]> bbs.cooldavid.org Git - net-next-2.6.git/blame - fs/ocfs2/xattr.c
git://bbs.cooldavid.org / net-next-2.6.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
ocfs2: Set the xattr name+value pair in one place
[net-next-2.6.git] / fs / ocfs2 / xattr.c
CommitLineData
f56654c4
TM		 1/* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * xattr.c
5 *
c3cb6827 6 * Copyright (C) 2004, 2008 Oracle. All rights reserved.
f56654c4 7 *
cf1d6c76 8 * CREDITS:
c3cb6827
TY		 9 * Lots of code in this file is copy from linux/fs/ext3/xattr.c.
10 * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
cf1d6c76 11 *
f56654c4
TM		 12 * This program is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU General Public
c3cb6827 14 * License version 2 as published by the Free Software Foundation.
f56654c4
TM		 15 *
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
f56654c4
TM		 20 */
21
cf1d6c76
TY		 22#include <linux/capability.h>
23#include <linux/fs.h>
24#include <linux/types.h>
25#include <linux/slab.h>
26#include <linux/highmem.h>
27#include <linux/pagemap.h>
28#include <linux/uio.h>
29#include <linux/sched.h>
30#include <linux/splice.h>
31#include <linux/mount.h>
32#include <linux/writeback.h>
33#include <linux/falloc.h>
01225596 34#include <linux/sort.h>
99219aea
MF		 35#include <linux/init.h>
36#include <linux/module.h>
37#include <linux/string.h>
923f7f31 38#include <linux/security.h>
cf1d6c76 39
f56654c4
TM		 40#define MLOG_MASK_PREFIX ML_XATTR
41#include <cluster/masklog.h>
42
43#include "ocfs2.h"
44#include "alloc.h"
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 57#include "xattr.h"
492a8a33 58#include "refcounttree.h"
0fe9b66c 59#include "acl.h"
cf1d6c76
TY		 60
61struct ocfs2_xattr_def_value_root {
62 struct ocfs2_xattr_value_root xv;
63 struct ocfs2_extent_rec er;
64};
65
0c044f0b 66struct ocfs2_xattr_bucket {
ba937127
JB		 67 /* The inode these xattrs are associated with */
68 struct inode *bu_inode;
69
70 /* The actual buffers that make up the bucket */
4ac6032d 71 struct buffer_head *bu_bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
ba937127
JB		 72
73 /* How many blocks make up one bucket for this filesystem */
74 int bu_blocks;
0c044f0b
TM		 75};
76
78f30c31 77struct ocfs2_xattr_set_ctxt {
85db90e7 78 handle_t *handle;
78f30c31
TM		 79 struct ocfs2_alloc_context *meta_ac;
80 struct ocfs2_alloc_context *data_ac;
81 struct ocfs2_cached_dealloc_ctxt dealloc;
82};
83
cf1d6c76
TY		 84#define OCFS2_XATTR_ROOT_SIZE (sizeof(struct ocfs2_xattr_def_value_root))
85#define OCFS2_XATTR_INLINE_SIZE 80
4442f518 86#define OCFS2_XATTR_HEADER_GAP 4
534eaddd
TY		 87#define OCFS2_XATTR_FREE_IN_IBODY (OCFS2_MIN_XATTR_INLINE_SIZE \
88 - sizeof(struct ocfs2_xattr_header) \
4442f518 89 - OCFS2_XATTR_HEADER_GAP)
89c38bd0
TY		 90#define OCFS2_XATTR_FREE_IN_BLOCK(ptr) ((ptr)->i_sb->s_blocksize \
91 - sizeof(struct ocfs2_xattr_block) \
92 - sizeof(struct ocfs2_xattr_header) \
4442f518 93 - OCFS2_XATTR_HEADER_GAP)
cf1d6c76
TY		 94
95static struct ocfs2_xattr_def_value_root def_xv = {
96 .xv.xr_list.l_count = cpu_to_le16(1),
97};
98
99struct xattr_handler *ocfs2_xattr_handlers[] = {
100 &ocfs2_xattr_user_handler,
929fb014
TY		 101 &ocfs2_xattr_acl_access_handler,
102 &ocfs2_xattr_acl_default_handler,
cf1d6c76 103 &ocfs2_xattr_trusted_handler,
923f7f31 104 &ocfs2_xattr_security_handler,
cf1d6c76
TY		 105 NULL
106};
107
c988fd04 108static struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
cf1d6c76 109 [OCFS2_XATTR_INDEX_USER] = &ocfs2_xattr_user_handler,
929fb014
TY		 110 [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS]
111 = &ocfs2_xattr_acl_access_handler,
112 [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT]
113 = &ocfs2_xattr_acl_default_handler,
cf1d6c76 114 [OCFS2_XATTR_INDEX_TRUSTED] = &ocfs2_xattr_trusted_handler,
923f7f31 115 [OCFS2_XATTR_INDEX_SECURITY] = &ocfs2_xattr_security_handler,
cf1d6c76
TY		 116};
117
118struct ocfs2_xattr_info {
6b240ff6
JB		 119 int xi_name_index;
120 const char *xi_name;
18853b95 121 int xi_name_len;
6b240ff6
JB		 122 const void *xi_value;
123 size_t xi_value_len;
cf1d6c76
TY		 124};
125
126struct ocfs2_xattr_search {
127 struct buffer_head *inode_bh;
128 /*
129 * xattr_bh point to the block buffer head which has extended attribute
130 * when extended attribute in inode, xattr_bh is equal to inode_bh.
131 */
132 struct buffer_head *xattr_bh;
133 struct ocfs2_xattr_header *header;
ba937127 134 struct ocfs2_xattr_bucket *bucket;
cf1d6c76
TY		 135 void *base;
136 void *end;
137 struct ocfs2_xattr_entry *here;
138 int not_found;
139};
140
11179f2c
JB		 141/* Operations on struct ocfs2_xa_entry */
142struct ocfs2_xa_loc;
143struct ocfs2_xa_loc_operations {
144 /*
145 * Return a pointer to the appropriate buffer in loc->xl_storage
146 * at the given offset from loc->xl_header.
147 */
148 void *(*xlo_offset_pointer)(struct ocfs2_xa_loc *loc, int offset);
149
69a3e539
JB		 150 /* Can we reuse the existing entry for the new value? */
151 int (*xlo_can_reuse)(struct ocfs2_xa_loc *loc,
152 struct ocfs2_xattr_info *xi);
153
154 /* How much space is needed for the new value? */
155 int (*xlo_check_space)(struct ocfs2_xa_loc *loc,
156 struct ocfs2_xattr_info *xi);
157
158 /*
159 * Return the offset of the first name+value pair. This is
160 * the start of our downward-filling free space.
161 */
162 int (*xlo_get_free_start)(struct ocfs2_xa_loc *loc);
163
11179f2c
JB		 164 /*
165 * Remove the name+value at this location. Do whatever is
166 * appropriate with the remaining name+value pairs.
167 */
168 void (*xlo_wipe_namevalue)(struct ocfs2_xa_loc *loc);
69a3e539
JB		 169
170 /* Fill xl_entry with a new entry */
171 void (*xlo_add_entry)(struct ocfs2_xa_loc *loc, u32 name_hash);
172
173 /* Add name+value storage to an entry */
174 void (*xlo_add_namevalue)(struct ocfs2_xa_loc *loc, int size);
11179f2c
JB		 175};
176
177/*
178 * Describes an xattr entry location. This is a memory structure
179 * tracking the on-disk structure.
180 */
181struct ocfs2_xa_loc {
182 /* The ocfs2_xattr_header inside the on-disk storage. Not NULL. */
183 struct ocfs2_xattr_header *xl_header;
184
185 /* Bytes from xl_header to the end of the storage */
186 int xl_size;
187
188 /*
189 * The ocfs2_xattr_entry this location describes. If this is
190 * NULL, this location describes the on-disk structure where it
191 * would have been.
192 */
193 struct ocfs2_xattr_entry *xl_entry;
194
195 /*
196 * Internal housekeeping
197 */
198
199 /* Buffer(s) containing this entry */
200 void *xl_storage;
201
202 /* Operations on the storage backing this location */
203 const struct ocfs2_xa_loc_operations *xl_ops;
204};
205
199799a3
JB		 206/*
207 * Convenience functions to calculate how much space is needed for a
208 * given name+value pair
209 */
210static int namevalue_size(int name_len, uint64_t value_len)
211{
212 if (value_len > OCFS2_XATTR_INLINE_SIZE)
213 return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
214 else
215 return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
216}
217
218static int namevalue_size_xi(struct ocfs2_xattr_info *xi)
219{
220 return namevalue_size(xi->xi_name_len, xi->xi_value_len);
221}
222
223static int namevalue_size_xe(struct ocfs2_xattr_entry *xe)
224{
225 u64 value_len = le64_to_cpu(xe->xe_value_size);
226
227 BUG_ON((value_len > OCFS2_XATTR_INLINE_SIZE) &&
228 ocfs2_xattr_is_local(xe));
229 return namevalue_size(xe->xe_name_len, value_len);
230}
231
232
fd68a894 233static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
589dc260
TM		 234 struct ocfs2_xattr_header *xh,
235 int index,
236 int *block_off,
237 int *new_offset);
238
54f443f4
JB		 239static int ocfs2_xattr_block_find(struct inode *inode,
240 int name_index,
241 const char *name,
242 struct ocfs2_xattr_search *xs);
589dc260
TM		 243static int ocfs2_xattr_index_block_find(struct inode *inode,
244 struct buffer_head *root_bh,
245 int name_index,
246 const char *name,
247 struct ocfs2_xattr_search *xs);
248
0c044f0b 249static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
47bca495 250 struct buffer_head *blk_bh,
0c044f0b
TM		 251 char *buffer,
252 size_t buffer_size);
253
01225596 254static int ocfs2_xattr_create_index_block(struct inode *inode,
78f30c31
TM		 255 struct ocfs2_xattr_search *xs,
256 struct ocfs2_xattr_set_ctxt *ctxt);
01225596
TM		 257
258static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
259 struct ocfs2_xattr_info *xi,
78f30c31
TM		 260 struct ocfs2_xattr_search *xs,
261 struct ocfs2_xattr_set_ctxt *ctxt);
01225596 262
47bca495
TM		 263typedef int (xattr_tree_rec_func)(struct inode *inode,
264 struct buffer_head *root_bh,
265 u64 blkno, u32 cpos, u32 len, void *para);
266static int ocfs2_iterate_xattr_index_block(struct inode *inode,
267 struct buffer_head *root_bh,
268 xattr_tree_rec_func *rec_func,
269 void *para);
270static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
271 struct ocfs2_xattr_bucket *bucket,
272 void *para);
273static int ocfs2_rm_xattr_cluster(struct inode *inode,
274 struct buffer_head *root_bh,
275 u64 blkno,
276 u32 cpos,
277 u32 len,
278 void *para);
279
c58b6032
JB		 280static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
281 u64 src_blk, u64 last_blk, u64 to_blk,
282 unsigned int start_bucket,
283 u32 *first_hash);
492a8a33
TM		 284static int ocfs2_prepare_refcount_xattr(struct inode *inode,
285 struct ocfs2_dinode *di,
286 struct ocfs2_xattr_info *xi,
287 struct ocfs2_xattr_search *xis,
288 struct ocfs2_xattr_search *xbs,
289 struct ocfs2_refcount_tree **ref_tree,
290 int *meta_need,
291 int *credits);
ce9c5a54
TM		 292static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
293 struct ocfs2_xattr_bucket *bucket,
294 int offset,
295 struct ocfs2_xattr_value_root **xv,
296 struct buffer_head **bh);
a3944256 297
0030e001
TY		 298static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
299{
300 return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
301}
302
303static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
304{
305 return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
306}
307
308static inline u16 ocfs2_xattr_max_xe_in_bucket(struct super_block *sb)
309{
310 u16 len = sb->s_blocksize -
311 offsetof(struct ocfs2_xattr_header, xh_entries);
312
313 return len / sizeof(struct ocfs2_xattr_entry);
314}
315
9c7759aa 316#define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
51def39f 317#define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
3e632946 318#define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
9c7759aa 319
ba937127 320static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
6dde41d9 321{
ba937127
JB		 322 struct ocfs2_xattr_bucket *bucket;
323 int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
6dde41d9 324
ba937127
JB		 325 BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
326
327 bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
328 if (bucket) {
329 bucket->bu_inode = inode;
330 bucket->bu_blocks = blks;
331 }
332
333 return bucket;
334}
335
336static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
337{
338 int i;
339
340 for (i = 0; i < bucket->bu_blocks; i++) {
6dde41d9
JB		 341 brelse(bucket->bu_bhs[i]);
342 bucket->bu_bhs[i] = NULL;
343 }
344}
345
ba937127
JB		 346static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
347{
348 if (bucket) {
349 ocfs2_xattr_bucket_relse(bucket);
350 bucket->bu_inode = NULL;
351 kfree(bucket);
352 }
353}
354
784b816a
JB		 355/*
356 * A bucket that has never been written to disk doesn't need to be
357 * read. We just need the buffer_heads. Don't call this for
358 * buckets that are already on disk. ocfs2_read_xattr_bucket() initializes
359 * them fully.
360 */
ba937127 361static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
784b816a
JB		 362 u64 xb_blkno)
363{
364 int i, rc = 0;
784b816a 365
ba937127
JB		 366 for (i = 0; i < bucket->bu_blocks; i++) {
367 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
368 xb_blkno + i);
784b816a
JB		 369 if (!bucket->bu_bhs[i]) {
370 rc = -EIO;
371 mlog_errno(rc);
372 break;
373 }
374
8cb471e8 375 if (!ocfs2_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
757055ad 376 bucket->bu_bhs[i]))
8cb471e8 377 ocfs2_set_new_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
757055ad 378 bucket->bu_bhs[i]);
784b816a
JB		 379 }
380
381 if (rc)
ba937127 382 ocfs2_xattr_bucket_relse(bucket);
784b816a
JB		 383 return rc;
384}
385
386/* Read the xattr bucket at xb_blkno */
ba937127 387static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
784b816a
JB		 388 u64 xb_blkno)
389{
ba937127 390 int rc;
784b816a 391
8cb471e8 392 rc = ocfs2_read_blocks(INODE_CACHE(bucket->bu_inode), xb_blkno,
970e4936
JB		 393 bucket->bu_blocks, bucket->bu_bhs, 0,
394 NULL);
4d0e214e 395 if (!rc) {
c8b9cf9a 396 spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
4d0e214e
JB		 397 rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
398 bucket->bu_bhs,
399 bucket->bu_blocks,
400 &bucket_xh(bucket)->xh_check);
c8b9cf9a 401 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
4d0e214e
JB		 402 if (rc)
403 mlog_errno(rc);
404 }
405
784b816a 406 if (rc)
ba937127 407 ocfs2_xattr_bucket_relse(bucket);
784b816a
JB		 408 return rc;
409}
410
1224be02 411static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
1224be02
JB		 412 struct ocfs2_xattr_bucket *bucket,
413 int type)
414{
415 int i, rc = 0;
1224be02 416
ba937127 417 for (i = 0; i < bucket->bu_blocks; i++) {
0cf2f763
JB		 418 rc = ocfs2_journal_access(handle,
419 INODE_CACHE(bucket->bu_inode),
1224be02
JB		 420 bucket->bu_bhs[i], type);
421 if (rc) {
422 mlog_errno(rc);
423 break;
424 }
425 }
426
427 return rc;
428}
429
430static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
1224be02
JB		 431 struct ocfs2_xattr_bucket *bucket)
432{
ba937127 433 int i;
1224be02 434
c8b9cf9a 435 spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
4d0e214e
JB		 436 ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
437 bucket->bu_bhs, bucket->bu_blocks,
438 &bucket_xh(bucket)->xh_check);
c8b9cf9a 439 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
4d0e214e 440
ba937127 441 for (i = 0; i < bucket->bu_blocks; i++)
1224be02
JB		 442 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
443}
444
ba937127 445static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
4980c6da
JB		 446 struct ocfs2_xattr_bucket *src)
447{
448 int i;
ba937127
JB		 449 int blocksize = src->bu_inode->i_sb->s_blocksize;
450
451 BUG_ON(dest->bu_blocks != src->bu_blocks);
452 BUG_ON(dest->bu_inode != src->bu_inode);
4980c6da 453
ba937127 454 for (i = 0; i < src->bu_blocks; i++) {
4980c6da
JB		 455 memcpy(bucket_block(dest, i), bucket_block(src, i),
456 blocksize);
457 }
458}
1224be02 459
4ae1d69b
JB		 460static int ocfs2_validate_xattr_block(struct super_block *sb,
461 struct buffer_head *bh)
462{
d6b32bbb 463 int rc;
4ae1d69b
JB		 464 struct ocfs2_xattr_block *xb =
465 (struct ocfs2_xattr_block *)bh->b_data;
466
467 mlog(0, "Validating xattr block %llu\n",
468 (unsigned long long)bh->b_blocknr);
469
d6b32bbb
JB		 470 BUG_ON(!buffer_uptodate(bh));
471
472 /*
473 * If the ecc fails, we return the error but otherwise
474 * leave the filesystem running. We know any error is
475 * local to this block.
476 */
477 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
478 if (rc)
479 return rc;
480
481 /*
482 * Errors after here are fatal
483 */
484
4ae1d69b
JB		 485 if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
486 ocfs2_error(sb,
487 "Extended attribute block #%llu has bad "
488 "signature %.*s",
489 (unsigned long long)bh->b_blocknr, 7,
490 xb->xb_signature);
491 return -EINVAL;
492 }
493
494 if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
495 ocfs2_error(sb,
496 "Extended attribute block #%llu has an "
497 "invalid xb_blkno of %llu",
498 (unsigned long long)bh->b_blocknr,
499 (unsigned long long)le64_to_cpu(xb->xb_blkno));
500 return -EINVAL;
501 }
502
503 if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
504 ocfs2_error(sb,
505 "Extended attribute block #%llu has an invalid "
506 "xb_fs_generation of #%u",
507 (unsigned long long)bh->b_blocknr,
508 le32_to_cpu(xb->xb_fs_generation));
509 return -EINVAL;
510 }
511
512 return 0;
513}
514
515static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
516 struct buffer_head **bh)
517{
518 int rc;
519 struct buffer_head *tmp = *bh;
520
8cb471e8 521 rc = ocfs2_read_block(INODE_CACHE(inode), xb_blkno, &tmp,
970e4936 522 ocfs2_validate_xattr_block);
4ae1d69b
JB		 523
524 /* If ocfs2_read_block() got us a new bh, pass it up. */
525 if (!rc && !*bh)
526 *bh = tmp;
527
528 return rc;
529}
530
936b8834 531static inline const char *ocfs2_xattr_prefix(int name_index)
cf1d6c76
TY		 532{
533 struct xattr_handler *handler = NULL;
534
535 if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
536 handler = ocfs2_xattr_handler_map[name_index];
537
936b8834 538 return handler ? handler->prefix : NULL;
cf1d6c76
TY		 539}
540
40daa16a 541static u32 ocfs2_xattr_name_hash(struct inode *inode,
2057e5c6 542 const char *name,
40daa16a 543 int name_len)
cf1d6c76
TY		 544{
545 /* Get hash value of uuid from super block */
546 u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
547 int i;
548
cf1d6c76
TY		 549 /* hash extended attribute name */
550 for (i = 0; i < name_len; i++) {
551 hash = (hash << OCFS2_HASH_SHIFT) ^
552 (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
553 *name++;
554 }
555
556 return hash;
557}
558
559/*
560 * ocfs2_xattr_hash_entry()
561 *
562 * Compute the hash of an extended attribute.
563 */
564static void ocfs2_xattr_hash_entry(struct inode *inode,
565 struct ocfs2_xattr_header *header,
566 struct ocfs2_xattr_entry *entry)
567{
568 u32 hash = 0;
cf1d6c76 569 char *name = (char *)header + le16_to_cpu(entry->xe_name_offset);
cf1d6c76 570
2057e5c6 571 hash = ocfs2_xattr_name_hash(inode, name, entry->xe_name_len);
cf1d6c76
TY		 572 entry->xe_name_hash = cpu_to_le32(hash);
573
574 return;
575}
f56654c4 576
534eaddd
TY		 577static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
578{
199799a3
JB		 579 return namevalue_size(name_len, value_len) +
580 sizeof(struct ocfs2_xattr_entry);
581}
534eaddd 582
199799a3
JB		 583static int ocfs2_xi_entry_usage(struct ocfs2_xattr_info *xi)
584{
585 return namevalue_size_xi(xi) +
586 sizeof(struct ocfs2_xattr_entry);
587}
534eaddd 588
199799a3
JB		 589static int ocfs2_xe_entry_usage(struct ocfs2_xattr_entry *xe)
590{
591 return namevalue_size_xe(xe) +
592 sizeof(struct ocfs2_xattr_entry);
534eaddd
TY		 593}
594
595int ocfs2_calc_security_init(struct inode *dir,
596 struct ocfs2_security_xattr_info *si,
597 int *want_clusters,
598 int *xattr_credits,
599 struct ocfs2_alloc_context **xattr_ac)
600{
601 int ret = 0;
602 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
603 int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
604 si->value_len);
605
606 /*
607 * The max space of security xattr taken inline is
608 * 256(name) + 80(value) + 16(entry) = 352 bytes,
609 * So reserve one metadata block for it is ok.
610 */
611 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
612 s_size > OCFS2_XATTR_FREE_IN_IBODY) {
613 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
614 if (ret) {
615 mlog_errno(ret);
616 return ret;
617 }
618 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
619 }
620
621 /* reserve clusters for xattr value which will be set in B tree*/
0e445b6f
TY		 622 if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
623 int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
624 si->value_len);
625
626 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
627 new_clusters);
628 *want_clusters += new_clusters;
629 }
534eaddd
TY		 630 return ret;
631}
632
89c38bd0
TY		 633int ocfs2_calc_xattr_init(struct inode *dir,
634 struct buffer_head *dir_bh,
635 int mode,
636 struct ocfs2_security_xattr_info *si,
637 int *want_clusters,
638 int *xattr_credits,
9b7895ef 639 int *want_meta)
89c38bd0
TY		 640{
641 int ret = 0;
642 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
0e445b6f 643 int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
89c38bd0
TY		 644
645 if (si->enable)
646 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
647 si->value_len);
648
649 if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
650 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
651 OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
652 "", NULL, 0);
653 if (acl_len > 0) {
654 a_size = ocfs2_xattr_entry_real_size(0, acl_len);
655 if (S_ISDIR(mode))
656 a_size <<= 1;
657 } else if (acl_len != 0 && acl_len != -ENODATA) {
658 mlog_errno(ret);
659 return ret;
660 }
661 }
662
663 if (!(s_size + a_size))
664 return ret;
665
666 /*
667 * The max space of security xattr taken inline is
668 * 256(name) + 80(value) + 16(entry) = 352 bytes,
669 * The max space of acl xattr taken inline is
670 * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
671 * when blocksize = 512, may reserve one more cluser for
672 * xattr bucket, otherwise reserve one metadata block
673 * for them is ok.
6c9fd1dc
TY		 674 * If this is a new directory with inline data,
675 * we choose to reserve the entire inline area for
676 * directory contents and force an external xattr block.
89c38bd0
TY		 677 */
678 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
6c9fd1dc 679 (S_ISDIR(mode) && ocfs2_supports_inline_data(osb)) ||
89c38bd0 680 (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
9b7895ef 681 *want_meta = *want_meta + 1;
89c38bd0
TY		 682 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
683 }
684
685 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
686 (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
687 *want_clusters += 1;
688 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
689 }
690
0e445b6f
TY		 691 /*
692 * reserve credits and clusters for xattrs which has large value
693 * and have to be set outside
694 */
695 if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
696 new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
697 si->value_len);
698 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
699 new_clusters);
700 *want_clusters += new_clusters;
701 }
89c38bd0
TY		 702 if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
703 acl_len > OCFS2_XATTR_INLINE_SIZE) {
0e445b6f
TY		 704 /* for directory, it has DEFAULT and ACCESS two types of acls */
705 new_clusters = (S_ISDIR(mode) ? 2 : 1) *
706 ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
707 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
708 new_clusters);
709 *want_clusters += new_clusters;
89c38bd0
TY		 710 }
711
712 return ret;
713}
714
f56654c4
TM		 715static int ocfs2_xattr_extend_allocation(struct inode *inode,
716 u32 clusters_to_add,
19b801f4 717 struct ocfs2_xattr_value_buf *vb,
78f30c31 718 struct ocfs2_xattr_set_ctxt *ctxt)
f56654c4
TM		 719{
720 int status = 0;
85db90e7 721 handle_t *handle = ctxt->handle;
f56654c4 722 enum ocfs2_alloc_restarted why;
19b801f4 723 u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
f99b9b7c 724 struct ocfs2_extent_tree et;
f56654c4
TM		 725
726 mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
727
5e404e9e 728 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
f99b9b7c 729
0cf2f763 730 status = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
2a50a743 731 OCFS2_JOURNAL_ACCESS_WRITE);
f56654c4
TM		 732 if (status < 0) {
733 mlog_errno(status);
734 goto leave;
735 }
736
19b801f4 737 prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
cbee7e1a
JB		 738 status = ocfs2_add_clusters_in_btree(handle,
739 &et,
f56654c4
TM		 740 &logical_start,
741 clusters_to_add,
742 0,
78f30c31
TM		 743 ctxt->data_ac,
744 ctxt->meta_ac,
f99b9b7c 745 &why);
85db90e7
TM		 746 if (status < 0) {
747 mlog_errno(status);
f56654c4
TM		 748 goto leave;
749 }
750
19b801f4 751 status = ocfs2_journal_dirty(handle, vb->vb_bh);
f56654c4
TM		 752 if (status < 0) {
753 mlog_errno(status);
754 goto leave;
755 }
756
19b801f4 757 clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) - prev_clusters;
f56654c4 758
85db90e7
TM		 759 /*
760 * We should have already allocated enough space before the transaction,
761 * so no need to restart.
