projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
ocfs2: Add incompatible flag for extended attribute
[deliverable/linux.git] / fs / ocfs2 / xattr.c
CommitLineData
f56654c4
TM		 1/* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * xattr.c
5 *
6 * Copyright (C) 2008 Oracle. All rights reserved.
7 *
cf1d6c76
TY		 8 * CREDITS:
9 * Lots of code in this file is taken from ext3.
10 *
f56654c4
TM		 11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public
13 * License as published by the Free Software Foundation; either
14 * version 2 of the License, or (at your option) any later version.
15 *
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.
20 *
21 * You should have received a copy of the GNU General Public
22 * License along with this program; if not, write to the
23 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
24 * Boston, MA 021110-1307, USA.
25 */
26
cf1d6c76
TY		 27#include <linux/capability.h>
28#include <linux/fs.h>
29#include <linux/types.h>
30#include <linux/slab.h>
31#include <linux/highmem.h>
32#include <linux/pagemap.h>
33#include <linux/uio.h>
34#include <linux/sched.h>
35#include <linux/splice.h>
36#include <linux/mount.h>
37#include <linux/writeback.h>
38#include <linux/falloc.h>
01225596 39#include <linux/sort.h>
cf1d6c76 40
f56654c4
TM		 41#define MLOG_MASK_PREFIX ML_XATTR
42#include <cluster/masklog.h>
43
44#include "ocfs2.h"
45#include "alloc.h"
46#include "dlmglue.h"
47#include "file.h"
cf1d6c76
TY		 48#include "symlink.h"
49#include "sysfile.h"
f56654c4
TM		 50#include "inode.h"
51#include "journal.h"
52#include "ocfs2_fs.h"
53#include "suballoc.h"
54#include "uptodate.h"
55#include "buffer_head_io.h"
0c044f0b 56#include "super.h"
cf1d6c76
TY		 57#include "xattr.h"
58
59
60struct ocfs2_xattr_def_value_root {
61 struct ocfs2_xattr_value_root xv;
62 struct ocfs2_extent_rec er;
63};
64
0c044f0b
TM		 65struct ocfs2_xattr_bucket {
66 struct buffer_head *bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
67 struct ocfs2_xattr_header *xh;
68};
69
cf1d6c76
TY		 70#define OCFS2_XATTR_ROOT_SIZE (sizeof(struct ocfs2_xattr_def_value_root))
71#define OCFS2_XATTR_INLINE_SIZE 80
72
73static struct ocfs2_xattr_def_value_root def_xv = {
74 .xv.xr_list.l_count = cpu_to_le16(1),
75};
76
77struct xattr_handler *ocfs2_xattr_handlers[] = {
78 &ocfs2_xattr_user_handler,
79 &ocfs2_xattr_trusted_handler,
80 NULL
81};
82
83static struct xattr_handler *ocfs2_xattr_handler_map[] = {
84 [OCFS2_XATTR_INDEX_USER] = &ocfs2_xattr_user_handler,
85 [OCFS2_XATTR_INDEX_TRUSTED] = &ocfs2_xattr_trusted_handler,
86};
87
88struct ocfs2_xattr_info {
89 int name_index;
90 const char *name;
91 const void *value;
92 size_t value_len;
93};
94
95struct ocfs2_xattr_search {
96 struct buffer_head *inode_bh;
97 /*
98 * xattr_bh point to the block buffer head which has extended attribute
99 * when extended attribute in inode, xattr_bh is equal to inode_bh.
100 */
101 struct buffer_head *xattr_bh;
102 struct ocfs2_xattr_header *header;
589dc260 103 struct ocfs2_xattr_bucket bucket;
cf1d6c76
TY		 104 void *base;
105 void *end;
106 struct ocfs2_xattr_entry *here;
107 int not_found;
108};
109
589dc260
TM		 110static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
111 struct ocfs2_xattr_header *xh,
112 int index,
113 int *block_off,
114 int *new_offset);
115
116static int ocfs2_xattr_index_block_find(struct inode *inode,
117 struct buffer_head *root_bh,
118 int name_index,
119 const char *name,
120 struct ocfs2_xattr_search *xs);
121
0c044f0b
TM		 122static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
123 struct ocfs2_xattr_tree_root *xt,
124 char *buffer,
125 size_t buffer_size);
126
01225596
TM		 127static int ocfs2_xattr_create_index_block(struct inode *inode,
128 struct ocfs2_xattr_search *xs);
129
130static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
131 struct ocfs2_xattr_info *xi,
132 struct ocfs2_xattr_search *xs);
133
a3944256
TM		 134static int ocfs2_delete_xattr_index_block(struct inode *inode,
135 struct buffer_head *xb_bh);
136
cf1d6c76
TY		 137static inline struct xattr_handler *ocfs2_xattr_handler(int name_index)
138{
139 struct xattr_handler *handler = NULL;
140
141 if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
142 handler = ocfs2_xattr_handler_map[name_index];
143
144 return handler;
145}
146
147static inline u32 ocfs2_xattr_name_hash(struct inode *inode,
148 char *prefix,
149 int prefix_len,
150 char *name,
151 int name_len)
152{
153 /* Get hash value of uuid from super block */
154 u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
155 int i;
156
157 /* hash extended attribute prefix */
158 for (i = 0; i < prefix_len; i++) {
159 hash = (hash << OCFS2_HASH_SHIFT) ^
160 (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
161 *prefix++;
162 }
163 /* hash extended attribute name */
164 for (i = 0; i < name_len; i++) {
165 hash = (hash << OCFS2_HASH_SHIFT) ^
166 (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
167 *name++;
168 }
169
170 return hash;
171}
172
173/*
174 * ocfs2_xattr_hash_entry()
175 *
176 * Compute the hash of an extended attribute.
177 */
178static void ocfs2_xattr_hash_entry(struct inode *inode,
179 struct ocfs2_xattr_header *header,
180 struct ocfs2_xattr_entry *entry)
181{
182 u32 hash = 0;
183 struct xattr_handler *handler =
184 ocfs2_xattr_handler(ocfs2_xattr_get_type(entry));
185 char *prefix = handler->prefix;
186 char *name = (char *)header + le16_to_cpu(entry->xe_name_offset);
187 int prefix_len = strlen(handler->prefix);
188
189 hash = ocfs2_xattr_name_hash(inode, prefix, prefix_len, name,
190 entry->xe_name_len);
191 entry->xe_name_hash = cpu_to_le32(hash);
192
193 return;
194}
f56654c4
TM		 195
196static int ocfs2_xattr_extend_allocation(struct inode *inode,
197 u32 clusters_to_add,
198 struct buffer_head *xattr_bh,
199 struct ocfs2_xattr_value_root *xv)
200{
201 int status = 0;
202 int restart_func = 0;
203 int credits = 0;
204 handle_t *handle = NULL;
205 struct ocfs2_alloc_context *data_ac = NULL;
206 struct ocfs2_alloc_context *meta_ac = NULL;
207 enum ocfs2_alloc_restarted why;
208 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
209 struct ocfs2_extent_list *root_el = &xv->xr_list;
210 u32 prev_clusters, logical_start = le32_to_cpu(xv->xr_clusters);
211
212 mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
213
214restart_all:
215
216 status = ocfs2_lock_allocators(inode, xattr_bh, root_el,
217 clusters_to_add, 0, &data_ac,
218 &meta_ac, OCFS2_XATTR_VALUE_EXTENT, xv);
219 if (status) {
220 mlog_errno(status);
221 goto leave;
222 }
223
224 credits = ocfs2_calc_extend_credits(osb->sb, root_el, clusters_to_add);
225 handle = ocfs2_start_trans(osb, credits);
226 if (IS_ERR(handle)) {
227 status = PTR_ERR(handle);
228 handle = NULL;
229 mlog_errno(status);
230 goto leave;
231 }
232
233restarted_transaction:
234 status = ocfs2_journal_access(handle, inode, xattr_bh,
235 OCFS2_JOURNAL_ACCESS_WRITE);
236 if (status < 0) {
237 mlog_errno(status);
238 goto leave;
239 }
240
241 prev_clusters = le32_to_cpu(xv->xr_clusters);
242 status = ocfs2_add_clusters_in_btree(osb,
243 inode,
244 &logical_start,
245 clusters_to_add,
246 0,
247 xattr_bh,
248 root_el,
249 handle,
250 data_ac,
251 meta_ac,
252 &why,
253 OCFS2_XATTR_VALUE_EXTENT,
254 xv);
255 if ((status < 0) && (status != -EAGAIN)) {
256 if (status != -ENOSPC)
257 mlog_errno(status);
258 goto leave;
259 }
260
261 status = ocfs2_journal_dirty(handle, xattr_bh);
262 if (status < 0) {
263 mlog_errno(status);
264 goto leave;
265 }
266
267 clusters_to_add -= le32_to_cpu(xv->xr_clusters) - prev_clusters;
268
269 if (why != RESTART_NONE && clusters_to_add) {
270 if (why == RESTART_META) {
271 mlog(0, "restarting function.\n");
272 restart_func = 1;
273 } else {
274 BUG_ON(why != RESTART_TRANS);
275
276 mlog(0, "restarting transaction.\n");
277 /* TODO: This can be more intelligent. */
278 credits = ocfs2_calc_extend_credits(osb->sb,
279 root_el,
280 clusters_to_add);
281 status = ocfs2_extend_trans(handle, credits);
282 if (status < 0) {
283 /* handle still has to be committed at
284 * this point. */
285 status = -ENOMEM;
286 mlog_errno(status);
287 goto leave;
288 }
289 goto restarted_transaction;
290 }
291 }
292
293leave:
294 if (handle) {
295 ocfs2_commit_trans(osb, handle);
296 handle = NULL;
297 }
298 if (data_ac) {
299 ocfs2_free_alloc_context(data_ac);
300 data_ac = NULL;
301 }
302 if (meta_ac) {
303 ocfs2_free_alloc_context(meta_ac);
304 meta_ac = NULL;
305 }
306 if ((!status) && restart_func) {
307 restart_func = 0;
308 goto restart_all;
309 }
310
311 return status;
312}
313
314static int __ocfs2_remove_xattr_range(struct inode *inode,
315 struct buffer_head *root_bh,
316 struct ocfs2_xattr_value_root *xv,
317 u32 cpos, u32 phys_cpos, u32 len,
318 struct ocfs2_cached_dealloc_ctxt *dealloc)
319{
320 int ret;
321 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
322 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
323 struct inode *tl_inode = osb->osb_tl_inode;
324 handle_t *handle;
325 struct ocfs2_alloc_context *meta_ac = NULL;
326
327 ret = ocfs2_lock_allocators(inode, root_bh, &xv->xr_list,
328 0, 1, NULL, &meta_ac,
329 OCFS2_XATTR_VALUE_EXTENT, xv);
330 if (ret) {
331 mlog_errno(ret);
332 return ret;
333 }
334
335 mutex_lock(&tl_inode->i_mutex);
336
337 if (ocfs2_truncate_log_needs_flush(osb)) {
338 ret = __ocfs2_flush_truncate_log(osb);
339 if (ret < 0) {
340 mlog_errno(ret);
341 goto out;
342 }
343 }
344
345 handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
346 if (IS_ERR(handle)) {
347 ret = PTR_ERR(handle);
348 mlog_errno(ret);
349 goto out;
350 }
351
352 ret = ocfs2_journal_access(handle, inode, root_bh,
353 OCFS2_JOURNAL_ACCESS_WRITE);
354 if (ret) {
355 mlog_errno(ret);
356 goto out_commit;
357 }
358
359 ret = ocfs2_remove_extent(inode, root_bh, cpos, len, handle, meta_ac,
360 dealloc, OCFS2_XATTR_VALUE_EXTENT, xv);
361 if (ret) {
362 mlog_errno(ret);
363 goto out_commit;
364 }
365
366 le32_add_cpu(&xv->xr_clusters, -len);
367
368 ret = ocfs2_journal_dirty(handle, root_bh);
369 if (ret) {
370 mlog_errno(ret);
371 goto out_commit;
372 }
373
374 ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len);
375 if (ret)
376 mlog_errno(ret);
377
378out_commit:
379 ocfs2_commit_trans(osb, handle);
380out:
381 mutex_unlock(&tl_inode->i_mutex);
382
383 if (meta_ac)
384 ocfs2_free_alloc_context(meta_ac);
385
386 return ret;
387}
388
389static int ocfs2_xattr_shrink_size(struct inode *inode,
390 u32 old_clusters,
391 u32 new_clusters,
392 struct buffer_head *root_bh,
393 struct ocfs2_xattr_value_root *xv)
394{
395 int ret = 0;
396 u32 trunc_len, cpos, phys_cpos, alloc_size;
397 u64 block;
398 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
399 struct ocfs2_cached_dealloc_ctxt dealloc;
400
401 ocfs2_init_dealloc_ctxt(&dealloc);
402
403 if (old_clusters <= new_clusters)
404 return 0;
405
406 cpos = new_clusters;
407 trunc_len = old_clusters - new_clusters;
408 while (trunc_len) {
409 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
410 &alloc_size, &xv->xr_list);
411 if (ret) {
412 mlog_errno(ret);
413 goto out;
414 }
415
416 if (alloc_size > trunc_len)
417 alloc_size = trunc_len;
418
419 ret = __ocfs2_remove_xattr_range(inode, root_bh, xv, cpos,
420 phys_cpos, alloc_size,
421 &dealloc);
422 if (ret) {
423 mlog_errno(ret);
424 goto out;
425 }
426
427 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
428 ocfs2_remove_xattr_clusters_from_cache(inode, block,
429 alloc_size);
430 cpos += alloc_size;
431 trunc_len -= alloc_size;
432 }
433
434out:
435 ocfs2_schedule_truncate_log_flush(osb, 1);
436 ocfs2_run_deallocs(osb, &dealloc);
437
438 return ret;
439}
440
441static int ocfs2_xattr_value_truncate(struct inode *inode,
442 struct buffer_head *root_bh,
443 struct ocfs2_xattr_value_root *xv,
444 int len)
445{
446 int ret;
447 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
448 u32 old_clusters = le32_to_cpu(xv->xr_clusters);
449
450 if (new_clusters == old_clusters)
451 return 0;
452
453 if (new_clusters > old_clusters)
454 ret = ocfs2_xattr_extend_allocation(inode,
455 new_clusters - old_clusters,
456 root_bh, xv);
457 else
458 ret = ocfs2_xattr_shrink_size(inode,
459 old_clusters, new_clusters,
460 root_bh, xv);
461
462 return ret;
463}
cf1d6c76
TY		 464
465static int ocfs2_xattr_list_entries(struct inode *inode,
466 struct ocfs2_xattr_header *header,
467 char *buffer, size_t buffer_size)
468{
469 size_t rest = buffer_size;
470 int i;
471
472 for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
473 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
474 struct xattr_handler *handler =
475 ocfs2_xattr_handler(ocfs2_xattr_get_type(entry));
476
477 if (handler) {
478 size_t size = handler->list(inode, buffer, rest,
479 ((char *)header +
480 le16_to_cpu(entry->xe_name_offset)),
481 entry->xe_name_len);
482 if (buffer) {
483 if (size > rest)
484 return -ERANGE;
485 buffer += size;
486 }
487 rest -= size;
488 }
489 }
490
491 return buffer_size - rest;
492}
493
494static int ocfs2_xattr_ibody_list(struct inode *inode,
495 struct ocfs2_dinode *di,
496 char *buffer,
497 size_t buffer_size)
498{
499 struct ocfs2_xattr_header *header = NULL;
500 struct ocfs2_inode_info *oi = OCFS2_I(inode);
501 int ret = 0;
502
503 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
504 return ret;
505
506 header = (struct ocfs2_xattr_header *)
507 ((void *)di + inode->i_sb->s_blocksize -
508 le16_to_cpu(di->i_xattr_inline_size));
509
510 ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
511
512 return ret;
513}
514
515static int ocfs2_xattr_block_list(struct inode *inode,
516 struct ocfs2_dinode *di,
517 char *buffer,
518 size_t buffer_size)
519{
520 struct buffer_head *blk_bh = NULL;
0c044f0b 521 struct ocfs2_xattr_block *xb;
cf1d6c76
TY		 522 int ret = 0;
523
524 if (!di->i_xattr_loc)
525 return ret;
526
527 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
528 le64_to_cpu(di->i_xattr_loc),
529 &blk_bh, OCFS2_BH_CACHED, inode);
530 if (ret < 0) {
531 mlog_errno(ret);
532 return ret;
533 }
534 /*Verify the signature of xattr block*/
535 if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
536 strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
537 ret = -EFAULT;
538 goto cleanup;
539 }
540
0c044f0b 541 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
cf1d6c76 542
0c044f0b
TM		 543 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
544 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
545 ret = ocfs2_xattr_list_entries(inode, header,
546 buffer, buffer_size);
547 } else {
548 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
549 ret = ocfs2_xattr_tree_list_index_block(inode, xt,
550 buffer, buffer_size);
551 }
cf1d6c76
TY		 552cleanup:
553 brelse(blk_bh);
554
555 return ret;
556}
557
558ssize_t ocfs2_listxattr(struct dentry *dentry,
559 char *buffer,
560 size_t size)
561{
562 int ret = 0, i_ret = 0, b_ret = 0;
563 struct buffer_head *di_bh = NULL;
564 struct ocfs2_dinode *di = NULL;
565 struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
566
8154da3d
TY		 567 if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
568 return -EOPNOTSUPP;
569
cf1d6c76
TY		 570 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
571 return ret;
572
573 ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
574 if (ret < 0) {
575 mlog_errno(ret);
576 return ret;
577 }
578
579 di = (struct ocfs2_dinode *)di_bh->b_data;
580
581 down_read(&oi->ip_xattr_sem);
582 i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
583 if (i_ret < 0)
584 b_ret = 0;
585 else {
586 if (buffer) {
587 buffer += i_ret;
588 size -= i_ret;
589 }
590 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
591 buffer, size);
592 if (b_ret < 0)
593 i_ret = 0;
594 }
595 up_read(&oi->ip_xattr_sem);
596 ocfs2_inode_unlock(dentry->d_inode, 0);
597
598 brelse(di_bh);
599
600 return i_ret + b_ret;
601}
602
603static int ocfs2_xattr_find_entry(int name_index,
604 const char *name,
605 struct ocfs2_xattr_search *xs)
606{
607 struct ocfs2_xattr_entry *entry;
608 size_t name_len;
609 int i, cmp = 1;
610
611 if (name == NULL)
612 return -EINVAL;
613
614 name_len = strlen(name);
615 entry = xs->here;
616 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
617 cmp = name_index - ocfs2_xattr_get_type(entry);
618 if (!cmp)
619 cmp = name_len - entry->xe_name_len;
620 if (!cmp)
621 cmp = memcmp(name, (xs->base +
622 le16_to_cpu(entry->xe_name_offset)),
623 name_len);
624 if (cmp == 0)
625 break;
626 entry += 1;
627 }
628 xs->here = entry;
629
630 return cmp ? -ENODATA : 0;
631}
632
633static int ocfs2_xattr_get_value_outside(struct inode *inode,
589dc260 634 struct ocfs2_xattr_value_root *xv,
cf1d6c76
TY		 635 void *buffer,
636 size_t len)
637{
638 u32 cpos, p_cluster, num_clusters, bpc, clusters;
639 u64 blkno;
640 int i, ret = 0;
641 size_t cplen, blocksize;
642 struct buffer_head *bh = NULL;
cf1d6c76
TY		 643 struct ocfs2_extent_list *el;
644
cf1d6c76
TY		 645 el = &xv->xr_list;
646 clusters = le32_to_cpu(xv->xr_clusters);
647 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
648 blocksize = inode->i_sb->s_blocksize;
649
650 cpos = 0;
651 while (cpos < clusters) {
652 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
653 &num_clusters, el);
654 if (ret) {
655 mlog_errno(ret);
656 goto out;
657 }
658
659 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
660 /* Copy ocfs2_xattr_value */
661 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
662 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), blkno,
663 &bh, OCFS2_BH_CACHED, inode);
664 if (ret) {
665 mlog_errno(ret);
666 goto out;
667 }
668
