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 {
122 const void *xi_value;
126 struct ocfs2_xattr_search {
127 struct buffer_head *inode_bh;
129 * xattr_bh point to the block buffer head which has extended attribute
130 * when extended attribute in inode, xattr_bh is equal to inode_bh.
132 struct buffer_head *xattr_bh;
133 struct ocfs2_xattr_header *header;
134 struct ocfs2_xattr_bucket *bucket;
137 struct ocfs2_xattr_entry *here;
141 /* Operations on struct ocfs2_xa_entry */
143 struct ocfs2_xa_loc_operations {
147 int (*xlo_journal_access)(handle_t *handle, struct ocfs2_xa_loc *loc,
149 void (*xlo_journal_dirty)(handle_t *handle, struct ocfs2_xa_loc *loc);
152 * Return a pointer to the appropriate buffer in loc->xl_storage
153 * at the given offset from loc->xl_header.
155 void *(*xlo_offset_pointer)(struct ocfs2_xa_loc *loc, int offset);
157 /* Can we reuse the existing entry for the new value? */
158 int (*xlo_can_reuse)(struct ocfs2_xa_loc *loc,
159 struct ocfs2_xattr_info *xi);
161 /* How much space is needed for the new value? */
162 int (*xlo_check_space)(struct ocfs2_xa_loc *loc,
163 struct ocfs2_xattr_info *xi);
166 * Return the offset of the first name+value pair. This is
167 * the start of our downward-filling free space.
169 int (*xlo_get_free_start)(struct ocfs2_xa_loc *loc);
172 * Remove the name+value at this location. Do whatever is
173 * appropriate with the remaining name+value pairs.
175 void (*xlo_wipe_namevalue)(struct ocfs2_xa_loc *loc);
177 /* Fill xl_entry with a new entry */
178 void (*xlo_add_entry)(struct ocfs2_xa_loc *loc, u32 name_hash);
180 /* Add name+value storage to an entry */
181 void (*xlo_add_namevalue)(struct ocfs2_xa_loc *loc, int size);
184 * Initialize the value buf's access and bh fields for this entry.
185 * ocfs2_xa_fill_value_buf() will handle the xv pointer.
187 void (*xlo_fill_value_buf)(struct ocfs2_xa_loc *loc,
188 struct ocfs2_xattr_value_buf *vb);
192 * Describes an xattr entry location. This is a memory structure
193 * tracking the on-disk structure.
195 struct ocfs2_xa_loc {
196 /* This xattr belongs to this inode */
197 struct inode *xl_inode;
199 /* The ocfs2_xattr_header inside the on-disk storage. Not NULL. */
200 struct ocfs2_xattr_header *xl_header;
202 /* Bytes from xl_header to the end of the storage */
206 * The ocfs2_xattr_entry this location describes. If this is
207 * NULL, this location describes the on-disk structure where it
210 struct ocfs2_xattr_entry *xl_entry;
213 * Internal housekeeping
216 /* Buffer(s) containing this entry */
219 /* Operations on the storage backing this location */
220 const struct ocfs2_xa_loc_operations *xl_ops;
224 * Convenience functions to calculate how much space is needed for a
225 * given name+value pair
227 static int namevalue_size(int name_len, uint64_t value_len)
229 if (value_len > OCFS2_XATTR_INLINE_SIZE)
230 return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
232 return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
235 static int namevalue_size_xi(struct ocfs2_xattr_info *xi)
237 return namevalue_size(xi->xi_name_len, xi->xi_value_len);
240 static int namevalue_size_xe(struct ocfs2_xattr_entry *xe)
242 u64 value_len = le64_to_cpu(xe->xe_value_size);
244 BUG_ON((value_len > OCFS2_XATTR_INLINE_SIZE) &&
245 ocfs2_xattr_is_local(xe));
246 return namevalue_size(xe->xe_name_len, value_len);
250 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
251 struct ocfs2_xattr_header *xh,
256 static int ocfs2_xattr_block_find(struct inode *inode,
259 struct ocfs2_xattr_search *xs);
260 static int ocfs2_xattr_index_block_find(struct inode *inode,
261 struct buffer_head *root_bh,
264 struct ocfs2_xattr_search *xs);
266 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
267 struct buffer_head *blk_bh,
271 static int ocfs2_xattr_create_index_block(struct inode *inode,
272 struct ocfs2_xattr_search *xs,
273 struct ocfs2_xattr_set_ctxt *ctxt);
275 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
276 struct ocfs2_xattr_info *xi,
277 struct ocfs2_xattr_search *xs,
278 struct ocfs2_xattr_set_ctxt *ctxt);
280 typedef int (xattr_tree_rec_func)(struct inode *inode,
281 struct buffer_head *root_bh,
282 u64 blkno, u32 cpos, u32 len, void *para);
283 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
284 struct buffer_head *root_bh,
285 xattr_tree_rec_func *rec_func,
287 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
288 struct ocfs2_xattr_bucket *bucket,
290 static int ocfs2_rm_xattr_cluster(struct inode *inode,
291 struct buffer_head *root_bh,
297 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
298 u64 src_blk, u64 last_blk, u64 to_blk,
299 unsigned int start_bucket,
301 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
302 struct ocfs2_dinode *di,
303 struct ocfs2_xattr_info *xi,
304 struct ocfs2_xattr_search *xis,
305 struct ocfs2_xattr_search *xbs,
306 struct ocfs2_refcount_tree **ref_tree,
309 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
310 struct ocfs2_xattr_bucket *bucket,
312 struct ocfs2_xattr_value_root **xv,
313 struct buffer_head **bh);
315 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
317 return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
320 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
322 return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
325 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
326 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
327 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
329 static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
331 struct ocfs2_xattr_bucket *bucket;
332 int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
334 BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
336 bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
338 bucket->bu_inode = inode;
339 bucket->bu_blocks = blks;
345 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
349 for (i = 0; i < bucket->bu_blocks; i++) {
350 brelse(bucket->bu_bhs[i]);
351 bucket->bu_bhs[i] = NULL;
355 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
358 ocfs2_xattr_bucket_relse(bucket);
359 bucket->bu_inode = NULL;
365 * A bucket that has never been written to disk doesn't need to be
366 * read. We just need the buffer_heads. Don't call this for
367 * buckets that are already on disk. ocfs2_read_xattr_bucket() initializes
370 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
375 for (i = 0; i < bucket->bu_blocks; i++) {
376 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
378 if (!bucket->bu_bhs[i]) {
384 if (!ocfs2_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
386 ocfs2_set_new_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
391 ocfs2_xattr_bucket_relse(bucket);
395 /* Read the xattr bucket at xb_blkno */
396 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
401 rc = ocfs2_read_blocks(INODE_CACHE(bucket->bu_inode), xb_blkno,
402 bucket->bu_blocks, bucket->bu_bhs, 0,
405 spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
406 rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
409 &bucket_xh(bucket)->xh_check);
410 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
416 ocfs2_xattr_bucket_relse(bucket);
420 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
421 struct ocfs2_xattr_bucket *bucket,
426 for (i = 0; i < bucket->bu_blocks; i++) {
427 rc = ocfs2_journal_access(handle,
428 INODE_CACHE(bucket->bu_inode),
429 bucket->bu_bhs[i], type);
439 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
440 struct ocfs2_xattr_bucket *bucket)
444 spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
445 ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
446 bucket->bu_bhs, bucket->bu_blocks,
447 &bucket_xh(bucket)->xh_check);
448 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
450 for (i = 0; i < bucket->bu_blocks; i++)
451 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
454 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
455 struct ocfs2_xattr_bucket *src)
458 int blocksize = src->bu_inode->i_sb->s_blocksize;
460 BUG_ON(dest->bu_blocks != src->bu_blocks);
461 BUG_ON(dest->bu_inode != src->bu_inode);
463 for (i = 0; i < src->bu_blocks; i++) {
464 memcpy(bucket_block(dest, i), bucket_block(src, i),
469 static int ocfs2_validate_xattr_block(struct super_block *sb,
470 struct buffer_head *bh)
473 struct ocfs2_xattr_block *xb =
474 (struct ocfs2_xattr_block *)bh->b_data;
476 mlog(0, "Validating xattr block %llu\n",
477 (unsigned long long)bh->b_blocknr);
479 BUG_ON(!buffer_uptodate(bh));
482 * If the ecc fails, we return the error but otherwise
483 * leave the filesystem running. We know any error is
484 * local to this block.
486 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
491 * Errors after here are fatal
494 if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
496 "Extended attribute block #%llu has bad "
498 (unsigned long long)bh->b_blocknr, 7,
503 if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
505 "Extended attribute block #%llu has an "
506 "invalid xb_blkno of %llu",
507 (unsigned long long)bh->b_blocknr,
508 (unsigned long long)le64_to_cpu(xb->xb_blkno));
512 if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
514 "Extended attribute block #%llu has an invalid "
515 "xb_fs_generation of #%u",
516 (unsigned long long)bh->b_blocknr,
517 le32_to_cpu(xb->xb_fs_generation));
524 static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
525 struct buffer_head **bh)
528 struct buffer_head *tmp = *bh;
530 rc = ocfs2_read_block(INODE_CACHE(inode), xb_blkno, &tmp,
531 ocfs2_validate_xattr_block);
533 /* If ocfs2_read_block() got us a new bh, pass it up. */
540 static inline const char *ocfs2_xattr_prefix(int name_index)
542 struct xattr_handler *handler = NULL;
544 if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
545 handler = ocfs2_xattr_handler_map[name_index];
547 return handler ? handler->prefix : NULL;
550 static u32 ocfs2_xattr_name_hash(struct inode *inode,
554 /* Get hash value of uuid from super block */
555 u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
558 /* hash extended attribute name */
559 for (i = 0; i < name_len; i++) {
560 hash = (hash << OCFS2_HASH_SHIFT) ^
561 (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
568 static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
570 return namevalue_size(name_len, value_len) +
571 sizeof(struct ocfs2_xattr_entry);
574 static int ocfs2_xi_entry_usage(struct ocfs2_xattr_info *xi)
576 return namevalue_size_xi(xi) +
577 sizeof(struct ocfs2_xattr_entry);
580 static int ocfs2_xe_entry_usage(struct ocfs2_xattr_entry *xe)
582 return namevalue_size_xe(xe) +
583 sizeof(struct ocfs2_xattr_entry);
586 int ocfs2_calc_security_init(struct inode *dir,
587 struct ocfs2_security_xattr_info *si,
590 struct ocfs2_alloc_context **xattr_ac)
593 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
594 int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
598 * The max space of security xattr taken inline is
599 * 256(name) + 80(value) + 16(entry) = 352 bytes,
600 * So reserve one metadata block for it is ok.
602 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
603 s_size > OCFS2_XATTR_FREE_IN_IBODY) {
604 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
609 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
612 /* reserve clusters for xattr value which will be set in B tree*/
613 if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
614 int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
617 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
619 *want_clusters += new_clusters;
624 int ocfs2_calc_xattr_init(struct inode *dir,
625 struct buffer_head *dir_bh,
627 struct ocfs2_security_xattr_info *si,
633 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
634 int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
637 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
640 if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
641 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
642 OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
645 a_size = ocfs2_xattr_entry_real_size(0, acl_len);
648 } else if (acl_len != 0 && acl_len != -ENODATA) {
654 if (!(s_size + a_size))
658 * The max space of security xattr taken inline is
659 * 256(name) + 80(value) + 16(entry) = 352 bytes,
660 * The max space of acl xattr taken inline is
661 * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
662 * when blocksize = 512, may reserve one more cluser for
663 * xattr bucket, otherwise reserve one metadata block
665 * If this is a new directory with inline data,
666 * we choose to reserve the entire inline area for
667 * directory contents and force an external xattr block.
669 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
670 (S_ISDIR(mode) && ocfs2_supports_inline_data(osb)) ||
671 (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
672 *want_meta = *want_meta + 1;
673 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
676 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
677 (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
679 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
683 * reserve credits and clusters for xattrs which has large value
684 * and have to be set outside
686 if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
687 new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
689 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
691 *want_clusters += new_clusters;
693 if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
694 acl_len > OCFS2_XATTR_INLINE_SIZE) {
695 /* for directory, it has DEFAULT and ACCESS two types of acls */
696 new_clusters = (S_ISDIR(mode) ? 2 : 1) *
697 ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
698 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
700 *want_clusters += new_clusters;
706 static int ocfs2_xattr_extend_allocation(struct inode *inode,
708 struct ocfs2_xattr_value_buf *vb,
709 struct ocfs2_xattr_set_ctxt *ctxt)
712 handle_t *handle = ctxt->handle;
713 enum ocfs2_alloc_restarted why;
714 u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
715 struct ocfs2_extent_tree et;
717 mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
719 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
721 status = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
722 OCFS2_JOURNAL_ACCESS_WRITE);
728 prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
729 status = ocfs2_add_clusters_in_btree(handle,
742 ocfs2_journal_dirty(handle, vb->vb_bh);
744 clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) - prev_clusters;
747 * We should have already allocated enough space before the transaction,
748 * so no need to restart.
750 BUG_ON(why != RESTART_NONE || clusters_to_add);
757 static int __ocfs2_remove_xattr_range(struct inode *inode,
758 struct ocfs2_xattr_value_buf *vb,
759 u32 cpos, u32 phys_cpos, u32 len,
760 unsigned int ext_flags,
761 struct ocfs2_xattr_set_ctxt *ctxt)
764 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
765 handle_t *handle = ctxt->handle;
766 struct ocfs2_extent_tree et;
768 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
770 ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
771 OCFS2_JOURNAL_ACCESS_WRITE);
777 ret = ocfs2_remove_extent(handle, &et, cpos, len, ctxt->meta_ac,
784 le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
785 ocfs2_journal_dirty(handle, vb->vb_bh);
787 if (ext_flags & OCFS2_EXT_REFCOUNTED)
788 ret = ocfs2_decrease_refcount(inode, handle,
789 ocfs2_blocks_to_clusters(inode->i_sb,
791 len, ctxt->meta_ac, &ctxt->dealloc, 1);
793 ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc,
802 static int ocfs2_xattr_shrink_size(struct inode *inode,
805 struct ocfs2_xattr_value_buf *vb,
806 struct ocfs2_xattr_set_ctxt *ctxt)
809 unsigned int ext_flags;
810 u32 trunc_len, cpos, phys_cpos, alloc_size;
813 if (old_clusters <= new_clusters)
817 trunc_len = old_clusters - new_clusters;
819 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
821 &vb->vb_xv->xr_list, &ext_flags);
827 if (alloc_size > trunc_len)
828 alloc_size = trunc_len;
830 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
831 phys_cpos, alloc_size,
838 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
839 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode),
842 trunc_len -= alloc_size;
849 static int ocfs2_xattr_value_truncate(struct inode *inode,
850 struct ocfs2_xattr_value_buf *vb,
852 struct ocfs2_xattr_set_ctxt *ctxt)
855 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
856 u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
858 if (new_clusters == old_clusters)
861 if (new_clusters > old_clusters)
862 ret = ocfs2_xattr_extend_allocation(inode,
863 new_clusters - old_clusters,
866 ret = ocfs2_xattr_shrink_size(inode,
867 old_clusters, new_clusters,
873 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
874 size_t *result, const char *prefix,
875 const char *name, int name_len)
877 char *p = buffer + *result;
878 int prefix_len = strlen(prefix);
879 int total_len = prefix_len + name_len + 1;
881 *result += total_len;
883 /* we are just looking for how big our buffer needs to be */
890 memcpy(p, prefix, prefix_len);
891 memcpy(p + prefix_len, name, name_len);
892 p[prefix_len + name_len] = '\0';
897 static int ocfs2_xattr_list_entries(struct inode *inode,
898 struct ocfs2_xattr_header *header,
899 char *buffer, size_t buffer_size)
903 const char *prefix, *name;
905 for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
906 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
907 type = ocfs2_xattr_get_type(entry);
908 prefix = ocfs2_xattr_prefix(type);
911 name = (const char *)header +
912 le16_to_cpu(entry->xe_name_offset);
914 ret = ocfs2_xattr_list_entry(buffer, buffer_size,
915 &result, prefix, name,
925 int ocfs2_has_inline_xattr_value_outside(struct inode *inode,
926 struct ocfs2_dinode *di)
928 struct ocfs2_xattr_header *xh;
931 xh = (struct ocfs2_xattr_header *)
932 ((void *)di + inode->i_sb->s_blocksize -
933 le16_to_cpu(di->i_xattr_inline_size));
935 for (i = 0; i < le16_to_cpu(xh->xh_count); i++)
936 if (!ocfs2_xattr_is_local(&xh->xh_entries[i]))
942 static int ocfs2_xattr_ibody_list(struct inode *inode,
943 struct ocfs2_dinode *di,
947 struct ocfs2_xattr_header *header = NULL;
948 struct ocfs2_inode_info *oi = OCFS2_I(inode);
951 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
954 header = (struct ocfs2_xattr_header *)
955 ((void *)di + inode->i_sb->s_blocksize -
956 le16_to_cpu(di->i_xattr_inline_size));
958 ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
963 static int ocfs2_xattr_block_list(struct inode *inode,
964 struct ocfs2_dinode *di,
968 struct buffer_head *blk_bh = NULL;
969 struct ocfs2_xattr_block *xb;
972 if (!di->i_xattr_loc)
975 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
982 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
983 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
984 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
985 ret = ocfs2_xattr_list_entries(inode, header,
986 buffer, buffer_size);
988 ret = ocfs2_xattr_tree_list_index_block(inode, blk_bh,
989 buffer, buffer_size);
996 ssize_t ocfs2_listxattr(struct dentry *dentry,
1000 int ret = 0, i_ret = 0, b_ret = 0;
1001 struct buffer_head *di_bh = NULL;
1002 struct ocfs2_dinode *di = NULL;
1003 struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
1005 if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
1008 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1011 ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
1017 di = (struct ocfs2_dinode *)di_bh->b_data;
1019 down_read(&oi->ip_xattr_sem);
1020 i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
1028 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
1033 up_read(&oi->ip_xattr_sem);
1034 ocfs2_inode_unlock(dentry->d_inode, 0);
1038 return i_ret + b_ret;
1041 static int ocfs2_xattr_find_entry(int name_index,
1043 struct ocfs2_xattr_search *xs)
1045 struct ocfs2_xattr_entry *entry;
1052 name_len = strlen(name);
1054 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
1055 cmp = name_index - ocfs2_xattr_get_type(entry);
1057 cmp = name_len - entry->xe_name_len;
1059 cmp = memcmp(name, (xs->base +
1060 le16_to_cpu(entry->xe_name_offset)),
1068 return cmp ? -ENODATA : 0;
1071 static int ocfs2_xattr_get_value_outside(struct inode *inode,
1072 struct ocfs2_xattr_value_root *xv,
1076 u32 cpos, p_cluster, num_clusters, bpc, clusters;
1079 size_t cplen, blocksize;
1080 struct buffer_head *bh = NULL;
1081 struct ocfs2_extent_list *el;
1084 clusters = le32_to_cpu(xv->xr_clusters);
1085 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1086 blocksize = inode->i_sb->s_blocksize;
1089 while (cpos < clusters) {
1090 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1091 &num_clusters, el, NULL);
1097 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1098 /* Copy ocfs2_xattr_value */
1099 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1100 ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1107 cplen = len >= blocksize ? blocksize : len;
1108 memcpy(buffer, bh->b_data, cplen);
1117 cpos += num_clusters;
1123 static int ocfs2_xattr_ibody_get(struct inode *inode,
1128 struct ocfs2_xattr_search *xs)
1130 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1131 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1132 struct ocfs2_xattr_value_root *xv;
1136 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1139 xs->end = (void *)di + inode->i_sb->s_blocksize;
1140 xs->header = (struct ocfs2_xattr_header *)
1141 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1142 xs->base = (void *)xs->header;
1143 xs->here = xs->header->xh_entries;
1145 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1148 size = le64_to_cpu(xs->here->xe_value_size);
1150 if (size > buffer_size)
1152 if (ocfs2_xattr_is_local(xs->here)) {
1153 memcpy(buffer, (void *)xs->base +
1154 le16_to_cpu(xs->here->xe_name_offset) +
1155 OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1157 xv = (struct ocfs2_xattr_value_root *)
1158 (xs->base + le16_to_cpu(
1159 xs->here->xe_name_offset) +
1160 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1161 ret = ocfs2_xattr_get_value_outside(inode, xv,
1173 static int ocfs2_xattr_block_get(struct inode *inode,
1178 struct ocfs2_xattr_search *xs)
1180 struct ocfs2_xattr_block *xb;
1181 struct ocfs2_xattr_value_root *xv;
1183 int ret = -ENODATA, name_offset, name_len, i;
1184 int uninitialized_var(block_off);
1186 xs->bucket = ocfs2_xattr_bucket_new(inode);
1193 ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1199 if (xs->not_found) {
1204 xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1205 size = le64_to_cpu(xs->here->xe_value_size);
1208 if (size > buffer_size)
1211 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1212 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1213 i = xs->here - xs->header->xh_entries;
1215 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1216 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
1217 bucket_xh(xs->bucket),
1221 xs->base = bucket_block(xs->bucket, block_off);
1223 if (ocfs2_xattr_is_local(xs->here)) {
1224 memcpy(buffer, (void *)xs->base +
1225 name_offset + name_len, size);
1227 xv = (struct ocfs2_xattr_value_root *)
1228 (xs->base + name_offset + name_len);
1229 ret = ocfs2_xattr_get_value_outside(inode, xv,
1239 ocfs2_xattr_bucket_free(xs->bucket);
1241 brelse(xs->xattr_bh);
1242 xs->xattr_bh = NULL;
1246 int ocfs2_xattr_get_nolock(struct inode *inode,
1247 struct buffer_head *di_bh,
1254 struct ocfs2_dinode *di = NULL;
1255 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1256 struct ocfs2_xattr_search xis = {
1257 .not_found = -ENODATA,
1259 struct ocfs2_xattr_search xbs = {
1260 .not_found = -ENODATA,
1263 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1266 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1269 xis.inode_bh = xbs.inode_bh = di_bh;
1270 di = (struct ocfs2_dinode *)di_bh->b_data;
1272 down_read(&oi->ip_xattr_sem);
1273 ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1275 if (ret == -ENODATA && di->i_xattr_loc)
1276 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1278 up_read(&oi->ip_xattr_sem);
1283 /* ocfs2_xattr_get()
1285 * Copy an extended attribute into the buffer provided.
