Linux 3.16-rc2
[linux/fpc-iii.git] / fs / ocfs2 / xattr.c
blob016f01df382514fd6dc93ce1e7386c147909c54d
1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
4 * xattr.c
6 * Copyright (C) 2004, 2008 Oracle. All rights reserved.
8 * CREDITS:
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>
23 #include <linux/fs.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>
42 #include "ocfs2.h"
43 #include "alloc.h"
44 #include "blockcheck.h"
45 #include "dlmglue.h"
46 #include "file.h"
47 #include "symlink.h"
48 #include "sysfile.h"
49 #include "inode.h"
50 #include "journal.h"
51 #include "ocfs2_fs.h"
52 #include "suballoc.h"
53 #include "uptodate.h"
54 #include "buffer_head_io.h"
55 #include "super.h"
56 #include "xattr.h"
57 #include "refcounttree.h"
58 #include "acl.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 */
74 int bu_blocks;
77 struct ocfs2_xattr_set_ctxt {
78 handle_t *handle;
79 struct ocfs2_alloc_context *meta_ac;
80 struct ocfs2_alloc_context *data_ac;
81 struct ocfs2_cached_dealloc_ctxt dealloc;
82 int set_abort;
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,
106 NULL
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 {
120 int xi_name_index;
121 const char *xi_name;
122 int xi_name_len;
123 const void *xi_value;
124 size_t xi_value_len;
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;
136 void *base;
137 void *end;
138 struct ocfs2_xattr_entry *here;
139 int not_found;
142 /* Operations on struct ocfs2_xa_entry */
143 struct ocfs2_xa_loc;
144 struct ocfs2_xa_loc_operations {
146 * Journal functions
148 int (*xlo_journal_access)(handle_t *handle, struct ocfs2_xa_loc *loc,
149 int type);
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 */
204 int xl_size;
207 * The ocfs2_xattr_entry this location describes. If this is
208 * NULL, this location describes the on-disk structure where it
209 * would have been.
211 struct ocfs2_xattr_entry *xl_entry;
214 * Internal housekeeping
217 /* Buffer(s) containing this entry */
218 void *xl_storage;
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;
232 else
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,
253 int index,
254 int *block_off,
255 int *new_offset);
257 static int ocfs2_xattr_block_find(struct inode *inode,
258 int name_index,
259 const char *name,
260 struct ocfs2_xattr_search *xs);
261 static int ocfs2_xattr_index_block_find(struct inode *inode,
262 struct buffer_head *root_bh,
263 int name_index,
264 const char *name,
265 struct ocfs2_xattr_search *xs);
267 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
268 struct buffer_head *blk_bh,
269 char *buffer,
270 size_t buffer_size);
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,
287 void *para);
288 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
289 struct ocfs2_xattr_bucket *bucket,
290 void *para);
291 static int ocfs2_rm_xattr_cluster(struct inode *inode,
292 struct buffer_head *root_bh,
293 u64 blkno,
294 u32 cpos,
295 u32 len,
296 void *para);
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,
301 u32 *first_hash);
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,
308 int *meta_need,
309 int *credits);
310 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
311 struct ocfs2_xattr_bucket *bucket,
312 int offset,
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);
338 if (bucket) {
339 bucket->bu_inode = inode;
340 bucket->bu_blocks = blks;
343 return bucket;
346 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
348 int i;
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)
358 if (bucket) {
359 ocfs2_xattr_bucket_relse(bucket);
360 bucket->bu_inode = NULL;
361 kfree(bucket);
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
369 * them fully.
371 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
372 u64 xb_blkno, int new)
374 int i, rc = 0;
376 for (i = 0; i < bucket->bu_blocks; i++) {
377 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
378 xb_blkno + i);
379 if (!bucket->bu_bhs[i]) {
380 rc = -ENOMEM;
381 mlog_errno(rc);
382 break;
385 if (!ocfs2_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
386 bucket->bu_bhs[i])) {
387 if (new)
388 ocfs2_set_new_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
389 bucket->bu_bhs[i]);
390 else {
391 set_buffer_uptodate(bucket->bu_bhs[i]);
392 ocfs2_set_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
393 bucket->bu_bhs[i]);
398 if (rc)
399 ocfs2_xattr_bucket_relse(bucket);
400 return rc;
403 /* Read the xattr bucket at xb_blkno */
404 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
405 u64 xb_blkno)
407 int rc;
409 rc = ocfs2_read_blocks(INODE_CACHE(bucket->bu_inode), xb_blkno,
410 bucket->bu_blocks, bucket->bu_bhs, 0,
411 NULL);
412 if (!rc) {
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,
415 bucket->bu_bhs,
416 bucket->bu_blocks,
417 &bucket_xh(bucket)->xh_check);
418 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
419 if (rc)
420 mlog_errno(rc);
423 if (rc)
424 ocfs2_xattr_bucket_relse(bucket);
425 return rc;
428 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
429 struct ocfs2_xattr_bucket *bucket,
430 int type)
432 int i, rc = 0;
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);
438 if (rc) {
439 mlog_errno(rc);
440 break;
444 return rc;
447 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
448 struct ocfs2_xattr_bucket *bucket)
450 int i;
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)
465 int i;
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),
473 blocksize);
477 static int ocfs2_validate_xattr_block(struct super_block *sb,
478 struct buffer_head *bh)
480 int rc;
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);
494 if (rc)
495 return rc;
498 * Errors after here are fatal
501 if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
502 ocfs2_error(sb,
503 "Extended attribute block #%llu has bad "
504 "signature %.*s",
505 (unsigned long long)bh->b_blocknr, 7,
506 xb->xb_signature);
507 return -EINVAL;
510 if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
511 ocfs2_error(sb,
512 "Extended attribute block #%llu has an "
513 "invalid xb_blkno of %llu",
514 (unsigned long long)bh->b_blocknr,
515 (unsigned long long)le64_to_cpu(xb->xb_blkno));
516 return -EINVAL;
519 if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
520 ocfs2_error(sb,
521 "Extended attribute block #%llu has an invalid "
522 "xb_fs_generation of #%u",
523 (unsigned long long)bh->b_blocknr,
524 le32_to_cpu(xb->xb_fs_generation));
525 return -EINVAL;
528 return 0;
531 static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
532 struct buffer_head **bh)
534 int rc;
535 struct buffer_head *tmp = *bh;
537 rc = ocfs2_read_block(INODE_CACHE(inode), xb_blkno, &tmp,
538 ocfs2_validate_xattr_block);
540 /* If ocfs2_read_block() got us a new bh, pass it up. */
541 if (!rc && !*bh)
542 *bh = tmp;
544 return rc;
547 static inline const char *ocfs2_xattr_prefix(int name_index)
549 const struct xattr_handler *handler = NULL;
551 if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
552 handler = ocfs2_xattr_handler_map[name_index];
554 return handler ? handler->prefix : NULL;
557 static u32 ocfs2_xattr_name_hash(struct inode *inode,
558 const char *name,
559 int name_len)
561 /* Get hash value of uuid from super block */
562 u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
563 int i;
565 /* hash extended attribute name */
566 for (i = 0; i < name_len; i++) {
567 hash = (hash << OCFS2_HASH_SHIFT) ^
568 (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
569 *name++;
572 return hash;
575 static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
577 return namevalue_size(name_len, value_len) +
578 sizeof(struct ocfs2_xattr_entry);
581 static int ocfs2_xi_entry_usage(struct ocfs2_xattr_info *xi)
583 return namevalue_size_xi(xi) +
584 sizeof(struct ocfs2_xattr_entry);
587 static int ocfs2_xe_entry_usage(struct ocfs2_xattr_entry *xe)
589 return namevalue_size_xe(xe) +
590 sizeof(struct ocfs2_xattr_entry);
593 int ocfs2_calc_security_init(struct inode *dir,
594 struct ocfs2_security_xattr_info *si,
595 int *want_clusters,
596 int *xattr_credits,
597 struct ocfs2_alloc_context **xattr_ac)
599 int ret = 0;
600 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
601 int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
602 si->value_len);
605 * The max space of security xattr taken inline is
606 * 256(name) + 80(value) + 16(entry) = 352 bytes,
607 * So reserve one metadata block for it is ok.
609 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
610 s_size > OCFS2_XATTR_FREE_IN_IBODY) {
611 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
612 if (ret) {
613 mlog_errno(ret);
614 return ret;
616 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
619 /* reserve clusters for xattr value which will be set in B tree*/
620 if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
621 int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
622 si->value_len);
624 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
625 new_clusters);
626 *want_clusters += new_clusters;
628 return ret;
631 int ocfs2_calc_xattr_init(struct inode *dir,
632 struct buffer_head *dir_bh,
633 umode_t mode,
634 struct ocfs2_security_xattr_info *si,
635 int *want_clusters,
636 int *xattr_credits,
637 int *want_meta)
639 int ret = 0;
640 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
641 int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
643 if (si->enable)
644 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
645 si->value_len);
647 if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
648 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
649 OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
650 "", NULL, 0);
651 if (acl_len > 0) {
652 a_size = ocfs2_xattr_entry_real_size(0, acl_len);
653 if (S_ISDIR(mode))
654 a_size <<= 1;
655 } else if (acl_len != 0 && acl_len != -ENODATA) {
656 mlog_errno(ret);
657 return ret;
661 if (!(s_size + a_size))
662 return ret;
665 * The max space of security xattr taken inline is
666 * 256(name) + 80(value) + 16(entry) = 352 bytes,
667 * The max space of acl xattr taken inline is
668 * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
669 * when blocksize = 512, may reserve one more cluser for
670 * xattr bucket, otherwise reserve one metadata block
671 * for them is ok.
672 * If this is a new directory with inline data,
673 * we choose to reserve the entire inline area for
674 * directory contents and force an external xattr block.
676 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
677 (S_ISDIR(mode) && ocfs2_supports_inline_data(osb)) ||
678 (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
679 *want_meta = *want_meta + 1;
680 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
683 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
684 (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
685 *want_clusters += 1;
686 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
690 * reserve credits and clusters for xattrs which has large value
691 * and have to be set outside
693 if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
694 new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
695 si->value_len);
696 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
697 new_clusters);
698 *want_clusters += new_clusters;
700 if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
701 acl_len > OCFS2_XATTR_INLINE_SIZE) {
702 /* for directory, it has DEFAULT and ACCESS two types of acls */
703 new_clusters = (S_ISDIR(mode) ? 2 : 1) *
704 ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
705 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
706 new_clusters);
707 *want_clusters += new_clusters;
710 return ret;
713 static int ocfs2_xattr_extend_allocation(struct inode *inode,
714 u32 clusters_to_add,
715 struct ocfs2_xattr_value_buf *vb,
716 struct ocfs2_xattr_set_ctxt *ctxt)
718 int status = 0, credits;
719 handle_t *handle = ctxt->handle;
720 enum ocfs2_alloc_restarted why;
721 u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
722 struct ocfs2_extent_tree et;
724 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
726 while (clusters_to_add) {
727 trace_ocfs2_xattr_extend_allocation(clusters_to_add);
729 status = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
730 OCFS2_JOURNAL_ACCESS_WRITE);
731 if (status < 0) {
732 mlog_errno(status);
733 break;
736 prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
737 status = ocfs2_add_clusters_in_btree(handle,
738 &et,
739 &logical_start,
740 clusters_to_add,
742 ctxt->data_ac,
743 ctxt->meta_ac,
744 &why);
745 if ((status < 0) && (status != -EAGAIN)) {
746 if (status != -ENOSPC)
747 mlog_errno(status);
748 break;
751 ocfs2_journal_dirty(handle, vb->vb_bh);
753 clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) -
754 prev_clusters;
756 if (why != RESTART_NONE && clusters_to_add) {
758 * We can only fail in case the alloc file doesn't give
759 * up enough clusters.
761 BUG_ON(why == RESTART_META);
763 credits = ocfs2_calc_extend_credits(inode->i_sb,
764 &vb->vb_xv->xr_list);
765 status = ocfs2_extend_trans(handle, credits);
766 if (status < 0) {
767 status = -ENOMEM;
768 mlog_errno(status);
769 break;
774 return status;
777 static int __ocfs2_remove_xattr_range(struct inode *inode,
778 struct ocfs2_xattr_value_buf *vb,
779 u32 cpos, u32 phys_cpos, u32 len,
780 unsigned int ext_flags,
781 struct ocfs2_xattr_set_ctxt *ctxt)
783 int ret;
784 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
785 handle_t *handle = ctxt->handle;
786 struct ocfs2_extent_tree et;
788 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
790 ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
791 OCFS2_JOURNAL_ACCESS_WRITE);
792 if (ret) {
793 mlog_errno(ret);
794 goto out;
797 ret = ocfs2_remove_extent(handle, &et, cpos, len, ctxt->meta_ac,
798 &ctxt->dealloc);
799 if (ret) {
800 mlog_errno(ret);
801 goto out;
804 le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
805 ocfs2_journal_dirty(handle, vb->vb_bh);
807 if (ext_flags & OCFS2_EXT_REFCOUNTED)
808 ret = ocfs2_decrease_refcount(inode, handle,
809 ocfs2_blocks_to_clusters(inode->i_sb,
810 phys_blkno),
811 len, ctxt->meta_ac, &ctxt->dealloc, 1);
812 else
813 ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc,
814 phys_blkno, len);
815 if (ret)
816 mlog_errno(ret);
818 out:
819 return ret;
822 static int ocfs2_xattr_shrink_size(struct inode *inode,
823 u32 old_clusters,
824 u32 new_clusters,
825 struct ocfs2_xattr_value_buf *vb,
826 struct ocfs2_xattr_set_ctxt *ctxt)
828 int ret = 0;
829 unsigned int ext_flags;
830 u32 trunc_len, cpos, phys_cpos, alloc_size;
831 u64 block;
833 if (old_clusters <= new_clusters)
834 return 0;
836 cpos = new_clusters;
837 trunc_len = old_clusters - new_clusters;
838 while (trunc_len) {
839 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
840 &alloc_size,
841 &vb->vb_xv->xr_list, &ext_flags);
842 if (ret) {
843 mlog_errno(ret);
844 goto out;
847 if (alloc_size > trunc_len)
848 alloc_size = trunc_len;
850 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
851 phys_cpos, alloc_size,
852 ext_flags, ctxt);
853 if (ret) {
854 mlog_errno(ret);
855 goto out;
858 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
859 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode),
860 block, alloc_size);
861 cpos += alloc_size;
862 trunc_len -= alloc_size;
865 out:
866 return ret;
869 static int ocfs2_xattr_value_truncate(struct inode *inode,
870 struct ocfs2_xattr_value_buf *vb,
871 int len,
872 struct ocfs2_xattr_set_ctxt *ctxt)
874 int ret;
875 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
876 u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
878 if (new_clusters == old_clusters)
879 return 0;
881 if (new_clusters > old_clusters)
882 ret = ocfs2_xattr_extend_allocation(inode,
883 new_clusters - old_clusters,
884 vb, ctxt);
885 else
886 ret = ocfs2_xattr_shrink_size(inode,
887 old_clusters, new_clusters,
888 vb, ctxt);
890 return ret;
893 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
894 size_t *result, const char *prefix,
895 const char *name, int name_len)
897 char *p = buffer + *result;
898 int prefix_len = strlen(prefix);
899 int total_len = prefix_len + name_len + 1;
901 *result += total_len;
903 /* we are just looking for how big our buffer needs to be */
904 if (!size)
905 return 0;
907 if (*result > size)
908 return -ERANGE;
910 memcpy(p, prefix, prefix_len);
911 memcpy(p + prefix_len, name, name_len);
912 p[prefix_len + name_len] = '\0';
914 return 0;
917 static int ocfs2_xattr_list_entries(struct inode *inode,
918 struct ocfs2_xattr_header *header,
919 char *buffer, size_t buffer_size)
921 size_t result = 0;
922 int i, type, ret;
923 const char *prefix, *name;
925 for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
926 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
927 type = ocfs2_xattr_get_type(entry);
928 prefix = ocfs2_xattr_prefix(type);
930 if (prefix) {
931 name = (const char *)header +
932 le16_to_cpu(entry->xe_name_offset);
934 ret = ocfs2_xattr_list_entry(buffer, buffer_size,
935 &result, prefix, name,
936 entry->xe_name_len);
937 if (ret)
938 return ret;
942 return result;
945 int ocfs2_has_inline_xattr_value_outside(struct inode *inode,
946 struct ocfs2_dinode *di)
948 struct ocfs2_xattr_header *xh;
949 int i;
951 xh = (struct ocfs2_xattr_header *)
952 ((void *)di + inode->i_sb->s_blocksize -
953 le16_to_cpu(di->i_xattr_inline_size));
955 for (i = 0; i < le16_to_cpu(xh->xh_count); i++)
956 if (!ocfs2_xattr_is_local(&xh->xh_entries[i]))
957 return 1;
959 return 0;
962 static int ocfs2_xattr_ibody_list(struct inode *inode,
963 struct ocfs2_dinode *di,
964 char *buffer,
965 size_t buffer_size)
967 struct ocfs2_xattr_header *header = NULL;
968 struct ocfs2_inode_info *oi = OCFS2_I(inode);
969 int ret = 0;
971 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
972 return ret;
974 header = (struct ocfs2_xattr_header *)
975 ((void *)di + inode->i_sb->s_blocksize -
976 le16_to_cpu(di->i_xattr_inline_size));
978 ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
980 return ret;
983 static int ocfs2_xattr_block_list(struct inode *inode,
984 struct ocfs2_dinode *di,
985 char *buffer,
986 size_t buffer_size)
988 struct buffer_head *blk_bh = NULL;
989 struct ocfs2_xattr_block *xb;
990 int ret = 0;
992 if (!di->i_xattr_loc)
993 return ret;
995 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
996 &blk_bh);
997 if (ret < 0) {
998 mlog_errno(ret);
999 return ret;
1002 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1003 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1004 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
1005 ret = ocfs2_xattr_list_entries(inode, header,
1006 buffer, buffer_size);
1007 } else
1008 ret = ocfs2_xattr_tree_list_index_block(inode, blk_bh,
1009 buffer, buffer_size);
1011 brelse(blk_bh);
1013 return ret;
1016 ssize_t ocfs2_listxattr(struct dentry *dentry,
1017 char *buffer,
1018 size_t size)
1020 int ret = 0, i_ret = 0, b_ret = 0;
1021 struct buffer_head *di_bh = NULL;
1022 struct ocfs2_dinode *di = NULL;
1023 struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
1025 if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
1026 return -EOPNOTSUPP;
1028 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1029 return ret;
1031 ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
1032 if (ret < 0) {
1033 mlog_errno(ret);
1034 return ret;
1037 di = (struct ocfs2_dinode *)di_bh->b_data;
1039 down_read(&oi->ip_xattr_sem);
1040 i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
1041 if (i_ret < 0)
1042 b_ret = 0;
1043 else {
1044 if (buffer) {
1045 buffer += i_ret;
1046 size -= i_ret;
1048 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
1049 buffer, size);
1050 if (b_ret < 0)
1051 i_ret = 0;
1053 up_read(&oi->ip_xattr_sem);
1054 ocfs2_inode_unlock(dentry->d_inode, 0);
1056 brelse(di_bh);
1058 return i_ret + b_ret;
1061 static int ocfs2_xattr_find_entry(int name_index,
1062 const char *name,
1063 struct ocfs2_xattr_search *xs)
1065 struct ocfs2_xattr_entry *entry;
1066 size_t name_len;
1067 int i, cmp = 1;
1069 if (name == NULL)
1070 return -EINVAL;
1072 name_len = strlen(name);
1073 entry = xs->here;
1074 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
1075 cmp = name_index - ocfs2_xattr_get_type(entry);
1076 if (!cmp)
1077 cmp = name_len - entry->xe_name_len;
1078 if (!cmp)
1079 cmp = memcmp(name, (xs->base +
1080 le16_to_cpu(entry->xe_name_offset)),
1081 name_len);
1082 if (cmp == 0)
1083 break;
1084 entry += 1;
1086 xs->here = entry;
1088 return cmp ? -ENODATA : 0;
1091 static int ocfs2_xattr_get_value_outside(struct inode *inode,
1092 struct ocfs2_xattr_value_root *xv,
1093 void *buffer,
1094 size_t len)
1096 u32 cpos, p_cluster, num_clusters, bpc, clusters;
1097 u64 blkno;
1098 int i, ret = 0;
1099 size_t cplen, blocksize;
1100 struct buffer_head *bh = NULL;
1101 struct ocfs2_extent_list *el;
1103 el = &xv->xr_list;
1104 clusters = le32_to_cpu(xv->xr_clusters);
1105 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1106 blocksize = inode->i_sb->s_blocksize;
1108 cpos = 0;
1109 while (cpos < clusters) {
1110 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1111 &num_clusters, el, NULL);
1112 if (ret) {
1113 mlog_errno(ret);
1114 goto out;
1117 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1118 /* Copy ocfs2_xattr_value */
1119 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1120 ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1121 &bh, NULL);
1122 if (ret) {
1123 mlog_errno(ret);
1124 goto out;
1127 cplen = len >= blocksize ? blocksize : len;
1128 memcpy(buffer, bh->b_data, cplen);
1129 len -= cplen;
1130 buffer += cplen;
1132 brelse(bh);
1133 bh = NULL;
1134 if (len == 0)
1135 break;
1137 cpos += num_clusters;
1139 out:
1140 return ret;
1143 static int ocfs2_xattr_ibody_get(struct inode *inode,
1144 int name_index,
1145 const char *name,
1146 void *buffer,
1147 size_t buffer_size,
1148 struct ocfs2_xattr_search *xs)
1150 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1151 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1152 struct ocfs2_xattr_value_root *xv;
1153 size_t size;
1154 int ret = 0;
1156 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1157 return -ENODATA;
1159 xs->end = (void *)di + inode->i_sb->s_blocksize;
1160 xs->header = (struct ocfs2_xattr_header *)
1161 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1162 xs->base = (void *)xs->header;
1163 xs->here = xs->header->xh_entries;
1165 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1166 if (ret)
1167 return ret;
1168 size = le64_to_cpu(xs->here->xe_value_size);
1169 if (buffer) {
1170 if (size > buffer_size)
1171 return -ERANGE;
1172 if (ocfs2_xattr_is_local(xs->here)) {
1173 memcpy(buffer, (void *)xs->base +
1174 le16_to_cpu(xs->here->xe_name_offset) +
1175 OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1176 } else {
1177 xv = (struct ocfs2_xattr_value_root *)
1178 (xs->base + le16_to_cpu(
1179 xs->here->xe_name_offset) +
1180 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1181 ret = ocfs2_xattr_get_value_outside(inode, xv,
1182 buffer, size);
1183 if (ret < 0) {
1184 mlog_errno(ret);
1185 return ret;
1190 return size;
1193 static int ocfs2_xattr_block_get(struct inode *inode,
1194 int name_index,
1195 const char *name,
1196 void *buffer,
1197 size_t buffer_size,
1198 struct ocfs2_xattr_search *xs)
1200 struct ocfs2_xattr_block *xb;
1201 struct ocfs2_xattr_value_root *xv;
1202 size_t size;
1203 int ret = -ENODATA, name_offset, name_len, i;
1204 int uninitialized_var(block_off);
1206 xs->bucket = ocfs2_xattr_bucket_new(inode);
1207 if (!xs->bucket) {
1208 ret = -ENOMEM;
1209 mlog_errno(ret);
1210 goto cleanup;
1213 ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1214 if (ret) {
1215 mlog_errno(ret);
1216 goto cleanup;
1219 if (xs->not_found) {
1220 ret = -ENODATA;
1221 goto cleanup;
1224 xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1225 size = le64_to_cpu(xs->here->xe_value_size);
1226 if (buffer) {
1227 ret = -ERANGE;
1228 if (size > buffer_size)
1229 goto cleanup;
1231 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1232 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1233 i = xs->here - xs->header->xh_entries;
1235 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1236 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
1237 bucket_xh(xs->bucket),
1239 &block_off,
1240 &name_offset);
1241 xs->base = bucket_block(xs->bucket, block_off);
1243 if (ocfs2_xattr_is_local(xs->here)) {
1244 memcpy(buffer, (void *)xs->base +
1245 name_offset + name_len, size);
1246 } else {
1247 xv = (struct ocfs2_xattr_value_root *)
1248 (xs->base + name_offset + name_len);
1249 ret = ocfs2_xattr_get_value_outside(inode, xv,
1250 buffer, size);
1251 if (ret < 0) {
1252 mlog_errno(ret);
1253 goto cleanup;
1257 ret = size;
1258 cleanup:
1259 ocfs2_xattr_bucket_free(xs->bucket);
1261 brelse(xs->xattr_bh);
1262 xs->xattr_bh = NULL;
1263 return ret;
1266 int ocfs2_xattr_get_nolock(struct inode *inode,
1267 struct buffer_head *di_bh,
1268 int name_index,
1269 const char *name,
1270 void *buffer,
1271 size_t buffer_size)
1273 int ret;
1274 struct ocfs2_dinode *di = NULL;
1275 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1276 struct ocfs2_xattr_search xis = {
1277 .not_found = -ENODATA,
1279 struct ocfs2_xattr_search xbs = {
1280 .not_found = -ENODATA,
1283 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1284 return -EOPNOTSUPP;
1286 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1287 ret = -ENODATA;
1289 xis.inode_bh = xbs.inode_bh = di_bh;
1290 di = (struct ocfs2_dinode *)di_bh->b_data;
1292 ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1293 buffer_size, &xis);
1294 if (ret == -ENODATA && di->i_xattr_loc)
1295 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1296 buffer_size, &xbs);
1298 return ret;
1301 /* ocfs2_xattr_get()
1303 * Copy an extended attribute into the buffer provided.
1304 * Buffer is NULL to compute the size of buffer required.
1306 static int ocfs2_xattr_get(struct inode *inode,
1307 int name_index,
1308 const char *name,
1309 void *buffer,
1310 size_t buffer_size)
1312 int ret;
1313 struct buffer_head *di_bh = NULL;
1315 ret = ocfs2_inode_lock(inode, &di_bh, 0);
1316 if (ret < 0) {
1317 mlog_errno(ret);
1318 return ret;
1320 down_read(&OCFS2_I(inode)->ip_xattr_sem);
1321 ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1322 name, buffer, buffer_size);
1323 up_read(&OCFS2_I(inode)->ip_xattr_sem);
1325 ocfs2_inode_unlock(inode, 0);
1327 brelse(di_bh);
1329 return ret;
1332 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1333 handle_t *handle,
1334 struct ocfs2_xattr_value_buf *vb,
1335 const void *value,
1336 int value_len)
1338 int ret = 0, i, cp_len;
1339 u16 blocksize = inode->i_sb->s_blocksize;
1340 u32 p_cluster, num_clusters;
1341 u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1342 u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1343 u64 blkno;
1344 struct buffer_head *bh = NULL;
1345 unsigned int ext_flags;
1346 struct ocfs2_xattr_value_root *xv = vb->vb_xv;
1348 BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1350 while (cpos < clusters) {
1351 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1352 &num_clusters, &xv->xr_list,
1353 &ext_flags);
1354 if (ret) {
1355 mlog_errno(ret);
1356 goto out;
1359 BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1361 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1363 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1364 ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1365 &bh, NULL);
1366 if (ret) {
1367 mlog_errno(ret);
1368 goto out;
1371 ret = ocfs2_journal_access(handle,
1372 INODE_CACHE(inode),
1374 OCFS2_JOURNAL_ACCESS_WRITE);
1375 if (ret < 0) {
1376 mlog_errno(ret);
1377 goto out;
1380 cp_len = value_len > blocksize ? blocksize : value_len;
1381 memcpy(bh->b_data, value, cp_len);
1382 value_len -= cp_len;
1383 value += cp_len;
1384 if (cp_len < blocksize)
1385 memset(bh->b_data + cp_len, 0,
1386 blocksize - cp_len);
1388 ocfs2_journal_dirty(handle, bh);
1389 brelse(bh);
1390 bh = NULL;
1393 * XXX: do we need to empty all the following
1394 * blocks in this cluster?