669 cplen = len >= blocksize ? blocksize : len;
670 memcpy(buffer, bh->b_data, cplen);
671 len -= cplen;
672 buffer += cplen;
673
674 brelse(bh);
675 bh = NULL;
676 if (len == 0)
677 break;
678 }
679 cpos += num_clusters;
680 }
681out:
682 return ret;
683}
684
685static int ocfs2_xattr_ibody_get(struct inode *inode,
686 int name_index,
687 const char *name,
688 void *buffer,
689 size_t buffer_size,
690 struct ocfs2_xattr_search *xs)
691{
692 struct ocfs2_inode_info *oi = OCFS2_I(inode);
693 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
589dc260 694 struct ocfs2_xattr_value_root *xv;
cf1d6c76
TY		 695 size_t size;
696 int ret = 0;
697
698 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
699 return -ENODATA;
700
701 xs->end = (void *)di + inode->i_sb->s_blocksize;
702 xs->header = (struct ocfs2_xattr_header *)
703 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
704 xs->base = (void *)xs->header;
705 xs->here = xs->header->xh_entries;
706
707 ret = ocfs2_xattr_find_entry(name_index, name, xs);
708 if (ret)
709 return ret;
710 size = le64_to_cpu(xs->here->xe_value_size);
711 if (buffer) {
712 if (size > buffer_size)
713 return -ERANGE;
714 if (ocfs2_xattr_is_local(xs->here)) {
715 memcpy(buffer, (void *)xs->base +
716 le16_to_cpu(xs->here->xe_name_offset) +
717 OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
718 } else {
589dc260
TM		 719 xv = (struct ocfs2_xattr_value_root *)
720 (xs->base + le16_to_cpu(
721 xs->here->xe_name_offset) +
722 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
723 ret = ocfs2_xattr_get_value_outside(inode, xv,
cf1d6c76
TY		 724 buffer, size);
725 if (ret < 0) {
726 mlog_errno(ret);
727 return ret;
728 }
729 }
730 }
731
732 return size;
733}
734
735static int ocfs2_xattr_block_get(struct inode *inode,
736 int name_index,
737 const char *name,
738 void *buffer,
739 size_t buffer_size,
740 struct ocfs2_xattr_search *xs)
741{
742 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
743 struct buffer_head *blk_bh = NULL;
744 struct ocfs2_xattr_block *xb;
589dc260 745 struct ocfs2_xattr_value_root *xv;
cf1d6c76 746 size_t size;
589dc260 747 int ret = -ENODATA, name_offset, name_len, block_off, i;
cf1d6c76
TY		 748
749 if (!di->i_xattr_loc)
750 return ret;
751
589dc260
TM		 752 memset(&xs->bucket, 0, sizeof(xs->bucket));
753
cf1d6c76
TY		 754 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
755 le64_to_cpu(di->i_xattr_loc),
756 &blk_bh, OCFS2_BH_CACHED, inode);
757 if (ret < 0) {
758 mlog_errno(ret);
759 return ret;
760 }
761 /*Verify the signature of xattr block*/
762 if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
763 strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
764 ret = -EFAULT;
765 goto cleanup;
766 }
767
768 xs->xattr_bh = blk_bh;
769 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
cf1d6c76 770
589dc260
TM		 771 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
772 xs->header = &xb->xb_attrs.xb_header;
773 xs->base = (void *)xs->header;
774 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
775 xs->here = xs->header->xh_entries;
776
777 ret = ocfs2_xattr_find_entry(name_index, name, xs);
778 } else
779 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
780 name_index,
781 name, xs);
782
cf1d6c76
TY		 783 if (ret)
784 goto cleanup;
785 size = le64_to_cpu(xs->here->xe_value_size);
786 if (buffer) {
787 ret = -ERANGE;
788 if (size > buffer_size)
789 goto cleanup;
589dc260
TM		 790
791 name_offset = le16_to_cpu(xs->here->xe_name_offset);
792 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
793 i = xs->here - xs->header->xh_entries;
794
795 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
796 ret = ocfs2_xattr_bucket_get_name_value(inode,
797 xs->bucket.xh,
798 i,
799 &block_off,
800 &name_offset);
801 xs->base = xs->bucket.bhs[block_off]->b_data;
802 }
cf1d6c76
TY		 803 if (ocfs2_xattr_is_local(xs->here)) {
804 memcpy(buffer, (void *)xs->base +
589dc260 805 name_offset + name_len, size);
cf1d6c76 806 } else {
589dc260
TM		 807 xv = (struct ocfs2_xattr_value_root *)
808 (xs->base + name_offset + name_len);
809 ret = ocfs2_xattr_get_value_outside(inode, xv,
cf1d6c76
TY		 810 buffer, size);
811 if (ret < 0) {
812 mlog_errno(ret);
813 goto cleanup;
814 }
815 }
816 }
817 ret = size;
818cleanup:
589dc260
TM		 819 for (i = 0; i < OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET; i++)
820 brelse(xs->bucket.bhs[i]);
821 memset(&xs->bucket, 0, sizeof(xs->bucket));
cf1d6c76 822
589dc260 823 brelse(blk_bh);
cf1d6c76
TY		 824 return ret;
825}
826
827/* ocfs2_xattr_get()
828 *
829 * Copy an extended attribute into the buffer provided.
830 * Buffer is NULL to compute the size of buffer required.
831 */
832int ocfs2_xattr_get(struct inode *inode,
833 int name_index,
834 const char *name,
835 void *buffer,
836 size_t buffer_size)
837{
838 int ret;
839 struct ocfs2_dinode *di = NULL;
840 struct buffer_head *di_bh = NULL;
841 struct ocfs2_inode_info *oi = OCFS2_I(inode);
842 struct ocfs2_xattr_search xis = {
843 .not_found = -ENODATA,
844 };
845 struct ocfs2_xattr_search xbs = {
846 .not_found = -ENODATA,
847 };
848
8154da3d
TY		 849 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
850 return -EOPNOTSUPP;
851
cf1d6c76
TY		 852 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
853 ret = -ENODATA;
854
855 ret = ocfs2_inode_lock(inode, &di_bh, 0);
856 if (ret < 0) {
857 mlog_errno(ret);
858 return ret;
859 }
860 xis.inode_bh = xbs.inode_bh = di_bh;
861 di = (struct ocfs2_dinode *)di_bh->b_data;
862
863 down_read(&oi->ip_xattr_sem);
864 ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
865 buffer_size, &xis);
866 if (ret == -ENODATA)
867 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
868 buffer_size, &xbs);
869 up_read(&oi->ip_xattr_sem);
870 ocfs2_inode_unlock(inode, 0);
871
872 brelse(di_bh);
873
874 return ret;
875}
876
877static int __ocfs2_xattr_set_value_outside(struct inode *inode,
878 struct ocfs2_xattr_value_root *xv,
879 const void *value,
880 int value_len)
881{
882 int ret = 0, i, cp_len, credits;
883 u16 blocksize = inode->i_sb->s_blocksize;
884 u32 p_cluster, num_clusters;
885 u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
886 u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
887 u64 blkno;
888 struct buffer_head *bh = NULL;
889 handle_t *handle;
890
891 BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
892
893 credits = clusters * bpc;
894 handle = ocfs2_start_trans(OCFS2_SB(inode->i_sb), credits);
895 if (IS_ERR(handle)) {
896 ret = PTR_ERR(handle);
897 mlog_errno(ret);
898 goto out;
899 }
900
901 while (cpos < clusters) {
902 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
903 &num_clusters, &xv->xr_list);
904 if (ret) {
905 mlog_errno(ret);
906 goto out_commit;
907 }
908
909 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
910
911 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
912 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), blkno,
913 &bh, OCFS2_BH_CACHED, inode);
914 if (ret) {
915 mlog_errno(ret);
916 goto out_commit;
917 }
918
919 ret = ocfs2_journal_access(handle,
920 inode,
921 bh,
922 OCFS2_JOURNAL_ACCESS_WRITE);
923 if (ret < 0) {
924 mlog_errno(ret);
925 goto out_commit;
926 }
927
928 cp_len = value_len > blocksize ? blocksize : value_len;
929 memcpy(bh->b_data, value, cp_len);
930 value_len -= cp_len;
931 value += cp_len;
932 if (cp_len < blocksize)
933 memset(bh->b_data + cp_len, 0,
934 blocksize - cp_len);
935
936 ret = ocfs2_journal_dirty(handle, bh);
937 if (ret < 0) {
938 mlog_errno(ret);
939 goto out_commit;
940 }
941 brelse(bh);
942 bh = NULL;
943
944 /*
945 * XXX: do we need to empty all the following
946 * blocks in this cluster?
947 */
948 if (!value_len)
949 break;
950 }
951 cpos += num_clusters;
952 }
953out_commit:
954 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
955out:
956 brelse(bh);
957
958 return ret;
959}
960
961static int ocfs2_xattr_cleanup(struct inode *inode,
962 struct ocfs2_xattr_info *xi,
963 struct ocfs2_xattr_search *xs,
964 size_t offs)
965{
966 handle_t *handle = NULL;
967 int ret = 0;
968 size_t name_len = strlen(xi->name);
969 void *val = xs->base + offs;
970 size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
971
972 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
973 OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
974 if (IS_ERR(handle)) {
975 ret = PTR_ERR(handle);
976 mlog_errno(ret);
977 goto out;
978 }
979 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
980 OCFS2_JOURNAL_ACCESS_WRITE);
981 if (ret) {
982 mlog_errno(ret);
983 goto out_commit;
984 }
985 /* Decrease xattr count */
986 le16_add_cpu(&xs->header->xh_count, -1);
987 /* Remove the xattr entry and tree root which has already be set*/
988 memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
989 memset(val, 0, size);
990
991 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
992 if (ret < 0)
993 mlog_errno(ret);
994out_commit:
995 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
996out:
997 return ret;
998}
999
1000static int ocfs2_xattr_update_entry(struct inode *inode,
1001 struct ocfs2_xattr_info *xi,
1002 struct ocfs2_xattr_search *xs,
1003 size_t offs)
1004{
1005 handle_t *handle = NULL;
1006 int ret = 0;
1007
1008 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1009 OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1010 if (IS_ERR(handle)) {
1011 ret = PTR_ERR(handle);
1012 mlog_errno(ret);
1013 goto out;
1014 }
1015 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1016 OCFS2_JOURNAL_ACCESS_WRITE);
1017 if (ret) {
1018 mlog_errno(ret);
1019 goto out_commit;
1020 }
1021
1022 xs->here->xe_name_offset = cpu_to_le16(offs);
1023 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1024 if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1025 ocfs2_xattr_set_local(xs->here, 1);
1026 else
1027 ocfs2_xattr_set_local(xs->here, 0);
1028 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1029
1030 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1031 if (ret < 0)
1032 mlog_errno(ret);
1033out_commit:
1034 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1035out:
1036 return ret;
1037}
1038
1039/*
1040 * ocfs2_xattr_set_value_outside()
1041 *
1042 * Set large size value in B tree.
1043 */
1044static int ocfs2_xattr_set_value_outside(struct inode *inode,
1045 struct ocfs2_xattr_info *xi,
1046 struct ocfs2_xattr_search *xs,
1047 size_t offs)
1048{
1049 size_t name_len = strlen(xi->name);
1050 void *val = xs->base + offs;
1051 struct ocfs2_xattr_value_root *xv = NULL;
1052 size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1053 int ret = 0;
1054
1055 memset(val, 0, size);
1056 memcpy(val, xi->name, name_len);
1057 xv = (struct ocfs2_xattr_value_root *)
1058 (val + OCFS2_XATTR_SIZE(name_len));
1059 xv->xr_clusters = 0;
1060 xv->xr_last_eb_blk = 0;
1061 xv->xr_list.l_tree_depth = 0;
1062 xv->xr_list.l_count = cpu_to_le16(1);
1063 xv->xr_list.l_next_free_rec = 0;
1064
1065 ret = ocfs2_xattr_value_truncate(inode, xs->xattr_bh, xv,
1066 xi->value_len);
1067 if (ret < 0) {
1068 mlog_errno(ret);
1069 return ret;
1070 }
1071 ret = __ocfs2_xattr_set_value_outside(inode, xv, xi->value,
1072 xi->value_len);
1073 if (ret < 0) {
1074 mlog_errno(ret);
1075 return ret;
1076 }
1077 ret = ocfs2_xattr_update_entry(inode, xi, xs, offs);
1078 if (ret < 0)
1079 mlog_errno(ret);
1080
1081 return ret;
1082}
1083
1084/*
1085 * ocfs2_xattr_set_entry_local()
1086 *
1087 * Set, replace or remove extended attribute in local.
1088 */
1089static void ocfs2_xattr_set_entry_local(struct inode *inode,
1090 struct ocfs2_xattr_info *xi,
1091 struct ocfs2_xattr_search *xs,
1092 struct ocfs2_xattr_entry *last,
1093 size_t min_offs)
1094{
1095 size_t name_len = strlen(xi->name);
1096 int i;
1097
1098 if (xi->value && xs->not_found) {
1099 /* Insert the new xattr entry. */
1100 le16_add_cpu(&xs->header->xh_count, 1);
1101 ocfs2_xattr_set_type(last, xi->name_index);
1102 ocfs2_xattr_set_local(last, 1);
1103 last->xe_name_len = name_len;
1104 } else {
1105 void *first_val;
1106 void *val;
1107 size_t offs, size;
1108
1109 first_val = xs->base + min_offs;
1110 offs = le16_to_cpu(xs->here->xe_name_offset);
1111 val = xs->base + offs;
1112
1113 if (le64_to_cpu(xs->here->xe_value_size) >
1114 OCFS2_XATTR_INLINE_SIZE)
1115 size = OCFS2_XATTR_SIZE(name_len) +
1116 OCFS2_XATTR_ROOT_SIZE;
1117 else
1118 size = OCFS2_XATTR_SIZE(name_len) +
1119 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1120
1121 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1122 OCFS2_XATTR_SIZE(xi->value_len)) {
1123 /* The old and the new value have the
1124 same size. Just replace the value. */
1125 ocfs2_xattr_set_local(xs->here, 1);
1126 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1127 /* Clear value bytes. */
1128 memset(val + OCFS2_XATTR_SIZE(name_len),
1129 0,
1130 OCFS2_XATTR_SIZE(xi->value_len));
1131 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1132 xi->value,
1133 xi->value_len);
1134 return;
1135 }
1136 /* Remove the old name+value. */
1137 memmove(first_val + size, first_val, val - first_val);
1138 memset(first_val, 0, size);
1139 xs->here->xe_name_hash = 0;
1140 xs->here->xe_name_offset = 0;
1141 ocfs2_xattr_set_local(xs->here, 1);
1142 xs->here->xe_value_size = 0;
1143
1144 min_offs += size;
1145
1146 /* Adjust all value offsets. */
1147 last = xs->header->xh_entries;
1148 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1149 size_t o = le16_to_cpu(last->xe_name_offset);
1150
1151 if (o < offs)
1152 last->xe_name_offset = cpu_to_le16(o + size);
1153 last += 1;
1154 }
1155
1156 if (!xi->value) {
1157 /* Remove the old entry. */
1158 last -= 1;
1159 memmove(xs->here, xs->here + 1,
1160 (void *)last - (void *)xs->here);
1161 memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1162 le16_add_cpu(&xs->header->xh_count, -1);
1163 }
1164 }
1165 if (xi->value) {
1166 /* Insert the new name+value. */
1167 size_t size = OCFS2_XATTR_SIZE(name_len) +
1168 OCFS2_XATTR_SIZE(xi->value_len);
1169 void *val = xs->base + min_offs - size;
1170
1171 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1172 memset(val, 0, size);
1173 memcpy(val, xi->name, name_len);
1174 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1175 xi->value,
1176 xi->value_len);
1177 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1178 ocfs2_xattr_set_local(xs->here, 1);
1179 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1180 }
1181
1182 return;
1183}
1184
1185/*
1186 * ocfs2_xattr_set_entry()
1187 *
1188 * Set extended attribute entry into inode or block.
1189 *
1190 * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1191 * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1192 * then set value in B tree with set_value_outside().
1193 */
1194static int ocfs2_xattr_set_entry(struct inode *inode,
1195 struct ocfs2_xattr_info *xi,
1196 struct ocfs2_xattr_search *xs,
1197 int flag)
1198{
1199 struct ocfs2_xattr_entry *last;
1200 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1201 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1202 size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1203 size_t size_l = 0;
1204 handle_t *handle = NULL;
1205 int free, i, ret;
1206 struct ocfs2_xattr_info xi_l = {
1207 .name_index = xi->name_index,
1208 .name = xi->name,
1209 .value = xi->value,
1210 .value_len = xi->value_len,
1211 };
1212
1213 /* Compute min_offs, last and free space. */
1214 last = xs->header->xh_entries;
1215
1216 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1217 size_t offs = le16_to_cpu(last->xe_name_offset);
1218 if (offs < min_offs)
1219 min_offs = offs;
1220 last += 1;
1221 }
1222
1223 free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
1224 if (free < 0)
1225 return -EFAULT;
1226
1227 if (!xs->not_found) {
1228 size_t size = 0;
1229 if (ocfs2_xattr_is_local(xs->here))
1230 size = OCFS2_XATTR_SIZE(name_len) +
1231 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1232 else
1233 size = OCFS2_XATTR_SIZE(name_len) +
1234 OCFS2_XATTR_ROOT_SIZE;
1235 free += (size + sizeof(struct ocfs2_xattr_entry));
1236 }
1237 /* Check free space in inode or block */
1238 if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1239 if (free < sizeof(struct ocfs2_xattr_entry) +
1240 OCFS2_XATTR_SIZE(name_len) +
1241 OCFS2_XATTR_ROOT_SIZE) {
1242 ret = -ENOSPC;
1243 goto out;
1244 }
1245 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1246 xi_l.value = (void *)&def_xv;
1247 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1248 } else if (xi->value) {
1249 if (free < sizeof(struct ocfs2_xattr_entry) +
1250 OCFS2_XATTR_SIZE(name_len) +
1251 OCFS2_XATTR_SIZE(xi->value_len)) {
1252 ret = -ENOSPC;
1253 goto out;
1254 }
1255 }
1256
1257 if (!xs->not_found) {
1258 /* For existing extended attribute */
1259 size_t size = OCFS2_XATTR_SIZE(name_len) +
1260 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1261 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1262 void *val = xs->base + offs;
1263
1264 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1265 /* Replace existing local xattr with tree root */
1266 ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1267 offs);
1268 if (ret < 0)
1269 mlog_errno(ret);
1270 goto out;
1271 } else if (!ocfs2_xattr_is_local(xs->here)) {
1272 /* For existing xattr which has value outside */
1273 struct ocfs2_xattr_value_root *xv = NULL;
1274 xv = (struct ocfs2_xattr_value_root *)(val +
1275 OCFS2_XATTR_SIZE(name_len));
1276
1277 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1278 /*
1279 * If new value need set outside also,
1280 * first truncate old value to new value,
1281 * then set new value with set_value_outside().
