projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
ocfs2: Pass ocfs2_caching_info to ocfs2_read_extent_block().
[deliverable/linux.git] / fs / ocfs2 / xattr.c
CommitLineData
f56654c4
TM		 1/* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * xattr.c
5 *
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"
d6b32bbb 45#include "blockcheck.h"
f56654c4
TM		 46#include "dlmglue.h"
47#include "file.h"
cf1d6c76
TY		 48#include "symlink.h"
49#include "sysfile.h"
f56654c4
TM		 50#include "inode.h"
51#include "journal.h"
52#include "ocfs2_fs.h"
53#include "suballoc.h"
54#include "uptodate.h"
55#include "buffer_head_io.h"
0c044f0b 56#include "super.h"
cf1d6c76
TY		 57#include "xattr.h"
58
59
60struct ocfs2_xattr_def_value_root {
61 struct ocfs2_xattr_value_root xv;
62 struct ocfs2_extent_rec er;
63};
64
0c044f0b 65struct ocfs2_xattr_bucket {
ba937127
JB		 66 /* The inode these xattrs are associated with */
67 struct inode *bu_inode;
68
69 /* The actual buffers that make up the bucket */
4ac6032d 70 struct buffer_head *bu_bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
ba937127
JB		 71
72 /* How many blocks make up one bucket for this filesystem */
73 int bu_blocks;
0c044f0b
TM		 74};
75
78f30c31 76struct ocfs2_xattr_set_ctxt {
85db90e7 77 handle_t *handle;
78f30c31
TM		 78 struct ocfs2_alloc_context *meta_ac;
79 struct ocfs2_alloc_context *data_ac;
80 struct ocfs2_cached_dealloc_ctxt dealloc;
81};
82
cf1d6c76
TY		 83#define OCFS2_XATTR_ROOT_SIZE (sizeof(struct ocfs2_xattr_def_value_root))
84#define OCFS2_XATTR_INLINE_SIZE 80
4442f518 85#define OCFS2_XATTR_HEADER_GAP 4
534eaddd
TY		 86#define OCFS2_XATTR_FREE_IN_IBODY (OCFS2_MIN_XATTR_INLINE_SIZE \
87 - sizeof(struct ocfs2_xattr_header) \
4442f518 88 - OCFS2_XATTR_HEADER_GAP)
89c38bd0
TY		 89#define OCFS2_XATTR_FREE_IN_BLOCK(ptr) ((ptr)->i_sb->s_blocksize \
90 - sizeof(struct ocfs2_xattr_block) \
91 - sizeof(struct ocfs2_xattr_header) \
4442f518 92 - OCFS2_XATTR_HEADER_GAP)
cf1d6c76
TY		 93
94static struct ocfs2_xattr_def_value_root def_xv = {
95 .xv.xr_list.l_count = cpu_to_le16(1),
96};
97
98struct xattr_handler *ocfs2_xattr_handlers[] = {
99 &ocfs2_xattr_user_handler,
929fb014
TY		 100#ifdef CONFIG_OCFS2_FS_POSIX_ACL
101 &ocfs2_xattr_acl_access_handler,
102 &ocfs2_xattr_acl_default_handler,
103#endif
cf1d6c76 104 &ocfs2_xattr_trusted_handler,
923f7f31 105 &ocfs2_xattr_security_handler,
cf1d6c76
TY		 106 NULL
107};
108
c988fd04 109static struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
cf1d6c76 110 [OCFS2_XATTR_INDEX_USER] = &ocfs2_xattr_user_handler,
929fb014
TY		 111#ifdef CONFIG_OCFS2_FS_POSIX_ACL
112 [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS]
113 = &ocfs2_xattr_acl_access_handler,
114 [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT]
115 = &ocfs2_xattr_acl_default_handler,
116#endif
cf1d6c76 117 [OCFS2_XATTR_INDEX_TRUSTED] = &ocfs2_xattr_trusted_handler,
923f7f31 118 [OCFS2_XATTR_INDEX_SECURITY] = &ocfs2_xattr_security_handler,
cf1d6c76
TY		 119};
120
121struct ocfs2_xattr_info {
122 int name_index;
123 const char *name;
124 const void *value;
125 size_t value_len;
126};
127
128struct ocfs2_xattr_search {
129 struct buffer_head *inode_bh;
130 /*
131 * xattr_bh point to the block buffer head which has extended attribute
132 * when extended attribute in inode, xattr_bh is equal to inode_bh.
133 */
134 struct buffer_head *xattr_bh;
135 struct ocfs2_xattr_header *header;
ba937127 136 struct ocfs2_xattr_bucket *bucket;
cf1d6c76
TY		 137 void *base;
138 void *end;
139 struct ocfs2_xattr_entry *here;
140 int not_found;
141};
142
589dc260
TM		 143static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
144 struct ocfs2_xattr_header *xh,
145 int index,
146 int *block_off,
147 int *new_offset);
148
54f443f4
JB		 149static int ocfs2_xattr_block_find(struct inode *inode,
150 int name_index,
151 const char *name,
152 struct ocfs2_xattr_search *xs);
589dc260
TM		 153static int ocfs2_xattr_index_block_find(struct inode *inode,
154 struct buffer_head *root_bh,
155 int name_index,
156 const char *name,
157 struct ocfs2_xattr_search *xs);
158
0c044f0b
TM		 159static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
160 struct ocfs2_xattr_tree_root *xt,
161 char *buffer,
162 size_t buffer_size);
163
01225596 164static int ocfs2_xattr_create_index_block(struct inode *inode,
78f30c31
TM		 165 struct ocfs2_xattr_search *xs,
166 struct ocfs2_xattr_set_ctxt *ctxt);
01225596
TM		 167
168static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
169 struct ocfs2_xattr_info *xi,
78f30c31
TM		 170 struct ocfs2_xattr_search *xs,
171 struct ocfs2_xattr_set_ctxt *ctxt);
01225596 172
a3944256
TM		 173static int ocfs2_delete_xattr_index_block(struct inode *inode,
174 struct buffer_head *xb_bh);
c58b6032
JB		 175static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
176 u64 src_blk, u64 last_blk, u64 to_blk,
177 unsigned int start_bucket,
178 u32 *first_hash);
a3944256 179
0030e001
TY		 180static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
181{
182 return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
183}
184
185static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
186{
187 return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
188}
189
190static inline u16 ocfs2_xattr_max_xe_in_bucket(struct super_block *sb)
191{
192 u16 len = sb->s_blocksize -
193 offsetof(struct ocfs2_xattr_header, xh_entries);
194
195 return len / sizeof(struct ocfs2_xattr_entry);
196}
197
9c7759aa 198#define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
51def39f 199#define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
3e632946 200#define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
9c7759aa 201
ba937127 202static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
6dde41d9 203{
ba937127
JB		 204 struct ocfs2_xattr_bucket *bucket;
205 int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
6dde41d9 206
ba937127
JB		 207 BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
208
209 bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
210 if (bucket) {
211 bucket->bu_inode = inode;
212 bucket->bu_blocks = blks;
213 }
214
215 return bucket;
216}
217
218static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
219{
220 int i;
221
222 for (i = 0; i < bucket->bu_blocks; i++) {
6dde41d9
JB		 223 brelse(bucket->bu_bhs[i]);
224 bucket->bu_bhs[i] = NULL;
225 }
226}
227
ba937127
JB		 228static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
229{
230 if (bucket) {
231 ocfs2_xattr_bucket_relse(bucket);
232 bucket->bu_inode = NULL;
233 kfree(bucket);
234 }
235}
236
784b816a
JB		 237/*
238 * A bucket that has never been written to disk doesn't need to be
239 * read. We just need the buffer_heads. Don't call this for
240 * buckets that are already on disk. ocfs2_read_xattr_bucket() initializes
241 * them fully.
242 */
ba937127 243static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
784b816a
JB		 244 u64 xb_blkno)
245{
246 int i, rc = 0;
784b816a 247
ba937127
JB		 248 for (i = 0; i < bucket->bu_blocks; i++) {
249 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
250 xb_blkno + i);
784b816a
JB		 251 if (!bucket->bu_bhs[i]) {
252 rc = -EIO;
253 mlog_errno(rc);
254 break;
255 }
256
8cb471e8 257 if (!ocfs2_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
757055ad 258 bucket->bu_bhs[i]))
8cb471e8 259 ocfs2_set_new_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
757055ad 260 bucket->bu_bhs[i]);
784b816a
JB		 261 }
262
263 if (rc)
ba937127 264 ocfs2_xattr_bucket_relse(bucket);
784b816a
JB		 265 return rc;
266}
267
268/* Read the xattr bucket at xb_blkno */
ba937127 269static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
784b816a
JB		 270 u64 xb_blkno)
271{
ba937127 272 int rc;
784b816a 273
8cb471e8 274 rc = ocfs2_read_blocks(INODE_CACHE(bucket->bu_inode), xb_blkno,
970e4936
JB		 275 bucket->bu_blocks, bucket->bu_bhs, 0,
276 NULL);
4d0e214e 277 if (!rc) {
c8b9cf9a 278 spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
4d0e214e
JB		 279 rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
280 bucket->bu_bhs,
281 bucket->bu_blocks,
282 &bucket_xh(bucket)->xh_check);
c8b9cf9a 283 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
4d0e214e
JB		 284 if (rc)
285 mlog_errno(rc);
286 }
287
784b816a 288 if (rc)
ba937127 289 ocfs2_xattr_bucket_relse(bucket);
784b816a
JB		 290 return rc;
291}
292
1224be02 293static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
1224be02
JB		 294 struct ocfs2_xattr_bucket *bucket,
295 int type)
296{
297 int i, rc = 0;
1224be02 298
ba937127 299 for (i = 0; i < bucket->bu_blocks; i++) {
0cf2f763
JB		 300 rc = ocfs2_journal_access(handle,
301 INODE_CACHE(bucket->bu_inode),
1224be02
JB		 302 bucket->bu_bhs[i], type);
303 if (rc) {
304 mlog_errno(rc);
305 break;
306 }
307 }
308
309 return rc;
310}
311
312static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
1224be02
JB		 313 struct ocfs2_xattr_bucket *bucket)
314{
ba937127 315 int i;
1224be02 316
c8b9cf9a 317 spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
4d0e214e
JB		 318 ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
319 bucket->bu_bhs, bucket->bu_blocks,
320 &bucket_xh(bucket)->xh_check);
c8b9cf9a 321 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
4d0e214e 322
ba937127 323 for (i = 0; i < bucket->bu_blocks; i++)
1224be02
JB		 324 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
325}
326
ba937127 327static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
4980c6da
JB		 328 struct ocfs2_xattr_bucket *src)
329{
330 int i;
ba937127
JB		 331 int blocksize = src->bu_inode->i_sb->s_blocksize;
332
333 BUG_ON(dest->bu_blocks != src->bu_blocks);
334 BUG_ON(dest->bu_inode != src->bu_inode);
4980c6da 335
ba937127 336 for (i = 0; i < src->bu_blocks; i++) {
4980c6da
JB		 337 memcpy(bucket_block(dest, i), bucket_block(src, i),
338 blocksize);
339 }
340}
1224be02 341
4ae1d69b
JB		 342static int ocfs2_validate_xattr_block(struct super_block *sb,
343 struct buffer_head *bh)
344{
d6b32bbb 345 int rc;
4ae1d69b
JB		 346 struct ocfs2_xattr_block *xb =
347 (struct ocfs2_xattr_block *)bh->b_data;
348
349 mlog(0, "Validating xattr block %llu\n",
350 (unsigned long long)bh->b_blocknr);
351
d6b32bbb
JB		 352 BUG_ON(!buffer_uptodate(bh));
353
354 /*
355 * If the ecc fails, we return the error but otherwise
356 * leave the filesystem running. We know any error is
357 * local to this block.
358 */
359 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
360 if (rc)
361 return rc;
362
363 /*
364 * Errors after here are fatal
365 */
366
4ae1d69b
JB		 367 if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
368 ocfs2_error(sb,
369 "Extended attribute block #%llu has bad "
370 "signature %.*s",
371 (unsigned long long)bh->b_blocknr, 7,
372 xb->xb_signature);
373 return -EINVAL;
374 }
375
376 if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
377 ocfs2_error(sb,
378 "Extended attribute block #%llu has an "
379 "invalid xb_blkno of %llu",
380 (unsigned long long)bh->b_blocknr,
381 (unsigned long long)le64_to_cpu(xb->xb_blkno));
382 return -EINVAL;
383 }
384
385 if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
386 ocfs2_error(sb,
387 "Extended attribute block #%llu has an invalid "
388 "xb_fs_generation of #%u",
389 (unsigned long long)bh->b_blocknr,
390 le32_to_cpu(xb->xb_fs_generation));
391 return -EINVAL;
392 }
393
394 return 0;
395}
396
397static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
398 struct buffer_head **bh)
399{
400 int rc;
401 struct buffer_head *tmp = *bh;
402
8cb471e8 403 rc = ocfs2_read_block(INODE_CACHE(inode), xb_blkno, &tmp,
970e4936 404 ocfs2_validate_xattr_block);
4ae1d69b
JB		 405
406 /* If ocfs2_read_block() got us a new bh, pass it up. */
407 if (!rc && !*bh)
408 *bh = tmp;
409
410 return rc;
411}
412
936b8834 413static inline const char *ocfs2_xattr_prefix(int name_index)
cf1d6c76
TY		 414{
415 struct xattr_handler *handler = NULL;
416
417 if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
418 handler = ocfs2_xattr_handler_map[name_index];
419
936b8834 420 return handler ? handler->prefix : NULL;
cf1d6c76
TY		 421}
422
40daa16a 423static u32 ocfs2_xattr_name_hash(struct inode *inode,
2057e5c6 424 const char *name,
40daa16a 425 int name_len)
cf1d6c76
TY		 426{
427 /* Get hash value of uuid from super block */
428 u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
429 int i;
430
cf1d6c76
TY		 431 /* hash extended attribute name */
432 for (i = 0; i < name_len; i++) {
433 hash = (hash << OCFS2_HASH_SHIFT) ^
434 (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
435 *name++;
436 }
437
438 return hash;
439}
440
441/*
442 * ocfs2_xattr_hash_entry()
443 *
444 * Compute the hash of an extended attribute.
445 */
446static void ocfs2_xattr_hash_entry(struct inode *inode,
447 struct ocfs2_xattr_header *header,
448 struct ocfs2_xattr_entry *entry)
449{
450 u32 hash = 0;
cf1d6c76 451 char *name = (char *)header + le16_to_cpu(entry->xe_name_offset);
cf1d6c76 452
2057e5c6 453 hash = ocfs2_xattr_name_hash(inode, name, entry->xe_name_len);
cf1d6c76
TY		 454 entry->xe_name_hash = cpu_to_le32(hash);
455
456 return;
457}
f56654c4 458
534eaddd
TY		 459static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
460{
461 int size = 0;
462
463 if (value_len <= OCFS2_XATTR_INLINE_SIZE)
464 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
465 else
466 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
467 size += sizeof(struct ocfs2_xattr_entry);
468
469 return size;
470}
471
472int ocfs2_calc_security_init(struct inode *dir,
473 struct ocfs2_security_xattr_info *si,
474 int *want_clusters,
475 int *xattr_credits,
476 struct ocfs2_alloc_context **xattr_ac)
477{
478 int ret = 0;
479 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
480 int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
481 si->value_len);
482
483 /*
484 * The max space of security xattr taken inline is
485 * 256(name) + 80(value) + 16(entry) = 352 bytes,
486 * So reserve one metadata block for it is ok.
487 */
488 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
489 s_size > OCFS2_XATTR_FREE_IN_IBODY) {
490 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
491 if (ret) {
492 mlog_errno(ret);
493 return ret;
494 }
495 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
496 }
497
498 /* reserve clusters for xattr value which will be set in B tree*/
0e445b6f
TY		 499 if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
500 int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
501 si->value_len);
502
503 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
504 new_clusters);
505 *want_clusters += new_clusters;
506 }
534eaddd
TY		 507 return ret;
508}
509
89c38bd0
TY		 510int ocfs2_calc_xattr_init(struct inode *dir,
511 struct buffer_head *dir_bh,
512 int mode,
513 struct ocfs2_security_xattr_info *si,
514 int *want_clusters,
515 int *xattr_credits,
9b7895ef 516 int *want_meta)
89c38bd0
TY		 517{
518 int ret = 0;
519 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
0e445b6f 520 int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
89c38bd0
TY		 521
522 if (si->enable)
523 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
524 si->value_len);
525
526 if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
527 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
528 OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
529 "", NULL, 0);
530 if (acl_len > 0) {
531 a_size = ocfs2_xattr_entry_real_size(0, acl_len);
532 if (S_ISDIR(mode))
533 a_size <<= 1;
534 } else if (acl_len != 0 && acl_len != -ENODATA) {
535 mlog_errno(ret);
536 return ret;
537 }
538 }
539
540 if (!(s_size + a_size))
541 return ret;
542
543 /*
544 * The max space of security xattr taken inline is
545 * 256(name) + 80(value) + 16(entry) = 352 bytes,
546 * The max space of acl xattr taken inline is
547 * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
548 * when blocksize = 512, may reserve one more cluser for
549 * xattr bucket, otherwise reserve one metadata block
550 * for them is ok.
6c9fd1dc
TY		 551 * If this is a new directory with inline data,
552 * we choose to reserve the entire inline area for
553 * directory contents and force an external xattr block.
89c38bd0
TY		 554 */
555 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
6c9fd1dc 556 (S_ISDIR(mode) && ocfs2_supports_inline_data(osb)) ||
89c38bd0 557 (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
9b7895ef 558 *want_meta = *want_meta + 1;
89c38bd0
TY		 559 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
560 }
561
562 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
563 (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
564 *want_clusters += 1;
565 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
566 }
567
0e445b6f
TY		 568 /*
569 * reserve credits and clusters for xattrs which has large value
570 * and have to be set outside
571 */
572 if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
573 new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
574 si->value_len);
575 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
576 new_clusters);
577 *want_clusters += new_clusters;
578 }
89c38bd0
TY		 579 if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
580 acl_len > OCFS2_XATTR_INLINE_SIZE) {
0e445b6f
TY		 581 /* for directory, it has DEFAULT and ACCESS two types of acls */
582 new_clusters = (S_ISDIR(mode) ? 2 : 1) *
583 ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
584 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
585 new_clusters);
586 *want_clusters += new_clusters;
89c38bd0
TY		 587 }
588
589 return ret;
590}
591
f56654c4
TM		 592static int ocfs2_xattr_extend_allocation(struct inode *inode,
593 u32 clusters_to_add,
19b801f4 594 struct ocfs2_xattr_value_buf *vb,
78f30c31 595 struct ocfs2_xattr_set_ctxt *ctxt)
f56654c4
TM		 596{
597 int status = 0;
85db90e7 598 handle_t *handle = ctxt->handle;
f56654c4
TM		 599 enum ocfs2_alloc_restarted why;
600 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
19b801f4 601 u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
f99b9b7c 602 struct ocfs2_extent_tree et;
f56654c4
TM		 603
604 mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
605
19b801f4 606 ocfs2_init_xattr_value_extent_tree(&et, inode, vb);
f99b9b7c 607
0cf2f763 608 status = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
2a50a743 609 OCFS2_JOURNAL_ACCESS_WRITE);
f56654c4
TM		 610 if (status < 0) {
611 mlog_errno(status);
612 goto leave;
613 }
614
19b801f4 615 prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
f56654c4
TM		 616 status = ocfs2_add_clusters_in_btree(osb,
617 inode,
618 &logical_start,
619 clusters_to_add,
620 0,
f99b9b7c 621 &et,
f56654c4 622 handle,
78f30c31
TM		 623 ctxt->data_ac,
624 ctxt->meta_ac,
f99b9b7c 625 &why);
85db90e7
TM		 626 if (status < 0) {
627 mlog_errno(status);
f56654c4
TM		 628 goto leave;
629 }
630
19b801f4 631 status = ocfs2_journal_dirty(handle, vb->vb_bh);
f56654c4
TM		 632 if (status < 0) {
633 mlog_errno(status);
634 goto leave;
635 }
636
19b801f4 637 clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) - prev_clusters;
f56654c4 638
85db90e7
TM		 639 /*
640 * We should have already allocated enough space before the transaction,
641 * so no need to restart.
