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)
1577 struct ocfs2_xattr_header *xh = loc->xl_header;
1578 struct ocfs2_xattr_entry *entry = loc->xl_entry;
1580 ocfs2_xa_wipe_namevalue(loc);
1581 loc->xl_entry = NULL;
1583 le16_add_cpu(&xh->xh_count, -1);
1584 count = le16_to_cpu(xh->xh_count);
1587 * Only zero out the entry if there are more remaining. This is
1588 * important for an empty bucket, as it keeps track of the
1589 * bucket's hash value. It doesn't hurt empty block storage.
1592 index = ((char *)entry - (char *)&xh->xh_entries) /
1593 sizeof(struct ocfs2_xattr_entry);
1594 memmove(&xh->xh_entries[index], &xh->xh_entries[index + 1],
1595 (count - index) * sizeof(struct ocfs2_xattr_entry));
1596 memset(&xh->xh_entries[count], 0,
1597 sizeof(struct ocfs2_xattr_entry));
1601 static void ocfs2_init_dinode_xa_loc(struct ocfs2_xa_loc *loc,
1602 struct inode *inode,
1603 struct buffer_head *bh,
1604 struct ocfs2_xattr_entry *entry)
1606 struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1608 loc->xl_ops = &ocfs2_xa_block_loc_ops;
1609 loc->xl_storage = bh;
1610 loc->xl_entry = entry;
1612 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1613 loc->xl_size = le16_to_cpu(di->i_xattr_inline_size);
1616 loc->xl_size = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1619 (struct ocfs2_xattr_header *)(bh->b_data + bh->b_size -
1623 static void ocfs2_init_xattr_block_xa_loc(struct ocfs2_xa_loc *loc,
1624 struct buffer_head *bh,
1625 struct ocfs2_xattr_entry *entry)
1627 struct ocfs2_xattr_block *xb =
1628 (struct ocfs2_xattr_block *)bh->b_data;
1630 BUG_ON(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED);
1632 loc->xl_ops = &ocfs2_xa_block_loc_ops;
1633 loc->xl_storage = bh;
1634 loc->xl_header = &(xb->xb_attrs.xb_header);
1635 loc->xl_entry = entry;
1636 loc->xl_size = bh->b_size - offsetof(struct ocfs2_xattr_block,
1637 xb_attrs.xb_header);
1640 static void ocfs2_init_xattr_bucket_xa_loc(struct ocfs2_xa_loc *loc,
1641 struct ocfs2_xattr_bucket *bucket,
1642 struct ocfs2_xattr_entry *entry)
1644 loc->xl_ops = &ocfs2_xa_bucket_loc_ops;
1645 loc->xl_storage = bucket;
1646 loc->xl_header = bucket_xh(bucket);
1647 loc->xl_entry = entry;
1648 loc->xl_size = OCFS2_XATTR_BUCKET_SIZE;
1652 * ocfs2_xattr_set_entry_local()
1654 * Set, replace or remove extended attribute in local.
1656 static void ocfs2_xattr_set_entry_local(struct inode *inode,
1657 struct ocfs2_xattr_info *xi,
1658 struct ocfs2_xattr_search *xs,
1659 struct ocfs2_xattr_entry *last,
1662 size_t name_len = strlen(xi->name);
1663 struct ocfs2_xa_loc loc;
1665 if (xs->xattr_bh == xs->inode_bh)
1666 ocfs2_init_dinode_xa_loc(&loc, inode, xs->inode_bh,
1667 xs->not_found ? NULL : xs->here);
1669 ocfs2_init_xattr_block_xa_loc(&loc, xs->xattr_bh,
1670 xs->not_found ? NULL : xs->here);
1671 if (xi->value && xs->not_found) {
1672 /* Insert the new xattr entry. */
1673 le16_add_cpu(&xs->header->xh_count, 1);
1674 ocfs2_xattr_set_type(last, xi->name_index);
1675 ocfs2_xattr_set_local(last, 1);
1676 last->xe_name_len = name_len;
1682 first_val = xs->base + min_offs;
1683 offs = le16_to_cpu(xs->here->xe_name_offset);
1684 val = xs->base + offs;
1686 if (le64_to_cpu(xs->here->xe_value_size) >
1687 OCFS2_XATTR_INLINE_SIZE)
1688 size = OCFS2_XATTR_SIZE(name_len) +
1689 OCFS2_XATTR_ROOT_SIZE;
1691 size = OCFS2_XATTR_SIZE(name_len) +
1692 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1694 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1695 OCFS2_XATTR_SIZE(xi->value_len)) {
1696 /* The old and the new value have the
1697 same size. Just replace the value. */
1698 ocfs2_xattr_set_local(xs->here, 1);
1699 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1700 /* Clear value bytes. */
1701 memset(val + OCFS2_XATTR_SIZE(name_len),
1703 OCFS2_XATTR_SIZE(xi->value_len));
1704 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1711 ocfs2_xa_remove_entry(&loc);
1713 ocfs2_xa_wipe_namevalue(&loc);
1718 /* Insert the new name+value. */
1719 size_t size = OCFS2_XATTR_SIZE(name_len) +
1720 OCFS2_XATTR_SIZE(xi->value_len);
1721 void *val = xs->base + min_offs - size;
1723 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1724 memset(val, 0, size);
1725 memcpy(val, xi->name, name_len);
1726 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1729 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1730 ocfs2_xattr_set_local(xs->here, 1);
1731 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1738 * ocfs2_xattr_set_entry()
1740 * Set extended attribute entry into inode or block.
1742 * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1743 * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1744 * then set value in B tree with set_value_outside().
1746 static int ocfs2_xattr_set_entry(struct inode *inode,
1747 struct ocfs2_xattr_info *xi,
1748 struct ocfs2_xattr_search *xs,
1749 struct ocfs2_xattr_set_ctxt *ctxt,
1752 struct ocfs2_xattr_entry *last;
1753 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1754 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1755 size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1757 handle_t *handle = ctxt->handle;
1759 struct ocfs2_xattr_info xi_l = {
1760 .name_index = xi->name_index,
1763 .value_len = xi->value_len,
1765 struct ocfs2_xattr_value_buf vb = {
1766 .vb_bh = xs->xattr_bh,
1767 .vb_access = ocfs2_journal_access_di,
1770 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1771 BUG_ON(xs->xattr_bh == xs->inode_bh);
1772 vb.vb_access = ocfs2_journal_access_xb;
1774 BUG_ON(xs->xattr_bh != xs->inode_bh);
1776 /* Compute min_offs, last and free space. */
1777 last = xs->header->xh_entries;
1779 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1780 size_t offs = le16_to_cpu(last->xe_name_offset);
1781 if (offs < min_offs)
1786 free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
1790 if (!xs->not_found) {
1792 if (ocfs2_xattr_is_local(xs->here))
1793 size = OCFS2_XATTR_SIZE(name_len) +
1794 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1796 size = OCFS2_XATTR_SIZE(name_len) +
1797 OCFS2_XATTR_ROOT_SIZE;
1798 free += (size + sizeof(struct ocfs2_xattr_entry));
1800 /* Check free space in inode or block */
1801 if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1802 if (free < sizeof(struct ocfs2_xattr_entry) +
1803 OCFS2_XATTR_SIZE(name_len) +
1804 OCFS2_XATTR_ROOT_SIZE) {
1808 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1809 xi_l.value = (void *)&def_xv;
1810 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1811 } else if (xi->value) {
1812 if (free < sizeof(struct ocfs2_xattr_entry) +
1813 OCFS2_XATTR_SIZE(name_len) +
1814 OCFS2_XATTR_SIZE(xi->value_len)) {
1820 if (!xs->not_found) {
1821 /* For existing extended attribute */
1822 size_t size = OCFS2_XATTR_SIZE(name_len) +
1823 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1824 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1825 void *val = xs->base + offs;
1827 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1828 /* Replace existing local xattr with tree root */
1829 ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1834 } else if (!ocfs2_xattr_is_local(xs->here)) {
1835 /* For existing xattr which has value outside */
1836 vb.vb_xv = (struct ocfs2_xattr_value_root *)
1837 (val + OCFS2_XATTR_SIZE(name_len));
1839 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1841 * If new value need set outside also,
1842 * first truncate old value to new value,
1843 * then set new value with set_value_outside().
1845 ret = ocfs2_xattr_value_truncate(inode,
1854 ret = ocfs2_xattr_update_entry(inode,
1865 ret = __ocfs2_xattr_set_value_outside(inode,
1875 * If new value need set in local,
1876 * just trucate old value to zero.
1878 ret = ocfs2_xattr_value_truncate(inode,
1888 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), xs->inode_bh,
1889 OCFS2_JOURNAL_ACCESS_WRITE);
1895 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1896 ret = vb.vb_access(handle, INODE_CACHE(inode), vb.vb_bh,
1897 OCFS2_JOURNAL_ACCESS_WRITE);
1905 * Set value in local, include set tree root in local.
1906 * This is the first step for value size >INLINE_SIZE.
1908 ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1910 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1911 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1918 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1919 (flag & OCFS2_INLINE_XATTR_FL)) {
1920 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1921 unsigned int xattrsize = osb->s_xattr_inline_size;
1924 * Adjust extent record count or inline data size
1925 * to reserve space for extended attribute.
1927 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1928 struct ocfs2_inline_data *idata = &di->id2.i_data;
1929 le16_add_cpu(&idata->id_count, -xattrsize);
1930 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1931 struct ocfs2_extent_list *el = &di->id2.i_list;
1932 le16_add_cpu(&el->l_count, -(xattrsize /
1933 sizeof(struct ocfs2_extent_rec)));
1935 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1937 /* Update xattr flag */
1938 spin_lock(&oi->ip_lock);
1939 oi->ip_dyn_features |= flag;
1940 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1941 spin_unlock(&oi->ip_lock);
1943 ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1947 if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1949 * Set value outside in B tree.
1950 * This is the second step for value size > INLINE_SIZE.
1952 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1953 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, ctxt,
1960 * If set value outside failed, we have to clean
1961 * the junk tree root we have already set in local.
1963 ret2 = ocfs2_xattr_cleanup(inode, ctxt->handle,
1974 * In xattr remove, if it is stored outside and refcounted, we may have
1975 * the chance to split the refcount tree. So need the allocators.
1977 static int ocfs2_lock_xattr_remove_allocators(struct inode *inode,
1978 struct ocfs2_xattr_value_root *xv,
1979 struct ocfs2_caching_info *ref_ci,
1980 struct buffer_head *ref_root_bh,
1981 struct ocfs2_alloc_context **meta_ac,
1984 int ret, meta_add = 0;
1985 u32 p_cluster, num_clusters;
1986 unsigned int ext_flags;
1989 ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
1998 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
2001 ret = ocfs2_refcounted_xattr_delete_need(inode, ref_ci,
2003 &meta_add, ref_credits);
2009 ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
2018 static int ocfs2_remove_value_outside(struct inode*inode,
2019 struct ocfs2_xattr_value_buf *vb,
2020 struct ocfs2_xattr_header *header,
2021 struct ocfs2_caching_info *ref_ci,
2022 struct buffer_head *ref_root_bh)
2024 int ret = 0, i, ref_credits;
2025 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2026 struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2029 ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
2031 for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
2032 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
2034 if (ocfs2_xattr_is_local(entry))
2037 val = (void *)header +
2038 le16_to_cpu(entry->xe_name_offset);
2039 vb->vb_xv = (struct ocfs2_xattr_value_root *)
2040 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
2042 ret = ocfs2_lock_xattr_remove_allocators(inode, vb->vb_xv,
2043 ref_ci, ref_root_bh,
2047 ctxt.handle = ocfs2_start_trans(osb, ref_credits +
2048 ocfs2_remove_extent_credits(osb->sb));
2049 if (IS_ERR(ctxt.handle)) {
2050 ret = PTR_ERR(ctxt.handle);
2055 ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
2061 ocfs2_commit_trans(osb, ctxt.handle);
2063 ocfs2_free_alloc_context(ctxt.meta_ac);
2064 ctxt.meta_ac = NULL;
2069 ocfs2_free_alloc_context(ctxt.meta_ac);
2070 ocfs2_schedule_truncate_log_flush(osb, 1);
2071 ocfs2_run_deallocs(osb, &ctxt.dealloc);
2075 static int ocfs2_xattr_ibody_remove(struct inode *inode,
2076 struct buffer_head *di_bh,
2077 struct ocfs2_caching_info *ref_ci,
2078 struct buffer_head *ref_root_bh)
2081 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2082 struct ocfs2_xattr_header *header;
2084 struct ocfs2_xattr_value_buf vb = {
2086 .vb_access = ocfs2_journal_access_di,
2089 header = (struct ocfs2_xattr_header *)
2090 ((void *)di + inode->i_sb->s_blocksize -
2091 le16_to_cpu(di->i_xattr_inline_size));
2093 ret = ocfs2_remove_value_outside(inode, &vb, header,
2094 ref_ci, ref_root_bh);
2099 struct ocfs2_rm_xattr_bucket_para {
2100 struct ocfs2_caching_info *ref_ci;
2101 struct buffer_head *ref_root_bh;
2104 static int ocfs2_xattr_block_remove(struct inode *inode,
2105 struct buffer_head *blk_bh,
2106 struct ocfs2_caching_info *ref_ci,
2107 struct buffer_head *ref_root_bh)
2109 struct ocfs2_xattr_block *xb;
2111 struct ocfs2_xattr_value_buf vb = {
2113 .vb_access = ocfs2_journal_access_xb,
2115 struct ocfs2_rm_xattr_bucket_para args = {
2117 .ref_root_bh = ref_root_bh,
2120 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2121 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2122 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
2123 ret = ocfs2_remove_value_outside(inode, &vb, header,
2124 ref_ci, ref_root_bh);
2126 ret = ocfs2_iterate_xattr_index_block(inode,
2128 ocfs2_rm_xattr_cluster,
2134 static int ocfs2_xattr_free_block(struct inode *inode,
2136 struct ocfs2_caching_info *ref_ci,
2137 struct buffer_head *ref_root_bh)
2139 struct inode *xb_alloc_inode;
2140 struct buffer_head *xb_alloc_bh = NULL;
2141 struct buffer_head *blk_bh = NULL;
2142 struct ocfs2_xattr_block *xb;
2143 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2149 ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
2155 ret = ocfs2_xattr_block_remove(inode, blk_bh, ref_ci, ref_root_bh);
2161 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2162 blk = le64_to_cpu(xb->xb_blkno);
2163 bit = le16_to_cpu(xb->xb_suballoc_bit);
2164 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2166 xb_alloc_inode = ocfs2_get_system_file_inode(osb,
2167 EXTENT_ALLOC_SYSTEM_INODE,
2168 le16_to_cpu(xb->xb_suballoc_slot));
2169 if (!xb_alloc_inode) {
2174 mutex_lock(&xb_alloc_inode->i_mutex);
2176 ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
2182 handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
2183 if (IS_ERR(handle)) {
2184 ret = PTR_ERR(handle);
2189 ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
2194 ocfs2_commit_trans(osb, handle);
2196 ocfs2_inode_unlock(xb_alloc_inode, 1);
2197 brelse(xb_alloc_bh);
2199 mutex_unlock(&xb_alloc_inode->i_mutex);
2200 iput(xb_alloc_inode);
2207 * ocfs2_xattr_remove()
2209 * Free extended attribute resources associated with this inode.
2211 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
2213 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2214 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2215 struct ocfs2_refcount_tree *ref_tree = NULL;
2216 struct buffer_head *ref_root_bh = NULL;
2217 struct ocfs2_caching_info *ref_ci = NULL;
2221 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2224 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
2227 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
2228 ret = ocfs2_lock_refcount_tree(OCFS2_SB(inode->i_sb),
2229 le64_to_cpu(di->i_refcount_loc),
2230 1, &ref_tree, &ref_root_bh);
2235 ref_ci = &ref_tree->rf_ci;
2239 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2240 ret = ocfs2_xattr_ibody_remove(inode, di_bh,
2241 ref_ci, ref_root_bh);
2248 if (di->i_xattr_loc) {
2249 ret = ocfs2_xattr_free_block(inode,
2250 le64_to_cpu(di->i_xattr_loc),
2251 ref_ci, ref_root_bh);
2258 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
2259 OCFS2_INODE_UPDATE_CREDITS);
2260 if (IS_ERR(handle)) {
2261 ret = PTR_ERR(handle);
2265 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2266 OCFS2_JOURNAL_ACCESS_WRITE);
2272 di->i_xattr_loc = 0;
2274 spin_lock(&oi->ip_lock);
2275 oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
2276 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2277 spin_unlock(&oi->ip_lock);
2279 ret = ocfs2_journal_dirty(handle, di_bh);
2283 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
2286 ocfs2_unlock_refcount_tree(OCFS2_SB(inode->i_sb), ref_tree, 1);
2287 brelse(ref_root_bh);
2291 static int ocfs2_xattr_has_space_inline(struct inode *inode,
2292 struct ocfs2_dinode *di)
2294 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2295 unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
2298 if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
2301 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2302 struct ocfs2_inline_data *idata = &di->id2.i_data;
2303 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
2304 } else if (ocfs2_inode_is_fast_symlink(inode)) {
2305 free = ocfs2_fast_symlink_chars(inode->i_sb) -
2306 le64_to_cpu(di->i_size);
2308 struct ocfs2_extent_list *el = &di->id2.i_list;
2309 free = (le16_to_cpu(el->l_count) -
2310 le16_to_cpu(el->l_next_free_rec)) *
2311 sizeof(struct ocfs2_extent_rec);
2313 if (free >= xattrsize)
2320 * ocfs2_xattr_ibody_find()
2322 * Find extended attribute in inode block and
2323 * fill search info into struct ocfs2_xattr_search.