1286 * Buffer is NULL to compute the size of buffer required.
1288 static int ocfs2_xattr_get(struct inode *inode,
1295 struct buffer_head *di_bh = NULL;
1297 ret = ocfs2_inode_lock(inode, &di_bh, 0);
1302 ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1303 name, buffer, buffer_size);
1305 ocfs2_inode_unlock(inode, 0);
1312 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1314 struct ocfs2_xattr_value_buf *vb,
1318 int ret = 0, i, cp_len;
1319 u16 blocksize = inode->i_sb->s_blocksize;
1320 u32 p_cluster, num_clusters;
1321 u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1322 u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1324 struct buffer_head *bh = NULL;
1325 unsigned int ext_flags;
1326 struct ocfs2_xattr_value_root *xv = vb->vb_xv;
1328 BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1330 while (cpos < clusters) {
1331 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1332 &num_clusters, &xv->xr_list,
1339 BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1341 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1343 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1344 ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1351 ret = ocfs2_journal_access(handle,
1354 OCFS2_JOURNAL_ACCESS_WRITE);
1360 cp_len = value_len > blocksize ? blocksize : value_len;
1361 memcpy(bh->b_data, value, cp_len);
1362 value_len -= cp_len;
1364 if (cp_len < blocksize)
1365 memset(bh->b_data + cp_len, 0,
1366 blocksize - cp_len);
1368 ocfs2_journal_dirty(handle, bh);
1373 * XXX: do we need to empty all the following
1374 * blocks in this cluster?
1379 cpos += num_clusters;
1387 static int ocfs2_xa_check_space_helper(int needed_space, int free_start,
1395 free_space = free_start -
1396 sizeof(struct ocfs2_xattr_header) -
1397 (num_entries * sizeof(struct ocfs2_xattr_entry)) -
1398 OCFS2_XATTR_HEADER_GAP;
1401 if (free_space < needed_space)
1407 static int ocfs2_xa_journal_access(handle_t *handle, struct ocfs2_xa_loc *loc,
1410 return loc->xl_ops->xlo_journal_access(handle, loc, type);
1413 static void ocfs2_xa_journal_dirty(handle_t *handle, struct ocfs2_xa_loc *loc)
1415 loc->xl_ops->xlo_journal_dirty(handle, loc);
1418 /* Give a pointer into the storage for the given offset */
1419 static void *ocfs2_xa_offset_pointer(struct ocfs2_xa_loc *loc, int offset)
1421 BUG_ON(offset >= loc->xl_size);
1422 return loc->xl_ops->xlo_offset_pointer(loc, offset);
1426 * Wipe the name+value pair and allow the storage to reclaim it. This
1427 * must be followed by either removal of the entry or a call to
1428 * ocfs2_xa_add_namevalue().
1430 static void ocfs2_xa_wipe_namevalue(struct ocfs2_xa_loc *loc)
1432 loc->xl_ops->xlo_wipe_namevalue(loc);
1436 * Find lowest offset to a name+value pair. This is the start of our
1437 * downward-growing free space.
1439 static int ocfs2_xa_get_free_start(struct ocfs2_xa_loc *loc)
1441 return loc->xl_ops->xlo_get_free_start(loc);
1444 /* Can we reuse loc->xl_entry for xi? */
1445 static int ocfs2_xa_can_reuse_entry(struct ocfs2_xa_loc *loc,
1446 struct ocfs2_xattr_info *xi)
1448 return loc->xl_ops->xlo_can_reuse(loc, xi);
1451 /* How much free space is needed to set the new value */
1452 static int ocfs2_xa_check_space(struct ocfs2_xa_loc *loc,
1453 struct ocfs2_xattr_info *xi)
1455 return loc->xl_ops->xlo_check_space(loc, xi);
1458 static void ocfs2_xa_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1460 loc->xl_ops->xlo_add_entry(loc, name_hash);
1461 loc->xl_entry->xe_name_hash = cpu_to_le32(name_hash);
1463 * We can't leave the new entry's xe_name_offset at zero or
1464 * add_namevalue() will go nuts. We set it to the size of our
1465 * storage so that it can never be less than any other entry.
1467 loc->xl_entry->xe_name_offset = cpu_to_le16(loc->xl_size);
1470 static void ocfs2_xa_add_namevalue(struct ocfs2_xa_loc *loc,
1471 struct ocfs2_xattr_info *xi)
1473 int size = namevalue_size_xi(xi);
1477 loc->xl_ops->xlo_add_namevalue(loc, size);
1478 loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
1479 loc->xl_entry->xe_name_len = xi->xi_name_len;
1480 ocfs2_xattr_set_type(loc->xl_entry, xi->xi_name_index);
1481 ocfs2_xattr_set_local(loc->xl_entry,
1482 xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE);
1484 nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1485 nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
1486 memset(nameval_buf, 0, size);
1487 memcpy(nameval_buf, xi->xi_name, xi->xi_name_len);
1490 static void ocfs2_xa_fill_value_buf(struct ocfs2_xa_loc *loc,
1491 struct ocfs2_xattr_value_buf *vb)
1493 int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1494 int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
1496 /* Value bufs are for value trees */
1497 BUG_ON(ocfs2_xattr_is_local(loc->xl_entry));
1498 BUG_ON(namevalue_size_xe(loc->xl_entry) !=
1499 (name_size + OCFS2_XATTR_ROOT_SIZE));
1501 loc->xl_ops->xlo_fill_value_buf(loc, vb);
1503 (struct ocfs2_xattr_value_root *)ocfs2_xa_offset_pointer(loc,
1508 static int ocfs2_xa_block_journal_access(handle_t *handle,
1509 struct ocfs2_xa_loc *loc, int type)
1511 struct buffer_head *bh = loc->xl_storage;
1512 ocfs2_journal_access_func access;
1514 if (loc->xl_size == (bh->b_size -
1515 offsetof(struct ocfs2_xattr_block,
1516 xb_attrs.xb_header)))
1517 access = ocfs2_journal_access_xb;
1519 access = ocfs2_journal_access_di;
1520 return access(handle, INODE_CACHE(loc->xl_inode), bh, type);
1523 static void ocfs2_xa_block_journal_dirty(handle_t *handle,
1524 struct ocfs2_xa_loc *loc)
1526 struct buffer_head *bh = loc->xl_storage;
1528 ocfs2_journal_dirty(handle, bh);
1531 static void *ocfs2_xa_block_offset_pointer(struct ocfs2_xa_loc *loc,
1534 return (char *)loc->xl_header + offset;
1537 static int ocfs2_xa_block_can_reuse(struct ocfs2_xa_loc *loc,
1538 struct ocfs2_xattr_info *xi)
1541 * Block storage is strict. If the sizes aren't exact, we will
1542 * remove the old one and reinsert the new.
1544 return namevalue_size_xe(loc->xl_entry) ==
1545 namevalue_size_xi(xi);
1548 static int ocfs2_xa_block_get_free_start(struct ocfs2_xa_loc *loc)
1550 struct ocfs2_xattr_header *xh = loc->xl_header;
1551 int i, count = le16_to_cpu(xh->xh_count);
1552 int offset, free_start = loc->xl_size;
1554 for (i = 0; i < count; i++) {
1555 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1556 if (offset < free_start)
1557 free_start = offset;
1563 static int ocfs2_xa_block_check_space(struct ocfs2_xa_loc *loc,
1564 struct ocfs2_xattr_info *xi)
1566 int count = le16_to_cpu(loc->xl_header->xh_count);
1567 int free_start = ocfs2_xa_get_free_start(loc);
1568 int needed_space = ocfs2_xi_entry_usage(xi);
1571 * Block storage will reclaim the original entry before inserting
1572 * the new value, so we only need the difference. If the new
1573 * entry is smaller than the old one, we don't need anything.
1575 if (loc->xl_entry) {
1576 /* Don't need space if we're reusing! */
1577 if (ocfs2_xa_can_reuse_entry(loc, xi))
1580 needed_space -= ocfs2_xe_entry_usage(loc->xl_entry);
1582 if (needed_space < 0)
1584 return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1588 * Block storage for xattrs keeps the name+value pairs compacted. When
1589 * we remove one, we have to shift any that preceded it towards the end.
1591 static void ocfs2_xa_block_wipe_namevalue(struct ocfs2_xa_loc *loc)
1594 int namevalue_offset, first_namevalue_offset, namevalue_size;
1595 struct ocfs2_xattr_entry *entry = loc->xl_entry;
1596 struct ocfs2_xattr_header *xh = loc->xl_header;
1597 int count = le16_to_cpu(xh->xh_count);
1599 namevalue_offset = le16_to_cpu(entry->xe_name_offset);
1600 namevalue_size = namevalue_size_xe(entry);
1601 first_namevalue_offset = ocfs2_xa_get_free_start(loc);
1603 /* Shift the name+value pairs */
1604 memmove((char *)xh + first_namevalue_offset + namevalue_size,
1605 (char *)xh + first_namevalue_offset,
1606 namevalue_offset - first_namevalue_offset);
1607 memset((char *)xh + first_namevalue_offset, 0, namevalue_size);
1609 /* Now tell xh->xh_entries about it */
1610 for (i = 0; i < count; i++) {
1611 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1612 if (offset <= namevalue_offset)
1613 le16_add_cpu(&xh->xh_entries[i].xe_name_offset,
1618 * Note that we don't update xh_free_start or xh_name_value_len
1619 * because they're not used in block-stored xattrs.
1623 static void ocfs2_xa_block_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1625 int count = le16_to_cpu(loc->xl_header->xh_count);
1626 loc->xl_entry = &(loc->xl_header->xh_entries[count]);
1627 le16_add_cpu(&loc->xl_header->xh_count, 1);
1628 memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1631 static void ocfs2_xa_block_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1633 int free_start = ocfs2_xa_get_free_start(loc);
1635 loc->xl_entry->xe_name_offset = cpu_to_le16(free_start - size);
1638 static void ocfs2_xa_block_fill_value_buf(struct ocfs2_xa_loc *loc,
1639 struct ocfs2_xattr_value_buf *vb)
1641 struct buffer_head *bh = loc->xl_storage;
1643 if (loc->xl_size == (bh->b_size -
1644 offsetof(struct ocfs2_xattr_block,
1645 xb_attrs.xb_header)))
1646 vb->vb_access = ocfs2_journal_access_xb;
1648 vb->vb_access = ocfs2_journal_access_di;
1653 * Operations for xattrs stored in blocks. This includes inline inode
1654 * storage and unindexed ocfs2_xattr_blocks.
1656 static const struct ocfs2_xa_loc_operations ocfs2_xa_block_loc_ops = {
1657 .xlo_journal_access = ocfs2_xa_block_journal_access,
1658 .xlo_journal_dirty = ocfs2_xa_block_journal_dirty,
1659 .xlo_offset_pointer = ocfs2_xa_block_offset_pointer,
1660 .xlo_check_space = ocfs2_xa_block_check_space,
1661 .xlo_can_reuse = ocfs2_xa_block_can_reuse,
1662 .xlo_get_free_start = ocfs2_xa_block_get_free_start,
1663 .xlo_wipe_namevalue = ocfs2_xa_block_wipe_namevalue,
1664 .xlo_add_entry = ocfs2_xa_block_add_entry,
1665 .xlo_add_namevalue = ocfs2_xa_block_add_namevalue,
1666 .xlo_fill_value_buf = ocfs2_xa_block_fill_value_buf,
1669 static int ocfs2_xa_bucket_journal_access(handle_t *handle,
1670 struct ocfs2_xa_loc *loc, int type)
1672 struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1674 return ocfs2_xattr_bucket_journal_access(handle, bucket, type);
1677 static void ocfs2_xa_bucket_journal_dirty(handle_t *handle,
1678 struct ocfs2_xa_loc *loc)
1680 struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1682 ocfs2_xattr_bucket_journal_dirty(handle, bucket);
1685 static void *ocfs2_xa_bucket_offset_pointer(struct ocfs2_xa_loc *loc,
1688 struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1689 int block, block_offset;
1691 /* The header is at the front of the bucket */
1692 block = offset >> loc->xl_inode->i_sb->s_blocksize_bits;
1693 block_offset = offset % loc->xl_inode->i_sb->s_blocksize;
1695 return bucket_block(bucket, block) + block_offset;
1698 static int ocfs2_xa_bucket_can_reuse(struct ocfs2_xa_loc *loc,
1699 struct ocfs2_xattr_info *xi)
1701 return namevalue_size_xe(loc->xl_entry) >=
1702 namevalue_size_xi(xi);
1705 static int ocfs2_xa_bucket_get_free_start(struct ocfs2_xa_loc *loc)
1707 struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1708 return le16_to_cpu(bucket_xh(bucket)->xh_free_start);
1711 static int ocfs2_bucket_align_free_start(struct super_block *sb,
1712 int free_start, int size)
1715 * We need to make sure that the name+value pair fits within
1718 if (((free_start - size) >> sb->s_blocksize_bits) !=
1719 ((free_start - 1) >> sb->s_blocksize_bits))
1720 free_start -= free_start % sb->s_blocksize;
1725 static int ocfs2_xa_bucket_check_space(struct ocfs2_xa_loc *loc,
1726 struct ocfs2_xattr_info *xi)
1729 int count = le16_to_cpu(loc->xl_header->xh_count);
1730 int free_start = ocfs2_xa_get_free_start(loc);
1731 int needed_space = ocfs2_xi_entry_usage(xi);
1732 int size = namevalue_size_xi(xi);
1733 struct super_block *sb = loc->xl_inode->i_sb;
1736 * Bucket storage does not reclaim name+value pairs it cannot
1737 * reuse. They live as holes until the bucket fills, and then
1738 * the bucket is defragmented. However, the bucket can reclaim
1739 * the ocfs2_xattr_entry.
1741 if (loc->xl_entry) {
1742 /* Don't need space if we're reusing! */
1743 if (ocfs2_xa_can_reuse_entry(loc, xi))
1746 needed_space -= sizeof(struct ocfs2_xattr_entry);
1748 BUG_ON(needed_space < 0);
1750 if (free_start < size) {
1755 * First we check if it would fit in the first place.
1756 * Below, we align the free start to a block. This may
1757 * slide us below the minimum gap. By checking unaligned
1758 * first, we avoid that error.
1760 rc = ocfs2_xa_check_space_helper(needed_space, free_start,
1764 free_start = ocfs2_bucket_align_free_start(sb, free_start,
1767 return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1770 static void ocfs2_xa_bucket_wipe_namevalue(struct ocfs2_xa_loc *loc)
1772 le16_add_cpu(&loc->xl_header->xh_name_value_len,
1773 -namevalue_size_xe(loc->xl_entry));
1776 static void ocfs2_xa_bucket_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1778 struct ocfs2_xattr_header *xh = loc->xl_header;
1779 int count = le16_to_cpu(xh->xh_count);
1780 int low = 0, high = count - 1, tmp;
1781 struct ocfs2_xattr_entry *tmp_xe;
1784 * We keep buckets sorted by name_hash, so we need to find
1787 while (low <= high && count) {
1788 tmp = (low + high) / 2;
1789 tmp_xe = &xh->xh_entries[tmp];
1791 if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
1793 else if (name_hash < le32_to_cpu(tmp_xe->xe_name_hash))
1802 memmove(&xh->xh_entries[low + 1],
1803 &xh->xh_entries[low],
1804 ((count - low) * sizeof(struct ocfs2_xattr_entry)));
1806 le16_add_cpu(&xh->xh_count, 1);
1807 loc->xl_entry = &xh->xh_entries[low];
1808 memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1811 static void ocfs2_xa_bucket_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1813 int free_start = ocfs2_xa_get_free_start(loc);
1814 struct ocfs2_xattr_header *xh = loc->xl_header;
1815 struct super_block *sb = loc->xl_inode->i_sb;
1818 free_start = ocfs2_bucket_align_free_start(sb, free_start, size);
1819 nameval_offset = free_start - size;
1820 loc->xl_entry->xe_name_offset = cpu_to_le16(nameval_offset);
1821 xh->xh_free_start = cpu_to_le16(nameval_offset);
1822 le16_add_cpu(&xh->xh_name_value_len, size);
1826 static void ocfs2_xa_bucket_fill_value_buf(struct ocfs2_xa_loc *loc,
1827 struct ocfs2_xattr_value_buf *vb)
1829 struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1830 struct super_block *sb = loc->xl_inode->i_sb;
1831 int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1832 int size = namevalue_size_xe(loc->xl_entry);
1833 int block_offset = nameval_offset >> sb->s_blocksize_bits;
1835 /* Values are not allowed to straddle block boundaries */
1836 BUG_ON(block_offset !=
1837 ((nameval_offset + size - 1) >> sb->s_blocksize_bits));
1838 /* We expect the bucket to be filled in */
1839 BUG_ON(!bucket->bu_bhs[block_offset]);
1841 vb->vb_access = ocfs2_journal_access;
1842 vb->vb_bh = bucket->bu_bhs[block_offset];
1845 /* Operations for xattrs stored in buckets. */
1846 static const struct ocfs2_xa_loc_operations ocfs2_xa_bucket_loc_ops = {
1847 .xlo_journal_access = ocfs2_xa_bucket_journal_access,
1848 .xlo_journal_dirty = ocfs2_xa_bucket_journal_dirty,
1849 .xlo_offset_pointer = ocfs2_xa_bucket_offset_pointer,
1850 .xlo_check_space = ocfs2_xa_bucket_check_space,
1851 .xlo_can_reuse = ocfs2_xa_bucket_can_reuse,
1852 .xlo_get_free_start = ocfs2_xa_bucket_get_free_start,
1853 .xlo_wipe_namevalue = ocfs2_xa_bucket_wipe_namevalue,
1854 .xlo_add_entry = ocfs2_xa_bucket_add_entry,
1855 .xlo_add_namevalue = ocfs2_xa_bucket_add_namevalue,
1856 .xlo_fill_value_buf = ocfs2_xa_bucket_fill_value_buf,
1859 static unsigned int ocfs2_xa_value_clusters(struct ocfs2_xa_loc *loc)
1861 struct ocfs2_xattr_value_buf vb;
1863 if (ocfs2_xattr_is_local(loc->xl_entry))
1866 ocfs2_xa_fill_value_buf(loc, &vb);
1867 return le32_to_cpu(vb.vb_xv->xr_clusters);
1870 static int ocfs2_xa_value_truncate(struct ocfs2_xa_loc *loc, u64 bytes,
1871 struct ocfs2_xattr_set_ctxt *ctxt)
1873 int trunc_rc, access_rc;
1874 struct ocfs2_xattr_value_buf vb;
1876 ocfs2_xa_fill_value_buf(loc, &vb);
1877 trunc_rc = ocfs2_xattr_value_truncate(loc->xl_inode, &vb, bytes,
1881 * The caller of ocfs2_xa_value_truncate() has already called
1882 * ocfs2_xa_journal_access on the loc. However, The truncate code
1883 * calls ocfs2_extend_trans(). This may commit the previous
1884 * transaction and open a new one. If this is a bucket, truncate
1885 * could leave only vb->vb_bh set up for journaling. Meanwhile,
1886 * the caller is expecting to dirty the entire bucket. So we must
1887 * reset the journal work. We do this even if truncate has failed,
1888 * as it could have failed after committing the extend.
1890 access_rc = ocfs2_xa_journal_access(ctxt->handle, loc,
1891 OCFS2_JOURNAL_ACCESS_WRITE);
1893 /* Errors in truncate take precedence */
1894 return trunc_rc ? trunc_rc : access_rc;
1897 static void ocfs2_xa_remove_entry(struct ocfs2_xa_loc *loc)
1900 struct ocfs2_xattr_header *xh = loc->xl_header;
1901 struct ocfs2_xattr_entry *entry = loc->xl_entry;
1903 ocfs2_xa_wipe_namevalue(loc);
1904 loc->xl_entry = NULL;
1906 le16_add_cpu(&xh->xh_count, -1);
1907 count = le16_to_cpu(xh->xh_count);
1910 * Only zero out the entry if there are more remaining. This is
1911 * important for an empty bucket, as it keeps track of the
1912 * bucket's hash value. It doesn't hurt empty block storage.