1396 if (!value_len)
1397 break;
1399 cpos += num_clusters;
1401 out:
1402 brelse(bh);
1404 return ret;
1407 static int ocfs2_xa_check_space_helper(int needed_space, int free_start,
1408 int num_entries)
1410 int free_space;
1412 if (!needed_space)
1413 return 0;
1415 free_space = free_start -
1416 sizeof(struct ocfs2_xattr_header) -
1417 (num_entries * sizeof(struct ocfs2_xattr_entry)) -
1418 OCFS2_XATTR_HEADER_GAP;
1419 if (free_space < 0)
1420 return -EIO;
1421 if (free_space < needed_space)
1422 return -ENOSPC;
1424 return 0;
1427 static int ocfs2_xa_journal_access(handle_t *handle, struct ocfs2_xa_loc *loc,
1428 int type)
1430 return loc->xl_ops->xlo_journal_access(handle, loc, type);
1433 static void ocfs2_xa_journal_dirty(handle_t *handle, struct ocfs2_xa_loc *loc)
1435 loc->xl_ops->xlo_journal_dirty(handle, loc);
1438 /* Give a pointer into the storage for the given offset */
1439 static void *ocfs2_xa_offset_pointer(struct ocfs2_xa_loc *loc, int offset)
1441 BUG_ON(offset >= loc->xl_size);
1442 return loc->xl_ops->xlo_offset_pointer(loc, offset);
1446 * Wipe the name+value pair and allow the storage to reclaim it. This
1447 * must be followed by either removal of the entry or a call to
1448 * ocfs2_xa_add_namevalue().
1450 static void ocfs2_xa_wipe_namevalue(struct ocfs2_xa_loc *loc)
1452 loc->xl_ops->xlo_wipe_namevalue(loc);
1456 * Find lowest offset to a name+value pair. This is the start of our
1457 * downward-growing free space.
1459 static int ocfs2_xa_get_free_start(struct ocfs2_xa_loc *loc)
1461 return loc->xl_ops->xlo_get_free_start(loc);
1464 /* Can we reuse loc->xl_entry for xi? */
1465 static int ocfs2_xa_can_reuse_entry(struct ocfs2_xa_loc *loc,
1466 struct ocfs2_xattr_info *xi)
1468 return loc->xl_ops->xlo_can_reuse(loc, xi);
1471 /* How much free space is needed to set the new value */
1472 static int ocfs2_xa_check_space(struct ocfs2_xa_loc *loc,
1473 struct ocfs2_xattr_info *xi)
1475 return loc->xl_ops->xlo_check_space(loc, xi);
1478 static void ocfs2_xa_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1480 loc->xl_ops->xlo_add_entry(loc, name_hash);
1481 loc->xl_entry->xe_name_hash = cpu_to_le32(name_hash);
1483 * We can't leave the new entry's xe_name_offset at zero or
1484 * add_namevalue() will go nuts. We set it to the size of our
1485 * storage so that it can never be less than any other entry.
1487 loc->xl_entry->xe_name_offset = cpu_to_le16(loc->xl_size);
1490 static void ocfs2_xa_add_namevalue(struct ocfs2_xa_loc *loc,
1491 struct ocfs2_xattr_info *xi)
1493 int size = namevalue_size_xi(xi);
1494 int nameval_offset;
1495 char *nameval_buf;
1497 loc->xl_ops->xlo_add_namevalue(loc, size);
1498 loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
1499 loc->xl_entry->xe_name_len = xi->xi_name_len;
1500 ocfs2_xattr_set_type(loc->xl_entry, xi->xi_name_index);
1501 ocfs2_xattr_set_local(loc->xl_entry,
1502 xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE);
1504 nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1505 nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
1506 memset(nameval_buf, 0, size);
1507 memcpy(nameval_buf, xi->xi_name, xi->xi_name_len);
1510 static void ocfs2_xa_fill_value_buf(struct ocfs2_xa_loc *loc,
1511 struct ocfs2_xattr_value_buf *vb)
1513 int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1514 int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
1516 /* Value bufs are for value trees */
1517 BUG_ON(ocfs2_xattr_is_local(loc->xl_entry));
1518 BUG_ON(namevalue_size_xe(loc->xl_entry) !=
1519 (name_size + OCFS2_XATTR_ROOT_SIZE));
1521 loc->xl_ops->xlo_fill_value_buf(loc, vb);
1522 vb->vb_xv =
1523 (struct ocfs2_xattr_value_root *)ocfs2_xa_offset_pointer(loc,
1524 nameval_offset +
1525 name_size);
1528 static int ocfs2_xa_block_journal_access(handle_t *handle,
1529 struct ocfs2_xa_loc *loc, int type)
1531 struct buffer_head *bh = loc->xl_storage;
1532 ocfs2_journal_access_func access;
1534 if (loc->xl_size == (bh->b_size -
1535 offsetof(struct ocfs2_xattr_block,
1536 xb_attrs.xb_header)))
1537 access = ocfs2_journal_access_xb;
1538 else
1539 access = ocfs2_journal_access_di;
1540 return access(handle, INODE_CACHE(loc->xl_inode), bh, type);
1543 static void ocfs2_xa_block_journal_dirty(handle_t *handle,
1544 struct ocfs2_xa_loc *loc)
1546 struct buffer_head *bh = loc->xl_storage;
1548 ocfs2_journal_dirty(handle, bh);
1551 static void *ocfs2_xa_block_offset_pointer(struct ocfs2_xa_loc *loc,
1552 int offset)
1554 return (char *)loc->xl_header + offset;
1557 static int ocfs2_xa_block_can_reuse(struct ocfs2_xa_loc *loc,
1558 struct ocfs2_xattr_info *xi)
1561 * Block storage is strict. If the sizes aren't exact, we will
1562 * remove the old one and reinsert the new.
1564 return namevalue_size_xe(loc->xl_entry) ==
1565 namevalue_size_xi(xi);
1568 static int ocfs2_xa_block_get_free_start(struct ocfs2_xa_loc *loc)
1570 struct ocfs2_xattr_header *xh = loc->xl_header;
1571 int i, count = le16_to_cpu(xh->xh_count);
1572 int offset, free_start = loc->xl_size;
1574 for (i = 0; i < count; i++) {
1575 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1576 if (offset < free_start)
1577 free_start = offset;
1580 return free_start;
1583 static int ocfs2_xa_block_check_space(struct ocfs2_xa_loc *loc,
1584 struct ocfs2_xattr_info *xi)
1586 int count = le16_to_cpu(loc->xl_header->xh_count);
1587 int free_start = ocfs2_xa_get_free_start(loc);
1588 int needed_space = ocfs2_xi_entry_usage(xi);
1591 * Block storage will reclaim the original entry before inserting
1592 * the new value, so we only need the difference. If the new
1593 * entry is smaller than the old one, we don't need anything.
1595 if (loc->xl_entry) {
1596 /* Don't need space if we're reusing! */
1597 if (ocfs2_xa_can_reuse_entry(loc, xi))
1598 needed_space = 0;
1599 else
1600 needed_space -= ocfs2_xe_entry_usage(loc->xl_entry);
1602 if (needed_space < 0)
1603 needed_space = 0;
1604 return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1608 * Block storage for xattrs keeps the name+value pairs compacted. When
1609 * we remove one, we have to shift any that preceded it towards the end.
1611 static void ocfs2_xa_block_wipe_namevalue(struct ocfs2_xa_loc *loc)
1613 int i, offset;
1614 int namevalue_offset, first_namevalue_offset, namevalue_size;
1615 struct ocfs2_xattr_entry *entry = loc->xl_entry;
1616 struct ocfs2_xattr_header *xh = loc->xl_header;
1617 int count = le16_to_cpu(xh->xh_count);
1619 namevalue_offset = le16_to_cpu(entry->xe_name_offset);
1620 namevalue_size = namevalue_size_xe(entry);
1621 first_namevalue_offset = ocfs2_xa_get_free_start(loc);
1623 /* Shift the name+value pairs */
1624 memmove((char *)xh + first_namevalue_offset + namevalue_size,
1625 (char *)xh + first_namevalue_offset,
1626 namevalue_offset - first_namevalue_offset);
1627 memset((char *)xh + first_namevalue_offset, 0, namevalue_size);
1629 /* Now tell xh->xh_entries about it */
1630 for (i = 0; i < count; i++) {
1631 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1632 if (offset <= namevalue_offset)
1633 le16_add_cpu(&xh->xh_entries[i].xe_name_offset,
1634 namevalue_size);
1638 * Note that we don't update xh_free_start or xh_name_value_len
1639 * because they're not used in block-stored xattrs.
1643 static void ocfs2_xa_block_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1645 int count = le16_to_cpu(loc->xl_header->xh_count);
1646 loc->xl_entry = &(loc->xl_header->xh_entries[count]);
1647 le16_add_cpu(&loc->xl_header->xh_count, 1);
1648 memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1651 static void ocfs2_xa_block_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1653 int free_start = ocfs2_xa_get_free_start(loc);
1655 loc->xl_entry->xe_name_offset = cpu_to_le16(free_start - size);
1658 static void ocfs2_xa_block_fill_value_buf(struct ocfs2_xa_loc *loc,
1659 struct ocfs2_xattr_value_buf *vb)
1661 struct buffer_head *bh = loc->xl_storage;
1663 if (loc->xl_size == (bh->b_size -
1664 offsetof(struct ocfs2_xattr_block,
1665 xb_attrs.xb_header)))
1666 vb->vb_access = ocfs2_journal_access_xb;
1667 else
1668 vb->vb_access = ocfs2_journal_access_di;
1669 vb->vb_bh = bh;
1673 * Operations for xattrs stored in blocks. This includes inline inode
1674 * storage and unindexed ocfs2_xattr_blocks.
1676 static const struct ocfs2_xa_loc_operations ocfs2_xa_block_loc_ops = {
1677 .xlo_journal_access = ocfs2_xa_block_journal_access,
1678 .xlo_journal_dirty = ocfs2_xa_block_journal_dirty,
1679 .xlo_offset_pointer = ocfs2_xa_block_offset_pointer,
1680 .xlo_check_space = ocfs2_xa_block_check_space,
1681 .xlo_can_reuse = ocfs2_xa_block_can_reuse,
1682 .xlo_get_free_start = ocfs2_xa_block_get_free_start,
1683 .xlo_wipe_namevalue = ocfs2_xa_block_wipe_namevalue,
1684 .xlo_add_entry = ocfs2_xa_block_add_entry,
1685 .xlo_add_namevalue = ocfs2_xa_block_add_namevalue,
1686 .xlo_fill_value_buf = ocfs2_xa_block_fill_value_buf,
1689 static int ocfs2_xa_bucket_journal_access(handle_t *handle,
1690 struct ocfs2_xa_loc *loc, int type)
1692 struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1694 return ocfs2_xattr_bucket_journal_access(handle, bucket, type);
1697 static void ocfs2_xa_bucket_journal_dirty(handle_t *handle,
1698 struct ocfs2_xa_loc *loc)
1700 struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1702 ocfs2_xattr_bucket_journal_dirty(handle, bucket);
1705 static void *ocfs2_xa_bucket_offset_pointer(struct ocfs2_xa_loc *loc,
1706 int offset)
1708 struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1709 int block, block_offset;
1711 /* The header is at the front of the bucket */
1712 block = offset >> loc->xl_inode->i_sb->s_blocksize_bits;
1713 block_offset = offset % loc->xl_inode->i_sb->s_blocksize;
1715 return bucket_block(bucket, block) + block_offset;
1718 static int ocfs2_xa_bucket_can_reuse(struct ocfs2_xa_loc *loc,
1719 struct ocfs2_xattr_info *xi)
1721 return namevalue_size_xe(loc->xl_entry) >=
1722 namevalue_size_xi(xi);
1725 static int ocfs2_xa_bucket_get_free_start(struct ocfs2_xa_loc *loc)
1727 struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1728 return le16_to_cpu(bucket_xh(bucket)->xh_free_start);
1731 static int ocfs2_bucket_align_free_start(struct super_block *sb,
1732 int free_start, int size)
1735 * We need to make sure that the name+value pair fits within
1736 * one block.
1738 if (((free_start - size) >> sb->s_blocksize_bits) !=
1739 ((free_start - 1) >> sb->s_blocksize_bits))
1740 free_start -= free_start % sb->s_blocksize;
1742 return free_start;
1745 static int ocfs2_xa_bucket_check_space(struct ocfs2_xa_loc *loc,
1746 struct ocfs2_xattr_info *xi)
1748 int rc;
1749 int count = le16_to_cpu(loc->xl_header->xh_count);
1750 int free_start = ocfs2_xa_get_free_start(loc);
1751 int needed_space = ocfs2_xi_entry_usage(xi);
1752 int size = namevalue_size_xi(xi);
1753 struct super_block *sb = loc->xl_inode->i_sb;
1756 * Bucket storage does not reclaim name+value pairs it cannot
1757 * reuse. They live as holes until the bucket fills, and then
1758 * the bucket is defragmented. However, the bucket can reclaim
1759 * the ocfs2_xattr_entry.
1761 if (loc->xl_entry) {
1762 /* Don't need space if we're reusing! */
1763 if (ocfs2_xa_can_reuse_entry(loc, xi))
1764 needed_space = 0;
1765 else
1766 needed_space -= sizeof(struct ocfs2_xattr_entry);
1768 BUG_ON(needed_space < 0);
1770 if (free_start < size) {
1771 if (needed_space)
1772 return -ENOSPC;
1773 } else {
1775 * First we check if it would fit in the first place.
1776 * Below, we align the free start to a block. This may
1777 * slide us below the minimum gap. By checking unaligned
1778 * first, we avoid that error.
1780 rc = ocfs2_xa_check_space_helper(needed_space, free_start,
1781 count);
1782 if (rc)
1783 return rc;
1784 free_start = ocfs2_bucket_align_free_start(sb, free_start,
1785 size);
1787 return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1790 static void ocfs2_xa_bucket_wipe_namevalue(struct ocfs2_xa_loc *loc)
1792 le16_add_cpu(&loc->xl_header->xh_name_value_len,
1793 -namevalue_size_xe(loc->xl_entry));
1796 static void ocfs2_xa_bucket_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1798 struct ocfs2_xattr_header *xh = loc->xl_header;
1799 int count = le16_to_cpu(xh->xh_count);
1800 int low = 0, high = count - 1, tmp;
1801 struct ocfs2_xattr_entry *tmp_xe;
1804 * We keep buckets sorted by name_hash, so we need to find
1805 * our insert place.
1807 while (low <= high && count) {
1808 tmp = (low + high) / 2;
1809 tmp_xe = &xh->xh_entries[tmp];
1811 if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
1812 low = tmp + 1;
1813 else if (name_hash < le32_to_cpu(tmp_xe->xe_name_hash))
1814 high = tmp - 1;
1815 else {
1816 low = tmp;
1817 break;
1821 if (low != count)
1822 memmove(&xh->xh_entries[low + 1],
1823 &xh->xh_entries[low],
1824 ((count - low) * sizeof(struct ocfs2_xattr_entry)));
1826 le16_add_cpu(&xh->xh_count, 1);
1827 loc->xl_entry = &xh->xh_entries[low];
1828 memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1831 static void ocfs2_xa_bucket_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1833 int free_start = ocfs2_xa_get_free_start(loc);
1834 struct ocfs2_xattr_header *xh = loc->xl_header;
1835 struct super_block *sb = loc->xl_inode->i_sb;
1836 int nameval_offset;
1838 free_start = ocfs2_bucket_align_free_start(sb, free_start, size);
1839 nameval_offset = free_start - size;
1840 loc->xl_entry->xe_name_offset = cpu_to_le16(nameval_offset);
1841 xh->xh_free_start = cpu_to_le16(nameval_offset);
1842 le16_add_cpu(&xh->xh_name_value_len, size);
1846 static void ocfs2_xa_bucket_fill_value_buf(struct ocfs2_xa_loc *loc,
1847 struct ocfs2_xattr_value_buf *vb)
1849 struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1850 struct super_block *sb = loc->xl_inode->i_sb;
1851 int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1852 int size = namevalue_size_xe(loc->xl_entry);
1853 int block_offset = nameval_offset >> sb->s_blocksize_bits;
1855 /* Values are not allowed to straddle block boundaries */
1856 BUG_ON(block_offset !=
1857 ((nameval_offset + size - 1) >> sb->s_blocksize_bits));
1858 /* We expect the bucket to be filled in */
1859 BUG_ON(!bucket->bu_bhs[block_offset]);
1861 vb->vb_access = ocfs2_journal_access;
1862 vb->vb_bh = bucket->bu_bhs[block_offset];
1865 /* Operations for xattrs stored in buckets. */
1866 static const struct ocfs2_xa_loc_operations ocfs2_xa_bucket_loc_ops = {
1867 .xlo_journal_access = ocfs2_xa_bucket_journal_access,
1868 .xlo_journal_dirty = ocfs2_xa_bucket_journal_dirty,
1869 .xlo_offset_pointer = ocfs2_xa_bucket_offset_pointer,
1870 .xlo_check_space = ocfs2_xa_bucket_check_space,
1871 .xlo_can_reuse = ocfs2_xa_bucket_can_reuse,
1872 .xlo_get_free_start = ocfs2_xa_bucket_get_free_start,
1873 .xlo_wipe_namevalue = ocfs2_xa_bucket_wipe_namevalue,
1874 .xlo_add_entry = ocfs2_xa_bucket_add_entry,
1875 .xlo_add_namevalue = ocfs2_xa_bucket_add_namevalue,
1876 .xlo_fill_value_buf = ocfs2_xa_bucket_fill_value_buf,
1879 static unsigned int ocfs2_xa_value_clusters(struct ocfs2_xa_loc *loc)
1881 struct ocfs2_xattr_value_buf vb;
1883 if (ocfs2_xattr_is_local(loc->xl_entry))
1884 return 0;
1886 ocfs2_xa_fill_value_buf(loc, &vb);
1887 return le32_to_cpu(vb.vb_xv->xr_clusters);
1890 static int ocfs2_xa_value_truncate(struct ocfs2_xa_loc *loc, u64 bytes,
1891 struct ocfs2_xattr_set_ctxt *ctxt)
1893 int trunc_rc, access_rc;
1894 struct ocfs2_xattr_value_buf vb;
1896 ocfs2_xa_fill_value_buf(loc, &vb);
1897 trunc_rc = ocfs2_xattr_value_truncate(loc->xl_inode, &vb, bytes,
1898 ctxt);
1901 * The caller of ocfs2_xa_value_truncate() has already called
1902 * ocfs2_xa_journal_access on the loc. However, The truncate code
1903 * calls ocfs2_extend_trans(). This may commit the previous
1904 * transaction and open a new one. If this is a bucket, truncate
1905 * could leave only vb->vb_bh set up for journaling. Meanwhile,
1906 * the caller is expecting to dirty the entire bucket. So we must
1907 * reset the journal work. We do this even if truncate has failed,
1908 * as it could have failed after committing the extend.
1910 access_rc = ocfs2_xa_journal_access(ctxt->handle, loc,
1911 OCFS2_JOURNAL_ACCESS_WRITE);
1913 /* Errors in truncate take precedence */
1914 return trunc_rc ? trunc_rc : access_rc;
1917 static void ocfs2_xa_remove_entry(struct ocfs2_xa_loc *loc)
1919 int index, count;
1920 struct ocfs2_xattr_header *xh = loc->xl_header;
1921 struct ocfs2_xattr_entry *entry = loc->xl_entry;
1923 ocfs2_xa_wipe_namevalue(loc);
1924 loc->xl_entry = NULL;
1926 le16_add_cpu(&xh->xh_count, -1);
1927 count = le16_to_cpu(xh->xh_count);
1930 * Only zero out the entry if there are more remaining. This is
1931 * important for an empty bucket, as it keeps track of the
1932 * bucket's hash value. It doesn't hurt empty block storage.
1934 if (count) {
1935 index = ((char *)entry - (char *)&xh->xh_entries) /
1936 sizeof(struct ocfs2_xattr_entry);
1937 memmove(&xh->xh_entries[index], &xh->xh_entries[index + 1],
1938 (count - index) * sizeof(struct ocfs2_xattr_entry));
1939 memset(&xh->xh_entries[count], 0,
1940 sizeof(struct ocfs2_xattr_entry));
1945 * If we have a problem adjusting the size of an external value during
1946 * ocfs2_xa_prepare_entry() or ocfs2_xa_remove(), we may have an xattr
1947 * in an intermediate state. For example, the value may be partially
1948 * truncated.
1950 * If the value tree hasn't changed, the extend/truncate went nowhere.
1951 * We have nothing to do. The caller can treat it as a straight error.
1953 * If the value tree got partially truncated, we now have a corrupted
1954 * extended attribute. We're going to wipe its entry and leak the
1955 * clusters. Better to leak some storage than leave a corrupt entry.
1957 * If the value tree grew, it obviously didn't grow enough for the
1958 * new entry. We're not going to try and reclaim those clusters either.
1959 * If there was already an external value there (orig_clusters != 0),
1960 * the new clusters are attached safely and we can just leave the old
1961 * value in place. If there was no external value there, we remove
1962 * the entry.
1964 * This way, the xattr block we store in the journal will be consistent.
1965 * If the size change broke because of the journal, no changes will hit
1966 * disk anyway.
1968 static void ocfs2_xa_cleanup_value_truncate(struct ocfs2_xa_loc *loc,
1969 const char *what,
1970 unsigned int orig_clusters)
1972 unsigned int new_clusters = ocfs2_xa_value_clusters(loc);
1973 char *nameval_buf = ocfs2_xa_offset_pointer(loc,
1974 le16_to_cpu(loc->xl_entry->xe_name_offset));
1976 if (new_clusters < orig_clusters) {
1977 mlog(ML_ERROR,
1978 "Partial truncate while %s xattr %.*s. Leaking "
1979 "%u clusters and removing the entry\n",
1980 what, loc->xl_entry->xe_name_len, nameval_buf,
1981 orig_clusters - new_clusters);
1982 ocfs2_xa_remove_entry(loc);
1983 } else if (!orig_clusters) {
1984 mlog(ML_ERROR,
1985 "Unable to allocate an external value for xattr "
1986 "%.*s safely. Leaking %u clusters and removing the "
1987 "entry\n",
1988 loc->xl_entry->xe_name_len, nameval_buf,
1989 new_clusters - orig_clusters);
1990 ocfs2_xa_remove_entry(loc);
1991 } else if (new_clusters > orig_clusters)
1992 mlog(ML_ERROR,
1993 "Unable to grow xattr %.*s safely. %u new clusters "
1994 "have been added, but the value will not be "
1995 "modified\n",
1996 loc->xl_entry->xe_name_len, nameval_buf,
1997 new_clusters - orig_clusters);
2000 static int ocfs2_xa_remove(struct ocfs2_xa_loc *loc,
2001 struct ocfs2_xattr_set_ctxt *ctxt)
2003 int rc = 0;
2004 unsigned int orig_clusters;
2006 if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2007 orig_clusters = ocfs2_xa_value_clusters(loc);
2008 rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2009 if (rc) {
2010 mlog_errno(rc);
2012 * Since this is remove, we can return 0 if
2013 * ocfs2_xa_cleanup_value_truncate() is going to
2014 * wipe the entry anyway. So we check the
2015 * cluster count as well.
2017 if (orig_clusters != ocfs2_xa_value_clusters(loc))
2018 rc = 0;
2019 ocfs2_xa_cleanup_value_truncate(loc, "removing",
2020 orig_clusters);
2021 if (rc)
2022 goto out;
2026 ocfs2_xa_remove_entry(loc);
2028 out:
2029 return rc;
2032 static void ocfs2_xa_install_value_root(struct ocfs2_xa_loc *loc)
2034 int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
2035 char *nameval_buf;
2037 nameval_buf = ocfs2_xa_offset_pointer(loc,
2038 le16_to_cpu(loc->xl_entry->xe_name_offset));
2039 memcpy(nameval_buf + name_size, &def_xv, OCFS2_XATTR_ROOT_SIZE);
2043 * Take an existing entry and make it ready for the new value. This
2044 * won't allocate space, but it may free space. It should be ready for
2045 * ocfs2_xa_prepare_entry() to finish the work.
2047 static int ocfs2_xa_reuse_entry(struct ocfs2_xa_loc *loc,
2048 struct ocfs2_xattr_info *xi,
2049 struct ocfs2_xattr_set_ctxt *ctxt)
2051 int rc = 0;
2052 int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2053 unsigned int orig_clusters;
2054 char *nameval_buf;
2055 int xe_local = ocfs2_xattr_is_local(loc->xl_entry);
2056 int xi_local = xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE;
2058 BUG_ON(OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len) !=
2059 name_size);
2061 nameval_buf = ocfs2_xa_offset_pointer(loc,
2062 le16_to_cpu(loc->xl_entry->xe_name_offset));
2063 if (xe_local) {
2064 memset(nameval_buf + name_size, 0,
2065 namevalue_size_xe(loc->xl_entry) - name_size);
2066 if (!xi_local)
2067 ocfs2_xa_install_value_root(loc);
2068 } else {
2069 orig_clusters = ocfs2_xa_value_clusters(loc);
2070 if (xi_local) {
2071 rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2072 if (rc < 0)
2073 mlog_errno(rc);
2074 else
2075 memset(nameval_buf + name_size, 0,
2076 namevalue_size_xe(loc->xl_entry) -
2077 name_size);
2078 } else if (le64_to_cpu(loc->xl_entry->xe_value_size) >
2079 xi->xi_value_len) {
2080 rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len,
2081 ctxt);
2082 if (rc < 0)
2083 mlog_errno(rc);
2086 if (rc) {
2087 ocfs2_xa_cleanup_value_truncate(loc, "reusing",
2088 orig_clusters);
2089 goto out;
2093 loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
2094 ocfs2_xattr_set_local(loc->xl_entry, xi_local);
2096 out:
2097 return rc;
2101 * Prepares loc->xl_entry to receive the new xattr. This includes
2102 * properly setting up the name+value pair region. If loc->xl_entry
2103 * already exists, it will take care of modifying it appropriately.