762 */
763 BUG_ON(why != RESTART_NONE || clusters_to_add);
f56654c4
TM		 764
765leave:
f56654c4
TM		 766
767 return status;
768}
769
770static int __ocfs2_remove_xattr_range(struct inode *inode,
d72cc72d 771 struct ocfs2_xattr_value_buf *vb,
f56654c4 772 u32 cpos, u32 phys_cpos, u32 len,
492a8a33 773 unsigned int ext_flags,
78f30c31 774 struct ocfs2_xattr_set_ctxt *ctxt)
f56654c4
TM		 775{
776 int ret;
777 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
85db90e7 778 handle_t *handle = ctxt->handle;
f99b9b7c
JB		 779 struct ocfs2_extent_tree et;
780
5e404e9e 781 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
f56654c4 782
0cf2f763 783 ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
d72cc72d 784 OCFS2_JOURNAL_ACCESS_WRITE);
f56654c4
TM		 785 if (ret) {
786 mlog_errno(ret);
85db90e7 787 goto out;
f56654c4
TM		 788 }
789
dbdcf6a4 790 ret = ocfs2_remove_extent(handle, &et, cpos, len, ctxt->meta_ac,
78f30c31 791 &ctxt->dealloc);
f56654c4
TM		 792 if (ret) {
793 mlog_errno(ret);
85db90e7 794 goto out;
f56654c4
TM		 795 }
796
d72cc72d 797 le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
f56654c4 798
d72cc72d 799 ret = ocfs2_journal_dirty(handle, vb->vb_bh);
f56654c4
TM		 800 if (ret) {
801 mlog_errno(ret);
85db90e7 802 goto out;
f56654c4
TM		 803 }
804
492a8a33
TM		 805 if (ext_flags & OCFS2_EXT_REFCOUNTED)
806 ret = ocfs2_decrease_refcount(inode, handle,
807 ocfs2_blocks_to_clusters(inode->i_sb,
808 phys_blkno),
809 len, ctxt->meta_ac, &ctxt->dealloc, 1);
810 else
811 ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc,
812 phys_blkno, len);
f56654c4
TM		 813 if (ret)
814 mlog_errno(ret);
815
f56654c4 816out:
f56654c4
TM		 817 return ret;
818}
819
820static int ocfs2_xattr_shrink_size(struct inode *inode,
821 u32 old_clusters,
822 u32 new_clusters,
19b801f4 823 struct ocfs2_xattr_value_buf *vb,
78f30c31 824 struct ocfs2_xattr_set_ctxt *ctxt)
f56654c4
TM		 825{
826 int ret = 0;
492a8a33 827 unsigned int ext_flags;
f56654c4
TM		 828 u32 trunc_len, cpos, phys_cpos, alloc_size;
829 u64 block;
f56654c4
TM		 830
831 if (old_clusters <= new_clusters)
832 return 0;
833
834 cpos = new_clusters;
835 trunc_len = old_clusters - new_clusters;
836 while (trunc_len) {
837 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
d72cc72d 838 &alloc_size,
492a8a33 839 &vb->vb_xv->xr_list, &ext_flags);
f56654c4
TM		 840 if (ret) {
841 mlog_errno(ret);
842 goto out;
843 }
844
845 if (alloc_size > trunc_len)
846 alloc_size = trunc_len;
847
19b801f4 848 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
f56654c4 849 phys_cpos, alloc_size,
492a8a33 850 ext_flags, ctxt);
f56654c4
TM		 851 if (ret) {
852 mlog_errno(ret);
853 goto out;
854 }
855
856 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
8cb471e8
JB		 857 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode),
858 block, alloc_size);
f56654c4
TM		 859 cpos += alloc_size;
860 trunc_len -= alloc_size;
861 }
862
863out:
f56654c4
TM		 864 return ret;
865}
866
867static int ocfs2_xattr_value_truncate(struct inode *inode,
b3e5d379 868 struct ocfs2_xattr_value_buf *vb,
78f30c31
TM		 869 int len,
870 struct ocfs2_xattr_set_ctxt *ctxt)
f56654c4
TM		 871{
872 int ret;
873 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
b3e5d379 874 u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
f56654c4
TM		 875
876 if (new_clusters == old_clusters)
877 return 0;
878
879 if (new_clusters > old_clusters)
880 ret = ocfs2_xattr_extend_allocation(inode,
881 new_clusters - old_clusters,
b3e5d379 882 vb, ctxt);
f56654c4
TM		 883 else
884 ret = ocfs2_xattr_shrink_size(inode,
885 old_clusters, new_clusters,
b3e5d379 886 vb, ctxt);
f56654c4
TM		 887
888 return ret;
889}
cf1d6c76 890
936b8834
TM		 891static int ocfs2_xattr_list_entry(char *buffer, size_t size,
892 size_t *result, const char *prefix,
893 const char *name, int name_len)
894{
895 char *p = buffer + *result;
896 int prefix_len = strlen(prefix);
897 int total_len = prefix_len + name_len + 1;
898
899 *result += total_len;
900
901 /* we are just looking for how big our buffer needs to be */
902 if (!size)
903 return 0;
904
905 if (*result > size)
906 return -ERANGE;
907
908 memcpy(p, prefix, prefix_len);
909 memcpy(p + prefix_len, name, name_len);
910 p[prefix_len + name_len] = '\0';
911
912 return 0;
913}
914
cf1d6c76
TY		 915static int ocfs2_xattr_list_entries(struct inode *inode,
916 struct ocfs2_xattr_header *header,
917 char *buffer, size_t buffer_size)
918{
936b8834
TM		 919 size_t result = 0;
920 int i, type, ret;
921 const char *prefix, *name;
cf1d6c76
TY		 922
923 for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
924 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
936b8834
TM		 925 type = ocfs2_xattr_get_type(entry);
926 prefix = ocfs2_xattr_prefix(type);
927
928 if (prefix) {
929 name = (const char *)header +
930 le16_to_cpu(entry->xe_name_offset);
931
932 ret = ocfs2_xattr_list_entry(buffer, buffer_size,
933 &result, prefix, name,
934 entry->xe_name_len);
935 if (ret)
936 return ret;
cf1d6c76
TY		 937 }
938 }
939
936b8834 940 return result;
cf1d6c76
TY		 941}
942
8b2c0dba
TM		 943int ocfs2_has_inline_xattr_value_outside(struct inode *inode,
944 struct ocfs2_dinode *di)
945{
946 struct ocfs2_xattr_header *xh;
947 int i;
948
949 xh = (struct ocfs2_xattr_header *)
950 ((void *)di + inode->i_sb->s_blocksize -
951 le16_to_cpu(di->i_xattr_inline_size));
952
953 for (i = 0; i < le16_to_cpu(xh->xh_count); i++)
954 if (!ocfs2_xattr_is_local(&xh->xh_entries[i]))
955 return 1;
956
957 return 0;
958}
959
cf1d6c76
TY		 960static int ocfs2_xattr_ibody_list(struct inode *inode,
961 struct ocfs2_dinode *di,
962 char *buffer,
963 size_t buffer_size)
964{
965 struct ocfs2_xattr_header *header = NULL;
966 struct ocfs2_inode_info *oi = OCFS2_I(inode);
967 int ret = 0;
968
969 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
970 return ret;
971
972 header = (struct ocfs2_xattr_header *)
973 ((void *)di + inode->i_sb->s_blocksize -
974 le16_to_cpu(di->i_xattr_inline_size));
975
976 ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
977
978 return ret;
979}
980
981static int ocfs2_xattr_block_list(struct inode *inode,
982 struct ocfs2_dinode *di,
983 char *buffer,
984 size_t buffer_size)
985{
986 struct buffer_head *blk_bh = NULL;
0c044f0b 987 struct ocfs2_xattr_block *xb;
cf1d6c76
TY		 988 int ret = 0;
989
990 if (!di->i_xattr_loc)
991 return ret;
992
4ae1d69b
JB		 993 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
994 &blk_bh);
cf1d6c76
TY		 995 if (ret < 0) {
996 mlog_errno(ret);
997 return ret;
998 }
cf1d6c76 999
0c044f0b 1000 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
0c044f0b
TM		 1001 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1002 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
1003 ret = ocfs2_xattr_list_entries(inode, header,
1004 buffer, buffer_size);
47bca495
TM		 1005 } else
1006 ret = ocfs2_xattr_tree_list_index_block(inode, blk_bh,
0c044f0b 1007 buffer, buffer_size);
4ae1d69b 1008
cf1d6c76
TY		 1009 brelse(blk_bh);
1010
1011 return ret;
1012}
1013
1014ssize_t ocfs2_listxattr(struct dentry *dentry,
1015 char *buffer,
1016 size_t size)
1017{
1018 int ret = 0, i_ret = 0, b_ret = 0;
1019 struct buffer_head *di_bh = NULL;
1020 struct ocfs2_dinode *di = NULL;
1021 struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
1022
8154da3d
TY		 1023 if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
1024 return -EOPNOTSUPP;
1025
cf1d6c76
TY		 1026 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1027 return ret;
1028
1029 ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
1030 if (ret < 0) {
1031 mlog_errno(ret);
1032 return ret;
1033 }
1034
1035 di = (struct ocfs2_dinode *)di_bh->b_data;
1036
1037 down_read(&oi->ip_xattr_sem);
1038 i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
1039 if (i_ret < 0)
1040 b_ret = 0;
1041 else {
1042 if (buffer) {
1043 buffer += i_ret;
1044 size -= i_ret;
1045 }
1046 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
1047 buffer, size);
1048 if (b_ret < 0)
1049 i_ret = 0;
1050 }
1051 up_read(&oi->ip_xattr_sem);
1052 ocfs2_inode_unlock(dentry->d_inode, 0);
1053
1054 brelse(di_bh);
1055
1056 return i_ret + b_ret;
1057}
1058
1059static int ocfs2_xattr_find_entry(int name_index,
1060 const char *name,
1061 struct ocfs2_xattr_search *xs)
1062{
1063 struct ocfs2_xattr_entry *entry;
1064 size_t name_len;
1065 int i, cmp = 1;
1066
1067 if (name == NULL)
1068 return -EINVAL;
1069
1070 name_len = strlen(name);
1071 entry = xs->here;
1072 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
1073 cmp = name_index - ocfs2_xattr_get_type(entry);
1074 if (!cmp)
1075 cmp = name_len - entry->xe_name_len;
1076 if (!cmp)
1077 cmp = memcmp(name, (xs->base +
1078 le16_to_cpu(entry->xe_name_offset)),
1079 name_len);
1080 if (cmp == 0)
1081 break;
1082 entry += 1;
1083 }
1084 xs->here = entry;
1085
1086 return cmp ? -ENODATA : 0;
1087}
1088
1089static int ocfs2_xattr_get_value_outside(struct inode *inode,
589dc260 1090 struct ocfs2_xattr_value_root *xv,
cf1d6c76
TY		 1091 void *buffer,
1092 size_t len)
1093{
1094 u32 cpos, p_cluster, num_clusters, bpc, clusters;
1095 u64 blkno;
1096 int i, ret = 0;
1097 size_t cplen, blocksize;
1098 struct buffer_head *bh = NULL;
cf1d6c76
TY		 1099 struct ocfs2_extent_list *el;
1100
cf1d6c76
TY		 1101 el = &xv->xr_list;
1102 clusters = le32_to_cpu(xv->xr_clusters);
1103 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1104 blocksize = inode->i_sb->s_blocksize;
1105
1106 cpos = 0;
1107 while (cpos < clusters) {
1108 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1061f9c1 1109 &num_clusters, el, NULL);
cf1d6c76
TY		 1110 if (ret) {
1111 mlog_errno(ret);
1112 goto out;
1113 }
1114
1115 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1116 /* Copy ocfs2_xattr_value */
1117 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
8cb471e8
JB		 1118 ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1119 &bh, NULL);
cf1d6c76
TY		 1120 if (ret) {
1121 mlog_errno(ret);
1122 goto out;
1123 }
1124
1125 cplen = len >= blocksize ? blocksize : len;
1126 memcpy(buffer, bh->b_data, cplen);
1127 len -= cplen;
1128 buffer += cplen;
1129
1130 brelse(bh);
1131 bh = NULL;
1132 if (len == 0)
1133 break;
1134 }
1135 cpos += num_clusters;
1136 }
1137out:
1138 return ret;
1139}
1140
1141static int ocfs2_xattr_ibody_get(struct inode *inode,
1142 int name_index,
1143 const char *name,
1144 void *buffer,
1145 size_t buffer_size,
1146 struct ocfs2_xattr_search *xs)
1147{
1148 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1149 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
589dc260 1150 struct ocfs2_xattr_value_root *xv;
cf1d6c76
TY		 1151 size_t size;
1152 int ret = 0;
1153
1154 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1155 return -ENODATA;
1156
1157 xs->end = (void *)di + inode->i_sb->s_blocksize;
1158 xs->header = (struct ocfs2_xattr_header *)
1159 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1160 xs->base = (void *)xs->header;
1161 xs->here = xs->header->xh_entries;
1162
1163 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1164 if (ret)
1165 return ret;
1166 size = le64_to_cpu(xs->here->xe_value_size);
1167 if (buffer) {
1168 if (size > buffer_size)
1169 return -ERANGE;
1170 if (ocfs2_xattr_is_local(xs->here)) {
1171 memcpy(buffer, (void *)xs->base +
1172 le16_to_cpu(xs->here->xe_name_offset) +
1173 OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1174 } else {
589dc260
TM		 1175 xv = (struct ocfs2_xattr_value_root *)
1176 (xs->base + le16_to_cpu(
1177 xs->here->xe_name_offset) +
1178 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1179 ret = ocfs2_xattr_get_value_outside(inode, xv,
cf1d6c76
TY		 1180 buffer, size);
1181 if (ret < 0) {
1182 mlog_errno(ret);
1183 return ret;
1184 }
1185 }
1186 }
1187
1188 return size;
1189}
1190
1191static int ocfs2_xattr_block_get(struct inode *inode,
1192 int name_index,
1193 const char *name,
1194 void *buffer,
1195 size_t buffer_size,
1196 struct ocfs2_xattr_search *xs)
1197{
cf1d6c76 1198 struct ocfs2_xattr_block *xb;
589dc260 1199 struct ocfs2_xattr_value_root *xv;
cf1d6c76 1200 size_t size;
44d8e4e1
SM		 1201 int ret = -ENODATA, name_offset, name_len, i;
1202 int uninitialized_var(block_off);
cf1d6c76 1203
ba937127
JB		 1204 xs->bucket = ocfs2_xattr_bucket_new(inode);
1205 if (!xs->bucket) {
1206 ret = -ENOMEM;
1207 mlog_errno(ret);
1208 goto cleanup;
1209 }
589dc260 1210
54f443f4
JB		 1211 ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1212 if (ret) {
cf1d6c76 1213 mlog_errno(ret);
cf1d6c76
TY		 1214 goto cleanup;
1215 }
1216
6c1e183e
TY		 1217 if (xs->not_found) {
1218 ret = -ENODATA;
1219 goto cleanup;
1220 }
1221
54f443f4 1222 xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
cf1d6c76
TY		 1223 size = le64_to_cpu(xs->here->xe_value_size);
1224 if (buffer) {
1225 ret = -ERANGE;
1226 if (size > buffer_size)
1227 goto cleanup;
589dc260
TM		 1228
1229 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1230 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1231 i = xs->here - xs->header->xh_entries;
1232
1233 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
fd68a894 1234 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
ba937127 1235 bucket_xh(xs->bucket),
589dc260
TM		 1236 i,
1237 &block_off,
1238 &name_offset);
ba937127 1239 xs->base = bucket_block(xs->bucket, block_off);
589dc260 1240 }
cf1d6c76
TY		 1241 if (ocfs2_xattr_is_local(xs->here)) {
1242 memcpy(buffer, (void *)xs->base +
589dc260 1243 name_offset + name_len, size);
cf1d6c76 1244 } else {
589dc260
TM		 1245 xv = (struct ocfs2_xattr_value_root *)
1246 (xs->base + name_offset + name_len);
1247 ret = ocfs2_xattr_get_value_outside(inode, xv,
cf1d6c76
TY		 1248 buffer, size);
1249 if (ret < 0) {
1250 mlog_errno(ret);
1251 goto cleanup;
1252 }
1253 }
1254 }
1255 ret = size;
1256cleanup:
ba937127 1257 ocfs2_xattr_bucket_free(xs->bucket);
cf1d6c76 1258
54f443f4
JB		 1259 brelse(xs->xattr_bh);
1260 xs->xattr_bh = NULL;
cf1d6c76
TY		 1261 return ret;
1262}
1263
4e3e9d02
TY		 1264int ocfs2_xattr_get_nolock(struct inode *inode,
1265 struct buffer_head *di_bh,
0030e001
TY		 1266 int name_index,
1267 const char *name,
1268 void *buffer,
1269 size_t buffer_size)
cf1d6c76
TY		 1270{
1271 int ret;
1272 struct ocfs2_dinode *di = NULL;
cf1d6c76
TY		 1273 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1274 struct ocfs2_xattr_search xis = {
1275 .not_found = -ENODATA,
1276 };
1277 struct ocfs2_xattr_search xbs = {
1278 .not_found = -ENODATA,
1279 };
1280
8154da3d
TY		 1281 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1282 return -EOPNOTSUPP;
1283
cf1d6c76
TY		 1284 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1285 ret = -ENODATA;
1286
cf1d6c76
TY		 1287 xis.inode_bh = xbs.inode_bh = di_bh;
1288 di = (struct ocfs2_dinode *)di_bh->b_data;
1289
1290 down_read(&oi->ip_xattr_sem);
1291 ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1292 buffer_size, &xis);
6c1e183e 1293 if (ret == -ENODATA && di->i_xattr_loc)
cf1d6c76
TY		 1294 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1295 buffer_size, &xbs);
1296 up_read(&oi->ip_xattr_sem);
4e3e9d02
TY		 1297
1298 return ret;
1299}
1300
1301/* ocfs2_xattr_get()
1302 *
1303 * Copy an extended attribute into the buffer provided.
1304 * Buffer is NULL to compute the size of buffer required.
1305 */
1306static int ocfs2_xattr_get(struct inode *inode,
1307 int name_index,
1308 const char *name,
1309 void *buffer,
1310 size_t buffer_size)
1311{
1312 int ret;
1313 struct buffer_head *di_bh = NULL;
1314
1315 ret = ocfs2_inode_lock(inode, &di_bh, 0);
1316 if (ret < 0) {
1317 mlog_errno(ret);
1318 return ret;
1319 }
1320 ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1321 name, buffer, buffer_size);
1322
cf1d6c76
TY		 1323 ocfs2_inode_unlock(inode, 0);
1324
1325 brelse(di_bh);
1326
1327 return ret;
1328}
1329
1330static int __ocfs2_xattr_set_value_outside(struct inode *inode,
85db90e7 1331 handle_t *handle,
492a8a33 1332 struct ocfs2_xattr_value_buf *vb,
cf1d6c76
TY		 1333 const void *value,
1334 int value_len)
1335{
71d548a6 1336 int ret = 0, i, cp_len;
cf1d6c76
TY		 1337 u16 blocksize = inode->i_sb->s_blocksize;
1338 u32 p_cluster, num_clusters;
1339 u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1340 u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1341 u64 blkno;
1342 struct buffer_head *bh = NULL;
492a8a33
TM		 1343 unsigned int ext_flags;
1344 struct ocfs2_xattr_value_root *xv = vb->vb_xv;
cf1d6c76
TY		 1345
1346 BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1347
cf1d6c76
TY		 1348 while (cpos < clusters) {
1349 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1061f9c1 1350 &num_clusters, &xv->xr_list,
492a8a33 1351 &ext_flags);
cf1d6c76
TY		 1352 if (ret) {
1353 mlog_errno(ret);
85db90e7 1354 goto out;
cf1d6c76
TY		 1355 }
1356
492a8a33
TM		 1357 BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1358
cf1d6c76
TY		 1359 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1360
1361 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
8cb471e8
JB		 1362 ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1363 &bh, NULL);
cf1d6c76
TY		 1364 if (ret) {
1365 mlog_errno(ret);
85db90e7 1366 goto out;
cf1d6c76
TY		 1367 }
1368
1369 ret = ocfs2_journal_access(handle,
0cf2f763 1370 INODE_CACHE(inode),
cf1d6c76
TY		 1371 bh,
1372 OCFS2_JOURNAL_ACCESS_WRITE);
1373 if (ret < 0) {
1374 mlog_errno(ret);
85db90e7 1375 goto out;
cf1d6c76
TY		 1376 }
1377
1378 cp_len = value_len > blocksize ? blocksize : value_len;
1379 memcpy(bh->b_data, value, cp_len);
1380 value_len -= cp_len;
1381 value += cp_len;
1382 if (cp_len < blocksize)
1383 memset(bh->b_data + cp_len, 0,
1384 blocksize - cp_len);
1385
1386 ret = ocfs2_journal_dirty(handle, bh);
1387 if (ret < 0) {
1388 mlog_errno(ret);
85db90e7 1389 goto out;
cf1d6c76
TY		 1390 }
1391 brelse(bh);
1392 bh = NULL;
1393
1394 /*
1395 * XXX: do we need to empty all the following
1396 * blocks in this cluster?
1397 */
1398 if (!value_len)
1399 break;
1400 }
1401 cpos += num_clusters;
1402 }
cf1d6c76
TY		 1403out:
1404 brelse(bh);
1405
1406 return ret;
1407}
1408
1409static int ocfs2_xattr_cleanup(struct inode *inode,
85db90e7 1410 handle_t *handle,
cf1d6c76
TY		 1411 struct ocfs2_xattr_info *xi,
1412 struct ocfs2_xattr_search *xs,
512620f4 1413 struct ocfs2_xattr_value_buf *vb,
cf1d6c76
TY		 1414 size_t offs)
1415{
cf1d6c76 1416 int ret = 0;
cf1d6c76 1417 void *val = xs->base + offs;
199799a3 1418 size_t size = namevalue_size_xi(xi);
cf1d6c76 1419
0cf2f763 1420 ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
512620f4 1421 OCFS2_JOURNAL_ACCESS_WRITE);
cf1d6c76
TY		 1422 if (ret) {
1423 mlog_errno(ret);
85db90e7 1424 goto out;
cf1d6c76
TY		 1425 }
1426 /* Decrease xattr count */
1427 le16_add_cpu(&xs->header->xh_count, -1);
1428 /* Remove the xattr entry and tree root which has already be set*/
1429 memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
1430 memset(val, 0, size);
1431
512620f4 1432 ret = ocfs2_journal_dirty(handle, vb->vb_bh);
cf1d6c76
TY		 1433 if (ret < 0)
1434 mlog_errno(ret);
cf1d6c76
TY		 1435out:
1436 return ret;
1437}
1438
1439static int ocfs2_xattr_update_entry(struct inode *inode,
85db90e7 1440 handle_t *handle,
cf1d6c76
TY		 1441 struct ocfs2_xattr_info *xi,
1442 struct ocfs2_xattr_search *xs,
0c748e95 1443 struct ocfs2_xattr_value_buf *vb,
cf1d6c76
TY		 1444 size_t offs)
1445{
85db90e7 1446 int ret;
cf1d6c76 1447
0cf2f763 1448 ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
0c748e95 1449 OCFS2_JOURNAL_ACCESS_WRITE);
cf1d6c76
TY		 1450 if (ret) {
1451 mlog_errno(ret);
85db90e7 1452 goto out;
cf1d6c76
TY		 1453 }
1454
1455 xs->here->xe_name_offset = cpu_to_le16(offs);
6b240ff6
JB		 1456 xs->here->xe_value_size = cpu_to_le64(xi->xi_value_len);
1457 if (xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE)
cf1d6c76
TY		 1458 ocfs2_xattr_set_local(xs->here, 1);
1459 else
1460 ocfs2_xattr_set_local(xs->here, 0);
1461 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1462
0c748e95 1463 ret = ocfs2_journal_dirty(handle, vb->vb_bh);
cf1d6c76
TY		 1464 if (ret < 0)
1465 mlog_errno(ret);
cf1d6c76
TY		 1466out:
1467 return ret;
1468}
1469
1470/*
1471 * ocfs2_xattr_set_value_outside()
1472 *
1473 * Set large size value in B tree.
1474 */
1475static int ocfs2_xattr_set_value_outside(struct inode *inode,
1476 struct ocfs2_xattr_info *xi,
1477 struct ocfs2_xattr_search *xs,
78f30c31 1478 struct ocfs2_xattr_set_ctxt *ctxt,
512620f4 1479 struct ocfs2_xattr_value_buf *vb,
cf1d6c76
TY		 1480 size_t offs)
1481{
cf1d6c76
TY		 1482 void *val = xs->base + offs;
1483 struct ocfs2_xattr_value_root *xv = NULL;
199799a3 1484 size_t size = namevalue_size_xi(xi);
cf1d6c76
TY		 1485 int ret = 0;
1486
1487 memset(val, 0, size);
18853b95 1488 memcpy(val, xi->xi_name, xi->xi_name_len);
cf1d6c76 1489 xv = (struct ocfs2_xattr_value_root *)
18853b95 1490 (val + OCFS2_XATTR_SIZE(xi->xi_name_len));
cf1d6c76
TY		 1491 xv->xr_clusters = 0;
1492 xv->xr_last_eb_blk = 0;
1493 xv->xr_list.l_tree_depth = 0;
1494 xv->xr_list.l_count = cpu_to_le16(1);
1495 xv->xr_list.l_next_free_rec = 0;
512620f4 1496 vb->vb_xv = xv;
cf1d6c76 1497
6b240ff6 1498 ret = ocfs2_xattr_value_truncate(inode, vb, xi->xi_value_len, ctxt);
cf1d6c76
TY		 1499 if (ret < 0) {
1500 mlog_errno(ret);
1501 return ret;
1502 }
512620f4 1503 ret = ocfs2_xattr_update_entry(inode, ctxt->handle, xi, xs, vb, offs);
cf1d6c76
TY		 1504 if (ret < 0) {
1505 mlog_errno(ret);
1506 return ret;
1507 }
492a8a33 1508 ret = __ocfs2_xattr_set_value_outside(inode, ctxt->handle, vb,
6b240ff6 1509 xi->xi_value, xi->xi_value_len);
cf1d6c76
TY		 1510 if (ret < 0)
1511 mlog_errno(ret);
1512
1513 return ret;
1514}
1515
69a3e539
JB		 1516static int ocfs2_xa_check_space_helper(int needed_space, int free_start,
1517 int num_entries)
1518{
1519 int free_space;
1520
1521 if (!needed_space)
1522 return 0;
1523
1524 free_space = free_start -
1525 sizeof(struct ocfs2_xattr_header) -
1526 (num_entries * sizeof(struct ocfs2_xattr_entry)) -
1527 OCFS2_XATTR_HEADER_GAP;
1528 if (free_space < 0)
1529 return -EIO;
1530 if (free_space < needed_space)
1531 return -ENOSPC;
1532
1533 return 0;
1534}
1535
1536/* Give a pointer into the storage for the given offset */
1537static void *ocfs2_xa_offset_pointer(struct ocfs2_xa_loc *loc, int offset)
1538{
1539 BUG_ON(offset >= loc->xl_size);
1540 return loc->xl_ops->xlo_offset_pointer(loc, offset);
1541}
1542
11179f2c
JB		 1543/*
1544 * Wipe the name+value pair and allow the storage to reclaim it. This
1545 * must be followed by either removal of the entry or a call to
1546 * ocfs2_xa_add_namevalue().
1547 */
1548static void ocfs2_xa_wipe_namevalue(struct ocfs2_xa_loc *loc)
1549{
1550 loc->xl_ops->xlo_wipe_namevalue(loc);
1551}
1552
69a3e539
JB		 1553/*
1554 * Find lowest offset to a name+value pair. This is the start of our
1555 * downward-growing free space.
1556 */
1557static int ocfs2_xa_get_free_start(struct ocfs2_xa_loc *loc)
1558{
1559 return loc->xl_ops->xlo_get_free_start(loc);
1560}
1561
1562/* Can we reuse loc->xl_entry for xi? */
1563static int ocfs2_xa_can_reuse_entry(struct ocfs2_xa_loc *loc,
1564 struct ocfs2_xattr_info *xi)
1565{
1566 return loc->xl_ops->xlo_can_reuse(loc, xi);
1567}
1568
1569/* How much free space is needed to set the new value */
1570static int ocfs2_xa_check_space(struct ocfs2_xa_loc *loc,
1571 struct ocfs2_xattr_info *xi)
1572{
1573 return loc->xl_ops->xlo_check_space(loc, xi);
1574}
1575
1576static void ocfs2_xa_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1577{
1578 loc->xl_ops->xlo_add_entry(loc, name_hash);
1579 loc->xl_entry->xe_name_hash = cpu_to_le32(name_hash);
1580 /*
1581 * We can't leave the new entry's xe_name_offset at zero or
1582 * add_namevalue() will go nuts. We set it to the size of our
1583 * storage so that it can never be less than any other entry.
1584 */
1585 loc->xl_entry->xe_name_offset = cpu_to_le16(loc->xl_size);
1586}
1587
1588static void ocfs2_xa_add_namevalue(struct ocfs2_xa_loc *loc,
1589 struct ocfs2_xattr_info *xi)
1590{
1591 int size = namevalue_size_xi(xi);
1592 int nameval_offset;
1593 char *nameval_buf;
1594
1595 loc->xl_ops->xlo_add_namevalue(loc, size);
1596 loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
1597 loc->xl_entry->xe_name_len = xi->xi_name_len;
1598 ocfs2_xattr_set_type(loc->xl_entry, xi->xi_name_index);
1599 ocfs2_xattr_set_local(loc->xl_entry,
1600 xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE);
1601
1602 nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1603 nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
1604 memset(nameval_buf, 0, size);
1605 memcpy(nameval_buf, xi->xi_name, xi->xi_name_len);
1606}
1607
11179f2c
JB		 1608static void *ocfs2_xa_block_offset_pointer(struct ocfs2_xa_loc *loc,
1609 int offset)
1610{
11179f2c
JB		 1611 return (char *)loc->xl_header + offset;
1612}
1613
69a3e539
JB		 1614static int ocfs2_xa_block_can_reuse(struct ocfs2_xa_loc *loc,
1615 struct ocfs2_xattr_info *xi)
1616{
1617 /*
1618 * Block storage is strict. If the sizes aren't exact, we will
1619 * remove the old one and reinsert the new.
1620 */
1621 return namevalue_size_xe(loc->xl_entry) ==
1622 namevalue_size_xi(xi);
1623}
1624
1625static int ocfs2_xa_block_get_free_start(struct ocfs2_xa_loc *loc)
1626{
1627 struct ocfs2_xattr_header *xh = loc->xl_header;
1628 int i, count = le16_to_cpu(xh->xh_count);
1629 int offset, free_start = loc->xl_size;
1630
1631 for (i = 0; i < count; i++) {
1632 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1633 if (offset < free_start)
1634 free_start = offset;
1635 }
1636
1637 return free_start;
1638}
1639
1640static int ocfs2_xa_block_check_space(struct ocfs2_xa_loc *loc,
1641 struct ocfs2_xattr_info *xi)
1642{
1643 int count = le16_to_cpu(loc->xl_header->xh_count);
1644 int free_start = ocfs2_xa_get_free_start(loc);
1645 int needed_space = ocfs2_xi_entry_usage(xi);
1646
1647 /*
1648 * Block storage will reclaim the original entry before inserting
1649 * the new value, so we only need the difference. If the new
1650 * entry is smaller than the old one, we don't need anything.
1651 */
1652 if (loc->xl_entry) {
1653 /* Don't need space if we're reusing! */
1654 if (ocfs2_xa_can_reuse_entry(loc, xi))
1655 needed_space = 0;
1656 else
1657 needed_space -= ocfs2_xe_entry_usage(loc->xl_entry);
1658 }
1659 if (needed_space < 0)
1660 needed_space = 0;
1661 return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1662}
1663
11179f2c
JB		 1664/*
1665 * Block storage for xattrs keeps the name+value pairs compacted. When
1666 * we remove one, we have to shift any that preceded it towards the end.
1667 */
1668static void ocfs2_xa_block_wipe_namevalue(struct ocfs2_xa_loc *loc)
1669{
1670 int i, offset;
1671 int namevalue_offset, first_namevalue_offset, namevalue_size;
1672 struct ocfs2_xattr_entry *entry = loc->xl_entry;
1673 struct ocfs2_xattr_header *xh = loc->xl_header;
11179f2c
JB		 1674 int count = le16_to_cpu(xh->xh_count);
1675
1676 namevalue_offset = le16_to_cpu(entry->xe_name_offset);
199799a3 1677 namevalue_size = namevalue_size_xe(entry);
69a3e539 1678 first_namevalue_offset = ocfs2_xa_get_free_start(loc);
11179f2c
JB		 1679
1680 /* Shift the name+value pairs */
1681 memmove((char *)xh + first_namevalue_offset + namevalue_size,
1682 (char *)xh + first_namevalue_offset,
1683 namevalue_offset - first_namevalue_offset);
1684 memset((char *)xh + first_namevalue_offset, 0, namevalue_size);
1685
1686 /* Now tell xh->xh_entries about it */
1687 for (i = 0; i < count; i++) {
1688 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1689 if (offset < namevalue_offset)
1690 le16_add_cpu(&xh->xh_entries[i].xe_name_offset,
1691 namevalue_size);
1692 }
1693
1694 /*
1695 * Note that we don't update xh_free_start or xh_name_value_len
1696 * because they're not used in block-stored xattrs.
1697 */
1698}
1699
69a3e539
JB		 1700static void ocfs2_xa_block_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1701{
1702 int count = le16_to_cpu(loc->xl_header->xh_count);
1703 loc->xl_entry = &(loc->xl_header->xh_entries[count]);
1704 le16_add_cpu(&loc->xl_header->xh_count, 1);
1705 memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1706}
1707
1708static void ocfs2_xa_block_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1709{
1710 int free_start = ocfs2_xa_get_free_start(loc);
1711
1712 loc->xl_entry->xe_name_offset = cpu_to_le16(free_start - size);
1713}
1714
11179f2c
JB		 1715/*
1716 * Operations for xattrs stored in blocks. This includes inline inode
1717 * storage and unindexed ocfs2_xattr_blocks.