642 */
643 BUG_ON(why != RESTART_NONE || clusters_to_add);
f56654c4
TM		 644
645leave:
f56654c4
TM		 646
647 return status;
648}
649
650static int __ocfs2_remove_xattr_range(struct inode *inode,
d72cc72d 651 struct ocfs2_xattr_value_buf *vb,
f56654c4 652 u32 cpos, u32 phys_cpos, u32 len,
78f30c31 653 struct ocfs2_xattr_set_ctxt *ctxt)
f56654c4
TM		 654{
655 int ret;
656 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
85db90e7 657 handle_t *handle = ctxt->handle;
f99b9b7c
JB		 658 struct ocfs2_extent_tree et;
659
d72cc72d 660 ocfs2_init_xattr_value_extent_tree(&et, inode, vb);
f56654c4 661
0cf2f763 662 ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
d72cc72d 663 OCFS2_JOURNAL_ACCESS_WRITE);
f56654c4
TM		 664 if (ret) {
665 mlog_errno(ret);
85db90e7 666 goto out;
f56654c4
TM		 667 }
668
78f30c31
TM		 669 ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, ctxt->meta_ac,
670 &ctxt->dealloc);
f56654c4
TM		 671 if (ret) {
672 mlog_errno(ret);
85db90e7 673 goto out;
f56654c4
TM		 674 }
675
d72cc72d 676 le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
f56654c4 677
d72cc72d 678 ret = ocfs2_journal_dirty(handle, vb->vb_bh);
f56654c4
TM		 679 if (ret) {
680 mlog_errno(ret);
85db90e7 681 goto out;
f56654c4
TM		 682 }
683
78f30c31 684 ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc, phys_blkno, len);
f56654c4
TM		 685 if (ret)
686 mlog_errno(ret);
687
f56654c4 688out:
f56654c4
TM		 689 return ret;
690}
691
692static int ocfs2_xattr_shrink_size(struct inode *inode,
693 u32 old_clusters,
694 u32 new_clusters,
19b801f4 695 struct ocfs2_xattr_value_buf *vb,
78f30c31 696 struct ocfs2_xattr_set_ctxt *ctxt)
f56654c4
TM		 697{
698 int ret = 0;
699 u32 trunc_len, cpos, phys_cpos, alloc_size;
700 u64 block;
f56654c4
TM		 701
702 if (old_clusters <= new_clusters)
703 return 0;
704
705 cpos = new_clusters;
706 trunc_len = old_clusters - new_clusters;
707 while (trunc_len) {
708 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
d72cc72d 709 &alloc_size,
19b801f4 710 &vb->vb_xv->xr_list);
f56654c4
TM		 711 if (ret) {
712 mlog_errno(ret);
713 goto out;
714 }
715
716 if (alloc_size > trunc_len)
717 alloc_size = trunc_len;
718
19b801f4 719 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
f56654c4 720 phys_cpos, alloc_size,
78f30c31 721 ctxt);
f56654c4
TM		 722 if (ret) {
723 mlog_errno(ret);
724 goto out;
725 }
726
727 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
8cb471e8
JB		 728 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode),
729 block, alloc_size);
f56654c4
TM		 730 cpos += alloc_size;
731 trunc_len -= alloc_size;
732 }
733
734out:
f56654c4
TM		 735 return ret;
736}
737
738static int ocfs2_xattr_value_truncate(struct inode *inode,
b3e5d379 739 struct ocfs2_xattr_value_buf *vb,
78f30c31
TM		 740 int len,
741 struct ocfs2_xattr_set_ctxt *ctxt)
f56654c4
TM		 742{
743 int ret;
744 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
b3e5d379 745 u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
f56654c4
TM		 746
747 if (new_clusters == old_clusters)
748 return 0;
749
750 if (new_clusters > old_clusters)
751 ret = ocfs2_xattr_extend_allocation(inode,
752 new_clusters - old_clusters,
b3e5d379 753 vb, ctxt);
f56654c4
TM		 754 else
755 ret = ocfs2_xattr_shrink_size(inode,
756 old_clusters, new_clusters,
b3e5d379 757 vb, ctxt);
f56654c4
TM		 758
759 return ret;
760}
cf1d6c76 761
936b8834
TM		 762static int ocfs2_xattr_list_entry(char *buffer, size_t size,
763 size_t *result, const char *prefix,
764 const char *name, int name_len)
765{
766 char *p = buffer + *result;
767 int prefix_len = strlen(prefix);
768 int total_len = prefix_len + name_len + 1;
769
770 *result += total_len;
771
772 /* we are just looking for how big our buffer needs to be */
773 if (!size)
774 return 0;
775
776 if (*result > size)
777 return -ERANGE;
778
779 memcpy(p, prefix, prefix_len);
780 memcpy(p + prefix_len, name, name_len);
781 p[prefix_len + name_len] = '\0';
782
783 return 0;
784}
785
cf1d6c76
TY		 786static int ocfs2_xattr_list_entries(struct inode *inode,
787 struct ocfs2_xattr_header *header,
788 char *buffer, size_t buffer_size)
789{
936b8834
TM		 790 size_t result = 0;
791 int i, type, ret;
792 const char *prefix, *name;
cf1d6c76
TY		 793
794 for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
795 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
936b8834
TM		 796 type = ocfs2_xattr_get_type(entry);
797 prefix = ocfs2_xattr_prefix(type);
798
799 if (prefix) {
800 name = (const char *)header +
801 le16_to_cpu(entry->xe_name_offset);
802
803 ret = ocfs2_xattr_list_entry(buffer, buffer_size,
804 &result, prefix, name,
805 entry->xe_name_len);
806 if (ret)
807 return ret;
cf1d6c76
TY		 808 }
809 }
810
936b8834 811 return result;
cf1d6c76
TY		 812}
813
814static int ocfs2_xattr_ibody_list(struct inode *inode,
815 struct ocfs2_dinode *di,
816 char *buffer,
817 size_t buffer_size)
818{
819 struct ocfs2_xattr_header *header = NULL;
820 struct ocfs2_inode_info *oi = OCFS2_I(inode);
821 int ret = 0;
822
823 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
824 return ret;
825
826 header = (struct ocfs2_xattr_header *)
827 ((void *)di + inode->i_sb->s_blocksize -
828 le16_to_cpu(di->i_xattr_inline_size));
829
830 ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
831
832 return ret;
833}
834
835static int ocfs2_xattr_block_list(struct inode *inode,
836 struct ocfs2_dinode *di,
837 char *buffer,
838 size_t buffer_size)
839{
840 struct buffer_head *blk_bh = NULL;
0c044f0b 841 struct ocfs2_xattr_block *xb;
cf1d6c76
TY		 842 int ret = 0;
843
844 if (!di->i_xattr_loc)
845 return ret;
846
4ae1d69b
JB		 847 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
848 &blk_bh);
cf1d6c76
TY		 849 if (ret < 0) {
850 mlog_errno(ret);
851 return ret;
852 }
cf1d6c76 853
0c044f0b 854 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
0c044f0b
TM		 855 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
856 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
857 ret = ocfs2_xattr_list_entries(inode, header,
858 buffer, buffer_size);
859 } else {
860 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
861 ret = ocfs2_xattr_tree_list_index_block(inode, xt,
862 buffer, buffer_size);
863 }
4ae1d69b 864
cf1d6c76
TY		 865 brelse(blk_bh);
866
867 return ret;
868}
869
870ssize_t ocfs2_listxattr(struct dentry *dentry,
871 char *buffer,
872 size_t size)
873{
874 int ret = 0, i_ret = 0, b_ret = 0;
875 struct buffer_head *di_bh = NULL;
876 struct ocfs2_dinode *di = NULL;
877 struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
878
8154da3d
TY		 879 if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
880 return -EOPNOTSUPP;
881
cf1d6c76
TY		 882 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
883 return ret;
884
885 ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
886 if (ret < 0) {
887 mlog_errno(ret);
888 return ret;
889 }
890
891 di = (struct ocfs2_dinode *)di_bh->b_data;
892
893 down_read(&oi->ip_xattr_sem);
894 i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
895 if (i_ret < 0)
896 b_ret = 0;
897 else {
898 if (buffer) {
899 buffer += i_ret;
900 size -= i_ret;
901 }
902 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
903 buffer, size);
904 if (b_ret < 0)
905 i_ret = 0;
906 }
907 up_read(&oi->ip_xattr_sem);
908 ocfs2_inode_unlock(dentry->d_inode, 0);
909
910 brelse(di_bh);
911
912 return i_ret + b_ret;
913}
914
915static int ocfs2_xattr_find_entry(int name_index,
916 const char *name,
917 struct ocfs2_xattr_search *xs)
918{
919 struct ocfs2_xattr_entry *entry;
920 size_t name_len;
921 int i, cmp = 1;
922
923 if (name == NULL)
924 return -EINVAL;
925
926 name_len = strlen(name);
927 entry = xs->here;
928 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
929 cmp = name_index - ocfs2_xattr_get_type(entry);
930 if (!cmp)
931 cmp = name_len - entry->xe_name_len;
932 if (!cmp)
933 cmp = memcmp(name, (xs->base +
934 le16_to_cpu(entry->xe_name_offset)),
935 name_len);
936 if (cmp == 0)
937 break;
938 entry += 1;
939 }
940 xs->here = entry;
941
942 return cmp ? -ENODATA : 0;
943}
944
945static int ocfs2_xattr_get_value_outside(struct inode *inode,
589dc260 946 struct ocfs2_xattr_value_root *xv,
cf1d6c76
TY		 947 void *buffer,
948 size_t len)
949{
950 u32 cpos, p_cluster, num_clusters, bpc, clusters;
951 u64 blkno;
952 int i, ret = 0;
953 size_t cplen, blocksize;
954 struct buffer_head *bh = NULL;
cf1d6c76
TY		 955 struct ocfs2_extent_list *el;
956
cf1d6c76
TY		 957 el = &xv->xr_list;
958 clusters = le32_to_cpu(xv->xr_clusters);
959 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
960 blocksize = inode->i_sb->s_blocksize;
961
962 cpos = 0;
963 while (cpos < clusters) {
964 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
965 &num_clusters, el);
966 if (ret) {
967 mlog_errno(ret);
968 goto out;
969 }
970
971 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
972 /* Copy ocfs2_xattr_value */
973 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
8cb471e8
JB		 974 ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
975 &bh, NULL);
cf1d6c76
TY		 976 if (ret) {
977 mlog_errno(ret);
978 goto out;
979 }
980
981 cplen = len >= blocksize ? blocksize : len;
982 memcpy(buffer, bh->b_data, cplen);
983 len -= cplen;
984 buffer += cplen;
985
986 brelse(bh);
987 bh = NULL;
988 if (len == 0)
989 break;
990 }
991 cpos += num_clusters;
992 }
993out:
994 return ret;
995}
996
997static int ocfs2_xattr_ibody_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{
1004 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1005 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
589dc260 1006 struct ocfs2_xattr_value_root *xv;
cf1d6c76
TY		 1007 size_t size;
1008 int ret = 0;
1009
1010 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1011 return -ENODATA;
1012
1013 xs->end = (void *)di + inode->i_sb->s_blocksize;
1014 xs->header = (struct ocfs2_xattr_header *)
1015 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1016 xs->base = (void *)xs->header;
1017 xs->here = xs->header->xh_entries;
1018
1019 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1020 if (ret)
1021 return ret;
1022 size = le64_to_cpu(xs->here->xe_value_size);
1023 if (buffer) {
1024 if (size > buffer_size)
1025 return -ERANGE;
1026 if (ocfs2_xattr_is_local(xs->here)) {
1027 memcpy(buffer, (void *)xs->base +
1028 le16_to_cpu(xs->here->xe_name_offset) +
1029 OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1030 } else {
589dc260
TM		 1031 xv = (struct ocfs2_xattr_value_root *)
1032 (xs->base + le16_to_cpu(
1033 xs->here->xe_name_offset) +
1034 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1035 ret = ocfs2_xattr_get_value_outside(inode, xv,
cf1d6c76
TY		 1036 buffer, size);
1037 if (ret < 0) {
1038 mlog_errno(ret);
1039 return ret;
1040 }
1041 }
1042 }
1043
1044 return size;
1045}
1046
1047static int ocfs2_xattr_block_get(struct inode *inode,
1048 int name_index,
1049 const char *name,
1050 void *buffer,
1051 size_t buffer_size,
1052 struct ocfs2_xattr_search *xs)
1053{
cf1d6c76 1054 struct ocfs2_xattr_block *xb;
589dc260 1055 struct ocfs2_xattr_value_root *xv;
cf1d6c76 1056 size_t size;
44d8e4e1
SM		 1057 int ret = -ENODATA, name_offset, name_len, i;
1058 int uninitialized_var(block_off);
cf1d6c76 1059
ba937127
JB		 1060 xs->bucket = ocfs2_xattr_bucket_new(inode);
1061 if (!xs->bucket) {
1062 ret = -ENOMEM;
1063 mlog_errno(ret);
1064 goto cleanup;
1065 }
589dc260 1066
54f443f4
JB		 1067 ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1068 if (ret) {
cf1d6c76 1069 mlog_errno(ret);
cf1d6c76
TY		 1070 goto cleanup;
1071 }
1072
6c1e183e
TY		 1073 if (xs->not_found) {
1074 ret = -ENODATA;
1075 goto cleanup;
1076 }
1077
54f443f4 1078 xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
cf1d6c76
TY		 1079 size = le64_to_cpu(xs->here->xe_value_size);
1080 if (buffer) {
1081 ret = -ERANGE;
1082 if (size > buffer_size)
1083 goto cleanup;
589dc260
TM		 1084
1085 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1086 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1087 i = xs->here - xs->header->xh_entries;
1088
1089 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1090 ret = ocfs2_xattr_bucket_get_name_value(inode,
ba937127 1091 bucket_xh(xs->bucket),
589dc260
TM		 1092 i,
1093 &block_off,
1094 &name_offset);
ba937127 1095 xs->base = bucket_block(xs->bucket, block_off);
589dc260 1096 }
cf1d6c76
TY		 1097 if (ocfs2_xattr_is_local(xs->here)) {
1098 memcpy(buffer, (void *)xs->base +
589dc260 1099 name_offset + name_len, size);
cf1d6c76 1100 } else {
589dc260
TM		 1101 xv = (struct ocfs2_xattr_value_root *)
1102 (xs->base + name_offset + name_len);
1103 ret = ocfs2_xattr_get_value_outside(inode, xv,
cf1d6c76
TY		 1104 buffer, size);
1105 if (ret < 0) {
1106 mlog_errno(ret);
1107 goto cleanup;
1108 }
1109 }
1110 }
1111 ret = size;
1112cleanup:
ba937127 1113 ocfs2_xattr_bucket_free(xs->bucket);
cf1d6c76 1114
54f443f4
JB		 1115 brelse(xs->xattr_bh);
1116 xs->xattr_bh = NULL;
cf1d6c76
TY		 1117 return ret;
1118}
1119
4e3e9d02
TY		 1120int ocfs2_xattr_get_nolock(struct inode *inode,
1121 struct buffer_head *di_bh,
0030e001
TY		 1122 int name_index,
1123 const char *name,
1124 void *buffer,
1125 size_t buffer_size)
cf1d6c76
TY		 1126{
1127 int ret;
1128 struct ocfs2_dinode *di = NULL;
cf1d6c76
TY		 1129 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1130 struct ocfs2_xattr_search xis = {
1131 .not_found = -ENODATA,
1132 };
1133 struct ocfs2_xattr_search xbs = {
1134 .not_found = -ENODATA,
1135 };
1136
8154da3d
TY		 1137 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1138 return -EOPNOTSUPP;
1139
cf1d6c76
TY		 1140 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1141 ret = -ENODATA;
1142
cf1d6c76
TY		 1143 xis.inode_bh = xbs.inode_bh = di_bh;
1144 di = (struct ocfs2_dinode *)di_bh->b_data;
1145
1146 down_read(&oi->ip_xattr_sem);
1147 ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1148 buffer_size, &xis);
6c1e183e 1149 if (ret == -ENODATA && di->i_xattr_loc)
cf1d6c76
TY		 1150 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1151 buffer_size, &xbs);
1152 up_read(&oi->ip_xattr_sem);
4e3e9d02
TY		 1153
1154 return ret;
1155}
1156
1157/* ocfs2_xattr_get()
1158 *
1159 * Copy an extended attribute into the buffer provided.
1160 * Buffer is NULL to compute the size of buffer required.
1161 */
1162static int ocfs2_xattr_get(struct inode *inode,
1163 int name_index,
1164 const char *name,
1165 void *buffer,
1166 size_t buffer_size)
1167{
1168 int ret;
1169 struct buffer_head *di_bh = NULL;
1170
1171 ret = ocfs2_inode_lock(inode, &di_bh, 0);
1172 if (ret < 0) {
1173 mlog_errno(ret);
1174 return ret;
1175 }
1176 ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1177 name, buffer, buffer_size);
1178
cf1d6c76
TY		 1179 ocfs2_inode_unlock(inode, 0);
1180
1181 brelse(di_bh);
1182
1183 return ret;
1184}
1185
1186static int __ocfs2_xattr_set_value_outside(struct inode *inode,
85db90e7 1187 handle_t *handle,
cf1d6c76
TY		 1188 struct ocfs2_xattr_value_root *xv,
1189 const void *value,
1190 int value_len)
1191{
71d548a6 1192 int ret = 0, i, cp_len;
cf1d6c76
TY		 1193 u16 blocksize = inode->i_sb->s_blocksize;
1194 u32 p_cluster, num_clusters;
1195 u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1196 u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1197 u64 blkno;
1198 struct buffer_head *bh = NULL;
cf1d6c76
TY		 1199
1200 BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1201
cf1d6c76
TY		 1202 while (cpos < clusters) {
1203 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1204 &num_clusters, &xv->xr_list);
1205 if (ret) {
1206 mlog_errno(ret);
85db90e7 1207 goto out;
cf1d6c76
TY		 1208 }
1209
1210 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1211
1212 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
8cb471e8
JB		 1213 ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1214 &bh, NULL);
cf1d6c76
TY		 1215 if (ret) {
1216 mlog_errno(ret);
85db90e7 1217 goto out;
cf1d6c76
TY		 1218 }
1219
1220 ret = ocfs2_journal_access(handle,
0cf2f763 1221 INODE_CACHE(inode),
cf1d6c76
TY		 1222 bh,
1223 OCFS2_JOURNAL_ACCESS_WRITE);
1224 if (ret < 0) {
1225 mlog_errno(ret);
85db90e7 1226 goto out;
cf1d6c76
TY		 1227 }
1228
1229 cp_len = value_len > blocksize ? blocksize : value_len;
1230 memcpy(bh->b_data, value, cp_len);
1231 value_len -= cp_len;
1232 value += cp_len;
1233 if (cp_len < blocksize)
1234 memset(bh->b_data + cp_len, 0,
1235 blocksize - cp_len);
1236
1237 ret = ocfs2_journal_dirty(handle, bh);
1238 if (ret < 0) {
1239 mlog_errno(ret);
85db90e7 1240 goto out;
cf1d6c76
TY		 1241 }
1242 brelse(bh);
1243 bh = NULL;
1244
1245 /*
1246 * XXX: do we need to empty all the following
1247 * blocks in this cluster?
1248 */
1249 if (!value_len)
1250 break;
1251 }
1252 cpos += num_clusters;
1253 }
cf1d6c76
TY		 1254out:
1255 brelse(bh);
1256
1257 return ret;
1258}
1259
1260static int ocfs2_xattr_cleanup(struct inode *inode,
85db90e7 1261 handle_t *handle,
cf1d6c76
TY		 1262 struct ocfs2_xattr_info *xi,
1263 struct ocfs2_xattr_search *xs,
512620f4 1264 struct ocfs2_xattr_value_buf *vb,
cf1d6c76
TY		 1265 size_t offs)
1266{
cf1d6c76
TY		 1267 int ret = 0;
1268 size_t name_len = strlen(xi->name);
1269 void *val = xs->base + offs;
1270 size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1271
0cf2f763 1272 ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
512620f4 1273 OCFS2_JOURNAL_ACCESS_WRITE);
cf1d6c76
TY		 1274 if (ret) {
1275 mlog_errno(ret);
85db90e7 1276 goto out;
cf1d6c76
TY		 1277 }
1278 /* Decrease xattr count */
1279 le16_add_cpu(&xs->header->xh_count, -1);
1280 /* Remove the xattr entry and tree root which has already be set*/
1281 memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
1282 memset(val, 0, size);
1283
512620f4 1284 ret = ocfs2_journal_dirty(handle, vb->vb_bh);
cf1d6c76
TY		 1285 if (ret < 0)
1286 mlog_errno(ret);
cf1d6c76
TY		 1287out:
1288 return ret;
1289}
1290
1291static int ocfs2_xattr_update_entry(struct inode *inode,
85db90e7 1292 handle_t *handle,
cf1d6c76
TY		 1293 struct ocfs2_xattr_info *xi,
1294 struct ocfs2_xattr_search *xs,
0c748e95 1295 struct ocfs2_xattr_value_buf *vb,
cf1d6c76
TY		 1296 size_t offs)
1297{
85db90e7 1298 int ret;
cf1d6c76 1299
0cf2f763 1300 ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
0c748e95 1301 OCFS2_JOURNAL_ACCESS_WRITE);
cf1d6c76
TY		 1302 if (ret) {
1303 mlog_errno(ret);
85db90e7 1304 goto out;
cf1d6c76
TY		 1305 }
1306
1307 xs->here->xe_name_offset = cpu_to_le16(offs);
1308 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1309 if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1310 ocfs2_xattr_set_local(xs->here, 1);
1311 else
1312 ocfs2_xattr_set_local(xs->here, 0);
1313 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1314
0c748e95 1315 ret = ocfs2_journal_dirty(handle, vb->vb_bh);
cf1d6c76
TY		 1316 if (ret < 0)
1317 mlog_errno(ret);
cf1d6c76
TY		 1318out:
1319 return ret;
1320}
1321
1322/*
1323 * ocfs2_xattr_set_value_outside()
1324 *
1325 * Set large size value in B tree.
1326 */
1327static int ocfs2_xattr_set_value_outside(struct inode *inode,
1328 struct ocfs2_xattr_info *xi,
1329 struct ocfs2_xattr_search *xs,
78f30c31 1330 struct ocfs2_xattr_set_ctxt *ctxt,
512620f4 1331 struct ocfs2_xattr_value_buf *vb,
cf1d6c76
TY		 1332 size_t offs)
1333{
1334 size_t name_len = strlen(xi->name);
1335 void *val = xs->base + offs;
1336 struct ocfs2_xattr_value_root *xv = NULL;
1337 size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1338 int ret = 0;
1339
1340 memset(val, 0, size);
1341 memcpy(val, xi->name, name_len);
1342 xv = (struct ocfs2_xattr_value_root *)
1343 (val + OCFS2_XATTR_SIZE(name_len));
1344 xv->xr_clusters = 0;
1345 xv->xr_last_eb_blk = 0;
1346 xv->xr_list.l_tree_depth = 0;
1347 xv->xr_list.l_count = cpu_to_le16(1);
1348 xv->xr_list.l_next_free_rec = 0;
512620f4 1349 vb->vb_xv = xv;
cf1d6c76 1350
512620f4 1351 ret = ocfs2_xattr_value_truncate(inode, vb, xi->value_len, ctxt);
cf1d6c76
TY		 1352 if (ret < 0) {
1353 mlog_errno(ret);
1354 return ret;
1355 }
512620f4 1356 ret = ocfs2_xattr_update_entry(inode, ctxt->handle, xi, xs, vb, offs);
cf1d6c76
TY		 1357 if (ret < 0) {
1358 mlog_errno(ret);
1359 return ret;
1360 }
512620f4 1361 ret = __ocfs2_xattr_set_value_outside(inode, ctxt->handle, vb->vb_xv,
85db90e7 1362 xi->value, xi->value_len);
cf1d6c76
TY		 1363 if (ret < 0)
1364 mlog_errno(ret);
1365
1366 return ret;
1367}
1368
1369/*
1370 * ocfs2_xattr_set_entry_local()
1371 *
1372 * Set, replace or remove extended attribute in local.
1373 */
1374static void ocfs2_xattr_set_entry_local(struct inode *inode,
1375 struct ocfs2_xattr_info *xi,
1376 struct ocfs2_xattr_search *xs,
1377 struct ocfs2_xattr_entry *last,
1378 size_t min_offs)
1379{
1380 size_t name_len = strlen(xi->name);
1381 int i;
1382
1383 if (xi->value && xs->not_found) {
1384 /* Insert the new xattr entry. */
1385 le16_add_cpu(&xs->header->xh_count, 1);
1386 ocfs2_xattr_set_type(last, xi->name_index);
1387 ocfs2_xattr_set_local(last, 1);
1388 last->xe_name_len = name_len;
1389 } else {
1390 void *first_val;
1391 void *val;
1392 size_t offs, size;
1393
1394 first_val = xs->base + min_offs;
1395 offs = le16_to_cpu(xs->here->xe_name_offset);
1396 val = xs->base + offs;
1397
1398 if (le64_to_cpu(xs->here->xe_value_size) >
1399 OCFS2_XATTR_INLINE_SIZE)
1400 size = OCFS2_XATTR_SIZE(name_len) +
1401 OCFS2_XATTR_ROOT_SIZE;
1402 else
1403 size = OCFS2_XATTR_SIZE(name_len) +
1404 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1405
1406 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1407 OCFS2_XATTR_SIZE(xi->value_len)) {
1408 /* The old and the new value have the
1409 same size. Just replace the value. */
1410 ocfs2_xattr_set_local(xs->here, 1);
1411 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1412 /* Clear value bytes. */
1413 memset(val + OCFS2_XATTR_SIZE(name_len),
1414 0,
1415 OCFS2_XATTR_SIZE(xi->value_len));
1416 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1417 xi->value,
1418 xi->value_len);
1419 return;
1420 }
1421 /* Remove the old name+value. */
1422 memmove(first_val + size, first_val, val - first_val);
1423 memset(first_val, 0, size);
1424 xs->here->xe_name_hash = 0;
1425 xs->here->xe_name_offset = 0;
1426 ocfs2_xattr_set_local(xs->here, 1);
1427 xs->here->xe_value_size = 0;
1428
1429 min_offs += size;
1430
1431 /* Adjust all value offsets. */
1432 last = xs->header->xh_entries;
1433 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1434 size_t o = le16_to_cpu(last->xe_name_offset);
1435
1436 if (o < offs)
1437 last->xe_name_offset = cpu_to_le16(o + size);
1438 last += 1;
1439 }
1440
1441 if (!xi->value) {
1442 /* Remove the old entry. */
1443 last -= 1;
1444 memmove(xs->here, xs->here + 1,
1445 (void *)last - (void *)xs->here);
1446 memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1447 le16_add_cpu(&xs->header->xh_count, -1);
1448 }
1449 }
1450 if (xi->value) {
1451 /* Insert the new name+value. */
1452 size_t size = OCFS2_XATTR_SIZE(name_len) +
1453 OCFS2_XATTR_SIZE(xi->value_len);
1454 void *val = xs->base + min_offs - size;
1455
1456 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1457 memset(val, 0, size);
1458 memcpy(val, xi->name, name_len);
1459 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1460 xi->value,
1461 xi->value_len);
1462 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1463 ocfs2_xattr_set_local(xs->here, 1);
1464 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1465 }
1466
1467 return;
1468}
1469
1470/*
1471 * ocfs2_xattr_set_entry()
1472 *
1473 * Set extended attribute entry into inode or block.
