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 #include <cluster/masklog.h>
44 #include "blockcheck.h"
54 #include "buffer_head_io.h"
57 #include "refcounttree.h"
59 #include "ocfs2_trace.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;
85 #define OCFS2_XATTR_ROOT_SIZE (sizeof(struct ocfs2_xattr_def_value_root))
86 #define OCFS2_XATTR_INLINE_SIZE 80
87 #define OCFS2_XATTR_HEADER_GAP 4
88 #define OCFS2_XATTR_FREE_IN_IBODY (OCFS2_MIN_XATTR_INLINE_SIZE \
89 - sizeof(struct ocfs2_xattr_header) \
90 - OCFS2_XATTR_HEADER_GAP)
91 #define OCFS2_XATTR_FREE_IN_BLOCK(ptr) ((ptr)->i_sb->s_blocksize \
92 - sizeof(struct ocfs2_xattr_block) \
93 - sizeof(struct ocfs2_xattr_header) \
94 - OCFS2_XATTR_HEADER_GAP)
96 static struct ocfs2_xattr_def_value_root def_xv = {
97 .xv.xr_list.l_count = cpu_to_le16(1),
100 const struct xattr_handler *ocfs2_xattr_handlers[] = {
101 &ocfs2_xattr_user_handler,
102 &posix_acl_access_xattr_handler,
103 &posix_acl_default_xattr_handler,
104 &ocfs2_xattr_trusted_handler,
105 &ocfs2_xattr_security_handler,
109 static const struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
110 [OCFS2_XATTR_INDEX_USER] = &ocfs2_xattr_user_handler,
111 [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS]
112 = &posix_acl_access_xattr_handler,
113 [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT]
114 = &posix_acl_default_xattr_handler,
115 [OCFS2_XATTR_INDEX_TRUSTED] = &ocfs2_xattr_trusted_handler,
116 [OCFS2_XATTR_INDEX_SECURITY] = &ocfs2_xattr_security_handler,
119 struct ocfs2_xattr_info {
123 const void *xi_value;
127 struct ocfs2_xattr_search {
128 struct buffer_head *inode_bh;
130 * xattr_bh point to the block buffer head which has extended attribute
131 * when extended attribute in inode, xattr_bh is equal to inode_bh.
133 struct buffer_head *xattr_bh;
134 struct ocfs2_xattr_header *header;
135 struct ocfs2_xattr_bucket *bucket;
138 struct ocfs2_xattr_entry *here;
142 /* Operations on struct ocfs2_xa_entry */
144 struct ocfs2_xa_loc_operations {
148 int (*xlo_journal_access)(handle_t *handle, struct ocfs2_xa_loc *loc,
150 void (*xlo_journal_dirty)(handle_t *handle, struct ocfs2_xa_loc *loc);
153 * Return a pointer to the appropriate buffer in loc->xl_storage
154 * at the given offset from loc->xl_header.
156 void *(*xlo_offset_pointer)(struct ocfs2_xa_loc *loc, int offset);
158 /* Can we reuse the existing entry for the new value? */
159 int (*xlo_can_reuse)(struct ocfs2_xa_loc *loc,
160 struct ocfs2_xattr_info *xi);
162 /* How much space is needed for the new value? */
163 int (*xlo_check_space)(struct ocfs2_xa_loc *loc,
164 struct ocfs2_xattr_info *xi);
167 * Return the offset of the first name+value pair. This is
168 * the start of our downward-filling free space.
170 int (*xlo_get_free_start)(struct ocfs2_xa_loc *loc);
173 * Remove the name+value at this location. Do whatever is
174 * appropriate with the remaining name+value pairs.
176 void (*xlo_wipe_namevalue)(struct ocfs2_xa_loc *loc);
178 /* Fill xl_entry with a new entry */
179 void (*xlo_add_entry)(struct ocfs2_xa_loc *loc, u32 name_hash);
181 /* Add name+value storage to an entry */
182 void (*xlo_add_namevalue)(struct ocfs2_xa_loc *loc, int size);
185 * Initialize the value buf's access and bh fields for this entry.
186 * ocfs2_xa_fill_value_buf() will handle the xv pointer.
188 void (*xlo_fill_value_buf)(struct ocfs2_xa_loc *loc,
189 struct ocfs2_xattr_value_buf *vb);
193 * Describes an xattr entry location. This is a memory structure
194 * tracking the on-disk structure.
196 struct ocfs2_xa_loc {
197 /* This xattr belongs to this inode */
198 struct inode *xl_inode;
200 /* The ocfs2_xattr_header inside the on-disk storage. Not NULL. */
201 struct ocfs2_xattr_header *xl_header;
203 /* Bytes from xl_header to the end of the storage */
207 * The ocfs2_xattr_entry this location describes. If this is
208 * NULL, this location describes the on-disk structure where it
211 struct ocfs2_xattr_entry *xl_entry;
214 * Internal housekeeping
217 /* Buffer(s) containing this entry */
220 /* Operations on the storage backing this location */
221 const struct ocfs2_xa_loc_operations *xl_ops;
225 * Convenience functions to calculate how much space is needed for a
226 * given name+value pair
228 static int namevalue_size(int name_len, uint64_t value_len)
230 if (value_len > OCFS2_XATTR_INLINE_SIZE)
231 return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
233 return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
236 static int namevalue_size_xi(struct ocfs2_xattr_info *xi)
238 return namevalue_size(xi->xi_name_len, xi->xi_value_len);
241 static int namevalue_size_xe(struct ocfs2_xattr_entry *xe)
243 u64 value_len = le64_to_cpu(xe->xe_value_size);
245 BUG_ON((value_len > OCFS2_XATTR_INLINE_SIZE) &&
246 ocfs2_xattr_is_local(xe));
247 return namevalue_size(xe->xe_name_len, value_len);
251 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
252 struct ocfs2_xattr_header *xh,
257 static int ocfs2_xattr_block_find(struct inode *inode,
260 struct ocfs2_xattr_search *xs);
261 static int ocfs2_xattr_index_block_find(struct inode *inode,
262 struct buffer_head *root_bh,
265 struct ocfs2_xattr_search *xs);
267 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
268 struct buffer_head *blk_bh,
272 static int ocfs2_xattr_create_index_block(struct inode *inode,
273 struct ocfs2_xattr_search *xs,
274 struct ocfs2_xattr_set_ctxt *ctxt);
276 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
277 struct ocfs2_xattr_info *xi,
278 struct ocfs2_xattr_search *xs,
279 struct ocfs2_xattr_set_ctxt *ctxt);
281 typedef int (xattr_tree_rec_func)(struct inode *inode,
282 struct buffer_head *root_bh,
283 u64 blkno, u32 cpos, u32 len, void *para);
284 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
285 struct buffer_head *root_bh,
286 xattr_tree_rec_func *rec_func,
288 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
289 struct ocfs2_xattr_bucket *bucket,
291 static int ocfs2_rm_xattr_cluster(struct inode *inode,
292 struct buffer_head *root_bh,
298 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
299 u64 src_blk, u64 last_blk, u64 to_blk,
300 unsigned int start_bucket,
302 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
303 struct ocfs2_dinode *di,
304 struct ocfs2_xattr_info *xi,
305 struct ocfs2_xattr_search *xis,
306 struct ocfs2_xattr_search *xbs,
307 struct ocfs2_refcount_tree **ref_tree,
310 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
311 struct ocfs2_xattr_bucket *bucket,
313 struct ocfs2_xattr_value_root **xv,
314 struct buffer_head **bh);
316 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
318 return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
321 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
323 return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
326 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
327 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
328 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
330 static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
332 struct ocfs2_xattr_bucket *bucket;
333 int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
335 BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
337 bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
339 bucket->bu_inode = inode;
340 bucket->bu_blocks = blks;
346 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
350 for (i = 0; i < bucket->bu_blocks; i++) {
351 brelse(bucket->bu_bhs[i]);
352 bucket->bu_bhs[i] = NULL;
356 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
359 ocfs2_xattr_bucket_relse(bucket);
360 bucket->bu_inode = NULL;
366 * A bucket that has never been written to disk doesn't need to be
367 * read. We just need the buffer_heads. Don't call this for
368 * buckets that are already on disk. ocfs2_read_xattr_bucket() initializes
371 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
372 u64 xb_blkno, int new)
376 for (i = 0; i < bucket->bu_blocks; i++) {
377 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
379 if (!bucket->bu_bhs[i]) {
385 if (!ocfs2_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
386 bucket->bu_bhs[i])) {
388 ocfs2_set_new_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
391 set_buffer_uptodate(bucket->bu_bhs[i]);
392 ocfs2_set_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
399 ocfs2_xattr_bucket_relse(bucket);
403 /* Read the xattr bucket at xb_blkno */
404 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
409 rc = ocfs2_read_blocks(INODE_CACHE(bucket->bu_inode), xb_blkno,
410 bucket->bu_blocks, bucket->bu_bhs, 0,
413 spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
414 rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
417 &bucket_xh(bucket)->xh_check);
418 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
424 ocfs2_xattr_bucket_relse(bucket);
428 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
429 struct ocfs2_xattr_bucket *bucket,
434 for (i = 0; i < bucket->bu_blocks; i++) {
435 rc = ocfs2_journal_access(handle,
436 INODE_CACHE(bucket->bu_inode),
437 bucket->bu_bhs[i], type);
447 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
448 struct ocfs2_xattr_bucket *bucket)
452 spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
453 ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
454 bucket->bu_bhs, bucket->bu_blocks,
455 &bucket_xh(bucket)->xh_check);
456 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
458 for (i = 0; i < bucket->bu_blocks; i++)
459 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
462 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
463 struct ocfs2_xattr_bucket *src)
466 int blocksize = src->bu_inode->i_sb->s_blocksize;
468 BUG_ON(dest->bu_blocks != src->bu_blocks);
469 BUG_ON(dest->bu_inode != src->bu_inode);
471 for (i = 0; i < src->bu_blocks; i++) {
472 memcpy(bucket_block(dest, i), bucket_block(src, i),
477 static int ocfs2_validate_xattr_block(struct super_block *sb,
478 struct buffer_head *bh)
481 struct ocfs2_xattr_block *xb =
482 (struct ocfs2_xattr_block *)bh->b_data;
484 trace_ocfs2_validate_xattr_block((unsigned long long)bh->b_blocknr);
486 BUG_ON(!buffer_uptodate(bh));
489 * If the ecc fails, we return the error but otherwise
490 * leave the filesystem running. We know any error is
491 * local to this block.
493 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
498 * Errors after here are fatal
501 if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
502 return ocfs2_error(sb,
503 "Extended attribute block #%llu has bad signature %.*s\n",
504 (unsigned long long)bh->b_blocknr, 7,
508 if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
509 return ocfs2_error(sb,
510 "Extended attribute block #%llu has an invalid xb_blkno of %llu\n",
511 (unsigned long long)bh->b_blocknr,
512 (unsigned long long)le64_to_cpu(xb->xb_blkno));
515 if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
516 return ocfs2_error(sb,
517 "Extended attribute block #%llu has an invalid xb_fs_generation of #%u\n",
518 (unsigned long long)bh->b_blocknr,
519 le32_to_cpu(xb->xb_fs_generation));
525 static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
526 struct buffer_head **bh)
529 struct buffer_head *tmp = *bh;
531 rc = ocfs2_read_block(INODE_CACHE(inode), xb_blkno, &tmp,
532 ocfs2_validate_xattr_block);
534 /* If ocfs2_read_block() got us a new bh, pass it up. */
541 static inline const char *ocfs2_xattr_prefix(int name_index)
543 const struct xattr_handler *handler = NULL;
545 if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
546 handler = ocfs2_xattr_handler_map[name_index];
547 return handler ? xattr_prefix(handler) : 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 down_read(&OCFS2_I(dir)->ip_xattr_sem);
642 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
643 OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
645 up_read(&OCFS2_I(dir)->ip_xattr_sem);
647 a_size = ocfs2_xattr_entry_real_size(0, acl_len);
650 } else if (acl_len != 0 && acl_len != -ENODATA) {
657 if (!(s_size + a_size))
661 * The max space of security xattr taken inline is
662 * 256(name) + 80(value) + 16(entry) = 352 bytes,
663 * The max space of acl xattr taken inline is
664 * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
665 * when blocksize = 512, may reserve one more cluser for
666 * xattr bucket, otherwise reserve one metadata block
668 * If this is a new directory with inline data,
669 * we choose to reserve the entire inline area for
670 * directory contents and force an external xattr block.
672 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
673 (S_ISDIR(mode) && ocfs2_supports_inline_data(osb)) ||
674 (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
675 *want_meta = *want_meta + 1;
676 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
679 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
680 (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
682 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
686 * reserve credits and clusters for xattrs which has large value
687 * and have to be set outside
689 if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
690 new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
692 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
694 *want_clusters += new_clusters;
696 if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
697 acl_len > OCFS2_XATTR_INLINE_SIZE) {
698 /* for directory, it has DEFAULT and ACCESS two types of acls */
699 new_clusters = (S_ISDIR(mode) ? 2 : 1) *
700 ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
701 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
703 *want_clusters += new_clusters;
709 static int ocfs2_xattr_extend_allocation(struct inode *inode,
711 struct ocfs2_xattr_value_buf *vb,
712 struct ocfs2_xattr_set_ctxt *ctxt)
714 int status = 0, credits;
715 handle_t *handle = ctxt->handle;
716 enum ocfs2_alloc_restarted why;
717 u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
718 struct ocfs2_extent_tree et;
720 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
722 while (clusters_to_add) {
723 trace_ocfs2_xattr_extend_allocation(clusters_to_add);
725 status = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
726 OCFS2_JOURNAL_ACCESS_WRITE);
732 prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
733 status = ocfs2_add_clusters_in_btree(handle,
741 if ((status < 0) && (status != -EAGAIN)) {
742 if (status != -ENOSPC)
747 ocfs2_journal_dirty(handle, vb->vb_bh);
749 clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) -
752 if (why != RESTART_NONE && clusters_to_add) {
754 * We can only fail in case the alloc file doesn't give
755 * up enough clusters.
757 BUG_ON(why == RESTART_META);
759 credits = ocfs2_calc_extend_credits(inode->i_sb,
760 &vb->vb_xv->xr_list);
761 status = ocfs2_extend_trans(handle, credits);
773 static int __ocfs2_remove_xattr_range(struct inode *inode,
774 struct ocfs2_xattr_value_buf *vb,
775 u32 cpos, u32 phys_cpos, u32 len,
776 unsigned int ext_flags,
777 struct ocfs2_xattr_set_ctxt *ctxt)
780 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
781 handle_t *handle = ctxt->handle;
782 struct ocfs2_extent_tree et;
784 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
786 ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
787 OCFS2_JOURNAL_ACCESS_WRITE);
793 ret = ocfs2_remove_extent(handle, &et, cpos, len, ctxt->meta_ac,
800 le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
801 ocfs2_journal_dirty(handle, vb->vb_bh);
803 if (ext_flags & OCFS2_EXT_REFCOUNTED)
804 ret = ocfs2_decrease_refcount(inode, handle,
805 ocfs2_blocks_to_clusters(inode->i_sb,
807 len, ctxt->meta_ac, &ctxt->dealloc, 1);
809 ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc,
818 static int ocfs2_xattr_shrink_size(struct inode *inode,
821 struct ocfs2_xattr_value_buf *vb,
822 struct ocfs2_xattr_set_ctxt *ctxt)
825 unsigned int ext_flags;
826 u32 trunc_len, cpos, phys_cpos, alloc_size;
829 if (old_clusters <= new_clusters)
833 trunc_len = old_clusters - new_clusters;
835 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
837 &vb->vb_xv->xr_list, &ext_flags);
843 if (alloc_size > trunc_len)
844 alloc_size = trunc_len;
846 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
847 phys_cpos, alloc_size,
854 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
855 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode),
858 trunc_len -= alloc_size;
865 static int ocfs2_xattr_value_truncate(struct inode *inode,
866 struct ocfs2_xattr_value_buf *vb,
868 struct ocfs2_xattr_set_ctxt *ctxt)
871 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
872 u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
874 if (new_clusters == old_clusters)
877 if (new_clusters > old_clusters)
878 ret = ocfs2_xattr_extend_allocation(inode,
879 new_clusters - old_clusters,
882 ret = ocfs2_xattr_shrink_size(inode,
883 old_clusters, new_clusters,
889 static int ocfs2_xattr_list_entry(struct super_block *sb,
890 char *buffer, size_t size,
891 size_t *result, int type,
892 const char *name, int name_len)
894 char *p = buffer + *result;
900 case OCFS2_XATTR_INDEX_USER:
901 if (OCFS2_SB(sb)->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
905 case OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS:
906 case OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT:
907 if (!(sb->s_flags & SB_POSIXACL))
911 case OCFS2_XATTR_INDEX_TRUSTED:
912 if (!capable(CAP_SYS_ADMIN))
917 prefix = ocfs2_xattr_prefix(type);
920 prefix_len = strlen(prefix);
921 total_len = prefix_len + name_len + 1;
922 *result += total_len;
924 /* we are just looking for how big our buffer needs to be */
931 memcpy(p, prefix, prefix_len);
932 memcpy(p + prefix_len, name, name_len);
933 p[prefix_len + name_len] = '\0';
938 static int ocfs2_xattr_list_entries(struct inode *inode,
939 struct ocfs2_xattr_header *header,
940 char *buffer, size_t buffer_size)
946 for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
947 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
948 type = ocfs2_xattr_get_type(entry);
949 name = (const char *)header +
950 le16_to_cpu(entry->xe_name_offset);
952 ret = ocfs2_xattr_list_entry(inode->i_sb,
963 int ocfs2_has_inline_xattr_value_outside(struct inode *inode,
964 struct ocfs2_dinode *di)
966 struct ocfs2_xattr_header *xh;
969 xh = (struct ocfs2_xattr_header *)
970 ((void *)di + inode->i_sb->s_blocksize -
971 le16_to_cpu(di->i_xattr_inline_size));
973 for (i = 0; i < le16_to_cpu(xh->xh_count); i++)
974 if (!ocfs2_xattr_is_local(&xh->xh_entries[i]))
980 static int ocfs2_xattr_ibody_list(struct inode *inode,
981 struct ocfs2_dinode *di,
985 struct ocfs2_xattr_header *header = NULL;
986 struct ocfs2_inode_info *oi = OCFS2_I(inode);
989 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
992 header = (struct ocfs2_xattr_header *)
993 ((void *)di + inode->i_sb->s_blocksize -
994 le16_to_cpu(di->i_xattr_inline_size));
996 ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
1001 static int ocfs2_xattr_block_list(struct inode *inode,
1002 struct ocfs2_dinode *di,
1006 struct buffer_head *blk_bh = NULL;
1007 struct ocfs2_xattr_block *xb;
1010 if (!di->i_xattr_loc)
1013 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
1020 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1021 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1022 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
1023 ret = ocfs2_xattr_list_entries(inode, header,
1024 buffer, buffer_size);
1026 ret = ocfs2_xattr_tree_list_index_block(inode, blk_bh,
1027 buffer, buffer_size);
1034 ssize_t ocfs2_listxattr(struct dentry *dentry,
1038 int ret = 0, i_ret = 0, b_ret = 0;
1039 struct buffer_head *di_bh = NULL;
1040 struct ocfs2_dinode *di = NULL;
1041 struct ocfs2_inode_info *oi = OCFS2_I(d_inode(dentry));
1043 if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
1046 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1049 ret = ocfs2_inode_lock(d_inode(dentry), &di_bh, 0);
1055 di = (struct ocfs2_dinode *)di_bh->b_data;
1057 down_read(&oi->ip_xattr_sem);
1058 i_ret = ocfs2_xattr_ibody_list(d_inode(dentry), di, buffer, size);
1066 b_ret = ocfs2_xattr_block_list(d_inode(dentry), di,
1071 up_read(&oi->ip_xattr_sem);
1072 ocfs2_inode_unlock(d_inode(dentry), 0);
1076 return i_ret + b_ret;
1079 static int ocfs2_xattr_find_entry(int name_index,
1081 struct ocfs2_xattr_search *xs)
1083 struct ocfs2_xattr_entry *entry;
1090 name_len = strlen(name);
1092 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
1093 cmp = name_index - ocfs2_xattr_get_type(entry);
1095 cmp = name_len - entry->xe_name_len;
1097 cmp = memcmp(name, (xs->base +
1098 le16_to_cpu(entry->xe_name_offset)),
1106 return cmp ? -ENODATA : 0;
1109 static int ocfs2_xattr_get_value_outside(struct inode *inode,
1110 struct ocfs2_xattr_value_root *xv,
1114 u32 cpos, p_cluster, num_clusters, bpc, clusters;
1117 size_t cplen, blocksize;
1118 struct buffer_head *bh = NULL;
1119 struct ocfs2_extent_list *el;
1122 clusters = le32_to_cpu(xv->xr_clusters);
1123 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1124 blocksize = inode->i_sb->s_blocksize;
1127 while (cpos < clusters) {
1128 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1129 &num_clusters, el, NULL);
1135 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1136 /* Copy ocfs2_xattr_value */
1137 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1138 ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1145 cplen = len >= blocksize ? blocksize : len;
1146 memcpy(buffer, bh->b_data, cplen);
1155 cpos += num_clusters;
1161 static int ocfs2_xattr_ibody_get(struct inode *inode,
1166 struct ocfs2_xattr_search *xs)
1168 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1169 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1170 struct ocfs2_xattr_value_root *xv;
1174 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1177 xs->end = (void *)di + inode->i_sb->s_blocksize;
1178 xs->header = (struct ocfs2_xattr_header *)
1179 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1180 xs->base = (void *)xs->header;
1181 xs->here = xs->header->xh_entries;
1183 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1186 size = le64_to_cpu(xs->here->xe_value_size);
1188 if (size > buffer_size)
1190 if (ocfs2_xattr_is_local(xs->here)) {
1191 memcpy(buffer, (void *)xs->base +
1192 le16_to_cpu(xs->here->xe_name_offset) +
1193 OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1195 xv = (struct ocfs2_xattr_value_root *)
1196 (xs->base + le16_to_cpu(
1197 xs->here->xe_name_offset) +
1198 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1199 ret = ocfs2_xattr_get_value_outside(inode, xv,
1211 static int ocfs2_xattr_block_get(struct inode *inode,
1216 struct ocfs2_xattr_search *xs)
1218 struct ocfs2_xattr_block *xb;
1219 struct ocfs2_xattr_value_root *xv;
1221 int ret = -ENODATA, name_offset, name_len, i;
1222 int uninitialized_var(block_off);
1224 xs->bucket = ocfs2_xattr_bucket_new(inode);
1231 ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1237 if (xs->not_found) {
1242 xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1243 size = le64_to_cpu(xs->here->xe_value_size);
1246 if (size > buffer_size)
1249 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1250 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1251 i = xs->here - xs->header->xh_entries;
1253 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1254 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
1255 bucket_xh(xs->bucket),
1263 xs->base = bucket_block(xs->bucket, block_off);
1265 if (ocfs2_xattr_is_local(xs->here)) {
1266 memcpy(buffer, (void *)xs->base +
1267 name_offset + name_len, size);
1269 xv = (struct ocfs2_xattr_value_root *)
1270 (xs->base + name_offset + name_len);
1271 ret = ocfs2_xattr_get_value_outside(inode, xv,
1281 ocfs2_xattr_bucket_free(xs->bucket);
1283 brelse(xs->xattr_bh);
1284 xs->xattr_bh = NULL;
1288 int ocfs2_xattr_get_nolock(struct inode *inode,
1289 struct buffer_head *di_bh,
1296 struct ocfs2_dinode *di = NULL;
1297 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1298 struct ocfs2_xattr_search xis = {
1299 .not_found = -ENODATA,
1301 struct ocfs2_xattr_search xbs = {
1302 .not_found = -ENODATA,
1305 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1308 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1311 xis.inode_bh = xbs.inode_bh = di_bh;
1312 di = (struct ocfs2_dinode *)di_bh->b_data;
1314 ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1316 if (ret == -ENODATA && di->i_xattr_loc)
1317 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1323 /* ocfs2_xattr_get()
1325 * Copy an extended attribute into the buffer provided.