2105 * Note that this modifies the data. You did journal_access already,
2106 * right?
2108 static int ocfs2_xa_prepare_entry(struct ocfs2_xa_loc *loc,
2109 struct ocfs2_xattr_info *xi,
2110 u32 name_hash,
2111 struct ocfs2_xattr_set_ctxt *ctxt)
2113 int rc = 0;
2114 unsigned int orig_clusters;
2115 __le64 orig_value_size = 0;
2117 rc = ocfs2_xa_check_space(loc, xi);
2118 if (rc)
2119 goto out;
2121 if (loc->xl_entry) {
2122 if (ocfs2_xa_can_reuse_entry(loc, xi)) {
2123 orig_value_size = loc->xl_entry->xe_value_size;
2124 rc = ocfs2_xa_reuse_entry(loc, xi, ctxt);
2125 if (rc)
2126 goto out;
2127 goto alloc_value;
2130 if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2131 orig_clusters = ocfs2_xa_value_clusters(loc);
2132 rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2133 if (rc) {
2134 mlog_errno(rc);
2135 ocfs2_xa_cleanup_value_truncate(loc,
2136 "overwriting",
2137 orig_clusters);
2138 goto out;
2141 ocfs2_xa_wipe_namevalue(loc);
2142 } else
2143 ocfs2_xa_add_entry(loc, name_hash);
2146 * If we get here, we have a blank entry. Fill it. We grow our
2147 * name+value pair back from the end.
2149 ocfs2_xa_add_namevalue(loc, xi);
2150 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
2151 ocfs2_xa_install_value_root(loc);
2153 alloc_value:
2154 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2155 orig_clusters = ocfs2_xa_value_clusters(loc);
2156 rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len, ctxt);
2157 if (rc < 0) {
2158 ctxt->set_abort = 1;
2159 ocfs2_xa_cleanup_value_truncate(loc, "growing",
2160 orig_clusters);
2162 * If we were growing an existing value,
2163 * ocfs2_xa_cleanup_value_truncate() won't remove
2164 * the entry. We need to restore the original value
2165 * size.
2167 if (loc->xl_entry) {
2168 BUG_ON(!orig_value_size);
2169 loc->xl_entry->xe_value_size = orig_value_size;
2171 mlog_errno(rc);
2175 out:
2176 return rc;
2180 * Store the value portion of the name+value pair. This will skip
2181 * values that are stored externally. Their tree roots were set up
2182 * by ocfs2_xa_prepare_entry().
2184 static int ocfs2_xa_store_value(struct ocfs2_xa_loc *loc,
2185 struct ocfs2_xattr_info *xi,
2186 struct ocfs2_xattr_set_ctxt *ctxt)
2188 int rc = 0;
2189 int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
2190 int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2191 char *nameval_buf;
2192 struct ocfs2_xattr_value_buf vb;
2194 nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
2195 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2196 ocfs2_xa_fill_value_buf(loc, &vb);
2197 rc = __ocfs2_xattr_set_value_outside(loc->xl_inode,
2198 ctxt->handle, &vb,
2199 xi->xi_value,
2200 xi->xi_value_len);
2201 } else
2202 memcpy(nameval_buf + name_size, xi->xi_value, xi->xi_value_len);
2204 return rc;
2207 static int ocfs2_xa_set(struct ocfs2_xa_loc *loc,
2208 struct ocfs2_xattr_info *xi,
2209 struct ocfs2_xattr_set_ctxt *ctxt)
2211 int ret;
2212 u32 name_hash = ocfs2_xattr_name_hash(loc->xl_inode, xi->xi_name,
2213 xi->xi_name_len);
2215 ret = ocfs2_xa_journal_access(ctxt->handle, loc,
2216 OCFS2_JOURNAL_ACCESS_WRITE);
2217 if (ret) {
2218 mlog_errno(ret);
2219 goto out;
2223 * From here on out, everything is going to modify the buffer a
2224 * little. Errors are going to leave the xattr header in a
2225 * sane state. Thus, even with errors we dirty the sucker.
2228 /* Don't worry, we are never called with !xi_value and !xl_entry */
2229 if (!xi->xi_value) {
2230 ret = ocfs2_xa_remove(loc, ctxt);
2231 goto out_dirty;
2234 ret = ocfs2_xa_prepare_entry(loc, xi, name_hash, ctxt);
2235 if (ret) {
2236 if (ret != -ENOSPC)
2237 mlog_errno(ret);
2238 goto out_dirty;
2241 ret = ocfs2_xa_store_value(loc, xi, ctxt);
2242 if (ret)
2243 mlog_errno(ret);
2245 out_dirty:
2246 ocfs2_xa_journal_dirty(ctxt->handle, loc);
2248 out:
2249 return ret;
2252 static void ocfs2_init_dinode_xa_loc(struct ocfs2_xa_loc *loc,
2253 struct inode *inode,
2254 struct buffer_head *bh,
2255 struct ocfs2_xattr_entry *entry)
2257 struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2259 BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_XATTR_FL));
2261 loc->xl_inode = inode;
2262 loc->xl_ops = &ocfs2_xa_block_loc_ops;
2263 loc->xl_storage = bh;
2264 loc->xl_entry = entry;
2265 loc->xl_size = le16_to_cpu(di->i_xattr_inline_size);
2266 loc->xl_header =
2267 (struct ocfs2_xattr_header *)(bh->b_data + bh->b_size -
2268 loc->xl_size);
2271 static void ocfs2_init_xattr_block_xa_loc(struct ocfs2_xa_loc *loc,
2272 struct inode *inode,
2273 struct buffer_head *bh,
2274 struct ocfs2_xattr_entry *entry)
2276 struct ocfs2_xattr_block *xb =
2277 (struct ocfs2_xattr_block *)bh->b_data;
2279 BUG_ON(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED);
2281 loc->xl_inode = inode;
2282 loc->xl_ops = &ocfs2_xa_block_loc_ops;
2283 loc->xl_storage = bh;
2284 loc->xl_header = &(xb->xb_attrs.xb_header);
2285 loc->xl_entry = entry;
2286 loc->xl_size = bh->b_size - offsetof(struct ocfs2_xattr_block,
2287 xb_attrs.xb_header);
2290 static void ocfs2_init_xattr_bucket_xa_loc(struct ocfs2_xa_loc *loc,
2291 struct ocfs2_xattr_bucket *bucket,
2292 struct ocfs2_xattr_entry *entry)
2294 loc->xl_inode = bucket->bu_inode;
2295 loc->xl_ops = &ocfs2_xa_bucket_loc_ops;
2296 loc->xl_storage = bucket;
2297 loc->xl_header = bucket_xh(bucket);
2298 loc->xl_entry = entry;
2299 loc->xl_size = OCFS2_XATTR_BUCKET_SIZE;
2303 * In xattr remove, if it is stored outside and refcounted, we may have
2304 * the chance to split the refcount tree. So need the allocators.
2306 static int ocfs2_lock_xattr_remove_allocators(struct inode *inode,
2307 struct ocfs2_xattr_value_root *xv,
2308 struct ocfs2_caching_info *ref_ci,
2309 struct buffer_head *ref_root_bh,
2310 struct ocfs2_alloc_context **meta_ac,
2311 int *ref_credits)
2313 int ret, meta_add = 0;
2314 u32 p_cluster, num_clusters;
2315 unsigned int ext_flags;
2317 *ref_credits = 0;
2318 ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
2319 &num_clusters,
2320 &xv->xr_list,
2321 &ext_flags);
2322 if (ret) {
2323 mlog_errno(ret);
2324 goto out;
2327 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
2328 goto out;
2330 ret = ocfs2_refcounted_xattr_delete_need(inode, ref_ci,
2331 ref_root_bh, xv,
2332 &meta_add, ref_credits);
2333 if (ret) {
2334 mlog_errno(ret);
2335 goto out;
2338 ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
2339 meta_add, meta_ac);
2340 if (ret)
2341 mlog_errno(ret);
2343 out:
2344 return ret;
2347 static int ocfs2_remove_value_outside(struct inode*inode,
2348 struct ocfs2_xattr_value_buf *vb,
2349 struct ocfs2_xattr_header *header,
2350 struct ocfs2_caching_info *ref_ci,
2351 struct buffer_head *ref_root_bh)
2353 int ret = 0, i, ref_credits;
2354 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2355 struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2356 void *val;
2358 ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
2360 for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
2361 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
2363 if (ocfs2_xattr_is_local(entry))
2364 continue;
2366 val = (void *)header +
2367 le16_to_cpu(entry->xe_name_offset);
2368 vb->vb_xv = (struct ocfs2_xattr_value_root *)
2369 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
2371 ret = ocfs2_lock_xattr_remove_allocators(inode, vb->vb_xv,
2372 ref_ci, ref_root_bh,
2373 &ctxt.meta_ac,
2374 &ref_credits);
2376 ctxt.handle = ocfs2_start_trans(osb, ref_credits +
2377 ocfs2_remove_extent_credits(osb->sb));
2378 if (IS_ERR(ctxt.handle)) {
2379 ret = PTR_ERR(ctxt.handle);
2380 mlog_errno(ret);
2381 break;
2384 ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
2386 ocfs2_commit_trans(osb, ctxt.handle);
2387 if (ctxt.meta_ac) {
2388 ocfs2_free_alloc_context(ctxt.meta_ac);
2389 ctxt.meta_ac = NULL;
2392 if (ret < 0) {
2393 mlog_errno(ret);
2394 break;
2399 if (ctxt.meta_ac)
2400 ocfs2_free_alloc_context(ctxt.meta_ac);
2401 ocfs2_schedule_truncate_log_flush(osb, 1);
2402 ocfs2_run_deallocs(osb, &ctxt.dealloc);
2403 return ret;
2406 static int ocfs2_xattr_ibody_remove(struct inode *inode,
2407 struct buffer_head *di_bh,
2408 struct ocfs2_caching_info *ref_ci,
2409 struct buffer_head *ref_root_bh)
2412 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2413 struct ocfs2_xattr_header *header;
2414 int ret;
2415 struct ocfs2_xattr_value_buf vb = {
2416 .vb_bh = di_bh,
2417 .vb_access = ocfs2_journal_access_di,
2420 header = (struct ocfs2_xattr_header *)
2421 ((void *)di + inode->i_sb->s_blocksize -
2422 le16_to_cpu(di->i_xattr_inline_size));
2424 ret = ocfs2_remove_value_outside(inode, &vb, header,
2425 ref_ci, ref_root_bh);
2427 return ret;
2430 struct ocfs2_rm_xattr_bucket_para {
2431 struct ocfs2_caching_info *ref_ci;
2432 struct buffer_head *ref_root_bh;
2435 static int ocfs2_xattr_block_remove(struct inode *inode,
2436 struct buffer_head *blk_bh,
2437 struct ocfs2_caching_info *ref_ci,
2438 struct buffer_head *ref_root_bh)
2440 struct ocfs2_xattr_block *xb;
2441 int ret = 0;
2442 struct ocfs2_xattr_value_buf vb = {
2443 .vb_bh = blk_bh,
2444 .vb_access = ocfs2_journal_access_xb,
2446 struct ocfs2_rm_xattr_bucket_para args = {
2447 .ref_ci = ref_ci,
2448 .ref_root_bh = ref_root_bh,
2451 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2452 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2453 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
2454 ret = ocfs2_remove_value_outside(inode, &vb, header,
2455 ref_ci, ref_root_bh);
2456 } else
2457 ret = ocfs2_iterate_xattr_index_block(inode,
2458 blk_bh,
2459 ocfs2_rm_xattr_cluster,
2460 &args);
2462 return ret;
2465 static int ocfs2_xattr_free_block(struct inode *inode,
2466 u64 block,
2467 struct ocfs2_caching_info *ref_ci,
2468 struct buffer_head *ref_root_bh)
2470 struct inode *xb_alloc_inode;
2471 struct buffer_head *xb_alloc_bh = NULL;
2472 struct buffer_head *blk_bh = NULL;
2473 struct ocfs2_xattr_block *xb;
2474 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2475 handle_t *handle;
2476 int ret = 0;
2477 u64 blk, bg_blkno;
2478 u16 bit;
2480 ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
2481 if (ret < 0) {
2482 mlog_errno(ret);
2483 goto out;
2486 ret = ocfs2_xattr_block_remove(inode, blk_bh, ref_ci, ref_root_bh);
2487 if (ret < 0) {
2488 mlog_errno(ret);
2489 goto out;
2492 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2493 blk = le64_to_cpu(xb->xb_blkno);
2494 bit = le16_to_cpu(xb->xb_suballoc_bit);
2495 if (xb->xb_suballoc_loc)
2496 bg_blkno = le64_to_cpu(xb->xb_suballoc_loc);
2497 else
2498 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2500 xb_alloc_inode = ocfs2_get_system_file_inode(osb,
2501 EXTENT_ALLOC_SYSTEM_INODE,
2502 le16_to_cpu(xb->xb_suballoc_slot));
2503 if (!xb_alloc_inode) {
2504 ret = -ENOMEM;
2505 mlog_errno(ret);
2506 goto out;
2508 mutex_lock(&xb_alloc_inode->i_mutex);
2510 ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
2511 if (ret < 0) {
2512 mlog_errno(ret);
2513 goto out_mutex;
2516 handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
2517 if (IS_ERR(handle)) {
2518 ret = PTR_ERR(handle);
2519 mlog_errno(ret);
2520 goto out_unlock;
2523 ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
2524 bit, bg_blkno, 1);
2525 if (ret < 0)
2526 mlog_errno(ret);
2528 ocfs2_commit_trans(osb, handle);
2529 out_unlock:
2530 ocfs2_inode_unlock(xb_alloc_inode, 1);
2531 brelse(xb_alloc_bh);
2532 out_mutex:
2533 mutex_unlock(&xb_alloc_inode->i_mutex);
2534 iput(xb_alloc_inode);
2535 out:
2536 brelse(blk_bh);
2537 return ret;
2541 * ocfs2_xattr_remove()
2543 * Free extended attribute resources associated with this inode.
2545 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
2547 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2548 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2549 struct ocfs2_refcount_tree *ref_tree = NULL;
2550 struct buffer_head *ref_root_bh = NULL;
2551 struct ocfs2_caching_info *ref_ci = NULL;
2552 handle_t *handle;
2553 int ret;
2555 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2556 return 0;
2558 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
2559 return 0;
2561 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
2562 ret = ocfs2_lock_refcount_tree(OCFS2_SB(inode->i_sb),
2563 le64_to_cpu(di->i_refcount_loc),
2564 1, &ref_tree, &ref_root_bh);
2565 if (ret) {
2566 mlog_errno(ret);
2567 goto out;
2569 ref_ci = &ref_tree->rf_ci;
2573 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2574 ret = ocfs2_xattr_ibody_remove(inode, di_bh,
2575 ref_ci, ref_root_bh);
2576 if (ret < 0) {
2577 mlog_errno(ret);
2578 goto out;
2582 if (di->i_xattr_loc) {
2583 ret = ocfs2_xattr_free_block(inode,
2584 le64_to_cpu(di->i_xattr_loc),
2585 ref_ci, ref_root_bh);
2586 if (ret < 0) {
2587 mlog_errno(ret);
2588 goto out;
2592 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
2593 OCFS2_INODE_UPDATE_CREDITS);
2594 if (IS_ERR(handle)) {
2595 ret = PTR_ERR(handle);
2596 mlog_errno(ret);
2597 goto out;
2599 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2600 OCFS2_JOURNAL_ACCESS_WRITE);
2601 if (ret) {
2602 mlog_errno(ret);
2603 goto out_commit;
2606 di->i_xattr_loc = 0;
2608 spin_lock(&oi->ip_lock);
2609 oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
2610 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2611 spin_unlock(&oi->ip_lock);
2612 ocfs2_update_inode_fsync_trans(handle, inode, 0);
2614 ocfs2_journal_dirty(handle, di_bh);
2615 out_commit:
2616 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
2617 out:
2618 if (ref_tree)
2619 ocfs2_unlock_refcount_tree(OCFS2_SB(inode->i_sb), ref_tree, 1);
2620 brelse(ref_root_bh);
2621 return ret;
2624 static int ocfs2_xattr_has_space_inline(struct inode *inode,
2625 struct ocfs2_dinode *di)
2627 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2628 unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
2629 int free;
2631 if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
2632 return 0;
2634 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2635 struct ocfs2_inline_data *idata = &di->id2.i_data;
2636 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
2637 } else if (ocfs2_inode_is_fast_symlink(inode)) {
2638 free = ocfs2_fast_symlink_chars(inode->i_sb) -
2639 le64_to_cpu(di->i_size);
2640 } else {
2641 struct ocfs2_extent_list *el = &di->id2.i_list;
2642 free = (le16_to_cpu(el->l_count) -
2643 le16_to_cpu(el->l_next_free_rec)) *
2644 sizeof(struct ocfs2_extent_rec);
2646 if (free >= xattrsize)
2647 return 1;
2649 return 0;
2653 * ocfs2_xattr_ibody_find()
2655 * Find extended attribute in inode block and
2656 * fill search info into struct ocfs2_xattr_search.
2658 static int ocfs2_xattr_ibody_find(struct inode *inode,
2659 int name_index,
2660 const char *name,
2661 struct ocfs2_xattr_search *xs)
2663 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2664 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2665 int ret;
2666 int has_space = 0;
2668 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2669 return 0;
2671 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2672 down_read(&oi->ip_alloc_sem);
2673 has_space = ocfs2_xattr_has_space_inline(inode, di);
2674 up_read(&oi->ip_alloc_sem);
2675 if (!has_space)
2676 return 0;
2679 xs->xattr_bh = xs->inode_bh;
2680 xs->end = (void *)di + inode->i_sb->s_blocksize;
2681 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
2682 xs->header = (struct ocfs2_xattr_header *)
2683 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
2684 else
2685 xs->header = (struct ocfs2_xattr_header *)
2686 (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
2687 xs->base = (void *)xs->header;
2688 xs->here = xs->header->xh_entries;
2690 /* Find the named attribute. */
2691 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2692 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2693 if (ret && ret != -ENODATA)
2694 return ret;
2695 xs->not_found = ret;
2698 return 0;
2701 static int ocfs2_xattr_ibody_init(struct inode *inode,
2702 struct buffer_head *di_bh,
2703 struct ocfs2_xattr_set_ctxt *ctxt)
2705 int ret;
2706 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2707 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2708 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2709 unsigned int xattrsize = osb->s_xattr_inline_size;
2711 if (!ocfs2_xattr_has_space_inline(inode, di)) {
2712 ret = -ENOSPC;
2713 goto out;
2716 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode), di_bh,
2717 OCFS2_JOURNAL_ACCESS_WRITE);
2718 if (ret) {
2719 mlog_errno(ret);
2720 goto out;
2724 * Adjust extent record count or inline data size
2725 * to reserve space for extended attribute.
2727 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2728 struct ocfs2_inline_data *idata = &di->id2.i_data;
2729 le16_add_cpu(&idata->id_count, -xattrsize);
2730 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
2731 struct ocfs2_extent_list *el = &di->id2.i_list;
2732 le16_add_cpu(&el->l_count, -(xattrsize /
2733 sizeof(struct ocfs2_extent_rec)));
2735 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
2737 spin_lock(&oi->ip_lock);
2738 oi->ip_dyn_features |= OCFS2_INLINE_XATTR_FL|OCFS2_HAS_XATTR_FL;
2739 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2740 spin_unlock(&oi->ip_lock);
2742 ocfs2_journal_dirty(ctxt->handle, di_bh);
2744 out:
2745 return ret;
2749 * ocfs2_xattr_ibody_set()
2751 * Set, replace or remove an extended attribute into inode block.
2754 static int ocfs2_xattr_ibody_set(struct inode *inode,
2755 struct ocfs2_xattr_info *xi,
2756 struct ocfs2_xattr_search *xs,
2757 struct ocfs2_xattr_set_ctxt *ctxt)
2759 int ret;
2760 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2761 struct ocfs2_xa_loc loc;
2763 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2764 return -ENOSPC;
2766 down_write(&oi->ip_alloc_sem);
2767 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2768 ret = ocfs2_xattr_ibody_init(inode, xs->inode_bh, ctxt);
2769 if (ret) {
2770 if (ret != -ENOSPC)
2771 mlog_errno(ret);
2772 goto out;
2776 ocfs2_init_dinode_xa_loc(&loc, inode, xs->inode_bh,
2777 xs->not_found ? NULL : xs->here);
2778 ret = ocfs2_xa_set(&loc, xi, ctxt);
2779 if (ret) {
2780 if (ret != -ENOSPC)
2781 mlog_errno(ret);
2782 goto out;
2784 xs->here = loc.xl_entry;
2786 out:
2787 up_write(&oi->ip_alloc_sem);
2789 return ret;
2793 * ocfs2_xattr_block_find()
2795 * Find extended attribute in external block and
2796 * fill search info into struct ocfs2_xattr_search.
2798 static int ocfs2_xattr_block_find(struct inode *inode,
2799 int name_index,
2800 const char *name,
2801 struct ocfs2_xattr_search *xs)
2803 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2804 struct buffer_head *blk_bh = NULL;
2805 struct ocfs2_xattr_block *xb;
2806 int ret = 0;
2808 if (!di->i_xattr_loc)
2809 return ret;
2811 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2812 &blk_bh);
2813 if (ret < 0) {
2814 mlog_errno(ret);
2815 return ret;
2818 xs->xattr_bh = blk_bh;
2819 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2821 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2822 xs->header = &xb->xb_attrs.xb_header;
2823 xs->base = (void *)xs->header;
2824 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2825 xs->here = xs->header->xh_entries;
2827 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2828 } else
2829 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2830 name_index,
2831 name, xs);
2833 if (ret && ret != -ENODATA) {
2834 xs->xattr_bh = NULL;
2835 goto cleanup;
2837 xs->not_found = ret;
2838 return 0;
2839 cleanup:
2840 brelse(blk_bh);
2842 return ret;
2845 static int ocfs2_create_xattr_block(struct inode *inode,
2846 struct buffer_head *inode_bh,
2847 struct ocfs2_xattr_set_ctxt *ctxt,
2848 int indexed,
2849 struct buffer_head **ret_bh)
2851 int ret;
2852 u16 suballoc_bit_start;
2853 u32 num_got;
2854 u64 suballoc_loc, first_blkno;
2855 struct ocfs2_dinode *di = (struct ocfs2_dinode *)inode_bh->b_data;
2856 struct buffer_head *new_bh = NULL;
2857 struct ocfs2_xattr_block *xblk;
2859 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
2860 inode_bh, OCFS2_JOURNAL_ACCESS_CREATE);
2861 if (ret < 0) {
2862 mlog_errno(ret);
2863 goto end;
2866 ret = ocfs2_claim_metadata(ctxt->handle, ctxt->meta_ac, 1,
2867 &suballoc_loc, &suballoc_bit_start,
2868 &num_got, &first_blkno);
2869 if (ret < 0) {
2870 mlog_errno(ret);
2871 goto end;
2874 new_bh = sb_getblk(inode->i_sb, first_blkno);
2875 if (!new_bh) {
2876 ret = -ENOMEM;
2877 mlog_errno(ret);
2878 goto end;
2881 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2883 ret = ocfs2_journal_access_xb(ctxt->handle, INODE_CACHE(inode),
2884 new_bh,
2885 OCFS2_JOURNAL_ACCESS_CREATE);
2886 if (ret < 0) {
2887 mlog_errno(ret);
2888 goto end;
2891 /* Initialize ocfs2_xattr_block */
2892 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2893 memset(xblk, 0, inode->i_sb->s_blocksize);
2894 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2895 xblk->xb_suballoc_slot = cpu_to_le16(ctxt->meta_ac->ac_alloc_slot);
2896 xblk->xb_suballoc_loc = cpu_to_le64(suballoc_loc);
2897 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2898 xblk->xb_fs_generation =
2899 cpu_to_le32(OCFS2_SB(inode->i_sb)->fs_generation);
2900 xblk->xb_blkno = cpu_to_le64(first_blkno);
2901 if (indexed) {
2902 struct ocfs2_xattr_tree_root *xr = &xblk->xb_attrs.xb_root;
2903 xr->xt_clusters = cpu_to_le32(1);
2904 xr->xt_last_eb_blk = 0;
2905 xr->xt_list.l_tree_depth = 0;
2906 xr->xt_list.l_count = cpu_to_le16(
2907 ocfs2_xattr_recs_per_xb(inode->i_sb));
2908 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2909 xblk->xb_flags = cpu_to_le16(OCFS2_XATTR_INDEXED);
2911 ocfs2_journal_dirty(ctxt->handle, new_bh);
2913 /* Add it to the inode */
2914 di->i_xattr_loc = cpu_to_le64(first_blkno);
2916 spin_lock(&OCFS2_I(inode)->ip_lock);
2917 OCFS2_I(inode)->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
2918 di->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
2919 spin_unlock(&OCFS2_I(inode)->ip_lock);
2921 ocfs2_journal_dirty(ctxt->handle, inode_bh);
2923 *ret_bh = new_bh;
2924 new_bh = NULL;
2926 end:
2927 brelse(new_bh);
2928 return ret;
2932 * ocfs2_xattr_block_set()
2934 * Set, replace or remove an extended attribute into external block.