1474 *
1475 * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1476 * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1477 * then set value in B tree with set_value_outside().
1478 */
1479static int ocfs2_xattr_set_entry(struct inode *inode,
1480 struct ocfs2_xattr_info *xi,
1481 struct ocfs2_xattr_search *xs,
78f30c31 1482 struct ocfs2_xattr_set_ctxt *ctxt,
cf1d6c76
TY		 1483 int flag)
1484{
1485 struct ocfs2_xattr_entry *last;
1486 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1487 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1488 size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1489 size_t size_l = 0;
85db90e7 1490 handle_t *handle = ctxt->handle;
cf1d6c76
TY		 1491 int free, i, ret;
1492 struct ocfs2_xattr_info xi_l = {
1493 .name_index = xi->name_index,
1494 .name = xi->name,
1495 .value = xi->value,
1496 .value_len = xi->value_len,
1497 };
512620f4
JB		 1498 struct ocfs2_xattr_value_buf vb = {
1499 .vb_bh = xs->xattr_bh,
1500 .vb_access = ocfs2_journal_access_di,
1501 };
1502
1503 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1504 BUG_ON(xs->xattr_bh == xs->inode_bh);
1505 vb.vb_access = ocfs2_journal_access_xb;
1506 } else
1507 BUG_ON(xs->xattr_bh != xs->inode_bh);
cf1d6c76
TY		 1508
1509 /* Compute min_offs, last and free space. */
1510 last = xs->header->xh_entries;
1511
1512 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1513 size_t offs = le16_to_cpu(last->xe_name_offset);
1514 if (offs < min_offs)
1515 min_offs = offs;
1516 last += 1;
1517 }
1518
4442f518 1519 free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
cf1d6c76 1520 if (free < 0)
b37c4d84 1521 return -EIO;
cf1d6c76
TY		 1522
1523 if (!xs->not_found) {
1524 size_t size = 0;
1525 if (ocfs2_xattr_is_local(xs->here))
1526 size = OCFS2_XATTR_SIZE(name_len) +
1527 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1528 else
1529 size = OCFS2_XATTR_SIZE(name_len) +
1530 OCFS2_XATTR_ROOT_SIZE;
1531 free += (size + sizeof(struct ocfs2_xattr_entry));
1532 }
1533 /* Check free space in inode or block */
1534 if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1535 if (free < sizeof(struct ocfs2_xattr_entry) +
1536 OCFS2_XATTR_SIZE(name_len) +
1537 OCFS2_XATTR_ROOT_SIZE) {
1538 ret = -ENOSPC;
1539 goto out;
1540 }
1541 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1542 xi_l.value = (void *)&def_xv;
1543 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1544 } else if (xi->value) {
1545 if (free < sizeof(struct ocfs2_xattr_entry) +
1546 OCFS2_XATTR_SIZE(name_len) +
1547 OCFS2_XATTR_SIZE(xi->value_len)) {
1548 ret = -ENOSPC;
1549 goto out;
1550 }
1551 }
1552
1553 if (!xs->not_found) {
1554 /* For existing extended attribute */
1555 size_t size = OCFS2_XATTR_SIZE(name_len) +
1556 OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1557 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1558 void *val = xs->base + offs;
1559
1560 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1561 /* Replace existing local xattr with tree root */
1562 ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
512620f4 1563 ctxt, &vb, offs);
cf1d6c76
TY		 1564 if (ret < 0)
1565 mlog_errno(ret);
1566 goto out;
1567 } else if (!ocfs2_xattr_is_local(xs->here)) {
1568 /* For existing xattr which has value outside */
512620f4
JB		 1569 vb.vb_xv = (struct ocfs2_xattr_value_root *)
1570 (val + OCFS2_XATTR_SIZE(name_len));
cf1d6c76
TY		 1571
1572 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1573 /*
1574 * If new value need set outside also,
1575 * first truncate old value to new value,
1576 * then set new value with set_value_outside().
1577 */
1578 ret = ocfs2_xattr_value_truncate(inode,
b3e5d379 1579 &vb,
78f30c31
TM		 1580 xi->value_len,
1581 ctxt);
cf1d6c76
TY		 1582 if (ret < 0) {
1583 mlog_errno(ret);
1584 goto out;
1585 }
1586
85db90e7
TM		 1587 ret = ocfs2_xattr_update_entry(inode,
1588 handle,
1589 xi,
1590 xs,
0c748e95 1591 &vb,
85db90e7 1592 offs);
cf1d6c76
TY		 1593 if (ret < 0) {
1594 mlog_errno(ret);
1595 goto out;
1596 }
1597
85db90e7
TM		 1598 ret = __ocfs2_xattr_set_value_outside(inode,
1599 handle,
b3e5d379 1600 vb.vb_xv,
85db90e7
TM		 1601 xi->value,
1602 xi->value_len);
cf1d6c76
TY		 1603 if (ret < 0)
1604 mlog_errno(ret);
1605 goto out;
1606 } else {
1607 /*
1608 * If new value need set in local,
1609 * just trucate old value to zero.
1610 */
1611 ret = ocfs2_xattr_value_truncate(inode,
b3e5d379 1612 &vb,
85db90e7
TM		 1613 0,
1614 ctxt);
cf1d6c76
TY		 1615 if (ret < 0)
1616 mlog_errno(ret);
1617 }
1618 }
1619 }
1620
0cf2f763 1621 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), xs->inode_bh,
512620f4 1622 OCFS2_JOURNAL_ACCESS_WRITE);
cf1d6c76
TY		 1623 if (ret) {
1624 mlog_errno(ret);
85db90e7 1625 goto out;
cf1d6c76
TY		 1626 }
1627
1628 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
0cf2f763 1629 ret = vb.vb_access(handle, INODE_CACHE(inode), vb.vb_bh,
512620f4 1630 OCFS2_JOURNAL_ACCESS_WRITE);
cf1d6c76
TY		 1631 if (ret) {
1632 mlog_errno(ret);
85db90e7 1633 goto out;
cf1d6c76
TY		 1634 }
1635 }
1636
1637 /*
1638 * Set value in local, include set tree root in local.
1639 * This is the first step for value size >INLINE_SIZE.
1640 */
1641 ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1642
1643 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1644 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1645 if (ret < 0) {
1646 mlog_errno(ret);
85db90e7 1647 goto out;
cf1d6c76
TY		 1648 }
1649 }
1650
1651 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1652 (flag & OCFS2_INLINE_XATTR_FL)) {
1653 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1654 unsigned int xattrsize = osb->s_xattr_inline_size;
1655
1656 /*
1657 * Adjust extent record count or inline data size
1658 * to reserve space for extended attribute.
1659 */
1660 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1661 struct ocfs2_inline_data *idata = &di->id2.i_data;
1662 le16_add_cpu(&idata->id_count, -xattrsize);
1663 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1664 struct ocfs2_extent_list *el = &di->id2.i_list;
1665 le16_add_cpu(&el->l_count, -(xattrsize /
1666 sizeof(struct ocfs2_extent_rec)));
1667 }
1668 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1669 }
1670 /* Update xattr flag */
1671 spin_lock(&oi->ip_lock);
1672 oi->ip_dyn_features |= flag;
1673 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1674 spin_unlock(&oi->ip_lock);
cf1d6c76
TY		 1675
1676 ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1677 if (ret < 0)
1678 mlog_errno(ret);
1679
cf1d6c76
TY		 1680 if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1681 /*
1682 * Set value outside in B tree.
1683 * This is the second step for value size > INLINE_SIZE.
1684 */
1685 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
512620f4
JB		 1686 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, ctxt,
1687 &vb, offs);
cf1d6c76
TY		 1688 if (ret < 0) {
1689 int ret2;
1690
1691 mlog_errno(ret);
1692 /*
1693 * If set value outside failed, we have to clean
1694 * the junk tree root we have already set in local.
1695 */
85db90e7 1696 ret2 = ocfs2_xattr_cleanup(inode, ctxt->handle,
512620f4 1697 xi, xs, &vb, offs);
cf1d6c76
TY		 1698 if (ret2 < 0)
1699 mlog_errno(ret2);
1700 }
1701 }
1702out:
1703 return ret;
cf1d6c76
TY		 1704}
1705
cf1d6c76 1706static int ocfs2_remove_value_outside(struct inode*inode,
4311901d 1707 struct ocfs2_xattr_value_buf *vb,
cf1d6c76
TY		 1708 struct ocfs2_xattr_header *header)
1709{
1710 int ret = 0, i;
78f30c31
TM		 1711 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1712 struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
1713
1714 ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
cf1d6c76 1715
a90714c1
JK		 1716 ctxt.handle = ocfs2_start_trans(osb,
1717 ocfs2_remove_extent_credits(osb->sb));
85db90e7
TM		 1718 if (IS_ERR(ctxt.handle)) {
1719 ret = PTR_ERR(ctxt.handle);
1720 mlog_errno(ret);
1721 goto out;
1722 }
1723
cf1d6c76
TY		 1724 for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1725 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1726
1727 if (!ocfs2_xattr_is_local(entry)) {
cf1d6c76
TY		 1728 void *val;
1729
1730 val = (void *)header +
1731 le16_to_cpu(entry->xe_name_offset);
4311901d 1732 vb->vb_xv = (struct ocfs2_xattr_value_root *)
cf1d6c76 1733 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
4311901d 1734 ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
cf1d6c76
TY		 1735 if (ret < 0) {
1736 mlog_errno(ret);
78f30c31 1737 break;
cf1d6c76
TY		 1738 }
1739 }
1740 }
1741
85db90e7 1742 ocfs2_commit_trans(osb, ctxt.handle);
78f30c31
TM		 1743 ocfs2_schedule_truncate_log_flush(osb, 1);
1744 ocfs2_run_deallocs(osb, &ctxt.dealloc);
85db90e7 1745out:
cf1d6c76
TY		 1746 return ret;
1747}
1748
1749static int ocfs2_xattr_ibody_remove(struct inode *inode,
1750 struct buffer_head *di_bh)
1751{
1752
1753 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1754 struct ocfs2_xattr_header *header;
1755 int ret;
4311901d
JB		 1756 struct ocfs2_xattr_value_buf vb = {
1757 .vb_bh = di_bh,
1758 .vb_access = ocfs2_journal_access_di,
1759 };
cf1d6c76
TY		 1760
1761 header = (struct ocfs2_xattr_header *)
1762 ((void *)di + inode->i_sb->s_blocksize -
1763 le16_to_cpu(di->i_xattr_inline_size));
1764
4311901d 1765 ret = ocfs2_remove_value_outside(inode, &vb, header);
cf1d6c76
TY		 1766
1767 return ret;
1768}
1769
1770static int ocfs2_xattr_block_remove(struct inode *inode,
1771 struct buffer_head *blk_bh)
1772{
1773 struct ocfs2_xattr_block *xb;
cf1d6c76 1774 int ret = 0;
4311901d
JB		 1775 struct ocfs2_xattr_value_buf vb = {
1776 .vb_bh = blk_bh,
1777 .vb_access = ocfs2_journal_access_xb,
1778 };
cf1d6c76
TY		 1779
1780 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
a3944256
TM		 1781 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1782 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
4311901d 1783 ret = ocfs2_remove_value_outside(inode, &vb, header);
a3944256
TM		 1784 } else
1785 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
cf1d6c76
TY		 1786
1787 return ret;
1788}
1789
08413899
TM		 1790static int ocfs2_xattr_free_block(struct inode *inode,
1791 u64 block)
1792{
1793 struct inode *xb_alloc_inode;
1794 struct buffer_head *xb_alloc_bh = NULL;
1795 struct buffer_head *blk_bh = NULL;
1796 struct ocfs2_xattr_block *xb;
1797 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1798 handle_t *handle;
1799 int ret = 0;
1800 u64 blk, bg_blkno;
1801 u16 bit;
1802
4ae1d69b 1803 ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
08413899
TM		 1804 if (ret < 0) {
1805 mlog_errno(ret);
1806 goto out;
1807 }
1808
08413899
TM		 1809 ret = ocfs2_xattr_block_remove(inode, blk_bh);
1810 if (ret < 0) {
1811 mlog_errno(ret);
1812 goto out;
1813 }
1814
4ae1d69b 1815 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
08413899
TM		 1816 blk = le64_to_cpu(xb->xb_blkno);
1817 bit = le16_to_cpu(xb->xb_suballoc_bit);
1818 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1819
1820 xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1821 EXTENT_ALLOC_SYSTEM_INODE,
1822 le16_to_cpu(xb->xb_suballoc_slot));
1823 if (!xb_alloc_inode) {
1824 ret = -ENOMEM;
1825 mlog_errno(ret);
1826 goto out;
1827 }
1828 mutex_lock(&xb_alloc_inode->i_mutex);
1829
1830 ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1831 if (ret < 0) {
1832 mlog_errno(ret);
1833 goto out_mutex;
1834 }
1835
1836 handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
1837 if (IS_ERR(handle)) {
1838 ret = PTR_ERR(handle);
1839 mlog_errno(ret);
1840 goto out_unlock;
1841 }
1842
1843 ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1844 bit, bg_blkno, 1);
1845 if (ret < 0)
1846 mlog_errno(ret);
1847
1848 ocfs2_commit_trans(osb, handle);
1849out_unlock:
1850 ocfs2_inode_unlock(xb_alloc_inode, 1);
1851 brelse(xb_alloc_bh);
1852out_mutex:
1853 mutex_unlock(&xb_alloc_inode->i_mutex);
1854 iput(xb_alloc_inode);
1855out:
1856 brelse(blk_bh);
1857 return ret;
1858}
1859
cf1d6c76
TY		 1860/*
1861 * ocfs2_xattr_remove()
1862 *
1863 * Free extended attribute resources associated with this inode.
1864 */
1865int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1866{
cf1d6c76
TY		 1867 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1868 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1869 handle_t *handle;
1870 int ret;
1871
8154da3d
TY		 1872 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1873 return 0;
1874
cf1d6c76
TY		 1875 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1876 return 0;
1877
1878 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1879 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1880 if (ret < 0) {
1881 mlog_errno(ret);
1882 goto out;
1883 }
1884 }
cf1d6c76 1885
08413899
TM		 1886 if (di->i_xattr_loc) {
1887 ret = ocfs2_xattr_free_block(inode,
1888 le64_to_cpu(di->i_xattr_loc));
cf1d6c76
TY		 1889 if (ret < 0) {
1890 mlog_errno(ret);
1891 goto out;
1892 }
1893 }
1894
1895 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1896 OCFS2_INODE_UPDATE_CREDITS);
1897 if (IS_ERR(handle)) {
1898 ret = PTR_ERR(handle);
1899 mlog_errno(ret);
1900 goto out;
1901 }
0cf2f763 1902 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
84008972 1903 OCFS2_JOURNAL_ACCESS_WRITE);
cf1d6c76
TY		 1904 if (ret) {
1905 mlog_errno(ret);
1906 goto out_commit;
1907 }
1908
08413899 1909 di->i_xattr_loc = 0;
cf1d6c76
TY		 1910
1911 spin_lock(&oi->ip_lock);
1912 oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1913 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1914 spin_unlock(&oi->ip_lock);
1915
1916 ret = ocfs2_journal_dirty(handle, di_bh);
1917 if (ret < 0)
1918 mlog_errno(ret);
1919out_commit:
1920 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1921out:
cf1d6c76
TY		 1922 return ret;
1923}
1924
1925static int ocfs2_xattr_has_space_inline(struct inode *inode,
1926 struct ocfs2_dinode *di)
1927{
1928 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1929 unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1930 int free;
1931
1932 if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1933 return 0;
1934
1935 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1936 struct ocfs2_inline_data *idata = &di->id2.i_data;
1937 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1938 } else if (ocfs2_inode_is_fast_symlink(inode)) {
1939 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1940 le64_to_cpu(di->i_size);
1941 } else {
1942 struct ocfs2_extent_list *el = &di->id2.i_list;
1943 free = (le16_to_cpu(el->l_count) -
1944 le16_to_cpu(el->l_next_free_rec)) *
1945 sizeof(struct ocfs2_extent_rec);
1946 }
1947 if (free >= xattrsize)
1948 return 1;
1949
1950 return 0;
1951}
1952
1953/*
1954 * ocfs2_xattr_ibody_find()
1955 *
1956 * Find extended attribute in inode block and
1957 * fill search info into struct ocfs2_xattr_search.
1958 */
1959static int ocfs2_xattr_ibody_find(struct inode *inode,
1960 int name_index,
1961 const char *name,
1962 struct ocfs2_xattr_search *xs)
1963{
1964 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1965 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1966 int ret;
1967 int has_space = 0;
1968
1969 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1970 return 0;
1971
1972 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1973 down_read(&oi->ip_alloc_sem);
1974 has_space = ocfs2_xattr_has_space_inline(inode, di);
1975 up_read(&oi->ip_alloc_sem);
1976 if (!has_space)
1977 return 0;
1978 }
1979
1980 xs->xattr_bh = xs->inode_bh;
1981 xs->end = (void *)di + inode->i_sb->s_blocksize;
1982 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1983 xs->header = (struct ocfs2_xattr_header *)
1984 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1985 else
1986 xs->header = (struct ocfs2_xattr_header *)
1987 (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
1988 xs->base = (void *)xs->header;
1989 xs->here = xs->header->xh_entries;
1990
1991 /* Find the named attribute. */
1992 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1993 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1994 if (ret && ret != -ENODATA)
1995 return ret;
1996 xs->not_found = ret;
1997 }
1998
1999 return 0;
2000}
2001
2002/*
2003 * ocfs2_xattr_ibody_set()
2004 *
2005 * Set, replace or remove an extended attribute into inode block.
2006 *
2007 */
2008static int ocfs2_xattr_ibody_set(struct inode *inode,
2009 struct ocfs2_xattr_info *xi,
78f30c31
TM		 2010 struct ocfs2_xattr_search *xs,
2011 struct ocfs2_xattr_set_ctxt *ctxt)
cf1d6c76
TY		 2012{
2013 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2014 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2015 int ret;
2016
2017 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2018 return -ENOSPC;
2019
2020 down_write(&oi->ip_alloc_sem);
2021 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2022 if (!ocfs2_xattr_has_space_inline(inode, di)) {
2023 ret = -ENOSPC;
2024 goto out;
2025 }
2026 }
2027
78f30c31 2028 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
cf1d6c76
TY		 2029 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
2030out:
2031 up_write(&oi->ip_alloc_sem);
2032
2033 return ret;
2034}
2035
2036/*
2037 * ocfs2_xattr_block_find()
2038 *
2039 * Find extended attribute in external block and
2040 * fill search info into struct ocfs2_xattr_search.
2041 */
2042static int ocfs2_xattr_block_find(struct inode *inode,
2043 int name_index,
2044 const char *name,
2045 struct ocfs2_xattr_search *xs)
2046{
2047 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2048 struct buffer_head *blk_bh = NULL;
589dc260 2049 struct ocfs2_xattr_block *xb;
cf1d6c76
TY		 2050 int ret = 0;
2051
2052 if (!di->i_xattr_loc)
2053 return ret;
2054
4ae1d69b
JB		 2055 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2056 &blk_bh);
cf1d6c76
TY		 2057 if (ret < 0) {
2058 mlog_errno(ret);
2059 return ret;
2060 }
f6087fb7 2061
cf1d6c76 2062 xs->xattr_bh = blk_bh;
4ae1d69b 2063 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
589dc260
TM		 2064
2065 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2066 xs->header = &xb->xb_attrs.xb_header;
2067 xs->base = (void *)xs->header;
2068 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2069 xs->here = xs->header->xh_entries;
2070
2071 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2072 } else
2073 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2074 name_index,
2075 name, xs);
cf1d6c76 2076
cf1d6c76
TY		 2077 if (ret && ret != -ENODATA) {
2078 xs->xattr_bh = NULL;
2079 goto cleanup;
2080 }
2081 xs->not_found = ret;
2082 return 0;
cf1d6c76
TY		 2083cleanup:
2084 brelse(blk_bh);
2085
2086 return ret;
2087}
2088
2089/*
2090 * ocfs2_xattr_block_set()
2091 *
2092 * Set, replace or remove an extended attribute into external block.
2093 *
2094 */
2095static int ocfs2_xattr_block_set(struct inode *inode,
2096 struct ocfs2_xattr_info *xi,
78f30c31
TM		 2097 struct ocfs2_xattr_search *xs,
2098 struct ocfs2_xattr_set_ctxt *ctxt)
cf1d6c76
TY		 2099{
2100 struct buffer_head *new_bh = NULL;
2101 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2102 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
85db90e7 2103 handle_t *handle = ctxt->handle;
cf1d6c76
TY		 2104 struct ocfs2_xattr_block *xblk = NULL;
2105 u16 suballoc_bit_start;
2106 u32 num_got;
2107 u64 first_blkno;
2108 int ret;
2109
2110 if (!xs->xattr_bh) {
0cf2f763
JB		 2111 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2112 xs->inode_bh,
84008972 2113 OCFS2_JOURNAL_ACCESS_CREATE);
cf1d6c76
TY		 2114 if (ret < 0) {
2115 mlog_errno(ret);
85db90e7 2116 goto end;
cf1d6c76
TY		 2117 }
2118
78f30c31 2119 ret = ocfs2_claim_metadata(osb, handle, ctxt->meta_ac, 1,
cf1d6c76
TY		 2120 &suballoc_bit_start, &num_got,
2121 &first_blkno);
2122 if (ret < 0) {
2123 mlog_errno(ret);
85db90e7 2124 goto end;
cf1d6c76
TY		 2125 }
2126
2127 new_bh = sb_getblk(inode->i_sb, first_blkno);
8cb471e8 2128 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
cf1d6c76 2129
0cf2f763
JB		 2130 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode),
2131 new_bh,
84008972 2132 OCFS2_JOURNAL_ACCESS_CREATE);
cf1d6c76
TY		 2133 if (ret < 0) {
2134 mlog_errno(ret);
85db90e7 2135 goto end;
cf1d6c76
TY		 2136 }
2137
2138 /* Initialize ocfs2_xattr_block */
2139 xs->xattr_bh = new_bh;
2140 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2141 memset(xblk, 0, inode->i_sb->s_blocksize);
2142 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2143 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
2144 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2145 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
2146 xblk->xb_blkno = cpu_to_le64(first_blkno);
2147
2148 xs->header = &xblk->xb_attrs.xb_header;
2149 xs->base = (void *)xs->header;
2150 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2151 xs->here = xs->header->xh_entries;
2152
cf1d6c76
TY		 2153 ret = ocfs2_journal_dirty(handle, new_bh);
2154 if (ret < 0) {
2155 mlog_errno(ret);
85db90e7 2156 goto end;
cf1d6c76
TY		 2157 }
2158 di->i_xattr_loc = cpu_to_le64(first_blkno);
85db90e7 2159 ocfs2_journal_dirty(handle, xs->inode_bh);
01225596
TM		 2160 } else
2161 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2162
2163 if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2164 /* Set extended attribute into external block */
78f30c31
TM		 2165 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2166 OCFS2_HAS_XATTR_FL);
01225596
TM		 2167 if (!ret || ret != -ENOSPC)
2168 goto end;
2169
78f30c31 2170 ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
01225596
TM		 2171 if (ret)
2172 goto end;
cf1d6c76
TY		 2173 }
2174
78f30c31 2175 ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
01225596
TM		 2176
2177end:
cf1d6c76
TY		 2178
2179 return ret;
2180}
2181
78f30c31
TM		 2182/* Check whether the new xattr can be inserted into the inode. */
2183static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2184 struct ocfs2_xattr_info *xi,
2185 struct ocfs2_xattr_search *xs)
2186{
2187 u64 value_size;
2188 struct ocfs2_xattr_entry *last;
2189 int free, i;
2190 size_t min_offs = xs->end - xs->base;
2191
2192 if (!xs->header)
2193 return 0;
2194
2195 last = xs->header->xh_entries;
2196
2197 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
2198 size_t offs = le16_to_cpu(last->xe_name_offset);
2199 if (offs < min_offs)
2200 min_offs = offs;
2201 last += 1;
2202 }
2203
4442f518 2204 free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
78f30c31
TM		 2205 if (free < 0)
2206 return 0;
2207
2208 BUG_ON(!xs->not_found);
2209
2210 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2211 value_size = OCFS2_XATTR_ROOT_SIZE;
2212 else
2213 value_size = OCFS2_XATTR_SIZE(xi->value_len);
2214
2215 if (free >= sizeof(struct ocfs2_xattr_entry) +
2216 OCFS2_XATTR_SIZE(strlen(xi->name)) + value_size)
2217 return 1;
2218
2219 return 0;
2220}
2221
2222static int ocfs2_calc_xattr_set_need(struct inode *inode,
2223 struct ocfs2_dinode *di,
2224 struct ocfs2_xattr_info *xi,
2225 struct ocfs2_xattr_search *xis,
2226 struct ocfs2_xattr_search *xbs,
2227 int *clusters_need,
85db90e7
TM		 2228 int *meta_need,
2229 int *credits_need)
78f30c31
TM		 2230{
2231 int ret = 0, old_in_xb = 0;
85db90e7 2232 int clusters_add = 0, meta_add = 0, credits = 0;
78f30c31
TM		 2233 struct buffer_head *bh = NULL;
2234 struct ocfs2_xattr_block *xb = NULL;
2235 struct ocfs2_xattr_entry *xe = NULL;
2236 struct ocfs2_xattr_value_root *xv = NULL;
2237 char *base = NULL;
2238 int name_offset, name_len = 0;
2239 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2240 xi->value_len);
2241 u64 value_size;
2242
71d548a6
TM		 2243 /*
2244 * Calculate the clusters we need to write.