1718 */
1719static const struct ocfs2_xa_loc_operations ocfs2_xa_block_loc_ops = {
1720 .xlo_offset_pointer = ocfs2_xa_block_offset_pointer,
69a3e539
JB		 1721 .xlo_check_space = ocfs2_xa_block_check_space,
1722 .xlo_can_reuse = ocfs2_xa_block_can_reuse,
1723 .xlo_get_free_start = ocfs2_xa_block_get_free_start,
11179f2c 1724 .xlo_wipe_namevalue = ocfs2_xa_block_wipe_namevalue,
69a3e539
JB		 1725 .xlo_add_entry = ocfs2_xa_block_add_entry,
1726 .xlo_add_namevalue = ocfs2_xa_block_add_namevalue,
11179f2c
JB		 1727};
1728
1729static void *ocfs2_xa_bucket_offset_pointer(struct ocfs2_xa_loc *loc,
1730 int offset)
1731{
1732 struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1733 int block, block_offset;
1734
11179f2c
JB		 1735 /* The header is at the front of the bucket */
1736 block = offset >> bucket->bu_inode->i_sb->s_blocksize_bits;
1737 block_offset = offset % bucket->bu_inode->i_sb->s_blocksize;
1738
1739 return bucket_block(bucket, block) + block_offset;
1740}
1741
69a3e539
JB		 1742static int ocfs2_xa_bucket_can_reuse(struct ocfs2_xa_loc *loc,
1743 struct ocfs2_xattr_info *xi)
1744{
1745 return namevalue_size_xe(loc->xl_entry) >=
1746 namevalue_size_xi(xi);
1747}
1748
1749static int ocfs2_xa_bucket_get_free_start(struct ocfs2_xa_loc *loc)
1750{
1751 struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1752 return le16_to_cpu(bucket_xh(bucket)->xh_free_start);
1753}
1754
1755static int ocfs2_bucket_align_free_start(struct super_block *sb,
1756 int free_start, int size)
1757{
1758 /*
1759 * We need to make sure that the name+value pair fits within
1760 * one block.
1761 */
1762 if (((free_start - size) >> sb->s_blocksize_bits) !=
1763 ((free_start - 1) >> sb->s_blocksize_bits))
1764 free_start -= free_start % sb->s_blocksize;
1765
1766 return free_start;
1767}
1768
1769static int ocfs2_xa_bucket_check_space(struct ocfs2_xa_loc *loc,
1770 struct ocfs2_xattr_info *xi)
1771{
1772 int rc;
1773 int count = le16_to_cpu(loc->xl_header->xh_count);
1774 int free_start = ocfs2_xa_get_free_start(loc);
1775 int needed_space = ocfs2_xi_entry_usage(xi);
1776 int size = namevalue_size_xi(xi);
1777 struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1778 struct super_block *sb = bucket->bu_inode->i_sb;
1779
1780 /*
1781 * Bucket storage does not reclaim name+value pairs it cannot
1782 * reuse. They live as holes until the bucket fills, and then
1783 * the bucket is defragmented. However, the bucket can reclaim
1784 * the ocfs2_xattr_entry.
1785 */
1786 if (loc->xl_entry) {
1787 /* Don't need space if we're reusing! */
1788 if (ocfs2_xa_can_reuse_entry(loc, xi))
1789 needed_space = 0;
1790 else
1791 needed_space -= sizeof(struct ocfs2_xattr_entry);
1792 }
1793 BUG_ON(needed_space < 0);
1794
1795 if (free_start < size) {
1796 if (needed_space)
1797 return -ENOSPC;
1798 } else {
1799 /*
1800 * First we check if it would fit in the first place.
1801 * Below, we align the free start to a block. This may
1802 * slide us below the minimum gap. By checking unaligned
1803 * first, we avoid that error.
1804 */
1805 rc = ocfs2_xa_check_space_helper(needed_space, free_start,
1806 count);
1807 if (rc)
1808 return rc;
1809 free_start = ocfs2_bucket_align_free_start(sb, free_start,
1810 size);
1811 }
1812 return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1813}
1814
11179f2c
JB		 1815static void ocfs2_xa_bucket_wipe_namevalue(struct ocfs2_xa_loc *loc)
1816{
199799a3
JB		 1817 le16_add_cpu(&loc->xl_header->xh_name_value_len,
1818 -namevalue_size_xe(loc->xl_entry));
11179f2c
JB		 1819}
1820
69a3e539
JB		 1821static void ocfs2_xa_bucket_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1822{
1823 struct ocfs2_xattr_header *xh = loc->xl_header;
1824 int count = le16_to_cpu(xh->xh_count);
1825 int low = 0, high = count - 1, tmp;
1826 struct ocfs2_xattr_entry *tmp_xe;
1827
1828 /*
1829 * We keep buckets sorted by name_hash, so we need to find
1830 * our insert place.
1831 */
1832 while (low <= high && count) {
1833 tmp = (low + high) / 2;
1834 tmp_xe = &xh->xh_entries[tmp];
1835
1836 if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
1837 low = tmp + 1;
1838 else if (name_hash < le32_to_cpu(tmp_xe->xe_name_hash))
1839 high = tmp - 1;
1840 else {
1841 low = tmp;
1842 break;
1843 }
1844 }
1845
1846 if (low != count)
1847 memmove(&xh->xh_entries[low + 1],
1848 &xh->xh_entries[low],
1849 ((count - low) * sizeof(struct ocfs2_xattr_entry)));
1850
1851 le16_add_cpu(&xh->xh_count, 1);
1852 loc->xl_entry = &xh->xh_entries[low];
1853 memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1854}
1855
1856static void ocfs2_xa_bucket_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1857{
1858 int free_start = ocfs2_xa_get_free_start(loc);
1859 struct ocfs2_xattr_header *xh = loc->xl_header;
1860 struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1861 struct super_block *sb = bucket->bu_inode->i_sb;
1862 int nameval_offset;
1863
1864 free_start = ocfs2_bucket_align_free_start(sb, free_start, size);
1865 nameval_offset = free_start - size;
1866 loc->xl_entry->xe_name_offset = cpu_to_le16(nameval_offset);
1867 xh->xh_free_start = cpu_to_le16(nameval_offset);
1868 le16_add_cpu(&xh->xh_name_value_len, size);
1869
1870}
1871
11179f2c
JB		 1872/* Operations for xattrs stored in buckets. */
1873static const struct ocfs2_xa_loc_operations ocfs2_xa_bucket_loc_ops = {
1874 .xlo_offset_pointer = ocfs2_xa_bucket_offset_pointer,
69a3e539
JB		 1875 .xlo_check_space = ocfs2_xa_bucket_check_space,
1876 .xlo_can_reuse = ocfs2_xa_bucket_can_reuse,
1877 .xlo_get_free_start = ocfs2_xa_bucket_get_free_start,
11179f2c 1878 .xlo_wipe_namevalue = ocfs2_xa_bucket_wipe_namevalue,
69a3e539
JB		 1879 .xlo_add_entry = ocfs2_xa_bucket_add_entry,
1880 .xlo_add_namevalue = ocfs2_xa_bucket_add_namevalue,
11179f2c
JB		 1881};
1882
1883static void ocfs2_xa_remove_entry(struct ocfs2_xa_loc *loc)
1884{
bde1e540
JB		 1885 int index, count;
1886 struct ocfs2_xattr_header *xh = loc->xl_header;
1887 struct ocfs2_xattr_entry *entry = loc->xl_entry;
1888
11179f2c 1889 ocfs2_xa_wipe_namevalue(loc);
bde1e540
JB		 1890 loc->xl_entry = NULL;
1891
1892 le16_add_cpu(&xh->xh_count, -1);
1893 count = le16_to_cpu(xh->xh_count);
1894
1895 /*
1896 * Only zero out the entry if there are more remaining. This is
1897 * important for an empty bucket, as it keeps track of the
1898 * bucket's hash value. It doesn't hurt empty block storage.
1899 */
1900 if (count) {
1901 index = ((char *)entry - (char *)&xh->xh_entries) /
1902 sizeof(struct ocfs2_xattr_entry);
1903 memmove(&xh->xh_entries[index], &xh->xh_entries[index + 1],
1904 (count - index) * sizeof(struct ocfs2_xattr_entry));
1905 memset(&xh->xh_entries[count], 0,
1906 sizeof(struct ocfs2_xattr_entry));
1907 }
11179f2c
JB		 1908}
1909
69a3e539
JB		 1910/*
1911 * Prepares loc->xl_entry to receive the new xattr. This includes
1912 * properly setting up the name+value pair region. If loc->xl_entry
1913 * already exists, it will take care of modifying it appropriately.
1914 * This also includes deleting entries, but don't call this to remove
1915 * a non-existant entry. That's just a bug.
1916 *
1917 * Note that this modifies the data. You did journal_access already,
1918 * right?
1919 */
1920static int ocfs2_xa_prepare_entry(struct ocfs2_xa_loc *loc,
1921 struct ocfs2_xattr_info *xi,
1922 u32 name_hash)
1923{
1924 int rc = 0;
1925 int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
1926 char *nameval_buf;
1927
1928 if (!xi->xi_value) {
1929 ocfs2_xa_remove_entry(loc);
1930 goto out;
1931 }
1932
1933 rc = ocfs2_xa_check_space(loc, xi);
1934 if (rc)
1935 goto out;
1936
1937 if (loc->xl_entry) {
1938 if (ocfs2_xa_can_reuse_entry(loc, xi)) {
1939 nameval_buf = ocfs2_xa_offset_pointer(loc,
1940 le16_to_cpu(loc->xl_entry->xe_name_offset));
1941 memset(nameval_buf + name_size, 0,
1942 namevalue_size_xe(loc->xl_entry) - name_size);
1943 loc->xl_entry->xe_value_size =
1944 cpu_to_le64(xi->xi_value_len);
1945 goto out;
1946 }
1947
1948 ocfs2_xa_wipe_namevalue(loc);
1949 } else
1950 ocfs2_xa_add_entry(loc, name_hash);
1951
1952 /*
1953 * If we get here, we have a blank entry. Fill it. We grow our
1954 * name+value pair back from the end.
1955 */
1956 ocfs2_xa_add_namevalue(loc, xi);
1957
1958out:
1959 return rc;
1960}
1961
1962/*
1963 * Store the value portion of the name+value pair. This is either an
1964 * inline value or the tree root of an external value.
1965 */
1966static void ocfs2_xa_store_inline_value(struct ocfs2_xa_loc *loc,
1967 struct ocfs2_xattr_info *xi)
1968{
1969 int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1970 int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
1971 int size = namevalue_size_xi(xi);
1972 char *nameval_buf;
1973
1974 if (!xi->xi_value)
1975 return;
1976
1977 nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
1978 memcpy(nameval_buf + name_size, xi->xi_value, size - name_size);
1979}
1980
11179f2c
JB		 1981static void ocfs2_init_dinode_xa_loc(struct ocfs2_xa_loc *loc,
1982 struct inode *inode,
1983 struct buffer_head *bh,
1984 struct ocfs2_xattr_entry *entry)
1985{
1986 struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1987
1988 loc->xl_ops = &ocfs2_xa_block_loc_ops;
1989 loc->xl_storage = bh;
1990 loc->xl_entry = entry;
1991
1992 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1993 loc->xl_size = le16_to_cpu(di->i_xattr_inline_size);
1994 else {
1995 BUG_ON(entry);
1996 loc->xl_size = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1997 }
1998 loc->xl_header =
1999 (struct ocfs2_xattr_header *)(bh->b_data + bh->b_size -
2000 loc->xl_size);
2001}
2002
2003static void ocfs2_init_xattr_block_xa_loc(struct ocfs2_xa_loc *loc,
2004 struct buffer_head *bh,
2005 struct ocfs2_xattr_entry *entry)
2006{
2007 struct ocfs2_xattr_block *xb =
2008 (struct ocfs2_xattr_block *)bh->b_data;
2009
2010 BUG_ON(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED);
2011
2012 loc->xl_ops = &ocfs2_xa_block_loc_ops;
2013 loc->xl_storage = bh;
2014 loc->xl_header = &(xb->xb_attrs.xb_header);
2015 loc->xl_entry = entry;
2016 loc->xl_size = bh->b_size - offsetof(struct ocfs2_xattr_block,
2017 xb_attrs.xb_header);
2018}
2019
2020static void ocfs2_init_xattr_bucket_xa_loc(struct ocfs2_xa_loc *loc,
2021 struct ocfs2_xattr_bucket *bucket,
2022 struct ocfs2_xattr_entry *entry)
2023{
2024 loc->xl_ops = &ocfs2_xa_bucket_loc_ops;
2025 loc->xl_storage = bucket;
2026 loc->xl_header = bucket_xh(bucket);
2027 loc->xl_entry = entry;
2028 loc->xl_size = OCFS2_XATTR_BUCKET_SIZE;
2029}
2030
cf1d6c76
TY		 2031
2032/*
2033 * ocfs2_xattr_set_entry()
2034 *
2035 * Set extended attribute entry into inode or block.
2036 *
2037 * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
69a3e539 2038 * We first insert tree root(ocfs2_xattr_value_root) like a normal value,
cf1d6c76
TY		 2039 * then set value in B tree with set_value_outside().
2040 */
2041static int ocfs2_xattr_set_entry(struct inode *inode,
2042 struct ocfs2_xattr_info *xi,
2043 struct ocfs2_xattr_search *xs,
78f30c31 2044 struct ocfs2_xattr_set_ctxt *ctxt,
cf1d6c76
TY		 2045 int flag)
2046{
2047 struct ocfs2_xattr_entry *last;
2048 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2049 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
18853b95 2050 size_t min_offs = xs->end - xs->base;
cf1d6c76 2051 size_t size_l = 0;
85db90e7 2052 handle_t *handle = ctxt->handle;
cf1d6c76 2053 int free, i, ret;
69a3e539
JB		 2054 u32 name_hash = ocfs2_xattr_name_hash(inode, xi->xi_name,
2055 xi->xi_name_len);
2056 struct ocfs2_xa_loc loc;
cf1d6c76 2057 struct ocfs2_xattr_info xi_l = {
6b240ff6
JB		 2058 .xi_name_index = xi->xi_name_index,
2059 .xi_name = xi->xi_name,
18853b95 2060 .xi_name_len = xi->xi_name_len,
6b240ff6
JB		 2061 .xi_value = xi->xi_value,
2062 .xi_value_len = xi->xi_value_len,
cf1d6c76 2063 };
512620f4
JB		 2064 struct ocfs2_xattr_value_buf vb = {
2065 .vb_bh = xs->xattr_bh,
2066 .vb_access = ocfs2_journal_access_di,
2067 };
2068
2069 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
2070 BUG_ON(xs->xattr_bh == xs->inode_bh);
2071 vb.vb_access = ocfs2_journal_access_xb;
2072 } else
2073 BUG_ON(xs->xattr_bh != xs->inode_bh);
cf1d6c76
TY		 2074
2075 /* Compute min_offs, last and free space. */
2076 last = xs->header->xh_entries;
2077
2078 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
2079 size_t offs = le16_to_cpu(last->xe_name_offset);
2080 if (offs < min_offs)
2081 min_offs = offs;
2082 last += 1;
2083 }
2084
4442f518 2085 free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
cf1d6c76 2086 if (free < 0)
b37c4d84 2087 return -EIO;
cf1d6c76 2088
199799a3
JB		 2089 if (!xs->not_found)
2090 free += ocfs2_xe_entry_usage(xs->here);
2091
cf1d6c76 2092 /* Check free space in inode or block */
199799a3
JB		 2093 if (xi->xi_value) {
2094 if (free < ocfs2_xi_entry_usage(xi)) {
cf1d6c76
TY		 2095 ret = -ENOSPC;
2096 goto out;
2097 }
199799a3
JB		 2098 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2099 size_l = namevalue_size_xi(xi);
2100 xi_l.xi_value = (void *)&def_xv;
2101 xi_l.xi_value_len = OCFS2_XATTR_ROOT_SIZE;
cf1d6c76
TY		 2102 }
2103 }
2104
2105 if (!xs->not_found) {
2106 /* For existing extended attribute */
199799a3 2107 size_t size = namevalue_size_xe(xs->here);
cf1d6c76
TY		 2108 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
2109 void *val = xs->base + offs;
2110
2111 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
2112 /* Replace existing local xattr with tree root */
2113 ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
512620f4 2114 ctxt, &vb, offs);
cf1d6c76
TY		 2115 if (ret < 0)
2116 mlog_errno(ret);
2117 goto out;
2118 } else if (!ocfs2_xattr_is_local(xs->here)) {
2119 /* For existing xattr which has value outside */
512620f4 2120 vb.vb_xv = (struct ocfs2_xattr_value_root *)
18853b95 2121 (val + OCFS2_XATTR_SIZE(xi->xi_name_len));
cf1d6c76 2122
6b240ff6 2123 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
cf1d6c76
TY		 2124 /*
2125 * If new value need set outside also,
2126 * first truncate old value to new value,
2127 * then set new value with set_value_outside().
2128 */
2129 ret = ocfs2_xattr_value_truncate(inode,
6b240ff6
JB		 2130 &vb,
2131 xi->xi_value_len,
2132 ctxt);
cf1d6c76
TY		 2133 if (ret < 0) {
2134 mlog_errno(ret);
2135 goto out;
2136 }
2137
85db90e7
TM		 2138 ret = ocfs2_xattr_update_entry(inode,
2139 handle,
2140 xi,
2141 xs,
0c748e95 2142 &vb,
85db90e7 2143 offs);
cf1d6c76
TY		 2144 if (ret < 0) {
2145 mlog_errno(ret);
2146 goto out;
2147 }
2148
85db90e7 2149 ret = __ocfs2_xattr_set_value_outside(inode,
6b240ff6
JB		 2150 handle,
2151 &vb,
2152 xi->xi_value,
2153 xi->xi_value_len);
cf1d6c76
TY		 2154 if (ret < 0)
2155 mlog_errno(ret);
2156 goto out;
2157 } else {
2158 /*
2159 * If new value need set in local,
2160 * just trucate old value to zero.
2161 */
2162 ret = ocfs2_xattr_value_truncate(inode,
b3e5d379 2163 &vb,
85db90e7
TM		 2164 0,
2165 ctxt);
cf1d6c76
TY		 2166 if (ret < 0)
2167 mlog_errno(ret);
2168 }
2169 }
2170 }
2171
0cf2f763 2172 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), xs->inode_bh,
512620f4 2173 OCFS2_JOURNAL_ACCESS_WRITE);
cf1d6c76
TY		 2174 if (ret) {
2175 mlog_errno(ret);
85db90e7 2176 goto out;
cf1d6c76
TY		 2177 }
2178
2179 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
0cf2f763 2180 ret = vb.vb_access(handle, INODE_CACHE(inode), vb.vb_bh,
512620f4 2181 OCFS2_JOURNAL_ACCESS_WRITE);
cf1d6c76
TY		 2182 if (ret) {
2183 mlog_errno(ret);
85db90e7 2184 goto out;
cf1d6c76
TY		 2185 }
2186 }
2187
69a3e539
JB		 2188 if (xs->xattr_bh == xs->inode_bh)
2189 ocfs2_init_dinode_xa_loc(&loc, inode, xs->inode_bh,
2190 xs->not_found ? NULL : xs->here);
2191 else
2192 ocfs2_init_xattr_block_xa_loc(&loc, xs->xattr_bh,
2193 xs->not_found ? NULL : xs->here);
2194
cf1d6c76 2195 /*
69a3e539
JB		 2196 * Prepare our entry and insert the inline value. This will
2197 * be a value tree root for values that are larger than
2198 * OCFS2_XATTR_INLINE_SIZE.
cf1d6c76 2199 */
69a3e539
JB		 2200 ret = ocfs2_xa_prepare_entry(&loc, xi, name_hash);
2201 if (ret) {
2202 if (ret != -ENOSPC)
2203 mlog_errno(ret);
2204 goto out;
2205 }
2206 /* XXX For now, until we make ocfs2_xa_prepare_entry() primary */
2207 BUG_ON(ret == -ENOSPC);
2208 ocfs2_xa_store_inline_value(&loc, xi);
2209 xs->here = loc.xl_entry;
cf1d6c76
TY		 2210
2211 if (!(flag & OCFS2_INLINE_XATTR_FL)) {
2212 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
2213 if (ret < 0) {
2214 mlog_errno(ret);
85db90e7 2215 goto out;
cf1d6c76
TY		 2216 }
2217 }
2218
2219 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
2220 (flag & OCFS2_INLINE_XATTR_FL)) {
2221 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2222 unsigned int xattrsize = osb->s_xattr_inline_size;
2223
2224 /*
2225 * Adjust extent record count or inline data size
2226 * to reserve space for extended attribute.
2227 */
2228 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2229 struct ocfs2_inline_data *idata = &di->id2.i_data;
2230 le16_add_cpu(&idata->id_count, -xattrsize);
2231 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
2232 struct ocfs2_extent_list *el = &di->id2.i_list;
2233 le16_add_cpu(&el->l_count, -(xattrsize /
2234 sizeof(struct ocfs2_extent_rec)));
2235 }
2236 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
2237 }
2238 /* Update xattr flag */
2239 spin_lock(&oi->ip_lock);
2240 oi->ip_dyn_features |= flag;
2241 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2242 spin_unlock(&oi->ip_lock);
cf1d6c76
TY		 2243
2244 ret = ocfs2_journal_dirty(handle, xs->inode_bh);
2245 if (ret < 0)
2246 mlog_errno(ret);
2247
6b240ff6 2248 if (!ret && xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
cf1d6c76
TY		 2249 /*
2250 * Set value outside in B tree.
2251 * This is the second step for value size > INLINE_SIZE.
2252 */
2253 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
512620f4
JB		 2254 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, ctxt,
2255 &vb, offs);
cf1d6c76
TY		 2256 if (ret < 0) {
2257 int ret2;
2258
2259 mlog_errno(ret);
2260 /*
2261 * If set value outside failed, we have to clean
2262 * the junk tree root we have already set in local.
2263 */
85db90e7 2264 ret2 = ocfs2_xattr_cleanup(inode, ctxt->handle,
512620f4 2265 xi, xs, &vb, offs);
cf1d6c76
TY		 2266 if (ret2 < 0)
2267 mlog_errno(ret2);
2268 }
2269 }
2270out:
2271 return ret;
cf1d6c76
TY		 2272}
2273
ce9c5a54
TM		 2274/*
2275 * In xattr remove, if it is stored outside and refcounted, we may have
2276 * the chance to split the refcount tree. So need the allocators.
2277 */
2278static int ocfs2_lock_xattr_remove_allocators(struct inode *inode,
2279 struct ocfs2_xattr_value_root *xv,
2280 struct ocfs2_caching_info *ref_ci,
2281 struct buffer_head *ref_root_bh,
2282 struct ocfs2_alloc_context **meta_ac,
2283 int *ref_credits)
2284{
2285 int ret, meta_add = 0;
2286 u32 p_cluster, num_clusters;
2287 unsigned int ext_flags;
2288
2289 *ref_credits = 0;
2290 ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
2291 &num_clusters,
2292 &xv->xr_list,
2293 &ext_flags);
2294 if (ret) {
2295 mlog_errno(ret);
2296 goto out;
2297 }
2298
2299 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
2300 goto out;
2301
2302 ret = ocfs2_refcounted_xattr_delete_need(inode, ref_ci,
2303 ref_root_bh, xv,
2304 &meta_add, ref_credits);
2305 if (ret) {
2306 mlog_errno(ret);
2307 goto out;
2308 }
2309
2310 ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
2311 meta_add, meta_ac);
2312 if (ret)
2313 mlog_errno(ret);
2314
2315out:
2316 return ret;
2317}
2318
cf1d6c76 2319static int ocfs2_remove_value_outside(struct inode*inode,
4311901d 2320 struct ocfs2_xattr_value_buf *vb,
ce9c5a54
TM		 2321 struct ocfs2_xattr_header *header,
2322 struct ocfs2_caching_info *ref_ci,
2323 struct buffer_head *ref_root_bh)
cf1d6c76 2324{
ce9c5a54 2325 int ret = 0, i, ref_credits;
78f30c31
TM		 2326 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2327 struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
ce9c5a54 2328 void *val;
78f30c31
TM		 2329
2330 ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
cf1d6c76
TY		 2331
2332 for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
2333 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
2334
ce9c5a54
TM		 2335 if (ocfs2_xattr_is_local(entry))
2336 continue;
cf1d6c76 2337
ce9c5a54
TM		 2338 val = (void *)header +
2339 le16_to_cpu(entry->xe_name_offset);
2340 vb->vb_xv = (struct ocfs2_xattr_value_root *)
2341 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
2342
2343 ret = ocfs2_lock_xattr_remove_allocators(inode, vb->vb_xv,
2344 ref_ci, ref_root_bh,
2345 &ctxt.meta_ac,
2346 &ref_credits);
2347
2348 ctxt.handle = ocfs2_start_trans(osb, ref_credits +
2349 ocfs2_remove_extent_credits(osb->sb));
2350 if (IS_ERR(ctxt.handle)) {
2351 ret = PTR_ERR(ctxt.handle);
2352 mlog_errno(ret);
2353 break;
2354 }
2355
2356 ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
2357 if (ret < 0) {
2358 mlog_errno(ret);
2359 break;
2360 }
2361
2362 ocfs2_commit_trans(osb, ctxt.handle);
2363 if (ctxt.meta_ac) {
2364 ocfs2_free_alloc_context(ctxt.meta_ac);
2365 ctxt.meta_ac = NULL;
cf1d6c76
TY		 2366 }
2367 }
2368
ce9c5a54
TM		 2369 if (ctxt.meta_ac)
2370 ocfs2_free_alloc_context(ctxt.meta_ac);
78f30c31
TM		 2371 ocfs2_schedule_truncate_log_flush(osb, 1);
2372 ocfs2_run_deallocs(osb, &ctxt.dealloc);
cf1d6c76
TY		 2373 return ret;
2374}
2375
2376static int ocfs2_xattr_ibody_remove(struct inode *inode,
ce9c5a54
TM		 2377 struct buffer_head *di_bh,
2378 struct ocfs2_caching_info *ref_ci,
2379 struct buffer_head *ref_root_bh)
cf1d6c76
TY		 2380{
2381
2382 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2383 struct ocfs2_xattr_header *header;
2384 int ret;
4311901d
JB		 2385 struct ocfs2_xattr_value_buf vb = {
2386 .vb_bh = di_bh,
2387 .vb_access = ocfs2_journal_access_di,
2388 };
cf1d6c76
TY		 2389
2390 header = (struct ocfs2_xattr_header *)
2391 ((void *)di + inode->i_sb->s_blocksize -
2392 le16_to_cpu(di->i_xattr_inline_size));
2393
ce9c5a54
TM		 2394 ret = ocfs2_remove_value_outside(inode, &vb, header,
2395 ref_ci, ref_root_bh);
cf1d6c76
TY		 2396
2397 return ret;
2398}
2399
ce9c5a54
TM		 2400struct ocfs2_rm_xattr_bucket_para {
2401 struct ocfs2_caching_info *ref_ci;
2402 struct buffer_head *ref_root_bh;
2403};
2404
cf1d6c76 2405static int ocfs2_xattr_block_remove(struct inode *inode,
ce9c5a54
TM		 2406 struct buffer_head *blk_bh,
2407 struct ocfs2_caching_info *ref_ci,
2408 struct buffer_head *ref_root_bh)
cf1d6c76
TY		 2409{
2410 struct ocfs2_xattr_block *xb;
cf1d6c76 2411 int ret = 0;
4311901d
JB		 2412 struct ocfs2_xattr_value_buf vb = {
2413 .vb_bh = blk_bh,
2414 .vb_access = ocfs2_journal_access_xb,
2415 };
ce9c5a54
TM		 2416 struct ocfs2_rm_xattr_bucket_para args = {
2417 .ref_ci = ref_ci,
2418 .ref_root_bh = ref_root_bh,
2419 };
cf1d6c76
TY		 2420
2421 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
a3944256
TM		 2422 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2423 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
ce9c5a54
TM		 2424 ret = ocfs2_remove_value_outside(inode, &vb, header,
2425 ref_ci, ref_root_bh);
a3944256 2426 } else
47bca495
TM		 2427 ret = ocfs2_iterate_xattr_index_block(inode,
2428 blk_bh,
2429 ocfs2_rm_xattr_cluster,
ce9c5a54 2430 &args);
cf1d6c76
TY		 2431
2432 return ret;
2433}
2434
08413899 2435static int ocfs2_xattr_free_block(struct inode *inode,
ce9c5a54
TM		 2436 u64 block,
2437 struct ocfs2_caching_info *ref_ci,
2438 struct buffer_head *ref_root_bh)
08413899
TM		 2439{
2440 struct inode *xb_alloc_inode;
2441 struct buffer_head *xb_alloc_bh = NULL;
2442 struct buffer_head *blk_bh = NULL;
2443 struct ocfs2_xattr_block *xb;
2444 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2445 handle_t *handle;
2446 int ret = 0;
2447 u64 blk, bg_blkno;
2448 u16 bit;
2449
4ae1d69b 2450 ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
08413899
TM		 2451 if (ret < 0) {
2452 mlog_errno(ret);
2453 goto out;
2454 }
2455
ce9c5a54 2456 ret = ocfs2_xattr_block_remove(inode, blk_bh, ref_ci, ref_root_bh);
08413899
TM		 2457 if (ret < 0) {
2458 mlog_errno(ret);
2459 goto out;
2460 }
2461
4ae1d69b 2462 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
08413899
TM		 2463 blk = le64_to_cpu(xb->xb_blkno);
2464 bit = le16_to_cpu(xb->xb_suballoc_bit);
2465 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2466
2467 xb_alloc_inode = ocfs2_get_system_file_inode(osb,
2468 EXTENT_ALLOC_SYSTEM_INODE,
2469 le16_to_cpu(xb->xb_suballoc_slot));
2470 if (!xb_alloc_inode) {
2471 ret = -ENOMEM;
2472 mlog_errno(ret);
2473 goto out;
2474 }
2475 mutex_lock(&xb_alloc_inode->i_mutex);
2476
2477 ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
2478 if (ret < 0) {
2479 mlog_errno(ret);
2480 goto out_mutex;
2481 }
2482
2483 handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
2484 if (IS_ERR(handle)) {
2485 ret = PTR_ERR(handle);
2486 mlog_errno(ret);
2487 goto out_unlock;
2488 }
2489
2490 ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
2491 bit, bg_blkno, 1);
2492 if (ret < 0)
2493 mlog_errno(ret);
2494
2495 ocfs2_commit_trans(osb, handle);
2496out_unlock:
2497 ocfs2_inode_unlock(xb_alloc_inode, 1);
2498 brelse(xb_alloc_bh);
2499out_mutex:
2500 mutex_unlock(&xb_alloc_inode->i_mutex);
2501 iput(xb_alloc_inode);
2502out:
2503 brelse(blk_bh);
2504 return ret;
2505}
2506
cf1d6c76
TY		 2507/*
2508 * ocfs2_xattr_remove()
2509 *
2510 * Free extended attribute resources associated with this inode.