2937 static int ocfs2_xattr_block_set(struct inode *inode,
2938 struct ocfs2_xattr_info *xi,
2939 struct ocfs2_xattr_search *xs,
2940 struct ocfs2_xattr_set_ctxt *ctxt)
2942 struct buffer_head *new_bh = NULL;
2943 struct ocfs2_xattr_block *xblk = NULL;
2944 int ret;
2945 struct ocfs2_xa_loc loc;
2947 if (!xs->xattr_bh) {
2948 ret = ocfs2_create_xattr_block(inode, xs->inode_bh, ctxt,
2949 0, &new_bh);
2950 if (ret) {
2951 mlog_errno(ret);
2952 goto end;
2955 xs->xattr_bh = new_bh;
2956 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2957 xs->header = &xblk->xb_attrs.xb_header;
2958 xs->base = (void *)xs->header;
2959 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2960 xs->here = xs->header->xh_entries;
2961 } else
2962 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2964 if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2965 ocfs2_init_xattr_block_xa_loc(&loc, inode, xs->xattr_bh,
2966 xs->not_found ? NULL : xs->here);
2968 ret = ocfs2_xa_set(&loc, xi, ctxt);
2969 if (!ret)
2970 xs->here = loc.xl_entry;
2971 else if ((ret != -ENOSPC) || ctxt->set_abort)
2972 goto end;
2973 else {
2974 ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2975 if (ret)
2976 goto end;
2980 if (le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)
2981 ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2983 end:
2984 return ret;
2987 /* Check whether the new xattr can be inserted into the inode. */
2988 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2989 struct ocfs2_xattr_info *xi,
2990 struct ocfs2_xattr_search *xs)
2992 struct ocfs2_xattr_entry *last;
2993 int free, i;
2994 size_t min_offs = xs->end - xs->base;
2996 if (!xs->header)
2997 return 0;
2999 last = xs->header->xh_entries;
3001 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
3002 size_t offs = le16_to_cpu(last->xe_name_offset);
3003 if (offs < min_offs)
3004 min_offs = offs;
3005 last += 1;
3008 free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
3009 if (free < 0)
3010 return 0;
3012 BUG_ON(!xs->not_found);
3014 if (free >= (sizeof(struct ocfs2_xattr_entry) + namevalue_size_xi(xi)))
3015 return 1;
3017 return 0;
3020 static int ocfs2_calc_xattr_set_need(struct inode *inode,
3021 struct ocfs2_dinode *di,
3022 struct ocfs2_xattr_info *xi,
3023 struct ocfs2_xattr_search *xis,
3024 struct ocfs2_xattr_search *xbs,
3025 int *clusters_need,
3026 int *meta_need,
3027 int *credits_need)
3029 int ret = 0, old_in_xb = 0;
3030 int clusters_add = 0, meta_add = 0, credits = 0;
3031 struct buffer_head *bh = NULL;
3032 struct ocfs2_xattr_block *xb = NULL;
3033 struct ocfs2_xattr_entry *xe = NULL;
3034 struct ocfs2_xattr_value_root *xv = NULL;
3035 char *base = NULL;
3036 int name_offset, name_len = 0;
3037 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
3038 xi->xi_value_len);
3039 u64 value_size;
3042 * Calculate the clusters we need to write.
3043 * No matter whether we replace an old one or add a new one,
3044 * we need this for writing.
3046 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
3047 credits += new_clusters *
3048 ocfs2_clusters_to_blocks(inode->i_sb, 1);
3050 if (xis->not_found && xbs->not_found) {
3051 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3053 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3054 clusters_add += new_clusters;
3055 credits += ocfs2_calc_extend_credits(inode->i_sb,
3056 &def_xv.xv.xr_list);
3059 goto meta_guess;
3062 if (!xis->not_found) {
3063 xe = xis->here;
3064 name_offset = le16_to_cpu(xe->xe_name_offset);
3065 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3066 base = xis->base;
3067 credits += OCFS2_INODE_UPDATE_CREDITS;
3068 } else {
3069 int i, block_off = 0;
3070 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3071 xe = xbs->here;
3072 name_offset = le16_to_cpu(xe->xe_name_offset);
3073 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3074 i = xbs->here - xbs->header->xh_entries;
3075 old_in_xb = 1;
3077 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3078 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3079 bucket_xh(xbs->bucket),
3080 i, &block_off,
3081 &name_offset);
3082 base = bucket_block(xbs->bucket, block_off);
3083 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3084 } else {
3085 base = xbs->base;
3086 credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
3091 * delete a xattr doesn't need metadata and cluster allocation.
3092 * so just calculate the credits and return.
3094 * The credits for removing the value tree will be extended
3095 * by ocfs2_remove_extent itself.
3097 if (!xi->xi_value) {
3098 if (!ocfs2_xattr_is_local(xe))
3099 credits += ocfs2_remove_extent_credits(inode->i_sb);
3101 goto out;
3104 /* do cluster allocation guess first. */
3105 value_size = le64_to_cpu(xe->xe_value_size);
3107 if (old_in_xb) {
3109 * In xattr set, we always try to set the xe in inode first,
3110 * so if it can be inserted into inode successfully, the old
3111 * one will be removed from the xattr block, and this xattr
3112 * will be inserted into inode as a new xattr in inode.
3114 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
3115 clusters_add += new_clusters;
3116 credits += ocfs2_remove_extent_credits(inode->i_sb) +
3117 OCFS2_INODE_UPDATE_CREDITS;
3118 if (!ocfs2_xattr_is_local(xe))
3119 credits += ocfs2_calc_extend_credits(
3120 inode->i_sb,
3121 &def_xv.xv.xr_list);
3122 goto out;
3126 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3127 /* the new values will be stored outside. */
3128 u32 old_clusters = 0;
3130 if (!ocfs2_xattr_is_local(xe)) {
3131 old_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
3132 value_size);
3133 xv = (struct ocfs2_xattr_value_root *)
3134 (base + name_offset + name_len);
3135 value_size = OCFS2_XATTR_ROOT_SIZE;
3136 } else
3137 xv = &def_xv.xv;
3139 if (old_clusters >= new_clusters) {
3140 credits += ocfs2_remove_extent_credits(inode->i_sb);
3141 goto out;
3142 } else {
3143 meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
3144 clusters_add += new_clusters - old_clusters;
3145 credits += ocfs2_calc_extend_credits(inode->i_sb,
3146 &xv->xr_list);
3147 if (value_size >= OCFS2_XATTR_ROOT_SIZE)
3148 goto out;
3150 } else {
3152 * Now the new value will be stored inside. So if the new
3153 * value is smaller than the size of value root or the old
3154 * value, we don't need any allocation, otherwise we have
3155 * to guess metadata allocation.
3157 if ((ocfs2_xattr_is_local(xe) &&
3158 (value_size >= xi->xi_value_len)) ||
3159 (!ocfs2_xattr_is_local(xe) &&
3160 OCFS2_XATTR_ROOT_SIZE >= xi->xi_value_len))
3161 goto out;
3164 meta_guess:
3165 /* calculate metadata allocation. */
3166 if (di->i_xattr_loc) {
3167 if (!xbs->xattr_bh) {
3168 ret = ocfs2_read_xattr_block(inode,
3169 le64_to_cpu(di->i_xattr_loc),
3170 &bh);
3171 if (ret) {
3172 mlog_errno(ret);
3173 goto out;
3176 xb = (struct ocfs2_xattr_block *)bh->b_data;
3177 } else
3178 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3181 * If there is already an xattr tree, good, we can calculate
3182 * like other b-trees. Otherwise we may have the chance of
3183 * create a tree, the credit calculation is borrowed from
3184 * ocfs2_calc_extend_credits with root_el = NULL. And the
3185 * new tree will be cluster based, so no meta is needed.
3187 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3188 struct ocfs2_extent_list *el =
3189 &xb->xb_attrs.xb_root.xt_list;
3190 meta_add += ocfs2_extend_meta_needed(el);
3191 credits += ocfs2_calc_extend_credits(inode->i_sb,
3192 el);
3193 } else
3194 credits += OCFS2_SUBALLOC_ALLOC + 1;
3197 * This cluster will be used either for new bucket or for
3198 * new xattr block.
3199 * If the cluster size is the same as the bucket size, one
3200 * more is needed since we may need to extend the bucket
3201 * also.
3203 clusters_add += 1;
3204 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3205 if (OCFS2_XATTR_BUCKET_SIZE ==
3206 OCFS2_SB(inode->i_sb)->s_clustersize) {
3207 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3208 clusters_add += 1;
3210 } else {
3211 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
3212 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3213 struct ocfs2_extent_list *el = &def_xv.xv.xr_list;
3214 meta_add += ocfs2_extend_meta_needed(el);
3215 credits += ocfs2_calc_extend_credits(inode->i_sb,
3216 el);
3217 } else {
3218 meta_add += 1;
3221 out:
3222 if (clusters_need)
3223 *clusters_need = clusters_add;
3224 if (meta_need)
3225 *meta_need = meta_add;
3226 if (credits_need)
3227 *credits_need = credits;
3228 brelse(bh);
3229 return ret;
3232 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
3233 struct ocfs2_dinode *di,
3234 struct ocfs2_xattr_info *xi,
3235 struct ocfs2_xattr_search *xis,
3236 struct ocfs2_xattr_search *xbs,
3237 struct ocfs2_xattr_set_ctxt *ctxt,
3238 int extra_meta,
3239 int *credits)
3241 int clusters_add, meta_add, ret;
3242 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3244 memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
3246 ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
3248 ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
3249 &clusters_add, &meta_add, credits);
3250 if (ret) {
3251 mlog_errno(ret);
3252 return ret;
3255 meta_add += extra_meta;
3256 trace_ocfs2_init_xattr_set_ctxt(xi->xi_name, meta_add,
3257 clusters_add, *credits);
3259 if (meta_add) {
3260 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
3261 &ctxt->meta_ac);
3262 if (ret) {
3263 mlog_errno(ret);
3264 goto out;
3268 if (clusters_add) {
3269 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
3270 if (ret)
3271 mlog_errno(ret);
3273 out:
3274 if (ret) {
3275 if (ctxt->meta_ac) {
3276 ocfs2_free_alloc_context(ctxt->meta_ac);
3277 ctxt->meta_ac = NULL;
3281 * We cannot have an error and a non null ctxt->data_ac.
3285 return ret;
3288 static int __ocfs2_xattr_set_handle(struct inode *inode,
3289 struct ocfs2_dinode *di,
3290 struct ocfs2_xattr_info *xi,
3291 struct ocfs2_xattr_search *xis,
3292 struct ocfs2_xattr_search *xbs,
3293 struct ocfs2_xattr_set_ctxt *ctxt)
3295 int ret = 0, credits, old_found;
3297 if (!xi->xi_value) {
3298 /* Remove existing extended attribute */
3299 if (!xis->not_found)
3300 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3301 else if (!xbs->not_found)
3302 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3303 } else {
3304 /* We always try to set extended attribute into inode first*/
3305 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3306 if (!ret && !xbs->not_found) {
3308 * If succeed and that extended attribute existing in
3309 * external block, then we will remove it.
3311 xi->xi_value = NULL;
3312 xi->xi_value_len = 0;
3314 old_found = xis->not_found;
3315 xis->not_found = -ENODATA;
3316 ret = ocfs2_calc_xattr_set_need(inode,
3319 xis,
3320 xbs,
3321 NULL,
3322 NULL,
3323 &credits);
3324 xis->not_found = old_found;
3325 if (ret) {
3326 mlog_errno(ret);
3327 goto out;
3330 ret = ocfs2_extend_trans(ctxt->handle, credits);
3331 if (ret) {
3332 mlog_errno(ret);
3333 goto out;
3335 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3336 } else if ((ret == -ENOSPC) && !ctxt->set_abort) {
3337 if (di->i_xattr_loc && !xbs->xattr_bh) {
3338 ret = ocfs2_xattr_block_find(inode,
3339 xi->xi_name_index,
3340 xi->xi_name, xbs);
3341 if (ret)
3342 goto out;
3344 old_found = xis->not_found;
3345 xis->not_found = -ENODATA;
3346 ret = ocfs2_calc_xattr_set_need(inode,
3349 xis,
3350 xbs,
3351 NULL,
3352 NULL,
3353 &credits);
3354 xis->not_found = old_found;
3355 if (ret) {
3356 mlog_errno(ret);
3357 goto out;
3360 ret = ocfs2_extend_trans(ctxt->handle, credits);
3361 if (ret) {
3362 mlog_errno(ret);
3363 goto out;
3367 * If no space in inode, we will set extended attribute
3368 * into external block.
3370 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3371 if (ret)
3372 goto out;
3373 if (!xis->not_found) {
3375 * If succeed and that extended attribute
3376 * existing in inode, we will remove it.
3378 xi->xi_value = NULL;
3379 xi->xi_value_len = 0;
3380 xbs->not_found = -ENODATA;
3381 ret = ocfs2_calc_xattr_set_need(inode,
3384 xis,
3385 xbs,
3386 NULL,
3387 NULL,
3388 &credits);
3389 if (ret) {
3390 mlog_errno(ret);
3391 goto out;
3394 ret = ocfs2_extend_trans(ctxt->handle, credits);
3395 if (ret) {
3396 mlog_errno(ret);
3397 goto out;
3399 ret = ocfs2_xattr_ibody_set(inode, xi,
3400 xis, ctxt);
3405 if (!ret) {
3406 /* Update inode ctime. */
3407 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
3408 xis->inode_bh,
3409 OCFS2_JOURNAL_ACCESS_WRITE);
3410 if (ret) {
3411 mlog_errno(ret);
3412 goto out;
3415 inode->i_ctime = CURRENT_TIME;
3416 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
3417 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
3418 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
3420 out:
3421 return ret;
3425 * This function only called duing creating inode
3426 * for init security/acl xattrs of the new inode.
3427 * All transanction credits have been reserved in mknod.
3429 int ocfs2_xattr_set_handle(handle_t *handle,
3430 struct inode *inode,
3431 struct buffer_head *di_bh,
3432 int name_index,
3433 const char *name,
3434 const void *value,
3435 size_t value_len,
3436 int flags,
3437 struct ocfs2_alloc_context *meta_ac,
3438 struct ocfs2_alloc_context *data_ac)
3440 struct ocfs2_dinode *di;
3441 int ret;
3443 struct ocfs2_xattr_info xi = {
3444 .xi_name_index = name_index,
3445 .xi_name = name,
3446 .xi_name_len = strlen(name),
3447 .xi_value = value,
3448 .xi_value_len = value_len,
3451 struct ocfs2_xattr_search xis = {
3452 .not_found = -ENODATA,
3455 struct ocfs2_xattr_search xbs = {
3456 .not_found = -ENODATA,
3459 struct ocfs2_xattr_set_ctxt ctxt = {
3460 .handle = handle,
3461 .meta_ac = meta_ac,
3462 .data_ac = data_ac,
3465 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3466 return -EOPNOTSUPP;
3469 * In extreme situation, may need xattr bucket when
3470 * block size is too small. And we have already reserved
3471 * the credits for bucket in mknod.
3473 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
3474 xbs.bucket = ocfs2_xattr_bucket_new(inode);
3475 if (!xbs.bucket) {
3476 mlog_errno(-ENOMEM);
3477 return -ENOMEM;
3481 xis.inode_bh = xbs.inode_bh = di_bh;
3482 di = (struct ocfs2_dinode *)di_bh->b_data;
3484 down_write(&OCFS2_I(inode)->ip_xattr_sem);
3486 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3487 if (ret)
3488 goto cleanup;
3489 if (xis.not_found) {
3490 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3491 if (ret)
3492 goto cleanup;
3495 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3497 cleanup:
3498 up_write(&OCFS2_I(inode)->ip_xattr_sem);
3499 brelse(xbs.xattr_bh);
3500 ocfs2_xattr_bucket_free(xbs.bucket);
3502 return ret;
3506 * ocfs2_xattr_set()
3508 * Set, replace or remove an extended attribute for this inode.
3509 * value is NULL to remove an existing extended attribute, else either
3510 * create or replace an extended attribute.
3512 int ocfs2_xattr_set(struct inode *inode,
3513 int name_index,
3514 const char *name,
3515 const void *value,
3516 size_t value_len,
3517 int flags)
3519 struct buffer_head *di_bh = NULL;
3520 struct ocfs2_dinode *di;
3521 int ret, credits, ref_meta = 0, ref_credits = 0;
3522 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3523 struct inode *tl_inode = osb->osb_tl_inode;
3524 struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, NULL, };
3525 struct ocfs2_refcount_tree *ref_tree = NULL;
3527 struct ocfs2_xattr_info xi = {
3528 .xi_name_index = name_index,
3529 .xi_name = name,
3530 .xi_name_len = strlen(name),
3531 .xi_value = value,
3532 .xi_value_len = value_len,
3535 struct ocfs2_xattr_search xis = {
3536 .not_found = -ENODATA,
3539 struct ocfs2_xattr_search xbs = {
3540 .not_found = -ENODATA,
3543 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3544 return -EOPNOTSUPP;
3547 * Only xbs will be used on indexed trees. xis doesn't need a
3548 * bucket.
3550 xbs.bucket = ocfs2_xattr_bucket_new(inode);
3551 if (!xbs.bucket) {
3552 mlog_errno(-ENOMEM);
3553 return -ENOMEM;
3556 ret = ocfs2_inode_lock(inode, &di_bh, 1);
3557 if (ret < 0) {
3558 mlog_errno(ret);
3559 goto cleanup_nolock;
3561 xis.inode_bh = xbs.inode_bh = di_bh;
3562 di = (struct ocfs2_dinode *)di_bh->b_data;
3564 down_write(&OCFS2_I(inode)->ip_xattr_sem);
3566 * Scan inode and external block to find the same name
3567 * extended attribute and collect search information.
3569 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3570 if (ret)
3571 goto cleanup;
3572 if (xis.not_found) {
3573 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3574 if (ret)
3575 goto cleanup;
3578 if (xis.not_found && xbs.not_found) {
3579 ret = -ENODATA;
3580 if (flags & XATTR_REPLACE)
3581 goto cleanup;
3582 ret = 0;
3583 if (!value)
3584 goto cleanup;
3585 } else {
3586 ret = -EEXIST;
3587 if (flags & XATTR_CREATE)
3588 goto cleanup;
3591 /* Check whether the value is refcounted and do some preparation. */
3592 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL &&
3593 (!xis.not_found || !xbs.not_found)) {
3594 ret = ocfs2_prepare_refcount_xattr(inode, di, &xi,
3595 &xis, &xbs, &ref_tree,
3596 &ref_meta, &ref_credits);
3597 if (ret) {
3598 mlog_errno(ret);
3599 goto cleanup;
3603 mutex_lock(&tl_inode->i_mutex);
3605 if (ocfs2_truncate_log_needs_flush(osb)) {
3606 ret = __ocfs2_flush_truncate_log(osb);
3607 if (ret < 0) {
3608 mutex_unlock(&tl_inode->i_mutex);
3609 mlog_errno(ret);
3610 goto cleanup;
3613 mutex_unlock(&tl_inode->i_mutex);
3615 ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
3616 &xbs, &ctxt, ref_meta, &credits);
3617 if (ret) {
3618 mlog_errno(ret);
3619 goto cleanup;
3622 /* we need to update inode's ctime field, so add credit for it. */
3623 credits += OCFS2_INODE_UPDATE_CREDITS;
3624 ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
3625 if (IS_ERR(ctxt.handle)) {
3626 ret = PTR_ERR(ctxt.handle);
3627 mlog_errno(ret);
3628 goto out_free_ac;
3631 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3632 ocfs2_update_inode_fsync_trans(ctxt.handle, inode, 0);
3634 ocfs2_commit_trans(osb, ctxt.handle);
3636 out_free_ac:
3637 if (ctxt.data_ac)
3638 ocfs2_free_alloc_context(ctxt.data_ac);
3639 if (ctxt.meta_ac)
3640 ocfs2_free_alloc_context(ctxt.meta_ac);
3641 if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
3642 ocfs2_schedule_truncate_log_flush(osb, 1);
3643 ocfs2_run_deallocs(osb, &ctxt.dealloc);
3645 cleanup:
3646 if (ref_tree)
3647 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3648 up_write(&OCFS2_I(inode)->ip_xattr_sem);
3649 if (!value && !ret) {
3650 ret = ocfs2_try_remove_refcount_tree(inode, di_bh);
3651 if (ret)
3652 mlog_errno(ret);
3654 ocfs2_inode_unlock(inode, 1);
3655 cleanup_nolock:
3656 brelse(di_bh);
3657 brelse(xbs.xattr_bh);
3658 ocfs2_xattr_bucket_free(xbs.bucket);
3660 return ret;
3664 * Find the xattr extent rec which may contains name_hash.
3665 * e_cpos will be the first name hash of the xattr rec.
3666 * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
3668 static int ocfs2_xattr_get_rec(struct inode *inode,
3669 u32 name_hash,
3670 u64 *p_blkno,
3671 u32 *e_cpos,
3672 u32 *num_clusters,
3673 struct ocfs2_extent_list *el)
3675 int ret = 0, i;
3676 struct buffer_head *eb_bh = NULL;
3677 struct ocfs2_extent_block *eb;
3678 struct ocfs2_extent_rec *rec = NULL;
3679 u64 e_blkno = 0;
3681 if (el->l_tree_depth) {
3682 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, name_hash,
3683 &eb_bh);
3684 if (ret) {
3685 mlog_errno(ret);
3686 goto out;
3689 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
3690 el = &eb->h_list;
3692 if (el->l_tree_depth) {
3693 ocfs2_error(inode->i_sb,
3694 "Inode %lu has non zero tree depth in "
3695 "xattr tree block %llu\n", inode->i_ino,
3696 (unsigned long long)eb_bh->b_blocknr);
3697 ret = -EROFS;
3698 goto out;
3702 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
3703 rec = &el->l_recs[i];
3705 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
3706 e_blkno = le64_to_cpu(rec->e_blkno);
3707 break;
3711 if (!e_blkno) {
3712 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
3713 "record (%u, %u, 0) in xattr", inode->i_ino,
3714 le32_to_cpu(rec->e_cpos),
3715 ocfs2_rec_clusters(el, rec));
3716 ret = -EROFS;
3717 goto out;
3720 *p_blkno = le64_to_cpu(rec->e_blkno);
3721 *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
3722 if (e_cpos)
3723 *e_cpos = le32_to_cpu(rec->e_cpos);
3724 out:
3725 brelse(eb_bh);
3726 return ret;
3729 typedef int (xattr_bucket_func)(struct inode *inode,
3730 struct ocfs2_xattr_bucket *bucket,
3731 void *para);
3733 static int ocfs2_find_xe_in_bucket(struct inode *inode,
3734 struct ocfs2_xattr_bucket *bucket,
3735 int name_index,
3736 const char *name,
3737 u32 name_hash,
3738 u16 *xe_index,
3739 int *found)
3741 int i, ret = 0, cmp = 1, block_off, new_offset;
3742 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3743 size_t name_len = strlen(name);
3744 struct ocfs2_xattr_entry *xe = NULL;
3745 char *xe_name;
3748 * We don't use binary search in the bucket because there
3749 * may be multiple entries with the same name hash.
3751 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3752 xe = &xh->xh_entries[i];
3754 if (name_hash > le32_to_cpu(xe->xe_name_hash))
3755 continue;
3756 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
3757 break;
3759 cmp = name_index - ocfs2_xattr_get_type(xe);
3760 if (!cmp)
3761 cmp = name_len - xe->xe_name_len;
3762 if (cmp)
3763 continue;
3765 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3768 &block_off,
3769 &new_offset);
3770 if (ret) {
3771 mlog_errno(ret);
3772 break;
3776 xe_name = bucket_block(bucket, block_off) + new_offset;
3777 if (!memcmp(name, xe_name, name_len)) {
3778 *xe_index = i;
3779 *found = 1;
3780 ret = 0;
3781 break;
3785 return ret;
3789 * Find the specified xattr entry in a series of buckets.
3790 * This series start from p_blkno and last for num_clusters.
3791 * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
3792 * the num of the valid buckets.
3794 * Return the buffer_head this xattr should reside in. And if the xattr's
3795 * hash is in the gap of 2 buckets, return the lower bucket.
3797 static int ocfs2_xattr_bucket_find(struct inode *inode,
3798 int name_index,
3799 const char *name,
3800 u32 name_hash,
3801 u64 p_blkno,
3802 u32 first_hash,
3803 u32 num_clusters,
3804 struct ocfs2_xattr_search *xs)
3806 int ret, found = 0;
3807 struct ocfs2_xattr_header *xh = NULL;
3808 struct ocfs2_xattr_entry *xe = NULL;
3809 u16 index = 0;
3810 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3811 int low_bucket = 0, bucket, high_bucket;
3812 struct ocfs2_xattr_bucket *search;
3813 u32 last_hash;
3814 u64 blkno, lower_blkno = 0;
3816 search = ocfs2_xattr_bucket_new(inode);
3817 if (!search) {
3818 ret = -ENOMEM;
3819 mlog_errno(ret);
3820 goto out;
3823 ret = ocfs2_read_xattr_bucket(search, p_blkno);
3824 if (ret) {
3825 mlog_errno(ret);
3826 goto out;
3829 xh = bucket_xh(search);
3830 high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
3831 while (low_bucket <= high_bucket) {
3832 ocfs2_xattr_bucket_relse(search);
3834 bucket = (low_bucket + high_bucket) / 2;
3835 blkno = p_blkno + bucket * blk_per_bucket;
3836 ret = ocfs2_read_xattr_bucket(search, blkno);
3837 if (ret) {
3838 mlog_errno(ret);
3839 goto out;
3842 xh = bucket_xh(search);
3843 xe = &xh->xh_entries[0];
3844 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3845 high_bucket = bucket - 1;
3846 continue;
3850 * Check whether the hash of the last entry in our
3851 * bucket is larger than the search one. for an empty
3852 * bucket, the last one is also the first one.
3854 if (xh->xh_count)
3855 xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3857 last_hash = le32_to_cpu(xe->xe_name_hash);
3859 /* record lower_blkno which may be the insert place. */
3860 lower_blkno = blkno;
3862 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3863 low_bucket = bucket + 1;
3864 continue;
3867 /* the searched xattr should reside in this bucket if exists. */
3868 ret = ocfs2_find_xe_in_bucket(inode, search,
3869 name_index, name, name_hash,
3870 &index, &found);
3871 if (ret) {
3872 mlog_errno(ret);
3873 goto out;
3875 break;
3879 * Record the bucket we have found.
3880 * When the xattr's hash value is in the gap of 2 buckets, we will
3881 * always set it to the previous bucket.
3883 if (!lower_blkno)
3884 lower_blkno = p_blkno;
3886 /* This should be in cache - we just read it during the search */
3887 ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3888 if (ret) {
3889 mlog_errno(ret);
3890 goto out;
3893 xs->header = bucket_xh(xs->bucket);
3894 xs->base = bucket_block(xs->bucket, 0);
3895 xs->end = xs->base + inode->i_sb->s_blocksize;
3897 if (found) {
3898 xs->here = &xs->header->xh_entries[index];
3899 trace_ocfs2_xattr_bucket_find(OCFS2_I(inode)->ip_blkno,
3900 name, name_index, name_hash,
3901 (unsigned long long)bucket_blkno(xs->bucket),
3902 index);
3903 } else
3904 ret = -ENODATA;
3906 out:
3907 ocfs2_xattr_bucket_free(search);
3908 return ret;
3911 static int ocfs2_xattr_index_block_find(struct inode *inode,
3912 struct buffer_head *root_bh,
3913 int name_index,
3914 const char *name,
3915 struct ocfs2_xattr_search *xs)
3917 int ret;
3918 struct ocfs2_xattr_block *xb =
3919 (struct ocfs2_xattr_block *)root_bh->b_data;
3920 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3921 struct ocfs2_extent_list *el = &xb_root->xt_list;
3922 u64 p_blkno = 0;
3923 u32 first_hash, num_clusters = 0;
3924 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3926 if (le16_to_cpu(el->l_next_free_rec) == 0)
3927 return -ENODATA;
3929 trace_ocfs2_xattr_index_block_find(OCFS2_I(inode)->ip_blkno,
3930 name, name_index, name_hash,
3931 (unsigned long long)root_bh->b_blocknr,
3932 -1);
3934 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3935 &num_clusters, el);
3936 if (ret) {
3937 mlog_errno(ret);
3938 goto out;
3941 BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3943 trace_ocfs2_xattr_index_block_find_rec(OCFS2_I(inode)->ip_blkno,
3944 name, name_index, first_hash,
3945 (unsigned long long)p_blkno,
3946 num_clusters);
3948 ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3949 p_blkno, first_hash, num_clusters, xs);
3951 out:
3952 return ret;
3955 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3956 u64 blkno,
3957 u32 clusters,
3958 xattr_bucket_func *func,
3959 void *para)
3961 int i, ret = 0;
3962 u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3963 u32 num_buckets = clusters * bpc;
3964 struct ocfs2_xattr_bucket *bucket;
3966 bucket = ocfs2_xattr_bucket_new(inode);
3967 if (!bucket) {
3968 mlog_errno(-ENOMEM);
3969 return -ENOMEM;
3972 trace_ocfs2_iterate_xattr_buckets(
3973 (unsigned long long)OCFS2_I(inode)->ip_blkno,
3974 (unsigned long long)blkno, clusters);
3976 for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3977 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3978 if (ret) {
3979 mlog_errno(ret);
3980 break;
3984 * The real bucket num in this series of blocks is stored
3985 * in the 1st bucket.
