]> bbs.cooldavid.org Git - net-next-2.6.git/blame - fs/ocfs2/xattr.c
git://bbs.cooldavid.org / net-next-2.6.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
ocfs2: Dirty the entire bucket in ocfs2_bucket_value_truncate()
[net-next-2.6.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 *
c3cb6827 6 * Copyright (C) 2004, 2008 Oracle. All rights reserved.
f56654c4 7 *
cf1d6c76 8 * CREDITS:
c3cb6827
TY		 9 * Lots of code in this file is copy from linux/fs/ext3/xattr.c.
10 * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
cf1d6c76 11 *
f56654c4
TM		 12 * This program is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU General Public
c3cb6827 14 * License version 2 as published by the Free Software Foundation.
f56654c4
TM		 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.
f56654c4
TM		 20 */
21
cf1d6c76
TY		 22#include <linux/capability.h>
23#include <linux/fs.h>
24#include <linux/types.h>
25#include <linux/slab.h>
26#include <linux/highmem.h>
27#include <linux/pagemap.h>
28#include <linux/uio.h>
29#include <linux/sched.h>
30#include <linux/splice.h>
31#include <linux/mount.h>
32#include <linux/writeback.h>
33#include <linux/falloc.h>
01225596 34#include <linux/sort.h>
99219aea
MF		 35#include <linux/init.h>
36#include <linux/module.h>
37#include <linux/string.h>
923f7f31 38#include <linux/security.h>
cf1d6c76 39
f56654c4
TM		 40#define MLOG_MASK_PREFIX ML_XATTR
41#include <cluster/masklog.h>
42
43#include "ocfs2.h"
44#include "alloc.h"
45#include "dlmglue.h"
46#include "file.h"
cf1d6c76
TY		 47#include "symlink.h"
48#include "sysfile.h"
f56654c4
TM		 49#include "inode.h"
50#include "journal.h"
51#include "ocfs2_fs.h"
52#include "suballoc.h"
53#include "uptodate.h"
54#include "buffer_head_io.h"
0c044f0b 55#include "super.h"
cf1d6c76
TY		 56#include "xattr.h"
57
58
59struct ocfs2_xattr_def_value_root {
60 struct ocfs2_xattr_value_root xv;
61 struct ocfs2_extent_rec er;
62};
63
0c044f0b 64struct ocfs2_xattr_bucket {
ba937127
JB		 65 /* The inode these xattrs are associated with */
66 struct inode *bu_inode;
67
68 /* The actual buffers that make up the bucket */
4ac6032d 69 struct buffer_head *bu_bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
ba937127
JB		 70
71 /* How many blocks make up one bucket for this filesystem */
72 int bu_blocks;
0c044f0b
TM		 73};
74
78f30c31 75struct ocfs2_xattr_set_ctxt {
85db90e7 76 handle_t *handle;
78f30c31
TM		 77 struct ocfs2_alloc_context *meta_ac;
78 struct ocfs2_alloc_context *data_ac;
79 struct ocfs2_cached_dealloc_ctxt dealloc;
80};
81
cf1d6c76
TY		 82#define OCFS2_XATTR_ROOT_SIZE (sizeof(struct ocfs2_xattr_def_value_root))
83#define OCFS2_XATTR_INLINE_SIZE 80
534eaddd
TY		 84#define OCFS2_XATTR_FREE_IN_IBODY (OCFS2_MIN_XATTR_INLINE_SIZE \
85 - sizeof(struct ocfs2_xattr_header) \
86 - sizeof(__u32))
89c38bd0
TY		 87#define OCFS2_XATTR_FREE_IN_BLOCK(ptr) ((ptr)->i_sb->s_blocksize \
88 - sizeof(struct ocfs2_xattr_block) \
89 - sizeof(struct ocfs2_xattr_header) \
90 - sizeof(__u32))
cf1d6c76
TY		 91
92static struct ocfs2_xattr_def_value_root def_xv = {
93 .xv.xr_list.l_count = cpu_to_le16(1),
94};
95
96struct xattr_handler *ocfs2_xattr_handlers[] = {
97 &ocfs2_xattr_user_handler,
929fb014
TY		 98#ifdef CONFIG_OCFS2_FS_POSIX_ACL
99 &ocfs2_xattr_acl_access_handler,
100 &ocfs2_xattr_acl_default_handler,
101#endif
cf1d6c76 102 &ocfs2_xattr_trusted_handler,
923f7f31 103 &ocfs2_xattr_security_handler,
cf1d6c76
TY		 104 NULL
105};
106
c988fd04 107static struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
cf1d6c76 108 [OCFS2_XATTR_INDEX_USER] = &ocfs2_xattr_user_handler,
929fb014
TY		 109#ifdef CONFIG_OCFS2_FS_POSIX_ACL
110 [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS]
111 = &ocfs2_xattr_acl_access_handler,
112 [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT]
113 = &ocfs2_xattr_acl_default_handler,
114#endif
cf1d6c76 115 [OCFS2_XATTR_INDEX_TRUSTED] = &ocfs2_xattr_trusted_handler,
923f7f31 116 [OCFS2_XATTR_INDEX_SECURITY] = &ocfs2_xattr_security_handler,
cf1d6c76
TY		 117};
118
119struct ocfs2_xattr_info {
120 int name_index;
121 const char *name;
122 const void *value;
123 size_t value_len;
124};
125
126struct ocfs2_xattr_search {
127 struct buffer_head *inode_bh;
128 /*
129 * xattr_bh point to the block buffer head which has extended attribute
130 * when extended attribute in inode, xattr_bh is equal to inode_bh.
131 */
132 struct buffer_head *xattr_bh;
133 struct ocfs2_xattr_header *header;
ba937127 134 struct ocfs2_xattr_bucket *bucket;
cf1d6c76
TY		 135 void *base;
136 void *end;
137 struct ocfs2_xattr_entry *here;
138 int not_found;
139};
140
589dc260
TM		 141static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
142 struct ocfs2_xattr_header *xh,
143 int index,
144 int *block_off,
145 int *new_offset);
146
54f443f4
JB		 147static int ocfs2_xattr_block_find(struct inode *inode,
148 int name_index,
149 const char *name,
150 struct ocfs2_xattr_search *xs);
589dc260
TM		 151static int ocfs2_xattr_index_block_find(struct inode *inode,
152 struct buffer_head *root_bh,
153 int name_index,
154 const char *name,
155 struct ocfs2_xattr_search *xs);
156
0c044f0b
TM		 157static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
158 struct ocfs2_xattr_tree_root *xt,
159 char *buffer,
160 size_t buffer_size);
161
01225596 162static int ocfs2_xattr_create_index_block(struct inode *inode,
78f30c31
TM		 163 struct ocfs2_xattr_search *xs,
164 struct ocfs2_xattr_set_ctxt *ctxt);
01225596
TM		 165
166static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
167 struct ocfs2_xattr_info *xi,
78f30c31
TM		 168 struct ocfs2_xattr_search *xs,
169 struct ocfs2_xattr_set_ctxt *ctxt);
01225596 170
a3944256
TM		 171static int ocfs2_delete_xattr_index_block(struct inode *inode,
172 struct buffer_head *xb_bh);
173
0030e001
TY		 174static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
175{
176 return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
177}
178
179static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
180{
181 return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
182}
183
184static inline u16 ocfs2_xattr_max_xe_in_bucket(struct super_block *sb)
185{
186 u16 len = sb->s_blocksize -
187 offsetof(struct ocfs2_xattr_header, xh_entries);
188
189 return len / sizeof(struct ocfs2_xattr_entry);
190}
191
9c7759aa 192#define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
51def39f 193#define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
3e632946 194#define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
9c7759aa 195
ba937127 196static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
6dde41d9 197{
ba937127
JB		 198 struct ocfs2_xattr_bucket *bucket;
199 int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
6dde41d9 200
ba937127
JB		 201 BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
202
203 bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
204 if (bucket) {
205 bucket->bu_inode = inode;
206 bucket->bu_blocks = blks;
207 }
208
209 return bucket;
210}
211
212static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
213{
214 int i;
215
216 for (i = 0; i < bucket->bu_blocks; i++) {
6dde41d9
JB		 217 brelse(bucket->bu_bhs[i]);
218 bucket->bu_bhs[i] = NULL;
219 }
220}
221
ba937127
JB		 222static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
223{
224 if (bucket) {
225 ocfs2_xattr_bucket_relse(bucket);
226 bucket->bu_inode = NULL;
227 kfree(bucket);
228 }
229}
230
784b816a
JB		 231/*
232 * A bucket that has never been written to disk doesn't need to be
233 * read. We just need the buffer_heads. Don't call this for
234 * buckets that are already on disk. ocfs2_read_xattr_bucket() initializes
235 * them fully.
236 */
ba937127 237static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
784b816a
JB		 238 u64 xb_blkno)
239{
240 int i, rc = 0;
784b816a 241
ba937127
JB		 242 for (i = 0; i < bucket->bu_blocks; i++) {
243 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
244 xb_blkno + i);
784b816a
JB		 245 if (!bucket->bu_bhs[i]) {
246 rc = -EIO;
247 mlog_errno(rc);
248 break;
249 }
250
757055ad
TM		 251 if (!ocfs2_buffer_uptodate(bucket->bu_inode,
252 bucket->bu_bhs[i]))
253 ocfs2_set_new_buffer_uptodate(bucket->bu_inode,
254 bucket->bu_bhs[i]);
784b816a
JB		 255 }
256
257 if (rc)
ba937127 258 ocfs2_xattr_bucket_relse(bucket);
784b816a
JB		 259 return rc;
260}
261
262/* Read the xattr bucket at xb_blkno */
ba937127 263static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
784b816a
JB		 264 u64 xb_blkno)
265{
ba937127 266 int rc;
784b816a 267
ba937127 268 rc = ocfs2_read_blocks(bucket->bu_inode, xb_blkno,
970e4936
JB		 269 bucket->bu_blocks, bucket->bu_bhs, 0,
270 NULL);
784b816a 271 if (rc)
ba937127 272 ocfs2_xattr_bucket_relse(bucket);
784b816a
JB		 273 return rc;
274}
275
1224be02 276static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
1224be02
JB		 277 struct ocfs2_xattr_bucket *bucket,
278 int type)
279{
280 int i, rc = 0;
1224be02 281
ba937127
JB		 282 for (i = 0; i < bucket->bu_blocks; i++) {
283 rc = ocfs2_journal_access(handle, bucket->bu_inode,
1224be02
JB		 284 bucket->bu_bhs[i], type);
285 if (rc) {
286 mlog_errno(rc);
287 break;
288 }
289 }
290
291 return rc;
292}
293
294static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
1224be02
JB		 295 struct ocfs2_xattr_bucket *bucket)
296{
ba937127 297 int i;
1224be02 298
ba937127 299 for (i = 0; i < bucket->bu_blocks; i++)
1224be02
JB		 300 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
301}
302
ba937127 303static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
4980c6da
JB		 304 struct ocfs2_xattr_bucket *src)
305{
306 int i;
ba937127
JB		 307 int blocksize = src->bu_inode->i_sb->s_blocksize;
308
309 BUG_ON(dest->bu_blocks != src->bu_blocks);
310 BUG_ON(dest->bu_inode != src->bu_inode);
4980c6da 311
ba937127 312 for (i = 0; i < src->bu_blocks; i++) {
4980c6da
JB		 313 memcpy(bucket_block(dest, i), bucket_block(src, i),
314 blocksize);
315 }
316}
1224be02 317
4ae1d69b
JB		 318static int ocfs2_validate_xattr_block(struct super_block *sb,
319 struct buffer_head *bh)
320{
321 struct ocfs2_xattr_block *xb =
322 (struct ocfs2_xattr_block *)bh->b_data;
323
324 mlog(0, "Validating xattr block %llu\n",
325 (unsigned long long)bh->b_blocknr);
326
327 if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
328 ocfs2_error(sb,
329 "Extended attribute block #%llu has bad "
330 "signature %.*s",
331 (unsigned long long)bh->b_blocknr, 7,
332 xb->xb_signature);
333 return -EINVAL;
334 }
335
336 if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
337 ocfs2_error(sb,
338 "Extended attribute block #%llu has an "
339 "invalid xb_blkno of %llu",
340 (unsigned long long)bh->b_blocknr,
341 (unsigned long long)le64_to_cpu(xb->xb_blkno));
342 return -EINVAL;
343 }
344
345 if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
346 ocfs2_error(sb,
347 "Extended attribute block #%llu has an invalid "
348 "xb_fs_generation of #%u",
349 (unsigned long long)bh->b_blocknr,
350 le32_to_cpu(xb->xb_fs_generation));
351 return -EINVAL;
352 }
353
354 return 0;
355}
356
357static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
358 struct buffer_head **bh)
359{
360 int rc;
361 struct buffer_head *tmp = *bh;
362
970e4936
JB		 363 rc = ocfs2_read_block(inode, xb_blkno, &tmp,
364 ocfs2_validate_xattr_block);
4ae1d69b
JB		 365
366 /* If ocfs2_read_block() got us a new bh, pass it up. */
367 if (!rc && !*bh)
368 *bh = tmp;
369
370 return rc;
371}
372
936b8834 373static inline const char *ocfs2_xattr_prefix(int name_index)
cf1d6c76
TY		 374{
375 struct xattr_handler *handler = NULL;
376
377 if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
378 handler = ocfs2_xattr_handler_map[name_index];
379
936b8834 380 return handler ? handler->prefix : NULL;
cf1d6c76
TY		 381}
382
40daa16a 383static u32 ocfs2_xattr_name_hash(struct inode *inode,
2057e5c6 384 const char *name,
40daa16a 385 int name_len)
cf1d6c76
TY		 386{
387 /* Get hash value of uuid from super block */
388 u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
389 int i;
390
cf1d6c76
TY		 391 /* hash extended attribute name */
392 for (i = 0; i < name_len; i++) {
393 hash = (hash << OCFS2_HASH_SHIFT) ^
394 (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
395 *name++;
396 }
397
398 return hash;
399}
400
401/*
402 * ocfs2_xattr_hash_entry()
403 *
404 * Compute the hash of an extended attribute.
405 */
406static void ocfs2_xattr_hash_entry(struct inode *inode,
407 struct ocfs2_xattr_header *header,
408 struct ocfs2_xattr_entry *entry)
409{
410 u32 hash = 0;
cf1d6c76 411 char *name = (char *)header + le16_to_cpu(entry->xe_name_offset);
cf1d6c76 412
2057e5c6 413 hash = ocfs2_xattr_name_hash(inode, name, entry->xe_name_len);
cf1d6c76
TY		 414 entry->xe_name_hash = cpu_to_le32(hash);
415
416 return;
417}
f56654c4 418
534eaddd
TY		 419static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
420{
421 int size = 0;
422
423 if (value_len <= OCFS2_XATTR_INLINE_SIZE)
424 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
425 else
426 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
427 size += sizeof(struct ocfs2_xattr_entry);
428
429 return size;
430}
431
432int ocfs2_calc_security_init(struct inode *dir,
433 struct ocfs2_security_xattr_info *si,
434 int *want_clusters,
435 int *xattr_credits,
436 struct ocfs2_alloc_context **xattr_ac)
437{
438 int ret = 0;
439 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
440 int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
441 si->value_len);
442
443 /*
444 * The max space of security xattr taken inline is
445 * 256(name) + 80(value) + 16(entry) = 352 bytes,
446 * So reserve one metadata block for it is ok.
447 */
448 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
449 s_size > OCFS2_XATTR_FREE_IN_IBODY) {
450 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
451 if (ret) {
452 mlog_errno(ret);
453 return ret;
454 }
455 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
456 }
457
458 /* reserve clusters for xattr value which will be set in B tree*/
459 if (si->value_len > OCFS2_XATTR_INLINE_SIZE)
460 *want_clusters += ocfs2_clusters_for_bytes(dir->i_sb,
461 si->value_len);
462 return ret;
463}
464
89c38bd0
TY		 465int ocfs2_calc_xattr_init(struct inode *dir,
466 struct buffer_head *dir_bh,
467 int mode,
468 struct ocfs2_security_xattr_info *si,
469 int *want_clusters,
470 int *xattr_credits,
471 struct ocfs2_alloc_context **xattr_ac)
472{
473 int ret = 0;
474 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
475 int s_size = 0;
476 int a_size = 0;
477 int acl_len = 0;
478
479 if (si->enable)
480 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
481 si->value_len);
482
483 if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
484 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
485 OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
486 "", NULL, 0);
487 if (acl_len > 0) {
488 a_size = ocfs2_xattr_entry_real_size(0, acl_len);
489 if (S_ISDIR(mode))
490 a_size <<= 1;
491 } else if (acl_len != 0 && acl_len != -ENODATA) {
492 mlog_errno(ret);
493 return ret;
494 }
495 }
496
497 if (!(s_size + a_size))
498 return ret;
499
500 /*
501 * The max space of security xattr taken inline is
502 * 256(name) + 80(value) + 16(entry) = 352 bytes,
503 * The max space of acl xattr taken inline is
504 * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
505 * when blocksize = 512, may reserve one more cluser for
506 * xattr bucket, otherwise reserve one metadata block
507 * for them is ok.
508 */
509 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
510 (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
511 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
512 if (ret) {
513 mlog_errno(ret);
514 return ret;
515 }
516 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
517 }
518
519 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
520 (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
521 *want_clusters += 1;
522 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
523 }
524
525 /* reserve clusters for xattr value which will be set in B tree*/
526 if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE)
527 *want_clusters += ocfs2_clusters_for_bytes(dir->i_sb,
528 si->value_len);
529 if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
530 acl_len > OCFS2_XATTR_INLINE_SIZE) {
531 *want_clusters += ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
532 if (S_ISDIR(mode))
533 *want_clusters += ocfs2_clusters_for_bytes(dir->i_sb,
534 acl_len);
535 }
536
537 return ret;
538}
539
f56654c4
TM		 540static int ocfs2_xattr_extend_allocation(struct inode *inode,
541 u32 clusters_to_add,
542 struct buffer_head *xattr_bh,
78f30c31
TM		 543 struct ocfs2_xattr_value_root *xv,
544 struct ocfs2_xattr_set_ctxt *ctxt)
f56654c4
TM		 545{
546 int status = 0;
85db90e7 547 handle_t *handle = ctxt->handle;
f56654c4
TM		 548 enum ocfs2_alloc_restarted why;
549 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
f56654c4 550 u32 prev_clusters, logical_start = le32_to_cpu(xv->xr_clusters);
f99b9b7c 551 struct ocfs2_extent_tree et;
f56654c4
TM		 552
553 mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
554
8d6220d6 555 ocfs2_init_xattr_value_extent_tree(&et, inode, xattr_bh, xv);
f99b9b7c 556
f56654c4
TM		 557 status = ocfs2_journal_access(handle, inode, xattr_bh,
558 OCFS2_JOURNAL_ACCESS_WRITE);
559 if (status < 0) {
560 mlog_errno(status);
561 goto leave;
562 }
563
564 prev_clusters = le32_to_cpu(xv->xr_clusters);
565 status = ocfs2_add_clusters_in_btree(osb,
566 inode,
567 &logical_start,
568 clusters_to_add,
569 0,
f99b9b7c 570 &et,
f56654c4 571 handle,
78f30c31
TM		 572 ctxt->data_ac,
573 ctxt->meta_ac,
f99b9b7c 574 &why);
85db90e7
TM		 575 if (status < 0) {
576 mlog_errno(status);
f56654c4
TM		 577 goto leave;
578 }
579
580 status = ocfs2_journal_dirty(handle, xattr_bh);
581 if (status < 0) {
582 mlog_errno(status);
583 goto leave;
584 }
585
586 clusters_to_add -= le32_to_cpu(xv->xr_clusters) - prev_clusters;
587
85db90e7
TM		 588 /*
589 * We should have already allocated enough space before the transaction,
590 * so no need to restart.