2325 static int ocfs2_xattr_ibody_find(struct inode *inode,
2328 struct ocfs2_xattr_search *xs)
2330 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2331 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2335 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2338 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2339 down_read(&oi->ip_alloc_sem);
2340 has_space = ocfs2_xattr_has_space_inline(inode, di);
2341 up_read(&oi->ip_alloc_sem);
2346 xs->xattr_bh = xs->inode_bh;
2347 xs->end = (void *)di + inode->i_sb->s_blocksize;
2348 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
2349 xs->header = (struct ocfs2_xattr_header *)
2350 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
2352 xs->header = (struct ocfs2_xattr_header *)
2353 (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
2354 xs->base = (void *)xs->header;
2355 xs->here = xs->header->xh_entries;
2357 /* Find the named attribute. */
2358 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2359 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2360 if (ret && ret != -ENODATA)
2362 xs->not_found = ret;
2369 * ocfs2_xattr_ibody_set()
2371 * Set, replace or remove an extended attribute into inode block.
2374 static int ocfs2_xattr_ibody_set(struct inode *inode,
2375 struct ocfs2_xattr_info *xi,
2376 struct ocfs2_xattr_search *xs,
2377 struct ocfs2_xattr_set_ctxt *ctxt)
2379 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2380 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2383 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2386 down_write(&oi->ip_alloc_sem);
2387 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2388 if (!ocfs2_xattr_has_space_inline(inode, di)) {
2394 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2395 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
2397 up_write(&oi->ip_alloc_sem);
2403 * ocfs2_xattr_block_find()
2405 * Find extended attribute in external block and
2406 * fill search info into struct ocfs2_xattr_search.
2408 static int ocfs2_xattr_block_find(struct inode *inode,
2411 struct ocfs2_xattr_search *xs)
2413 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2414 struct buffer_head *blk_bh = NULL;
2415 struct ocfs2_xattr_block *xb;
2418 if (!di->i_xattr_loc)
2421 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2428 xs->xattr_bh = blk_bh;
2429 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2431 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2432 xs->header = &xb->xb_attrs.xb_header;
2433 xs->base = (void *)xs->header;
2434 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2435 xs->here = xs->header->xh_entries;
2437 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2439 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2443 if (ret && ret != -ENODATA) {
2444 xs->xattr_bh = NULL;
2447 xs->not_found = ret;
2455 static int ocfs2_create_xattr_block(handle_t *handle,
2456 struct inode *inode,
2457 struct buffer_head *inode_bh,
2458 struct ocfs2_alloc_context *meta_ac,
2459 struct buffer_head **ret_bh,
2463 u16 suballoc_bit_start;
2466 struct ocfs2_dinode *di = (struct ocfs2_dinode *)inode_bh->b_data;
2467 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2468 struct buffer_head *new_bh = NULL;
2469 struct ocfs2_xattr_block *xblk;
2471 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), inode_bh,
2472 OCFS2_JOURNAL_ACCESS_CREATE);
2478 ret = ocfs2_claim_metadata(osb, handle, meta_ac, 1,
2479 &suballoc_bit_start, &num_got,
2486 new_bh = sb_getblk(inode->i_sb, first_blkno);
2487 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2489 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode),
2491 OCFS2_JOURNAL_ACCESS_CREATE);
2497 /* Initialize ocfs2_xattr_block */
2498 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2499 memset(xblk, 0, inode->i_sb->s_blocksize);
2500 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2501 xblk->xb_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
2502 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2503 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
2504 xblk->xb_blkno = cpu_to_le64(first_blkno);
2507 struct ocfs2_xattr_tree_root *xr = &xblk->xb_attrs.xb_root;
2508 xr->xt_clusters = cpu_to_le32(1);
2509 xr->xt_last_eb_blk = 0;
2510 xr->xt_list.l_tree_depth = 0;
2511 xr->xt_list.l_count = cpu_to_le16(
2512 ocfs2_xattr_recs_per_xb(inode->i_sb));
2513 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2514 xblk->xb_flags = cpu_to_le16(OCFS2_XATTR_INDEXED);
2517 ret = ocfs2_journal_dirty(handle, new_bh);
2522 di->i_xattr_loc = cpu_to_le64(first_blkno);
2523 ocfs2_journal_dirty(handle, inode_bh);
2534 * ocfs2_xattr_block_set()
2536 * Set, replace or remove an extended attribute into external block.
2539 static int ocfs2_xattr_block_set(struct inode *inode,
2540 struct ocfs2_xattr_info *xi,
2541 struct ocfs2_xattr_search *xs,
2542 struct ocfs2_xattr_set_ctxt *ctxt)
2544 struct buffer_head *new_bh = NULL;
2545 handle_t *handle = ctxt->handle;
2546 struct ocfs2_xattr_block *xblk = NULL;
2549 if (!xs->xattr_bh) {
2550 ret = ocfs2_create_xattr_block(handle, inode, xs->inode_bh,
2551 ctxt->meta_ac, &new_bh, 0);
2557 xs->xattr_bh = new_bh;
2558 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2559 xs->header = &xblk->xb_attrs.xb_header;
2560 xs->base = (void *)xs->header;
2561 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2562 xs->here = xs->header->xh_entries;
2564 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2566 if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2567 /* Set extended attribute into external block */
2568 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2569 OCFS2_HAS_XATTR_FL);
2570 if (!ret || ret != -ENOSPC)
2573 ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2578 ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2585 /* Check whether the new xattr can be inserted into the inode. */
2586 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2587 struct ocfs2_xattr_info *xi,
2588 struct ocfs2_xattr_search *xs)
2591 struct ocfs2_xattr_entry *last;
2593 size_t min_offs = xs->end - xs->base;
2598 last = xs->header->xh_entries;
2600 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
2601 size_t offs = le16_to_cpu(last->xe_name_offset);
2602 if (offs < min_offs)
2607 free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
2611 BUG_ON(!xs->not_found);
2613 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2614 value_size = OCFS2_XATTR_ROOT_SIZE;
2616 value_size = OCFS2_XATTR_SIZE(xi->value_len);
2618 if (free >= sizeof(struct ocfs2_xattr_entry) +
2619 OCFS2_XATTR_SIZE(strlen(xi->name)) + value_size)
2625 static int ocfs2_calc_xattr_set_need(struct inode *inode,
2626 struct ocfs2_dinode *di,
2627 struct ocfs2_xattr_info *xi,
2628 struct ocfs2_xattr_search *xis,
2629 struct ocfs2_xattr_search *xbs,
2634 int ret = 0, old_in_xb = 0;
2635 int clusters_add = 0, meta_add = 0, credits = 0;
2636 struct buffer_head *bh = NULL;
2637 struct ocfs2_xattr_block *xb = NULL;
2638 struct ocfs2_xattr_entry *xe = NULL;
2639 struct ocfs2_xattr_value_root *xv = NULL;
2641 int name_offset, name_len = 0;
2642 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2647 * Calculate the clusters we need to write.
2648 * No matter whether we replace an old one or add a new one,
2649 * we need this for writing.
2651 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2652 credits += new_clusters *
2653 ocfs2_clusters_to_blocks(inode->i_sb, 1);
2655 if (xis->not_found && xbs->not_found) {
2656 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2658 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2659 clusters_add += new_clusters;
2660 credits += ocfs2_calc_extend_credits(inode->i_sb,
2668 if (!xis->not_found) {
2670 name_offset = le16_to_cpu(xe->xe_name_offset);
2671 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2673 credits += OCFS2_INODE_UPDATE_CREDITS;
2675 int i, block_off = 0;
2676 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2678 name_offset = le16_to_cpu(xe->xe_name_offset);
2679 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2680 i = xbs->here - xbs->header->xh_entries;
2683 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2684 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
2685 bucket_xh(xbs->bucket),
2688 base = bucket_block(xbs->bucket, block_off);
2689 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2692 credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
2697 * delete a xattr doesn't need metadata and cluster allocation.
2698 * so just calculate the credits and return.
2700 * The credits for removing the value tree will be extended
2701 * by ocfs2_remove_extent itself.
2704 if (!ocfs2_xattr_is_local(xe))
2705 credits += ocfs2_remove_extent_credits(inode->i_sb);
2710 /* do cluster allocation guess first. */
2711 value_size = le64_to_cpu(xe->xe_value_size);
2715 * In xattr set, we always try to set the xe in inode first,
2716 * so if it can be inserted into inode successfully, the old
2717 * one will be removed from the xattr block, and this xattr
2718 * will be inserted into inode as a new xattr in inode.
2720 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
2721 clusters_add += new_clusters;
2722 credits += ocfs2_remove_extent_credits(inode->i_sb) +
2723 OCFS2_INODE_UPDATE_CREDITS;
2724 if (!ocfs2_xattr_is_local(xe))
2725 credits += ocfs2_calc_extend_credits(
2733 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2734 /* the new values will be stored outside. */
2735 u32 old_clusters = 0;
2737 if (!ocfs2_xattr_is_local(xe)) {
2738 old_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2740 xv = (struct ocfs2_xattr_value_root *)
2741 (base + name_offset + name_len);
2742 value_size = OCFS2_XATTR_ROOT_SIZE;
2746 if (old_clusters >= new_clusters) {
2747 credits += ocfs2_remove_extent_credits(inode->i_sb);
2750 meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
2751 clusters_add += new_clusters - old_clusters;
2752 credits += ocfs2_calc_extend_credits(inode->i_sb,
2756 if (value_size >= OCFS2_XATTR_ROOT_SIZE)
2761 * Now the new value will be stored inside. So if the new
2762 * value is smaller than the size of value root or the old
2763 * value, we don't need any allocation, otherwise we have
2764 * to guess metadata allocation.
2766 if ((ocfs2_xattr_is_local(xe) && value_size >= xi->value_len) ||
2767 (!ocfs2_xattr_is_local(xe) &&
2768 OCFS2_XATTR_ROOT_SIZE >= xi->value_len))
2773 /* calculate metadata allocation. */
2774 if (di->i_xattr_loc) {
2775 if (!xbs->xattr_bh) {
2776 ret = ocfs2_read_xattr_block(inode,
2777 le64_to_cpu(di->i_xattr_loc),
2784 xb = (struct ocfs2_xattr_block *)bh->b_data;
2786 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2789 * If there is already an xattr tree, good, we can calculate
2790 * like other b-trees. Otherwise we may have the chance of
2791 * create a tree, the credit calculation is borrowed from
2792 * ocfs2_calc_extend_credits with root_el = NULL. And the
2793 * new tree will be cluster based, so no meta is needed.
2795 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2796 struct ocfs2_extent_list *el =
2797 &xb->xb_attrs.xb_root.xt_list;
2798 meta_add += ocfs2_extend_meta_needed(el);
2799 credits += ocfs2_calc_extend_credits(inode->i_sb,
2802 credits += OCFS2_SUBALLOC_ALLOC + 1;
2805 * This cluster will be used either for new bucket or for
2807 * If the cluster size is the same as the bucket size, one
2808 * more is needed since we may need to extend the bucket
2812 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2813 if (OCFS2_XATTR_BUCKET_SIZE ==
2814 OCFS2_SB(inode->i_sb)->s_clustersize) {
2815 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2820 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
2824 *clusters_need = clusters_add;
2826 *meta_need = meta_add;
2828 *credits_need = credits;
2833 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
2834 struct ocfs2_dinode *di,
2835 struct ocfs2_xattr_info *xi,
2836 struct ocfs2_xattr_search *xis,
2837 struct ocfs2_xattr_search *xbs,
2838 struct ocfs2_xattr_set_ctxt *ctxt,
2842 int clusters_add, meta_add, ret;
2843 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2845 memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
2847 ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
2849 ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
2850 &clusters_add, &meta_add, credits);
2856 meta_add += extra_meta;
2857 mlog(0, "Set xattr %s, reserve meta blocks = %d, clusters = %d, "
2858 "credits = %d\n", xi->name, meta_add, clusters_add, *credits);
2861 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
2870 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
2876 if (ctxt->meta_ac) {
2877 ocfs2_free_alloc_context(ctxt->meta_ac);
2878 ctxt->meta_ac = NULL;
2882 * We cannot have an error and a non null ctxt->data_ac.
2889 static int __ocfs2_xattr_set_handle(struct inode *inode,
2890 struct ocfs2_dinode *di,
2891 struct ocfs2_xattr_info *xi,
2892 struct ocfs2_xattr_search *xis,
2893 struct ocfs2_xattr_search *xbs,
2894 struct ocfs2_xattr_set_ctxt *ctxt)
2896 int ret = 0, credits, old_found;
2899 /* Remove existing extended attribute */
2900 if (!xis->not_found)
2901 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2902 else if (!xbs->not_found)
2903 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2905 /* We always try to set extended attribute into inode first*/
2906 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2907 if (!ret && !xbs->not_found) {
2909 * If succeed and that extended attribute existing in
2910 * external block, then we will remove it.
2915 old_found = xis->not_found;
2916 xis->not_found = -ENODATA;
2917 ret = ocfs2_calc_xattr_set_need(inode,
2925 xis->not_found = old_found;
2931 ret = ocfs2_extend_trans(ctxt->handle, credits +
2932 ctxt->handle->h_buffer_credits);
2937 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2938 } else if (ret == -ENOSPC) {
2939 if (di->i_xattr_loc && !xbs->xattr_bh) {
2940 ret = ocfs2_xattr_block_find(inode,
2946 old_found = xis->not_found;
2947 xis->not_found = -ENODATA;
2948 ret = ocfs2_calc_xattr_set_need(inode,
2956 xis->not_found = old_found;
2962 ret = ocfs2_extend_trans(ctxt->handle, credits +
2963 ctxt->handle->h_buffer_credits);
2970 * If no space in inode, we will set extended attribute
2971 * into external block.
2973 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2976 if (!xis->not_found) {
2978 * If succeed and that extended attribute
2979 * existing in inode, we will remove it.
2983 xbs->not_found = -ENODATA;
2984 ret = ocfs2_calc_xattr_set_need(inode,
2997 ret = ocfs2_extend_trans(ctxt->handle, credits +
2998 ctxt->handle->h_buffer_credits);
3003 ret = ocfs2_xattr_ibody_set(inode, xi,
3010 /* Update inode ctime. */
3011 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
3013 OCFS2_JOURNAL_ACCESS_WRITE);
3019 inode->i_ctime = CURRENT_TIME;
3020 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
3021 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
3022 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
3029 * This function only called duing creating inode
3030 * for init security/acl xattrs of the new inode.
3031 * All transanction credits have been reserved in mknod.
3033 int ocfs2_xattr_set_handle(handle_t *handle,
3034 struct inode *inode,
3035 struct buffer_head *di_bh,
3041 struct ocfs2_alloc_context *meta_ac,
3042 struct ocfs2_alloc_context *data_ac)
3044 struct ocfs2_dinode *di;
3047 struct ocfs2_xattr_info xi = {
3048 .name_index = name_index,
3051 .value_len = value_len,
3054 struct ocfs2_xattr_search xis = {
3055 .not_found = -ENODATA,
3058 struct ocfs2_xattr_search xbs = {
3059 .not_found = -ENODATA,
3062 struct ocfs2_xattr_set_ctxt ctxt = {
3068 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3072 * In extreme situation, may need xattr bucket when
3073 * block size is too small. And we have already reserved
3074 * the credits for bucket in mknod.
3076 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
3077 xbs.bucket = ocfs2_xattr_bucket_new(inode);
3079 mlog_errno(-ENOMEM);
3084 xis.inode_bh = xbs.inode_bh = di_bh;
3085 di = (struct ocfs2_dinode *)di_bh->b_data;
3087 down_write(&OCFS2_I(inode)->ip_xattr_sem);
3089 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3092 if (xis.not_found) {
3093 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3098 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3101 up_write(&OCFS2_I(inode)->ip_xattr_sem);
3102 brelse(xbs.xattr_bh);
3103 ocfs2_xattr_bucket_free(xbs.bucket);
3111 * Set, replace or remove an extended attribute for this inode.
3112 * value is NULL to remove an existing extended attribute, else either
3113 * create or replace an extended attribute.