2245 * No matter whether we replace an old one or add a new one,
2246 * we need this for writing.
2247 */
2248 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2249 credits += new_clusters *
2250 ocfs2_clusters_to_blocks(inode->i_sb, 1);
2251
78f30c31 2252 if (xis->not_found && xbs->not_found) {
85db90e7
TM		 2253 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2254
2255 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
78f30c31 2256 clusters_add += new_clusters;
85db90e7
TM		 2257 credits += ocfs2_calc_extend_credits(inode->i_sb,
2258 &def_xv.xv.xr_list,
2259 new_clusters);
2260 }
78f30c31
TM		 2261
2262 goto meta_guess;
2263 }
2264
2265 if (!xis->not_found) {
2266 xe = xis->here;
2267 name_offset = le16_to_cpu(xe->xe_name_offset);
2268 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2269 base = xis->base;
85db90e7 2270 credits += OCFS2_INODE_UPDATE_CREDITS;
78f30c31 2271 } else {
970e4936 2272 int i, block_off = 0;
78f30c31
TM		 2273 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2274 xe = xbs->here;
2275 name_offset = le16_to_cpu(xe->xe_name_offset);
2276 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2277 i = xbs->here - xbs->header->xh_entries;
2278 old_in_xb = 1;
2279
2280 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2281 ret = ocfs2_xattr_bucket_get_name_value(inode,
2282 bucket_xh(xbs->bucket),
2283 i, &block_off,
2284 &name_offset);
2285 base = bucket_block(xbs->bucket, block_off);
85db90e7
TM		 2286 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2287 } else {
78f30c31 2288 base = xbs->base;
85db90e7
TM		 2289 credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
2290 }
2291 }
2292
2293 /*
2294 * delete a xattr doesn't need metadata and cluster allocation.
2295 * so just calculate the credits and return.
2296 *
2297 * The credits for removing the value tree will be extended
2298 * by ocfs2_remove_extent itself.
2299 */
2300 if (!xi->value) {
2301 if (!ocfs2_xattr_is_local(xe))
a90714c1 2302 credits += ocfs2_remove_extent_credits(inode->i_sb);
85db90e7
TM		 2303
2304 goto out;
78f30c31
TM		 2305 }
2306
2307 /* do cluster allocation guess first. */
2308 value_size = le64_to_cpu(xe->xe_value_size);
2309
2310 if (old_in_xb) {
2311 /*
2312 * In xattr set, we always try to set the xe in inode first,
2313 * so if it can be inserted into inode successfully, the old
2314 * one will be removed from the xattr block, and this xattr
2315 * will be inserted into inode as a new xattr in inode.
2316 */
2317 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
2318 clusters_add += new_clusters;
a90714c1 2319 credits += ocfs2_remove_extent_credits(inode->i_sb) +
85db90e7
TM		 2320 OCFS2_INODE_UPDATE_CREDITS;
2321 if (!ocfs2_xattr_is_local(xe))
2322 credits += ocfs2_calc_extend_credits(
2323 inode->i_sb,
2324 &def_xv.xv.xr_list,
2325 new_clusters);
78f30c31
TM		 2326 goto out;
2327 }
2328 }
2329
2330 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2331 /* the new values will be stored outside. */
2332 u32 old_clusters = 0;
2333
2334 if (!ocfs2_xattr_is_local(xe)) {
2335 old_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2336 value_size);
2337 xv = (struct ocfs2_xattr_value_root *)
2338 (base + name_offset + name_len);
97aff52a 2339 value_size = OCFS2_XATTR_ROOT_SIZE;
78f30c31
TM		 2340 } else
2341 xv = &def_xv.xv;
2342
85db90e7 2343 if (old_clusters >= new_clusters) {
a90714c1 2344 credits += ocfs2_remove_extent_credits(inode->i_sb);
78f30c31 2345 goto out;
85db90e7 2346 } else {
78f30c31
TM		 2347 meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
2348 clusters_add += new_clusters - old_clusters;
85db90e7
TM		 2349 credits += ocfs2_calc_extend_credits(inode->i_sb,
2350 &xv->xr_list,
2351 new_clusters -
2352 old_clusters);
97aff52a
TM		 2353 if (value_size >= OCFS2_XATTR_ROOT_SIZE)
2354 goto out;
78f30c31
TM		 2355 }
2356 } else {
2357 /*
2358 * Now the new value will be stored inside. So if the new
2359 * value is smaller than the size of value root or the old
2360 * value, we don't need any allocation, otherwise we have
2361 * to guess metadata allocation.
2362 */
2363 if ((ocfs2_xattr_is_local(xe) && value_size >= xi->value_len) ||
2364 (!ocfs2_xattr_is_local(xe) &&
2365 OCFS2_XATTR_ROOT_SIZE >= xi->value_len))
2366 goto out;
2367 }
2368
2369meta_guess:
2370 /* calculate metadata allocation. */
2371 if (di->i_xattr_loc) {
2372 if (!xbs->xattr_bh) {
4ae1d69b
JB		 2373 ret = ocfs2_read_xattr_block(inode,
2374 le64_to_cpu(di->i_xattr_loc),
2375 &bh);
78f30c31
TM		 2376 if (ret) {
2377 mlog_errno(ret);
2378 goto out;
2379 }
2380
2381 xb = (struct ocfs2_xattr_block *)bh->b_data;
2382 } else
2383 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2384
90cb546c
TM		 2385 /*
2386 * If there is already an xattr tree, good, we can calculate
2387 * like other b-trees. Otherwise we may have the chance of
2388 * create a tree, the credit calculation is borrowed from
2389 * ocfs2_calc_extend_credits with root_el = NULL. And the
2390 * new tree will be cluster based, so no meta is needed.
2391 */
78f30c31
TM		 2392 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2393 struct ocfs2_extent_list *el =
2394 &xb->xb_attrs.xb_root.xt_list;
2395 meta_add += ocfs2_extend_meta_needed(el);
85db90e7
TM		 2396 credits += ocfs2_calc_extend_credits(inode->i_sb,
2397 el, 1);
90cb546c
TM		 2398 } else
2399 credits += OCFS2_SUBALLOC_ALLOC + 1;
78f30c31
TM		 2400
2401 /*
2402 * This cluster will be used either for new bucket or for
2403 * new xattr block.
2404 * If the cluster size is the same as the bucket size, one
2405 * more is needed since we may need to extend the bucket
2406 * also.
2407 */
2408 clusters_add += 1;
85db90e7 2409 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
78f30c31 2410 if (OCFS2_XATTR_BUCKET_SIZE ==
85db90e7
TM		 2411 OCFS2_SB(inode->i_sb)->s_clustersize) {
2412 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
78f30c31 2413 clusters_add += 1;
85db90e7
TM		 2414 }
2415 } else {
78f30c31 2416 meta_add += 1;
85db90e7
TM		 2417 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
2418 }
78f30c31
TM		 2419out:
2420 if (clusters_need)
2421 *clusters_need = clusters_add;
2422 if (meta_need)
2423 *meta_need = meta_add;
85db90e7
TM		 2424 if (credits_need)
2425 *credits_need = credits;
78f30c31
TM		 2426 brelse(bh);
2427 return ret;
2428}
2429
2430static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
2431 struct ocfs2_dinode *di,
2432 struct ocfs2_xattr_info *xi,
2433 struct ocfs2_xattr_search *xis,
2434 struct ocfs2_xattr_search *xbs,
85db90e7
TM		 2435 struct ocfs2_xattr_set_ctxt *ctxt,
2436 int *credits)
78f30c31
TM		 2437{
2438 int clusters_add, meta_add, ret;
2439 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2440
2441 memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
2442
2443 ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
2444
2445 ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
85db90e7 2446 &clusters_add, &meta_add, credits);
78f30c31
TM		 2447 if (ret) {
2448 mlog_errno(ret);
2449 return ret;
2450 }
2451
85db90e7
TM		 2452 mlog(0, "Set xattr %s, reserve meta blocks = %d, clusters = %d, "
2453 "credits = %d\n", xi->name, meta_add, clusters_add, *credits);
78f30c31
TM		 2454
2455 if (meta_add) {
2456 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
2457 &ctxt->meta_ac);
2458 if (ret) {
2459 mlog_errno(ret);
2460 goto out;
2461 }
2462 }
2463
2464 if (clusters_add) {
2465 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
2466 if (ret)
2467 mlog_errno(ret);
2468 }
2469out:
2470 if (ret) {
2471 if (ctxt->meta_ac) {
2472 ocfs2_free_alloc_context(ctxt->meta_ac);
2473 ctxt->meta_ac = NULL;
2474 }
2475
2476 /*
2477 * We cannot have an error and a non null ctxt->data_ac.
2478 */
2479 }
2480
2481 return ret;
2482}
2483
85db90e7
TM		 2484static int __ocfs2_xattr_set_handle(struct inode *inode,
2485 struct ocfs2_dinode *di,
2486 struct ocfs2_xattr_info *xi,
2487 struct ocfs2_xattr_search *xis,
2488 struct ocfs2_xattr_search *xbs,
2489 struct ocfs2_xattr_set_ctxt *ctxt)
2490{
9f868f16 2491 int ret = 0, credits, old_found;
85db90e7
TM		 2492
2493 if (!xi->value) {
2494 /* Remove existing extended attribute */
2495 if (!xis->not_found)
2496 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2497 else if (!xbs->not_found)
2498 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2499 } else {
2500 /* We always try to set extended attribute into inode first*/
2501 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2502 if (!ret && !xbs->not_found) {
2503 /*
2504 * If succeed and that extended attribute existing in
2505 * external block, then we will remove it.
2506 */
2507 xi->value = NULL;
2508 xi->value_len = 0;
2509
9f868f16 2510 old_found = xis->not_found;
85db90e7
TM		 2511 xis->not_found = -ENODATA;
2512 ret = ocfs2_calc_xattr_set_need(inode,
2513 di,
2514 xi,
2515 xis,
2516 xbs,
2517 NULL,
2518 NULL,
2519 &credits);
9f868f16 2520 xis->not_found = old_found;
85db90e7
TM		 2521 if (ret) {
2522 mlog_errno(ret);
2523 goto out;
2524 }
2525
2526 ret = ocfs2_extend_trans(ctxt->handle, credits +
2527 ctxt->handle->h_buffer_credits);
2528 if (ret) {
2529 mlog_errno(ret);
2530 goto out;
2531 }
2532 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2533 } else if (ret == -ENOSPC) {
2534 if (di->i_xattr_loc && !xbs->xattr_bh) {
2535 ret = ocfs2_xattr_block_find(inode,
2536 xi->name_index,
2537 xi->name, xbs);
2538 if (ret)
2539 goto out;
2540
9f868f16 2541 old_found = xis->not_found;
85db90e7
TM		 2542 xis->not_found = -ENODATA;
2543 ret = ocfs2_calc_xattr_set_need(inode,
2544 di,
2545 xi,
2546 xis,
2547 xbs,
2548 NULL,
2549 NULL,
2550 &credits);
9f868f16 2551 xis->not_found = old_found;
85db90e7
TM		 2552 if (ret) {
2553 mlog_errno(ret);
2554 goto out;
2555 }
2556
2557 ret = ocfs2_extend_trans(ctxt->handle, credits +
2558 ctxt->handle->h_buffer_credits);
2559 if (ret) {
2560 mlog_errno(ret);
2561 goto out;
2562 }
2563 }
2564 /*
2565 * If no space in inode, we will set extended attribute
2566 * into external block.
2567 */
2568 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2569 if (ret)
2570 goto out;
2571 if (!xis->not_found) {
2572 /*
2573 * If succeed and that extended attribute
2574 * existing in inode, we will remove it.
2575 */
2576 xi->value = NULL;
2577 xi->value_len = 0;
2578 xbs->not_found = -ENODATA;
2579 ret = ocfs2_calc_xattr_set_need(inode,
2580 di,
2581 xi,
2582 xis,
2583 xbs,
2584 NULL,
2585 NULL,
2586 &credits);
2587 if (ret) {
2588 mlog_errno(ret);
2589 goto out;
2590 }
2591
2592 ret = ocfs2_extend_trans(ctxt->handle, credits +
2593 ctxt->handle->h_buffer_credits);
2594 if (ret) {
2595 mlog_errno(ret);
2596 goto out;
2597 }
2598 ret = ocfs2_xattr_ibody_set(inode, xi,
2599 xis, ctxt);
2600 }
2601 }
2602 }
2603
4b3f6209
TM		 2604 if (!ret) {
2605 /* Update inode ctime. */
0cf2f763 2606 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
89a907af
TM		 2607 xis->inode_bh,
2608 OCFS2_JOURNAL_ACCESS_WRITE);
4b3f6209
TM		 2609 if (ret) {
2610 mlog_errno(ret);
2611 goto out;
2612 }
2613
2614 inode->i_ctime = CURRENT_TIME;
2615 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
2616 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
2617 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
2618 }
85db90e7
TM		 2619out:
2620 return ret;
2621}
2622
6c3faba4
TY		 2623/*
2624 * This function only called duing creating inode
2625 * for init security/acl xattrs of the new inode.
008aafaf 2626 * All transanction credits have been reserved in mknod.
6c3faba4
TY		 2627 */
2628int ocfs2_xattr_set_handle(handle_t *handle,
2629 struct inode *inode,
2630 struct buffer_head *di_bh,
2631 int name_index,
2632 const char *name,
2633 const void *value,
2634 size_t value_len,
2635 int flags,
2636 struct ocfs2_alloc_context *meta_ac,
2637 struct ocfs2_alloc_context *data_ac)
2638{
2639 struct ocfs2_dinode *di;
2640 int ret;
2641
2642 struct ocfs2_xattr_info xi = {
2643 .name_index = name_index,
2644 .name = name,
2645 .value = value,
2646 .value_len = value_len,
2647 };
2648
2649 struct ocfs2_xattr_search xis = {
2650 .not_found = -ENODATA,
2651 };
2652
2653 struct ocfs2_xattr_search xbs = {
2654 .not_found = -ENODATA,
2655 };
2656
2657 struct ocfs2_xattr_set_ctxt ctxt = {
2658 .handle = handle,
2659 .meta_ac = meta_ac,
2660 .data_ac = data_ac,
2661 };
2662
2663 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2664 return -EOPNOTSUPP;
2665
008aafaf
TY		 2666 /*
2667 * In extreme situation, may need xattr bucket when
2668 * block size is too small. And we have already reserved
2669 * the credits for bucket in mknod.
2670 */
2671 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
2672 xbs.bucket = ocfs2_xattr_bucket_new(inode);
2673 if (!xbs.bucket) {
2674 mlog_errno(-ENOMEM);
2675 return -ENOMEM;
2676 }
2677 }
2678
6c3faba4
TY		 2679 xis.inode_bh = xbs.inode_bh = di_bh;
2680 di = (struct ocfs2_dinode *)di_bh->b_data;
2681
2682 down_write(&OCFS2_I(inode)->ip_xattr_sem);
2683
2684 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2685 if (ret)
2686 goto cleanup;
2687 if (xis.not_found) {
2688 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2689 if (ret)
2690 goto cleanup;
2691 }
2692
2693 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2694
2695cleanup:
2696 up_write(&OCFS2_I(inode)->ip_xattr_sem);
2697 brelse(xbs.xattr_bh);
008aafaf 2698 ocfs2_xattr_bucket_free(xbs.bucket);
6c3faba4
TY		 2699
2700 return ret;
2701}
2702
cf1d6c76
TY		 2703/*
2704 * ocfs2_xattr_set()
2705 *
2706 * Set, replace or remove an extended attribute for this inode.
2707 * value is NULL to remove an existing extended attribute, else either
2708 * create or replace an extended attribute.
2709 */
2710int ocfs2_xattr_set(struct inode *inode,
2711 int name_index,
2712 const char *name,
2713 const void *value,
2714 size_t value_len,
2715 int flags)
2716{
2717 struct buffer_head *di_bh = NULL;
2718 struct ocfs2_dinode *di;
85db90e7 2719 int ret, credits;
78f30c31 2720 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
85db90e7 2721 struct inode *tl_inode = osb->osb_tl_inode;
78f30c31 2722 struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
cf1d6c76
TY		 2723
2724 struct ocfs2_xattr_info xi = {
2725 .name_index = name_index,
2726 .name = name,
2727 .value = value,
2728 .value_len = value_len,
2729 };
2730
2731 struct ocfs2_xattr_search xis = {
2732 .not_found = -ENODATA,
2733 };
2734
2735 struct ocfs2_xattr_search xbs = {
2736 .not_found = -ENODATA,
2737 };
2738
8154da3d
TY		 2739 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2740 return -EOPNOTSUPP;
2741
ba937127
JB		 2742 /*
2743 * Only xbs will be used on indexed trees. xis doesn't need a
2744 * bucket.
2745 */
2746 xbs.bucket = ocfs2_xattr_bucket_new(inode);
2747 if (!xbs.bucket) {
2748 mlog_errno(-ENOMEM);
2749 return -ENOMEM;
2750 }
2751
cf1d6c76
TY		 2752 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2753 if (ret < 0) {
2754 mlog_errno(ret);
ba937127 2755 goto cleanup_nolock;
cf1d6c76
TY		 2756 }
2757 xis.inode_bh = xbs.inode_bh = di_bh;
2758 di = (struct ocfs2_dinode *)di_bh->b_data;
2759
2760 down_write(&OCFS2_I(inode)->ip_xattr_sem);
2761 /*
2762 * Scan inode and external block to find the same name
2763 * extended attribute and collect search infomation.
2764 */
2765 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2766 if (ret)
2767 goto cleanup;
2768 if (xis.not_found) {
2769 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2770 if (ret)
2771 goto cleanup;
2772 }
2773
2774 if (xis.not_found && xbs.not_found) {
2775 ret = -ENODATA;
2776 if (flags & XATTR_REPLACE)
2777 goto cleanup;
2778 ret = 0;
2779 if (!value)
2780 goto cleanup;
2781 } else {
2782 ret = -EEXIST;
2783 if (flags & XATTR_CREATE)
2784 goto cleanup;
2785 }
2786
85db90e7
TM		 2787
2788 mutex_lock(&tl_inode->i_mutex);
2789
2790 if (ocfs2_truncate_log_needs_flush(osb)) {
2791 ret = __ocfs2_flush_truncate_log(osb);
2792 if (ret < 0) {
2793 mutex_unlock(&tl_inode->i_mutex);
2794 mlog_errno(ret);
2795 goto cleanup;
2796 }
2797 }
2798 mutex_unlock(&tl_inode->i_mutex);
2799
2800 ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
2801 &xbs, &ctxt, &credits);
78f30c31
TM		 2802 if (ret) {
2803 mlog_errno(ret);
2804 goto cleanup;
2805 }
2806
4b3f6209
TM		 2807 /* we need to update inode's ctime field, so add credit for it. */
2808 credits += OCFS2_INODE_UPDATE_CREDITS;
85db90e7
TM		 2809 ctxt.handle = ocfs2_start_trans(osb, credits);
2810 if (IS_ERR(ctxt.handle)) {
2811 ret = PTR_ERR(ctxt.handle);
2812 mlog_errno(ret);
2813 goto cleanup;
cf1d6c76 2814 }
85db90e7
TM		 2815
2816 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2817
2818 ocfs2_commit_trans(osb, ctxt.handle);
2819
78f30c31
TM		 2820 if (ctxt.data_ac)
2821 ocfs2_free_alloc_context(ctxt.data_ac);
2822 if (ctxt.meta_ac)
2823 ocfs2_free_alloc_context(ctxt.meta_ac);
2824 if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
2825 ocfs2_schedule_truncate_log_flush(osb, 1);
2826 ocfs2_run_deallocs(osb, &ctxt.dealloc);
cf1d6c76
TY		 2827cleanup:
2828 up_write(&OCFS2_I(inode)->ip_xattr_sem);
2829 ocfs2_inode_unlock(inode, 1);
ba937127 2830cleanup_nolock:
cf1d6c76
TY		 2831 brelse(di_bh);
2832 brelse(xbs.xattr_bh);
ba937127 2833 ocfs2_xattr_bucket_free(xbs.bucket);
cf1d6c76
TY		 2834
2835 return ret;
2836}
2837
0c044f0b
TM		 2838/*
2839 * Find the xattr extent rec which may contains name_hash.
2840 * e_cpos will be the first name hash of the xattr rec.
