para;
6746 struct ocfs2_xattr_header *xh =
6747 (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6749 /* Add the credits for this bucket first. */
6750 metas->credits += bucket->bu_blocks;
6751 return ocfs2_value_metas_in_xattr_header(inode->i_sb, bucket->bu_bhs[0],
6752 xh, &metas->num_metas,
6753 &metas->credits, &metas->num_recs,
6754 ocfs2_value_tree_metas_in_bucket,
6759 * Given a xattr extent rec starting from blkno and having len clusters,
6760 * iterate all the buckets calculate how much metadata we need for reflinking
6761 * all the ocfs2_xattr_value_root and lock the allocators accordingly.
6763 static int ocfs2_lock_reflink_xattr_rec_allocators(
6764 struct ocfs2_reflink_xattr_tree_args *args,
6765 struct ocfs2_extent_tree *xt_et,
6766 u64 blkno, u32 len, int *credits,
6767 struct ocfs2_alloc_context **meta_ac,
6768 struct ocfs2_alloc_context **data_ac)
6770 int ret, num_free_extents;
6771 struct ocfs2_value_tree_metas metas;
6772 struct ocfs2_super *osb = OCFS2_SB(args->reflink->old_inode->i_sb);
6773 struct ocfs2_refcount_block *rb;
6775 memset(&metas, 0, sizeof(metas));
6777 ret = ocfs2_iterate_xattr_buckets(args->reflink->old_inode, blkno, len,
6778 ocfs2_calc_value_tree_metas, &metas);
6784 *credits = metas.credits;
6787 * Calculate we need for refcount tree change.
6789 * We need to add/modify num_recs in refcount tree, so just calculate
6790 * an approximate number we need for refcount tree change.
6791 * Sometimes we need to split the tree, and after split, half recs
6792 * will be moved to the new block, and a new block can only provide
6793 * half number of recs. So we multiple new blocks by 2.
6794 * In the end, we have to add credits for modifying the already
6795 * existed refcount block.
6797 rb = (struct ocfs2_refcount_block *)args->reflink->ref_root_bh->b_data;
6799 (metas.num_recs + ocfs2_refcount_recs_per_rb(osb->sb) - 1) /
6800 ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6801 metas.num_metas += metas.num_recs;
6802 *credits += metas.num_recs +
6803 metas.num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6804 if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6805 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6806 le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6810 /* count in the xattr tree change. */
6811 num_free_extents = ocfs2_num_free_extents(osb, xt_et);
6812 if (num_free_extents < 0) {
6813 ret = num_free_extents;
6818 if (num_free_extents < len)
6819 metas.num_metas += ocfs2_extend_meta_needed(xt_et->et_root_el);
6821 *credits += ocfs2_calc_extend_credits(osb->sb,
6822 xt_et->et_root_el, len);
6824 if (metas.num_metas) {
6825 ret = ocfs2_reserve_new_metadata_blocks(osb, metas.num_metas,
6834 ret = ocfs2_reserve_clusters(osb, len, data_ac);
6841 ocfs2_free_alloc_context(*meta_ac);
6849 static int ocfs2_reflink_xattr_buckets(handle_t *handle,
6850 u64 blkno, u64 new_blkno, u32 clusters,
6851 struct ocfs2_alloc_context *meta_ac,
6852 struct ocfs2_alloc_context *data_ac,
6853 struct ocfs2_reflink_xattr_tree_args *args)
6856 struct super_block *sb = args->reflink->old_inode->i_sb;
6857 u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
6858 u32 num_buckets = clusters * bpc;
6859 int bpb = args->old_bucket->bu_blocks;
6860 struct ocfs2_xattr_value_buf vb = {
6861 .vb_access = ocfs2_journal_access,
6864 for (i = 0; i < num_buckets; i++, blkno += bpb, new_blkno += bpb) {
6865 ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6871 ret = ocfs2_init_xattr_bucket(args->new_bucket, new_blkno);
6878 * The real bucket num in this series of blocks is stored
6879 * in the 1st bucket.