3115 int ocfs2_xattr_set(struct inode *inode,
3122 struct buffer_head *di_bh = NULL;
3123 struct ocfs2_dinode *di;
3124 int ret, credits, ref_meta = 0, ref_credits = 0;
3125 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3126 struct inode *tl_inode = osb->osb_tl_inode;
3127 struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
3128 struct ocfs2_refcount_tree *ref_tree = NULL;
3130 struct ocfs2_xattr_info xi = {
3131 .name_index = name_index,
3134 .value_len = value_len,
3137 struct ocfs2_xattr_search xis = {
3138 .not_found = -ENODATA,
3141 struct ocfs2_xattr_search xbs = {
3142 .not_found = -ENODATA,
3145 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3149 * Only xbs will be used on indexed trees. xis doesn't need a
3152 xbs.bucket = ocfs2_xattr_bucket_new(inode);
3154 mlog_errno(-ENOMEM);
3158 ret = ocfs2_inode_lock(inode, &di_bh, 1);
3161 goto cleanup_nolock;
3163 xis.inode_bh = xbs.inode_bh = di_bh;
3164 di = (struct ocfs2_dinode *)di_bh->b_data;
3166 down_write(&OCFS2_I(inode)->ip_xattr_sem);
3168 * Scan inode and external block to find the same name
3169 * extended attribute and collect search infomation.
3171 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3174 if (xis.not_found) {
3175 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3180 if (xis.not_found && xbs.not_found) {
3182 if (flags & XATTR_REPLACE)
3189 if (flags & XATTR_CREATE)
3193 /* Check whether the value is refcounted and do some prepartion. */
3194 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL &&
3195 (!xis.not_found || !xbs.not_found)) {
3196 ret = ocfs2_prepare_refcount_xattr(inode, di, &xi,
3197 &xis, &xbs, &ref_tree,
3198 &ref_meta, &ref_credits);
3205 mutex_lock(&tl_inode->i_mutex);
3207 if (ocfs2_truncate_log_needs_flush(osb)) {
3208 ret = __ocfs2_flush_truncate_log(osb);
3210 mutex_unlock(&tl_inode->i_mutex);
3215 mutex_unlock(&tl_inode->i_mutex);
3217 ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
3218 &xbs, &ctxt, ref_meta, &credits);
3224 /* we need to update inode's ctime field, so add credit for it. */
3225 credits += OCFS2_INODE_UPDATE_CREDITS;
3226 ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
3227 if (IS_ERR(ctxt.handle)) {
3228 ret = PTR_ERR(ctxt.handle);
3233 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3235 ocfs2_commit_trans(osb, ctxt.handle);
3238 ocfs2_free_alloc_context(ctxt.data_ac);
3240 ocfs2_free_alloc_context(ctxt.meta_ac);
3241 if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
3242 ocfs2_schedule_truncate_log_flush(osb, 1);
3243 ocfs2_run_deallocs(osb, &ctxt.dealloc);
3247 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3248 up_write(&OCFS2_I(inode)->ip_xattr_sem);
3249 if (!value && !ret) {
3250 ret = ocfs2_try_remove_refcount_tree(inode, di_bh);
3254 ocfs2_inode_unlock(inode, 1);
3257 brelse(xbs.xattr_bh);
3258 ocfs2_xattr_bucket_free(xbs.bucket);
3264 * Find the xattr extent rec which may contains name_hash.
3265 * e_cpos will be the first name hash of the xattr rec.
3266 * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
3268 static int ocfs2_xattr_get_rec(struct inode *inode,
3273 struct ocfs2_extent_list *el)
3276 struct buffer_head *eb_bh = NULL;
3277 struct ocfs2_extent_block *eb;
3278 struct ocfs2_extent_rec *rec = NULL;
3281 if (el->l_tree_depth) {
3282 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, name_hash,
3289 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
3292 if (el->l_tree_depth) {
3293 ocfs2_error(inode->i_sb,
3294 "Inode %lu has non zero tree depth in "
3295 "xattr tree block %llu\n", inode->i_ino,
3296 (unsigned long long)eb_bh->b_blocknr);
3302 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
3303 rec = &el->l_recs[i];
3305 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
3306 e_blkno = le64_to_cpu(rec->e_blkno);
3312 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
3313 "record (%u, %u, 0) in xattr", inode->i_ino,
3314 le32_to_cpu(rec->e_cpos),
3315 ocfs2_rec_clusters(el, rec));
3320 *p_blkno = le64_to_cpu(rec->e_blkno);
3321 *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
3323 *e_cpos = le32_to_cpu(rec->e_cpos);
3329 typedef int (xattr_bucket_func)(struct inode *inode,
3330 struct ocfs2_xattr_bucket *bucket,
3333 static int ocfs2_find_xe_in_bucket(struct inode *inode,
3334 struct ocfs2_xattr_bucket *bucket,
3341 int i, ret = 0, cmp = 1, block_off, new_offset;
3342 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3343 size_t name_len = strlen(name);
3344 struct ocfs2_xattr_entry *xe = NULL;
3348 * We don't use binary search in the bucket because there
3349 * may be multiple entries with the same name hash.
3351 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3352 xe = &xh->xh_entries[i];
3354 if (name_hash > le32_to_cpu(xe->xe_name_hash))
3356 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
3359 cmp = name_index - ocfs2_xattr_get_type(xe);
3361 cmp = name_len - xe->xe_name_len;
3365 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3376 xe_name = bucket_block(bucket, block_off) + new_offset;
3377 if (!memcmp(name, xe_name, name_len)) {
3389 * Find the specified xattr entry in a series of buckets.
3390 * This series start from p_blkno and last for num_clusters.
3391 * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
3392 * the num of the valid buckets.
3394 * Return the buffer_head this xattr should reside in. And if the xattr's
3395 * hash is in the gap of 2 buckets, return the lower bucket.
3397 static int ocfs2_xattr_bucket_find(struct inode *inode,
3404 struct ocfs2_xattr_search *xs)
3407 struct ocfs2_xattr_header *xh = NULL;
3408 struct ocfs2_xattr_entry *xe = NULL;
3410 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3411 int low_bucket = 0, bucket, high_bucket;
3412 struct ocfs2_xattr_bucket *search;
3414 u64 blkno, lower_blkno = 0;
3416 search = ocfs2_xattr_bucket_new(inode);
3423 ret = ocfs2_read_xattr_bucket(search, p_blkno);
3429 xh = bucket_xh(search);
3430 high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
3431 while (low_bucket <= high_bucket) {
3432 ocfs2_xattr_bucket_relse(search);
3434 bucket = (low_bucket + high_bucket) / 2;
3435 blkno = p_blkno + bucket * blk_per_bucket;
3436 ret = ocfs2_read_xattr_bucket(search, blkno);
3442 xh = bucket_xh(search);
3443 xe = &xh->xh_entries[0];
3444 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3445 high_bucket = bucket - 1;
3450 * Check whether the hash of the last entry in our
3451 * bucket is larger than the search one. for an empty
3452 * bucket, the last one is also the first one.
3455 xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3457 last_hash = le32_to_cpu(xe->xe_name_hash);
3459 /* record lower_blkno which may be the insert place. */
3460 lower_blkno = blkno;
3462 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3463 low_bucket = bucket + 1;
3467 /* the searched xattr should reside in this bucket if exists. */
3468 ret = ocfs2_find_xe_in_bucket(inode, search,
3469 name_index, name, name_hash,
3479 * Record the bucket we have found.
3480 * When the xattr's hash value is in the gap of 2 buckets, we will
3481 * always set it to the previous bucket.
3484 lower_blkno = p_blkno;
3486 /* This should be in cache - we just read it during the search */
3487 ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3493 xs->header = bucket_xh(xs->bucket);
3494 xs->base = bucket_block(xs->bucket, 0);
3495 xs->end = xs->base + inode->i_sb->s_blocksize;
3498 xs->here = &xs->header->xh_entries[index];
3499 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
3500 (unsigned long long)bucket_blkno(xs->bucket), index);
3505 ocfs2_xattr_bucket_free(search);
3509 static int ocfs2_xattr_index_block_find(struct inode *inode,
3510 struct buffer_head *root_bh,
3513 struct ocfs2_xattr_search *xs)
3516 struct ocfs2_xattr_block *xb =
3517 (struct ocfs2_xattr_block *)root_bh->b_data;
3518 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3519 struct ocfs2_extent_list *el = &xb_root->xt_list;
3521 u32 first_hash, num_clusters = 0;
3522 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3524 if (le16_to_cpu(el->l_next_free_rec) == 0)
3527 mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
3528 name, name_hash, name_index);
3530 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3537 BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3539 mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
3540 "in the rec is %u\n", num_clusters, (unsigned long long)p_blkno,
3543 ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3544 p_blkno, first_hash, num_clusters, xs);
3550 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3553 xattr_bucket_func *func,
3557 u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3558 u32 num_buckets = clusters * bpc;
3559 struct ocfs2_xattr_bucket *bucket;
3561 bucket = ocfs2_xattr_bucket_new(inode);
3563 mlog_errno(-ENOMEM);
3567 mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
3568 clusters, (unsigned long long)blkno);
3570 for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3571 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3578 * The real bucket num in this series of blocks is stored
3579 * in the 1st bucket.
3582 num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3584 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
3585 (unsigned long long)blkno,
3586 le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3588 ret = func(inode, bucket, para);
3589 if (ret && ret != -ERANGE)
3591 /* Fall through to bucket_relse() */
3594 ocfs2_xattr_bucket_relse(bucket);
3599 ocfs2_xattr_bucket_free(bucket);
3603 struct ocfs2_xattr_tree_list {
3609 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
3610 struct ocfs2_xattr_header *xh,
3617 if (index < 0 || index >= le16_to_cpu(xh->xh_count))
3620 name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
3622 *block_off = name_offset >> sb->s_blocksize_bits;
3623 *new_offset = name_offset % sb->s_blocksize;
3628 static int ocfs2_list_xattr_bucket(struct inode *inode,
3629 struct ocfs2_xattr_bucket *bucket,
3633 struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
3634 int i, block_off, new_offset;
3635 const char *prefix, *name;
3637 for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
3638 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
3639 type = ocfs2_xattr_get_type(entry);
3640 prefix = ocfs2_xattr_prefix(type);
3643 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3651 name = (const char *)bucket_block(bucket, block_off) +
3653 ret = ocfs2_xattr_list_entry(xl->buffer,
3657 entry->xe_name_len);
3666 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
3667 struct buffer_head *blk_bh,
3668 xattr_tree_rec_func *rec_func,
3671 struct ocfs2_xattr_block *xb =
3672 (struct ocfs2_xattr_block *)blk_bh->b_data;
3673 struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
3675 u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
3678 if (!el->l_next_free_rec || !rec_func)
3681 while (name_hash > 0) {
3682 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
3683 &e_cpos, &num_clusters, el);
3689 ret = rec_func(inode, blk_bh, p_blkno, e_cpos,
3690 num_clusters, para);
3700 name_hash = e_cpos - 1;
3707 static int ocfs2_list_xattr_tree_rec(struct inode *inode,
3708 struct buffer_head *root_bh,
3709 u64 blkno, u32 cpos, u32 len, void *para)
3711 return ocfs2_iterate_xattr_buckets(inode, blkno, len,
3712 ocfs2_list_xattr_bucket, para);
3715 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
3716 struct buffer_head *blk_bh,
3721 struct ocfs2_xattr_tree_list xl = {
3723 .buffer_size = buffer_size,
3727 ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
3728 ocfs2_list_xattr_tree_rec, &xl);
3739 static int cmp_xe(const void *a, const void *b)
3741 const struct ocfs2_xattr_entry *l = a, *r = b;
3742 u32 l_hash = le32_to_cpu(l->xe_name_hash);
3743 u32 r_hash = le32_to_cpu(r->xe_name_hash);
3745 if (l_hash > r_hash)
3747 if (l_hash < r_hash)
3752 static void swap_xe(void *a, void *b, int size)
3754 struct ocfs2_xattr_entry *l = a, *r = b, tmp;
3757 memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
3758 memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
3762 * When the ocfs2_xattr_block is filled up, new bucket will be created
3763 * and all the xattr entries will be moved to the new bucket.
3764 * The header goes at the start of the bucket, and the names+values are
3765 * filled from the end. This is why *target starts as the last buffer.
3766 * Note: we need to sort the entries since they are not saved in order
3767 * in the ocfs2_xattr_block.
3769 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
3770 struct buffer_head *xb_bh,
3771 struct ocfs2_xattr_bucket *bucket)
3773 int i, blocksize = inode->i_sb->s_blocksize;
3774 int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3775 u16 offset, size, off_change;
3776 struct ocfs2_xattr_entry *xe;
3777 struct ocfs2_xattr_block *xb =
3778 (struct ocfs2_xattr_block *)xb_bh->b_data;
3779 struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
3780 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3781 u16 count = le16_to_cpu(xb_xh->xh_count);
3782 char *src = xb_bh->b_data;
3783 char *target = bucket_block(bucket, blks - 1);
3785 mlog(0, "cp xattr from block %llu to bucket %llu\n",
3786 (unsigned long long)xb_bh->b_blocknr,
3787 (unsigned long long)bucket_blkno(bucket));
3789 for (i = 0; i < blks; i++)
3790 memset(bucket_block(bucket, i), 0, blocksize);
3793 * Since the xe_name_offset is based on ocfs2_xattr_header,
3794 * there is a offset change corresponding to the change of
3795 * ocfs2_xattr_header's position.
3797 off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3798 xe = &xb_xh->xh_entries[count - 1];
3799 offset = le16_to_cpu(xe->xe_name_offset) + off_change;
3800 size = blocksize - offset;
3802 /* copy all the names and values. */
3803 memcpy(target + offset, src + offset, size);
3805 /* Init new header now. */
3806 xh->xh_count = xb_xh->xh_count;
3807 xh->xh_num_buckets = cpu_to_le16(1);
3808 xh->xh_name_value_len = cpu_to_le16(size);
3809 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
3811 /* copy all the entries. */
3812 target = bucket_block(bucket, 0);
3813 offset = offsetof(struct ocfs2_xattr_header, xh_entries);
3814 size = count * sizeof(struct ocfs2_xattr_entry);
3815 memcpy(target + offset, (char *)xb_xh + offset, size);
3817 /* Change the xe offset for all the xe because of the move. */
3818 off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
3819 offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3820 for (i = 0; i < count; i++)
3821 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
3823 mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
3824 offset, size, off_change);
3826 sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
3831 * After we move xattr from block to index btree, we have to
3832 * update ocfs2_xattr_search to the new xe and base.
3834 * When the entry is in xattr block, xattr_bh indicates the storage place.
3835 * While if the entry is in index b-tree, "bucket" indicates the
3836 * real place of the xattr.
3838 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
3839 struct ocfs2_xattr_search *xs,
3840 struct buffer_head *old_bh)
3842 char *buf = old_bh->b_data;
3843 struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
3844 struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
3847 xs->header = bucket_xh(xs->bucket);
3848 xs->base = bucket_block(xs->bucket, 0);
3849 xs->end = xs->base + inode->i_sb->s_blocksize;
3854 i = xs->here - old_xh->xh_entries;
3855 xs->here = &xs->header->xh_entries[i];
3858 static int ocfs2_xattr_create_index_block(struct inode *inode,
3859 struct ocfs2_xattr_search *xs,
3860 struct ocfs2_xattr_set_ctxt *ctxt)
3865 handle_t *handle = ctxt->handle;
3866 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3867 struct ocfs2_inode_info *oi = OCFS2_I(inode);
3868 struct buffer_head *xb_bh = xs->xattr_bh;
3869 struct ocfs2_xattr_block *xb =
3870 (struct ocfs2_xattr_block *)xb_bh->b_data;
3871 struct ocfs2_xattr_tree_root *xr;
3872 u16 xb_flags = le16_to_cpu(xb->xb_flags);
3874 mlog(0, "create xattr index block for %llu\n",
3875 (unsigned long long)xb_bh->b_blocknr);
3877 BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
3878 BUG_ON(!xs->bucket);
3882 * We can use this lock for now, and maybe move to a dedicated mutex
3883 * if performance becomes a problem later.
3885 down_write(&oi->ip_alloc_sem);
3887 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), xb_bh,
3888 OCFS2_JOURNAL_ACCESS_WRITE);
3894 ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac,
3895 1, 1, &bit_off, &len);
3902 * The bucket may spread in many blocks, and
3903 * we will only touch the 1st block and the last block
3904 * in the whole bucket(one for entry and one for data).
3906 blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
3908 mlog(0, "allocate 1 cluster from %llu to xattr block\n",
3909 (unsigned long long)blkno);
3911 ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
3917 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
3918 OCFS2_JOURNAL_ACCESS_CREATE);
3924 ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
3925 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
3927 ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
3929 /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
3930 memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
3931 offsetof(struct ocfs2_xattr_block, xb_attrs));
3933 xr = &xb->xb_attrs.xb_root;
3934 xr->xt_clusters = cpu_to_le32(1);
3935 xr->xt_last_eb_blk = 0;
3936 xr->xt_list.l_tree_depth = 0;
3937 xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
3938 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
3940 xr->xt_list.l_recs[0].e_cpos = 0;
3941 xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
3942 xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
3944 xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
3946 ocfs2_journal_dirty(handle, xb_bh);
3949 up_write(&oi->ip_alloc_sem);
3954 static int cmp_xe_offset(const void *a, const void *b)
3956 const struct ocfs2_xattr_entry *l = a, *r = b;
3957 u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
3958 u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
3960 if (l_name_offset < r_name_offset)
3962 if (l_name_offset > r_name_offset)
3968 * defrag a xattr bucket if we find that the bucket has some
3969 * holes beteen name/value pairs.
3970 * We will move all the name/value pairs to the end of the bucket
3971 * so that we can spare some space for insertion.