591 */
592 BUG_ON(why != RESTART_NONE || clusters_to_add);
f56654c4
TM		 593
594leave:
f56654c4
TM		 595
596 return status;
597}
598
599static int __ocfs2_remove_xattr_range(struct inode *inode,
600 struct buffer_head *root_bh,
601 struct ocfs2_xattr_value_root *xv,
602 u32 cpos, u32 phys_cpos, u32 len,
78f30c31 603 struct ocfs2_xattr_set_ctxt *ctxt)
f56654c4
TM		 604{
605 int ret;
606 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
85db90e7 607 handle_t *handle = ctxt->handle;
f99b9b7c
JB		 608 struct ocfs2_extent_tree et;
609
8d6220d6 610 ocfs2_init_xattr_value_extent_tree(&et, inode, root_bh, xv);
f56654c4 611
f56654c4
TM		 612 ret = ocfs2_journal_access(handle, inode, root_bh,
613 OCFS2_JOURNAL_ACCESS_WRITE);
614 if (ret) {
615 mlog_errno(ret);
85db90e7 616 goto out;
f56654c4
TM		 617 }
618
78f30c31
TM		 619 ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, ctxt->meta_ac,
620 &ctxt->dealloc);
f56654c4
TM		 621 if (ret) {
622 mlog_errno(ret);
85db90e7 623 goto out;
f56654c4
TM		 624 }
625
626 le32_add_cpu(&xv->xr_clusters, -len);
627
628 ret = ocfs2_journal_dirty(handle, root_bh);
629 if (ret) {
630 mlog_errno(ret);
85db90e7 631 goto out;
f56654c4
TM		 632 }
633
78f30c31 634 ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc, phys_blkno, len);
f56654c4
TM		 635 if (ret)
636 mlog_errno(ret);
637
f56654c4 638out:
f56654c4
TM		 639 return ret;
640}
641
642static int ocfs2_xattr_shrink_size(struct inode *inode,
643 u32 old_clusters,
644 u32 new_clusters,
645 struct buffer_head *root_bh,
78f30c31
TM		 646 struct ocfs2_xattr_value_root *xv,
647 struct ocfs2_xattr_set_ctxt *ctxt)
f56654c4
TM		 648{
649 int ret = 0;
650 u32 trunc_len, cpos, phys_cpos, alloc_size;
651 u64 block;
f56654c4
TM		 652
653 if (old_clusters <= new_clusters)
654 return 0;
655
656 cpos = new_clusters;
657 trunc_len = old_clusters - new_clusters;
658 while (trunc_len) {
659 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
660 &alloc_size, &xv->xr_list);
661 if (ret) {
662 mlog_errno(ret);
663 goto out;
664 }
665
666 if (alloc_size > trunc_len)
667 alloc_size = trunc_len;
668
669 ret = __ocfs2_remove_xattr_range(inode, root_bh, xv, cpos,
670 phys_cpos, alloc_size,
78f30c31 671 ctxt);
f56654c4
TM		 672 if (ret) {
673 mlog_errno(ret);
674 goto out;
675 }
676
677 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
678 ocfs2_remove_xattr_clusters_from_cache(inode, block,
679 alloc_size);
680 cpos += alloc_size;
681 trunc_len -= alloc_size;
682 }
683
684out:
f56654c4
TM		 685 return ret;
686}
687
688static int ocfs2_xattr_value_truncate(struct inode *inode,
689 struct buffer_head *root_bh,
690 struct ocfs2_xattr_value_root *xv,
78f30c31
TM		 691 int len,
692 struct ocfs2_xattr_set_ctxt *ctxt)
f56654c4
TM		 693{
694 int ret;
695 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
696 u32 old_clusters = le32_to_cpu(xv->xr_clusters);
697
698 if (new_clusters == old_clusters)
699 return 0;
700
701 if (new_clusters > old_clusters)
702 ret = ocfs2_xattr_extend_allocation(inode,
703 new_clusters - old_clusters,
78f30c31 704 root_bh, xv, ctxt);
f56654c4
TM		 705 else
706 ret = ocfs2_xattr_shrink_size(inode,
707 old_clusters, new_clusters,
78f30c31 708 root_bh, xv, ctxt);
f56654c4
TM		 709
710 return ret;
711}
cf1d6c76 712
936b8834
TM		 713static int ocfs2_xattr_list_entry(char *buffer, size_t size,
714 size_t *result, const char *prefix,
715 const char *name, int name_len)
716{
717 char *p = buffer + *result;
718 int prefix_len = strlen(prefix);
719 int total_len = prefix_len + name_len + 1;
720
721 *result += total_len;
722
723 /* we are just looking for how big our buffer needs to be */
724 if (!size)
725 return 0;
726
727 if (*result > size)
728 return -ERANGE;
729
730 memcpy(p, prefix, prefix_len);
731 memcpy(p + prefix_len, name, name_len);
732 p[prefix_len + name_len] = '\0';
733
734 return 0;
735}
736
cf1d6c76
TY		 737static int ocfs2_xattr_list_entries(struct inode *inode,
738 struct ocfs2_xattr_header *header,
739 char *buffer, size_t buffer_size)
740{
936b8834
TM		 741 size_t result = 0;
742 int i, type, ret;
743 const char *prefix, *name;
cf1d6c76
TY		 744
745 for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
746 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
936b8834
TM		 747 type = ocfs2_xattr_get_type(entry);
748 prefix = ocfs2_xattr_prefix(type);
749
750 if (prefix) {
751 name = (const char *)header +
752 le16_to_cpu(entry->xe_name_offset);
753
754 ret = ocfs2_xattr_list_entry(buffer, buffer_size,
755 &result, prefix, name,
756 entry->xe_name_len);
757 if (ret)
758 return ret;
cf1d6c76
TY		 759 }
760 }
761
936b8834 762 return result;
cf1d6c76
TY		 763}
764
765static int ocfs2_xattr_ibody_list(struct inode *inode,
766 struct ocfs2_dinode *di,
767 char *buffer,
768 size_t buffer_size)
769{
770 struct ocfs2_xattr_header *header = NULL;
771 struct ocfs2_inode_info *oi = OCFS2_I(inode);
772 int ret = 0;
773
774 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
775 return ret;
776
777 header = (struct ocfs2_xattr_header *)
778 ((void *)di + inode->i_sb->s_blocksize -
779 le16_to_cpu(di->i_xattr_inline_size));
780
781 ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
782
783 return ret;
784}
785
786static int ocfs2_xattr_block_list(struct inode *inode,
787 struct ocfs2_dinode *di,
788 char *buffer,
789 size_t buffer_size)
790{
791 struct buffer_head *blk_bh = NULL;
0c044f0b 792 struct ocfs2_xattr_block *xb;
cf1d6c76
TY		 793 int ret = 0;
794
795 if (!di->i_xattr_loc)
796 return ret;
797
4ae1d69b
JB		 798 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
799 &blk_bh);
cf1d6c76
TY		 800 if (ret < 0) {
801 mlog_errno(ret);
802 return ret;
803 }
cf1d6c76 804
0c044f0b 805 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
0c044f0b
TM		 806 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
807 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
808 ret = ocfs2_xattr_list_entries(inode, header,
809 buffer, buffer_size);
810 } else {
811 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
812 ret = ocfs2_xattr_tree_list_index_block(inode, xt,
813 buffer, buffer_size);
814 }
4ae1d69b 815
cf1d6c76
TY		 816 brelse(blk_bh);
817
818 return ret;
819}
820
821ssize_t ocfs2_listxattr(struct dentry *dentry,
822 char *buffer,
823 size_t size)
824{
825 int ret = 0, i_ret = 0, b_ret = 0;
826 struct buffer_head *di_bh = NULL;
827 struct ocfs2_dinode *di = NULL;
828 struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
829
8154da3d
TY		 830 if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
831 return -EOPNOTSUPP;
832
cf1d6c76
TY		 833 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
834 return ret;
835
836 ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
837 if (ret < 0) {
838 mlog_errno(ret);
839 return ret;
840 }
841
842 di = (struct ocfs2_dinode *)di_bh->b_data;
843
844 down_read(&oi->ip_xattr_sem);
845 i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
846 if (i_ret < 0)
847 b_ret = 0;
848 else {
849 if (buffer) {
850 buffer += i_ret;
851 size -= i_ret;
852 }
853 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
854 buffer, size);
855 if (b_ret < 0)
856 i_ret = 0;
857 }
858 up_read(&oi->ip_xattr_sem);
859 ocfs2_inode_unlock(dentry->d_inode, 0);
860
861 brelse(di_bh);
862
863 return i_ret + b_ret;
864}
865
866static int ocfs2_xattr_find_entry(int name_index,
867 const char *name,
868 struct ocfs2_xattr_search *xs)
869{
870 struct ocfs2_xattr_entry *entry;
871 size_t name_len;
872 int i, cmp = 1;
873
874 if (name == NULL)
875 return -EINVAL;
876
877 name_len = strlen(name);
878 entry = xs->here;
879 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
880 cmp = name_index - ocfs2_xattr_get_type(entry);
881 if (!cmp)
882 cmp = name_len - entry->xe_name_len;
883 if (!cmp)
884 cmp = memcmp(name, (xs->base +
885 le16_to_cpu(entry->xe_name_offset)),
886 name_len);
887 if (cmp == 0)
888 break;
889 entry += 1;
890 }
891 xs->here = entry;
892
893 return cmp ? -ENODATA : 0;
894}
895
896static int ocfs2_xattr_get_value_outside(struct inode *inode,
589dc260 897 struct ocfs2_xattr_value_root *xv,
cf1d6c76
TY		 898 void *buffer,
899 size_t len)
900{
901 u32 cpos, p_cluster, num_clusters, bpc, clusters;
902 u64 blkno;
903 int i, ret = 0;
904 size_t cplen, blocksize;
905 struct buffer_head *bh = NULL;
cf1d6c76
TY		 906 struct ocfs2_extent_list *el;
907
cf1d6c76
TY		 908 el = &xv->xr_list;
909 clusters = le32_to_cpu(xv->xr_clusters);
910 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
911 blocksize = inode->i_sb->s_blocksize;
912
913 cpos = 0;
914 while (cpos < clusters) {
915 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
916 &num_clusters, el);
917 if (ret) {
918 mlog_errno(ret);
919 goto out;
920 }
921
922 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
923 /* Copy ocfs2_xattr_value */
924 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
970e4936 925 ret = ocfs2_read_block(inode, blkno, &bh, NULL);
cf1d6c76
TY		 926 if (ret) {
927 mlog_errno(ret);
928 goto out;
929 }
930
931 cplen = len >= blocksize ? blocksize : len;
932 memcpy(buffer, bh->b_data, cplen);
933 len -= cplen;
934 buffer += cplen;
935
936 brelse(bh);
937 bh = NULL;
938 if (len == 0)
939 break;
940 }
941 cpos += num_clusters;
942 }
943out:
944 return ret;
945}
946
947static int ocfs2_xattr_ibody_get(struct inode *inode,
948 int name_index,
949 const char *name,
950 void *buffer,
951 size_t buffer_size,
952 struct ocfs2_xattr_search *xs)
953{
954 struct ocfs2_inode_info *oi = OCFS2_I(inode);
955 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
589dc260 956 struct ocfs2_xattr_value_root *xv;
cf1d6c76
TY		 957 size_t size;
958 int ret = 0;
959
960 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
961 return -ENODATA;
962
963 xs->end = (void *)di + inode->i_sb->s_blocksize;
964 xs->header = (struct ocfs2_xattr_header *)
965 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
966 xs->base = (void *)xs->header;
967 xs->here = xs->header->xh_entries;
968
969 ret = ocfs2_xattr_find_entry(name_index, name, xs);
970 if (ret)
971 return ret;
972 size = le64_to_cpu(xs->here->xe_value_size);
973 if (buffer) {
974 if (size > buffer_size)
975 return -ERANGE;
976 if (ocfs2_xattr_is_local(xs->here)) {
977 memcpy(buffer, (void *)xs->base +
978 le16_to_cpu(xs->here->xe_name_offset) +
979 OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
980 } else {
589dc260
TM		 981 xv = (struct ocfs2_xattr_value_root *)
982 (xs->base + le16_to_cpu(
983 xs->here->xe_name_offset) +
984 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
985 ret = ocfs2_xattr_get_value_outside(inode, xv,
cf1d6c76
TY		 986 buffer, size);
987 if (ret < 0) {
988 mlog_errno(ret);
989 return ret;
990 }
991 }
992 }
993
994 return size;
995}
996
997static int ocfs2_xattr_block_get(struct inode *inode,
998 int name_index,
999 const char *name,
1000 void *buffer,
1001 size_t buffer_size,
1002 struct ocfs2_xattr_search *xs)
1003{
cf1d6c76 1004 struct ocfs2_xattr_block *xb;
589dc260 1005 struct ocfs2_xattr_value_root *xv;
cf1d6c76 1006 size_t size;
589dc260 1007 int ret = -ENODATA, name_offset, name_len, block_off, i;
cf1d6c76 1008
ba937127
JB		 1009 xs->bucket = ocfs2_xattr_bucket_new(inode);
1010 if (!xs->bucket) {
1011 ret = -ENOMEM;
1012 mlog_errno(ret);
1013 goto cleanup;
1014 }
589dc260 1015
54f443f4
JB		 1016 ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1017 if (ret) {
cf1d6c76 1018 mlog_errno(ret);
cf1d6c76
TY		 1019 goto cleanup;
1020 }
1021
6c1e183e
TY		 1022 if (xs->not_found) {
1023 ret = -ENODATA;
1024 goto cleanup;
1025 }
1026
54f443f4 1027 xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
cf1d6c76
TY		 1028 size = le64_to_cpu(xs->here->xe_value_size);
1029 if (buffer) {
1030 ret = -ERANGE;
1031 if (size > buffer_size)
1032 goto cleanup;
589dc260
TM		 1033
1034 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1035 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1036 i = xs->here - xs->header->xh_entries;
1037
1038 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1039 ret = ocfs2_xattr_bucket_get_name_value(inode,
ba937127 1040 bucket_xh(xs->bucket),
589dc260
TM		 1041 i,
1042 &block_off,
1043 &name_offset);
ba937127 1044 xs->base = bucket_block(xs->bucket, block_off);
589dc260 1045 }
cf1d6c76
TY		 1046 if (ocfs2_xattr_is_local(xs->here)) {
1047 memcpy(buffer, (void *)xs->base +
589dc260 1048 name_offset + name_len, size);
cf1d6c76 1049 } else {
589dc260
TM		 1050 xv = (struct ocfs2_xattr_value_root *)
1051 (xs->base + name_offset + name_len);
1052 ret = ocfs2_xattr_get_value_outside(inode, xv,
cf1d6c76
TY		 1053 buffer, size);
1054 if (ret < 0) {
1055 mlog_errno(ret);
1056 goto cleanup;
1057 }
1058 }
1059 }
1060 ret = size;
1061cleanup:
ba937127 1062 ocfs2_xattr_bucket_free(xs->bucket);
cf1d6c76 1063
54f443f4
JB		 1064 brelse(xs->xattr_bh);
1065 xs->xattr_bh = NULL;
cf1d6c76
TY		 1066 return ret;
1067}
1068
4e3e9d02
TY		 1069int ocfs2_xattr_get_nolock(struct inode *inode,
1070 struct buffer_head *di_bh,
0030e001
TY		 1071 int name_index,
1072 const char *name,
1073 void *buffer,
1074 size_t buffer_size)
cf1d6c76
TY		 1075{
1076 int ret;
1077 struct ocfs2_dinode *di = NULL;
cf1d6c76
TY		 1078 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1079 struct ocfs2_xattr_search xis = {
1080 .not_found = -ENODATA,
1081 };
1082 struct ocfs2_xattr_search xbs = {
1083 .not_found = -ENODATA,
1084 };
1085
8154da3d
TY		 1086 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1087 return -EOPNOTSUPP;
1088
cf1d6c76
TY		 1089 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1090 ret = -ENODATA;
1091
cf1d6c76
TY		 1092 xis.inode_bh = xbs.inode_bh = di_bh;
1093 di = (struct ocfs2_dinode *)di_bh->b_data;
1094
1095 down_read(&oi->ip_xattr_sem);
1096 ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1097 buffer_size, &xis);
6c1e183e 1098 if (ret == -ENODATA && di->i_xattr_loc)
cf1d6c76
TY		 1099 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1100 buffer_size, &xbs);
1101 up_read(&oi->ip_xattr_sem);
4e3e9d02
TY		 1102
1103 return ret;
1104}
1105
1106/* ocfs2_xattr_get()
1107 *
1108 * Copy an extended attribute into the buffer provided.
1109 * Buffer is NULL to compute the size of buffer required.
1110 */
1111static int ocfs2_xattr_get(struct inode *inode,
1112 int name_index,
1113 const char *name,
1114 void *buffer,
1115 size_t buffer_size)
1116{
1117 int ret;
1118 struct buffer_head *di_bh = NULL;
1119
1120 ret = ocfs2_inode_lock(inode, &di_bh, 0);
1121 if (ret < 0) {
1122 mlog_errno(ret);
1123 return ret;
1124 }
1125 ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1126 name, buffer, buffer_size);
1127
cf1d6c76
TY		 1128 ocfs2_inode_unlock(inode, 0);
1129
1130 brelse(di_bh);
1131
1132 return ret;
1133}
1134
1135static int __ocfs2_xattr_set_value_outside(struct inode *inode,
85db90e7 1136 handle_t *handle,
cf1d6c76
TY		 1137 struct ocfs2_xattr_value_root *xv,
1138 const void *value,
1139 int value_len)
1140{
1141 int ret = 0, i, cp_len, credits;
1142 u16 blocksize = inode->i_sb->s_blocksize;
1143 u32 p_cluster, num_clusters;
1144 u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1145 u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1146 u64 blkno;
1147 struct buffer_head *bh = NULL;
cf1d6c76
TY		 1148
1149 BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1150
85db90e7
TM		 1151 /*
1152 * In __ocfs2_xattr_set_value_outside has already been dirtied,
1153 * so we don't need to worry about whether ocfs2_extend_trans
1154 * will create a new transactio for us or not.
1155 */
cf1d6c76 1156 credits = clusters * bpc;
85db90e7
TM		 1157 ret = ocfs2_extend_trans(handle, credits);
1158 if (ret) {
cf1d6c76
TY		 1159 mlog_errno(ret);
1160 goto out;
1161 }
1162
1163 while (cpos < clusters) {
1164 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1165 &num_clusters, &xv->xr_list);
1166 if (ret) {
1167 mlog_errno(ret);
85db90e7 1168 goto out;
cf1d6c76
TY		 1169 }
1170
1171 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1172
1173 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
970e4936 1174 ret = ocfs2_read_block(inode, blkno, &bh, NULL);
cf1d6c76
TY		 1175 if (ret) {
1176 mlog_errno(ret);
85db90e7 1177 goto out;
cf1d6c76
TY		 1178 }
1179
1180 ret = ocfs2_journal_access(handle,
1181 inode,
1182 bh,
1183 OCFS2_JOURNAL_ACCESS_WRITE);
1184 if (ret < 0) {
1185 mlog_errno(ret);
85db90e7 1186 goto out;
cf1d6c76
TY		 1187 }
1188
1189 cp_len = value_len > blocksize ? blocksize : value_len;
1190 memcpy(bh->b_data, value, cp_len);
1191 value_len -= cp_len;
1192 value += cp_len;
1193 if (cp_len < blocksize)
1194 memset(bh->b_data + cp_len, 0,
1195 blocksize - cp_len);
1196
1197 ret = ocfs2_journal_dirty(handle, bh);
1198 if (ret < 0) {
1199 mlog_errno(ret);
85db90e7 1200 goto out;
cf1d6c76
TY		 1201 }
1202 brelse(bh);
1203 bh = NULL;
1204
1205 /*
1206 * XXX: do we need to empty all the following
1207 * blocks in this cluster?
1208 */
1209 if (!value_len)
1210 break;
1211 }
1212 cpos += num_clusters;
1213 }
cf1d6c76
TY		 1214out:
1215 brelse(bh);
1216
1217 return ret;
1218}
1219
1220static int ocfs2_xattr_cleanup(struct inode *inode,
85db90e7 1221 handle_t *handle,
cf1d6c76
TY		 1222 struct ocfs2_xattr_info *xi,
1223 struct ocfs2_xattr_search *xs,
1224 size_t offs)
1225{
cf1d6c76
TY		 1226 int ret = 0;
1227 size_t name_len = strlen(xi->name);
1228 void *val = xs->base + offs;
1229 size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1230
cf1d6c76
TY		 1231 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1232 OCFS2_JOURNAL_ACCESS_WRITE);
1233 if (ret) {
1234 mlog_errno(ret);
85db90e7 1235 goto out;
cf1d6c76
TY		 1236 }
1237 /* Decrease xattr count */
1238 le16_add_cpu(&xs->header->xh_count, -1);
1239 /* Remove the xattr entry and tree root which has already be set*/
1240 memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
1241 memset(val, 0, size);
1242
1243 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1244 if (ret < 0)
1245 mlog_errno(ret);
cf1d6c76
TY		 1246out:
1247 return ret;
1248}
1249
1250static int ocfs2_xattr_update_entry(struct inode *inode,
85db90e7 1251 handle_t *handle,
cf1d6c76
TY		 1252 struct ocfs2_xattr_info *xi,
1253 struct ocfs2_xattr_search *xs,
1254 size_t offs)
1255{
85db90e7 1256 int ret;
cf1d6c76 1257
cf1d6c76
TY		 1258 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1259 OCFS2_JOURNAL_ACCESS_WRITE);
1260 if (ret) {
1261 mlog_errno(ret);
85db90e7 1262 goto out;
cf1d6c76
TY		 1263 }
1264
1265 xs->here->xe_name_offset = cpu_to_le16(offs);
1266 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1267 if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1268 ocfs2_xattr_set_local(xs->here, 1);
1269 else
1270 ocfs2_xattr_set_local(xs->here, 0);
1271 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1272
1273 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1274 if (ret < 0)
1275 mlog_errno(ret);
cf1d6c76
TY		 1276out:
1277 return ret;
1278}
1279
1280/*
1281 * ocfs2_xattr_set_value_outside()
1282 *
1283 * Set large size value in B tree.
1284 */
1285static int ocfs2_xattr_set_value_outside(struct inode *inode,
1286 struct ocfs2_xattr_info *xi,
1287 struct ocfs2_xattr_search *xs,
78f30c31 1288 struct ocfs2_xattr_set_ctxt *ctxt,
cf1d6c76
TY		 1289 size_t offs)
1290{
1291 size_t name_len = strlen(xi->name);
1292 void *val = xs->base + offs;
1293 struct ocfs2_xattr_value_root *xv = NULL;
1294 size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1295 int ret = 0;
1296
1297 memset(val, 0, size);
1298 memcpy(val, xi->name, name_len);
1299 xv = (struct ocfs2_xattr_value_root *)
1300 (val + OCFS2_XATTR_SIZE(name_len));
1301 xv->xr_clusters = 0;
1302 xv->xr_last_eb_blk = 0;
1303 xv->xr_list.l_tree_depth = 0;
1304 xv->xr_list.l_count = cpu_to_le16(1);
1305 xv->xr_list.l_next_free_rec = 0;
1306
1307 ret = ocfs2_xattr_value_truncate(inode, xs->xattr_bh, xv,
78f30c31 1308 xi->value_len, ctxt);
cf1d6c76
TY		 1309 if (ret < 0) {
1310 mlog_errno(ret);
1311 return ret;
1312 }
85db90e7 1313 ret = ocfs2_xattr_update_entry(inode, ctxt->handle, xi, xs, offs);
cf1d6c76
TY		 1314 if (ret < 0) {
1315 mlog_errno(ret);
1316 return ret;
1317 }
85db90e7
TM		 1318 ret = __ocfs2_xattr_set_value_outside(inode, ctxt->handle, xv,
1319 xi->value, xi->value_len);
cf1d6c76
TY		 1320 if (ret < 0)
1321 mlog_errno(ret);
1322
1323 return ret;
1324}
1325
1326/*
1327 * ocfs2_xattr_set_entry_local()
1328 *
1329 * Set, replace or remove extended attribute in local.