1326 * Buffer is NULL to compute the size of buffer required.
1328 static int ocfs2_xattr_get(struct inode *inode,
1335 struct buffer_head *di_bh = NULL;
1336 struct ocfs2_lock_holder oh;
1338 had_lock = ocfs2_inode_lock_tracker(inode, &di_bh, 0, &oh);
1340 mlog_errno(had_lock);
1343 down_read(&OCFS2_I(inode)->ip_xattr_sem);
1344 ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1345 name, buffer, buffer_size);
1346 up_read(&OCFS2_I(inode)->ip_xattr_sem);
1348 ocfs2_inode_unlock_tracker(inode, 0, &oh, had_lock);
1355 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1357 struct ocfs2_xattr_value_buf *vb,
1361 int ret = 0, i, cp_len;
1362 u16 blocksize = inode->i_sb->s_blocksize;
1363 u32 p_cluster, num_clusters;
1364 u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1365 u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1367 struct buffer_head *bh = NULL;
1368 unsigned int ext_flags;
1369 struct ocfs2_xattr_value_root *xv = vb->vb_xv;
1371 BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1373 while (cpos < clusters) {
1374 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1375 &num_clusters, &xv->xr_list,
1382 BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1384 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1386 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1387 ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1394 ret = ocfs2_journal_access(handle,
1397 OCFS2_JOURNAL_ACCESS_WRITE);
1403 cp_len = value_len > blocksize ? blocksize : value_len;
1404 memcpy(bh->b_data, value, cp_len);
1405 value_len -= cp_len;
1407 if (cp_len < blocksize)
1408 memset(bh->b_data + cp_len, 0,
1409 blocksize - cp_len);
1411 ocfs2_journal_dirty(handle, bh);
1416 * XXX: do we need to empty all the following
1417 * blocks in this cluster?
1422 cpos += num_clusters;
1430 static int ocfs2_xa_check_space_helper(int needed_space, int free_start,
1438 free_space = free_start -
1439 sizeof(struct ocfs2_xattr_header) -
1440 (num_entries * sizeof(struct ocfs2_xattr_entry)) -
1441 OCFS2_XATTR_HEADER_GAP;
1444 if (free_space < needed_space)
1450 static int ocfs2_xa_journal_access(handle_t *handle, struct ocfs2_xa_loc *loc,
1453 return loc->xl_ops->xlo_journal_access(handle, loc, type);
1456 static void ocfs2_xa_journal_dirty(handle_t *handle, struct ocfs2_xa_loc *loc)
1458 loc->xl_ops->xlo_journal_dirty(handle, loc);
1461 /* Give a pointer into the storage for the given offset */
1462 static void *ocfs2_xa_offset_pointer(struct ocfs2_xa_loc *loc, int offset)
1464 BUG_ON(offset >= loc->xl_size);
1465 return loc->xl_ops->xlo_offset_pointer(loc, offset);
1469 * Wipe the name+value pair and allow the storage to reclaim it. This
1470 * must be followed by either removal of the entry or a call to
1471 * ocfs2_xa_add_namevalue().
1473 static void ocfs2_xa_wipe_namevalue(struct ocfs2_xa_loc *loc)
1475 loc->xl_ops->xlo_wipe_namevalue(loc);
1479 * Find lowest offset to a name+value pair. This is the start of our
1480 * downward-growing free space.
1482 static int ocfs2_xa_get_free_start(struct ocfs2_xa_loc *loc)
1484 return loc->xl_ops->xlo_get_free_start(loc);
1487 /* Can we reuse loc->xl_entry for xi? */
1488 static int ocfs2_xa_can_reuse_entry(struct ocfs2_xa_loc *loc,
1489 struct ocfs2_xattr_info *xi)
1491 return loc->xl_ops->xlo_can_reuse(loc, xi);
1494 /* How much free space is needed to set the new value */
1495 static int ocfs2_xa_check_space(struct ocfs2_xa_loc *loc,
1496 struct ocfs2_xattr_info *xi)
1498 return loc->xl_ops->xlo_check_space(loc, xi);
1501 static void ocfs2_xa_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1503 loc->xl_ops->xlo_add_entry(loc, name_hash);
1504 loc->xl_entry->xe_name_hash = cpu_to_le32(name_hash);
1506 * We can't leave the new entry's xe_name_offset at zero or
1507 * add_namevalue() will go nuts. We set it to the size of our
1508 * storage so that it can never be less than any other entry.
1510 loc->xl_entry->xe_name_offset = cpu_to_le16(loc->xl_size);
1513 static void ocfs2_xa_add_namevalue(struct ocfs2_xa_loc *loc,
1514 struct ocfs2_xattr_info *xi)
1516 int size = namevalue_size_xi(xi);
1520 loc->xl_ops->xlo_add_namevalue(loc, size);
1521 loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
1522 loc->xl_entry->xe_name_len = xi->xi_name_len;
1523 ocfs2_xattr_set_type(loc->xl_entry, xi->xi_name_index);
1524 ocfs2_xattr_set_local(loc->xl_entry,
1525 xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE);
1527 nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1528 nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
1529 memset(nameval_buf, 0, size);
1530 memcpy(nameval_buf, xi->xi_name, xi->xi_name_len);
1533 static void ocfs2_xa_fill_value_buf(struct ocfs2_xa_loc *loc,
1534 struct ocfs2_xattr_value_buf *vb)
1536 int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1537 int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
1539 /* Value bufs are for value trees */
1540 BUG_ON(ocfs2_xattr_is_local(loc->xl_entry));
1541 BUG_ON(namevalue_size_xe(loc->xl_entry) !=
1542 (name_size + OCFS2_XATTR_ROOT_SIZE));
1544 loc->xl_ops->xlo_fill_value_buf(loc, vb);
1546 (struct ocfs2_xattr_value_root *)ocfs2_xa_offset_pointer(loc,
1551 static int ocfs2_xa_block_journal_access(handle_t *handle,
1552 struct ocfs2_xa_loc *loc, int type)
1554 struct buffer_head *bh = loc->xl_storage;
1555 ocfs2_journal_access_func access;
1557 if (loc->xl_size == (bh->b_size -
1558 offsetof(struct ocfs2_xattr_block,
1559 xb_attrs.xb_header)))
1560 access = ocfs2_journal_access_xb;
1562 access = ocfs2_journal_access_di;
1563 return access(handle, INODE_CACHE(loc->xl_inode), bh, type);
1566 static void ocfs2_xa_block_journal_dirty(handle_t *handle,
1567 struct ocfs2_xa_loc *loc)
1569 struct buffer_head *bh = loc->xl_storage;
1571 ocfs2_journal_dirty(handle, bh);
1574 static void *ocfs2_xa_block_offset_pointer(struct ocfs2_xa_loc *loc,
1577 return (char *)loc->xl_header + offset;
1580 static int ocfs2_xa_block_can_reuse(struct ocfs2_xa_loc *loc,
1581 struct ocfs2_xattr_info *xi)
1584 * Block storage is strict. If the sizes aren't exact, we will
1585 * remove the old one and reinsert the new.
1587 return namevalue_size_xe(loc->xl_entry) ==
1588 namevalue_size_xi(xi);
1591 static int ocfs2_xa_block_get_free_start(struct ocfs2_xa_loc *loc)
1593 struct ocfs2_xattr_header *xh = loc->xl_header;
1594 int i, count = le16_to_cpu(xh->xh_count);
1595 int offset, free_start = loc->xl_size;
1597 for (i = 0; i < count; i++) {
1598 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1599 if (offset < free_start)
1600 free_start = offset;
1606 static int ocfs2_xa_block_check_space(struct ocfs2_xa_loc *loc,
1607 struct ocfs2_xattr_info *xi)
1609 int count = le16_to_cpu(loc->xl_header->xh_count);
1610 int free_start = ocfs2_xa_get_free_start(loc);
1611 int needed_space = ocfs2_xi_entry_usage(xi);
1614 * Block storage will reclaim the original entry before inserting
1615 * the new value, so we only need the difference. If the new
1616 * entry is smaller than the old one, we don't need anything.
1618 if (loc->xl_entry) {
1619 /* Don't need space if we're reusing! */
1620 if (ocfs2_xa_can_reuse_entry(loc, xi))
1623 needed_space -= ocfs2_xe_entry_usage(loc->xl_entry);
1625 if (needed_space < 0)
1627 return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1631 * Block storage for xattrs keeps the name+value pairs compacted. When
1632 * we remove one, we have to shift any that preceded it towards the end.
1634 static void ocfs2_xa_block_wipe_namevalue(struct ocfs2_xa_loc *loc)
1637 int namevalue_offset, first_namevalue_offset, namevalue_size;
1638 struct ocfs2_xattr_entry *entry = loc->xl_entry;
1639 struct ocfs2_xattr_header *xh = loc->xl_header;
1640 int count = le16_to_cpu(xh->xh_count);
1642 namevalue_offset = le16_to_cpu(entry->xe_name_offset);
1643 namevalue_size = namevalue_size_xe(entry);
1644 first_namevalue_offset = ocfs2_xa_get_free_start(loc);
1646 /* Shift the name+value pairs */
1647 memmove((char *)xh + first_namevalue_offset + namevalue_size,
1648 (char *)xh + first_namevalue_offset,
1649 namevalue_offset - first_namevalue_offset);
1650 memset((char *)xh + first_namevalue_offset, 0, namevalue_size);
1652 /* Now tell xh->xh_entries about it */
1653 for (i = 0; i < count; i++) {
1654 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1655 if (offset <= namevalue_offset)
1656 le16_add_cpu(&xh->xh_entries[i].xe_name_offset,
1661 * Note that we don't update xh_free_start or xh_name_value_len
1662 * because they're not used in block-stored xattrs.
1666 static void ocfs2_xa_block_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1668 int count = le16_to_cpu(loc->xl_header->xh_count);
1669 loc->xl_entry = &(loc->xl_header->xh_entries[count]);
1670 le16_add_cpu(&loc->xl_header->xh_count, 1);
1671 memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1674 static void ocfs2_xa_block_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1676 int free_start = ocfs2_xa_get_free_start(loc);
1678 loc->xl_entry->xe_name_offset = cpu_to_le16(free_start - size);
1681 static void ocfs2_xa_block_fill_value_buf(struct ocfs2_xa_loc *loc,
1682 struct ocfs2_xattr_value_buf *vb)
1684 struct buffer_head *bh = loc->xl_storage;
1686 if (loc->xl_size == (bh->b_size -
1687 offsetof(struct ocfs2_xattr_block,
1688 xb_attrs.xb_header)))
1689 vb->vb_access = ocfs2_journal_access_xb;
1691 vb->vb_access = ocfs2_journal_access_di;
1696 * Operations for xattrs stored in blocks. This includes inline inode
1697 * storage and unindexed ocfs2_xattr_blocks.
1699 static const struct ocfs2_xa_loc_operations ocfs2_xa_block_loc_ops = {
1700 .xlo_journal_access = ocfs2_xa_block_journal_access,
1701 .xlo_journal_dirty = ocfs2_xa_block_journal_dirty,
1702 .xlo_offset_pointer = ocfs2_xa_block_offset_pointer,
1703 .xlo_check_space = ocfs2_xa_block_check_space,
1704 .xlo_can_reuse = ocfs2_xa_block_can_reuse,
1705 .xlo_get_free_start = ocfs2_xa_block_get_free_start,
1706 .xlo_wipe_namevalue = ocfs2_xa_block_wipe_namevalue,
1707 .xlo_add_entry = ocfs2_xa_block_add_entry,
1708 .xlo_add_namevalue = ocfs2_xa_block_add_namevalue,
1709 .xlo_fill_value_buf = ocfs2_xa_block_fill_value_buf,
1712 static int ocfs2_xa_bucket_journal_access(handle_t *handle,
1713 struct ocfs2_xa_loc *loc, int type)
1715 struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1717 return ocfs2_xattr_bucket_journal_access(handle, bucket, type);
1720 static void ocfs2_xa_bucket_journal_dirty(handle_t *handle,
1721 struct ocfs2_xa_loc *loc)
1723 struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1725 ocfs2_xattr_bucket_journal_dirty(handle, bucket);
1728 static void *ocfs2_xa_bucket_offset_pointer(struct ocfs2_xa_loc *loc,
1731 struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1732 int block, block_offset;
1734 /* The header is at the front of the bucket */
1735 block = offset >> loc->xl_inode->i_sb->s_blocksize_bits;
1736 block_offset = offset % loc->xl_inode->i_sb->s_blocksize;
1738 return bucket_block(bucket, block) + block_offset;
1741 static int ocfs2_xa_bucket_can_reuse(struct ocfs2_xa_loc *loc,
1742 struct ocfs2_xattr_info *xi)
1744 return namevalue_size_xe(loc->xl_entry) >=
1745 namevalue_size_xi(xi);
1748 static int ocfs2_xa_bucket_get_free_start(struct ocfs2_xa_loc *loc)
1750 struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1751 return le16_to_cpu(bucket_xh(bucket)->xh_free_start);
1754 static int ocfs2_bucket_align_free_start(struct super_block *sb,
1755 int free_start, int size)
1758 * We need to make sure that the name+value pair fits within
1761 if (((free_start - size) >> sb->s_blocksize_bits) !=
1762 ((free_start - 1) >> sb->s_blocksize_bits))
1763 free_start -= free_start % sb->s_blocksize;
1768 static int ocfs2_xa_bucket_check_space(struct ocfs2_xa_loc *loc,
1769 struct ocfs2_xattr_info *xi)
1772 int count = le16_to_cpu(loc->xl_header->xh_count);
1773 int free_start = ocfs2_xa_get_free_start(loc);
1774 int needed_space = ocfs2_xi_entry_usage(xi);
1775 int size = namevalue_size_xi(xi);
1776 struct super_block *sb = loc->xl_inode->i_sb;
1779 * Bucket storage does not reclaim name+value pairs it cannot
1780 * reuse. They live as holes until the bucket fills, and then
1781 * the bucket is defragmented. However, the bucket can reclaim
1782 * the ocfs2_xattr_entry.
1784 if (loc->xl_entry) {
1785 /* Don't need space if we're reusing! */
1786 if (ocfs2_xa_can_reuse_entry(loc, xi))
1789 needed_space -= sizeof(struct ocfs2_xattr_entry);
1791 BUG_ON(needed_space < 0);
1793 if (free_start < size) {
1798 * First we check if it would fit in the first place.
1799 * Below, we align the free start to a block. This may
1800 * slide us below the minimum gap. By checking unaligned
1801 * first, we avoid that error.
1803 rc = ocfs2_xa_check_space_helper(needed_space, free_start,
1807 free_start = ocfs2_bucket_align_free_start(sb, free_start,
1810 return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1813 static void ocfs2_xa_bucket_wipe_namevalue(struct ocfs2_xa_loc *loc)
1815 le16_add_cpu(&loc->xl_header->xh_name_value_len,
1816 -namevalue_size_xe(loc->xl_entry));
1819 static void ocfs2_xa_bucket_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1821 struct ocfs2_xattr_header *xh = loc->xl_header;
1822 int count = le16_to_cpu(xh->xh_count);
1823 int low = 0, high = count - 1, tmp;
1824 struct ocfs2_xattr_entry *tmp_xe;
1827 * We keep buckets sorted by name_hash, so we need to find
1830 while (low <= high && count) {
1831 tmp = (low + high) / 2;
1832 tmp_xe = &xh->xh_entries[tmp];
1834 if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
1836 else if (name_hash < le32_to_cpu(tmp_xe->xe_name_hash))
1845 memmove(&xh->xh_entries[low + 1],
1846 &xh->xh_entries[low],
1847 ((count - low) * sizeof(struct ocfs2_xattr_entry)));
1849 le16_add_cpu(&xh->xh_count, 1);
1850 loc->xl_entry = &xh->xh_entries[low];
1851 memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1854 static void ocfs2_xa_bucket_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1856 int free_start = ocfs2_xa_get_free_start(loc);
1857 struct ocfs2_xattr_header *xh = loc->xl_header;
1858 struct super_block *sb = loc->xl_inode->i_sb;
1861 free_start = ocfs2_bucket_align_free_start(sb, free_start, size);
1862 nameval_offset = free_start - size;
1863 loc->xl_entry->xe_name_offset = cpu_to_le16(nameval_offset);
1864 xh->xh_free_start = cpu_to_le16(nameval_offset);
1865 le16_add_cpu(&xh->xh_name_value_len, size);
1869 static void ocfs2_xa_bucket_fill_value_buf(struct ocfs2_xa_loc *loc,
1870 struct ocfs2_xattr_value_buf *vb)
1872 struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1873 struct super_block *sb = loc->xl_inode->i_sb;
1874 int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1875 int size = namevalue_size_xe(loc->xl_entry);
1876 int block_offset = nameval_offset >> sb->s_blocksize_bits;
1878 /* Values are not allowed to straddle block boundaries */
1879 BUG_ON(block_offset !=
1880 ((nameval_offset + size - 1) >> sb->s_blocksize_bits));
1881 /* We expect the bucket to be filled in */
1882 BUG_ON(!bucket->bu_bhs[block_offset]);
1884 vb->vb_access = ocfs2_journal_access;
1885 vb->vb_bh = bucket->bu_bhs[block_offset];
1888 /* Operations for xattrs stored in buckets. */
1889 static const struct ocfs2_xa_loc_operations ocfs2_xa_bucket_loc_ops = {
1890 .xlo_journal_access = ocfs2_xa_bucket_journal_access,
1891 .xlo_journal_dirty = ocfs2_xa_bucket_journal_dirty,
1892 .xlo_offset_pointer = ocfs2_xa_bucket_offset_pointer,
1893 .xlo_check_space = ocfs2_xa_bucket_check_space,
1894 .xlo_can_reuse = ocfs2_xa_bucket_can_reuse,
1895 .xlo_get_free_start = ocfs2_xa_bucket_get_free_start,
1896 .xlo_wipe_namevalue = ocfs2_xa_bucket_wipe_namevalue,
1897 .xlo_add_entry = ocfs2_xa_bucket_add_entry,
1898 .xlo_add_namevalue = ocfs2_xa_bucket_add_namevalue,
1899 .xlo_fill_value_buf = ocfs2_xa_bucket_fill_value_buf,
1902 static unsigned int ocfs2_xa_value_clusters(struct ocfs2_xa_loc *loc)
1904 struct ocfs2_xattr_value_buf vb;
1906 if (ocfs2_xattr_is_local(loc->xl_entry))
1909 ocfs2_xa_fill_value_buf(loc, &vb);
1910 return le32_to_cpu(vb.vb_xv->xr_clusters);
1913 static int ocfs2_xa_value_truncate(struct ocfs2_xa_loc *loc, u64 bytes,
1914 struct ocfs2_xattr_set_ctxt *ctxt)
1916 int trunc_rc, access_rc;
1917 struct ocfs2_xattr_value_buf vb;
1919 ocfs2_xa_fill_value_buf(loc, &vb);
1920 trunc_rc = ocfs2_xattr_value_truncate(loc->xl_inode, &vb, bytes,
1924 * The caller of ocfs2_xa_value_truncate() has already called
1925 * ocfs2_xa_journal_access on the loc. However, The truncate code
1926 * calls ocfs2_extend_trans(). This may commit the previous
1927 * transaction and open a new one. If this is a bucket, truncate
1928 * could leave only vb->vb_bh set up for journaling. Meanwhile,
1929 * the caller is expecting to dirty the entire bucket. So we must
1930 * reset the journal work. We do this even if truncate has failed,
1931 * as it could have failed after committing the extend.