3973 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
3975 struct ocfs2_xattr_bucket *bucket)
3978 size_t end, offset, len, value_len;
3979 struct ocfs2_xattr_header *xh;
3980 char *entries, *buf, *bucket_buf = NULL;
3981 u64 blkno = bucket_blkno(bucket);
3983 size_t blocksize = inode->i_sb->s_blocksize;
3984 struct ocfs2_xattr_entry *xe;
3987 * In order to make the operation more efficient and generic,
3988 * we copy all the blocks into a contiguous memory and do the
3989 * defragment there, so if anything is error, we will not touch
3992 bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
3999 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4000 memcpy(buf, bucket_block(bucket, i), blocksize);
4002 ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
4003 OCFS2_JOURNAL_ACCESS_WRITE);
4009 xh = (struct ocfs2_xattr_header *)bucket_buf;
4010 entries = (char *)xh->xh_entries;
4011 xh_free_start = le16_to_cpu(xh->xh_free_start);
4013 mlog(0, "adjust xattr bucket in %llu, count = %u, "
4014 "xh_free_start = %u, xh_name_value_len = %u.\n",
4015 (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
4016 xh_free_start, le16_to_cpu(xh->xh_name_value_len));
4019 * sort all the entries by their offset.
4020 * the largest will be the first, so that we can
4021 * move them to the end one by one.
4023 sort(entries, le16_to_cpu(xh->xh_count),
4024 sizeof(struct ocfs2_xattr_entry),
4025 cmp_xe_offset, swap_xe);
4027 /* Move all name/values to the end of the bucket. */
4028 xe = xh->xh_entries;
4029 end = OCFS2_XATTR_BUCKET_SIZE;
4030 for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
4031 offset = le16_to_cpu(xe->xe_name_offset);
4032 if (ocfs2_xattr_is_local(xe))
4033 value_len = OCFS2_XATTR_SIZE(
4034 le64_to_cpu(xe->xe_value_size));
4036 value_len = OCFS2_XATTR_ROOT_SIZE;
4037 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
4040 * We must make sure that the name/value pair
4041 * exist in the same block. So adjust end to
4042 * the previous block end if needed.
4044 if (((end - len) / blocksize !=
4045 (end - 1) / blocksize))
4046 end = end - end % blocksize;
4048 if (end > offset + len) {
4049 memmove(bucket_buf + end - len,
4050 bucket_buf + offset, len);
4051 xe->xe_name_offset = cpu_to_le16(end - len);
4054 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
4055 "bucket %llu\n", (unsigned long long)blkno);
4060 mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
4061 "bucket %llu\n", (unsigned long long)blkno);
4063 if (xh_free_start == end)
4066 memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
4067 xh->xh_free_start = cpu_to_le16(end);
4069 /* sort the entries by their name_hash. */
4070 sort(entries, le16_to_cpu(xh->xh_count),
4071 sizeof(struct ocfs2_xattr_entry),
4075 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4076 memcpy(bucket_block(bucket, i), buf, blocksize);
4077 ocfs2_xattr_bucket_journal_dirty(handle, bucket);
4085 * prev_blkno points to the start of an existing extent. new_blkno
4086 * points to a newly allocated extent. Because we know each of our
4087 * clusters contains more than bucket, we can easily split one cluster
4088 * at a bucket boundary. So we take the last cluster of the existing
4089 * extent and split it down the middle. We move the last half of the
4090 * buckets in the last cluster of the existing extent over to the new
4093 * first_bh is the buffer at prev_blkno so we can update the existing
4094 * extent's bucket count. header_bh is the bucket were we were hoping
4095 * to insert our xattr. If the bucket move places the target in the new
4096 * extent, we'll update first_bh and header_bh after modifying the old
4099 * first_hash will be set as the 1st xe's name_hash in the new extent.
4101 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
4103 struct ocfs2_xattr_bucket *first,
4104 struct ocfs2_xattr_bucket *target,
4110 struct super_block *sb = inode->i_sb;
4111 int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
4112 int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
4113 int to_move = num_buckets / 2;
4115 u64 last_cluster_blkno = bucket_blkno(first) +
4116 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
4118 BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
4119 BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
4121 mlog(0, "move half of xattrs in cluster %llu to %llu\n",
4122 (unsigned long long)last_cluster_blkno, (unsigned long long)new_blkno);
4124 ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
4125 last_cluster_blkno, new_blkno,
4126 to_move, first_hash);
4132 /* This is the first bucket that got moved */
4133 src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
4136 * If the target bucket was part of the moved buckets, we need to
4137 * update first and target.
4139 if (bucket_blkno(target) >= src_blkno) {
4140 /* Find the block for the new target bucket */
4141 src_blkno = new_blkno +
4142 (bucket_blkno(target) - src_blkno);
4144 ocfs2_xattr_bucket_relse(first);
4145 ocfs2_xattr_bucket_relse(target);
4148 * These shouldn't fail - the buffers are in the
4149 * journal from ocfs2_cp_xattr_bucket().
4151 ret = ocfs2_read_xattr_bucket(first, new_blkno);
4156 ret = ocfs2_read_xattr_bucket(target, src_blkno);
4167 * Find the suitable pos when we divide a bucket into 2.
4168 * We have to make sure the xattrs with the same hash value exist
4169 * in the same bucket.
4171 * If this ocfs2_xattr_header covers more than one hash value, find a
4172 * place where the hash value changes. Try to find the most even split.
4173 * The most common case is that all entries have different hash values,
4174 * and the first check we make will find a place to split.
4176 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
4178 struct ocfs2_xattr_entry *entries = xh->xh_entries;
4179 int count = le16_to_cpu(xh->xh_count);
4180 int delta, middle = count / 2;
4183 * We start at the middle. Each step gets farther away in both
4184 * directions. We therefore hit the change in hash value
4185 * nearest to the middle. Note that this loop does not execute for
4188 for (delta = 0; delta < middle; delta++) {
4189 /* Let's check delta earlier than middle */
4190 if (cmp_xe(&entries[middle - delta - 1],
4191 &entries[middle - delta]))
4192 return middle - delta;
4194 /* For even counts, don't walk off the end */
4195 if ((middle + delta + 1) == count)
4198 /* Now try delta past middle */
4199 if (cmp_xe(&entries[middle + delta],
4200 &entries[middle + delta + 1]))
4201 return middle + delta + 1;
4204 /* Every entry had the same hash */
4209 * Move some xattrs in old bucket(blk) to new bucket(new_blk).
4210 * first_hash will record the 1st hash of the new bucket.
4212 * Normally half of the xattrs will be moved. But we have to make
4213 * sure that the xattrs with the same hash value are stored in the
4214 * same bucket. If all the xattrs in this bucket have the same hash
4215 * value, the new bucket will be initialized as an empty one and the
4216 * first_hash will be initialized as (hash_value+1).
4218 static int ocfs2_divide_xattr_bucket(struct inode *inode,
4223 int new_bucket_head)
4226 int count, start, len, name_value_len = 0, xe_len, name_offset = 0;
4227 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4228 struct ocfs2_xattr_header *xh;
4229 struct ocfs2_xattr_entry *xe;
4230 int blocksize = inode->i_sb->s_blocksize;
4232 mlog(0, "move some of xattrs from bucket %llu to %llu\n",
4233 (unsigned long long)blk, (unsigned long long)new_blk);
4235 s_bucket = ocfs2_xattr_bucket_new(inode);
4236 t_bucket = ocfs2_xattr_bucket_new(inode);
4237 if (!s_bucket || !t_bucket) {
4243 ret = ocfs2_read_xattr_bucket(s_bucket, blk);
4249 ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
4250 OCFS2_JOURNAL_ACCESS_WRITE);
4257 * Even if !new_bucket_head, we're overwriting t_bucket. Thus,
4258 * there's no need to read it.
4260 ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
4267 * Hey, if we're overwriting t_bucket, what difference does
4268 * ACCESS_CREATE vs ACCESS_WRITE make? See the comment in the
4269 * same part of ocfs2_cp_xattr_bucket().
4271 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4273 OCFS2_JOURNAL_ACCESS_CREATE :
4274 OCFS2_JOURNAL_ACCESS_WRITE);
4280 xh = bucket_xh(s_bucket);
4281 count = le16_to_cpu(xh->xh_count);
4282 start = ocfs2_xattr_find_divide_pos(xh);
4284 if (start == count) {
4285 xe = &xh->xh_entries[start-1];
4288 * initialized a new empty bucket here.
4289 * The hash value is set as one larger than
4290 * that of the last entry in the previous bucket.
4292 for (i = 0; i < t_bucket->bu_blocks; i++)
4293 memset(bucket_block(t_bucket, i), 0, blocksize);
4295 xh = bucket_xh(t_bucket);
4296 xh->xh_free_start = cpu_to_le16(blocksize);
4297 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
4298 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
4300 goto set_num_buckets;
4303 /* copy the whole bucket to the new first. */
4304 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4306 /* update the new bucket. */
4307 xh = bucket_xh(t_bucket);
4310 * Calculate the total name/value len and xh_free_start for
4311 * the old bucket first.
4313 name_offset = OCFS2_XATTR_BUCKET_SIZE;
4315 for (i = 0; i < start; i++) {
4316 xe = &xh->xh_entries[i];
4317 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
4318 if (ocfs2_xattr_is_local(xe))
4320 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4322 xe_len += OCFS2_XATTR_ROOT_SIZE;
4323 name_value_len += xe_len;
4324 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
4325 name_offset = le16_to_cpu(xe->xe_name_offset);
4329 * Now begin the modification to the new bucket.
4331 * In the new bucket, We just move the xattr entry to the beginning
4332 * and don't touch the name/value. So there will be some holes in the
4333 * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
4336 xe = &xh->xh_entries[start];
4337 len = sizeof(struct ocfs2_xattr_entry) * (count - start);
4338 mlog(0, "mv xattr entry len %d from %d to %d\n", len,
4339 (int)((char *)xe - (char *)xh),
4340 (int)((char *)xh->xh_entries - (char *)xh));
4341 memmove((char *)xh->xh_entries, (char *)xe, len);
4342 xe = &xh->xh_entries[count - start];
4343 len = sizeof(struct ocfs2_xattr_entry) * start;
4344 memset((char *)xe, 0, len);
4346 le16_add_cpu(&xh->xh_count, -start);
4347 le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
4349 /* Calculate xh_free_start for the new bucket. */
4350 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4351 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4352 xe = &xh->xh_entries[i];
4353 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
4354 if (ocfs2_xattr_is_local(xe))
4356 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4358 xe_len += OCFS2_XATTR_ROOT_SIZE;
4359 if (le16_to_cpu(xe->xe_name_offset) <
4360 le16_to_cpu(xh->xh_free_start))
4361 xh->xh_free_start = xe->xe_name_offset;
4365 /* set xh->xh_num_buckets for the new xh. */
4366 if (new_bucket_head)
4367 xh->xh_num_buckets = cpu_to_le16(1);
4369 xh->xh_num_buckets = 0;
4371 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4373 /* store the first_hash of the new bucket. */
4375 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
4378 * Now only update the 1st block of the old bucket. If we
4379 * just added a new empty bucket, there is no need to modify
4385 xh = bucket_xh(s_bucket);
4386 memset(&xh->xh_entries[start], 0,
4387 sizeof(struct ocfs2_xattr_entry) * (count - start));
4388 xh->xh_count = cpu_to_le16(start);
4389 xh->xh_free_start = cpu_to_le16(name_offset);
4390 xh->xh_name_value_len = cpu_to_le16(name_value_len);
4392 ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
4395 ocfs2_xattr_bucket_free(s_bucket);
4396 ocfs2_xattr_bucket_free(t_bucket);
4402 * Copy xattr from one bucket to another bucket.
4404 * The caller must make sure that the journal transaction
4405 * has enough space for journaling.
4407 static int ocfs2_cp_xattr_bucket(struct inode *inode,
4414 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4416 BUG_ON(s_blkno == t_blkno);
4418 mlog(0, "cp bucket %llu to %llu, target is %d\n",
4419 (unsigned long long)s_blkno, (unsigned long long)t_blkno,
4422 s_bucket = ocfs2_xattr_bucket_new(inode);
4423 t_bucket = ocfs2_xattr_bucket_new(inode);
4424 if (!s_bucket || !t_bucket) {
4430 ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
4435 * Even if !t_is_new, we're overwriting t_bucket. Thus,
4436 * there's no need to read it.
4438 ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
4443 * Hey, if we're overwriting t_bucket, what difference does
4444 * ACCESS_CREATE vs ACCESS_WRITE make? Well, if we allocated a new
4445 * cluster to fill, we came here from
4446 * ocfs2_mv_xattr_buckets(), and it is really new -
4447 * ACCESS_CREATE is required. But we also might have moved data
4448 * out of t_bucket before extending back into it.
4449 * ocfs2_add_new_xattr_bucket() can do this - its call to
4450 * ocfs2_add_new_xattr_cluster() may have created a new extent
4451 * and copied out the end of the old extent. Then it re-extends
4452 * the old extent back to create space for new xattrs. That's
4453 * how we get here, and the bucket isn't really new.
4455 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4457 OCFS2_JOURNAL_ACCESS_CREATE :
4458 OCFS2_JOURNAL_ACCESS_WRITE);
4462 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4463 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4466 ocfs2_xattr_bucket_free(t_bucket);
4467 ocfs2_xattr_bucket_free(s_bucket);
4473 * src_blk points to the start of an existing extent. last_blk points to
4474 * last cluster in that extent. to_blk points to a newly allocated
4475 * extent. We copy the buckets from the cluster at last_blk to the new
4476 * extent. If start_bucket is non-zero, we skip that many buckets before
4477 * we start copying. The new extent's xh_num_buckets gets set to the
4478 * number of buckets we copied. The old extent's xh_num_buckets shrinks
4479 * by the same amount.
4481 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4482 u64 src_blk, u64 last_blk, u64 to_blk,
4483 unsigned int start_bucket,
4486 int i, ret, credits;
4487 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4488 int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4489 int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4490 struct ocfs2_xattr_bucket *old_first, *new_first;
4492 mlog(0, "mv xattrs from cluster %llu to %llu\n",
4493 (unsigned long long)last_blk, (unsigned long long)to_blk);
4495 BUG_ON(start_bucket >= num_buckets);
4497 num_buckets -= start_bucket;
4498 last_blk += (start_bucket * blks_per_bucket);
4501 /* The first bucket of the original extent */
4502 old_first = ocfs2_xattr_bucket_new(inode);
4503 /* The first bucket of the new extent */
4504 new_first = ocfs2_xattr_bucket_new(inode);
4505 if (!old_first || !new_first) {
4511 ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4518 * We need to update the first bucket of the old extent and all
4519 * the buckets going to the new extent.
4521 credits = ((num_buckets + 1) * blks_per_bucket) +
4522 handle->h_buffer_credits;
4523 ret = ocfs2_extend_trans(handle, credits);
4529 ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4530 OCFS2_JOURNAL_ACCESS_WRITE);
4536 for (i = 0; i < num_buckets; i++) {
4537 ret = ocfs2_cp_xattr_bucket(inode, handle,
4538 last_blk + (i * blks_per_bucket),
4539 to_blk + (i * blks_per_bucket),
4548 * Get the new bucket ready before we dirty anything
4549 * (This actually shouldn't fail, because we already dirtied
4550 * it once in ocfs2_cp_xattr_bucket()).
4552 ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4557 ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4558 OCFS2_JOURNAL_ACCESS_WRITE);
4564 /* Now update the headers */
4565 le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4566 ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4568 bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4569 ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4572 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4575 ocfs2_xattr_bucket_free(new_first);
4576 ocfs2_xattr_bucket_free(old_first);
4581 * Move some xattrs in this cluster to the new cluster.
4582 * This function should only be called when bucket size == cluster size.
4583 * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4585 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4591 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4592 int ret, credits = 2 * blk_per_bucket + handle->h_buffer_credits;
4594 BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4596 ret = ocfs2_extend_trans(handle, credits);
4602 /* Move half of the xattr in start_blk to the next bucket. */
4603 return ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4604 new_blk, first_hash, 1);
4608 * Move some xattrs from the old cluster to the new one since they are not
4609 * contiguous in ocfs2 xattr tree.
4611 * new_blk starts a new separate cluster, and we will move some xattrs from
4612 * prev_blk to it. v_start will be set as the first name hash value in this
4613 * new cluster so that it can be used as e_cpos during tree insertion and
4614 * don't collide with our original b-tree operations. first_bh and header_bh
4615 * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4616 * to extend the insert bucket.
4618 * The problem is how much xattr should we move to the new one and when should
4619 * we update first_bh and header_bh?
4620 * 1. If cluster size > bucket size, that means the previous cluster has more
4621 * than 1 bucket, so just move half nums of bucket into the new cluster and
4622 * update the first_bh and header_bh if the insert bucket has been moved
4623 * to the new cluster.
4624 * 2. If cluster_size == bucket_size:
4625 * a) If the previous extent rec has more than one cluster and the insert
4626 * place isn't in the last cluster, copy the entire last cluster to the
4627 * new one. This time, we don't need to upate the first_bh and header_bh
4628 * since they will not be moved into the new cluster.