3987 if (i == 0)
3988 num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3990 trace_ocfs2_iterate_xattr_bucket((unsigned long long)blkno,
3991 le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3992 if (func) {
3993 ret = func(inode, bucket, para);
3994 if (ret && ret != -ERANGE)
3995 mlog_errno(ret);
3996 /* Fall through to bucket_relse() */
3999 ocfs2_xattr_bucket_relse(bucket);
4000 if (ret)
4001 break;
4004 ocfs2_xattr_bucket_free(bucket);
4005 return ret;
4008 struct ocfs2_xattr_tree_list {
4009 char *buffer;
4010 size_t buffer_size;
4011 size_t result;
4014 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
4015 struct ocfs2_xattr_header *xh,
4016 int index,
4017 int *block_off,
4018 int *new_offset)
4020 u16 name_offset;
4022 if (index < 0 || index >= le16_to_cpu(xh->xh_count))
4023 return -EINVAL;
4025 name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
4027 *block_off = name_offset >> sb->s_blocksize_bits;
4028 *new_offset = name_offset % sb->s_blocksize;
4030 return 0;
4033 static int ocfs2_list_xattr_bucket(struct inode *inode,
4034 struct ocfs2_xattr_bucket *bucket,
4035 void *para)
4037 int ret = 0, type;
4038 struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
4039 int i, block_off, new_offset;
4040 const char *prefix, *name;
4042 for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
4043 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
4044 type = ocfs2_xattr_get_type(entry);
4045 prefix = ocfs2_xattr_prefix(type);
4047 if (prefix) {
4048 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
4049 bucket_xh(bucket),
4051 &block_off,
4052 &new_offset);
4053 if (ret)
4054 break;
4056 name = (const char *)bucket_block(bucket, block_off) +
4057 new_offset;
4058 ret = ocfs2_xattr_list_entry(xl->buffer,
4059 xl->buffer_size,
4060 &xl->result,
4061 prefix, name,
4062 entry->xe_name_len);
4063 if (ret)
4064 break;
4068 return ret;
4071 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
4072 struct buffer_head *blk_bh,
4073 xattr_tree_rec_func *rec_func,
4074 void *para)
4076 struct ocfs2_xattr_block *xb =
4077 (struct ocfs2_xattr_block *)blk_bh->b_data;
4078 struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4079 int ret = 0;
4080 u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
4081 u64 p_blkno = 0;
4083 if (!el->l_next_free_rec || !rec_func)
4084 return 0;
4086 while (name_hash > 0) {
4087 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4088 &e_cpos, &num_clusters, el);
4089 if (ret) {
4090 mlog_errno(ret);
4091 break;
4094 ret = rec_func(inode, blk_bh, p_blkno, e_cpos,
4095 num_clusters, para);
4096 if (ret) {
4097 if (ret != -ERANGE)
4098 mlog_errno(ret);
4099 break;
4102 if (e_cpos == 0)
4103 break;
4105 name_hash = e_cpos - 1;
4108 return ret;
4112 static int ocfs2_list_xattr_tree_rec(struct inode *inode,
4113 struct buffer_head *root_bh,
4114 u64 blkno, u32 cpos, u32 len, void *para)
4116 return ocfs2_iterate_xattr_buckets(inode, blkno, len,
4117 ocfs2_list_xattr_bucket, para);
4120 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
4121 struct buffer_head *blk_bh,
4122 char *buffer,
4123 size_t buffer_size)
4125 int ret;
4126 struct ocfs2_xattr_tree_list xl = {
4127 .buffer = buffer,
4128 .buffer_size = buffer_size,
4129 .result = 0,
4132 ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
4133 ocfs2_list_xattr_tree_rec, &xl);
4134 if (ret) {
4135 mlog_errno(ret);
4136 goto out;
4139 ret = xl.result;
4140 out:
4141 return ret;
4144 static int cmp_xe(const void *a, const void *b)
4146 const struct ocfs2_xattr_entry *l = a, *r = b;
4147 u32 l_hash = le32_to_cpu(l->xe_name_hash);
4148 u32 r_hash = le32_to_cpu(r->xe_name_hash);
4150 if (l_hash > r_hash)
4151 return 1;
4152 if (l_hash < r_hash)
4153 return -1;
4154 return 0;
4157 static void swap_xe(void *a, void *b, int size)
4159 struct ocfs2_xattr_entry *l = a, *r = b, tmp;
4161 tmp = *l;
4162 memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
4163 memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
4167 * When the ocfs2_xattr_block is filled up, new bucket will be created
4168 * and all the xattr entries will be moved to the new bucket.
4169 * The header goes at the start of the bucket, and the names+values are
4170 * filled from the end. This is why *target starts as the last buffer.
4171 * Note: we need to sort the entries since they are not saved in order
4172 * in the ocfs2_xattr_block.
4174 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
4175 struct buffer_head *xb_bh,
4176 struct ocfs2_xattr_bucket *bucket)
4178 int i, blocksize = inode->i_sb->s_blocksize;
4179 int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4180 u16 offset, size, off_change;
4181 struct ocfs2_xattr_entry *xe;
4182 struct ocfs2_xattr_block *xb =
4183 (struct ocfs2_xattr_block *)xb_bh->b_data;
4184 struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
4185 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4186 u16 count = le16_to_cpu(xb_xh->xh_count);
4187 char *src = xb_bh->b_data;
4188 char *target = bucket_block(bucket, blks - 1);
4190 trace_ocfs2_cp_xattr_block_to_bucket_begin(
4191 (unsigned long long)xb_bh->b_blocknr,
4192 (unsigned long long)bucket_blkno(bucket));
4194 for (i = 0; i < blks; i++)
4195 memset(bucket_block(bucket, i), 0, blocksize);
4198 * Since the xe_name_offset is based on ocfs2_xattr_header,
4199 * there is a offset change corresponding to the change of
4200 * ocfs2_xattr_header's position.
4202 off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4203 xe = &xb_xh->xh_entries[count - 1];
4204 offset = le16_to_cpu(xe->xe_name_offset) + off_change;
4205 size = blocksize - offset;
4207 /* copy all the names and values. */
4208 memcpy(target + offset, src + offset, size);
4210 /* Init new header now. */
4211 xh->xh_count = xb_xh->xh_count;
4212 xh->xh_num_buckets = cpu_to_le16(1);
4213 xh->xh_name_value_len = cpu_to_le16(size);
4214 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
4216 /* copy all the entries. */
4217 target = bucket_block(bucket, 0);
4218 offset = offsetof(struct ocfs2_xattr_header, xh_entries);
4219 size = count * sizeof(struct ocfs2_xattr_entry);
4220 memcpy(target + offset, (char *)xb_xh + offset, size);
4222 /* Change the xe offset for all the xe because of the move. */
4223 off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
4224 offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4225 for (i = 0; i < count; i++)
4226 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
4228 trace_ocfs2_cp_xattr_block_to_bucket_end(offset, size, off_change);
4230 sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
4231 cmp_xe, swap_xe);
4235 * After we move xattr from block to index btree, we have to
4236 * update ocfs2_xattr_search to the new xe and base.
4238 * When the entry is in xattr block, xattr_bh indicates the storage place.
4239 * While if the entry is in index b-tree, "bucket" indicates the
4240 * real place of the xattr.
4242 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
4243 struct ocfs2_xattr_search *xs,
4244 struct buffer_head *old_bh)
4246 char *buf = old_bh->b_data;
4247 struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
4248 struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
4249 int i;
4251 xs->header = bucket_xh(xs->bucket);
4252 xs->base = bucket_block(xs->bucket, 0);
4253 xs->end = xs->base + inode->i_sb->s_blocksize;
4255 if (xs->not_found)
4256 return;
4258 i = xs->here - old_xh->xh_entries;
4259 xs->here = &xs->header->xh_entries[i];
4262 static int ocfs2_xattr_create_index_block(struct inode *inode,
4263 struct ocfs2_xattr_search *xs,
4264 struct ocfs2_xattr_set_ctxt *ctxt)
4266 int ret;
4267 u32 bit_off, len;
4268 u64 blkno;
4269 handle_t *handle = ctxt->handle;
4270 struct ocfs2_inode_info *oi = OCFS2_I(inode);
4271 struct buffer_head *xb_bh = xs->xattr_bh;
4272 struct ocfs2_xattr_block *xb =
4273 (struct ocfs2_xattr_block *)xb_bh->b_data;
4274 struct ocfs2_xattr_tree_root *xr;
4275 u16 xb_flags = le16_to_cpu(xb->xb_flags);
4277 trace_ocfs2_xattr_create_index_block_begin(
4278 (unsigned long long)xb_bh->b_blocknr);
4280 BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
4281 BUG_ON(!xs->bucket);
4284 * XXX:
4285 * We can use this lock for now, and maybe move to a dedicated mutex
4286 * if performance becomes a problem later.
4288 down_write(&oi->ip_alloc_sem);
4290 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), xb_bh,
4291 OCFS2_JOURNAL_ACCESS_WRITE);
4292 if (ret) {
4293 mlog_errno(ret);
4294 goto out;
4297 ret = __ocfs2_claim_clusters(handle, ctxt->data_ac,
4298 1, 1, &bit_off, &len);
4299 if (ret) {
4300 mlog_errno(ret);
4301 goto out;
4305 * The bucket may spread in many blocks, and
4306 * we will only touch the 1st block and the last block
4307 * in the whole bucket(one for entry and one for data).
4309 blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
4311 trace_ocfs2_xattr_create_index_block((unsigned long long)blkno);
4313 ret = ocfs2_init_xattr_bucket(xs->bucket, blkno, 1);
4314 if (ret) {
4315 mlog_errno(ret);
4316 goto out;
4319 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4320 OCFS2_JOURNAL_ACCESS_CREATE);
4321 if (ret) {
4322 mlog_errno(ret);
4323 goto out;
4326 ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
4327 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4329 ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
4331 /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
4332 memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
4333 offsetof(struct ocfs2_xattr_block, xb_attrs));
4335 xr = &xb->xb_attrs.xb_root;
4336 xr->xt_clusters = cpu_to_le32(1);
4337 xr->xt_last_eb_blk = 0;
4338 xr->xt_list.l_tree_depth = 0;
4339 xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
4340 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
4342 xr->xt_list.l_recs[0].e_cpos = 0;
4343 xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
4344 xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
4346 xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
4348 ocfs2_journal_dirty(handle, xb_bh);
4350 out:
4351 up_write(&oi->ip_alloc_sem);
4353 return ret;
4356 static int cmp_xe_offset(const void *a, const void *b)
4358 const struct ocfs2_xattr_entry *l = a, *r = b;
4359 u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
4360 u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
4362 if (l_name_offset < r_name_offset)
4363 return 1;
4364 if (l_name_offset > r_name_offset)
4365 return -1;
4366 return 0;
4370 * defrag a xattr bucket if we find that the bucket has some
4371 * holes beteen name/value pairs.
4372 * We will move all the name/value pairs to the end of the bucket
4373 * so that we can spare some space for insertion.
4375 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
4376 handle_t *handle,
4377 struct ocfs2_xattr_bucket *bucket)
4379 int ret, i;
4380 size_t end, offset, len;
4381 struct ocfs2_xattr_header *xh;
4382 char *entries, *buf, *bucket_buf = NULL;
4383 u64 blkno = bucket_blkno(bucket);
4384 u16 xh_free_start;
4385 size_t blocksize = inode->i_sb->s_blocksize;
4386 struct ocfs2_xattr_entry *xe;
4389 * In order to make the operation more efficient and generic,
4390 * we copy all the blocks into a contiguous memory and do the
4391 * defragment there, so if anything is error, we will not touch
4392 * the real block.
4394 bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
4395 if (!bucket_buf) {
4396 ret = -EIO;
4397 goto out;
4400 buf = bucket_buf;
4401 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4402 memcpy(buf, bucket_block(bucket, i), blocksize);
4404 ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
4405 OCFS2_JOURNAL_ACCESS_WRITE);
4406 if (ret < 0) {
4407 mlog_errno(ret);
4408 goto out;
4411 xh = (struct ocfs2_xattr_header *)bucket_buf;
4412 entries = (char *)xh->xh_entries;
4413 xh_free_start = le16_to_cpu(xh->xh_free_start);
4415 trace_ocfs2_defrag_xattr_bucket(
4416 (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
4417 xh_free_start, le16_to_cpu(xh->xh_name_value_len));
4420 * sort all the entries by their offset.
4421 * the largest will be the first, so that we can
4422 * move them to the end one by one.
4424 sort(entries, le16_to_cpu(xh->xh_count),
4425 sizeof(struct ocfs2_xattr_entry),
4426 cmp_xe_offset, swap_xe);
4428 /* Move all name/values to the end of the bucket. */
4429 xe = xh->xh_entries;
4430 end = OCFS2_XATTR_BUCKET_SIZE;
4431 for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
4432 offset = le16_to_cpu(xe->xe_name_offset);
4433 len = namevalue_size_xe(xe);
4436 * We must make sure that the name/value pair
4437 * exist in the same block. So adjust end to
4438 * the previous block end if needed.
4440 if (((end - len) / blocksize !=
4441 (end - 1) / blocksize))
4442 end = end - end % blocksize;
4444 if (end > offset + len) {
4445 memmove(bucket_buf + end - len,
4446 bucket_buf + offset, len);
4447 xe->xe_name_offset = cpu_to_le16(end - len);
4450 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
4451 "bucket %llu\n", (unsigned long long)blkno);
4453 end -= len;
4456 mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
4457 "bucket %llu\n", (unsigned long long)blkno);
4459 if (xh_free_start == end)
4460 goto out;
4462 memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
4463 xh->xh_free_start = cpu_to_le16(end);
4465 /* sort the entries by their name_hash. */
4466 sort(entries, le16_to_cpu(xh->xh_count),
4467 sizeof(struct ocfs2_xattr_entry),
4468 cmp_xe, swap_xe);
4470 buf = bucket_buf;
4471 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4472 memcpy(bucket_block(bucket, i), buf, blocksize);
4473 ocfs2_xattr_bucket_journal_dirty(handle, bucket);
4475 out:
4476 kfree(bucket_buf);
4477 return ret;
4481 * prev_blkno points to the start of an existing extent. new_blkno
4482 * points to a newly allocated extent. Because we know each of our
4483 * clusters contains more than bucket, we can easily split one cluster
4484 * at a bucket boundary. So we take the last cluster of the existing
4485 * extent and split it down the middle. We move the last half of the
4486 * buckets in the last cluster of the existing extent over to the new
4487 * extent.
4489 * first_bh is the buffer at prev_blkno so we can update the existing
4490 * extent's bucket count. header_bh is the bucket were we were hoping
4491 * to insert our xattr. If the bucket move places the target in the new
4492 * extent, we'll update first_bh and header_bh after modifying the old
4493 * extent.
4495 * first_hash will be set as the 1st xe's name_hash in the new extent.
4497 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
4498 handle_t *handle,
4499 struct ocfs2_xattr_bucket *first,
4500 struct ocfs2_xattr_bucket *target,
4501 u64 new_blkno,
4502 u32 num_clusters,
4503 u32 *first_hash)
4505 int ret;
4506 struct super_block *sb = inode->i_sb;
4507 int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
4508 int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
4509 int to_move = num_buckets / 2;
4510 u64 src_blkno;
4511 u64 last_cluster_blkno = bucket_blkno(first) +
4512 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
4514 BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
4515 BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
4517 trace_ocfs2_mv_xattr_bucket_cross_cluster(
4518 (unsigned long long)last_cluster_blkno,
4519 (unsigned long long)new_blkno);
4521 ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
4522 last_cluster_blkno, new_blkno,
4523 to_move, first_hash);
4524 if (ret) {
4525 mlog_errno(ret);
4526 goto out;
4529 /* This is the first bucket that got moved */
4530 src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
4533 * If the target bucket was part of the moved buckets, we need to
4534 * update first and target.
4536 if (bucket_blkno(target) >= src_blkno) {
4537 /* Find the block for the new target bucket */
4538 src_blkno = new_blkno +
4539 (bucket_blkno(target) - src_blkno);
4541 ocfs2_xattr_bucket_relse(first);
4542 ocfs2_xattr_bucket_relse(target);
4545 * These shouldn't fail - the buffers are in the
4546 * journal from ocfs2_cp_xattr_bucket().
4548 ret = ocfs2_read_xattr_bucket(first, new_blkno);
4549 if (ret) {
4550 mlog_errno(ret);
4551 goto out;
4553 ret = ocfs2_read_xattr_bucket(target, src_blkno);
4554 if (ret)
4555 mlog_errno(ret);
4559 out:
4560 return ret;
4564 * Find the suitable pos when we divide a bucket into 2.
4565 * We have to make sure the xattrs with the same hash value exist
4566 * in the same bucket.
4568 * If this ocfs2_xattr_header covers more than one hash value, find a
4569 * place where the hash value changes. Try to find the most even split.
4570 * The most common case is that all entries have different hash values,
4571 * and the first check we make will find a place to split.
4573 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
4575 struct ocfs2_xattr_entry *entries = xh->xh_entries;
4576 int count = le16_to_cpu(xh->xh_count);
4577 int delta, middle = count / 2;
4580 * We start at the middle. Each step gets farther away in both
4581 * directions. We therefore hit the change in hash value
4582 * nearest to the middle. Note that this loop does not execute for
4583 * count < 2.
4585 for (delta = 0; delta < middle; delta++) {
4586 /* Let's check delta earlier than middle */
4587 if (cmp_xe(&entries[middle - delta - 1],
4588 &entries[middle - delta]))
4589 return middle - delta;
4591 /* For even counts, don't walk off the end */
4592 if ((middle + delta + 1) == count)
4593 continue;
4595 /* Now try delta past middle */
4596 if (cmp_xe(&entries[middle + delta],
4597 &entries[middle + delta + 1]))
4598 return middle + delta + 1;
4601 /* Every entry had the same hash */
4602 return count;
4606 * Move some xattrs in old bucket(blk) to new bucket(new_blk).
4607 * first_hash will record the 1st hash of the new bucket.
4609 * Normally half of the xattrs will be moved. But we have to make
4610 * sure that the xattrs with the same hash value are stored in the
4611 * same bucket. If all the xattrs in this bucket have the same hash
4612 * value, the new bucket will be initialized as an empty one and the
4613 * first_hash will be initialized as (hash_value+1).
4615 static int ocfs2_divide_xattr_bucket(struct inode *inode,
4616 handle_t *handle,
4617 u64 blk,
4618 u64 new_blk,
4619 u32 *first_hash,
4620 int new_bucket_head)
4622 int ret, i;
4623 int count, start, len, name_value_len = 0, name_offset = 0;
4624 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4625 struct ocfs2_xattr_header *xh;
4626 struct ocfs2_xattr_entry *xe;
4627 int blocksize = inode->i_sb->s_blocksize;
4629 trace_ocfs2_divide_xattr_bucket_begin((unsigned long long)blk,
4630 (unsigned long long)new_blk);
4632 s_bucket = ocfs2_xattr_bucket_new(inode);
4633 t_bucket = ocfs2_xattr_bucket_new(inode);
4634 if (!s_bucket || !t_bucket) {
4635 ret = -ENOMEM;
4636 mlog_errno(ret);
4637 goto out;
4640 ret = ocfs2_read_xattr_bucket(s_bucket, blk);
4641 if (ret) {
4642 mlog_errno(ret);
4643 goto out;
4646 ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
4647 OCFS2_JOURNAL_ACCESS_WRITE);
4648 if (ret) {
4649 mlog_errno(ret);
4650 goto out;
4654 * Even if !new_bucket_head, we're overwriting t_bucket. Thus,
4655 * there's no need to read it.
4657 ret = ocfs2_init_xattr_bucket(t_bucket, new_blk, new_bucket_head);
4658 if (ret) {
4659 mlog_errno(ret);
4660 goto out;
4664 * Hey, if we're overwriting t_bucket, what difference does
4665 * ACCESS_CREATE vs ACCESS_WRITE make? See the comment in the
4666 * same part of ocfs2_cp_xattr_bucket().
4668 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4669 new_bucket_head ?
4670 OCFS2_JOURNAL_ACCESS_CREATE :
4671 OCFS2_JOURNAL_ACCESS_WRITE);
4672 if (ret) {
4673 mlog_errno(ret);
4674 goto out;
4677 xh = bucket_xh(s_bucket);
4678 count = le16_to_cpu(xh->xh_count);
4679 start = ocfs2_xattr_find_divide_pos(xh);
4681 if (start == count) {
4682 xe = &xh->xh_entries[start-1];
4685 * initialized a new empty bucket here.
4686 * The hash value is set as one larger than
4687 * that of the last entry in the previous bucket.
4689 for (i = 0; i < t_bucket->bu_blocks; i++)
4690 memset(bucket_block(t_bucket, i), 0, blocksize);
4692 xh = bucket_xh(t_bucket);
4693 xh->xh_free_start = cpu_to_le16(blocksize);
4694 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
4695 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
4697 goto set_num_buckets;
4700 /* copy the whole bucket to the new first. */
4701 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4703 /* update the new bucket. */
4704 xh = bucket_xh(t_bucket);
4707 * Calculate the total name/value len and xh_free_start for
4708 * the old bucket first.
4710 name_offset = OCFS2_XATTR_BUCKET_SIZE;
4711 name_value_len = 0;
4712 for (i = 0; i < start; i++) {
4713 xe = &xh->xh_entries[i];
4714 name_value_len += namevalue_size_xe(xe);
4715 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
4716 name_offset = le16_to_cpu(xe->xe_name_offset);
4720 * Now begin the modification to the new bucket.
4722 * In the new bucket, We just move the xattr entry to the beginning
4723 * and don't touch the name/value. So there will be some holes in the
4724 * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
4725 * called.
4727 xe = &xh->xh_entries[start];
4728 len = sizeof(struct ocfs2_xattr_entry) * (count - start);
4729 trace_ocfs2_divide_xattr_bucket_move(len,
4730 (int)((char *)xe - (char *)xh),
4731 (int)((char *)xh->xh_entries - (char *)xh));
4732 memmove((char *)xh->xh_entries, (char *)xe, len);
4733 xe = &xh->xh_entries[count - start];
4734 len = sizeof(struct ocfs2_xattr_entry) * start;
4735 memset((char *)xe, 0, len);
4737 le16_add_cpu(&xh->xh_count, -start);
4738 le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
4740 /* Calculate xh_free_start for the new bucket. */
4741 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4742 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4743 xe = &xh->xh_entries[i];
4744 if (le16_to_cpu(xe->xe_name_offset) <
4745 le16_to_cpu(xh->xh_free_start))
4746 xh->xh_free_start = xe->xe_name_offset;
4749 set_num_buckets:
4750 /* set xh->xh_num_buckets for the new xh. */
4751 if (new_bucket_head)
4752 xh->xh_num_buckets = cpu_to_le16(1);
4753 else
4754 xh->xh_num_buckets = 0;
4756 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4758 /* store the first_hash of the new bucket. */
4759 if (first_hash)
4760 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
4763 * Now only update the 1st block of the old bucket. If we
4764 * just added a new empty bucket, there is no need to modify
4765 * it.
4767 if (start == count)
4768 goto out;
4770 xh = bucket_xh(s_bucket);
4771 memset(&xh->xh_entries[start], 0,
4772 sizeof(struct ocfs2_xattr_entry) * (count - start));
4773 xh->xh_count = cpu_to_le16(start);
4774 xh->xh_free_start = cpu_to_le16(name_offset);
4775 xh->xh_name_value_len = cpu_to_le16(name_value_len);
4777 ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
4779 out:
4780 ocfs2_xattr_bucket_free(s_bucket);
4781 ocfs2_xattr_bucket_free(t_bucket);
4783 return ret;
4787 * Copy xattr from one bucket to another bucket.
4789 * The caller must make sure that the journal transaction
4790 * has enough space for journaling.
4792 static int ocfs2_cp_xattr_bucket(struct inode *inode,
4793 handle_t *handle,
4794 u64 s_blkno,
4795 u64 t_blkno,
4796 int t_is_new)
4798 int ret;
4799 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4801 BUG_ON(s_blkno == t_blkno);
4803 trace_ocfs2_cp_xattr_bucket((unsigned long long)s_blkno,
4804 (unsigned long long)t_blkno,
4805 t_is_new);
4807 s_bucket = ocfs2_xattr_bucket_new(inode);
4808 t_bucket = ocfs2_xattr_bucket_new(inode);
4809 if (!s_bucket || !t_bucket) {
4810 ret = -ENOMEM;
4811 mlog_errno(ret);
4812 goto out;
4815 ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
4816 if (ret)
4817 goto out;
4820 * Even if !t_is_new, we're overwriting t_bucket. Thus,
4821 * there's no need to read it.
4823 ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno, t_is_new);
4824 if (ret)
4825 goto out;
4828 * Hey, if we're overwriting t_bucket, what difference does
4829 * ACCESS_CREATE vs ACCESS_WRITE make? Well, if we allocated a new
4830 * cluster to fill, we came here from
4831 * ocfs2_mv_xattr_buckets(), and it is really new -
4832 * ACCESS_CREATE is required. But we also might have moved data
4833 * out of t_bucket before extending back into it.
4834 * ocfs2_add_new_xattr_bucket() can do this - its call to
4835 * ocfs2_add_new_xattr_cluster() may have created a new extent
4836 * and copied out the end of the old extent. Then it re-extends
4837 * the old extent back to create space for new xattrs. That's
4838 * how we get here, and the bucket isn't really new.
4840 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4841 t_is_new ?