1933 access_rc = ocfs2_xa_journal_access(ctxt->handle, loc,
1934 OCFS2_JOURNAL_ACCESS_WRITE);
1936 /* Errors in truncate take precedence */
1937 return trunc_rc ? trunc_rc : access_rc;
1940 static void ocfs2_xa_remove_entry(struct ocfs2_xa_loc *loc)
1943 struct ocfs2_xattr_header *xh = loc->xl_header;
1944 struct ocfs2_xattr_entry *entry = loc->xl_entry;
1946 ocfs2_xa_wipe_namevalue(loc);
1947 loc->xl_entry = NULL;
1949 le16_add_cpu(&xh->xh_count, -1);
1950 count = le16_to_cpu(xh->xh_count);
1953 * Only zero out the entry if there are more remaining. This is
1954 * important for an empty bucket, as it keeps track of the
1955 * bucket's hash value. It doesn't hurt empty block storage.
1958 index = ((char *)entry - (char *)&xh->xh_entries) /
1959 sizeof(struct ocfs2_xattr_entry);
1960 memmove(&xh->xh_entries[index], &xh->xh_entries[index + 1],
1961 (count - index) * sizeof(struct ocfs2_xattr_entry));
1962 memset(&xh->xh_entries[count], 0,
1963 sizeof(struct ocfs2_xattr_entry));
1968 * If we have a problem adjusting the size of an external value during
1969 * ocfs2_xa_prepare_entry() or ocfs2_xa_remove(), we may have an xattr
1970 * in an intermediate state. For example, the value may be partially
1973 * If the value tree hasn't changed, the extend/truncate went nowhere.
1974 * We have nothing to do. The caller can treat it as a straight error.
1976 * If the value tree got partially truncated, we now have a corrupted
1977 * extended attribute. We're going to wipe its entry and leak the
1978 * clusters. Better to leak some storage than leave a corrupt entry.
1980 * If the value tree grew, it obviously didn't grow enough for the
1981 * new entry. We're not going to try and reclaim those clusters either.
1982 * If there was already an external value there (orig_clusters != 0),
1983 * the new clusters are attached safely and we can just leave the old
1984 * value in place. If there was no external value there, we remove
1987 * This way, the xattr block we store in the journal will be consistent.
1988 * If the size change broke because of the journal, no changes will hit
1991 static void ocfs2_xa_cleanup_value_truncate(struct ocfs2_xa_loc *loc,
1993 unsigned int orig_clusters)
1995 unsigned int new_clusters = ocfs2_xa_value_clusters(loc);
1996 char *nameval_buf = ocfs2_xa_offset_pointer(loc,
1997 le16_to_cpu(loc->xl_entry->xe_name_offset));
1999 if (new_clusters < orig_clusters) {
2001 "Partial truncate while %s xattr %.*s. Leaking "
2002 "%u clusters and removing the entry\n",
2003 what, loc->xl_entry->xe_name_len, nameval_buf,
2004 orig_clusters - new_clusters);
2005 ocfs2_xa_remove_entry(loc);
2006 } else if (!orig_clusters) {
2008 "Unable to allocate an external value for xattr "
2009 "%.*s safely. Leaking %u clusters and removing the "
2011 loc->xl_entry->xe_name_len, nameval_buf,
2012 new_clusters - orig_clusters);
2013 ocfs2_xa_remove_entry(loc);
2014 } else if (new_clusters > orig_clusters)
2016 "Unable to grow xattr %.*s safely. %u new clusters "
2017 "have been added, but the value will not be "
2019 loc->xl_entry->xe_name_len, nameval_buf,
2020 new_clusters - orig_clusters);
2023 static int ocfs2_xa_remove(struct ocfs2_xa_loc *loc,
2024 struct ocfs2_xattr_set_ctxt *ctxt)
2027 unsigned int orig_clusters;
2029 if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2030 orig_clusters = ocfs2_xa_value_clusters(loc);
2031 rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2035 * Since this is remove, we can return 0 if
2036 * ocfs2_xa_cleanup_value_truncate() is going to
2037 * wipe the entry anyway. So we check the
2038 * cluster count as well.
2040 if (orig_clusters != ocfs2_xa_value_clusters(loc))
2042 ocfs2_xa_cleanup_value_truncate(loc, "removing",
2049 ocfs2_xa_remove_entry(loc);
2055 static void ocfs2_xa_install_value_root(struct ocfs2_xa_loc *loc)
2057 int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
2060 nameval_buf = ocfs2_xa_offset_pointer(loc,
2061 le16_to_cpu(loc->xl_entry->xe_name_offset));
2062 memcpy(nameval_buf + name_size, &def_xv, OCFS2_XATTR_ROOT_SIZE);
2066 * Take an existing entry and make it ready for the new value. This
2067 * won't allocate space, but it may free space. It should be ready for
2068 * ocfs2_xa_prepare_entry() to finish the work.
2070 static int ocfs2_xa_reuse_entry(struct ocfs2_xa_loc *loc,
2071 struct ocfs2_xattr_info *xi,
2072 struct ocfs2_xattr_set_ctxt *ctxt)
2075 int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2076 unsigned int orig_clusters;
2078 int xe_local = ocfs2_xattr_is_local(loc->xl_entry);
2079 int xi_local = xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE;
2081 BUG_ON(OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len) !=
2084 nameval_buf = ocfs2_xa_offset_pointer(loc,
2085 le16_to_cpu(loc->xl_entry->xe_name_offset));
2087 memset(nameval_buf + name_size, 0,
2088 namevalue_size_xe(loc->xl_entry) - name_size);
2090 ocfs2_xa_install_value_root(loc);
2092 orig_clusters = ocfs2_xa_value_clusters(loc);
2094 rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2098 memset(nameval_buf + name_size, 0,
2099 namevalue_size_xe(loc->xl_entry) -
2101 } else if (le64_to_cpu(loc->xl_entry->xe_value_size) >
2103 rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len,
2110 ocfs2_xa_cleanup_value_truncate(loc, "reusing",
2116 loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
2117 ocfs2_xattr_set_local(loc->xl_entry, xi_local);
2124 * Prepares loc->xl_entry to receive the new xattr. This includes
2125 * properly setting up the name+value pair region. If loc->xl_entry
2126 * already exists, it will take care of modifying it appropriately.
2128 * Note that this modifies the data. You did journal_access already,
2131 static int ocfs2_xa_prepare_entry(struct ocfs2_xa_loc *loc,
2132 struct ocfs2_xattr_info *xi,
2134 struct ocfs2_xattr_set_ctxt *ctxt)
2137 unsigned int orig_clusters;
2138 __le64 orig_value_size = 0;
2140 rc = ocfs2_xa_check_space(loc, xi);
2144 if (loc->xl_entry) {
2145 if (ocfs2_xa_can_reuse_entry(loc, xi)) {
2146 orig_value_size = loc->xl_entry->xe_value_size;
2147 rc = ocfs2_xa_reuse_entry(loc, xi, ctxt);
2153 if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2154 orig_clusters = ocfs2_xa_value_clusters(loc);
2155 rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2158 ocfs2_xa_cleanup_value_truncate(loc,
2164 ocfs2_xa_wipe_namevalue(loc);
2166 ocfs2_xa_add_entry(loc, name_hash);
2169 * If we get here, we have a blank entry. Fill it. We grow our
2170 * name+value pair back from the end.
2172 ocfs2_xa_add_namevalue(loc, xi);
2173 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
2174 ocfs2_xa_install_value_root(loc);
2177 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2178 orig_clusters = ocfs2_xa_value_clusters(loc);
2179 rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len, ctxt);
2181 ctxt->set_abort = 1;
2182 ocfs2_xa_cleanup_value_truncate(loc, "growing",
2185 * If we were growing an existing value,
2186 * ocfs2_xa_cleanup_value_truncate() won't remove
2187 * the entry. We need to restore the original value
2190 if (loc->xl_entry) {
2191 BUG_ON(!orig_value_size);
2192 loc->xl_entry->xe_value_size = orig_value_size;
2203 * Store the value portion of the name+value pair. This will skip
2204 * values that are stored externally. Their tree roots were set up
2205 * by ocfs2_xa_prepare_entry().
2207 static int ocfs2_xa_store_value(struct ocfs2_xa_loc *loc,
2208 struct ocfs2_xattr_info *xi,
2209 struct ocfs2_xattr_set_ctxt *ctxt)
2212 int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
2213 int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2215 struct ocfs2_xattr_value_buf vb;
2217 nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
2218 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2219 ocfs2_xa_fill_value_buf(loc, &vb);
2220 rc = __ocfs2_xattr_set_value_outside(loc->xl_inode,
2225 memcpy(nameval_buf + name_size, xi->xi_value, xi->xi_value_len);
2230 static int ocfs2_xa_set(struct ocfs2_xa_loc *loc,
2231 struct ocfs2_xattr_info *xi,
2232 struct ocfs2_xattr_set_ctxt *ctxt)
2235 u32 name_hash = ocfs2_xattr_name_hash(loc->xl_inode, xi->xi_name,
2238 ret = ocfs2_xa_journal_access(ctxt->handle, loc,
2239 OCFS2_JOURNAL_ACCESS_WRITE);
2246 * From here on out, everything is going to modify the buffer a
2247 * little. Errors are going to leave the xattr header in a
2248 * sane state. Thus, even with errors we dirty the sucker.
2251 /* Don't worry, we are never called with !xi_value and !xl_entry */
2252 if (!xi->xi_value) {
2253 ret = ocfs2_xa_remove(loc, ctxt);
2257 ret = ocfs2_xa_prepare_entry(loc, xi, name_hash, ctxt);
2264 ret = ocfs2_xa_store_value(loc, xi, ctxt);
2269 ocfs2_xa_journal_dirty(ctxt->handle, loc);
2275 static void ocfs2_init_dinode_xa_loc(struct ocfs2_xa_loc *loc,
2276 struct inode *inode,
2277 struct buffer_head *bh,
2278 struct ocfs2_xattr_entry *entry)
2280 struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2282 BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_XATTR_FL));
2284 loc->xl_inode = inode;
2285 loc->xl_ops = &ocfs2_xa_block_loc_ops;
2286 loc->xl_storage = bh;
2287 loc->xl_entry = entry;
2288 loc->xl_size = le16_to_cpu(di->i_xattr_inline_size);
2290 (struct ocfs2_xattr_header *)(bh->b_data + bh->b_size -
2294 static void ocfs2_init_xattr_block_xa_loc(struct ocfs2_xa_loc *loc,
2295 struct inode *inode,
2296 struct buffer_head *bh,
2297 struct ocfs2_xattr_entry *entry)
2299 struct ocfs2_xattr_block *xb =
2300 (struct ocfs2_xattr_block *)bh->b_data;
2302 BUG_ON(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED);
2304 loc->xl_inode = inode;
2305 loc->xl_ops = &ocfs2_xa_block_loc_ops;
2306 loc->xl_storage = bh;
2307 loc->xl_header = &(xb->xb_attrs.xb_header);
2308 loc->xl_entry = entry;
2309 loc->xl_size = bh->b_size - offsetof(struct ocfs2_xattr_block,
2310 xb_attrs.xb_header);
2313 static void ocfs2_init_xattr_bucket_xa_loc(struct ocfs2_xa_loc *loc,
2314 struct ocfs2_xattr_bucket *bucket,
2315 struct ocfs2_xattr_entry *entry)
2317 loc->xl_inode = bucket->bu_inode;
2318 loc->xl_ops = &ocfs2_xa_bucket_loc_ops;
2319 loc->xl_storage = bucket;
2320 loc->xl_header = bucket_xh(bucket);
2321 loc->xl_entry = entry;
2322 loc->xl_size = OCFS2_XATTR_BUCKET_SIZE;
2326 * In xattr remove, if it is stored outside and refcounted, we may have
2327 * the chance to split the refcount tree. So need the allocators.
2329 static int ocfs2_lock_xattr_remove_allocators(struct inode *inode,
2330 struct ocfs2_xattr_value_root *xv,
2331 struct ocfs2_caching_info *ref_ci,
2332 struct buffer_head *ref_root_bh,
2333 struct ocfs2_alloc_context **meta_ac,
2336 int ret, meta_add = 0;
2337 u32 p_cluster, num_clusters;
2338 unsigned int ext_flags;
2341 ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
2350 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
2353 ret = ocfs2_refcounted_xattr_delete_need(inode, ref_ci,
2355 &meta_add, ref_credits);
2361 ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
2370 static int ocfs2_remove_value_outside(struct inode*inode,
2371 struct ocfs2_xattr_value_buf *vb,
2372 struct ocfs2_xattr_header *header,
2373 struct ocfs2_caching_info *ref_ci,
2374 struct buffer_head *ref_root_bh)
2376 int ret = 0, i, ref_credits;
2377 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2378 struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2381 ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
2383 for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
2384 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
2386 if (ocfs2_xattr_is_local(entry))
2389 val = (void *)header +
2390 le16_to_cpu(entry->xe_name_offset);
2391 vb->vb_xv = (struct ocfs2_xattr_value_root *)
2392 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
2394 ret = ocfs2_lock_xattr_remove_allocators(inode, vb->vb_xv,
2395 ref_ci, ref_root_bh,
2399 ctxt.handle = ocfs2_start_trans(osb, ref_credits +
2400 ocfs2_remove_extent_credits(osb->sb));
2401 if (IS_ERR(ctxt.handle)) {
2402 ret = PTR_ERR(ctxt.handle);
2407 ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
2409 ocfs2_commit_trans(osb, ctxt.handle);
2411 ocfs2_free_alloc_context(ctxt.meta_ac);
2412 ctxt.meta_ac = NULL;
2423 ocfs2_free_alloc_context(ctxt.meta_ac);
2424 ocfs2_schedule_truncate_log_flush(osb, 1);
2425 ocfs2_run_deallocs(osb, &ctxt.dealloc);
2429 static int ocfs2_xattr_ibody_remove(struct inode *inode,
2430 struct buffer_head *di_bh,
2431 struct ocfs2_caching_info *ref_ci,
2432 struct buffer_head *ref_root_bh)
2435 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2436 struct ocfs2_xattr_header *header;
2438 struct ocfs2_xattr_value_buf vb = {
2440 .vb_access = ocfs2_journal_access_di,
2443 header = (struct ocfs2_xattr_header *)
2444 ((void *)di + inode->i_sb->s_blocksize -
2445 le16_to_cpu(di->i_xattr_inline_size));
2447 ret = ocfs2_remove_value_outside(inode, &vb, header,
2448 ref_ci, ref_root_bh);
2453 struct ocfs2_rm_xattr_bucket_para {
2454 struct ocfs2_caching_info *ref_ci;
2455 struct buffer_head *ref_root_bh;
2458 static int ocfs2_xattr_block_remove(struct inode *inode,
2459 struct buffer_head *blk_bh,
2460 struct ocfs2_caching_info *ref_ci,
2461 struct buffer_head *ref_root_bh)
2463 struct ocfs2_xattr_block *xb;
2465 struct ocfs2_xattr_value_buf vb = {
2467 .vb_access = ocfs2_journal_access_xb,
2469 struct ocfs2_rm_xattr_bucket_para args = {
2471 .ref_root_bh = ref_root_bh,
2474 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2475 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2476 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
2477 ret = ocfs2_remove_value_outside(inode, &vb, header,
2478 ref_ci, ref_root_bh);
2480 ret = ocfs2_iterate_xattr_index_block(inode,
2482 ocfs2_rm_xattr_cluster,
2488 static int ocfs2_xattr_free_block(struct inode *inode,
2490 struct ocfs2_caching_info *ref_ci,
2491 struct buffer_head *ref_root_bh)
2493 struct inode *xb_alloc_inode;
2494 struct buffer_head *xb_alloc_bh = NULL;
2495 struct buffer_head *blk_bh = NULL;
2496 struct ocfs2_xattr_block *xb;
2497 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2503 ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
2509 ret = ocfs2_xattr_block_remove(inode, blk_bh, ref_ci, ref_root_bh);
2515 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2516 blk = le64_to_cpu(xb->xb_blkno);
2517 bit = le16_to_cpu(xb->xb_suballoc_bit);
2518 if (xb->xb_suballoc_loc)
2519 bg_blkno = le64_to_cpu(xb->xb_suballoc_loc);
2521 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2523 xb_alloc_inode = ocfs2_get_system_file_inode(osb,
2524 EXTENT_ALLOC_SYSTEM_INODE,
2525 le16_to_cpu(xb->xb_suballoc_slot));
2526 if (!xb_alloc_inode) {
2531 inode_lock(xb_alloc_inode);
2533 ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
2539 handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
2540 if (IS_ERR(handle)) {
2541 ret = PTR_ERR(handle);
2546 ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
2551 ocfs2_commit_trans(osb, handle);
2553 ocfs2_inode_unlock(xb_alloc_inode, 1);
2554 brelse(xb_alloc_bh);
2556 inode_unlock(xb_alloc_inode);
2557 iput(xb_alloc_inode);
2564 * ocfs2_xattr_remove()
2566 * Free extended attribute resources associated with this inode.
2568 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
2570 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2571 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2572 struct ocfs2_refcount_tree *ref_tree = NULL;
2573 struct buffer_head *ref_root_bh = NULL;
2574 struct ocfs2_caching_info *ref_ci = NULL;
2578 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2581 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
2584 if (ocfs2_is_refcount_inode(inode)) {
2585 ret = ocfs2_lock_refcount_tree(OCFS2_SB(inode->i_sb),
2586 le64_to_cpu(di->i_refcount_loc),
2587 1, &ref_tree, &ref_root_bh);
2592 ref_ci = &ref_tree->rf_ci;
2596 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2597 ret = ocfs2_xattr_ibody_remove(inode, di_bh,
2598 ref_ci, ref_root_bh);
2605 if (di->i_xattr_loc) {
2606 ret = ocfs2_xattr_free_block(inode,
2607 le64_to_cpu(di->i_xattr_loc),
2608 ref_ci, ref_root_bh);
2615 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
2616 OCFS2_INODE_UPDATE_CREDITS);
2617 if (IS_ERR(handle)) {
2618 ret = PTR_ERR(handle);
2622 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2623 OCFS2_JOURNAL_ACCESS_WRITE);
2629 di->i_xattr_loc = 0;
2631 spin_lock(&oi->ip_lock);
2632 oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
2633 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2634 spin_unlock(&oi->ip_lock);
2635 ocfs2_update_inode_fsync_trans(handle, inode, 0);
2637 ocfs2_journal_dirty(handle, di_bh);
2639 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
2642 ocfs2_unlock_refcount_tree(OCFS2_SB(inode->i_sb), ref_tree, 1);
2643 brelse(ref_root_bh);
2647 static int ocfs2_xattr_has_space_inline(struct inode *inode,
2648 struct ocfs2_dinode *di)
2650 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2651 unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
2654 if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
2657 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2658 struct ocfs2_inline_data *idata = &di->id2.i_data;
2659 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
2660 } else if (ocfs2_inode_is_fast_symlink(inode)) {
2661 free = ocfs2_fast_symlink_chars(inode->i_sb) -
2662 le64_to_cpu(di->i_size);
2664 struct ocfs2_extent_list *el = &di->id2.i_list;
2665 free = (le16_to_cpu(el->l_count) -
2666 le16_to_cpu(el->l_next_free_rec)) *
2667 sizeof(struct ocfs2_extent_rec);
2669 if (free >= xattrsize)
2676 * ocfs2_xattr_ibody_find()
2678 * Find extended attribute in inode block and
2679 * fill search info into struct ocfs2_xattr_search.
2681 static int ocfs2_xattr_ibody_find(struct inode *inode,
2684 struct ocfs2_xattr_search *xs)
2686 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2687 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2691 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2694 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2695 down_read(&oi->ip_alloc_sem);
2696 has_space = ocfs2_xattr_has_space_inline(inode, di);
2697 up_read(&oi->ip_alloc_sem);
2702 xs->xattr_bh = xs->inode_bh;
2703 xs->end = (void *)di + inode->i_sb->s_blocksize;
2704 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
2705 xs->header = (struct ocfs2_xattr_header *)
2706 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
2708 xs->header = (struct ocfs2_xattr_header *)
2709 (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
2710 xs->base = (void *)xs->header;
2711 xs->here = xs->header->xh_entries;
2713 /* Find the named attribute. */
2714 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2715 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2716 if (ret && ret != -ENODATA)
2718 xs->not_found = ret;
2724 static int ocfs2_xattr_ibody_init(struct inode *inode,
2725 struct buffer_head *di_bh,
2726 struct ocfs2_xattr_set_ctxt *ctxt)
2729 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2730 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2731 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2732 unsigned int xattrsize = osb->s_xattr_inline_size;
2734 if (!ocfs2_xattr_has_space_inline(inode, di)) {
2739 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode), di_bh,
2740 OCFS2_JOURNAL_ACCESS_WRITE);
2747 * Adjust extent record count or inline data size
2748 * to reserve space for extended attribute.
2750 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2751 struct ocfs2_inline_data *idata = &di->id2.i_data;
2752 le16_add_cpu(&idata->id_count, -xattrsize);
2753 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
2754 struct ocfs2_extent_list *el = &di->id2.i_list;
2755 le16_add_cpu(&el->l_count, -(xattrsize /
2756 sizeof(struct ocfs2_extent_rec)));
2758 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
2760 spin_lock(&oi->ip_lock);
2761 oi->ip_dyn_features |= OCFS2_INLINE_XATTR_FL|OCFS2_HAS_XATTR_FL;
2762 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2763 spin_unlock(&oi->ip_lock);
2765 ocfs2_journal_dirty(ctxt->handle, di_bh);
2772 * ocfs2_xattr_ibody_set()
2774 * Set, replace or remove an extended attribute into inode block.
2777 static int ocfs2_xattr_ibody_set(struct inode *inode,
2778 struct ocfs2_xattr_info *xi,
2779 struct ocfs2_xattr_search *xs,
2780 struct ocfs2_xattr_set_ctxt *ctxt)
2783 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2784 struct ocfs2_xa_loc loc;
2786 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2789 down_write(&oi->ip_alloc_sem);
2790 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2791 ret = ocfs2_xattr_ibody_init(inode, xs->inode_bh, ctxt);
2799 ocfs2_init_dinode_xa_loc(&loc, inode, xs->inode_bh,
2800 xs->not_found ? NULL : xs->here);
2801 ret = ocfs2_xa_set(&loc, xi, ctxt);
2807 xs->here = loc.xl_entry;
2810 up_write(&oi->ip_alloc_sem);
2816 * ocfs2_xattr_block_find()
2818 * Find extended attribute in external block and
2819 * fill search info into struct ocfs2_xattr_search.