4629 * b) Otherwise, move the bottom half of the xattrs in the last cluster into
4630 * the new one. And we set the extend flag to zero if the insert place is
4631 * moved into the new allocated cluster since no extend is needed.
4633 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
4635 struct ocfs2_xattr_bucket *first,
4636 struct ocfs2_xattr_bucket *target,
4644 mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
4645 (unsigned long long)bucket_blkno(first), prev_clusters,
4646 (unsigned long long)new_blk);
4648 if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
4649 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
4658 /* The start of the last cluster in the first extent */
4659 u64 last_blk = bucket_blkno(first) +
4660 ((prev_clusters - 1) *
4661 ocfs2_clusters_to_blocks(inode->i_sb, 1));
4663 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
4664 ret = ocfs2_mv_xattr_buckets(inode, handle,
4665 bucket_blkno(first),
4666 last_blk, new_blk, 0,
4671 ret = ocfs2_divide_xattr_cluster(inode, handle,
4677 if ((bucket_blkno(target) == last_blk) && extend)
4686 * Add a new cluster for xattr storage.
4688 * If the new cluster is contiguous with the previous one, it will be
4689 * appended to the same extent record, and num_clusters will be updated.
4690 * If not, we will insert a new extent for it and move some xattrs in
4691 * the last cluster into the new allocated one.
4692 * We also need to limit the maximum size of a btree leaf, otherwise we'll
4693 * lose the benefits of hashing because we'll have to search large leaves.
4694 * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
4697 * first_bh is the first block of the previous extent rec and header_bh
4698 * indicates the bucket we will insert the new xattrs. They will be updated
4699 * when the header_bh is moved into the new cluster.
4701 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
4702 struct buffer_head *root_bh,
4703 struct ocfs2_xattr_bucket *first,
4704 struct ocfs2_xattr_bucket *target,
4708 struct ocfs2_xattr_set_ctxt *ctxt)
4711 u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
4712 u32 prev_clusters = *num_clusters;
4713 u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
4715 handle_t *handle = ctxt->handle;
4716 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4717 struct ocfs2_extent_tree et;
4719 mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
4720 "previous xattr blkno = %llu\n",
4721 (unsigned long long)OCFS2_I(inode)->ip_blkno,
4722 prev_cpos, (unsigned long long)bucket_blkno(first));
4724 ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
4726 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
4727 OCFS2_JOURNAL_ACCESS_WRITE);
4733 ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac, 1,
4734 clusters_to_add, &bit_off, &num_bits);
4741 BUG_ON(num_bits > clusters_to_add);
4743 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
4744 mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
4745 num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
4747 if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
4748 (prev_clusters + num_bits) << osb->s_clustersize_bits <=
4749 OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
4751 * If this cluster is contiguous with the old one and
4752 * adding this new cluster, we don't surpass the limit of
4753 * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
4754 * initialized and used like other buckets in the previous
4756 * So add it as a contiguous one. The caller will handle
4759 v_start = prev_cpos + prev_clusters;
4760 *num_clusters = prev_clusters + num_bits;
4761 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
4764 ret = ocfs2_adjust_xattr_cross_cluster(inode,
4778 mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
4779 num_bits, (unsigned long long)block, v_start);
4780 ret = ocfs2_insert_extent(handle, &et, v_start, block,
4781 num_bits, 0, ctxt->meta_ac);
4787 ret = ocfs2_journal_dirty(handle, root_bh);
4796 * We are given an extent. 'first' is the bucket at the very front of
4797 * the extent. The extent has space for an additional bucket past
4798 * bucket_xh(first)->xh_num_buckets. 'target_blkno' is the block number
4799 * of the target bucket. We wish to shift every bucket past the target
4800 * down one, filling in that additional space. When we get back to the
4801 * target, we split the target between itself and the now-empty bucket
4802 * at target+1 (aka, target_blkno + blks_per_bucket).
4804 static int ocfs2_extend_xattr_bucket(struct inode *inode,
4806 struct ocfs2_xattr_bucket *first,
4811 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4812 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4814 u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
4816 mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
4817 "from %llu, len = %u\n", (unsigned long long)target_blk,
4818 (unsigned long long)bucket_blkno(first), num_clusters);
4820 /* The extent must have room for an additional bucket */
4821 BUG_ON(new_bucket >=
4822 (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
4824 /* end_blk points to the last existing bucket */
4825 end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
4828 * end_blk is the start of the last existing bucket.
4829 * Thus, (end_blk - target_blk) covers the target bucket and
4830 * every bucket after it up to, but not including, the last
4831 * existing bucket. Then we add the last existing bucket, the
4832 * new bucket, and the first bucket (3 * blk_per_bucket).
4834 credits = (end_blk - target_blk) + (3 * blk_per_bucket) +
4835 handle->h_buffer_credits;
4836 ret = ocfs2_extend_trans(handle, credits);
4842 ret = ocfs2_xattr_bucket_journal_access(handle, first,
4843 OCFS2_JOURNAL_ACCESS_WRITE);
4849 while (end_blk != target_blk) {
4850 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
4851 end_blk + blk_per_bucket, 0);
4854 end_blk -= blk_per_bucket;
4857 /* Move half of the xattr in target_blkno to the next bucket. */
4858 ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
4859 target_blk + blk_per_bucket, NULL, 0);
4861 le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
4862 ocfs2_xattr_bucket_journal_dirty(handle, first);
4869 * Add new xattr bucket in an extent record and adjust the buckets
4870 * accordingly. xb_bh is the ocfs2_xattr_block, and target is the
4871 * bucket we want to insert into.
4873 * In the easy case, we will move all the buckets after target down by
4874 * one. Half of target's xattrs will be moved to the next bucket.
4876 * If current cluster is full, we'll allocate a new one. This may not
4877 * be contiguous. The underlying calls will make sure that there is
4878 * space for the insert, shifting buckets around if necessary.
4879 * 'target' may be moved by those calls.
4881 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
4882 struct buffer_head *xb_bh,
4883 struct ocfs2_xattr_bucket *target,
4884 struct ocfs2_xattr_set_ctxt *ctxt)
4886 struct ocfs2_xattr_block *xb =
4887 (struct ocfs2_xattr_block *)xb_bh->b_data;
4888 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
4889 struct ocfs2_extent_list *el = &xb_root->xt_list;
4891 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
4892 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4893 int ret, num_buckets, extend = 1;
4895 u32 e_cpos, num_clusters;
4896 /* The bucket at the front of the extent */
4897 struct ocfs2_xattr_bucket *first;
4899 mlog(0, "Add new xattr bucket starting from %llu\n",
4900 (unsigned long long)bucket_blkno(target));
4902 /* The first bucket of the original extent */
4903 first = ocfs2_xattr_bucket_new(inode);
4910 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
4917 ret = ocfs2_read_xattr_bucket(first, p_blkno);
4923 num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
4924 if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
4926 * This can move first+target if the target bucket moves
4927 * to the new extent.
4929 ret = ocfs2_add_new_xattr_cluster(inode,
4944 ret = ocfs2_extend_xattr_bucket(inode,
4947 bucket_blkno(target),
4954 ocfs2_xattr_bucket_free(first);
4959 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
4960 struct ocfs2_xattr_bucket *bucket,
4963 int block_off = offs >> inode->i_sb->s_blocksize_bits;
4965 offs = offs % inode->i_sb->s_blocksize;
4966 return bucket_block(bucket, block_off) + offs;
4970 * Handle the normal xattr set, including replace, delete and new.
4972 * Note: "local" indicates the real data's locality. So we can't
4973 * just its bucket locality by its length.
4975 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
4976 struct ocfs2_xattr_info *xi,
4977 struct ocfs2_xattr_search *xs,
4981 struct ocfs2_xattr_entry *last, *xe;
4982 int name_len = strlen(xi->name);
4983 struct ocfs2_xattr_header *xh = xs->header;
4984 u16 count = le16_to_cpu(xh->xh_count), start;
4985 size_t blocksize = inode->i_sb->s_blocksize;
4987 size_t offs, size, new_size;
4988 struct ocfs2_xa_loc loc;
4990 ocfs2_init_xattr_bucket_xa_loc(&loc, xs->bucket,
4991 xs->not_found ? NULL : xs->here);
4992 last = &xh->xh_entries[count];
4993 if (!xs->not_found) {
4995 offs = le16_to_cpu(xe->xe_name_offset);
4996 if (ocfs2_xattr_is_local(xe))
4997 size = OCFS2_XATTR_SIZE(name_len) +
4998 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
5000 size = OCFS2_XATTR_SIZE(name_len) +
5001 OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
5004 * If the new value will be stored outside, xi->value has been
5005 * initalized as an empty ocfs2_xattr_value_root, and the same
5006 * goes with xi->value_len, so we can set new_size safely here.
5007 * See ocfs2_xattr_set_in_bucket.
5009 new_size = OCFS2_XATTR_SIZE(name_len) +
5010 OCFS2_XATTR_SIZE(xi->value_len);
5013 ocfs2_xa_wipe_namevalue(&loc);
5014 if (new_size > size)
5015 goto set_new_name_value;
5017 /* Now replace the old value with new one. */
5019 xe->xe_value_size = cpu_to_le64(xi->value_len);
5021 xe->xe_value_size = 0;
5023 val = ocfs2_xattr_bucket_get_val(inode,
5025 memset(val + OCFS2_XATTR_SIZE(name_len), 0,
5026 size - OCFS2_XATTR_SIZE(name_len));
5027 if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
5028 memcpy(val + OCFS2_XATTR_SIZE(name_len),
5029 xi->value, xi->value_len);
5031 le16_add_cpu(&xh->xh_name_value_len, new_size);
5032 ocfs2_xattr_set_local(xe, local);
5035 ocfs2_xa_remove_entry(&loc);
5038 cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
5043 /* find a new entry for insert. */
5044 int low = 0, high = count - 1, tmp;
5045 struct ocfs2_xattr_entry *tmp_xe;
5047 while (low <= high && count) {
5048 tmp = (low + high) / 2;
5049 tmp_xe = &xh->xh_entries[tmp];
5051 if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
5053 else if (name_hash <
5054 le32_to_cpu(tmp_xe->xe_name_hash))
5062 xe = &xh->xh_entries[low];
5064 memmove(xe + 1, xe, (void *)last - (void *)xe);
5066 le16_add_cpu(&xh->xh_count, 1);
5067 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
5068 xe->xe_name_hash = cpu_to_le32(name_hash);
5069 xe->xe_name_len = name_len;
5070 ocfs2_xattr_set_type(xe, xi->name_index);
5074 /* Insert the new name+value. */
5075 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
5078 * We must make sure that the name/value pair
5079 * exists in the same block.
5081 offs = le16_to_cpu(xh->xh_free_start);
5082 start = offs - size;
5084 if (start >> inode->i_sb->s_blocksize_bits !=
5085 (offs - 1) >> inode->i_sb->s_blocksize_bits) {
5086 offs = offs - offs % blocksize;
5087 xh->xh_free_start = cpu_to_le16(offs);
5090 val = ocfs2_xattr_bucket_get_val(inode, xs->bucket, offs - size);
5091 xe->xe_name_offset = cpu_to_le16(offs - size);
5093 memset(val, 0, size);
5094 memcpy(val, xi->name, name_len);
5095 memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
5097 xe->xe_value_size = cpu_to_le64(xi->value_len);
5098 ocfs2_xattr_set_local(xe, local);
5100 le16_add_cpu(&xh->xh_free_start, -size);
5101 le16_add_cpu(&xh->xh_name_value_len, size);
5107 * Set the xattr entry in the specified bucket.
5108 * The bucket is indicated by xs->bucket and it should have the enough
5109 * space for the xattr insertion.
5111 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
5113 struct ocfs2_xattr_info *xi,
5114 struct ocfs2_xattr_search *xs,
5121 mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
5122 (unsigned long)xi->value_len, xi->name_index,
5123 (unsigned long long)bucket_blkno(xs->bucket));
5125 if (!xs->bucket->bu_bhs[1]) {
5126 blkno = bucket_blkno(xs->bucket);
5127 ocfs2_xattr_bucket_relse(xs->bucket);
5128 ret = ocfs2_read_xattr_bucket(xs->bucket, blkno);
5135 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
5136 OCFS2_JOURNAL_ACCESS_WRITE);
5142 ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local);
5143 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
5150 * Truncate the specified xe_off entry in xattr bucket.
5151 * bucket is indicated by header_bh and len is the new length.
5152 * Both the ocfs2_xattr_value_root and the entry will be updated here.
5154 * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
5156 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
5157 struct ocfs2_xattr_bucket *bucket,
5160 struct ocfs2_xattr_set_ctxt *ctxt)
5164 struct ocfs2_xattr_entry *xe;
5165 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5166 size_t blocksize = inode->i_sb->s_blocksize;
5167 struct ocfs2_xattr_value_buf vb = {
5168 .vb_access = ocfs2_journal_access,
5171 xe = &xh->xh_entries[xe_off];
5173 BUG_ON(!xe || ocfs2_xattr_is_local(xe));
5175 offset = le16_to_cpu(xe->xe_name_offset) +
5176 OCFS2_XATTR_SIZE(xe->xe_name_len);
5178 value_blk = offset / blocksize;
5180 /* We don't allow ocfs2_xattr_value to be stored in different block. */
5181 BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
5183 vb.vb_bh = bucket->bu_bhs[value_blk];
5186 vb.vb_xv = (struct ocfs2_xattr_value_root *)
5187 (vb.vb_bh->b_data + offset % blocksize);
5190 * From here on out we have to dirty the bucket. The generic
5191 * value calls only modify one of the bucket's bhs, but we need
5192 * to send the bucket at once. So if they error, they *could* have
5193 * modified something. We have to assume they did, and dirty
5194 * the whole bucket. This leaves us in a consistent state.
5196 mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
5197 xe_off, (unsigned long long)bucket_blkno(bucket), len);
5198 ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
5204 ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
5205 OCFS2_JOURNAL_ACCESS_WRITE);
5211 xe->xe_value_size = cpu_to_le64(len);
5213 ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
5219 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
5220 struct ocfs2_xattr_search *xs,
5222 struct ocfs2_xattr_set_ctxt *ctxt)
5225 struct ocfs2_xattr_entry *xe = xs->here;
5226 struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
5228 BUG_ON(!xs->bucket->bu_bhs[0] || !xe || ocfs2_xattr_is_local(xe));
5230 offset = xe - xh->xh_entries;
5231 ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket,
5239 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
5241 struct ocfs2_xattr_search *xs,
5245 int ret, offset, block_off;
5246 struct ocfs2_xattr_value_root *xv;
5247 struct ocfs2_xattr_entry *xe = xs->here;
5248 struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
5250 struct ocfs2_xattr_value_buf vb = {
5251 .vb_access = ocfs2_journal_access,
5254 BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
5256 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb, xh,
5257 xe - xh->xh_entries,
5265 base = bucket_block(xs->bucket, block_off);
5266 xv = (struct ocfs2_xattr_value_root *)(base + offset +
5267 OCFS2_XATTR_SIZE(xe->xe_name_len));
5270 vb.vb_bh = xs->bucket->bu_bhs[block_off];
5271 ret = __ocfs2_xattr_set_value_outside(inode, handle,
5272 &vb, val, value_len);
5279 static int ocfs2_rm_xattr_cluster(struct inode *inode,
5280 struct buffer_head *root_bh,
5287 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5288 struct inode *tl_inode = osb->osb_tl_inode;
5290 struct ocfs2_xattr_block *xb =
5291 (struct ocfs2_xattr_block *)root_bh->b_data;
5292 struct ocfs2_alloc_context *meta_ac = NULL;
5293 struct ocfs2_cached_dealloc_ctxt dealloc;
5294 struct ocfs2_extent_tree et;
5296 ret = ocfs2_iterate_xattr_buckets(inode, blkno, len,
5297 ocfs2_delete_xattr_in_bucket, para);
5303 ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5305 ocfs2_init_dealloc_ctxt(&dealloc);
5307 mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
5308 cpos, len, (unsigned long long)blkno);
5310 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode), blkno,
5313 ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
5319 mutex_lock(&tl_inode->i_mutex);
5321 if (ocfs2_truncate_log_needs_flush(osb)) {
5322 ret = __ocfs2_flush_truncate_log(osb);
5329 handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
5330 if (IS_ERR(handle)) {
5336 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5337 OCFS2_JOURNAL_ACCESS_WRITE);
5343 ret = ocfs2_remove_extent(handle, &et, cpos, len, meta_ac,
5350 le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
5352 ret = ocfs2_journal_dirty(handle, root_bh);
5358 ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
5363 ocfs2_commit_trans(osb, handle);
5365 ocfs2_schedule_truncate_log_flush(osb, 1);
5367 mutex_unlock(&tl_inode->i_mutex);
5370 ocfs2_free_alloc_context(meta_ac);
5372 ocfs2_run_deallocs(osb, &dealloc);
5377 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
5379 struct ocfs2_xattr_search *xs)
5381 struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
5382 struct ocfs2_xattr_entry *last = &xh->xh_entries[
5383 le16_to_cpu(xh->xh_count) - 1];
5386 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
5387 OCFS2_JOURNAL_ACCESS_WRITE);
5393 /* Remove the old entry. */
5394 memmove(xs->here, xs->here + 1,
5395 (void *)last - (void *)xs->here);
5396 memset(last, 0, sizeof(struct ocfs2_xattr_entry));
5397 le16_add_cpu(&xh->xh_count, -1);
5399 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
5403 * Set the xattr name/value in the bucket specified in xs.