2511 */
2512int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
2513{
cf1d6c76
TY		 2514 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2515 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
ce9c5a54
TM		 2516 struct ocfs2_refcount_tree *ref_tree = NULL;
2517 struct buffer_head *ref_root_bh = NULL;
2518 struct ocfs2_caching_info *ref_ci = NULL;
cf1d6c76
TY		 2519 handle_t *handle;
2520 int ret;
2521
8154da3d
TY		 2522 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2523 return 0;
2524
cf1d6c76
TY		 2525 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
2526 return 0;
2527
ce9c5a54
TM		 2528 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
2529 ret = ocfs2_lock_refcount_tree(OCFS2_SB(inode->i_sb),
2530 le64_to_cpu(di->i_refcount_loc),
2531 1, &ref_tree, &ref_root_bh);
2532 if (ret) {
2533 mlog_errno(ret);
2534 goto out;
2535 }
2536 ref_ci = &ref_tree->rf_ci;
2537
2538 }
2539
cf1d6c76 2540 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
ce9c5a54
TM		 2541 ret = ocfs2_xattr_ibody_remove(inode, di_bh,
2542 ref_ci, ref_root_bh);
cf1d6c76
TY		 2543 if (ret < 0) {
2544 mlog_errno(ret);
2545 goto out;
2546 }
2547 }
cf1d6c76 2548
08413899
TM		 2549 if (di->i_xattr_loc) {
2550 ret = ocfs2_xattr_free_block(inode,
ce9c5a54
TM		 2551 le64_to_cpu(di->i_xattr_loc),
2552 ref_ci, ref_root_bh);
cf1d6c76
TY		 2553 if (ret < 0) {
2554 mlog_errno(ret);
2555 goto out;
2556 }
2557 }
2558
2559 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
2560 OCFS2_INODE_UPDATE_CREDITS);
2561 if (IS_ERR(handle)) {
2562 ret = PTR_ERR(handle);
2563 mlog_errno(ret);
2564 goto out;
2565 }
0cf2f763 2566 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
84008972 2567 OCFS2_JOURNAL_ACCESS_WRITE);
cf1d6c76
TY		 2568 if (ret) {
2569 mlog_errno(ret);
2570 goto out_commit;
2571 }
2572
08413899 2573 di->i_xattr_loc = 0;
cf1d6c76
TY		 2574
2575 spin_lock(&oi->ip_lock);
2576 oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
2577 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2578 spin_unlock(&oi->ip_lock);
2579
2580 ret = ocfs2_journal_dirty(handle, di_bh);
2581 if (ret < 0)
2582 mlog_errno(ret);
2583out_commit:
2584 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
2585out:
ce9c5a54
TM		 2586 if (ref_tree)
2587 ocfs2_unlock_refcount_tree(OCFS2_SB(inode->i_sb), ref_tree, 1);
2588 brelse(ref_root_bh);
cf1d6c76
TY		 2589 return ret;
2590}
2591
2592static int ocfs2_xattr_has_space_inline(struct inode *inode,
2593 struct ocfs2_dinode *di)
2594{
2595 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2596 unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
2597 int free;
2598
2599 if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
2600 return 0;
2601
2602 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2603 struct ocfs2_inline_data *idata = &di->id2.i_data;
2604 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
2605 } else if (ocfs2_inode_is_fast_symlink(inode)) {
2606 free = ocfs2_fast_symlink_chars(inode->i_sb) -
2607 le64_to_cpu(di->i_size);
2608 } else {
2609 struct ocfs2_extent_list *el = &di->id2.i_list;
2610 free = (le16_to_cpu(el->l_count) -
2611 le16_to_cpu(el->l_next_free_rec)) *
2612 sizeof(struct ocfs2_extent_rec);
2613 }
2614 if (free >= xattrsize)
2615 return 1;
2616
2617 return 0;
2618}
2619
2620/*
2621 * ocfs2_xattr_ibody_find()
2622 *
2623 * Find extended attribute in inode block and
2624 * fill search info into struct ocfs2_xattr_search.
2625 */
2626static int ocfs2_xattr_ibody_find(struct inode *inode,
2627 int name_index,
2628 const char *name,
2629 struct ocfs2_xattr_search *xs)
2630{
2631 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2632 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2633 int ret;
2634 int has_space = 0;
2635
2636 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2637 return 0;
2638
2639 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2640 down_read(&oi->ip_alloc_sem);
2641 has_space = ocfs2_xattr_has_space_inline(inode, di);
2642 up_read(&oi->ip_alloc_sem);
2643 if (!has_space)
2644 return 0;
2645 }
2646
2647 xs->xattr_bh = xs->inode_bh;
2648 xs->end = (void *)di + inode->i_sb->s_blocksize;
2649 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
2650 xs->header = (struct ocfs2_xattr_header *)
2651 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
2652 else
2653 xs->header = (struct ocfs2_xattr_header *)
2654 (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
2655 xs->base = (void *)xs->header;
2656 xs->here = xs->header->xh_entries;
2657
2658 /* Find the named attribute. */
2659 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2660 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2661 if (ret && ret != -ENODATA)
2662 return ret;
2663 xs->not_found = ret;
2664 }
2665
2666 return 0;
2667}
2668
2669/*
2670 * ocfs2_xattr_ibody_set()
2671 *
2672 * Set, replace or remove an extended attribute into inode block.
2673 *
2674 */
2675static int ocfs2_xattr_ibody_set(struct inode *inode,
2676 struct ocfs2_xattr_info *xi,
78f30c31
TM		 2677 struct ocfs2_xattr_search *xs,
2678 struct ocfs2_xattr_set_ctxt *ctxt)
cf1d6c76
TY		 2679{
2680 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2681 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2682 int ret;
2683
2684 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2685 return -ENOSPC;
2686
2687 down_write(&oi->ip_alloc_sem);
2688 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2689 if (!ocfs2_xattr_has_space_inline(inode, di)) {
2690 ret = -ENOSPC;
2691 goto out;
2692 }
2693 }
2694
78f30c31 2695 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
cf1d6c76
TY		 2696 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
2697out:
2698 up_write(&oi->ip_alloc_sem);
2699
2700 return ret;
2701}
2702
2703/*
2704 * ocfs2_xattr_block_find()
2705 *
2706 * Find extended attribute in external block and
2707 * fill search info into struct ocfs2_xattr_search.
2708 */
2709static int ocfs2_xattr_block_find(struct inode *inode,
2710 int name_index,
2711 const char *name,
2712 struct ocfs2_xattr_search *xs)
2713{
2714 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2715 struct buffer_head *blk_bh = NULL;
589dc260 2716 struct ocfs2_xattr_block *xb;
cf1d6c76
TY		 2717 int ret = 0;
2718
2719 if (!di->i_xattr_loc)
2720 return ret;
2721
4ae1d69b
JB		 2722 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2723 &blk_bh);
cf1d6c76
TY		 2724 if (ret < 0) {
2725 mlog_errno(ret);
2726 return ret;
2727 }
f6087fb7 2728
cf1d6c76 2729 xs->xattr_bh = blk_bh;
4ae1d69b 2730 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
589dc260
TM		 2731
2732 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2733 xs->header = &xb->xb_attrs.xb_header;
2734 xs->base = (void *)xs->header;
2735 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2736 xs->here = xs->header->xh_entries;
2737
2738 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2739 } else
2740 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2741 name_index,
2742 name, xs);
cf1d6c76 2743
cf1d6c76
TY		 2744 if (ret && ret != -ENODATA) {
2745 xs->xattr_bh = NULL;
2746 goto cleanup;
2747 }
2748 xs->not_found = ret;
2749 return 0;
cf1d6c76
TY		 2750cleanup:
2751 brelse(blk_bh);
2752
2753 return ret;
2754}
2755
5aea1f0e
TM		 2756static int ocfs2_create_xattr_block(handle_t *handle,
2757 struct inode *inode,
2758 struct buffer_head *inode_bh,
2759 struct ocfs2_alloc_context *meta_ac,
a7fe7a3a
TM		 2760 struct buffer_head **ret_bh,
2761 int indexed)
5aea1f0e
TM		 2762{
2763 int ret;
2764 u16 suballoc_bit_start;
2765 u32 num_got;
2766 u64 first_blkno;
2767 struct ocfs2_dinode *di = (struct ocfs2_dinode *)inode_bh->b_data;
2768 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2769 struct buffer_head *new_bh = NULL;
2770 struct ocfs2_xattr_block *xblk;
2771
2772 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), inode_bh,
2773 OCFS2_JOURNAL_ACCESS_CREATE);
2774 if (ret < 0) {
2775 mlog_errno(ret);
2776 goto end;
2777 }
2778
2779 ret = ocfs2_claim_metadata(osb, handle, meta_ac, 1,
2780 &suballoc_bit_start, &num_got,
2781 &first_blkno);
2782 if (ret < 0) {
2783 mlog_errno(ret);
2784 goto end;
2785 }
2786
2787 new_bh = sb_getblk(inode->i_sb, first_blkno);
2788 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2789
2790 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode),
2791 new_bh,
2792 OCFS2_JOURNAL_ACCESS_CREATE);
2793 if (ret < 0) {
2794 mlog_errno(ret);
2795 goto end;
2796 }
2797
2798 /* Initialize ocfs2_xattr_block */
2799 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2800 memset(xblk, 0, inode->i_sb->s_blocksize);
2801 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
b89c5428 2802 xblk->xb_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
5aea1f0e
TM		 2803 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2804 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
2805 xblk->xb_blkno = cpu_to_le64(first_blkno);
2806
a7fe7a3a
TM		 2807 if (indexed) {
2808 struct ocfs2_xattr_tree_root *xr = &xblk->xb_attrs.xb_root;
2809 xr->xt_clusters = cpu_to_le32(1);
2810 xr->xt_last_eb_blk = 0;
2811 xr->xt_list.l_tree_depth = 0;
2812 xr->xt_list.l_count = cpu_to_le16(
2813 ocfs2_xattr_recs_per_xb(inode->i_sb));
2814 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2815 xblk->xb_flags = cpu_to_le16(OCFS2_XATTR_INDEXED);
2816 }
2817
5aea1f0e
TM		 2818 ret = ocfs2_journal_dirty(handle, new_bh);
2819 if (ret < 0) {
2820 mlog_errno(ret);
2821 goto end;
2822 }
2823 di->i_xattr_loc = cpu_to_le64(first_blkno);
2824 ocfs2_journal_dirty(handle, inode_bh);
2825
2826 *ret_bh = new_bh;
2827 new_bh = NULL;
2828
2829end:
2830 brelse(new_bh);
2831 return ret;
2832}
2833
cf1d6c76
TY		 2834/*
2835 * ocfs2_xattr_block_set()
2836 *
2837 * Set, replace or remove an extended attribute into external block.
2838 *
2839 */
2840static int ocfs2_xattr_block_set(struct inode *inode,
2841 struct ocfs2_xattr_info *xi,
78f30c31
TM		 2842 struct ocfs2_xattr_search *xs,
2843 struct ocfs2_xattr_set_ctxt *ctxt)
cf1d6c76
TY		 2844{
2845 struct buffer_head *new_bh = NULL;
85db90e7 2846 handle_t *handle = ctxt->handle;
cf1d6c76 2847 struct ocfs2_xattr_block *xblk = NULL;
cf1d6c76
TY		 2848 int ret;
2849
2850 if (!xs->xattr_bh) {
5aea1f0e 2851 ret = ocfs2_create_xattr_block(handle, inode, xs->inode_bh,
a7fe7a3a 2852 ctxt->meta_ac, &new_bh, 0);
5aea1f0e 2853 if (ret) {
cf1d6c76 2854 mlog_errno(ret);
85db90e7 2855 goto end;
cf1d6c76
TY		 2856 }
2857
cf1d6c76 2858 xs->xattr_bh = new_bh;
5aea1f0e 2859 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
cf1d6c76
TY		 2860 xs->header = &xblk->xb_attrs.xb_header;
2861 xs->base = (void *)xs->header;
2862 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2863 xs->here = xs->header->xh_entries;
01225596
TM		 2864 } else
2865 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2866
2867 if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2868 /* Set extended attribute into external block */
78f30c31
TM		 2869 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2870 OCFS2_HAS_XATTR_FL);
01225596
TM		 2871 if (!ret || ret != -ENOSPC)
2872 goto end;
2873
78f30c31 2874 ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
01225596
TM		 2875 if (ret)
2876 goto end;
cf1d6c76
TY		 2877 }
2878
78f30c31 2879 ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
01225596
TM		 2880
2881end:
cf1d6c76
TY		 2882
2883 return ret;
2884}
2885
78f30c31
TM		 2886/* Check whether the new xattr can be inserted into the inode. */
2887static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2888 struct ocfs2_xattr_info *xi,
2889 struct ocfs2_xattr_search *xs)
2890{
78f30c31
TM		 2891 struct ocfs2_xattr_entry *last;
2892 int free, i;
2893 size_t min_offs = xs->end - xs->base;
2894
2895 if (!xs->header)
2896 return 0;
2897
2898 last = xs->header->xh_entries;
2899
2900 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
2901 size_t offs = le16_to_cpu(last->xe_name_offset);
2902 if (offs < min_offs)
2903 min_offs = offs;
2904 last += 1;
2905 }
2906
4442f518 2907 free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
78f30c31
TM		 2908 if (free < 0)
2909 return 0;
2910
2911 BUG_ON(!xs->not_found);
2912
199799a3 2913 if (free >= (sizeof(struct ocfs2_xattr_entry) + namevalue_size_xi(xi)))
78f30c31
TM		 2914 return 1;
2915
2916 return 0;
2917}
2918
2919static int ocfs2_calc_xattr_set_need(struct inode *inode,
2920 struct ocfs2_dinode *di,
2921 struct ocfs2_xattr_info *xi,
2922 struct ocfs2_xattr_search *xis,
2923 struct ocfs2_xattr_search *xbs,
2924 int *clusters_need,
85db90e7
TM		 2925 int *meta_need,
2926 int *credits_need)
78f30c31
TM		 2927{
2928 int ret = 0, old_in_xb = 0;
85db90e7 2929 int clusters_add = 0, meta_add = 0, credits = 0;
78f30c31
TM		 2930 struct buffer_head *bh = NULL;
2931 struct ocfs2_xattr_block *xb = NULL;
2932 struct ocfs2_xattr_entry *xe = NULL;
2933 struct ocfs2_xattr_value_root *xv = NULL;
2934 char *base = NULL;
2935 int name_offset, name_len = 0;
2936 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
6b240ff6 2937 xi->xi_value_len);
78f30c31
TM		 2938 u64 value_size;
2939
71d548a6
TM		 2940 /*
2941 * Calculate the clusters we need to write.
2942 * No matter whether we replace an old one or add a new one,
2943 * we need this for writing.
2944 */
6b240ff6 2945 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
71d548a6
TM		 2946 credits += new_clusters *
2947 ocfs2_clusters_to_blocks(inode->i_sb, 1);
2948
78f30c31 2949 if (xis->not_found && xbs->not_found) {
85db90e7
TM		 2950 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2951
6b240ff6 2952 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
78f30c31 2953 clusters_add += new_clusters;
85db90e7
TM		 2954 credits += ocfs2_calc_extend_credits(inode->i_sb,
2955 &def_xv.xv.xr_list,
2956 new_clusters);
2957 }
78f30c31
TM		 2958
2959 goto meta_guess;
2960 }
2961
2962 if (!xis->not_found) {
2963 xe = xis->here;
2964 name_offset = le16_to_cpu(xe->xe_name_offset);
2965 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2966 base = xis->base;
85db90e7 2967 credits += OCFS2_INODE_UPDATE_CREDITS;
78f30c31 2968 } else {
970e4936 2969 int i, block_off = 0;
78f30c31
TM		 2970 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2971 xe = xbs->here;
2972 name_offset = le16_to_cpu(xe->xe_name_offset);
2973 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2974 i = xbs->here - xbs->header->xh_entries;
2975 old_in_xb = 1;
2976
2977 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
fd68a894 2978 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
78f30c31
TM		 2979 bucket_xh(xbs->bucket),
2980 i, &block_off,
2981 &name_offset);
2982 base = bucket_block(xbs->bucket, block_off);
85db90e7
TM		 2983 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2984 } else {
78f30c31 2985 base = xbs->base;
85db90e7
TM		 2986 credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
2987 }
2988 }
2989
2990 /*
2991 * delete a xattr doesn't need metadata and cluster allocation.
2992 * so just calculate the credits and return.
2993 *
2994 * The credits for removing the value tree will be extended
2995 * by ocfs2_remove_extent itself.
2996 */
6b240ff6 2997 if (!xi->xi_value) {
85db90e7 2998 if (!ocfs2_xattr_is_local(xe))
a90714c1 2999 credits += ocfs2_remove_extent_credits(inode->i_sb);
85db90e7
TM		 3000
3001 goto out;
78f30c31
TM		 3002 }
3003
3004 /* do cluster allocation guess first. */
3005 value_size = le64_to_cpu(xe->xe_value_size);
3006
3007 if (old_in_xb) {
3008 /*
3009 * In xattr set, we always try to set the xe in inode first,
3010 * so if it can be inserted into inode successfully, the old
3011 * one will be removed from the xattr block, and this xattr
3012 * will be inserted into inode as a new xattr in inode.
3013 */
3014 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
3015 clusters_add += new_clusters;
a90714c1 3016 credits += ocfs2_remove_extent_credits(inode->i_sb) +
85db90e7
TM		 3017 OCFS2_INODE_UPDATE_CREDITS;
3018 if (!ocfs2_xattr_is_local(xe))
3019 credits += ocfs2_calc_extend_credits(
3020 inode->i_sb,
3021 &def_xv.xv.xr_list,
3022 new_clusters);
78f30c31
TM		 3023 goto out;
3024 }
3025 }
3026
6b240ff6 3027 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
78f30c31
TM		 3028 /* the new values will be stored outside. */
3029 u32 old_clusters = 0;
3030
3031 if (!ocfs2_xattr_is_local(xe)) {
3032 old_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
3033 value_size);
3034 xv = (struct ocfs2_xattr_value_root *)
3035 (base + name_offset + name_len);
97aff52a 3036 value_size = OCFS2_XATTR_ROOT_SIZE;
78f30c31
TM		 3037 } else
3038 xv = &def_xv.xv;
3039
85db90e7 3040 if (old_clusters >= new_clusters) {
a90714c1 3041 credits += ocfs2_remove_extent_credits(inode->i_sb);
78f30c31 3042 goto out;
85db90e7 3043 } else {
78f30c31
TM		 3044 meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
3045 clusters_add += new_clusters - old_clusters;
85db90e7
TM		 3046 credits += ocfs2_calc_extend_credits(inode->i_sb,
3047 &xv->xr_list,
3048 new_clusters -
3049 old_clusters);
97aff52a
TM		 3050 if (value_size >= OCFS2_XATTR_ROOT_SIZE)
3051 goto out;
78f30c31
TM		 3052 }
3053 } else {
3054 /*
3055 * Now the new value will be stored inside. So if the new
3056 * value is smaller than the size of value root or the old
3057 * value, we don't need any allocation, otherwise we have
3058 * to guess metadata allocation.
3059 */
6b240ff6
JB		 3060 if ((ocfs2_xattr_is_local(xe) &&
3061 (value_size >= xi->xi_value_len)) ||
78f30c31 3062 (!ocfs2_xattr_is_local(xe) &&
6b240ff6 3063 OCFS2_XATTR_ROOT_SIZE >= xi->xi_value_len))
78f30c31
TM		 3064 goto out;
3065 }
3066
3067meta_guess:
3068 /* calculate metadata allocation. */
3069 if (di->i_xattr_loc) {
3070 if (!xbs->xattr_bh) {
4ae1d69b
JB		 3071 ret = ocfs2_read_xattr_block(inode,
3072 le64_to_cpu(di->i_xattr_loc),
3073 &bh);
78f30c31
TM		 3074 if (ret) {
3075 mlog_errno(ret);
3076 goto out;
3077 }
3078
3079 xb = (struct ocfs2_xattr_block *)bh->b_data;
3080 } else
3081 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3082
90cb546c
TM		 3083 /*
3084 * If there is already an xattr tree, good, we can calculate
3085 * like other b-trees. Otherwise we may have the chance of
3086 * create a tree, the credit calculation is borrowed from
3087 * ocfs2_calc_extend_credits with root_el = NULL. And the
3088 * new tree will be cluster based, so no meta is needed.
3089 */
78f30c31
TM		 3090 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3091 struct ocfs2_extent_list *el =
3092 &xb->xb_attrs.xb_root.xt_list;
3093 meta_add += ocfs2_extend_meta_needed(el);
85db90e7
TM		 3094 credits += ocfs2_calc_extend_credits(inode->i_sb,
3095 el, 1);
90cb546c
TM		 3096 } else
3097 credits += OCFS2_SUBALLOC_ALLOC + 1;
78f30c31
TM		 3098
3099 /*
3100 * This cluster will be used either for new bucket or for
3101 * new xattr block.
3102 * If the cluster size is the same as the bucket size, one
3103 * more is needed since we may need to extend the bucket
3104 * also.
3105 */
3106 clusters_add += 1;
85db90e7 3107 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
78f30c31 3108 if (OCFS2_XATTR_BUCKET_SIZE ==
85db90e7
TM		 3109 OCFS2_SB(inode->i_sb)->s_clustersize) {
3110 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
78f30c31 3111 clusters_add += 1;
85db90e7
TM		 3112 }
3113 } else {
78f30c31 3114 meta_add += 1;
85db90e7
TM		 3115 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
3116 }
78f30c31
TM		 3117out:
3118 if (clusters_need)
3119 *clusters_need = clusters_add;
3120 if (meta_need)
3121 *meta_need = meta_add;
85db90e7
TM		 3122 if (credits_need)
3123 *credits_need = credits;
78f30c31
TM		 3124 brelse(bh);
3125 return ret;
3126}
3127
3128static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
3129 struct ocfs2_dinode *di,
3130 struct ocfs2_xattr_info *xi,
3131 struct ocfs2_xattr_search *xis,
3132 struct ocfs2_xattr_search *xbs,
85db90e7 3133 struct ocfs2_xattr_set_ctxt *ctxt,
492a8a33 3134 int extra_meta,
85db90e7 3135 int *credits)
78f30c31
TM		 3136{
3137 int clusters_add, meta_add, ret;
3138 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3139
3140 memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
3141
3142 ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
3143
3144 ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
85db90e7 3145 &clusters_add, &meta_add, credits);
78f30c31
TM		 3146 if (ret) {
3147 mlog_errno(ret);
3148 return ret;
3149 }
3150
492a8a33 3151 meta_add += extra_meta;
85db90e7 3152 mlog(0, "Set xattr %s, reserve meta blocks = %d, clusters = %d, "
6b240ff6 3153 "credits = %d\n", xi->xi_name, meta_add, clusters_add, *credits);
78f30c31
TM		 3154
3155 if (meta_add) {
3156 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
3157 &ctxt->meta_ac);
3158 if (ret) {
3159 mlog_errno(ret);
3160 goto out;
3161 }
3162 }
3163
3164 if (clusters_add) {
3165 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
3166 if (ret)
3167 mlog_errno(ret);
3168 }
3169out:
3170 if (ret) {
3171 if (ctxt->meta_ac) {
3172 ocfs2_free_alloc_context(ctxt->meta_ac);
3173 ctxt->meta_ac = NULL;
3174 }
3175
3176 /*
3177 * We cannot have an error and a non null ctxt->data_ac.
3178 */
3179 }
3180
3181 return ret;
3182}
3183
85db90e7
TM		 3184static int __ocfs2_xattr_set_handle(struct inode *inode,
3185 struct ocfs2_dinode *di,
3186 struct ocfs2_xattr_info *xi,
3187 struct ocfs2_xattr_search *xis,
3188 struct ocfs2_xattr_search *xbs,
3189 struct ocfs2_xattr_set_ctxt *ctxt)
3190{
9f868f16 3191 int ret = 0, credits, old_found;
85db90e7 3192
6b240ff6 3193 if (!xi->xi_value) {
85db90e7
TM		 3194 /* Remove existing extended attribute */
3195 if (!xis->not_found)
3196 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3197 else if (!xbs->not_found)
3198 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3199 } else {
3200 /* We always try to set extended attribute into inode first*/
3201 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3202 if (!ret && !xbs->not_found) {
3203 /*
3204 * If succeed and that extended attribute existing in
3205 * external block, then we will remove it.
3206 */
6b240ff6
JB		 3207 xi->xi_value = NULL;
3208 xi->xi_value_len = 0;
85db90e7 3209
9f868f16 3210 old_found = xis->not_found;
85db90e7
TM		 3211 xis->not_found = -ENODATA;
3212 ret = ocfs2_calc_xattr_set_need(inode,
3213 di,
3214 xi,
3215 xis,
3216 xbs,
3217 NULL,
3218 NULL,
3219 &credits);
9f868f16 3220 xis->not_found = old_found;
85db90e7
TM		 3221 if (ret) {
3222 mlog_errno(ret);
3223 goto out;
3224 }
3225
3226 ret = ocfs2_extend_trans(ctxt->handle, credits +
3227 ctxt->handle->h_buffer_credits);
3228 if (ret) {
3229 mlog_errno(ret);
3230 goto out;
3231 }
3232 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3233 } else if (ret == -ENOSPC) {
3234 if (di->i_xattr_loc && !xbs->xattr_bh) {
3235 ret = ocfs2_xattr_block_find(inode,
6b240ff6
JB		 3236 xi->xi_name_index,
3237 xi->xi_name, xbs);
85db90e7
TM		 3238 if (ret)
3239 goto out;
3240
9f868f16 3241 old_found = xis->not_found;
85db90e7
TM		 3242 xis->not_found = -ENODATA;
3243 ret = ocfs2_calc_xattr_set_need(inode,
3244 di,
3245 xi,
3246 xis,
3247 xbs,
3248 NULL,
3249 NULL,
3250 &credits);
9f868f16 3251 xis->not_found = old_found;
85db90e7
TM		 3252 if (ret) {
3253 mlog_errno(ret);
3254 goto out;
3255 }
3256
3257 ret = ocfs2_extend_trans(ctxt->handle, credits +
3258 ctxt->handle->h_buffer_credits);
3259 if (ret) {
3260 mlog_errno(ret);
3261 goto out;
3262 }
3263 }
3264 /*
3265 * If no space in inode, we will set extended attribute
3266 * into external block.
3267 */
3268 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3269 if (ret)
3270 goto out;
3271 if (!xis->not_found) {
3272 /*
3273 * If succeed and that extended attribute
3274 * existing in inode, we will remove it.
3275 */
6b240ff6
JB		 3276 xi->xi_value = NULL;
3277 xi->xi_value_len = 0;
85db90e7
TM		 3278 xbs->not_found = -ENODATA;
3279 ret = ocfs2_calc_xattr_set_need(inode,
3280 di,
3281 xi,
3282 xis,
3283 xbs,
3284 NULL,
3285 NULL,
3286 &credits);
3287 if (ret) {
3288 mlog_errno(ret);
3289 goto out;
3290 }
3291
3292 ret = ocfs2_extend_trans(ctxt->handle, credits +
3293 ctxt->handle->h_buffer_credits);
3294 if (ret) {
3295 mlog_errno(ret);
3296 goto out;
3297 }
3298 ret = ocfs2_xattr_ibody_set(inode, xi,
3299 xis, ctxt);
3300 }
3301 }
3302 }
3303
4b3f6209
TM		 3304 if (!ret) {
3305 /* Update inode ctime. */
0cf2f763 3306 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
89a907af
TM		 3307 xis->inode_bh,
3308 OCFS2_JOURNAL_ACCESS_WRITE);
4b3f6209
TM		 3309 if (ret) {
3310 mlog_errno(ret);
3311 goto out;
3312 }
3313
3314 inode->i_ctime = CURRENT_TIME;
3315 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
3316 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
3317 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
3318 }
85db90e7
TM		 3319out:
3320 return ret;
3321}
3322
6c3faba4
TY		 3323/*
3324 * This function only called duing creating inode
3325 * for init security/acl xattrs of the new inode.
008aafaf 3326 * All transanction credits have been reserved in mknod.
6c3faba4
TY		 3327 */
3328int ocfs2_xattr_set_handle(handle_t *handle,
3329 struct inode *inode,
3330 struct buffer_head *di_bh,
3331 int name_index,
3332 const char *name,
3333 const void *value,
3334 size_t value_len,
3335 int flags,
3336 struct ocfs2_alloc_context *meta_ac,
3337 struct ocfs2_alloc_context *data_ac)
3338{
3339 struct ocfs2_dinode *di;
3340 int ret;
3341
3342 struct ocfs2_xattr_info xi = {
6b240ff6
JB		 3343 .xi_name_index = name_index,
3344 .xi_name = name,
18853b95 3345 .xi_name_len = strlen(name),
6b240ff6
JB		 3346 .xi_value = value,
3347 .xi_value_len = value_len,
6c3faba4
TY		 3348 };
3349
3350 struct ocfs2_xattr_search xis = {
3351 .not_found = -ENODATA,
3352 };
3353
3354 struct ocfs2_xattr_search xbs = {
3355 .not_found = -ENODATA,
3356 };
3357
3358 struct ocfs2_xattr_set_ctxt ctxt = {
3359 .handle = handle,
3360 .meta_ac = meta_ac,
3361 .data_ac = data_ac,
3362 };
3363
3364 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3365 return -EOPNOTSUPP;
3366
008aafaf
TY		 3367 /*
3368 * In extreme situation, may need xattr bucket when
3369 * block size is too small. And we have already reserved
3370 * the credits for bucket in mknod.