2821 static int ocfs2_xattr_block_find(struct inode *inode,
2824 struct ocfs2_xattr_search *xs)
2826 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2827 struct buffer_head *blk_bh = NULL;
2828 struct ocfs2_xattr_block *xb;
2831 if (!di->i_xattr_loc)
2834 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2841 xs->xattr_bh = blk_bh;
2842 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2844 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2845 xs->header = &xb->xb_attrs.xb_header;
2846 xs->base = (void *)xs->header;
2847 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2848 xs->here = xs->header->xh_entries;
2850 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2852 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2856 if (ret && ret != -ENODATA) {
2857 xs->xattr_bh = NULL;
2860 xs->not_found = ret;
2868 static int ocfs2_create_xattr_block(struct inode *inode,
2869 struct buffer_head *inode_bh,
2870 struct ocfs2_xattr_set_ctxt *ctxt,
2872 struct buffer_head **ret_bh)
2875 u16 suballoc_bit_start;
2877 u64 suballoc_loc, first_blkno;
2878 struct ocfs2_dinode *di = (struct ocfs2_dinode *)inode_bh->b_data;
2879 struct buffer_head *new_bh = NULL;
2880 struct ocfs2_xattr_block *xblk;
2882 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
2883 inode_bh, OCFS2_JOURNAL_ACCESS_CREATE);
2889 ret = ocfs2_claim_metadata(ctxt->handle, ctxt->meta_ac, 1,
2890 &suballoc_loc, &suballoc_bit_start,
2891 &num_got, &first_blkno);
2897 new_bh = sb_getblk(inode->i_sb, first_blkno);
2904 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2906 ret = ocfs2_journal_access_xb(ctxt->handle, INODE_CACHE(inode),
2908 OCFS2_JOURNAL_ACCESS_CREATE);
2914 /* Initialize ocfs2_xattr_block */
2915 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2916 memset(xblk, 0, inode->i_sb->s_blocksize);
2917 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2918 xblk->xb_suballoc_slot = cpu_to_le16(ctxt->meta_ac->ac_alloc_slot);
2919 xblk->xb_suballoc_loc = cpu_to_le64(suballoc_loc);
2920 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2921 xblk->xb_fs_generation =
2922 cpu_to_le32(OCFS2_SB(inode->i_sb)->fs_generation);
2923 xblk->xb_blkno = cpu_to_le64(first_blkno);
2925 struct ocfs2_xattr_tree_root *xr = &xblk->xb_attrs.xb_root;
2926 xr->xt_clusters = cpu_to_le32(1);
2927 xr->xt_last_eb_blk = 0;
2928 xr->xt_list.l_tree_depth = 0;
2929 xr->xt_list.l_count = cpu_to_le16(
2930 ocfs2_xattr_recs_per_xb(inode->i_sb));
2931 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2932 xblk->xb_flags = cpu_to_le16(OCFS2_XATTR_INDEXED);
2934 ocfs2_journal_dirty(ctxt->handle, new_bh);
2936 /* Add it to the inode */
2937 di->i_xattr_loc = cpu_to_le64(first_blkno);
2939 spin_lock(&OCFS2_I(inode)->ip_lock);
2940 OCFS2_I(inode)->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
2941 di->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
2942 spin_unlock(&OCFS2_I(inode)->ip_lock);
2944 ocfs2_journal_dirty(ctxt->handle, inode_bh);
2955 * ocfs2_xattr_block_set()
2957 * Set, replace or remove an extended attribute into external block.
2960 static int ocfs2_xattr_block_set(struct inode *inode,
2961 struct ocfs2_xattr_info *xi,
2962 struct ocfs2_xattr_search *xs,
2963 struct ocfs2_xattr_set_ctxt *ctxt)
2965 struct buffer_head *new_bh = NULL;
2966 struct ocfs2_xattr_block *xblk = NULL;
2968 struct ocfs2_xa_loc loc;
2970 if (!xs->xattr_bh) {
2971 ret = ocfs2_create_xattr_block(inode, xs->inode_bh, ctxt,
2978 xs->xattr_bh = new_bh;
2979 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2980 xs->header = &xblk->xb_attrs.xb_header;
2981 xs->base = (void *)xs->header;
2982 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2983 xs->here = xs->header->xh_entries;
2985 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2987 if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2988 ocfs2_init_xattr_block_xa_loc(&loc, inode, xs->xattr_bh,
2989 xs->not_found ? NULL : xs->here);
2991 ret = ocfs2_xa_set(&loc, xi, ctxt);
2993 xs->here = loc.xl_entry;
2994 else if ((ret != -ENOSPC) || ctxt->set_abort)
2997 ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
3003 if (le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)
3004 ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
3010 /* Check whether the new xattr can be inserted into the inode. */
3011 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
3012 struct ocfs2_xattr_info *xi,
3013 struct ocfs2_xattr_search *xs)
3015 struct ocfs2_xattr_entry *last;
3017 size_t min_offs = xs->end - xs->base;
3022 last = xs->header->xh_entries;
3024 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
3025 size_t offs = le16_to_cpu(last->xe_name_offset);
3026 if (offs < min_offs)
3031 free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
3035 BUG_ON(!xs->not_found);
3037 if (free >= (sizeof(struct ocfs2_xattr_entry) + namevalue_size_xi(xi)))
3043 static int ocfs2_calc_xattr_set_need(struct inode *inode,
3044 struct ocfs2_dinode *di,
3045 struct ocfs2_xattr_info *xi,
3046 struct ocfs2_xattr_search *xis,
3047 struct ocfs2_xattr_search *xbs,
3052 int ret = 0, old_in_xb = 0;
3053 int clusters_add = 0, meta_add = 0, credits = 0;
3054 struct buffer_head *bh = NULL;
3055 struct ocfs2_xattr_block *xb = NULL;
3056 struct ocfs2_xattr_entry *xe = NULL;
3057 struct ocfs2_xattr_value_root *xv = NULL;
3059 int name_offset, name_len = 0;
3060 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
3065 * Calculate the clusters we need to write.
3066 * No matter whether we replace an old one or add a new one,
3067 * we need this for writing.
3069 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
3070 credits += new_clusters *
3071 ocfs2_clusters_to_blocks(inode->i_sb, 1);
3073 if (xis->not_found && xbs->not_found) {
3074 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3076 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3077 clusters_add += new_clusters;
3078 credits += ocfs2_calc_extend_credits(inode->i_sb,
3079 &def_xv.xv.xr_list);
3085 if (!xis->not_found) {
3087 name_offset = le16_to_cpu(xe->xe_name_offset);
3088 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3090 credits += OCFS2_INODE_UPDATE_CREDITS;
3092 int i, block_off = 0;
3093 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3095 name_offset = le16_to_cpu(xe->xe_name_offset);
3096 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3097 i = xbs->here - xbs->header->xh_entries;
3100 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3101 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3102 bucket_xh(xbs->bucket),
3105 base = bucket_block(xbs->bucket, block_off);
3106 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3109 credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
3114 * delete a xattr doesn't need metadata and cluster allocation.
3115 * so just calculate the credits and return.
3117 * The credits for removing the value tree will be extended
3118 * by ocfs2_remove_extent itself.
3120 if (!xi->xi_value) {
3121 if (!ocfs2_xattr_is_local(xe))
3122 credits += ocfs2_remove_extent_credits(inode->i_sb);
3127 /* do cluster allocation guess first. */
3128 value_size = le64_to_cpu(xe->xe_value_size);
3132 * In xattr set, we always try to set the xe in inode first,
3133 * so if it can be inserted into inode successfully, the old
3134 * one will be removed from the xattr block, and this xattr
3135 * will be inserted into inode as a new xattr in inode.
3137 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
3138 clusters_add += new_clusters;
3139 credits += ocfs2_remove_extent_credits(inode->i_sb) +
3140 OCFS2_INODE_UPDATE_CREDITS;
3141 if (!ocfs2_xattr_is_local(xe))
3142 credits += ocfs2_calc_extend_credits(
3144 &def_xv.xv.xr_list);
3149 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3150 /* the new values will be stored outside. */
3151 u32 old_clusters = 0;
3153 if (!ocfs2_xattr_is_local(xe)) {
3154 old_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
3156 xv = (struct ocfs2_xattr_value_root *)
3157 (base + name_offset + name_len);
3158 value_size = OCFS2_XATTR_ROOT_SIZE;
3162 if (old_clusters >= new_clusters) {
3163 credits += ocfs2_remove_extent_credits(inode->i_sb);
3166 meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
3167 clusters_add += new_clusters - old_clusters;
3168 credits += ocfs2_calc_extend_credits(inode->i_sb,
3170 if (value_size >= OCFS2_XATTR_ROOT_SIZE)
3175 * Now the new value will be stored inside. So if the new
3176 * value is smaller than the size of value root or the old
3177 * value, we don't need any allocation, otherwise we have
3178 * to guess metadata allocation.
3180 if ((ocfs2_xattr_is_local(xe) &&
3181 (value_size >= xi->xi_value_len)) ||
3182 (!ocfs2_xattr_is_local(xe) &&
3183 OCFS2_XATTR_ROOT_SIZE >= xi->xi_value_len))
3188 /* calculate metadata allocation. */
3189 if (di->i_xattr_loc) {
3190 if (!xbs->xattr_bh) {
3191 ret = ocfs2_read_xattr_block(inode,
3192 le64_to_cpu(di->i_xattr_loc),
3199 xb = (struct ocfs2_xattr_block *)bh->b_data;
3201 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3204 * If there is already an xattr tree, good, we can calculate
3205 * like other b-trees. Otherwise we may have the chance of
3206 * create a tree, the credit calculation is borrowed from
3207 * ocfs2_calc_extend_credits with root_el = NULL. And the
3208 * new tree will be cluster based, so no meta is needed.
3210 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3211 struct ocfs2_extent_list *el =
3212 &xb->xb_attrs.xb_root.xt_list;
3213 meta_add += ocfs2_extend_meta_needed(el);
3214 credits += ocfs2_calc_extend_credits(inode->i_sb,
3217 credits += OCFS2_SUBALLOC_ALLOC + 1;
3220 * This cluster will be used either for new bucket or for
3222 * If the cluster size is the same as the bucket size, one
3223 * more is needed since we may need to extend the bucket
3227 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3228 if (OCFS2_XATTR_BUCKET_SIZE ==
3229 OCFS2_SB(inode->i_sb)->s_clustersize) {
3230 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3234 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
3235 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3236 struct ocfs2_extent_list *el = &def_xv.xv.xr_list;
3237 meta_add += ocfs2_extend_meta_needed(el);
3238 credits += ocfs2_calc_extend_credits(inode->i_sb,
3246 *clusters_need = clusters_add;
3248 *meta_need = meta_add;
3250 *credits_need = credits;
3255 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
3256 struct ocfs2_dinode *di,
3257 struct ocfs2_xattr_info *xi,
3258 struct ocfs2_xattr_search *xis,
3259 struct ocfs2_xattr_search *xbs,
3260 struct ocfs2_xattr_set_ctxt *ctxt,
3264 int clusters_add, meta_add, ret;
3265 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3267 memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
3269 ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
3271 ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
3272 &clusters_add, &meta_add, credits);
3278 meta_add += extra_meta;
3279 trace_ocfs2_init_xattr_set_ctxt(xi->xi_name, meta_add,
3280 clusters_add, *credits);
3283 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
3292 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
3298 if (ctxt->meta_ac) {
3299 ocfs2_free_alloc_context(ctxt->meta_ac);
3300 ctxt->meta_ac = NULL;
3304 * We cannot have an error and a non null ctxt->data_ac.
3311 static int __ocfs2_xattr_set_handle(struct inode *inode,
3312 struct ocfs2_dinode *di,
3313 struct ocfs2_xattr_info *xi,
3314 struct ocfs2_xattr_search *xis,
3315 struct ocfs2_xattr_search *xbs,
3316 struct ocfs2_xattr_set_ctxt *ctxt)
3318 int ret = 0, credits, old_found;
3320 if (!xi->xi_value) {
3321 /* Remove existing extended attribute */
3322 if (!xis->not_found)
3323 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3324 else if (!xbs->not_found)
3325 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3327 /* We always try to set extended attribute into inode first*/
3328 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3329 if (!ret && !xbs->not_found) {
3331 * If succeed and that extended attribute existing in
3332 * external block, then we will remove it.
3334 xi->xi_value = NULL;
3335 xi->xi_value_len = 0;
3337 old_found = xis->not_found;
3338 xis->not_found = -ENODATA;
3339 ret = ocfs2_calc_xattr_set_need(inode,
3347 xis->not_found = old_found;
3353 ret = ocfs2_extend_trans(ctxt->handle, credits);
3358 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3359 } else if ((ret == -ENOSPC) && !ctxt->set_abort) {
3360 if (di->i_xattr_loc && !xbs->xattr_bh) {
3361 ret = ocfs2_xattr_block_find(inode,
3367 old_found = xis->not_found;
3368 xis->not_found = -ENODATA;
3369 ret = ocfs2_calc_xattr_set_need(inode,
3377 xis->not_found = old_found;
3383 ret = ocfs2_extend_trans(ctxt->handle, credits);
3390 * If no space in inode, we will set extended attribute
3391 * into external block.
3393 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3396 if (!xis->not_found) {
3398 * If succeed and that extended attribute
3399 * existing in inode, we will remove it.
3401 xi->xi_value = NULL;
3402 xi->xi_value_len = 0;
3403 xbs->not_found = -ENODATA;
3404 ret = ocfs2_calc_xattr_set_need(inode,
3417 ret = ocfs2_extend_trans(ctxt->handle, credits);
3422 ret = ocfs2_xattr_ibody_set(inode, xi,
3429 /* Update inode ctime. */
3430 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
3432 OCFS2_JOURNAL_ACCESS_WRITE);
3438 inode->i_ctime = current_time(inode);
3439 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
3440 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
3441 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
3448 * This function only called duing creating inode
3449 * for init security/acl xattrs of the new inode.
3450 * All transanction credits have been reserved in mknod.
3452 int ocfs2_xattr_set_handle(handle_t *handle,
3453 struct inode *inode,
3454 struct buffer_head *di_bh,
3460 struct ocfs2_alloc_context *meta_ac,
3461 struct ocfs2_alloc_context *data_ac)
3463 struct ocfs2_dinode *di;
3466 struct ocfs2_xattr_info xi = {
3467 .xi_name_index = name_index,
3469 .xi_name_len = strlen(name),
3471 .xi_value_len = value_len,
3474 struct ocfs2_xattr_search xis = {
3475 .not_found = -ENODATA,
3478 struct ocfs2_xattr_search xbs = {
3479 .not_found = -ENODATA,
3482 struct ocfs2_xattr_set_ctxt ctxt = {
3488 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3492 * In extreme situation, may need xattr bucket when
3493 * block size is too small. And we have already reserved
3494 * the credits for bucket in mknod.
3496 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
3497 xbs.bucket = ocfs2_xattr_bucket_new(inode);
3499 mlog_errno(-ENOMEM);
3504 xis.inode_bh = xbs.inode_bh = di_bh;
3505 di = (struct ocfs2_dinode *)di_bh->b_data;
3507 down_write(&OCFS2_I(inode)->ip_xattr_sem);
3509 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3512 if (xis.not_found) {
3513 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3518 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3521 up_write(&OCFS2_I(inode)->ip_xattr_sem);
3522 brelse(xbs.xattr_bh);
3523 ocfs2_xattr_bucket_free(xbs.bucket);
3531 * Set, replace or remove an extended attribute for this inode.
3532 * value is NULL to remove an existing extended attribute, else either
3533 * create or replace an extended attribute.
3535 int ocfs2_xattr_set(struct inode *inode,
3542 struct buffer_head *di_bh = NULL;
3543 struct ocfs2_dinode *di;
3544 int ret, credits, had_lock, ref_meta = 0, ref_credits = 0;
3545 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3546 struct inode *tl_inode = osb->osb_tl_inode;
3547 struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, NULL, };
3548 struct ocfs2_refcount_tree *ref_tree = NULL;
3549 struct ocfs2_lock_holder oh;
3551 struct ocfs2_xattr_info xi = {
3552 .xi_name_index = name_index,
3554 .xi_name_len = strlen(name),
3556 .xi_value_len = value_len,
3559 struct ocfs2_xattr_search xis = {
3560 .not_found = -ENODATA,
3563 struct ocfs2_xattr_search xbs = {
3564 .not_found = -ENODATA,
3567 if (!ocfs2_supports_xattr(osb))
3571 * Only xbs will be used on indexed trees. xis doesn't need a
3574 xbs.bucket = ocfs2_xattr_bucket_new(inode);
3576 mlog_errno(-ENOMEM);
3580 had_lock = ocfs2_inode_lock_tracker(inode, &di_bh, 1, &oh);
3584 goto cleanup_nolock;
3586 xis.inode_bh = xbs.inode_bh = di_bh;
3587 di = (struct ocfs2_dinode *)di_bh->b_data;
3589 down_write(&OCFS2_I(inode)->ip_xattr_sem);
3591 * Scan inode and external block to find the same name
3592 * extended attribute and collect search information.
3594 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3597 if (xis.not_found) {
3598 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3603 if (xis.not_found && xbs.not_found) {
3605 if (flags & XATTR_REPLACE)
3612 if (flags & XATTR_CREATE)
3616 /* Check whether the value is refcounted and do some preparation. */
3617 if (ocfs2_is_refcount_inode(inode) &&
3618 (!xis.not_found || !xbs.not_found)) {
3619 ret = ocfs2_prepare_refcount_xattr(inode, di, &xi,
3620 &xis, &xbs, &ref_tree,
3621 &ref_meta, &ref_credits);
3628 inode_lock(tl_inode);
3630 if (ocfs2_truncate_log_needs_flush(osb)) {
3631 ret = __ocfs2_flush_truncate_log(osb);
3633 inode_unlock(tl_inode);
3638 inode_unlock(tl_inode);
3640 ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
3641 &xbs, &ctxt, ref_meta, &credits);
3647 /* we need to update inode's ctime field, so add credit for it. */
3648 credits += OCFS2_INODE_UPDATE_CREDITS;
3649 ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
3650 if (IS_ERR(ctxt.handle)) {
3651 ret = PTR_ERR(ctxt.handle);
3656 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3657 ocfs2_update_inode_fsync_trans(ctxt.handle, inode, 0);
3659 ocfs2_commit_trans(osb, ctxt.handle);
3663 ocfs2_free_alloc_context(ctxt.data_ac);
3665 ocfs2_free_alloc_context(ctxt.meta_ac);
3666 if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
3667 ocfs2_schedule_truncate_log_flush(osb, 1);
3668 ocfs2_run_deallocs(osb, &ctxt.dealloc);
3672 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3673 up_write(&OCFS2_I(inode)->ip_xattr_sem);
3674 if (!value && !ret) {
3675 ret = ocfs2_try_remove_refcount_tree(inode, di_bh);
3679 ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
3682 brelse(xbs.xattr_bh);
3683 ocfs2_xattr_bucket_free(xbs.bucket);
3689 * Find the xattr extent rec which may contains name_hash.
3690 * e_cpos will be the first name hash of the xattr rec.
3691 * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
3693 static int ocfs2_xattr_get_rec(struct inode *inode,
3698 struct ocfs2_extent_list *el)
3701 struct buffer_head *eb_bh = NULL;
3702 struct ocfs2_extent_block *eb;
3703 struct ocfs2_extent_rec *rec = NULL;
3706 if (el->l_tree_depth) {
3707 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, name_hash,
3714 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
3717 if (el->l_tree_depth) {
3718 ret = ocfs2_error(inode->i_sb,
3719 "Inode %lu has non zero tree depth in xattr tree block %llu\n",
3721 (unsigned long long)eb_bh->b_blocknr);
3726 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
3727 rec = &el->l_recs[i];
3729 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
3730 e_blkno = le64_to_cpu(rec->e_blkno);
3736 ret = ocfs2_error(inode->i_sb, "Inode %lu has bad extent record (%u, %u, 0) in xattr\n",
3738 le32_to_cpu(rec->e_cpos),
3739 ocfs2_rec_clusters(el, rec));
3743 *p_blkno = le64_to_cpu(rec->e_blkno);
3744 *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
3746 *e_cpos = le32_to_cpu(rec->e_cpos);
3752 typedef int (xattr_bucket_func)(struct inode *inode,
3753 struct ocfs2_xattr_bucket *bucket,
3756 static int ocfs2_find_xe_in_bucket(struct inode *inode,
3757 struct ocfs2_xattr_bucket *bucket,
3764 int i, ret = 0, cmp = 1, block_off, new_offset;
3765 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3766 size_t name_len = strlen(name);
3767 struct ocfs2_xattr_entry *xe = NULL;
3771 * We don't use binary search in the bucket because there
3772 * may be multiple entries with the same name hash.
3774 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3775 xe = &xh->xh_entries[i];
3777 if (name_hash > le32_to_cpu(xe->xe_name_hash))
3779 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
3782 cmp = name_index - ocfs2_xattr_get_type(xe);
3784 cmp = name_len - xe->xe_name_len;
3788 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3799 xe_name = bucket_block(bucket, block_off) + new_offset;
3800 if (!memcmp(name, xe_name, name_len)) {
3812 * Find the specified xattr entry in a series of buckets.
3813 * This series start from p_blkno and last for num_clusters.
3814 * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
3815 * the num of the valid buckets.
3817 * Return the buffer_head this xattr should reside in. And if the xattr's
3818 * hash is in the gap of 2 buckets, return the lower bucket.