5405 * As the new value in xi may be stored in the bucket or in an outside cluster,
5406 * we divide the whole process into 3 steps:
5407 * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
5408 * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
5409 * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
5410 * 4. If the clusters for the new outside value can't be allocated, we need
5411 * to free the xattr we allocated in set.
5413 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
5414 struct ocfs2_xattr_info *xi,
5415 struct ocfs2_xattr_search *xs,
5416 struct ocfs2_xattr_set_ctxt *ctxt)
5420 char *val = (char *)xi->value;
5421 struct ocfs2_xattr_entry *xe = xs->here;
5422 u32 name_hash = ocfs2_xattr_name_hash(inode, xi->name,
5425 if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
5427 * We need to truncate the xattr storage first.
5429 * If both the old and new value are stored to
5430 * outside block, we only need to truncate
5431 * the storage and then set the value outside.
5433 * If the new value should be stored within block,
5434 * we should free all the outside block first and
5435 * the modification to the xattr block will be done
5436 * by following steps.
5438 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
5439 value_len = xi->value_len;
5443 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
5450 goto set_value_outside;
5453 value_len = xi->value_len;
5454 /* So we have to handle the inside block change now. */
5455 if (value_len > OCFS2_XATTR_INLINE_SIZE) {
5457 * If the new value will be stored outside of block,
5458 * initalize a new empty value root and insert it first.
5461 xi->value = &def_xv;
5462 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
5465 ret = ocfs2_xattr_set_entry_in_bucket(inode, ctxt->handle, xi, xs,
5472 if (value_len <= OCFS2_XATTR_INLINE_SIZE)
5475 /* allocate the space now for the outside block storage. */
5476 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
5481 if (xs->not_found) {
5483 * We can't allocate enough clusters for outside
5484 * storage and we have allocated xattr already,
5485 * so need to remove it.
5487 ocfs2_xattr_bucket_remove_xs(inode, ctxt->handle, xs);
5493 ret = ocfs2_xattr_bucket_set_value_outside(inode, ctxt->handle,
5494 xs, val, value_len);
5500 * check whether the xattr bucket is filled up with the same hash value.
5501 * If we want to insert the xattr with the same hash, return -ENOSPC.
5502 * If we want to insert a xattr with different hash value, go ahead
5503 * and ocfs2_divide_xattr_bucket will handle this.
5505 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
5506 struct ocfs2_xattr_bucket *bucket,
5509 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5510 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5512 if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5515 if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5516 xh->xh_entries[0].xe_name_hash) {
5517 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5519 (unsigned long long)bucket_blkno(bucket),
5520 le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5527 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5528 struct ocfs2_xattr_info *xi,
5529 struct ocfs2_xattr_search *xs,
5530 struct ocfs2_xattr_set_ctxt *ctxt)
5532 struct ocfs2_xattr_header *xh;
5533 struct ocfs2_xattr_entry *xe;
5534 u16 count, header_size, xh_free_start;
5535 int free, max_free, need, old;
5536 size_t value_size = 0, name_len = strlen(xi->name);
5537 size_t blocksize = inode->i_sb->s_blocksize;
5538 int ret, allocation = 0;
5540 mlog_entry("Set xattr %s in xattr index block\n", xi->name);
5544 count = le16_to_cpu(xh->xh_count);
5545 xh_free_start = le16_to_cpu(xh->xh_free_start);
5546 header_size = sizeof(struct ocfs2_xattr_header) +
5547 count * sizeof(struct ocfs2_xattr_entry);
5548 max_free = OCFS2_XATTR_BUCKET_SIZE - header_size -
5549 le16_to_cpu(xh->xh_name_value_len) - OCFS2_XATTR_HEADER_GAP;
5551 mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
5552 "of %u which exceed block size\n",
5553 (unsigned long long)bucket_blkno(xs->bucket),
5556 if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
5557 value_size = OCFS2_XATTR_ROOT_SIZE;
5559 value_size = OCFS2_XATTR_SIZE(xi->value_len);
5562 need = sizeof(struct ocfs2_xattr_entry) +
5563 OCFS2_XATTR_SIZE(name_len) + value_size;
5565 need = value_size + OCFS2_XATTR_SIZE(name_len);
5568 * We only replace the old value if the new length is smaller
5569 * than the old one. Otherwise we will allocate new space in the
5570 * bucket to store it.
5573 if (ocfs2_xattr_is_local(xe))
5574 old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
5576 old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
5578 if (old >= value_size)
5582 free = xh_free_start - header_size - OCFS2_XATTR_HEADER_GAP;
5584 * We need to make sure the new name/value pair
5585 * can exist in the same block.
5587 if (xh_free_start % blocksize < need)
5588 free -= xh_free_start % blocksize;
5590 mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
5591 "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
5592 " %u\n", xs->not_found,
5593 (unsigned long long)bucket_blkno(xs->bucket),
5594 free, need, max_free, le16_to_cpu(xh->xh_free_start),
5595 le16_to_cpu(xh->xh_name_value_len));
5599 count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb))) {
5600 if (need <= max_free &&
5601 count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
5603 * We can create the space by defragment. Since only the
5604 * name/value will be moved, the xe shouldn't be changed
5607 ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5614 xh_free_start = le16_to_cpu(xh->xh_free_start);
5615 free = xh_free_start - header_size
5616 - OCFS2_XATTR_HEADER_GAP;
5617 if (xh_free_start % blocksize < need)
5618 free -= xh_free_start % blocksize;
5623 mlog(0, "Can't get enough space for xattr insert by "
5624 "defragment. Need %u bytes, but we have %d, so "
5625 "allocate new bucket for it.\n", need, free);
5629 * We have to add new buckets or clusters and one
5630 * allocation should leave us enough space for insert.
5635 * We do not allow for overlapping ranges between buckets. And
5636 * the maximum number of collisions we will allow for then is
5637 * one bucket's worth, so check it here whether we need to
5638 * add a new bucket for the insert.
5640 ret = ocfs2_check_xattr_bucket_collision(inode,
5648 ret = ocfs2_add_new_xattr_bucket(inode,
5658 * ocfs2_add_new_xattr_bucket() will have updated
5659 * xs->bucket if it moved, but it will not have updated
5660 * any of the other search fields. Thus, we drop it and
5661 * re-search. Everything should be cached, so it'll be
5664 ocfs2_xattr_bucket_relse(xs->bucket);
5665 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5668 if (ret && ret != -ENODATA)
5670 xs->not_found = ret;
5676 ret = ocfs2_xattr_set_in_bucket(inode, xi, xs, ctxt);
5682 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5683 struct ocfs2_xattr_bucket *bucket,
5686 int ret = 0, ref_credits;
5687 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5689 struct ocfs2_xattr_entry *xe;
5690 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5691 struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5692 int credits = ocfs2_remove_extent_credits(osb->sb) +
5693 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5694 struct ocfs2_xattr_value_root *xv;
5695 struct ocfs2_rm_xattr_bucket_para *args =
5696 (struct ocfs2_rm_xattr_bucket_para *)para;
5698 ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5700 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5701 xe = &xh->xh_entries[i];
5702 if (ocfs2_xattr_is_local(xe))
5705 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket,
5708 ret = ocfs2_lock_xattr_remove_allocators(inode, xv,
5714 ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
5715 if (IS_ERR(ctxt.handle)) {
5716 ret = PTR_ERR(ctxt.handle);
5721 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5724 ocfs2_commit_trans(osb, ctxt.handle);
5726 ocfs2_free_alloc_context(ctxt.meta_ac);
5727 ctxt.meta_ac = NULL;
5736 ocfs2_free_alloc_context(ctxt.meta_ac);
5737 ocfs2_schedule_truncate_log_flush(osb, 1);
5738 ocfs2_run_deallocs(osb, &ctxt.dealloc);
5743 * Whenever we modify a xattr value root in the bucket(e.g, CoW
5744 * or change the extent record flag), we need to recalculate
5745 * the metaecc for the whole bucket. So it is done here.
5748 * We have to give the extra credits for the caller.
5750 static int ocfs2_xattr_bucket_post_refcount(struct inode *inode,
5755 struct ocfs2_xattr_bucket *bucket =
5756 (struct ocfs2_xattr_bucket *)para;
5758 ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
5759 OCFS2_JOURNAL_ACCESS_WRITE);
5765 ocfs2_xattr_bucket_journal_dirty(handle, bucket);
5771 * Special action we need if the xattr value is refcounted.
5773 * 1. If the xattr is refcounted, lock the tree.
5774 * 2. CoW the xattr if we are setting the new value and the value
5775 * will be stored outside.
5776 * 3. In other case, decrease_refcount will work for us, so just
5777 * lock the refcount tree, calculate the meta and credits is OK.
5779 * We have to do CoW before ocfs2_init_xattr_set_ctxt since
5780 * currently CoW is a completed transaction, while this function
5781 * will also lock the allocators and let us deadlock. So we will
5782 * CoW the whole xattr value.
5784 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
5785 struct ocfs2_dinode *di,
5786 struct ocfs2_xattr_info *xi,
5787 struct ocfs2_xattr_search *xis,
5788 struct ocfs2_xattr_search *xbs,
5789 struct ocfs2_refcount_tree **ref_tree,
5794 struct ocfs2_xattr_block *xb;
5795 struct ocfs2_xattr_entry *xe;
5797 u32 p_cluster, num_clusters;
5798 unsigned int ext_flags;
5799 int name_offset, name_len;
5800 struct ocfs2_xattr_value_buf vb;
5801 struct ocfs2_xattr_bucket *bucket = NULL;
5802 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5803 struct ocfs2_post_refcount refcount;
5804 struct ocfs2_post_refcount *p = NULL;
5805 struct buffer_head *ref_root_bh = NULL;
5807 if (!xis->not_found) {
5809 name_offset = le16_to_cpu(xe->xe_name_offset);
5810 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5812 vb.vb_bh = xis->inode_bh;
5813 vb.vb_access = ocfs2_journal_access_di;
5815 int i, block_off = 0;
5816 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
5818 name_offset = le16_to_cpu(xe->xe_name_offset);
5819 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5820 i = xbs->here - xbs->header->xh_entries;
5822 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
5823 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
5824 bucket_xh(xbs->bucket),
5831 base = bucket_block(xbs->bucket, block_off);
5832 vb.vb_bh = xbs->bucket->bu_bhs[block_off];
5833 vb.vb_access = ocfs2_journal_access;
5835 if (ocfs2_meta_ecc(osb)) {
5836 /*create parameters for ocfs2_post_refcount. */
5837 bucket = xbs->bucket;
5838 refcount.credits = bucket->bu_blocks;
5839 refcount.para = bucket;
5841 ocfs2_xattr_bucket_post_refcount;
5846 vb.vb_bh = xbs->xattr_bh;
5847 vb.vb_access = ocfs2_journal_access_xb;
5851 if (ocfs2_xattr_is_local(xe))
5854 vb.vb_xv = (struct ocfs2_xattr_value_root *)
5855 (base + name_offset + name_len);
5857 ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
5858 &num_clusters, &vb.vb_xv->xr_list,
5866 * We just need to check the 1st extent record, since we always
5867 * CoW the whole xattr. So there shouldn't be a xattr with
5868 * some REFCOUNT extent recs after the 1st one.
5870 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
5873 ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
5874 1, ref_tree, &ref_root_bh);
5881 * If we are deleting the xattr or the new size will be stored inside,
5882 * cool, leave it there, the xattr truncate process will remove them
5883 * for us(it still needs the refcount tree lock and the meta, credits).
5884 * And the worse case is that every cluster truncate will split the
5885 * refcount tree, and make the original extent become 3. So we will need
5886 * 2 * cluster more extent recs at most.
5888 if (!xi->value || xi->value_len <= OCFS2_XATTR_INLINE_SIZE) {
5890 ret = ocfs2_refcounted_xattr_delete_need(inode,
5891 &(*ref_tree)->rf_ci,
5892 ref_root_bh, vb.vb_xv,
5899 ret = ocfs2_refcount_cow_xattr(inode, di, &vb,
5900 *ref_tree, ref_root_bh, 0,
5901 le32_to_cpu(vb.vb_xv->xr_clusters), p);
5906 brelse(ref_root_bh);
5911 * Add the REFCOUNTED flags for all the extent rec in ocfs2_xattr_value_root.
5912 * The physical clusters will be added to refcount tree.
5914 static int ocfs2_xattr_value_attach_refcount(struct inode *inode,
5915 struct ocfs2_xattr_value_root *xv,
5916 struct ocfs2_extent_tree *value_et,
5917 struct ocfs2_caching_info *ref_ci,
5918 struct buffer_head *ref_root_bh,
5919 struct ocfs2_cached_dealloc_ctxt *dealloc,
5920 struct ocfs2_post_refcount *refcount)
5923 u32 clusters = le32_to_cpu(xv->xr_clusters);
5924 u32 cpos, p_cluster, num_clusters;
5925 struct ocfs2_extent_list *el = &xv->xr_list;
5926 unsigned int ext_flags;
5929 while (cpos < clusters) {
5930 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
5931 &num_clusters, el, &ext_flags);
5933 cpos += num_clusters;
5934 if ((ext_flags & OCFS2_EXT_REFCOUNTED))
5939 ret = ocfs2_add_refcount_flag(inode, value_et,
5940 ref_ci, ref_root_bh,
5941 cpos - num_clusters,
5942 p_cluster, num_clusters,
5954 * Given a normal ocfs2_xattr_header, refcount all the entries which
5955 * have value stored outside.
5956 * Used for xattrs stored in inode and ocfs2_xattr_block.
5958 static int ocfs2_xattr_attach_refcount_normal(struct inode *inode,
5959 struct ocfs2_xattr_value_buf *vb,
5960 struct ocfs2_xattr_header *header,
5961 struct ocfs2_caching_info *ref_ci,
5962 struct buffer_head *ref_root_bh,
5963 struct ocfs2_cached_dealloc_ctxt *dealloc)
5966 struct ocfs2_xattr_entry *xe;
5967 struct ocfs2_xattr_value_root *xv;
5968 struct ocfs2_extent_tree et;
5971 for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
5972 xe = &header->xh_entries[i];
5974 if (ocfs2_xattr_is_local(xe))
5977 xv = (struct ocfs2_xattr_value_root *)((void *)header +
5978 le16_to_cpu(xe->xe_name_offset) +
5979 OCFS2_XATTR_SIZE(xe->xe_name_len));
5982 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
5984 ret = ocfs2_xattr_value_attach_refcount(inode, xv, &et,
5985 ref_ci, ref_root_bh,
5996 static int ocfs2_xattr_inline_attach_refcount(struct inode *inode,
5997 struct buffer_head *fe_bh,
5998 struct ocfs2_caching_info *ref_ci,
5999 struct buffer_head *ref_root_bh,
6000 struct ocfs2_cached_dealloc_ctxt *dealloc)
6002 struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
6003 struct ocfs2_xattr_header *header = (struct ocfs2_xattr_header *)
6004 (fe_bh->b_data + inode->i_sb->s_blocksize -
6005 le16_to_cpu(di->i_xattr_inline_size));
6006 struct ocfs2_xattr_value_buf vb = {
6008 .vb_access = ocfs2_journal_access_di,
6011 return ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
6012 ref_ci, ref_root_bh, dealloc);
6015 struct ocfs2_xattr_tree_value_refcount_para {
6016 struct ocfs2_caching_info *ref_ci;
6017 struct buffer_head *ref_root_bh;
6018 struct ocfs2_cached_dealloc_ctxt *dealloc;
6021 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
6022 struct ocfs2_xattr_bucket *bucket,
6024 struct ocfs2_xattr_value_root **xv,
6025 struct buffer_head **bh)
6027 int ret, block_off, name_offset;
6028 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
6029 struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
6032 ret = ocfs2_xattr_bucket_get_name_value(sb,
6042 base = bucket_block(bucket, block_off);
6044 *xv = (struct ocfs2_xattr_value_root *)(base + name_offset +
6045 OCFS2_XATTR_SIZE(xe->xe_name_len));
6048 *bh = bucket->bu_bhs[block_off];
6054 * For a given xattr bucket, refcount all the entries which
6055 * have value stored outside.