3371 */
3372 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
3373 xbs.bucket = ocfs2_xattr_bucket_new(inode);
3374 if (!xbs.bucket) {
3375 mlog_errno(-ENOMEM);
3376 return -ENOMEM;
3377 }
3378 }
3379
6c3faba4
TY		 3380 xis.inode_bh = xbs.inode_bh = di_bh;
3381 di = (struct ocfs2_dinode *)di_bh->b_data;
3382
3383 down_write(&OCFS2_I(inode)->ip_xattr_sem);
3384
3385 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3386 if (ret)
3387 goto cleanup;
3388 if (xis.not_found) {
3389 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3390 if (ret)
3391 goto cleanup;
3392 }
3393
3394 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3395
3396cleanup:
3397 up_write(&OCFS2_I(inode)->ip_xattr_sem);
3398 brelse(xbs.xattr_bh);
008aafaf 3399 ocfs2_xattr_bucket_free(xbs.bucket);
6c3faba4
TY		 3400
3401 return ret;
3402}
3403
cf1d6c76
TY		 3404/*
3405 * ocfs2_xattr_set()
3406 *
3407 * Set, replace or remove an extended attribute for this inode.
3408 * value is NULL to remove an existing extended attribute, else either
3409 * create or replace an extended attribute.
3410 */
3411int ocfs2_xattr_set(struct inode *inode,
3412 int name_index,
3413 const char *name,
3414 const void *value,
3415 size_t value_len,
3416 int flags)
3417{
3418 struct buffer_head *di_bh = NULL;
3419 struct ocfs2_dinode *di;
492a8a33 3420 int ret, credits, ref_meta = 0, ref_credits = 0;
78f30c31 3421 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
85db90e7 3422 struct inode *tl_inode = osb->osb_tl_inode;
78f30c31 3423 struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
492a8a33 3424 struct ocfs2_refcount_tree *ref_tree = NULL;
cf1d6c76
TY		 3425
3426 struct ocfs2_xattr_info xi = {
6b240ff6
JB		 3427 .xi_name_index = name_index,
3428 .xi_name = name,
18853b95 3429 .xi_name_len = strlen(name),
6b240ff6
JB		 3430 .xi_value = value,
3431 .xi_value_len = value_len,
cf1d6c76
TY		 3432 };
3433
3434 struct ocfs2_xattr_search xis = {
3435 .not_found = -ENODATA,
3436 };
3437
3438 struct ocfs2_xattr_search xbs = {
3439 .not_found = -ENODATA,
3440 };
3441
8154da3d
TY		 3442 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3443 return -EOPNOTSUPP;
3444
ba937127
JB		 3445 /*
3446 * Only xbs will be used on indexed trees. xis doesn't need a
3447 * bucket.
3448 */
3449 xbs.bucket = ocfs2_xattr_bucket_new(inode);
3450 if (!xbs.bucket) {
3451 mlog_errno(-ENOMEM);
3452 return -ENOMEM;
3453 }
3454
cf1d6c76
TY		 3455 ret = ocfs2_inode_lock(inode, &di_bh, 1);
3456 if (ret < 0) {
3457 mlog_errno(ret);
ba937127 3458 goto cleanup_nolock;
cf1d6c76
TY		 3459 }
3460 xis.inode_bh = xbs.inode_bh = di_bh;
3461 di = (struct ocfs2_dinode *)di_bh->b_data;
3462
3463 down_write(&OCFS2_I(inode)->ip_xattr_sem);
3464 /*
3465 * Scan inode and external block to find the same name
3466 * extended attribute and collect search infomation.
3467 */
3468 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3469 if (ret)
3470 goto cleanup;
3471 if (xis.not_found) {
3472 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3473 if (ret)
3474 goto cleanup;
3475 }
3476
3477 if (xis.not_found && xbs.not_found) {
3478 ret = -ENODATA;
3479 if (flags & XATTR_REPLACE)
3480 goto cleanup;
3481 ret = 0;
3482 if (!value)
3483 goto cleanup;
3484 } else {
3485 ret = -EEXIST;
3486 if (flags & XATTR_CREATE)
3487 goto cleanup;
3488 }
3489
492a8a33
TM		 3490 /* Check whether the value is refcounted and do some prepartion. */
3491 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL &&
3492 (!xis.not_found || !xbs.not_found)) {
3493 ret = ocfs2_prepare_refcount_xattr(inode, di, &xi,
3494 &xis, &xbs, &ref_tree,
3495 &ref_meta, &ref_credits);
3496 if (ret) {
3497 mlog_errno(ret);
3498 goto cleanup;
3499 }
3500 }
85db90e7
TM		 3501
3502 mutex_lock(&tl_inode->i_mutex);
3503
3504 if (ocfs2_truncate_log_needs_flush(osb)) {
3505 ret = __ocfs2_flush_truncate_log(osb);
3506 if (ret < 0) {
3507 mutex_unlock(&tl_inode->i_mutex);
3508 mlog_errno(ret);
3509 goto cleanup;
3510 }
3511 }
3512 mutex_unlock(&tl_inode->i_mutex);
3513
3514 ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
492a8a33 3515 &xbs, &ctxt, ref_meta, &credits);
78f30c31
TM		 3516 if (ret) {
3517 mlog_errno(ret);
3518 goto cleanup;
3519 }
3520
4b3f6209
TM		 3521 /* we need to update inode's ctime field, so add credit for it. */
3522 credits += OCFS2_INODE_UPDATE_CREDITS;
492a8a33 3523 ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
85db90e7
TM		 3524 if (IS_ERR(ctxt.handle)) {
3525 ret = PTR_ERR(ctxt.handle);
3526 mlog_errno(ret);
3527 goto cleanup;
cf1d6c76 3528 }
85db90e7
TM		 3529
3530 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3531
3532 ocfs2_commit_trans(osb, ctxt.handle);
3533
78f30c31
TM		 3534 if (ctxt.data_ac)
3535 ocfs2_free_alloc_context(ctxt.data_ac);
3536 if (ctxt.meta_ac)
3537 ocfs2_free_alloc_context(ctxt.meta_ac);
3538 if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
3539 ocfs2_schedule_truncate_log_flush(osb, 1);
3540 ocfs2_run_deallocs(osb, &ctxt.dealloc);
8b2c0dba 3541
cf1d6c76 3542cleanup:
492a8a33
TM		 3543 if (ref_tree)
3544 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
cf1d6c76 3545 up_write(&OCFS2_I(inode)->ip_xattr_sem);
8b2c0dba
TM		 3546 if (!value && !ret) {
3547 ret = ocfs2_try_remove_refcount_tree(inode, di_bh);
3548 if (ret)
3549 mlog_errno(ret);
3550 }
cf1d6c76 3551 ocfs2_inode_unlock(inode, 1);
ba937127 3552cleanup_nolock:
cf1d6c76
TY		 3553 brelse(di_bh);
3554 brelse(xbs.xattr_bh);
ba937127 3555 ocfs2_xattr_bucket_free(xbs.bucket);
cf1d6c76
TY		 3556
3557 return ret;
3558}
3559
0c044f0b
TM		 3560/*
3561 * Find the xattr extent rec which may contains name_hash.
3562 * e_cpos will be the first name hash of the xattr rec.
3563 * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
3564 */
3565static int ocfs2_xattr_get_rec(struct inode *inode,
3566 u32 name_hash,
3567 u64 *p_blkno,
3568 u32 *e_cpos,
3569 u32 *num_clusters,
3570 struct ocfs2_extent_list *el)
3571{
3572 int ret = 0, i;
3573 struct buffer_head *eb_bh = NULL;
3574 struct ocfs2_extent_block *eb;
3575 struct ocfs2_extent_rec *rec = NULL;
3576 u64 e_blkno = 0;
3577
3578 if (el->l_tree_depth) {
facdb77f
JB		 3579 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, name_hash,
3580 &eb_bh);
0c044f0b
TM		 3581 if (ret) {
3582 mlog_errno(ret);
3583 goto out;
3584 }
3585
3586 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
3587 el = &eb->h_list;
3588
3589 if (el->l_tree_depth) {
3590 ocfs2_error(inode->i_sb,
3591 "Inode %lu has non zero tree depth in "
3592 "xattr tree block %llu\n", inode->i_ino,
3593 (unsigned long long)eb_bh->b_blocknr);
3594 ret = -EROFS;
3595 goto out;
3596 }
3597 }
3598
3599 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
3600 rec = &el->l_recs[i];
3601
3602 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
3603 e_blkno = le64_to_cpu(rec->e_blkno);
3604 break;
3605 }
3606 }
3607
3608 if (!e_blkno) {
3609 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
3610 "record (%u, %u, 0) in xattr", inode->i_ino,
3611 le32_to_cpu(rec->e_cpos),
3612 ocfs2_rec_clusters(el, rec));
3613 ret = -EROFS;
3614 goto out;
3615 }
3616
3617 *p_blkno = le64_to_cpu(rec->e_blkno);
3618 *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
3619 if (e_cpos)
3620 *e_cpos = le32_to_cpu(rec->e_cpos);
3621out:
3622 brelse(eb_bh);
3623 return ret;
3624}
3625
3626typedef int (xattr_bucket_func)(struct inode *inode,
3627 struct ocfs2_xattr_bucket *bucket,
3628 void *para);
3629
589dc260 3630static int ocfs2_find_xe_in_bucket(struct inode *inode,
e2356a3f 3631 struct ocfs2_xattr_bucket *bucket,
589dc260
TM		 3632 int name_index,
3633 const char *name,
3634 u32 name_hash,
3635 u16 *xe_index,
3636 int *found)
3637{
3638 int i, ret = 0, cmp = 1, block_off, new_offset;
e2356a3f 3639 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
589dc260
TM		 3640 size_t name_len = strlen(name);
3641 struct ocfs2_xattr_entry *xe = NULL;
589dc260
TM		 3642 char *xe_name;
3643
3644 /*
3645 * We don't use binary search in the bucket because there
3646 * may be multiple entries with the same name hash.
3647 */
3648 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3649 xe = &xh->xh_entries[i];
3650
3651 if (name_hash > le32_to_cpu(xe->xe_name_hash))
3652 continue;
3653 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
3654 break;
3655
3656 cmp = name_index - ocfs2_xattr_get_type(xe);
3657 if (!cmp)
3658 cmp = name_len - xe->xe_name_len;
3659 if (cmp)
3660 continue;
3661
fd68a894 3662 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
589dc260
TM		 3663 xh,
3664 i,
3665 &block_off,
3666 &new_offset);
3667 if (ret) {
3668 mlog_errno(ret);
3669 break;
3670 }
3671
970e4936 3672
e2356a3f
JB		 3673 xe_name = bucket_block(bucket, block_off) + new_offset;
3674 if (!memcmp(name, xe_name, name_len)) {
589dc260
TM		 3675 *xe_index = i;
3676 *found = 1;
3677 ret = 0;
3678 break;
3679 }
3680 }
3681
3682 return ret;
3683}
3684
3685/*
3686 * Find the specified xattr entry in a series of buckets.
3687 * This series start from p_blkno and last for num_clusters.
3688 * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
3689 * the num of the valid buckets.
3690 *
3691 * Return the buffer_head this xattr should reside in. And if the xattr's
3692 * hash is in the gap of 2 buckets, return the lower bucket.
3693 */
3694static int ocfs2_xattr_bucket_find(struct inode *inode,
3695 int name_index,
3696 const char *name,
3697 u32 name_hash,
3698 u64 p_blkno,
3699 u32 first_hash,
3700 u32 num_clusters,
3701 struct ocfs2_xattr_search *xs)
3702{
3703 int ret, found = 0;
589dc260
TM		 3704 struct ocfs2_xattr_header *xh = NULL;
3705 struct ocfs2_xattr_entry *xe = NULL;
3706 u16 index = 0;
3707 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3708 int low_bucket = 0, bucket, high_bucket;
e2356a3f 3709 struct ocfs2_xattr_bucket *search;
589dc260 3710 u32 last_hash;
e2356a3f 3711 u64 blkno, lower_blkno = 0;
589dc260 3712
e2356a3f
JB		 3713 search = ocfs2_xattr_bucket_new(inode);
3714 if (!search) {
3715 ret = -ENOMEM;
3716 mlog_errno(ret);
3717 goto out;
3718 }
3719
3720 ret = ocfs2_read_xattr_bucket(search, p_blkno);
589dc260
TM		 3721 if (ret) {
3722 mlog_errno(ret);
3723 goto out;
3724 }
3725
e2356a3f 3726 xh = bucket_xh(search);
589dc260 3727 high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
589dc260 3728 while (low_bucket <= high_bucket) {
e2356a3f 3729 ocfs2_xattr_bucket_relse(search);
589dc260 3730
e2356a3f 3731 bucket = (low_bucket + high_bucket) / 2;
589dc260 3732 blkno = p_blkno + bucket * blk_per_bucket;
e2356a3f 3733 ret = ocfs2_read_xattr_bucket(search, blkno);
589dc260
TM		 3734 if (ret) {
3735 mlog_errno(ret);
3736 goto out;
3737 }
3738
e2356a3f 3739 xh = bucket_xh(search);
589dc260
TM		 3740 xe = &xh->xh_entries[0];
3741 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3742 high_bucket = bucket - 1;
3743 continue;
3744 }
3745
3746 /*
3747 * Check whether the hash of the last entry in our
5a095611
TM		 3748 * bucket is larger than the search one. for an empty
3749 * bucket, the last one is also the first one.
589dc260 3750 */
5a095611
TM		 3751 if (xh->xh_count)
3752 xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3753
589dc260
TM		 3754 last_hash = le32_to_cpu(xe->xe_name_hash);
3755
e2356a3f
JB		 3756 /* record lower_blkno which may be the insert place. */
3757 lower_blkno = blkno;
589dc260
TM		 3758
3759 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3760 low_bucket = bucket + 1;
3761 continue;
3762 }
3763
3764 /* the searched xattr should reside in this bucket if exists. */
e2356a3f 3765 ret = ocfs2_find_xe_in_bucket(inode, search,
589dc260
TM		 3766 name_index, name, name_hash,
3767 &index, &found);
3768 if (ret) {
3769 mlog_errno(ret);
3770 goto out;
3771 }
3772 break;
3773 }
3774
3775 /*
3776 * Record the bucket we have found.
3777 * When the xattr's hash value is in the gap of 2 buckets, we will
3778 * always set it to the previous bucket.
3779 */
e2356a3f
JB		 3780 if (!lower_blkno)
3781 lower_blkno = p_blkno;
3782
3783 /* This should be in cache - we just read it during the search */
3784 ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3785 if (ret) {
3786 mlog_errno(ret);
3787 goto out;
589dc260 3788 }
589dc260 3789
ba937127
JB		 3790 xs->header = bucket_xh(xs->bucket);
3791 xs->base = bucket_block(xs->bucket, 0);
589dc260
TM		 3792 xs->end = xs->base + inode->i_sb->s_blocksize;
3793
3794 if (found) {
589dc260
TM		 3795 xs->here = &xs->header->xh_entries[index];
3796 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
ba937127 3797 (unsigned long long)bucket_blkno(xs->bucket), index);
589dc260
TM		 3798 } else
3799 ret = -ENODATA;
3800
3801out:
e2356a3f 3802 ocfs2_xattr_bucket_free(search);
589dc260
TM		 3803 return ret;
3804}
3805
3806static int ocfs2_xattr_index_block_find(struct inode *inode,
3807 struct buffer_head *root_bh,
3808 int name_index,
3809 const char *name,
3810 struct ocfs2_xattr_search *xs)
3811{
3812 int ret;
3813 struct ocfs2_xattr_block *xb =
3814 (struct ocfs2_xattr_block *)root_bh->b_data;
3815 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3816 struct ocfs2_extent_list *el = &xb_root->xt_list;
3817 u64 p_blkno = 0;
3818 u32 first_hash, num_clusters = 0;
2057e5c6 3819 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
589dc260
TM		 3820
3821 if (le16_to_cpu(el->l_next_free_rec) == 0)
3822 return -ENODATA;
3823
3824 mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
3825 name, name_hash, name_index);
3826
3827 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3828 &num_clusters, el);
3829 if (ret) {
3830 mlog_errno(ret);
3831 goto out;
3832 }
3833
3834 BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3835
3836 mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
de29c085
MF		 3837 "in the rec is %u\n", num_clusters, (unsigned long long)p_blkno,
3838 first_hash);
589dc260
TM		 3839
3840 ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3841 p_blkno, first_hash, num_clusters, xs);
3842
3843out:
3844 return ret;
3845}
3846
0c044f0b
TM		 3847static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3848 u64 blkno,
3849 u32 clusters,
3850 xattr_bucket_func *func,
3851 void *para)
3852{
6dde41d9 3853 int i, ret = 0;
0c044f0b
TM		 3854 u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3855 u32 num_buckets = clusters * bpc;
ba937127 3856 struct ocfs2_xattr_bucket *bucket;
0c044f0b 3857
ba937127
JB		 3858 bucket = ocfs2_xattr_bucket_new(inode);
3859 if (!bucket) {
3860 mlog_errno(-ENOMEM);
3861 return -ENOMEM;
3862 }
0c044f0b
TM		 3863
3864 mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
de29c085 3865 clusters, (unsigned long long)blkno);
0c044f0b 3866
ba937127
JB		 3867 for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3868 ret = ocfs2_read_xattr_bucket(bucket, blkno);
0c044f0b
TM		 3869 if (ret) {
3870 mlog_errno(ret);
ba937127 3871 break;
0c044f0b
TM		 3872 }
3873
0c044f0b
TM		 3874 /*
3875 * The real bucket num in this series of blocks is stored
3876 * in the 1st bucket.
3877 */
3878 if (i == 0)
ba937127 3879 num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
0c044f0b 3880
de29c085
MF		 3881 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
3882 (unsigned long long)blkno,
ba937127 3883 le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
0c044f0b 3884 if (func) {
ba937127 3885 ret = func(inode, bucket, para);
a46fa684 3886 if (ret && ret != -ERANGE)
0c044f0b 3887 mlog_errno(ret);
ba937127 3888 /* Fall through to bucket_relse() */
0c044f0b
TM		 3889 }
3890
ba937127
JB		 3891 ocfs2_xattr_bucket_relse(bucket);
3892 if (ret)
3893 break;
0c044f0b
TM		 3894 }
3895
ba937127 3896 ocfs2_xattr_bucket_free(bucket);
0c044f0b
TM		 3897 return ret;
3898}
3899
3900struct ocfs2_xattr_tree_list {
3901 char *buffer;
3902 size_t buffer_size;
936b8834 3903 size_t result;
0c044f0b
TM		 3904};
3905
fd68a894 3906static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
0c044f0b
TM		 3907 struct ocfs2_xattr_header *xh,
3908 int index,
3909 int *block_off,
3910 int *new_offset)
3911{
3912 u16 name_offset;
3913
3914 if (index < 0 || index >= le16_to_cpu(xh->xh_count))
3915 return -EINVAL;
3916
3917 name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
3918
fd68a894
TM		 3919 *block_off = name_offset >> sb->s_blocksize_bits;
3920 *new_offset = name_offset % sb->s_blocksize;
0c044f0b
TM		 3921
3922 return 0;
3923}
3924
3925static int ocfs2_list_xattr_bucket(struct inode *inode,
3926 struct ocfs2_xattr_bucket *bucket,
3927 void *para)
3928{
936b8834 3929 int ret = 0, type;
0c044f0b 3930 struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
0c044f0b 3931 int i, block_off, new_offset;
936b8834 3932 const char *prefix, *name;
0c044f0b 3933
3e632946
JB		 3934 for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
3935 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
936b8834
TM		 3936 type = ocfs2_xattr_get_type(entry);
3937 prefix = ocfs2_xattr_prefix(type);
0c044f0b 3938
936b8834 3939 if (prefix) {
fd68a894 3940 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3e632946 3941 bucket_xh(bucket),
0c044f0b
TM		 3942 i,
3943 &block_off,
3944 &new_offset);
3945 if (ret)
3946 break;
936b8834 3947
51def39f 3948 name = (const char *)bucket_block(bucket, block_off) +
936b8834
TM		 3949 new_offset;
3950 ret = ocfs2_xattr_list_entry(xl->buffer,
3951 xl->buffer_size,
3952 &xl->result,
3953 prefix, name,
3954 entry->xe_name_len);
3955 if (ret)
3956 break;
0c044f0b
TM		 3957 }
3958 }
3959
3960 return ret;
3961}
3962
47bca495
TM		 3963static int ocfs2_iterate_xattr_index_block(struct inode *inode,
3964 struct buffer_head *blk_bh,
3965 xattr_tree_rec_func *rec_func,
3966 void *para)
0c044f0b 3967{
47bca495
TM		 3968 struct ocfs2_xattr_block *xb =
3969 (struct ocfs2_xattr_block *)blk_bh->b_data;
3970 struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
0c044f0b
TM		 3971 int ret = 0;
3972 u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
3973 u64 p_blkno = 0;
0c044f0b 3974
47bca495 3975 if (!el->l_next_free_rec || !rec_func)
0c044f0b
TM		 3976 return 0;
3977
3978 while (name_hash > 0) {
3979 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
3980 &e_cpos, &num_clusters, el);
3981 if (ret) {
3982 mlog_errno(ret);
47bca495 3983 break;
0c044f0b
TM		 3984 }
3985
47bca495
TM		 3986 ret = rec_func(inode, blk_bh, p_blkno, e_cpos,
3987 num_clusters, para);
0c044f0b 3988 if (ret) {
a46fa684
TM		 3989 if (ret != -ERANGE)
3990 mlog_errno(ret);
47bca495 3991 break;
0c044f0b
TM		 3992 }
3993
3994 if (e_cpos == 0)
3995 break;
3996
3997 name_hash = e_cpos - 1;
3998 }
3999
47bca495
TM		 4000 return ret;
4001
4002}
4003
4004static int ocfs2_list_xattr_tree_rec(struct inode *inode,
4005 struct buffer_head *root_bh,
4006 u64 blkno, u32 cpos, u32 len, void *para)
4007{
4008 return ocfs2_iterate_xattr_buckets(inode, blkno, len,
4009 ocfs2_list_xattr_bucket, para);
4010}
4011
4012static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
4013 struct buffer_head *blk_bh,
4014 char *buffer,
4015 size_t buffer_size)
4016{
4017 int ret;
4018 struct ocfs2_xattr_tree_list xl = {
4019 .buffer = buffer,
4020 .buffer_size = buffer_size,
4021 .result = 0,
4022 };
4023
4024 ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
4025 ocfs2_list_xattr_tree_rec, &xl);
4026 if (ret) {
4027 mlog_errno(ret);
4028 goto out;
4029 }
4030
936b8834 4031 ret = xl.result;
0c044f0b
TM		 4032out:
4033 return ret;
4034}
01225596
TM		 4035
4036static int cmp_xe(const void *a, const void *b)
4037{
4038 const struct ocfs2_xattr_entry *l = a, *r = b;
4039 u32 l_hash = le32_to_cpu(l->xe_name_hash);
4040 u32 r_hash = le32_to_cpu(r->xe_name_hash);
4041
4042 if (l_hash > r_hash)
4043 return 1;
4044 if (l_hash < r_hash)
4045 return -1;
4046 return 0;
4047}
4048
4049static void swap_xe(void *a, void *b, int size)
4050{
4051 struct ocfs2_xattr_entry *l = a, *r = b, tmp;
4052
4053 tmp = *l;
4054 memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
4055 memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
4056}
4057
4058/*
4059 * When the ocfs2_xattr_block is filled up, new bucket will be created
4060 * and all the xattr entries will be moved to the new bucket.
178eeac3
JB		 4061 * The header goes at the start of the bucket, and the names+values are
4062 * filled from the end. This is why *target starts as the last buffer.
01225596
TM		 4063 * Note: we need to sort the entries since they are not saved in order
4064 * in the ocfs2_xattr_block.
4065 */
4066static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
4067 struct buffer_head *xb_bh,
178eeac3 4068 struct ocfs2_xattr_bucket *bucket)
01225596
TM		 4069{
4070 int i, blocksize = inode->i_sb->s_blocksize;
178eeac3 4071 int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
01225596
TM		 4072 u16 offset, size, off_change;
4073 struct ocfs2_xattr_entry *xe;
4074 struct ocfs2_xattr_block *xb =
4075 (struct ocfs2_xattr_block *)xb_bh->b_data;
4076 struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
178eeac3 4077 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
01225596 4078 u16 count = le16_to_cpu(xb_xh->xh_count);
178eeac3
JB		 4079 char *src = xb_bh->b_data;
4080 char *target = bucket_block(bucket, blks - 1);
01225596
TM		 4081
4082 mlog(0, "cp xattr from block %llu to bucket %llu\n",
4083 (unsigned long long)xb_bh->b_blocknr,
178eeac3
JB		 4084 (unsigned long long)bucket_blkno(bucket));
4085
4086 for (i = 0; i < blks; i++)
4087 memset(bucket_block(bucket, i), 0, blocksize);
01225596 4088
01225596
TM		 4089 /*
4090 * Since the xe_name_offset is based on ocfs2_xattr_header,
4091 * there is a offset change corresponding to the change of
4092 * ocfs2_xattr_header's position.
4093 */
4094 off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4095 xe = &xb_xh->xh_entries[count - 1];
4096 offset = le16_to_cpu(xe->xe_name_offset) + off_change;
4097 size = blocksize - offset;
4098
4099 /* copy all the names and values. */
01225596
TM		 4100 memcpy(target + offset, src + offset, size);
4101
4102 /* Init new header now. */
4103 xh->xh_count = xb_xh->xh_count;
4104 xh->xh_num_buckets = cpu_to_le16(1);
4105 xh->xh_name_value_len = cpu_to_le16(size);
4106 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
4107
4108 /* copy all the entries. */
178eeac3 4109 target = bucket_block(bucket, 0);
01225596
TM		 4110 offset = offsetof(struct ocfs2_xattr_header, xh_entries);
4111 size = count * sizeof(struct ocfs2_xattr_entry);
4112 memcpy(target + offset, (char *)xb_xh + offset, size);
4113
4114 /* Change the xe offset for all the xe because of the move. */
4115 off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
4116 offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4117 for (i = 0; i < count; i++)
4118 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
4119
4120 mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
4121 offset, size, off_change);
4122
4123 sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
4124 cmp_xe, swap_xe);
4125}
4126
4127/*
4128 * After we move xattr from block to index btree, we have to
4129 * update ocfs2_xattr_search to the new xe and base.
4130 *
4131 * When the entry is in xattr block, xattr_bh indicates the storage place.
4132 * While if the entry is in index b-tree, "bucket" indicates the
4133 * real place of the xattr.
4134 */
178eeac3
JB		 4135static void ocfs2_xattr_update_xattr_search(struct inode *inode,
4136 struct ocfs2_xattr_search *xs,
4137 struct buffer_head *old_bh)
01225596 4138{
01225596
TM		 4139 char *buf = old_bh->b_data;
4140 struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
4141 struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
178eeac3 4142 int i;
01225596 4143
ba937127 4144 xs->header = bucket_xh(xs->bucket);
178eeac3 4145 xs->base = bucket_block(xs->bucket, 0);
01225596
TM		 4146 xs->end = xs->base + inode->i_sb->s_blocksize;
4147
178eeac3
JB		 4148 if (xs->not_found)
4149 return;
01225596 4150
178eeac3
JB		 4151 i = xs->here - old_xh->xh_entries;
4152 xs->here = &xs->header->xh_entries[i];
01225596
TM		 4153}
4154
4155static int ocfs2_xattr_create_index_block(struct inode *inode,
78f30c31
TM		 4156 struct ocfs2_xattr_search *xs,
4157 struct ocfs2_xattr_set_ctxt *ctxt)
01225596 4158{
85db90e7 4159 int ret;
01225596
TM		 4160 u32 bit_off, len;
4161 u64 blkno;
85db90e7 4162 handle_t *handle = ctxt->handle;
01225596
TM		 4163 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4164 struct ocfs2_inode_info *oi = OCFS2_I(inode);
01225596
TM		 4165 struct buffer_head *xb_bh = xs->xattr_bh;
4166 struct ocfs2_xattr_block *xb =
4167 (struct ocfs2_xattr_block *)xb_bh->b_data;
4168 struct ocfs2_xattr_tree_root *xr;
4169 u16 xb_flags = le16_to_cpu(xb->xb_flags);
01225596
TM		 4170
4171 mlog(0, "create xattr index block for %llu\n",
4172 (unsigned long long)xb_bh->b_blocknr);
4173
4174 BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
178eeac3 4175 BUG_ON(!xs->bucket);
01225596 4176
01225596
TM		 4177 /*
4178 * XXX:
4179 * We can use this lock for now, and maybe move to a dedicated mutex
4180 * if performance becomes a problem later.
4181 */
4182 down_write(&oi->ip_alloc_sem);
4183
0cf2f763 4184 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), xb_bh,
84008972 4185 OCFS2_JOURNAL_ACCESS_WRITE);
01225596
TM		 4186 if (ret) {
4187 mlog_errno(ret);
85db90e7 4188 goto out;
01225596
TM		 4189 }
4190
78f30c31
TM		 4191 ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac,
4192 1, 1, &bit_off, &len);
01225596
TM		 4193 if (ret) {
4194 mlog_errno(ret);
85db90e7 4195 goto out;
01225596
TM		 4196 }
4197
4198 /*
4199 * The bucket may spread in many blocks, and
4200 * we will only touch the 1st block and the last block
4201 * in the whole bucket(one for entry and one for data).