2841 * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2842 */
2843static int ocfs2_xattr_get_rec(struct inode *inode,
2844 u32 name_hash,
2845 u64 *p_blkno,
2846 u32 *e_cpos,
2847 u32 *num_clusters,
2848 struct ocfs2_extent_list *el)
2849{
2850 int ret = 0, i;
2851 struct buffer_head *eb_bh = NULL;
2852 struct ocfs2_extent_block *eb;
2853 struct ocfs2_extent_rec *rec = NULL;
2854 u64 e_blkno = 0;
2855
2856 if (el->l_tree_depth) {
2857 ret = ocfs2_find_leaf(inode, el, name_hash, &eb_bh);
2858 if (ret) {
2859 mlog_errno(ret);
2860 goto out;
2861 }
2862
2863 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2864 el = &eb->h_list;
2865
2866 if (el->l_tree_depth) {
2867 ocfs2_error(inode->i_sb,
2868 "Inode %lu has non zero tree depth in "
2869 "xattr tree block %llu\n", inode->i_ino,
2870 (unsigned long long)eb_bh->b_blocknr);
2871 ret = -EROFS;
2872 goto out;
2873 }
2874 }
2875
2876 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2877 rec = &el->l_recs[i];
2878
2879 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2880 e_blkno = le64_to_cpu(rec->e_blkno);
2881 break;
2882 }
2883 }
2884
2885 if (!e_blkno) {
2886 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2887 "record (%u, %u, 0) in xattr", inode->i_ino,
2888 le32_to_cpu(rec->e_cpos),
2889 ocfs2_rec_clusters(el, rec));
2890 ret = -EROFS;
2891 goto out;
2892 }
2893
2894 *p_blkno = le64_to_cpu(rec->e_blkno);
2895 *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2896 if (e_cpos)
2897 *e_cpos = le32_to_cpu(rec->e_cpos);
2898out:
2899 brelse(eb_bh);
2900 return ret;
2901}
2902
2903typedef int (xattr_bucket_func)(struct inode *inode,
2904 struct ocfs2_xattr_bucket *bucket,
2905 void *para);
2906
589dc260 2907static int ocfs2_find_xe_in_bucket(struct inode *inode,
e2356a3f 2908 struct ocfs2_xattr_bucket *bucket,
589dc260
TM		 2909 int name_index,
2910 const char *name,
2911 u32 name_hash,
2912 u16 *xe_index,
2913 int *found)
2914{
2915 int i, ret = 0, cmp = 1, block_off, new_offset;
e2356a3f 2916 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
589dc260
TM		 2917 size_t name_len = strlen(name);
2918 struct ocfs2_xattr_entry *xe = NULL;
589dc260
TM		 2919 char *xe_name;
2920
2921 /*
2922 * We don't use binary search in the bucket because there
2923 * may be multiple entries with the same name hash.
2924 */
2925 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2926 xe = &xh->xh_entries[i];
2927
2928 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2929 continue;
2930 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2931 break;
2932
2933 cmp = name_index - ocfs2_xattr_get_type(xe);
2934 if (!cmp)
2935 cmp = name_len - xe->xe_name_len;
2936 if (cmp)
2937 continue;
2938
2939 ret = ocfs2_xattr_bucket_get_name_value(inode,
2940 xh,
2941 i,
2942 &block_off,
2943 &new_offset);
2944 if (ret) {
2945 mlog_errno(ret);
2946 break;
2947 }
2948
970e4936 2949
e2356a3f
JB		 2950 xe_name = bucket_block(bucket, block_off) + new_offset;
2951 if (!memcmp(name, xe_name, name_len)) {
589dc260
TM		 2952 *xe_index = i;
2953 *found = 1;
2954 ret = 0;
2955 break;
2956 }
2957 }
2958
2959 return ret;
2960}
2961
2962/*
2963 * Find the specified xattr entry in a series of buckets.
2964 * This series start from p_blkno and last for num_clusters.
2965 * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2966 * the num of the valid buckets.
2967 *
2968 * Return the buffer_head this xattr should reside in. And if the xattr's
2969 * hash is in the gap of 2 buckets, return the lower bucket.
2970 */
2971static int ocfs2_xattr_bucket_find(struct inode *inode,
2972 int name_index,
2973 const char *name,
2974 u32 name_hash,
2975 u64 p_blkno,
2976 u32 first_hash,
2977 u32 num_clusters,
2978 struct ocfs2_xattr_search *xs)
2979{
2980 int ret, found = 0;
589dc260
TM		 2981 struct ocfs2_xattr_header *xh = NULL;
2982 struct ocfs2_xattr_entry *xe = NULL;
2983 u16 index = 0;
2984 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2985 int low_bucket = 0, bucket, high_bucket;
e2356a3f 2986 struct ocfs2_xattr_bucket *search;
589dc260 2987 u32 last_hash;
e2356a3f 2988 u64 blkno, lower_blkno = 0;
589dc260 2989
e2356a3f
JB		 2990 search = ocfs2_xattr_bucket_new(inode);
2991 if (!search) {
2992 ret = -ENOMEM;
2993 mlog_errno(ret);
2994 goto out;
2995 }
2996
2997 ret = ocfs2_read_xattr_bucket(search, p_blkno);
589dc260
TM		 2998 if (ret) {
2999 mlog_errno(ret);
3000 goto out;
3001 }
3002
e2356a3f 3003 xh = bucket_xh(search);
589dc260 3004 high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
589dc260 3005 while (low_bucket <= high_bucket) {
e2356a3f 3006 ocfs2_xattr_bucket_relse(search);
589dc260 3007
e2356a3f 3008 bucket = (low_bucket + high_bucket) / 2;
589dc260 3009 blkno = p_blkno + bucket * blk_per_bucket;
e2356a3f 3010 ret = ocfs2_read_xattr_bucket(search, blkno);
589dc260
TM		 3011 if (ret) {
3012 mlog_errno(ret);
3013 goto out;
3014 }
3015
e2356a3f 3016 xh = bucket_xh(search);
589dc260
TM		 3017 xe = &xh->xh_entries[0];
3018 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3019 high_bucket = bucket - 1;
3020 continue;
3021 }
3022
3023 /*
3024 * Check whether the hash of the last entry in our
5a095611
TM		 3025 * bucket is larger than the search one. for an empty
3026 * bucket, the last one is also the first one.
589dc260 3027 */
5a095611
TM		 3028 if (xh->xh_count)
3029 xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3030
589dc260
TM		 3031 last_hash = le32_to_cpu(xe->xe_name_hash);
3032
e2356a3f
JB		 3033 /* record lower_blkno which may be the insert place. */
3034 lower_blkno = blkno;
589dc260
TM		 3035
3036 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3037 low_bucket = bucket + 1;
3038 continue;
3039 }
3040
3041 /* the searched xattr should reside in this bucket if exists. */
e2356a3f 3042 ret = ocfs2_find_xe_in_bucket(inode, search,
589dc260
TM		 3043 name_index, name, name_hash,
3044 &index, &found);
3045 if (ret) {
3046 mlog_errno(ret);
3047 goto out;
3048 }
3049 break;
3050 }
3051
3052 /*
3053 * Record the bucket we have found.
3054 * When the xattr's hash value is in the gap of 2 buckets, we will
3055 * always set it to the previous bucket.
3056 */
e2356a3f
JB		 3057 if (!lower_blkno)
3058 lower_blkno = p_blkno;
3059
3060 /* This should be in cache - we just read it during the search */
3061 ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3062 if (ret) {
3063 mlog_errno(ret);
3064 goto out;
589dc260 3065 }
589dc260 3066
ba937127
JB		 3067 xs->header = bucket_xh(xs->bucket);
3068 xs->base = bucket_block(xs->bucket, 0);
589dc260
TM		 3069 xs->end = xs->base + inode->i_sb->s_blocksize;
3070
3071 if (found) {
589dc260
TM		 3072 xs->here = &xs->header->xh_entries[index];
3073 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
ba937127 3074 (unsigned long long)bucket_blkno(xs->bucket), index);
589dc260
TM		 3075 } else
3076 ret = -ENODATA;
3077
3078out:
e2356a3f 3079 ocfs2_xattr_bucket_free(search);
589dc260
TM		 3080 return ret;
3081}
3082
3083static int ocfs2_xattr_index_block_find(struct inode *inode,
3084 struct buffer_head *root_bh,
3085 int name_index,
3086 const char *name,
3087 struct ocfs2_xattr_search *xs)
3088{
3089 int ret;
3090 struct ocfs2_xattr_block *xb =
3091 (struct ocfs2_xattr_block *)root_bh->b_data;
3092 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3093 struct ocfs2_extent_list *el = &xb_root->xt_list;
3094 u64 p_blkno = 0;
3095 u32 first_hash, num_clusters = 0;
2057e5c6 3096 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
589dc260
TM		 3097
3098 if (le16_to_cpu(el->l_next_free_rec) == 0)
3099 return -ENODATA;
3100
3101 mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
3102 name, name_hash, name_index);
3103
3104 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3105 &num_clusters, el);
3106 if (ret) {
3107 mlog_errno(ret);
3108 goto out;
3109 }
3110
3111 BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3112
3113 mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
de29c085
MF		 3114 "in the rec is %u\n", num_clusters, (unsigned long long)p_blkno,
3115 first_hash);
589dc260
TM		 3116
3117 ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3118 p_blkno, first_hash, num_clusters, xs);
3119
3120out:
3121 return ret;
3122}
3123
0c044f0b
TM		 3124static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3125 u64 blkno,
3126 u32 clusters,
3127 xattr_bucket_func *func,
3128 void *para)
3129{
6dde41d9 3130 int i, ret = 0;
0c044f0b
TM		 3131 u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3132 u32 num_buckets = clusters * bpc;
ba937127 3133 struct ocfs2_xattr_bucket *bucket;
0c044f0b 3134
ba937127
JB		 3135 bucket = ocfs2_xattr_bucket_new(inode);
3136 if (!bucket) {
3137 mlog_errno(-ENOMEM);
3138 return -ENOMEM;
3139 }
0c044f0b
TM		 3140
3141 mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
de29c085 3142 clusters, (unsigned long long)blkno);
0c044f0b 3143
ba937127
JB		 3144 for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3145 ret = ocfs2_read_xattr_bucket(bucket, blkno);
0c044f0b
TM		 3146 if (ret) {
3147 mlog_errno(ret);
ba937127 3148 break;
0c044f0b
TM		 3149 }
3150
0c044f0b
TM		 3151 /*
3152 * The real bucket num in this series of blocks is stored
3153 * in the 1st bucket.
3154 */
3155 if (i == 0)
ba937127 3156 num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
0c044f0b 3157
de29c085
MF		 3158 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
3159 (unsigned long long)blkno,
ba937127 3160 le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
0c044f0b 3161 if (func) {
ba937127 3162 ret = func(inode, bucket, para);
a46fa684 3163 if (ret && ret != -ERANGE)
0c044f0b 3164 mlog_errno(ret);
ba937127 3165 /* Fall through to bucket_relse() */
0c044f0b
TM		 3166 }
3167
ba937127
JB		 3168 ocfs2_xattr_bucket_relse(bucket);
3169 if (ret)
3170 break;
0c044f0b
TM		 3171 }
3172
ba937127 3173 ocfs2_xattr_bucket_free(bucket);
0c044f0b
TM		 3174 return ret;
3175}
3176
3177struct ocfs2_xattr_tree_list {
3178 char *buffer;
3179 size_t buffer_size;
936b8834 3180 size_t result;
0c044f0b
TM		 3181};
3182
3183static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
3184 struct ocfs2_xattr_header *xh,
3185 int index,
3186 int *block_off,
3187 int *new_offset)
3188{
3189 u16 name_offset;
3190
3191 if (index < 0 || index >= le16_to_cpu(xh->xh_count))
3192 return -EINVAL;
3193
3194 name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
3195
3196 *block_off = name_offset >> inode->i_sb->s_blocksize_bits;
3197 *new_offset = name_offset % inode->i_sb->s_blocksize;
3198
3199 return 0;
3200}
3201
3202static int ocfs2_list_xattr_bucket(struct inode *inode,
3203 struct ocfs2_xattr_bucket *bucket,
3204 void *para)
3205{
936b8834 3206 int ret = 0, type;
0c044f0b 3207 struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
0c044f0b 3208 int i, block_off, new_offset;
936b8834 3209 const char *prefix, *name;
0c044f0b 3210
3e632946
JB		 3211 for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
3212 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
936b8834
TM		 3213 type = ocfs2_xattr_get_type(entry);
3214 prefix = ocfs2_xattr_prefix(type);
0c044f0b 3215
936b8834 3216 if (prefix) {
0c044f0b 3217 ret = ocfs2_xattr_bucket_get_name_value(inode,
3e632946 3218 bucket_xh(bucket),
0c044f0b
TM		 3219 i,
3220 &block_off,
3221 &new_offset);
3222 if (ret)
3223 break;
936b8834 3224
51def39f 3225 name = (const char *)bucket_block(bucket, block_off) +
936b8834
TM		 3226 new_offset;
3227 ret = ocfs2_xattr_list_entry(xl->buffer,
3228 xl->buffer_size,
3229 &xl->result,
3230 prefix, name,
3231 entry->xe_name_len);
3232 if (ret)
3233 break;
0c044f0b
TM		 3234 }
3235 }
3236
3237 return ret;
3238}
3239
3240static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
3241 struct ocfs2_xattr_tree_root *xt,
3242 char *buffer,
3243 size_t buffer_size)
3244{
3245 struct ocfs2_extent_list *el = &xt->xt_list;
3246 int ret = 0;
3247 u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
3248 u64 p_blkno = 0;
3249 struct ocfs2_xattr_tree_list xl = {
3250 .buffer = buffer,
3251 .buffer_size = buffer_size,
936b8834 3252 .result = 0,
0c044f0b
TM		 3253 };
3254
3255 if (le16_to_cpu(el->l_next_free_rec) == 0)
3256 return 0;
3257
3258 while (name_hash > 0) {
3259 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
3260 &e_cpos, &num_clusters, el);
3261 if (ret) {
3262 mlog_errno(ret);
3263 goto out;
3264 }
3265
3266 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
3267 ocfs2_list_xattr_bucket,
3268 &xl);
3269 if (ret) {
a46fa684
TM		 3270 if (ret != -ERANGE)
3271 mlog_errno(ret);
0c044f0b
TM		 3272 goto out;
3273 }
3274
3275 if (e_cpos == 0)
3276 break;
3277
3278 name_hash = e_cpos - 1;
3279 }
3280
936b8834 3281 ret = xl.result;
0c044f0b
TM		 3282out:
3283 return ret;
3284}
01225596
TM		 3285
3286static int cmp_xe(const void *a, const void *b)
3287{
3288 const struct ocfs2_xattr_entry *l = a, *r = b;
3289 u32 l_hash = le32_to_cpu(l->xe_name_hash);
3290 u32 r_hash = le32_to_cpu(r->xe_name_hash);
3291
3292 if (l_hash > r_hash)
3293 return 1;
3294 if (l_hash < r_hash)
3295 return -1;
3296 return 0;
3297}
3298
3299static void swap_xe(void *a, void *b, int size)
3300{
3301 struct ocfs2_xattr_entry *l = a, *r = b, tmp;
3302
3303 tmp = *l;
3304 memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
3305 memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
3306}
3307
3308/*
3309 * When the ocfs2_xattr_block is filled up, new bucket will be created
3310 * and all the xattr entries will be moved to the new bucket.
178eeac3
JB		 3311 * The header goes at the start of the bucket, and the names+values are
3312 * filled from the end. This is why *target starts as the last buffer.
01225596
TM		 3313 * Note: we need to sort the entries since they are not saved in order
3314 * in the ocfs2_xattr_block.
3315 */
3316static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
3317 struct buffer_head *xb_bh,
178eeac3 3318 struct ocfs2_xattr_bucket *bucket)
01225596
TM		 3319{
3320 int i, blocksize = inode->i_sb->s_blocksize;
178eeac3 3321 int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
01225596
TM		 3322 u16 offset, size, off_change;
3323 struct ocfs2_xattr_entry *xe;
3324 struct ocfs2_xattr_block *xb =
3325 (struct ocfs2_xattr_block *)xb_bh->b_data;
3326 struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
178eeac3 3327 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
01225596 3328 u16 count = le16_to_cpu(xb_xh->xh_count);
178eeac3
JB		 3329 char *src = xb_bh->b_data;
3330 char *target = bucket_block(bucket, blks - 1);
01225596
TM		 3331
3332 mlog(0, "cp xattr from block %llu to bucket %llu\n",
3333 (unsigned long long)xb_bh->b_blocknr,
178eeac3
JB		 3334 (unsigned long long)bucket_blkno(bucket));
3335
3336 for (i = 0; i < blks; i++)
3337 memset(bucket_block(bucket, i), 0, blocksize);
01225596 3338
01225596
TM		 3339 /*
3340 * Since the xe_name_offset is based on ocfs2_xattr_header,
3341 * there is a offset change corresponding to the change of
3342 * ocfs2_xattr_header's position.
3343 */
3344 off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3345 xe = &xb_xh->xh_entries[count - 1];
3346 offset = le16_to_cpu(xe->xe_name_offset) + off_change;
3347 size = blocksize - offset;
3348
3349 /* copy all the names and values. */
01225596
TM		 3350 memcpy(target + offset, src + offset, size);
3351
3352 /* Init new header now. */
3353 xh->xh_count = xb_xh->xh_count;
3354 xh->xh_num_buckets = cpu_to_le16(1);
3355 xh->xh_name_value_len = cpu_to_le16(size);
3356 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
3357
3358 /* copy all the entries. */
178eeac3 3359 target = bucket_block(bucket, 0);
01225596
TM		 3360 offset = offsetof(struct ocfs2_xattr_header, xh_entries);
3361 size = count * sizeof(struct ocfs2_xattr_entry);
3362 memcpy(target + offset, (char *)xb_xh + offset, size);
3363
3364 /* Change the xe offset for all the xe because of the move. */
3365 off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
3366 offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3367 for (i = 0; i < count; i++)
3368 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
3369
3370 mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
3371 offset, size, off_change);
3372
3373 sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
3374 cmp_xe, swap_xe);
3375}
3376
3377/*
3378 * After we move xattr from block to index btree, we have to
3379 * update ocfs2_xattr_search to the new xe and base.
3380 *
3381 * When the entry is in xattr block, xattr_bh indicates the storage place.
3382 * While if the entry is in index b-tree, "bucket" indicates the
3383 * real place of the xattr.
3384 */
178eeac3
JB		 3385static void ocfs2_xattr_update_xattr_search(struct inode *inode,
3386 struct ocfs2_xattr_search *xs,
3387 struct buffer_head *old_bh)
01225596 3388{
01225596
TM		 3389 char *buf = old_bh->b_data;
3390 struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
3391 struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
178eeac3 3392 int i;
01225596 3393
ba937127 3394 xs->header = bucket_xh(xs->bucket);
178eeac3 3395 xs->base = bucket_block(xs->bucket, 0);
01225596
TM		 3396 xs->end = xs->base + inode->i_sb->s_blocksize;
3397
178eeac3
JB		 3398 if (xs->not_found)
3399 return;
01225596 3400
178eeac3
JB		 3401 i = xs->here - old_xh->xh_entries;
3402 xs->here = &xs->header->xh_entries[i];
01225596
TM		 3403}
3404
3405static int ocfs2_xattr_create_index_block(struct inode *inode,
78f30c31
TM		 3406 struct ocfs2_xattr_search *xs,
3407 struct ocfs2_xattr_set_ctxt *ctxt)
01225596 3408{
85db90e7 3409 int ret;
01225596
TM		 3410 u32 bit_off, len;
3411 u64 blkno;
85db90e7 3412 handle_t *handle = ctxt->handle;
01225596
TM		 3413 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3414 struct ocfs2_inode_info *oi = OCFS2_I(inode);
01225596
TM		 3415 struct buffer_head *xb_bh = xs->xattr_bh;
3416 struct ocfs2_xattr_block *xb =
3417 (struct ocfs2_xattr_block *)xb_bh->b_data;
3418 struct ocfs2_xattr_tree_root *xr;
3419 u16 xb_flags = le16_to_cpu(xb->xb_flags);
01225596
TM		 3420
3421 mlog(0, "create xattr index block for %llu\n",
3422 (unsigned long long)xb_bh->b_blocknr);
3423
3424 BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
178eeac3 3425 BUG_ON(!xs->bucket);
01225596 3426
01225596
TM		 3427 /*
3428 * XXX:
3429 * We can use this lock for now, and maybe move to a dedicated mutex
3430 * if performance becomes a problem later.
3431 */
3432 down_write(&oi->ip_alloc_sem);
3433
0cf2f763 3434 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), xb_bh,
84008972 3435 OCFS2_JOURNAL_ACCESS_WRITE);
01225596
TM		 3436 if (ret) {
3437 mlog_errno(ret);
85db90e7 3438 goto out;
01225596
TM		 3439 }
3440
78f30c31
TM		 3441 ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac,
3442 1, 1, &bit_off, &len);
01225596
TM		 3443 if (ret) {
3444 mlog_errno(ret);
85db90e7 3445 goto out;
01225596
TM		 3446 }
3447
3448 /*
3449 * The bucket may spread in many blocks, and
3450 * we will only touch the 1st block and the last block
3451 * in the whole bucket(one for entry and one for data).
3452 */
3453 blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
3454
de29c085
MF		 3455 mlog(0, "allocate 1 cluster from %llu to xattr block\n",
3456 (unsigned long long)blkno);
01225596 3457
178eeac3
JB		 3458 ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
3459 if (ret) {
01225596 3460 mlog_errno(ret);
85db90e7 3461 goto out;
01225596
TM		 3462 }
3463
178eeac3
JB		 3464 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
3465 OCFS2_JOURNAL_ACCESS_CREATE);
01225596
TM		 3466 if (ret) {
3467 mlog_errno(ret);
85db90e7 3468 goto out;
01225596
TM		 3469 }
3470
178eeac3
JB		 3471 ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
3472 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
01225596 3473
178eeac3 3474 ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
01225596
TM		 3475
3476 /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
3477 memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
3478 offsetof(struct ocfs2_xattr_block, xb_attrs));
3479
3480 xr = &xb->xb_attrs.xb_root;
3481 xr->xt_clusters = cpu_to_le32(1);
3482 xr->xt_last_eb_blk = 0;
3483 xr->xt_list.l_tree_depth = 0;
3484 xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
3485 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
3486
3487 xr->xt_list.l_recs[0].e_cpos = 0;
3488 xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
3489 xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
3490
3491 xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
3492
85db90e7 3493 ocfs2_journal_dirty(handle, xb_bh);
01225596 3494
85db90e7 3495out:
01225596
TM		 3496 up_write(&oi->ip_alloc_sem);
3497
01225596
TM		 3498 return ret;
3499}
3500
3501static int cmp_xe_offset(const void *a, const void *b)
3502{
3503 const struct ocfs2_xattr_entry *l = a, *r = b;
3504 u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
3505 u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
3506
3507 if (l_name_offset < r_name_offset)
3508 return 1;
3509 if (l_name_offset > r_name_offset)
3510 return -1;
3511 return 0;
3512}
3513
3514/*
3515 * defrag a xattr bucket if we find that the bucket has some
3516 * holes beteen name/value pairs.
3517 * We will move all the name/value pairs to the end of the bucket
3518 * so that we can spare some space for insertion.
3519 */
3520static int ocfs2_defrag_xattr_bucket(struct inode *inode,
85db90e7 3521 handle_t *handle,
01225596
TM		 3522 struct ocfs2_xattr_bucket *bucket)
3523{
3524 int ret, i;
3525 size_t end, offset, len, value_len;
3526 struct ocfs2_xattr_header *xh;
3527 char *entries, *buf, *bucket_buf = NULL;
9c7759aa 3528 u64 blkno = bucket_blkno(bucket);
01225596 3529 u16 xh_free_start;
01225596 3530 size_t blocksize = inode->i_sb->s_blocksize;
01225596 3531 struct ocfs2_xattr_entry *xe;
01225596
TM		 3532
3533 /*
3534 * In order to make the operation more efficient and generic,
3535 * we copy all the blocks into a contiguous memory and do the
3536 * defragment there, so if anything is error, we will not touch
3537 * the real block.