6057 static int ocfs2_xattr_bucket_value_refcount(struct inode *inode,
6058 struct ocfs2_xattr_bucket *bucket,
6062 struct ocfs2_extent_tree et;
6063 struct ocfs2_xattr_tree_value_refcount_para *ref =
6064 (struct ocfs2_xattr_tree_value_refcount_para *)para;
6065 struct ocfs2_xattr_header *xh =
6066 (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6067 struct ocfs2_xattr_entry *xe;
6068 struct ocfs2_xattr_value_buf vb = {
6069 .vb_access = ocfs2_journal_access,
6071 struct ocfs2_post_refcount refcount = {
6072 .credits = bucket->bu_blocks,
6074 .func = ocfs2_xattr_bucket_post_refcount,
6076 struct ocfs2_post_refcount *p = NULL;
6078 /* We only need post_refcount if we support metaecc. */
6079 if (ocfs2_meta_ecc(OCFS2_SB(inode->i_sb)))
6082 mlog(0, "refcount bucket %llu, count = %u\n",
6083 (unsigned long long)bucket_blkno(bucket),
6084 le16_to_cpu(xh->xh_count));
6085 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6086 xe = &xh->xh_entries[i];
6088 if (ocfs2_xattr_is_local(xe))
6091 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket, i,
6092 &vb.vb_xv, &vb.vb_bh);
6098 ocfs2_init_xattr_value_extent_tree(&et,
6099 INODE_CACHE(inode), &vb);
6101 ret = ocfs2_xattr_value_attach_refcount(inode, vb.vb_xv,
6115 static int ocfs2_refcount_xattr_tree_rec(struct inode *inode,
6116 struct buffer_head *root_bh,
6117 u64 blkno, u32 cpos, u32 len, void *para)
6119 return ocfs2_iterate_xattr_buckets(inode, blkno, len,
6120 ocfs2_xattr_bucket_value_refcount,
6124 static int ocfs2_xattr_block_attach_refcount(struct inode *inode,
6125 struct buffer_head *blk_bh,
6126 struct ocfs2_caching_info *ref_ci,
6127 struct buffer_head *ref_root_bh,
6128 struct ocfs2_cached_dealloc_ctxt *dealloc)
6131 struct ocfs2_xattr_block *xb =
6132 (struct ocfs2_xattr_block *)blk_bh->b_data;
6134 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
6135 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
6136 struct ocfs2_xattr_value_buf vb = {
6138 .vb_access = ocfs2_journal_access_xb,
6141 ret = ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
6142 ref_ci, ref_root_bh,
6145 struct ocfs2_xattr_tree_value_refcount_para para = {
6147 .ref_root_bh = ref_root_bh,
6151 ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
6152 ocfs2_refcount_xattr_tree_rec,
6159 int ocfs2_xattr_attach_refcount_tree(struct inode *inode,
6160 struct buffer_head *fe_bh,
6161 struct ocfs2_caching_info *ref_ci,
6162 struct buffer_head *ref_root_bh,
6163 struct ocfs2_cached_dealloc_ctxt *dealloc)
6166 struct ocfs2_inode_info *oi = OCFS2_I(inode);
6167 struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
6168 struct buffer_head *blk_bh = NULL;
6170 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
6171 ret = ocfs2_xattr_inline_attach_refcount(inode, fe_bh,
6172 ref_ci, ref_root_bh,
6180 if (!di->i_xattr_loc)
6183 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
6190 ret = ocfs2_xattr_block_attach_refcount(inode, blk_bh, ref_ci,
6191 ref_root_bh, dealloc);
6201 typedef int (should_xattr_reflinked)(struct ocfs2_xattr_entry *xe);
6203 * Store the information we need in xattr reflink.
6204 * old_bh and new_bh are inode bh for the old and new inode.
6206 struct ocfs2_xattr_reflink {
6207 struct inode *old_inode;
6208 struct inode *new_inode;
6209 struct buffer_head *old_bh;
6210 struct buffer_head *new_bh;
6211 struct ocfs2_caching_info *ref_ci;
6212 struct buffer_head *ref_root_bh;
6213 struct ocfs2_cached_dealloc_ctxt *dealloc;
6214 should_xattr_reflinked *xattr_reflinked;
6218 * Given a xattr header and xe offset,
6219 * return the proper xv and the corresponding bh.
6220 * xattr in inode, block and xattr tree have different implementaions.
6222 typedef int (get_xattr_value_root)(struct super_block *sb,
6223 struct buffer_head *bh,
6224 struct ocfs2_xattr_header *xh,
6226 struct ocfs2_xattr_value_root **xv,
6227 struct buffer_head **ret_bh,
6231 * Calculate all the xattr value root metadata stored in this xattr header and
6232 * credits we need if we create them from the scratch.
6233 * We use get_xattr_value_root so that all types of xattr container can use it.
6235 static int ocfs2_value_metas_in_xattr_header(struct super_block *sb,
6236 struct buffer_head *bh,
6237 struct ocfs2_xattr_header *xh,
6238 int *metas, int *credits,
6240 get_xattr_value_root *func,
6244 struct ocfs2_xattr_value_root *xv;
6245 struct ocfs2_xattr_entry *xe;
6247 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6248 xe = &xh->xh_entries[i];
6249 if (ocfs2_xattr_is_local(xe))
6252 ret = func(sb, bh, xh, i, &xv, NULL, para);
6258 *metas += le16_to_cpu(xv->xr_list.l_tree_depth) *
6259 le16_to_cpu(xv->xr_list.l_next_free_rec);
6261 *credits += ocfs2_calc_extend_credits(sb,
6263 le32_to_cpu(xv->xr_clusters));
6266 * If the value is a tree with depth > 1, We don't go deep
6267 * to the extent block, so just calculate a maximum record num.
6269 if (!xv->xr_list.l_tree_depth)
6270 *num_recs += le16_to_cpu(xv->xr_list.l_next_free_rec);
6272 *num_recs += ocfs2_clusters_for_bytes(sb,
6279 /* Used by xattr inode and block to return the right xv and buffer_head. */
6280 static int ocfs2_get_xattr_value_root(struct super_block *sb,
6281 struct buffer_head *bh,
6282 struct ocfs2_xattr_header *xh,
6284 struct ocfs2_xattr_value_root **xv,
6285 struct buffer_head **ret_bh,
6288 struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
6290 *xv = (struct ocfs2_xattr_value_root *)((void *)xh +
6291 le16_to_cpu(xe->xe_name_offset) +
6292 OCFS2_XATTR_SIZE(xe->xe_name_len));
6301 * Lock the meta_ac and caculate how much credits we need for reflink xattrs.
6302 * It is only used for inline xattr and xattr block.
6304 static int ocfs2_reflink_lock_xattr_allocators(struct ocfs2_super *osb,
6305 struct ocfs2_xattr_header *xh,
6306 struct buffer_head *ref_root_bh,
6308 struct ocfs2_alloc_context **meta_ac)
6310 int ret, meta_add = 0, num_recs = 0;
6311 struct ocfs2_refcount_block *rb =
6312 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
6316 ret = ocfs2_value_metas_in_xattr_header(osb->sb, NULL, xh,
6317 &meta_add, credits, &num_recs,
6318 ocfs2_get_xattr_value_root,
6326 * We need to add/modify num_recs in refcount tree, so just calculate
6327 * an approximate number we need for refcount tree change.
6328 * Sometimes we need to split the tree, and after split, half recs
6329 * will be moved to the new block, and a new block can only provide
6330 * half number of recs. So we multiple new blocks by 2.
6332 num_recs = num_recs / ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6333 meta_add += num_recs;
6334 *credits += num_recs + num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6335 if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6336 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6337 le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6341 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add, meta_ac);
6350 * Given a xattr header, reflink all the xattrs in this container.
6351 * It can be used for inode, block and bucket.
6354 * Before we call this function, the caller has memcpy the xattr in
6355 * old_xh to the new_xh.
6357 * If args.xattr_reflinked is set, call it to decide whether the xe should
6358 * be reflinked or not. If not, remove it from the new xattr header.
6360 static int ocfs2_reflink_xattr_header(handle_t *handle,
6361 struct ocfs2_xattr_reflink *args,
6362 struct buffer_head *old_bh,
6363 struct ocfs2_xattr_header *xh,
6364 struct buffer_head *new_bh,
6365 struct ocfs2_xattr_header *new_xh,
6366 struct ocfs2_xattr_value_buf *vb,
6367 struct ocfs2_alloc_context *meta_ac,
6368 get_xattr_value_root *func,
6372 struct super_block *sb = args->old_inode->i_sb;
6373 struct buffer_head *value_bh;
6374 struct ocfs2_xattr_entry *xe, *last;
6375 struct ocfs2_xattr_value_root *xv, *new_xv;
6376 struct ocfs2_extent_tree data_et;
6377 u32 clusters, cpos, p_cluster, num_clusters;
6378 unsigned int ext_flags = 0;
6380 mlog(0, "reflink xattr in container %llu, count = %u\n",
6381 (unsigned long long)old_bh->b_blocknr, le16_to_cpu(xh->xh_count));
6383 last = &new_xh->xh_entries[le16_to_cpu(new_xh->xh_count)];
6384 for (i = 0, j = 0; i < le16_to_cpu(xh->xh_count); i++, j++) {
6385 xe = &xh->xh_entries[i];
6387 if (args->xattr_reflinked && !args->xattr_reflinked(xe)) {
6388 xe = &new_xh->xh_entries[j];
6390 le16_add_cpu(&new_xh->xh_count, -1);
6391 if (new_xh->xh_count) {
6393 (void *)last - (void *)xe);
6395 sizeof(struct ocfs2_xattr_entry));
6399 * We don't want j to increase in the next round since
6400 * it is already moved ahead.
6406 if (ocfs2_xattr_is_local(xe))
6409 ret = func(sb, old_bh, xh, i, &xv, NULL, para);
6415 ret = func(sb, new_bh, new_xh, j, &new_xv, &value_bh, para);
6422 * For the xattr which has l_tree_depth = 0, all the extent
6423 * recs have already be copied to the new xh with the
6424 * propriate OCFS2_EXT_REFCOUNTED flag we just need to
6425 * increase the refount count int the refcount tree.
6427 * For the xattr which has l_tree_depth > 0, we need
6428 * to initialize it to the empty default value root,
6429 * and then insert the extents one by one.
6431 if (xv->xr_list.l_tree_depth) {
6432 memcpy(new_xv, &def_xv, sizeof(def_xv));
6434 vb->vb_bh = value_bh;
6435 ocfs2_init_xattr_value_extent_tree(&data_et,
6436 INODE_CACHE(args->new_inode), vb);
6439 clusters = le32_to_cpu(xv->xr_clusters);
6441 while (cpos < clusters) {
6442 ret = ocfs2_xattr_get_clusters(args->old_inode,
6455 if (xv->xr_list.l_tree_depth) {
6456 ret = ocfs2_insert_extent(handle,
6458 ocfs2_clusters_to_blocks(
6459 args->old_inode->i_sb,
6461 num_clusters, ext_flags,
6469 ret = ocfs2_increase_refcount(handle, args->ref_ci,
6471 p_cluster, num_clusters,
6472 meta_ac, args->dealloc);
6478 cpos += num_clusters;
6486 static int ocfs2_reflink_xattr_inline(struct ocfs2_xattr_reflink *args)
6488 int ret = 0, credits = 0;
6490 struct ocfs2_super *osb = OCFS2_SB(args->old_inode->i_sb);
6491 struct ocfs2_dinode *di = (struct ocfs2_dinode *)args->old_bh->b_data;
6492 int inline_size = le16_to_cpu(di->i_xattr_inline_size);
6493 int header_off = osb->sb->s_blocksize - inline_size;
6494 struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)
6495 (args->old_bh->b_data + header_off);
6496 struct ocfs2_xattr_header *new_xh = (struct ocfs2_xattr_header *)
6497 (args->new_bh->b_data + header_off);
6498 struct ocfs2_alloc_context *meta_ac = NULL;
6499 struct ocfs2_inode_info *new_oi;
6500 struct ocfs2_dinode *new_di;
6501 struct ocfs2_xattr_value_buf vb = {
6502 .vb_bh = args->new_bh,
6503 .vb_access = ocfs2_journal_access_di,
6506 ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6507 &credits, &meta_ac);
6513 handle = ocfs2_start_trans(osb, credits);
6514 if (IS_ERR(handle)) {
6515 ret = PTR_ERR(handle);
6520 ret = ocfs2_journal_access_di(handle, INODE_CACHE(args->new_inode),
6521 args->new_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6527 memcpy(args->new_bh->b_data + header_off,
6528 args->old_bh->b_data + header_off, inline_size);
6530 new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6531 new_di->i_xattr_inline_size = cpu_to_le16(inline_size);
6533 ret = ocfs2_reflink_xattr_header(handle, args, args->old_bh, xh,
6534 args->new_bh, new_xh, &vb, meta_ac,
6535 ocfs2_get_xattr_value_root, NULL);
6541 new_oi = OCFS2_I(args->new_inode);
6542 spin_lock(&new_oi->ip_lock);
6543 new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL | OCFS2_INLINE_XATTR_FL;
6544 new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6545 spin_unlock(&new_oi->ip_lock);
6547 ocfs2_journal_dirty(handle, args->new_bh);
6550 ocfs2_commit_trans(osb, handle);
6554 ocfs2_free_alloc_context(meta_ac);
6558 static int ocfs2_create_empty_xattr_block(struct inode *inode,
6559 struct buffer_head *fe_bh,
6560 struct buffer_head **ret_bh,
6565 struct ocfs2_alloc_context *meta_ac;
6566 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6568 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
6574 handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_CREATE_CREDITS);
6575 if (IS_ERR(handle)) {
6576 ret = PTR_ERR(handle);
6581 mlog(0, "create new xattr block for inode %llu, index = %d\n",
6582 (unsigned long long)fe_bh->b_blocknr, indexed);
6583 ret = ocfs2_create_xattr_block(handle, inode, fe_bh,
6584 meta_ac, ret_bh, indexed);
6588 ocfs2_commit_trans(osb, handle);
6590 ocfs2_free_alloc_context(meta_ac);
6594 static int ocfs2_reflink_xattr_block(struct ocfs2_xattr_reflink *args,
6595 struct buffer_head *blk_bh,
6596 struct buffer_head *new_blk_bh)
6598 int ret = 0, credits = 0;
6600 struct ocfs2_inode_info *new_oi = OCFS2_I(args->new_inode);
6601 struct ocfs2_dinode *new_di;
6602 struct ocfs2_super *osb = OCFS2_SB(args->new_inode->i_sb);
6603 int header_off = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
6604 struct ocfs2_xattr_block *xb =
6605 (struct ocfs2_xattr_block *)blk_bh->b_data;
6606 struct ocfs2_xattr_header *xh = &xb->xb_attrs.xb_header;
6607 struct ocfs2_xattr_block *new_xb =
6608 (struct ocfs2_xattr_block *)new_blk_bh->b_data;
6609 struct ocfs2_xattr_header *new_xh = &new_xb->xb_attrs.xb_header;
6610 struct ocfs2_alloc_context *meta_ac;
6611 struct ocfs2_xattr_value_buf vb = {
6612 .vb_bh = new_blk_bh,
6613 .vb_access = ocfs2_journal_access_xb,
6616 ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6617 &credits, &meta_ac);
6623 /* One more credits in case we need to add xattr flags in new inode. */
6624 handle = ocfs2_start_trans(osb, credits + 1);
6625 if (IS_ERR(handle)) {
6626 ret = PTR_ERR(handle);
6631 if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6632 ret = ocfs2_journal_access_di(handle,
6633 INODE_CACHE(args->new_inode),
6635 OCFS2_JOURNAL_ACCESS_WRITE);
6642 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(args->new_inode),
6643 new_blk_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6649 memcpy(new_blk_bh->b_data + header_off, blk_bh->b_data + header_off,
6650 osb->sb->s_blocksize - header_off);
6652 ret = ocfs2_reflink_xattr_header(handle, args, blk_bh, xh,
6653 new_blk_bh, new_xh, &vb, meta_ac,
6654 ocfs2_get_xattr_value_root, NULL);
6660 ocfs2_journal_dirty(handle, new_blk_bh);
6662 if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6663 new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6664 spin_lock(&new_oi->ip_lock);
6665 new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
6666 new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6667 spin_unlock(&new_oi->ip_lock);
6669 ocfs2_journal_dirty(handle, args->new_bh);
6673 ocfs2_commit_trans(osb, handle);
6676 ocfs2_free_alloc_context(meta_ac);
6680 struct ocfs2_reflink_xattr_tree_args {
6681 struct ocfs2_xattr_reflink *reflink;
6682 struct buffer_head *old_blk_bh;
6683 struct buffer_head *new_blk_bh;
6684 struct ocfs2_xattr_bucket *old_bucket;
6685 struct ocfs2_xattr_bucket *new_bucket;
6690 * We have to handle the case that both old bucket and new bucket
6691 * will call this function to get the right ret_bh.
6692 * So The caller must give us the right bh.