1282 */
1283 ret = ocfs2_xattr_value_truncate(inode,
1284 xs->xattr_bh,
1285 xv,
1286 xi->value_len);
1287 if (ret < 0) {
1288 mlog_errno(ret);
1289 goto out;
1290 }
1291
1292 ret = __ocfs2_xattr_set_value_outside(inode,
1293 xv,
1294 xi->value,
1295 xi->value_len);
1296 if (ret < 0) {
1297 mlog_errno(ret);
1298 goto out;
1299 }
1300
1301 ret = ocfs2_xattr_update_entry(inode,
1302 xi,
1303 xs,
1304 offs);
1305 if (ret < 0)
1306 mlog_errno(ret);
1307 goto out;
1308 } else {
1309 /*
1310 * If new value need set in local,
1311 * just trucate old value to zero.
1312 */
1313 ret = ocfs2_xattr_value_truncate(inode,
1314 xs->xattr_bh,
1315 xv,
1316 0);
1317 if (ret < 0)
1318 mlog_errno(ret);
1319 }
1320 }
1321 }
1322
1323 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1324 OCFS2_INODE_UPDATE_CREDITS);
1325 if (IS_ERR(handle)) {
1326 ret = PTR_ERR(handle);
1327 mlog_errno(ret);
1328 goto out;
1329 }
1330
1331 ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
1332 OCFS2_JOURNAL_ACCESS_WRITE);
1333 if (ret) {
1334 mlog_errno(ret);
1335 goto out_commit;
1336 }
1337
1338 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1339 /*set extended attribue in external blcok*/
1340 ret = ocfs2_extend_trans(handle,
1341 OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1342 if (ret) {
1343 mlog_errno(ret);
1344 goto out_commit;
1345 }
1346 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1347 OCFS2_JOURNAL_ACCESS_WRITE);
1348 if (ret) {
1349 mlog_errno(ret);
1350 goto out_commit;
1351 }
1352 }
1353
1354 /*
1355 * Set value in local, include set tree root in local.
1356 * This is the first step for value size >INLINE_SIZE.
1357 */
1358 ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1359
1360 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1361 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1362 if (ret < 0) {
1363 mlog_errno(ret);
1364 goto out_commit;
1365 }
1366 }
1367
1368 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1369 (flag & OCFS2_INLINE_XATTR_FL)) {
1370 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1371 unsigned int xattrsize = osb->s_xattr_inline_size;
1372
1373 /*
1374 * Adjust extent record count or inline data size
1375 * to reserve space for extended attribute.
1376 */
1377 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1378 struct ocfs2_inline_data *idata = &di->id2.i_data;
1379 le16_add_cpu(&idata->id_count, -xattrsize);
1380 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1381 struct ocfs2_extent_list *el = &di->id2.i_list;
1382 le16_add_cpu(&el->l_count, -(xattrsize /
1383 sizeof(struct ocfs2_extent_rec)));
1384 }
1385 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1386 }
1387 /* Update xattr flag */
1388 spin_lock(&oi->ip_lock);
1389 oi->ip_dyn_features |= flag;
1390 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1391 spin_unlock(&oi->ip_lock);
1392 /* Update inode ctime */
1393 inode->i_ctime = CURRENT_TIME;
1394 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1395 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1396
1397 ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1398 if (ret < 0)
1399 mlog_errno(ret);
1400
1401out_commit:
1402 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1403
1404 if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1405 /*
1406 * Set value outside in B tree.
1407 * This is the second step for value size > INLINE_SIZE.
1408 */
1409 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1410 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, offs);
1411 if (ret < 0) {
1412 int ret2;
1413
1414 mlog_errno(ret);
1415 /*
1416 * If set value outside failed, we have to clean
1417 * the junk tree root we have already set in local.
1418 */
1419 ret2 = ocfs2_xattr_cleanup(inode, xi, xs, offs);
1420 if (ret2 < 0)
1421 mlog_errno(ret2);
1422 }
1423 }
1424out:
1425 return ret;
1426
1427}
1428
1429static int ocfs2_xattr_free_block(handle_t *handle,
1430 struct ocfs2_super *osb,
1431 struct ocfs2_xattr_block *xb)
1432{
1433 struct inode *xb_alloc_inode;
1434 struct buffer_head *xb_alloc_bh = NULL;
1435 u64 blk = le64_to_cpu(xb->xb_blkno);
1436 u16 bit = le16_to_cpu(xb->xb_suballoc_bit);
1437 u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1438 int ret = 0;
1439
1440 xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1441 EXTENT_ALLOC_SYSTEM_INODE,
1442 le16_to_cpu(xb->xb_suballoc_slot));
1443 if (!xb_alloc_inode) {
1444 ret = -ENOMEM;
1445 mlog_errno(ret);
1446 goto out;
1447 }
1448 mutex_lock(&xb_alloc_inode->i_mutex);
1449
1450 ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1451 if (ret < 0) {
1452 mlog_errno(ret);
1453 goto out_mutex;
1454 }
1455 ret = ocfs2_extend_trans(handle, OCFS2_SUBALLOC_FREE);
1456 if (ret < 0) {
1457 mlog_errno(ret);
1458 goto out_unlock;
1459 }
1460 ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1461 bit, bg_blkno, 1);
1462 if (ret < 0)
1463 mlog_errno(ret);
1464out_unlock:
1465 ocfs2_inode_unlock(xb_alloc_inode, 1);
1466 brelse(xb_alloc_bh);
1467out_mutex:
1468 mutex_unlock(&xb_alloc_inode->i_mutex);
1469 iput(xb_alloc_inode);
1470out:
1471 return ret;
1472}
1473
1474static int ocfs2_remove_value_outside(struct inode*inode,
1475 struct buffer_head *bh,
1476 struct ocfs2_xattr_header *header)
1477{
1478 int ret = 0, i;
1479
1480 for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1481 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1482
1483 if (!ocfs2_xattr_is_local(entry)) {
1484 struct ocfs2_xattr_value_root *xv;
1485 void *val;
1486
1487 val = (void *)header +
1488 le16_to_cpu(entry->xe_name_offset);
1489 xv = (struct ocfs2_xattr_value_root *)
1490 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
1491 ret = ocfs2_xattr_value_truncate(inode, bh, xv, 0);
1492 if (ret < 0) {
1493 mlog_errno(ret);
1494 return ret;
1495 }
1496 }
1497 }
1498
1499 return ret;
1500}
1501
1502static int ocfs2_xattr_ibody_remove(struct inode *inode,
1503 struct buffer_head *di_bh)
1504{
1505
1506 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1507 struct ocfs2_xattr_header *header;
1508 int ret;
1509
1510 header = (struct ocfs2_xattr_header *)
1511 ((void *)di + inode->i_sb->s_blocksize -
1512 le16_to_cpu(di->i_xattr_inline_size));
1513
1514 ret = ocfs2_remove_value_outside(inode, di_bh, header);
1515
1516 return ret;
1517}
1518
1519static int ocfs2_xattr_block_remove(struct inode *inode,
1520 struct buffer_head *blk_bh)
1521{
1522 struct ocfs2_xattr_block *xb;
cf1d6c76
TY		 1523 int ret = 0;
1524
1525 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
a3944256
TM		 1526 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1527 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1528 ret = ocfs2_remove_value_outside(inode, blk_bh, header);
1529 } else
1530 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
cf1d6c76
TY		 1531
1532 return ret;
1533}
1534
1535/*
1536 * ocfs2_xattr_remove()
1537 *
1538 * Free extended attribute resources associated with this inode.
1539 */
1540int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1541{
1542 struct ocfs2_xattr_block *xb;
1543 struct buffer_head *blk_bh = NULL;
1544 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1545 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1546 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1547 handle_t *handle;
1548 int ret;
1549
8154da3d
TY		 1550 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1551 return 0;
1552
cf1d6c76
TY		 1553 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1554 return 0;
1555
1556 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1557 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1558 if (ret < 0) {
1559 mlog_errno(ret);
1560 goto out;
1561 }
1562 }
1563 if (di->i_xattr_loc) {
1564 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
1565 le64_to_cpu(di->i_xattr_loc),
1566 &blk_bh, OCFS2_BH_CACHED, inode);
1567 if (ret < 0) {
1568 mlog_errno(ret);
1569 return ret;
1570 }
1571 /*Verify the signature of xattr block*/
1572 if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
1573 strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
1574 ret = -EFAULT;
1575 goto out;
1576 }
1577
1578 ret = ocfs2_xattr_block_remove(inode, blk_bh);
1579 if (ret < 0) {
1580 mlog_errno(ret);
1581 goto out;
1582 }
1583 }
1584
1585 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1586 OCFS2_INODE_UPDATE_CREDITS);
1587 if (IS_ERR(handle)) {
1588 ret = PTR_ERR(handle);
1589 mlog_errno(ret);
1590 goto out;
1591 }
1592 ret = ocfs2_journal_access(handle, inode, di_bh,
1593 OCFS2_JOURNAL_ACCESS_WRITE);
1594 if (ret) {
1595 mlog_errno(ret);
1596 goto out_commit;
1597 }
1598
1599 if (di->i_xattr_loc) {
1600 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1601 ocfs2_xattr_free_block(handle, osb, xb);
1602 di->i_xattr_loc = cpu_to_le64(0);
1603 }
1604
1605 spin_lock(&oi->ip_lock);
1606 oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1607 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1608 spin_unlock(&oi->ip_lock);
1609
1610 ret = ocfs2_journal_dirty(handle, di_bh);
1611 if (ret < 0)
1612 mlog_errno(ret);
1613out_commit:
1614 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1615out:
1616 brelse(blk_bh);
1617
1618 return ret;
1619}
1620
1621static int ocfs2_xattr_has_space_inline(struct inode *inode,
1622 struct ocfs2_dinode *di)
1623{
1624 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1625 unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1626 int free;
1627
1628 if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1629 return 0;
1630
1631 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1632 struct ocfs2_inline_data *idata = &di->id2.i_data;
1633 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1634 } else if (ocfs2_inode_is_fast_symlink(inode)) {
1635 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1636 le64_to_cpu(di->i_size);
1637 } else {
1638 struct ocfs2_extent_list *el = &di->id2.i_list;
1639 free = (le16_to_cpu(el->l_count) -
1640 le16_to_cpu(el->l_next_free_rec)) *
1641 sizeof(struct ocfs2_extent_rec);
1642 }
1643 if (free >= xattrsize)
1644 return 1;
1645
1646 return 0;
1647}
1648
1649/*
1650 * ocfs2_xattr_ibody_find()
1651 *
1652 * Find extended attribute in inode block and
1653 * fill search info into struct ocfs2_xattr_search.
1654 */
1655static int ocfs2_xattr_ibody_find(struct inode *inode,
1656 int name_index,
1657 const char *name,
1658 struct ocfs2_xattr_search *xs)
1659{
1660 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1661 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1662 int ret;
1663 int has_space = 0;
1664
1665 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1666 return 0;
1667
1668 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1669 down_read(&oi->ip_alloc_sem);
1670 has_space = ocfs2_xattr_has_space_inline(inode, di);
1671 up_read(&oi->ip_alloc_sem);
1672 if (!has_space)
1673 return 0;
1674 }
1675
1676 xs->xattr_bh = xs->inode_bh;
1677 xs->end = (void *)di + inode->i_sb->s_blocksize;
1678 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1679 xs->header = (struct ocfs2_xattr_header *)
1680 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1681 else
1682 xs->header = (struct ocfs2_xattr_header *)
1683 (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
1684 xs->base = (void *)xs->header;
1685 xs->here = xs->header->xh_entries;
1686
1687 /* Find the named attribute. */
1688 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1689 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1690 if (ret && ret != -ENODATA)
1691 return ret;
1692 xs->not_found = ret;
1693 }
1694
1695 return 0;
1696}
1697
1698/*
1699 * ocfs2_xattr_ibody_set()
1700 *
1701 * Set, replace or remove an extended attribute into inode block.
1702 *
1703 */
1704static int ocfs2_xattr_ibody_set(struct inode *inode,
1705 struct ocfs2_xattr_info *xi,
1706 struct ocfs2_xattr_search *xs)
1707{
1708 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1709 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1710 int ret;
1711
1712 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1713 return -ENOSPC;
1714
1715 down_write(&oi->ip_alloc_sem);
1716 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1717 if (!ocfs2_xattr_has_space_inline(inode, di)) {
1718 ret = -ENOSPC;
1719 goto out;
1720 }
1721 }
1722
1723 ret = ocfs2_xattr_set_entry(inode, xi, xs,
1724 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
1725out:
1726 up_write(&oi->ip_alloc_sem);
1727
1728 return ret;
1729}
1730
1731/*
1732 * ocfs2_xattr_block_find()
1733 *
1734 * Find extended attribute in external block and
1735 * fill search info into struct ocfs2_xattr_search.
1736 */
1737static int ocfs2_xattr_block_find(struct inode *inode,
1738 int name_index,
1739 const char *name,
1740 struct ocfs2_xattr_search *xs)
1741{
1742 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1743 struct buffer_head *blk_bh = NULL;
589dc260 1744 struct ocfs2_xattr_block *xb;
cf1d6c76
TY		 1745 int ret = 0;
1746
1747 if (!di->i_xattr_loc)
1748 return ret;
1749
1750 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
1751 le64_to_cpu(di->i_xattr_loc),
1752 &blk_bh, OCFS2_BH_CACHED, inode);
1753 if (ret < 0) {
1754 mlog_errno(ret);
1755 return ret;
1756 }
1757 /*Verify the signature of xattr block*/
1758 if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
1759 strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
1760 ret = -EFAULT;
1761 goto cleanup;
1762 }
1763
1764 xs->xattr_bh = blk_bh;
589dc260
TM		 1765 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1766
1767 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1768 xs->header = &xb->xb_attrs.xb_header;
1769 xs->base = (void *)xs->header;
1770 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
1771 xs->here = xs->header->xh_entries;
1772
1773 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1774 } else
1775 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
1776 name_index,
1777 name, xs);
cf1d6c76 1778
cf1d6c76
TY		 1779 if (ret && ret != -ENODATA) {
1780 xs->xattr_bh = NULL;
1781 goto cleanup;
1782 }
1783 xs->not_found = ret;
1784 return 0;
cf1d6c76
TY		 1785cleanup:
1786 brelse(blk_bh);
1787
1788 return ret;
1789}
1790
01225596
TM		 1791/*
1792 * When all the xattrs are deleted from index btree, the ocfs2_xattr_tree
1793 * will be erased and ocfs2_xattr_block will have its ocfs2_xattr_header
1794 * re-initialized.
1795 */
1796static int ocfs2_restore_xattr_block(struct inode *inode,
1797 struct ocfs2_xattr_search *xs)
1798{
1799 int ret;
1800 handle_t *handle;
1801 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1802 struct ocfs2_xattr_block *xb =
1803 (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1804 struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
1805 u16 xb_flags = le16_to_cpu(xb->xb_flags);
1806
1807 BUG_ON(!(xb_flags & OCFS2_XATTR_INDEXED) ||
1808 le16_to_cpu(el->l_next_free_rec) != 0);
1809
1810 handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1811 if (IS_ERR(handle)) {
1812 ret = PTR_ERR(handle);
1813 handle = NULL;
1814 goto out;
1815 }
1816
1817 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1818 OCFS2_JOURNAL_ACCESS_WRITE);
1819 if (ret < 0) {
1820 mlog_errno(ret);
1821 goto out_commit;
1822 }
1823
1824 memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
1825 offsetof(struct ocfs2_xattr_block, xb_attrs));
1826
1827 xb->xb_flags = cpu_to_le16(xb_flags & ~OCFS2_XATTR_INDEXED);
1828
1829 ocfs2_journal_dirty(handle, xs->xattr_bh);
1830
1831out_commit:
1832 ocfs2_commit_trans(osb, handle);
1833out:
1834 return ret;
1835}
1836
cf1d6c76
TY		 1837/*
1838 * ocfs2_xattr_block_set()
1839 *
1840 * Set, replace or remove an extended attribute into external block.
1841 *
1842 */
1843static int ocfs2_xattr_block_set(struct inode *inode,
1844 struct ocfs2_xattr_info *xi,
1845 struct ocfs2_xattr_search *xs)
1846{
1847 struct buffer_head *new_bh = NULL;
1848 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1849 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1850 struct ocfs2_alloc_context *meta_ac = NULL;
1851 handle_t *handle = NULL;
1852 struct ocfs2_xattr_block *xblk = NULL;
1853 u16 suballoc_bit_start;
1854 u32 num_got;
1855 u64 first_blkno;
1856 int ret;
1857
1858 if (!xs->xattr_bh) {
1859 /*
1860 * Alloc one external block for extended attribute
1861 * outside of inode.
1862 */
1863 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
1864 if (ret < 0) {
1865 mlog_errno(ret);
1866 goto out;
1867 }
1868 handle = ocfs2_start_trans(osb,
1869 OCFS2_XATTR_BLOCK_CREATE_CREDITS);
1870 if (IS_ERR(handle)) {
1871 ret = PTR_ERR(handle);
1872 mlog_errno(ret);
1873 goto out;
1874 }
1875 ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
1876 OCFS2_JOURNAL_ACCESS_CREATE);
1877 if (ret < 0) {
1878 mlog_errno(ret);
1879 goto out_commit;
1880 }
1881
1882 ret = ocfs2_claim_metadata(osb, handle, meta_ac, 1,
1883 &suballoc_bit_start, &num_got,
1884 &first_blkno);
1885 if (ret < 0) {
1886 mlog_errno(ret);
1887 goto out_commit;
1888 }
1889
1890 new_bh = sb_getblk(inode->i_sb, first_blkno);
1891 ocfs2_set_new_buffer_uptodate(inode, new_bh);
1892
1893 ret = ocfs2_journal_access(handle, inode, new_bh,
1894 OCFS2_JOURNAL_ACCESS_CREATE);
1895 if (ret < 0) {
1896 mlog_errno(ret);
1897 goto out_commit;
1898 }
1899
1900 /* Initialize ocfs2_xattr_block */
1901 xs->xattr_bh = new_bh;
1902 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
1903 memset(xblk, 0, inode->i_sb->s_blocksize);
1904 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
1905 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
1906 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1907 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
1908 xblk->xb_blkno = cpu_to_le64(first_blkno);
1909
1910 xs->header = &xblk->xb_attrs.xb_header;
1911 xs->base = (void *)xs->header;
1912 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
1913 xs->here = xs->header->xh_entries;
1914
1915
1916 ret = ocfs2_journal_dirty(handle, new_bh);
1917 if (ret < 0) {
1918 mlog_errno(ret);
1919 goto out_commit;
1920 }
1921 di->i_xattr_loc = cpu_to_le64(first_blkno);
1922 ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1923 if (ret < 0)
1924 mlog_errno(ret);
1925out_commit:
1926 ocfs2_commit_trans(osb, handle);
1927out:
1928 if (meta_ac)
1929 ocfs2_free_alloc_context(meta_ac);
1930 if (ret < 0)
1931 return ret;
01225596
TM		 1932 } else
1933 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1934
1935 if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
1936 /* Set extended attribute into external block */
1937 ret = ocfs2_xattr_set_entry(inode, xi, xs, OCFS2_HAS_XATTR_FL);
1938 if (!ret || ret != -ENOSPC)
1939 goto end;
1940
1941 ret = ocfs2_xattr_create_index_block(inode, xs);
1942 if (ret)
1943 goto end;
cf1d6c76
TY		 1944 }
1945
01225596
TM		 1946 ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs);
1947 if (!ret && xblk->xb_attrs.xb_root.xt_list.l_next_free_rec == 0)
1948 ret = ocfs2_restore_xattr_block(inode, xs);
1949
1950end:
cf1d6c76
TY		 1951
1952 return ret;
1953}
1954
1955/*
1956 * ocfs2_xattr_set()
1957 *
1958 * Set, replace or remove an extended attribute for this inode.