3538 */
3539 bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
3540 if (!bucket_buf) {
3541 ret = -EIO;
3542 goto out;
3543 }
3544
3545 buf = bucket_buf;
1c32a2fd
TM		 3546 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3547 memcpy(buf, bucket_block(bucket, i), blocksize);
01225596 3548
1c32a2fd 3549 ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
161d6f30
JB		 3550 OCFS2_JOURNAL_ACCESS_WRITE);
3551 if (ret < 0) {
3552 mlog_errno(ret);
85db90e7 3553 goto out;
01225596
TM		 3554 }
3555
3556 xh = (struct ocfs2_xattr_header *)bucket_buf;
3557 entries = (char *)xh->xh_entries;
3558 xh_free_start = le16_to_cpu(xh->xh_free_start);
3559
3560 mlog(0, "adjust xattr bucket in %llu, count = %u, "
3561 "xh_free_start = %u, xh_name_value_len = %u.\n",
de29c085
MF		 3562 (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
3563 xh_free_start, le16_to_cpu(xh->xh_name_value_len));
01225596
TM		 3564
3565 /*
3566 * sort all the entries by their offset.
3567 * the largest will be the first, so that we can
3568 * move them to the end one by one.
3569 */
3570 sort(entries, le16_to_cpu(xh->xh_count),
3571 sizeof(struct ocfs2_xattr_entry),
3572 cmp_xe_offset, swap_xe);
3573
3574 /* Move all name/values to the end of the bucket. */
3575 xe = xh->xh_entries;
3576 end = OCFS2_XATTR_BUCKET_SIZE;
3577 for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
3578 offset = le16_to_cpu(xe->xe_name_offset);
3579 if (ocfs2_xattr_is_local(xe))
3580 value_len = OCFS2_XATTR_SIZE(
3581 le64_to_cpu(xe->xe_value_size));
3582 else
3583 value_len = OCFS2_XATTR_ROOT_SIZE;
3584 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
3585
3586 /*
3587 * We must make sure that the name/value pair
3588 * exist in the same block. So adjust end to
3589 * the previous block end if needed.
3590 */
3591 if (((end - len) / blocksize !=
3592 (end - 1) / blocksize))
3593 end = end - end % blocksize;
3594
3595 if (end > offset + len) {
3596 memmove(bucket_buf + end - len,
3597 bucket_buf + offset, len);
3598 xe->xe_name_offset = cpu_to_le16(end - len);
3599 }
3600
3601 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
3602 "bucket %llu\n", (unsigned long long)blkno);
3603
3604 end -= len;
3605 }
3606
3607 mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
3608 "bucket %llu\n", (unsigned long long)blkno);
3609
3610 if (xh_free_start == end)
85db90e7 3611 goto out;
01225596
TM		 3612
3613 memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
3614 xh->xh_free_start = cpu_to_le16(end);
3615
3616 /* sort the entries by their name_hash. */
3617 sort(entries, le16_to_cpu(xh->xh_count),
3618 sizeof(struct ocfs2_xattr_entry),
3619 cmp_xe, swap_xe);
3620
3621 buf = bucket_buf;
1c32a2fd
TM		 3622 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3623 memcpy(bucket_block(bucket, i), buf, blocksize);
3624 ocfs2_xattr_bucket_journal_dirty(handle, bucket);
01225596 3625
01225596 3626out:
01225596
TM		 3627 kfree(bucket_buf);
3628 return ret;
3629}
3630
3631/*
b5c03e74
JB		 3632 * prev_blkno points to the start of an existing extent. new_blkno
3633 * points to a newly allocated extent. Because we know each of our
3634 * clusters contains more than bucket, we can easily split one cluster
3635 * at a bucket boundary. So we take the last cluster of the existing
3636 * extent and split it down the middle. We move the last half of the
3637 * buckets in the last cluster of the existing extent over to the new
3638 * extent.
01225596 3639 *
b5c03e74
JB		 3640 * first_bh is the buffer at prev_blkno so we can update the existing
3641 * extent's bucket count. header_bh is the bucket were we were hoping
3642 * to insert our xattr. If the bucket move places the target in the new
3643 * extent, we'll update first_bh and header_bh after modifying the old
3644 * extent.
3645 *
3646 * first_hash will be set as the 1st xe's name_hash in the new extent.
01225596
TM		 3647 */
3648static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3649 handle_t *handle,
41cb8148
JB		 3650 struct ocfs2_xattr_bucket *first,
3651 struct ocfs2_xattr_bucket *target,
01225596 3652 u64 new_blkno,
01225596
TM		 3653 u32 num_clusters,
3654 u32 *first_hash)
3655{
c58b6032 3656 int ret;
41cb8148
JB		 3657 struct super_block *sb = inode->i_sb;
3658 int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
3659 int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
b5c03e74 3660 int to_move = num_buckets / 2;
c58b6032 3661 u64 src_blkno;
41cb8148
JB		 3662 u64 last_cluster_blkno = bucket_blkno(first) +
3663 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
01225596 3664
41cb8148
JB		 3665 BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
3666 BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
01225596 3667
01225596 3668 mlog(0, "move half of xattrs in cluster %llu to %llu\n",
c58b6032 3669 (unsigned long long)last_cluster_blkno, (unsigned long long)new_blkno);
01225596 3670
41cb8148 3671 ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
c58b6032
JB		 3672 last_cluster_blkno, new_blkno,
3673 to_move, first_hash);
01225596
TM		 3674 if (ret) {
3675 mlog_errno(ret);
3676 goto out;
3677 }
3678
c58b6032
JB		 3679 /* This is the first bucket that got moved */
3680 src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
01225596 3681
b5c03e74 3682 /*
c58b6032 3683 * If the target bucket was part of the moved buckets, we need to
41cb8148 3684 * update first and target.
b5c03e74 3685 */
41cb8148 3686 if (bucket_blkno(target) >= src_blkno) {
b5c03e74
JB		 3687 /* Find the block for the new target bucket */
3688 src_blkno = new_blkno +
41cb8148
JB		 3689 (bucket_blkno(target) - src_blkno);
3690
3691 ocfs2_xattr_bucket_relse(first);
3692 ocfs2_xattr_bucket_relse(target);
01225596 3693
b5c03e74 3694 /*
c58b6032 3695 * These shouldn't fail - the buffers are in the
b5c03e74
JB		 3696 * journal from ocfs2_cp_xattr_bucket().
3697 */
41cb8148 3698 ret = ocfs2_read_xattr_bucket(first, new_blkno);
c58b6032
JB		 3699 if (ret) {
3700 mlog_errno(ret);
3701 goto out;
3702 }
41cb8148
JB		 3703 ret = ocfs2_read_xattr_bucket(target, src_blkno);
3704 if (ret)
b5c03e74 3705 mlog_errno(ret);
01225596 3706
01225596
TM		 3707 }
3708
01225596 3709out:
01225596
TM		 3710 return ret;
3711}
3712
01225596 3713/*
80bcaf34
TM		 3714 * Find the suitable pos when we divide a bucket into 2.
3715 * We have to make sure the xattrs with the same hash value exist
3716 * in the same bucket.
3717 *
3718 * If this ocfs2_xattr_header covers more than one hash value, find a
3719 * place where the hash value changes. Try to find the most even split.
3720 * The most common case is that all entries have different hash values,
3721 * and the first check we make will find a place to split.
3722 */
3723static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
3724{
3725 struct ocfs2_xattr_entry *entries = xh->xh_entries;
3726 int count = le16_to_cpu(xh->xh_count);
3727 int delta, middle = count / 2;
3728
3729 /*
3730 * We start at the middle. Each step gets farther away in both
3731 * directions. We therefore hit the change in hash value
3732 * nearest to the middle. Note that this loop does not execute for
3733 * count < 2.
3734 */
3735 for (delta = 0; delta < middle; delta++) {
3736 /* Let's check delta earlier than middle */
3737 if (cmp_xe(&entries[middle - delta - 1],
3738 &entries[middle - delta]))
3739 return middle - delta;
3740
3741 /* For even counts, don't walk off the end */
3742 if ((middle + delta + 1) == count)
3743 continue;
3744
3745 /* Now try delta past middle */
3746 if (cmp_xe(&entries[middle + delta],
3747 &entries[middle + delta + 1]))
3748 return middle + delta + 1;
3749 }
3750
3751 /* Every entry had the same hash */
3752 return count;
3753}
3754
3755/*
3756 * Move some xattrs in old bucket(blk) to new bucket(new_blk).
01225596 3757 * first_hash will record the 1st hash of the new bucket.
80bcaf34
TM		 3758 *
3759 * Normally half of the xattrs will be moved. But we have to make
3760 * sure that the xattrs with the same hash value are stored in the
3761 * same bucket. If all the xattrs in this bucket have the same hash
3762 * value, the new bucket will be initialized as an empty one and the
3763 * first_hash will be initialized as (hash_value+1).
01225596 3764 */
80bcaf34
TM		 3765static int ocfs2_divide_xattr_bucket(struct inode *inode,
3766 handle_t *handle,
3767 u64 blk,
3768 u64 new_blk,
3769 u32 *first_hash,
3770 int new_bucket_head)
01225596
TM		 3771{
3772 int ret, i;
80bcaf34 3773 int count, start, len, name_value_len = 0, xe_len, name_offset = 0;
ba937127 3774 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
01225596
TM		 3775 struct ocfs2_xattr_header *xh;
3776 struct ocfs2_xattr_entry *xe;
3777 int blocksize = inode->i_sb->s_blocksize;
3778
80bcaf34 3779 mlog(0, "move some of xattrs from bucket %llu to %llu\n",
de29c085 3780 (unsigned long long)blk, (unsigned long long)new_blk);
01225596 3781
ba937127
JB		 3782 s_bucket = ocfs2_xattr_bucket_new(inode);
3783 t_bucket = ocfs2_xattr_bucket_new(inode);
3784 if (!s_bucket || !t_bucket) {
3785 ret = -ENOMEM;
3786 mlog_errno(ret);
3787 goto out;
3788 }
01225596 3789
ba937127 3790 ret = ocfs2_read_xattr_bucket(s_bucket, blk);
01225596
TM		 3791 if (ret) {
3792 mlog_errno(ret);
3793 goto out;
3794 }
3795
ba937127 3796 ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
1224be02 3797 OCFS2_JOURNAL_ACCESS_WRITE);
01225596
TM		 3798 if (ret) {
3799 mlog_errno(ret);
3800 goto out;
3801 }
3802
784b816a
JB		 3803 /*
3804 * Even if !new_bucket_head, we're overwriting t_bucket. Thus,
3805 * there's no need to read it.
3806 */
ba937127 3807 ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
01225596
TM		 3808 if (ret) {
3809 mlog_errno(ret);
3810 goto out;
3811 }
3812
2b656c1d
JB		 3813 /*
3814 * Hey, if we're overwriting t_bucket, what difference does
3815 * ACCESS_CREATE vs ACCESS_WRITE make? See the comment in the
3816 * same part of ocfs2_cp_xattr_bucket().
3817 */
ba937127 3818 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
1224be02
JB		 3819 new_bucket_head ?
3820 OCFS2_JOURNAL_ACCESS_CREATE :
3821 OCFS2_JOURNAL_ACCESS_WRITE);
3822 if (ret) {
3823 mlog_errno(ret);
3824 goto out;
01225596
TM		 3825 }
3826
ba937127 3827 xh = bucket_xh(s_bucket);
80bcaf34
TM		 3828 count = le16_to_cpu(xh->xh_count);
3829 start = ocfs2_xattr_find_divide_pos(xh);
3830
3831 if (start == count) {
3832 xe = &xh->xh_entries[start-1];
3833
3834 /*
3835 * initialized a new empty bucket here.
3836 * The hash value is set as one larger than
3837 * that of the last entry in the previous bucket.
3838 */
ba937127
JB		 3839 for (i = 0; i < t_bucket->bu_blocks; i++)
3840 memset(bucket_block(t_bucket, i), 0, blocksize);
80bcaf34 3841
ba937127 3842 xh = bucket_xh(t_bucket);
80bcaf34
TM		 3843 xh->xh_free_start = cpu_to_le16(blocksize);
3844 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
3845 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
3846
3847 goto set_num_buckets;
3848 }
3849
01225596 3850 /* copy the whole bucket to the new first. */
ba937127 3851 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
01225596
TM		 3852
3853 /* update the new bucket. */
ba937127 3854 xh = bucket_xh(t_bucket);
01225596
TM		 3855
3856 /*
3857 * Calculate the total name/value len and xh_free_start for
3858 * the old bucket first.
3859 */
3860 name_offset = OCFS2_XATTR_BUCKET_SIZE;
3861 name_value_len = 0;
3862 for (i = 0; i < start; i++) {
3863 xe = &xh->xh_entries[i];
3864 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3865 if (ocfs2_xattr_is_local(xe))
3866 xe_len +=
3867 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3868 else
3869 xe_len += OCFS2_XATTR_ROOT_SIZE;
3870 name_value_len += xe_len;
3871 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3872 name_offset = le16_to_cpu(xe->xe_name_offset);
3873 }
3874
3875 /*
3876 * Now begin the modification to the new bucket.
3877 *
3878 * In the new bucket, We just move the xattr entry to the beginning
3879 * and don't touch the name/value. So there will be some holes in the
3880 * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3881 * called.
3882 */
3883 xe = &xh->xh_entries[start];
3884 len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3885 mlog(0, "mv xattr entry len %d from %d to %d\n", len,
ff1ec20e
MF		 3886 (int)((char *)xe - (char *)xh),
3887 (int)((char *)xh->xh_entries - (char *)xh));
01225596
TM		 3888 memmove((char *)xh->xh_entries, (char *)xe, len);
3889 xe = &xh->xh_entries[count - start];
3890 len = sizeof(struct ocfs2_xattr_entry) * start;
3891 memset((char *)xe, 0, len);
3892
3893 le16_add_cpu(&xh->xh_count, -start);
3894 le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3895
3896 /* Calculate xh_free_start for the new bucket. */
3897 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3898 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3899 xe = &xh->xh_entries[i];
3900 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3901 if (ocfs2_xattr_is_local(xe))
3902 xe_len +=
3903 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3904 else
3905 xe_len += OCFS2_XATTR_ROOT_SIZE;
3906 if (le16_to_cpu(xe->xe_name_offset) <
3907 le16_to_cpu(xh->xh_free_start))
3908 xh->xh_free_start = xe->xe_name_offset;
3909 }
3910
80bcaf34 3911set_num_buckets:
01225596
TM		 3912 /* set xh->xh_num_buckets for the new xh. */
3913 if (new_bucket_head)
3914 xh->xh_num_buckets = cpu_to_le16(1);
3915 else
3916 xh->xh_num_buckets = 0;
3917
ba937127 3918 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
01225596
TM		 3919
3920 /* store the first_hash of the new bucket. */
3921 if (first_hash)
3922 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3923
3924 /*
80bcaf34
TM		 3925 * Now only update the 1st block of the old bucket. If we
3926 * just added a new empty bucket, there is no need to modify
3927 * it.
01225596 3928 */
80bcaf34
TM		 3929 if (start == count)
3930 goto out;
3931
ba937127 3932 xh = bucket_xh(s_bucket);
01225596
TM		 3933 memset(&xh->xh_entries[start], 0,
3934 sizeof(struct ocfs2_xattr_entry) * (count - start));
3935 xh->xh_count = cpu_to_le16(start);
3936 xh->xh_free_start = cpu_to_le16(name_offset);
3937 xh->xh_name_value_len = cpu_to_le16(name_value_len);
3938
ba937127 3939 ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
01225596
TM		 3940
3941out:
ba937127
JB		 3942 ocfs2_xattr_bucket_free(s_bucket);
3943 ocfs2_xattr_bucket_free(t_bucket);
01225596
TM		 3944
3945 return ret;
3946}
3947
3948/*
3949 * Copy xattr from one bucket to another bucket.
3950 *
3951 * The caller must make sure that the journal transaction
3952 * has enough space for journaling.
3953 */
3954static int ocfs2_cp_xattr_bucket(struct inode *inode,
3955 handle_t *handle,
3956 u64 s_blkno,
3957 u64 t_blkno,
3958 int t_is_new)
3959{
4980c6da 3960 int ret;
ba937127 3961 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
01225596
TM		 3962
3963 BUG_ON(s_blkno == t_blkno);
3964
3965 mlog(0, "cp bucket %llu to %llu, target is %d\n",
de29c085
MF		 3966 (unsigned long long)s_blkno, (unsigned long long)t_blkno,
3967 t_is_new);
01225596 3968
ba937127
JB		 3969 s_bucket = ocfs2_xattr_bucket_new(inode);
3970 t_bucket = ocfs2_xattr_bucket_new(inode);
3971 if (!s_bucket || !t_bucket) {
3972 ret = -ENOMEM;
3973 mlog_errno(ret);
3974 goto out;
3975 }
92de109a 3976
ba937127 3977 ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
01225596
TM		 3978 if (ret)
3979 goto out;
3980
784b816a
JB		 3981 /*
3982 * Even if !t_is_new, we're overwriting t_bucket. Thus,
3983 * there's no need to read it.
3984 */
ba937127 3985 ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
01225596
TM		 3986 if (ret)
3987 goto out;
3988
2b656c1d
JB		 3989 /*
3990 * Hey, if we're overwriting t_bucket, what difference does
3991 * ACCESS_CREATE vs ACCESS_WRITE make? Well, if we allocated a new
874d65af
JB		 3992 * cluster to fill, we came here from
3993 * ocfs2_mv_xattr_buckets(), and it is really new -
3994 * ACCESS_CREATE is required. But we also might have moved data
3995 * out of t_bucket before extending back into it.
3996 * ocfs2_add_new_xattr_bucket() can do this - its call to
3997 * ocfs2_add_new_xattr_cluster() may have created a new extent
2b656c1d
JB		 3998 * and copied out the end of the old extent. Then it re-extends
3999 * the old extent back to create space for new xattrs. That's
4000 * how we get here, and the bucket isn't really new.
4001 */
ba937127 4002 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
1224be02
JB		 4003 t_is_new ?
4004 OCFS2_JOURNAL_ACCESS_CREATE :
4005 OCFS2_JOURNAL_ACCESS_WRITE);
4006 if (ret)
4007 goto out;
01225596 4008
ba937127
JB		 4009 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4010 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
01225596
TM		 4011
4012out:
ba937127
JB		 4013 ocfs2_xattr_bucket_free(t_bucket);
4014 ocfs2_xattr_bucket_free(s_bucket);
01225596
TM		 4015
4016 return ret;
4017}
4018
4019/*
874d65af
JB		 4020 * src_blk points to the start of an existing extent. last_blk points to
4021 * last cluster in that extent. to_blk points to a newly allocated
54ecb6b6
JB		 4022 * extent. We copy the buckets from the cluster at last_blk to the new
4023 * extent. If start_bucket is non-zero, we skip that many buckets before
4024 * we start copying. The new extent's xh_num_buckets gets set to the
4025 * number of buckets we copied. The old extent's xh_num_buckets shrinks
4026 * by the same amount.
01225596 4027 */
54ecb6b6
JB		 4028static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4029 u64 src_blk, u64 last_blk, u64 to_blk,
4030 unsigned int start_bucket,
4031 u32 *first_hash)
01225596
TM		 4032{
4033 int i, ret, credits;
4034 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
15d60929 4035 int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
01225596 4036 int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
15d60929 4037 struct ocfs2_xattr_bucket *old_first, *new_first;
01225596 4038
874d65af
JB		 4039 mlog(0, "mv xattrs from cluster %llu to %llu\n",
4040 (unsigned long long)last_blk, (unsigned long long)to_blk);
01225596 4041
54ecb6b6
JB		 4042 BUG_ON(start_bucket >= num_buckets);
4043 if (start_bucket) {
4044 num_buckets -= start_bucket;
4045 last_blk += (start_bucket * blks_per_bucket);
4046 }
4047
15d60929
JB		 4048 /* The first bucket of the original extent */
4049 old_first = ocfs2_xattr_bucket_new(inode);
4050 /* The first bucket of the new extent */
4051 new_first = ocfs2_xattr_bucket_new(inode);
4052 if (!old_first || !new_first) {
4053 ret = -ENOMEM;
4054 mlog_errno(ret);
4055 goto out;
4056 }
4057
874d65af 4058 ret = ocfs2_read_xattr_bucket(old_first, src_blk);
15d60929
JB		 4059 if (ret) {
4060 mlog_errno(ret);
4061 goto out;
4062 }
4063
01225596 4064 /*
54ecb6b6
JB		 4065 * We need to update the first bucket of the old extent and all
4066 * the buckets going to the new extent.
01225596 4067 */
54ecb6b6
JB		 4068 credits = ((num_buckets + 1) * blks_per_bucket) +
4069 handle->h_buffer_credits;
01225596
TM		 4070 ret = ocfs2_extend_trans(handle, credits);
4071 if (ret) {
4072 mlog_errno(ret);
4073 goto out;
4074 }
4075
15d60929
JB		 4076 ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4077 OCFS2_JOURNAL_ACCESS_WRITE);
01225596
TM		 4078 if (ret) {
4079 mlog_errno(ret);
4080 goto out;
4081 }
4082
4083 for (i = 0; i < num_buckets; i++) {
4084 ret = ocfs2_cp_xattr_bucket(inode, handle,
874d65af 4085 last_blk + (i * blks_per_bucket),
15d60929
JB		 4086 to_blk + (i * blks_per_bucket),
4087 1);
01225596
TM		 4088 if (ret) {
4089 mlog_errno(ret);
4090 goto out;
4091 }
01225596
TM		 4092 }
4093
15d60929
JB		 4094 /*
4095 * Get the new bucket ready before we dirty anything
4096 * (This actually shouldn't fail, because we already dirtied
4097 * it once in ocfs2_cp_xattr_bucket()).
4098 */
4099 ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4100 if (ret) {
01225596
TM		 4101 mlog_errno(ret);
4102 goto out;
4103 }
15d60929
JB		 4104 ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4105 OCFS2_JOURNAL_ACCESS_WRITE);
01225596
TM		 4106 if (ret) {
4107 mlog_errno(ret);
4108 goto out;
4109 }
4110
15d60929
JB		 4111 /* Now update the headers */
4112 le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4113 ocfs2_xattr_bucket_journal_dirty(handle, old_first);
01225596 4114
15d60929
JB		 4115 bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4116 ocfs2_xattr_bucket_journal_dirty(handle, new_first);
01225596
TM		 4117
4118 if (first_hash)
15d60929
JB		 4119 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4120
01225596 4121out:
15d60929
JB		 4122 ocfs2_xattr_bucket_free(new_first);
4123 ocfs2_xattr_bucket_free(old_first);
01225596
TM		 4124 return ret;
4125}
4126
4127/*
80bcaf34 4128 * Move some xattrs in this cluster to the new cluster.
01225596
TM		 4129 * This function should only be called when bucket size == cluster size.
4130 * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4131 */
80bcaf34
TM		 4132static int ocfs2_divide_xattr_cluster(struct inode *inode,
4133 handle_t *handle,
4134 u64 prev_blk,
4135 u64 new_blk,
4136 u32 *first_hash)
01225596
TM		 4137{
4138 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
85db90e7 4139 int ret, credits = 2 * blk_per_bucket + handle->h_buffer_credits;
01225596
TM		 4140
4141 BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4142
4143 ret = ocfs2_extend_trans(handle, credits);
4144 if (ret) {
4145 mlog_errno(ret);
4146 return ret;
4147 }
4148
4149 /* Move half of the xattr in start_blk to the next bucket. */
80bcaf34
TM		 4150 return ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4151 new_blk, first_hash, 1);
01225596
TM		 4152}
4153
4154/*
4155 * Move some xattrs from the old cluster to the new one since they are not
4156 * contiguous in ocfs2 xattr tree.
4157 *
4158 * new_blk starts a new separate cluster, and we will move some xattrs from
4159 * prev_blk to it. v_start will be set as the first name hash value in this
4160 * new cluster so that it can be used as e_cpos during tree insertion and
4161 * don't collide with our original b-tree operations. first_bh and header_bh
4162 * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4163 * to extend the insert bucket.
4164 *
4165 * The problem is how much xattr should we move to the new one and when should
4166 * we update first_bh and header_bh?