6882 num_buckets = le16_to_cpu(
6883 bucket_xh(args->old_bucket)->xh_num_buckets);
6885 ret = ocfs2_xattr_bucket_journal_access(handle,
6887 OCFS2_JOURNAL_ACCESS_CREATE);
6893 for (j = 0; j < bpb; j++)
6894 memcpy(bucket_block(args->new_bucket, j),
6895 bucket_block(args->old_bucket, j),
6898 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6900 ret = ocfs2_reflink_xattr_header(handle, args->reflink,
6901 args->old_bucket->bu_bhs[0],
6902 bucket_xh(args->old_bucket),
6903 args->new_bucket->bu_bhs[0],
6904 bucket_xh(args->new_bucket),
6906 ocfs2_get_reflink_xattr_value_root,
6914 * Re-access and dirty the bucket to calculate metaecc.
6915 * Because we may extend the transaction in reflink_xattr_header
6916 * which will let the already accessed block gone.
6918 ret = ocfs2_xattr_bucket_journal_access(handle,
6920 OCFS2_JOURNAL_ACCESS_WRITE);
6926 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6927 ocfs2_xattr_bucket_relse(args->old_bucket);
6928 ocfs2_xattr_bucket_relse(args->new_bucket);
6931 ocfs2_xattr_bucket_relse(args->old_bucket);
6932 ocfs2_xattr_bucket_relse(args->new_bucket);
6936 * Create the same xattr extent record in the new inode's xattr tree.
6938 static int ocfs2_reflink_xattr_rec(struct inode *inode,
6939 struct buffer_head *root_bh,
6945 int ret, credits = 0;
6946 u32 p_cluster, num_clusters;
6949 struct ocfs2_reflink_xattr_tree_args *args =
6950 (struct ocfs2_reflink_xattr_tree_args *)para;
6951 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6952 struct ocfs2_alloc_context *meta_ac = NULL;
6953 struct ocfs2_alloc_context *data_ac = NULL;
6954 struct ocfs2_extent_tree et;
6956 ocfs2_init_xattr_tree_extent_tree(&et,
6957 INODE_CACHE(args->reflink->new_inode),
6960 ret = ocfs2_lock_reflink_xattr_rec_allocators(args, &et, blkno,
6962 &meta_ac, &data_ac);
6968 handle = ocfs2_start_trans(osb, credits);
6969 if (IS_ERR(handle)) {
6970 ret = PTR_ERR(handle);
6975 ret = ocfs2_claim_clusters(osb, handle, data_ac,
6976 len, &p_cluster, &num_clusters);
6982 new_blkno = ocfs2_clusters_to_blocks(osb->sb, p_cluster);
6984 mlog(0, "reflink xattr buckets %llu to %llu, len %u\n",
6985 (unsigned long long)blkno, (unsigned long long)new_blkno, len);
6986 ret = ocfs2_reflink_xattr_buckets(handle, blkno, new_blkno, len,
6987 meta_ac, data_ac, args);
6993 mlog(0, "insert new xattr extent rec start %llu len %u to %u\n",
6994 (unsigned long long)new_blkno, len, cpos);
6995 ret = ocfs2_insert_extent(handle, &et, cpos, new_blkno,
7001 ocfs2_commit_trans(osb, handle);
7005 ocfs2_free_alloc_context(meta_ac);
7007 ocfs2_free_alloc_context(data_ac);
7012 * Create reflinked xattr buckets.
7013 * We will add bucket one by one, and refcount all the xattrs in the bucket
7014 * if they are stored outside.