4842 OCFS2_JOURNAL_ACCESS_CREATE :
4843 OCFS2_JOURNAL_ACCESS_WRITE);
4844 if (ret)
4845 goto out;
4847 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4848 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4850 out:
4851 ocfs2_xattr_bucket_free(t_bucket);
4852 ocfs2_xattr_bucket_free(s_bucket);
4854 return ret;
4858 * src_blk points to the start of an existing extent. last_blk points to
4859 * last cluster in that extent. to_blk points to a newly allocated
4860 * extent. We copy the buckets from the cluster at last_blk to the new
4861 * extent. If start_bucket is non-zero, we skip that many buckets before
4862 * we start copying. The new extent's xh_num_buckets gets set to the
4863 * number of buckets we copied. The old extent's xh_num_buckets shrinks
4864 * by the same amount.
4866 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4867 u64 src_blk, u64 last_blk, u64 to_blk,
4868 unsigned int start_bucket,
4869 u32 *first_hash)
4871 int i, ret, credits;
4872 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4873 int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4874 int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4875 struct ocfs2_xattr_bucket *old_first, *new_first;
4877 trace_ocfs2_mv_xattr_buckets((unsigned long long)last_blk,
4878 (unsigned long long)to_blk);
4880 BUG_ON(start_bucket >= num_buckets);
4881 if (start_bucket) {
4882 num_buckets -= start_bucket;
4883 last_blk += (start_bucket * blks_per_bucket);
4886 /* The first bucket of the original extent */
4887 old_first = ocfs2_xattr_bucket_new(inode);
4888 /* The first bucket of the new extent */
4889 new_first = ocfs2_xattr_bucket_new(inode);
4890 if (!old_first || !new_first) {
4891 ret = -ENOMEM;
4892 mlog_errno(ret);
4893 goto out;
4896 ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4897 if (ret) {
4898 mlog_errno(ret);
4899 goto out;
4903 * We need to update the first bucket of the old extent and all
4904 * the buckets going to the new extent.
4906 credits = ((num_buckets + 1) * blks_per_bucket);
4907 ret = ocfs2_extend_trans(handle, credits);
4908 if (ret) {
4909 mlog_errno(ret);
4910 goto out;
4913 ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4914 OCFS2_JOURNAL_ACCESS_WRITE);
4915 if (ret) {
4916 mlog_errno(ret);
4917 goto out;
4920 for (i = 0; i < num_buckets; i++) {
4921 ret = ocfs2_cp_xattr_bucket(inode, handle,
4922 last_blk + (i * blks_per_bucket),
4923 to_blk + (i * blks_per_bucket),
4925 if (ret) {
4926 mlog_errno(ret);
4927 goto out;
4932 * Get the new bucket ready before we dirty anything
4933 * (This actually shouldn't fail, because we already dirtied
4934 * it once in ocfs2_cp_xattr_bucket()).
4936 ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4937 if (ret) {
4938 mlog_errno(ret);
4939 goto out;
4941 ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4942 OCFS2_JOURNAL_ACCESS_WRITE);
4943 if (ret) {
4944 mlog_errno(ret);
4945 goto out;
4948 /* Now update the headers */
4949 le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4950 ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4952 bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4953 ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4955 if (first_hash)
4956 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4958 out:
4959 ocfs2_xattr_bucket_free(new_first);
4960 ocfs2_xattr_bucket_free(old_first);
4961 return ret;
4965 * Move some xattrs in this cluster to the new cluster.
4966 * This function should only be called when bucket size == cluster size.
4967 * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4969 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4970 handle_t *handle,
4971 u64 prev_blk,
4972 u64 new_blk,
4973 u32 *first_hash)
4975 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4976 int ret, credits = 2 * blk_per_bucket;
4978 BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4980 ret = ocfs2_extend_trans(handle, credits);
4981 if (ret) {
4982 mlog_errno(ret);
4983 return ret;
4986 /* Move half of the xattr in start_blk to the next bucket. */
4987 return ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4988 new_blk, first_hash, 1);
4992 * Move some xattrs from the old cluster to the new one since they are not
4993 * contiguous in ocfs2 xattr tree.
4995 * new_blk starts a new separate cluster, and we will move some xattrs from
4996 * prev_blk to it. v_start will be set as the first name hash value in this
4997 * new cluster so that it can be used as e_cpos during tree insertion and
4998 * don't collide with our original b-tree operations. first_bh and header_bh
4999 * will also be updated since they will be used in ocfs2_extend_xattr_bucket
5000 * to extend the insert bucket.
5002 * The problem is how much xattr should we move to the new one and when should
5003 * we update first_bh and header_bh?
5004 * 1. If cluster size > bucket size, that means the previous cluster has more
5005 * than 1 bucket, so just move half nums of bucket into the new cluster and
5006 * update the first_bh and header_bh if the insert bucket has been moved
5007 * to the new cluster.
5008 * 2. If cluster_size == bucket_size:
5009 * a) If the previous extent rec has more than one cluster and the insert
5010 * place isn't in the last cluster, copy the entire last cluster to the
5011 * new one. This time, we don't need to upate the first_bh and header_bh
5012 * since they will not be moved into the new cluster.
5013 * b) Otherwise, move the bottom half of the xattrs in the last cluster into
5014 * the new one. And we set the extend flag to zero if the insert place is
5015 * moved into the new allocated cluster since no extend is needed.
5017 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
5018 handle_t *handle,
5019 struct ocfs2_xattr_bucket *first,
5020 struct ocfs2_xattr_bucket *target,
5021 u64 new_blk,
5022 u32 prev_clusters,
5023 u32 *v_start,
5024 int *extend)
5026 int ret;
5028 trace_ocfs2_adjust_xattr_cross_cluster(
5029 (unsigned long long)bucket_blkno(first),
5030 (unsigned long long)new_blk, prev_clusters);
5032 if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
5033 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
5034 handle,
5035 first, target,
5036 new_blk,
5037 prev_clusters,
5038 v_start);
5039 if (ret)
5040 mlog_errno(ret);
5041 } else {
5042 /* The start of the last cluster in the first extent */
5043 u64 last_blk = bucket_blkno(first) +
5044 ((prev_clusters - 1) *
5045 ocfs2_clusters_to_blocks(inode->i_sb, 1));
5047 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
5048 ret = ocfs2_mv_xattr_buckets(inode, handle,
5049 bucket_blkno(first),
5050 last_blk, new_blk, 0,
5051 v_start);
5052 if (ret)
5053 mlog_errno(ret);
5054 } else {
5055 ret = ocfs2_divide_xattr_cluster(inode, handle,
5056 last_blk, new_blk,
5057 v_start);
5058 if (ret)
5059 mlog_errno(ret);
5061 if ((bucket_blkno(target) == last_blk) && extend)
5062 *extend = 0;
5066 return ret;
5070 * Add a new cluster for xattr storage.
5072 * If the new cluster is contiguous with the previous one, it will be
5073 * appended to the same extent record, and num_clusters will be updated.
5074 * If not, we will insert a new extent for it and move some xattrs in
5075 * the last cluster into the new allocated one.
5076 * We also need to limit the maximum size of a btree leaf, otherwise we'll
5077 * lose the benefits of hashing because we'll have to search large leaves.
5078 * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
5079 * if it's bigger).
5081 * first_bh is the first block of the previous extent rec and header_bh
5082 * indicates the bucket we will insert the new xattrs. They will be updated
5083 * when the header_bh is moved into the new cluster.
5085 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
5086 struct buffer_head *root_bh,
5087 struct ocfs2_xattr_bucket *first,
5088 struct ocfs2_xattr_bucket *target,
5089 u32 *num_clusters,
5090 u32 prev_cpos,
5091 int *extend,
5092 struct ocfs2_xattr_set_ctxt *ctxt)
5094 int ret;
5095 u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
5096 u32 prev_clusters = *num_clusters;
5097 u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
5098 u64 block;
5099 handle_t *handle = ctxt->handle;
5100 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5101 struct ocfs2_extent_tree et;
5103 trace_ocfs2_add_new_xattr_cluster_begin(
5104 (unsigned long long)OCFS2_I(inode)->ip_blkno,
5105 (unsigned long long)bucket_blkno(first),
5106 prev_cpos, prev_clusters);
5108 ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5110 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5111 OCFS2_JOURNAL_ACCESS_WRITE);
5112 if (ret < 0) {
5113 mlog_errno(ret);
5114 goto leave;
5117 ret = __ocfs2_claim_clusters(handle, ctxt->data_ac, 1,
5118 clusters_to_add, &bit_off, &num_bits);
5119 if (ret < 0) {
5120 if (ret != -ENOSPC)
5121 mlog_errno(ret);
5122 goto leave;
5125 BUG_ON(num_bits > clusters_to_add);
5127 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
5128 trace_ocfs2_add_new_xattr_cluster((unsigned long long)block, num_bits);
5130 if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
5131 (prev_clusters + num_bits) << osb->s_clustersize_bits <=
5132 OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
5134 * If this cluster is contiguous with the old one and
5135 * adding this new cluster, we don't surpass the limit of
5136 * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
5137 * initialized and used like other buckets in the previous
5138 * cluster.
5139 * So add it as a contiguous one. The caller will handle
5140 * its init process.
5142 v_start = prev_cpos + prev_clusters;
5143 *num_clusters = prev_clusters + num_bits;
5144 } else {
5145 ret = ocfs2_adjust_xattr_cross_cluster(inode,
5146 handle,
5147 first,
5148 target,
5149 block,
5150 prev_clusters,
5151 &v_start,
5152 extend);
5153 if (ret) {
5154 mlog_errno(ret);
5155 goto leave;
5159 trace_ocfs2_add_new_xattr_cluster_insert((unsigned long long)block,
5160 v_start, num_bits);
5161 ret = ocfs2_insert_extent(handle, &et, v_start, block,
5162 num_bits, 0, ctxt->meta_ac);
5163 if (ret < 0) {
5164 mlog_errno(ret);
5165 goto leave;
5168 ocfs2_journal_dirty(handle, root_bh);
5170 leave:
5171 return ret;
5175 * We are given an extent. 'first' is the bucket at the very front of
5176 * the extent. The extent has space for an additional bucket past
5177 * bucket_xh(first)->xh_num_buckets. 'target_blkno' is the block number
5178 * of the target bucket. We wish to shift every bucket past the target
5179 * down one, filling in that additional space. When we get back to the
5180 * target, we split the target between itself and the now-empty bucket
5181 * at target+1 (aka, target_blkno + blks_per_bucket).
5183 static int ocfs2_extend_xattr_bucket(struct inode *inode,
5184 handle_t *handle,
5185 struct ocfs2_xattr_bucket *first,
5186 u64 target_blk,
5187 u32 num_clusters)
5189 int ret, credits;
5190 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5191 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5192 u64 end_blk;
5193 u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
5195 trace_ocfs2_extend_xattr_bucket((unsigned long long)target_blk,
5196 (unsigned long long)bucket_blkno(first),
5197 num_clusters, new_bucket);
5199 /* The extent must have room for an additional bucket */
5200 BUG_ON(new_bucket >=
5201 (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
5203 /* end_blk points to the last existing bucket */
5204 end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
5207 * end_blk is the start of the last existing bucket.
5208 * Thus, (end_blk - target_blk) covers the target bucket and
5209 * every bucket after it up to, but not including, the last
5210 * existing bucket. Then we add the last existing bucket, the
5211 * new bucket, and the first bucket (3 * blk_per_bucket).
5213 credits = (end_blk - target_blk) + (3 * blk_per_bucket);
5214 ret = ocfs2_extend_trans(handle, credits);
5215 if (ret) {
5216 mlog_errno(ret);
5217 goto out;
5220 ret = ocfs2_xattr_bucket_journal_access(handle, first,
5221 OCFS2_JOURNAL_ACCESS_WRITE);
5222 if (ret) {
5223 mlog_errno(ret);
5224 goto out;
5227 while (end_blk != target_blk) {
5228 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
5229 end_blk + blk_per_bucket, 0);
5230 if (ret)
5231 goto out;
5232 end_blk -= blk_per_bucket;
5235 /* Move half of the xattr in target_blkno to the next bucket. */
5236 ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
5237 target_blk + blk_per_bucket, NULL, 0);
5239 le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
5240 ocfs2_xattr_bucket_journal_dirty(handle, first);
5242 out:
5243 return ret;
5247 * Add new xattr bucket in an extent record and adjust the buckets
5248 * accordingly. xb_bh is the ocfs2_xattr_block, and target is the
5249 * bucket we want to insert into.
5251 * In the easy case, we will move all the buckets after target down by
5252 * one. Half of target's xattrs will be moved to the next bucket.
5254 * If current cluster is full, we'll allocate a new one. This may not
5255 * be contiguous. The underlying calls will make sure that there is
5256 * space for the insert, shifting buckets around if necessary.
5257 * 'target' may be moved by those calls.
5259 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
5260 struct buffer_head *xb_bh,
5261 struct ocfs2_xattr_bucket *target,
5262 struct ocfs2_xattr_set_ctxt *ctxt)
5264 struct ocfs2_xattr_block *xb =
5265 (struct ocfs2_xattr_block *)xb_bh->b_data;
5266 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
5267 struct ocfs2_extent_list *el = &xb_root->xt_list;
5268 u32 name_hash =
5269 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
5270 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5271 int ret, num_buckets, extend = 1;
5272 u64 p_blkno;
5273 u32 e_cpos, num_clusters;
5274 /* The bucket at the front of the extent */
5275 struct ocfs2_xattr_bucket *first;
5277 trace_ocfs2_add_new_xattr_bucket(
5278 (unsigned long long)bucket_blkno(target));
5280 /* The first bucket of the original extent */
5281 first = ocfs2_xattr_bucket_new(inode);
5282 if (!first) {
5283 ret = -ENOMEM;
5284 mlog_errno(ret);
5285 goto out;
5288 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
5289 &num_clusters, el);
5290 if (ret) {
5291 mlog_errno(ret);
5292 goto out;
5295 ret = ocfs2_read_xattr_bucket(first, p_blkno);
5296 if (ret) {
5297 mlog_errno(ret);
5298 goto out;
5301 num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
5302 if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
5304 * This can move first+target if the target bucket moves
5305 * to the new extent.
5307 ret = ocfs2_add_new_xattr_cluster(inode,
5308 xb_bh,
5309 first,
5310 target,
5311 &num_clusters,
5312 e_cpos,
5313 &extend,
5314 ctxt);
5315 if (ret) {
5316 mlog_errno(ret);
5317 goto out;
5321 if (extend) {
5322 ret = ocfs2_extend_xattr_bucket(inode,
5323 ctxt->handle,
5324 first,
5325 bucket_blkno(target),
5326 num_clusters);
5327 if (ret)
5328 mlog_errno(ret);
5331 out:
5332 ocfs2_xattr_bucket_free(first);
5334 return ret;
5337 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
5338 struct ocfs2_xattr_bucket *bucket,
5339 int offs)
5341 int block_off = offs >> inode->i_sb->s_blocksize_bits;
5343 offs = offs % inode->i_sb->s_blocksize;
5344 return bucket_block(bucket, block_off) + offs;
5348 * Truncate the specified xe_off entry in xattr bucket.
5349 * bucket is indicated by header_bh and len is the new length.
5350 * Both the ocfs2_xattr_value_root and the entry will be updated here.
5352 * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
5354 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
5355 struct ocfs2_xattr_bucket *bucket,
5356 int xe_off,
5357 int len,
5358 struct ocfs2_xattr_set_ctxt *ctxt)
5360 int ret, offset;
5361 u64 value_blk;
5362 struct ocfs2_xattr_entry *xe;
5363 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5364 size_t blocksize = inode->i_sb->s_blocksize;
5365 struct ocfs2_xattr_value_buf vb = {
5366 .vb_access = ocfs2_journal_access,
5369 xe = &xh->xh_entries[xe_off];
5371 BUG_ON(!xe || ocfs2_xattr_is_local(xe));
5373 offset = le16_to_cpu(xe->xe_name_offset) +
5374 OCFS2_XATTR_SIZE(xe->xe_name_len);
5376 value_blk = offset / blocksize;
5378 /* We don't allow ocfs2_xattr_value to be stored in different block. */
5379 BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
5381 vb.vb_bh = bucket->bu_bhs[value_blk];
5382 BUG_ON(!vb.vb_bh);
5384 vb.vb_xv = (struct ocfs2_xattr_value_root *)
5385 (vb.vb_bh->b_data + offset % blocksize);
5388 * From here on out we have to dirty the bucket. The generic
5389 * value calls only modify one of the bucket's bhs, but we need
5390 * to send the bucket at once. So if they error, they *could* have
5391 * modified something. We have to assume they did, and dirty
5392 * the whole bucket. This leaves us in a consistent state.
5394 trace_ocfs2_xattr_bucket_value_truncate(
5395 (unsigned long long)bucket_blkno(bucket), xe_off, len);
5396 ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
5397 if (ret) {
5398 mlog_errno(ret);
5399 goto out;
5402 ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
5403 OCFS2_JOURNAL_ACCESS_WRITE);
5404 if (ret) {
5405 mlog_errno(ret);
5406 goto out;
5409 xe->xe_value_size = cpu_to_le64(len);
5411 ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
5413 out:
5414 return ret;
5417 static int ocfs2_rm_xattr_cluster(struct inode *inode,
5418 struct buffer_head *root_bh,
5419 u64 blkno,
5420 u32 cpos,
5421 u32 len,
5422 void *para)
5424 int ret;
5425 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5426 struct inode *tl_inode = osb->osb_tl_inode;
5427 handle_t *handle;
5428 struct ocfs2_xattr_block *xb =
5429 (struct ocfs2_xattr_block *)root_bh->b_data;
5430 struct ocfs2_alloc_context *meta_ac = NULL;
5431 struct ocfs2_cached_dealloc_ctxt dealloc;
5432 struct ocfs2_extent_tree et;
5434 ret = ocfs2_iterate_xattr_buckets(inode, blkno, len,
5435 ocfs2_delete_xattr_in_bucket, para);
5436 if (ret) {
5437 mlog_errno(ret);
5438 return ret;
5441 ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5443 ocfs2_init_dealloc_ctxt(&dealloc);
5445 trace_ocfs2_rm_xattr_cluster(
5446 (unsigned long long)OCFS2_I(inode)->ip_blkno,
5447 (unsigned long long)blkno, cpos, len);
5449 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode), blkno,
5450 len);
5452 ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
5453 if (ret) {
5454 mlog_errno(ret);
5455 return ret;
5458 mutex_lock(&tl_inode->i_mutex);
5460 if (ocfs2_truncate_log_needs_flush(osb)) {
5461 ret = __ocfs2_flush_truncate_log(osb);
5462 if (ret < 0) {
5463 mlog_errno(ret);
5464 goto out;
5468 handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
5469 if (IS_ERR(handle)) {
5470 ret = -ENOMEM;
5471 mlog_errno(ret);
5472 goto out;
5475 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5476 OCFS2_JOURNAL_ACCESS_WRITE);
5477 if (ret) {
5478 mlog_errno(ret);
5479 goto out_commit;
5482 ret = ocfs2_remove_extent(handle, &et, cpos, len, meta_ac,
5483 &dealloc);
5484 if (ret) {
5485 mlog_errno(ret);
5486 goto out_commit;
5489 le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
5490 ocfs2_journal_dirty(handle, root_bh);
5492 ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
5493 if (ret)
5494 mlog_errno(ret);
5495 ocfs2_update_inode_fsync_trans(handle, inode, 0);
5497 out_commit:
5498 ocfs2_commit_trans(osb, handle);
5499 out:
5500 ocfs2_schedule_truncate_log_flush(osb, 1);
5502 mutex_unlock(&tl_inode->i_mutex);
5504 if (meta_ac)
5505 ocfs2_free_alloc_context(meta_ac);
5507 ocfs2_run_deallocs(osb, &dealloc);
5509 return ret;
5513 * check whether the xattr bucket is filled up with the same hash value.
5514 * If we want to insert the xattr with the same hash, return -ENOSPC.
5515 * If we want to insert a xattr with different hash value, go ahead
5516 * and ocfs2_divide_xattr_bucket will handle this.
5518 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
5519 struct ocfs2_xattr_bucket *bucket,
5520 const char *name)
5522 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5523 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5525 if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5526 return 0;
5528 if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5529 xh->xh_entries[0].xe_name_hash) {
5530 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5531 "hash = %u\n",
5532 (unsigned long long)bucket_blkno(bucket),
5533 le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5534 return -ENOSPC;
5537 return 0;
5541 * Try to set the entry in the current bucket. If we fail, the caller
5542 * will handle getting us another bucket.
5544 static int ocfs2_xattr_set_entry_bucket(struct inode *inode,
5545 struct ocfs2_xattr_info *xi,
5546 struct ocfs2_xattr_search *xs,
5547 struct ocfs2_xattr_set_ctxt *ctxt)
5549 int ret;
5550 struct ocfs2_xa_loc loc;
5552 trace_ocfs2_xattr_set_entry_bucket(xi->xi_name);
5554 ocfs2_init_xattr_bucket_xa_loc(&loc, xs->bucket,
5555 xs->not_found ? NULL : xs->here);
5556 ret = ocfs2_xa_set(&loc, xi, ctxt);
5557 if (!ret) {
5558 xs->here = loc.xl_entry;
5559 goto out;
5561 if (ret != -ENOSPC) {
5562 mlog_errno(ret);
5563 goto out;
5566 /* Ok, we need space. Let's try defragmenting the bucket. */
5567 ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5568 xs->bucket);
5569 if (ret) {
5570 mlog_errno(ret);
5571 goto out;
5574 ret = ocfs2_xa_set(&loc, xi, ctxt);
5575 if (!ret) {
5576 xs->here = loc.xl_entry;
5577 goto out;
5579 if (ret != -ENOSPC)
5580 mlog_errno(ret);
5583 out:
5584 return ret;
5587 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5588 struct ocfs2_xattr_info *xi,
5589 struct ocfs2_xattr_search *xs,
5590 struct ocfs2_xattr_set_ctxt *ctxt)
5592 int ret;
5594 trace_ocfs2_xattr_set_entry_index_block(xi->xi_name);
5596 ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5597 if (!ret)
5598 goto out;
5599 if (ret != -ENOSPC) {
5600 mlog_errno(ret);
5601 goto out;
5604 /* Ack, need more space. Let's try to get another bucket! */
5607 * We do not allow for overlapping ranges between buckets. And
5608 * the maximum number of collisions we will allow for then is
5609 * one bucket's worth, so check it here whether we need to
5610 * add a new bucket for the insert.
5612 ret = ocfs2_check_xattr_bucket_collision(inode,
5613 xs->bucket,
5614 xi->xi_name);
5615 if (ret) {
5616 mlog_errno(ret);
5617 goto out;
5620 ret = ocfs2_add_new_xattr_bucket(inode,
5621 xs->xattr_bh,
5622 xs->bucket,
5623 ctxt);
5624 if (ret) {
5625 mlog_errno(ret);
5626 goto out;
5630 * ocfs2_add_new_xattr_bucket() will have updated
5631 * xs->bucket if it moved, but it will not have updated
5632 * any of the other search fields. Thus, we drop it and
5633 * re-search. Everything should be cached, so it'll be
5634 * quick.
5636 ocfs2_xattr_bucket_relse(xs->bucket);
5637 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5638 xi->xi_name_index,
5639 xi->xi_name, xs);
5640 if (ret && ret != -ENODATA)
5641 goto out;
5642 xs->not_found = ret;
5644 /* Ok, we have a new bucket, let's try again */
5645 ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5646 if (ret && (ret != -ENOSPC))
5647 mlog_errno(ret);
5649 out:
5650 return ret;
5653 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5654 struct ocfs2_xattr_bucket *bucket,
5655 void *para)
5657 int ret = 0, ref_credits;
5658 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5659 u16 i;
5660 struct ocfs2_xattr_entry *xe;
5661 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5662 struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5663 int credits = ocfs2_remove_extent_credits(osb->sb) +
5664 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5665 struct ocfs2_xattr_value_root *xv;
5666 struct ocfs2_rm_xattr_bucket_para *args =
5667 (struct ocfs2_rm_xattr_bucket_para *)para;
5669 ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5671 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5672 xe = &xh->xh_entries[i];
5673 if (ocfs2_xattr_is_local(xe))
5674 continue;
5676 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket,
5677 i, &xv, NULL);
5679 ret = ocfs2_lock_xattr_remove_allocators(inode, xv,
5680 args->ref_ci,
5681 args->ref_root_bh,
5682 &ctxt.meta_ac,
5683 &ref_credits);
5685 ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
5686 if (IS_ERR(ctxt.handle)) {
5687 ret = PTR_ERR(ctxt.handle);
5688 mlog_errno(ret);
5689 break;
5692 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5693 i, 0, &ctxt);
5695 ocfs2_commit_trans(osb, ctxt.handle);
5696 if (ctxt.meta_ac) {
5697 ocfs2_free_alloc_context(ctxt.meta_ac);
5698 ctxt.meta_ac = NULL;
5700 if (ret) {
5701 mlog_errno(ret);
5702 break;
5706 if (ctxt.meta_ac)
5707 ocfs2_free_alloc_context(ctxt.meta_ac);
5708 ocfs2_schedule_truncate_log_flush(osb, 1);
5709 ocfs2_run_deallocs(osb, &ctxt.dealloc);
5710 return ret;
5714 * Whenever we modify a xattr value root in the bucket(e.g, CoW
5715 * or change the extent record flag), we need to recalculate
5716 * the metaecc for the whole bucket. So it is done here.
5718 * Note:
5719 * We have to give the extra credits for the caller.
5721 static int ocfs2_xattr_bucket_post_refcount(struct inode *inode,
5722 handle_t *handle,
5723 void *para)
5725 int ret;
5726 struct ocfs2_xattr_bucket *bucket =
5727 (struct ocfs2_xattr_bucket *)para;
5729 ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
5730 OCFS2_JOURNAL_ACCESS_WRITE);
5731 if (ret) {
5732 mlog_errno(ret);
5733 return ret;
5736 ocfs2_xattr_bucket_journal_dirty(handle, bucket);
5738 return 0;
5742 * Special action we need if the xattr value is refcounted.
5744 * 1. If the xattr is refcounted, lock the tree.
5745 * 2. CoW the xattr if we are setting the new value and the value
5746 * will be stored outside.
5747 * 3. In other case, decrease_refcount will work for us, so just
5748 * lock the refcount tree, calculate the meta and credits is OK.
5750 * We have to do CoW before ocfs2_init_xattr_set_ctxt since
5751 * currently CoW is a completed transaction, while this function
5752 * will also lock the allocators and let us deadlock. So we will
5753 * CoW the whole xattr value.