4167 * 1. If cluster size > bucket size, that means the previous cluster has more
4168 * than 1 bucket, so just move half nums of bucket into the new cluster and
4169 * update the first_bh and header_bh if the insert bucket has been moved
4170 * to the new cluster.
4171 * 2. If cluster_size == bucket_size:
4172 * a) If the previous extent rec has more than one cluster and the insert
4173 * place isn't in the last cluster, copy the entire last cluster to the
4174 * new one. This time, we don't need to upate the first_bh and header_bh
4175 * since they will not be moved into the new cluster.
4176 * b) Otherwise, move the bottom half of the xattrs in the last cluster into
4177 * the new one. And we set the extend flag to zero if the insert place is
4178 * moved into the new allocated cluster since no extend is needed.
4179 */
4180static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
4181 handle_t *handle,
012ee910
JB		 4182 struct ocfs2_xattr_bucket *first,
4183 struct ocfs2_xattr_bucket *target,
01225596 4184 u64 new_blk,
01225596
TM		 4185 u32 prev_clusters,
4186 u32 *v_start,
4187 int *extend)
4188{
92cf3adf 4189 int ret;
01225596
TM		 4190
4191 mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
012ee910 4192 (unsigned long long)bucket_blkno(first), prev_clusters,
de29c085 4193 (unsigned long long)new_blk);
01225596 4194
41cb8148 4195 if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
01225596
TM		 4196 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
4197 handle,
41cb8148 4198 first, target,
01225596 4199 new_blk,
01225596
TM		 4200 prev_clusters,
4201 v_start);
012ee910 4202 if (ret)
41cb8148 4203 mlog_errno(ret);
41cb8148 4204 } else {
92cf3adf
JB		 4205 /* The start of the last cluster in the first extent */
4206 u64 last_blk = bucket_blkno(first) +
4207 ((prev_clusters - 1) *
4208 ocfs2_clusters_to_blocks(inode->i_sb, 1));
01225596 4209
012ee910 4210 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
874d65af 4211 ret = ocfs2_mv_xattr_buckets(inode, handle,
92cf3adf 4212 bucket_blkno(first),
54ecb6b6 4213 last_blk, new_blk, 0,
01225596 4214 v_start);
012ee910
JB		 4215 if (ret)
4216 mlog_errno(ret);
4217 } else {
80bcaf34
TM		 4218 ret = ocfs2_divide_xattr_cluster(inode, handle,
4219 last_blk, new_blk,
4220 v_start);
012ee910
JB		 4221 if (ret)
4222 mlog_errno(ret);
01225596 4223
92cf3adf 4224 if ((bucket_blkno(target) == last_blk) && extend)
01225596
TM		 4225 *extend = 0;
4226 }
4227 }
4228
4229 return ret;
4230}
4231
4232/*
4233 * Add a new cluster for xattr storage.
4234 *
4235 * If the new cluster is contiguous with the previous one, it will be
4236 * appended to the same extent record, and num_clusters will be updated.
4237 * If not, we will insert a new extent for it and move some xattrs in
4238 * the last cluster into the new allocated one.
4239 * We also need to limit the maximum size of a btree leaf, otherwise we'll
4240 * lose the benefits of hashing because we'll have to search large leaves.
4241 * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
4242 * if it's bigger).
4243 *
4244 * first_bh is the first block of the previous extent rec and header_bh
4245 * indicates the bucket we will insert the new xattrs. They will be updated
4246 * when the header_bh is moved into the new cluster.
4247 */
4248static int ocfs2_add_new_xattr_cluster(struct inode *inode,
4249 struct buffer_head *root_bh,
ed29c0ca
JB		 4250 struct ocfs2_xattr_bucket *first,
4251 struct ocfs2_xattr_bucket *target,
01225596
TM		 4252 u32 *num_clusters,
4253 u32 prev_cpos,
78f30c31
TM		 4254 int *extend,
4255 struct ocfs2_xattr_set_ctxt *ctxt)
01225596 4256{
85db90e7 4257 int ret;
01225596
TM		 4258 u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
4259 u32 prev_clusters = *num_clusters;
4260 u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
4261 u64 block;
85db90e7 4262 handle_t *handle = ctxt->handle;
01225596 4263 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
f99b9b7c 4264 struct ocfs2_extent_tree et;
01225596
TM		 4265
4266 mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
4267 "previous xattr blkno = %llu\n",
4268 (unsigned long long)OCFS2_I(inode)->ip_blkno,
ed29c0ca 4269 prev_cpos, (unsigned long long)bucket_blkno(first));
01225596 4270
8d6220d6 4271 ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
f99b9b7c 4272
0cf2f763 4273 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
84008972 4274 OCFS2_JOURNAL_ACCESS_WRITE);
01225596
TM		 4275 if (ret < 0) {
4276 mlog_errno(ret);
4277 goto leave;
4278 }
4279
78f30c31 4280 ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac, 1,
01225596
TM		 4281 clusters_to_add, &bit_off, &num_bits);
4282 if (ret < 0) {
4283 if (ret != -ENOSPC)
4284 mlog_errno(ret);
4285 goto leave;
4286 }
4287
4288 BUG_ON(num_bits > clusters_to_add);
4289
4290 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
4291 mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
4292 num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
4293
ed29c0ca 4294 if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
01225596
TM		 4295 (prev_clusters + num_bits) << osb->s_clustersize_bits <=
4296 OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
4297 /*
4298 * If this cluster is contiguous with the old one and
4299 * adding this new cluster, we don't surpass the limit of
4300 * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
4301 * initialized and used like other buckets in the previous
4302 * cluster.
4303 * So add it as a contiguous one. The caller will handle
4304 * its init process.
4305 */
4306 v_start = prev_cpos + prev_clusters;
4307 *num_clusters = prev_clusters + num_bits;
4308 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
4309 num_bits);
4310 } else {
4311 ret = ocfs2_adjust_xattr_cross_cluster(inode,
4312 handle,
012ee910
JB		 4313 first,
4314 target,
01225596 4315 block,
01225596
TM		 4316 prev_clusters,
4317 &v_start,
4318 extend);
4319 if (ret) {
4320 mlog_errno(ret);
4321 goto leave;
4322 }
4323 }
4324
4325 mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
de29c085 4326 num_bits, (unsigned long long)block, v_start);
f99b9b7c 4327 ret = ocfs2_insert_extent(osb, handle, inode, &et, v_start, block,
78f30c31 4328 num_bits, 0, ctxt->meta_ac);
01225596
TM		 4329 if (ret < 0) {
4330 mlog_errno(ret);
4331 goto leave;
4332 }
4333
4334 ret = ocfs2_journal_dirty(handle, root_bh);
85db90e7 4335 if (ret < 0)
01225596 4336 mlog_errno(ret);
01225596
TM		 4337
4338leave:
01225596
TM		 4339 return ret;
4340}
4341
4342/*
92de109a
JB		 4343 * We are given an extent. 'first' is the bucket at the very front of
4344 * the extent. The extent has space for an additional bucket past
4345 * bucket_xh(first)->xh_num_buckets. 'target_blkno' is the block number
4346 * of the target bucket. We wish to shift every bucket past the target
4347 * down one, filling in that additional space. When we get back to the
4348 * target, we split the target between itself and the now-empty bucket
4349 * at target+1 (aka, target_blkno + blks_per_bucket).
01225596
TM		 4350 */
4351static int ocfs2_extend_xattr_bucket(struct inode *inode,
85db90e7 4352 handle_t *handle,
92de109a
JB		 4353 struct ocfs2_xattr_bucket *first,
4354 u64 target_blk,
01225596
TM		 4355 u32 num_clusters)
4356{
4357 int ret, credits;
4358 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4359 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
92de109a
JB		 4360 u64 end_blk;
4361 u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
01225596
TM		 4362
4363 mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
92de109a
JB		 4364 "from %llu, len = %u\n", (unsigned long long)target_blk,
4365 (unsigned long long)bucket_blkno(first), num_clusters);
01225596 4366
92de109a
JB		 4367 /* The extent must have room for an additional bucket */
4368 BUG_ON(new_bucket >=
4369 (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
01225596 4370
92de109a
JB		 4371 /* end_blk points to the last existing bucket */
4372 end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
01225596
TM		 4373
4374 /*
92de109a
JB		 4375 * end_blk is the start of the last existing bucket.
4376 * Thus, (end_blk - target_blk) covers the target bucket and
4377 * every bucket after it up to, but not including, the last
4378 * existing bucket. Then we add the last existing bucket, the
4379 * new bucket, and the first bucket (3 * blk_per_bucket).
01225596 4380 */
92de109a 4381 credits = (end_blk - target_blk) + (3 * blk_per_bucket) +
85db90e7
TM		 4382 handle->h_buffer_credits;
4383 ret = ocfs2_extend_trans(handle, credits);
4384 if (ret) {
01225596
TM		 4385 mlog_errno(ret);
4386 goto out;
4387 }
4388
92de109a
JB		 4389 ret = ocfs2_xattr_bucket_journal_access(handle, first,
4390 OCFS2_JOURNAL_ACCESS_WRITE);
01225596
TM		 4391 if (ret) {
4392 mlog_errno(ret);
85db90e7 4393 goto out;
01225596
TM		 4394 }
4395
92de109a 4396 while (end_blk != target_blk) {
01225596
TM		 4397 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
4398 end_blk + blk_per_bucket, 0);
4399 if (ret)
85db90e7 4400 goto out;
01225596
TM		 4401 end_blk -= blk_per_bucket;
4402 }
4403
92de109a
JB		 4404 /* Move half of the xattr in target_blkno to the next bucket. */
4405 ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
4406 target_blk + blk_per_bucket, NULL, 0);
01225596 4407
92de109a
JB		 4408 le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
4409 ocfs2_xattr_bucket_journal_dirty(handle, first);
01225596 4410
01225596
TM		 4411out:
4412 return ret;
4413}
4414
4415/*
91f2033f
JB		 4416 * Add new xattr bucket in an extent record and adjust the buckets
4417 * accordingly. xb_bh is the ocfs2_xattr_block, and target is the
4418 * bucket we want to insert into.
01225596 4419 *
91f2033f
JB		 4420 * In the easy case, we will move all the buckets after target down by
4421 * one. Half of target's xattrs will be moved to the next bucket.
4422 *
4423 * If current cluster is full, we'll allocate a new one. This may not
4424 * be contiguous. The underlying calls will make sure that there is
4425 * space for the insert, shifting buckets around if necessary.
4426 * 'target' may be moved by those calls.
01225596
TM		 4427 */
4428static int ocfs2_add_new_xattr_bucket(struct inode *inode,
4429 struct buffer_head *xb_bh,
91f2033f 4430 struct ocfs2_xattr_bucket *target,
78f30c31 4431 struct ocfs2_xattr_set_ctxt *ctxt)
01225596 4432{
01225596
TM		 4433 struct ocfs2_xattr_block *xb =
4434 (struct ocfs2_xattr_block *)xb_bh->b_data;
4435 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
4436 struct ocfs2_extent_list *el = &xb_root->xt_list;
91f2033f
JB		 4437 u32 name_hash =
4438 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
ed29c0ca 4439 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
01225596
TM		 4440 int ret, num_buckets, extend = 1;
4441 u64 p_blkno;
4442 u32 e_cpos, num_clusters;
92de109a 4443 /* The bucket at the front of the extent */
91f2033f 4444 struct ocfs2_xattr_bucket *first;
01225596 4445
91f2033f
JB		 4446 mlog(0, "Add new xattr bucket starting from %llu\n",
4447 (unsigned long long)bucket_blkno(target));
01225596 4448
ed29c0ca 4449 /* The first bucket of the original extent */
92de109a 4450 first = ocfs2_xattr_bucket_new(inode);
91f2033f 4451 if (!first) {
92de109a
JB		 4452 ret = -ENOMEM;
4453 mlog_errno(ret);
4454 goto out;
4455 }
4456
01225596
TM		 4457 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
4458 &num_clusters, el);
4459 if (ret) {
4460 mlog_errno(ret);
4461 goto out;
4462 }
4463
ed29c0ca 4464 ret = ocfs2_read_xattr_bucket(first, p_blkno);
01225596
TM		 4465 if (ret) {
4466 mlog_errno(ret);
4467 goto out;
4468 }
4469
ed29c0ca
JB		 4470 num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
4471 if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
4472 /*
4473 * This can move first+target if the target bucket moves
4474 * to the new extent.
4475 */
01225596
TM		 4476 ret = ocfs2_add_new_xattr_cluster(inode,
4477 xb_bh,
ed29c0ca
JB		 4478 first,
4479 target,
01225596
TM		 4480 &num_clusters,
4481 e_cpos,
78f30c31
TM		 4482 &extend,
4483 ctxt);
01225596
TM		 4484 if (ret) {
4485 mlog_errno(ret);
4486 goto out;
4487 }
4488 }
4489
92de109a 4490 if (extend) {
01225596 4491 ret = ocfs2_extend_xattr_bucket(inode,
85db90e7 4492 ctxt->handle,
ed29c0ca
JB		 4493 first,
4494 bucket_blkno(target),
01225596 4495 num_clusters);
92de109a
JB		 4496 if (ret)
4497 mlog_errno(ret);
4498 }
4499
01225596 4500out:
92de109a 4501 ocfs2_xattr_bucket_free(first);
ed29c0ca 4502
01225596
TM		 4503 return ret;
4504}
4505
4506static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
4507 struct ocfs2_xattr_bucket *bucket,
4508 int offs)
4509{
4510 int block_off = offs >> inode->i_sb->s_blocksize_bits;
4511
4512 offs = offs % inode->i_sb->s_blocksize;
51def39f 4513 return bucket_block(bucket, block_off) + offs;
01225596
TM		 4514}
4515
4516/*
4517 * Handle the normal xattr set, including replace, delete and new.
01225596
TM		 4518 *
4519 * Note: "local" indicates the real data's locality. So we can't
4520 * just its bucket locality by its length.
4521 */
4522static void ocfs2_xattr_set_entry_normal(struct inode *inode,
4523 struct ocfs2_xattr_info *xi,
4524 struct ocfs2_xattr_search *xs,
4525 u32 name_hash,
5a095611 4526 int local)
01225596
TM		 4527{
4528 struct ocfs2_xattr_entry *last, *xe;
4529 int name_len = strlen(xi->name);
4530 struct ocfs2_xattr_header *xh = xs->header;
4531 u16 count = le16_to_cpu(xh->xh_count), start;
4532 size_t blocksize = inode->i_sb->s_blocksize;
4533 char *val;
4534 size_t offs, size, new_size;
4535
4536 last = &xh->xh_entries[count];
4537 if (!xs->not_found) {
4538 xe = xs->here;
4539 offs = le16_to_cpu(xe->xe_name_offset);
4540 if (ocfs2_xattr_is_local(xe))
4541 size = OCFS2_XATTR_SIZE(name_len) +
4542 OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4543 else
4544 size = OCFS2_XATTR_SIZE(name_len) +
4545 OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4546
4547 /*
4548 * If the new value will be stored outside, xi->value has been
4549 * initalized as an empty ocfs2_xattr_value_root, and the same
4550 * goes with xi->value_len, so we can set new_size safely here.
4551 * See ocfs2_xattr_set_in_bucket.
4552 */
4553 new_size = OCFS2_XATTR_SIZE(name_len) +
4554 OCFS2_XATTR_SIZE(xi->value_len);
4555
4556 le16_add_cpu(&xh->xh_name_value_len, -size);
4557 if (xi->value) {
4558 if (new_size > size)
4559 goto set_new_name_value;
4560
4561 /* Now replace the old value with new one. */
4562 if (local)
4563 xe->xe_value_size = cpu_to_le64(xi->value_len);
4564 else
4565 xe->xe_value_size = 0;
4566
4567 val = ocfs2_xattr_bucket_get_val(inode,
ba937127 4568 xs->bucket, offs);
01225596
TM		 4569 memset(val + OCFS2_XATTR_SIZE(name_len), 0,
4570 size - OCFS2_XATTR_SIZE(name_len));
4571 if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
4572 memcpy(val + OCFS2_XATTR_SIZE(name_len),
4573 xi->value, xi->value_len);
4574
4575 le16_add_cpu(&xh->xh_name_value_len, new_size);
4576 ocfs2_xattr_set_local(xe, local);
4577 return;
4578 } else {
5a095611
TM		 4579 /*
4580 * Remove the old entry if there is more than one.
4581 * We don't remove the last entry so that we can
4582 * use it to indicate the hash value of the empty
4583 * bucket.
4584 */
01225596 4585 last -= 1;
01225596 4586 le16_add_cpu(&xh->xh_count, -1);
5a095611
TM		 4587 if (xh->xh_count) {
4588 memmove(xe, xe + 1,
4589 (void *)last - (void *)xe);
4590 memset(last, 0,
4591 sizeof(struct ocfs2_xattr_entry));
4592 } else
4593 xh->xh_free_start =
4594 cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4595
01225596
TM		 4596 return;
4597 }
4598 } else {
4599 /* find a new entry for insert. */
4600 int low = 0, high = count - 1, tmp;
4601 struct ocfs2_xattr_entry *tmp_xe;
4602
5a095611 4603 while (low <= high && count) {
01225596
TM		 4604 tmp = (low + high) / 2;
4605 tmp_xe = &xh->xh_entries[tmp];
4606
4607 if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
4608 low = tmp + 1;
4609 else if (name_hash <
4610 le32_to_cpu(tmp_xe->xe_name_hash))
4611 high = tmp - 1;
06b240d8
TM		 4612 else {
4613 low = tmp;
01225596 4614 break;
06b240d8 4615 }
01225596
TM		 4616 }
4617
4618 xe = &xh->xh_entries[low];
4619 if (low != count)
4620 memmove(xe + 1, xe, (void *)last - (void *)xe);
4621
4622 le16_add_cpu(&xh->xh_count, 1);
4623 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
4624 xe->xe_name_hash = cpu_to_le32(name_hash);
4625 xe->xe_name_len = name_len;
4626 ocfs2_xattr_set_type(xe, xi->name_index);
4627 }
4628
4629set_new_name_value:
4630 /* Insert the new name+value. */
4631 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4632
4633 /*
4634 * We must make sure that the name/value pair
4635 * exists in the same block.
4636 */
4637 offs = le16_to_cpu(xh->xh_free_start);
4638 start = offs - size;
4639
4640 if (start >> inode->i_sb->s_blocksize_bits !=
4641 (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4642 offs = offs - offs % blocksize;
4643 xh->xh_free_start = cpu_to_le16(offs);
4644 }
4645
ba937127 4646 val = ocfs2_xattr_bucket_get_val(inode, xs->bucket, offs - size);
01225596
TM		 4647 xe->xe_name_offset = cpu_to_le16(offs - size);
4648
4649 memset(val, 0, size);
4650 memcpy(val, xi->name, name_len);
4651 memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4652
4653 xe->xe_value_size = cpu_to_le64(xi->value_len);
4654 ocfs2_xattr_set_local(xe, local);
4655 xs->here = xe;
4656 le16_add_cpu(&xh->xh_free_start, -size);
4657 le16_add_cpu(&xh->xh_name_value_len, size);
4658
4659 return;
4660}
4661
01225596
TM		 4662/*
4663 * Set the xattr entry in the specified bucket.
4664 * The bucket is indicated by xs->bucket and it should have the enough
4665 * space for the xattr insertion.
4666 */
4667static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
85db90e7 4668 handle_t *handle,
01225596
TM		 4669 struct ocfs2_xattr_info *xi,
4670 struct ocfs2_xattr_search *xs,
4671 u32 name_hash,
5a095611 4672 int local)
01225596 4673{
1224be02 4674 int ret;
02dbf38d 4675 u64 blkno;
01225596 4676
ff1ec20e
MF		 4677 mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4678 (unsigned long)xi->value_len, xi->name_index,
ba937127 4679 (unsigned long long)bucket_blkno(xs->bucket));
01225596 4680
ba937127 4681 if (!xs->bucket->bu_bhs[1]) {
02dbf38d
JB		 4682 blkno = bucket_blkno(xs->bucket);
4683 ocfs2_xattr_bucket_relse(xs->bucket);
4684 ret = ocfs2_read_xattr_bucket(xs->bucket, blkno);
01225596
TM		 4685 if (ret) {
4686 mlog_errno(ret);
4687 goto out;
4688 }
4689 }
4690
ba937127 4691 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
1224be02
JB		 4692 OCFS2_JOURNAL_ACCESS_WRITE);
4693 if (ret < 0) {
4694 mlog_errno(ret);
4695 goto out;
01225596
TM		 4696 }
4697
5a095611 4698 ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local);
ba937127 4699 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
01225596 4700
01225596 4701out:
01225596
TM		 4702 return ret;
4703}
4704
01225596
TM		 4705/*
4706 * Truncate the specified xe_off entry in xattr bucket.
4707 * bucket is indicated by header_bh and len is the new length.
4708 * Both the ocfs2_xattr_value_root and the entry will be updated here.
4709 *
4710 * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4711 */
4712static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
548b0f22 4713 struct ocfs2_xattr_bucket *bucket,
01225596 4714 int xe_off,
78f30c31
TM		 4715 int len,
4716 struct ocfs2_xattr_set_ctxt *ctxt)
01225596
TM		 4717{
4718 int ret, offset;
4719 u64 value_blk;
01225596 4720 struct ocfs2_xattr_entry *xe;
548b0f22 4721 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
01225596 4722 size_t blocksize = inode->i_sb->s_blocksize;
b3e5d379
JB		 4723 struct ocfs2_xattr_value_buf vb = {
4724 .vb_access = ocfs2_journal_access,
4725 };
01225596
TM		 4726
4727 xe = &xh->xh_entries[xe_off];
4728
4729 BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4730
4731 offset = le16_to_cpu(xe->xe_name_offset) +
4732 OCFS2_XATTR_SIZE(xe->xe_name_len);
4733
4734 value_blk = offset / blocksize;
4735
4736 /* We don't allow ocfs2_xattr_value to be stored in different block. */
4737 BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
01225596 4738
b3e5d379
JB		 4739 vb.vb_bh = bucket->bu_bhs[value_blk];
4740 BUG_ON(!vb.vb_bh);
01225596 4741
b3e5d379
JB		 4742 vb.vb_xv = (struct ocfs2_xattr_value_root *)
4743 (vb.vb_bh->b_data + offset % blocksize);
01225596 4744
548b0f22
JB		 4745 /*
4746 * From here on out we have to dirty the bucket. The generic
4747 * value calls only modify one of the bucket's bhs, but we need
4748 * to send the bucket at once. So if they error, they *could* have
4749 * modified something. We have to assume they did, and dirty
4750 * the whole bucket. This leaves us in a consistent state.