3820 static int ocfs2_xattr_bucket_find(struct inode *inode,
3827 struct ocfs2_xattr_search *xs)
3830 struct ocfs2_xattr_header *xh = NULL;
3831 struct ocfs2_xattr_entry *xe = NULL;
3833 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3834 int low_bucket = 0, bucket, high_bucket;
3835 struct ocfs2_xattr_bucket *search;
3836 u64 blkno, lower_blkno = 0;
3838 search = ocfs2_xattr_bucket_new(inode);
3845 ret = ocfs2_read_xattr_bucket(search, p_blkno);
3851 xh = bucket_xh(search);
3852 high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
3853 while (low_bucket <= high_bucket) {
3854 ocfs2_xattr_bucket_relse(search);
3856 bucket = (low_bucket + high_bucket) / 2;
3857 blkno = p_blkno + bucket * blk_per_bucket;
3858 ret = ocfs2_read_xattr_bucket(search, blkno);
3864 xh = bucket_xh(search);
3865 xe = &xh->xh_entries[0];
3866 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3867 high_bucket = bucket - 1;
3872 * Check whether the hash of the last entry in our
3873 * bucket is larger than the search one. for an empty
3874 * bucket, the last one is also the first one.
3877 xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3879 /* record lower_blkno which may be the insert place. */
3880 lower_blkno = blkno;
3882 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3883 low_bucket = bucket + 1;
3887 /* the searched xattr should reside in this bucket if exists. */
3888 ret = ocfs2_find_xe_in_bucket(inode, search,
3889 name_index, name, name_hash,
3899 * Record the bucket we have found.
3900 * When the xattr's hash value is in the gap of 2 buckets, we will
3901 * always set it to the previous bucket.
3904 lower_blkno = p_blkno;
3906 /* This should be in cache - we just read it during the search */
3907 ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3913 xs->header = bucket_xh(xs->bucket);
3914 xs->base = bucket_block(xs->bucket, 0);
3915 xs->end = xs->base + inode->i_sb->s_blocksize;
3918 xs->here = &xs->header->xh_entries[index];
3919 trace_ocfs2_xattr_bucket_find(OCFS2_I(inode)->ip_blkno,
3920 name, name_index, name_hash,
3921 (unsigned long long)bucket_blkno(xs->bucket),
3927 ocfs2_xattr_bucket_free(search);
3931 static int ocfs2_xattr_index_block_find(struct inode *inode,
3932 struct buffer_head *root_bh,
3935 struct ocfs2_xattr_search *xs)
3938 struct ocfs2_xattr_block *xb =
3939 (struct ocfs2_xattr_block *)root_bh->b_data;
3940 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3941 struct ocfs2_extent_list *el = &xb_root->xt_list;
3943 u32 first_hash, num_clusters = 0;
3944 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3946 if (le16_to_cpu(el->l_next_free_rec) == 0)
3949 trace_ocfs2_xattr_index_block_find(OCFS2_I(inode)->ip_blkno,
3950 name, name_index, name_hash,
3951 (unsigned long long)root_bh->b_blocknr,
3954 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3961 BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3963 trace_ocfs2_xattr_index_block_find_rec(OCFS2_I(inode)->ip_blkno,
3964 name, name_index, first_hash,
3965 (unsigned long long)p_blkno,
3968 ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3969 p_blkno, first_hash, num_clusters, xs);
3975 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3978 xattr_bucket_func *func,
3982 u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3983 u32 num_buckets = clusters * bpc;
3984 struct ocfs2_xattr_bucket *bucket;
3986 bucket = ocfs2_xattr_bucket_new(inode);
3988 mlog_errno(-ENOMEM);
3992 trace_ocfs2_iterate_xattr_buckets(
3993 (unsigned long long)OCFS2_I(inode)->ip_blkno,
3994 (unsigned long long)blkno, clusters);
3996 for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3997 ret = ocfs2_read_xattr_bucket(bucket, blkno);
4004 * The real bucket num in this series of blocks is stored
4005 * in the 1st bucket.
4008 num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
4010 trace_ocfs2_iterate_xattr_bucket((unsigned long long)blkno,
4011 le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
4013 ret = func(inode, bucket, para);
4014 if (ret && ret != -ERANGE)
4016 /* Fall through to bucket_relse() */
4019 ocfs2_xattr_bucket_relse(bucket);
4024 ocfs2_xattr_bucket_free(bucket);
4028 struct ocfs2_xattr_tree_list {
4034 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
4035 struct ocfs2_xattr_header *xh,
4042 if (index < 0 || index >= le16_to_cpu(xh->xh_count))
4045 name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
4047 *block_off = name_offset >> sb->s_blocksize_bits;
4048 *new_offset = name_offset % sb->s_blocksize;
4053 static int ocfs2_list_xattr_bucket(struct inode *inode,
4054 struct ocfs2_xattr_bucket *bucket,
4058 struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
4059 int i, block_off, new_offset;
4062 for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
4063 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
4064 type = ocfs2_xattr_get_type(entry);
4066 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
4074 name = (const char *)bucket_block(bucket, block_off) +
4076 ret = ocfs2_xattr_list_entry(inode->i_sb,
4081 entry->xe_name_len);
4089 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
4090 struct buffer_head *blk_bh,
4091 xattr_tree_rec_func *rec_func,
4094 struct ocfs2_xattr_block *xb =
4095 (struct ocfs2_xattr_block *)blk_bh->b_data;
4096 struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4098 u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
4101 if (!el->l_next_free_rec || !rec_func)
4104 while (name_hash > 0) {
4105 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4106 &e_cpos, &num_clusters, el);
4112 ret = rec_func(inode, blk_bh, p_blkno, e_cpos,
4113 num_clusters, para);
4123 name_hash = e_cpos - 1;
4130 static int ocfs2_list_xattr_tree_rec(struct inode *inode,
4131 struct buffer_head *root_bh,
4132 u64 blkno, u32 cpos, u32 len, void *para)
4134 return ocfs2_iterate_xattr_buckets(inode, blkno, len,
4135 ocfs2_list_xattr_bucket, para);
4138 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
4139 struct buffer_head *blk_bh,
4144 struct ocfs2_xattr_tree_list xl = {
4146 .buffer_size = buffer_size,
4150 ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
4151 ocfs2_list_xattr_tree_rec, &xl);
4162 static int cmp_xe(const void *a, const void *b)
4164 const struct ocfs2_xattr_entry *l = a, *r = b;
4165 u32 l_hash = le32_to_cpu(l->xe_name_hash);
4166 u32 r_hash = le32_to_cpu(r->xe_name_hash);
4168 if (l_hash > r_hash)
4170 if (l_hash < r_hash)
4175 static void swap_xe(void *a, void *b, int size)
4177 struct ocfs2_xattr_entry *l = a, *r = b, tmp;
4180 memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
4181 memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
4185 * When the ocfs2_xattr_block is filled up, new bucket will be created
4186 * and all the xattr entries will be moved to the new bucket.
4187 * The header goes at the start of the bucket, and the names+values are
4188 * filled from the end. This is why *target starts as the last buffer.
4189 * Note: we need to sort the entries since they are not saved in order
4190 * in the ocfs2_xattr_block.
4192 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
4193 struct buffer_head *xb_bh,
4194 struct ocfs2_xattr_bucket *bucket)
4196 int i, blocksize = inode->i_sb->s_blocksize;
4197 int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4198 u16 offset, size, off_change;
4199 struct ocfs2_xattr_entry *xe;
4200 struct ocfs2_xattr_block *xb =
4201 (struct ocfs2_xattr_block *)xb_bh->b_data;
4202 struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
4203 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4204 u16 count = le16_to_cpu(xb_xh->xh_count);
4205 char *src = xb_bh->b_data;
4206 char *target = bucket_block(bucket, blks - 1);
4208 trace_ocfs2_cp_xattr_block_to_bucket_begin(
4209 (unsigned long long)xb_bh->b_blocknr,
4210 (unsigned long long)bucket_blkno(bucket));
4212 for (i = 0; i < blks; i++)
4213 memset(bucket_block(bucket, i), 0, blocksize);
4216 * Since the xe_name_offset is based on ocfs2_xattr_header,
4217 * there is a offset change corresponding to the change of
4218 * ocfs2_xattr_header's position.
4220 off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4221 xe = &xb_xh->xh_entries[count - 1];
4222 offset = le16_to_cpu(xe->xe_name_offset) + off_change;
4223 size = blocksize - offset;
4225 /* copy all the names and values. */
4226 memcpy(target + offset, src + offset, size);
4228 /* Init new header now. */
4229 xh->xh_count = xb_xh->xh_count;
4230 xh->xh_num_buckets = cpu_to_le16(1);
4231 xh->xh_name_value_len = cpu_to_le16(size);
4232 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
4234 /* copy all the entries. */
4235 target = bucket_block(bucket, 0);
4236 offset = offsetof(struct ocfs2_xattr_header, xh_entries);
4237 size = count * sizeof(struct ocfs2_xattr_entry);
4238 memcpy(target + offset, (char *)xb_xh + offset, size);
4240 /* Change the xe offset for all the xe because of the move. */
4241 off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
4242 offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4243 for (i = 0; i < count; i++)
4244 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
4246 trace_ocfs2_cp_xattr_block_to_bucket_end(offset, size, off_change);
4248 sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
4253 * After we move xattr from block to index btree, we have to
4254 * update ocfs2_xattr_search to the new xe and base.
4256 * When the entry is in xattr block, xattr_bh indicates the storage place.
4257 * While if the entry is in index b-tree, "bucket" indicates the
4258 * real place of the xattr.
4260 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
4261 struct ocfs2_xattr_search *xs,
4262 struct buffer_head *old_bh)
4264 char *buf = old_bh->b_data;
4265 struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
4266 struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
4269 xs->header = bucket_xh(xs->bucket);
4270 xs->base = bucket_block(xs->bucket, 0);
4271 xs->end = xs->base + inode->i_sb->s_blocksize;
4276 i = xs->here - old_xh->xh_entries;
4277 xs->here = &xs->header->xh_entries[i];
4280 static int ocfs2_xattr_create_index_block(struct inode *inode,
4281 struct ocfs2_xattr_search *xs,
4282 struct ocfs2_xattr_set_ctxt *ctxt)
4287 handle_t *handle = ctxt->handle;
4288 struct ocfs2_inode_info *oi = OCFS2_I(inode);
4289 struct buffer_head *xb_bh = xs->xattr_bh;
4290 struct ocfs2_xattr_block *xb =
4291 (struct ocfs2_xattr_block *)xb_bh->b_data;
4292 struct ocfs2_xattr_tree_root *xr;
4293 u16 xb_flags = le16_to_cpu(xb->xb_flags);
4295 trace_ocfs2_xattr_create_index_block_begin(
4296 (unsigned long long)xb_bh->b_blocknr);
4298 BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
4299 BUG_ON(!xs->bucket);
4303 * We can use this lock for now, and maybe move to a dedicated mutex
4304 * if performance becomes a problem later.
4306 down_write(&oi->ip_alloc_sem);
4308 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), xb_bh,
4309 OCFS2_JOURNAL_ACCESS_WRITE);
4315 ret = __ocfs2_claim_clusters(handle, ctxt->data_ac,
4316 1, 1, &bit_off, &len);
4323 * The bucket may spread in many blocks, and
4324 * we will only touch the 1st block and the last block
4325 * in the whole bucket(one for entry and one for data).
4327 blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
4329 trace_ocfs2_xattr_create_index_block((unsigned long long)blkno);
4331 ret = ocfs2_init_xattr_bucket(xs->bucket, blkno, 1);
4337 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4338 OCFS2_JOURNAL_ACCESS_CREATE);
4344 ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
4345 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4347 ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
4349 /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
4350 memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
4351 offsetof(struct ocfs2_xattr_block, xb_attrs));
4353 xr = &xb->xb_attrs.xb_root;
4354 xr->xt_clusters = cpu_to_le32(1);
4355 xr->xt_last_eb_blk = 0;
4356 xr->xt_list.l_tree_depth = 0;
4357 xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
4358 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
4360 xr->xt_list.l_recs[0].e_cpos = 0;
4361 xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
4362 xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
4364 xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
4366 ocfs2_journal_dirty(handle, xb_bh);
4369 up_write(&oi->ip_alloc_sem);
4374 static int cmp_xe_offset(const void *a, const void *b)
4376 const struct ocfs2_xattr_entry *l = a, *r = b;
4377 u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
4378 u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
4380 if (l_name_offset < r_name_offset)
4382 if (l_name_offset > r_name_offset)
4388 * defrag a xattr bucket if we find that the bucket has some
4389 * holes beteen name/value pairs.
4390 * We will move all the name/value pairs to the end of the bucket
4391 * so that we can spare some space for insertion.
4393 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
4395 struct ocfs2_xattr_bucket *bucket)
4398 size_t end, offset, len;
4399 struct ocfs2_xattr_header *xh;
4400 char *entries, *buf, *bucket_buf = NULL;
4401 u64 blkno = bucket_blkno(bucket);
4403 size_t blocksize = inode->i_sb->s_blocksize;
4404 struct ocfs2_xattr_entry *xe;
4407 * In order to make the operation more efficient and generic,
4408 * we copy all the blocks into a contiguous memory and do the
4409 * defragment there, so if anything is error, we will not touch
4412 bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
4419 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4420 memcpy(buf, bucket_block(bucket, i), blocksize);
4422 ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
4423 OCFS2_JOURNAL_ACCESS_WRITE);
4429 xh = (struct ocfs2_xattr_header *)bucket_buf;
4430 entries = (char *)xh->xh_entries;
4431 xh_free_start = le16_to_cpu(xh->xh_free_start);
4433 trace_ocfs2_defrag_xattr_bucket(
4434 (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
4435 xh_free_start, le16_to_cpu(xh->xh_name_value_len));
4438 * sort all the entries by their offset.
4439 * the largest will be the first, so that we can
4440 * move them to the end one by one.
4442 sort(entries, le16_to_cpu(xh->xh_count),
4443 sizeof(struct ocfs2_xattr_entry),
4444 cmp_xe_offset, swap_xe);
4446 /* Move all name/values to the end of the bucket. */
4447 xe = xh->xh_entries;
4448 end = OCFS2_XATTR_BUCKET_SIZE;
4449 for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
4450 offset = le16_to_cpu(xe->xe_name_offset);
4451 len = namevalue_size_xe(xe);
4454 * We must make sure that the name/value pair
4455 * exist in the same block. So adjust end to
4456 * the previous block end if needed.
4458 if (((end - len) / blocksize !=
4459 (end - 1) / blocksize))
4460 end = end - end % blocksize;
4462 if (end > offset + len) {
4463 memmove(bucket_buf + end - len,
4464 bucket_buf + offset, len);
4465 xe->xe_name_offset = cpu_to_le16(end - len);
4468 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
4469 "bucket %llu\n", (unsigned long long)blkno);
4474 mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
4475 "bucket %llu\n", (unsigned long long)blkno);
4477 if (xh_free_start == end)
4480 memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
4481 xh->xh_free_start = cpu_to_le16(end);
4483 /* sort the entries by their name_hash. */
4484 sort(entries, le16_to_cpu(xh->xh_count),
4485 sizeof(struct ocfs2_xattr_entry),
4489 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4490 memcpy(bucket_block(bucket, i), buf, blocksize);
4491 ocfs2_xattr_bucket_journal_dirty(handle, bucket);
4499 * prev_blkno points to the start of an existing extent. new_blkno
4500 * points to a newly allocated extent. Because we know each of our
4501 * clusters contains more than bucket, we can easily split one cluster
4502 * at a bucket boundary. So we take the last cluster of the existing
4503 * extent and split it down the middle. We move the last half of the
4504 * buckets in the last cluster of the existing extent over to the new
4507 * first_bh is the buffer at prev_blkno so we can update the existing
4508 * extent's bucket count. header_bh is the bucket were we were hoping
4509 * to insert our xattr. If the bucket move places the target in the new
4510 * extent, we'll update first_bh and header_bh after modifying the old
4513 * first_hash will be set as the 1st xe's name_hash in the new extent.
4515 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
4517 struct ocfs2_xattr_bucket *first,
4518 struct ocfs2_xattr_bucket *target,
4524 struct super_block *sb = inode->i_sb;
4525 int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
4526 int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
4527 int to_move = num_buckets / 2;
4529 u64 last_cluster_blkno = bucket_blkno(first) +
4530 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
4532 BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
4533 BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
4535 trace_ocfs2_mv_xattr_bucket_cross_cluster(
4536 (unsigned long long)last_cluster_blkno,
4537 (unsigned long long)new_blkno);
4539 ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
4540 last_cluster_blkno, new_blkno,
4541 to_move, first_hash);
4547 /* This is the first bucket that got moved */
4548 src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
4551 * If the target bucket was part of the moved buckets, we need to
4552 * update first and target.
4554 if (bucket_blkno(target) >= src_blkno) {
4555 /* Find the block for the new target bucket */
4556 src_blkno = new_blkno +
4557 (bucket_blkno(target) - src_blkno);
4559 ocfs2_xattr_bucket_relse(first);
4560 ocfs2_xattr_bucket_relse(target);
4563 * These shouldn't fail - the buffers are in the
4564 * journal from ocfs2_cp_xattr_bucket().
4566 ret = ocfs2_read_xattr_bucket(first, new_blkno);
4571 ret = ocfs2_read_xattr_bucket(target, src_blkno);
4582 * Find the suitable pos when we divide a bucket into 2.
4583 * We have to make sure the xattrs with the same hash value exist
4584 * in the same bucket.
4586 * If this ocfs2_xattr_header covers more than one hash value, find a
4587 * place where the hash value changes. Try to find the most even split.
4588 * The most common case is that all entries have different hash values,
4589 * and the first check we make will find a place to split.
4591 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
4593 struct ocfs2_xattr_entry *entries = xh->xh_entries;
4594 int count = le16_to_cpu(xh->xh_count);
4595 int delta, middle = count / 2;
4598 * We start at the middle. Each step gets farther away in both
4599 * directions. We therefore hit the change in hash value
4600 * nearest to the middle. Note that this loop does not execute for
4603 for (delta = 0; delta < middle; delta++) {
4604 /* Let's check delta earlier than middle */
4605 if (cmp_xe(&entries[middle - delta - 1],
4606 &entries[middle - delta]))
4607 return middle - delta;
4609 /* For even counts, don't walk off the end */
4610 if ((middle + delta + 1) == count)
4613 /* Now try delta past middle */
4614 if (cmp_xe(&entries[middle + delta],
4615 &entries[middle + delta + 1]))
4616 return middle + delta + 1;
4619 /* Every entry had the same hash */
4624 * Move some xattrs in old bucket(blk) to new bucket(new_blk).
4625 * first_hash will record the 1st hash of the new bucket.
4627 * Normally half of the xattrs will be moved. But we have to make
4628 * sure that the xattrs with the same hash value are stored in the
4629 * same bucket. If all the xattrs in this bucket have the same hash
4630 * value, the new bucket will be initialized as an empty one and the
4631 * first_hash will be initialized as (hash_value+1).
4633 static int ocfs2_divide_xattr_bucket(struct inode *inode,
4638 int new_bucket_head)
4641 int count, start, len, name_value_len = 0, name_offset = 0;
4642 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4643 struct ocfs2_xattr_header *xh;
4644 struct ocfs2_xattr_entry *xe;
4645 int blocksize = inode->i_sb->s_blocksize;
4647 trace_ocfs2_divide_xattr_bucket_begin((unsigned long long)blk,
4648 (unsigned long long)new_blk);
4650 s_bucket = ocfs2_xattr_bucket_new(inode);
4651 t_bucket = ocfs2_xattr_bucket_new(inode);
4652 if (!s_bucket || !t_bucket) {
4658 ret = ocfs2_read_xattr_bucket(s_bucket, blk);
4664 ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
4665 OCFS2_JOURNAL_ACCESS_WRITE);
4672 * Even if !new_bucket_head, we're overwriting t_bucket. Thus,
4673 * there's no need to read it.
4675 ret = ocfs2_init_xattr_bucket(t_bucket, new_blk, new_bucket_head);
4682 * Hey, if we're overwriting t_bucket, what difference does
4683 * ACCESS_CREATE vs ACCESS_WRITE make? See the comment in the
4684 * same part of ocfs2_cp_xattr_bucket().
4686 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4688 OCFS2_JOURNAL_ACCESS_CREATE :
4689 OCFS2_JOURNAL_ACCESS_WRITE);
4695 xh = bucket_xh(s_bucket);
4696 count = le16_to_cpu(xh->xh_count);
4697 start = ocfs2_xattr_find_divide_pos(xh);
4699 if (start == count) {
4700 xe = &xh->xh_entries[start-1];
4703 * initialized a new empty bucket here.
4704 * The hash value is set as one larger than
4705 * that of the last entry in the previous bucket.
4707 for (i = 0; i < t_bucket->bu_blocks; i++)
4708 memset(bucket_block(t_bucket, i), 0, blocksize);
4710 xh = bucket_xh(t_bucket);
4711 xh->xh_free_start = cpu_to_le16(blocksize);
4712 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
4713 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
4715 goto set_num_buckets;
4718 /* copy the whole bucket to the new first. */
4719 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4721 /* update the new bucket. */
4722 xh = bucket_xh(t_bucket);
4725 * Calculate the total name/value len and xh_free_start for
4726 * the old bucket first.
4728 name_offset = OCFS2_XATTR_BUCKET_SIZE;
4730 for (i = 0; i < start; i++) {
4731 xe = &xh->xh_entries[i];
4732 name_value_len += namevalue_size_xe(xe);
4733 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
4734 name_offset = le16_to_cpu(xe->xe_name_offset);
4738 * Now begin the modification to the new bucket.