1330 */
1331static void ocfs2_xattr_set_entry_local(struct inode *inode,
1332 struct ocfs2_xattr_info *xi,
1333 struct ocfs2_xattr_search *xs,
1334 struct ocfs2_xattr_entry *last,
1335 size_t min_offs)
1336{
1337 size_t name_len = strlen(xi->name);
1338 int i;
1339
1340 if (xi->value && xs->not_found) {
1341 /* Insert the new xattr entry. */
1342 le16_add_cpu(&xs->header->xh_count, 1);
1343 ocfs2_xattr_set_type(last, xi->name_index);
1344 ocfs2_xattr_set_local(last, 1);
1345 last->xe_name_len = name_len;
1346 } else {
1347 void *first_val;
1348 void *val;
1349 size_t offs, size;
1350
1351 first_val = xs->base + min_offs;
1352 offs = le16_to_cpu(xs->here->xe_name_offset);
1353 val = xs->base + offs;
1354
1355 if (le64_to_cpu(xs->here->xe_value_size) >
1356 OCFS2_XATTR_INLINE_SIZE)
1357 size = OCFS2_XATTR_SIZE(name_len) +
1358 OCFS2_XATTR_ROOT_SIZE;
1359 else
1360 size = OCFS2_XATTR_SIZE(name_len) +
1361 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1362
1363 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1364 OCFS2_XATTR_SIZE(xi->value_len)) {
1365 /* The old and the new value have the
1366 same size. Just replace the value. */
1367 ocfs2_xattr_set_local(xs->here, 1);
1368 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1369 /* Clear value bytes. */
1370 memset(val + OCFS2_XATTR_SIZE(name_len),
1371 0,
1372 OCFS2_XATTR_SIZE(xi->value_len));
1373 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1374 xi->value,
1375 xi->value_len);
1376 return;
1377 }
1378 /* Remove the old name+value. */
1379 memmove(first_val + size, first_val, val - first_val);
1380 memset(first_val, 0, size);
1381 xs->here->xe_name_hash = 0;
1382 xs->here->xe_name_offset = 0;
1383 ocfs2_xattr_set_local(xs->here, 1);
1384 xs->here->xe_value_size = 0;
1385
1386 min_offs += size;
1387
1388 /* Adjust all value offsets. */
1389 last = xs->header->xh_entries;
1390 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1391 size_t o = le16_to_cpu(last->xe_name_offset);
1392
1393 if (o < offs)
1394 last->xe_name_offset = cpu_to_le16(o + size);
1395 last += 1;
1396 }
1397
1398 if (!xi->value) {
1399 /* Remove the old entry. */
1400 last -= 1;
1401 memmove(xs->here, xs->here + 1,
1402 (void *)last - (void *)xs->here);
1403 memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1404 le16_add_cpu(&xs->header->xh_count, -1);
1405 }
1406 }
1407 if (xi->value) {
1408 /* Insert the new name+value. */
1409 size_t size = OCFS2_XATTR_SIZE(name_len) +
1410 OCFS2_XATTR_SIZE(xi->value_len);
1411 void *val = xs->base + min_offs - size;
1412
1413 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1414 memset(val, 0, size);
1415 memcpy(val, xi->name, name_len);
1416 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1417 xi->value,
1418 xi->value_len);
1419 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1420 ocfs2_xattr_set_local(xs->here, 1);
1421 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1422 }
1423
1424 return;
1425}
1426
1427/*
1428 * ocfs2_xattr_set_entry()
1429 *
1430 * Set extended attribute entry into inode or block.
1431 *
1432 * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1433 * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1434 * then set value in B tree with set_value_outside().
1435 */
1436static int ocfs2_xattr_set_entry(struct inode *inode,
1437 struct ocfs2_xattr_info *xi,
1438 struct ocfs2_xattr_search *xs,
78f30c31 1439 struct ocfs2_xattr_set_ctxt *ctxt,
cf1d6c76
TY		 1440 int flag)
1441{
1442 struct ocfs2_xattr_entry *last;
1443 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1444 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1445 size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1446 size_t size_l = 0;
85db90e7 1447 handle_t *handle = ctxt->handle;
cf1d6c76
TY		 1448 int free, i, ret;
1449 struct ocfs2_xattr_info xi_l = {
1450 .name_index = xi->name_index,
1451 .name = xi->name,
1452 .value = xi->value,
1453 .value_len = xi->value_len,
1454 };
1455
1456 /* Compute min_offs, last and free space. */
1457 last = xs->header->xh_entries;
1458
1459 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1460 size_t offs = le16_to_cpu(last->xe_name_offset);
1461 if (offs < min_offs)
1462 min_offs = offs;
1463 last += 1;
1464 }
1465
1466 free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
1467 if (free < 0)
b37c4d84 1468 return -EIO;
cf1d6c76
TY		 1469
1470 if (!xs->not_found) {
1471 size_t size = 0;
1472 if (ocfs2_xattr_is_local(xs->here))
1473 size = OCFS2_XATTR_SIZE(name_len) +
1474 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1475 else
1476 size = OCFS2_XATTR_SIZE(name_len) +
1477 OCFS2_XATTR_ROOT_SIZE;
1478 free += (size + sizeof(struct ocfs2_xattr_entry));
1479 }
1480 /* Check free space in inode or block */
1481 if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1482 if (free < sizeof(struct ocfs2_xattr_entry) +
1483 OCFS2_XATTR_SIZE(name_len) +
1484 OCFS2_XATTR_ROOT_SIZE) {
1485 ret = -ENOSPC;
1486 goto out;
1487 }
1488 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1489 xi_l.value = (void *)&def_xv;
1490 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1491 } else if (xi->value) {
1492 if (free < sizeof(struct ocfs2_xattr_entry) +
1493 OCFS2_XATTR_SIZE(name_len) +
1494 OCFS2_XATTR_SIZE(xi->value_len)) {
1495 ret = -ENOSPC;
1496 goto out;
1497 }
1498 }
1499
1500 if (!xs->not_found) {
1501 /* For existing extended attribute */
1502 size_t size = OCFS2_XATTR_SIZE(name_len) +
1503 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1504 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1505 void *val = xs->base + offs;
1506
1507 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1508 /* Replace existing local xattr with tree root */
1509 ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
78f30c31 1510 ctxt, offs);
cf1d6c76
TY		 1511 if (ret < 0)
1512 mlog_errno(ret);
1513 goto out;
1514 } else if (!ocfs2_xattr_is_local(xs->here)) {
1515 /* For existing xattr which has value outside */
1516 struct ocfs2_xattr_value_root *xv = NULL;
1517 xv = (struct ocfs2_xattr_value_root *)(val +
1518 OCFS2_XATTR_SIZE(name_len));
1519
1520 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1521 /*
1522 * If new value need set outside also,
1523 * first truncate old value to new value,
1524 * then set new value with set_value_outside().
1525 */
1526 ret = ocfs2_xattr_value_truncate(inode,
1527 xs->xattr_bh,
1528 xv,
78f30c31
TM		 1529 xi->value_len,
1530 ctxt);
cf1d6c76
TY		 1531 if (ret < 0) {
1532 mlog_errno(ret);
1533 goto out;
1534 }
1535
85db90e7
TM		 1536 ret = ocfs2_xattr_update_entry(inode,
1537 handle,
1538 xi,
1539 xs,
1540 offs);
cf1d6c76
TY		 1541 if (ret < 0) {
1542 mlog_errno(ret);
1543 goto out;
1544 }
1545
85db90e7
TM		 1546 ret = __ocfs2_xattr_set_value_outside(inode,
1547 handle,
1548 xv,
1549 xi->value,
1550 xi->value_len);
cf1d6c76
TY		 1551 if (ret < 0)
1552 mlog_errno(ret);
1553 goto out;
1554 } else {
1555 /*
1556 * If new value need set in local,
1557 * just trucate old value to zero.
1558 */
1559 ret = ocfs2_xattr_value_truncate(inode,
85db90e7
TM		 1560 xs->xattr_bh,
1561 xv,
1562 0,
1563 ctxt);
cf1d6c76
TY		 1564 if (ret < 0)
1565 mlog_errno(ret);
1566 }
1567 }
1568 }
1569
cf1d6c76
TY		 1570 ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
1571 OCFS2_JOURNAL_ACCESS_WRITE);
1572 if (ret) {
1573 mlog_errno(ret);
85db90e7 1574 goto out;
cf1d6c76
TY		 1575 }
1576
1577 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
cf1d6c76
TY		 1578 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1579 OCFS2_JOURNAL_ACCESS_WRITE);
1580 if (ret) {
1581 mlog_errno(ret);
85db90e7 1582 goto out;
cf1d6c76
TY		 1583 }
1584 }
1585
1586 /*
1587 * Set value in local, include set tree root in local.
1588 * This is the first step for value size >INLINE_SIZE.
1589 */
1590 ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1591
1592 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1593 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1594 if (ret < 0) {
1595 mlog_errno(ret);
85db90e7 1596 goto out;
cf1d6c76
TY		 1597 }
1598 }
1599
1600 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1601 (flag & OCFS2_INLINE_XATTR_FL)) {
1602 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1603 unsigned int xattrsize = osb->s_xattr_inline_size;
1604
1605 /*
1606 * Adjust extent record count or inline data size
1607 * to reserve space for extended attribute.
1608 */
1609 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1610 struct ocfs2_inline_data *idata = &di->id2.i_data;
1611 le16_add_cpu(&idata->id_count, -xattrsize);
1612 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1613 struct ocfs2_extent_list *el = &di->id2.i_list;
1614 le16_add_cpu(&el->l_count, -(xattrsize /
1615 sizeof(struct ocfs2_extent_rec)));
1616 }
1617 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1618 }
1619 /* Update xattr flag */
1620 spin_lock(&oi->ip_lock);
1621 oi->ip_dyn_features |= flag;
1622 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1623 spin_unlock(&oi->ip_lock);
1624 /* Update inode ctime */
1625 inode->i_ctime = CURRENT_TIME;
1626 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1627 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1628
1629 ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1630 if (ret < 0)
1631 mlog_errno(ret);
1632
cf1d6c76
TY		 1633 if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1634 /*
1635 * Set value outside in B tree.
1636 * This is the second step for value size > INLINE_SIZE.
1637 */
1638 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
78f30c31 1639 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, ctxt, offs);
cf1d6c76
TY		 1640 if (ret < 0) {
1641 int ret2;
1642
1643 mlog_errno(ret);
1644 /*
1645 * If set value outside failed, we have to clean
1646 * the junk tree root we have already set in local.
1647 */
85db90e7
TM		 1648 ret2 = ocfs2_xattr_cleanup(inode, ctxt->handle,
1649 xi, xs, offs);
cf1d6c76
TY		 1650 if (ret2 < 0)
1651 mlog_errno(ret2);
1652 }
1653 }
1654out:
1655 return ret;
cf1d6c76
TY		 1656}
1657
cf1d6c76
TY		 1658static int ocfs2_remove_value_outside(struct inode*inode,
1659 struct buffer_head *bh,
1660 struct ocfs2_xattr_header *header)
1661{
1662 int ret = 0, i;
78f30c31
TM		 1663 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1664 struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
1665
1666 ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
cf1d6c76 1667
a90714c1
JK		 1668 ctxt.handle = ocfs2_start_trans(osb,
1669 ocfs2_remove_extent_credits(osb->sb));
85db90e7
TM		 1670 if (IS_ERR(ctxt.handle)) {
1671 ret = PTR_ERR(ctxt.handle);
1672 mlog_errno(ret);
1673 goto out;
1674 }
1675
cf1d6c76
TY		 1676 for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1677 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1678
1679 if (!ocfs2_xattr_is_local(entry)) {
1680 struct ocfs2_xattr_value_root *xv;
1681 void *val;
1682
1683 val = (void *)header +
1684 le16_to_cpu(entry->xe_name_offset);
1685 xv = (struct ocfs2_xattr_value_root *)
1686 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
78f30c31
TM		 1687 ret = ocfs2_xattr_value_truncate(inode, bh, xv,
1688 0, &ctxt);
cf1d6c76
TY		 1689 if (ret < 0) {
1690 mlog_errno(ret);
78f30c31 1691 break;
cf1d6c76
TY		 1692 }
1693 }
1694 }
1695
85db90e7 1696 ocfs2_commit_trans(osb, ctxt.handle);
78f30c31
TM		 1697 ocfs2_schedule_truncate_log_flush(osb, 1);
1698 ocfs2_run_deallocs(osb, &ctxt.dealloc);
85db90e7 1699out:
cf1d6c76
TY		 1700 return ret;
1701}
1702
1703static int ocfs2_xattr_ibody_remove(struct inode *inode,
1704 struct buffer_head *di_bh)
1705{
1706
1707 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1708 struct ocfs2_xattr_header *header;
1709 int ret;
1710
1711 header = (struct ocfs2_xattr_header *)
1712 ((void *)di + inode->i_sb->s_blocksize -
1713 le16_to_cpu(di->i_xattr_inline_size));
1714
1715 ret = ocfs2_remove_value_outside(inode, di_bh, header);
1716
1717 return ret;
1718}
1719
1720static int ocfs2_xattr_block_remove(struct inode *inode,
1721 struct buffer_head *blk_bh)
1722{
1723 struct ocfs2_xattr_block *xb;
cf1d6c76
TY		 1724 int ret = 0;
1725
1726 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
a3944256
TM		 1727 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1728 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1729 ret = ocfs2_remove_value_outside(inode, blk_bh, header);
1730 } else
1731 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
cf1d6c76
TY		 1732
1733 return ret;
1734}
1735
08413899
TM		 1736static int ocfs2_xattr_free_block(struct inode *inode,
1737 u64 block)
1738{
1739 struct inode *xb_alloc_inode;
1740 struct buffer_head *xb_alloc_bh = NULL;
1741 struct buffer_head *blk_bh = NULL;
1742 struct ocfs2_xattr_block *xb;
1743 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1744 handle_t *handle;
1745 int ret = 0;
1746 u64 blk, bg_blkno;
1747 u16 bit;
1748
4ae1d69b 1749 ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
08413899
TM		 1750 if (ret < 0) {
1751 mlog_errno(ret);
1752 goto out;
1753 }
1754
08413899
TM		 1755 ret = ocfs2_xattr_block_remove(inode, blk_bh);
1756 if (ret < 0) {
1757 mlog_errno(ret);
1758 goto out;
1759 }
1760
4ae1d69b 1761 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
08413899
TM		 1762 blk = le64_to_cpu(xb->xb_blkno);
1763 bit = le16_to_cpu(xb->xb_suballoc_bit);
1764 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1765
1766 xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1767 EXTENT_ALLOC_SYSTEM_INODE,
1768 le16_to_cpu(xb->xb_suballoc_slot));
1769 if (!xb_alloc_inode) {
1770 ret = -ENOMEM;
1771 mlog_errno(ret);
1772 goto out;
1773 }
1774 mutex_lock(&xb_alloc_inode->i_mutex);
1775
1776 ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1777 if (ret < 0) {
1778 mlog_errno(ret);
1779 goto out_mutex;
1780 }
1781
1782 handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
1783 if (IS_ERR(handle)) {
1784 ret = PTR_ERR(handle);
1785 mlog_errno(ret);
1786 goto out_unlock;
1787 }
1788
1789 ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1790 bit, bg_blkno, 1);
1791 if (ret < 0)
1792 mlog_errno(ret);
1793
1794 ocfs2_commit_trans(osb, handle);
1795out_unlock:
1796 ocfs2_inode_unlock(xb_alloc_inode, 1);
1797 brelse(xb_alloc_bh);
1798out_mutex:
1799 mutex_unlock(&xb_alloc_inode->i_mutex);
1800 iput(xb_alloc_inode);
1801out:
1802 brelse(blk_bh);
1803 return ret;
1804}
1805
cf1d6c76
TY		 1806/*
1807 * ocfs2_xattr_remove()
1808 *
1809 * Free extended attribute resources associated with this inode.
1810 */
1811int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1812{
cf1d6c76
TY		 1813 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1814 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1815 handle_t *handle;
1816 int ret;
1817
8154da3d
TY		 1818 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1819 return 0;
1820
cf1d6c76
TY		 1821 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1822 return 0;
1823
1824 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1825 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1826 if (ret < 0) {
1827 mlog_errno(ret);
1828 goto out;
1829 }
1830 }
cf1d6c76 1831
08413899
TM		 1832 if (di->i_xattr_loc) {
1833 ret = ocfs2_xattr_free_block(inode,
1834 le64_to_cpu(di->i_xattr_loc));
cf1d6c76
TY		 1835 if (ret < 0) {
1836 mlog_errno(ret);
1837 goto out;
1838 }
1839 }
1840
1841 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1842 OCFS2_INODE_UPDATE_CREDITS);
1843 if (IS_ERR(handle)) {
1844 ret = PTR_ERR(handle);
1845 mlog_errno(ret);
1846 goto out;
1847 }
1848 ret = ocfs2_journal_access(handle, inode, di_bh,
1849 OCFS2_JOURNAL_ACCESS_WRITE);
1850 if (ret) {
1851 mlog_errno(ret);
1852 goto out_commit;
1853 }
1854
08413899 1855 di->i_xattr_loc = 0;
cf1d6c76
TY		 1856
1857 spin_lock(&oi->ip_lock);
1858 oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1859 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1860 spin_unlock(&oi->ip_lock);
1861
1862 ret = ocfs2_journal_dirty(handle, di_bh);
1863 if (ret < 0)
1864 mlog_errno(ret);
1865out_commit:
1866 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1867out:
cf1d6c76
TY		 1868 return ret;
1869}
1870
1871static int ocfs2_xattr_has_space_inline(struct inode *inode,
1872 struct ocfs2_dinode *di)
1873{
1874 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1875 unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1876 int free;
1877
1878 if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1879 return 0;
1880
1881 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1882 struct ocfs2_inline_data *idata = &di->id2.i_data;
1883 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1884 } else if (ocfs2_inode_is_fast_symlink(inode)) {
1885 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1886 le64_to_cpu(di->i_size);
1887 } else {
1888 struct ocfs2_extent_list *el = &di->id2.i_list;
1889 free = (le16_to_cpu(el->l_count) -
1890 le16_to_cpu(el->l_next_free_rec)) *
1891 sizeof(struct ocfs2_extent_rec);
1892 }
1893 if (free >= xattrsize)
1894 return 1;
1895
1896 return 0;
1897}
1898
1899/*
1900 * ocfs2_xattr_ibody_find()
1901 *
1902 * Find extended attribute in inode block and
1903 * fill search info into struct ocfs2_xattr_search.
1904 */
1905static int ocfs2_xattr_ibody_find(struct inode *inode,
1906 int name_index,
1907 const char *name,
1908 struct ocfs2_xattr_search *xs)
1909{
1910 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1911 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1912 int ret;
1913 int has_space = 0;
1914
1915 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1916 return 0;
1917
1918 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1919 down_read(&oi->ip_alloc_sem);
1920 has_space = ocfs2_xattr_has_space_inline(inode, di);
1921 up_read(&oi->ip_alloc_sem);
1922 if (!has_space)
1923 return 0;
1924 }
1925
1926 xs->xattr_bh = xs->inode_bh;
1927 xs->end = (void *)di + inode->i_sb->s_blocksize;
1928 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1929 xs->header = (struct ocfs2_xattr_header *)
1930 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1931 else
1932 xs->header = (struct ocfs2_xattr_header *)
1933 (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
1934 xs->base = (void *)xs->header;
1935 xs->here = xs->header->xh_entries;
1936
1937 /* Find the named attribute. */
1938 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1939 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1940 if (ret && ret != -ENODATA)
1941 return ret;
1942 xs->not_found = ret;
1943 }
1944
1945 return 0;
1946}
1947
1948/*
1949 * ocfs2_xattr_ibody_set()
1950 *
1951 * Set, replace or remove an extended attribute into inode block.
1952 *
1953 */
1954static int ocfs2_xattr_ibody_set(struct inode *inode,
1955 struct ocfs2_xattr_info *xi,
78f30c31
TM		 1956 struct ocfs2_xattr_search *xs,
1957 struct ocfs2_xattr_set_ctxt *ctxt)
cf1d6c76
TY		 1958{
1959 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1960 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1961 int ret;
1962
1963 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1964 return -ENOSPC;
1965
1966 down_write(&oi->ip_alloc_sem);
1967 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1968 if (!ocfs2_xattr_has_space_inline(inode, di)) {
1969 ret = -ENOSPC;
1970 goto out;
1971 }
1972 }
1973
78f30c31 1974 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
cf1d6c76
TY		 1975 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
1976out:
1977 up_write(&oi->ip_alloc_sem);
1978
1979 return ret;
1980}
1981
1982/*
1983 * ocfs2_xattr_block_find()
1984 *
1985 * Find extended attribute in external block and
1986 * fill search info into struct ocfs2_xattr_search.
1987 */
1988static int ocfs2_xattr_block_find(struct inode *inode,
1989 int name_index,
1990 const char *name,
1991 struct ocfs2_xattr_search *xs)
1992{
1993 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1994 struct buffer_head *blk_bh = NULL;
589dc260 1995 struct ocfs2_xattr_block *xb;
cf1d6c76
TY		 1996 int ret = 0;
1997
1998 if (!di->i_xattr_loc)
1999 return ret;
2000
4ae1d69b
JB		 2001 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2002 &blk_bh);
cf1d6c76
TY		 2003 if (ret < 0) {
2004 mlog_errno(ret);
2005 return ret;
2006 }
f6087fb7 2007
cf1d6c76 2008 xs->xattr_bh = blk_bh;
4ae1d69b 2009 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
589dc260
TM		 2010
2011 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2012 xs->header = &xb->xb_attrs.xb_header;
2013 xs->base = (void *)xs->header;
2014 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2015 xs->here = xs->header->xh_entries;
2016
2017 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2018 } else
2019 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2020 name_index,
2021 name, xs);
cf1d6c76 2022
cf1d6c76
TY		 2023 if (ret && ret != -ENODATA) {
2024 xs->xattr_bh = NULL;
2025 goto cleanup;
2026 }
2027 xs->not_found = ret;
2028 return 0;
cf1d6c76
TY		 2029cleanup:
2030 brelse(blk_bh);
2031
2032 return ret;
2033}
2034
2035/*
2036 * ocfs2_xattr_block_set()
2037 *
2038 * Set, replace or remove an extended attribute into external block.