5755 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
5756 struct ocfs2_dinode *di,
5757 struct ocfs2_xattr_info *xi,
5758 struct ocfs2_xattr_search *xis,
5759 struct ocfs2_xattr_search *xbs,
5760 struct ocfs2_refcount_tree **ref_tree,
5761 int *meta_add,
5762 int *credits)
5764 int ret = 0;
5765 struct ocfs2_xattr_block *xb;
5766 struct ocfs2_xattr_entry *xe;
5767 char *base;
5768 u32 p_cluster, num_clusters;
5769 unsigned int ext_flags;
5770 int name_offset, name_len;
5771 struct ocfs2_xattr_value_buf vb;
5772 struct ocfs2_xattr_bucket *bucket = NULL;
5773 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5774 struct ocfs2_post_refcount refcount;
5775 struct ocfs2_post_refcount *p = NULL;
5776 struct buffer_head *ref_root_bh = NULL;
5778 if (!xis->not_found) {
5779 xe = xis->here;
5780 name_offset = le16_to_cpu(xe->xe_name_offset);
5781 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5782 base = xis->base;
5783 vb.vb_bh = xis->inode_bh;
5784 vb.vb_access = ocfs2_journal_access_di;
5785 } else {
5786 int i, block_off = 0;
5787 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
5788 xe = xbs->here;
5789 name_offset = le16_to_cpu(xe->xe_name_offset);
5790 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5791 i = xbs->here - xbs->header->xh_entries;
5793 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
5794 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
5795 bucket_xh(xbs->bucket),
5796 i, &block_off,
5797 &name_offset);
5798 if (ret) {
5799 mlog_errno(ret);
5800 goto out;
5802 base = bucket_block(xbs->bucket, block_off);
5803 vb.vb_bh = xbs->bucket->bu_bhs[block_off];
5804 vb.vb_access = ocfs2_journal_access;
5806 if (ocfs2_meta_ecc(osb)) {
5807 /*create parameters for ocfs2_post_refcount. */
5808 bucket = xbs->bucket;
5809 refcount.credits = bucket->bu_blocks;
5810 refcount.para = bucket;
5811 refcount.func =
5812 ocfs2_xattr_bucket_post_refcount;
5813 p = &refcount;
5815 } else {
5816 base = xbs->base;
5817 vb.vb_bh = xbs->xattr_bh;
5818 vb.vb_access = ocfs2_journal_access_xb;
5822 if (ocfs2_xattr_is_local(xe))
5823 goto out;
5825 vb.vb_xv = (struct ocfs2_xattr_value_root *)
5826 (base + name_offset + name_len);
5828 ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
5829 &num_clusters, &vb.vb_xv->xr_list,
5830 &ext_flags);
5831 if (ret) {
5832 mlog_errno(ret);
5833 goto out;
5837 * We just need to check the 1st extent record, since we always
5838 * CoW the whole xattr. So there shouldn't be a xattr with
5839 * some REFCOUNT extent recs after the 1st one.
5841 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
5842 goto out;
5844 ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
5845 1, ref_tree, &ref_root_bh);
5846 if (ret) {
5847 mlog_errno(ret);
5848 goto out;
5852 * If we are deleting the xattr or the new size will be stored inside,
5853 * cool, leave it there, the xattr truncate process will remove them
5854 * for us(it still needs the refcount tree lock and the meta, credits).
5855 * And the worse case is that every cluster truncate will split the
5856 * refcount tree, and make the original extent become 3. So we will need
5857 * 2 * cluster more extent recs at most.
5859 if (!xi->xi_value || xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE) {
5861 ret = ocfs2_refcounted_xattr_delete_need(inode,
5862 &(*ref_tree)->rf_ci,
5863 ref_root_bh, vb.vb_xv,
5864 meta_add, credits);
5865 if (ret)
5866 mlog_errno(ret);
5867 goto out;
5870 ret = ocfs2_refcount_cow_xattr(inode, di, &vb,
5871 *ref_tree, ref_root_bh, 0,
5872 le32_to_cpu(vb.vb_xv->xr_clusters), p);
5873 if (ret)
5874 mlog_errno(ret);
5876 out:
5877 brelse(ref_root_bh);
5878 return ret;
5882 * Add the REFCOUNTED flags for all the extent rec in ocfs2_xattr_value_root.
5883 * The physical clusters will be added to refcount tree.
5885 static int ocfs2_xattr_value_attach_refcount(struct inode *inode,
5886 struct ocfs2_xattr_value_root *xv,
5887 struct ocfs2_extent_tree *value_et,
5888 struct ocfs2_caching_info *ref_ci,
5889 struct buffer_head *ref_root_bh,
5890 struct ocfs2_cached_dealloc_ctxt *dealloc,
5891 struct ocfs2_post_refcount *refcount)
5893 int ret = 0;
5894 u32 clusters = le32_to_cpu(xv->xr_clusters);
5895 u32 cpos, p_cluster, num_clusters;
5896 struct ocfs2_extent_list *el = &xv->xr_list;
5897 unsigned int ext_flags;
5899 cpos = 0;
5900 while (cpos < clusters) {
5901 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
5902 &num_clusters, el, &ext_flags);
5903 if (ret) {
5904 mlog_errno(ret);
5905 break;
5908 cpos += num_clusters;
5909 if ((ext_flags & OCFS2_EXT_REFCOUNTED))
5910 continue;
5912 BUG_ON(!p_cluster);
5914 ret = ocfs2_add_refcount_flag(inode, value_et,
5915 ref_ci, ref_root_bh,
5916 cpos - num_clusters,
5917 p_cluster, num_clusters,
5918 dealloc, refcount);
5919 if (ret) {
5920 mlog_errno(ret);
5921 break;
5925 return ret;
5929 * Given a normal ocfs2_xattr_header, refcount all the entries which
5930 * have value stored outside.
5931 * Used for xattrs stored in inode and ocfs2_xattr_block.
5933 static int ocfs2_xattr_attach_refcount_normal(struct inode *inode,
5934 struct ocfs2_xattr_value_buf *vb,
5935 struct ocfs2_xattr_header *header,
5936 struct ocfs2_caching_info *ref_ci,
5937 struct buffer_head *ref_root_bh,
5938 struct ocfs2_cached_dealloc_ctxt *dealloc)
5941 struct ocfs2_xattr_entry *xe;
5942 struct ocfs2_xattr_value_root *xv;
5943 struct ocfs2_extent_tree et;
5944 int i, ret = 0;
5946 for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
5947 xe = &header->xh_entries[i];
5949 if (ocfs2_xattr_is_local(xe))
5950 continue;
5952 xv = (struct ocfs2_xattr_value_root *)((void *)header +
5953 le16_to_cpu(xe->xe_name_offset) +
5954 OCFS2_XATTR_SIZE(xe->xe_name_len));
5956 vb->vb_xv = xv;
5957 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
5959 ret = ocfs2_xattr_value_attach_refcount(inode, xv, &et,
5960 ref_ci, ref_root_bh,
5961 dealloc, NULL);
5962 if (ret) {
5963 mlog_errno(ret);
5964 break;
5968 return ret;
5971 static int ocfs2_xattr_inline_attach_refcount(struct inode *inode,
5972 struct buffer_head *fe_bh,
5973 struct ocfs2_caching_info *ref_ci,
5974 struct buffer_head *ref_root_bh,
5975 struct ocfs2_cached_dealloc_ctxt *dealloc)
5977 struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
5978 struct ocfs2_xattr_header *header = (struct ocfs2_xattr_header *)
5979 (fe_bh->b_data + inode->i_sb->s_blocksize -
5980 le16_to_cpu(di->i_xattr_inline_size));
5981 struct ocfs2_xattr_value_buf vb = {
5982 .vb_bh = fe_bh,
5983 .vb_access = ocfs2_journal_access_di,
5986 return ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
5987 ref_ci, ref_root_bh, dealloc);
5990 struct ocfs2_xattr_tree_value_refcount_para {
5991 struct ocfs2_caching_info *ref_ci;
5992 struct buffer_head *ref_root_bh;
5993 struct ocfs2_cached_dealloc_ctxt *dealloc;
5996 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
5997 struct ocfs2_xattr_bucket *bucket,
5998 int offset,
5999 struct ocfs2_xattr_value_root **xv,
6000 struct buffer_head **bh)
6002 int ret, block_off, name_offset;
6003 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
6004 struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
6005 void *base;
6007 ret = ocfs2_xattr_bucket_get_name_value(sb,
6008 bucket_xh(bucket),
6009 offset,
6010 &block_off,
6011 &name_offset);
6012 if (ret) {
6013 mlog_errno(ret);
6014 goto out;
6017 base = bucket_block(bucket, block_off);
6019 *xv = (struct ocfs2_xattr_value_root *)(base + name_offset +
6020 OCFS2_XATTR_SIZE(xe->xe_name_len));
6022 if (bh)
6023 *bh = bucket->bu_bhs[block_off];
6024 out:
6025 return ret;
6029 * For a given xattr bucket, refcount all the entries which
6030 * have value stored outside.
6032 static int ocfs2_xattr_bucket_value_refcount(struct inode *inode,
6033 struct ocfs2_xattr_bucket *bucket,
6034 void *para)
6036 int i, ret = 0;
6037 struct ocfs2_extent_tree et;
6038 struct ocfs2_xattr_tree_value_refcount_para *ref =
6039 (struct ocfs2_xattr_tree_value_refcount_para *)para;
6040 struct ocfs2_xattr_header *xh =
6041 (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6042 struct ocfs2_xattr_entry *xe;
6043 struct ocfs2_xattr_value_buf vb = {
6044 .vb_access = ocfs2_journal_access,
6046 struct ocfs2_post_refcount refcount = {
6047 .credits = bucket->bu_blocks,
6048 .para = bucket,
6049 .func = ocfs2_xattr_bucket_post_refcount,
6051 struct ocfs2_post_refcount *p = NULL;
6053 /* We only need post_refcount if we support metaecc. */
6054 if (ocfs2_meta_ecc(OCFS2_SB(inode->i_sb)))
6055 p = &refcount;
6057 trace_ocfs2_xattr_bucket_value_refcount(
6058 (unsigned long long)bucket_blkno(bucket),
6059 le16_to_cpu(xh->xh_count));
6060 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6061 xe = &xh->xh_entries[i];
6063 if (ocfs2_xattr_is_local(xe))
6064 continue;
6066 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket, i,
6067 &vb.vb_xv, &vb.vb_bh);
6068 if (ret) {
6069 mlog_errno(ret);
6070 break;
6073 ocfs2_init_xattr_value_extent_tree(&et,
6074 INODE_CACHE(inode), &vb);
6076 ret = ocfs2_xattr_value_attach_refcount(inode, vb.vb_xv,
6077 &et, ref->ref_ci,
6078 ref->ref_root_bh,
6079 ref->dealloc, p);
6080 if (ret) {
6081 mlog_errno(ret);
6082 break;
6086 return ret;
6090 static int ocfs2_refcount_xattr_tree_rec(struct inode *inode,
6091 struct buffer_head *root_bh,
6092 u64 blkno, u32 cpos, u32 len, void *para)
6094 return ocfs2_iterate_xattr_buckets(inode, blkno, len,
6095 ocfs2_xattr_bucket_value_refcount,
6096 para);
6099 static int ocfs2_xattr_block_attach_refcount(struct inode *inode,
6100 struct buffer_head *blk_bh,
6101 struct ocfs2_caching_info *ref_ci,
6102 struct buffer_head *ref_root_bh,
6103 struct ocfs2_cached_dealloc_ctxt *dealloc)
6105 int ret = 0;
6106 struct ocfs2_xattr_block *xb =
6107 (struct ocfs2_xattr_block *)blk_bh->b_data;
6109 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
6110 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
6111 struct ocfs2_xattr_value_buf vb = {
6112 .vb_bh = blk_bh,
6113 .vb_access = ocfs2_journal_access_xb,
6116 ret = ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
6117 ref_ci, ref_root_bh,
6118 dealloc);
6119 } else {
6120 struct ocfs2_xattr_tree_value_refcount_para para = {
6121 .ref_ci = ref_ci,
6122 .ref_root_bh = ref_root_bh,
6123 .dealloc = dealloc,
6126 ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
6127 ocfs2_refcount_xattr_tree_rec,
6128 &para);
6131 return ret;
6134 int ocfs2_xattr_attach_refcount_tree(struct inode *inode,
6135 struct buffer_head *fe_bh,
6136 struct ocfs2_caching_info *ref_ci,
6137 struct buffer_head *ref_root_bh,
6138 struct ocfs2_cached_dealloc_ctxt *dealloc)
6140 int ret = 0;
6141 struct ocfs2_inode_info *oi = OCFS2_I(inode);
6142 struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
6143 struct buffer_head *blk_bh = NULL;
6145 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
6146 ret = ocfs2_xattr_inline_attach_refcount(inode, fe_bh,
6147 ref_ci, ref_root_bh,
6148 dealloc);
6149 if (ret) {
6150 mlog_errno(ret);
6151 goto out;
6155 if (!di->i_xattr_loc)
6156 goto out;
6158 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
6159 &blk_bh);
6160 if (ret < 0) {
6161 mlog_errno(ret);
6162 goto out;
6165 ret = ocfs2_xattr_block_attach_refcount(inode, blk_bh, ref_ci,
6166 ref_root_bh, dealloc);
6167 if (ret)
6168 mlog_errno(ret);
6170 brelse(blk_bh);
6171 out:
6173 return ret;
6176 typedef int (should_xattr_reflinked)(struct ocfs2_xattr_entry *xe);
6178 * Store the information we need in xattr reflink.
6179 * old_bh and new_bh are inode bh for the old and new inode.
6181 struct ocfs2_xattr_reflink {
6182 struct inode *old_inode;
6183 struct inode *new_inode;
6184 struct buffer_head *old_bh;
6185 struct buffer_head *new_bh;
6186 struct ocfs2_caching_info *ref_ci;
6187 struct buffer_head *ref_root_bh;
6188 struct ocfs2_cached_dealloc_ctxt *dealloc;
6189 should_xattr_reflinked *xattr_reflinked;
6193 * Given a xattr header and xe offset,
6194 * return the proper xv and the corresponding bh.
6195 * xattr in inode, block and xattr tree have different implementaions.
6197 typedef int (get_xattr_value_root)(struct super_block *sb,
6198 struct buffer_head *bh,
6199 struct ocfs2_xattr_header *xh,
6200 int offset,
6201 struct ocfs2_xattr_value_root **xv,
6202 struct buffer_head **ret_bh,
6203 void *para);
6206 * Calculate all the xattr value root metadata stored in this xattr header and
6207 * credits we need if we create them from the scratch.
6208 * We use get_xattr_value_root so that all types of xattr container can use it.
6210 static int ocfs2_value_metas_in_xattr_header(struct super_block *sb,
6211 struct buffer_head *bh,
6212 struct ocfs2_xattr_header *xh,
6213 int *metas, int *credits,
6214 int *num_recs,
6215 get_xattr_value_root *func,
6216 void *para)
6218 int i, ret = 0;
6219 struct ocfs2_xattr_value_root *xv;
6220 struct ocfs2_xattr_entry *xe;
6222 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6223 xe = &xh->xh_entries[i];
6224 if (ocfs2_xattr_is_local(xe))
6225 continue;
6227 ret = func(sb, bh, xh, i, &xv, NULL, para);
6228 if (ret) {
6229 mlog_errno(ret);
6230 break;
6233 *metas += le16_to_cpu(xv->xr_list.l_tree_depth) *
6234 le16_to_cpu(xv->xr_list.l_next_free_rec);
6236 *credits += ocfs2_calc_extend_credits(sb,
6237 &def_xv.xv.xr_list);
6240 * If the value is a tree with depth > 1, We don't go deep
6241 * to the extent block, so just calculate a maximum record num.
6243 if (!xv->xr_list.l_tree_depth)
6244 *num_recs += le16_to_cpu(xv->xr_list.l_next_free_rec);
6245 else
6246 *num_recs += ocfs2_clusters_for_bytes(sb,
6247 XATTR_SIZE_MAX);
6250 return ret;
6253 /* Used by xattr inode and block to return the right xv and buffer_head. */
6254 static int ocfs2_get_xattr_value_root(struct super_block *sb,
6255 struct buffer_head *bh,
6256 struct ocfs2_xattr_header *xh,
6257 int offset,
6258 struct ocfs2_xattr_value_root **xv,
6259 struct buffer_head **ret_bh,
6260 void *para)
6262 struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
6264 *xv = (struct ocfs2_xattr_value_root *)((void *)xh +
6265 le16_to_cpu(xe->xe_name_offset) +
6266 OCFS2_XATTR_SIZE(xe->xe_name_len));
6268 if (ret_bh)
6269 *ret_bh = bh;
6271 return 0;
6275 * Lock the meta_ac and caculate how much credits we need for reflink xattrs.
6276 * It is only used for inline xattr and xattr block.
6278 static int ocfs2_reflink_lock_xattr_allocators(struct ocfs2_super *osb,
6279 struct ocfs2_xattr_header *xh,
6280 struct buffer_head *ref_root_bh,
6281 int *credits,
6282 struct ocfs2_alloc_context **meta_ac)
6284 int ret, meta_add = 0, num_recs = 0;
6285 struct ocfs2_refcount_block *rb =
6286 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
6288 *credits = 0;
6290 ret = ocfs2_value_metas_in_xattr_header(osb->sb, NULL, xh,
6291 &meta_add, credits, &num_recs,
6292 ocfs2_get_xattr_value_root,
6293 NULL);
6294 if (ret) {
6295 mlog_errno(ret);
6296 goto out;
6300 * We need to add/modify num_recs in refcount tree, so just calculate
6301 * an approximate number we need for refcount tree change.
6302 * Sometimes we need to split the tree, and after split, half recs
6303 * will be moved to the new block, and a new block can only provide
6304 * half number of recs. So we multiple new blocks by 2.
6306 num_recs = num_recs / ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6307 meta_add += num_recs;
6308 *credits += num_recs + num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6309 if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6310 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6311 le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6312 else
6313 *credits += 1;
6315 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add, meta_ac);
6316 if (ret)
6317 mlog_errno(ret);
6319 out:
6320 return ret;
6324 * Given a xattr header, reflink all the xattrs in this container.
6325 * It can be used for inode, block and bucket.
6327 * NOTE:
6328 * Before we call this function, the caller has memcpy the xattr in
6329 * old_xh to the new_xh.
6331 * If args.xattr_reflinked is set, call it to decide whether the xe should
6332 * be reflinked or not. If not, remove it from the new xattr header.
6334 static int ocfs2_reflink_xattr_header(handle_t *handle,
6335 struct ocfs2_xattr_reflink *args,
6336 struct buffer_head *old_bh,
6337 struct ocfs2_xattr_header *xh,
6338 struct buffer_head *new_bh,
6339 struct ocfs2_xattr_header *new_xh,
6340 struct ocfs2_xattr_value_buf *vb,
6341 struct ocfs2_alloc_context *meta_ac,
6342 get_xattr_value_root *func,
6343 void *para)
6345 int ret = 0, i, j;
6346 struct super_block *sb = args->old_inode->i_sb;
6347 struct buffer_head *value_bh;
6348 struct ocfs2_xattr_entry *xe, *last;
6349 struct ocfs2_xattr_value_root *xv, *new_xv;
6350 struct ocfs2_extent_tree data_et;
6351 u32 clusters, cpos, p_cluster, num_clusters;
6352 unsigned int ext_flags = 0;
6354 trace_ocfs2_reflink_xattr_header((unsigned long long)old_bh->b_blocknr,
6355 le16_to_cpu(xh->xh_count));
6357 last = &new_xh->xh_entries[le16_to_cpu(new_xh->xh_count)];
6358 for (i = 0, j = 0; i < le16_to_cpu(xh->xh_count); i++, j++) {
6359 xe = &xh->xh_entries[i];
6361 if (args->xattr_reflinked && !args->xattr_reflinked(xe)) {
6362 xe = &new_xh->xh_entries[j];
6364 le16_add_cpu(&new_xh->xh_count, -1);
6365 if (new_xh->xh_count) {
6366 memmove(xe, xe + 1,
6367 (void *)last - (void *)xe);
6368 memset(last, 0,
6369 sizeof(struct ocfs2_xattr_entry));
6373 * We don't want j to increase in the next round since
6374 * it is already moved ahead.
6376 j--;
6377 continue;
6380 if (ocfs2_xattr_is_local(xe))
6381 continue;
6383 ret = func(sb, old_bh, xh, i, &xv, NULL, para);
6384 if (ret) {
6385 mlog_errno(ret);
6386 break;
6389 ret = func(sb, new_bh, new_xh, j, &new_xv, &value_bh, para);
6390 if (ret) {
6391 mlog_errno(ret);
6392 break;
6396 * For the xattr which has l_tree_depth = 0, all the extent
6397 * recs have already be copied to the new xh with the
6398 * propriate OCFS2_EXT_REFCOUNTED flag we just need to
6399 * increase the refount count int the refcount tree.
6401 * For the xattr which has l_tree_depth > 0, we need
6402 * to initialize it to the empty default value root,
6403 * and then insert the extents one by one.
6405 if (xv->xr_list.l_tree_depth) {
6406 memcpy(new_xv, &def_xv, sizeof(def_xv));
6407 vb->vb_xv = new_xv;
6408 vb->vb_bh = value_bh;
6409 ocfs2_init_xattr_value_extent_tree(&data_et,
6410 INODE_CACHE(args->new_inode), vb);
6413 clusters = le32_to_cpu(xv->xr_clusters);
6414 cpos = 0;
6415 while (cpos < clusters) {
6416 ret = ocfs2_xattr_get_clusters(args->old_inode,
6417 cpos,
6418 &p_cluster,
6419 &num_clusters,
6420 &xv->xr_list,
6421 &ext_flags);
6422 if (ret) {
6423 mlog_errno(ret);
6424 goto out;
6427 BUG_ON(!p_cluster);
6429 if (xv->xr_list.l_tree_depth) {
6430 ret = ocfs2_insert_extent(handle,
6431 &data_et, cpos,
6432 ocfs2_clusters_to_blocks(
6433 args->old_inode->i_sb,
6434 p_cluster),
6435 num_clusters, ext_flags,
6436 meta_ac);
6437 if (ret) {
6438 mlog_errno(ret);
6439 goto out;
6443 ret = ocfs2_increase_refcount(handle, args->ref_ci,
6444 args->ref_root_bh,
6445 p_cluster, num_clusters,
6446 meta_ac, args->dealloc);
6447 if (ret) {
6448 mlog_errno(ret);
6449 goto out;
6452 cpos += num_clusters;
6456 out:
6457 return ret;
6460 static int ocfs2_reflink_xattr_inline(struct ocfs2_xattr_reflink *args)
6462 int ret = 0, credits = 0;
6463 handle_t *handle;
6464 struct ocfs2_super *osb = OCFS2_SB(args->old_inode->i_sb);
6465 struct ocfs2_dinode *di = (struct ocfs2_dinode *)args->old_bh->b_data;
6466 int inline_size = le16_to_cpu(di->i_xattr_inline_size);
6467 int header_off = osb->sb->s_blocksize - inline_size;
6468 struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)
6469 (args->old_bh->b_data + header_off);
6470 struct ocfs2_xattr_header *new_xh = (struct ocfs2_xattr_header *)
6471 (args->new_bh->b_data + header_off);
6472 struct ocfs2_alloc_context *meta_ac = NULL;
6473 struct ocfs2_inode_info *new_oi;
6474 struct ocfs2_dinode *new_di;
6475 struct ocfs2_xattr_value_buf vb = {
6476 .vb_bh = args->new_bh,
6477 .vb_access = ocfs2_journal_access_di,
6480 ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6481 &credits, &meta_ac);
6482 if (ret) {
6483 mlog_errno(ret);
6484 goto out;
6487 handle = ocfs2_start_trans(osb, credits);
6488 if (IS_ERR(handle)) {
6489 ret = PTR_ERR(handle);
6490 mlog_errno(ret);
6491 goto out;
6494 ret = ocfs2_journal_access_di(handle, INODE_CACHE(args->new_inode),
6495 args->new_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6496 if (ret) {
6497 mlog_errno(ret);
6498 goto out_commit;
6501 memcpy(args->new_bh->b_data + header_off,
6502 args->old_bh->b_data + header_off, inline_size);
6504 new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6505 new_di->i_xattr_inline_size = cpu_to_le16(inline_size);
6507 ret = ocfs2_reflink_xattr_header(handle, args, args->old_bh, xh,
6508 args->new_bh, new_xh, &vb, meta_ac,
6509 ocfs2_get_xattr_value_root, NULL);
6510 if (ret) {
6511 mlog_errno(ret);
6512 goto out_commit;
6515 new_oi = OCFS2_I(args->new_inode);
6517 * Adjust extent record count to reserve space for extended attribute.
6518 * Inline data count had been adjusted in ocfs2_duplicate_inline_data().