4740 * In the new bucket, We just move the xattr entry to the beginning
4741 * and don't touch the name/value. So there will be some holes in the
4742 * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
4745 xe = &xh->xh_entries[start];
4746 len = sizeof(struct ocfs2_xattr_entry) * (count - start);
4747 trace_ocfs2_divide_xattr_bucket_move(len,
4748 (int)((char *)xe - (char *)xh),
4749 (int)((char *)xh->xh_entries - (char *)xh));
4750 memmove((char *)xh->xh_entries, (char *)xe, len);
4751 xe = &xh->xh_entries[count - start];
4752 len = sizeof(struct ocfs2_xattr_entry) * start;
4753 memset((char *)xe, 0, len);
4755 le16_add_cpu(&xh->xh_count, -start);
4756 le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
4758 /* Calculate xh_free_start for the new bucket. */
4759 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4760 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4761 xe = &xh->xh_entries[i];
4762 if (le16_to_cpu(xe->xe_name_offset) <
4763 le16_to_cpu(xh->xh_free_start))
4764 xh->xh_free_start = xe->xe_name_offset;
4768 /* set xh->xh_num_buckets for the new xh. */
4769 if (new_bucket_head)
4770 xh->xh_num_buckets = cpu_to_le16(1);
4772 xh->xh_num_buckets = 0;
4774 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4776 /* store the first_hash of the new bucket. */
4778 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
4781 * Now only update the 1st block of the old bucket. If we
4782 * just added a new empty bucket, there is no need to modify
4788 xh = bucket_xh(s_bucket);
4789 memset(&xh->xh_entries[start], 0,
4790 sizeof(struct ocfs2_xattr_entry) * (count - start));
4791 xh->xh_count = cpu_to_le16(start);
4792 xh->xh_free_start = cpu_to_le16(name_offset);
4793 xh->xh_name_value_len = cpu_to_le16(name_value_len);
4795 ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
4798 ocfs2_xattr_bucket_free(s_bucket);
4799 ocfs2_xattr_bucket_free(t_bucket);
4805 * Copy xattr from one bucket to another bucket.
4807 * The caller must make sure that the journal transaction
4808 * has enough space for journaling.
4810 static int ocfs2_cp_xattr_bucket(struct inode *inode,
4817 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4819 BUG_ON(s_blkno == t_blkno);
4821 trace_ocfs2_cp_xattr_bucket((unsigned long long)s_blkno,
4822 (unsigned long long)t_blkno,
4825 s_bucket = ocfs2_xattr_bucket_new(inode);
4826 t_bucket = ocfs2_xattr_bucket_new(inode);
4827 if (!s_bucket || !t_bucket) {
4833 ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
4838 * Even if !t_is_new, we're overwriting t_bucket. Thus,
4839 * there's no need to read it.
4841 ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno, t_is_new);
4846 * Hey, if we're overwriting t_bucket, what difference does
4847 * ACCESS_CREATE vs ACCESS_WRITE make? Well, if we allocated a new
4848 * cluster to fill, we came here from
4849 * ocfs2_mv_xattr_buckets(), and it is really new -
4850 * ACCESS_CREATE is required. But we also might have moved data
4851 * out of t_bucket before extending back into it.
4852 * ocfs2_add_new_xattr_bucket() can do this - its call to
4853 * ocfs2_add_new_xattr_cluster() may have created a new extent
4854 * and copied out the end of the old extent. Then it re-extends
4855 * the old extent back to create space for new xattrs. That's
4856 * how we get here, and the bucket isn't really new.
4858 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4860 OCFS2_JOURNAL_ACCESS_CREATE :
4861 OCFS2_JOURNAL_ACCESS_WRITE);
4865 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4866 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4869 ocfs2_xattr_bucket_free(t_bucket);
4870 ocfs2_xattr_bucket_free(s_bucket);
4876 * src_blk points to the start of an existing extent. last_blk points to
4877 * last cluster in that extent. to_blk points to a newly allocated
4878 * extent. We copy the buckets from the cluster at last_blk to the new
4879 * extent. If start_bucket is non-zero, we skip that many buckets before
4880 * we start copying. The new extent's xh_num_buckets gets set to the
4881 * number of buckets we copied. The old extent's xh_num_buckets shrinks
4882 * by the same amount.
4884 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4885 u64 src_blk, u64 last_blk, u64 to_blk,
4886 unsigned int start_bucket,
4889 int i, ret, credits;
4890 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4891 int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4892 int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4893 struct ocfs2_xattr_bucket *old_first, *new_first;
4895 trace_ocfs2_mv_xattr_buckets((unsigned long long)last_blk,
4896 (unsigned long long)to_blk);
4898 BUG_ON(start_bucket >= num_buckets);
4900 num_buckets -= start_bucket;
4901 last_blk += (start_bucket * blks_per_bucket);
4904 /* The first bucket of the original extent */
4905 old_first = ocfs2_xattr_bucket_new(inode);
4906 /* The first bucket of the new extent */
4907 new_first = ocfs2_xattr_bucket_new(inode);
4908 if (!old_first || !new_first) {
4914 ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4921 * We need to update the first bucket of the old extent and all
4922 * the buckets going to the new extent.
4924 credits = ((num_buckets + 1) * blks_per_bucket);
4925 ret = ocfs2_extend_trans(handle, credits);
4931 ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4932 OCFS2_JOURNAL_ACCESS_WRITE);
4938 for (i = 0; i < num_buckets; i++) {
4939 ret = ocfs2_cp_xattr_bucket(inode, handle,
4940 last_blk + (i * blks_per_bucket),
4941 to_blk + (i * blks_per_bucket),
4950 * Get the new bucket ready before we dirty anything
4951 * (This actually shouldn't fail, because we already dirtied
4952 * it once in ocfs2_cp_xattr_bucket()).
4954 ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4959 ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4960 OCFS2_JOURNAL_ACCESS_WRITE);
4966 /* Now update the headers */
4967 le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4968 ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4970 bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4971 ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4974 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4977 ocfs2_xattr_bucket_free(new_first);
4978 ocfs2_xattr_bucket_free(old_first);
4983 * Move some xattrs in this cluster to the new cluster.
4984 * This function should only be called when bucket size == cluster size.
4985 * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4987 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4993 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4994 int ret, credits = 2 * blk_per_bucket;
4996 BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4998 ret = ocfs2_extend_trans(handle, credits);
5004 /* Move half of the xattr in start_blk to the next bucket. */
5005 return ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
5006 new_blk, first_hash, 1);
5010 * Move some xattrs from the old cluster to the new one since they are not
5011 * contiguous in ocfs2 xattr tree.
5013 * new_blk starts a new separate cluster, and we will move some xattrs from
5014 * prev_blk to it. v_start will be set as the first name hash value in this
5015 * new cluster so that it can be used as e_cpos during tree insertion and
5016 * don't collide with our original b-tree operations. first_bh and header_bh
5017 * will also be updated since they will be used in ocfs2_extend_xattr_bucket
5018 * to extend the insert bucket.
5020 * The problem is how much xattr should we move to the new one and when should
5021 * we update first_bh and header_bh?
5022 * 1. If cluster size > bucket size, that means the previous cluster has more
5023 * than 1 bucket, so just move half nums of bucket into the new cluster and
5024 * update the first_bh and header_bh if the insert bucket has been moved
5025 * to the new cluster.
5026 * 2. If cluster_size == bucket_size:
5027 * a) If the previous extent rec has more than one cluster and the insert
5028 * place isn't in the last cluster, copy the entire last cluster to the
5029 * new one. This time, we don't need to upate the first_bh and header_bh
5030 * since they will not be moved into the new cluster.
5031 * b) Otherwise, move the bottom half of the xattrs in the last cluster into
5032 * the new one. And we set the extend flag to zero if the insert place is
5033 * moved into the new allocated cluster since no extend is needed.
5035 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
5037 struct ocfs2_xattr_bucket *first,
5038 struct ocfs2_xattr_bucket *target,
5046 trace_ocfs2_adjust_xattr_cross_cluster(
5047 (unsigned long long)bucket_blkno(first),
5048 (unsigned long long)new_blk, prev_clusters);
5050 if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
5051 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
5060 /* The start of the last cluster in the first extent */
5061 u64 last_blk = bucket_blkno(first) +
5062 ((prev_clusters - 1) *
5063 ocfs2_clusters_to_blocks(inode->i_sb, 1));
5065 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
5066 ret = ocfs2_mv_xattr_buckets(inode, handle,
5067 bucket_blkno(first),
5068 last_blk, new_blk, 0,
5073 ret = ocfs2_divide_xattr_cluster(inode, handle,
5079 if ((bucket_blkno(target) == last_blk) && extend)
5088 * Add a new cluster for xattr storage.
5090 * If the new cluster is contiguous with the previous one, it will be
5091 * appended to the same extent record, and num_clusters will be updated.
5092 * If not, we will insert a new extent for it and move some xattrs in
5093 * the last cluster into the new allocated one.
5094 * We also need to limit the maximum size of a btree leaf, otherwise we'll
5095 * lose the benefits of hashing because we'll have to search large leaves.
5096 * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
5099 * first_bh is the first block of the previous extent rec and header_bh
5100 * indicates the bucket we will insert the new xattrs. They will be updated
5101 * when the header_bh is moved into the new cluster.
5103 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
5104 struct buffer_head *root_bh,
5105 struct ocfs2_xattr_bucket *first,
5106 struct ocfs2_xattr_bucket *target,
5110 struct ocfs2_xattr_set_ctxt *ctxt)
5113 u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
5114 u32 prev_clusters = *num_clusters;
5115 u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
5117 handle_t *handle = ctxt->handle;
5118 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5119 struct ocfs2_extent_tree et;
5121 trace_ocfs2_add_new_xattr_cluster_begin(
5122 (unsigned long long)OCFS2_I(inode)->ip_blkno,
5123 (unsigned long long)bucket_blkno(first),
5124 prev_cpos, prev_clusters);
5126 ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5128 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5129 OCFS2_JOURNAL_ACCESS_WRITE);
5135 ret = __ocfs2_claim_clusters(handle, ctxt->data_ac, 1,
5136 clusters_to_add, &bit_off, &num_bits);
5143 BUG_ON(num_bits > clusters_to_add);
5145 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
5146 trace_ocfs2_add_new_xattr_cluster((unsigned long long)block, num_bits);
5148 if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
5149 (prev_clusters + num_bits) << osb->s_clustersize_bits <=
5150 OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
5152 * If this cluster is contiguous with the old one and
5153 * adding this new cluster, we don't surpass the limit of
5154 * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
5155 * initialized and used like other buckets in the previous
5157 * So add it as a contiguous one. The caller will handle
5160 v_start = prev_cpos + prev_clusters;
5161 *num_clusters = prev_clusters + num_bits;
5163 ret = ocfs2_adjust_xattr_cross_cluster(inode,
5177 trace_ocfs2_add_new_xattr_cluster_insert((unsigned long long)block,
5179 ret = ocfs2_insert_extent(handle, &et, v_start, block,
5180 num_bits, 0, ctxt->meta_ac);
5186 ocfs2_journal_dirty(handle, root_bh);
5193 * We are given an extent. 'first' is the bucket at the very front of
5194 * the extent. The extent has space for an additional bucket past
5195 * bucket_xh(first)->xh_num_buckets. 'target_blkno' is the block number
5196 * of the target bucket. We wish to shift every bucket past the target
5197 * down one, filling in that additional space. When we get back to the
5198 * target, we split the target between itself and the now-empty bucket
5199 * at target+1 (aka, target_blkno + blks_per_bucket).
5201 static int ocfs2_extend_xattr_bucket(struct inode *inode,
5203 struct ocfs2_xattr_bucket *first,
5208 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5209 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5211 u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
5213 trace_ocfs2_extend_xattr_bucket((unsigned long long)target_blk,
5214 (unsigned long long)bucket_blkno(first),
5215 num_clusters, new_bucket);
5217 /* The extent must have room for an additional bucket */
5218 BUG_ON(new_bucket >=
5219 (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
5221 /* end_blk points to the last existing bucket */
5222 end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
5225 * end_blk is the start of the last existing bucket.
5226 * Thus, (end_blk - target_blk) covers the target bucket and
5227 * every bucket after it up to, but not including, the last
5228 * existing bucket. Then we add the last existing bucket, the
5229 * new bucket, and the first bucket (3 * blk_per_bucket).
5231 credits = (end_blk - target_blk) + (3 * blk_per_bucket);
5232 ret = ocfs2_extend_trans(handle, credits);
5238 ret = ocfs2_xattr_bucket_journal_access(handle, first,
5239 OCFS2_JOURNAL_ACCESS_WRITE);
5245 while (end_blk != target_blk) {
5246 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
5247 end_blk + blk_per_bucket, 0);
5250 end_blk -= blk_per_bucket;
5253 /* Move half of the xattr in target_blkno to the next bucket. */
5254 ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
5255 target_blk + blk_per_bucket, NULL, 0);
5257 le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
5258 ocfs2_xattr_bucket_journal_dirty(handle, first);
5265 * Add new xattr bucket in an extent record and adjust the buckets
5266 * accordingly. xb_bh is the ocfs2_xattr_block, and target is the
5267 * bucket we want to insert into.
5269 * In the easy case, we will move all the buckets after target down by
5270 * one. Half of target's xattrs will be moved to the next bucket.
5272 * If current cluster is full, we'll allocate a new one. This may not
5273 * be contiguous. The underlying calls will make sure that there is
5274 * space for the insert, shifting buckets around if necessary.
5275 * 'target' may be moved by those calls.
5277 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
5278 struct buffer_head *xb_bh,
5279 struct ocfs2_xattr_bucket *target,
5280 struct ocfs2_xattr_set_ctxt *ctxt)
5282 struct ocfs2_xattr_block *xb =
5283 (struct ocfs2_xattr_block *)xb_bh->b_data;
5284 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
5285 struct ocfs2_extent_list *el = &xb_root->xt_list;
5287 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
5288 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5289 int ret, num_buckets, extend = 1;
5291 u32 e_cpos, num_clusters;
5292 /* The bucket at the front of the extent */
5293 struct ocfs2_xattr_bucket *first;
5295 trace_ocfs2_add_new_xattr_bucket(
5296 (unsigned long long)bucket_blkno(target));
5298 /* The first bucket of the original extent */
5299 first = ocfs2_xattr_bucket_new(inode);
5306 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
5313 ret = ocfs2_read_xattr_bucket(first, p_blkno);
5319 num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
5320 if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
5322 * This can move first+target if the target bucket moves
5323 * to the new extent.
5325 ret = ocfs2_add_new_xattr_cluster(inode,
5340 ret = ocfs2_extend_xattr_bucket(inode,
5343 bucket_blkno(target),
5350 ocfs2_xattr_bucket_free(first);
5356 * Truncate the specified xe_off entry in xattr bucket.
5357 * bucket is indicated by header_bh and len is the new length.
5358 * Both the ocfs2_xattr_value_root and the entry will be updated here.
5360 * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
5362 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
5363 struct ocfs2_xattr_bucket *bucket,
5366 struct ocfs2_xattr_set_ctxt *ctxt)
5370 struct ocfs2_xattr_entry *xe;
5371 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5372 size_t blocksize = inode->i_sb->s_blocksize;
5373 struct ocfs2_xattr_value_buf vb = {
5374 .vb_access = ocfs2_journal_access,
5377 xe = &xh->xh_entries[xe_off];
5379 BUG_ON(!xe || ocfs2_xattr_is_local(xe));
5381 offset = le16_to_cpu(xe->xe_name_offset) +
5382 OCFS2_XATTR_SIZE(xe->xe_name_len);
5384 value_blk = offset / blocksize;
5386 /* We don't allow ocfs2_xattr_value to be stored in different block. */
5387 BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
5389 vb.vb_bh = bucket->bu_bhs[value_blk];
5392 vb.vb_xv = (struct ocfs2_xattr_value_root *)
5393 (vb.vb_bh->b_data + offset % blocksize);
5396 * From here on out we have to dirty the bucket. The generic
5397 * value calls only modify one of the bucket's bhs, but we need
5398 * to send the bucket at once. So if they error, they *could* have
5399 * modified something. We have to assume they did, and dirty
5400 * the whole bucket. This leaves us in a consistent state.
5402 trace_ocfs2_xattr_bucket_value_truncate(
5403 (unsigned long long)bucket_blkno(bucket), xe_off, len);
5404 ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
5410 ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
5411 OCFS2_JOURNAL_ACCESS_WRITE);
5417 xe->xe_value_size = cpu_to_le64(len);
5419 ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
5425 static int ocfs2_rm_xattr_cluster(struct inode *inode,
5426 struct buffer_head *root_bh,
5433 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5434 struct inode *tl_inode = osb->osb_tl_inode;
5436 struct ocfs2_xattr_block *xb =
5437 (struct ocfs2_xattr_block *)root_bh->b_data;
5438 struct ocfs2_alloc_context *meta_ac = NULL;
5439 struct ocfs2_cached_dealloc_ctxt dealloc;
5440 struct ocfs2_extent_tree et;
5442 ret = ocfs2_iterate_xattr_buckets(inode, blkno, len,
5443 ocfs2_delete_xattr_in_bucket, para);
5449 ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5451 ocfs2_init_dealloc_ctxt(&dealloc);
5453 trace_ocfs2_rm_xattr_cluster(
5454 (unsigned long long)OCFS2_I(inode)->ip_blkno,
5455 (unsigned long long)blkno, cpos, len);
5457 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode), blkno,
5460 ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
5466 inode_lock(tl_inode);
5468 if (ocfs2_truncate_log_needs_flush(osb)) {
5469 ret = __ocfs2_flush_truncate_log(osb);
5476 handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
5477 if (IS_ERR(handle)) {
5483 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5484 OCFS2_JOURNAL_ACCESS_WRITE);
5490 ret = ocfs2_remove_extent(handle, &et, cpos, len, meta_ac,
5497 le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
5498 ocfs2_journal_dirty(handle, root_bh);
5500 ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
5503 ocfs2_update_inode_fsync_trans(handle, inode, 0);
5506 ocfs2_commit_trans(osb, handle);
5508 ocfs2_schedule_truncate_log_flush(osb, 1);
5510 inode_unlock(tl_inode);
5513 ocfs2_free_alloc_context(meta_ac);
5515 ocfs2_run_deallocs(osb, &dealloc);
5521 * check whether the xattr bucket is filled up with the same hash value.
5522 * If we want to insert the xattr with the same hash, return -ENOSPC.
5523 * If we want to insert a xattr with different hash value, go ahead
5524 * and ocfs2_divide_xattr_bucket will handle this.
5526 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
5527 struct ocfs2_xattr_bucket *bucket,
5530 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5531 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5533 if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5536 if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5537 xh->xh_entries[0].xe_name_hash) {
5538 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5540 (unsigned long long)bucket_blkno(bucket),
5541 le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5549 * Try to set the entry in the current bucket. If we fail, the caller
5550 * will handle getting us another bucket.
5552 static int ocfs2_xattr_set_entry_bucket(struct inode *inode,
5553 struct ocfs2_xattr_info *xi,
5554 struct ocfs2_xattr_search *xs,
5555 struct ocfs2_xattr_set_ctxt *ctxt)
5558 struct ocfs2_xa_loc loc;
5560 trace_ocfs2_xattr_set_entry_bucket(xi->xi_name);
5562 ocfs2_init_xattr_bucket_xa_loc(&loc, xs->bucket,
5563 xs->not_found ? NULL : xs->here);
5564 ret = ocfs2_xa_set(&loc, xi, ctxt);
5566 xs->here = loc.xl_entry;
5569 if (ret != -ENOSPC) {
5574 /* Ok, we need space. Let's try defragmenting the bucket. */
5575 ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5582 ret = ocfs2_xa_set(&loc, xi, ctxt);
5584 xs->here = loc.xl_entry;
5595 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5596 struct ocfs2_xattr_info *xi,
5597 struct ocfs2_xattr_search *xs,
5598 struct ocfs2_xattr_set_ctxt *ctxt)
5602 trace_ocfs2_xattr_set_entry_index_block(xi->xi_name);
5604 ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5607 if (ret != -ENOSPC) {
5612 /* Ack, need more space. Let's try to get another bucket! */
5615 * We do not allow for overlapping ranges between buckets. And
5616 * the maximum number of collisions we will allow for then is
5617 * one bucket's worth, so check it here whether we need to
5618 * add a new bucket for the insert.
5620 ret = ocfs2_check_xattr_bucket_collision(inode,
5628 ret = ocfs2_add_new_xattr_bucket(inode,
5638 * ocfs2_add_new_xattr_bucket() will have updated
5639 * xs->bucket if it moved, but it will not have updated
5640 * any of the other search fields. Thus, we drop it and
5641 * re-search. Everything should be cached, so it'll be
5644 ocfs2_xattr_bucket_relse(xs->bucket);
5645 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5648 if (ret && ret != -ENODATA)
5650 xs->not_found = ret;
5652 /* Ok, we have a new bucket, let's try again */
5653 ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5654 if (ret && (ret != -ENOSPC))
5661 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5662 struct ocfs2_xattr_bucket *bucket,
5665 int ret = 0, ref_credits;
5666 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5668 struct ocfs2_xattr_entry *xe;
5669 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5670 struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5671 int credits = ocfs2_remove_extent_credits(osb->sb) +
5672 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5673 struct ocfs2_xattr_value_root *xv;
5674 struct ocfs2_rm_xattr_bucket_para *args =
5675 (struct ocfs2_rm_xattr_bucket_para *)para;
5677 ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5679 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5680 xe = &xh->xh_entries[i];
5681 if (ocfs2_xattr_is_local(xe))
5684 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket,
5691 ret = ocfs2_lock_xattr_remove_allocators(inode, xv,
5697 ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
5698 if (IS_ERR(ctxt.handle)) {
5699 ret = PTR_ERR(ctxt.handle);
5704 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5707 ocfs2_commit_trans(osb, ctxt.handle);
5709 ocfs2_free_alloc_context(ctxt.meta_ac);
5710 ctxt.meta_ac = NULL;
5719 ocfs2_free_alloc_context(ctxt.meta_ac);
5720 ocfs2_schedule_truncate_log_flush(osb, 1);
5721 ocfs2_run_deallocs(osb, &ctxt.dealloc);
5726 * Whenever we modify a xattr value root in the bucket(e.g, CoW
5727 * or change the extent record flag), we need to recalculate
5728 * the metaecc for the whole bucket. So it is done here.