1915 index = ((char *)entry - (char *)&xh->xh_entries) /
1916 sizeof(struct ocfs2_xattr_entry);
1917 memmove(&xh->xh_entries[index], &xh->xh_entries[index + 1],
1918 (count - index) * sizeof(struct ocfs2_xattr_entry));
1919 memset(&xh->xh_entries[count], 0,
1920 sizeof(struct ocfs2_xattr_entry));
1925 * If we have a problem adjusting the size of an external value during
1926 * ocfs2_xa_prepare_entry() or ocfs2_xa_remove(), we may have an xattr
1927 * in an intermediate state. For example, the value may be partially
1930 * If the value tree hasn't changed, the extend/truncate went nowhere.
1931 * We have nothing to do. The caller can treat it as a straight error.
1933 * If the value tree got partially truncated, we now have a corrupted
1934 * extended attribute. We're going to wipe its entry and leak the
1935 * clusters. Better to leak some storage than leave a corrupt entry.
1937 * If the value tree grew, it obviously didn't grow enough for the
1938 * new entry. We're not going to try and reclaim those clusters either.
1939 * If there was already an external value there (orig_clusters != 0),
1940 * the new clusters are attached safely and we can just leave the old
1941 * value in place. If there was no external value there, we remove
1944 * This way, the xattr block we store in the journal will be consistent.
1945 * If the size change broke because of the journal, no changes will hit
1948 static void ocfs2_xa_cleanup_value_truncate(struct ocfs2_xa_loc *loc,
1950 unsigned int orig_clusters)
1952 unsigned int new_clusters = ocfs2_xa_value_clusters(loc);
1953 char *nameval_buf = ocfs2_xa_offset_pointer(loc,
1954 le16_to_cpu(loc->xl_entry->xe_name_offset));
1956 if (new_clusters < orig_clusters) {
1958 "Partial truncate while %s xattr %.*s. Leaking "
1959 "%u clusters and removing the entry\n",
1960 what, loc->xl_entry->xe_name_len, nameval_buf,
1961 orig_clusters - new_clusters);
1962 ocfs2_xa_remove_entry(loc);
1963 } else if (!orig_clusters) {
1965 "Unable to allocate an external value for xattr "
1966 "%.*s safely. Leaking %u clusters and removing the "
1968 loc->xl_entry->xe_name_len, nameval_buf,
1969 new_clusters - orig_clusters);
1970 ocfs2_xa_remove_entry(loc);
1971 } else if (new_clusters > orig_clusters)
1973 "Unable to grow xattr %.*s safely. %u new clusters "
1974 "have been added, but the value will not be "
1976 loc->xl_entry->xe_name_len, nameval_buf,
1977 new_clusters - orig_clusters);
1980 static int ocfs2_xa_remove(struct ocfs2_xa_loc *loc,
1981 struct ocfs2_xattr_set_ctxt *ctxt)
1984 unsigned int orig_clusters;
1986 if (!ocfs2_xattr_is_local(loc->xl_entry)) {
1987 orig_clusters = ocfs2_xa_value_clusters(loc);
1988 rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
1992 * Since this is remove, we can return 0 if
1993 * ocfs2_xa_cleanup_value_truncate() is going to
1994 * wipe the entry anyway. So we check the
1995 * cluster count as well.
1997 if (orig_clusters != ocfs2_xa_value_clusters(loc))
1999 ocfs2_xa_cleanup_value_truncate(loc, "removing",
2006 ocfs2_xa_remove_entry(loc);
2012 static void ocfs2_xa_install_value_root(struct ocfs2_xa_loc *loc)
2014 int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
2017 nameval_buf = ocfs2_xa_offset_pointer(loc,
2018 le16_to_cpu(loc->xl_entry->xe_name_offset));
2019 memcpy(nameval_buf + name_size, &def_xv, OCFS2_XATTR_ROOT_SIZE);
2023 * Take an existing entry and make it ready for the new value. This
2024 * won't allocate space, but it may free space. It should be ready for
2025 * ocfs2_xa_prepare_entry() to finish the work.
2027 static int ocfs2_xa_reuse_entry(struct ocfs2_xa_loc *loc,
2028 struct ocfs2_xattr_info *xi,
2029 struct ocfs2_xattr_set_ctxt *ctxt)
2032 int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2033 unsigned int orig_clusters;
2035 int xe_local = ocfs2_xattr_is_local(loc->xl_entry);
2036 int xi_local = xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE;
2038 BUG_ON(OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len) !=
2041 nameval_buf = ocfs2_xa_offset_pointer(loc,
2042 le16_to_cpu(loc->xl_entry->xe_name_offset));
2044 memset(nameval_buf + name_size, 0,
2045 namevalue_size_xe(loc->xl_entry) - name_size);
2047 ocfs2_xa_install_value_root(loc);
2049 orig_clusters = ocfs2_xa_value_clusters(loc);
2051 rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2055 memset(nameval_buf + name_size, 0,
2056 namevalue_size_xe(loc->xl_entry) -
2058 } else if (le64_to_cpu(loc->xl_entry->xe_value_size) >
2060 rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len,
2067 ocfs2_xa_cleanup_value_truncate(loc, "reusing",
2073 loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
2074 ocfs2_xattr_set_local(loc->xl_entry, xi_local);
2081 * Prepares loc->xl_entry to receive the new xattr. This includes
2082 * properly setting up the name+value pair region. If loc->xl_entry
2083 * already exists, it will take care of modifying it appropriately.
2085 * Note that this modifies the data. You did journal_access already,
2088 static int ocfs2_xa_prepare_entry(struct ocfs2_xa_loc *loc,
2089 struct ocfs2_xattr_info *xi,
2091 struct ocfs2_xattr_set_ctxt *ctxt)
2094 unsigned int orig_clusters;
2095 __le64 orig_value_size = 0;
2097 rc = ocfs2_xa_check_space(loc, xi);
2101 if (loc->xl_entry) {
2102 if (ocfs2_xa_can_reuse_entry(loc, xi)) {
2103 orig_value_size = loc->xl_entry->xe_value_size;
2104 rc = ocfs2_xa_reuse_entry(loc, xi, ctxt);
2110 if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2111 orig_clusters = ocfs2_xa_value_clusters(loc);
2112 rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2115 ocfs2_xa_cleanup_value_truncate(loc,
2121 ocfs2_xa_wipe_namevalue(loc);
2123 ocfs2_xa_add_entry(loc, name_hash);
2126 * If we get here, we have a blank entry. Fill it. We grow our
2127 * name+value pair back from the end.
2129 ocfs2_xa_add_namevalue(loc, xi);
2130 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
2131 ocfs2_xa_install_value_root(loc);
2134 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2135 orig_clusters = ocfs2_xa_value_clusters(loc);
2136 rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len, ctxt);
2139 * If we tried to grow an existing external value,
2140 * ocfs2_xa_cleanuP-value_truncate() is going to
2141 * let it stand. We have to restore its original
2144 loc->xl_entry->xe_value_size = orig_value_size;
2145 ocfs2_xa_cleanup_value_truncate(loc, "growing",
2156 * Store the value portion of the name+value pair. This will skip
2157 * values that are stored externally. Their tree roots were set up
2158 * by ocfs2_xa_prepare_entry().
2160 static int ocfs2_xa_store_value(struct ocfs2_xa_loc *loc,
2161 struct ocfs2_xattr_info *xi,
2162 struct ocfs2_xattr_set_ctxt *ctxt)
2165 int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
2166 int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2168 struct ocfs2_xattr_value_buf vb;
2170 nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
2171 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2172 ocfs2_xa_fill_value_buf(loc, &vb);
2173 rc = __ocfs2_xattr_set_value_outside(loc->xl_inode,
2178 memcpy(nameval_buf + name_size, xi->xi_value, xi->xi_value_len);
2183 static int ocfs2_xa_set(struct ocfs2_xa_loc *loc,
2184 struct ocfs2_xattr_info *xi,
2185 struct ocfs2_xattr_set_ctxt *ctxt)
2188 u32 name_hash = ocfs2_xattr_name_hash(loc->xl_inode, xi->xi_name,
2191 ret = ocfs2_xa_journal_access(ctxt->handle, loc,
2192 OCFS2_JOURNAL_ACCESS_WRITE);
2199 * From here on out, everything is going to modify the buffer a
2200 * little. Errors are going to leave the xattr header in a
2201 * sane state. Thus, even with errors we dirty the sucker.
2204 /* Don't worry, we are never called with !xi_value and !xl_entry */
2205 if (!xi->xi_value) {
2206 ret = ocfs2_xa_remove(loc, ctxt);
2210 ret = ocfs2_xa_prepare_entry(loc, xi, name_hash, ctxt);
2217 ret = ocfs2_xa_store_value(loc, xi, ctxt);
2222 ocfs2_xa_journal_dirty(ctxt->handle, loc);
2228 static void ocfs2_init_dinode_xa_loc(struct ocfs2_xa_loc *loc,
2229 struct inode *inode,
2230 struct buffer_head *bh,
2231 struct ocfs2_xattr_entry *entry)
2233 struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2235 BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_XATTR_FL));
2237 loc->xl_inode = inode;
2238 loc->xl_ops = &ocfs2_xa_block_loc_ops;
2239 loc->xl_storage = bh;
2240 loc->xl_entry = entry;
2241 loc->xl_size = le16_to_cpu(di->i_xattr_inline_size);
2243 (struct ocfs2_xattr_header *)(bh->b_data + bh->b_size -
2247 static void ocfs2_init_xattr_block_xa_loc(struct ocfs2_xa_loc *loc,
2248 struct inode *inode,
2249 struct buffer_head *bh,
2250 struct ocfs2_xattr_entry *entry)
2252 struct ocfs2_xattr_block *xb =
2253 (struct ocfs2_xattr_block *)bh->b_data;
2255 BUG_ON(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED);
2257 loc->xl_inode = inode;
2258 loc->xl_ops = &ocfs2_xa_block_loc_ops;
2259 loc->xl_storage = bh;
2260 loc->xl_header = &(xb->xb_attrs.xb_header);
2261 loc->xl_entry = entry;
2262 loc->xl_size = bh->b_size - offsetof(struct ocfs2_xattr_block,
2263 xb_attrs.xb_header);
2266 static void ocfs2_init_xattr_bucket_xa_loc(struct ocfs2_xa_loc *loc,
2267 struct ocfs2_xattr_bucket *bucket,
2268 struct ocfs2_xattr_entry *entry)
2270 loc->xl_inode = bucket->bu_inode;
2271 loc->xl_ops = &ocfs2_xa_bucket_loc_ops;
2272 loc->xl_storage = bucket;
2273 loc->xl_header = bucket_xh(bucket);
2274 loc->xl_entry = entry;
2275 loc->xl_size = OCFS2_XATTR_BUCKET_SIZE;
2279 * In xattr remove, if it is stored outside and refcounted, we may have
2280 * the chance to split the refcount tree. So need the allocators.
2282 static int ocfs2_lock_xattr_remove_allocators(struct inode *inode,
2283 struct ocfs2_xattr_value_root *xv,
2284 struct ocfs2_caching_info *ref_ci,
2285 struct buffer_head *ref_root_bh,
2286 struct ocfs2_alloc_context **meta_ac,
2289 int ret, meta_add = 0;
2290 u32 p_cluster, num_clusters;
2291 unsigned int ext_flags;
2294 ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
2303 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
2306 ret = ocfs2_refcounted_xattr_delete_need(inode, ref_ci,
2308 &meta_add, ref_credits);
2314 ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
2323 static int ocfs2_remove_value_outside(struct inode*inode,
2324 struct ocfs2_xattr_value_buf *vb,
2325 struct ocfs2_xattr_header *header,
2326 struct ocfs2_caching_info *ref_ci,
2327 struct buffer_head *ref_root_bh)
2329 int ret = 0, i, ref_credits;
2330 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2331 struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2334 ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
2336 for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
2337 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
2339 if (ocfs2_xattr_is_local(entry))
2342 val = (void *)header +
2343 le16_to_cpu(entry->xe_name_offset);
2344 vb->vb_xv = (struct ocfs2_xattr_value_root *)
2345 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
2347 ret = ocfs2_lock_xattr_remove_allocators(inode, vb->vb_xv,
2348 ref_ci, ref_root_bh,
2352 ctxt.handle = ocfs2_start_trans(osb, ref_credits +
2353 ocfs2_remove_extent_credits(osb->sb));
2354 if (IS_ERR(ctxt.handle)) {
2355 ret = PTR_ERR(ctxt.handle);
2360 ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
2366 ocfs2_commit_trans(osb, ctxt.handle);
2368 ocfs2_free_alloc_context(ctxt.meta_ac);
2369 ctxt.meta_ac = NULL;
2374 ocfs2_free_alloc_context(ctxt.meta_ac);
2375 ocfs2_schedule_truncate_log_flush(osb, 1);
2376 ocfs2_run_deallocs(osb, &ctxt.dealloc);
2380 static int ocfs2_xattr_ibody_remove(struct inode *inode,
2381 struct buffer_head *di_bh,
2382 struct ocfs2_caching_info *ref_ci,
2383 struct buffer_head *ref_root_bh)
2386 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2387 struct ocfs2_xattr_header *header;
2389 struct ocfs2_xattr_value_buf vb = {
2391 .vb_access = ocfs2_journal_access_di,
2394 header = (struct ocfs2_xattr_header *)
2395 ((void *)di + inode->i_sb->s_blocksize -
2396 le16_to_cpu(di->i_xattr_inline_size));
2398 ret = ocfs2_remove_value_outside(inode, &vb, header,
2399 ref_ci, ref_root_bh);
2404 struct ocfs2_rm_xattr_bucket_para {
2405 struct ocfs2_caching_info *ref_ci;
2406 struct buffer_head *ref_root_bh;
2409 static int ocfs2_xattr_block_remove(struct inode *inode,
2410 struct buffer_head *blk_bh,
2411 struct ocfs2_caching_info *ref_ci,
2412 struct buffer_head *ref_root_bh)
2414 struct ocfs2_xattr_block *xb;
2416 struct ocfs2_xattr_value_buf vb = {
2418 .vb_access = ocfs2_journal_access_xb,
2420 struct ocfs2_rm_xattr_bucket_para args = {
2422 .ref_root_bh = ref_root_bh,
2425 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2426 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2427 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
2428 ret = ocfs2_remove_value_outside(inode, &vb, header,
2429 ref_ci, ref_root_bh);
2431 ret = ocfs2_iterate_xattr_index_block(inode,
2433 ocfs2_rm_xattr_cluster,
2439 static int ocfs2_xattr_free_block(struct inode *inode,
2441 struct ocfs2_caching_info *ref_ci,
2442 struct buffer_head *ref_root_bh)
2444 struct inode *xb_alloc_inode;
2445 struct buffer_head *xb_alloc_bh = NULL;
2446 struct buffer_head *blk_bh = NULL;
2447 struct ocfs2_xattr_block *xb;
2448 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2454 ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
2460 ret = ocfs2_xattr_block_remove(inode, blk_bh, ref_ci, ref_root_bh);
2466 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2467 blk = le64_to_cpu(xb->xb_blkno);
2468 bit = le16_to_cpu(xb->xb_suballoc_bit);
2469 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2471 xb_alloc_inode = ocfs2_get_system_file_inode(osb,
2472 EXTENT_ALLOC_SYSTEM_INODE,
2473 le16_to_cpu(xb->xb_suballoc_slot));
2474 if (!xb_alloc_inode) {
2479 mutex_lock(&xb_alloc_inode->i_mutex);
2481 ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
2487 handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
2488 if (IS_ERR(handle)) {
2489 ret = PTR_ERR(handle);
2494 ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
2499 ocfs2_commit_trans(osb, handle);
2501 ocfs2_inode_unlock(xb_alloc_inode, 1);
2502 brelse(xb_alloc_bh);
2504 mutex_unlock(&xb_alloc_inode->i_mutex);
2505 iput(xb_alloc_inode);
2512 * ocfs2_xattr_remove()
2514 * Free extended attribute resources associated with this inode.
2516 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
2518 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2519 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2520 struct ocfs2_refcount_tree *ref_tree = NULL;
2521 struct buffer_head *ref_root_bh = NULL;
2522 struct ocfs2_caching_info *ref_ci = NULL;
2526 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2529 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
2532 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
2533 ret = ocfs2_lock_refcount_tree(OCFS2_SB(inode->i_sb),
2534 le64_to_cpu(di->i_refcount_loc),
2535 1, &ref_tree, &ref_root_bh);
2540 ref_ci = &ref_tree->rf_ci;
2544 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2545 ret = ocfs2_xattr_ibody_remove(inode, di_bh,
2546 ref_ci, ref_root_bh);
2553 if (di->i_xattr_loc) {
2554 ret = ocfs2_xattr_free_block(inode,
2555 le64_to_cpu(di->i_xattr_loc),
2556 ref_ci, ref_root_bh);
2563 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
2564 OCFS2_INODE_UPDATE_CREDITS);
2565 if (IS_ERR(handle)) {
2566 ret = PTR_ERR(handle);
2570 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2571 OCFS2_JOURNAL_ACCESS_WRITE);
2577 di->i_xattr_loc = 0;
2579 spin_lock(&oi->ip_lock);
2580 oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
2581 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2582 spin_unlock(&oi->ip_lock);
2584 ocfs2_journal_dirty(handle, di_bh);
2586 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
2589 ocfs2_unlock_refcount_tree(OCFS2_SB(inode->i_sb), ref_tree, 1);
2590 brelse(ref_root_bh);
2594 static int ocfs2_xattr_has_space_inline(struct inode *inode,
2595 struct ocfs2_dinode *di)
2597 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2598 unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
2601 if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
2604 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2605 struct ocfs2_inline_data *idata = &di->id2.i_data;
2606 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
2607 } else if (ocfs2_inode_is_fast_symlink(inode)) {
2608 free = ocfs2_fast_symlink_chars(inode->i_sb) -
2609 le64_to_cpu(di->i_size);
2611 struct ocfs2_extent_list *el = &di->id2.i_list;
2612 free = (le16_to_cpu(el->l_count) -
2613 le16_to_cpu(el->l_next_free_rec)) *
2614 sizeof(struct ocfs2_extent_rec);
2616 if (free >= xattrsize)
2623 * ocfs2_xattr_ibody_find()
2625 * Find extended attribute in inode block and
2626 * fill search info into struct ocfs2_xattr_search.
2628 static int ocfs2_xattr_ibody_find(struct inode *inode,
2631 struct ocfs2_xattr_search *xs)
2633 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2634 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2638 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2641 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2642 down_read(&oi->ip_alloc_sem);
2643 has_space = ocfs2_xattr_has_space_inline(inode, di);
2644 up_read(&oi->ip_alloc_sem);
2649 xs->xattr_bh = xs->inode_bh;
2650 xs->end = (void *)di + inode->i_sb->s_blocksize;
2651 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
2652 xs->header = (struct ocfs2_xattr_header *)
2653 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
2655 xs->header = (struct ocfs2_xattr_header *)
2656 (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
2657 xs->base = (void *)xs->header;
2658 xs->here = xs->header->xh_entries;
2660 /* Find the named attribute. */
2661 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2662 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2663 if (ret && ret != -ENODATA)
2665 xs->not_found = ret;
2671 static int ocfs2_xattr_ibody_init(struct inode *inode,
2672 struct buffer_head *di_bh,
2673 struct ocfs2_xattr_set_ctxt *ctxt)
2676 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2677 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2678 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2679 unsigned int xattrsize = osb->s_xattr_inline_size;
2681 if (!ocfs2_xattr_has_space_inline(inode, di)) {
2686 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode), di_bh,
2687 OCFS2_JOURNAL_ACCESS_WRITE);
2694 * Adjust extent record count or inline data size
2695 * to reserve space for extended attribute.
2697 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2698 struct ocfs2_inline_data *idata = &di->id2.i_data;
2699 le16_add_cpu(&idata->id_count, -xattrsize);
2700 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
2701 struct ocfs2_extent_list *el = &di->id2.i_list;
2702 le16_add_cpu(&el->l_count, -(xattrsize /
2703 sizeof(struct ocfs2_extent_rec)));
2705 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
2707 spin_lock(&oi->ip_lock);
2708 oi->ip_dyn_features |= OCFS2_INLINE_XATTR_FL|OCFS2_HAS_XATTR_FL;
2709 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2710 spin_unlock(&oi->ip_lock);
2712 ocfs2_journal_dirty(ctxt->handle, di_bh);
2719 * ocfs2_xattr_ibody_set()
2721 * Set, replace or remove an extended attribute into inode block.
2724 static int ocfs2_xattr_ibody_set(struct inode *inode,
2725 struct ocfs2_xattr_info *xi,
2726 struct ocfs2_xattr_search *xs,
2727 struct ocfs2_xattr_set_ctxt *ctxt)
2730 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2731 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2732 struct ocfs2_xa_loc loc;
2734 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2737 down_write(&oi->ip_alloc_sem);
2738 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2739 if (!ocfs2_xattr_has_space_inline(inode, di)) {
2745 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2746 ret = ocfs2_xattr_ibody_init(inode, xs->inode_bh, ctxt);
2754 ocfs2_init_dinode_xa_loc(&loc, inode, xs->inode_bh,
2755 xs->not_found ? NULL : xs->here);
2756 ret = ocfs2_xa_set(&loc, xi, ctxt);
2762 xs->here = loc.xl_entry;
2765 up_write(&oi->ip_alloc_sem);
2771 * ocfs2_xattr_block_find()
2773 * Find extended attribute in external block and
2774 * fill search info into struct ocfs2_xattr_search.