2039 *
2040 */
2041static int ocfs2_xattr_block_set(struct inode *inode,
2042 struct ocfs2_xattr_info *xi,
78f30c31
TM		 2043 struct ocfs2_xattr_search *xs,
2044 struct ocfs2_xattr_set_ctxt *ctxt)
cf1d6c76
TY		 2045{
2046 struct buffer_head *new_bh = NULL;
2047 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2048 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
85db90e7 2049 handle_t *handle = ctxt->handle;
cf1d6c76
TY		 2050 struct ocfs2_xattr_block *xblk = NULL;
2051 u16 suballoc_bit_start;
2052 u32 num_got;
2053 u64 first_blkno;
2054 int ret;
2055
2056 if (!xs->xattr_bh) {
cf1d6c76
TY		 2057 ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
2058 OCFS2_JOURNAL_ACCESS_CREATE);
2059 if (ret < 0) {
2060 mlog_errno(ret);
85db90e7 2061 goto end;
cf1d6c76
TY		 2062 }
2063
78f30c31 2064 ret = ocfs2_claim_metadata(osb, handle, ctxt->meta_ac, 1,
cf1d6c76
TY		 2065 &suballoc_bit_start, &num_got,
2066 &first_blkno);
2067 if (ret < 0) {
2068 mlog_errno(ret);
85db90e7 2069 goto end;
cf1d6c76
TY		 2070 }
2071
2072 new_bh = sb_getblk(inode->i_sb, first_blkno);
2073 ocfs2_set_new_buffer_uptodate(inode, new_bh);
2074
2075 ret = ocfs2_journal_access(handle, inode, new_bh,
2076 OCFS2_JOURNAL_ACCESS_CREATE);
2077 if (ret < 0) {
2078 mlog_errno(ret);
85db90e7 2079 goto end;
cf1d6c76
TY		 2080 }
2081
2082 /* Initialize ocfs2_xattr_block */
2083 xs->xattr_bh = new_bh;
2084 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2085 memset(xblk, 0, inode->i_sb->s_blocksize);
2086 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2087 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
2088 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2089 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
2090 xblk->xb_blkno = cpu_to_le64(first_blkno);
2091
2092 xs->header = &xblk->xb_attrs.xb_header;
2093 xs->base = (void *)xs->header;
2094 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2095 xs->here = xs->header->xh_entries;
2096
cf1d6c76
TY		 2097 ret = ocfs2_journal_dirty(handle, new_bh);
2098 if (ret < 0) {
2099 mlog_errno(ret);
85db90e7 2100 goto end;
cf1d6c76
TY		 2101 }
2102 di->i_xattr_loc = cpu_to_le64(first_blkno);
85db90e7 2103 ocfs2_journal_dirty(handle, xs->inode_bh);
01225596
TM		 2104 } else
2105 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2106
2107 if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2108 /* Set extended attribute into external block */
78f30c31
TM		 2109 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2110 OCFS2_HAS_XATTR_FL);
01225596
TM		 2111 if (!ret || ret != -ENOSPC)
2112 goto end;
2113
78f30c31 2114 ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
01225596
TM		 2115 if (ret)
2116 goto end;
cf1d6c76
TY		 2117 }
2118
78f30c31 2119 ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
01225596
TM		 2120
2121end:
cf1d6c76
TY		 2122
2123 return ret;
2124}
2125
78f30c31
TM		 2126/* Check whether the new xattr can be inserted into the inode. */
2127static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2128 struct ocfs2_xattr_info *xi,
2129 struct ocfs2_xattr_search *xs)
2130{
2131 u64 value_size;
2132 struct ocfs2_xattr_entry *last;
2133 int free, i;
2134 size_t min_offs = xs->end - xs->base;
2135
2136 if (!xs->header)
2137 return 0;
2138
2139 last = xs->header->xh_entries;
2140
2141 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
2142 size_t offs = le16_to_cpu(last->xe_name_offset);
2143 if (offs < min_offs)
2144 min_offs = offs;
2145 last += 1;
2146 }
2147
2148 free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
2149 if (free < 0)
2150 return 0;
2151
2152 BUG_ON(!xs->not_found);
2153
2154 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2155 value_size = OCFS2_XATTR_ROOT_SIZE;
2156 else
2157 value_size = OCFS2_XATTR_SIZE(xi->value_len);
2158
2159 if (free >= sizeof(struct ocfs2_xattr_entry) +
2160 OCFS2_XATTR_SIZE(strlen(xi->name)) + value_size)
2161 return 1;
2162
2163 return 0;
2164}
2165
2166static int ocfs2_calc_xattr_set_need(struct inode *inode,
2167 struct ocfs2_dinode *di,
2168 struct ocfs2_xattr_info *xi,
2169 struct ocfs2_xattr_search *xis,
2170 struct ocfs2_xattr_search *xbs,
2171 int *clusters_need,
85db90e7
TM		 2172 int *meta_need,
2173 int *credits_need)
78f30c31
TM		 2174{
2175 int ret = 0, old_in_xb = 0;
85db90e7 2176 int clusters_add = 0, meta_add = 0, credits = 0;
78f30c31
TM		 2177 struct buffer_head *bh = NULL;
2178 struct ocfs2_xattr_block *xb = NULL;
2179 struct ocfs2_xattr_entry *xe = NULL;
2180 struct ocfs2_xattr_value_root *xv = NULL;
2181 char *base = NULL;
2182 int name_offset, name_len = 0;
2183 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2184 xi->value_len);
2185 u64 value_size;
2186
78f30c31 2187 if (xis->not_found && xbs->not_found) {
85db90e7
TM		 2188 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2189
2190 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
78f30c31 2191 clusters_add += new_clusters;
85db90e7
TM		 2192 credits += ocfs2_calc_extend_credits(inode->i_sb,
2193 &def_xv.xv.xr_list,
2194 new_clusters);
2195 }
78f30c31
TM		 2196
2197 goto meta_guess;
2198 }
2199
2200 if (!xis->not_found) {
2201 xe = xis->here;
2202 name_offset = le16_to_cpu(xe->xe_name_offset);
2203 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2204 base = xis->base;
85db90e7 2205 credits += OCFS2_INODE_UPDATE_CREDITS;
78f30c31 2206 } else {
970e4936 2207 int i, block_off = 0;
78f30c31
TM		 2208 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2209 xe = xbs->here;
2210 name_offset = le16_to_cpu(xe->xe_name_offset);
2211 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2212 i = xbs->here - xbs->header->xh_entries;
2213 old_in_xb = 1;
2214
2215 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2216 ret = ocfs2_xattr_bucket_get_name_value(inode,
2217 bucket_xh(xbs->bucket),
2218 i, &block_off,
2219 &name_offset);
2220 base = bucket_block(xbs->bucket, block_off);
85db90e7
TM		 2221 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2222 } else {
78f30c31 2223 base = xbs->base;
85db90e7
TM		 2224 credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
2225 }
2226 }
2227
2228 /*
2229 * delete a xattr doesn't need metadata and cluster allocation.
2230 * so just calculate the credits and return.
2231 *
2232 * The credits for removing the value tree will be extended
2233 * by ocfs2_remove_extent itself.
2234 */
2235 if (!xi->value) {
2236 if (!ocfs2_xattr_is_local(xe))
a90714c1 2237 credits += ocfs2_remove_extent_credits(inode->i_sb);
85db90e7
TM		 2238
2239 goto out;
78f30c31
TM		 2240 }
2241
2242 /* do cluster allocation guess first. */
2243 value_size = le64_to_cpu(xe->xe_value_size);
2244
2245 if (old_in_xb) {
2246 /*
2247 * In xattr set, we always try to set the xe in inode first,
2248 * so if it can be inserted into inode successfully, the old
2249 * one will be removed from the xattr block, and this xattr
2250 * will be inserted into inode as a new xattr in inode.
2251 */
2252 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
2253 clusters_add += new_clusters;
a90714c1 2254 credits += ocfs2_remove_extent_credits(inode->i_sb) +
85db90e7
TM		 2255 OCFS2_INODE_UPDATE_CREDITS;
2256 if (!ocfs2_xattr_is_local(xe))
2257 credits += ocfs2_calc_extend_credits(
2258 inode->i_sb,
2259 &def_xv.xv.xr_list,
2260 new_clusters);
78f30c31
TM		 2261 goto out;
2262 }
2263 }
2264
2265 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2266 /* the new values will be stored outside. */
2267 u32 old_clusters = 0;
2268
2269 if (!ocfs2_xattr_is_local(xe)) {
2270 old_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2271 value_size);
2272 xv = (struct ocfs2_xattr_value_root *)
2273 (base + name_offset + name_len);
97aff52a 2274 value_size = OCFS2_XATTR_ROOT_SIZE;
78f30c31
TM		 2275 } else
2276 xv = &def_xv.xv;
2277
85db90e7 2278 if (old_clusters >= new_clusters) {
a90714c1 2279 credits += ocfs2_remove_extent_credits(inode->i_sb);
78f30c31 2280 goto out;
85db90e7 2281 } else {
78f30c31
TM		 2282 meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
2283 clusters_add += new_clusters - old_clusters;
85db90e7
TM		 2284 credits += ocfs2_calc_extend_credits(inode->i_sb,
2285 &xv->xr_list,
2286 new_clusters -
2287 old_clusters);
97aff52a
TM		 2288 if (value_size >= OCFS2_XATTR_ROOT_SIZE)
2289 goto out;
78f30c31
TM		 2290 }
2291 } else {
2292 /*
2293 * Now the new value will be stored inside. So if the new
2294 * value is smaller than the size of value root or the old
2295 * value, we don't need any allocation, otherwise we have
2296 * to guess metadata allocation.
2297 */
2298 if ((ocfs2_xattr_is_local(xe) && value_size >= xi->value_len) ||
2299 (!ocfs2_xattr_is_local(xe) &&
2300 OCFS2_XATTR_ROOT_SIZE >= xi->value_len))
2301 goto out;
2302 }
2303
2304meta_guess:
2305 /* calculate metadata allocation. */
2306 if (di->i_xattr_loc) {
2307 if (!xbs->xattr_bh) {
4ae1d69b
JB		 2308 ret = ocfs2_read_xattr_block(inode,
2309 le64_to_cpu(di->i_xattr_loc),
2310 &bh);
78f30c31
TM		 2311 if (ret) {
2312 mlog_errno(ret);
2313 goto out;
2314 }
2315
2316 xb = (struct ocfs2_xattr_block *)bh->b_data;
2317 } else
2318 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2319
2320 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2321 struct ocfs2_extent_list *el =
2322 &xb->xb_attrs.xb_root.xt_list;
2323 meta_add += ocfs2_extend_meta_needed(el);
85db90e7
TM		 2324 credits += ocfs2_calc_extend_credits(inode->i_sb,
2325 el, 1);
78f30c31
TM		 2326 }
2327
2328 /*
2329 * This cluster will be used either for new bucket or for
2330 * new xattr block.
2331 * If the cluster size is the same as the bucket size, one
2332 * more is needed since we may need to extend the bucket
2333 * also.
2334 */
2335 clusters_add += 1;
85db90e7 2336 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
78f30c31 2337 if (OCFS2_XATTR_BUCKET_SIZE ==
85db90e7
TM		 2338 OCFS2_SB(inode->i_sb)->s_clustersize) {
2339 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
78f30c31 2340 clusters_add += 1;
85db90e7
TM		 2341 }
2342 } else {
78f30c31 2343 meta_add += 1;
85db90e7
TM		 2344 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
2345 }
78f30c31
TM		 2346out:
2347 if (clusters_need)
2348 *clusters_need = clusters_add;
2349 if (meta_need)
2350 *meta_need = meta_add;
85db90e7
TM		 2351 if (credits_need)
2352 *credits_need = credits;
78f30c31
TM		 2353 brelse(bh);
2354 return ret;
2355}
2356
2357static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
2358 struct ocfs2_dinode *di,
2359 struct ocfs2_xattr_info *xi,
2360 struct ocfs2_xattr_search *xis,
2361 struct ocfs2_xattr_search *xbs,
85db90e7
TM		 2362 struct ocfs2_xattr_set_ctxt *ctxt,
2363 int *credits)
78f30c31
TM		 2364{
2365 int clusters_add, meta_add, ret;
2366 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2367
2368 memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
2369
2370 ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
2371
2372 ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
85db90e7 2373 &clusters_add, &meta_add, credits);
78f30c31
TM		 2374 if (ret) {
2375 mlog_errno(ret);
2376 return ret;
2377 }
2378
85db90e7
TM		 2379 mlog(0, "Set xattr %s, reserve meta blocks = %d, clusters = %d, "
2380 "credits = %d\n", xi->name, meta_add, clusters_add, *credits);
78f30c31
TM		 2381
2382 if (meta_add) {
2383 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
2384 &ctxt->meta_ac);
2385 if (ret) {
2386 mlog_errno(ret);
2387 goto out;
2388 }
2389 }
2390
2391 if (clusters_add) {
2392 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
2393 if (ret)
2394 mlog_errno(ret);
2395 }
2396out:
2397 if (ret) {
2398 if (ctxt->meta_ac) {
2399 ocfs2_free_alloc_context(ctxt->meta_ac);
2400 ctxt->meta_ac = NULL;
2401 }
2402
2403 /*
2404 * We cannot have an error and a non null ctxt->data_ac.
2405 */
2406 }
2407
2408 return ret;
2409}
2410
85db90e7
TM		 2411static int __ocfs2_xattr_set_handle(struct inode *inode,
2412 struct ocfs2_dinode *di,
2413 struct ocfs2_xattr_info *xi,
2414 struct ocfs2_xattr_search *xis,
2415 struct ocfs2_xattr_search *xbs,
2416 struct ocfs2_xattr_set_ctxt *ctxt)
2417{
9f868f16 2418 int ret = 0, credits, old_found;
85db90e7
TM		 2419
2420 if (!xi->value) {
2421 /* Remove existing extended attribute */
2422 if (!xis->not_found)
2423 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2424 else if (!xbs->not_found)
2425 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2426 } else {
2427 /* We always try to set extended attribute into inode first*/
2428 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2429 if (!ret && !xbs->not_found) {
2430 /*
2431 * If succeed and that extended attribute existing in
2432 * external block, then we will remove it.
2433 */
2434 xi->value = NULL;
2435 xi->value_len = 0;
2436
9f868f16 2437 old_found = xis->not_found;
85db90e7
TM		 2438 xis->not_found = -ENODATA;
2439 ret = ocfs2_calc_xattr_set_need(inode,
2440 di,
2441 xi,
2442 xis,
2443 xbs,
2444 NULL,
2445 NULL,
2446 &credits);
9f868f16 2447 xis->not_found = old_found;
85db90e7
TM		 2448 if (ret) {
2449 mlog_errno(ret);
2450 goto out;
2451 }
2452
2453 ret = ocfs2_extend_trans(ctxt->handle, credits +
2454 ctxt->handle->h_buffer_credits);
2455 if (ret) {
2456 mlog_errno(ret);
2457 goto out;
2458 }
2459 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2460 } else if (ret == -ENOSPC) {
2461 if (di->i_xattr_loc && !xbs->xattr_bh) {
2462 ret = ocfs2_xattr_block_find(inode,
2463 xi->name_index,
2464 xi->name, xbs);
2465 if (ret)
2466 goto out;
2467
9f868f16 2468 old_found = xis->not_found;
85db90e7
TM		 2469 xis->not_found = -ENODATA;
2470 ret = ocfs2_calc_xattr_set_need(inode,
2471 di,
2472 xi,
2473 xis,
2474 xbs,
2475 NULL,
2476 NULL,
2477 &credits);
9f868f16 2478 xis->not_found = old_found;
85db90e7
TM		 2479 if (ret) {
2480 mlog_errno(ret);
2481 goto out;
2482 }
2483
2484 ret = ocfs2_extend_trans(ctxt->handle, credits +
2485 ctxt->handle->h_buffer_credits);
2486 if (ret) {
2487 mlog_errno(ret);
2488 goto out;
2489 }
2490 }
2491 /*
2492 * If no space in inode, we will set extended attribute
2493 * into external block.
2494 */
2495 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2496 if (ret)
2497 goto out;
2498 if (!xis->not_found) {
2499 /*
2500 * If succeed and that extended attribute
2501 * existing in inode, we will remove it.
2502 */
2503 xi->value = NULL;
2504 xi->value_len = 0;
2505 xbs->not_found = -ENODATA;
2506 ret = ocfs2_calc_xattr_set_need(inode,
2507 di,
2508 xi,
2509 xis,
2510 xbs,
2511 NULL,
2512 NULL,
2513 &credits);
2514 if (ret) {
2515 mlog_errno(ret);
2516 goto out;
2517 }
2518
2519 ret = ocfs2_extend_trans(ctxt->handle, credits +
2520 ctxt->handle->h_buffer_credits);
2521 if (ret) {
2522 mlog_errno(ret);
2523 goto out;
2524 }
2525 ret = ocfs2_xattr_ibody_set(inode, xi,
2526 xis, ctxt);
2527 }
2528 }
2529 }
2530
2531out:
2532 return ret;
2533}
2534
6c3faba4
TY		 2535/*
2536 * This function only called duing creating inode
2537 * for init security/acl xattrs of the new inode.
2538 * The xattrs could be put into ibody or extent block,
2539 * xattr bucket would not be use in this case.
2540 * transanction credits also be reserved in here.
2541 */
2542int ocfs2_xattr_set_handle(handle_t *handle,
2543 struct inode *inode,
2544 struct buffer_head *di_bh,
2545 int name_index,
2546 const char *name,
2547 const void *value,
2548 size_t value_len,
2549 int flags,
2550 struct ocfs2_alloc_context *meta_ac,
2551 struct ocfs2_alloc_context *data_ac)
2552{
2553 struct ocfs2_dinode *di;
2554 int ret;
2555
2556 struct ocfs2_xattr_info xi = {
2557 .name_index = name_index,
2558 .name = name,
2559 .value = value,
2560 .value_len = value_len,
2561 };
2562
2563 struct ocfs2_xattr_search xis = {
2564 .not_found = -ENODATA,
2565 };
2566
2567 struct ocfs2_xattr_search xbs = {
2568 .not_found = -ENODATA,
2569 };
2570
2571 struct ocfs2_xattr_set_ctxt ctxt = {
2572 .handle = handle,
2573 .meta_ac = meta_ac,
2574 .data_ac = data_ac,
2575 };
2576
2577 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2578 return -EOPNOTSUPP;
2579
2580 xis.inode_bh = xbs.inode_bh = di_bh;
2581 di = (struct ocfs2_dinode *)di_bh->b_data;
2582
2583 down_write(&OCFS2_I(inode)->ip_xattr_sem);
2584
2585 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2586 if (ret)
2587 goto cleanup;
2588 if (xis.not_found) {
2589 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2590 if (ret)
2591 goto cleanup;
2592 }
2593
2594 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2595
2596cleanup:
2597 up_write(&OCFS2_I(inode)->ip_xattr_sem);
2598 brelse(xbs.xattr_bh);
2599
2600 return ret;
2601}
2602
cf1d6c76
TY		 2603/*
2604 * ocfs2_xattr_set()
2605 *
2606 * Set, replace or remove an extended attribute for this inode.
2607 * value is NULL to remove an existing extended attribute, else either
2608 * create or replace an extended attribute.
2609 */
2610int ocfs2_xattr_set(struct inode *inode,
2611 int name_index,
2612 const char *name,
2613 const void *value,
2614 size_t value_len,
2615 int flags)
2616{
2617 struct buffer_head *di_bh = NULL;
2618 struct ocfs2_dinode *di;
85db90e7 2619 int ret, credits;
78f30c31 2620 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
85db90e7 2621 struct inode *tl_inode = osb->osb_tl_inode;
78f30c31 2622 struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
cf1d6c76
TY		 2623
2624 struct ocfs2_xattr_info xi = {
2625 .name_index = name_index,
2626 .name = name,
2627 .value = value,
2628 .value_len = value_len,
2629 };
2630
2631 struct ocfs2_xattr_search xis = {
2632 .not_found = -ENODATA,
2633 };
2634
2635 struct ocfs2_xattr_search xbs = {
2636 .not_found = -ENODATA,
2637 };
2638
8154da3d
TY		 2639 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2640 return -EOPNOTSUPP;
2641
ba937127
JB		 2642 /*
2643 * Only xbs will be used on indexed trees. xis doesn't need a
2644 * bucket.
2645 */
2646 xbs.bucket = ocfs2_xattr_bucket_new(inode);
2647 if (!xbs.bucket) {
2648 mlog_errno(-ENOMEM);
2649 return -ENOMEM;
2650 }
2651
cf1d6c76
TY		 2652 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2653 if (ret < 0) {
2654 mlog_errno(ret);
ba937127 2655 goto cleanup_nolock;
cf1d6c76
TY		 2656 }
2657 xis.inode_bh = xbs.inode_bh = di_bh;
2658 di = (struct ocfs2_dinode *)di_bh->b_data;
2659
2660 down_write(&OCFS2_I(inode)->ip_xattr_sem);
2661 /*
2662 * Scan inode and external block to find the same name
2663 * extended attribute and collect search infomation.
2664 */
2665 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2666 if (ret)
2667 goto cleanup;
2668 if (xis.not_found) {
2669 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2670 if (ret)
2671 goto cleanup;
2672 }
2673
2674 if (xis.not_found && xbs.not_found) {
2675 ret = -ENODATA;
2676 if (flags & XATTR_REPLACE)
2677 goto cleanup;
2678 ret = 0;
2679 if (!value)
2680 goto cleanup;
2681 } else {
2682 ret = -EEXIST;
2683 if (flags & XATTR_CREATE)
2684 goto cleanup;
2685 }
2686
85db90e7
TM		 2687
2688 mutex_lock(&tl_inode->i_mutex);
2689
2690 if (ocfs2_truncate_log_needs_flush(osb)) {
2691 ret = __ocfs2_flush_truncate_log(osb);
2692 if (ret < 0) {
2693 mutex_unlock(&tl_inode->i_mutex);
2694 mlog_errno(ret);
2695 goto cleanup;
2696 }
2697 }
2698 mutex_unlock(&tl_inode->i_mutex);
2699
2700 ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
2701 &xbs, &ctxt, &credits);
78f30c31
TM		 2702 if (ret) {
2703 mlog_errno(ret);
2704 goto cleanup;
2705 }
2706
85db90e7
TM		 2707 ctxt.handle = ocfs2_start_trans(osb, credits);
2708 if (IS_ERR(ctxt.handle)) {
2709 ret = PTR_ERR(ctxt.handle);
2710 mlog_errno(ret);
2711 goto cleanup;
cf1d6c76 2712 }
85db90e7
TM		 2713
2714 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2715
2716 ocfs2_commit_trans(osb, ctxt.handle);
2717
78f30c31
TM		 2718 if (ctxt.data_ac)
2719 ocfs2_free_alloc_context(ctxt.data_ac);
2720 if (ctxt.meta_ac)
2721 ocfs2_free_alloc_context(ctxt.meta_ac);
2722 if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
2723 ocfs2_schedule_truncate_log_flush(osb, 1);
2724 ocfs2_run_deallocs(osb, &ctxt.dealloc);
cf1d6c76
TY		 2725cleanup:
2726 up_write(&OCFS2_I(inode)->ip_xattr_sem);
2727 ocfs2_inode_unlock(inode, 1);
ba937127 2728cleanup_nolock:
cf1d6c76
TY		 2729 brelse(di_bh);
2730 brelse(xbs.xattr_bh);
ba937127 2731 ocfs2_xattr_bucket_free(xbs.bucket);
cf1d6c76
TY		 2732
2733 return ret;
2734}
2735
0c044f0b
TM		 2736/*
2737 * Find the xattr extent rec which may contains name_hash.
2738 * e_cpos will be the first name hash of the xattr rec.