5731 * We have to give the extra credits for the caller.
5733 static int ocfs2_xattr_bucket_post_refcount(struct inode *inode,
5738 struct ocfs2_xattr_bucket *bucket =
5739 (struct ocfs2_xattr_bucket *)para;
5741 ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
5742 OCFS2_JOURNAL_ACCESS_WRITE);
5748 ocfs2_xattr_bucket_journal_dirty(handle, bucket);
5754 * Special action we need if the xattr value is refcounted.
5756 * 1. If the xattr is refcounted, lock the tree.
5757 * 2. CoW the xattr if we are setting the new value and the value
5758 * will be stored outside.
5759 * 3. In other case, decrease_refcount will work for us, so just
5760 * lock the refcount tree, calculate the meta and credits is OK.
5762 * We have to do CoW before ocfs2_init_xattr_set_ctxt since
5763 * currently CoW is a completed transaction, while this function
5764 * will also lock the allocators and let us deadlock. So we will
5765 * CoW the whole xattr value.
5767 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
5768 struct ocfs2_dinode *di,
5769 struct ocfs2_xattr_info *xi,
5770 struct ocfs2_xattr_search *xis,
5771 struct ocfs2_xattr_search *xbs,
5772 struct ocfs2_refcount_tree **ref_tree,
5777 struct ocfs2_xattr_block *xb;
5778 struct ocfs2_xattr_entry *xe;
5780 u32 p_cluster, num_clusters;
5781 unsigned int ext_flags;
5782 int name_offset, name_len;
5783 struct ocfs2_xattr_value_buf vb;
5784 struct ocfs2_xattr_bucket *bucket = NULL;
5785 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5786 struct ocfs2_post_refcount refcount;
5787 struct ocfs2_post_refcount *p = NULL;
5788 struct buffer_head *ref_root_bh = NULL;
5790 if (!xis->not_found) {
5792 name_offset = le16_to_cpu(xe->xe_name_offset);
5793 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5795 vb.vb_bh = xis->inode_bh;
5796 vb.vb_access = ocfs2_journal_access_di;
5798 int i, block_off = 0;
5799 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
5801 name_offset = le16_to_cpu(xe->xe_name_offset);
5802 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5803 i = xbs->here - xbs->header->xh_entries;
5805 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
5806 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
5807 bucket_xh(xbs->bucket),
5814 base = bucket_block(xbs->bucket, block_off);
5815 vb.vb_bh = xbs->bucket->bu_bhs[block_off];
5816 vb.vb_access = ocfs2_journal_access;
5818 if (ocfs2_meta_ecc(osb)) {
5819 /*create parameters for ocfs2_post_refcount. */
5820 bucket = xbs->bucket;
5821 refcount.credits = bucket->bu_blocks;
5822 refcount.para = bucket;
5824 ocfs2_xattr_bucket_post_refcount;
5829 vb.vb_bh = xbs->xattr_bh;
5830 vb.vb_access = ocfs2_journal_access_xb;
5834 if (ocfs2_xattr_is_local(xe))
5837 vb.vb_xv = (struct ocfs2_xattr_value_root *)
5838 (base + name_offset + name_len);
5840 ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
5841 &num_clusters, &vb.vb_xv->xr_list,
5849 * We just need to check the 1st extent record, since we always
5850 * CoW the whole xattr. So there shouldn't be a xattr with
5851 * some REFCOUNT extent recs after the 1st one.
5853 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
5856 ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
5857 1, ref_tree, &ref_root_bh);
5864 * If we are deleting the xattr or the new size will be stored inside,
5865 * cool, leave it there, the xattr truncate process will remove them
5866 * for us(it still needs the refcount tree lock and the meta, credits).
5867 * And the worse case is that every cluster truncate will split the
5868 * refcount tree, and make the original extent become 3. So we will need
5869 * 2 * cluster more extent recs at most.
5871 if (!xi->xi_value || xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE) {
5873 ret = ocfs2_refcounted_xattr_delete_need(inode,
5874 &(*ref_tree)->rf_ci,
5875 ref_root_bh, vb.vb_xv,
5882 ret = ocfs2_refcount_cow_xattr(inode, di, &vb,
5883 *ref_tree, ref_root_bh, 0,
5884 le32_to_cpu(vb.vb_xv->xr_clusters), p);
5889 brelse(ref_root_bh);
5894 * Add the REFCOUNTED flags for all the extent rec in ocfs2_xattr_value_root.
5895 * The physical clusters will be added to refcount tree.
5897 static int ocfs2_xattr_value_attach_refcount(struct inode *inode,
5898 struct ocfs2_xattr_value_root *xv,
5899 struct ocfs2_extent_tree *value_et,
5900 struct ocfs2_caching_info *ref_ci,
5901 struct buffer_head *ref_root_bh,
5902 struct ocfs2_cached_dealloc_ctxt *dealloc,
5903 struct ocfs2_post_refcount *refcount)
5906 u32 clusters = le32_to_cpu(xv->xr_clusters);
5907 u32 cpos, p_cluster, num_clusters;
5908 struct ocfs2_extent_list *el = &xv->xr_list;
5909 unsigned int ext_flags;
5912 while (cpos < clusters) {
5913 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
5914 &num_clusters, el, &ext_flags);
5920 cpos += num_clusters;
5921 if ((ext_flags & OCFS2_EXT_REFCOUNTED))
5926 ret = ocfs2_add_refcount_flag(inode, value_et,
5927 ref_ci, ref_root_bh,
5928 cpos - num_clusters,
5929 p_cluster, num_clusters,
5941 * Given a normal ocfs2_xattr_header, refcount all the entries which
5942 * have value stored outside.
5943 * Used for xattrs stored in inode and ocfs2_xattr_block.
5945 static int ocfs2_xattr_attach_refcount_normal(struct inode *inode,
5946 struct ocfs2_xattr_value_buf *vb,
5947 struct ocfs2_xattr_header *header,
5948 struct ocfs2_caching_info *ref_ci,
5949 struct buffer_head *ref_root_bh,
5950 struct ocfs2_cached_dealloc_ctxt *dealloc)
5953 struct ocfs2_xattr_entry *xe;
5954 struct ocfs2_xattr_value_root *xv;
5955 struct ocfs2_extent_tree et;
5958 for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
5959 xe = &header->xh_entries[i];
5961 if (ocfs2_xattr_is_local(xe))
5964 xv = (struct ocfs2_xattr_value_root *)((void *)header +
5965 le16_to_cpu(xe->xe_name_offset) +
5966 OCFS2_XATTR_SIZE(xe->xe_name_len));
5969 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
5971 ret = ocfs2_xattr_value_attach_refcount(inode, xv, &et,
5972 ref_ci, ref_root_bh,
5983 static int ocfs2_xattr_inline_attach_refcount(struct inode *inode,
5984 struct buffer_head *fe_bh,
5985 struct ocfs2_caching_info *ref_ci,
5986 struct buffer_head *ref_root_bh,
5987 struct ocfs2_cached_dealloc_ctxt *dealloc)
5989 struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
5990 struct ocfs2_xattr_header *header = (struct ocfs2_xattr_header *)
5991 (fe_bh->b_data + inode->i_sb->s_blocksize -
5992 le16_to_cpu(di->i_xattr_inline_size));
5993 struct ocfs2_xattr_value_buf vb = {
5995 .vb_access = ocfs2_journal_access_di,
5998 return ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
5999 ref_ci, ref_root_bh, dealloc);
6002 struct ocfs2_xattr_tree_value_refcount_para {
6003 struct ocfs2_caching_info *ref_ci;
6004 struct buffer_head *ref_root_bh;
6005 struct ocfs2_cached_dealloc_ctxt *dealloc;
6008 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
6009 struct ocfs2_xattr_bucket *bucket,
6011 struct ocfs2_xattr_value_root **xv,
6012 struct buffer_head **bh)
6014 int ret, block_off, name_offset;
6015 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
6016 struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
6019 ret = ocfs2_xattr_bucket_get_name_value(sb,
6029 base = bucket_block(bucket, block_off);
6031 *xv = (struct ocfs2_xattr_value_root *)(base + name_offset +
6032 OCFS2_XATTR_SIZE(xe->xe_name_len));
6035 *bh = bucket->bu_bhs[block_off];
6041 * For a given xattr bucket, refcount all the entries which
6042 * have value stored outside.
6044 static int ocfs2_xattr_bucket_value_refcount(struct inode *inode,
6045 struct ocfs2_xattr_bucket *bucket,
6049 struct ocfs2_extent_tree et;
6050 struct ocfs2_xattr_tree_value_refcount_para *ref =
6051 (struct ocfs2_xattr_tree_value_refcount_para *)para;
6052 struct ocfs2_xattr_header *xh =
6053 (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6054 struct ocfs2_xattr_entry *xe;
6055 struct ocfs2_xattr_value_buf vb = {
6056 .vb_access = ocfs2_journal_access,
6058 struct ocfs2_post_refcount refcount = {
6059 .credits = bucket->bu_blocks,
6061 .func = ocfs2_xattr_bucket_post_refcount,
6063 struct ocfs2_post_refcount *p = NULL;
6065 /* We only need post_refcount if we support metaecc. */
6066 if (ocfs2_meta_ecc(OCFS2_SB(inode->i_sb)))
6069 trace_ocfs2_xattr_bucket_value_refcount(
6070 (unsigned long long)bucket_blkno(bucket),
6071 le16_to_cpu(xh->xh_count));
6072 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6073 xe = &xh->xh_entries[i];
6075 if (ocfs2_xattr_is_local(xe))
6078 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket, i,
6079 &vb.vb_xv, &vb.vb_bh);
6085 ocfs2_init_xattr_value_extent_tree(&et,
6086 INODE_CACHE(inode), &vb);
6088 ret = ocfs2_xattr_value_attach_refcount(inode, vb.vb_xv,
6102 static int ocfs2_refcount_xattr_tree_rec(struct inode *inode,
6103 struct buffer_head *root_bh,
6104 u64 blkno, u32 cpos, u32 len, void *para)
6106 return ocfs2_iterate_xattr_buckets(inode, blkno, len,
6107 ocfs2_xattr_bucket_value_refcount,
6111 static int ocfs2_xattr_block_attach_refcount(struct inode *inode,
6112 struct buffer_head *blk_bh,
6113 struct ocfs2_caching_info *ref_ci,
6114 struct buffer_head *ref_root_bh,
6115 struct ocfs2_cached_dealloc_ctxt *dealloc)
6118 struct ocfs2_xattr_block *xb =
6119 (struct ocfs2_xattr_block *)blk_bh->b_data;
6121 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
6122 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
6123 struct ocfs2_xattr_value_buf vb = {
6125 .vb_access = ocfs2_journal_access_xb,
6128 ret = ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
6129 ref_ci, ref_root_bh,
6132 struct ocfs2_xattr_tree_value_refcount_para para = {
6134 .ref_root_bh = ref_root_bh,
6138 ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
6139 ocfs2_refcount_xattr_tree_rec,
6146 int ocfs2_xattr_attach_refcount_tree(struct inode *inode,
6147 struct buffer_head *fe_bh,
6148 struct ocfs2_caching_info *ref_ci,
6149 struct buffer_head *ref_root_bh,
6150 struct ocfs2_cached_dealloc_ctxt *dealloc)
6153 struct ocfs2_inode_info *oi = OCFS2_I(inode);
6154 struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
6155 struct buffer_head *blk_bh = NULL;
6157 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
6158 ret = ocfs2_xattr_inline_attach_refcount(inode, fe_bh,
6159 ref_ci, ref_root_bh,
6167 if (!di->i_xattr_loc)
6170 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
6177 ret = ocfs2_xattr_block_attach_refcount(inode, blk_bh, ref_ci,
6178 ref_root_bh, dealloc);
6188 typedef int (should_xattr_reflinked)(struct ocfs2_xattr_entry *xe);
6190 * Store the information we need in xattr reflink.
6191 * old_bh and new_bh are inode bh for the old and new inode.
6193 struct ocfs2_xattr_reflink {
6194 struct inode *old_inode;
6195 struct inode *new_inode;
6196 struct buffer_head *old_bh;
6197 struct buffer_head *new_bh;
6198 struct ocfs2_caching_info *ref_ci;
6199 struct buffer_head *ref_root_bh;
6200 struct ocfs2_cached_dealloc_ctxt *dealloc;
6201 should_xattr_reflinked *xattr_reflinked;
6205 * Given a xattr header and xe offset,
6206 * return the proper xv and the corresponding bh.
6207 * xattr in inode, block and xattr tree have different implementaions.
6209 typedef int (get_xattr_value_root)(struct super_block *sb,
6210 struct buffer_head *bh,
6211 struct ocfs2_xattr_header *xh,
6213 struct ocfs2_xattr_value_root **xv,
6214 struct buffer_head **ret_bh,
6218 * Calculate all the xattr value root metadata stored in this xattr header and
6219 * credits we need if we create them from the scratch.
6220 * We use get_xattr_value_root so that all types of xattr container can use it.
6222 static int ocfs2_value_metas_in_xattr_header(struct super_block *sb,
6223 struct buffer_head *bh,
6224 struct ocfs2_xattr_header *xh,
6225 int *metas, int *credits,
6227 get_xattr_value_root *func,
6231 struct ocfs2_xattr_value_root *xv;
6232 struct ocfs2_xattr_entry *xe;
6234 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6235 xe = &xh->xh_entries[i];
6236 if (ocfs2_xattr_is_local(xe))
6239 ret = func(sb, bh, xh, i, &xv, NULL, para);
6245 *metas += le16_to_cpu(xv->xr_list.l_tree_depth) *
6246 le16_to_cpu(xv->xr_list.l_next_free_rec);
6248 *credits += ocfs2_calc_extend_credits(sb,
6249 &def_xv.xv.xr_list);
6252 * If the value is a tree with depth > 1, We don't go deep
6253 * to the extent block, so just calculate a maximum record num.
6255 if (!xv->xr_list.l_tree_depth)
6256 *num_recs += le16_to_cpu(xv->xr_list.l_next_free_rec);
6258 *num_recs += ocfs2_clusters_for_bytes(sb,
6265 /* Used by xattr inode and block to return the right xv and buffer_head. */
6266 static int ocfs2_get_xattr_value_root(struct super_block *sb,
6267 struct buffer_head *bh,
6268 struct ocfs2_xattr_header *xh,
6270 struct ocfs2_xattr_value_root **xv,
6271 struct buffer_head **ret_bh,
6274 struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
6276 *xv = (struct ocfs2_xattr_value_root *)((void *)xh +
6277 le16_to_cpu(xe->xe_name_offset) +
6278 OCFS2_XATTR_SIZE(xe->xe_name_len));
6287 * Lock the meta_ac and caculate how much credits we need for reflink xattrs.
6288 * It is only used for inline xattr and xattr block.
6290 static int ocfs2_reflink_lock_xattr_allocators(struct ocfs2_super *osb,
6291 struct ocfs2_xattr_header *xh,
6292 struct buffer_head *ref_root_bh,
6294 struct ocfs2_alloc_context **meta_ac)
6296 int ret, meta_add = 0, num_recs = 0;
6297 struct ocfs2_refcount_block *rb =
6298 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
6302 ret = ocfs2_value_metas_in_xattr_header(osb->sb, NULL, xh,
6303 &meta_add, credits, &num_recs,
6304 ocfs2_get_xattr_value_root,
6312 * We need to add/modify num_recs in refcount tree, so just calculate
6313 * an approximate number we need for refcount tree change.
6314 * Sometimes we need to split the tree, and after split, half recs
6315 * will be moved to the new block, and a new block can only provide
6316 * half number of recs. So we multiple new blocks by 2.
6318 num_recs = num_recs / ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6319 meta_add += num_recs;
6320 *credits += num_recs + num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6321 if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6322 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6323 le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6327 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add, meta_ac);
6336 * Given a xattr header, reflink all the xattrs in this container.
6337 * It can be used for inode, block and bucket.
6340 * Before we call this function, the caller has memcpy the xattr in
6341 * old_xh to the new_xh.
6343 * If args.xattr_reflinked is set, call it to decide whether the xe should
6344 * be reflinked or not. If not, remove it from the new xattr header.
6346 static int ocfs2_reflink_xattr_header(handle_t *handle,
6347 struct ocfs2_xattr_reflink *args,
6348 struct buffer_head *old_bh,
6349 struct ocfs2_xattr_header *xh,
6350 struct buffer_head *new_bh,
6351 struct ocfs2_xattr_header *new_xh,
6352 struct ocfs2_xattr_value_buf *vb,
6353 struct ocfs2_alloc_context *meta_ac,
6354 get_xattr_value_root *func,
6358 struct super_block *sb = args->old_inode->i_sb;
6359 struct buffer_head *value_bh;
6360 struct ocfs2_xattr_entry *xe, *last;
6361 struct ocfs2_xattr_value_root *xv, *new_xv;
6362 struct ocfs2_extent_tree data_et;
6363 u32 clusters, cpos, p_cluster, num_clusters;
6364 unsigned int ext_flags = 0;
6366 trace_ocfs2_reflink_xattr_header((unsigned long long)old_bh->b_blocknr,
6367 le16_to_cpu(xh->xh_count));
6369 last = &new_xh->xh_entries[le16_to_cpu(new_xh->xh_count)];
6370 for (i = 0, j = 0; i < le16_to_cpu(xh->xh_count); i++, j++) {
6371 xe = &xh->xh_entries[i];
6373 if (args->xattr_reflinked && !args->xattr_reflinked(xe)) {
6374 xe = &new_xh->xh_entries[j];
6376 le16_add_cpu(&new_xh->xh_count, -1);
6377 if (new_xh->xh_count) {
6379 (void *)last - (void *)xe);
6381 sizeof(struct ocfs2_xattr_entry));
6385 * We don't want j to increase in the next round since
6386 * it is already moved ahead.
6392 if (ocfs2_xattr_is_local(xe))
6395 ret = func(sb, old_bh, xh, i, &xv, NULL, para);
6401 ret = func(sb, new_bh, new_xh, j, &new_xv, &value_bh, para);
6408 * For the xattr which has l_tree_depth = 0, all the extent
6409 * recs have already be copied to the new xh with the
6410 * propriate OCFS2_EXT_REFCOUNTED flag we just need to
6411 * increase the refount count int the refcount tree.
6413 * For the xattr which has l_tree_depth > 0, we need
6414 * to initialize it to the empty default value root,
6415 * and then insert the extents one by one.
6417 if (xv->xr_list.l_tree_depth) {
6418 memcpy(new_xv, &def_xv, OCFS2_XATTR_ROOT_SIZE);
6420 vb->vb_bh = value_bh;
6421 ocfs2_init_xattr_value_extent_tree(&data_et,
6422 INODE_CACHE(args->new_inode), vb);
6425 clusters = le32_to_cpu(xv->xr_clusters);
6427 while (cpos < clusters) {
6428 ret = ocfs2_xattr_get_clusters(args->old_inode,
6441 if (xv->xr_list.l_tree_depth) {
6442 ret = ocfs2_insert_extent(handle,
6444 ocfs2_clusters_to_blocks(
6445 args->old_inode->i_sb,
6447 num_clusters, ext_flags,
6455 ret = ocfs2_increase_refcount(handle, args->ref_ci,
6457 p_cluster, num_clusters,
6458 meta_ac, args->dealloc);
6464 cpos += num_clusters;
6472 static int ocfs2_reflink_xattr_inline(struct ocfs2_xattr_reflink *args)
6474 int ret = 0, credits = 0;
6476 struct ocfs2_super *osb = OCFS2_SB(args->old_inode->i_sb);
6477 struct ocfs2_dinode *di = (struct ocfs2_dinode *)args->old_bh->b_data;
6478 int inline_size = le16_to_cpu(di->i_xattr_inline_size);
6479 int header_off = osb->sb->s_blocksize - inline_size;
6480 struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)
6481 (args->old_bh->b_data + header_off);
6482 struct ocfs2_xattr_header *new_xh = (struct ocfs2_xattr_header *)
6483 (args->new_bh->b_data + header_off);
6484 struct ocfs2_alloc_context *meta_ac = NULL;
6485 struct ocfs2_inode_info *new_oi;
6486 struct ocfs2_dinode *new_di;
6487 struct ocfs2_xattr_value_buf vb = {
6488 .vb_bh = args->new_bh,
6489 .vb_access = ocfs2_journal_access_di,
6492 ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6493 &credits, &meta_ac);
6499 handle = ocfs2_start_trans(osb, credits);
6500 if (IS_ERR(handle)) {
6501 ret = PTR_ERR(handle);
6506 ret = ocfs2_journal_access_di(handle, INODE_CACHE(args->new_inode),
6507 args->new_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6513 memcpy(args->new_bh->b_data + header_off,
6514 args->old_bh->b_data + header_off, inline_size);
6516 new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6517 new_di->i_xattr_inline_size = cpu_to_le16(inline_size);
6519 ret = ocfs2_reflink_xattr_header(handle, args, args->old_bh, xh,
6520 args->new_bh, new_xh, &vb, meta_ac,
6521 ocfs2_get_xattr_value_root, NULL);
6527 new_oi = OCFS2_I(args->new_inode);
6529 * Adjust extent record count to reserve space for extended attribute.
6530 * Inline data count had been adjusted in ocfs2_duplicate_inline_data().