7016 static int ocfs2_reflink_xattr_tree(struct ocfs2_xattr_reflink *args,
7017 struct buffer_head *blk_bh,
7018 struct buffer_head *new_blk_bh)
7021 struct ocfs2_reflink_xattr_tree_args para;
7023 memset(¶, 0, sizeof(para));
7024 para.reflink = args;
7025 para.old_blk_bh = blk_bh;
7026 para.new_blk_bh = new_blk_bh;
7028 para.old_bucket = ocfs2_xattr_bucket_new(args->old_inode);
7029 if (!para.old_bucket) {
7030 mlog_errno(-ENOMEM);
7034 para.new_bucket = ocfs2_xattr_bucket_new(args->new_inode);
7035 if (!para.new_bucket) {
7041 ret = ocfs2_iterate_xattr_index_block(args->old_inode, blk_bh,
7042 ocfs2_reflink_xattr_rec,
7048 ocfs2_xattr_bucket_free(para.old_bucket);
7049 ocfs2_xattr_bucket_free(para.new_bucket);
7053 static int ocfs2_reflink_xattr_in_block(struct ocfs2_xattr_reflink *args,
7054 struct buffer_head *blk_bh)
7056 int ret, indexed = 0;
7057 struct buffer_head *new_blk_bh = NULL;
7058 struct ocfs2_xattr_block *xb =
7059 (struct ocfs2_xattr_block *)blk_bh->b_data;
7062 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)
7065 ret = ocfs2_create_empty_xattr_block(args->new_inode, args->new_bh,
7066 &new_blk_bh, indexed);
7072 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED))
7073 ret = ocfs2_reflink_xattr_block(args, blk_bh, new_blk_bh);
7075 ret = ocfs2_reflink_xattr_tree(args, blk_bh, new_blk_bh);
7084 static int ocfs2_reflink_xattr_no_security(struct ocfs2_xattr_entry *xe)
7086 int type = ocfs2_xattr_get_type(xe);
7088 return type != OCFS2_XATTR_INDEX_SECURITY &&
7089 type != OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS &&
7090 type != OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT;
7093 int ocfs2_reflink_xattrs(struct inode *old_inode,
7094 struct buffer_head *old_bh,
7095 struct inode *new_inode,
7096 struct buffer_head *new_bh,
7097 bool preserve_security)
7100 struct ocfs2_xattr_reflink args;
7101 struct ocfs2_inode_info *oi = OCFS2_I(old_inode);
7102 struct ocfs2_dinode *di = (struct ocfs2_dinode *)old_bh->b_data;
7103 struct buffer_head *blk_bh = NULL;
7104 struct ocfs2_cached_dealloc_ctxt dealloc;
7105 struct ocfs2_refcount_tree *ref_tree;
7106 struct buffer_head *ref_root_bh = NULL;
7108 ret = ocfs2_lock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7109 le64_to_cpu(di->i_refcount_loc),
7110 1, &ref_tree, &ref_root_bh);
7116 ocfs2_init_dealloc_ctxt(&dealloc);
7118 args.old_inode = old_inode;
7119 args.new_inode = new_inode;
7120 args.old_bh = old_bh;
7121 args.new_bh = new_bh;
7122 args.ref_ci = &ref_tree->rf_ci;
7123 args.ref_root_bh = ref_root_bh;
7124 args.dealloc = &dealloc;
7125 if (preserve_security)
7126 args.xattr_reflinked = NULL;
7128 args.xattr_reflinked = ocfs2_reflink_xattr_no_security;
7130 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
7131 ret = ocfs2_reflink_xattr_inline(&args);
7138 if (!di->i_xattr_loc)
7141 ret = ocfs2_read_xattr_block(old_inode, le64_to_cpu(di->i_xattr_loc),
7148 ret = ocfs2_reflink_xattr_in_block(&args, blk_bh);
7155 ocfs2_unlock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7157 brelse(ref_root_bh);
7159 if (ocfs2_dealloc_has_cluster(&dealloc)) {
7160 ocfs2_schedule_truncate_log_flush(OCFS2_SB(old_inode->i_sb), 1);
7161 ocfs2_run_deallocs(OCFS2_SB(old_inode->i_sb), &dealloc);
7169 * Initialize security and acl for a already created inode.
7170 * Used for reflink a non-preserve-security file.
7172 * It uses common api like ocfs2_xattr_set, so the caller
7173 * must not hold any lock expect i_mutex.