4202 */
4203 blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
4204
de29c085
MF		 4205 mlog(0, "allocate 1 cluster from %llu to xattr block\n",
4206 (unsigned long long)blkno);
01225596 4207
178eeac3
JB		 4208 ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
4209 if (ret) {
01225596 4210 mlog_errno(ret);
85db90e7 4211 goto out;
01225596
TM		 4212 }
4213
178eeac3
JB		 4214 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4215 OCFS2_JOURNAL_ACCESS_CREATE);
01225596
TM		 4216 if (ret) {
4217 mlog_errno(ret);
85db90e7 4218 goto out;
01225596
TM		 4219 }
4220
178eeac3
JB		 4221 ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
4222 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
01225596 4223
178eeac3 4224 ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
01225596
TM		 4225
4226 /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
4227 memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
4228 offsetof(struct ocfs2_xattr_block, xb_attrs));
4229
4230 xr = &xb->xb_attrs.xb_root;
4231 xr->xt_clusters = cpu_to_le32(1);
4232 xr->xt_last_eb_blk = 0;
4233 xr->xt_list.l_tree_depth = 0;
4234 xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
4235 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
4236
4237 xr->xt_list.l_recs[0].e_cpos = 0;
4238 xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
4239 xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
4240
4241 xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
4242
85db90e7 4243 ocfs2_journal_dirty(handle, xb_bh);
01225596 4244
85db90e7 4245out:
01225596
TM		 4246 up_write(&oi->ip_alloc_sem);
4247
01225596
TM		 4248 return ret;
4249}
4250
4251static int cmp_xe_offset(const void *a, const void *b)
4252{
4253 const struct ocfs2_xattr_entry *l = a, *r = b;
4254 u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
4255 u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
4256
4257 if (l_name_offset < r_name_offset)
4258 return 1;
4259 if (l_name_offset > r_name_offset)
4260 return -1;
4261 return 0;
4262}
4263
4264/*
4265 * defrag a xattr bucket if we find that the bucket has some
4266 * holes beteen name/value pairs.
4267 * We will move all the name/value pairs to the end of the bucket
4268 * so that we can spare some space for insertion.
4269 */
4270static int ocfs2_defrag_xattr_bucket(struct inode *inode,
85db90e7 4271 handle_t *handle,
01225596
TM		 4272 struct ocfs2_xattr_bucket *bucket)
4273{
4274 int ret, i;
199799a3 4275 size_t end, offset, len;
01225596
TM		 4276 struct ocfs2_xattr_header *xh;
4277 char *entries, *buf, *bucket_buf = NULL;
9c7759aa 4278 u64 blkno = bucket_blkno(bucket);
01225596 4279 u16 xh_free_start;
01225596 4280 size_t blocksize = inode->i_sb->s_blocksize;
01225596 4281 struct ocfs2_xattr_entry *xe;
01225596
TM		 4282
4283 /*
4284 * In order to make the operation more efficient and generic,
4285 * we copy all the blocks into a contiguous memory and do the
4286 * defragment there, so if anything is error, we will not touch
4287 * the real block.
4288 */
4289 bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
4290 if (!bucket_buf) {
4291 ret = -EIO;
4292 goto out;
4293 }
4294
4295 buf = bucket_buf;
1c32a2fd
TM		 4296 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4297 memcpy(buf, bucket_block(bucket, i), blocksize);
01225596 4298
1c32a2fd 4299 ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
161d6f30
JB		 4300 OCFS2_JOURNAL_ACCESS_WRITE);
4301 if (ret < 0) {
4302 mlog_errno(ret);
85db90e7 4303 goto out;
01225596
TM		 4304 }
4305
4306 xh = (struct ocfs2_xattr_header *)bucket_buf;
4307 entries = (char *)xh->xh_entries;
4308 xh_free_start = le16_to_cpu(xh->xh_free_start);
4309
4310 mlog(0, "adjust xattr bucket in %llu, count = %u, "
4311 "xh_free_start = %u, xh_name_value_len = %u.\n",
de29c085
MF		 4312 (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
4313 xh_free_start, le16_to_cpu(xh->xh_name_value_len));
01225596
TM		 4314
4315 /*
4316 * sort all the entries by their offset.
4317 * the largest will be the first, so that we can
4318 * move them to the end one by one.
4319 */
4320 sort(entries, le16_to_cpu(xh->xh_count),
4321 sizeof(struct ocfs2_xattr_entry),
4322 cmp_xe_offset, swap_xe);
4323
4324 /* Move all name/values to the end of the bucket. */
4325 xe = xh->xh_entries;
4326 end = OCFS2_XATTR_BUCKET_SIZE;
4327 for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
4328 offset = le16_to_cpu(xe->xe_name_offset);
199799a3 4329 len = namevalue_size_xe(xe);
01225596
TM		 4330
4331 /*
4332 * We must make sure that the name/value pair
4333 * exist in the same block. So adjust end to
4334 * the previous block end if needed.
4335 */
4336 if (((end - len) / blocksize !=
4337 (end - 1) / blocksize))
4338 end = end - end % blocksize;
4339
4340 if (end > offset + len) {
4341 memmove(bucket_buf + end - len,
4342 bucket_buf + offset, len);
4343 xe->xe_name_offset = cpu_to_le16(end - len);
4344 }
4345
4346 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
4347 "bucket %llu\n", (unsigned long long)blkno);
4348
4349 end -= len;
4350 }
4351
4352 mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
4353 "bucket %llu\n", (unsigned long long)blkno);
4354
4355 if (xh_free_start == end)
85db90e7 4356 goto out;
01225596
TM		 4357
4358 memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
4359 xh->xh_free_start = cpu_to_le16(end);
4360
4361 /* sort the entries by their name_hash. */
4362 sort(entries, le16_to_cpu(xh->xh_count),
4363 sizeof(struct ocfs2_xattr_entry),
4364 cmp_xe, swap_xe);
4365
4366 buf = bucket_buf;
1c32a2fd
TM		 4367 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4368 memcpy(bucket_block(bucket, i), buf, blocksize);
4369 ocfs2_xattr_bucket_journal_dirty(handle, bucket);
01225596 4370
01225596 4371out:
01225596
TM		 4372 kfree(bucket_buf);
4373 return ret;
4374}
4375
4376/*
b5c03e74
JB		 4377 * prev_blkno points to the start of an existing extent. new_blkno
4378 * points to a newly allocated extent. Because we know each of our
4379 * clusters contains more than bucket, we can easily split one cluster
4380 * at a bucket boundary. So we take the last cluster of the existing
4381 * extent and split it down the middle. We move the last half of the
4382 * buckets in the last cluster of the existing extent over to the new
4383 * extent.
01225596 4384 *
b5c03e74
JB		 4385 * first_bh is the buffer at prev_blkno so we can update the existing
4386 * extent's bucket count. header_bh is the bucket were we were hoping
4387 * to insert our xattr. If the bucket move places the target in the new
4388 * extent, we'll update first_bh and header_bh after modifying the old
4389 * extent.
4390 *
4391 * first_hash will be set as the 1st xe's name_hash in the new extent.
01225596
TM		 4392 */
4393static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
4394 handle_t *handle,
41cb8148
JB		 4395 struct ocfs2_xattr_bucket *first,
4396 struct ocfs2_xattr_bucket *target,
01225596 4397 u64 new_blkno,
01225596
TM		 4398 u32 num_clusters,
4399 u32 *first_hash)
4400{
c58b6032 4401 int ret;
41cb8148
JB		 4402 struct super_block *sb = inode->i_sb;
4403 int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
4404 int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
b5c03e74 4405 int to_move = num_buckets / 2;
c58b6032 4406 u64 src_blkno;
41cb8148
JB		 4407 u64 last_cluster_blkno = bucket_blkno(first) +
4408 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
01225596 4409
41cb8148
JB		 4410 BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
4411 BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
01225596 4412
01225596 4413 mlog(0, "move half of xattrs in cluster %llu to %llu\n",
c58b6032 4414 (unsigned long long)last_cluster_blkno, (unsigned long long)new_blkno);
01225596 4415
41cb8148 4416 ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
c58b6032
JB		 4417 last_cluster_blkno, new_blkno,
4418 to_move, first_hash);
01225596
TM		 4419 if (ret) {
4420 mlog_errno(ret);
4421 goto out;
4422 }
4423
c58b6032
JB		 4424 /* This is the first bucket that got moved */
4425 src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
01225596 4426
b5c03e74 4427 /*
c58b6032 4428 * If the target bucket was part of the moved buckets, we need to
41cb8148 4429 * update first and target.
b5c03e74 4430 */
41cb8148 4431 if (bucket_blkno(target) >= src_blkno) {
b5c03e74
JB		 4432 /* Find the block for the new target bucket */
4433 src_blkno = new_blkno +
41cb8148
JB		 4434 (bucket_blkno(target) - src_blkno);
4435
4436 ocfs2_xattr_bucket_relse(first);
4437 ocfs2_xattr_bucket_relse(target);
01225596 4438
b5c03e74 4439 /*
c58b6032 4440 * These shouldn't fail - the buffers are in the
b5c03e74
JB		 4441 * journal from ocfs2_cp_xattr_bucket().
4442 */
41cb8148 4443 ret = ocfs2_read_xattr_bucket(first, new_blkno);
c58b6032
JB		 4444 if (ret) {
4445 mlog_errno(ret);
4446 goto out;
4447 }
41cb8148
JB		 4448 ret = ocfs2_read_xattr_bucket(target, src_blkno);
4449 if (ret)
b5c03e74 4450 mlog_errno(ret);
01225596 4451
01225596
TM		 4452 }
4453
01225596 4454out:
01225596
TM		 4455 return ret;
4456}
4457
01225596 4458/*
80bcaf34
TM		 4459 * Find the suitable pos when we divide a bucket into 2.
4460 * We have to make sure the xattrs with the same hash value exist
4461 * in the same bucket.
4462 *
4463 * If this ocfs2_xattr_header covers more than one hash value, find a
4464 * place where the hash value changes. Try to find the most even split.
4465 * The most common case is that all entries have different hash values,
4466 * and the first check we make will find a place to split.
4467 */
4468static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
4469{
4470 struct ocfs2_xattr_entry *entries = xh->xh_entries;
4471 int count = le16_to_cpu(xh->xh_count);
4472 int delta, middle = count / 2;
4473
4474 /*
4475 * We start at the middle. Each step gets farther away in both
4476 * directions. We therefore hit the change in hash value
4477 * nearest to the middle. Note that this loop does not execute for
4478 * count < 2.
4479 */
4480 for (delta = 0; delta < middle; delta++) {
4481 /* Let's check delta earlier than middle */
4482 if (cmp_xe(&entries[middle - delta - 1],
4483 &entries[middle - delta]))
4484 return middle - delta;
4485
4486 /* For even counts, don't walk off the end */
4487 if ((middle + delta + 1) == count)
4488 continue;
4489
4490 /* Now try delta past middle */
4491 if (cmp_xe(&entries[middle + delta],
4492 &entries[middle + delta + 1]))
4493 return middle + delta + 1;
4494 }
4495
4496 /* Every entry had the same hash */
4497 return count;
4498}
4499
4500/*
4501 * Move some xattrs in old bucket(blk) to new bucket(new_blk).
01225596 4502 * first_hash will record the 1st hash of the new bucket.
80bcaf34
TM		 4503 *
4504 * Normally half of the xattrs will be moved. But we have to make
4505 * sure that the xattrs with the same hash value are stored in the
4506 * same bucket. If all the xattrs in this bucket have the same hash
4507 * value, the new bucket will be initialized as an empty one and the
4508 * first_hash will be initialized as (hash_value+1).
01225596 4509 */
80bcaf34
TM		 4510static int ocfs2_divide_xattr_bucket(struct inode *inode,
4511 handle_t *handle,
4512 u64 blk,
4513 u64 new_blk,
4514 u32 *first_hash,
4515 int new_bucket_head)
01225596
TM		 4516{
4517 int ret, i;
199799a3 4518 int count, start, len, name_value_len = 0, name_offset = 0;
ba937127 4519 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
01225596
TM		 4520 struct ocfs2_xattr_header *xh;
4521 struct ocfs2_xattr_entry *xe;
4522 int blocksize = inode->i_sb->s_blocksize;
4523
80bcaf34 4524 mlog(0, "move some of xattrs from bucket %llu to %llu\n",
de29c085 4525 (unsigned long long)blk, (unsigned long long)new_blk);
01225596 4526
ba937127
JB		 4527 s_bucket = ocfs2_xattr_bucket_new(inode);
4528 t_bucket = ocfs2_xattr_bucket_new(inode);
4529 if (!s_bucket || !t_bucket) {
4530 ret = -ENOMEM;
4531 mlog_errno(ret);
4532 goto out;
4533 }
01225596 4534
ba937127 4535 ret = ocfs2_read_xattr_bucket(s_bucket, blk);
01225596
TM		 4536 if (ret) {
4537 mlog_errno(ret);
4538 goto out;
4539 }
4540
ba937127 4541 ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
1224be02 4542 OCFS2_JOURNAL_ACCESS_WRITE);
01225596
TM		 4543 if (ret) {
4544 mlog_errno(ret);
4545 goto out;
4546 }
4547
784b816a
JB		 4548 /*
4549 * Even if !new_bucket_head, we're overwriting t_bucket. Thus,
4550 * there's no need to read it.
4551 */
ba937127 4552 ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
01225596
TM		 4553 if (ret) {
4554 mlog_errno(ret);
4555 goto out;
4556 }
4557
2b656c1d
JB		 4558 /*
4559 * Hey, if we're overwriting t_bucket, what difference does
4560 * ACCESS_CREATE vs ACCESS_WRITE make? See the comment in the
4561 * same part of ocfs2_cp_xattr_bucket().
4562 */
ba937127 4563 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
1224be02
JB		 4564 new_bucket_head ?
4565 OCFS2_JOURNAL_ACCESS_CREATE :
4566 OCFS2_JOURNAL_ACCESS_WRITE);
4567 if (ret) {
4568 mlog_errno(ret);
4569 goto out;
01225596
TM		 4570 }
4571
ba937127 4572 xh = bucket_xh(s_bucket);
80bcaf34
TM		 4573 count = le16_to_cpu(xh->xh_count);
4574 start = ocfs2_xattr_find_divide_pos(xh);
4575
4576 if (start == count) {
4577 xe = &xh->xh_entries[start-1];
4578
4579 /*
4580 * initialized a new empty bucket here.
4581 * The hash value is set as one larger than
4582 * that of the last entry in the previous bucket.
4583 */
ba937127
JB		 4584 for (i = 0; i < t_bucket->bu_blocks; i++)
4585 memset(bucket_block(t_bucket, i), 0, blocksize);
80bcaf34 4586
ba937127 4587 xh = bucket_xh(t_bucket);
80bcaf34
TM		 4588 xh->xh_free_start = cpu_to_le16(blocksize);
4589 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
4590 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
4591
4592 goto set_num_buckets;
4593 }
4594
01225596 4595 /* copy the whole bucket to the new first. */
ba937127 4596 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
01225596
TM		 4597
4598 /* update the new bucket. */
ba937127 4599 xh = bucket_xh(t_bucket);
01225596
TM		 4600
4601 /*
4602 * Calculate the total name/value len and xh_free_start for
4603 * the old bucket first.
4604 */
4605 name_offset = OCFS2_XATTR_BUCKET_SIZE;
4606 name_value_len = 0;
4607 for (i = 0; i < start; i++) {
4608 xe = &xh->xh_entries[i];
199799a3 4609 name_value_len += namevalue_size_xe(xe);
01225596
TM		 4610 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
4611 name_offset = le16_to_cpu(xe->xe_name_offset);
4612 }
4613
4614 /*
4615 * Now begin the modification to the new bucket.
4616 *
4617 * In the new bucket, We just move the xattr entry to the beginning
4618 * and don't touch the name/value. So there will be some holes in the
4619 * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
4620 * called.
4621 */
4622 xe = &xh->xh_entries[start];
4623 len = sizeof(struct ocfs2_xattr_entry) * (count - start);
4624 mlog(0, "mv xattr entry len %d from %d to %d\n", len,
ff1ec20e
MF		 4625 (int)((char *)xe - (char *)xh),
4626 (int)((char *)xh->xh_entries - (char *)xh));
01225596
TM		 4627 memmove((char *)xh->xh_entries, (char *)xe, len);
4628 xe = &xh->xh_entries[count - start];
4629 len = sizeof(struct ocfs2_xattr_entry) * start;
4630 memset((char *)xe, 0, len);
4631
4632 le16_add_cpu(&xh->xh_count, -start);
4633 le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
4634
4635 /* Calculate xh_free_start for the new bucket. */
4636 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4637 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4638 xe = &xh->xh_entries[i];
01225596
TM		 4639 if (le16_to_cpu(xe->xe_name_offset) <
4640 le16_to_cpu(xh->xh_free_start))
4641 xh->xh_free_start = xe->xe_name_offset;
4642 }
4643
80bcaf34 4644set_num_buckets:
01225596
TM		 4645 /* set xh->xh_num_buckets for the new xh. */
4646 if (new_bucket_head)
4647 xh->xh_num_buckets = cpu_to_le16(1);
4648 else
4649 xh->xh_num_buckets = 0;
4650
ba937127 4651 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
01225596
TM		 4652
4653 /* store the first_hash of the new bucket. */
4654 if (first_hash)
4655 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
4656
4657 /*
80bcaf34
TM		 4658 * Now only update the 1st block of the old bucket. If we
4659 * just added a new empty bucket, there is no need to modify
4660 * it.
01225596 4661 */
80bcaf34
TM		 4662 if (start == count)
4663 goto out;
4664
ba937127 4665 xh = bucket_xh(s_bucket);
01225596
TM		 4666 memset(&xh->xh_entries[start], 0,
4667 sizeof(struct ocfs2_xattr_entry) * (count - start));
4668 xh->xh_count = cpu_to_le16(start);
4669 xh->xh_free_start = cpu_to_le16(name_offset);
4670 xh->xh_name_value_len = cpu_to_le16(name_value_len);
4671
ba937127 4672 ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
01225596
TM		 4673
4674out:
ba937127
JB		 4675 ocfs2_xattr_bucket_free(s_bucket);
4676 ocfs2_xattr_bucket_free(t_bucket);
01225596
TM		 4677
4678 return ret;
4679}
4680
4681/*
4682 * Copy xattr from one bucket to another bucket.
4683 *
4684 * The caller must make sure that the journal transaction
4685 * has enough space for journaling.
4686 */
4687static int ocfs2_cp_xattr_bucket(struct inode *inode,
4688 handle_t *handle,
4689 u64 s_blkno,
4690 u64 t_blkno,
4691 int t_is_new)
4692{
4980c6da 4693 int ret;
ba937127 4694 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
01225596
TM		 4695
4696 BUG_ON(s_blkno == t_blkno);
4697
4698 mlog(0, "cp bucket %llu to %llu, target is %d\n",
de29c085
MF		 4699 (unsigned long long)s_blkno, (unsigned long long)t_blkno,
4700 t_is_new);
01225596 4701
ba937127
JB		 4702 s_bucket = ocfs2_xattr_bucket_new(inode);
4703 t_bucket = ocfs2_xattr_bucket_new(inode);
4704 if (!s_bucket || !t_bucket) {
4705 ret = -ENOMEM;
4706 mlog_errno(ret);
4707 goto out;
4708 }
92de109a 4709
ba937127 4710 ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
01225596
TM		 4711 if (ret)
4712 goto out;
4713
784b816a
JB		 4714 /*
4715 * Even if !t_is_new, we're overwriting t_bucket. Thus,
4716 * there's no need to read it.
4717 */
ba937127 4718 ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
01225596
TM		 4719 if (ret)
4720 goto out;
4721
2b656c1d
JB		 4722 /*
4723 * Hey, if we're overwriting t_bucket, what difference does
4724 * ACCESS_CREATE vs ACCESS_WRITE make? Well, if we allocated a new
874d65af
JB		 4725 * cluster to fill, we came here from
4726 * ocfs2_mv_xattr_buckets(), and it is really new -
4727 * ACCESS_CREATE is required. But we also might have moved data
4728 * out of t_bucket before extending back into it.
4729 * ocfs2_add_new_xattr_bucket() can do this - its call to
4730 * ocfs2_add_new_xattr_cluster() may have created a new extent
2b656c1d
JB		 4731 * and copied out the end of the old extent. Then it re-extends
4732 * the old extent back to create space for new xattrs. That's
4733 * how we get here, and the bucket isn't really new.
4734 */
ba937127 4735 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
1224be02
JB		 4736 t_is_new ?
4737 OCFS2_JOURNAL_ACCESS_CREATE :
4738 OCFS2_JOURNAL_ACCESS_WRITE);
4739 if (ret)
4740 goto out;
01225596 4741
ba937127
JB		 4742 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4743 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
01225596
TM		 4744
4745out:
ba937127
JB		 4746 ocfs2_xattr_bucket_free(t_bucket);
4747 ocfs2_xattr_bucket_free(s_bucket);
01225596
TM		 4748
4749 return ret;
4750}
4751
4752/*
874d65af
JB		 4753 * src_blk points to the start of an existing extent. last_blk points to
4754 * last cluster in that extent. to_blk points to a newly allocated
54ecb6b6
JB		 4755 * extent. We copy the buckets from the cluster at last_blk to the new
4756 * extent. If start_bucket is non-zero, we skip that many buckets before
4757 * we start copying. The new extent's xh_num_buckets gets set to the
4758 * number of buckets we copied. The old extent's xh_num_buckets shrinks
4759 * by the same amount.
01225596 4760 */
54ecb6b6
JB		 4761static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4762 u64 src_blk, u64 last_blk, u64 to_blk,
4763 unsigned int start_bucket,
4764 u32 *first_hash)
01225596
TM		 4765{
4766 int i, ret, credits;
4767 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
15d60929 4768 int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
01225596 4769 int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
15d60929 4770 struct ocfs2_xattr_bucket *old_first, *new_first;
01225596 4771
874d65af
JB		 4772 mlog(0, "mv xattrs from cluster %llu to %llu\n",
4773 (unsigned long long)last_blk, (unsigned long long)to_blk);
01225596 4774
54ecb6b6
JB		 4775 BUG_ON(start_bucket >= num_buckets);
4776 if (start_bucket) {
4777 num_buckets -= start_bucket;
4778 last_blk += (start_bucket * blks_per_bucket);
4779 }
4780
15d60929
JB		 4781 /* The first bucket of the original extent */
4782 old_first = ocfs2_xattr_bucket_new(inode);
4783 /* The first bucket of the new extent */
4784 new_first = ocfs2_xattr_bucket_new(inode);
4785 if (!old_first || !new_first) {
4786 ret = -ENOMEM;
4787 mlog_errno(ret);
4788 goto out;
4789 }
4790
874d65af 4791 ret = ocfs2_read_xattr_bucket(old_first, src_blk);
15d60929
JB		 4792 if (ret) {
4793 mlog_errno(ret);
4794 goto out;
4795 }
4796
01225596 4797 /*
54ecb6b6
JB		 4798 * We need to update the first bucket of the old extent and all
4799 * the buckets going to the new extent.
01225596 4800 */
54ecb6b6
JB		 4801 credits = ((num_buckets + 1) * blks_per_bucket) +
4802 handle->h_buffer_credits;
01225596
TM		 4803 ret = ocfs2_extend_trans(handle, credits);
4804 if (ret) {
4805 mlog_errno(ret);
4806 goto out;
4807 }
4808
15d60929
JB		 4809 ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4810 OCFS2_JOURNAL_ACCESS_WRITE);
01225596
TM		 4811 if (ret) {
4812 mlog_errno(ret);
4813 goto out;
4814 }
4815
4816 for (i = 0; i < num_buckets; i++) {
4817 ret = ocfs2_cp_xattr_bucket(inode, handle,
874d65af 4818 last_blk + (i * blks_per_bucket),
15d60929
JB		 4819 to_blk + (i * blks_per_bucket),
4820 1);
01225596
TM		 4821 if (ret) {
4822 mlog_errno(ret);
4823 goto out;
4824 }
01225596
TM		 4825 }
4826
15d60929
JB		 4827 /*
4828 * Get the new bucket ready before we dirty anything
4829 * (This actually shouldn't fail, because we already dirtied
4830 * it once in ocfs2_cp_xattr_bucket()).
4831 */
4832 ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4833 if (ret) {
01225596
TM		 4834 mlog_errno(ret);
4835 goto out;
4836 }
15d60929
JB		 4837 ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4838 OCFS2_JOURNAL_ACCESS_WRITE);
01225596
TM		 4839 if (ret) {
4840 mlog_errno(ret);
4841 goto out;
4842 }
4843
15d60929
JB		 4844 /* Now update the headers */
4845 le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4846 ocfs2_xattr_bucket_journal_dirty(handle, old_first);
01225596 4847
15d60929
JB		 4848 bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4849 ocfs2_xattr_bucket_journal_dirty(handle, new_first);
01225596
TM		 4850
4851 if (first_hash)
15d60929
JB		 4852 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4853
01225596 4854out:
15d60929
JB		 4855 ocfs2_xattr_bucket_free(new_first);
4856 ocfs2_xattr_bucket_free(old_first);
01225596
TM		 4857 return ret;
4858}
4859
4860/*
80bcaf34 4861 * Move some xattrs in this cluster to the new cluster.
01225596
TM		 4862 * This function should only be called when bucket size == cluster size.
4863 * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4864 */
80bcaf34
TM		 4865static int ocfs2_divide_xattr_cluster(struct inode *inode,
4866 handle_t *handle,
4867 u64 prev_blk,
4868 u64 new_blk,
4869 u32 *first_hash)
01225596
TM		 4870{
4871 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
85db90e7 4872 int ret, credits = 2 * blk_per_bucket + handle->h_buffer_credits;
01225596
TM		 4873
4874 BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4875
4876 ret = ocfs2_extend_trans(handle, credits);
4877 if (ret) {
4878 mlog_errno(ret);
4879 return ret;
4880 }
4881
4882 /* Move half of the xattr in start_blk to the next bucket. */
80bcaf34
TM		 4883 return ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4884 new_blk, first_hash, 1);
01225596
TM		 4885}
4886
4887/*
4888 * Move some xattrs from the old cluster to the new one since they are not
4889 * contiguous in ocfs2 xattr tree.
4890 *
4891 * new_blk starts a new separate cluster, and we will move some xattrs from
4892 * prev_blk to it. v_start will be set as the first name hash value in this
4893 * new cluster so that it can be used as e_cpos during tree insertion and
4894 * don't collide with our original b-tree operations. first_bh and header_bh
4895 * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4896 * to extend the insert bucket.
4897 *
4898 * The problem is how much xattr should we move to the new one and when should
4899 * we update first_bh and header_bh?
4900 * 1. If cluster size > bucket size, that means the previous cluster has more
4901 * than 1 bucket, so just move half nums of bucket into the new cluster and
4902 * update the first_bh and header_bh if the insert bucket has been moved
4903 * to the new cluster.
4904 * 2. If cluster_size == bucket_size:
4905 * a) If the previous extent rec has more than one cluster and the insert
4906 * place isn't in the last cluster, copy the entire last cluster to the
4907 * new one. This time, we don't need to upate the first_bh and header_bh
4908 * since they will not be moved into the new cluster.
4909 * b) Otherwise, move the bottom half of the xattrs in the last cluster into
4910 * the new one. And we set the extend flag to zero if the insert place is
4911 * moved into the new allocated cluster since no extend is needed.
4912 */
4913static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
4914 handle_t *handle,
012ee910
JB		 4915 struct ocfs2_xattr_bucket *first,
4916 struct ocfs2_xattr_bucket *target,
01225596 4917 u64 new_blk,
01225596
TM		 4918 u32 prev_clusters,
4919 u32 *v_start,
4920 int *extend)
4921{
92cf3adf 4922 int ret;
01225596
TM		 4923
4924 mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
012ee910 4925 (unsigned long long)bucket_blkno(first), prev_clusters,
de29c085 4926 (unsigned long long)new_blk);
01225596 4927
41cb8148 4928 if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
01225596
TM		 4929 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
4930 handle,
41cb8148 4931 first, target,
01225596 4932 new_blk,
01225596
TM		 4933 prev_clusters,
4934 v_start);
012ee910 4935 if (ret)
41cb8148 4936 mlog_errno(ret);
41cb8148 4937 } else {
92cf3adf
JB		 4938 /* The start of the last cluster in the first extent */
4939 u64 last_blk = bucket_blkno(first) +
4940 ((prev_clusters - 1) *
4941 ocfs2_clusters_to_blocks(inode->i_sb, 1));
01225596 4942
012ee910 4943 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
874d65af 4944 ret = ocfs2_mv_xattr_buckets(inode, handle,
92cf3adf 4945 bucket_blkno(first),
54ecb6b6 4946 last_blk, new_blk, 0,
01225596 4947 v_start);
012ee910
JB		 4948 if (ret)
4949 mlog_errno(ret);
4950 } else {
80bcaf34
TM		 4951 ret = ocfs2_divide_xattr_cluster(inode, handle,
4952 last_blk, new_blk,
4953 v_start);
012ee910
JB		 4954 if (ret)
4955 mlog_errno(ret);
01225596 4956
92cf3adf 4957 if ((bucket_blkno(target) == last_blk) && extend)
01225596
TM		 4958 *extend = 0;
4959 }
4960 }
4961
4962 return ret;
4963}
4964
4965/*
4966 * Add a new cluster for xattr storage.
4967 *
4968 * If the new cluster is contiguous with the previous one, it will be
4969 * appended to the same extent record, and num_clusters will be updated.
4970 * If not, we will insert a new extent for it and move some xattrs in
4971 * the last cluster into the new allocated one.
4972 * We also need to limit the maximum size of a btree leaf, otherwise we'll
4973 * lose the benefits of hashing because we'll have to search large leaves.
4974 * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
4975 * if it's bigger).
4976 *
4977 * first_bh is the first block of the previous extent rec and header_bh
4978 * indicates the bucket we will insert the new xattrs. They will be updated
4979 * when the header_bh is moved into the new cluster.