2776 static int ocfs2_xattr_block_find(struct inode *inode,
2779 struct ocfs2_xattr_search *xs)
2781 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2782 struct buffer_head *blk_bh = NULL;
2783 struct ocfs2_xattr_block *xb;
2786 if (!di->i_xattr_loc)
2789 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2796 xs->xattr_bh = blk_bh;
2797 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2799 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2800 xs->header = &xb->xb_attrs.xb_header;
2801 xs->base = (void *)xs->header;
2802 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2803 xs->here = xs->header->xh_entries;
2805 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2807 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2811 if (ret && ret != -ENODATA) {
2812 xs->xattr_bh = NULL;
2815 xs->not_found = ret;
2823 static int ocfs2_create_xattr_block(struct inode *inode,
2824 struct buffer_head *inode_bh,
2825 struct ocfs2_xattr_set_ctxt *ctxt,
2827 struct buffer_head **ret_bh)
2830 u16 suballoc_bit_start;
2833 struct ocfs2_dinode *di = (struct ocfs2_dinode *)inode_bh->b_data;
2834 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2835 struct buffer_head *new_bh = NULL;
2836 struct ocfs2_xattr_block *xblk;
2838 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
2839 inode_bh, OCFS2_JOURNAL_ACCESS_CREATE);
2845 ret = ocfs2_claim_metadata(osb, ctxt->handle, ctxt->meta_ac, 1,
2846 &suballoc_bit_start, &num_got,
2853 new_bh = sb_getblk(inode->i_sb, first_blkno);
2854 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2856 ret = ocfs2_journal_access_xb(ctxt->handle, INODE_CACHE(inode),
2858 OCFS2_JOURNAL_ACCESS_CREATE);
2864 /* Initialize ocfs2_xattr_block */
2865 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2866 memset(xblk, 0, inode->i_sb->s_blocksize);
2867 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2868 xblk->xb_suballoc_slot = cpu_to_le16(ctxt->meta_ac->ac_alloc_slot);
2869 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2870 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
2871 xblk->xb_blkno = cpu_to_le64(first_blkno);
2873 struct ocfs2_xattr_tree_root *xr = &xblk->xb_attrs.xb_root;
2874 xr->xt_clusters = cpu_to_le32(1);
2875 xr->xt_last_eb_blk = 0;
2876 xr->xt_list.l_tree_depth = 0;
2877 xr->xt_list.l_count = cpu_to_le16(
2878 ocfs2_xattr_recs_per_xb(inode->i_sb));
2879 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2880 xblk->xb_flags = cpu_to_le16(OCFS2_XATTR_INDEXED);
2882 ocfs2_journal_dirty(ctxt->handle, new_bh);
2884 /* Add it to the inode */
2885 di->i_xattr_loc = cpu_to_le64(first_blkno);
2887 spin_lock(&OCFS2_I(inode)->ip_lock);
2888 OCFS2_I(inode)->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
2889 di->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
2890 spin_unlock(&OCFS2_I(inode)->ip_lock);
2892 ocfs2_journal_dirty(ctxt->handle, inode_bh);
2903 * ocfs2_xattr_block_set()
2905 * Set, replace or remove an extended attribute into external block.
2908 static int ocfs2_xattr_block_set(struct inode *inode,
2909 struct ocfs2_xattr_info *xi,
2910 struct ocfs2_xattr_search *xs,
2911 struct ocfs2_xattr_set_ctxt *ctxt)
2913 struct buffer_head *new_bh = NULL;
2914 struct ocfs2_xattr_block *xblk = NULL;
2916 struct ocfs2_xa_loc loc;
2918 if (!xs->xattr_bh) {
2919 ret = ocfs2_create_xattr_block(inode, xs->inode_bh, ctxt,
2926 xs->xattr_bh = new_bh;
2927 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2928 xs->header = &xblk->xb_attrs.xb_header;
2929 xs->base = (void *)xs->header;
2930 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2931 xs->here = xs->header->xh_entries;
2933 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2935 if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2936 ocfs2_init_xattr_block_xa_loc(&loc, inode, xs->xattr_bh,
2937 xs->not_found ? NULL : xs->here);
2939 ret = ocfs2_xa_set(&loc, xi, ctxt);
2941 xs->here = loc.xl_entry;
2942 else if (ret != -ENOSPC)
2945 ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2951 if (le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)
2952 ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2958 /* Check whether the new xattr can be inserted into the inode. */
2959 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2960 struct ocfs2_xattr_info *xi,
2961 struct ocfs2_xattr_search *xs)
2963 struct ocfs2_xattr_entry *last;
2965 size_t min_offs = xs->end - xs->base;
2970 last = xs->header->xh_entries;
2972 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
2973 size_t offs = le16_to_cpu(last->xe_name_offset);
2974 if (offs < min_offs)
2979 free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
2983 BUG_ON(!xs->not_found);
2985 if (free >= (sizeof(struct ocfs2_xattr_entry) + namevalue_size_xi(xi)))
2991 static int ocfs2_calc_xattr_set_need(struct inode *inode,
2992 struct ocfs2_dinode *di,
2993 struct ocfs2_xattr_info *xi,
2994 struct ocfs2_xattr_search *xis,
2995 struct ocfs2_xattr_search *xbs,
3000 int ret = 0, old_in_xb = 0;
3001 int clusters_add = 0, meta_add = 0, credits = 0;
3002 struct buffer_head *bh = NULL;
3003 struct ocfs2_xattr_block *xb = NULL;
3004 struct ocfs2_xattr_entry *xe = NULL;
3005 struct ocfs2_xattr_value_root *xv = NULL;
3007 int name_offset, name_len = 0;
3008 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
3013 * Calculate the clusters we need to write.
3014 * No matter whether we replace an old one or add a new one,
3015 * we need this for writing.
3017 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
3018 credits += new_clusters *
3019 ocfs2_clusters_to_blocks(inode->i_sb, 1);
3021 if (xis->not_found && xbs->not_found) {
3022 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3024 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3025 clusters_add += new_clusters;
3026 credits += ocfs2_calc_extend_credits(inode->i_sb,
3034 if (!xis->not_found) {
3036 name_offset = le16_to_cpu(xe->xe_name_offset);
3037 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3039 credits += OCFS2_INODE_UPDATE_CREDITS;
3041 int i, block_off = 0;
3042 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3044 name_offset = le16_to_cpu(xe->xe_name_offset);
3045 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3046 i = xbs->here - xbs->header->xh_entries;
3049 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3050 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3051 bucket_xh(xbs->bucket),
3054 base = bucket_block(xbs->bucket, block_off);
3055 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3058 credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
3063 * delete a xattr doesn't need metadata and cluster allocation.
3064 * so just calculate the credits and return.
3066 * The credits for removing the value tree will be extended
3067 * by ocfs2_remove_extent itself.
3069 if (!xi->xi_value) {
3070 if (!ocfs2_xattr_is_local(xe))
3071 credits += ocfs2_remove_extent_credits(inode->i_sb);
3076 /* do cluster allocation guess first. */
3077 value_size = le64_to_cpu(xe->xe_value_size);
3081 * In xattr set, we always try to set the xe in inode first,
3082 * so if it can be inserted into inode successfully, the old
3083 * one will be removed from the xattr block, and this xattr
3084 * will be inserted into inode as a new xattr in inode.
3086 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
3087 clusters_add += new_clusters;
3088 credits += ocfs2_remove_extent_credits(inode->i_sb) +
3089 OCFS2_INODE_UPDATE_CREDITS;
3090 if (!ocfs2_xattr_is_local(xe))
3091 credits += ocfs2_calc_extend_credits(
3099 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3100 /* the new values will be stored outside. */
3101 u32 old_clusters = 0;
3103 if (!ocfs2_xattr_is_local(xe)) {
3104 old_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
3106 xv = (struct ocfs2_xattr_value_root *)
3107 (base + name_offset + name_len);
3108 value_size = OCFS2_XATTR_ROOT_SIZE;
3112 if (old_clusters >= new_clusters) {
3113 credits += ocfs2_remove_extent_credits(inode->i_sb);
3116 meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
3117 clusters_add += new_clusters - old_clusters;
3118 credits += ocfs2_calc_extend_credits(inode->i_sb,
3122 if (value_size >= OCFS2_XATTR_ROOT_SIZE)
3127 * Now the new value will be stored inside. So if the new
3128 * value is smaller than the size of value root or the old
3129 * value, we don't need any allocation, otherwise we have
3130 * to guess metadata allocation.
3132 if ((ocfs2_xattr_is_local(xe) &&
3133 (value_size >= xi->xi_value_len)) ||
3134 (!ocfs2_xattr_is_local(xe) &&
3135 OCFS2_XATTR_ROOT_SIZE >= xi->xi_value_len))
3140 /* calculate metadata allocation. */
3141 if (di->i_xattr_loc) {
3142 if (!xbs->xattr_bh) {
3143 ret = ocfs2_read_xattr_block(inode,
3144 le64_to_cpu(di->i_xattr_loc),
3151 xb = (struct ocfs2_xattr_block *)bh->b_data;
3153 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3156 * If there is already an xattr tree, good, we can calculate
3157 * like other b-trees. Otherwise we may have the chance of
3158 * create a tree, the credit calculation is borrowed from
3159 * ocfs2_calc_extend_credits with root_el = NULL. And the
3160 * new tree will be cluster based, so no meta is needed.
3162 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3163 struct ocfs2_extent_list *el =
3164 &xb->xb_attrs.xb_root.xt_list;
3165 meta_add += ocfs2_extend_meta_needed(el);
3166 credits += ocfs2_calc_extend_credits(inode->i_sb,
3169 credits += OCFS2_SUBALLOC_ALLOC + 1;
3172 * This cluster will be used either for new bucket or for
3174 * If the cluster size is the same as the bucket size, one
3175 * more is needed since we may need to extend the bucket
3179 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3180 if (OCFS2_XATTR_BUCKET_SIZE ==
3181 OCFS2_SB(inode->i_sb)->s_clustersize) {
3182 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3187 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
3191 *clusters_need = clusters_add;
3193 *meta_need = meta_add;
3195 *credits_need = credits;
3200 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
3201 struct ocfs2_dinode *di,
3202 struct ocfs2_xattr_info *xi,
3203 struct ocfs2_xattr_search *xis,
3204 struct ocfs2_xattr_search *xbs,
3205 struct ocfs2_xattr_set_ctxt *ctxt,
3209 int clusters_add, meta_add, ret;
3210 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3212 memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
3214 ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
3216 ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
3217 &clusters_add, &meta_add, credits);
3223 meta_add += extra_meta;
3224 mlog(0, "Set xattr %s, reserve meta blocks = %d, clusters = %d, "
3225 "credits = %d\n", xi->xi_name, meta_add, clusters_add, *credits);
3228 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
3237 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
3243 if (ctxt->meta_ac) {
3244 ocfs2_free_alloc_context(ctxt->meta_ac);
3245 ctxt->meta_ac = NULL;
3249 * We cannot have an error and a non null ctxt->data_ac.
3256 static int __ocfs2_xattr_set_handle(struct inode *inode,
3257 struct ocfs2_dinode *di,
3258 struct ocfs2_xattr_info *xi,
3259 struct ocfs2_xattr_search *xis,
3260 struct ocfs2_xattr_search *xbs,
3261 struct ocfs2_xattr_set_ctxt *ctxt)
3263 int ret = 0, credits, old_found;
3265 if (!xi->xi_value) {
3266 /* Remove existing extended attribute */
3267 if (!xis->not_found)
3268 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3269 else if (!xbs->not_found)
3270 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3272 /* We always try to set extended attribute into inode first*/
3273 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3274 if (!ret && !xbs->not_found) {
3276 * If succeed and that extended attribute existing in
3277 * external block, then we will remove it.
3279 xi->xi_value = NULL;
3280 xi->xi_value_len = 0;
3282 old_found = xis->not_found;
3283 xis->not_found = -ENODATA;
3284 ret = ocfs2_calc_xattr_set_need(inode,
3292 xis->not_found = old_found;
3298 ret = ocfs2_extend_trans(ctxt->handle, credits);
3303 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3304 } else if (ret == -ENOSPC) {
3305 if (di->i_xattr_loc && !xbs->xattr_bh) {
3306 ret = ocfs2_xattr_block_find(inode,
3312 old_found = xis->not_found;
3313 xis->not_found = -ENODATA;
3314 ret = ocfs2_calc_xattr_set_need(inode,
3322 xis->not_found = old_found;
3328 ret = ocfs2_extend_trans(ctxt->handle, credits);
3335 * If no space in inode, we will set extended attribute
3336 * into external block.
3338 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3341 if (!xis->not_found) {
3343 * If succeed and that extended attribute
3344 * existing in inode, we will remove it.
3346 xi->xi_value = NULL;
3347 xi->xi_value_len = 0;
3348 xbs->not_found = -ENODATA;
3349 ret = ocfs2_calc_xattr_set_need(inode,
3362 ret = ocfs2_extend_trans(ctxt->handle, credits);
3367 ret = ocfs2_xattr_ibody_set(inode, xi,
3374 /* Update inode ctime. */
3375 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
3377 OCFS2_JOURNAL_ACCESS_WRITE);
3383 inode->i_ctime = CURRENT_TIME;
3384 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
3385 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
3386 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
3393 * This function only called duing creating inode
3394 * for init security/acl xattrs of the new inode.
3395 * All transanction credits have been reserved in mknod.
3397 int ocfs2_xattr_set_handle(handle_t *handle,
3398 struct inode *inode,
3399 struct buffer_head *di_bh,
3405 struct ocfs2_alloc_context *meta_ac,
3406 struct ocfs2_alloc_context *data_ac)
3408 struct ocfs2_dinode *di;
3411 struct ocfs2_xattr_info xi = {
3412 .xi_name_index = name_index,
3414 .xi_name_len = strlen(name),
3416 .xi_value_len = value_len,
3419 struct ocfs2_xattr_search xis = {
3420 .not_found = -ENODATA,
3423 struct ocfs2_xattr_search xbs = {
3424 .not_found = -ENODATA,
3427 struct ocfs2_xattr_set_ctxt ctxt = {
3433 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3437 * In extreme situation, may need xattr bucket when
3438 * block size is too small. And we have already reserved
3439 * the credits for bucket in mknod.
3441 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
3442 xbs.bucket = ocfs2_xattr_bucket_new(inode);
3444 mlog_errno(-ENOMEM);
3449 xis.inode_bh = xbs.inode_bh = di_bh;
3450 di = (struct ocfs2_dinode *)di_bh->b_data;
3452 down_write(&OCFS2_I(inode)->ip_xattr_sem);
3454 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3457 if (xis.not_found) {
3458 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3463 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3466 up_write(&OCFS2_I(inode)->ip_xattr_sem);
3467 brelse(xbs.xattr_bh);
3468 ocfs2_xattr_bucket_free(xbs.bucket);
3476 * Set, replace or remove an extended attribute for this inode.
3477 * value is NULL to remove an existing extended attribute, else either
3478 * create or replace an extended attribute.
3480 int ocfs2_xattr_set(struct inode *inode,
3487 struct buffer_head *di_bh = NULL;
3488 struct ocfs2_dinode *di;
3489 int ret, credits, ref_meta = 0, ref_credits = 0;
3490 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3491 struct inode *tl_inode = osb->osb_tl_inode;
3492 struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
3493 struct ocfs2_refcount_tree *ref_tree = NULL;
3495 struct ocfs2_xattr_info xi = {
3496 .xi_name_index = name_index,
3498 .xi_name_len = strlen(name),
3500 .xi_value_len = value_len,
3503 struct ocfs2_xattr_search xis = {
3504 .not_found = -ENODATA,
3507 struct ocfs2_xattr_search xbs = {
3508 .not_found = -ENODATA,
3511 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3515 * Only xbs will be used on indexed trees. xis doesn't need a
3518 xbs.bucket = ocfs2_xattr_bucket_new(inode);
3520 mlog_errno(-ENOMEM);
3524 ret = ocfs2_inode_lock(inode, &di_bh, 1);
3527 goto cleanup_nolock;
3529 xis.inode_bh = xbs.inode_bh = di_bh;
3530 di = (struct ocfs2_dinode *)di_bh->b_data;
3532 down_write(&OCFS2_I(inode)->ip_xattr_sem);
3534 * Scan inode and external block to find the same name
3535 * extended attribute and collect search infomation.
3537 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3540 if (xis.not_found) {
3541 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3546 if (xis.not_found && xbs.not_found) {
3548 if (flags & XATTR_REPLACE)
3555 if (flags & XATTR_CREATE)
3559 /* Check whether the value is refcounted and do some prepartion. */
3560 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL &&
3561 (!xis.not_found || !xbs.not_found)) {
3562 ret = ocfs2_prepare_refcount_xattr(inode, di, &xi,
3563 &xis, &xbs, &ref_tree,
3564 &ref_meta, &ref_credits);
3571 mutex_lock(&tl_inode->i_mutex);
3573 if (ocfs2_truncate_log_needs_flush(osb)) {
3574 ret = __ocfs2_flush_truncate_log(osb);
3576 mutex_unlock(&tl_inode->i_mutex);
3581 mutex_unlock(&tl_inode->i_mutex);
3583 ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
3584 &xbs, &ctxt, ref_meta, &credits);
3590 /* we need to update inode's ctime field, so add credit for it. */
3591 credits += OCFS2_INODE_UPDATE_CREDITS;
3592 ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
3593 if (IS_ERR(ctxt.handle)) {
3594 ret = PTR_ERR(ctxt.handle);
3599 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3601 ocfs2_commit_trans(osb, ctxt.handle);
3604 ocfs2_free_alloc_context(ctxt.data_ac);
3606 ocfs2_free_alloc_context(ctxt.meta_ac);
3607 if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
3608 ocfs2_schedule_truncate_log_flush(osb, 1);
3609 ocfs2_run_deallocs(osb, &ctxt.dealloc);
3613 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3614 up_write(&OCFS2_I(inode)->ip_xattr_sem);
3615 if (!value && !ret) {
3616 ret = ocfs2_try_remove_refcount_tree(inode, di_bh);
3620 ocfs2_inode_unlock(inode, 1);
3623 brelse(xbs.xattr_bh);
3624 ocfs2_xattr_bucket_free(xbs.bucket);
3630 * Find the xattr extent rec which may contains name_hash.
3631 * e_cpos will be the first name hash of the xattr rec.
3632 * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
3634 static int ocfs2_xattr_get_rec(struct inode *inode,
3639 struct ocfs2_extent_list *el)
3642 struct buffer_head *eb_bh = NULL;
3643 struct ocfs2_extent_block *eb;
3644 struct ocfs2_extent_rec *rec = NULL;
3647 if (el->l_tree_depth) {
3648 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, name_hash,
3655 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
3658 if (el->l_tree_depth) {
3659 ocfs2_error(inode->i_sb,
3660 "Inode %lu has non zero tree depth in "
3661 "xattr tree block %llu\n", inode->i_ino,
3662 (unsigned long long)eb_bh->b_blocknr);
3668 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
3669 rec = &el->l_recs[i];
3671 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
3672 e_blkno = le64_to_cpu(rec->e_blkno);
3678 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
3679 "record (%u, %u, 0) in xattr", inode->i_ino,
3680 le32_to_cpu(rec->e_cpos),
3681 ocfs2_rec_clusters(el, rec));
3686 *p_blkno = le64_to_cpu(rec->e_blkno);
3687 *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
3689 *e_cpos = le32_to_cpu(rec->e_cpos);
3695 typedef int (xattr_bucket_func)(struct inode *inode,
3696 struct ocfs2_xattr_bucket *bucket,
3699 static int ocfs2_find_xe_in_bucket(struct inode *inode,
3700 struct ocfs2_xattr_bucket *bucket,
3707 int i, ret = 0, cmp = 1, block_off, new_offset;
3708 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3709 size_t name_len = strlen(name);
3710 struct ocfs2_xattr_entry *xe = NULL;
3714 * We don't use binary search in the bucket because there
3715 * may be multiple entries with the same name hash.
3717 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3718 xe = &xh->xh_entries[i];
3720 if (name_hash > le32_to_cpu(xe->xe_name_hash))
3722 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
3725 cmp = name_index - ocfs2_xattr_get_type(xe);
3727 cmp = name_len - xe->xe_name_len;
3731 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3742 xe_name = bucket_block(bucket, block_off) + new_offset;
3743 if (!memcmp(name, xe_name, name_len)) {
3755 * Find the specified xattr entry in a series of buckets.
3756 * This series start from p_blkno and last for num_clusters.
3757 * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
3758 * the num of the valid buckets.
3760 * Return the buffer_head this xattr should reside in. And if the xattr's
3761 * hash is in the gap of 2 buckets, return the lower bucket.
3763 static int ocfs2_xattr_bucket_find(struct inode *inode,
3770 struct ocfs2_xattr_search *xs)
3773 struct ocfs2_xattr_header *xh = NULL;
3774 struct ocfs2_xattr_entry *xe = NULL;
3776 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3777 int low_bucket = 0, bucket, high_bucket;
3778 struct ocfs2_xattr_bucket *search;
3780 u64 blkno, lower_blkno = 0;
3782 search = ocfs2_xattr_bucket_new(inode);
3789 ret = ocfs2_read_xattr_bucket(search, p_blkno);
3795 xh = bucket_xh(search);
3796 high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
3797 while (low_bucket <= high_bucket) {
3798 ocfs2_xattr_bucket_relse(search);
3800 bucket = (low_bucket + high_bucket) / 2;
3801 blkno = p_blkno + bucket * blk_per_bucket;
3802 ret = ocfs2_read_xattr_bucket(search, blkno);
3808 xh = bucket_xh(search);
3809 xe = &xh->xh_entries[0];
3810 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3811 high_bucket = bucket - 1;
3816 * Check whether the hash of the last entry in our
3817 * bucket is larger than the search one. for an empty
3818 * bucket, the last one is also the first one.