2739 * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2740 */
2741static int ocfs2_xattr_get_rec(struct inode *inode,
2742 u32 name_hash,
2743 u64 *p_blkno,
2744 u32 *e_cpos,
2745 u32 *num_clusters,
2746 struct ocfs2_extent_list *el)
2747{
2748 int ret = 0, i;
2749 struct buffer_head *eb_bh = NULL;
2750 struct ocfs2_extent_block *eb;
2751 struct ocfs2_extent_rec *rec = NULL;
2752 u64 e_blkno = 0;
2753
2754 if (el->l_tree_depth) {
2755 ret = ocfs2_find_leaf(inode, el, name_hash, &eb_bh);
2756 if (ret) {
2757 mlog_errno(ret);
2758 goto out;
2759 }
2760
2761 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2762 el = &eb->h_list;
2763
2764 if (el->l_tree_depth) {
2765 ocfs2_error(inode->i_sb,
2766 "Inode %lu has non zero tree depth in "
2767 "xattr tree block %llu\n", inode->i_ino,
2768 (unsigned long long)eb_bh->b_blocknr);
2769 ret = -EROFS;
2770 goto out;
2771 }
2772 }
2773
2774 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2775 rec = &el->l_recs[i];
2776
2777 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2778 e_blkno = le64_to_cpu(rec->e_blkno);
2779 break;
2780 }
2781 }
2782
2783 if (!e_blkno) {
2784 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2785 "record (%u, %u, 0) in xattr", inode->i_ino,
2786 le32_to_cpu(rec->e_cpos),
2787 ocfs2_rec_clusters(el, rec));
2788 ret = -EROFS;
2789 goto out;
2790 }
2791
2792 *p_blkno = le64_to_cpu(rec->e_blkno);
2793 *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2794 if (e_cpos)
2795 *e_cpos = le32_to_cpu(rec->e_cpos);
2796out:
2797 brelse(eb_bh);
2798 return ret;
2799}
2800
2801typedef int (xattr_bucket_func)(struct inode *inode,
2802 struct ocfs2_xattr_bucket *bucket,
2803 void *para);
2804
589dc260 2805static int ocfs2_find_xe_in_bucket(struct inode *inode,
e2356a3f 2806 struct ocfs2_xattr_bucket *bucket,
589dc260
TM		 2807 int name_index,
2808 const char *name,
2809 u32 name_hash,
2810 u16 *xe_index,
2811 int *found)
2812{
2813 int i, ret = 0, cmp = 1, block_off, new_offset;
e2356a3f 2814 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
589dc260
TM		 2815 size_t name_len = strlen(name);
2816 struct ocfs2_xattr_entry *xe = NULL;
589dc260
TM		 2817 char *xe_name;
2818
2819 /*
2820 * We don't use binary search in the bucket because there
2821 * may be multiple entries with the same name hash.
2822 */
2823 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2824 xe = &xh->xh_entries[i];
2825
2826 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2827 continue;
2828 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2829 break;
2830
2831 cmp = name_index - ocfs2_xattr_get_type(xe);
2832 if (!cmp)
2833 cmp = name_len - xe->xe_name_len;
2834 if (cmp)
2835 continue;
2836
2837 ret = ocfs2_xattr_bucket_get_name_value(inode,
2838 xh,
2839 i,
2840 &block_off,
2841 &new_offset);
2842 if (ret) {
2843 mlog_errno(ret);
2844 break;
2845 }
2846
970e4936 2847
e2356a3f
JB		 2848 xe_name = bucket_block(bucket, block_off) + new_offset;
2849 if (!memcmp(name, xe_name, name_len)) {
589dc260
TM		 2850 *xe_index = i;
2851 *found = 1;
2852 ret = 0;
2853 break;
2854 }
2855 }
2856
2857 return ret;
2858}
2859
2860/*
2861 * Find the specified xattr entry in a series of buckets.
2862 * This series start from p_blkno and last for num_clusters.
2863 * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2864 * the num of the valid buckets.
2865 *
2866 * Return the buffer_head this xattr should reside in. And if the xattr's
2867 * hash is in the gap of 2 buckets, return the lower bucket.
2868 */
2869static int ocfs2_xattr_bucket_find(struct inode *inode,
2870 int name_index,
2871 const char *name,
2872 u32 name_hash,
2873 u64 p_blkno,
2874 u32 first_hash,
2875 u32 num_clusters,
2876 struct ocfs2_xattr_search *xs)
2877{
2878 int ret, found = 0;
589dc260
TM		 2879 struct ocfs2_xattr_header *xh = NULL;
2880 struct ocfs2_xattr_entry *xe = NULL;
2881 u16 index = 0;
2882 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2883 int low_bucket = 0, bucket, high_bucket;
e2356a3f 2884 struct ocfs2_xattr_bucket *search;
589dc260 2885 u32 last_hash;
e2356a3f 2886 u64 blkno, lower_blkno = 0;
589dc260 2887
e2356a3f
JB		 2888 search = ocfs2_xattr_bucket_new(inode);
2889 if (!search) {
2890 ret = -ENOMEM;
2891 mlog_errno(ret);
2892 goto out;
2893 }
2894
2895 ret = ocfs2_read_xattr_bucket(search, p_blkno);
589dc260
TM		 2896 if (ret) {
2897 mlog_errno(ret);
2898 goto out;
2899 }
2900
e2356a3f 2901 xh = bucket_xh(search);
589dc260 2902 high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
589dc260 2903 while (low_bucket <= high_bucket) {
e2356a3f 2904 ocfs2_xattr_bucket_relse(search);
589dc260 2905
e2356a3f 2906 bucket = (low_bucket + high_bucket) / 2;
589dc260 2907 blkno = p_blkno + bucket * blk_per_bucket;
e2356a3f 2908 ret = ocfs2_read_xattr_bucket(search, blkno);
589dc260
TM		 2909 if (ret) {
2910 mlog_errno(ret);
2911 goto out;
2912 }
2913
e2356a3f 2914 xh = bucket_xh(search);
589dc260
TM		 2915 xe = &xh->xh_entries[0];
2916 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
2917 high_bucket = bucket - 1;
2918 continue;
2919 }
2920
2921 /*
2922 * Check whether the hash of the last entry in our
5a095611
TM		 2923 * bucket is larger than the search one. for an empty
2924 * bucket, the last one is also the first one.
589dc260 2925 */
5a095611
TM		 2926 if (xh->xh_count)
2927 xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
2928
589dc260
TM		 2929 last_hash = le32_to_cpu(xe->xe_name_hash);
2930
e2356a3f
JB		 2931 /* record lower_blkno which may be the insert place. */
2932 lower_blkno = blkno;
589dc260
TM		 2933
2934 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
2935 low_bucket = bucket + 1;
2936 continue;
2937 }
2938
2939 /* the searched xattr should reside in this bucket if exists. */
e2356a3f 2940 ret = ocfs2_find_xe_in_bucket(inode, search,
589dc260
TM		 2941 name_index, name, name_hash,
2942 &index, &found);
2943 if (ret) {
2944 mlog_errno(ret);
2945 goto out;
2946 }
2947 break;
2948 }
2949
2950 /*
2951 * Record the bucket we have found.
2952 * When the xattr's hash value is in the gap of 2 buckets, we will
2953 * always set it to the previous bucket.
2954 */
e2356a3f
JB		 2955 if (!lower_blkno)
2956 lower_blkno = p_blkno;
2957
2958 /* This should be in cache - we just read it during the search */
2959 ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
2960 if (ret) {
2961 mlog_errno(ret);
2962 goto out;
589dc260 2963 }
589dc260 2964
ba937127
JB		 2965 xs->header = bucket_xh(xs->bucket);
2966 xs->base = bucket_block(xs->bucket, 0);
589dc260
TM		 2967 xs->end = xs->base + inode->i_sb->s_blocksize;
2968
2969 if (found) {
589dc260
TM		 2970 xs->here = &xs->header->xh_entries[index];
2971 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
ba937127 2972 (unsigned long long)bucket_blkno(xs->bucket), index);
589dc260
TM		 2973 } else
2974 ret = -ENODATA;
2975
2976out:
e2356a3f 2977 ocfs2_xattr_bucket_free(search);
589dc260
TM		 2978 return ret;
2979}
2980
2981static int ocfs2_xattr_index_block_find(struct inode *inode,
2982 struct buffer_head *root_bh,
2983 int name_index,
2984 const char *name,
2985 struct ocfs2_xattr_search *xs)
2986{
2987 int ret;
2988 struct ocfs2_xattr_block *xb =
2989 (struct ocfs2_xattr_block *)root_bh->b_data;
2990 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
2991 struct ocfs2_extent_list *el = &xb_root->xt_list;
2992 u64 p_blkno = 0;
2993 u32 first_hash, num_clusters = 0;
2057e5c6 2994 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
589dc260
TM		 2995
2996 if (le16_to_cpu(el->l_next_free_rec) == 0)
2997 return -ENODATA;
2998
2999 mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
3000 name, name_hash, name_index);
3001
3002 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3003 &num_clusters, el);
3004 if (ret) {
3005 mlog_errno(ret);
3006 goto out;
3007 }
3008
3009 BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3010
3011 mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
de29c085
MF		 3012 "in the rec is %u\n", num_clusters, (unsigned long long)p_blkno,
3013 first_hash);
589dc260
TM		 3014
3015 ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3016 p_blkno, first_hash, num_clusters, xs);
3017
3018out:
3019 return ret;
3020}
3021
0c044f0b
TM		 3022static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3023 u64 blkno,
3024 u32 clusters,
3025 xattr_bucket_func *func,
3026 void *para)
3027{
6dde41d9 3028 int i, ret = 0;
0c044f0b
TM		 3029 u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3030 u32 num_buckets = clusters * bpc;
ba937127 3031 struct ocfs2_xattr_bucket *bucket;
0c044f0b 3032
ba937127
JB		 3033 bucket = ocfs2_xattr_bucket_new(inode);
3034 if (!bucket) {
3035 mlog_errno(-ENOMEM);
3036 return -ENOMEM;
3037 }
0c044f0b
TM		 3038
3039 mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
de29c085 3040 clusters, (unsigned long long)blkno);
0c044f0b 3041
ba937127
JB		 3042 for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3043 ret = ocfs2_read_xattr_bucket(bucket, blkno);
0c044f0b
TM		 3044 if (ret) {
3045 mlog_errno(ret);
ba937127 3046 break;
0c044f0b
TM		 3047 }
3048
0c044f0b
TM		 3049 /*
3050 * The real bucket num in this series of blocks is stored
3051 * in the 1st bucket.
3052 */
3053 if (i == 0)
ba937127 3054 num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
0c044f0b 3055
de29c085
MF		 3056 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
3057 (unsigned long long)blkno,
ba937127 3058 le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
0c044f0b 3059 if (func) {
ba937127
JB		 3060 ret = func(inode, bucket, para);
3061 if (ret)
0c044f0b 3062 mlog_errno(ret);
ba937127 3063 /* Fall through to bucket_relse() */
0c044f0b
TM		 3064 }
3065
ba937127
JB		 3066 ocfs2_xattr_bucket_relse(bucket);
3067 if (ret)
3068 break;
0c044f0b
TM		 3069 }
3070
ba937127 3071 ocfs2_xattr_bucket_free(bucket);
0c044f0b
TM		 3072 return ret;
3073}
3074
3075struct ocfs2_xattr_tree_list {
3076 char *buffer;
3077 size_t buffer_size;
936b8834 3078 size_t result;
0c044f0b
TM		 3079};
3080
3081static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
3082 struct ocfs2_xattr_header *xh,
3083 int index,
3084 int *block_off,
3085 int *new_offset)
3086{
3087 u16 name_offset;
3088
3089 if (index < 0 || index >= le16_to_cpu(xh->xh_count))
3090 return -EINVAL;
3091
3092 name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
3093
3094 *block_off = name_offset >> inode->i_sb->s_blocksize_bits;
3095 *new_offset = name_offset % inode->i_sb->s_blocksize;
3096
3097 return 0;
3098}
3099
3100static int ocfs2_list_xattr_bucket(struct inode *inode,
3101 struct ocfs2_xattr_bucket *bucket,
3102 void *para)
3103{
936b8834 3104 int ret = 0, type;
0c044f0b 3105 struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
0c044f0b 3106 int i, block_off, new_offset;
936b8834 3107 const char *prefix, *name;
0c044f0b 3108
3e632946
JB		 3109 for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
3110 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
936b8834
TM		 3111 type = ocfs2_xattr_get_type(entry);
3112 prefix = ocfs2_xattr_prefix(type);
0c044f0b 3113
936b8834 3114 if (prefix) {
0c044f0b 3115 ret = ocfs2_xattr_bucket_get_name_value(inode,
3e632946 3116 bucket_xh(bucket),
0c044f0b
TM		 3117 i,
3118 &block_off,
3119 &new_offset);
3120 if (ret)
3121 break;
936b8834 3122
51def39f 3123 name = (const char *)bucket_block(bucket, block_off) +
936b8834
TM		 3124 new_offset;
3125 ret = ocfs2_xattr_list_entry(xl->buffer,
3126 xl->buffer_size,
3127 &xl->result,
3128 prefix, name,
3129 entry->xe_name_len);
3130 if (ret)
3131 break;
0c044f0b
TM		 3132 }
3133 }
3134
3135 return ret;
3136}
3137
3138static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
3139 struct ocfs2_xattr_tree_root *xt,
3140 char *buffer,
3141 size_t buffer_size)
3142{
3143 struct ocfs2_extent_list *el = &xt->xt_list;
3144 int ret = 0;
3145 u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
3146 u64 p_blkno = 0;
3147 struct ocfs2_xattr_tree_list xl = {
3148 .buffer = buffer,
3149 .buffer_size = buffer_size,
936b8834 3150 .result = 0,
0c044f0b
TM		 3151 };
3152
3153 if (le16_to_cpu(el->l_next_free_rec) == 0)
3154 return 0;
3155
3156 while (name_hash > 0) {
3157 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
3158 &e_cpos, &num_clusters, el);
3159 if (ret) {
3160 mlog_errno(ret);
3161 goto out;
3162 }
3163
3164 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
3165 ocfs2_list_xattr_bucket,
3166 &xl);
3167 if (ret) {
3168 mlog_errno(ret);
3169 goto out;
3170 }
3171
3172 if (e_cpos == 0)
3173 break;
3174
3175 name_hash = e_cpos - 1;
3176 }
3177
936b8834 3178 ret = xl.result;
0c044f0b
TM		 3179out:
3180 return ret;
3181}
01225596
TM		 3182
3183static int cmp_xe(const void *a, const void *b)
3184{
3185 const struct ocfs2_xattr_entry *l = a, *r = b;
3186 u32 l_hash = le32_to_cpu(l->xe_name_hash);
3187 u32 r_hash = le32_to_cpu(r->xe_name_hash);
3188
3189 if (l_hash > r_hash)
3190 return 1;
3191 if (l_hash < r_hash)
3192 return -1;
3193 return 0;
3194}
3195
3196static void swap_xe(void *a, void *b, int size)
3197{
3198 struct ocfs2_xattr_entry *l = a, *r = b, tmp;
3199
3200 tmp = *l;
3201 memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
3202 memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
3203}
3204
3205/*
3206 * When the ocfs2_xattr_block is filled up, new bucket will be created
3207 * and all the xattr entries will be moved to the new bucket.
178eeac3
JB		 3208 * The header goes at the start of the bucket, and the names+values are
3209 * filled from the end. This is why *target starts as the last buffer.
01225596
TM		 3210 * Note: we need to sort the entries since they are not saved in order
3211 * in the ocfs2_xattr_block.
3212 */
3213static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
3214 struct buffer_head *xb_bh,
178eeac3 3215 struct ocfs2_xattr_bucket *bucket)
01225596
TM		 3216{
3217 int i, blocksize = inode->i_sb->s_blocksize;
178eeac3 3218 int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
01225596
TM		 3219 u16 offset, size, off_change;
3220 struct ocfs2_xattr_entry *xe;
3221 struct ocfs2_xattr_block *xb =
3222 (struct ocfs2_xattr_block *)xb_bh->b_data;
3223 struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
178eeac3 3224 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
01225596 3225 u16 count = le16_to_cpu(xb_xh->xh_count);
178eeac3
JB		 3226 char *src = xb_bh->b_data;
3227 char *target = bucket_block(bucket, blks - 1);
01225596
TM		 3228
3229 mlog(0, "cp xattr from block %llu to bucket %llu\n",
3230 (unsigned long long)xb_bh->b_blocknr,
178eeac3
JB		 3231 (unsigned long long)bucket_blkno(bucket));
3232
3233 for (i = 0; i < blks; i++)
3234 memset(bucket_block(bucket, i), 0, blocksize);
01225596 3235
01225596
TM		 3236 /*
3237 * Since the xe_name_offset is based on ocfs2_xattr_header,
3238 * there is a offset change corresponding to the change of
3239 * ocfs2_xattr_header's position.
3240 */
3241 off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3242 xe = &xb_xh->xh_entries[count - 1];
3243 offset = le16_to_cpu(xe->xe_name_offset) + off_change;
3244 size = blocksize - offset;
3245
3246 /* copy all the names and values. */
01225596
TM		 3247 memcpy(target + offset, src + offset, size);
3248
3249 /* Init new header now. */
3250 xh->xh_count = xb_xh->xh_count;
3251 xh->xh_num_buckets = cpu_to_le16(1);
3252 xh->xh_name_value_len = cpu_to_le16(size);
3253 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
3254
3255 /* copy all the entries. */
178eeac3 3256 target = bucket_block(bucket, 0);
01225596
TM		 3257 offset = offsetof(struct ocfs2_xattr_header, xh_entries);
3258 size = count * sizeof(struct ocfs2_xattr_entry);
3259 memcpy(target + offset, (char *)xb_xh + offset, size);
3260
3261 /* Change the xe offset for all the xe because of the move. */
3262 off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
3263 offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3264 for (i = 0; i < count; i++)
3265 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
3266
3267 mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
3268 offset, size, off_change);
3269
3270 sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
3271 cmp_xe, swap_xe);
3272}
3273
3274/*
3275 * After we move xattr from block to index btree, we have to
3276 * update ocfs2_xattr_search to the new xe and base.
3277 *
3278 * When the entry is in xattr block, xattr_bh indicates the storage place.
3279 * While if the entry is in index b-tree, "bucket" indicates the
3280 * real place of the xattr.
3281 */
178eeac3
JB		 3282static void ocfs2_xattr_update_xattr_search(struct inode *inode,
3283 struct ocfs2_xattr_search *xs,
3284 struct buffer_head *old_bh)
01225596 3285{
01225596
TM		 3286 char *buf = old_bh->b_data;
3287 struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
3288 struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
178eeac3 3289 int i;
01225596 3290
ba937127 3291 xs->header = bucket_xh(xs->bucket);
178eeac3 3292 xs->base = bucket_block(xs->bucket, 0);
01225596
TM		 3293 xs->end = xs->base + inode->i_sb->s_blocksize;
3294
178eeac3
JB		 3295 if (xs->not_found)
3296 return;
01225596 3297
178eeac3
JB		 3298 i = xs->here - old_xh->xh_entries;
3299 xs->here = &xs->header->xh_entries[i];
01225596
TM		 3300}
3301
3302static int ocfs2_xattr_create_index_block(struct inode *inode,
78f30c31
TM		 3303 struct ocfs2_xattr_search *xs,
3304 struct ocfs2_xattr_set_ctxt *ctxt)
01225596 3305{
85db90e7 3306 int ret;
01225596
TM		 3307 u32 bit_off, len;
3308 u64 blkno;
85db90e7 3309 handle_t *handle = ctxt->handle;
01225596
TM		 3310 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3311 struct ocfs2_inode_info *oi = OCFS2_I(inode);
01225596
TM		 3312 struct buffer_head *xb_bh = xs->xattr_bh;
3313 struct ocfs2_xattr_block *xb =
3314 (struct ocfs2_xattr_block *)xb_bh->b_data;
3315 struct ocfs2_xattr_tree_root *xr;
3316 u16 xb_flags = le16_to_cpu(xb->xb_flags);
01225596
TM		 3317
3318 mlog(0, "create xattr index block for %llu\n",
3319 (unsigned long long)xb_bh->b_blocknr);
3320
3321 BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
178eeac3 3322 BUG_ON(!xs->bucket);
01225596 3323
01225596
TM		 3324 /*
3325 * XXX:
3326 * We can use this lock for now, and maybe move to a dedicated mutex
3327 * if performance becomes a problem later.
3328 */
3329 down_write(&oi->ip_alloc_sem);
3330
01225596
TM		 3331 ret = ocfs2_journal_access(handle, inode, xb_bh,
3332 OCFS2_JOURNAL_ACCESS_WRITE);
3333 if (ret) {
3334 mlog_errno(ret);
85db90e7 3335 goto out;
01225596
TM		 3336 }
3337
78f30c31
TM		 3338 ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac,
3339 1, 1, &bit_off, &len);
01225596
TM		 3340 if (ret) {
3341 mlog_errno(ret);
85db90e7 3342 goto out;
01225596
TM		 3343 }
3344
3345 /*
3346 * The bucket may spread in many blocks, and
3347 * we will only touch the 1st block and the last block
3348 * in the whole bucket(one for entry and one for data).
3349 */
3350 blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
3351
de29c085
MF		 3352 mlog(0, "allocate 1 cluster from %llu to xattr block\n",
3353 (unsigned long long)blkno);
01225596 3354
178eeac3
JB		 3355 ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
3356 if (ret) {
01225596 3357 mlog_errno(ret);
85db90e7 3358 goto out;
01225596
TM		 3359 }
3360
178eeac3
JB		 3361 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
3362 OCFS2_JOURNAL_ACCESS_CREATE);
01225596
TM		 3363 if (ret) {
3364 mlog_errno(ret);
85db90e7 3365 goto out;
01225596
TM		 3366 }
3367
178eeac3
JB		 3368 ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
3369 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
01225596 3370
178eeac3 3371 ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
01225596
TM		 3372
3373 /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
3374 memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
3375 offsetof(struct ocfs2_xattr_block, xb_attrs));
3376
3377 xr = &xb->xb_attrs.xb_root;
3378 xr->xt_clusters = cpu_to_le32(1);
3379 xr->xt_last_eb_blk = 0;
3380 xr->xt_list.l_tree_depth = 0;
3381 xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
3382 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
3383
3384 xr->xt_list.l_recs[0].e_cpos = 0;
3385 xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
3386 xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
3387
3388 xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
3389
85db90e7 3390 ocfs2_journal_dirty(handle, xb_bh);
01225596 3391
85db90e7 3392out:
01225596
TM		 3393 up_write(&oi->ip_alloc_sem);
3394
01225596
TM		 3395 return ret;
3396}
3397
3398static int cmp_xe_offset(const void *a, const void *b)
3399{
3400 const struct ocfs2_xattr_entry *l = a, *r = b;
3401 u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
3402 u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
3403
3404 if (l_name_offset < r_name_offset)
3405 return 1;
3406 if (l_name_offset > r_name_offset)
3407 return -1;
3408 return 0;
3409}
3410
3411/*
3412 * defrag a xattr bucket if we find that the bucket has some
3413 * holes beteen name/value pairs.
3414 * We will move all the name/value pairs to the end of the bucket
3415 * so that we can spare some space for insertion.
3416 */
3417static int ocfs2_defrag_xattr_bucket(struct inode *inode,
85db90e7 3418 handle_t *handle,
01225596
TM		 3419 struct ocfs2_xattr_bucket *bucket)
3420{
3421 int ret, i;
3422 size_t end, offset, len, value_len;
3423 struct ocfs2_xattr_header *xh;
3424 char *entries, *buf, *bucket_buf = NULL;
9c7759aa 3425 u64 blkno = bucket_blkno(bucket);
01225596 3426 u16 xh_free_start;
01225596 3427 size_t blocksize = inode->i_sb->s_blocksize;
01225596 3428 struct ocfs2_xattr_entry *xe;
01225596
TM		 3429
3430 /*
3431 * In order to make the operation more efficient and generic,
3432 * we copy all the blocks into a contiguous memory and do the
3433 * defragment there, so if anything is error, we will not touch
3434 * the real block.