6694 static int ocfs2_get_reflink_xattr_value_root(struct super_block *sb,
6695 struct buffer_head *bh,
6696 struct ocfs2_xattr_header *xh,
6698 struct ocfs2_xattr_value_root **xv,
6699 struct buffer_head **ret_bh,
6702 struct ocfs2_reflink_xattr_tree_args *args =
6703 (struct ocfs2_reflink_xattr_tree_args *)para;
6704 struct ocfs2_xattr_bucket *bucket;
6706 if (bh == args->old_bucket->bu_bhs[0])
6707 bucket = args->old_bucket;
6709 bucket = args->new_bucket;
6711 return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6715 struct ocfs2_value_tree_metas {
6721 static int ocfs2_value_tree_metas_in_bucket(struct super_block *sb,
6722 struct buffer_head *bh,
6723 struct ocfs2_xattr_header *xh,
6725 struct ocfs2_xattr_value_root **xv,
6726 struct buffer_head **ret_bh,
6729 struct ocfs2_xattr_bucket *bucket =
6730 (struct ocfs2_xattr_bucket *)para;
6732 return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6736 static int ocfs2_calc_value_tree_metas(struct inode *inode,
6737 struct ocfs2_xattr_bucket *bucket,
6740 struct ocfs2_value_tree_metas *metas =
6741 (struct ocfs2_value_tree_metas *)para;
6742 struct ocfs2_xattr_header *xh =
6743 (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6745 /* Add the credits for this bucket first. */
6746 metas->credits += bucket->bu_blocks;
6747 return ocfs2_value_metas_in_xattr_header(inode->i_sb, bucket->bu_bhs[0],
6748 xh, &metas->num_metas,
6749 &metas->credits, &metas->num_recs,
6750 ocfs2_value_tree_metas_in_bucket,
6755 * Given a xattr extent rec starting from blkno and having len clusters,
6756 * iterate all the buckets calculate how much metadata we need for reflinking
6757 * all the ocfs2_xattr_value_root and lock the allocators accordingly.
6759 static int ocfs2_lock_reflink_xattr_rec_allocators(
6760 struct ocfs2_reflink_xattr_tree_args *args,
6761 struct ocfs2_extent_tree *xt_et,
6762 u64 blkno, u32 len, int *credits,
6763 struct ocfs2_alloc_context **meta_ac,
6764 struct ocfs2_alloc_context **data_ac)
6766 int ret, num_free_extents;
6767 struct ocfs2_value_tree_metas metas;
6768 struct ocfs2_super *osb = OCFS2_SB(args->reflink->old_inode->i_sb);
6769 struct ocfs2_refcount_block *rb;
6771 memset(&metas, 0, sizeof(metas));
6773 ret = ocfs2_iterate_xattr_buckets(args->reflink->old_inode, blkno, len,
6774 ocfs2_calc_value_tree_metas, &metas);
6780 *credits = metas.credits;
6783 * Calculate we need for refcount tree change.
6785 * We need to add/modify num_recs in refcount tree, so just calculate
6786 * an approximate number we need for refcount tree change.
6787 * Sometimes we need to split the tree, and after split, half recs
6788 * will be moved to the new block, and a new block can only provide
6789 * half number of recs. So we multiple new blocks by 2.
6790 * In the end, we have to add credits for modifying the already
6791 * existed refcount block.
6793 rb = (struct ocfs2_refcount_block *)args->reflink->ref_root_bh->b_data;
6795 (metas.num_recs + ocfs2_refcount_recs_per_rb(osb->sb) - 1) /
6796 ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6797 metas.num_metas += metas.num_recs;
6798 *credits += metas.num_recs +
6799 metas.num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6800 if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6801 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6802 le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6806 /* count in the xattr tree change. */
6807 num_free_extents = ocfs2_num_free_extents(osb, xt_et);
6808 if (num_free_extents < 0) {
6809 ret = num_free_extents;
6814 if (num_free_extents < len)
6815 metas.num_metas += ocfs2_extend_meta_needed(xt_et->et_root_el);
6817 *credits += ocfs2_calc_extend_credits(osb->sb,
6818 xt_et->et_root_el, len);
6820 if (metas.num_metas) {
6821 ret = ocfs2_reserve_new_metadata_blocks(osb, metas.num_metas,
6830 ret = ocfs2_reserve_clusters(osb, len, data_ac);
6837 ocfs2_free_alloc_context(*meta_ac);
6845 static int ocfs2_reflink_xattr_buckets(handle_t *handle,
6846 u64 blkno, u64 new_blkno, u32 clusters,
6847 struct ocfs2_alloc_context *meta_ac,
6848 struct ocfs2_alloc_context *data_ac,
6849 struct ocfs2_reflink_xattr_tree_args *args)
6852 struct super_block *sb = args->reflink->old_inode->i_sb;
6853 u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
6854 u32 num_buckets = clusters * bpc;
6855 int bpb = args->old_bucket->bu_blocks;
6856 struct ocfs2_xattr_value_buf vb = {
6857 .vb_access = ocfs2_journal_access,
6860 for (i = 0; i < num_buckets; i++, blkno += bpb, new_blkno += bpb) {
6861 ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6867 ret = ocfs2_init_xattr_bucket(args->new_bucket, new_blkno);
6874 * The real bucket num in this series of blocks is stored
6875 * in the 1st bucket.
6878 num_buckets = le16_to_cpu(
6879 bucket_xh(args->old_bucket)->xh_num_buckets);
6881 ret = ocfs2_xattr_bucket_journal_access(handle,
6883 OCFS2_JOURNAL_ACCESS_CREATE);
6889 for (j = 0; j < bpb; j++)
6890 memcpy(bucket_block(args->new_bucket, j),
6891 bucket_block(args->old_bucket, j),
6894 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6896 ret = ocfs2_reflink_xattr_header(handle, args->reflink,
6897 args->old_bucket->bu_bhs[0],
6898 bucket_xh(args->old_bucket),
6899 args->new_bucket->bu_bhs[0],
6900 bucket_xh(args->new_bucket),
6902 ocfs2_get_reflink_xattr_value_root,
6910 * Re-access and dirty the bucket to calculate metaecc.
6911 * Because we may extend the transaction in reflink_xattr_header
6912 * which will let the already accessed block gone.
6914 ret = ocfs2_xattr_bucket_journal_access(handle,
6916 OCFS2_JOURNAL_ACCESS_WRITE);
6922 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6923 ocfs2_xattr_bucket_relse(args->old_bucket);
6924 ocfs2_xattr_bucket_relse(args->new_bucket);
6927 ocfs2_xattr_bucket_relse(args->old_bucket);
6928 ocfs2_xattr_bucket_relse(args->new_bucket);
6932 * Create the same xattr extent record in the new inode's xattr tree.
6934 static int ocfs2_reflink_xattr_rec(struct inode *inode,
6935 struct buffer_head *root_bh,
6941 int ret, credits = 0;
6942 u32 p_cluster, num_clusters;
6945 struct ocfs2_reflink_xattr_tree_args *args =
6946 (struct ocfs2_reflink_xattr_tree_args *)para;
6947 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6948 struct ocfs2_alloc_context *meta_ac = NULL;
6949 struct ocfs2_alloc_context *data_ac = NULL;
6950 struct ocfs2_extent_tree et;
6952 ocfs2_init_xattr_tree_extent_tree(&et,
6953 INODE_CACHE(args->reflink->new_inode),
6956 ret = ocfs2_lock_reflink_xattr_rec_allocators(args, &et, blkno,
6958 &meta_ac, &data_ac);
6964 handle = ocfs2_start_trans(osb, credits);
6965 if (IS_ERR(handle)) {
6966 ret = PTR_ERR(handle);
6971 ret = ocfs2_claim_clusters(osb, handle, data_ac,
6972 len, &p_cluster, &num_clusters);
6978 new_blkno = ocfs2_clusters_to_blocks(osb->sb, p_cluster);
6980 mlog(0, "reflink xattr buckets %llu to %llu, len %u\n",
6981 (unsigned long long)blkno, (unsigned long long)new_blkno, len);
6982 ret = ocfs2_reflink_xattr_buckets(handle, blkno, new_blkno, len,
6983 meta_ac, data_ac, args);
6989 mlog(0, "insert new xattr extent rec start %llu len %u to %u\n",
6990 (unsigned long long)new_blkno, len, cpos);
6991 ret = ocfs2_insert_extent(handle, &et, cpos, new_blkno,
6997 ocfs2_commit_trans(osb, handle);
7001 ocfs2_free_alloc_context(meta_ac);
7003 ocfs2_free_alloc_context(data_ac);
7008 * Create reflinked xattr buckets.
7009 * We will add bucket one by one, and refcount all the xattrs in the bucket
7010 * if they are stored outside.
7012 static int ocfs2_reflink_xattr_tree(struct ocfs2_xattr_reflink *args,
7013 struct buffer_head *blk_bh,
7014 struct buffer_head *new_blk_bh)
7017 struct ocfs2_reflink_xattr_tree_args para;
7019 memset(¶, 0, sizeof(para));
7020 para.reflink = args;
7021 para.old_blk_bh = blk_bh;
7022 para.new_blk_bh = new_blk_bh;
7024 para.old_bucket = ocfs2_xattr_bucket_new(args->old_inode);
7025 if (!para.old_bucket) {
7026 mlog_errno(-ENOMEM);
7030 para.new_bucket = ocfs2_xattr_bucket_new(args->new_inode);
7031 if (!para.new_bucket) {
7037 ret = ocfs2_iterate_xattr_index_block(args->old_inode, blk_bh,
7038 ocfs2_reflink_xattr_rec,
7044 ocfs2_xattr_bucket_free(para.old_bucket);
7045 ocfs2_xattr_bucket_free(para.new_bucket);
7049 static int ocfs2_reflink_xattr_in_block(struct ocfs2_xattr_reflink *args,
7050 struct buffer_head *blk_bh)
7052 int ret, indexed = 0;
7053 struct buffer_head *new_blk_bh = NULL;
7054 struct ocfs2_xattr_block *xb =
7055 (struct ocfs2_xattr_block *)blk_bh->b_data;
7058 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)
7061 ret = ocfs2_create_empty_xattr_block(args->new_inode, args->new_bh,
7062 &new_blk_bh, indexed);
7068 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED))
7069 ret = ocfs2_reflink_xattr_block(args, blk_bh, new_blk_bh);
7071 ret = ocfs2_reflink_xattr_tree(args, blk_bh, new_blk_bh);
7080 static int ocfs2_reflink_xattr_no_security(struct ocfs2_xattr_entry *xe)
7082 int type = ocfs2_xattr_get_type(xe);
7084 return type != OCFS2_XATTR_INDEX_SECURITY &&
7085 type != OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS &&
7086 type != OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT;
7089 int ocfs2_reflink_xattrs(struct inode *old_inode,
7090 struct buffer_head *old_bh,
7091 struct inode *new_inode,
7092 struct buffer_head *new_bh,
7093 bool preserve_security)
7096 struct ocfs2_xattr_reflink args;
7097 struct ocfs2_inode_info *oi = OCFS2_I(old_inode);
7098 struct ocfs2_dinode *di = (struct ocfs2_dinode *)old_bh->b_data;
7099 struct buffer_head *blk_bh = NULL;
7100 struct ocfs2_cached_dealloc_ctxt dealloc;
7101 struct ocfs2_refcount_tree *ref_tree;
7102 struct buffer_head *ref_root_bh = NULL;
7104 ret = ocfs2_lock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7105 le64_to_cpu(di->i_refcount_loc),
7106 1, &ref_tree, &ref_root_bh);
7112 ocfs2_init_dealloc_ctxt(&dealloc);
7114 args.old_inode = old_inode;
7115 args.new_inode = new_inode;
7116 args.old_bh = old_bh;
7117 args.new_bh = new_bh;
7118 args.ref_ci = &ref_tree->rf_ci;
7119 args.ref_root_bh = ref_root_bh;
7120 args.dealloc = &dealloc;
7121 if (preserve_security)
7122 args.xattr_reflinked = NULL;
7124 args.xattr_reflinked = ocfs2_reflink_xattr_no_security;
7126 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
7127 ret = ocfs2_reflink_xattr_inline(&args);
7134 if (!di->i_xattr_loc)
7137 ret = ocfs2_read_xattr_block(old_inode, le64_to_cpu(di->i_xattr_loc),
7144 ret = ocfs2_reflink_xattr_in_block(&args, blk_bh);
7151 ocfs2_unlock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7153 brelse(ref_root_bh);
7155 if (ocfs2_dealloc_has_cluster(&dealloc)) {
7156 ocfs2_schedule_truncate_log_flush(OCFS2_SB(old_inode->i_sb), 1);
7157 ocfs2_run_deallocs(OCFS2_SB(old_inode->i_sb), &dealloc);
7165 * Initialize security and acl for a already created inode.
7166 * Used for reflink a non-preserve-security file.
7168 * It uses common api like ocfs2_xattr_set, so the caller
7169 * must not hold any lock expect i_mutex.
7171 int ocfs2_init_security_and_acl(struct inode *dir,
7172 struct inode *inode)
7175 struct buffer_head *dir_bh = NULL;
7176 struct ocfs2_security_xattr_info si = {
7180 ret = ocfs2_init_security_get(inode, dir, &si);
7182 ret = ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY,
7183 si.name, si.value, si.value_len,
7189 } else if (ret != -EOPNOTSUPP) {
7194 ret = ocfs2_inode_lock(dir, &dir_bh, 0);
7200 ret = ocfs2_init_acl(NULL, inode, dir, NULL, dir_bh, NULL, NULL);
7204 ocfs2_inode_unlock(dir, 0);
7210 * 'security' attributes support
7212 static size_t ocfs2_xattr_security_list(struct dentry *dentry, char *list,
7213 size_t list_size, const char *name,
7214 size_t name_len, int type)
7216 const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
7217 const size_t total_len = prefix_len + name_len + 1;
7219 if (list && total_len <= list_size) {
7220 memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
7221 memcpy(list + prefix_len, name, name_len);
7222 list[prefix_len + name_len] = '\0';
7227 static int ocfs2_xattr_security_get(struct dentry *dentry, const char *name,
7228 void *buffer, size_t size, int type)
7230 if (strcmp(name, "") == 0)
7232 return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_SECURITY,
7233 name, buffer, size);
7236 static int ocfs2_xattr_security_set(struct dentry *dentry, const char *name,
7237 const void *value, size_t size, int flags, int type)
7239 if (strcmp(name, "") == 0)
7242 return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_SECURITY,
7243 name, value, size, flags);
7246 int ocfs2_init_security_get(struct inode *inode,
7248 struct ocfs2_security_xattr_info *si)
7250 /* check whether ocfs2 support feature xattr */
7251 if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
7253 return security_inode_init_security(inode, dir, &si->name, &si->value,
7257 int ocfs2_init_security_set(handle_t *handle,
7258 struct inode *inode,
7259 struct buffer_head *di_bh,
7260 struct ocfs2_security_xattr_info *si,
7261 struct ocfs2_alloc_context *xattr_ac,
7262 struct ocfs2_alloc_context *data_ac)
7264 return ocfs2_xattr_set_handle(handle, inode, di_bh,
7265 OCFS2_XATTR_INDEX_SECURITY,
7266 si->name, si->value, si->value_len, 0,
7270 struct xattr_handler ocfs2_xattr_security_handler = {
7271 .prefix = XATTR_SECURITY_PREFIX,
7272 .list = ocfs2_xattr_security_list,
7273 .get = ocfs2_xattr_security_get,
7274 .set = ocfs2_xattr_security_set,
7278 * 'trusted' attributes support
7280 static size_t ocfs2_xattr_trusted_list(struct dentry *dentry, char *list,
7281 size_t list_size, const char *name,
7282 size_t name_len, int type)
7284 const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
7285 const size_t total_len = prefix_len + name_len + 1;
7287 if (list && total_len <= list_size) {
7288 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
7289 memcpy(list + prefix_len, name, name_len);
7290 list[prefix_len + name_len] = '\0';
7295 static int ocfs2_xattr_trusted_get(struct dentry *dentry, const char *name,
7296 void *buffer, size_t size, int type)
7298 if (strcmp(name, "") == 0)
7300 return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_TRUSTED,
7301 name, buffer, size);
7304 static int ocfs2_xattr_trusted_set(struct dentry *dentry, const char *name,
7305 const void *value, size_t size, int flags, int type)
7307 if (strcmp(name, "") == 0)
7310 return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_TRUSTED,
7311 name, value, size, flags);
7314 struct xattr_handler ocfs2_xattr_trusted_handler = {
7315 .prefix = XATTR_TRUSTED_PREFIX,
7316 .list = ocfs2_xattr_trusted_list,
7317 .get = ocfs2_xattr_trusted_get,
7318 .set = ocfs2_xattr_trusted_set,
7322 * 'user' attributes support
7324 static size_t ocfs2_xattr_user_list(struct dentry *dentry, char *list,
7325 size_t list_size, const char *name,
7326 size_t name_len, int type)
7328 const size_t prefix_len = XATTR_USER_PREFIX_LEN;
7329 const size_t total_len = prefix_len + name_len + 1;
7330 struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7332 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7335 if (list && total_len <= list_size) {
7336 memcpy(list, XATTR_USER_PREFIX, prefix_len);
7337 memcpy(list + prefix_len, name, name_len);
7338 list[prefix_len + name_len] = '\0';
7343 static int ocfs2_xattr_user_get(struct dentry *dentry, const char *name,
7344 void *buffer, size_t size, int type)
7346 struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7348 if (strcmp(name, "") == 0)
7350 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7352 return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_USER, name,
7356 static int ocfs2_xattr_user_set(struct dentry *dentry, const char *name,
7357 const void *value, size_t size, int flags, int type)
7359 struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7361 if (strcmp(name, "") == 0)
7363 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7366 return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_USER,
7367 name, value, size, flags);
7370 struct xattr_handler ocfs2_xattr_user_handler = {
7371 .prefix = XATTR_USER_PREFIX,
7372 .list = ocfs2_xattr_user_list,
7373 .get = ocfs2_xattr_user_get,
7374 .set = ocfs2_xattr_user_set,