1959 * value is NULL to remove an existing extended attribute, else either
1960 * create or replace an extended attribute.
1961 */
1962int ocfs2_xattr_set(struct inode *inode,
1963 int name_index,
1964 const char *name,
1965 const void *value,
1966 size_t value_len,
1967 int flags)
1968{
1969 struct buffer_head *di_bh = NULL;
1970 struct ocfs2_dinode *di;
1971 int ret;
01225596 1972 u16 i, blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
cf1d6c76
TY		 1973
1974 struct ocfs2_xattr_info xi = {
1975 .name_index = name_index,
1976 .name = name,
1977 .value = value,
1978 .value_len = value_len,
1979 };
1980
1981 struct ocfs2_xattr_search xis = {
1982 .not_found = -ENODATA,
1983 };
1984
1985 struct ocfs2_xattr_search xbs = {
1986 .not_found = -ENODATA,
1987 };
1988
8154da3d
TY		 1989 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1990 return -EOPNOTSUPP;
1991
cf1d6c76
TY		 1992 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1993 if (ret < 0) {
1994 mlog_errno(ret);
1995 return ret;
1996 }
1997 xis.inode_bh = xbs.inode_bh = di_bh;
1998 di = (struct ocfs2_dinode *)di_bh->b_data;
1999
2000 down_write(&OCFS2_I(inode)->ip_xattr_sem);
2001 /*
2002 * Scan inode and external block to find the same name
2003 * extended attribute and collect search infomation.
2004 */
2005 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2006 if (ret)
2007 goto cleanup;
2008 if (xis.not_found) {
2009 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2010 if (ret)
2011 goto cleanup;
2012 }
2013
2014 if (xis.not_found && xbs.not_found) {
2015 ret = -ENODATA;
2016 if (flags & XATTR_REPLACE)
2017 goto cleanup;
2018 ret = 0;
2019 if (!value)
2020 goto cleanup;
2021 } else {
2022 ret = -EEXIST;
2023 if (flags & XATTR_CREATE)
2024 goto cleanup;
2025 }
2026
2027 if (!value) {
2028 /* Remove existing extended attribute */
2029 if (!xis.not_found)
2030 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2031 else if (!xbs.not_found)
2032 ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2033 } else {
2034 /* We always try to set extended attribute into inode first*/
2035 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2036 if (!ret && !xbs.not_found) {
2037 /*
2038 * If succeed and that extended attribute existing in
2039 * external block, then we will remove it.
2040 */
2041 xi.value = NULL;
2042 xi.value_len = 0;
2043 ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2044 } else if (ret == -ENOSPC) {
2045 if (di->i_xattr_loc && !xbs.xattr_bh) {
2046 ret = ocfs2_xattr_block_find(inode, name_index,
2047 name, &xbs);
2048 if (ret)
2049 goto cleanup;
2050 }
2051 /*
2052 * If no space in inode, we will set extended attribute
2053 * into external block.
2054 */
2055 ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2056 if (ret)
2057 goto cleanup;
2058 if (!xis.not_found) {
2059 /*
2060 * If succeed and that extended attribute
2061 * existing in inode, we will remove it.
2062 */
2063 xi.value = NULL;
2064 xi.value_len = 0;
2065 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2066 }
2067 }
2068 }
2069cleanup:
2070 up_write(&OCFS2_I(inode)->ip_xattr_sem);
2071 ocfs2_inode_unlock(inode, 1);
2072 brelse(di_bh);
2073 brelse(xbs.xattr_bh);
01225596
TM		 2074 for (i = 0; i < blk_per_bucket; i++)
2075 brelse(xbs.bucket.bhs[i]);
cf1d6c76
TY		 2076
2077 return ret;
2078}
2079
589dc260
TM		 2080static inline u32 ocfs2_xattr_hash_by_name(struct inode *inode,
2081 int name_index,
2082 const char *suffix_name)
2083{
2084 struct xattr_handler *handler = ocfs2_xattr_handler(name_index);
2085 char *prefix = handler->prefix;
2086 int prefix_len = strlen(handler->prefix);
2087
2088 return ocfs2_xattr_name_hash(inode, prefix, prefix_len,
2089 (char *)suffix_name, strlen(suffix_name));
2090}
2091
0c044f0b
TM		 2092/*
2093 * Find the xattr extent rec which may contains name_hash.
2094 * e_cpos will be the first name hash of the xattr rec.
2095 * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2096 */
2097static int ocfs2_xattr_get_rec(struct inode *inode,
2098 u32 name_hash,
2099 u64 *p_blkno,
2100 u32 *e_cpos,
2101 u32 *num_clusters,
2102 struct ocfs2_extent_list *el)
2103{
2104 int ret = 0, i;
2105 struct buffer_head *eb_bh = NULL;
2106 struct ocfs2_extent_block *eb;
2107 struct ocfs2_extent_rec *rec = NULL;
2108 u64 e_blkno = 0;
2109
2110 if (el->l_tree_depth) {
2111 ret = ocfs2_find_leaf(inode, el, name_hash, &eb_bh);
2112 if (ret) {
2113 mlog_errno(ret);
2114 goto out;
2115 }
2116
2117 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2118 el = &eb->h_list;
2119
2120 if (el->l_tree_depth) {
2121 ocfs2_error(inode->i_sb,
2122 "Inode %lu has non zero tree depth in "
2123 "xattr tree block %llu\n", inode->i_ino,
2124 (unsigned long long)eb_bh->b_blocknr);
2125 ret = -EROFS;
2126 goto out;
2127 }
2128 }
2129
2130 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2131 rec = &el->l_recs[i];
2132
2133 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2134 e_blkno = le64_to_cpu(rec->e_blkno);
2135 break;
2136 }
2137 }
2138
2139 if (!e_blkno) {
2140 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2141 "record (%u, %u, 0) in xattr", inode->i_ino,
2142 le32_to_cpu(rec->e_cpos),
2143 ocfs2_rec_clusters(el, rec));
2144 ret = -EROFS;
2145 goto out;
2146 }
2147
2148 *p_blkno = le64_to_cpu(rec->e_blkno);
2149 *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2150 if (e_cpos)
2151 *e_cpos = le32_to_cpu(rec->e_cpos);
2152out:
2153 brelse(eb_bh);
2154 return ret;
2155}
2156
2157typedef int (xattr_bucket_func)(struct inode *inode,
2158 struct ocfs2_xattr_bucket *bucket,
2159 void *para);
2160
589dc260
TM		 2161static int ocfs2_find_xe_in_bucket(struct inode *inode,
2162 struct buffer_head *header_bh,
2163 int name_index,
2164 const char *name,
2165 u32 name_hash,
2166 u16 *xe_index,
2167 int *found)
2168{
2169 int i, ret = 0, cmp = 1, block_off, new_offset;
2170 struct ocfs2_xattr_header *xh =
2171 (struct ocfs2_xattr_header *)header_bh->b_data;
2172 size_t name_len = strlen(name);
2173 struct ocfs2_xattr_entry *xe = NULL;
2174 struct buffer_head *name_bh = NULL;
2175 char *xe_name;
2176
2177 /*
2178 * We don't use binary search in the bucket because there
2179 * may be multiple entries with the same name hash.
2180 */
2181 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2182 xe = &xh->xh_entries[i];
2183
2184 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2185 continue;
2186 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2187 break;
2188
2189 cmp = name_index - ocfs2_xattr_get_type(xe);
2190 if (!cmp)
2191 cmp = name_len - xe->xe_name_len;
2192 if (cmp)
2193 continue;
2194
2195 ret = ocfs2_xattr_bucket_get_name_value(inode,
2196 xh,
2197 i,
2198 &block_off,
2199 &new_offset);
2200 if (ret) {
2201 mlog_errno(ret);
2202 break;
2203 }
2204
2205 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
2206 header_bh->b_blocknr + block_off,
2207 &name_bh, OCFS2_BH_CACHED, inode);
2208 if (ret) {
2209 mlog_errno(ret);
2210 break;
2211 }
2212 xe_name = name_bh->b_data + new_offset;
2213
2214 cmp = memcmp(name, xe_name, name_len);
2215 brelse(name_bh);
2216 name_bh = NULL;
2217
2218 if (cmp == 0) {
2219 *xe_index = i;
2220 *found = 1;
2221 ret = 0;
2222 break;
2223 }
2224 }
2225
2226 return ret;
2227}
2228
2229/*
2230 * Find the specified xattr entry in a series of buckets.
2231 * This series start from p_blkno and last for num_clusters.
2232 * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2233 * the num of the valid buckets.
2234 *
2235 * Return the buffer_head this xattr should reside in. And if the xattr's
2236 * hash is in the gap of 2 buckets, return the lower bucket.
2237 */
2238static int ocfs2_xattr_bucket_find(struct inode *inode,
2239 int name_index,
2240 const char *name,
2241 u32 name_hash,
2242 u64 p_blkno,
2243 u32 first_hash,
2244 u32 num_clusters,
2245 struct ocfs2_xattr_search *xs)
2246{
2247 int ret, found = 0;
2248 struct buffer_head *bh = NULL;
2249 struct buffer_head *lower_bh = NULL;
2250 struct ocfs2_xattr_header *xh = NULL;
2251 struct ocfs2_xattr_entry *xe = NULL;
2252 u16 index = 0;
2253 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2254 int low_bucket = 0, bucket, high_bucket;
2255 u32 last_hash;
2256 u64 blkno;
2257
2258 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno,
2259 &bh, OCFS2_BH_CACHED, inode);
2260 if (ret) {
2261 mlog_errno(ret);
2262 goto out;
2263 }
2264
2265 xh = (struct ocfs2_xattr_header *)bh->b_data;
2266 high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
2267
2268 while (low_bucket <= high_bucket) {
2269 brelse(bh);
2270 bh = NULL;
2271 bucket = (low_bucket + high_bucket) / 2;
2272
2273 blkno = p_blkno + bucket * blk_per_bucket;
2274
2275 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), blkno,
2276 &bh, OCFS2_BH_CACHED, inode);
2277 if (ret) {
2278 mlog_errno(ret);
2279 goto out;
2280 }
2281
2282 xh = (struct ocfs2_xattr_header *)bh->b_data;
2283 xe = &xh->xh_entries[0];
2284 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
2285 high_bucket = bucket - 1;
2286 continue;
2287 }
2288
2289 /*
2290 * Check whether the hash of the last entry in our
2291 * bucket is larger than the search one.
2292 */
2293 xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
2294 last_hash = le32_to_cpu(xe->xe_name_hash);
2295
2296 /* record lower_bh which may be the insert place. */
2297 brelse(lower_bh);
2298 lower_bh = bh;
2299 bh = NULL;
2300
2301 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
2302 low_bucket = bucket + 1;
2303 continue;
2304 }
2305
2306 /* the searched xattr should reside in this bucket if exists. */
2307 ret = ocfs2_find_xe_in_bucket(inode, lower_bh,
2308 name_index, name, name_hash,
2309 &index, &found);
2310 if (ret) {
2311 mlog_errno(ret);
2312 goto out;
2313 }
2314 break;
2315 }
2316
2317 /*
2318 * Record the bucket we have found.
2319 * When the xattr's hash value is in the gap of 2 buckets, we will
2320 * always set it to the previous bucket.
2321 */
2322 if (!lower_bh) {
2323 /*
2324 * We can't find any bucket whose first name_hash is less
2325 * than the find name_hash.
2326 */
2327 BUG_ON(bh->b_blocknr != p_blkno);
2328 lower_bh = bh;
2329 bh = NULL;
2330 }
2331 xs->bucket.bhs[0] = lower_bh;
2332 xs->bucket.xh = (struct ocfs2_xattr_header *)
2333 xs->bucket.bhs[0]->b_data;
2334 lower_bh = NULL;
2335
2336 xs->header = xs->bucket.xh;
2337 xs->base = xs->bucket.bhs[0]->b_data;
2338 xs->end = xs->base + inode->i_sb->s_blocksize;
2339
2340 if (found) {
2341 /*
2342 * If we have found the xattr enty, read all the blocks in
2343 * this bucket.
2344 */
2345 ret = ocfs2_read_blocks(OCFS2_SB(inode->i_sb),
2346 xs->bucket.bhs[0]->b_blocknr + 1,
2347 blk_per_bucket - 1, &xs->bucket.bhs[1],
2348 OCFS2_BH_CACHED, inode);
2349 if (ret) {
2350 mlog_errno(ret);
2351 goto out;
2352 }
2353
2354 xs->here = &xs->header->xh_entries[index];
2355 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
2356 (unsigned long long)xs->bucket.bhs[0]->b_blocknr, index);
2357 } else
2358 ret = -ENODATA;
2359
2360out:
2361 brelse(bh);
2362 brelse(lower_bh);
2363 return ret;
2364}
2365
2366static int ocfs2_xattr_index_block_find(struct inode *inode,
2367 struct buffer_head *root_bh,
2368 int name_index,
2369 const char *name,
2370 struct ocfs2_xattr_search *xs)
2371{
2372 int ret;
2373 struct ocfs2_xattr_block *xb =
2374 (struct ocfs2_xattr_block *)root_bh->b_data;
2375 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
2376 struct ocfs2_extent_list *el = &xb_root->xt_list;
2377 u64 p_blkno = 0;
2378 u32 first_hash, num_clusters = 0;
2379 u32 name_hash = ocfs2_xattr_hash_by_name(inode, name_index, name);
2380
2381 if (le16_to_cpu(el->l_next_free_rec) == 0)
2382 return -ENODATA;
2383
2384 mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
2385 name, name_hash, name_index);
2386
2387 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
2388 &num_clusters, el);
2389 if (ret) {
2390 mlog_errno(ret);
2391 goto out;
2392 }
2393
2394 BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
2395
2396 mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
2397 "in the rec is %u\n", num_clusters, p_blkno, first_hash);
2398
2399 ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
2400 p_blkno, first_hash, num_clusters, xs);
2401
2402out:
2403 return ret;
2404}
2405
0c044f0b
TM		 2406static int ocfs2_iterate_xattr_buckets(struct inode *inode,
2407 u64 blkno,
2408 u32 clusters,
2409 xattr_bucket_func *func,
2410 void *para)
2411{
2412 int i, j, ret = 0;
2413 int blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2414 u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
2415 u32 num_buckets = clusters * bpc;
2416 struct ocfs2_xattr_bucket bucket;
2417
2418 memset(&bucket, 0, sizeof(bucket));
2419
2420 mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
2421 clusters, blkno);
2422
2423 for (i = 0; i < num_buckets; i++, blkno += blk_per_bucket) {
2424 ret = ocfs2_read_blocks(OCFS2_SB(inode->i_sb),
2425 blkno, blk_per_bucket,
2426 bucket.bhs, OCFS2_BH_CACHED, inode);
2427 if (ret) {
2428 mlog_errno(ret);
2429 goto out;
2430 }
2431
2432 bucket.xh = (struct ocfs2_xattr_header *)bucket.bhs[0]->b_data;
2433 /*
2434 * The real bucket num in this series of blocks is stored
2435 * in the 1st bucket.
2436 */
2437 if (i == 0)
2438 num_buckets = le16_to_cpu(bucket.xh->xh_num_buckets);
2439
2440 mlog(0, "iterating xattr bucket %llu\n", blkno);
2441 if (func) {
2442 ret = func(inode, &bucket, para);
2443 if (ret) {
2444 mlog_errno(ret);
2445 break;
2446 }
2447 }
2448
2449 for (j = 0; j < blk_per_bucket; j++)
2450 brelse(bucket.bhs[j]);
2451 memset(&bucket, 0, sizeof(bucket));
2452 }
2453
2454out:
2455 for (j = 0; j < blk_per_bucket; j++)
2456 brelse(bucket.bhs[j]);
2457
2458 return ret;
2459}
2460
2461struct ocfs2_xattr_tree_list {
2462 char *buffer;
2463 size_t buffer_size;
2464};
2465
2466static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
2467 struct ocfs2_xattr_header *xh,
2468 int index,
2469 int *block_off,
2470 int *new_offset)
2471{
2472 u16 name_offset;
2473
2474 if (index < 0 || index >= le16_to_cpu(xh->xh_count))
2475 return -EINVAL;
2476
2477 name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
2478
2479 *block_off = name_offset >> inode->i_sb->s_blocksize_bits;
2480 *new_offset = name_offset % inode->i_sb->s_blocksize;
2481
2482 return 0;
2483}
2484
2485static int ocfs2_list_xattr_bucket(struct inode *inode,
2486 struct ocfs2_xattr_bucket *bucket,
2487 void *para)
2488{
2489 int ret = 0;
2490 struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
2491 size_t size;
2492 int i, block_off, new_offset;
2493
2494 for (i = 0 ; i < le16_to_cpu(bucket->xh->xh_count); i++) {
2495 struct ocfs2_xattr_entry *entry = &bucket->xh->xh_entries[i];
2496 struct xattr_handler *handler =
2497 ocfs2_xattr_handler(ocfs2_xattr_get_type(entry));
2498
2499 if (handler) {
2500 ret = ocfs2_xattr_bucket_get_name_value(inode,
2501 bucket->xh,
2502 i,
2503 &block_off,
2504 &new_offset);
2505 if (ret)
2506 break;
2507 size = handler->list(inode, xl->buffer, xl->buffer_size,
2508 bucket->bhs[block_off]->b_data +
2509 new_offset,
2510 entry->xe_name_len);
2511 if (xl->buffer) {
2512 if (size > xl->buffer_size)
2513 return -ERANGE;
2514 xl->buffer += size;
2515 }
2516 xl->buffer_size -= size;
2517 }
2518 }
2519
2520 return ret;
2521}
2522
2523static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
2524 struct ocfs2_xattr_tree_root *xt,
2525 char *buffer,
2526 size_t buffer_size)
2527{
2528 struct ocfs2_extent_list *el = &xt->xt_list;
2529 int ret = 0;
2530 u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
2531 u64 p_blkno = 0;
2532 struct ocfs2_xattr_tree_list xl = {
2533 .buffer = buffer,
2534 .buffer_size = buffer_size,
2535 };
2536
2537 if (le16_to_cpu(el->l_next_free_rec) == 0)
2538 return 0;
2539
2540 while (name_hash > 0) {
2541 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
2542 &e_cpos, &num_clusters, el);
2543 if (ret) {
2544 mlog_errno(ret);
2545 goto out;
2546 }
2547
2548 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
2549 ocfs2_list_xattr_bucket,
2550 &xl);
2551 if (ret) {
2552 mlog_errno(ret);
2553 goto out;
2554 }
2555
2556 if (e_cpos == 0)
2557 break;
2558
2559 name_hash = e_cpos - 1;
2560 }
2561
2562 ret = buffer_size - xl.buffer_size;
2563out:
2564 return ret;
2565}
01225596
TM		 2566
2567static int cmp_xe(const void *a, const void *b)
2568{
2569 const struct ocfs2_xattr_entry *l = a, *r = b;
2570 u32 l_hash = le32_to_cpu(l->xe_name_hash);
2571 u32 r_hash = le32_to_cpu(r->xe_name_hash);
2572
2573 if (l_hash > r_hash)
2574 return 1;
2575 if (l_hash < r_hash)
2576 return -1;
2577 return 0;
2578}
2579
2580static void swap_xe(void *a, void *b, int size)
2581{
2582 struct ocfs2_xattr_entry *l = a, *r = b, tmp;
2583
2584 tmp = *l;
2585 memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
2586 memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
2587}
2588
2589/*
2590 * When the ocfs2_xattr_block is filled up, new bucket will be created
2591 * and all the xattr entries will be moved to the new bucket.