4751 */
4752 mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4753 xe_off, (unsigned long long)bucket_blkno(bucket), len);
b3e5d379 4754 ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
01225596
TM		 4755 if (ret) {
4756 mlog_errno(ret);
554e7f9e
TM		 4757 goto out;
4758 }
4759
4760 ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
4761 OCFS2_JOURNAL_ACCESS_WRITE);
4762 if (ret) {
4763 mlog_errno(ret);
4764 goto out;
01225596
TM		 4765 }
4766
548b0f22
JB		 4767 xe->xe_value_size = cpu_to_le64(len);
4768
548b0f22
JB		 4769 ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
4770
01225596 4771out:
01225596
TM		 4772 return ret;
4773}
4774
4775static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
78f30c31
TM		 4776 struct ocfs2_xattr_search *xs,
4777 int len,
4778 struct ocfs2_xattr_set_ctxt *ctxt)
01225596
TM		 4779{
4780 int ret, offset;
4781 struct ocfs2_xattr_entry *xe = xs->here;
4782 struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4783
ba937127 4784 BUG_ON(!xs->bucket->bu_bhs[0] || !xe || ocfs2_xattr_is_local(xe));
01225596
TM		 4785
4786 offset = xe - xh->xh_entries;
548b0f22 4787 ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket,
78f30c31 4788 offset, len, ctxt);
01225596
TM		 4789 if (ret)
4790 mlog_errno(ret);
4791
4792 return ret;
4793}
4794
4795static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
85db90e7 4796 handle_t *handle,
01225596
TM		 4797 struct ocfs2_xattr_search *xs,
4798 char *val,
4799 int value_len)
4800{
712e53e4 4801 int ret, offset, block_off;
01225596
TM		 4802 struct ocfs2_xattr_value_root *xv;
4803 struct ocfs2_xattr_entry *xe = xs->here;
712e53e4
TM		 4804 struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
4805 void *base;
01225596
TM		 4806
4807 BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4808
712e53e4
TM		 4809 ret = ocfs2_xattr_bucket_get_name_value(inode, xh,
4810 xe - xh->xh_entries,
4811 &block_off,
4812 &offset);
4813 if (ret) {
4814 mlog_errno(ret);
4815 goto out;
4816 }
01225596 4817
712e53e4
TM		 4818 base = bucket_block(xs->bucket, block_off);
4819 xv = (struct ocfs2_xattr_value_root *)(base + offset +
4820 OCFS2_XATTR_SIZE(xe->xe_name_len));
01225596 4821
712e53e4
TM		 4822 ret = __ocfs2_xattr_set_value_outside(inode, handle,
4823 xv, val, value_len);
4824 if (ret)
4825 mlog_errno(ret);
4826out:
4827 return ret;
01225596
TM		 4828}
4829
01225596
TM		 4830static int ocfs2_rm_xattr_cluster(struct inode *inode,
4831 struct buffer_head *root_bh,
4832 u64 blkno,
4833 u32 cpos,
4834 u32 len)
4835{
4836 int ret;
4837 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4838 struct inode *tl_inode = osb->osb_tl_inode;
4839 handle_t *handle;
4840 struct ocfs2_xattr_block *xb =
4841 (struct ocfs2_xattr_block *)root_bh->b_data;
01225596
TM		 4842 struct ocfs2_alloc_context *meta_ac = NULL;
4843 struct ocfs2_cached_dealloc_ctxt dealloc;
f99b9b7c
JB		 4844 struct ocfs2_extent_tree et;
4845
8d6220d6 4846 ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
01225596
TM		 4847
4848 ocfs2_init_dealloc_ctxt(&dealloc);
4849
4850 mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4851 cpos, len, (unsigned long long)blkno);
4852
8cb471e8
JB		 4853 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode), blkno,
4854 len);
01225596 4855
f99b9b7c 4856 ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
01225596
TM		 4857 if (ret) {
4858 mlog_errno(ret);
4859 return ret;
4860 }
4861
4862 mutex_lock(&tl_inode->i_mutex);
4863
4864 if (ocfs2_truncate_log_needs_flush(osb)) {
4865 ret = __ocfs2_flush_truncate_log(osb);
4866 if (ret < 0) {
4867 mlog_errno(ret);
4868 goto out;
4869 }
4870 }
4871
a90714c1 4872 handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
d3264799 4873 if (IS_ERR(handle)) {
01225596
TM		 4874 ret = -ENOMEM;
4875 mlog_errno(ret);
4876 goto out;
4877 }
4878
0cf2f763 4879 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
84008972 4880 OCFS2_JOURNAL_ACCESS_WRITE);
01225596
TM		 4881 if (ret) {
4882 mlog_errno(ret);
4883 goto out_commit;
4884 }
4885
f99b9b7c
JB		 4886 ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
4887 &dealloc);
01225596
TM		 4888 if (ret) {
4889 mlog_errno(ret);
4890 goto out_commit;
4891 }
4892
4893 le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4894
4895 ret = ocfs2_journal_dirty(handle, root_bh);
4896 if (ret) {
4897 mlog_errno(ret);
4898 goto out_commit;
4899 }
4900
4901 ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4902 if (ret)
4903 mlog_errno(ret);
4904
4905out_commit:
4906 ocfs2_commit_trans(osb, handle);
4907out:
4908 ocfs2_schedule_truncate_log_flush(osb, 1);
4909
4910 mutex_unlock(&tl_inode->i_mutex);
4911
4912 if (meta_ac)
4913 ocfs2_free_alloc_context(meta_ac);
4914
4915 ocfs2_run_deallocs(osb, &dealloc);
4916
4917 return ret;
4918}
4919
01225596 4920static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
85db90e7 4921 handle_t *handle,
01225596
TM		 4922 struct ocfs2_xattr_search *xs)
4923{
ba937127 4924 struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
01225596
TM		 4925 struct ocfs2_xattr_entry *last = &xh->xh_entries[
4926 le16_to_cpu(xh->xh_count) - 1];
4927 int ret = 0;
4928
ba937127 4929 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
1224be02 4930 OCFS2_JOURNAL_ACCESS_WRITE);
01225596
TM		 4931 if (ret) {
4932 mlog_errno(ret);
85db90e7 4933 return;
01225596
TM		 4934 }
4935
4936 /* Remove the old entry. */
4937 memmove(xs->here, xs->here + 1,
4938 (void *)last - (void *)xs->here);
4939 memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4940 le16_add_cpu(&xh->xh_count, -1);
4941
ba937127 4942 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
01225596
TM		 4943}
4944
4945/*
4946 * Set the xattr name/value in the bucket specified in xs.
4947 *
4948 * As the new value in xi may be stored in the bucket or in an outside cluster,
4949 * we divide the whole process into 3 steps:
4950 * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4951 * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4952 * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4953 * 4. If the clusters for the new outside value can't be allocated, we need
4954 * to free the xattr we allocated in set.
4955 */
4956static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4957 struct ocfs2_xattr_info *xi,
78f30c31
TM		 4958 struct ocfs2_xattr_search *xs,
4959 struct ocfs2_xattr_set_ctxt *ctxt)
01225596 4960{
5a095611 4961 int ret, local = 1;
01225596
TM		 4962 size_t value_len;
4963 char *val = (char *)xi->value;
4964 struct ocfs2_xattr_entry *xe = xs->here;
2057e5c6
TM		 4965 u32 name_hash = ocfs2_xattr_name_hash(inode, xi->name,
4966 strlen(xi->name));
01225596
TM		 4967
4968 if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
4969 /*
4970 * We need to truncate the xattr storage first.
4971 *
4972 * If both the old and new value are stored to
4973 * outside block, we only need to truncate
4974 * the storage and then set the value outside.
4975 *
4976 * If the new value should be stored within block,
4977 * we should free all the outside block first and
4978 * the modification to the xattr block will be done
4979 * by following steps.
4980 */
4981 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4982 value_len = xi->value_len;
4983 else
4984 value_len = 0;
4985
4986 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
78f30c31
TM		 4987 value_len,
4988 ctxt);
01225596
TM		 4989 if (ret)
4990 goto out;
4991
4992 if (value_len)
4993 goto set_value_outside;
4994 }
4995
4996 value_len = xi->value_len;
4997 /* So we have to handle the inside block change now. */
4998 if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4999 /*
5000 * If the new value will be stored outside of block,
5001 * initalize a new empty value root and insert it first.
5002 */
5003 local = 0;
5004 xi->value = &def_xv;
5005 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
5006 }
5007
85db90e7
TM		 5008 ret = ocfs2_xattr_set_entry_in_bucket(inode, ctxt->handle, xi, xs,
5009 name_hash, local);
01225596
TM		 5010 if (ret) {
5011 mlog_errno(ret);
5012 goto out;
5013 }
5014
5a095611
TM		 5015 if (value_len <= OCFS2_XATTR_INLINE_SIZE)
5016 goto out;
01225596 5017
5a095611
TM		 5018 /* allocate the space now for the outside block storage. */
5019 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
78f30c31 5020 value_len, ctxt);
5a095611
TM		 5021 if (ret) {
5022 mlog_errno(ret);
5023
5024 if (xs->not_found) {
5025 /*
5026 * We can't allocate enough clusters for outside
5027 * storage and we have allocated xattr already,
5028 * so need to remove it.
5029 */
85db90e7 5030 ocfs2_xattr_bucket_remove_xs(inode, ctxt->handle, xs);
01225596 5031 }
01225596
TM		 5032 goto out;
5033 }
5034
5035set_value_outside:
85db90e7
TM		 5036 ret = ocfs2_xattr_bucket_set_value_outside(inode, ctxt->handle,
5037 xs, val, value_len);
01225596
TM		 5038out:
5039 return ret;
5040}
5041
80bcaf34
TM		 5042/*
5043 * check whether the xattr bucket is filled up with the same hash value.
5044 * If we want to insert the xattr with the same hash, return -ENOSPC.
5045 * If we want to insert a xattr with different hash value, go ahead
5046 * and ocfs2_divide_xattr_bucket will handle this.
5047 */
01225596 5048static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
80bcaf34
TM		 5049 struct ocfs2_xattr_bucket *bucket,
5050 const char *name)
01225596 5051{
3e632946 5052 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
80bcaf34
TM		 5053 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5054
5055 if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5056 return 0;
01225596
TM		 5057
5058 if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5059 xh->xh_entries[0].xe_name_hash) {
5060 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5061 "hash = %u\n",
9c7759aa 5062 (unsigned long long)bucket_blkno(bucket),
01225596
TM		 5063 le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5064 return -ENOSPC;
5065 }
5066
5067 return 0;
5068}
5069
5070static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5071 struct ocfs2_xattr_info *xi,
78f30c31
TM		 5072 struct ocfs2_xattr_search *xs,
5073 struct ocfs2_xattr_set_ctxt *ctxt)
01225596
TM		 5074{
5075 struct ocfs2_xattr_header *xh;
5076 struct ocfs2_xattr_entry *xe;
5077 u16 count, header_size, xh_free_start;
6dde41d9 5078 int free, max_free, need, old;
01225596
TM		 5079 size_t value_size = 0, name_len = strlen(xi->name);
5080 size_t blocksize = inode->i_sb->s_blocksize;
5081 int ret, allocation = 0;
01225596
TM		 5082
5083 mlog_entry("Set xattr %s in xattr index block\n", xi->name);
5084
5085try_again:
5086 xh = xs->header;
5087 count = le16_to_cpu(xh->xh_count);
5088 xh_free_start = le16_to_cpu(xh->xh_free_start);
5089 header_size = sizeof(struct ocfs2_xattr_header) +
5090 count * sizeof(struct ocfs2_xattr_entry);
4442f518
TY		 5091 max_free = OCFS2_XATTR_BUCKET_SIZE - header_size -
5092 le16_to_cpu(xh->xh_name_value_len) - OCFS2_XATTR_HEADER_GAP;
01225596
TM		 5093
5094 mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
5095 "of %u which exceed block size\n",
ba937127 5096 (unsigned long long)bucket_blkno(xs->bucket),
01225596
TM		 5097 header_size);
5098
5099 if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
5100 value_size = OCFS2_XATTR_ROOT_SIZE;
5101 else if (xi->value)
5102 value_size = OCFS2_XATTR_SIZE(xi->value_len);
5103
5104 if (xs->not_found)
5105 need = sizeof(struct ocfs2_xattr_entry) +
5106 OCFS2_XATTR_SIZE(name_len) + value_size;
5107 else {
5108 need = value_size + OCFS2_XATTR_SIZE(name_len);
5109
5110 /*
5111 * We only replace the old value if the new length is smaller
5112 * than the old one. Otherwise we will allocate new space in the
5113 * bucket to store it.
5114 */
5115 xe = xs->here;
5116 if (ocfs2_xattr_is_local(xe))
5117 old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
5118 else
5119 old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
5120
5121 if (old >= value_size)
5122 need = 0;
5123 }
5124
4442f518 5125 free = xh_free_start - header_size - OCFS2_XATTR_HEADER_GAP;
01225596
TM		 5126 /*
5127 * We need to make sure the new name/value pair
5128 * can exist in the same block.
5129 */
5130 if (xh_free_start % blocksize < need)
5131 free -= xh_free_start % blocksize;
5132
5133 mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
5134 "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
5135 " %u\n", xs->not_found,
ba937127 5136 (unsigned long long)bucket_blkno(xs->bucket),
01225596
TM		 5137 free, need, max_free, le16_to_cpu(xh->xh_free_start),
5138 le16_to_cpu(xh->xh_name_value_len));
5139
976331d8
TM		 5140 if (free < need ||
5141 (xs->not_found &&
5142 count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb))) {
01225596
TM		 5143 if (need <= max_free &&
5144 count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
5145 /*
5146 * We can create the space by defragment. Since only the
5147 * name/value will be moved, the xe shouldn't be changed
5148 * in xs.
5149 */
85db90e7
TM		 5150 ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5151 xs->bucket);
01225596
TM		 5152 if (ret) {
5153 mlog_errno(ret);
5154 goto out;
5155 }
5156
5157 xh_free_start = le16_to_cpu(xh->xh_free_start);
4442f518
TY		 5158 free = xh_free_start - header_size
5159 - OCFS2_XATTR_HEADER_GAP;
01225596
TM		 5160 if (xh_free_start % blocksize < need)
5161 free -= xh_free_start % blocksize;
5162
5163 if (free >= need)
5164 goto xattr_set;
5165
5166 mlog(0, "Can't get enough space for xattr insert by "
5167 "defragment. Need %u bytes, but we have %d, so "
5168 "allocate new bucket for it.\n", need, free);
5169 }
5170
5171 /*
5172 * We have to add new buckets or clusters and one
5173 * allocation should leave us enough space for insert.
5174 */
5175 BUG_ON(allocation);
5176
5177 /*
5178 * We do not allow for overlapping ranges between buckets. And
5179 * the maximum number of collisions we will allow for then is
5180 * one bucket's worth, so check it here whether we need to
5181 * add a new bucket for the insert.
5182 */
80bcaf34 5183 ret = ocfs2_check_xattr_bucket_collision(inode,
ba937127 5184 xs->bucket,
80bcaf34 5185 xi->name);
01225596
TM		 5186 if (ret) {
5187 mlog_errno(ret);
5188 goto out;
5189 }
5190
5191 ret = ocfs2_add_new_xattr_bucket(inode,
5192 xs->xattr_bh,
91f2033f 5193 xs->bucket,
78f30c31 5194 ctxt);
01225596
TM		 5195 if (ret) {
5196 mlog_errno(ret);
5197 goto out;
5198 }
5199
91f2033f
JB		 5200 /*
5201 * ocfs2_add_new_xattr_bucket() will have updated
5202 * xs->bucket if it moved, but it will not have updated
5203 * any of the other search fields. Thus, we drop it and
5204 * re-search. Everything should be cached, so it'll be
5205 * quick.
5206 */
ba937127 5207 ocfs2_xattr_bucket_relse(xs->bucket);
01225596
TM		 5208 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5209 xi->name_index,
5210 xi->name, xs);
5211 if (ret && ret != -ENODATA)
5212 goto out;
5213 xs->not_found = ret;
5214 allocation = 1;
5215 goto try_again;
5216 }
5217
5218xattr_set:
78f30c31 5219 ret = ocfs2_xattr_set_in_bucket(inode, xi, xs, ctxt);
01225596
TM		 5220out:
5221 mlog_exit(ret);
5222 return ret;
5223}
a3944256
TM		 5224
5225static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5226 struct ocfs2_xattr_bucket *bucket,
5227 void *para)
5228{
5229 int ret = 0;
3e632946 5230 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
a3944256
TM		 5231 u16 i;
5232 struct ocfs2_xattr_entry *xe;
78f30c31
TM		 5233 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5234 struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
548b0f22
JB		 5235 int credits = ocfs2_remove_extent_credits(osb->sb) +
5236 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5237
78f30c31
TM		 5238
5239 ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
a3944256
TM		 5240
5241 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5242 xe = &xh->xh_entries[i];
5243 if (ocfs2_xattr_is_local(xe))
5244 continue;
5245
88c3b062
TM		 5246 ctxt.handle = ocfs2_start_trans(osb, credits);
5247 if (IS_ERR(ctxt.handle)) {
5248 ret = PTR_ERR(ctxt.handle);
5249 mlog_errno(ret);
5250 break;
5251 }
5252
548b0f22 5253 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
78f30c31 5254 i, 0, &ctxt);
88c3b062
TM		 5255
5256 ocfs2_commit_trans(osb, ctxt.handle);
a3944256
TM		 5257 if (ret) {
5258 mlog_errno(ret);
5259 break;
5260 }
5261 }
5262
78f30c31
TM		 5263 ocfs2_schedule_truncate_log_flush(osb, 1);
5264 ocfs2_run_deallocs(osb, &ctxt.dealloc);
a3944256
TM		 5265 return ret;
5266}
5267
5268static int ocfs2_delete_xattr_index_block(struct inode *inode,
5269 struct buffer_head *xb_bh)
5270{
5271 struct ocfs2_xattr_block *xb =
5272 (struct ocfs2_xattr_block *)xb_bh->b_data;
5273 struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
5274 int ret = 0;
5275 u32 name_hash = UINT_MAX, e_cpos, num_clusters;
5276 u64 p_blkno;
5277
5278 if (le16_to_cpu(el->l_next_free_rec) == 0)
5279 return 0;
5280
5281 while (name_hash > 0) {
5282 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
5283 &e_cpos, &num_clusters, el);
5284 if (ret) {
5285 mlog_errno(ret);
5286 goto out;
5287 }
5288
5289 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
5290 ocfs2_delete_xattr_in_bucket,
5291 NULL);
5292 if (ret) {
5293 mlog_errno(ret);
5294 goto out;
5295 }
5296
5297 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
5298 p_blkno, e_cpos, num_clusters);
5299 if (ret) {
5300 mlog_errno(ret);
5301 break;
5302 }
5303
5304 if (e_cpos == 0)
5305 break;
5306
5307 name_hash = e_cpos - 1;
5308 }
5309
5310out:
5311 return ret;
5312}
99219aea 5313
923f7f31
TY		 5314/*
5315 * 'security' attributes support
5316 */
5317static size_t ocfs2_xattr_security_list(struct inode *inode, char *list,
5318 size_t list_size, const char *name,
5319 size_t name_len)
5320{
5321 const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
5322 const size_t total_len = prefix_len + name_len + 1;
5323
5324 if (list && total_len <= list_size) {
5325 memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
5326 memcpy(list + prefix_len, name, name_len);
5327 list[prefix_len + name_len] = '\0';
5328 }
5329 return total_len;
5330}
5331
5332static int ocfs2_xattr_security_get(struct inode *inode, const char *name,
5333 void *buffer, size_t size)
5334{
5335 if (strcmp(name, "") == 0)
5336 return -EINVAL;
5337 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_SECURITY, name,
5338 buffer, size);
5339}
5340
5341static int ocfs2_xattr_security_set(struct inode *inode, const char *name,
5342 const void *value, size_t size, int flags)
5343{
5344 if (strcmp(name, "") == 0)
5345 return -EINVAL;
5346
5347 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY, name, value,
5348 size, flags);
5349}
5350
534eaddd
TY		 5351int ocfs2_init_security_get(struct inode *inode,
5352 struct inode *dir,
5353 struct ocfs2_security_xattr_info *si)
5354{
38d59ef6
TY		 5355 /* check whether ocfs2 support feature xattr */
5356 if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
5357 return -EOPNOTSUPP;
534eaddd
TY		 5358 return security_inode_init_security(inode, dir, &si->name, &si->value,
5359 &si->value_len);
5360}
5361
5362int ocfs2_init_security_set(handle_t *handle,
5363 struct inode *inode,
5364 struct buffer_head *di_bh,
5365 struct ocfs2_security_xattr_info *si,
5366 struct ocfs2_alloc_context *xattr_ac,
5367 struct ocfs2_alloc_context *data_ac)
5368{
5369 return ocfs2_xattr_set_handle(handle, inode, di_bh,
5370 OCFS2_XATTR_INDEX_SECURITY,
5371 si->name, si->value, si->value_len, 0,
5372 xattr_ac, data_ac);
5373}
5374
923f7f31
TY		 5375struct xattr_handler ocfs2_xattr_security_handler = {
5376 .prefix = XATTR_SECURITY_PREFIX,
5377 .list = ocfs2_xattr_security_list,
5378 .get = ocfs2_xattr_security_get,
5379 .set = ocfs2_xattr_security_set,
5380};
5381
99219aea
MF		 5382/*
5383 * 'trusted' attributes support
5384 */
99219aea
MF		 5385static size_t ocfs2_xattr_trusted_list(struct inode *inode, char *list,
5386 size_t list_size, const char *name,
5387 size_t name_len)
5388{
ceb1eba3 5389 const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
99219aea
MF		 5390 const size_t total_len = prefix_len + name_len + 1;
5391
5392 if (list && total_len <= list_size) {
5393 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
5394 memcpy(list + prefix_len, name, name_len);
5395 list[prefix_len + name_len] = '\0';
5396 }
5397 return total_len;
5398}
5399
5400static int ocfs2_xattr_trusted_get(struct inode *inode, const char *name,
5401 void *buffer, size_t size)
5402{
5403 if (strcmp(name, "") == 0)
5404 return -EINVAL;
5405 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED, name,
5406 buffer, size);
5407}
5408
5409static int ocfs2_xattr_trusted_set(struct inode *inode, const char *name,
5410 const void *value, size_t size, int flags)
5411{
5412 if (strcmp(name, "") == 0)
5413 return -EINVAL;
5414
5415 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED, name, value,
5416 size, flags);
5417}
5418
5419struct xattr_handler ocfs2_xattr_trusted_handler = {
5420 .prefix = XATTR_TRUSTED_PREFIX,
5421 .list = ocfs2_xattr_trusted_list,
5422 .get = ocfs2_xattr_trusted_get,
5423 .set = ocfs2_xattr_trusted_set,
5424};
5425
99219aea
MF		 5426/*
5427 * 'user' attributes support
5428 */
99219aea
MF		 5429static size_t ocfs2_xattr_user_list(struct inode *inode, char *list,
5430 size_t list_size, const char *name,
5431 size_t name_len)
5432{
ceb1eba3 5433 const size_t prefix_len = XATTR_USER_PREFIX_LEN;
99219aea
MF		 5434 const size_t total_len = prefix_len + name_len + 1;
5435 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5436
5437 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5438 return 0;
5439
5440 if (list && total_len <= list_size) {
5441 memcpy(list, XATTR_USER_PREFIX, prefix_len);
5442 memcpy(list + prefix_len, name, name_len);
5443 list[prefix_len + name_len] = '\0';
5444 }
5445 return total_len;
5446}
5447
5448static int ocfs2_xattr_user_get(struct inode *inode, const char *name,
5449 void *buffer, size_t size)
5450{
5451 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5452
5453 if (strcmp(name, "") == 0)
5454 return -EINVAL;
5455 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5456 return -EOPNOTSUPP;
5457 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
5458 buffer, size);
5459}
5460
5461static int ocfs2_xattr_user_set(struct inode *inode, const char *name,
5462 const void *value, size_t size, int flags)
5463{
5464 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5465
5466 if (strcmp(name, "") == 0)
5467 return -EINVAL;
5468 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5469 return -EOPNOTSUPP;
5470
5471 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER, name, value,
5472 size, flags);
5473}
5474
5475struct xattr_handler ocfs2_xattr_user_handler = {
5476 .prefix = XATTR_USER_PREFIX,
5477 .list = ocfs2_xattr_user_list,
5478 .get = ocfs2_xattr_user_get,
5479 .set = ocfs2_xattr_user_set,
5480};
Linux kernel - Deliverables
This page took 0.551517 seconds and 5 git commands to generate.