3821 xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3823 last_hash = le32_to_cpu(xe->xe_name_hash);
3825 /* record lower_blkno which may be the insert place. */
3826 lower_blkno = blkno;
3828 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3829 low_bucket = bucket + 1;
3833 /* the searched xattr should reside in this bucket if exists. */
3834 ret = ocfs2_find_xe_in_bucket(inode, search,
3835 name_index, name, name_hash,
3845 * Record the bucket we have found.
3846 * When the xattr's hash value is in the gap of 2 buckets, we will
3847 * always set it to the previous bucket.
3850 lower_blkno = p_blkno;
3852 /* This should be in cache - we just read it during the search */
3853 ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3859 xs->header = bucket_xh(xs->bucket);
3860 xs->base = bucket_block(xs->bucket, 0);
3861 xs->end = xs->base + inode->i_sb->s_blocksize;
3864 xs->here = &xs->header->xh_entries[index];
3865 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
3866 (unsigned long long)bucket_blkno(xs->bucket), index);
3871 ocfs2_xattr_bucket_free(search);
3875 static int ocfs2_xattr_index_block_find(struct inode *inode,
3876 struct buffer_head *root_bh,
3879 struct ocfs2_xattr_search *xs)
3882 struct ocfs2_xattr_block *xb =
3883 (struct ocfs2_xattr_block *)root_bh->b_data;
3884 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3885 struct ocfs2_extent_list *el = &xb_root->xt_list;
3887 u32 first_hash, num_clusters = 0;
3888 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3890 if (le16_to_cpu(el->l_next_free_rec) == 0)
3893 mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
3894 name, name_hash, name_index);
3896 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3903 BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3905 mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
3906 "in the rec is %u\n", num_clusters, (unsigned long long)p_blkno,
3909 ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3910 p_blkno, first_hash, num_clusters, xs);
3916 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3919 xattr_bucket_func *func,
3923 u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3924 u32 num_buckets = clusters * bpc;
3925 struct ocfs2_xattr_bucket *bucket;
3927 bucket = ocfs2_xattr_bucket_new(inode);
3929 mlog_errno(-ENOMEM);
3933 mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
3934 clusters, (unsigned long long)blkno);
3936 for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3937 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3944 * The real bucket num in this series of blocks is stored
3945 * in the 1st bucket.
3948 num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3950 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
3951 (unsigned long long)blkno,
3952 le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3954 ret = func(inode, bucket, para);
3955 if (ret && ret != -ERANGE)
3957 /* Fall through to bucket_relse() */
3960 ocfs2_xattr_bucket_relse(bucket);
3965 ocfs2_xattr_bucket_free(bucket);
3969 struct ocfs2_xattr_tree_list {
3975 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
3976 struct ocfs2_xattr_header *xh,
3983 if (index < 0 || index >= le16_to_cpu(xh->xh_count))
3986 name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
3988 *block_off = name_offset >> sb->s_blocksize_bits;
3989 *new_offset = name_offset % sb->s_blocksize;
3994 static int ocfs2_list_xattr_bucket(struct inode *inode,
3995 struct ocfs2_xattr_bucket *bucket,
3999 struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
4000 int i, block_off, new_offset;
4001 const char *prefix, *name;
4003 for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
4004 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
4005 type = ocfs2_xattr_get_type(entry);
4006 prefix = ocfs2_xattr_prefix(type);
4009 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
4017 name = (const char *)bucket_block(bucket, block_off) +
4019 ret = ocfs2_xattr_list_entry(xl->buffer,
4023 entry->xe_name_len);
4032 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
4033 struct buffer_head *blk_bh,
4034 xattr_tree_rec_func *rec_func,
4037 struct ocfs2_xattr_block *xb =
4038 (struct ocfs2_xattr_block *)blk_bh->b_data;
4039 struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4041 u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
4044 if (!el->l_next_free_rec || !rec_func)
4047 while (name_hash > 0) {
4048 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4049 &e_cpos, &num_clusters, el);
4055 ret = rec_func(inode, blk_bh, p_blkno, e_cpos,
4056 num_clusters, para);
4066 name_hash = e_cpos - 1;
4073 static int ocfs2_list_xattr_tree_rec(struct inode *inode,
4074 struct buffer_head *root_bh,
4075 u64 blkno, u32 cpos, u32 len, void *para)
4077 return ocfs2_iterate_xattr_buckets(inode, blkno, len,
4078 ocfs2_list_xattr_bucket, para);
4081 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
4082 struct buffer_head *blk_bh,
4087 struct ocfs2_xattr_tree_list xl = {
4089 .buffer_size = buffer_size,
4093 ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
4094 ocfs2_list_xattr_tree_rec, &xl);
4105 static int cmp_xe(const void *a, const void *b)
4107 const struct ocfs2_xattr_entry *l = a, *r = b;
4108 u32 l_hash = le32_to_cpu(l->xe_name_hash);
4109 u32 r_hash = le32_to_cpu(r->xe_name_hash);
4111 if (l_hash > r_hash)
4113 if (l_hash < r_hash)
4118 static void swap_xe(void *a, void *b, int size)
4120 struct ocfs2_xattr_entry *l = a, *r = b, tmp;
4123 memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
4124 memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
4128 * When the ocfs2_xattr_block is filled up, new bucket will be created
4129 * and all the xattr entries will be moved to the new bucket.
4130 * The header goes at the start of the bucket, and the names+values are
4131 * filled from the end. This is why *target starts as the last buffer.
4132 * Note: we need to sort the entries since they are not saved in order
4133 * in the ocfs2_xattr_block.
4135 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
4136 struct buffer_head *xb_bh,
4137 struct ocfs2_xattr_bucket *bucket)
4139 int i, blocksize = inode->i_sb->s_blocksize;
4140 int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4141 u16 offset, size, off_change;
4142 struct ocfs2_xattr_entry *xe;
4143 struct ocfs2_xattr_block *xb =
4144 (struct ocfs2_xattr_block *)xb_bh->b_data;
4145 struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
4146 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4147 u16 count = le16_to_cpu(xb_xh->xh_count);
4148 char *src = xb_bh->b_data;
4149 char *target = bucket_block(bucket, blks - 1);
4151 mlog(0, "cp xattr from block %llu to bucket %llu\n",
4152 (unsigned long long)xb_bh->b_blocknr,
4153 (unsigned long long)bucket_blkno(bucket));
4155 for (i = 0; i < blks; i++)
4156 memset(bucket_block(bucket, i), 0, blocksize);
4159 * Since the xe_name_offset is based on ocfs2_xattr_header,
4160 * there is a offset change corresponding to the change of
4161 * ocfs2_xattr_header's position.
4163 off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4164 xe = &xb_xh->xh_entries[count - 1];
4165 offset = le16_to_cpu(xe->xe_name_offset) + off_change;
4166 size = blocksize - offset;
4168 /* copy all the names and values. */
4169 memcpy(target + offset, src + offset, size);
4171 /* Init new header now. */
4172 xh->xh_count = xb_xh->xh_count;
4173 xh->xh_num_buckets = cpu_to_le16(1);
4174 xh->xh_name_value_len = cpu_to_le16(size);
4175 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
4177 /* copy all the entries. */
4178 target = bucket_block(bucket, 0);
4179 offset = offsetof(struct ocfs2_xattr_header, xh_entries);
4180 size = count * sizeof(struct ocfs2_xattr_entry);
4181 memcpy(target + offset, (char *)xb_xh + offset, size);
4183 /* Change the xe offset for all the xe because of the move. */
4184 off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
4185 offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4186 for (i = 0; i < count; i++)
4187 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
4189 mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
4190 offset, size, off_change);
4192 sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
4197 * After we move xattr from block to index btree, we have to
4198 * update ocfs2_xattr_search to the new xe and base.
4200 * When the entry is in xattr block, xattr_bh indicates the storage place.
4201 * While if the entry is in index b-tree, "bucket" indicates the
4202 * real place of the xattr.
4204 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
4205 struct ocfs2_xattr_search *xs,
4206 struct buffer_head *old_bh)
4208 char *buf = old_bh->b_data;
4209 struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
4210 struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
4213 xs->header = bucket_xh(xs->bucket);
4214 xs->base = bucket_block(xs->bucket, 0);
4215 xs->end = xs->base + inode->i_sb->s_blocksize;
4220 i = xs->here - old_xh->xh_entries;
4221 xs->here = &xs->header->xh_entries[i];
4224 static int ocfs2_xattr_create_index_block(struct inode *inode,
4225 struct ocfs2_xattr_search *xs,
4226 struct ocfs2_xattr_set_ctxt *ctxt)
4231 handle_t *handle = ctxt->handle;
4232 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4233 struct ocfs2_inode_info *oi = OCFS2_I(inode);
4234 struct buffer_head *xb_bh = xs->xattr_bh;
4235 struct ocfs2_xattr_block *xb =
4236 (struct ocfs2_xattr_block *)xb_bh->b_data;
4237 struct ocfs2_xattr_tree_root *xr;
4238 u16 xb_flags = le16_to_cpu(xb->xb_flags);
4240 mlog(0, "create xattr index block for %llu\n",
4241 (unsigned long long)xb_bh->b_blocknr);
4243 BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
4244 BUG_ON(!xs->bucket);
4248 * We can use this lock for now, and maybe move to a dedicated mutex
4249 * if performance becomes a problem later.
4251 down_write(&oi->ip_alloc_sem);
4253 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), xb_bh,
4254 OCFS2_JOURNAL_ACCESS_WRITE);
4260 ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac,
4261 1, 1, &bit_off, &len);
4268 * The bucket may spread in many blocks, and
4269 * we will only touch the 1st block and the last block
4270 * in the whole bucket(one for entry and one for data).
4272 blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
4274 mlog(0, "allocate 1 cluster from %llu to xattr block\n",
4275 (unsigned long long)blkno);
4277 ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
4283 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4284 OCFS2_JOURNAL_ACCESS_CREATE);
4290 ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
4291 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4293 ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
4295 /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
4296 memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
4297 offsetof(struct ocfs2_xattr_block, xb_attrs));
4299 xr = &xb->xb_attrs.xb_root;
4300 xr->xt_clusters = cpu_to_le32(1);
4301 xr->xt_last_eb_blk = 0;
4302 xr->xt_list.l_tree_depth = 0;
4303 xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
4304 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
4306 xr->xt_list.l_recs[0].e_cpos = 0;
4307 xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
4308 xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
4310 xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
4312 ocfs2_journal_dirty(handle, xb_bh);
4315 up_write(&oi->ip_alloc_sem);
4320 static int cmp_xe_offset(const void *a, const void *b)
4322 const struct ocfs2_xattr_entry *l = a, *r = b;
4323 u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
4324 u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
4326 if (l_name_offset < r_name_offset)
4328 if (l_name_offset > r_name_offset)
4334 * defrag a xattr bucket if we find that the bucket has some
4335 * holes beteen name/value pairs.
4336 * We will move all the name/value pairs to the end of the bucket
4337 * so that we can spare some space for insertion.
4339 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
4341 struct ocfs2_xattr_bucket *bucket)
4344 size_t end, offset, len;
4345 struct ocfs2_xattr_header *xh;
4346 char *entries, *buf, *bucket_buf = NULL;
4347 u64 blkno = bucket_blkno(bucket);
4349 size_t blocksize = inode->i_sb->s_blocksize;
4350 struct ocfs2_xattr_entry *xe;
4353 * In order to make the operation more efficient and generic,
4354 * we copy all the blocks into a contiguous memory and do the
4355 * defragment there, so if anything is error, we will not touch
4358 bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
4365 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4366 memcpy(buf, bucket_block(bucket, i), blocksize);
4368 ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
4369 OCFS2_JOURNAL_ACCESS_WRITE);
4375 xh = (struct ocfs2_xattr_header *)bucket_buf;
4376 entries = (char *)xh->xh_entries;
4377 xh_free_start = le16_to_cpu(xh->xh_free_start);
4379 mlog(0, "adjust xattr bucket in %llu, count = %u, "
4380 "xh_free_start = %u, xh_name_value_len = %u.\n",
4381 (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
4382 xh_free_start, le16_to_cpu(xh->xh_name_value_len));
4385 * sort all the entries by their offset.
4386 * the largest will be the first, so that we can
4387 * move them to the end one by one.
4389 sort(entries, le16_to_cpu(xh->xh_count),
4390 sizeof(struct ocfs2_xattr_entry),
4391 cmp_xe_offset, swap_xe);
4393 /* Move all name/values to the end of the bucket. */
4394 xe = xh->xh_entries;
4395 end = OCFS2_XATTR_BUCKET_SIZE;
4396 for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
4397 offset = le16_to_cpu(xe->xe_name_offset);
4398 len = namevalue_size_xe(xe);
4401 * We must make sure that the name/value pair
4402 * exist in the same block. So adjust end to
4403 * the previous block end if needed.
4405 if (((end - len) / blocksize !=
4406 (end - 1) / blocksize))
4407 end = end - end % blocksize;
4409 if (end > offset + len) {
4410 memmove(bucket_buf + end - len,
4411 bucket_buf + offset, len);
4412 xe->xe_name_offset = cpu_to_le16(end - len);
4415 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
4416 "bucket %llu\n", (unsigned long long)blkno);
4421 mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
4422 "bucket %llu\n", (unsigned long long)blkno);
4424 if (xh_free_start == end)
4427 memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
4428 xh->xh_free_start = cpu_to_le16(end);
4430 /* sort the entries by their name_hash. */
4431 sort(entries, le16_to_cpu(xh->xh_count),
4432 sizeof(struct ocfs2_xattr_entry),
4436 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4437 memcpy(bucket_block(bucket, i), buf, blocksize);
4438 ocfs2_xattr_bucket_journal_dirty(handle, bucket);
4446 * prev_blkno points to the start of an existing extent. new_blkno
4447 * points to a newly allocated extent. Because we know each of our
4448 * clusters contains more than bucket, we can easily split one cluster
4449 * at a bucket boundary. So we take the last cluster of the existing
4450 * extent and split it down the middle. We move the last half of the
4451 * buckets in the last cluster of the existing extent over to the new
4454 * first_bh is the buffer at prev_blkno so we can update the existing
4455 * extent's bucket count. header_bh is the bucket were we were hoping
4456 * to insert our xattr. If the bucket move places the target in the new
4457 * extent, we'll update first_bh and header_bh after modifying the old
4460 * first_hash will be set as the 1st xe's name_hash in the new extent.
4462 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
4464 struct ocfs2_xattr_bucket *first,
4465 struct ocfs2_xattr_bucket *target,
4471 struct super_block *sb = inode->i_sb;
4472 int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
4473 int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
4474 int to_move = num_buckets / 2;
4476 u64 last_cluster_blkno = bucket_blkno(first) +
4477 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
4479 BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
4480 BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
4482 mlog(0, "move half of xattrs in cluster %llu to %llu\n",
4483 (unsigned long long)last_cluster_blkno, (unsigned long long)new_blkno);
4485 ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
4486 last_cluster_blkno, new_blkno,
4487 to_move, first_hash);
4493 /* This is the first bucket that got moved */
4494 src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
4497 * If the target bucket was part of the moved buckets, we need to
4498 * update first and target.
4500 if (bucket_blkno(target) >= src_blkno) {
4501 /* Find the block for the new target bucket */
4502 src_blkno = new_blkno +
4503 (bucket_blkno(target) - src_blkno);
4505 ocfs2_xattr_bucket_relse(first);
4506 ocfs2_xattr_bucket_relse(target);
4509 * These shouldn't fail - the buffers are in the
4510 * journal from ocfs2_cp_xattr_bucket().
4512 ret = ocfs2_read_xattr_bucket(first, new_blkno);
4517 ret = ocfs2_read_xattr_bucket(target, src_blkno);
4528 * Find the suitable pos when we divide a bucket into 2.
4529 * We have to make sure the xattrs with the same hash value exist
4530 * in the same bucket.
4532 * If this ocfs2_xattr_header covers more than one hash value, find a
4533 * place where the hash value changes. Try to find the most even split.
4534 * The most common case is that all entries have different hash values,
4535 * and the first check we make will find a place to split.
4537 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
4539 struct ocfs2_xattr_entry *entries = xh->xh_entries;
4540 int count = le16_to_cpu(xh->xh_count);
4541 int delta, middle = count / 2;
4544 * We start at the middle. Each step gets farther away in both
4545 * directions. We therefore hit the change in hash value
4546 * nearest to the middle. Note that this loop does not execute for
4549 for (delta = 0; delta < middle; delta++) {
4550 /* Let's check delta earlier than middle */
4551 if (cmp_xe(&entries[middle - delta - 1],
4552 &entries[middle - delta]))
4553 return middle - delta;
4555 /* For even counts, don't walk off the end */
4556 if ((middle + delta + 1) == count)
4559 /* Now try delta past middle */
4560 if (cmp_xe(&entries[middle + delta],
4561 &entries[middle + delta + 1]))
4562 return middle + delta + 1;
4565 /* Every entry had the same hash */
4570 * Move some xattrs in old bucket(blk) to new bucket(new_blk).
4571 * first_hash will record the 1st hash of the new bucket.
4573 * Normally half of the xattrs will be moved. But we have to make
4574 * sure that the xattrs with the same hash value are stored in the
4575 * same bucket. If all the xattrs in this bucket have the same hash
4576 * value, the new bucket will be initialized as an empty one and the
4577 * first_hash will be initialized as (hash_value+1).
4579 static int ocfs2_divide_xattr_bucket(struct inode *inode,
4584 int new_bucket_head)
4587 int count, start, len, name_value_len = 0, name_offset = 0;
4588 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4589 struct ocfs2_xattr_header *xh;
4590 struct ocfs2_xattr_entry *xe;
4591 int blocksize = inode->i_sb->s_blocksize;
4593 mlog(0, "move some of xattrs from bucket %llu to %llu\n",
4594 (unsigned long long)blk, (unsigned long long)new_blk);
4596 s_bucket = ocfs2_xattr_bucket_new(inode);
4597 t_bucket = ocfs2_xattr_bucket_new(inode);
4598 if (!s_bucket || !t_bucket) {
4604 ret = ocfs2_read_xattr_bucket(s_bucket, blk);
4610 ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
4611 OCFS2_JOURNAL_ACCESS_WRITE);
4618 * Even if !new_bucket_head, we're overwriting t_bucket. Thus,
4619 * there's no need to read it.
4621 ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
4628 * Hey, if we're overwriting t_bucket, what difference does
4629 * ACCESS_CREATE vs ACCESS_WRITE make? See the comment in the
4630 * same part of ocfs2_cp_xattr_bucket().
4632 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4634 OCFS2_JOURNAL_ACCESS_CREATE :
4635 OCFS2_JOURNAL_ACCESS_WRITE);
4641 xh = bucket_xh(s_bucket);
4642 count = le16_to_cpu(xh->xh_count);
4643 start = ocfs2_xattr_find_divide_pos(xh);
4645 if (start == count) {
4646 xe = &xh->xh_entries[start-1];
4649 * initialized a new empty bucket here.
4650 * The hash value is set as one larger than
4651 * that of the last entry in the previous bucket.
4653 for (i = 0; i < t_bucket->bu_blocks; i++)
4654 memset(bucket_block(t_bucket, i), 0, blocksize);
4656 xh = bucket_xh(t_bucket);
4657 xh->xh_free_start = cpu_to_le16(blocksize);
4658 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
4659 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
4661 goto set_num_buckets;
4664 /* copy the whole bucket to the new first. */
4665 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4667 /* update the new bucket. */
4668 xh = bucket_xh(t_bucket);
4671 * Calculate the total name/value len and xh_free_start for
4672 * the old bucket first.
4674 name_offset = OCFS2_XATTR_BUCKET_SIZE;
4676 for (i = 0; i < start; i++) {
4677 xe = &xh->xh_entries[i];
4678 name_value_len += namevalue_size_xe(xe);
4679 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
4680 name_offset = le16_to_cpu(xe->xe_name_offset);
4684 * Now begin the modification to the new bucket.
4686 * In the new bucket, We just move the xattr entry to the beginning
4687 * and don't touch the name/value. So there will be some holes in the
4688 * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
4691 xe = &xh->xh_entries[start];
4692 len = sizeof(struct ocfs2_xattr_entry) * (count - start);
4693 mlog(0, "mv xattr entry len %d from %d to %d\n", len,
4694 (int)((char *)xe - (char *)xh),
4695 (int)((char *)xh->xh_entries - (char *)xh));
4696 memmove((char *)xh->xh_entries, (char *)xe, len);
4697 xe = &xh->xh_entries[count - start];
4698 len = sizeof(struct ocfs2_xattr_entry) * start;
4699 memset((char *)xe, 0, len);
4701 le16_add_cpu(&xh->xh_count, -start);
4702 le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
4704 /* Calculate xh_free_start for the new bucket. */
4705 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4706 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4707 xe = &xh->xh_entries[i];
4708 if (le16_to_cpu(xe->xe_name_offset) <
4709 le16_to_cpu(xh->xh_free_start))
4710 xh->xh_free_start = xe->xe_name_offset;
4714 /* set xh->xh_num_buckets for the new xh. */
4715 if (new_bucket_head)
4716 xh->xh_num_buckets = cpu_to_le16(1);
4718 xh->xh_num_buckets = 0;
4720 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4722 /* store the first_hash of the new bucket. */
4724 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
4727 * Now only update the 1st block of the old bucket. If we
4728 * just added a new empty bucket, there is no need to modify
4734 xh = bucket_xh(s_bucket);
4735 memset(&xh->xh_entries[start], 0,
4736 sizeof(struct ocfs2_xattr_entry) * (count - start));
4737 xh->xh_count = cpu_to_le16(start);
4738 xh->xh_free_start = cpu_to_le16(name_offset);
4739 xh->xh_name_value_len = cpu_to_le16(name_value_len);
4741 ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
4744 ocfs2_xattr_bucket_free(s_bucket);
4745 ocfs2_xattr_bucket_free(t_bucket);
4751 * Copy xattr from one bucket to another bucket.