4980 */
4981static int ocfs2_add_new_xattr_cluster(struct inode *inode,
4982 struct buffer_head *root_bh,
ed29c0ca
JB		 4983 struct ocfs2_xattr_bucket *first,
4984 struct ocfs2_xattr_bucket *target,
01225596
TM		 4985 u32 *num_clusters,
4986 u32 prev_cpos,
78f30c31
TM		 4987 int *extend,
4988 struct ocfs2_xattr_set_ctxt *ctxt)
01225596 4989{
85db90e7 4990 int ret;
01225596
TM		 4991 u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
4992 u32 prev_clusters = *num_clusters;
4993 u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
4994 u64 block;
85db90e7 4995 handle_t *handle = ctxt->handle;
01225596 4996 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
f99b9b7c 4997 struct ocfs2_extent_tree et;
01225596
TM		 4998
4999 mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
5000 "previous xattr blkno = %llu\n",
5001 (unsigned long long)OCFS2_I(inode)->ip_blkno,
ed29c0ca 5002 prev_cpos, (unsigned long long)bucket_blkno(first));
01225596 5003
5e404e9e 5004 ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
f99b9b7c 5005
0cf2f763 5006 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
84008972 5007 OCFS2_JOURNAL_ACCESS_WRITE);
01225596
TM		 5008 if (ret < 0) {
5009 mlog_errno(ret);
5010 goto leave;
5011 }
5012
78f30c31 5013 ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac, 1,
01225596
TM		 5014 clusters_to_add, &bit_off, &num_bits);
5015 if (ret < 0) {
5016 if (ret != -ENOSPC)
5017 mlog_errno(ret);
5018 goto leave;
5019 }
5020
5021 BUG_ON(num_bits > clusters_to_add);
5022
5023 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
5024 mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
5025 num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
5026
ed29c0ca 5027 if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
01225596
TM		 5028 (prev_clusters + num_bits) << osb->s_clustersize_bits <=
5029 OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
5030 /*
5031 * If this cluster is contiguous with the old one and
5032 * adding this new cluster, we don't surpass the limit of
5033 * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
5034 * initialized and used like other buckets in the previous
5035 * cluster.
5036 * So add it as a contiguous one. The caller will handle
5037 * its init process.
5038 */
5039 v_start = prev_cpos + prev_clusters;
5040 *num_clusters = prev_clusters + num_bits;
5041 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
5042 num_bits);
5043 } else {
5044 ret = ocfs2_adjust_xattr_cross_cluster(inode,
5045 handle,
012ee910
JB		 5046 first,
5047 target,
01225596 5048 block,
01225596
TM		 5049 prev_clusters,
5050 &v_start,
5051 extend);
5052 if (ret) {
5053 mlog_errno(ret);
5054 goto leave;
5055 }
5056 }
5057
5058 mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
de29c085 5059 num_bits, (unsigned long long)block, v_start);
cc79d8c1 5060 ret = ocfs2_insert_extent(handle, &et, v_start, block,
78f30c31 5061 num_bits, 0, ctxt->meta_ac);
01225596
TM		 5062 if (ret < 0) {
5063 mlog_errno(ret);
5064 goto leave;
5065 }
5066
5067 ret = ocfs2_journal_dirty(handle, root_bh);
85db90e7 5068 if (ret < 0)
01225596 5069 mlog_errno(ret);
01225596
TM		 5070
5071leave:
01225596
TM		 5072 return ret;
5073}
5074
5075/*
92de109a
JB		 5076 * We are given an extent. 'first' is the bucket at the very front of
5077 * the extent. The extent has space for an additional bucket past
5078 * bucket_xh(first)->xh_num_buckets. 'target_blkno' is the block number
5079 * of the target bucket. We wish to shift every bucket past the target
5080 * down one, filling in that additional space. When we get back to the
5081 * target, we split the target between itself and the now-empty bucket
5082 * at target+1 (aka, target_blkno + blks_per_bucket).
01225596
TM		 5083 */
5084static int ocfs2_extend_xattr_bucket(struct inode *inode,
85db90e7 5085 handle_t *handle,
92de109a
JB		 5086 struct ocfs2_xattr_bucket *first,
5087 u64 target_blk,
01225596
TM		 5088 u32 num_clusters)
5089{
5090 int ret, credits;
5091 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5092 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
92de109a
JB		 5093 u64 end_blk;
5094 u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
01225596
TM		 5095
5096 mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
92de109a
JB		 5097 "from %llu, len = %u\n", (unsigned long long)target_blk,
5098 (unsigned long long)bucket_blkno(first), num_clusters);
01225596 5099
92de109a
JB		 5100 /* The extent must have room for an additional bucket */
5101 BUG_ON(new_bucket >=
5102 (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
01225596 5103
92de109a
JB		 5104 /* end_blk points to the last existing bucket */
5105 end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
01225596
TM		 5106
5107 /*
92de109a
JB		 5108 * end_blk is the start of the last existing bucket.
5109 * Thus, (end_blk - target_blk) covers the target bucket and
5110 * every bucket after it up to, but not including, the last
5111 * existing bucket. Then we add the last existing bucket, the
5112 * new bucket, and the first bucket (3 * blk_per_bucket).
01225596 5113 */
92de109a 5114 credits = (end_blk - target_blk) + (3 * blk_per_bucket) +
85db90e7
TM		 5115 handle->h_buffer_credits;
5116 ret = ocfs2_extend_trans(handle, credits);
5117 if (ret) {
01225596
TM		 5118 mlog_errno(ret);
5119 goto out;
5120 }
5121
92de109a
JB		 5122 ret = ocfs2_xattr_bucket_journal_access(handle, first,
5123 OCFS2_JOURNAL_ACCESS_WRITE);
01225596
TM		 5124 if (ret) {
5125 mlog_errno(ret);
85db90e7 5126 goto out;
01225596
TM		 5127 }
5128
92de109a 5129 while (end_blk != target_blk) {
01225596
TM		 5130 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
5131 end_blk + blk_per_bucket, 0);
5132 if (ret)
85db90e7 5133 goto out;
01225596
TM		 5134 end_blk -= blk_per_bucket;
5135 }
5136
92de109a
JB		 5137 /* Move half of the xattr in target_blkno to the next bucket. */
5138 ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
5139 target_blk + blk_per_bucket, NULL, 0);
01225596 5140
92de109a
JB		 5141 le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
5142 ocfs2_xattr_bucket_journal_dirty(handle, first);
01225596 5143
01225596
TM		 5144out:
5145 return ret;
5146}
5147
5148/*
91f2033f
JB		 5149 * Add new xattr bucket in an extent record and adjust the buckets
5150 * accordingly. xb_bh is the ocfs2_xattr_block, and target is the
5151 * bucket we want to insert into.
01225596 5152 *
91f2033f
JB		 5153 * In the easy case, we will move all the buckets after target down by
5154 * one. Half of target's xattrs will be moved to the next bucket.
5155 *
5156 * If current cluster is full, we'll allocate a new one. This may not
5157 * be contiguous. The underlying calls will make sure that there is
5158 * space for the insert, shifting buckets around if necessary.
5159 * 'target' may be moved by those calls.
01225596
TM		 5160 */
5161static int ocfs2_add_new_xattr_bucket(struct inode *inode,
5162 struct buffer_head *xb_bh,
91f2033f 5163 struct ocfs2_xattr_bucket *target,
78f30c31 5164 struct ocfs2_xattr_set_ctxt *ctxt)
01225596 5165{
01225596
TM		 5166 struct ocfs2_xattr_block *xb =
5167 (struct ocfs2_xattr_block *)xb_bh->b_data;
5168 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
5169 struct ocfs2_extent_list *el = &xb_root->xt_list;
91f2033f
JB		 5170 u32 name_hash =
5171 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
ed29c0ca 5172 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
01225596
TM		 5173 int ret, num_buckets, extend = 1;
5174 u64 p_blkno;
5175 u32 e_cpos, num_clusters;
92de109a 5176 /* The bucket at the front of the extent */
91f2033f 5177 struct ocfs2_xattr_bucket *first;
01225596 5178
91f2033f
JB		 5179 mlog(0, "Add new xattr bucket starting from %llu\n",
5180 (unsigned long long)bucket_blkno(target));
01225596 5181
ed29c0ca 5182 /* The first bucket of the original extent */
92de109a 5183 first = ocfs2_xattr_bucket_new(inode);
91f2033f 5184 if (!first) {
92de109a
JB		 5185 ret = -ENOMEM;
5186 mlog_errno(ret);
5187 goto out;
5188 }
5189
01225596
TM		 5190 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
5191 &num_clusters, el);
5192 if (ret) {
5193 mlog_errno(ret);
5194 goto out;
5195 }
5196
ed29c0ca 5197 ret = ocfs2_read_xattr_bucket(first, p_blkno);
01225596
TM		 5198 if (ret) {
5199 mlog_errno(ret);
5200 goto out;
5201 }
5202
ed29c0ca
JB		 5203 num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
5204 if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
5205 /*
5206 * This can move first+target if the target bucket moves
5207 * to the new extent.
5208 */
01225596
TM		 5209 ret = ocfs2_add_new_xattr_cluster(inode,
5210 xb_bh,
ed29c0ca
JB		 5211 first,
5212 target,
01225596
TM		 5213 &num_clusters,
5214 e_cpos,
78f30c31
TM		 5215 &extend,
5216 ctxt);
01225596
TM		 5217 if (ret) {
5218 mlog_errno(ret);
5219 goto out;
5220 }
5221 }
5222
92de109a 5223 if (extend) {
01225596 5224 ret = ocfs2_extend_xattr_bucket(inode,
85db90e7 5225 ctxt->handle,
ed29c0ca
JB		 5226 first,
5227 bucket_blkno(target),
01225596 5228 num_clusters);
92de109a
JB		 5229 if (ret)
5230 mlog_errno(ret);
5231 }
5232
01225596 5233out:
92de109a 5234 ocfs2_xattr_bucket_free(first);
ed29c0ca 5235
01225596
TM		 5236 return ret;
5237}
5238
5239static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
5240 struct ocfs2_xattr_bucket *bucket,
5241 int offs)
5242{
5243 int block_off = offs >> inode->i_sb->s_blocksize_bits;
5244
5245 offs = offs % inode->i_sb->s_blocksize;
51def39f 5246 return bucket_block(bucket, block_off) + offs;
01225596
TM		 5247}
5248
01225596
TM		 5249/*
5250 * Set the xattr entry in the specified bucket.
5251 * The bucket is indicated by xs->bucket and it should have the enough
5252 * space for the xattr insertion.
5253 */
5254static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
85db90e7 5255 handle_t *handle,
01225596
TM		 5256 struct ocfs2_xattr_info *xi,
5257 struct ocfs2_xattr_search *xs,
5258 u32 name_hash,
5a095611 5259 int local)
01225596 5260{
1224be02 5261 int ret;
02dbf38d 5262 u64 blkno;
69a3e539 5263 struct ocfs2_xa_loc loc;
01225596 5264
ff1ec20e 5265 mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
6b240ff6 5266 (unsigned long)xi->xi_value_len, xi->xi_name_index,
ba937127 5267 (unsigned long long)bucket_blkno(xs->bucket));
01225596 5268
ba937127 5269 if (!xs->bucket->bu_bhs[1]) {
02dbf38d
JB		 5270 blkno = bucket_blkno(xs->bucket);
5271 ocfs2_xattr_bucket_relse(xs->bucket);
5272 ret = ocfs2_read_xattr_bucket(xs->bucket, blkno);
01225596
TM		 5273 if (ret) {
5274 mlog_errno(ret);
5275 goto out;
5276 }
5277 }
5278
ba937127 5279 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
1224be02
JB		 5280 OCFS2_JOURNAL_ACCESS_WRITE);
5281 if (ret < 0) {
5282 mlog_errno(ret);
5283 goto out;
01225596
TM		 5284 }
5285
69a3e539
JB		 5286 ocfs2_init_xattr_bucket_xa_loc(&loc, xs->bucket,
5287 xs->not_found ? NULL : xs->here);
5288 ret = ocfs2_xa_prepare_entry(&loc, xi, name_hash);
5289 if (ret) {
5290 if (ret != -ENOSPC)
5291 mlog_errno(ret);
5292 goto out;
5293 }
5294 /* XXX For now, until we make ocfs2_xa_prepare_entry() primary */
5295 BUG_ON(ret == -ENOSPC);
5296 ocfs2_xa_store_inline_value(&loc, xi);
5297 xs->here = loc.xl_entry;
5298
ba937127 5299 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
01225596 5300
01225596 5301out:
01225596
TM		 5302 return ret;
5303}
5304
01225596
TM		 5305/*
5306 * Truncate the specified xe_off entry in xattr bucket.
5307 * bucket is indicated by header_bh and len is the new length.
5308 * Both the ocfs2_xattr_value_root and the entry will be updated here.
5309 *
5310 * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
5311 */
5312static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
548b0f22 5313 struct ocfs2_xattr_bucket *bucket,
01225596 5314 int xe_off,
78f30c31
TM		 5315 int len,
5316 struct ocfs2_xattr_set_ctxt *ctxt)
01225596
TM		 5317{
5318 int ret, offset;
5319 u64 value_blk;
01225596 5320 struct ocfs2_xattr_entry *xe;
548b0f22 5321 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
01225596 5322 size_t blocksize = inode->i_sb->s_blocksize;
b3e5d379
JB		 5323 struct ocfs2_xattr_value_buf vb = {
5324 .vb_access = ocfs2_journal_access,
5325 };
01225596
TM		 5326
5327 xe = &xh->xh_entries[xe_off];
5328
5329 BUG_ON(!xe || ocfs2_xattr_is_local(xe));
5330
5331 offset = le16_to_cpu(xe->xe_name_offset) +
5332 OCFS2_XATTR_SIZE(xe->xe_name_len);
5333
5334 value_blk = offset / blocksize;
5335
5336 /* We don't allow ocfs2_xattr_value to be stored in different block. */
5337 BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
01225596 5338
b3e5d379
JB		 5339 vb.vb_bh = bucket->bu_bhs[value_blk];
5340 BUG_ON(!vb.vb_bh);
01225596 5341
b3e5d379
JB		 5342 vb.vb_xv = (struct ocfs2_xattr_value_root *)
5343 (vb.vb_bh->b_data + offset % blocksize);
01225596 5344
548b0f22
JB		 5345 /*
5346 * From here on out we have to dirty the bucket. The generic
5347 * value calls only modify one of the bucket's bhs, but we need
5348 * to send the bucket at once. So if they error, they *could* have
5349 * modified something. We have to assume they did, and dirty
5350 * the whole bucket. This leaves us in a consistent state.
5351 */
5352 mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
5353 xe_off, (unsigned long long)bucket_blkno(bucket), len);
b3e5d379 5354 ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
01225596
TM		 5355 if (ret) {
5356 mlog_errno(ret);
554e7f9e
TM		 5357 goto out;
5358 }
5359
5360 ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
5361 OCFS2_JOURNAL_ACCESS_WRITE);
5362 if (ret) {
5363 mlog_errno(ret);
5364 goto out;
01225596
TM		 5365 }
5366
548b0f22
JB		 5367 xe->xe_value_size = cpu_to_le64(len);
5368
548b0f22
JB		 5369 ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
5370
01225596 5371out:
01225596
TM		 5372 return ret;
5373}
5374
5375static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
78f30c31
TM		 5376 struct ocfs2_xattr_search *xs,
5377 int len,
5378 struct ocfs2_xattr_set_ctxt *ctxt)
01225596
TM		 5379{
5380 int ret, offset;
5381 struct ocfs2_xattr_entry *xe = xs->here;
5382 struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
5383
ba937127 5384 BUG_ON(!xs->bucket->bu_bhs[0] || !xe || ocfs2_xattr_is_local(xe));
01225596
TM		 5385
5386 offset = xe - xh->xh_entries;
548b0f22 5387 ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket,
78f30c31 5388 offset, len, ctxt);
01225596
TM		 5389 if (ret)
5390 mlog_errno(ret);
5391
5392 return ret;
5393}
5394
5395static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
85db90e7 5396 handle_t *handle,
01225596
TM		 5397 struct ocfs2_xattr_search *xs,
5398 char *val,
5399 int value_len)
5400{
712e53e4 5401 int ret, offset, block_off;
01225596
TM		 5402 struct ocfs2_xattr_value_root *xv;
5403 struct ocfs2_xattr_entry *xe = xs->here;
712e53e4
TM		 5404 struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
5405 void *base;
492a8a33
TM		 5406 struct ocfs2_xattr_value_buf vb = {
5407 .vb_access = ocfs2_journal_access,
5408 };
01225596
TM		 5409
5410 BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
5411
fd68a894 5412 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb, xh,
712e53e4
TM		 5413 xe - xh->xh_entries,
5414 &block_off,
5415 &offset);
5416 if (ret) {
5417 mlog_errno(ret);
5418 goto out;
5419 }
01225596 5420
712e53e4
TM		 5421 base = bucket_block(xs->bucket, block_off);
5422 xv = (struct ocfs2_xattr_value_root *)(base + offset +
5423 OCFS2_XATTR_SIZE(xe->xe_name_len));
01225596 5424
492a8a33
TM		 5425 vb.vb_xv = xv;
5426 vb.vb_bh = xs->bucket->bu_bhs[block_off];
712e53e4 5427 ret = __ocfs2_xattr_set_value_outside(inode, handle,
492a8a33 5428 &vb, val, value_len);
712e53e4
TM		 5429 if (ret)
5430 mlog_errno(ret);
5431out:
5432 return ret;
01225596
TM		 5433}
5434
01225596
TM		 5435static int ocfs2_rm_xattr_cluster(struct inode *inode,
5436 struct buffer_head *root_bh,
5437 u64 blkno,
5438 u32 cpos,
47bca495
TM		 5439 u32 len,
5440 void *para)
01225596
TM		 5441{
5442 int ret;
5443 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5444 struct inode *tl_inode = osb->osb_tl_inode;
5445 handle_t *handle;
5446 struct ocfs2_xattr_block *xb =
5447 (struct ocfs2_xattr_block *)root_bh->b_data;
01225596
TM		 5448 struct ocfs2_alloc_context *meta_ac = NULL;
5449 struct ocfs2_cached_dealloc_ctxt dealloc;
f99b9b7c
JB		 5450 struct ocfs2_extent_tree et;
5451
47bca495 5452 ret = ocfs2_iterate_xattr_buckets(inode, blkno, len,
ce9c5a54 5453 ocfs2_delete_xattr_in_bucket, para);
47bca495
TM		 5454 if (ret) {
5455 mlog_errno(ret);
5456 return ret;
5457 }
5458
5e404e9e 5459 ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
01225596
TM		 5460
5461 ocfs2_init_dealloc_ctxt(&dealloc);
5462
5463 mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
5464 cpos, len, (unsigned long long)blkno);
5465
8cb471e8
JB		 5466 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode), blkno,
5467 len);
01225596 5468
f99b9b7c 5469 ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
01225596
TM		 5470 if (ret) {
5471 mlog_errno(ret);
5472 return ret;
5473 }
5474
5475 mutex_lock(&tl_inode->i_mutex);
5476
5477 if (ocfs2_truncate_log_needs_flush(osb)) {
5478 ret = __ocfs2_flush_truncate_log(osb);
5479 if (ret < 0) {
5480 mlog_errno(ret);
5481 goto out;
5482 }
5483 }
5484
a90714c1 5485 handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
d3264799 5486 if (IS_ERR(handle)) {
01225596
TM		 5487 ret = -ENOMEM;
5488 mlog_errno(ret);
5489 goto out;
5490 }
5491
0cf2f763 5492 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
84008972 5493 OCFS2_JOURNAL_ACCESS_WRITE);
01225596
TM		 5494 if (ret) {
5495 mlog_errno(ret);
5496 goto out_commit;
5497 }
5498
dbdcf6a4 5499 ret = ocfs2_remove_extent(handle, &et, cpos, len, meta_ac,
f99b9b7c 5500 &dealloc);
01225596
TM		 5501 if (ret) {
5502 mlog_errno(ret);
5503 goto out_commit;
5504 }
5505
5506 le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
5507
5508 ret = ocfs2_journal_dirty(handle, root_bh);
5509 if (ret) {
5510 mlog_errno(ret);
5511 goto out_commit;
5512 }
5513
5514 ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
5515 if (ret)
5516 mlog_errno(ret);
5517
5518out_commit:
5519 ocfs2_commit_trans(osb, handle);
5520out:
5521 ocfs2_schedule_truncate_log_flush(osb, 1);
5522
5523 mutex_unlock(&tl_inode->i_mutex);
5524
5525 if (meta_ac)
5526 ocfs2_free_alloc_context(meta_ac);
5527
5528 ocfs2_run_deallocs(osb, &dealloc);
5529
5530 return ret;
5531}
5532
01225596 5533static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
85db90e7 5534 handle_t *handle,
01225596
TM		 5535 struct ocfs2_xattr_search *xs)
5536{
ba937127 5537 struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
01225596
TM		 5538 struct ocfs2_xattr_entry *last = &xh->xh_entries[
5539 le16_to_cpu(xh->xh_count) - 1];
5540 int ret = 0;
5541
ba937127 5542 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
1224be02 5543 OCFS2_JOURNAL_ACCESS_WRITE);
01225596
TM		 5544 if (ret) {
5545 mlog_errno(ret);
85db90e7 5546 return;
01225596
TM		 5547 }
5548
5549 /* Remove the old entry. */
5550 memmove(xs->here, xs->here + 1,
5551 (void *)last - (void *)xs->here);
5552 memset(last, 0, sizeof(struct ocfs2_xattr_entry));
5553 le16_add_cpu(&xh->xh_count, -1);
5554
ba937127 5555 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
01225596
TM		 5556}
5557
5558/*
5559 * Set the xattr name/value in the bucket specified in xs.
5560 *
5561 * As the new value in xi may be stored in the bucket or in an outside cluster,
5562 * we divide the whole process into 3 steps:
5563 * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
5564 * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
5565 * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
5566 * 4. If the clusters for the new outside value can't be allocated, we need
5567 * to free the xattr we allocated in set.
5568 */
5569static int ocfs2_xattr_set_in_bucket(struct inode *inode,
5570 struct ocfs2_xattr_info *xi,
78f30c31
TM		 5571 struct ocfs2_xattr_search *xs,
5572 struct ocfs2_xattr_set_ctxt *ctxt)
01225596 5573{
5a095611 5574 int ret, local = 1;
01225596 5575 size_t value_len;
6b240ff6 5576 char *val = (char *)xi->xi_value;
01225596 5577 struct ocfs2_xattr_entry *xe = xs->here;
6b240ff6 5578 u32 name_hash = ocfs2_xattr_name_hash(inode, xi->xi_name,
18853b95 5579 xi->xi_name_len);
01225596
TM		 5580
5581 if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
5582 /*
5583 * We need to truncate the xattr storage first.
5584 *
5585 * If both the old and new value are stored to
5586 * outside block, we only need to truncate
5587 * the storage and then set the value outside.
5588 *
5589 * If the new value should be stored within block,
5590 * we should free all the outside block first and
5591 * the modification to the xattr block will be done
5592 * by following steps.
5593 */
6b240ff6
JB		 5594 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
5595 value_len = xi->xi_value_len;
01225596
TM		 5596 else
5597 value_len = 0;
5598
5599 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
78f30c31
TM		 5600 value_len,
5601 ctxt);
01225596
TM		 5602 if (ret)
5603 goto out;
5604
5605 if (value_len)
5606 goto set_value_outside;
5607 }
5608
6b240ff6 5609 value_len = xi->xi_value_len;
01225596
TM		 5610 /* So we have to handle the inside block change now. */
5611 if (value_len > OCFS2_XATTR_INLINE_SIZE) {
5612 /*
5613 * If the new value will be stored outside of block,
5614 * initalize a new empty value root and insert it first.
5615 */
5616 local = 0;
6b240ff6
JB		 5617 xi->xi_value = &def_xv;
5618 xi->xi_value_len = OCFS2_XATTR_ROOT_SIZE;
01225596
TM		 5619 }
5620
85db90e7
TM		 5621 ret = ocfs2_xattr_set_entry_in_bucket(inode, ctxt->handle, xi, xs,
5622 name_hash, local);
01225596
TM		 5623 if (ret) {
5624 mlog_errno(ret);
5625 goto out;
5626 }
5627
5a095611
TM		 5628 if (value_len <= OCFS2_XATTR_INLINE_SIZE)
5629 goto out;
01225596 5630
5a095611
TM		 5631 /* allocate the space now for the outside block storage. */
5632 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
78f30c31 5633 value_len, ctxt);
5a095611
TM		 5634 if (ret) {
5635 mlog_errno(ret);
5636
5637 if (xs->not_found) {
5638 /*
5639 * We can't allocate enough clusters for outside
5640 * storage and we have allocated xattr already,
5641 * so need to remove it.
5642 */
85db90e7 5643 ocfs2_xattr_bucket_remove_xs(inode, ctxt->handle, xs);
01225596 5644 }
01225596
TM		 5645 goto out;
5646 }
5647
5648set_value_outside:
85db90e7
TM		 5649 ret = ocfs2_xattr_bucket_set_value_outside(inode, ctxt->handle,
5650 xs, val, value_len);
01225596
TM		 5651out:
5652 return ret;
5653}
5654
80bcaf34
TM		 5655/*
5656 * check whether the xattr bucket is filled up with the same hash value.
5657 * If we want to insert the xattr with the same hash, return -ENOSPC.
5658 * If we want to insert a xattr with different hash value, go ahead
5659 * and ocfs2_divide_xattr_bucket will handle this.
5660 */
01225596 5661static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
80bcaf34
TM		 5662 struct ocfs2_xattr_bucket *bucket,
5663 const char *name)
01225596 5664{
3e632946 5665 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
80bcaf34
TM		 5666 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5667
5668 if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5669 return 0;
01225596
TM		 5670
5671 if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5672 xh->xh_entries[0].xe_name_hash) {
5673 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5674 "hash = %u\n",
9c7759aa 5675 (unsigned long long)bucket_blkno(bucket),
01225596
TM		 5676 le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5677 return -ENOSPC;
5678 }
5679
5680 return 0;
5681}
5682
5683static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5684 struct ocfs2_xattr_info *xi,
78f30c31
TM		 5685 struct ocfs2_xattr_search *xs,
5686 struct ocfs2_xattr_set_ctxt *ctxt)
01225596
TM		 5687{
5688 struct ocfs2_xattr_header *xh;
5689 struct ocfs2_xattr_entry *xe;
5690 u16 count, header_size, xh_free_start;
6dde41d9 5691 int free, max_free, need, old;
18853b95 5692 size_t value_size = 0;
01225596
TM		 5693 size_t blocksize = inode->i_sb->s_blocksize;
5694 int ret, allocation = 0;
01225596 5695
6b240ff6 5696 mlog_entry("Set xattr %s in xattr index block\n", xi->xi_name);
01225596
TM		 5697
5698try_again:
5699 xh = xs->header;
5700 count = le16_to_cpu(xh->xh_count);
5701 xh_free_start = le16_to_cpu(xh->xh_free_start);
5702 header_size = sizeof(struct ocfs2_xattr_header) +
5703 count * sizeof(struct ocfs2_xattr_entry);
4442f518
TY		 5704 max_free = OCFS2_XATTR_BUCKET_SIZE - header_size -
5705 le16_to_cpu(xh->xh_name_value_len) - OCFS2_XATTR_HEADER_GAP;
01225596
TM		 5706
5707 mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
5708 "of %u which exceed block size\n",
ba937127 5709 (unsigned long long)bucket_blkno(xs->bucket),
01225596
TM		 5710 header_size);
5711
6b240ff6 5712 if (xi->xi_value && xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
01225596 5713 value_size = OCFS2_XATTR_ROOT_SIZE;
6b240ff6
JB		 5714 else if (xi->xi_value)
5715 value_size = OCFS2_XATTR_SIZE(xi->xi_value_len);
01225596
TM		 5716
5717 if (xs->not_found)
5718 need = sizeof(struct ocfs2_xattr_entry) +
18853b95 5719 OCFS2_XATTR_SIZE(xi->xi_name_len) + value_size;
01225596 5720 else {
18853b95 5721 need = value_size + OCFS2_XATTR_SIZE(xi->xi_name_len);
01225596
TM		 5722
5723 /*
5724 * We only replace the old value if the new length is smaller
5725 * than the old one. Otherwise we will allocate new space in the
5726 * bucket to store it.
5727 */
5728 xe = xs->here;
5729 if (ocfs2_xattr_is_local(xe))
5730 old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
5731 else
5732 old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
5733
5734 if (old >= value_size)
5735 need = 0;
5736 }
5737
4442f518 5738 free = xh_free_start - header_size - OCFS2_XATTR_HEADER_GAP;
01225596
TM		 5739 /*
5740 * We need to make sure the new name/value pair
5741 * can exist in the same block.
5742 */
5743 if (xh_free_start % blocksize < need)
5744 free -= xh_free_start % blocksize;
5745
5746 mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
5747 "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
5748 " %u\n", xs->not_found,
ba937127 5749 (unsigned long long)bucket_blkno(xs->bucket),
01225596
TM		 5750 free, need, max_free, le16_to_cpu(xh->xh_free_start),
5751 le16_to_cpu(xh->xh_name_value_len));
5752
976331d8
TM		 5753 if (free < need ||
5754 (xs->not_found &&
5755 count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb))) {
01225596
TM		 5756 if (need <= max_free &&
5757 count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
5758 /*
5759 * We can create the space by defragment. Since only the
5760 * name/value will be moved, the xe shouldn't be changed
5761 * in xs.
5762 */
85db90e7
TM		 5763 ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5764 xs->bucket);
01225596
TM		 5765 if (ret) {
5766 mlog_errno(ret);
5767 goto out;
5768 }
5769
5770 xh_free_start = le16_to_cpu(xh->xh_free_start);
4442f518
TY		 5771 free = xh_free_start - header_size
5772 - OCFS2_XATTR_HEADER_GAP;
01225596
TM		 5773 if (xh_free_start % blocksize < need)
5774 free -= xh_free_start % blocksize;
5775
5776 if (free >= need)
5777 goto xattr_set;
5778
5779 mlog(0, "Can't get enough space for xattr insert by "
5780 "defragment. Need %u bytes, but we have %d, so "
5781 "allocate new bucket for it.\n", need, free);
5782 }
5783
5784 /*
5785 * We have to add new buckets or clusters and one
5786 * allocation should leave us enough space for insert.
5787 */
5788 BUG_ON(allocation);
5789
5790 /*
5791 * We do not allow for overlapping ranges between buckets. And
5792 * the maximum number of collisions we will allow for then is
5793 * one bucket's worth, so check it here whether we need to
5794 * add a new bucket for the insert.