3435 */
3436 bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
3437 if (!bucket_buf) {
3438 ret = -EIO;
3439 goto out;
3440 }
3441
3442 buf = bucket_buf;
1c32a2fd
TM		 3443 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3444 memcpy(buf, bucket_block(bucket, i), blocksize);
01225596 3445
1c32a2fd 3446 ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
161d6f30
JB		 3447 OCFS2_JOURNAL_ACCESS_WRITE);
3448 if (ret < 0) {
3449 mlog_errno(ret);
85db90e7 3450 goto out;
01225596
TM		 3451 }
3452
3453 xh = (struct ocfs2_xattr_header *)bucket_buf;
3454 entries = (char *)xh->xh_entries;
3455 xh_free_start = le16_to_cpu(xh->xh_free_start);
3456
3457 mlog(0, "adjust xattr bucket in %llu, count = %u, "
3458 "xh_free_start = %u, xh_name_value_len = %u.\n",
de29c085
MF		 3459 (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
3460 xh_free_start, le16_to_cpu(xh->xh_name_value_len));
01225596
TM		 3461
3462 /*
3463 * sort all the entries by their offset.
3464 * the largest will be the first, so that we can
3465 * move them to the end one by one.
3466 */
3467 sort(entries, le16_to_cpu(xh->xh_count),
3468 sizeof(struct ocfs2_xattr_entry),
3469 cmp_xe_offset, swap_xe);
3470
3471 /* Move all name/values to the end of the bucket. */
3472 xe = xh->xh_entries;
3473 end = OCFS2_XATTR_BUCKET_SIZE;
3474 for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
3475 offset = le16_to_cpu(xe->xe_name_offset);
3476 if (ocfs2_xattr_is_local(xe))
3477 value_len = OCFS2_XATTR_SIZE(
3478 le64_to_cpu(xe->xe_value_size));
3479 else
3480 value_len = OCFS2_XATTR_ROOT_SIZE;
3481 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
3482
3483 /*
3484 * We must make sure that the name/value pair
3485 * exist in the same block. So adjust end to
3486 * the previous block end if needed.
3487 */
3488 if (((end - len) / blocksize !=
3489 (end - 1) / blocksize))
3490 end = end - end % blocksize;
3491
3492 if (end > offset + len) {
3493 memmove(bucket_buf + end - len,
3494 bucket_buf + offset, len);
3495 xe->xe_name_offset = cpu_to_le16(end - len);
3496 }
3497
3498 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
3499 "bucket %llu\n", (unsigned long long)blkno);
3500
3501 end -= len;
3502 }
3503
3504 mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
3505 "bucket %llu\n", (unsigned long long)blkno);
3506
3507 if (xh_free_start == end)
85db90e7 3508 goto out;
01225596
TM		 3509
3510 memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
3511 xh->xh_free_start = cpu_to_le16(end);
3512
3513 /* sort the entries by their name_hash. */
3514 sort(entries, le16_to_cpu(xh->xh_count),
3515 sizeof(struct ocfs2_xattr_entry),
3516 cmp_xe, swap_xe);
3517
3518 buf = bucket_buf;
1c32a2fd
TM		 3519 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3520 memcpy(bucket_block(bucket, i), buf, blocksize);
3521 ocfs2_xattr_bucket_journal_dirty(handle, bucket);
01225596 3522
01225596 3523out:
01225596
TM		 3524 kfree(bucket_buf);
3525 return ret;
3526}
3527
3528/*
3529 * Move half nums of the xattr bucket in the previous cluster to this new
3530 * cluster. We only touch the last cluster of the previous extend record.
3531 *
3532 * first_bh is the first buffer_head of a series of bucket in the same
3533 * extent rec and header_bh is the header of one bucket in this cluster.
3534 * They will be updated if we move the data header_bh contains to the new
3535 * cluster. first_hash will be set as the 1st xe's name_hash of the new cluster.
3536 */
3537static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3538 handle_t *handle,
3539 struct buffer_head **first_bh,
3540 struct buffer_head **header_bh,
3541 u64 new_blkno,
3542 u64 prev_blkno,
3543 u32 num_clusters,
3544 u32 *first_hash)
3545{
3546 int i, ret, credits;
3547 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3548 int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3549 int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3550 int blocksize = inode->i_sb->s_blocksize;
3551 struct buffer_head *old_bh, *new_bh, *prev_bh, *new_first_bh = NULL;
3552 struct ocfs2_xattr_header *new_xh;
3553 struct ocfs2_xattr_header *xh =
3554 (struct ocfs2_xattr_header *)((*first_bh)->b_data);
3555
3556 BUG_ON(le16_to_cpu(xh->xh_num_buckets) < num_buckets);
3557 BUG_ON(OCFS2_XATTR_BUCKET_SIZE == osb->s_clustersize);
3558
3559 prev_bh = *first_bh;
3560 get_bh(prev_bh);
3561 xh = (struct ocfs2_xattr_header *)prev_bh->b_data;
3562
3563 prev_blkno += (num_clusters - 1) * bpc + bpc / 2;
3564
3565 mlog(0, "move half of xattrs in cluster %llu to %llu\n",
de29c085 3566 (unsigned long long)prev_blkno, (unsigned long long)new_blkno);
01225596
TM		 3567
3568 /*
3569 * We need to update the 1st half of the new cluster and
3570 * 1 more for the update of the 1st bucket of the previous
3571 * extent record.
3572 */
85db90e7 3573 credits = bpc / 2 + 1 + handle->h_buffer_credits;
01225596
TM		 3574 ret = ocfs2_extend_trans(handle, credits);
3575 if (ret) {
3576 mlog_errno(ret);
3577 goto out;
3578 }
3579
3580 ret = ocfs2_journal_access(handle, inode, prev_bh,
3581 OCFS2_JOURNAL_ACCESS_WRITE);
3582 if (ret) {
3583 mlog_errno(ret);
3584 goto out;
3585 }
3586
3587 for (i = 0; i < bpc / 2; i++, prev_blkno++, new_blkno++) {
3588 old_bh = new_bh = NULL;
3589 new_bh = sb_getblk(inode->i_sb, new_blkno);
3590 if (!new_bh) {
3591 ret = -EIO;
3592 mlog_errno(ret);
3593 goto out;
3594 }
3595
3596 ocfs2_set_new_buffer_uptodate(inode, new_bh);
3597
3598 ret = ocfs2_journal_access(handle, inode, new_bh,
3599 OCFS2_JOURNAL_ACCESS_CREATE);
3600 if (ret < 0) {
3601 mlog_errno(ret);
3602 brelse(new_bh);
3603 goto out;
3604 }
3605
970e4936 3606 ret = ocfs2_read_block(inode, prev_blkno, &old_bh, NULL);
01225596
TM		 3607 if (ret < 0) {
3608 mlog_errno(ret);
3609 brelse(new_bh);
3610 goto out;
3611 }
3612
3613 memcpy(new_bh->b_data, old_bh->b_data, blocksize);
3614
3615 if (i == 0) {
3616 new_xh = (struct ocfs2_xattr_header *)new_bh->b_data;
3617 new_xh->xh_num_buckets = cpu_to_le16(num_buckets / 2);
3618
3619 if (first_hash)
3620 *first_hash = le32_to_cpu(
3621 new_xh->xh_entries[0].xe_name_hash);
3622 new_first_bh = new_bh;
3623 get_bh(new_first_bh);
3624 }
3625
3626 ocfs2_journal_dirty(handle, new_bh);
3627
3628 if (*header_bh == old_bh) {
3629 brelse(*header_bh);
3630 *header_bh = new_bh;
3631 get_bh(*header_bh);
3632
3633 brelse(*first_bh);
3634 *first_bh = new_first_bh;
3635 get_bh(*first_bh);
3636 }
3637 brelse(new_bh);
3638 brelse(old_bh);
3639 }
3640
3641 le16_add_cpu(&xh->xh_num_buckets, -(num_buckets / 2));
3642
3643 ocfs2_journal_dirty(handle, prev_bh);
3644out:
3645 brelse(prev_bh);
3646 brelse(new_first_bh);
3647 return ret;
3648}
3649
01225596 3650/*
80bcaf34
TM		 3651 * Find the suitable pos when we divide a bucket into 2.
3652 * We have to make sure the xattrs with the same hash value exist
3653 * in the same bucket.
3654 *
3655 * If this ocfs2_xattr_header covers more than one hash value, find a
3656 * place where the hash value changes. Try to find the most even split.
3657 * The most common case is that all entries have different hash values,
3658 * and the first check we make will find a place to split.
3659 */
3660static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
3661{
3662 struct ocfs2_xattr_entry *entries = xh->xh_entries;
3663 int count = le16_to_cpu(xh->xh_count);
3664 int delta, middle = count / 2;
3665
3666 /*
3667 * We start at the middle. Each step gets farther away in both
3668 * directions. We therefore hit the change in hash value
3669 * nearest to the middle. Note that this loop does not execute for
3670 * count < 2.
3671 */
3672 for (delta = 0; delta < middle; delta++) {
3673 /* Let's check delta earlier than middle */
3674 if (cmp_xe(&entries[middle - delta - 1],
3675 &entries[middle - delta]))
3676 return middle - delta;
3677
3678 /* For even counts, don't walk off the end */
3679 if ((middle + delta + 1) == count)
3680 continue;
3681
3682 /* Now try delta past middle */
3683 if (cmp_xe(&entries[middle + delta],
3684 &entries[middle + delta + 1]))
3685 return middle + delta + 1;
3686 }
3687
3688 /* Every entry had the same hash */
3689 return count;
3690}
3691
3692/*
3693 * Move some xattrs in old bucket(blk) to new bucket(new_blk).
01225596 3694 * first_hash will record the 1st hash of the new bucket.
80bcaf34
TM		 3695 *
3696 * Normally half of the xattrs will be moved. But we have to make
3697 * sure that the xattrs with the same hash value are stored in the
3698 * same bucket. If all the xattrs in this bucket have the same hash
3699 * value, the new bucket will be initialized as an empty one and the
3700 * first_hash will be initialized as (hash_value+1).
01225596 3701 */
80bcaf34
TM		 3702static int ocfs2_divide_xattr_bucket(struct inode *inode,
3703 handle_t *handle,
3704 u64 blk,
3705 u64 new_blk,
3706 u32 *first_hash,
3707 int new_bucket_head)
01225596
TM		 3708{
3709 int ret, i;
80bcaf34 3710 int count, start, len, name_value_len = 0, xe_len, name_offset = 0;
ba937127 3711 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
01225596
TM		 3712 struct ocfs2_xattr_header *xh;
3713 struct ocfs2_xattr_entry *xe;
3714 int blocksize = inode->i_sb->s_blocksize;
3715
80bcaf34 3716 mlog(0, "move some of xattrs from bucket %llu to %llu\n",
de29c085 3717 (unsigned long long)blk, (unsigned long long)new_blk);
01225596 3718
ba937127
JB		 3719 s_bucket = ocfs2_xattr_bucket_new(inode);
3720 t_bucket = ocfs2_xattr_bucket_new(inode);
3721 if (!s_bucket || !t_bucket) {
3722 ret = -ENOMEM;
3723 mlog_errno(ret);
3724 goto out;
3725 }
01225596 3726
ba937127 3727 ret = ocfs2_read_xattr_bucket(s_bucket, blk);
01225596
TM		 3728 if (ret) {
3729 mlog_errno(ret);
3730 goto out;
3731 }
3732
ba937127 3733 ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
1224be02 3734 OCFS2_JOURNAL_ACCESS_WRITE);
01225596
TM		 3735 if (ret) {
3736 mlog_errno(ret);
3737 goto out;
3738 }
3739
784b816a
JB		 3740 /*
3741 * Even if !new_bucket_head, we're overwriting t_bucket. Thus,
3742 * there's no need to read it.
3743 */
ba937127 3744 ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
01225596
TM		 3745 if (ret) {
3746 mlog_errno(ret);
3747 goto out;
3748 }
3749
ba937127 3750 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
1224be02
JB		 3751 new_bucket_head ?
3752 OCFS2_JOURNAL_ACCESS_CREATE :
3753 OCFS2_JOURNAL_ACCESS_WRITE);
3754 if (ret) {
3755 mlog_errno(ret);
3756 goto out;
01225596
TM		 3757 }
3758
ba937127 3759 xh = bucket_xh(s_bucket);
80bcaf34
TM		 3760 count = le16_to_cpu(xh->xh_count);
3761 start = ocfs2_xattr_find_divide_pos(xh);
3762
3763 if (start == count) {
3764 xe = &xh->xh_entries[start-1];
3765
3766 /*
3767 * initialized a new empty bucket here.
3768 * The hash value is set as one larger than
3769 * that of the last entry in the previous bucket.
3770 */
ba937127
JB		 3771 for (i = 0; i < t_bucket->bu_blocks; i++)
3772 memset(bucket_block(t_bucket, i), 0, blocksize);
80bcaf34 3773
ba937127 3774 xh = bucket_xh(t_bucket);
80bcaf34
TM		 3775 xh->xh_free_start = cpu_to_le16(blocksize);
3776 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
3777 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
3778
3779 goto set_num_buckets;
3780 }
3781
01225596 3782 /* copy the whole bucket to the new first. */
ba937127 3783 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
01225596
TM		 3784
3785 /* update the new bucket. */
ba937127 3786 xh = bucket_xh(t_bucket);
01225596
TM		 3787
3788 /*
3789 * Calculate the total name/value len and xh_free_start for
3790 * the old bucket first.
3791 */
3792 name_offset = OCFS2_XATTR_BUCKET_SIZE;
3793 name_value_len = 0;
3794 for (i = 0; i < start; i++) {
3795 xe = &xh->xh_entries[i];
3796 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3797 if (ocfs2_xattr_is_local(xe))
3798 xe_len +=
3799 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3800 else
3801 xe_len += OCFS2_XATTR_ROOT_SIZE;
3802 name_value_len += xe_len;
3803 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3804 name_offset = le16_to_cpu(xe->xe_name_offset);
3805 }
3806
3807 /*
3808 * Now begin the modification to the new bucket.
3809 *
3810 * In the new bucket, We just move the xattr entry to the beginning
3811 * and don't touch the name/value. So there will be some holes in the
3812 * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3813 * called.
3814 */
3815 xe = &xh->xh_entries[start];
3816 len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3817 mlog(0, "mv xattr entry len %d from %d to %d\n", len,
ff1ec20e
MF		 3818 (int)((char *)xe - (char *)xh),
3819 (int)((char *)xh->xh_entries - (char *)xh));
01225596
TM		 3820 memmove((char *)xh->xh_entries, (char *)xe, len);
3821 xe = &xh->xh_entries[count - start];
3822 len = sizeof(struct ocfs2_xattr_entry) * start;
3823 memset((char *)xe, 0, len);
3824
3825 le16_add_cpu(&xh->xh_count, -start);
3826 le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3827
3828 /* Calculate xh_free_start for the new bucket. */
3829 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3830 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3831 xe = &xh->xh_entries[i];
3832 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3833 if (ocfs2_xattr_is_local(xe))
3834 xe_len +=
3835 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3836 else
3837 xe_len += OCFS2_XATTR_ROOT_SIZE;
3838 if (le16_to_cpu(xe->xe_name_offset) <
3839 le16_to_cpu(xh->xh_free_start))
3840 xh->xh_free_start = xe->xe_name_offset;
3841 }
3842
80bcaf34 3843set_num_buckets:
01225596
TM		 3844 /* set xh->xh_num_buckets for the new xh. */
3845 if (new_bucket_head)
3846 xh->xh_num_buckets = cpu_to_le16(1);
3847 else
3848 xh->xh_num_buckets = 0;
3849
ba937127 3850 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
01225596
TM		 3851
3852 /* store the first_hash of the new bucket. */
3853 if (first_hash)
3854 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3855
3856 /*
80bcaf34
TM		 3857 * Now only update the 1st block of the old bucket. If we
3858 * just added a new empty bucket, there is no need to modify
3859 * it.
01225596 3860 */
80bcaf34
TM		 3861 if (start == count)
3862 goto out;
3863
ba937127 3864 xh = bucket_xh(s_bucket);
01225596
TM		 3865 memset(&xh->xh_entries[start], 0,
3866 sizeof(struct ocfs2_xattr_entry) * (count - start));
3867 xh->xh_count = cpu_to_le16(start);
3868 xh->xh_free_start = cpu_to_le16(name_offset);
3869 xh->xh_name_value_len = cpu_to_le16(name_value_len);
3870
ba937127 3871 ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
01225596
TM		 3872
3873out:
ba937127
JB		 3874 ocfs2_xattr_bucket_free(s_bucket);
3875 ocfs2_xattr_bucket_free(t_bucket);
01225596
TM		 3876
3877 return ret;
3878}
3879
3880/*
3881 * Copy xattr from one bucket to another bucket.
3882 *
3883 * The caller must make sure that the journal transaction
3884 * has enough space for journaling.
3885 */
3886static int ocfs2_cp_xattr_bucket(struct inode *inode,
3887 handle_t *handle,
3888 u64 s_blkno,
3889 u64 t_blkno,
3890 int t_is_new)
3891{
4980c6da 3892 int ret;
ba937127 3893 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
01225596
TM		 3894
3895 BUG_ON(s_blkno == t_blkno);
3896
3897 mlog(0, "cp bucket %llu to %llu, target is %d\n",
de29c085
MF		 3898 (unsigned long long)s_blkno, (unsigned long long)t_blkno,
3899 t_is_new);
01225596 3900
ba937127
JB		 3901 s_bucket = ocfs2_xattr_bucket_new(inode);
3902 t_bucket = ocfs2_xattr_bucket_new(inode);
3903 if (!s_bucket || !t_bucket) {
3904 ret = -ENOMEM;
3905 mlog_errno(ret);
3906 goto out;
3907 }
3908
3909 ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
01225596
TM		 3910 if (ret)
3911 goto out;
3912
784b816a
JB		 3913 /*
3914 * Even if !t_is_new, we're overwriting t_bucket. Thus,
3915 * there's no need to read it.
3916 */
ba937127 3917 ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
01225596
TM		 3918 if (ret)
3919 goto out;
3920
ba937127 3921 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
1224be02
JB		 3922 t_is_new ?
3923 OCFS2_JOURNAL_ACCESS_CREATE :
3924 OCFS2_JOURNAL_ACCESS_WRITE);
3925 if (ret)
3926 goto out;
01225596 3927
ba937127
JB		 3928 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
3929 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
01225596
TM		 3930
3931out:
ba937127
JB		 3932 ocfs2_xattr_bucket_free(t_bucket);
3933 ocfs2_xattr_bucket_free(s_bucket);
01225596
TM		 3934
3935 return ret;
3936}
3937
3938/*
3939 * Copy one xattr cluster from src_blk to to_blk.
3940 * The to_blk will become the first bucket header of the cluster, so its
3941 * xh_num_buckets will be initialized as the bucket num in the cluster.
3942 */
3943static int ocfs2_cp_xattr_cluster(struct inode *inode,
3944 handle_t *handle,
3945 struct buffer_head *first_bh,
3946 u64 src_blk,
3947 u64 to_blk,
3948 u32 *first_hash)
3949{
3950 int i, ret, credits;
3951 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3952 int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3953 int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3954 struct buffer_head *bh = NULL;
3955 struct ocfs2_xattr_header *xh;
3956 u64 to_blk_start = to_blk;
3957
de29c085
MF		 3958 mlog(0, "cp xattrs from cluster %llu to %llu\n",
3959 (unsigned long long)src_blk, (unsigned long long)to_blk);
01225596
TM		 3960
3961 /*
3962 * We need to update the new cluster and 1 more for the update of
3963 * the 1st bucket of the previous extent rec.
3964 */
85db90e7 3965 credits = bpc + 1 + handle->h_buffer_credits;
01225596
TM		 3966 ret = ocfs2_extend_trans(handle, credits);
3967 if (ret) {
3968 mlog_errno(ret);
3969 goto out;
3970 }
3971
3972 ret = ocfs2_journal_access(handle, inode, first_bh,
3973 OCFS2_JOURNAL_ACCESS_WRITE);
3974 if (ret) {
3975 mlog_errno(ret);
3976 goto out;
3977 }
3978
3979 for (i = 0; i < num_buckets; i++) {
3980 ret = ocfs2_cp_xattr_bucket(inode, handle,
3981 src_blk, to_blk, 1);
3982 if (ret) {
3983 mlog_errno(ret);
3984 goto out;
3985 }
3986
3987 src_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3988 to_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3989 }
3990
3991 /* update the old bucket header. */
3992 xh = (struct ocfs2_xattr_header *)first_bh->b_data;
3993 le16_add_cpu(&xh->xh_num_buckets, -num_buckets);
3994
3995 ocfs2_journal_dirty(handle, first_bh);
3996
3997 /* update the new bucket header. */
970e4936 3998 ret = ocfs2_read_block(inode, to_blk_start, &bh, NULL);
01225596
TM		 3999 if (ret < 0) {
4000 mlog_errno(ret);
4001 goto out;
4002 }
4003
4004 ret = ocfs2_journal_access(handle, inode, bh,
4005 OCFS2_JOURNAL_ACCESS_WRITE);
4006 if (ret) {
4007 mlog_errno(ret);
4008 goto out;
4009 }
4010
4011 xh = (struct ocfs2_xattr_header *)bh->b_data;
4012 xh->xh_num_buckets = cpu_to_le16(num_buckets);
4013
4014 ocfs2_journal_dirty(handle, bh);
4015
4016 if (first_hash)
4017 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
4018out:
4019 brelse(bh);
4020 return ret;
4021}
4022
4023/*
80bcaf34 4024 * Move some xattrs in this cluster to the new cluster.
01225596
TM		 4025 * This function should only be called when bucket size == cluster size.
4026 * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4027 */
80bcaf34
TM		 4028static int ocfs2_divide_xattr_cluster(struct inode *inode,
4029 handle_t *handle,
4030 u64 prev_blk,
4031 u64 new_blk,
4032 u32 *first_hash)
01225596
TM		 4033{
4034 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
85db90e7 4035 int ret, credits = 2 * blk_per_bucket + handle->h_buffer_credits;
01225596
TM		 4036
4037 BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4038
4039 ret = ocfs2_extend_trans(handle, credits);
4040 if (ret) {
4041 mlog_errno(ret);
4042 return ret;
4043 }
4044
4045 /* Move half of the xattr in start_blk to the next bucket. */
80bcaf34
TM		 4046 return ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4047 new_blk, first_hash, 1);
01225596
TM		 4048}
4049
4050/*
4051 * Move some xattrs from the old cluster to the new one since they are not
4052 * contiguous in ocfs2 xattr tree.