4753 * The caller must make sure that the journal transaction
4754 * has enough space for journaling.
4756 static int ocfs2_cp_xattr_bucket(struct inode *inode,
4763 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4765 BUG_ON(s_blkno == t_blkno);
4767 mlog(0, "cp bucket %llu to %llu, target is %d\n",
4768 (unsigned long long)s_blkno, (unsigned long long)t_blkno,
4771 s_bucket = ocfs2_xattr_bucket_new(inode);
4772 t_bucket = ocfs2_xattr_bucket_new(inode);
4773 if (!s_bucket || !t_bucket) {
4779 ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
4784 * Even if !t_is_new, we're overwriting t_bucket. Thus,
4785 * there's no need to read it.
4787 ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
4792 * Hey, if we're overwriting t_bucket, what difference does
4793 * ACCESS_CREATE vs ACCESS_WRITE make? Well, if we allocated a new
4794 * cluster to fill, we came here from
4795 * ocfs2_mv_xattr_buckets(), and it is really new -
4796 * ACCESS_CREATE is required. But we also might have moved data
4797 * out of t_bucket before extending back into it.
4798 * ocfs2_add_new_xattr_bucket() can do this - its call to
4799 * ocfs2_add_new_xattr_cluster() may have created a new extent
4800 * and copied out the end of the old extent. Then it re-extends
4801 * the old extent back to create space for new xattrs. That's
4802 * how we get here, and the bucket isn't really new.
4804 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4806 OCFS2_JOURNAL_ACCESS_CREATE :
4807 OCFS2_JOURNAL_ACCESS_WRITE);
4811 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4812 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4815 ocfs2_xattr_bucket_free(t_bucket);
4816 ocfs2_xattr_bucket_free(s_bucket);
4822 * src_blk points to the start of an existing extent. last_blk points to
4823 * last cluster in that extent. to_blk points to a newly allocated
4824 * extent. We copy the buckets from the cluster at last_blk to the new
4825 * extent. If start_bucket is non-zero, we skip that many buckets before
4826 * we start copying. The new extent's xh_num_buckets gets set to the
4827 * number of buckets we copied. The old extent's xh_num_buckets shrinks
4828 * by the same amount.
4830 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4831 u64 src_blk, u64 last_blk, u64 to_blk,
4832 unsigned int start_bucket,
4835 int i, ret, credits;
4836 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4837 int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4838 int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4839 struct ocfs2_xattr_bucket *old_first, *new_first;
4841 mlog(0, "mv xattrs from cluster %llu to %llu\n",
4842 (unsigned long long)last_blk, (unsigned long long)to_blk);
4844 BUG_ON(start_bucket >= num_buckets);
4846 num_buckets -= start_bucket;
4847 last_blk += (start_bucket * blks_per_bucket);
4850 /* The first bucket of the original extent */
4851 old_first = ocfs2_xattr_bucket_new(inode);
4852 /* The first bucket of the new extent */
4853 new_first = ocfs2_xattr_bucket_new(inode);
4854 if (!old_first || !new_first) {
4860 ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4867 * We need to update the first bucket of the old extent and all
4868 * the buckets going to the new extent.
4870 credits = ((num_buckets + 1) * blks_per_bucket);
4871 ret = ocfs2_extend_trans(handle, credits);
4877 ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4878 OCFS2_JOURNAL_ACCESS_WRITE);
4884 for (i = 0; i < num_buckets; i++) {
4885 ret = ocfs2_cp_xattr_bucket(inode, handle,
4886 last_blk + (i * blks_per_bucket),
4887 to_blk + (i * blks_per_bucket),
4896 * Get the new bucket ready before we dirty anything
4897 * (This actually shouldn't fail, because we already dirtied
4898 * it once in ocfs2_cp_xattr_bucket()).
4900 ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4905 ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4906 OCFS2_JOURNAL_ACCESS_WRITE);
4912 /* Now update the headers */
4913 le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4914 ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4916 bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4917 ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4920 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4923 ocfs2_xattr_bucket_free(new_first);
4924 ocfs2_xattr_bucket_free(old_first);
4929 * Move some xattrs in this cluster to the new cluster.
4930 * This function should only be called when bucket size == cluster size.
4931 * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4933 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4939 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4940 int ret, credits = 2 * blk_per_bucket;
4942 BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4944 ret = ocfs2_extend_trans(handle, credits);
4950 /* Move half of the xattr in start_blk to the next bucket. */
4951 return ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4952 new_blk, first_hash, 1);
4956 * Move some xattrs from the old cluster to the new one since they are not
4957 * contiguous in ocfs2 xattr tree.
4959 * new_blk starts a new separate cluster, and we will move some xattrs from
4960 * prev_blk to it. v_start will be set as the first name hash value in this
4961 * new cluster so that it can be used as e_cpos during tree insertion and
4962 * don't collide with our original b-tree operations. first_bh and header_bh
4963 * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4964 * to extend the insert bucket.
4966 * The problem is how much xattr should we move to the new one and when should
4967 * we update first_bh and header_bh?
4968 * 1. If cluster size > bucket size, that means the previous cluster has more
4969 * than 1 bucket, so just move half nums of bucket into the new cluster and
4970 * update the first_bh and header_bh if the insert bucket has been moved
4971 * to the new cluster.
4972 * 2. If cluster_size == bucket_size:
4973 * a) If the previous extent rec has more than one cluster and the insert
4974 * place isn't in the last cluster, copy the entire last cluster to the
4975 * new one. This time, we don't need to upate the first_bh and header_bh
4976 * since they will not be moved into the new cluster.
4977 * b) Otherwise, move the bottom half of the xattrs in the last cluster into
4978 * the new one. And we set the extend flag to zero if the insert place is
4979 * moved into the new allocated cluster since no extend is needed.
4981 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
4983 struct ocfs2_xattr_bucket *first,
4984 struct ocfs2_xattr_bucket *target,
4992 mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
4993 (unsigned long long)bucket_blkno(first), prev_clusters,
4994 (unsigned long long)new_blk);
4996 if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
4997 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
5006 /* The start of the last cluster in the first extent */
5007 u64 last_blk = bucket_blkno(first) +
5008 ((prev_clusters - 1) *
5009 ocfs2_clusters_to_blocks(inode->i_sb, 1));
5011 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
5012 ret = ocfs2_mv_xattr_buckets(inode, handle,
5013 bucket_blkno(first),
5014 last_blk, new_blk, 0,
5019 ret = ocfs2_divide_xattr_cluster(inode, handle,
5025 if ((bucket_blkno(target) == last_blk) && extend)
5034 * Add a new cluster for xattr storage.
5036 * If the new cluster is contiguous with the previous one, it will be
5037 * appended to the same extent record, and num_clusters will be updated.
5038 * If not, we will insert a new extent for it and move some xattrs in
5039 * the last cluster into the new allocated one.
5040 * We also need to limit the maximum size of a btree leaf, otherwise we'll
5041 * lose the benefits of hashing because we'll have to search large leaves.
5042 * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
5045 * first_bh is the first block of the previous extent rec and header_bh
5046 * indicates the bucket we will insert the new xattrs. They will be updated
5047 * when the header_bh is moved into the new cluster.
5049 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
5050 struct buffer_head *root_bh,
5051 struct ocfs2_xattr_bucket *first,
5052 struct ocfs2_xattr_bucket *target,
5056 struct ocfs2_xattr_set_ctxt *ctxt)
5059 u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
5060 u32 prev_clusters = *num_clusters;
5061 u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
5063 handle_t *handle = ctxt->handle;
5064 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5065 struct ocfs2_extent_tree et;
5067 mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
5068 "previous xattr blkno = %llu\n",
5069 (unsigned long long)OCFS2_I(inode)->ip_blkno,
5070 prev_cpos, (unsigned long long)bucket_blkno(first));
5072 ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5074 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5075 OCFS2_JOURNAL_ACCESS_WRITE);
5081 ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac, 1,
5082 clusters_to_add, &bit_off, &num_bits);
5089 BUG_ON(num_bits > clusters_to_add);
5091 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
5092 mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
5093 num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
5095 if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
5096 (prev_clusters + num_bits) << osb->s_clustersize_bits <=
5097 OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
5099 * If this cluster is contiguous with the old one and
5100 * adding this new cluster, we don't surpass the limit of
5101 * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
5102 * initialized and used like other buckets in the previous
5104 * So add it as a contiguous one. The caller will handle
5107 v_start = prev_cpos + prev_clusters;
5108 *num_clusters = prev_clusters + num_bits;
5109 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
5112 ret = ocfs2_adjust_xattr_cross_cluster(inode,
5126 mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
5127 num_bits, (unsigned long long)block, v_start);
5128 ret = ocfs2_insert_extent(handle, &et, v_start, block,
5129 num_bits, 0, ctxt->meta_ac);
5135 ocfs2_journal_dirty(handle, root_bh);
5142 * We are given an extent. 'first' is the bucket at the very front of
5143 * the extent. The extent has space for an additional bucket past
5144 * bucket_xh(first)->xh_num_buckets. 'target_blkno' is the block number
5145 * of the target bucket. We wish to shift every bucket past the target
5146 * down one, filling in that additional space. When we get back to the
5147 * target, we split the target between itself and the now-empty bucket
5148 * at target+1 (aka, target_blkno + blks_per_bucket).
5150 static int ocfs2_extend_xattr_bucket(struct inode *inode,
5152 struct ocfs2_xattr_bucket *first,
5157 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5158 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5160 u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
5162 mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
5163 "from %llu, len = %u\n", (unsigned long long)target_blk,
5164 (unsigned long long)bucket_blkno(first), num_clusters);
5166 /* The extent must have room for an additional bucket */
5167 BUG_ON(new_bucket >=
5168 (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
5170 /* end_blk points to the last existing bucket */
5171 end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
5174 * end_blk is the start of the last existing bucket.
5175 * Thus, (end_blk - target_blk) covers the target bucket and
5176 * every bucket after it up to, but not including, the last
5177 * existing bucket. Then we add the last existing bucket, the
5178 * new bucket, and the first bucket (3 * blk_per_bucket).
5180 credits = (end_blk - target_blk) + (3 * blk_per_bucket);
5181 ret = ocfs2_extend_trans(handle, credits);
5187 ret = ocfs2_xattr_bucket_journal_access(handle, first,
5188 OCFS2_JOURNAL_ACCESS_WRITE);
5194 while (end_blk != target_blk) {
5195 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
5196 end_blk + blk_per_bucket, 0);
5199 end_blk -= blk_per_bucket;
5202 /* Move half of the xattr in target_blkno to the next bucket. */
5203 ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
5204 target_blk + blk_per_bucket, NULL, 0);
5206 le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
5207 ocfs2_xattr_bucket_journal_dirty(handle, first);
5214 * Add new xattr bucket in an extent record and adjust the buckets
5215 * accordingly. xb_bh is the ocfs2_xattr_block, and target is the
5216 * bucket we want to insert into.
5218 * In the easy case, we will move all the buckets after target down by
5219 * one. Half of target's xattrs will be moved to the next bucket.
5221 * If current cluster is full, we'll allocate a new one. This may not
5222 * be contiguous. The underlying calls will make sure that there is
5223 * space for the insert, shifting buckets around if necessary.
5224 * 'target' may be moved by those calls.
5226 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
5227 struct buffer_head *xb_bh,
5228 struct ocfs2_xattr_bucket *target,
5229 struct ocfs2_xattr_set_ctxt *ctxt)
5231 struct ocfs2_xattr_block *xb =
5232 (struct ocfs2_xattr_block *)xb_bh->b_data;
5233 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
5234 struct ocfs2_extent_list *el = &xb_root->xt_list;
5236 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
5237 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5238 int ret, num_buckets, extend = 1;
5240 u32 e_cpos, num_clusters;
5241 /* The bucket at the front of the extent */
5242 struct ocfs2_xattr_bucket *first;
5244 mlog(0, "Add new xattr bucket starting from %llu\n",
5245 (unsigned long long)bucket_blkno(target));
5247 /* The first bucket of the original extent */
5248 first = ocfs2_xattr_bucket_new(inode);
5255 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
5262 ret = ocfs2_read_xattr_bucket(first, p_blkno);
5268 num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
5269 if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
5271 * This can move first+target if the target bucket moves
5272 * to the new extent.
5274 ret = ocfs2_add_new_xattr_cluster(inode,
5289 ret = ocfs2_extend_xattr_bucket(inode,
5292 bucket_blkno(target),
5299 ocfs2_xattr_bucket_free(first);
5304 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
5305 struct ocfs2_xattr_bucket *bucket,
5308 int block_off = offs >> inode->i_sb->s_blocksize_bits;
5310 offs = offs % inode->i_sb->s_blocksize;
5311 return bucket_block(bucket, block_off) + offs;
5315 * Truncate the specified xe_off entry in xattr bucket.
5316 * bucket is indicated by header_bh and len is the new length.
5317 * Both the ocfs2_xattr_value_root and the entry will be updated here.
5319 * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
5321 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
5322 struct ocfs2_xattr_bucket *bucket,
5325 struct ocfs2_xattr_set_ctxt *ctxt)
5329 struct ocfs2_xattr_entry *xe;
5330 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5331 size_t blocksize = inode->i_sb->s_blocksize;
5332 struct ocfs2_xattr_value_buf vb = {
5333 .vb_access = ocfs2_journal_access,
5336 xe = &xh->xh_entries[xe_off];
5338 BUG_ON(!xe || ocfs2_xattr_is_local(xe));
5340 offset = le16_to_cpu(xe->xe_name_offset) +
5341 OCFS2_XATTR_SIZE(xe->xe_name_len);
5343 value_blk = offset / blocksize;
5345 /* We don't allow ocfs2_xattr_value to be stored in different block. */
5346 BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
5348 vb.vb_bh = bucket->bu_bhs[value_blk];
5351 vb.vb_xv = (struct ocfs2_xattr_value_root *)
5352 (vb.vb_bh->b_data + offset % blocksize);
5355 * From here on out we have to dirty the bucket. The generic
5356 * value calls only modify one of the bucket's bhs, but we need
5357 * to send the bucket at once. So if they error, they *could* have
5358 * modified something. We have to assume they did, and dirty
5359 * the whole bucket. This leaves us in a consistent state.
5361 mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
5362 xe_off, (unsigned long long)bucket_blkno(bucket), len);
5363 ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
5369 ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
5370 OCFS2_JOURNAL_ACCESS_WRITE);
5376 xe->xe_value_size = cpu_to_le64(len);
5378 ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
5384 static int ocfs2_rm_xattr_cluster(struct inode *inode,
5385 struct buffer_head *root_bh,
5392 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5393 struct inode *tl_inode = osb->osb_tl_inode;
5395 struct ocfs2_xattr_block *xb =
5396 (struct ocfs2_xattr_block *)root_bh->b_data;
5397 struct ocfs2_alloc_context *meta_ac = NULL;
5398 struct ocfs2_cached_dealloc_ctxt dealloc;
5399 struct ocfs2_extent_tree et;
5401 ret = ocfs2_iterate_xattr_buckets(inode, blkno, len,
5402 ocfs2_delete_xattr_in_bucket, para);
5408 ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5410 ocfs2_init_dealloc_ctxt(&dealloc);
5412 mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
5413 cpos, len, (unsigned long long)blkno);
5415 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode), blkno,
5418 ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
5424 mutex_lock(&tl_inode->i_mutex);
5426 if (ocfs2_truncate_log_needs_flush(osb)) {
5427 ret = __ocfs2_flush_truncate_log(osb);
5434 handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
5435 if (IS_ERR(handle)) {
5441 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5442 OCFS2_JOURNAL_ACCESS_WRITE);
5448 ret = ocfs2_remove_extent(handle, &et, cpos, len, meta_ac,
5455 le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
5456 ocfs2_journal_dirty(handle, root_bh);
5458 ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
5463 ocfs2_commit_trans(osb, handle);
5465 ocfs2_schedule_truncate_log_flush(osb, 1);
5467 mutex_unlock(&tl_inode->i_mutex);
5470 ocfs2_free_alloc_context(meta_ac);
5472 ocfs2_run_deallocs(osb, &dealloc);
5478 * check whether the xattr bucket is filled up with the same hash value.
5479 * If we want to insert the xattr with the same hash, return -ENOSPC.
5480 * If we want to insert a xattr with different hash value, go ahead
5481 * and ocfs2_divide_xattr_bucket will handle this.
5483 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
5484 struct ocfs2_xattr_bucket *bucket,
5487 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5488 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5490 if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5493 if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5494 xh->xh_entries[0].xe_name_hash) {
5495 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5497 (unsigned long long)bucket_blkno(bucket),
5498 le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5506 * Try to set the entry in the current bucket. If we fail, the caller
5507 * will handle getting us another bucket.
5509 static int ocfs2_xattr_set_entry_bucket(struct inode *inode,
5510 struct ocfs2_xattr_info *xi,
5511 struct ocfs2_xattr_search *xs,
5512 struct ocfs2_xattr_set_ctxt *ctxt)
5515 struct ocfs2_xa_loc loc;
5517 mlog_entry("Set xattr %s in xattr bucket\n", xi->xi_name);
5519 ocfs2_init_xattr_bucket_xa_loc(&loc, xs->bucket,
5520 xs->not_found ? NULL : xs->here);
5521 ret = ocfs2_xa_set(&loc, xi, ctxt);
5523 xs->here = loc.xl_entry;
5526 if (ret != -ENOSPC) {
5531 /* Ok, we need space. Let's try defragmenting the bucket. */
5532 ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5539 ret = ocfs2_xa_set(&loc, xi, ctxt);
5541 xs->here = loc.xl_entry;
5553 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5554 struct ocfs2_xattr_info *xi,
5555 struct ocfs2_xattr_search *xs,
5556 struct ocfs2_xattr_set_ctxt *ctxt)
5560 mlog_entry("Set xattr %s in xattr index block\n", xi->xi_name);
5562 ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5565 if (ret != -ENOSPC) {
5570 /* Ack, need more space. Let's try to get another bucket! */
5573 * We do not allow for overlapping ranges between buckets. And
5574 * the maximum number of collisions we will allow for then is
5575 * one bucket's worth, so check it here whether we need to
5576 * add a new bucket for the insert.
5578 ret = ocfs2_check_xattr_bucket_collision(inode,
5586 ret = ocfs2_add_new_xattr_bucket(inode,
5596 * ocfs2_add_new_xattr_bucket() will have updated
5597 * xs->bucket if it moved, but it will not have updated
5598 * any of the other search fields. Thus, we drop it and
5599 * re-search. Everything should be cached, so it'll be
5602 ocfs2_xattr_bucket_relse(xs->bucket);
5603 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5606 if (ret && ret != -ENODATA)
5608 xs->not_found = ret;
5610 /* Ok, we have a new bucket, let's try again */
5611 ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5612 if (ret && (ret != -ENOSPC))
5620 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5621 struct ocfs2_xattr_bucket *bucket,
5624 int ret = 0, ref_credits;
5625 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5627 struct ocfs2_xattr_entry *xe;
5628 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5629 struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5630 int credits = ocfs2_remove_extent_credits(osb->sb) +
5631 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5632 struct ocfs2_xattr_value_root *xv;
5633 struct ocfs2_rm_xattr_bucket_para *args =
5634 (struct ocfs2_rm_xattr_bucket_para *)para;
5636 ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5638 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5639 xe = &xh->xh_entries[i];
5640 if (ocfs2_xattr_is_local(xe))
5643 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket,
5646 ret = ocfs2_lock_xattr_remove_allocators(inode, xv,
5652 ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
5653 if (IS_ERR(ctxt.handle)) {
5654 ret = PTR_ERR(ctxt.handle);
5659 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5662 ocfs2_commit_trans(osb, ctxt.handle);
5664 ocfs2_free_alloc_context(ctxt.meta_ac);
5665 ctxt.meta_ac = NULL;
5674 ocfs2_free_alloc_context(ctxt.meta_ac);
5675 ocfs2_schedule_truncate_log_flush(osb, 1);
5676 ocfs2_run_deallocs(osb, &ctxt.dealloc);
5681 * Whenever we modify a xattr value root in the bucket(e.g, CoW
5682 * or change the extent record flag), we need to recalculate
5683 * the metaecc for the whole bucket. So it is done here.
5686 * We have to give the extra credits for the caller.
5688 static int ocfs2_xattr_bucket_post_refcount(struct inode *inode,
5693 struct ocfs2_xattr_bucket *bucket =
5694 (struct ocfs2_xattr_bucket *)para;
5696 ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
5697 OCFS2_JOURNAL_ACCESS_WRITE);
5703 ocfs2_xattr_bucket_journal_dirty(handle, bucket);
5709 * Special action we need if the xattr value is refcounted.
5711 * 1. If the xattr is refcounted, lock the tree.