6532 if (!(new_oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) &&
6533 !(ocfs2_inode_is_fast_symlink(args->new_inode))) {
6534 struct ocfs2_extent_list *el = &new_di->id2.i_list;
6535 le16_add_cpu(&el->l_count, -(inline_size /
6536 sizeof(struct ocfs2_extent_rec)));
6538 spin_lock(&new_oi->ip_lock);
6539 new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL | OCFS2_INLINE_XATTR_FL;
6540 new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6541 spin_unlock(&new_oi->ip_lock);
6543 ocfs2_journal_dirty(handle, args->new_bh);
6546 ocfs2_commit_trans(osb, handle);
6550 ocfs2_free_alloc_context(meta_ac);
6554 static int ocfs2_create_empty_xattr_block(struct inode *inode,
6555 struct buffer_head *fe_bh,
6556 struct buffer_head **ret_bh,
6560 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6561 struct ocfs2_xattr_set_ctxt ctxt;
6563 memset(&ctxt, 0, sizeof(ctxt));
6564 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &ctxt.meta_ac);
6570 ctxt.handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_CREATE_CREDITS);
6571 if (IS_ERR(ctxt.handle)) {
6572 ret = PTR_ERR(ctxt.handle);
6577 trace_ocfs2_create_empty_xattr_block(
6578 (unsigned long long)fe_bh->b_blocknr, indexed);
6579 ret = ocfs2_create_xattr_block(inode, fe_bh, &ctxt, indexed,
6584 ocfs2_commit_trans(osb, ctxt.handle);
6586 ocfs2_free_alloc_context(ctxt.meta_ac);
6590 static int ocfs2_reflink_xattr_block(struct ocfs2_xattr_reflink *args,
6591 struct buffer_head *blk_bh,
6592 struct buffer_head *new_blk_bh)
6594 int ret = 0, credits = 0;
6596 struct ocfs2_inode_info *new_oi = OCFS2_I(args->new_inode);
6597 struct ocfs2_dinode *new_di;
6598 struct ocfs2_super *osb = OCFS2_SB(args->new_inode->i_sb);
6599 int header_off = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
6600 struct ocfs2_xattr_block *xb =
6601 (struct ocfs2_xattr_block *)blk_bh->b_data;
6602 struct ocfs2_xattr_header *xh = &xb->xb_attrs.xb_header;
6603 struct ocfs2_xattr_block *new_xb =
6604 (struct ocfs2_xattr_block *)new_blk_bh->b_data;
6605 struct ocfs2_xattr_header *new_xh = &new_xb->xb_attrs.xb_header;
6606 struct ocfs2_alloc_context *meta_ac;
6607 struct ocfs2_xattr_value_buf vb = {
6608 .vb_bh = new_blk_bh,
6609 .vb_access = ocfs2_journal_access_xb,
6612 ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6613 &credits, &meta_ac);
6619 /* One more credits in case we need to add xattr flags in new inode. */
6620 handle = ocfs2_start_trans(osb, credits + 1);
6621 if (IS_ERR(handle)) {
6622 ret = PTR_ERR(handle);
6627 if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6628 ret = ocfs2_journal_access_di(handle,
6629 INODE_CACHE(args->new_inode),
6631 OCFS2_JOURNAL_ACCESS_WRITE);
6638 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(args->new_inode),
6639 new_blk_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6645 memcpy(new_blk_bh->b_data + header_off, blk_bh->b_data + header_off,
6646 osb->sb->s_blocksize - header_off);
6648 ret = ocfs2_reflink_xattr_header(handle, args, blk_bh, xh,
6649 new_blk_bh, new_xh, &vb, meta_ac,
6650 ocfs2_get_xattr_value_root, NULL);
6656 ocfs2_journal_dirty(handle, new_blk_bh);
6658 if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6659 new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6660 spin_lock(&new_oi->ip_lock);
6661 new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
6662 new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6663 spin_unlock(&new_oi->ip_lock);
6665 ocfs2_journal_dirty(handle, args->new_bh);
6669 ocfs2_commit_trans(osb, handle);
6672 ocfs2_free_alloc_context(meta_ac);
6676 struct ocfs2_reflink_xattr_tree_args {
6677 struct ocfs2_xattr_reflink *reflink;
6678 struct buffer_head *old_blk_bh;
6679 struct buffer_head *new_blk_bh;
6680 struct ocfs2_xattr_bucket *old_bucket;
6681 struct ocfs2_xattr_bucket *new_bucket;
6686 * We have to handle the case that both old bucket and new bucket
6687 * will call this function to get the right ret_bh.
6688 * So The caller must give us the right bh.
6690 static int ocfs2_get_reflink_xattr_value_root(struct super_block *sb,
6691 struct buffer_head *bh,
6692 struct ocfs2_xattr_header *xh,
6694 struct ocfs2_xattr_value_root **xv,
6695 struct buffer_head **ret_bh,
6698 struct ocfs2_reflink_xattr_tree_args *args =
6699 (struct ocfs2_reflink_xattr_tree_args *)para;
6700 struct ocfs2_xattr_bucket *bucket;
6702 if (bh == args->old_bucket->bu_bhs[0])
6703 bucket = args->old_bucket;
6705 bucket = args->new_bucket;
6707 return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6711 struct ocfs2_value_tree_metas {
6717 static int ocfs2_value_tree_metas_in_bucket(struct super_block *sb,
6718 struct buffer_head *bh,
6719 struct ocfs2_xattr_header *xh,
6721 struct ocfs2_xattr_value_root **xv,
6722 struct buffer_head **ret_bh,
6725 struct ocfs2_xattr_bucket *bucket =
6726 (struct ocfs2_xattr_bucket *)para;
6728 return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6732 static int ocfs2_calc_value_tree_metas(struct inode *inode,
6733 struct ocfs2_xattr_bucket *bucket,
6736 struct ocfs2_value_tree_metas *metas =
6737 (struct ocfs2_value_tree_metas *)para;
6738 struct ocfs2_xattr_header *xh =
6739 (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6741 /* Add the credits for this bucket first. */
6742 metas->credits += bucket->bu_blocks;
6743 return ocfs2_value_metas_in_xattr_header(inode->i_sb, bucket->bu_bhs[0],
6744 xh, &metas->num_metas,
6745 &metas->credits, &metas->num_recs,
6746 ocfs2_value_tree_metas_in_bucket,
6751 * Given a xattr extent rec starting from blkno and having len clusters,
6752 * iterate all the buckets calculate how much metadata we need for reflinking
6753 * all the ocfs2_xattr_value_root and lock the allocators accordingly.
6755 static int ocfs2_lock_reflink_xattr_rec_allocators(
6756 struct ocfs2_reflink_xattr_tree_args *args,
6757 struct ocfs2_extent_tree *xt_et,
6758 u64 blkno, u32 len, int *credits,
6759 struct ocfs2_alloc_context **meta_ac,
6760 struct ocfs2_alloc_context **data_ac)
6762 int ret, num_free_extents;
6763 struct ocfs2_value_tree_metas metas;
6764 struct ocfs2_super *osb = OCFS2_SB(args->reflink->old_inode->i_sb);
6765 struct ocfs2_refcount_block *rb;
6767 memset(&metas, 0, sizeof(metas));
6769 ret = ocfs2_iterate_xattr_buckets(args->reflink->old_inode, blkno, len,
6770 ocfs2_calc_value_tree_metas, &metas);
6776 *credits = metas.credits;
6779 * Calculate we need for refcount tree change.
6781 * We need to add/modify num_recs in refcount tree, so just calculate
6782 * an approximate number we need for refcount tree change.
6783 * Sometimes we need to split the tree, and after split, half recs
6784 * will be moved to the new block, and a new block can only provide
6785 * half number of recs. So we multiple new blocks by 2.
6786 * In the end, we have to add credits for modifying the already
6787 * existed refcount block.
6789 rb = (struct ocfs2_refcount_block *)args->reflink->ref_root_bh->b_data;
6791 (metas.num_recs + ocfs2_refcount_recs_per_rb(osb->sb) - 1) /
6792 ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6793 metas.num_metas += metas.num_recs;
6794 *credits += metas.num_recs +
6795 metas.num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6796 if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6797 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6798 le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6802 /* count in the xattr tree change. */
6803 num_free_extents = ocfs2_num_free_extents(xt_et);
6804 if (num_free_extents < 0) {
6805 ret = num_free_extents;
6810 if (num_free_extents < len)
6811 metas.num_metas += ocfs2_extend_meta_needed(xt_et->et_root_el);
6813 *credits += ocfs2_calc_extend_credits(osb->sb,
6816 if (metas.num_metas) {
6817 ret = ocfs2_reserve_new_metadata_blocks(osb, metas.num_metas,
6826 ret = ocfs2_reserve_clusters(osb, len, data_ac);
6833 ocfs2_free_alloc_context(*meta_ac);
6841 static int ocfs2_reflink_xattr_bucket(handle_t *handle,
6842 u64 blkno, u64 new_blkno, u32 clusters,
6843 u32 *cpos, int num_buckets,
6844 struct ocfs2_alloc_context *meta_ac,
6845 struct ocfs2_alloc_context *data_ac,
6846 struct ocfs2_reflink_xattr_tree_args *args)
6849 struct super_block *sb = args->reflink->old_inode->i_sb;
6850 int bpb = args->old_bucket->bu_blocks;
6851 struct ocfs2_xattr_value_buf vb = {
6852 .vb_access = ocfs2_journal_access,
6855 for (i = 0; i < num_buckets; i++, blkno += bpb, new_blkno += bpb) {
6856 ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6862 ret = ocfs2_init_xattr_bucket(args->new_bucket, new_blkno, 1);
6868 ret = ocfs2_xattr_bucket_journal_access(handle,
6870 OCFS2_JOURNAL_ACCESS_CREATE);
6876 for (j = 0; j < bpb; j++)
6877 memcpy(bucket_block(args->new_bucket, j),
6878 bucket_block(args->old_bucket, j),
6882 * Record the start cpos so that we can use it to initialize
6883 * our xattr tree we also set the xh_num_bucket for the new
6887 *cpos = le32_to_cpu(bucket_xh(args->new_bucket)->
6888 xh_entries[0].xe_name_hash);
6889 bucket_xh(args->new_bucket)->xh_num_buckets =
6890 cpu_to_le16(num_buckets);
6893 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6895 ret = ocfs2_reflink_xattr_header(handle, args->reflink,
6896 args->old_bucket->bu_bhs[0],
6897 bucket_xh(args->old_bucket),
6898 args->new_bucket->bu_bhs[0],
6899 bucket_xh(args->new_bucket),
6901 ocfs2_get_reflink_xattr_value_root,
6909 * Re-access and dirty the bucket to calculate metaecc.
6910 * Because we may extend the transaction in reflink_xattr_header
6911 * which will let the already accessed block gone.
6913 ret = ocfs2_xattr_bucket_journal_access(handle,
6915 OCFS2_JOURNAL_ACCESS_WRITE);
6921 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6923 ocfs2_xattr_bucket_relse(args->old_bucket);
6924 ocfs2_xattr_bucket_relse(args->new_bucket);
6927 ocfs2_xattr_bucket_relse(args->old_bucket);
6928 ocfs2_xattr_bucket_relse(args->new_bucket);
6932 static int ocfs2_reflink_xattr_buckets(handle_t *handle,
6933 struct inode *inode,
6934 struct ocfs2_reflink_xattr_tree_args *args,
6935 struct ocfs2_extent_tree *et,
6936 struct ocfs2_alloc_context *meta_ac,
6937 struct ocfs2_alloc_context *data_ac,
6938 u64 blkno, u32 cpos, u32 len)
6940 int ret, first_inserted = 0;
6941 u32 p_cluster, num_clusters, reflink_cpos = 0;
6943 unsigned int num_buckets, reflink_buckets;
6945 ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
6947 ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6952 num_buckets = le16_to_cpu(bucket_xh(args->old_bucket)->xh_num_buckets);
6953 ocfs2_xattr_bucket_relse(args->old_bucket);
6955 while (len && num_buckets) {
6956 ret = ocfs2_claim_clusters(handle, data_ac,
6957 1, &p_cluster, &num_clusters);
6963 new_blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
6964 reflink_buckets = min(num_buckets, bpc * num_clusters);
6966 ret = ocfs2_reflink_xattr_bucket(handle, blkno,
6967 new_blkno, num_clusters,
6968 &reflink_cpos, reflink_buckets,
6969 meta_ac, data_ac, args);
6976 * For the 1st allocated cluster, we make it use the same cpos
6977 * so that the xattr tree looks the same as the original one
6980 if (!first_inserted) {
6981 reflink_cpos = cpos;
6984 ret = ocfs2_insert_extent(handle, et, reflink_cpos, new_blkno,
6985 num_clusters, 0, meta_ac);
6989 trace_ocfs2_reflink_xattr_buckets((unsigned long long)new_blkno,
6990 num_clusters, reflink_cpos);
6992 len -= num_clusters;
6993 blkno += ocfs2_clusters_to_blocks(inode->i_sb, num_clusters);
6994 num_buckets -= reflink_buckets;
7001 * Create the same xattr extent record in the new inode's xattr tree.
7003 static int ocfs2_reflink_xattr_rec(struct inode *inode,
7004 struct buffer_head *root_bh,
7010 int ret, credits = 0;
7012 struct ocfs2_reflink_xattr_tree_args *args =
7013 (struct ocfs2_reflink_xattr_tree_args *)para;
7014 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7015 struct ocfs2_alloc_context *meta_ac = NULL;
7016 struct ocfs2_alloc_context *data_ac = NULL;
7017 struct ocfs2_extent_tree et;
7019 trace_ocfs2_reflink_xattr_rec((unsigned long long)blkno, len);
7021 ocfs2_init_xattr_tree_extent_tree(&et,
7022 INODE_CACHE(args->reflink->new_inode),
7025 ret = ocfs2_lock_reflink_xattr_rec_allocators(args, &et, blkno,
7027 &meta_ac, &data_ac);
7033 handle = ocfs2_start_trans(osb, credits);
7034 if (IS_ERR(handle)) {
7035 ret = PTR_ERR(handle);
7040 ret = ocfs2_reflink_xattr_buckets(handle, inode, args, &et,
7046 ocfs2_commit_trans(osb, handle);
7050 ocfs2_free_alloc_context(meta_ac);
7052 ocfs2_free_alloc_context(data_ac);
7057 * Create reflinked xattr buckets.
7058 * We will add bucket one by one, and refcount all the xattrs in the bucket
7059 * if they are stored outside.
7061 static int ocfs2_reflink_xattr_tree(struct ocfs2_xattr_reflink *args,
7062 struct buffer_head *blk_bh,
7063 struct buffer_head *new_blk_bh)
7066 struct ocfs2_reflink_xattr_tree_args para;
7068 memset(¶, 0, sizeof(para));
7069 para.reflink = args;
7070 para.old_blk_bh = blk_bh;
7071 para.new_blk_bh = new_blk_bh;
7073 para.old_bucket = ocfs2_xattr_bucket_new(args->old_inode);
7074 if (!para.old_bucket) {
7075 mlog_errno(-ENOMEM);
7079 para.new_bucket = ocfs2_xattr_bucket_new(args->new_inode);
7080 if (!para.new_bucket) {
7086 ret = ocfs2_iterate_xattr_index_block(args->old_inode, blk_bh,
7087 ocfs2_reflink_xattr_rec,
7093 ocfs2_xattr_bucket_free(para.old_bucket);
7094 ocfs2_xattr_bucket_free(para.new_bucket);
7098 static int ocfs2_reflink_xattr_in_block(struct ocfs2_xattr_reflink *args,
7099 struct buffer_head *blk_bh)
7101 int ret, indexed = 0;
7102 struct buffer_head *new_blk_bh = NULL;
7103 struct ocfs2_xattr_block *xb =
7104 (struct ocfs2_xattr_block *)blk_bh->b_data;
7107 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)
7110 ret = ocfs2_create_empty_xattr_block(args->new_inode, args->new_bh,
7111 &new_blk_bh, indexed);
7118 ret = ocfs2_reflink_xattr_block(args, blk_bh, new_blk_bh);
7120 ret = ocfs2_reflink_xattr_tree(args, blk_bh, new_blk_bh);
7129 static int ocfs2_reflink_xattr_no_security(struct ocfs2_xattr_entry *xe)
7131 int type = ocfs2_xattr_get_type(xe);
7133 return type != OCFS2_XATTR_INDEX_SECURITY &&
7134 type != OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS &&
7135 type != OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT;
7138 int ocfs2_reflink_xattrs(struct inode *old_inode,
7139 struct buffer_head *old_bh,
7140 struct inode *new_inode,
7141 struct buffer_head *new_bh,
7142 bool preserve_security)
7145 struct ocfs2_xattr_reflink args;
7146 struct ocfs2_inode_info *oi = OCFS2_I(old_inode);
7147 struct ocfs2_dinode *di = (struct ocfs2_dinode *)old_bh->b_data;
7148 struct buffer_head *blk_bh = NULL;
7149 struct ocfs2_cached_dealloc_ctxt dealloc;
7150 struct ocfs2_refcount_tree *ref_tree;
7151 struct buffer_head *ref_root_bh = NULL;
7153 ret = ocfs2_lock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7154 le64_to_cpu(di->i_refcount_loc),
7155 1, &ref_tree, &ref_root_bh);
7161 ocfs2_init_dealloc_ctxt(&dealloc);
7163 args.old_inode = old_inode;
7164 args.new_inode = new_inode;
7165 args.old_bh = old_bh;
7166 args.new_bh = new_bh;
7167 args.ref_ci = &ref_tree->rf_ci;
7168 args.ref_root_bh = ref_root_bh;
7169 args.dealloc = &dealloc;
7170 if (preserve_security)
7171 args.xattr_reflinked = NULL;
7173 args.xattr_reflinked = ocfs2_reflink_xattr_no_security;
7175 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
7176 ret = ocfs2_reflink_xattr_inline(&args);
7183 if (!di->i_xattr_loc)
7186 ret = ocfs2_read_xattr_block(old_inode, le64_to_cpu(di->i_xattr_loc),
7193 ret = ocfs2_reflink_xattr_in_block(&args, blk_bh);
7200 ocfs2_unlock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7202 brelse(ref_root_bh);
7204 if (ocfs2_dealloc_has_cluster(&dealloc)) {
7205 ocfs2_schedule_truncate_log_flush(OCFS2_SB(old_inode->i_sb), 1);
7206 ocfs2_run_deallocs(OCFS2_SB(old_inode->i_sb), &dealloc);
7214 * Initialize security and acl for a already created inode.
7215 * Used for reflink a non-preserve-security file.
7217 * It uses common api like ocfs2_xattr_set, so the caller
7218 * must not hold any lock expect i_mutex.
7220 int ocfs2_init_security_and_acl(struct inode *dir,
7221 struct inode *inode,
7222 const struct qstr *qstr)
7225 struct buffer_head *dir_bh = NULL;
7227 ret = ocfs2_init_security_get(inode, dir, qstr, NULL);
7233 ret = ocfs2_inode_lock(dir, &dir_bh, 0);
7238 ret = ocfs2_init_acl(NULL, inode, dir, NULL, dir_bh, NULL, NULL);
7242 ocfs2_inode_unlock(dir, 0);
7249 * 'security' attributes support
7251 static int ocfs2_xattr_security_get(const struct xattr_handler *handler,
7252 struct dentry *unused, struct inode *inode,
7253 const char *name, void *buffer, size_t size)
7255 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_SECURITY,
7256 name, buffer, size);
7259 static int ocfs2_xattr_security_set(const struct xattr_handler *handler,
7260 struct dentry *unused, struct inode *inode,
7261 const char *name, const void *value,
7262 size_t size, int flags)
7264 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY,
7265 name, value, size, flags);
7268 static int ocfs2_initxattrs(struct inode *inode, const struct xattr *xattr_array,
7271 const struct xattr *xattr;
7274 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
7275 err = ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY,
7276 xattr->name, xattr->value,
7277 xattr->value_len, XATTR_CREATE);
7284 int ocfs2_init_security_get(struct inode *inode,
7286 const struct qstr *qstr,
7287 struct ocfs2_security_xattr_info *si)
7289 /* check whether ocfs2 support feature xattr */
7290 if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
7293 return security_old_inode_init_security(inode, dir, qstr,
7294 &si->name, &si->value,
7297 return security_inode_init_security(inode, dir, qstr,
7298 &ocfs2_initxattrs, NULL);
7301 int ocfs2_init_security_set(handle_t *handle,
7302 struct inode *inode,
7303 struct buffer_head *di_bh,
7304 struct ocfs2_security_xattr_info *si,
7305 struct ocfs2_alloc_context *xattr_ac,
7306 struct ocfs2_alloc_context *data_ac)
7308 return ocfs2_xattr_set_handle(handle, inode, di_bh,
7309 OCFS2_XATTR_INDEX_SECURITY,
7310 si->name, si->value, si->value_len, 0,
7314 const struct xattr_handler ocfs2_xattr_security_handler = {
7315 .prefix = XATTR_SECURITY_PREFIX,
7316 .get = ocfs2_xattr_security_get,
7317 .set = ocfs2_xattr_security_set,
7321 * 'trusted' attributes support
7323 static int ocfs2_xattr_trusted_get(const struct xattr_handler *handler,
7324 struct dentry *unused, struct inode *inode,
7325 const char *name, void *buffer, size_t size)
7327 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED,
7328 name, buffer, size);
7331 static int ocfs2_xattr_trusted_set(const struct xattr_handler *handler,
7332 struct dentry *unused, struct inode *inode,
7333 const char *name, const void *value,
7334 size_t size, int flags)
7336 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED,
7337 name, value, size, flags);
7340 const struct xattr_handler ocfs2_xattr_trusted_handler = {
7341 .prefix = XATTR_TRUSTED_PREFIX,
7342 .get = ocfs2_xattr_trusted_get,
7343 .set = ocfs2_xattr_trusted_set,
7347 * 'user' attributes support
7349 static int ocfs2_xattr_user_get(const struct xattr_handler *handler,
7350 struct dentry *unused, struct inode *inode,
7351 const char *name, void *buffer, size_t size)
7353 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7355 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7357 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
7361 static int ocfs2_xattr_user_set(const struct xattr_handler *handler,
7362 struct dentry *unused, struct inode *inode,
7363 const char *name, const void *value,
7364 size_t size, int flags)
7366 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7368 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7371 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER,
7372 name, value, size, flags);
7375 const struct xattr_handler ocfs2_xattr_user_handler = {
7376 .prefix = XATTR_USER_PREFIX,
7377 .get = ocfs2_xattr_user_get,
7378 .set = ocfs2_xattr_user_set,