4053 *
4054 * new_blk starts a new separate cluster, and we will move some xattrs from
4055 * prev_blk to it. v_start will be set as the first name hash value in this
4056 * new cluster so that it can be used as e_cpos during tree insertion and
4057 * don't collide with our original b-tree operations. first_bh and header_bh
4058 * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4059 * to extend the insert bucket.
4060 *
4061 * The problem is how much xattr should we move to the new one and when should
4062 * we update first_bh and header_bh?
4063 * 1. If cluster size > bucket size, that means the previous cluster has more
4064 * than 1 bucket, so just move half nums of bucket into the new cluster and
4065 * update the first_bh and header_bh if the insert bucket has been moved
4066 * to the new cluster.
4067 * 2. If cluster_size == bucket_size:
4068 * a) If the previous extent rec has more than one cluster and the insert
4069 * place isn't in the last cluster, copy the entire last cluster to the
4070 * new one. This time, we don't need to upate the first_bh and header_bh
4071 * since they will not be moved into the new cluster.
4072 * b) Otherwise, move the bottom half of the xattrs in the last cluster into
4073 * the new one. And we set the extend flag to zero if the insert place is
4074 * moved into the new allocated cluster since no extend is needed.
4075 */
4076static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
4077 handle_t *handle,
4078 struct buffer_head **first_bh,
4079 struct buffer_head **header_bh,
4080 u64 new_blk,
4081 u64 prev_blk,
4082 u32 prev_clusters,
4083 u32 *v_start,
4084 int *extend)
4085{
4086 int ret = 0;
4087 int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
4088
4089 mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
de29c085
MF		 4090 (unsigned long long)prev_blk, prev_clusters,
4091 (unsigned long long)new_blk);
01225596
TM		 4092
4093 if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1)
4094 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
4095 handle,
4096 first_bh,
4097 header_bh,
4098 new_blk,
4099 prev_blk,
4100 prev_clusters,
4101 v_start);
4102 else {
4103 u64 last_blk = prev_blk + bpc * (prev_clusters - 1);
4104
4105 if (prev_clusters > 1 && (*header_bh)->b_blocknr != last_blk)
4106 ret = ocfs2_cp_xattr_cluster(inode, handle, *first_bh,
4107 last_blk, new_blk,
4108 v_start);
4109 else {
80bcaf34
TM		 4110 ret = ocfs2_divide_xattr_cluster(inode, handle,
4111 last_blk, new_blk,
4112 v_start);
01225596
TM		 4113
4114 if ((*header_bh)->b_blocknr == last_blk && extend)
4115 *extend = 0;
4116 }
4117 }
4118
4119 return ret;
4120}
4121
4122/*
4123 * Add a new cluster for xattr storage.
4124 *
4125 * If the new cluster is contiguous with the previous one, it will be
4126 * appended to the same extent record, and num_clusters will be updated.
4127 * If not, we will insert a new extent for it and move some xattrs in
4128 * the last cluster into the new allocated one.
4129 * We also need to limit the maximum size of a btree leaf, otherwise we'll
4130 * lose the benefits of hashing because we'll have to search large leaves.
4131 * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
4132 * if it's bigger).
4133 *
4134 * first_bh is the first block of the previous extent rec and header_bh
4135 * indicates the bucket we will insert the new xattrs. They will be updated
4136 * when the header_bh is moved into the new cluster.
4137 */
4138static int ocfs2_add_new_xattr_cluster(struct inode *inode,
4139 struct buffer_head *root_bh,
4140 struct buffer_head **first_bh,
4141 struct buffer_head **header_bh,
4142 u32 *num_clusters,
4143 u32 prev_cpos,
4144 u64 prev_blkno,
78f30c31
TM		 4145 int *extend,
4146 struct ocfs2_xattr_set_ctxt *ctxt)
01225596 4147{
85db90e7 4148 int ret;
01225596
TM		 4149 u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
4150 u32 prev_clusters = *num_clusters;
4151 u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
4152 u64 block;
85db90e7 4153 handle_t *handle = ctxt->handle;
01225596 4154 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
f99b9b7c 4155 struct ocfs2_extent_tree et;
01225596
TM		 4156
4157 mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
4158 "previous xattr blkno = %llu\n",
4159 (unsigned long long)OCFS2_I(inode)->ip_blkno,
de29c085 4160 prev_cpos, (unsigned long long)prev_blkno);
01225596 4161
8d6220d6 4162 ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
f99b9b7c 4163
01225596
TM		 4164 ret = ocfs2_journal_access(handle, inode, root_bh,
4165 OCFS2_JOURNAL_ACCESS_WRITE);
4166 if (ret < 0) {
4167 mlog_errno(ret);
4168 goto leave;
4169 }
4170
78f30c31 4171 ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac, 1,
01225596
TM		 4172 clusters_to_add, &bit_off, &num_bits);
4173 if (ret < 0) {
4174 if (ret != -ENOSPC)
4175 mlog_errno(ret);
4176 goto leave;
4177 }
4178
4179 BUG_ON(num_bits > clusters_to_add);
4180
4181 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
4182 mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
4183 num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
4184
4185 if (prev_blkno + prev_clusters * bpc == block &&
4186 (prev_clusters + num_bits) << osb->s_clustersize_bits <=
4187 OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
4188 /*
4189 * If this cluster is contiguous with the old one and
4190 * adding this new cluster, we don't surpass the limit of
4191 * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
4192 * initialized and used like other buckets in the previous
4193 * cluster.
4194 * So add it as a contiguous one. The caller will handle
4195 * its init process.
4196 */
4197 v_start = prev_cpos + prev_clusters;
4198 *num_clusters = prev_clusters + num_bits;
4199 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
4200 num_bits);
4201 } else {
4202 ret = ocfs2_adjust_xattr_cross_cluster(inode,
4203 handle,
4204 first_bh,
4205 header_bh,
4206 block,
4207 prev_blkno,
4208 prev_clusters,
4209 &v_start,
4210 extend);
4211 if (ret) {
4212 mlog_errno(ret);
4213 goto leave;
4214 }
4215 }
4216
4217 mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
de29c085 4218 num_bits, (unsigned long long)block, v_start);
f99b9b7c 4219 ret = ocfs2_insert_extent(osb, handle, inode, &et, v_start, block,
78f30c31 4220 num_bits, 0, ctxt->meta_ac);
01225596
TM		 4221 if (ret < 0) {
4222 mlog_errno(ret);
4223 goto leave;
4224 }
4225
4226 ret = ocfs2_journal_dirty(handle, root_bh);
85db90e7 4227 if (ret < 0)
01225596 4228 mlog_errno(ret);
01225596
TM		 4229
4230leave:
01225596
TM		 4231 return ret;
4232}
4233
4234/*
4235 * Extend a new xattr bucket and move xattrs to the end one by one until
4236 * We meet with start_bh. Only move half of the xattrs to the bucket after it.
4237 */
4238static int ocfs2_extend_xattr_bucket(struct inode *inode,
85db90e7 4239 handle_t *handle,
01225596
TM		 4240 struct buffer_head *first_bh,
4241 struct buffer_head *start_bh,
4242 u32 num_clusters)
4243{
4244 int ret, credits;
4245 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4246 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4247 u64 start_blk = start_bh->b_blocknr, end_blk;
4248 u32 num_buckets = num_clusters * ocfs2_xattr_buckets_per_cluster(osb);
01225596
TM		 4249 struct ocfs2_xattr_header *first_xh =
4250 (struct ocfs2_xattr_header *)first_bh->b_data;
4251 u16 bucket = le16_to_cpu(first_xh->xh_num_buckets);
4252
4253 mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
de29c085 4254 "from %llu, len = %u\n", (unsigned long long)start_blk,
01225596
TM		 4255 (unsigned long long)first_bh->b_blocknr, num_clusters);
4256
4257 BUG_ON(bucket >= num_buckets);
4258
4259 end_blk = first_bh->b_blocknr + (bucket - 1) * blk_per_bucket;
4260
4261 /*
4262 * We will touch all the buckets after the start_bh(include it).
1224be02 4263 * Then we add one more bucket.
01225596 4264 */
85db90e7
TM		 4265 credits = end_blk - start_blk + 3 * blk_per_bucket + 1 +
4266 handle->h_buffer_credits;
4267 ret = ocfs2_extend_trans(handle, credits);
4268 if (ret) {
01225596
TM		 4269 mlog_errno(ret);
4270 goto out;
4271 }
4272
4273 ret = ocfs2_journal_access(handle, inode, first_bh,
4274 OCFS2_JOURNAL_ACCESS_WRITE);
4275 if (ret) {
4276 mlog_errno(ret);
85db90e7 4277 goto out;
01225596
TM		 4278 }
4279
4280 while (end_blk != start_blk) {
4281 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
4282 end_blk + blk_per_bucket, 0);
4283 if (ret)
85db90e7 4284 goto out;
01225596
TM		 4285 end_blk -= blk_per_bucket;
4286 }
4287
4288 /* Move half of the xattr in start_blk to the next bucket. */
80bcaf34
TM		 4289 ret = ocfs2_divide_xattr_bucket(inode, handle, start_blk,
4290 start_blk + blk_per_bucket, NULL, 0);
01225596
TM		 4291
4292 le16_add_cpu(&first_xh->xh_num_buckets, 1);
4293 ocfs2_journal_dirty(handle, first_bh);
4294
01225596
TM		 4295out:
4296 return ret;
4297}
4298
4299/*
4300 * Add new xattr bucket in an extent record and adjust the buckets accordingly.
4301 * xb_bh is the ocfs2_xattr_block.
4302 * We will move all the buckets starting from header_bh to the next place. As
4303 * for this one, half num of its xattrs will be moved to the next one.
4304 *
4305 * We will allocate a new cluster if current cluster is full and adjust
4306 * header_bh and first_bh if the insert place is moved to the new cluster.
4307 */
4308static int ocfs2_add_new_xattr_bucket(struct inode *inode,
4309 struct buffer_head *xb_bh,
78f30c31
TM		 4310 struct buffer_head *header_bh,
4311 struct ocfs2_xattr_set_ctxt *ctxt)
01225596
TM		 4312{
4313 struct ocfs2_xattr_header *first_xh = NULL;
4314 struct buffer_head *first_bh = NULL;
4315 struct ocfs2_xattr_block *xb =
4316 (struct ocfs2_xattr_block *)xb_bh->b_data;
4317 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
4318 struct ocfs2_extent_list *el = &xb_root->xt_list;
4319 struct ocfs2_xattr_header *xh =
4320 (struct ocfs2_xattr_header *)header_bh->b_data;
4321 u32 name_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
4322 struct super_block *sb = inode->i_sb;
4323 struct ocfs2_super *osb = OCFS2_SB(sb);
4324 int ret, num_buckets, extend = 1;
4325 u64 p_blkno;
4326 u32 e_cpos, num_clusters;
4327
4328 mlog(0, "Add new xattr bucket starting form %llu\n",
4329 (unsigned long long)header_bh->b_blocknr);
4330
4331 /*
4332 * Add refrence for header_bh here because it may be
4333 * changed in ocfs2_add_new_xattr_cluster and we need
4334 * to free it in the end.
4335 */
4336 get_bh(header_bh);
4337
4338 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
4339 &num_clusters, el);
4340 if (ret) {
4341 mlog_errno(ret);
4342 goto out;
4343 }
4344
970e4936 4345 ret = ocfs2_read_block(inode, p_blkno, &first_bh, NULL);
01225596
TM		 4346 if (ret) {
4347 mlog_errno(ret);
4348 goto out;
4349 }
4350
4351 num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
4352 first_xh = (struct ocfs2_xattr_header *)first_bh->b_data;
4353
4354 if (num_buckets == le16_to_cpu(first_xh->xh_num_buckets)) {
4355 ret = ocfs2_add_new_xattr_cluster(inode,
4356 xb_bh,
4357 &first_bh,
4358 &header_bh,
4359 &num_clusters,
4360 e_cpos,
4361 p_blkno,
78f30c31
TM		 4362 &extend,
4363 ctxt);
01225596
TM		 4364 if (ret) {
4365 mlog_errno(ret);
4366 goto out;
4367 }
4368 }
4369
4370 if (extend)
4371 ret = ocfs2_extend_xattr_bucket(inode,
85db90e7 4372 ctxt->handle,
01225596
TM		 4373 first_bh,
4374 header_bh,
4375 num_clusters);
4376 if (ret)
4377 mlog_errno(ret);
4378out:
4379 brelse(first_bh);
4380 brelse(header_bh);
4381 return ret;
4382}
4383
4384static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
4385 struct ocfs2_xattr_bucket *bucket,
4386 int offs)
4387{
4388 int block_off = offs >> inode->i_sb->s_blocksize_bits;
4389
4390 offs = offs % inode->i_sb->s_blocksize;
51def39f 4391 return bucket_block(bucket, block_off) + offs;
01225596
TM		 4392}
4393
4394/*
4395 * Handle the normal xattr set, including replace, delete and new.
01225596
TM		 4396 *
4397 * Note: "local" indicates the real data's locality. So we can't
4398 * just its bucket locality by its length.
4399 */
4400static void ocfs2_xattr_set_entry_normal(struct inode *inode,
4401 struct ocfs2_xattr_info *xi,
4402 struct ocfs2_xattr_search *xs,
4403 u32 name_hash,
5a095611 4404 int local)
01225596
TM		 4405{
4406 struct ocfs2_xattr_entry *last, *xe;
4407 int name_len = strlen(xi->name);
4408 struct ocfs2_xattr_header *xh = xs->header;
4409 u16 count = le16_to_cpu(xh->xh_count), start;
4410 size_t blocksize = inode->i_sb->s_blocksize;
4411 char *val;
4412 size_t offs, size, new_size;
4413
4414 last = &xh->xh_entries[count];
4415 if (!xs->not_found) {
4416 xe = xs->here;
4417 offs = le16_to_cpu(xe->xe_name_offset);
4418 if (ocfs2_xattr_is_local(xe))
4419 size = OCFS2_XATTR_SIZE(name_len) +
4420 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4421 else
4422 size = OCFS2_XATTR_SIZE(name_len) +
4423 OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4424
4425 /*
4426 * If the new value will be stored outside, xi->value has been
4427 * initalized as an empty ocfs2_xattr_value_root, and the same
4428 * goes with xi->value_len, so we can set new_size safely here.
4429 * See ocfs2_xattr_set_in_bucket.
4430 */
4431 new_size = OCFS2_XATTR_SIZE(name_len) +
4432 OCFS2_XATTR_SIZE(xi->value_len);
4433
4434 le16_add_cpu(&xh->xh_name_value_len, -size);
4435 if (xi->value) {
4436 if (new_size > size)
4437 goto set_new_name_value;
4438
4439 /* Now replace the old value with new one. */
4440 if (local)
4441 xe->xe_value_size = cpu_to_le64(xi->value_len);
4442 else
4443 xe->xe_value_size = 0;
4444
4445 val = ocfs2_xattr_bucket_get_val(inode,
ba937127 4446 xs->bucket, offs);
01225596
TM		 4447 memset(val + OCFS2_XATTR_SIZE(name_len), 0,
4448 size - OCFS2_XATTR_SIZE(name_len));
4449 if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
4450 memcpy(val + OCFS2_XATTR_SIZE(name_len),
4451 xi->value, xi->value_len);
4452
4453 le16_add_cpu(&xh->xh_name_value_len, new_size);
4454 ocfs2_xattr_set_local(xe, local);
4455 return;
4456 } else {
5a095611
TM		 4457 /*
4458 * Remove the old entry if there is more than one.
4459 * We don't remove the last entry so that we can
4460 * use it to indicate the hash value of the empty
4461 * bucket.
4462 */
01225596 4463 last -= 1;
01225596 4464 le16_add_cpu(&xh->xh_count, -1);
5a095611
TM		 4465 if (xh->xh_count) {
4466 memmove(xe, xe + 1,
4467 (void *)last - (void *)xe);
4468 memset(last, 0,
4469 sizeof(struct ocfs2_xattr_entry));
4470 } else
4471 xh->xh_free_start =
4472 cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4473
01225596
TM		 4474 return;
4475 }
4476 } else {
4477 /* find a new entry for insert. */
4478 int low = 0, high = count - 1, tmp;
4479 struct ocfs2_xattr_entry *tmp_xe;
4480
5a095611 4481 while (low <= high && count) {
01225596
TM		 4482 tmp = (low + high) / 2;
4483 tmp_xe = &xh->xh_entries[tmp];
4484
4485 if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
4486 low = tmp + 1;
4487 else if (name_hash <
4488 le32_to_cpu(tmp_xe->xe_name_hash))
4489 high = tmp - 1;
06b240d8
TM		 4490 else {
4491 low = tmp;
01225596 4492 break;
06b240d8 4493 }
01225596
TM		 4494 }
4495
4496 xe = &xh->xh_entries[low];
4497 if (low != count)
4498 memmove(xe + 1, xe, (void *)last - (void *)xe);
4499
4500 le16_add_cpu(&xh->xh_count, 1);
4501 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
4502 xe->xe_name_hash = cpu_to_le32(name_hash);
4503 xe->xe_name_len = name_len;
4504 ocfs2_xattr_set_type(xe, xi->name_index);
4505 }
4506
4507set_new_name_value:
4508 /* Insert the new name+value. */
4509 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4510
4511 /*
4512 * We must make sure that the name/value pair
4513 * exists in the same block.
4514 */
4515 offs = le16_to_cpu(xh->xh_free_start);
4516 start = offs - size;
4517
4518 if (start >> inode->i_sb->s_blocksize_bits !=
4519 (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4520 offs = offs - offs % blocksize;
4521 xh->xh_free_start = cpu_to_le16(offs);
4522 }
4523
ba937127 4524 val = ocfs2_xattr_bucket_get_val(inode, xs->bucket, offs - size);
01225596
TM		 4525 xe->xe_name_offset = cpu_to_le16(offs - size);
4526
4527 memset(val, 0, size);
4528 memcpy(val, xi->name, name_len);
4529 memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4530
4531 xe->xe_value_size = cpu_to_le64(xi->value_len);
4532 ocfs2_xattr_set_local(xe, local);
4533 xs->here = xe;
4534 le16_add_cpu(&xh->xh_free_start, -size);
4535 le16_add_cpu(&xh->xh_name_value_len, size);
4536
4537 return;
4538}
4539
01225596
TM		 4540/*
4541 * Set the xattr entry in the specified bucket.
4542 * The bucket is indicated by xs->bucket and it should have the enough
4543 * space for the xattr insertion.
4544 */
4545static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
85db90e7 4546 handle_t *handle,
01225596
TM		 4547 struct ocfs2_xattr_info *xi,
4548 struct ocfs2_xattr_search *xs,
4549 u32 name_hash,
5a095611 4550 int local)
01225596 4551{
1224be02 4552 int ret;
02dbf38d 4553 u64 blkno;
01225596 4554
ff1ec20e
MF		 4555 mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4556 (unsigned long)xi->value_len, xi->name_index,
ba937127 4557 (unsigned long long)bucket_blkno(xs->bucket));
01225596 4558
ba937127 4559 if (!xs->bucket->bu_bhs[1]) {
02dbf38d
JB		 4560 blkno = bucket_blkno(xs->bucket);
4561 ocfs2_xattr_bucket_relse(xs->bucket);
4562 ret = ocfs2_read_xattr_bucket(xs->bucket, blkno);
01225596
TM		 4563 if (ret) {
4564 mlog_errno(ret);
4565 goto out;
4566 }
4567 }
4568
ba937127 4569 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
1224be02
JB		 4570 OCFS2_JOURNAL_ACCESS_WRITE);
4571 if (ret < 0) {
4572 mlog_errno(ret);
4573 goto out;
01225596
TM		 4574 }
4575
5a095611 4576 ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local);
ba937127 4577 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
01225596 4578
01225596 4579out:
01225596
TM		 4580 return ret;
4581}
4582
01225596
TM		 4583/*
4584 * Truncate the specified xe_off entry in xattr bucket.
4585 * bucket is indicated by header_bh and len is the new length.
4586 * Both the ocfs2_xattr_value_root and the entry will be updated here.
4587 *
4588 * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4589 */
4590static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
548b0f22 4591 struct ocfs2_xattr_bucket *bucket,
01225596 4592 int xe_off,
78f30c31
TM		 4593 int len,
4594 struct ocfs2_xattr_set_ctxt *ctxt)
01225596
TM		 4595{
4596 int ret, offset;
4597 u64 value_blk;
4598 struct buffer_head *value_bh = NULL;
4599 struct ocfs2_xattr_value_root *xv;
4600 struct ocfs2_xattr_entry *xe;
548b0f22 4601 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
01225596
TM		 4602 size_t blocksize = inode->i_sb->s_blocksize;
4603
4604 xe = &xh->xh_entries[xe_off];
4605
4606 BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4607
4608 offset = le16_to_cpu(xe->xe_name_offset) +
4609 OCFS2_XATTR_SIZE(xe->xe_name_len);
4610
4611 value_blk = offset / blocksize;
4612
4613 /* We don't allow ocfs2_xattr_value to be stored in different block. */
4614 BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
01225596 4615
548b0f22
JB		 4616 value_bh = bucket->bu_bhs[value_blk];
4617 BUG_ON(!value_bh);
01225596
TM		 4618
4619 xv = (struct ocfs2_xattr_value_root *)
4620 (value_bh->b_data + offset % blocksize);
4621
548b0f22
JB		 4622 ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
4623 OCFS2_JOURNAL_ACCESS_WRITE);
01225596
TM		 4624 if (ret) {
4625 mlog_errno(ret);
4626 goto out;
4627 }
4628
548b0f22
JB		 4629 /*
4630 * From here on out we have to dirty the bucket. The generic
4631 * value calls only modify one of the bucket's bhs, but we need
4632 * to send the bucket at once. So if they error, they *could* have
4633 * modified something. We have to assume they did, and dirty
4634 * the whole bucket. This leaves us in a consistent state.