5712 * 2. CoW the xattr if we are setting the new value and the value
5713 * will be stored outside.
5714 * 3. In other case, decrease_refcount will work for us, so just
5715 * lock the refcount tree, calculate the meta and credits is OK.
5717 * We have to do CoW before ocfs2_init_xattr_set_ctxt since
5718 * currently CoW is a completed transaction, while this function
5719 * will also lock the allocators and let us deadlock. So we will
5720 * CoW the whole xattr value.
5722 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
5723 struct ocfs2_dinode *di,
5724 struct ocfs2_xattr_info *xi,
5725 struct ocfs2_xattr_search *xis,
5726 struct ocfs2_xattr_search *xbs,
5727 struct ocfs2_refcount_tree **ref_tree,
5732 struct ocfs2_xattr_block *xb;
5733 struct ocfs2_xattr_entry *xe;
5735 u32 p_cluster, num_clusters;
5736 unsigned int ext_flags;
5737 int name_offset, name_len;
5738 struct ocfs2_xattr_value_buf vb;
5739 struct ocfs2_xattr_bucket *bucket = NULL;
5740 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5741 struct ocfs2_post_refcount refcount;
5742 struct ocfs2_post_refcount *p = NULL;
5743 struct buffer_head *ref_root_bh = NULL;
5745 if (!xis->not_found) {
5747 name_offset = le16_to_cpu(xe->xe_name_offset);
5748 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5750 vb.vb_bh = xis->inode_bh;
5751 vb.vb_access = ocfs2_journal_access_di;
5753 int i, block_off = 0;
5754 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
5756 name_offset = le16_to_cpu(xe->xe_name_offset);
5757 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5758 i = xbs->here - xbs->header->xh_entries;
5760 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
5761 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
5762 bucket_xh(xbs->bucket),
5769 base = bucket_block(xbs->bucket, block_off);
5770 vb.vb_bh = xbs->bucket->bu_bhs[block_off];
5771 vb.vb_access = ocfs2_journal_access;
5773 if (ocfs2_meta_ecc(osb)) {
5774 /*create parameters for ocfs2_post_refcount. */
5775 bucket = xbs->bucket;
5776 refcount.credits = bucket->bu_blocks;
5777 refcount.para = bucket;
5779 ocfs2_xattr_bucket_post_refcount;
5784 vb.vb_bh = xbs->xattr_bh;
5785 vb.vb_access = ocfs2_journal_access_xb;
5789 if (ocfs2_xattr_is_local(xe))
5792 vb.vb_xv = (struct ocfs2_xattr_value_root *)
5793 (base + name_offset + name_len);
5795 ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
5796 &num_clusters, &vb.vb_xv->xr_list,
5804 * We just need to check the 1st extent record, since we always
5805 * CoW the whole xattr. So there shouldn't be a xattr with
5806 * some REFCOUNT extent recs after the 1st one.
5808 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
5811 ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
5812 1, ref_tree, &ref_root_bh);
5819 * If we are deleting the xattr or the new size will be stored inside,
5820 * cool, leave it there, the xattr truncate process will remove them
5821 * for us(it still needs the refcount tree lock and the meta, credits).
5822 * And the worse case is that every cluster truncate will split the
5823 * refcount tree, and make the original extent become 3. So we will need
5824 * 2 * cluster more extent recs at most.
5826 if (!xi->xi_value || xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE) {
5828 ret = ocfs2_refcounted_xattr_delete_need(inode,
5829 &(*ref_tree)->rf_ci,
5830 ref_root_bh, vb.vb_xv,
5837 ret = ocfs2_refcount_cow_xattr(inode, di, &vb,
5838 *ref_tree, ref_root_bh, 0,
5839 le32_to_cpu(vb.vb_xv->xr_clusters), p);
5844 brelse(ref_root_bh);
5849 * Add the REFCOUNTED flags for all the extent rec in ocfs2_xattr_value_root.
5850 * The physical clusters will be added to refcount tree.
5852 static int ocfs2_xattr_value_attach_refcount(struct inode *inode,
5853 struct ocfs2_xattr_value_root *xv,
5854 struct ocfs2_extent_tree *value_et,
5855 struct ocfs2_caching_info *ref_ci,
5856 struct buffer_head *ref_root_bh,
5857 struct ocfs2_cached_dealloc_ctxt *dealloc,
5858 struct ocfs2_post_refcount *refcount)
5861 u32 clusters = le32_to_cpu(xv->xr_clusters);
5862 u32 cpos, p_cluster, num_clusters;
5863 struct ocfs2_extent_list *el = &xv->xr_list;
5864 unsigned int ext_flags;
5867 while (cpos < clusters) {
5868 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
5869 &num_clusters, el, &ext_flags);
5871 cpos += num_clusters;
5872 if ((ext_flags & OCFS2_EXT_REFCOUNTED))
5877 ret = ocfs2_add_refcount_flag(inode, value_et,
5878 ref_ci, ref_root_bh,
5879 cpos - num_clusters,
5880 p_cluster, num_clusters,
5892 * Given a normal ocfs2_xattr_header, refcount all the entries which
5893 * have value stored outside.
5894 * Used for xattrs stored in inode and ocfs2_xattr_block.
5896 static int ocfs2_xattr_attach_refcount_normal(struct inode *inode,
5897 struct ocfs2_xattr_value_buf *vb,
5898 struct ocfs2_xattr_header *header,
5899 struct ocfs2_caching_info *ref_ci,
5900 struct buffer_head *ref_root_bh,
5901 struct ocfs2_cached_dealloc_ctxt *dealloc)
5904 struct ocfs2_xattr_entry *xe;
5905 struct ocfs2_xattr_value_root *xv;
5906 struct ocfs2_extent_tree et;
5909 for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
5910 xe = &header->xh_entries[i];
5912 if (ocfs2_xattr_is_local(xe))
5915 xv = (struct ocfs2_xattr_value_root *)((void *)header +
5916 le16_to_cpu(xe->xe_name_offset) +
5917 OCFS2_XATTR_SIZE(xe->xe_name_len));
5920 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
5922 ret = ocfs2_xattr_value_attach_refcount(inode, xv, &et,
5923 ref_ci, ref_root_bh,
5934 static int ocfs2_xattr_inline_attach_refcount(struct inode *inode,
5935 struct buffer_head *fe_bh,
5936 struct ocfs2_caching_info *ref_ci,
5937 struct buffer_head *ref_root_bh,
5938 struct ocfs2_cached_dealloc_ctxt *dealloc)
5940 struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
5941 struct ocfs2_xattr_header *header = (struct ocfs2_xattr_header *)
5942 (fe_bh->b_data + inode->i_sb->s_blocksize -
5943 le16_to_cpu(di->i_xattr_inline_size));
5944 struct ocfs2_xattr_value_buf vb = {
5946 .vb_access = ocfs2_journal_access_di,
5949 return ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
5950 ref_ci, ref_root_bh, dealloc);
5953 struct ocfs2_xattr_tree_value_refcount_para {
5954 struct ocfs2_caching_info *ref_ci;
5955 struct buffer_head *ref_root_bh;
5956 struct ocfs2_cached_dealloc_ctxt *dealloc;
5959 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
5960 struct ocfs2_xattr_bucket *bucket,
5962 struct ocfs2_xattr_value_root **xv,
5963 struct buffer_head **bh)
5965 int ret, block_off, name_offset;
5966 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5967 struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
5970 ret = ocfs2_xattr_bucket_get_name_value(sb,
5980 base = bucket_block(bucket, block_off);
5982 *xv = (struct ocfs2_xattr_value_root *)(base + name_offset +
5983 OCFS2_XATTR_SIZE(xe->xe_name_len));
5986 *bh = bucket->bu_bhs[block_off];
5992 * For a given xattr bucket, refcount all the entries which
5993 * have value stored outside.
5995 static int ocfs2_xattr_bucket_value_refcount(struct inode *inode,
5996 struct ocfs2_xattr_bucket *bucket,
6000 struct ocfs2_extent_tree et;
6001 struct ocfs2_xattr_tree_value_refcount_para *ref =
6002 (struct ocfs2_xattr_tree_value_refcount_para *)para;
6003 struct ocfs2_xattr_header *xh =
6004 (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6005 struct ocfs2_xattr_entry *xe;
6006 struct ocfs2_xattr_value_buf vb = {
6007 .vb_access = ocfs2_journal_access,
6009 struct ocfs2_post_refcount refcount = {
6010 .credits = bucket->bu_blocks,
6012 .func = ocfs2_xattr_bucket_post_refcount,
6014 struct ocfs2_post_refcount *p = NULL;
6016 /* We only need post_refcount if we support metaecc. */
6017 if (ocfs2_meta_ecc(OCFS2_SB(inode->i_sb)))
6020 mlog(0, "refcount bucket %llu, count = %u\n",
6021 (unsigned long long)bucket_blkno(bucket),
6022 le16_to_cpu(xh->xh_count));
6023 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6024 xe = &xh->xh_entries[i];
6026 if (ocfs2_xattr_is_local(xe))
6029 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket, i,
6030 &vb.vb_xv, &vb.vb_bh);
6036 ocfs2_init_xattr_value_extent_tree(&et,
6037 INODE_CACHE(inode), &vb);
6039 ret = ocfs2_xattr_value_attach_refcount(inode, vb.vb_xv,
6053 static int ocfs2_refcount_xattr_tree_rec(struct inode *inode,
6054 struct buffer_head *root_bh,
6055 u64 blkno, u32 cpos, u32 len, void *para)
6057 return ocfs2_iterate_xattr_buckets(inode, blkno, len,
6058 ocfs2_xattr_bucket_value_refcount,
6062 static int ocfs2_xattr_block_attach_refcount(struct inode *inode,
6063 struct buffer_head *blk_bh,
6064 struct ocfs2_caching_info *ref_ci,
6065 struct buffer_head *ref_root_bh,
6066 struct ocfs2_cached_dealloc_ctxt *dealloc)
6069 struct ocfs2_xattr_block *xb =
6070 (struct ocfs2_xattr_block *)blk_bh->b_data;
6072 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
6073 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
6074 struct ocfs2_xattr_value_buf vb = {
6076 .vb_access = ocfs2_journal_access_xb,
6079 ret = ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
6080 ref_ci, ref_root_bh,
6083 struct ocfs2_xattr_tree_value_refcount_para para = {
6085 .ref_root_bh = ref_root_bh,
6089 ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
6090 ocfs2_refcount_xattr_tree_rec,
6097 int ocfs2_xattr_attach_refcount_tree(struct inode *inode,
6098 struct buffer_head *fe_bh,
6099 struct ocfs2_caching_info *ref_ci,
6100 struct buffer_head *ref_root_bh,
6101 struct ocfs2_cached_dealloc_ctxt *dealloc)
6104 struct ocfs2_inode_info *oi = OCFS2_I(inode);
6105 struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
6106 struct buffer_head *blk_bh = NULL;
6108 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
6109 ret = ocfs2_xattr_inline_attach_refcount(inode, fe_bh,
6110 ref_ci, ref_root_bh,
6118 if (!di->i_xattr_loc)
6121 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
6128 ret = ocfs2_xattr_block_attach_refcount(inode, blk_bh, ref_ci,
6129 ref_root_bh, dealloc);
6139 typedef int (should_xattr_reflinked)(struct ocfs2_xattr_entry *xe);
6141 * Store the information we need in xattr reflink.
6142 * old_bh and new_bh are inode bh for the old and new inode.
6144 struct ocfs2_xattr_reflink {
6145 struct inode *old_inode;
6146 struct inode *new_inode;
6147 struct buffer_head *old_bh;
6148 struct buffer_head *new_bh;
6149 struct ocfs2_caching_info *ref_ci;
6150 struct buffer_head *ref_root_bh;
6151 struct ocfs2_cached_dealloc_ctxt *dealloc;
6152 should_xattr_reflinked *xattr_reflinked;
6156 * Given a xattr header and xe offset,
6157 * return the proper xv and the corresponding bh.
6158 * xattr in inode, block and xattr tree have different implementaions.
6160 typedef int (get_xattr_value_root)(struct super_block *sb,
6161 struct buffer_head *bh,
6162 struct ocfs2_xattr_header *xh,
6164 struct ocfs2_xattr_value_root **xv,
6165 struct buffer_head **ret_bh,
6169 * Calculate all the xattr value root metadata stored in this xattr header and
6170 * credits we need if we create them from the scratch.
6171 * We use get_xattr_value_root so that all types of xattr container can use it.
6173 static int ocfs2_value_metas_in_xattr_header(struct super_block *sb,
6174 struct buffer_head *bh,
6175 struct ocfs2_xattr_header *xh,
6176 int *metas, int *credits,
6178 get_xattr_value_root *func,
6182 struct ocfs2_xattr_value_root *xv;
6183 struct ocfs2_xattr_entry *xe;
6185 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6186 xe = &xh->xh_entries[i];
6187 if (ocfs2_xattr_is_local(xe))
6190 ret = func(sb, bh, xh, i, &xv, NULL, para);
6196 *metas += le16_to_cpu(xv->xr_list.l_tree_depth) *
6197 le16_to_cpu(xv->xr_list.l_next_free_rec);
6199 *credits += ocfs2_calc_extend_credits(sb,
6201 le32_to_cpu(xv->xr_clusters));
6204 * If the value is a tree with depth > 1, We don't go deep
6205 * to the extent block, so just calculate a maximum record num.
6207 if (!xv->xr_list.l_tree_depth)
6208 *num_recs += le16_to_cpu(xv->xr_list.l_next_free_rec);
6210 *num_recs += ocfs2_clusters_for_bytes(sb,
6217 /* Used by xattr inode and block to return the right xv and buffer_head. */
6218 static int ocfs2_get_xattr_value_root(struct super_block *sb,
6219 struct buffer_head *bh,
6220 struct ocfs2_xattr_header *xh,
6222 struct ocfs2_xattr_value_root **xv,
6223 struct buffer_head **ret_bh,
6226 struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
6228 *xv = (struct ocfs2_xattr_value_root *)((void *)xh +
6229 le16_to_cpu(xe->xe_name_offset) +
6230 OCFS2_XATTR_SIZE(xe->xe_name_len));
6239 * Lock the meta_ac and caculate how much credits we need for reflink xattrs.
6240 * It is only used for inline xattr and xattr block.
6242 static int ocfs2_reflink_lock_xattr_allocators(struct ocfs2_super *osb,
6243 struct ocfs2_xattr_header *xh,
6244 struct buffer_head *ref_root_bh,
6246 struct ocfs2_alloc_context **meta_ac)
6248 int ret, meta_add = 0, num_recs = 0;
6249 struct ocfs2_refcount_block *rb =
6250 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
6254 ret = ocfs2_value_metas_in_xattr_header(osb->sb, NULL, xh,
6255 &meta_add, credits, &num_recs,
6256 ocfs2_get_xattr_value_root,
6264 * We need to add/modify num_recs in refcount tree, so just calculate
6265 * an approximate number we need for refcount tree change.
6266 * Sometimes we need to split the tree, and after split, half recs
6267 * will be moved to the new block, and a new block can only provide
6268 * half number of recs. So we multiple new blocks by 2.
6270 num_recs = num_recs / ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6271 meta_add += num_recs;
6272 *credits += num_recs + num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6273 if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6274 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6275 le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6279 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add, meta_ac);
6288 * Given a xattr header, reflink all the xattrs in this container.
6289 * It can be used for inode, block and bucket.
6292 * Before we call this function, the caller has memcpy the xattr in
6293 * old_xh to the new_xh.
6295 * If args.xattr_reflinked is set, call it to decide whether the xe should
6296 * be reflinked or not. If not, remove it from the new xattr header.
6298 static int ocfs2_reflink_xattr_header(handle_t *handle,
6299 struct ocfs2_xattr_reflink *args,
6300 struct buffer_head *old_bh,
6301 struct ocfs2_xattr_header *xh,
6302 struct buffer_head *new_bh,
6303 struct ocfs2_xattr_header *new_xh,
6304 struct ocfs2_xattr_value_buf *vb,
6305 struct ocfs2_alloc_context *meta_ac,
6306 get_xattr_value_root *func,
6310 struct super_block *sb = args->old_inode->i_sb;
6311 struct buffer_head *value_bh;
6312 struct ocfs2_xattr_entry *xe, *last;
6313 struct ocfs2_xattr_value_root *xv, *new_xv;
6314 struct ocfs2_extent_tree data_et;
6315 u32 clusters, cpos, p_cluster, num_clusters;
6316 unsigned int ext_flags = 0;
6318 mlog(0, "reflink xattr in container %llu, count = %u\n",
6319 (unsigned long long)old_bh->b_blocknr, le16_to_cpu(xh->xh_count));
6321 last = &new_xh->xh_entries[le16_to_cpu(new_xh->xh_count)];
6322 for (i = 0, j = 0; i < le16_to_cpu(xh->xh_count); i++, j++) {
6323 xe = &xh->xh_entries[i];
6325 if (args->xattr_reflinked && !args->xattr_reflinked(xe)) {
6326 xe = &new_xh->xh_entries[j];
6328 le16_add_cpu(&new_xh->xh_count, -1);
6329 if (new_xh->xh_count) {
6331 (void *)last - (void *)xe);
6333 sizeof(struct ocfs2_xattr_entry));
6337 * We don't want j to increase in the next round since
6338 * it is already moved ahead.
6344 if (ocfs2_xattr_is_local(xe))
6347 ret = func(sb, old_bh, xh, i, &xv, NULL, para);
6353 ret = func(sb, new_bh, new_xh, j, &new_xv, &value_bh, para);
6360 * For the xattr which has l_tree_depth = 0, all the extent
6361 * recs have already be copied to the new xh with the
6362 * propriate OCFS2_EXT_REFCOUNTED flag we just need to
6363 * increase the refount count int the refcount tree.
6365 * For the xattr which has l_tree_depth > 0, we need
6366 * to initialize it to the empty default value root,
6367 * and then insert the extents one by one.
6369 if (xv->xr_list.l_tree_depth) {
6370 memcpy(new_xv, &def_xv, sizeof(def_xv));
6372 vb->vb_bh = value_bh;
6373 ocfs2_init_xattr_value_extent_tree(&data_et,
6374 INODE_CACHE(args->new_inode), vb);
6377 clusters = le32_to_cpu(xv->xr_clusters);
6379 while (cpos < clusters) {
6380 ret = ocfs2_xattr_get_clusters(args->old_inode,
6393 if (xv->xr_list.l_tree_depth) {
6394 ret = ocfs2_insert_extent(handle,
6396 ocfs2_clusters_to_blocks(
6397 args->old_inode->i_sb,
6399 num_clusters, ext_flags,
6407 ret = ocfs2_increase_refcount(handle, args->ref_ci,
6409 p_cluster, num_clusters,
6410 meta_ac, args->dealloc);
6416 cpos += num_clusters;
6424 static int ocfs2_reflink_xattr_inline(struct ocfs2_xattr_reflink *args)
6426 int ret = 0, credits = 0;
6428 struct ocfs2_super *osb = OCFS2_SB(args->old_inode->i_sb);
6429 struct ocfs2_dinode *di = (struct ocfs2_dinode *)args->old_bh->b_data;
6430 int inline_size = le16_to_cpu(di->i_xattr_inline_size);
6431 int header_off = osb->sb->s_blocksize - inline_size;
6432 struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)
6433 (args->old_bh->b_data + header_off);
6434 struct ocfs2_xattr_header *new_xh = (struct ocfs2_xattr_header *)
6435 (args->new_bh->b_data + header_off);
6436 struct ocfs2_alloc_context *meta_ac = NULL;
6437 struct ocfs2_inode_info *new_oi;
6438 struct ocfs2_dinode *new_di;
6439 struct ocfs2_xattr_value_buf vb = {
6440 .vb_bh = args->new_bh,
6441 .vb_access = ocfs2_journal_access_di,
6444 ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6445 &credits, &meta_ac);
6451 handle = ocfs2_start_trans(osb, credits);
6452 if (IS_ERR(handle)) {
6453 ret = PTR_ERR(handle);
6458 ret = ocfs2_journal_access_di(handle, INODE_CACHE(args->new_inode),
6459 args->new_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6465 memcpy(args->new_bh->b_data + header_off,
6466 args->old_bh->b_data + header_off, inline_size);
6468 new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6469 new_di->i_xattr_inline_size = cpu_to_le16(inline_size);
6471 ret = ocfs2_reflink_xattr_header(handle, args, args->old_bh, xh,
6472 args->new_bh, new_xh, &vb, meta_ac,
6473 ocfs2_get_xattr_value_root, NULL);
6479 new_oi = OCFS2_I(args->new_inode);
6480 spin_lock(&new_oi->ip_lock);
6481 new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL | OCFS2_INLINE_XATTR_FL;
6482 new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6483 spin_unlock(&new_oi->ip_lock);
6485 ocfs2_journal_dirty(handle, args->new_bh);
6488 ocfs2_commit_trans(osb, handle);
6492 ocfs2_free_alloc_context(meta_ac);
6496 static int ocfs2_create_empty_xattr_block(struct inode *inode,
6497 struct buffer_head *fe_bh,
6498 struct buffer_head **ret_bh,
6502 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6503 struct ocfs2_xattr_set_ctxt ctxt;
6505 memset(&ctxt, 0, sizeof(ctxt));
6506 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &ctxt.meta_ac);
6512 ctxt.handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_CREATE_CREDITS);
6513 if (IS_ERR(ctxt.handle)) {
6514 ret = PTR_ERR(ctxt.handle);
6519 mlog(0, "create new xattr block for inode %llu, index = %d\n",
6520 (unsigned long long)fe_bh->b_blocknr, indexed);
6521 ret = ocfs2_create_xattr_block(inode, fe_bh, &ctxt, indexed,
6526 ocfs2_commit_trans(osb, ctxt.handle);
6528 ocfs2_free_alloc_context(ctxt.meta_ac);
6532 static int ocfs2_reflink_xattr_block(struct ocfs2_xattr_reflink *args,
6533 struct buffer_head *blk_bh,
6534 struct buffer_head *new_blk_bh)
6536 int ret = 0, credits = 0;
6538 struct ocfs2_inode_info *new_oi = OCFS2_I(args->new_inode);
6539 struct ocfs2_dinode *new_di;
6540 struct ocfs2_super *osb = OCFS2_SB(args->new_inode->i_sb);
6541 int header_off = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
6542 struct ocfs2_xattr_block *xb =
6543 (struct ocfs2_xattr_block *)blk_bh->b_data;
6544 struct ocfs2_xattr_header *xh = &xb->xb_attrs.xb_header;
6545 struct ocfs2_xattr_block *new_xb =
6546 (struct ocfs2_xattr_block *)new_blk_bh->b_data;
6547 struct ocfs2_xattr_header *new_xh = &new_xb->xb_attrs.xb_header;
6548 struct ocfs2_alloc_context *meta_ac;
6549 struct ocfs2_xattr_value_buf vb = {
6550 .vb_bh = new_blk_bh,
6551 .vb_access = ocfs2_journal_access_xb,
6554 ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6555 &credits, &meta_ac);
6561 /* One more credits in case we need to add xattr flags in new inode. */
6562 handle = ocfs2_start_trans(osb, credits + 1);
6563 if (IS_ERR(handle)) {
6564 ret = PTR_ERR(handle);
6569 if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6570 ret = ocfs2_journal_access_di(handle,
6571 INODE_CACHE(args->new_inode),
6573 OCFS2_JOURNAL_ACCESS_WRITE);
6580 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(args->new_inode),
6581 new_blk_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6587 memcpy(new_blk_bh->b_data + header_off, blk_bh->b_data + header_off,
6588 osb->sb->s_blocksize - header_off);
6590 ret = ocfs2_reflink_xattr_header(handle, args, blk_bh, xh,
6591 new_blk_bh, new_xh, &vb, meta_ac,
6592 ocfs2_get_xattr_value_root, NULL);
6598 ocfs2_journal_dirty(handle, new_blk_bh);
6600 if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6601 new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6602 spin_lock(&new_oi->ip_lock);
6603 new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
6604 new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6605 spin_unlock(&new_oi->ip_lock);
6607 ocfs2_journal_dirty(handle, args->new_bh);
6611 ocfs2_commit_trans(osb, handle);
6614 ocfs2_free_alloc_context(meta_ac);
6618 struct ocfs2_reflink_xattr_tree_args {
6619 struct ocfs2_xattr_reflink *reflink;
6620 struct buffer_head *old_blk_bh;
6621 struct buffer_head *new_blk_bh;
6622 struct ocfs2_xattr_bucket *old_bucket;
6623 struct ocfs2_xattr_bucket *new_bucket;
6628 * We have to handle the case that both old bucket and new bucket
6629 * will call this function to get the right ret_bh.