2592 * Note: we need to sort the entries since they are not saved in order
2593 * in the ocfs2_xattr_block.
2594 */
2595static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
2596 struct buffer_head *xb_bh,
2597 struct buffer_head *xh_bh,
2598 struct buffer_head *data_bh)
2599{
2600 int i, blocksize = inode->i_sb->s_blocksize;
2601 u16 offset, size, off_change;
2602 struct ocfs2_xattr_entry *xe;
2603 struct ocfs2_xattr_block *xb =
2604 (struct ocfs2_xattr_block *)xb_bh->b_data;
2605 struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
2606 struct ocfs2_xattr_header *xh =
2607 (struct ocfs2_xattr_header *)xh_bh->b_data;
2608 u16 count = le16_to_cpu(xb_xh->xh_count);
2609 char *target = xh_bh->b_data, *src = xb_bh->b_data;
2610
2611 mlog(0, "cp xattr from block %llu to bucket %llu\n",
2612 (unsigned long long)xb_bh->b_blocknr,
2613 (unsigned long long)xh_bh->b_blocknr);
2614
2615 memset(xh_bh->b_data, 0, blocksize);
2616 if (data_bh)
2617 memset(data_bh->b_data, 0, blocksize);
2618 /*
2619 * Since the xe_name_offset is based on ocfs2_xattr_header,
2620 * there is a offset change corresponding to the change of
2621 * ocfs2_xattr_header's position.
2622 */
2623 off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
2624 xe = &xb_xh->xh_entries[count - 1];
2625 offset = le16_to_cpu(xe->xe_name_offset) + off_change;
2626 size = blocksize - offset;
2627
2628 /* copy all the names and values. */
2629 if (data_bh)
2630 target = data_bh->b_data;
2631 memcpy(target + offset, src + offset, size);
2632
2633 /* Init new header now. */
2634 xh->xh_count = xb_xh->xh_count;
2635 xh->xh_num_buckets = cpu_to_le16(1);
2636 xh->xh_name_value_len = cpu_to_le16(size);
2637 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
2638
2639 /* copy all the entries. */
2640 target = xh_bh->b_data;
2641 offset = offsetof(struct ocfs2_xattr_header, xh_entries);
2642 size = count * sizeof(struct ocfs2_xattr_entry);
2643 memcpy(target + offset, (char *)xb_xh + offset, size);
2644
2645 /* Change the xe offset for all the xe because of the move. */
2646 off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
2647 offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
2648 for (i = 0; i < count; i++)
2649 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
2650
2651 mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
2652 offset, size, off_change);
2653
2654 sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
2655 cmp_xe, swap_xe);
2656}
2657
2658/*
2659 * After we move xattr from block to index btree, we have to
2660 * update ocfs2_xattr_search to the new xe and base.
2661 *
2662 * When the entry is in xattr block, xattr_bh indicates the storage place.
2663 * While if the entry is in index b-tree, "bucket" indicates the
2664 * real place of the xattr.
2665 */
2666static int ocfs2_xattr_update_xattr_search(struct inode *inode,
2667 struct ocfs2_xattr_search *xs,
2668 struct buffer_head *old_bh,
2669 struct buffer_head *new_bh)
2670{
2671 int ret = 0;
2672 char *buf = old_bh->b_data;
2673 struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
2674 struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
2675 int i, blocksize = inode->i_sb->s_blocksize;
2676 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2677
2678 xs->bucket.bhs[0] = new_bh;
2679 get_bh(new_bh);
2680 xs->bucket.xh = (struct ocfs2_xattr_header *)xs->bucket.bhs[0]->b_data;
2681 xs->header = xs->bucket.xh;
2682
2683 xs->base = new_bh->b_data;
2684 xs->end = xs->base + inode->i_sb->s_blocksize;
2685
2686 if (!xs->not_found) {
2687 if (OCFS2_XATTR_BUCKET_SIZE != blocksize) {
2688 ret = ocfs2_read_blocks(OCFS2_SB(inode->i_sb),
2689 xs->bucket.bhs[0]->b_blocknr + 1,
2690 blk_per_bucket - 1, &xs->bucket.bhs[1],
2691 OCFS2_BH_CACHED, inode);
2692 if (ret) {
2693 mlog_errno(ret);
2694 return ret;
2695 }
2696
2697 i = xs->here - old_xh->xh_entries;
2698 xs->here = &xs->header->xh_entries[i];
2699 }
2700 }
2701
2702 return ret;
2703}
2704
2705static int ocfs2_xattr_create_index_block(struct inode *inode,
2706 struct ocfs2_xattr_search *xs)
2707{
2708 int ret, credits = OCFS2_SUBALLOC_ALLOC;
2709 u32 bit_off, len;
2710 u64 blkno;
2711 handle_t *handle;
2712 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2713 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2714 struct ocfs2_alloc_context *data_ac;
2715 struct buffer_head *xh_bh = NULL, *data_bh = NULL;
2716 struct buffer_head *xb_bh = xs->xattr_bh;
2717 struct ocfs2_xattr_block *xb =
2718 (struct ocfs2_xattr_block *)xb_bh->b_data;
2719 struct ocfs2_xattr_tree_root *xr;
2720 u16 xb_flags = le16_to_cpu(xb->xb_flags);
2721 u16 bpb = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2722
2723 mlog(0, "create xattr index block for %llu\n",
2724 (unsigned long long)xb_bh->b_blocknr);
2725
2726 BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
2727
2728 ret = ocfs2_reserve_clusters(osb, 1, &data_ac);
2729 if (ret) {
2730 mlog_errno(ret);
2731 goto out;
2732 }
2733
2734 /*
2735 * XXX:
2736 * We can use this lock for now, and maybe move to a dedicated mutex
2737 * if performance becomes a problem later.
2738 */
2739 down_write(&oi->ip_alloc_sem);
2740
2741 /*
2742 * 3 more credits, one for xattr block update, one for the 1st block
2743 * of the new xattr bucket and one for the value/data.
2744 */
2745 credits += 3;
2746 handle = ocfs2_start_trans(osb, credits);
2747 if (IS_ERR(handle)) {
2748 ret = PTR_ERR(handle);
2749 mlog_errno(ret);
2750 goto out_sem;
2751 }
2752
2753 ret = ocfs2_journal_access(handle, inode, xb_bh,
2754 OCFS2_JOURNAL_ACCESS_WRITE);
2755 if (ret) {
2756 mlog_errno(ret);
2757 goto out_commit;
2758 }
2759
2760 ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off, &len);
2761 if (ret) {
2762 mlog_errno(ret);
2763 goto out_commit;
2764 }
2765
2766 /*
2767 * The bucket may spread in many blocks, and
2768 * we will only touch the 1st block and the last block
2769 * in the whole bucket(one for entry and one for data).
2770 */
2771 blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
2772
2773 mlog(0, "allocate 1 cluster from %llu to xattr block\n", blkno);
2774
2775 xh_bh = sb_getblk(inode->i_sb, blkno);
2776 if (!xh_bh) {
2777 ret = -EIO;
2778 mlog_errno(ret);
2779 goto out_commit;
2780 }
2781
2782 ocfs2_set_new_buffer_uptodate(inode, xh_bh);
2783
2784 ret = ocfs2_journal_access(handle, inode, xh_bh,
2785 OCFS2_JOURNAL_ACCESS_CREATE);
2786 if (ret) {
2787 mlog_errno(ret);
2788 goto out_commit;
2789 }
2790
2791 if (bpb > 1) {
2792 data_bh = sb_getblk(inode->i_sb, blkno + bpb - 1);
2793 if (!data_bh) {
2794 ret = -EIO;
2795 mlog_errno(ret);
2796 goto out_commit;
2797 }
2798
2799 ocfs2_set_new_buffer_uptodate(inode, data_bh);
2800
2801 ret = ocfs2_journal_access(handle, inode, data_bh,
2802 OCFS2_JOURNAL_ACCESS_CREATE);
2803 if (ret) {
2804 mlog_errno(ret);
2805 goto out_commit;
2806 }
2807 }
2808
2809 ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xh_bh, data_bh);
2810
2811 ocfs2_journal_dirty(handle, xh_bh);
2812 if (data_bh)
2813 ocfs2_journal_dirty(handle, data_bh);
2814
2815 ocfs2_xattr_update_xattr_search(inode, xs, xb_bh, xh_bh);
2816
2817 /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
2818 memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
2819 offsetof(struct ocfs2_xattr_block, xb_attrs));
2820
2821 xr = &xb->xb_attrs.xb_root;
2822 xr->xt_clusters = cpu_to_le32(1);
2823 xr->xt_last_eb_blk = 0;
2824 xr->xt_list.l_tree_depth = 0;
2825 xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
2826 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2827
2828 xr->xt_list.l_recs[0].e_cpos = 0;
2829 xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
2830 xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
2831
2832 xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
2833
2834 ret = ocfs2_journal_dirty(handle, xb_bh);
2835 if (ret) {
2836 mlog_errno(ret);
2837 goto out_commit;
2838 }
2839
2840out_commit:
2841 ocfs2_commit_trans(osb, handle);
2842
2843out_sem:
2844 up_write(&oi->ip_alloc_sem);
2845
2846out:
2847 if (data_ac)
2848 ocfs2_free_alloc_context(data_ac);
2849
2850 brelse(xh_bh);
2851 brelse(data_bh);
2852
2853 return ret;
2854}
2855
2856static int cmp_xe_offset(const void *a, const void *b)
2857{
2858 const struct ocfs2_xattr_entry *l = a, *r = b;
2859 u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
2860 u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
2861
2862 if (l_name_offset < r_name_offset)
2863 return 1;
2864 if (l_name_offset > r_name_offset)
2865 return -1;
2866 return 0;
2867}
2868
2869/*
2870 * defrag a xattr bucket if we find that the bucket has some
2871 * holes beteen name/value pairs.
2872 * We will move all the name/value pairs to the end of the bucket
2873 * so that we can spare some space for insertion.
2874 */
2875static int ocfs2_defrag_xattr_bucket(struct inode *inode,
2876 struct ocfs2_xattr_bucket *bucket)
2877{
2878 int ret, i;
2879 size_t end, offset, len, value_len;
2880 struct ocfs2_xattr_header *xh;
2881 char *entries, *buf, *bucket_buf = NULL;
2882 u64 blkno = bucket->bhs[0]->b_blocknr;
2883 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2884 u16 xh_free_start;
2885 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2886 size_t blocksize = inode->i_sb->s_blocksize;
2887 handle_t *handle;
2888 struct buffer_head **bhs;
2889 struct ocfs2_xattr_entry *xe;
2890
2891 bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
2892 GFP_NOFS);
2893 if (!bhs)
2894 return -ENOMEM;
2895
2896 ret = ocfs2_read_blocks(osb, blkno, blk_per_bucket, bhs,
2897 OCFS2_BH_CACHED, inode);
2898 if (ret)
2899 goto out;
2900
2901 /*
2902 * In order to make the operation more efficient and generic,
2903 * we copy all the blocks into a contiguous memory and do the
2904 * defragment there, so if anything is error, we will not touch
2905 * the real block.
2906 */
2907 bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
2908 if (!bucket_buf) {
2909 ret = -EIO;
2910 goto out;
2911 }
2912
2913 buf = bucket_buf;
2914 for (i = 0; i < blk_per_bucket; i++, buf += blocksize)
2915 memcpy(buf, bhs[i]->b_data, blocksize);
2916
2917 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), blk_per_bucket);
2918 if (IS_ERR(handle)) {
2919 ret = PTR_ERR(handle);
2920 handle = NULL;
2921 mlog_errno(ret);
2922 goto out;
2923 }
2924
2925 for (i = 0; i < blk_per_bucket; i++) {
2926 ret = ocfs2_journal_access(handle, inode, bhs[i],
2927 OCFS2_JOURNAL_ACCESS_WRITE);
2928 if (ret < 0) {
2929 mlog_errno(ret);
2930 goto commit;
2931 }
2932 }
2933
2934 xh = (struct ocfs2_xattr_header *)bucket_buf;
2935 entries = (char *)xh->xh_entries;
2936 xh_free_start = le16_to_cpu(xh->xh_free_start);
2937
2938 mlog(0, "adjust xattr bucket in %llu, count = %u, "
2939 "xh_free_start = %u, xh_name_value_len = %u.\n",
2940 blkno, le16_to_cpu(xh->xh_count), xh_free_start,
2941 le16_to_cpu(xh->xh_name_value_len));
2942
2943 /*
2944 * sort all the entries by their offset.
2945 * the largest will be the first, so that we can
2946 * move them to the end one by one.
2947 */
2948 sort(entries, le16_to_cpu(xh->xh_count),
2949 sizeof(struct ocfs2_xattr_entry),
2950 cmp_xe_offset, swap_xe);
2951
2952 /* Move all name/values to the end of the bucket. */
2953 xe = xh->xh_entries;
2954 end = OCFS2_XATTR_BUCKET_SIZE;
2955 for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
2956 offset = le16_to_cpu(xe->xe_name_offset);
2957 if (ocfs2_xattr_is_local(xe))
2958 value_len = OCFS2_XATTR_SIZE(
2959 le64_to_cpu(xe->xe_value_size));
2960 else
2961 value_len = OCFS2_XATTR_ROOT_SIZE;
2962 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
2963
2964 /*
2965 * We must make sure that the name/value pair
2966 * exist in the same block. So adjust end to
2967 * the previous block end if needed.
2968 */
2969 if (((end - len) / blocksize !=
2970 (end - 1) / blocksize))
2971 end = end - end % blocksize;
2972
2973 if (end > offset + len) {
2974 memmove(bucket_buf + end - len,
2975 bucket_buf + offset, len);
2976 xe->xe_name_offset = cpu_to_le16(end - len);
2977 }
2978
2979 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
2980 "bucket %llu\n", (unsigned long long)blkno);
2981
2982 end -= len;
2983 }
2984
2985 mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
2986 "bucket %llu\n", (unsigned long long)blkno);
2987
2988 if (xh_free_start == end)
2989 goto commit;
2990
2991 memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
2992 xh->xh_free_start = cpu_to_le16(end);
2993
2994 /* sort the entries by their name_hash. */
2995 sort(entries, le16_to_cpu(xh->xh_count),
2996 sizeof(struct ocfs2_xattr_entry),
2997 cmp_xe, swap_xe);
2998
2999 buf = bucket_buf;
3000 for (i = 0; i < blk_per_bucket; i++, buf += blocksize) {
3001 memcpy(bhs[i]->b_data, buf, blocksize);
3002 ocfs2_journal_dirty(handle, bhs[i]);
3003 }
3004
3005commit:
3006 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
3007out:
3008
3009 if (bhs) {
3010 for (i = 0; i < blk_per_bucket; i++)
3011 brelse(bhs[i]);
3012 }
3013 kfree(bhs);
3014
3015 kfree(bucket_buf);
3016 return ret;
3017}
3018
3019/*
3020 * Move half nums of the xattr bucket in the previous cluster to this new
3021 * cluster. We only touch the last cluster of the previous extend record.
3022 *
3023 * first_bh is the first buffer_head of a series of bucket in the same
3024 * extent rec and header_bh is the header of one bucket in this cluster.
3025 * They will be updated if we move the data header_bh contains to the new
3026 * cluster. first_hash will be set as the 1st xe's name_hash of the new cluster.
3027 */
3028static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3029 handle_t *handle,
3030 struct buffer_head **first_bh,
3031 struct buffer_head **header_bh,
3032 u64 new_blkno,
3033 u64 prev_blkno,
3034 u32 num_clusters,
3035 u32 *first_hash)
3036{
3037 int i, ret, credits;
3038 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3039 int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3040 int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3041 int blocksize = inode->i_sb->s_blocksize;
3042 struct buffer_head *old_bh, *new_bh, *prev_bh, *new_first_bh = NULL;
3043 struct ocfs2_xattr_header *new_xh;
3044 struct ocfs2_xattr_header *xh =
3045 (struct ocfs2_xattr_header *)((*first_bh)->b_data);
3046
3047 BUG_ON(le16_to_cpu(xh->xh_num_buckets) < num_buckets);
3048 BUG_ON(OCFS2_XATTR_BUCKET_SIZE == osb->s_clustersize);
3049
3050 prev_bh = *first_bh;
3051 get_bh(prev_bh);
3052 xh = (struct ocfs2_xattr_header *)prev_bh->b_data;
3053
3054 prev_blkno += (num_clusters - 1) * bpc + bpc / 2;
3055
3056 mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3057 prev_blkno, new_blkno);
3058
3059 /*
3060 * We need to update the 1st half of the new cluster and
3061 * 1 more for the update of the 1st bucket of the previous
3062 * extent record.