7175 int ocfs2_init_security_and_acl(struct inode *dir,
7176 struct inode *inode)
7179 struct buffer_head *dir_bh = NULL;
7180 struct ocfs2_security_xattr_info si = {
7184 ret = ocfs2_init_security_get(inode, dir, &si);
7186 ret = ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY,
7187 si.name, si.value, si.value_len,
7193 } else if (ret != -EOPNOTSUPP) {
7198 ret = ocfs2_inode_lock(dir, &dir_bh, 0);
7204 ret = ocfs2_init_acl(NULL, inode, dir, NULL, dir_bh, NULL, NULL);
7208 ocfs2_inode_unlock(dir, 0);
7214 * 'security' attributes support
7216 static size_t ocfs2_xattr_security_list(struct dentry *dentry, char *list,
7217 size_t list_size, const char *name,
7218 size_t name_len, int type)
7220 const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
7221 const size_t total_len = prefix_len + name_len + 1;
7223 if (list && total_len <= list_size) {
7224 memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
7225 memcpy(list + prefix_len, name, name_len);
7226 list[prefix_len + name_len] = '\0';
7231 static int ocfs2_xattr_security_get(struct dentry *dentry, const char *name,
7232 void *buffer, size_t size, int type)
7234 if (strcmp(name, "") == 0)
7236 return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_SECURITY,
7237 name, buffer, size);
7240 static int ocfs2_xattr_security_set(struct dentry *dentry, const char *name,
7241 const void *value, size_t size, int flags, int type)
7243 if (strcmp(name, "") == 0)
7246 return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_SECURITY,
7247 name, value, size, flags);
7250 int ocfs2_init_security_get(struct inode *inode,
7252 struct ocfs2_security_xattr_info *si)
7254 /* check whether ocfs2 support feature xattr */
7255 if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
7257 return security_inode_init_security(inode, dir, &si->name, &si->value,
7261 int ocfs2_init_security_set(handle_t *handle,
7262 struct inode *inode,
7263 struct buffer_head *di_bh,
7264 struct ocfs2_security_xattr_info *si,
7265 struct ocfs2_alloc_context *xattr_ac,
7266 struct ocfs2_alloc_context *data_ac)
7268 return ocfs2_xattr_set_handle(handle, inode, di_bh,
7269 OCFS2_XATTR_INDEX_SECURITY,
7270 si->name, si->value, si->value_len, 0,
7274 struct xattr_handler ocfs2_xattr_security_handler = {
7275 .prefix = XATTR_SECURITY_PREFIX,
7276 .list = ocfs2_xattr_security_list,
7277 .get = ocfs2_xattr_security_get,
7278 .set = ocfs2_xattr_security_set,
7282 * 'trusted' attributes support
7284 static size_t ocfs2_xattr_trusted_list(struct dentry *dentry, char *list,
7285 size_t list_size, const char *name,
7286 size_t name_len, int type)
7288 const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
7289 const size_t total_len = prefix_len + name_len + 1;
7291 if (list && total_len <= list_size) {
7292 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
7293 memcpy(list + prefix_len, name, name_len);
7294 list[prefix_len + name_len] = '\0';
7299 static int ocfs2_xattr_trusted_get(struct dentry *dentry, const char *name,
7300 void *buffer, size_t size, int type)
7302 if (strcmp(name, "") == 0)
7304 return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_TRUSTED,
7305 name, buffer, size);
7308 static int ocfs2_xattr_trusted_set(struct dentry *dentry, const char *name,
7309 const void *value, size_t size, int flags, int type)
7311 if (strcmp(name, "") == 0)
7314 return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_TRUSTED,
7315 name, value, size, flags);
7318 struct xattr_handler ocfs2_xattr_trusted_handler = {
7319 .prefix = XATTR_TRUSTED_PREFIX,
7320 .list = ocfs2_xattr_trusted_list,
7321 .get = ocfs2_xattr_trusted_get,
7322 .set = ocfs2_xattr_trusted_set,
7326 * 'user' attributes support
7328 static size_t ocfs2_xattr_user_list(struct dentry *dentry, char *list,
7329 size_t list_size, const char *name,
7330 size_t name_len, int type)
7332 const size_t prefix_len = XATTR_USER_PREFIX_LEN;
7333 const size_t total_len = prefix_len + name_len + 1;
7334 struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7336 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7339 if (list && total_len <= list_size) {
7340 memcpy(list, XATTR_USER_PREFIX, prefix_len);
7341 memcpy(list + prefix_len, name, name_len);
7342 list[prefix_len + name_len] = '\0';
7347 static int ocfs2_xattr_user_get(struct dentry *dentry, const char *name,
7348 void *buffer, size_t size, int type)
7350 struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7352 if (strcmp(name, "") == 0)
7354 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7356 return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_USER, name,
7360 static int ocfs2_xattr_user_set(struct dentry *dentry, const char *name,
7361 const void *value, size_t size, int flags, int type)
7363 struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7365 if (strcmp(name, "") == 0)
7367 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7370 return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_USER,
7371 name, value, size, flags);
7374 struct xattr_handler ocfs2_xattr_user_handler = {
7375 .prefix = XATTR_USER_PREFIX,
7376 .list = ocfs2_xattr_user_list,
7377 .get = ocfs2_xattr_user_get,
7378 .set = ocfs2_xattr_user_set,