6630 * So The caller must give us the right bh.
6632 static int ocfs2_get_reflink_xattr_value_root(struct super_block *sb,
6633 struct buffer_head *bh,
6634 struct ocfs2_xattr_header *xh,
6636 struct ocfs2_xattr_value_root **xv,
6637 struct buffer_head **ret_bh,
6640 struct ocfs2_reflink_xattr_tree_args *args =
6641 (struct ocfs2_reflink_xattr_tree_args *)para;
6642 struct ocfs2_xattr_bucket *bucket;
6644 if (bh == args->old_bucket->bu_bhs[0])
6645 bucket = args->old_bucket;
6647 bucket = args->new_bucket;
6649 return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6653 struct ocfs2_value_tree_metas {
6659 static int ocfs2_value_tree_metas_in_bucket(struct super_block *sb,
6660 struct buffer_head *bh,
6661 struct ocfs2_xattr_header *xh,
6663 struct ocfs2_xattr_value_root **xv,
6664 struct buffer_head **ret_bh,
6667 struct ocfs2_xattr_bucket *bucket =
6668 (struct ocfs2_xattr_bucket *)para;
6670 return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6674 static int ocfs2_calc_value_tree_metas(struct inode *inode,
6675 struct ocfs2_xattr_bucket *bucket,
6678 struct ocfs2_value_tree_metas *metas =
6679 (struct ocfs2_value_tree_metas *)para;
6680 struct ocfs2_xattr_header *xh =
6681 (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6683 /* Add the credits for this bucket first. */
6684 metas->credits += bucket->bu_blocks;
6685 return ocfs2_value_metas_in_xattr_header(inode->i_sb, bucket->bu_bhs[0],
6686 xh, &metas->num_metas,
6687 &metas->credits, &metas->num_recs,
6688 ocfs2_value_tree_metas_in_bucket,
6693 * Given a xattr extent rec starting from blkno and having len clusters,
6694 * iterate all the buckets calculate how much metadata we need for reflinking
6695 * all the ocfs2_xattr_value_root and lock the allocators accordingly.
6697 static int ocfs2_lock_reflink_xattr_rec_allocators(
6698 struct ocfs2_reflink_xattr_tree_args *args,
6699 struct ocfs2_extent_tree *xt_et,
6700 u64 blkno, u32 len, int *credits,
6701 struct ocfs2_alloc_context **meta_ac,
6702 struct ocfs2_alloc_context **data_ac)
6704 int ret, num_free_extents;
6705 struct ocfs2_value_tree_metas metas;
6706 struct ocfs2_super *osb = OCFS2_SB(args->reflink->old_inode->i_sb);
6707 struct ocfs2_refcount_block *rb;
6709 memset(&metas, 0, sizeof(metas));
6711 ret = ocfs2_iterate_xattr_buckets(args->reflink->old_inode, blkno, len,
6712 ocfs2_calc_value_tree_metas, &metas);
6718 *credits = metas.credits;
6721 * Calculate we need for refcount tree change.
6723 * We need to add/modify num_recs in refcount tree, so just calculate
6724 * an approximate number we need for refcount tree change.
6725 * Sometimes we need to split the tree, and after split, half recs
6726 * will be moved to the new block, and a new block can only provide
6727 * half number of recs. So we multiple new blocks by 2.
6728 * In the end, we have to add credits for modifying the already
6729 * existed refcount block.
6731 rb = (struct ocfs2_refcount_block *)args->reflink->ref_root_bh->b_data;
6733 (metas.num_recs + ocfs2_refcount_recs_per_rb(osb->sb) - 1) /
6734 ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6735 metas.num_metas += metas.num_recs;
6736 *credits += metas.num_recs +
6737 metas.num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6738 if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6739 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6740 le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6744 /* count in the xattr tree change. */
6745 num_free_extents = ocfs2_num_free_extents(osb, xt_et);
6746 if (num_free_extents < 0) {
6747 ret = num_free_extents;
6752 if (num_free_extents < len)
6753 metas.num_metas += ocfs2_extend_meta_needed(xt_et->et_root_el);
6755 *credits += ocfs2_calc_extend_credits(osb->sb,
6756 xt_et->et_root_el, len);
6758 if (metas.num_metas) {
6759 ret = ocfs2_reserve_new_metadata_blocks(osb, metas.num_metas,
6768 ret = ocfs2_reserve_clusters(osb, len, data_ac);
6775 ocfs2_free_alloc_context(*meta_ac);
6783 static int ocfs2_reflink_xattr_buckets(handle_t *handle,
6784 u64 blkno, u64 new_blkno, u32 clusters,
6785 struct ocfs2_alloc_context *meta_ac,
6786 struct ocfs2_alloc_context *data_ac,
6787 struct ocfs2_reflink_xattr_tree_args *args)
6790 struct super_block *sb = args->reflink->old_inode->i_sb;
6791 u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
6792 u32 num_buckets = clusters * bpc;
6793 int bpb = args->old_bucket->bu_blocks;
6794 struct ocfs2_xattr_value_buf vb = {
6795 .vb_access = ocfs2_journal_access,
6798 for (i = 0; i < num_buckets; i++, blkno += bpb, new_blkno += bpb) {
6799 ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6805 ret = ocfs2_init_xattr_bucket(args->new_bucket, new_blkno);
6812 * The real bucket num in this series of blocks is stored
6813 * in the 1st bucket.
6816 num_buckets = le16_to_cpu(
6817 bucket_xh(args->old_bucket)->xh_num_buckets);
6819 ret = ocfs2_xattr_bucket_journal_access(handle,
6821 OCFS2_JOURNAL_ACCESS_CREATE);
6827 for (j = 0; j < bpb; j++)
6828 memcpy(bucket_block(args->new_bucket, j),
6829 bucket_block(args->old_bucket, j),
6832 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6834 ret = ocfs2_reflink_xattr_header(handle, args->reflink,
6835 args->old_bucket->bu_bhs[0],
6836 bucket_xh(args->old_bucket),
6837 args->new_bucket->bu_bhs[0],
6838 bucket_xh(args->new_bucket),
6840 ocfs2_get_reflink_xattr_value_root,
6848 * Re-access and dirty the bucket to calculate metaecc.
6849 * Because we may extend the transaction in reflink_xattr_header
6850 * which will let the already accessed block gone.
6852 ret = ocfs2_xattr_bucket_journal_access(handle,
6854 OCFS2_JOURNAL_ACCESS_WRITE);
6860 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6861 ocfs2_xattr_bucket_relse(args->old_bucket);
6862 ocfs2_xattr_bucket_relse(args->new_bucket);
6865 ocfs2_xattr_bucket_relse(args->old_bucket);
6866 ocfs2_xattr_bucket_relse(args->new_bucket);
6870 * Create the same xattr extent record in the new inode's xattr tree.
6872 static int ocfs2_reflink_xattr_rec(struct inode *inode,
6873 struct buffer_head *root_bh,
6879 int ret, credits = 0;
6880 u32 p_cluster, num_clusters;
6883 struct ocfs2_reflink_xattr_tree_args *args =
6884 (struct ocfs2_reflink_xattr_tree_args *)para;
6885 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6886 struct ocfs2_alloc_context *meta_ac = NULL;
6887 struct ocfs2_alloc_context *data_ac = NULL;
6888 struct ocfs2_extent_tree et;
6890 ocfs2_init_xattr_tree_extent_tree(&et,
6891 INODE_CACHE(args->reflink->new_inode),
6894 ret = ocfs2_lock_reflink_xattr_rec_allocators(args, &et, blkno,
6896 &meta_ac, &data_ac);
6902 handle = ocfs2_start_trans(osb, credits);
6903 if (IS_ERR(handle)) {
6904 ret = PTR_ERR(handle);
6909 ret = ocfs2_claim_clusters(osb, handle, data_ac,
6910 len, &p_cluster, &num_clusters);
6916 new_blkno = ocfs2_clusters_to_blocks(osb->sb, p_cluster);
6918 mlog(0, "reflink xattr buckets %llu to %llu, len %u\n",
6919 (unsigned long long)blkno, (unsigned long long)new_blkno, len);
6920 ret = ocfs2_reflink_xattr_buckets(handle, blkno, new_blkno, len,
6921 meta_ac, data_ac, args);
6927 mlog(0, "insert new xattr extent rec start %llu len %u to %u\n",
6928 (unsigned long long)new_blkno, len, cpos);
6929 ret = ocfs2_insert_extent(handle, &et, cpos, new_blkno,
6935 ocfs2_commit_trans(osb, handle);
6939 ocfs2_free_alloc_context(meta_ac);
6941 ocfs2_free_alloc_context(data_ac);
6946 * Create reflinked xattr buckets.
6947 * We will add bucket one by one, and refcount all the xattrs in the bucket
6948 * if they are stored outside.
6950 static int ocfs2_reflink_xattr_tree(struct ocfs2_xattr_reflink *args,
6951 struct buffer_head *blk_bh,
6952 struct buffer_head *new_blk_bh)
6955 struct ocfs2_reflink_xattr_tree_args para;
6957 memset(¶, 0, sizeof(para));
6958 para.reflink = args;
6959 para.old_blk_bh = blk_bh;
6960 para.new_blk_bh = new_blk_bh;
6962 para.old_bucket = ocfs2_xattr_bucket_new(args->old_inode);
6963 if (!para.old_bucket) {
6964 mlog_errno(-ENOMEM);
6968 para.new_bucket = ocfs2_xattr_bucket_new(args->new_inode);
6969 if (!para.new_bucket) {
6975 ret = ocfs2_iterate_xattr_index_block(args->old_inode, blk_bh,
6976 ocfs2_reflink_xattr_rec,
6982 ocfs2_xattr_bucket_free(para.old_bucket);
6983 ocfs2_xattr_bucket_free(para.new_bucket);
6987 static int ocfs2_reflink_xattr_in_block(struct ocfs2_xattr_reflink *args,
6988 struct buffer_head *blk_bh)
6990 int ret, indexed = 0;
6991 struct buffer_head *new_blk_bh = NULL;
6992 struct ocfs2_xattr_block *xb =
6993 (struct ocfs2_xattr_block *)blk_bh->b_data;
6996 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)
6999 ret = ocfs2_create_empty_xattr_block(args->new_inode, args->new_bh,
7000 &new_blk_bh, indexed);
7006 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED))
7007 ret = ocfs2_reflink_xattr_block(args, blk_bh, new_blk_bh);
7009 ret = ocfs2_reflink_xattr_tree(args, blk_bh, new_blk_bh);
7018 static int ocfs2_reflink_xattr_no_security(struct ocfs2_xattr_entry *xe)
7020 int type = ocfs2_xattr_get_type(xe);
7022 return type != OCFS2_XATTR_INDEX_SECURITY &&
7023 type != OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS &&
7024 type != OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT;
7027 int ocfs2_reflink_xattrs(struct inode *old_inode,
7028 struct buffer_head *old_bh,
7029 struct inode *new_inode,
7030 struct buffer_head *new_bh,
7031 bool preserve_security)
7034 struct ocfs2_xattr_reflink args;
7035 struct ocfs2_inode_info *oi = OCFS2_I(old_inode);
7036 struct ocfs2_dinode *di = (struct ocfs2_dinode *)old_bh->b_data;
7037 struct buffer_head *blk_bh = NULL;
7038 struct ocfs2_cached_dealloc_ctxt dealloc;
7039 struct ocfs2_refcount_tree *ref_tree;
7040 struct buffer_head *ref_root_bh = NULL;
7042 ret = ocfs2_lock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7043 le64_to_cpu(di->i_refcount_loc),
7044 1, &ref_tree, &ref_root_bh);
7050 ocfs2_init_dealloc_ctxt(&dealloc);
7052 args.old_inode = old_inode;
7053 args.new_inode = new_inode;
7054 args.old_bh = old_bh;
7055 args.new_bh = new_bh;
7056 args.ref_ci = &ref_tree->rf_ci;
7057 args.ref_root_bh = ref_root_bh;
7058 args.dealloc = &dealloc;
7059 if (preserve_security)
7060 args.xattr_reflinked = NULL;
7062 args.xattr_reflinked = ocfs2_reflink_xattr_no_security;
7064 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
7065 ret = ocfs2_reflink_xattr_inline(&args);
7072 if (!di->i_xattr_loc)
7075 ret = ocfs2_read_xattr_block(old_inode, le64_to_cpu(di->i_xattr_loc),
7082 ret = ocfs2_reflink_xattr_in_block(&args, blk_bh);
7089 ocfs2_unlock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7091 brelse(ref_root_bh);
7093 if (ocfs2_dealloc_has_cluster(&dealloc)) {
7094 ocfs2_schedule_truncate_log_flush(OCFS2_SB(old_inode->i_sb), 1);
7095 ocfs2_run_deallocs(OCFS2_SB(old_inode->i_sb), &dealloc);
7103 * Initialize security and acl for a already created inode.
7104 * Used for reflink a non-preserve-security file.
7106 * It uses common api like ocfs2_xattr_set, so the caller
7107 * must not hold any lock expect i_mutex.
7109 int ocfs2_init_security_and_acl(struct inode *dir,
7110 struct inode *inode)
7113 struct buffer_head *dir_bh = NULL;
7114 struct ocfs2_security_xattr_info si = {
7118 ret = ocfs2_init_security_get(inode, dir, &si);
7120 ret = ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY,
7121 si.name, si.value, si.value_len,
7127 } else if (ret != -EOPNOTSUPP) {
7132 ret = ocfs2_inode_lock(dir, &dir_bh, 0);
7138 ret = ocfs2_init_acl(NULL, inode, dir, NULL, dir_bh, NULL, NULL);
7142 ocfs2_inode_unlock(dir, 0);
7148 * 'security' attributes support
7150 static size_t ocfs2_xattr_security_list(struct dentry *dentry, char *list,
7151 size_t list_size, const char *name,
7152 size_t name_len, int type)
7154 const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
7155 const size_t total_len = prefix_len + name_len + 1;
7157 if (list && total_len <= list_size) {
7158 memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
7159 memcpy(list + prefix_len, name, name_len);
7160 list[prefix_len + name_len] = '\0';
7165 static int ocfs2_xattr_security_get(struct dentry *dentry, const char *name,
7166 void *buffer, size_t size, int type)
7168 if (strcmp(name, "") == 0)
7170 return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_SECURITY,
7171 name, buffer, size);
7174 static int ocfs2_xattr_security_set(struct dentry *dentry, const char *name,
7175 const void *value, size_t size, int flags, int type)
7177 if (strcmp(name, "") == 0)
7180 return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_SECURITY,
7181 name, value, size, flags);
7184 int ocfs2_init_security_get(struct inode *inode,
7186 struct ocfs2_security_xattr_info *si)
7188 /* check whether ocfs2 support feature xattr */
7189 if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
7191 return security_inode_init_security(inode, dir, &si->name, &si->value,
7195 int ocfs2_init_security_set(handle_t *handle,
7196 struct inode *inode,
7197 struct buffer_head *di_bh,
7198 struct ocfs2_security_xattr_info *si,
7199 struct ocfs2_alloc_context *xattr_ac,
7200 struct ocfs2_alloc_context *data_ac)
7202 return ocfs2_xattr_set_handle(handle, inode, di_bh,
7203 OCFS2_XATTR_INDEX_SECURITY,
7204 si->name, si->value, si->value_len, 0,
7208 struct xattr_handler ocfs2_xattr_security_handler = {
7209 .prefix = XATTR_SECURITY_PREFIX,
7210 .list = ocfs2_xattr_security_list,
7211 .get = ocfs2_xattr_security_get,
7212 .set = ocfs2_xattr_security_set,
7216 * 'trusted' attributes support
7218 static size_t ocfs2_xattr_trusted_list(struct dentry *dentry, char *list,
7219 size_t list_size, const char *name,
7220 size_t name_len, int type)
7222 const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
7223 const size_t total_len = prefix_len + name_len + 1;
7225 if (list && total_len <= list_size) {
7226 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
7227 memcpy(list + prefix_len, name, name_len);
7228 list[prefix_len + name_len] = '\0';
7233 static int ocfs2_xattr_trusted_get(struct dentry *dentry, const char *name,
7234 void *buffer, size_t size, int type)
7236 if (strcmp(name, "") == 0)
7238 return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_TRUSTED,
7239 name, buffer, size);
7242 static int ocfs2_xattr_trusted_set(struct dentry *dentry, const char *name,
7243 const void *value, size_t size, int flags, int type)
7245 if (strcmp(name, "") == 0)
7248 return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_TRUSTED,
7249 name, value, size, flags);
7252 struct xattr_handler ocfs2_xattr_trusted_handler = {
7253 .prefix = XATTR_TRUSTED_PREFIX,
7254 .list = ocfs2_xattr_trusted_list,
7255 .get = ocfs2_xattr_trusted_get,
7256 .set = ocfs2_xattr_trusted_set,
7260 * 'user' attributes support
7262 static size_t ocfs2_xattr_user_list(struct dentry *dentry, char *list,
7263 size_t list_size, const char *name,
7264 size_t name_len, int type)
7266 const size_t prefix_len = XATTR_USER_PREFIX_LEN;
7267 const size_t total_len = prefix_len + name_len + 1;
7268 struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7270 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7273 if (list && total_len <= list_size) {
7274 memcpy(list, XATTR_USER_PREFIX, prefix_len);
7275 memcpy(list + prefix_len, name, name_len);
7276 list[prefix_len + name_len] = '\0';
7281 static int ocfs2_xattr_user_get(struct dentry *dentry, const char *name,
7282 void *buffer, size_t size, int type)
7284 struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7286 if (strcmp(name, "") == 0)
7288 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7290 return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_USER, name,
7294 static int ocfs2_xattr_user_set(struct dentry *dentry, const char *name,
7295 const void *value, size_t size, int flags, int type)
7297 struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7299 if (strcmp(name, "") == 0)
7301 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7304 return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_USER,
7305 name, value, size, flags);
7308 struct xattr_handler ocfs2_xattr_user_handler = {
7309 .prefix = XATTR_USER_PREFIX,
7310 .list = ocfs2_xattr_user_list,
7311 .get = ocfs2_xattr_user_get,
7312 .set = ocfs2_xattr_user_set,