3063 */
3064 credits = bpc / 2 + 1;
3065 ret = ocfs2_extend_trans(handle, credits);
3066 if (ret) {
3067 mlog_errno(ret);
3068 goto out;
3069 }
3070
3071 ret = ocfs2_journal_access(handle, inode, prev_bh,
3072 OCFS2_JOURNAL_ACCESS_WRITE);
3073 if (ret) {
3074 mlog_errno(ret);
3075 goto out;
3076 }
3077
3078 for (i = 0; i < bpc / 2; i++, prev_blkno++, new_blkno++) {
3079 old_bh = new_bh = NULL;
3080 new_bh = sb_getblk(inode->i_sb, new_blkno);
3081 if (!new_bh) {
3082 ret = -EIO;
3083 mlog_errno(ret);
3084 goto out;
3085 }
3086
3087 ocfs2_set_new_buffer_uptodate(inode, new_bh);
3088
3089 ret = ocfs2_journal_access(handle, inode, new_bh,
3090 OCFS2_JOURNAL_ACCESS_CREATE);
3091 if (ret < 0) {
3092 mlog_errno(ret);
3093 brelse(new_bh);
3094 goto out;
3095 }
3096
3097 ret = ocfs2_read_block(osb, prev_blkno,
3098 &old_bh, OCFS2_BH_CACHED, inode);
3099 if (ret < 0) {
3100 mlog_errno(ret);
3101 brelse(new_bh);
3102 goto out;
3103 }
3104
3105 memcpy(new_bh->b_data, old_bh->b_data, blocksize);
3106
3107 if (i == 0) {
3108 new_xh = (struct ocfs2_xattr_header *)new_bh->b_data;
3109 new_xh->xh_num_buckets = cpu_to_le16(num_buckets / 2);
3110
3111 if (first_hash)
3112 *first_hash = le32_to_cpu(
3113 new_xh->xh_entries[0].xe_name_hash);
3114 new_first_bh = new_bh;
3115 get_bh(new_first_bh);
3116 }
3117
3118 ocfs2_journal_dirty(handle, new_bh);
3119
3120 if (*header_bh == old_bh) {
3121 brelse(*header_bh);
3122 *header_bh = new_bh;
3123 get_bh(*header_bh);
3124
3125 brelse(*first_bh);
3126 *first_bh = new_first_bh;
3127 get_bh(*first_bh);
3128 }
3129 brelse(new_bh);
3130 brelse(old_bh);
3131 }
3132
3133 le16_add_cpu(&xh->xh_num_buckets, -(num_buckets / 2));
3134
3135 ocfs2_journal_dirty(handle, prev_bh);
3136out:
3137 brelse(prev_bh);
3138 brelse(new_first_bh);
3139 return ret;
3140}
3141
3142static int ocfs2_read_xattr_bucket(struct inode *inode,
3143 u64 blkno,
3144 struct buffer_head **bhs,
3145 int new)
3146{
3147 int ret = 0;
3148 u16 i, blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3149
3150 if (!new)
3151 return ocfs2_read_blocks(OCFS2_SB(inode->i_sb), blkno,
3152 blk_per_bucket, bhs,
3153 OCFS2_BH_CACHED, inode);
3154
3155 for (i = 0; i < blk_per_bucket; i++) {
3156 bhs[i] = sb_getblk(inode->i_sb, blkno + i);
3157 if (bhs[i] == NULL) {
3158 ret = -EIO;
3159 mlog_errno(ret);
3160 break;
3161 }
3162 ocfs2_set_new_buffer_uptodate(inode, bhs[i]);
3163 }
3164
3165 return ret;
3166}
3167
3168/*
3169 * Move half num of the xattrs in old bucket(blk) to new bucket(new_blk).
3170 * first_hash will record the 1st hash of the new bucket.
3171 */
3172static int ocfs2_half_xattr_bucket(struct inode *inode,
3173 handle_t *handle,
3174 u64 blk,
3175 u64 new_blk,
3176 u32 *first_hash,
3177 int new_bucket_head)
3178{
3179 int ret, i;
3180 u16 count, start, len, name_value_len, xe_len, name_offset;
3181 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3182 struct buffer_head **s_bhs, **t_bhs = NULL;
3183 struct ocfs2_xattr_header *xh;
3184 struct ocfs2_xattr_entry *xe;
3185 int blocksize = inode->i_sb->s_blocksize;
3186
3187 mlog(0, "move half of xattrs from bucket %llu to %llu\n",
3188 blk, new_blk);
3189
3190 s_bhs = kcalloc(blk_per_bucket, sizeof(struct buffer_head *), GFP_NOFS);
3191 if (!s_bhs)
3192 return -ENOMEM;
3193
3194 ret = ocfs2_read_xattr_bucket(inode, blk, s_bhs, 0);
3195 if (ret) {
3196 mlog_errno(ret);
3197 goto out;
3198 }
3199
3200 ret = ocfs2_journal_access(handle, inode, s_bhs[0],
3201 OCFS2_JOURNAL_ACCESS_WRITE);
3202 if (ret) {
3203 mlog_errno(ret);
3204 goto out;
3205 }
3206
3207 t_bhs = kcalloc(blk_per_bucket, sizeof(struct buffer_head *), GFP_NOFS);
3208 if (!t_bhs) {
3209 ret = -ENOMEM;
3210 goto out;
3211 }
3212
3213 ret = ocfs2_read_xattr_bucket(inode, new_blk, t_bhs, new_bucket_head);
3214 if (ret) {
3215 mlog_errno(ret);
3216 goto out;
3217 }
3218
3219 for (i = 0; i < blk_per_bucket; i++) {
3220 ret = ocfs2_journal_access(handle, inode, t_bhs[i],
3221 OCFS2_JOURNAL_ACCESS_CREATE);
3222 if (ret) {
3223 mlog_errno(ret);
3224 goto out;
3225 }
3226 }
3227
3228 /* copy the whole bucket to the new first. */
3229 for (i = 0; i < blk_per_bucket; i++)
3230 memcpy(t_bhs[i]->b_data, s_bhs[i]->b_data, blocksize);
3231
3232 /* update the new bucket. */
3233 xh = (struct ocfs2_xattr_header *)t_bhs[0]->b_data;
3234 count = le16_to_cpu(xh->xh_count);
3235 start = count / 2;
3236
3237 /*
3238 * Calculate the total name/value len and xh_free_start for
3239 * the old bucket first.
3240 */
3241 name_offset = OCFS2_XATTR_BUCKET_SIZE;
3242 name_value_len = 0;
3243 for (i = 0; i < start; i++) {
3244 xe = &xh->xh_entries[i];
3245 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3246 if (ocfs2_xattr_is_local(xe))
3247 xe_len +=
3248 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3249 else
3250 xe_len += OCFS2_XATTR_ROOT_SIZE;
3251 name_value_len += xe_len;
3252 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3253 name_offset = le16_to_cpu(xe->xe_name_offset);
3254 }
3255
3256 /*
3257 * Now begin the modification to the new bucket.
3258 *
3259 * In the new bucket, We just move the xattr entry to the beginning
3260 * and don't touch the name/value. So there will be some holes in the
3261 * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3262 * called.
3263 */
3264 xe = &xh->xh_entries[start];
3265 len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3266 mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3267 (char *)xe - (char *)xh, (char *)xh->xh_entries - (char *)xh);
3268 memmove((char *)xh->xh_entries, (char *)xe, len);
3269 xe = &xh->xh_entries[count - start];
3270 len = sizeof(struct ocfs2_xattr_entry) * start;
3271 memset((char *)xe, 0, len);
3272
3273 le16_add_cpu(&xh->xh_count, -start);
3274 le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3275
3276 /* Calculate xh_free_start for the new bucket. */
3277 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3278 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3279 xe = &xh->xh_entries[i];
3280 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3281 if (ocfs2_xattr_is_local(xe))
3282 xe_len +=
3283 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3284 else
3285 xe_len += OCFS2_XATTR_ROOT_SIZE;
3286 if (le16_to_cpu(xe->xe_name_offset) <
3287 le16_to_cpu(xh->xh_free_start))
3288 xh->xh_free_start = xe->xe_name_offset;
3289 }
3290
3291 /* set xh->xh_num_buckets for the new xh. */
3292 if (new_bucket_head)
3293 xh->xh_num_buckets = cpu_to_le16(1);
3294 else
3295 xh->xh_num_buckets = 0;
3296
3297 for (i = 0; i < blk_per_bucket; i++) {
3298 ocfs2_journal_dirty(handle, t_bhs[i]);
3299 if (ret)
3300 mlog_errno(ret);
3301 }
3302
3303 /* store the first_hash of the new bucket. */
3304 if (first_hash)
3305 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3306
3307 /*
3308 * Now only update the 1st block of the old bucket.
3309 * Please note that the entry has been sorted already above.
3310 */
3311 xh = (struct ocfs2_xattr_header *)s_bhs[0]->b_data;
3312 memset(&xh->xh_entries[start], 0,
3313 sizeof(struct ocfs2_xattr_entry) * (count - start));
3314 xh->xh_count = cpu_to_le16(start);
3315 xh->xh_free_start = cpu_to_le16(name_offset);
3316 xh->xh_name_value_len = cpu_to_le16(name_value_len);
3317
3318 ocfs2_journal_dirty(handle, s_bhs[0]);
3319 if (ret)
3320 mlog_errno(ret);
3321
3322out:
3323 if (s_bhs) {
3324 for (i = 0; i < blk_per_bucket; i++)
3325 brelse(s_bhs[i]);
3326 }
3327 kfree(s_bhs);
3328
3329 if (t_bhs) {
3330 for (i = 0; i < blk_per_bucket; i++)
3331 brelse(t_bhs[i]);
3332 }
3333 kfree(t_bhs);
3334
3335 return ret;
3336}
3337
3338/*
3339 * Copy xattr from one bucket to another bucket.
3340 *
3341 * The caller must make sure that the journal transaction
3342 * has enough space for journaling.
3343 */
3344static int ocfs2_cp_xattr_bucket(struct inode *inode,
3345 handle_t *handle,
3346 u64 s_blkno,
3347 u64 t_blkno,
3348 int t_is_new)
3349{
3350 int ret, i;
3351 int blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3352 int blocksize = inode->i_sb->s_blocksize;
3353 struct buffer_head **s_bhs, **t_bhs = NULL;
3354
3355 BUG_ON(s_blkno == t_blkno);
3356
3357 mlog(0, "cp bucket %llu to %llu, target is %d\n",
3358 s_blkno, t_blkno, t_is_new);
3359
3360 s_bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
3361 GFP_NOFS);
3362 if (!s_bhs)
3363 return -ENOMEM;
3364
3365 ret = ocfs2_read_xattr_bucket(inode, s_blkno, s_bhs, 0);
3366 if (ret)
3367 goto out;
3368
3369 t_bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
3370 GFP_NOFS);
3371 if (!t_bhs) {
3372 ret = -ENOMEM;
3373 goto out;
3374 }
3375
3376 ret = ocfs2_read_xattr_bucket(inode, t_blkno, t_bhs, t_is_new);
3377 if (ret)
3378 goto out;
3379
3380 for (i = 0; i < blk_per_bucket; i++) {
3381 ret = ocfs2_journal_access(handle, inode, t_bhs[i],
3382 OCFS2_JOURNAL_ACCESS_WRITE);
3383 if (ret)
3384 goto out;
3385 }
3386
3387 for (i = 0; i < blk_per_bucket; i++) {
3388 memcpy(t_bhs[i]->b_data, s_bhs[i]->b_data, blocksize);
3389 ocfs2_journal_dirty(handle, t_bhs[i]);
3390 }
3391
3392out:
3393 if (s_bhs) {
3394 for (i = 0; i < blk_per_bucket; i++)
3395 brelse(s_bhs[i]);
3396 }
3397 kfree(s_bhs);
3398
3399 if (t_bhs) {
3400 for (i = 0; i < blk_per_bucket; i++)
3401 brelse(t_bhs[i]);
3402 }
3403 kfree(t_bhs);
3404
3405 return ret;
3406}
3407
3408/*
3409 * Copy one xattr cluster from src_blk to to_blk.
3410 * The to_blk will become the first bucket header of the cluster, so its
3411 * xh_num_buckets will be initialized as the bucket num in the cluster.
3412 */
3413static int ocfs2_cp_xattr_cluster(struct inode *inode,
3414 handle_t *handle,
3415 struct buffer_head *first_bh,
3416 u64 src_blk,
3417 u64 to_blk,
3418 u32 *first_hash)
3419{
3420 int i, ret, credits;
3421 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3422 int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3423 int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3424 struct buffer_head *bh = NULL;
3425 struct ocfs2_xattr_header *xh;
3426 u64 to_blk_start = to_blk;
3427
3428 mlog(0, "cp xattrs from cluster %llu to %llu\n", src_blk, to_blk);
3429
3430 /*
3431 * We need to update the new cluster and 1 more for the update of
3432 * the 1st bucket of the previous extent rec.
3433 */
3434 credits = bpc + 1;
3435 ret = ocfs2_extend_trans(handle, credits);
3436 if (ret) {
3437 mlog_errno(ret);
3438 goto out;
3439 }
3440
3441 ret = ocfs2_journal_access(handle, inode, first_bh,
3442 OCFS2_JOURNAL_ACCESS_WRITE);
3443 if (ret) {
3444 mlog_errno(ret);
3445 goto out;
3446 }
3447
3448 for (i = 0; i < num_buckets; i++) {
3449 ret = ocfs2_cp_xattr_bucket(inode, handle,
3450 src_blk, to_blk, 1);
3451 if (ret) {
3452 mlog_errno(ret);
3453 goto out;
3454 }
3455
3456 src_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3457 to_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3458 }
3459
3460 /* update the old bucket header. */
3461 xh = (struct ocfs2_xattr_header *)first_bh->b_data;
3462 le16_add_cpu(&xh->xh_num_buckets, -num_buckets);
3463
3464 ocfs2_journal_dirty(handle, first_bh);
3465
3466 /* update the new bucket header. */
3467 ret = ocfs2_read_block(osb, to_blk_start, &bh, OCFS2_BH_CACHED, inode);
3468 if (ret < 0) {
3469 mlog_errno(ret);
3470 goto out;
3471 }
3472
3473 ret = ocfs2_journal_access(handle, inode, bh,
3474 OCFS2_JOURNAL_ACCESS_WRITE);
3475 if (ret) {
3476 mlog_errno(ret);
3477 goto out;
3478 }
3479
3480 xh = (struct ocfs2_xattr_header *)bh->b_data;
3481 xh->xh_num_buckets = cpu_to_le16(num_buckets);
3482
3483 ocfs2_journal_dirty(handle, bh);
3484
3485 if (first_hash)
3486 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3487out:
3488 brelse(bh);
3489 return ret;
3490}
3491
3492/*
3493 * Move half of the xattrs in this cluster to the new cluster.
3494 * This function should only be called when bucket size == cluster size.
3495 * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
3496 */
3497static int ocfs2_half_xattr_cluster(struct inode *inode,
3498 handle_t *handle,
3499 u64 prev_blk,
3500 u64 new_blk,
3501 u32 *first_hash)
3502{
3503 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3504 int ret, credits = 2 * blk_per_bucket;
3505
3506 BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
3507
3508 ret = ocfs2_extend_trans(handle, credits);
3509 if (ret) {
3510 mlog_errno(ret);
3511 return ret;
3512 }
3513
3514 /* Move half of the xattr in start_blk to the next bucket. */
3515 return ocfs2_half_xattr_bucket(inode, handle, prev_blk,
3516 new_blk, first_hash, 1);
3517}
3518
3519/*
3520 * Move some xattrs from the old cluster to the new one since they are not
3521 * contiguous in ocfs2 xattr tree.
3522 *
3523 * new_blk starts a new separate cluster, and we will move some xattrs from
3524 * prev_blk to it. v_start will be set as the first name hash value in this
3525 * new cluster so that it can be used as e_cpos during tree insertion and
3526 * don't collide with our original b-tree operations. first_bh and header_bh
3527 * will also be updated since they will be used in ocfs2_extend_xattr_bucket
3528 * to extend the insert bucket.
3529 *
3530 * The problem is how much xattr should we move to the new one and when should
3531 * we update first_bh and header_bh?
3532 * 1. If cluster size > bucket size, that means the previous cluster has more
3533 * than 1 bucket, so just move half nums of bucket into the new cluster and
3534 * update the first_bh and header_bh if the insert bucket has been moved
3535 * to the new cluster.
3536 * 2. If cluster_size == bucket_size:
3537 * a) If the previous extent rec has more than one cluster and the insert
3538 * place isn't in the last cluster, copy the entire last cluster to the
3539 * new one. This time, we don't need to upate the first_bh and header_bh
3540 * since they will not be moved into the new cluster.
3541 * b) Otherwise, move the bottom half of the xattrs in the last cluster into
3542 * the new one. And we set the extend flag to zero if the insert place is
3543 * moved into the new allocated cluster since no extend is needed.
3544 */
3545static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
3546 handle_t *handle,
3547 struct buffer_head **first_bh,
3548 struct buffer_head **header_bh,
3549 u64 new_blk,
3550 u64 prev_blk,
3551 u32 prev_clusters,
3552 u32 *v_start,
3553 int *extend)
3554{
3555 int ret = 0;
3556 int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3557
3558 mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
3559 prev_blk, prev_clusters, new_blk);
3560
3561 if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1)
3562 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
3563 handle,
3564 first_bh,
3565 header_bh,
3566 new_blk,
3567 prev_blk,
3568 prev_clusters,
3569 v_start);
3570 else {
3571 u64 last_blk = prev_blk + bpc * (prev_clusters - 1);
3572
3573 if (prev_clusters > 1 && (*header_bh)->b_blocknr != last_blk)
3574 ret = ocfs2_cp_xattr_cluster(inode, handle, *first_bh,
3575 last_blk, new_blk,
3576 v_start);
3577 else {
3578 ret = ocfs2_half_xattr_cluster(inode, handle,
3579 last_blk, new_blk,
3580 v_start);
3581
3582 if ((*header_bh)->b_blocknr == last_blk && extend)
3583 *extend = 0;
3584 }
3585 }
3586
3587 return ret;
3588}
3589
3590/*
3591 * Add a new cluster for xattr storage.
3592 *
3593 * If the new cluster is contiguous with the previous one, it will be
3594 * appended to the same extent record, and num_clusters will be updated.
3595 * If not, we will insert a new extent for it and move some xattrs in
3596 * the last cluster into the new allocated one.
3597 * We also need to limit the maximum size of a btree leaf, otherwise we'll
3598 * lose the benefits of hashing because we'll have to search large leaves.
3599 * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
3600 * if it's bigger).
3601 *
3602 * first_bh is the first block of the previous extent rec and header_bh
3603 * indicates the bucket we will insert the new xattrs. They will be updated
3604 * when the header_bh is moved into the new cluster.
3605 */
3606static int ocfs2_add_new_xattr_cluster(struct inode *inode,
3607 struct buffer_head *root_bh,
3608 struct buffer_head **first_bh,
3609 struct buffer_head **header_bh,
3610 u32 *num_clusters,
3611 u32 prev_cpos,
3612 u64 prev_blkno,
3613 int *extend)
3614{
3615 int ret, credits;
3616 u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3617 u32 prev_clusters = *num_clusters;
3618 u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
3619 u64 block;
3620 handle_t *handle = NULL;
3621 struct ocfs2_alloc_context *data_ac = NULL;
3622 struct ocfs2_alloc_context *meta_ac = NULL;
3623 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3624 struct ocfs2_xattr_block *xb =
3625 (struct ocfs2_xattr_block *)root_bh->b_data;
3626 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3627 struct ocfs2_extent_list *root_el = &xb_root->xt_list;
3628 enum ocfs2_extent_tree_type type = OCFS2_XATTR_TREE_EXTENT;
3629
3630 mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
3631 "previous xattr blkno = %llu\n",
3632 (unsigned long long)OCFS2_I(inode)->ip_blkno,
3633 prev_cpos, prev_blkno);
3634
3635 ret = ocfs2_lock_allocators(inode, root_bh, root_el,
3636 clusters_to_add, 0, &data_ac,
3637 &meta_ac, type, NULL);
3638 if (ret) {
3639 mlog_errno(ret);
3640 goto leave;
3641 }
3642
3643 credits = ocfs2_calc_extend_credits(osb->sb, root_el, clusters_to_add);
3644 handle = ocfs2_start_trans(osb, credits);
3645 if (IS_ERR(handle)) {
3646 ret = PTR_ERR(handle);
3647 handle = NULL;
3648 mlog_errno(ret);
3649 goto leave;
3650 }
3651
3652 ret = ocfs2_journal_access(handle, inode, root_bh,
3653 OCFS2_JOURNAL_ACCESS_WRITE);
3654 if (ret < 0) {
3655 mlog_errno(ret);
3656 goto leave;
3657 }
3658
3659 ret = __ocfs2_claim_clusters(osb, handle, data_ac, 1,
3660 clusters_to_add, &bit_off, &num_bits);
3661 if (ret < 0) {
3662 if (ret != -ENOSPC)
3663 mlog_errno(ret);
3664 goto leave;
3665 }
3666
3667 BUG_ON(num_bits > clusters_to_add);
3668
3669 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
3670 mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
3671 num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
3672
3673 if (prev_blkno + prev_clusters * bpc == block &&
3674 (prev_clusters + num_bits) << osb->s_clustersize_bits <=
3675 OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
3676 /*
3677 * If this cluster is contiguous with the old one and
3678 * adding this new cluster, we don't surpass the limit of
3679 * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
3680 * initialized and used like other buckets in the previous
3681 * cluster.