4635 */
4636 mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4637 xe_off, (unsigned long long)bucket_blkno(bucket), len);
4638 ret = ocfs2_xattr_value_truncate(inode, value_bh, xv, len, ctxt);
01225596
TM		 4639 if (ret) {
4640 mlog_errno(ret);
548b0f22 4641 goto out_dirty;
01225596
TM		 4642 }
4643
548b0f22
JB		 4644 xe->xe_value_size = cpu_to_le64(len);
4645
4646out_dirty:
4647 ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
4648
01225596 4649out:
01225596
TM		 4650 return ret;
4651}
4652
4653static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
78f30c31
TM		 4654 struct ocfs2_xattr_search *xs,
4655 int len,
4656 struct ocfs2_xattr_set_ctxt *ctxt)
01225596
TM		 4657{
4658 int ret, offset;
4659 struct ocfs2_xattr_entry *xe = xs->here;
4660 struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4661
ba937127 4662 BUG_ON(!xs->bucket->bu_bhs[0] || !xe || ocfs2_xattr_is_local(xe));
01225596
TM		 4663
4664 offset = xe - xh->xh_entries;
548b0f22 4665 ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket,
78f30c31 4666 offset, len, ctxt);
01225596
TM		 4667 if (ret)
4668 mlog_errno(ret);
4669
4670 return ret;
4671}
4672
4673static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
85db90e7 4674 handle_t *handle,
01225596
TM		 4675 struct ocfs2_xattr_search *xs,
4676 char *val,
4677 int value_len)
4678{
4679 int offset;
4680 struct ocfs2_xattr_value_root *xv;
4681 struct ocfs2_xattr_entry *xe = xs->here;
4682
4683 BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4684
4685 offset = le16_to_cpu(xe->xe_name_offset) +
4686 OCFS2_XATTR_SIZE(xe->xe_name_len);
4687
4688 xv = (struct ocfs2_xattr_value_root *)(xs->base + offset);
4689
85db90e7
TM		 4690 return __ocfs2_xattr_set_value_outside(inode, handle,
4691 xv, val, value_len);
01225596
TM		 4692}
4693
01225596
TM		 4694static int ocfs2_rm_xattr_cluster(struct inode *inode,
4695 struct buffer_head *root_bh,
4696 u64 blkno,
4697 u32 cpos,
4698 u32 len)
4699{
4700 int ret;
4701 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4702 struct inode *tl_inode = osb->osb_tl_inode;
4703 handle_t *handle;
4704 struct ocfs2_xattr_block *xb =
4705 (struct ocfs2_xattr_block *)root_bh->b_data;
01225596
TM		 4706 struct ocfs2_alloc_context *meta_ac = NULL;
4707 struct ocfs2_cached_dealloc_ctxt dealloc;
f99b9b7c
JB		 4708 struct ocfs2_extent_tree et;
4709
8d6220d6 4710 ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
01225596
TM		 4711
4712 ocfs2_init_dealloc_ctxt(&dealloc);
4713
4714 mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4715 cpos, len, (unsigned long long)blkno);
4716
4717 ocfs2_remove_xattr_clusters_from_cache(inode, blkno, len);
4718
f99b9b7c 4719 ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
01225596
TM		 4720 if (ret) {
4721 mlog_errno(ret);
4722 return ret;
4723 }
4724
4725 mutex_lock(&tl_inode->i_mutex);
4726
4727 if (ocfs2_truncate_log_needs_flush(osb)) {
4728 ret = __ocfs2_flush_truncate_log(osb);
4729 if (ret < 0) {
4730 mlog_errno(ret);
4731 goto out;
4732 }
4733 }
4734
a90714c1 4735 handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
d3264799 4736 if (IS_ERR(handle)) {
01225596
TM		 4737 ret = -ENOMEM;
4738 mlog_errno(ret);
4739 goto out;
4740 }
4741
4742 ret = ocfs2_journal_access(handle, inode, root_bh,
4743 OCFS2_JOURNAL_ACCESS_WRITE);
4744 if (ret) {
4745 mlog_errno(ret);
4746 goto out_commit;
4747 }
4748
f99b9b7c
JB		 4749 ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
4750 &dealloc);
01225596
TM		 4751 if (ret) {
4752 mlog_errno(ret);
4753 goto out_commit;
4754 }
4755
4756 le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4757
4758 ret = ocfs2_journal_dirty(handle, root_bh);
4759 if (ret) {
4760 mlog_errno(ret);
4761 goto out_commit;
4762 }
4763
4764 ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4765 if (ret)
4766 mlog_errno(ret);
4767
4768out_commit:
4769 ocfs2_commit_trans(osb, handle);
4770out:
4771 ocfs2_schedule_truncate_log_flush(osb, 1);
4772
4773 mutex_unlock(&tl_inode->i_mutex);
4774
4775 if (meta_ac)
4776 ocfs2_free_alloc_context(meta_ac);
4777
4778 ocfs2_run_deallocs(osb, &dealloc);
4779
4780 return ret;
4781}
4782
01225596 4783static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
85db90e7 4784 handle_t *handle,
01225596
TM		 4785 struct ocfs2_xattr_search *xs)
4786{
ba937127 4787 struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
01225596
TM		 4788 struct ocfs2_xattr_entry *last = &xh->xh_entries[
4789 le16_to_cpu(xh->xh_count) - 1];
4790 int ret = 0;
4791
ba937127 4792 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
1224be02 4793 OCFS2_JOURNAL_ACCESS_WRITE);
01225596
TM		 4794 if (ret) {
4795 mlog_errno(ret);
85db90e7 4796 return;
01225596
TM		 4797 }
4798
4799 /* Remove the old entry. */
4800 memmove(xs->here, xs->here + 1,
4801 (void *)last - (void *)xs->here);
4802 memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4803 le16_add_cpu(&xh->xh_count, -1);
4804
ba937127 4805 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
01225596
TM		 4806}
4807
4808/*
4809 * Set the xattr name/value in the bucket specified in xs.
4810 *
4811 * As the new value in xi may be stored in the bucket or in an outside cluster,
4812 * we divide the whole process into 3 steps:
4813 * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4814 * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4815 * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4816 * 4. If the clusters for the new outside value can't be allocated, we need
4817 * to free the xattr we allocated in set.
4818 */
4819static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4820 struct ocfs2_xattr_info *xi,
78f30c31
TM		 4821 struct ocfs2_xattr_search *xs,
4822 struct ocfs2_xattr_set_ctxt *ctxt)
01225596 4823{
5a095611 4824 int ret, local = 1;
01225596
TM		 4825 size_t value_len;
4826 char *val = (char *)xi->value;
4827 struct ocfs2_xattr_entry *xe = xs->here;
2057e5c6
TM		 4828 u32 name_hash = ocfs2_xattr_name_hash(inode, xi->name,
4829 strlen(xi->name));
01225596
TM		 4830
4831 if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
4832 /*
4833 * We need to truncate the xattr storage first.
4834 *
4835 * If both the old and new value are stored to
4836 * outside block, we only need to truncate
4837 * the storage and then set the value outside.
4838 *
4839 * If the new value should be stored within block,
4840 * we should free all the outside block first and
4841 * the modification to the xattr block will be done
4842 * by following steps.
4843 */
4844 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4845 value_len = xi->value_len;
4846 else
4847 value_len = 0;
4848
4849 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
78f30c31
TM		 4850 value_len,
4851 ctxt);
01225596
TM		 4852 if (ret)
4853 goto out;
4854
4855 if (value_len)
4856 goto set_value_outside;
4857 }
4858
4859 value_len = xi->value_len;
4860 /* So we have to handle the inside block change now. */
4861 if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4862 /*
4863 * If the new value will be stored outside of block,
4864 * initalize a new empty value root and insert it first.
4865 */
4866 local = 0;
4867 xi->value = &def_xv;
4868 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
4869 }
4870
85db90e7
TM		 4871 ret = ocfs2_xattr_set_entry_in_bucket(inode, ctxt->handle, xi, xs,
4872 name_hash, local);
01225596
TM		 4873 if (ret) {
4874 mlog_errno(ret);
4875 goto out;
4876 }
4877
5a095611
TM		 4878 if (value_len <= OCFS2_XATTR_INLINE_SIZE)
4879 goto out;
01225596 4880
5a095611
TM		 4881 /* allocate the space now for the outside block storage. */
4882 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
78f30c31 4883 value_len, ctxt);
5a095611
TM		 4884 if (ret) {
4885 mlog_errno(ret);
4886
4887 if (xs->not_found) {
4888 /*
4889 * We can't allocate enough clusters for outside
4890 * storage and we have allocated xattr already,
4891 * so need to remove it.
4892 */
85db90e7 4893 ocfs2_xattr_bucket_remove_xs(inode, ctxt->handle, xs);
01225596 4894 }
01225596
TM		 4895 goto out;
4896 }
4897
4898set_value_outside:
85db90e7
TM		 4899 ret = ocfs2_xattr_bucket_set_value_outside(inode, ctxt->handle,
4900 xs, val, value_len);
01225596
TM		 4901out:
4902 return ret;
4903}
4904
80bcaf34
TM		 4905/*
4906 * check whether the xattr bucket is filled up with the same hash value.
4907 * If we want to insert the xattr with the same hash, return -ENOSPC.
4908 * If we want to insert a xattr with different hash value, go ahead
4909 * and ocfs2_divide_xattr_bucket will handle this.
4910 */
01225596 4911static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
80bcaf34
TM		 4912 struct ocfs2_xattr_bucket *bucket,
4913 const char *name)
01225596 4914{
3e632946 4915 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
80bcaf34
TM		 4916 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
4917
4918 if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
4919 return 0;
01225596
TM		 4920
4921 if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
4922 xh->xh_entries[0].xe_name_hash) {
4923 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
4924 "hash = %u\n",
9c7759aa 4925 (unsigned long long)bucket_blkno(bucket),
01225596
TM		 4926 le32_to_cpu(xh->xh_entries[0].xe_name_hash));
4927 return -ENOSPC;
4928 }
4929
4930 return 0;
4931}
4932
4933static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
4934 struct ocfs2_xattr_info *xi,
78f30c31
TM		 4935 struct ocfs2_xattr_search *xs,
4936 struct ocfs2_xattr_set_ctxt *ctxt)
01225596
TM		 4937{
4938 struct ocfs2_xattr_header *xh;
4939 struct ocfs2_xattr_entry *xe;
4940 u16 count, header_size, xh_free_start;
6dde41d9 4941 int free, max_free, need, old;
01225596
TM		 4942 size_t value_size = 0, name_len = strlen(xi->name);
4943 size_t blocksize = inode->i_sb->s_blocksize;
4944 int ret, allocation = 0;
01225596
TM		 4945
4946 mlog_entry("Set xattr %s in xattr index block\n", xi->name);
4947
4948try_again:
4949 xh = xs->header;
4950 count = le16_to_cpu(xh->xh_count);
4951 xh_free_start = le16_to_cpu(xh->xh_free_start);
4952 header_size = sizeof(struct ocfs2_xattr_header) +
4953 count * sizeof(struct ocfs2_xattr_entry);
4954 max_free = OCFS2_XATTR_BUCKET_SIZE -
4955 le16_to_cpu(xh->xh_name_value_len) - header_size;
4956
4957 mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
4958 "of %u which exceed block size\n",
ba937127 4959 (unsigned long long)bucket_blkno(xs->bucket),
01225596
TM		 4960 header_size);
4961
4962 if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4963 value_size = OCFS2_XATTR_ROOT_SIZE;
4964 else if (xi->value)
4965 value_size = OCFS2_XATTR_SIZE(xi->value_len);
4966
4967 if (xs->not_found)
4968 need = sizeof(struct ocfs2_xattr_entry) +
4969 OCFS2_XATTR_SIZE(name_len) + value_size;
4970 else {
4971 need = value_size + OCFS2_XATTR_SIZE(name_len);
4972
4973 /*
4974 * We only replace the old value if the new length is smaller
4975 * than the old one. Otherwise we will allocate new space in the
4976 * bucket to store it.
4977 */
4978 xe = xs->here;
4979 if (ocfs2_xattr_is_local(xe))
4980 old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4981 else
4982 old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4983
4984 if (old >= value_size)
4985 need = 0;
4986 }
4987
4988 free = xh_free_start - header_size;
4989 /*
4990 * We need to make sure the new name/value pair
4991 * can exist in the same block.
4992 */
4993 if (xh_free_start % blocksize < need)
4994 free -= xh_free_start % blocksize;
4995
4996 mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
4997 "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
4998 " %u\n", xs->not_found,
ba937127 4999 (unsigned long long)bucket_blkno(xs->bucket),
01225596
TM		 5000 free, need, max_free, le16_to_cpu(xh->xh_free_start),
5001 le16_to_cpu(xh->xh_name_value_len));
5002
976331d8
TM		 5003 if (free < need ||
5004 (xs->not_found &&
5005 count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb))) {
01225596
TM		 5006 if (need <= max_free &&
5007 count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
5008 /*
5009 * We can create the space by defragment. Since only the
5010 * name/value will be moved, the xe shouldn't be changed
5011 * in xs.
5012 */
85db90e7
TM		 5013 ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5014 xs->bucket);
01225596
TM		 5015 if (ret) {
5016 mlog_errno(ret);
5017 goto out;
5018 }
5019
5020 xh_free_start = le16_to_cpu(xh->xh_free_start);
5021 free = xh_free_start - header_size;
5022 if (xh_free_start % blocksize < need)
5023 free -= xh_free_start % blocksize;
5024
5025 if (free >= need)
5026 goto xattr_set;
5027
5028 mlog(0, "Can't get enough space for xattr insert by "
5029 "defragment. Need %u bytes, but we have %d, so "
5030 "allocate new bucket for it.\n", need, free);
5031 }
5032
5033 /*
5034 * We have to add new buckets or clusters and one
5035 * allocation should leave us enough space for insert.
5036 */
5037 BUG_ON(allocation);
5038
5039 /*
5040 * We do not allow for overlapping ranges between buckets. And
5041 * the maximum number of collisions we will allow for then is
5042 * one bucket's worth, so check it here whether we need to
5043 * add a new bucket for the insert.
5044 */
80bcaf34 5045 ret = ocfs2_check_xattr_bucket_collision(inode,
ba937127 5046 xs->bucket,
80bcaf34 5047 xi->name);
01225596
TM		 5048 if (ret) {
5049 mlog_errno(ret);
5050 goto out;
5051 }
5052
5053 ret = ocfs2_add_new_xattr_bucket(inode,
5054 xs->xattr_bh,
78f30c31
TM		 5055 xs->bucket->bu_bhs[0],
5056 ctxt);
01225596
TM		 5057 if (ret) {
5058 mlog_errno(ret);
5059 goto out;
5060 }
5061
ba937127 5062 ocfs2_xattr_bucket_relse(xs->bucket);
01225596
TM		 5063
5064 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5065 xi->name_index,
5066 xi->name, xs);
5067 if (ret && ret != -ENODATA)
5068 goto out;
5069 xs->not_found = ret;
5070 allocation = 1;
5071 goto try_again;
5072 }
5073
5074xattr_set:
78f30c31 5075 ret = ocfs2_xattr_set_in_bucket(inode, xi, xs, ctxt);
01225596
TM		 5076out:
5077 mlog_exit(ret);
5078 return ret;
5079}
a3944256
TM		 5080
5081static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5082 struct ocfs2_xattr_bucket *bucket,
5083 void *para)
5084{
5085 int ret = 0;
3e632946 5086 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
a3944256
TM		 5087 u16 i;
5088 struct ocfs2_xattr_entry *xe;
78f30c31
TM		 5089 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5090 struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
548b0f22
JB		 5091 int credits = ocfs2_remove_extent_credits(osb->sb) +
5092 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5093
78f30c31
TM		 5094
5095 ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
a3944256 5096
548b0f22 5097 ctxt.handle = ocfs2_start_trans(osb, credits);
85db90e7
TM		 5098 if (IS_ERR(ctxt.handle)) {
5099 ret = PTR_ERR(ctxt.handle);
5100 mlog_errno(ret);
5101 goto out;
5102 }
5103
a3944256
TM		 5104 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5105 xe = &xh->xh_entries[i];
5106 if (ocfs2_xattr_is_local(xe))
5107 continue;
5108
548b0f22 5109 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
78f30c31 5110 i, 0, &ctxt);
a3944256
TM		 5111 if (ret) {
5112 mlog_errno(ret);
5113 break;
5114 }
5115 }
5116
85db90e7 5117 ret = ocfs2_commit_trans(osb, ctxt.handle);
78f30c31
TM		 5118 ocfs2_schedule_truncate_log_flush(osb, 1);
5119 ocfs2_run_deallocs(osb, &ctxt.dealloc);
85db90e7 5120out:
a3944256
TM		 5121 return ret;
5122}
5123
5124static int ocfs2_delete_xattr_index_block(struct inode *inode,
5125 struct buffer_head *xb_bh)
5126{
5127 struct ocfs2_xattr_block *xb =
5128 (struct ocfs2_xattr_block *)xb_bh->b_data;
5129 struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
5130 int ret = 0;
5131 u32 name_hash = UINT_MAX, e_cpos, num_clusters;
5132 u64 p_blkno;
5133
5134 if (le16_to_cpu(el->l_next_free_rec) == 0)
5135 return 0;
5136
5137 while (name_hash > 0) {
5138 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
5139 &e_cpos, &num_clusters, el);
5140 if (ret) {
5141 mlog_errno(ret);
5142 goto out;
5143 }
5144
5145 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
5146 ocfs2_delete_xattr_in_bucket,
5147 NULL);
5148 if (ret) {
5149 mlog_errno(ret);
5150 goto out;
5151 }
5152
5153 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
5154 p_blkno, e_cpos, num_clusters);
5155 if (ret) {
5156 mlog_errno(ret);
5157 break;
5158 }
5159
5160 if (e_cpos == 0)
5161 break;
5162
5163 name_hash = e_cpos - 1;
5164 }
5165
5166out:
5167 return ret;
5168}
99219aea 5169
923f7f31
TY		 5170/*
5171 * 'security' attributes support
5172 */
5173static size_t ocfs2_xattr_security_list(struct inode *inode, char *list,
5174 size_t list_size, const char *name,
5175 size_t name_len)
5176{
5177 const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
5178 const size_t total_len = prefix_len + name_len + 1;
5179
5180 if (list && total_len <= list_size) {
5181 memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
5182 memcpy(list + prefix_len, name, name_len);
5183 list[prefix_len + name_len] = '\0';
5184 }
5185 return total_len;
5186}
5187
5188static int ocfs2_xattr_security_get(struct inode *inode, const char *name,
5189 void *buffer, size_t size)
5190{
5191 if (strcmp(name, "") == 0)
5192 return -EINVAL;
5193 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_SECURITY, name,
5194 buffer, size);
5195}
5196
5197static int ocfs2_xattr_security_set(struct inode *inode, const char *name,
5198 const void *value, size_t size, int flags)
5199{
5200 if (strcmp(name, "") == 0)
5201 return -EINVAL;
5202
5203 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY, name, value,
5204 size, flags);
5205}
5206
534eaddd
TY		 5207int ocfs2_init_security_get(struct inode *inode,
5208 struct inode *dir,
5209 struct ocfs2_security_xattr_info *si)
5210{
5211 return security_inode_init_security(inode, dir, &si->name, &si->value,
5212 &si->value_len);
5213}
5214
5215int ocfs2_init_security_set(handle_t *handle,
5216 struct inode *inode,
5217 struct buffer_head *di_bh,
5218 struct ocfs2_security_xattr_info *si,
5219 struct ocfs2_alloc_context *xattr_ac,
5220 struct ocfs2_alloc_context *data_ac)
5221{
5222 return ocfs2_xattr_set_handle(handle, inode, di_bh,
5223 OCFS2_XATTR_INDEX_SECURITY,
5224 si->name, si->value, si->value_len, 0,
5225 xattr_ac, data_ac);
5226}
5227
923f7f31
TY		 5228struct xattr_handler ocfs2_xattr_security_handler = {
5229 .prefix = XATTR_SECURITY_PREFIX,
5230 .list = ocfs2_xattr_security_list,
5231 .get = ocfs2_xattr_security_get,
5232 .set = ocfs2_xattr_security_set,
5233};
5234
99219aea
MF		 5235/*
5236 * 'trusted' attributes support
5237 */
99219aea
MF		 5238static size_t ocfs2_xattr_trusted_list(struct inode *inode, char *list,
5239 size_t list_size, const char *name,
5240 size_t name_len)
5241{
ceb1eba3 5242 const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
99219aea
MF		 5243 const size_t total_len = prefix_len + name_len + 1;
5244
5245 if (list && total_len <= list_size) {
5246 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
5247 memcpy(list + prefix_len, name, name_len);
5248 list[prefix_len + name_len] = '\0';
5249 }
5250 return total_len;
5251}
5252
5253static int ocfs2_xattr_trusted_get(struct inode *inode, const char *name,
5254 void *buffer, size_t size)
5255{
5256 if (strcmp(name, "") == 0)
5257 return -EINVAL;
5258 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED, name,
5259 buffer, size);
5260}
5261
5262static int ocfs2_xattr_trusted_set(struct inode *inode, const char *name,
5263 const void *value, size_t size, int flags)
5264{
5265 if (strcmp(name, "") == 0)
5266 return -EINVAL;
5267
5268 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED, name, value,
5269 size, flags);
5270}
5271
5272struct xattr_handler ocfs2_xattr_trusted_handler = {
5273 .prefix = XATTR_TRUSTED_PREFIX,
5274 .list = ocfs2_xattr_trusted_list,
5275 .get = ocfs2_xattr_trusted_get,
5276 .set = ocfs2_xattr_trusted_set,
5277};
5278
99219aea
MF		 5279/*
5280 * 'user' attributes support
5281 */
99219aea
MF		 5282static size_t ocfs2_xattr_user_list(struct inode *inode, char *list,
5283 size_t list_size, const char *name,
5284 size_t name_len)
5285{
ceb1eba3 5286 const size_t prefix_len = XATTR_USER_PREFIX_LEN;
99219aea
MF		 5287 const size_t total_len = prefix_len + name_len + 1;
5288 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5289
5290 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5291 return 0;
5292
5293 if (list && total_len <= list_size) {
5294 memcpy(list, XATTR_USER_PREFIX, prefix_len);
5295 memcpy(list + prefix_len, name, name_len);
5296 list[prefix_len + name_len] = '\0';
5297 }
5298 return total_len;
5299}
5300
5301static int ocfs2_xattr_user_get(struct inode *inode, const char *name,
5302 void *buffer, size_t size)
5303{
5304 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5305
5306 if (strcmp(name, "") == 0)
5307 return -EINVAL;
5308 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5309 return -EOPNOTSUPP;
5310 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
5311 buffer, size);
5312}
5313
5314static int ocfs2_xattr_user_set(struct inode *inode, const char *name,
5315 const void *value, size_t size, int flags)
5316{
5317 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5318
5319 if (strcmp(name, "") == 0)
5320 return -EINVAL;
5321 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5322 return -EOPNOTSUPP;
5323
5324 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER, name, value,
5325 size, flags);
5326}
5327
5328struct xattr_handler ocfs2_xattr_user_handler = {
5329 .prefix = XATTR_USER_PREFIX,
5330 .list = ocfs2_xattr_user_list,
5331 .get = ocfs2_xattr_user_get,
5332 .set = ocfs2_xattr_user_set,
5333};
Clone of davem's net-next-2.6 tree