6520 if (!(new_oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) &&
6521 !(ocfs2_inode_is_fast_symlink(args->new_inode))) {
6522 struct ocfs2_extent_list *el = &new_di->id2.i_list;
6523 le16_add_cpu(&el->l_count, -(inline_size /
6524 sizeof(struct ocfs2_extent_rec)));
6526 spin_lock(&new_oi->ip_lock);
6527 new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL | OCFS2_INLINE_XATTR_FL;
6528 new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6529 spin_unlock(&new_oi->ip_lock);
6531 ocfs2_journal_dirty(handle, args->new_bh);
6533 out_commit:
6534 ocfs2_commit_trans(osb, handle);
6536 out:
6537 if (meta_ac)
6538 ocfs2_free_alloc_context(meta_ac);
6539 return ret;
6542 static int ocfs2_create_empty_xattr_block(struct inode *inode,
6543 struct buffer_head *fe_bh,
6544 struct buffer_head **ret_bh,
6545 int indexed)
6547 int ret;
6548 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6549 struct ocfs2_xattr_set_ctxt ctxt;
6551 memset(&ctxt, 0, sizeof(ctxt));
6552 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &ctxt.meta_ac);
6553 if (ret < 0) {
6554 mlog_errno(ret);
6555 return ret;
6558 ctxt.handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_CREATE_CREDITS);
6559 if (IS_ERR(ctxt.handle)) {
6560 ret = PTR_ERR(ctxt.handle);
6561 mlog_errno(ret);
6562 goto out;
6565 trace_ocfs2_create_empty_xattr_block(
6566 (unsigned long long)fe_bh->b_blocknr, indexed);
6567 ret = ocfs2_create_xattr_block(inode, fe_bh, &ctxt, indexed,
6568 ret_bh);
6569 if (ret)
6570 mlog_errno(ret);
6572 ocfs2_commit_trans(osb, ctxt.handle);
6573 out:
6574 ocfs2_free_alloc_context(ctxt.meta_ac);
6575 return ret;
6578 static int ocfs2_reflink_xattr_block(struct ocfs2_xattr_reflink *args,
6579 struct buffer_head *blk_bh,
6580 struct buffer_head *new_blk_bh)
6582 int ret = 0, credits = 0;
6583 handle_t *handle;
6584 struct ocfs2_inode_info *new_oi = OCFS2_I(args->new_inode);
6585 struct ocfs2_dinode *new_di;
6586 struct ocfs2_super *osb = OCFS2_SB(args->new_inode->i_sb);
6587 int header_off = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
6588 struct ocfs2_xattr_block *xb =
6589 (struct ocfs2_xattr_block *)blk_bh->b_data;
6590 struct ocfs2_xattr_header *xh = &xb->xb_attrs.xb_header;
6591 struct ocfs2_xattr_block *new_xb =
6592 (struct ocfs2_xattr_block *)new_blk_bh->b_data;
6593 struct ocfs2_xattr_header *new_xh = &new_xb->xb_attrs.xb_header;
6594 struct ocfs2_alloc_context *meta_ac;
6595 struct ocfs2_xattr_value_buf vb = {
6596 .vb_bh = new_blk_bh,
6597 .vb_access = ocfs2_journal_access_xb,
6600 ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6601 &credits, &meta_ac);
6602 if (ret) {
6603 mlog_errno(ret);
6604 return ret;
6607 /* One more credits in case we need to add xattr flags in new inode. */
6608 handle = ocfs2_start_trans(osb, credits + 1);
6609 if (IS_ERR(handle)) {
6610 ret = PTR_ERR(handle);
6611 mlog_errno(ret);
6612 goto out;
6615 if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6616 ret = ocfs2_journal_access_di(handle,
6617 INODE_CACHE(args->new_inode),
6618 args->new_bh,
6619 OCFS2_JOURNAL_ACCESS_WRITE);
6620 if (ret) {
6621 mlog_errno(ret);
6622 goto out_commit;
6626 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(args->new_inode),
6627 new_blk_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6628 if (ret) {
6629 mlog_errno(ret);
6630 goto out_commit;
6633 memcpy(new_blk_bh->b_data + header_off, blk_bh->b_data + header_off,
6634 osb->sb->s_blocksize - header_off);
6636 ret = ocfs2_reflink_xattr_header(handle, args, blk_bh, xh,
6637 new_blk_bh, new_xh, &vb, meta_ac,
6638 ocfs2_get_xattr_value_root, NULL);
6639 if (ret) {
6640 mlog_errno(ret);
6641 goto out_commit;
6644 ocfs2_journal_dirty(handle, new_blk_bh);
6646 if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6647 new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6648 spin_lock(&new_oi->ip_lock);
6649 new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
6650 new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6651 spin_unlock(&new_oi->ip_lock);
6653 ocfs2_journal_dirty(handle, args->new_bh);
6656 out_commit:
6657 ocfs2_commit_trans(osb, handle);
6659 out:
6660 ocfs2_free_alloc_context(meta_ac);
6661 return ret;
6664 struct ocfs2_reflink_xattr_tree_args {
6665 struct ocfs2_xattr_reflink *reflink;
6666 struct buffer_head *old_blk_bh;
6667 struct buffer_head *new_blk_bh;
6668 struct ocfs2_xattr_bucket *old_bucket;
6669 struct ocfs2_xattr_bucket *new_bucket;
6673 * NOTE:
6674 * We have to handle the case that both old bucket and new bucket
6675 * will call this function to get the right ret_bh.
6676 * So The caller must give us the right bh.
6678 static int ocfs2_get_reflink_xattr_value_root(struct super_block *sb,
6679 struct buffer_head *bh,
6680 struct ocfs2_xattr_header *xh,
6681 int offset,
6682 struct ocfs2_xattr_value_root **xv,
6683 struct buffer_head **ret_bh,
6684 void *para)
6686 struct ocfs2_reflink_xattr_tree_args *args =
6687 (struct ocfs2_reflink_xattr_tree_args *)para;
6688 struct ocfs2_xattr_bucket *bucket;
6690 if (bh == args->old_bucket->bu_bhs[0])
6691 bucket = args->old_bucket;
6692 else
6693 bucket = args->new_bucket;
6695 return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6696 xv, ret_bh);
6699 struct ocfs2_value_tree_metas {
6700 int num_metas;
6701 int credits;
6702 int num_recs;
6705 static int ocfs2_value_tree_metas_in_bucket(struct super_block *sb,
6706 struct buffer_head *bh,
6707 struct ocfs2_xattr_header *xh,
6708 int offset,
6709 struct ocfs2_xattr_value_root **xv,
6710 struct buffer_head **ret_bh,
6711 void *para)
6713 struct ocfs2_xattr_bucket *bucket =
6714 (struct ocfs2_xattr_bucket *)para;
6716 return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6717 xv, ret_bh);
6720 static int ocfs2_calc_value_tree_metas(struct inode *inode,
6721 struct ocfs2_xattr_bucket *bucket,
6722 void *para)
6724 struct ocfs2_value_tree_metas *metas =
6725 (struct ocfs2_value_tree_metas *)para;
6726 struct ocfs2_xattr_header *xh =
6727 (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6729 /* Add the credits for this bucket first. */
6730 metas->credits += bucket->bu_blocks;
6731 return ocfs2_value_metas_in_xattr_header(inode->i_sb, bucket->bu_bhs[0],
6732 xh, &metas->num_metas,
6733 &metas->credits, &metas->num_recs,
6734 ocfs2_value_tree_metas_in_bucket,
6735 bucket);
6739 * Given a xattr extent rec starting from blkno and having len clusters,
6740 * iterate all the buckets calculate how much metadata we need for reflinking
6741 * all the ocfs2_xattr_value_root and lock the allocators accordingly.
6743 static int ocfs2_lock_reflink_xattr_rec_allocators(
6744 struct ocfs2_reflink_xattr_tree_args *args,
6745 struct ocfs2_extent_tree *xt_et,
6746 u64 blkno, u32 len, int *credits,
6747 struct ocfs2_alloc_context **meta_ac,
6748 struct ocfs2_alloc_context **data_ac)
6750 int ret, num_free_extents;
6751 struct ocfs2_value_tree_metas metas;
6752 struct ocfs2_super *osb = OCFS2_SB(args->reflink->old_inode->i_sb);
6753 struct ocfs2_refcount_block *rb;
6755 memset(&metas, 0, sizeof(metas));
6757 ret = ocfs2_iterate_xattr_buckets(args->reflink->old_inode, blkno, len,
6758 ocfs2_calc_value_tree_metas, &metas);
6759 if (ret) {
6760 mlog_errno(ret);
6761 goto out;
6764 *credits = metas.credits;
6767 * Calculate we need for refcount tree change.
6769 * We need to add/modify num_recs in refcount tree, so just calculate
6770 * an approximate number we need for refcount tree change.
6771 * Sometimes we need to split the tree, and after split, half recs
6772 * will be moved to the new block, and a new block can only provide
6773 * half number of recs. So we multiple new blocks by 2.
6774 * In the end, we have to add credits for modifying the already
6775 * existed refcount block.
6777 rb = (struct ocfs2_refcount_block *)args->reflink->ref_root_bh->b_data;
6778 metas.num_recs =
6779 (metas.num_recs + ocfs2_refcount_recs_per_rb(osb->sb) - 1) /
6780 ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6781 metas.num_metas += metas.num_recs;
6782 *credits += metas.num_recs +
6783 metas.num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6784 if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6785 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6786 le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6787 else
6788 *credits += 1;
6790 /* count in the xattr tree change. */
6791 num_free_extents = ocfs2_num_free_extents(osb, xt_et);
6792 if (num_free_extents < 0) {
6793 ret = num_free_extents;
6794 mlog_errno(ret);
6795 goto out;
6798 if (num_free_extents < len)
6799 metas.num_metas += ocfs2_extend_meta_needed(xt_et->et_root_el);
6801 *credits += ocfs2_calc_extend_credits(osb->sb,
6802 xt_et->et_root_el);
6804 if (metas.num_metas) {
6805 ret = ocfs2_reserve_new_metadata_blocks(osb, metas.num_metas,
6806 meta_ac);
6807 if (ret) {
6808 mlog_errno(ret);
6809 goto out;
6813 if (len) {
6814 ret = ocfs2_reserve_clusters(osb, len, data_ac);
6815 if (ret)
6816 mlog_errno(ret);
6818 out:
6819 if (ret) {
6820 if (*meta_ac) {
6821 ocfs2_free_alloc_context(*meta_ac);
6822 *meta_ac = NULL;
6826 return ret;
6829 static int ocfs2_reflink_xattr_bucket(handle_t *handle,
6830 u64 blkno, u64 new_blkno, u32 clusters,
6831 u32 *cpos, int num_buckets,
6832 struct ocfs2_alloc_context *meta_ac,
6833 struct ocfs2_alloc_context *data_ac,
6834 struct ocfs2_reflink_xattr_tree_args *args)
6836 int i, j, ret = 0;
6837 struct super_block *sb = args->reflink->old_inode->i_sb;
6838 int bpb = args->old_bucket->bu_blocks;
6839 struct ocfs2_xattr_value_buf vb = {
6840 .vb_access = ocfs2_journal_access,
6843 for (i = 0; i < num_buckets; i++, blkno += bpb, new_blkno += bpb) {
6844 ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6845 if (ret) {
6846 mlog_errno(ret);
6847 break;
6850 ret = ocfs2_init_xattr_bucket(args->new_bucket, new_blkno, 1);
6851 if (ret) {
6852 mlog_errno(ret);
6853 break;
6856 ret = ocfs2_xattr_bucket_journal_access(handle,
6857 args->new_bucket,
6858 OCFS2_JOURNAL_ACCESS_CREATE);
6859 if (ret) {
6860 mlog_errno(ret);
6861 break;
6864 for (j = 0; j < bpb; j++)
6865 memcpy(bucket_block(args->new_bucket, j),
6866 bucket_block(args->old_bucket, j),
6867 sb->s_blocksize);
6870 * Record the start cpos so that we can use it to initialize
6871 * our xattr tree we also set the xh_num_bucket for the new
6872 * bucket.
6874 if (i == 0) {
6875 *cpos = le32_to_cpu(bucket_xh(args->new_bucket)->
6876 xh_entries[0].xe_name_hash);
6877 bucket_xh(args->new_bucket)->xh_num_buckets =
6878 cpu_to_le16(num_buckets);
6881 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6883 ret = ocfs2_reflink_xattr_header(handle, args->reflink,
6884 args->old_bucket->bu_bhs[0],
6885 bucket_xh(args->old_bucket),
6886 args->new_bucket->bu_bhs[0],
6887 bucket_xh(args->new_bucket),
6888 &vb, meta_ac,
6889 ocfs2_get_reflink_xattr_value_root,
6890 args);
6891 if (ret) {
6892 mlog_errno(ret);
6893 break;
6897 * Re-access and dirty the bucket to calculate metaecc.
6898 * Because we may extend the transaction in reflink_xattr_header
6899 * which will let the already accessed block gone.
6901 ret = ocfs2_xattr_bucket_journal_access(handle,
6902 args->new_bucket,
6903 OCFS2_JOURNAL_ACCESS_WRITE);
6904 if (ret) {
6905 mlog_errno(ret);
6906 break;
6909 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6911 ocfs2_xattr_bucket_relse(args->old_bucket);
6912 ocfs2_xattr_bucket_relse(args->new_bucket);
6915 ocfs2_xattr_bucket_relse(args->old_bucket);
6916 ocfs2_xattr_bucket_relse(args->new_bucket);
6917 return ret;
6920 static int ocfs2_reflink_xattr_buckets(handle_t *handle,
6921 struct inode *inode,
6922 struct ocfs2_reflink_xattr_tree_args *args,
6923 struct ocfs2_extent_tree *et,
6924 struct ocfs2_alloc_context *meta_ac,
6925 struct ocfs2_alloc_context *data_ac,
6926 u64 blkno, u32 cpos, u32 len)
6928 int ret, first_inserted = 0;
6929 u32 p_cluster, num_clusters, reflink_cpos = 0;
6930 u64 new_blkno;
6931 unsigned int num_buckets, reflink_buckets;
6932 unsigned int bpc =
6933 ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
6935 ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6936 if (ret) {
6937 mlog_errno(ret);
6938 goto out;
6940 num_buckets = le16_to_cpu(bucket_xh(args->old_bucket)->xh_num_buckets);
6941 ocfs2_xattr_bucket_relse(args->old_bucket);
6943 while (len && num_buckets) {
6944 ret = ocfs2_claim_clusters(handle, data_ac,
6945 1, &p_cluster, &num_clusters);
6946 if (ret) {
6947 mlog_errno(ret);
6948 goto out;
6951 new_blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
6952 reflink_buckets = min(num_buckets, bpc * num_clusters);
6954 ret = ocfs2_reflink_xattr_bucket(handle, blkno,
6955 new_blkno, num_clusters,
6956 &reflink_cpos, reflink_buckets,
6957 meta_ac, data_ac, args);
6958 if (ret) {
6959 mlog_errno(ret);
6960 goto out;
6964 * For the 1st allocated cluster, we make it use the same cpos
6965 * so that the xattr tree looks the same as the original one
6966 * in the most case.
6968 if (!first_inserted) {
6969 reflink_cpos = cpos;
6970 first_inserted = 1;
6972 ret = ocfs2_insert_extent(handle, et, reflink_cpos, new_blkno,
6973 num_clusters, 0, meta_ac);
6974 if (ret)
6975 mlog_errno(ret);
6977 trace_ocfs2_reflink_xattr_buckets((unsigned long long)new_blkno,
6978 num_clusters, reflink_cpos);
6980 len -= num_clusters;
6981 blkno += ocfs2_clusters_to_blocks(inode->i_sb, num_clusters);
6982 num_buckets -= reflink_buckets;
6984 out:
6985 return ret;
6989 * Create the same xattr extent record in the new inode's xattr tree.
6991 static int ocfs2_reflink_xattr_rec(struct inode *inode,
6992 struct buffer_head *root_bh,
6993 u64 blkno,
6994 u32 cpos,
6995 u32 len,
6996 void *para)
6998 int ret, credits = 0;
6999 handle_t *handle;
7000 struct ocfs2_reflink_xattr_tree_args *args =
7001 (struct ocfs2_reflink_xattr_tree_args *)para;
7002 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7003 struct ocfs2_alloc_context *meta_ac = NULL;
7004 struct ocfs2_alloc_context *data_ac = NULL;
7005 struct ocfs2_extent_tree et;
7007 trace_ocfs2_reflink_xattr_rec((unsigned long long)blkno, len);
7009 ocfs2_init_xattr_tree_extent_tree(&et,
7010 INODE_CACHE(args->reflink->new_inode),
7011 args->new_blk_bh);
7013 ret = ocfs2_lock_reflink_xattr_rec_allocators(args, &et, blkno,
7014 len, &credits,
7015 &meta_ac, &data_ac);
7016 if (ret) {
7017 mlog_errno(ret);
7018 goto out;
7021 handle = ocfs2_start_trans(osb, credits);
7022 if (IS_ERR(handle)) {
7023 ret = PTR_ERR(handle);
7024 mlog_errno(ret);
7025 goto out;
7028 ret = ocfs2_reflink_xattr_buckets(handle, inode, args, &et,
7029 meta_ac, data_ac,
7030 blkno, cpos, len);
7031 if (ret)
7032 mlog_errno(ret);
7034 ocfs2_commit_trans(osb, handle);
7036 out:
7037 if (meta_ac)
7038 ocfs2_free_alloc_context(meta_ac);
7039 if (data_ac)
7040 ocfs2_free_alloc_context(data_ac);
7041 return ret;
7045 * Create reflinked xattr buckets.
7046 * We will add bucket one by one, and refcount all the xattrs in the bucket
7047 * if they are stored outside.
7049 static int ocfs2_reflink_xattr_tree(struct ocfs2_xattr_reflink *args,
7050 struct buffer_head *blk_bh,
7051 struct buffer_head *new_blk_bh)
7053 int ret;
7054 struct ocfs2_reflink_xattr_tree_args para;
7056 memset(&para, 0, sizeof(para));
7057 para.reflink = args;
7058 para.old_blk_bh = blk_bh;
7059 para.new_blk_bh = new_blk_bh;
7061 para.old_bucket = ocfs2_xattr_bucket_new(args->old_inode);
7062 if (!para.old_bucket) {
7063 mlog_errno(-ENOMEM);
7064 return -ENOMEM;
7067 para.new_bucket = ocfs2_xattr_bucket_new(args->new_inode);
7068 if (!para.new_bucket) {
7069 ret = -ENOMEM;
7070 mlog_errno(ret);
7071 goto out;
7074 ret = ocfs2_iterate_xattr_index_block(args->old_inode, blk_bh,
7075 ocfs2_reflink_xattr_rec,
7076 &para);
7077 if (ret)
7078 mlog_errno(ret);
7080 out:
7081 ocfs2_xattr_bucket_free(para.old_bucket);
7082 ocfs2_xattr_bucket_free(para.new_bucket);
7083 return ret;
7086 static int ocfs2_reflink_xattr_in_block(struct ocfs2_xattr_reflink *args,
7087 struct buffer_head *blk_bh)
7089 int ret, indexed = 0;
7090 struct buffer_head *new_blk_bh = NULL;
7091 struct ocfs2_xattr_block *xb =
7092 (struct ocfs2_xattr_block *)blk_bh->b_data;
7095 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)
7096 indexed = 1;
7098 ret = ocfs2_create_empty_xattr_block(args->new_inode, args->new_bh,
7099 &new_blk_bh, indexed);
7100 if (ret) {
7101 mlog_errno(ret);
7102 goto out;
7105 if (!indexed)
7106 ret = ocfs2_reflink_xattr_block(args, blk_bh, new_blk_bh);
7107 else
7108 ret = ocfs2_reflink_xattr_tree(args, blk_bh, new_blk_bh);
7109 if (ret)
7110 mlog_errno(ret);
7112 out:
7113 brelse(new_blk_bh);
7114 return ret;
7117 static int ocfs2_reflink_xattr_no_security(struct ocfs2_xattr_entry *xe)
7119 int type = ocfs2_xattr_get_type(xe);
7121 return type != OCFS2_XATTR_INDEX_SECURITY &&
7122 type != OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS &&
7123 type != OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT;
7126 int ocfs2_reflink_xattrs(struct inode *old_inode,
7127 struct buffer_head *old_bh,
7128 struct inode *new_inode,
7129 struct buffer_head *new_bh,
7130 bool preserve_security)
7132 int ret;
7133 struct ocfs2_xattr_reflink args;
7134 struct ocfs2_inode_info *oi = OCFS2_I(old_inode);
7135 struct ocfs2_dinode *di = (struct ocfs2_dinode *)old_bh->b_data;
7136 struct buffer_head *blk_bh = NULL;
7137 struct ocfs2_cached_dealloc_ctxt dealloc;
7138 struct ocfs2_refcount_tree *ref_tree;
7139 struct buffer_head *ref_root_bh = NULL;
7141 ret = ocfs2_lock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7142 le64_to_cpu(di->i_refcount_loc),
7143 1, &ref_tree, &ref_root_bh);
7144 if (ret) {
7145 mlog_errno(ret);
7146 goto out;
7149 ocfs2_init_dealloc_ctxt(&dealloc);
7151 args.old_inode = old_inode;
7152 args.new_inode = new_inode;
7153 args.old_bh = old_bh;
7154 args.new_bh = new_bh;
7155 args.ref_ci = &ref_tree->rf_ci;
7156 args.ref_root_bh = ref_root_bh;
7157 args.dealloc = &dealloc;
7158 if (preserve_security)
7159 args.xattr_reflinked = NULL;
7160 else
7161 args.xattr_reflinked = ocfs2_reflink_xattr_no_security;
7163 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
7164 ret = ocfs2_reflink_xattr_inline(&args);
7165 if (ret) {
7166 mlog_errno(ret);
7167 goto out_unlock;
7171 if (!di->i_xattr_loc)
7172 goto out_unlock;
7174 ret = ocfs2_read_xattr_block(old_inode, le64_to_cpu(di->i_xattr_loc),
7175 &blk_bh);
7176 if (ret < 0) {
7177 mlog_errno(ret);
7178 goto out_unlock;
7181 ret = ocfs2_reflink_xattr_in_block(&args, blk_bh);
7182 if (ret)
7183 mlog_errno(ret);
7185 brelse(blk_bh);
7187 out_unlock:
7188 ocfs2_unlock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7189 ref_tree, 1);
7190 brelse(ref_root_bh);
7192 if (ocfs2_dealloc_has_cluster(&dealloc)) {
7193 ocfs2_schedule_truncate_log_flush(OCFS2_SB(old_inode->i_sb), 1);
7194 ocfs2_run_deallocs(OCFS2_SB(old_inode->i_sb), &dealloc);
7197 out:
7198 return ret;
7202 * Initialize security and acl for a already created inode.
7203 * Used for reflink a non-preserve-security file.
7205 * It uses common api like ocfs2_xattr_set, so the caller
7206 * must not hold any lock expect i_mutex.
7208 int ocfs2_init_security_and_acl(struct inode *dir,
7209 struct inode *inode,
7210 const struct qstr *qstr,
7211 struct posix_acl *default_acl,
7212 struct posix_acl *acl)
7214 struct buffer_head *dir_bh = NULL;
7215 int ret = 0;
7217 ret = ocfs2_init_security_get(inode, dir, qstr, NULL);
7218 if (ret) {
7219 mlog_errno(ret);
7220 goto leave;
7223 ret = ocfs2_inode_lock(dir, &dir_bh, 0);
7224 if (ret) {
7225 mlog_errno(ret);
7226 goto leave;
7229 if (!ret && default_acl)
7230 ret = ocfs2_iop_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
7231 if (!ret && acl)
7232 ret = ocfs2_iop_set_acl(inode, acl, ACL_TYPE_ACCESS);
7234 ocfs2_inode_unlock(dir, 0);
7235 brelse(dir_bh);
7236 leave:
7237 return ret;
7240 * 'security' attributes support
7242 static size_t ocfs2_xattr_security_list(struct dentry *dentry, char *list,
7243 size_t list_size, const char *name,
7244 size_t name_len, int type)
7246 const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
7247 const size_t total_len = prefix_len + name_len + 1;
7249 if (list && total_len <= list_size) {
7250 memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
7251 memcpy(list + prefix_len, name, name_len);
7252 list[prefix_len + name_len] = '\0';
7254 return total_len;
7257 static int ocfs2_xattr_security_get(struct dentry *dentry, const char *name,
7258 void *buffer, size_t size, int type)
7260 if (strcmp(name, "") == 0)
7261 return -EINVAL;
7262 return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_SECURITY,
7263 name, buffer, size);
7266 static int ocfs2_xattr_security_set(struct dentry *dentry, const char *name,
7267 const void *value, size_t size, int flags, int type)
7269 if (strcmp(name, "") == 0)
7270 return -EINVAL;
7272 return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_SECURITY,
7273 name, value, size, flags);
7276 int ocfs2_initxattrs(struct inode *inode, const struct xattr *xattr_array,
7277 void *fs_info)
7279 const struct xattr *xattr;
7280 int err = 0;
7282 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
7283 err = ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY,
7284 xattr->name, xattr->value,
7285 xattr->value_len, XATTR_CREATE);
7286 if (err)
7287 break;
7289 return err;
7292 int ocfs2_init_security_get(struct inode *inode,
7293 struct inode *dir,
7294 const struct qstr *qstr,
7295 struct ocfs2_security_xattr_info *si)
7297 /* check whether ocfs2 support feature xattr */
7298 if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
7299 return -EOPNOTSUPP;
7300 if (si)
7301 return security_old_inode_init_security(inode, dir, qstr,
7302 &si->name, &si->value,
7303 &si->value_len);
7305 return security_inode_init_security(inode, dir, qstr,
7306 &ocfs2_initxattrs, NULL);
7309 int ocfs2_init_security_set(handle_t *handle,
7310 struct inode *inode,
7311 struct buffer_head *di_bh,
7312 struct ocfs2_security_xattr_info *si,
7313 struct ocfs2_alloc_context *xattr_ac,
7314 struct ocfs2_alloc_context *data_ac)
7316 return ocfs2_xattr_set_handle(handle, inode, di_bh,
7317 OCFS2_XATTR_INDEX_SECURITY,
7318 si->name, si->value, si->value_len, 0,
7319 xattr_ac, data_ac);
7322 const struct xattr_handler ocfs2_xattr_security_handler = {
7323 .prefix = XATTR_SECURITY_PREFIX,
7324 .list = ocfs2_xattr_security_list,
7325 .get = ocfs2_xattr_security_get,
7326 .set = ocfs2_xattr_security_set,
7330 * 'trusted' attributes support
7332 static size_t ocfs2_xattr_trusted_list(struct dentry *dentry, char *list,
7333 size_t list_size, const char *name,
7334 size_t name_len, int type)
7336 const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
7337 const size_t total_len = prefix_len + name_len + 1;
7339 if (list && total_len <= list_size) {
7340 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
7341 memcpy(list + prefix_len, name, name_len);
7342 list[prefix_len + name_len] = '\0';
7344 return total_len;
7347 static int ocfs2_xattr_trusted_get(struct dentry *dentry, const char *name,
7348 void *buffer, size_t size, int type)
7350 if (strcmp(name, "") == 0)
7351 return -EINVAL;
7352 return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_TRUSTED,
7353 name, buffer, size);
7356 static int ocfs2_xattr_trusted_set(struct dentry *dentry, const char *name,
7357 const void *value, size_t size, int flags, int type)
7359 if (strcmp(name, "") == 0)
7360 return -EINVAL;
7362 return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_TRUSTED,
7363 name, value, size, flags);
7366 const struct xattr_handler ocfs2_xattr_trusted_handler = {
7367 .prefix = XATTR_TRUSTED_PREFIX,
7368 .list = ocfs2_xattr_trusted_list,
7369 .get = ocfs2_xattr_trusted_get,
7370 .set = ocfs2_xattr_trusted_set,
7374 * 'user' attributes support
7376 static size_t ocfs2_xattr_user_list(struct dentry *dentry, char *list,
7377 size_t list_size, const char *name,
7378 size_t name_len, int type)
7380 const size_t prefix_len = XATTR_USER_PREFIX_LEN;
7381 const size_t total_len = prefix_len + name_len + 1;
7382 struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7384 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7385 return 0;
7387 if (list && total_len <= list_size) {
7388 memcpy(list, XATTR_USER_PREFIX, prefix_len);
7389 memcpy(list + prefix_len, name, name_len);
7390 list[prefix_len + name_len] = '\0';
7392 return total_len;
7395 static int ocfs2_xattr_user_get(struct dentry *dentry, const char *name,
7396 void *buffer, size_t size, int type)
7398 struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7400 if (strcmp(name, "") == 0)
7401 return -EINVAL;
7402 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7403 return -EOPNOTSUPP;
7404 return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_USER, name,
7405 buffer, size);
7408 static int ocfs2_xattr_user_set(struct dentry *dentry, const char *name,
7409 const void *value, size_t size, int flags, int type)
7411 struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7413 if (strcmp(name, "") == 0)
7414 return -EINVAL;
7415 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7416 return -EOPNOTSUPP;
7418 return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_USER,
7419 name, value, size, flags);
7422 const struct xattr_handler ocfs2_xattr_user_handler = {
7423 .prefix = XATTR_USER_PREFIX,
7424 .list = ocfs2_xattr_user_list,
7425 .get = ocfs2_xattr_user_get,
7426 .set = ocfs2_xattr_user_set,