3682 * So add it as a contiguous one. The caller will handle
3683 * its init process.
3684 */
3685 v_start = prev_cpos + prev_clusters;
3686 *num_clusters = prev_clusters + num_bits;
3687 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
3688 num_bits);
3689 } else {
3690 ret = ocfs2_adjust_xattr_cross_cluster(inode,
3691 handle,
3692 first_bh,
3693 header_bh,
3694 block,
3695 prev_blkno,
3696 prev_clusters,
3697 &v_start,
3698 extend);
3699 if (ret) {
3700 mlog_errno(ret);
3701 goto leave;
3702 }
3703 }
3704
3705 mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
3706 num_bits, block, v_start);
3707 ret = ocfs2_xattr_tree_insert_extent(osb, handle, inode, root_bh,
3708 v_start, block, num_bits,
3709 0, meta_ac);
3710 if (ret < 0) {
3711 mlog_errno(ret);
3712 goto leave;
3713 }
3714
3715 ret = ocfs2_journal_dirty(handle, root_bh);
3716 if (ret < 0) {
3717 mlog_errno(ret);
3718 goto leave;
3719 }
3720
3721leave:
3722 if (handle)
3723 ocfs2_commit_trans(osb, handle);
3724 if (data_ac)
3725 ocfs2_free_alloc_context(data_ac);
3726 if (meta_ac)
3727 ocfs2_free_alloc_context(meta_ac);
3728
3729 return ret;
3730}
3731
3732/*
3733 * Extend a new xattr bucket and move xattrs to the end one by one until
3734 * We meet with start_bh. Only move half of the xattrs to the bucket after it.
3735 */
3736static int ocfs2_extend_xattr_bucket(struct inode *inode,
3737 struct buffer_head *first_bh,
3738 struct buffer_head *start_bh,
3739 u32 num_clusters)
3740{
3741 int ret, credits;
3742 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3743 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3744 u64 start_blk = start_bh->b_blocknr, end_blk;
3745 u32 num_buckets = num_clusters * ocfs2_xattr_buckets_per_cluster(osb);
3746 handle_t *handle;
3747 struct ocfs2_xattr_header *first_xh =
3748 (struct ocfs2_xattr_header *)first_bh->b_data;
3749 u16 bucket = le16_to_cpu(first_xh->xh_num_buckets);
3750
3751 mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
3752 "from %llu, len = %u\n", start_blk,
3753 (unsigned long long)first_bh->b_blocknr, num_clusters);
3754
3755 BUG_ON(bucket >= num_buckets);
3756
3757 end_blk = first_bh->b_blocknr + (bucket - 1) * blk_per_bucket;
3758
3759 /*
3760 * We will touch all the buckets after the start_bh(include it).
3761 * Add one more bucket and modify the first_bh.
3762 */
3763 credits = end_blk - start_blk + 2 * blk_per_bucket + 1;
3764 handle = ocfs2_start_trans(osb, credits);
3765 if (IS_ERR(handle)) {
3766 ret = PTR_ERR(handle);
3767 handle = NULL;
3768 mlog_errno(ret);
3769 goto out;
3770 }
3771
3772 ret = ocfs2_journal_access(handle, inode, first_bh,
3773 OCFS2_JOURNAL_ACCESS_WRITE);
3774 if (ret) {
3775 mlog_errno(ret);
3776 goto commit;
3777 }
3778
3779 while (end_blk != start_blk) {
3780 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
3781 end_blk + blk_per_bucket, 0);
3782 if (ret)
3783 goto commit;
3784 end_blk -= blk_per_bucket;
3785 }
3786
3787 /* Move half of the xattr in start_blk to the next bucket. */
3788 ret = ocfs2_half_xattr_bucket(inode, handle, start_blk,
3789 start_blk + blk_per_bucket, NULL, 0);
3790
3791 le16_add_cpu(&first_xh->xh_num_buckets, 1);
3792 ocfs2_journal_dirty(handle, first_bh);
3793
3794commit:
3795 ocfs2_commit_trans(osb, handle);
3796out:
3797 return ret;
3798}
3799
3800/*
3801 * Add new xattr bucket in an extent record and adjust the buckets accordingly.
3802 * xb_bh is the ocfs2_xattr_block.
3803 * We will move all the buckets starting from header_bh to the next place. As
3804 * for this one, half num of its xattrs will be moved to the next one.
3805 *
3806 * We will allocate a new cluster if current cluster is full and adjust
3807 * header_bh and first_bh if the insert place is moved to the new cluster.
3808 */
3809static int ocfs2_add_new_xattr_bucket(struct inode *inode,
3810 struct buffer_head *xb_bh,
3811 struct buffer_head *header_bh)
3812{
3813 struct ocfs2_xattr_header *first_xh = NULL;
3814 struct buffer_head *first_bh = NULL;
3815 struct ocfs2_xattr_block *xb =
3816 (struct ocfs2_xattr_block *)xb_bh->b_data;
3817 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3818 struct ocfs2_extent_list *el = &xb_root->xt_list;
3819 struct ocfs2_xattr_header *xh =
3820 (struct ocfs2_xattr_header *)header_bh->b_data;
3821 u32 name_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3822 struct super_block *sb = inode->i_sb;
3823 struct ocfs2_super *osb = OCFS2_SB(sb);
3824 int ret, num_buckets, extend = 1;
3825 u64 p_blkno;
3826 u32 e_cpos, num_clusters;
3827
3828 mlog(0, "Add new xattr bucket starting form %llu\n",
3829 (unsigned long long)header_bh->b_blocknr);
3830
3831 /*
3832 * Add refrence for header_bh here because it may be
3833 * changed in ocfs2_add_new_xattr_cluster and we need
3834 * to free it in the end.
3835 */
3836 get_bh(header_bh);
3837
3838 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
3839 &num_clusters, el);
3840 if (ret) {
3841 mlog_errno(ret);
3842 goto out;
3843 }
3844
3845 ret = ocfs2_read_block(osb, p_blkno,
3846 &first_bh, OCFS2_BH_CACHED, inode);
3847 if (ret) {
3848 mlog_errno(ret);
3849 goto out;
3850 }
3851
3852 num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
3853 first_xh = (struct ocfs2_xattr_header *)first_bh->b_data;
3854
3855 if (num_buckets == le16_to_cpu(first_xh->xh_num_buckets)) {
3856 ret = ocfs2_add_new_xattr_cluster(inode,
3857 xb_bh,
3858 &first_bh,
3859 &header_bh,
3860 &num_clusters,
3861 e_cpos,
3862 p_blkno,
3863 &extend);
3864 if (ret) {
3865 mlog_errno(ret);
3866 goto out;
3867 }
3868 }
3869
3870 if (extend)
3871 ret = ocfs2_extend_xattr_bucket(inode,
3872 first_bh,
3873 header_bh,
3874 num_clusters);
3875 if (ret)
3876 mlog_errno(ret);
3877out:
3878 brelse(first_bh);
3879 brelse(header_bh);
3880 return ret;
3881}
3882
3883static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
3884 struct ocfs2_xattr_bucket *bucket,
3885 int offs)
3886{
3887 int block_off = offs >> inode->i_sb->s_blocksize_bits;
3888
3889 offs = offs % inode->i_sb->s_blocksize;
3890 return bucket->bhs[block_off]->b_data + offs;
3891}
3892
3893/*
3894 * Handle the normal xattr set, including replace, delete and new.
3895 * When the bucket is empty, "is_empty" is set and the caller can
3896 * free this bucket.
3897 *
3898 * Note: "local" indicates the real data's locality. So we can't
3899 * just its bucket locality by its length.
3900 */
3901static void ocfs2_xattr_set_entry_normal(struct inode *inode,
3902 struct ocfs2_xattr_info *xi,
3903 struct ocfs2_xattr_search *xs,
3904 u32 name_hash,
3905 int local,
3906 int *is_empty)
3907{
3908 struct ocfs2_xattr_entry *last, *xe;
3909 int name_len = strlen(xi->name);
3910 struct ocfs2_xattr_header *xh = xs->header;
3911 u16 count = le16_to_cpu(xh->xh_count), start;
3912 size_t blocksize = inode->i_sb->s_blocksize;
3913 char *val;
3914 size_t offs, size, new_size;
3915
3916 last = &xh->xh_entries[count];
3917 if (!xs->not_found) {
3918 xe = xs->here;
3919 offs = le16_to_cpu(xe->xe_name_offset);
3920 if (ocfs2_xattr_is_local(xe))
3921 size = OCFS2_XATTR_SIZE(name_len) +
3922 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3923 else
3924 size = OCFS2_XATTR_SIZE(name_len) +
3925 OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
3926
3927 /*
3928 * If the new value will be stored outside, xi->value has been
3929 * initalized as an empty ocfs2_xattr_value_root, and the same
3930 * goes with xi->value_len, so we can set new_size safely here.
3931 * See ocfs2_xattr_set_in_bucket.
3932 */
3933 new_size = OCFS2_XATTR_SIZE(name_len) +
3934 OCFS2_XATTR_SIZE(xi->value_len);
3935
3936 le16_add_cpu(&xh->xh_name_value_len, -size);
3937 if (xi->value) {
3938 if (new_size > size)
3939 goto set_new_name_value;
3940
3941 /* Now replace the old value with new one. */
3942 if (local)
3943 xe->xe_value_size = cpu_to_le64(xi->value_len);
3944 else
3945 xe->xe_value_size = 0;
3946
3947 val = ocfs2_xattr_bucket_get_val(inode,
3948 &xs->bucket, offs);
3949 memset(val + OCFS2_XATTR_SIZE(name_len), 0,
3950 size - OCFS2_XATTR_SIZE(name_len));
3951 if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
3952 memcpy(val + OCFS2_XATTR_SIZE(name_len),
3953 xi->value, xi->value_len);
3954
3955 le16_add_cpu(&xh->xh_name_value_len, new_size);
3956 ocfs2_xattr_set_local(xe, local);
3957 return;
3958 } else {
3959 /* Remove the old entry. */
3960 last -= 1;
3961 memmove(xe, xe + 1,
3962 (void *)last - (void *)xe);
3963 memset(last, 0, sizeof(struct ocfs2_xattr_entry));
3964 le16_add_cpu(&xh->xh_count, -1);
3965 if (xh->xh_count == 0 && is_empty)
3966 *is_empty = 1;
3967 return;
3968 }
3969 } else {
3970 /* find a new entry for insert. */
3971 int low = 0, high = count - 1, tmp;
3972 struct ocfs2_xattr_entry *tmp_xe;
3973
3974 while (low <= high) {
3975 tmp = (low + high) / 2;
3976 tmp_xe = &xh->xh_entries[tmp];
3977
3978 if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
3979 low = tmp + 1;
3980 else if (name_hash <
3981 le32_to_cpu(tmp_xe->xe_name_hash))
3982 high = tmp - 1;
3983 else
3984 break;
3985 }
3986
3987 xe = &xh->xh_entries[low];
3988 if (low != count)
3989 memmove(xe + 1, xe, (void *)last - (void *)xe);
3990
3991 le16_add_cpu(&xh->xh_count, 1);
3992 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
3993 xe->xe_name_hash = cpu_to_le32(name_hash);
3994 xe->xe_name_len = name_len;
3995 ocfs2_xattr_set_type(xe, xi->name_index);
3996 }
3997
3998set_new_name_value:
3999 /* Insert the new name+value. */
4000 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4001
4002 /*
4003 * We must make sure that the name/value pair
4004 * exists in the same block.
4005 */
4006 offs = le16_to_cpu(xh->xh_free_start);
4007 start = offs - size;
4008
4009 if (start >> inode->i_sb->s_blocksize_bits !=
4010 (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4011 offs = offs - offs % blocksize;
4012 xh->xh_free_start = cpu_to_le16(offs);
4013 }
4014
4015 val = ocfs2_xattr_bucket_get_val(inode,
4016 &xs->bucket, offs - size);
4017 xe->xe_name_offset = cpu_to_le16(offs - size);
4018
4019 memset(val, 0, size);
4020 memcpy(val, xi->name, name_len);
4021 memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4022
4023 xe->xe_value_size = cpu_to_le64(xi->value_len);
4024 ocfs2_xattr_set_local(xe, local);
4025 xs->here = xe;
4026 le16_add_cpu(&xh->xh_free_start, -size);
4027 le16_add_cpu(&xh->xh_name_value_len, size);
4028
4029 return;
4030}
4031
4032static int ocfs2_xattr_bucket_handle_journal(struct inode *inode,
4033 handle_t *handle,
4034 struct ocfs2_xattr_search *xs,
4035 struct buffer_head **bhs,
4036 u16 bh_num)
4037{
4038 int ret = 0, off, block_off;
4039 struct ocfs2_xattr_entry *xe = xs->here;
4040
4041 /*
4042 * First calculate all the blocks we should journal_access
4043 * and journal_dirty. The first block should always be touched.
4044 */
4045 ret = ocfs2_journal_dirty(handle, bhs[0]);
4046 if (ret)
4047 mlog_errno(ret);
4048
4049 /* calc the data. */
4050 off = le16_to_cpu(xe->xe_name_offset);
4051 block_off = off >> inode->i_sb->s_blocksize_bits;
4052 ret = ocfs2_journal_dirty(handle, bhs[block_off]);
4053 if (ret)
4054 mlog_errno(ret);
4055
4056 return ret;
4057}
4058
4059/*
4060 * Set the xattr entry in the specified bucket.
4061 * The bucket is indicated by xs->bucket and it should have the enough
4062 * space for the xattr insertion.
4063 */
4064static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4065 struct ocfs2_xattr_info *xi,
4066 struct ocfs2_xattr_search *xs,
4067 u32 name_hash,
4068 int local,
4069 int *bucket_empty)
4070{
4071 int i, ret;
4072 handle_t *handle = NULL;
4073 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4074 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4075
4076 mlog(0, "Set xattr entry len = %d index = %d in bucket %llu\n",
4077 xi->value_len, xi->name_index,
4078 (unsigned long long)xs->bucket.bhs[0]->b_blocknr);
4079
4080 if (!xs->bucket.bhs[1]) {
4081 ret = ocfs2_read_blocks(osb,
4082 xs->bucket.bhs[0]->b_blocknr + 1,
4083 blk_per_bucket - 1, &xs->bucket.bhs[1],
4084 OCFS2_BH_CACHED, inode);
4085 if (ret) {
4086 mlog_errno(ret);
4087 goto out;
4088 }
4089 }
4090
4091 handle = ocfs2_start_trans(osb, blk_per_bucket);
4092 if (IS_ERR(handle)) {
4093 ret = PTR_ERR(handle);
4094 handle = NULL;
4095 mlog_errno(ret);
4096 goto out;
4097 }
4098
4099 for (i = 0; i < blk_per_bucket; i++) {
4100 ret = ocfs2_journal_access(handle, inode, xs->bucket.bhs[i],
4101 OCFS2_JOURNAL_ACCESS_WRITE);
4102 if (ret < 0) {
4103 mlog_errno(ret);
4104 goto out;
4105 }
4106 }
4107
4108 ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash,
4109 local, bucket_empty);
4110
4111 /*Only dirty the blocks we have touched in set xattr. */
4112 ret = ocfs2_xattr_bucket_handle_journal(inode, handle, xs,
4113 xs->bucket.bhs, blk_per_bucket);
4114 if (ret)
4115 mlog_errno(ret);
4116out:
4117 ocfs2_commit_trans(osb, handle);
4118
4119 return ret;
4120}
4121
4122static int ocfs2_xattr_value_update_size(struct inode *inode,
4123 struct buffer_head *xe_bh,
4124 struct ocfs2_xattr_entry *xe,
4125 u64 new_size)
4126{
4127 int ret;
4128 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4129 handle_t *handle = NULL;
4130
4131 handle = ocfs2_start_trans(osb, 1);
4132 if (handle == NULL) {
4133 ret = -ENOMEM;
4134 mlog_errno(ret);
4135 goto out;
4136 }
4137
4138 ret = ocfs2_journal_access(handle, inode, xe_bh,
4139 OCFS2_JOURNAL_ACCESS_WRITE);
4140 if (ret < 0) {
4141 mlog_errno(ret);
4142 goto out_commit;
4143 }
4144
4145 xe->xe_value_size = cpu_to_le64(new_size);
4146
4147 ret = ocfs2_journal_dirty(handle, xe_bh);
4148 if (ret < 0)
4149 mlog_errno(ret);
4150
4151out_commit:
4152 ocfs2_commit_trans(osb, handle);
4153out:
4154 return ret;
4155}
4156
4157/*
4158 * Truncate the specified xe_off entry in xattr bucket.
4159 * bucket is indicated by header_bh and len is the new length.
4160 * Both the ocfs2_xattr_value_root and the entry will be updated here.
4161 *
4162 * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4163 */
4164static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4165 struct buffer_head *header_bh,
4166 int xe_off,
4167 int len)
4168{
4169 int ret, offset;
4170 u64 value_blk;
4171 struct buffer_head *value_bh = NULL;
4172 struct ocfs2_xattr_value_root *xv;
4173 struct ocfs2_xattr_entry *xe;
4174 struct ocfs2_xattr_header *xh =
4175 (struct ocfs2_xattr_header *)header_bh->b_data;
4176 size_t blocksize = inode->i_sb->s_blocksize;
4177
4178 xe = &xh->xh_entries[xe_off];
4179
4180 BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4181
4182 offset = le16_to_cpu(xe->xe_name_offset) +
4183 OCFS2_XATTR_SIZE(xe->xe_name_len);
4184
4185 value_blk = offset / blocksize;
4186
4187 /* We don't allow ocfs2_xattr_value to be stored in different block. */
4188 BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4189 value_blk += header_bh->b_blocknr;
4190
4191 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), value_blk,
4192 &value_bh, OCFS2_BH_CACHED, inode);
4193 if (ret) {
4194 mlog_errno(ret);
4195 goto out;
4196 }
4197
4198 xv = (struct ocfs2_xattr_value_root *)
4199 (value_bh->b_data + offset % blocksize);
4200
4201 mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4202 xe_off, (unsigned long long)header_bh->b_blocknr, len);
4203 ret = ocfs2_xattr_value_truncate(inode, value_bh, xv, len);
4204 if (ret) {
4205 mlog_errno(ret);
4206 goto out;
4207 }
4208
4209 ret = ocfs2_xattr_value_update_size(inode, header_bh, xe, len);
4210 if (ret) {
4211 mlog_errno(ret);
4212 goto out;
4213 }
4214
4215out:
4216 brelse(value_bh);
4217 return ret;
4218}
4219
4220static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4221 struct ocfs2_xattr_search *xs,
4222 int len)
4223{
4224 int ret, offset;
4225 struct ocfs2_xattr_entry *xe = xs->here;
4226 struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4227
4228 BUG_ON(!xs->bucket.bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4229
4230 offset = xe - xh->xh_entries;
4231 ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket.bhs[0],
4232 offset, len);
4233 if (ret)
4234 mlog_errno(ret);
4235
4236 return ret;
4237}
4238
4239static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4240 struct ocfs2_xattr_search *xs,
4241 char *val,
4242 int value_len)
4243{
4244 int offset;
4245 struct ocfs2_xattr_value_root *xv;
4246 struct ocfs2_xattr_entry *xe = xs->here;
4247
4248 BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4249
4250 offset = le16_to_cpu(xe->xe_name_offset) +
4251 OCFS2_XATTR_SIZE(xe->xe_name_len);
4252
4253 xv = (struct ocfs2_xattr_value_root *)(xs->base + offset);
4254
4255 return __ocfs2_xattr_set_value_outside(inode, xv, val, value_len);
4256}
4257
4258/*
4259 * Remove the xattr bucket pointed by bucket_bh.
4260 * All the buckets after it in the same xattr extent rec will be
4261 * move forward one by one.
4262 */
4263static int ocfs2_rm_xattr_bucket(struct inode *inode,
4264 struct buffer_head *first_bh,
4265 struct ocfs2_xattr_bucket *bucket)
4266{
4267 int ret = 0, credits;
4268 struct ocfs2_xattr_header *xh =
4269 (struct ocfs2_xattr_header *)first_bh->b_data;
4270 u16 bucket_num = le16_to_cpu(xh->xh_num_buckets);
4271 u64 end, start = bucket->bhs[0]->b_blocknr;
4272 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4273 handle_t *handle;
4274 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4275
4276 end = first_bh->b_blocknr + (bucket_num - 1) * blk_per_bucket;
4277
4278 mlog(0, "rm xattr bucket %llu\n", start);
4279 /*
4280 * We need to update the first xattr_header and all the buckets starting
4281 * from start in this xattr rec.
4282 *
4283 * XXX: Should we empty the old last bucket here?
4284 */
4285 credits = 1 + end - start;
4286 handle = ocfs2_start_trans(osb, credits);
4287 if (IS_ERR(handle)) {
4288 ret = PTR_ERR(handle);
4289 mlog_errno(ret);
4290 return ret;
4291 }
4292
4293 ret = ocfs2_journal_access(handle, inode, first_bh,
4294 OCFS2_JOURNAL_ACCESS_WRITE);
4295 if (ret) {
4296 mlog_errno(ret);
4297 goto out_commit;
4298 }
4299
4300
4301 while (start < end) {
4302 ret = ocfs2_cp_xattr_bucket(inode, handle,
4303 start + blk_per_bucket,
4304 start, 0);
4305 if (ret) {
4306 mlog_errno(ret);
4307 goto out_commit;
4308 }
4309 start += blk_per_bucket;
4310 }
4311
4312 /* update the first_bh. */
4313 xh->xh_num_buckets = cpu_to_le16(bucket_num - 1);
4314 ocfs2_journal_dirty(handle, first_bh);
4315
4316out_commit:
4317 ocfs2_commit_trans(osb, handle);
4318 return ret;
4319}
4320
4321static int ocfs2_rm_xattr_cluster(struct inode *inode,
4322 struct buffer_head *root_bh,
4323 u64 blkno,
4324 u32 cpos,
4325 u32 len)
4326{
4327 int ret;
4328 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4329 struct inode *tl_inode = osb->osb_tl_inode;
4330 handle_t *handle;
4331 struct ocfs2_xattr_block *xb =
4332 (struct ocfs2_xattr_block *)root_bh->b_data;
4333 struct ocfs2_extent_list *root_el = &xb->xb_attrs.xb_root.xt_list;
4334 struct ocfs2_alloc_context *meta_ac = NULL;
4335 struct ocfs2_cached_dealloc_ctxt dealloc;
4336
4337 ocfs2_init_dealloc_ctxt(&dealloc);
4338
4339 mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4340 cpos, len, (unsigned long long)blkno);
4341
4342 ocfs2_remove_xattr_clusters_from_cache(inode, blkno, len);
4343
4344 ret = ocfs2_lock_allocators(inode, root_bh, root_el,
4345 0, 1, NULL, &meta_ac,
4346 OCFS2_XATTR_TREE_EXTENT, NULL);
4347 if (ret) {
4348 mlog_errno(ret);
4349 return ret;
4350 }
4351
4352 mutex_lock(&tl_inode->i_mutex);
4353
4354 if (ocfs2_truncate_log_needs_flush(osb)) {
4355 ret = __ocfs2_flush_truncate_log(osb);
4356 if (ret < 0) {
4357 mlog_errno(ret);
4358 goto out;
4359 }
4360 }
4361
4362 handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
4363 if (handle == NULL) {
4364 ret = -ENOMEM;
4365 mlog_errno(ret);
4366 goto out;
4367 }
4368
4369 ret = ocfs2_journal_access(handle, inode, root_bh,
4370 OCFS2_JOURNAL_ACCESS_WRITE);
4371 if (ret) {
4372 mlog_errno(ret);
4373 goto out_commit;
4374 }
4375
4376 ret = ocfs2_remove_extent(inode, root_bh, cpos, len, handle, meta_ac,
4377 &dealloc, OCFS2_XATTR_TREE_EXTENT, NULL);
4378 if (ret) {
4379 mlog_errno(ret);
4380 goto out_commit;
4381 }
4382
4383 le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4384
4385 ret = ocfs2_journal_dirty(handle, root_bh);
4386 if (ret) {
4387 mlog_errno(ret);
4388 goto out_commit;
4389 }
4390
4391 ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4392 if (ret)
4393 mlog_errno(ret);
4394
4395out_commit:
4396 ocfs2_commit_trans(osb, handle);
4397out:
4398 ocfs2_schedule_truncate_log_flush(osb, 1);
4399
4400 mutex_unlock(&tl_inode->i_mutex);
4401
4402 if (meta_ac)
4403 ocfs2_free_alloc_context(meta_ac);
4404
4405 ocfs2_run_deallocs(osb, &dealloc);
4406
4407 return ret;
4408}
4409
4410/*
4411 * Free the xattr bucket indicated by xs->bucket and if all the buckets
4412 * in the clusters is free, free the clusters also.
4413 */
4414static int ocfs2_xattr_bucket_shrink(struct inode *inode,
4415 struct ocfs2_xattr_info *xi,
4416 struct ocfs2_xattr_search *xs,
4417 u32 name_hash)
4418{
4419 int ret;
4420 u32 e_cpos, num_clusters;
4421 u64 p_blkno;
4422 struct buffer_head *first_bh = NULL;
4423 struct ocfs2_xattr_header *first_xh;
4424 struct ocfs2_xattr_block *xb =
4425 (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
4426
4427 BUG_ON(xs->header->xh_count != 0);
4428
4429 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4430 &e_cpos, &num_clusters,
4431 &xb->xb_attrs.xb_root.xt_list);
4432 if (ret) {
4433 mlog_errno(ret);
4434 return ret;
4435 }
4436
4437 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno,
4438 &first_bh, OCFS2_BH_CACHED, inode);
4439 if (ret) {
4440 mlog_errno(ret);
4441 return ret;
4442 }
4443
4444 ret = ocfs2_rm_xattr_bucket(inode, first_bh, &xs->bucket);
4445 if (ret) {
4446 mlog_errno(ret);
4447 goto out;
4448 }
4449
4450 first_xh = (struct ocfs2_xattr_header *)first_bh->b_data;
4451 if (first_xh->xh_num_buckets == 0)
4452 ret = ocfs2_rm_xattr_cluster(inode, xs->xattr_bh,
4453 p_blkno, e_cpos,
4454 num_clusters);
4455
4456out:
4457 brelse(first_bh);
4458 return ret;
4459}
4460
4461static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4462 struct ocfs2_xattr_search *xs)
4463{
4464 handle_t *handle = NULL;
4465 struct ocfs2_xattr_header *xh = xs->bucket.xh;
4466 struct ocfs2_xattr_entry *last = &xh->xh_entries[
4467 le16_to_cpu(xh->xh_count) - 1];
4468 int ret = 0;
4469
4470 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), 1);
4471 if (IS_ERR(handle)) {
4472 ret = PTR_ERR(handle);
4473 mlog_errno(ret);
4474 return;
4475 }
4476
4477 ret = ocfs2_journal_access(handle, inode, xs->bucket.bhs[0],
4478 OCFS2_JOURNAL_ACCESS_WRITE);
4479 if (ret) {
4480 mlog_errno(ret);
4481 goto out_commit;
4482 }
4483
4484 /* Remove the old entry. */
4485 memmove(xs->here, xs->here + 1,
4486 (void *)last - (void *)xs->here);
4487 memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4488 le16_add_cpu(&xh->xh_count, -1);
4489
4490 ret = ocfs2_journal_dirty(handle, xs->bucket.bhs[0]);
4491 if (ret < 0)
4492 mlog_errno(ret);
4493out_commit:
4494 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
4495}
4496
4497/*
4498 * Set the xattr name/value in the bucket specified in xs.
4499 *
4500 * As the new value in xi may be stored in the bucket or in an outside cluster,
4501 * we divide the whole process into 3 steps:
4502 * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4503 * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4504 * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4505 * 4. If the clusters for the new outside value can't be allocated, we need
4506 * to free the xattr we allocated in set.
4507 */
4508static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4509 struct ocfs2_xattr_info *xi,
4510 struct ocfs2_xattr_search *xs)
4511{
4512 int ret, local = 1, bucket_empty = 0;
4513 size_t value_len;
4514 char *val = (char *)xi->value;
4515 struct ocfs2_xattr_entry *xe = xs->here;
4516 u32 name_hash = ocfs2_xattr_hash_by_name(inode,
4517 xi->name_index, xi->name);
4518
4519 if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
4520 /*
4521 * We need to truncate the xattr storage first.
4522 *
4523 * If both the old and new value are stored to
4524 * outside block, we only need to truncate
4525 * the storage and then set the value outside.
4526 *
4527 * If the new value should be stored within block,
4528 * we should free all the outside block first and
4529 * the modification to the xattr block will be done
4530 * by following steps.
4531 */
4532 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4533 value_len = xi->value_len;
4534 else
4535 value_len = 0;
4536
4537 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4538 value_len);
4539 if (ret)
4540 goto out;
4541
4542 if (value_len)
4543 goto set_value_outside;
4544 }
4545
4546 value_len = xi->value_len;
4547 /* So we have to handle the inside block change now. */
4548 if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4549 /*
4550 * If the new value will be stored outside of block,
4551 * initalize a new empty value root and insert it first.
4552 */
4553 local = 0;
4554 xi->value = &def_xv;
4555 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
4556 }
4557
4558 ret = ocfs2_xattr_set_entry_in_bucket(inode, xi, xs, name_hash,
4559 local, &bucket_empty);
4560 if (ret) {
4561 mlog_errno(ret);
4562 goto out;
4563 }
4564
4565 if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4566 /* allocate the space now for the outside block storage. */
4567 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4568 value_len);
4569 if (ret) {
4570 mlog_errno(ret);
4571
4572 if (xs->not_found) {
4573 /*
4574 * We can't allocate enough clusters for outside
4575 * storage and we have allocated xattr already,
4576 * so need to remove it.
4577 */
4578 ocfs2_xattr_bucket_remove_xs(inode, xs);
4579 }
4580 goto out;
4581 }
4582 } else {
4583 if (bucket_empty)
4584 ret = ocfs2_xattr_bucket_shrink(inode, xi,
4585 xs, name_hash);
4586 goto out;
4587 }
4588
4589set_value_outside:
4590 ret = ocfs2_xattr_bucket_set_value_outside(inode, xs, val, value_len);
4591out:
4592 return ret;
4593}
4594
4595/* check whether the xattr bucket is filled up with the same hash value. */
4596static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
4597 struct ocfs2_xattr_bucket *bucket)
4598{
4599 struct ocfs2_xattr_header *xh = bucket->xh;
4600
4601 if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
4602 xh->xh_entries[0].xe_name_hash) {
4603 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
4604 "hash = %u\n",
4605 (unsigned long long)bucket->bhs[0]->b_blocknr,
4606 le32_to_cpu(xh->xh_entries[0].xe_name_hash));
4607 return -ENOSPC;
4608 }
4609
4610 return 0;
4611}
4612
4613static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
4614 struct ocfs2_xattr_info *xi,
4615 struct ocfs2_xattr_search *xs)
4616{
4617 struct ocfs2_xattr_header *xh;
4618 struct ocfs2_xattr_entry *xe;
4619 u16 count, header_size, xh_free_start;
4620 int i, free, max_free, need, old;
4621 size_t value_size = 0, name_len = strlen(xi->name);
4622 size_t blocksize = inode->i_sb->s_blocksize;
4623 int ret, allocation = 0;
4624 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4625
4626 mlog_entry("Set xattr %s in xattr index block\n", xi->name);
4627
4628try_again:
4629 xh = xs->header;
4630 count = le16_to_cpu(xh->xh_count);
4631 xh_free_start = le16_to_cpu(xh->xh_free_start);
4632 header_size = sizeof(struct ocfs2_xattr_header) +
4633 count * sizeof(struct ocfs2_xattr_entry);
4634 max_free = OCFS2_XATTR_BUCKET_SIZE -
4635 le16_to_cpu(xh->xh_name_value_len) - header_size;
4636
4637 mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
4638 "of %u which exceed block size\n",
4639 (unsigned long long)xs->bucket.bhs[0]->b_blocknr,
4640 header_size);
4641
4642 if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4643 value_size = OCFS2_XATTR_ROOT_SIZE;
4644 else if (xi->value)
4645 value_size = OCFS2_XATTR_SIZE(xi->value_len);
4646
4647 if (xs->not_found)
4648 need = sizeof(struct ocfs2_xattr_entry) +
4649 OCFS2_XATTR_SIZE(name_len) + value_size;
4650 else {
4651 need = value_size + OCFS2_XATTR_SIZE(name_len);
4652
4653 /*
4654 * We only replace the old value if the new length is smaller
4655 * than the old one. Otherwise we will allocate new space in the
4656 * bucket to store it.
4657 */
4658 xe = xs->here;
4659 if (ocfs2_xattr_is_local(xe))
4660 old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4661 else
4662 old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4663
4664 if (old >= value_size)
4665 need = 0;
4666 }
4667
4668 free = xh_free_start - header_size;
4669 /*
4670 * We need to make sure the new name/value pair
4671 * can exist in the same block.
4672 */
4673 if (xh_free_start % blocksize < need)
4674 free -= xh_free_start % blocksize;
4675
4676 mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
4677 "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
4678 " %u\n", xs->not_found,
4679 (unsigned long long)xs->bucket.bhs[0]->b_blocknr,
4680 free, need, max_free, le16_to_cpu(xh->xh_free_start),
4681 le16_to_cpu(xh->xh_name_value_len));
4682
4683 if (free < need || count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
4684 if (need <= max_free &&
4685 count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
4686 /*
4687 * We can create the space by defragment. Since only the
4688 * name/value will be moved, the xe shouldn't be changed
4689 * in xs.
4690 */
4691 ret = ocfs2_defrag_xattr_bucket(inode, &xs->bucket);
4692 if (ret) {
4693 mlog_errno(ret);
4694 goto out;
4695 }
4696
4697 xh_free_start = le16_to_cpu(xh->xh_free_start);
4698 free = xh_free_start - header_size;
4699 if (xh_free_start % blocksize < need)
4700 free -= xh_free_start % blocksize;
4701
4702 if (free >= need)
4703 goto xattr_set;
4704
4705 mlog(0, "Can't get enough space for xattr insert by "
4706 "defragment. Need %u bytes, but we have %d, so "
4707 "allocate new bucket for it.\n", need, free);
4708 }
4709
4710 /*
4711 * We have to add new buckets or clusters and one
4712 * allocation should leave us enough space for insert.
4713 */
4714 BUG_ON(allocation);
4715
4716 /*
4717 * We do not allow for overlapping ranges between buckets. And
4718 * the maximum number of collisions we will allow for then is
4719 * one bucket's worth, so check it here whether we need to
4720 * add a new bucket for the insert.
4721 */
4722 ret = ocfs2_check_xattr_bucket_collision(inode, &xs->bucket);
4723 if (ret) {
4724 mlog_errno(ret);
4725 goto out;
4726 }
4727
4728 ret = ocfs2_add_new_xattr_bucket(inode,
4729 xs->xattr_bh,
4730 xs->bucket.bhs[0]);
4731 if (ret) {
4732 mlog_errno(ret);
4733 goto out;
4734 }
4735
4736 for (i = 0; i < blk_per_bucket; i++)
4737 brelse(xs->bucket.bhs[i]);
4738
4739 memset(&xs->bucket, 0, sizeof(xs->bucket));
4740
4741 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
4742 xi->name_index,
4743 xi->name, xs);
4744 if (ret && ret != -ENODATA)
4745 goto out;
4746 xs->not_found = ret;
4747 allocation = 1;
4748 goto try_again;
4749 }
4750
4751xattr_set:
4752 ret = ocfs2_xattr_set_in_bucket(inode, xi, xs);
4753out:
4754 mlog_exit(ret);
4755 return ret;
4756}
a3944256
TM		 4757
4758static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
4759 struct ocfs2_xattr_bucket *bucket,
4760 void *para)
4761{
4762 int ret = 0;
4763 struct ocfs2_xattr_header *xh = bucket->xh;
4764 u16 i;
4765 struct ocfs2_xattr_entry *xe;
4766
4767 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4768 xe = &xh->xh_entries[i];
4769 if (ocfs2_xattr_is_local(xe))
4770 continue;
4771
4772 ret = ocfs2_xattr_bucket_value_truncate(inode,
4773 bucket->bhs[0],
4774 i, 0);
4775 if (ret) {
4776 mlog_errno(ret);
4777 break;
4778 }
4779 }
4780
4781 return ret;
4782}
4783
4784static int ocfs2_delete_xattr_index_block(struct inode *inode,
4785 struct buffer_head *xb_bh)
4786{
4787 struct ocfs2_xattr_block *xb =
4788 (struct ocfs2_xattr_block *)xb_bh->b_data;
4789 struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4790 int ret = 0;
4791 u32 name_hash = UINT_MAX, e_cpos, num_clusters;
4792 u64 p_blkno;
4793
4794 if (le16_to_cpu(el->l_next_free_rec) == 0)
4795 return 0;
4796
4797 while (name_hash > 0) {
4798 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4799 &e_cpos, &num_clusters, el);
4800 if (ret) {
4801 mlog_errno(ret);
4802 goto out;
4803 }
4804
4805 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
4806 ocfs2_delete_xattr_in_bucket,
4807 NULL);
4808 if (ret) {
4809 mlog_errno(ret);
4810 goto out;
4811 }
4812
4813 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
4814 p_blkno, e_cpos, num_clusters);
4815 if (ret) {
4816 mlog_errno(ret);
4817 break;
4818 }
4819
4820 if (e_cpos == 0)
4821 break;
4822
4823 name_hash = e_cpos - 1;
4824 }
4825
4826out:
4827 return ret;
4828}
Linux kernel - Deliverables
This page took 0.211364 seconds and 5 git commands to generate.