5795 */
80bcaf34 5796 ret = ocfs2_check_xattr_bucket_collision(inode,
ba937127 5797 xs->bucket,
6b240ff6 5798 xi->xi_name);
01225596
TM		 5799 if (ret) {
5800 mlog_errno(ret);
5801 goto out;
5802 }
5803
5804 ret = ocfs2_add_new_xattr_bucket(inode,
5805 xs->xattr_bh,
91f2033f 5806 xs->bucket,
78f30c31 5807 ctxt);
01225596
TM		 5808 if (ret) {
5809 mlog_errno(ret);
5810 goto out;
5811 }
5812
91f2033f
JB		 5813 /*
5814 * ocfs2_add_new_xattr_bucket() will have updated
5815 * xs->bucket if it moved, but it will not have updated
5816 * any of the other search fields. Thus, we drop it and
5817 * re-search. Everything should be cached, so it'll be
5818 * quick.
5819 */
ba937127 5820 ocfs2_xattr_bucket_relse(xs->bucket);
01225596 5821 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
6b240ff6
JB		 5822 xi->xi_name_index,
5823 xi->xi_name, xs);
01225596
TM		 5824 if (ret && ret != -ENODATA)
5825 goto out;
5826 xs->not_found = ret;
5827 allocation = 1;
5828 goto try_again;
5829 }
5830
5831xattr_set:
78f30c31 5832 ret = ocfs2_xattr_set_in_bucket(inode, xi, xs, ctxt);
01225596
TM		 5833out:
5834 mlog_exit(ret);
5835 return ret;
5836}
a3944256
TM		 5837
5838static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5839 struct ocfs2_xattr_bucket *bucket,
5840 void *para)
5841{
ce9c5a54 5842 int ret = 0, ref_credits;
3e632946 5843 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
a3944256
TM		 5844 u16 i;
5845 struct ocfs2_xattr_entry *xe;
78f30c31
TM		 5846 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5847 struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
548b0f22
JB		 5848 int credits = ocfs2_remove_extent_credits(osb->sb) +
5849 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
ce9c5a54
TM		 5850 struct ocfs2_xattr_value_root *xv;
5851 struct ocfs2_rm_xattr_bucket_para *args =
5852 (struct ocfs2_rm_xattr_bucket_para *)para;
78f30c31
TM		 5853
5854 ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
a3944256
TM		 5855
5856 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5857 xe = &xh->xh_entries[i];
5858 if (ocfs2_xattr_is_local(xe))
5859 continue;
5860
ce9c5a54
TM		 5861 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket,
5862 i, &xv, NULL);
5863
5864 ret = ocfs2_lock_xattr_remove_allocators(inode, xv,
5865 args->ref_ci,
5866 args->ref_root_bh,
5867 &ctxt.meta_ac,
5868 &ref_credits);
5869
5870 ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
88c3b062
TM		 5871 if (IS_ERR(ctxt.handle)) {
5872 ret = PTR_ERR(ctxt.handle);
5873 mlog_errno(ret);
5874 break;
5875 }
5876
548b0f22 5877 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
78f30c31 5878 i, 0, &ctxt);
88c3b062
TM		 5879
5880 ocfs2_commit_trans(osb, ctxt.handle);
ce9c5a54
TM		 5881 if (ctxt.meta_ac) {
5882 ocfs2_free_alloc_context(ctxt.meta_ac);
5883 ctxt.meta_ac = NULL;
5884 }
a3944256
TM		 5885 if (ret) {
5886 mlog_errno(ret);
5887 break;
5888 }
5889 }
5890
ce9c5a54
TM		 5891 if (ctxt.meta_ac)
5892 ocfs2_free_alloc_context(ctxt.meta_ac);
78f30c31
TM		 5893 ocfs2_schedule_truncate_log_flush(osb, 1);
5894 ocfs2_run_deallocs(osb, &ctxt.dealloc);
a3944256
TM		 5895 return ret;
5896}
5897
492a8a33
TM		 5898/*
5899 * Whenever we modify a xattr value root in the bucket(e.g, CoW
5900 * or change the extent record flag), we need to recalculate
5901 * the metaecc for the whole bucket. So it is done here.
5902 *
5903 * Note:
5904 * We have to give the extra credits for the caller.
5905 */
5906static int ocfs2_xattr_bucket_post_refcount(struct inode *inode,
5907 handle_t *handle,
5908 void *para)
5909{
5910 int ret;
5911 struct ocfs2_xattr_bucket *bucket =
5912 (struct ocfs2_xattr_bucket *)para;
5913
5914 ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
5915 OCFS2_JOURNAL_ACCESS_WRITE);
5916 if (ret) {
5917 mlog_errno(ret);
5918 return ret;
5919 }
5920
5921 ocfs2_xattr_bucket_journal_dirty(handle, bucket);
5922
5923 return 0;
5924}
5925
5926/*
5927 * Special action we need if the xattr value is refcounted.
5928 *
5929 * 1. If the xattr is refcounted, lock the tree.
5930 * 2. CoW the xattr if we are setting the new value and the value
5931 * will be stored outside.
5932 * 3. In other case, decrease_refcount will work for us, so just
5933 * lock the refcount tree, calculate the meta and credits is OK.
5934 *
5935 * We have to do CoW before ocfs2_init_xattr_set_ctxt since
5936 * currently CoW is a completed transaction, while this function
5937 * will also lock the allocators and let us deadlock. So we will
5938 * CoW the whole xattr value.
5939 */
5940static int ocfs2_prepare_refcount_xattr(struct inode *inode,
5941 struct ocfs2_dinode *di,
5942 struct ocfs2_xattr_info *xi,
5943 struct ocfs2_xattr_search *xis,
5944 struct ocfs2_xattr_search *xbs,
5945 struct ocfs2_refcount_tree **ref_tree,
5946 int *meta_add,
5947 int *credits)
5948{
5949 int ret = 0;
5950 struct ocfs2_xattr_block *xb;
5951 struct ocfs2_xattr_entry *xe;
5952 char *base;
5953 u32 p_cluster, num_clusters;
5954 unsigned int ext_flags;
5955 int name_offset, name_len;
5956 struct ocfs2_xattr_value_buf vb;
5957 struct ocfs2_xattr_bucket *bucket = NULL;
5958 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5959 struct ocfs2_post_refcount refcount;
5960 struct ocfs2_post_refcount *p = NULL;
5961 struct buffer_head *ref_root_bh = NULL;
5962
5963 if (!xis->not_found) {
5964 xe = xis->here;
5965 name_offset = le16_to_cpu(xe->xe_name_offset);
5966 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5967 base = xis->base;
5968 vb.vb_bh = xis->inode_bh;
5969 vb.vb_access = ocfs2_journal_access_di;
5970 } else {
5971 int i, block_off = 0;
5972 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
5973 xe = xbs->here;
5974 name_offset = le16_to_cpu(xe->xe_name_offset);
5975 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5976 i = xbs->here - xbs->header->xh_entries;
5977
5978 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
fd68a894 5979 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
492a8a33
TM		 5980 bucket_xh(xbs->bucket),
5981 i, &block_off,
5982 &name_offset);
5983 if (ret) {
5984 mlog_errno(ret);
5985 goto out;
5986 }
5987 base = bucket_block(xbs->bucket, block_off);
5988 vb.vb_bh = xbs->bucket->bu_bhs[block_off];
5989 vb.vb_access = ocfs2_journal_access;
5990
5991 if (ocfs2_meta_ecc(osb)) {
5992 /*create parameters for ocfs2_post_refcount. */
5993 bucket = xbs->bucket;
5994 refcount.credits = bucket->bu_blocks;
5995 refcount.para = bucket;
5996 refcount.func =
5997 ocfs2_xattr_bucket_post_refcount;
5998 p = &refcount;
5999 }
6000 } else {
6001 base = xbs->base;
6002 vb.vb_bh = xbs->xattr_bh;
6003 vb.vb_access = ocfs2_journal_access_xb;
6004 }
6005 }
6006
6007 if (ocfs2_xattr_is_local(xe))
6008 goto out;
6009
6010 vb.vb_xv = (struct ocfs2_xattr_value_root *)
6011 (base + name_offset + name_len);
6012
6013 ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
6014 &num_clusters, &vb.vb_xv->xr_list,
6015 &ext_flags);
6016 if (ret) {
6017 mlog_errno(ret);
6018 goto out;
6019 }
6020
6021 /*
6022 * We just need to check the 1st extent record, since we always
6023 * CoW the whole xattr. So there shouldn't be a xattr with
6024 * some REFCOUNT extent recs after the 1st one.
6025 */
6026 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
6027 goto out;
6028
6029 ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
6030 1, ref_tree, &ref_root_bh);
6031 if (ret) {
6032 mlog_errno(ret);
6033 goto out;
6034 }
6035
6036 /*
6037 * If we are deleting the xattr or the new size will be stored inside,
6038 * cool, leave it there, the xattr truncate process will remove them
6039 * for us(it still needs the refcount tree lock and the meta, credits).
6040 * And the worse case is that every cluster truncate will split the
6041 * refcount tree, and make the original extent become 3. So we will need
6042 * 2 * cluster more extent recs at most.
6043 */
6b240ff6 6044 if (!xi->xi_value || xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE) {
492a8a33
TM		 6045
6046 ret = ocfs2_refcounted_xattr_delete_need(inode,
6047 &(*ref_tree)->rf_ci,
6048 ref_root_bh, vb.vb_xv,
6049 meta_add, credits);
6050 if (ret)
6051 mlog_errno(ret);
6052 goto out;
6053 }
6054
6055 ret = ocfs2_refcount_cow_xattr(inode, di, &vb,
6056 *ref_tree, ref_root_bh, 0,
6057 le32_to_cpu(vb.vb_xv->xr_clusters), p);
6058 if (ret)
6059 mlog_errno(ret);
6060
6061out:
6062 brelse(ref_root_bh);
6063 return ret;
6064}
6065
0129241e
TM		 6066/*
6067 * Add the REFCOUNTED flags for all the extent rec in ocfs2_xattr_value_root.
6068 * The physical clusters will be added to refcount tree.
6069 */
6070static int ocfs2_xattr_value_attach_refcount(struct inode *inode,
6071 struct ocfs2_xattr_value_root *xv,
6072 struct ocfs2_extent_tree *value_et,
6073 struct ocfs2_caching_info *ref_ci,
6074 struct buffer_head *ref_root_bh,
6075 struct ocfs2_cached_dealloc_ctxt *dealloc,
6076 struct ocfs2_post_refcount *refcount)
6077{
6078 int ret = 0;
6079 u32 clusters = le32_to_cpu(xv->xr_clusters);
6080 u32 cpos, p_cluster, num_clusters;
6081 struct ocfs2_extent_list *el = &xv->xr_list;
6082 unsigned int ext_flags;
6083
6084 cpos = 0;
6085 while (cpos < clusters) {
6086 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
6087 &num_clusters, el, &ext_flags);
6088
6089 cpos += num_clusters;
6090 if ((ext_flags & OCFS2_EXT_REFCOUNTED))
6091 continue;
6092
6093 BUG_ON(!p_cluster);
6094
6095 ret = ocfs2_add_refcount_flag(inode, value_et,
6096 ref_ci, ref_root_bh,
6097 cpos - num_clusters,
6098 p_cluster, num_clusters,
6099 dealloc, refcount);
6100 if (ret) {
6101 mlog_errno(ret);
6102 break;
6103 }
6104 }
6105
6106 return ret;
6107}
6108
6109/*
6110 * Given a normal ocfs2_xattr_header, refcount all the entries which
6111 * have value stored outside.
6112 * Used for xattrs stored in inode and ocfs2_xattr_block.
6113 */
6114static int ocfs2_xattr_attach_refcount_normal(struct inode *inode,
6115 struct ocfs2_xattr_value_buf *vb,
6116 struct ocfs2_xattr_header *header,
6117 struct ocfs2_caching_info *ref_ci,
6118 struct buffer_head *ref_root_bh,
6119 struct ocfs2_cached_dealloc_ctxt *dealloc)
6120{
6121
6122 struct ocfs2_xattr_entry *xe;
6123 struct ocfs2_xattr_value_root *xv;
6124 struct ocfs2_extent_tree et;
6125 int i, ret = 0;
6126
6127 for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
6128 xe = &header->xh_entries[i];
6129
6130 if (ocfs2_xattr_is_local(xe))
6131 continue;
6132
6133 xv = (struct ocfs2_xattr_value_root *)((void *)header +
6134 le16_to_cpu(xe->xe_name_offset) +
6135 OCFS2_XATTR_SIZE(xe->xe_name_len));
6136
6137 vb->vb_xv = xv;
6138 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
6139
6140 ret = ocfs2_xattr_value_attach_refcount(inode, xv, &et,
6141 ref_ci, ref_root_bh,
6142 dealloc, NULL);
6143 if (ret) {
6144 mlog_errno(ret);
6145 break;
6146 }
6147 }
6148
6149 return ret;
6150}
6151
6152static int ocfs2_xattr_inline_attach_refcount(struct inode *inode,
6153 struct buffer_head *fe_bh,
6154 struct ocfs2_caching_info *ref_ci,
6155 struct buffer_head *ref_root_bh,
6156 struct ocfs2_cached_dealloc_ctxt *dealloc)
6157{
6158 struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
6159 struct ocfs2_xattr_header *header = (struct ocfs2_xattr_header *)
6160 (fe_bh->b_data + inode->i_sb->s_blocksize -
6161 le16_to_cpu(di->i_xattr_inline_size));
6162 struct ocfs2_xattr_value_buf vb = {
6163 .vb_bh = fe_bh,
6164 .vb_access = ocfs2_journal_access_di,
6165 };
6166
6167 return ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
6168 ref_ci, ref_root_bh, dealloc);
6169}
6170
6171struct ocfs2_xattr_tree_value_refcount_para {
6172 struct ocfs2_caching_info *ref_ci;
6173 struct buffer_head *ref_root_bh;
6174 struct ocfs2_cached_dealloc_ctxt *dealloc;
6175};
6176
6177static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
6178 struct ocfs2_xattr_bucket *bucket,
6179 int offset,
6180 struct ocfs2_xattr_value_root **xv,
6181 struct buffer_head **bh)
6182{
6183 int ret, block_off, name_offset;
6184 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
6185 struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
6186 void *base;
6187
6188 ret = ocfs2_xattr_bucket_get_name_value(sb,
6189 bucket_xh(bucket),
6190 offset,
6191 &block_off,
6192 &name_offset);
6193 if (ret) {
6194 mlog_errno(ret);
6195 goto out;
6196 }
6197
6198 base = bucket_block(bucket, block_off);
6199
6200 *xv = (struct ocfs2_xattr_value_root *)(base + name_offset +
6201 OCFS2_XATTR_SIZE(xe->xe_name_len));
6202
6203 if (bh)
6204 *bh = bucket->bu_bhs[block_off];
6205out:
6206 return ret;
6207}
6208
6209/*
6210 * For a given xattr bucket, refcount all the entries which
6211 * have value stored outside.
6212 */
6213static int ocfs2_xattr_bucket_value_refcount(struct inode *inode,
6214 struct ocfs2_xattr_bucket *bucket,
6215 void *para)
6216{
6217 int i, ret = 0;
6218 struct ocfs2_extent_tree et;
6219 struct ocfs2_xattr_tree_value_refcount_para *ref =
6220 (struct ocfs2_xattr_tree_value_refcount_para *)para;
6221 struct ocfs2_xattr_header *xh =
6222 (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6223 struct ocfs2_xattr_entry *xe;
6224 struct ocfs2_xattr_value_buf vb = {
6225 .vb_access = ocfs2_journal_access,
6226 };
6227 struct ocfs2_post_refcount refcount = {
6228 .credits = bucket->bu_blocks,
6229 .para = bucket,
6230 .func = ocfs2_xattr_bucket_post_refcount,
6231 };
6232 struct ocfs2_post_refcount *p = NULL;
6233
6234 /* We only need post_refcount if we support metaecc. */
6235 if (ocfs2_meta_ecc(OCFS2_SB(inode->i_sb)))
6236 p = &refcount;
6237
6238 mlog(0, "refcount bucket %llu, count = %u\n",
6239 (unsigned long long)bucket_blkno(bucket),
6240 le16_to_cpu(xh->xh_count));
6241 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6242 xe = &xh->xh_entries[i];
6243
6244 if (ocfs2_xattr_is_local(xe))
6245 continue;
6246
6247 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket, i,
6248 &vb.vb_xv, &vb.vb_bh);
6249 if (ret) {
6250 mlog_errno(ret);
6251 break;
6252 }
6253
6254 ocfs2_init_xattr_value_extent_tree(&et,
6255 INODE_CACHE(inode), &vb);
6256
6257 ret = ocfs2_xattr_value_attach_refcount(inode, vb.vb_xv,
6258 &et, ref->ref_ci,
6259 ref->ref_root_bh,
6260 ref->dealloc, p);
6261 if (ret) {
6262 mlog_errno(ret);
6263 break;
6264 }
6265 }
6266
6267 return ret;
6268
6269}
6270
6271static int ocfs2_refcount_xattr_tree_rec(struct inode *inode,
6272 struct buffer_head *root_bh,
6273 u64 blkno, u32 cpos, u32 len, void *para)
6274{
6275 return ocfs2_iterate_xattr_buckets(inode, blkno, len,
6276 ocfs2_xattr_bucket_value_refcount,
6277 para);
6278}
6279
6280static int ocfs2_xattr_block_attach_refcount(struct inode *inode,
6281 struct buffer_head *blk_bh,
6282 struct ocfs2_caching_info *ref_ci,
6283 struct buffer_head *ref_root_bh,
6284 struct ocfs2_cached_dealloc_ctxt *dealloc)
6285{
6286 int ret = 0;
6287 struct ocfs2_xattr_block *xb =
6288 (struct ocfs2_xattr_block *)blk_bh->b_data;
6289
6290 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
6291 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
6292 struct ocfs2_xattr_value_buf vb = {
6293 .vb_bh = blk_bh,
6294 .vb_access = ocfs2_journal_access_xb,
6295 };
6296
6297 ret = ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
6298 ref_ci, ref_root_bh,
6299 dealloc);
6300 } else {
6301 struct ocfs2_xattr_tree_value_refcount_para para = {
6302 .ref_ci = ref_ci,
6303 .ref_root_bh = ref_root_bh,
6304 .dealloc = dealloc,
6305 };
6306
6307 ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
6308 ocfs2_refcount_xattr_tree_rec,
6309 &para);
6310 }
6311
6312 return ret;
6313}
6314
6315int ocfs2_xattr_attach_refcount_tree(struct inode *inode,
6316 struct buffer_head *fe_bh,
6317 struct ocfs2_caching_info *ref_ci,
6318 struct buffer_head *ref_root_bh,
6319 struct ocfs2_cached_dealloc_ctxt *dealloc)
6320{
6321 int ret = 0;
6322 struct ocfs2_inode_info *oi = OCFS2_I(inode);
6323 struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
6324 struct buffer_head *blk_bh = NULL;
6325
6326 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
6327 ret = ocfs2_xattr_inline_attach_refcount(inode, fe_bh,
6328 ref_ci, ref_root_bh,
6329 dealloc);
6330 if (ret) {
6331 mlog_errno(ret);
6332 goto out;
6333 }
6334 }
6335
6336 if (!di->i_xattr_loc)
6337 goto out;
6338
6339 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
6340 &blk_bh);
6341 if (ret < 0) {
6342 mlog_errno(ret);
6343 goto out;
6344 }
6345
6346 ret = ocfs2_xattr_block_attach_refcount(inode, blk_bh, ref_ci,
6347 ref_root_bh, dealloc);
6348 if (ret)
6349 mlog_errno(ret);
6350
6351 brelse(blk_bh);
6352out:
6353
6354 return ret;
6355}
6356
0fe9b66c 6357typedef int (should_xattr_reflinked)(struct ocfs2_xattr_entry *xe);
2999d12f
TM		 6358/*
6359 * Store the information we need in xattr reflink.
6360 * old_bh and new_bh are inode bh for the old and new inode.
6361 */
6362struct ocfs2_xattr_reflink {
6363 struct inode *old_inode;
6364 struct inode *new_inode;
6365 struct buffer_head *old_bh;
6366 struct buffer_head *new_bh;
6367 struct ocfs2_caching_info *ref_ci;
6368 struct buffer_head *ref_root_bh;
6369 struct ocfs2_cached_dealloc_ctxt *dealloc;
0fe9b66c 6370 should_xattr_reflinked *xattr_reflinked;
2999d12f
TM		 6371};
6372
6373/*
6374 * Given a xattr header and xe offset,
6375 * return the proper xv and the corresponding bh.
6376 * xattr in inode, block and xattr tree have different implementaions.
6377 */
6378typedef int (get_xattr_value_root)(struct super_block *sb,
6379 struct buffer_head *bh,
6380 struct ocfs2_xattr_header *xh,
6381 int offset,
6382 struct ocfs2_xattr_value_root **xv,
6383 struct buffer_head **ret_bh,
6384 void *para);
6385
6386/*
6387 * Calculate all the xattr value root metadata stored in this xattr header and
6388 * credits we need if we create them from the scratch.
6389 * We use get_xattr_value_root so that all types of xattr container can use it.
6390 */
6391static int ocfs2_value_metas_in_xattr_header(struct super_block *sb,
6392 struct buffer_head *bh,
6393 struct ocfs2_xattr_header *xh,
6394 int *metas, int *credits,
6395 int *num_recs,
6396 get_xattr_value_root *func,
6397 void *para)
6398{
6399 int i, ret = 0;
6400 struct ocfs2_xattr_value_root *xv;
6401 struct ocfs2_xattr_entry *xe;
6402
6403 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6404 xe = &xh->xh_entries[i];
6405 if (ocfs2_xattr_is_local(xe))
6406 continue;
6407
6408 ret = func(sb, bh, xh, i, &xv, NULL, para);
6409 if (ret) {
6410 mlog_errno(ret);
6411 break;
6412 }
6413
6414 *metas += le16_to_cpu(xv->xr_list.l_tree_depth) *
6415 le16_to_cpu(xv->xr_list.l_next_free_rec);
6416
6417 *credits += ocfs2_calc_extend_credits(sb,
6418 &def_xv.xv.xr_list,
6419 le32_to_cpu(xv->xr_clusters));
6420
6421 /*
6422 * If the value is a tree with depth > 1, We don't go deep
6423 * to the extent block, so just calculate a maximum record num.
6424 */
6425 if (!xv->xr_list.l_tree_depth)
8ff6af88 6426 *num_recs += le16_to_cpu(xv->xr_list.l_next_free_rec);
2999d12f
TM		 6427 else
6428 *num_recs += ocfs2_clusters_for_bytes(sb,
6429 XATTR_SIZE_MAX);
6430 }
6431
6432 return ret;
6433}
6434
6435/* Used by xattr inode and block to return the right xv and buffer_head. */
6436static int ocfs2_get_xattr_value_root(struct super_block *sb,
6437 struct buffer_head *bh,
6438 struct ocfs2_xattr_header *xh,
6439 int offset,
6440 struct ocfs2_xattr_value_root **xv,
6441 struct buffer_head **ret_bh,
6442 void *para)
6443{
6444 struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
6445
6446 *xv = (struct ocfs2_xattr_value_root *)((void *)xh +
6447 le16_to_cpu(xe->xe_name_offset) +
6448 OCFS2_XATTR_SIZE(xe->xe_name_len));
6449
6450 if (ret_bh)
6451 *ret_bh = bh;
6452
6453 return 0;
6454}
6455
6456/*
6457 * Lock the meta_ac and caculate how much credits we need for reflink xattrs.
6458 * It is only used for inline xattr and xattr block.
6459 */
6460static int ocfs2_reflink_lock_xattr_allocators(struct ocfs2_super *osb,
6461 struct ocfs2_xattr_header *xh,
6462 struct buffer_head *ref_root_bh,
6463 int *credits,
6464 struct ocfs2_alloc_context **meta_ac)
6465{
6466 int ret, meta_add = 0, num_recs = 0;
6467 struct ocfs2_refcount_block *rb =
6468 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
6469
6470 *credits = 0;
6471
6472 ret = ocfs2_value_metas_in_xattr_header(osb->sb, NULL, xh,
6473 &meta_add, credits, &num_recs,
6474 ocfs2_get_xattr_value_root,
6475 NULL);
6476 if (ret) {
6477 mlog_errno(ret);
6478 goto out;
6479 }
6480
6481 /*
6482 * We need to add/modify num_recs in refcount tree, so just calculate
6483 * an approximate number we need for refcount tree change.
6484 * Sometimes we need to split the tree, and after split, half recs
6485 * will be moved to the new block, and a new block can only provide
6486 * half number of recs. So we multiple new blocks by 2.
6487 */
6488 num_recs = num_recs / ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6489 meta_add += num_recs;
6490 *credits += num_recs + num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6491 if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6492 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6493 le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6494 else
6495 *credits += 1;
6496
6497 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add, meta_ac);
6498 if (ret)
6499 mlog_errno(ret);
6500
6501out:
6502 return ret;
6503}
6504
6505/*
6506 * Given a xattr header, reflink all the xattrs in this container.
6507 * It can be used for inode, block and bucket.
6508 *
6509 * NOTE:
6510 * Before we call this function, the caller has memcpy the xattr in
6511 * old_xh to the new_xh.
0fe9b66c
TM		 6512 *
6513 * If args.xattr_reflinked is set, call it to decide whether the xe should
6514 * be reflinked or not. If not, remove it from the new xattr header.
2999d12f
TM		 6515 */
6516static int ocfs2_reflink_xattr_header(handle_t *handle,
6517 struct ocfs2_xattr_reflink *args,
6518 struct buffer_head *old_bh,
6519 struct ocfs2_xattr_header *xh,
6520 struct buffer_head *new_bh,
6521 struct ocfs2_xattr_header *new_xh,
6522 struct ocfs2_xattr_value_buf *vb,
6523 struct ocfs2_alloc_context *meta_ac,
6524 get_xattr_value_root *func,
6525 void *para)
6526{
0fe9b66c 6527 int ret = 0, i, j;
2999d12f
TM		 6528 struct super_block *sb = args->old_inode->i_sb;
6529 struct buffer_head *value_bh;
0fe9b66c 6530 struct ocfs2_xattr_entry *xe, *last;
2999d12f
TM		 6531 struct ocfs2_xattr_value_root *xv, *new_xv;
6532 struct ocfs2_extent_tree data_et;
6533 u32 clusters, cpos, p_cluster, num_clusters;
6534 unsigned int ext_flags = 0;
6535
6536 mlog(0, "reflink xattr in container %llu, count = %u\n",
6537 (unsigned long long)old_bh->b_blocknr, le16_to_cpu(xh->xh_count));
0fe9b66c
TM		 6538
6539 last = &new_xh->xh_entries[le16_to_cpu(new_xh->xh_count)];
6540 for (i = 0, j = 0; i < le16_to_cpu(xh->xh_count); i++, j++) {
2999d12f
TM		 6541 xe = &xh->xh_entries[i];
6542
0fe9b66c
TM		 6543 if (args->xattr_reflinked && !args->xattr_reflinked(xe)) {
6544 xe = &new_xh->xh_entries[j];
6545
6546 le16_add_cpu(&new_xh->xh_count, -1);
6547 if (new_xh->xh_count) {
6548 memmove(xe, xe + 1,
6549 (void *)last - (void *)xe);
6550 memset(last, 0,
6551 sizeof(struct ocfs2_xattr_entry));
6552 }
6553
6554 /*
6555 * We don't want j to increase in the next round since
6556 * it is already moved ahead.
6557 */
6558 j--;
6559 continue;
6560 }
6561
2999d12f
TM		 6562 if (ocfs2_xattr_is_local(xe))
6563 continue;
6564
6565 ret = func(sb, old_bh, xh, i, &xv, NULL, para);
6566 if (ret) {
6567 mlog_errno(ret);
6568 break;
6569 }
6570
0fe9b66c 6571 ret = func(sb, new_bh, new_xh, j, &new_xv, &value_bh, para);
2999d12f
TM		 6572 if (ret) {
6573 mlog_errno(ret);
6574 break;
6575 }
6576
6577 /*
6578 * For the xattr which has l_tree_depth = 0, all the extent
6579 * recs have already be copied to the new xh with the
6580 * propriate OCFS2_EXT_REFCOUNTED flag we just need to
6581 * increase the refount count int the refcount tree.
6582 *
6583 * For the xattr which has l_tree_depth > 0, we need
6584 * to initialize it to the empty default value root,
6585 * and then insert the extents one by one.
6586 */
6587 if (xv->xr_list.l_tree_depth) {
6588 memcpy(new_xv, &def_xv, sizeof(def_xv));
6589 vb->vb_xv = new_xv;
6590 vb->vb_bh = value_bh;
6591 ocfs2_init_xattr_value_extent_tree(&data_et,
6592 INODE_CACHE(args->new_inode), vb);
6593 }
6594
6595 clusters = le32_to_cpu(xv->xr_clusters);
6596 cpos = 0;
6597 while (cpos < clusters) {
6598 ret = ocfs2_xattr_get_clusters(args->old_inode,
6599 cpos,
6600 &p_cluster,
6601 &num_clusters,
6602 &xv->xr_list,
6603 &ext_flags);
6604 if (ret) {
6605 mlog_errno(ret);
6606 goto out;
6607 }
6608
6609 BUG_ON(!p_cluster);
6610
6611 if (xv->xr_list.l_tree_depth) {
6612 ret = ocfs2_insert_extent(handle,
6613 &data_et, cpos,
6614 ocfs2_clusters_to_blocks(
6615 args->old_inode->i_sb,
6616 p_cluster),
6617 num_clusters, ext_flags,
6618 meta_ac);
6619 if (ret) {
6620 mlog_errno(ret);
6621 goto out;
6622 }
6623 }
6624
6625 ret = ocfs2_increase_refcount(handle, args->ref_ci,
6626 args->ref_root_bh,
6627 p_cluster, num_clusters,
6628 meta_ac, args->dealloc);
6629 if (ret) {
6630 mlog_errno(ret);
6631 goto out;
6632 }
6633
6634 cpos += num_clusters;
6635 }
6636 }
6637
6638out:
6639 return ret;
6640}
6641
6642static int ocfs2_reflink_xattr_inline(struct ocfs2_xattr_reflink *args)
6643{
6644 int ret = 0, credits = 0;
6645 handle_t *handle;
6646 struct ocfs2_super *osb = OCFS2_SB(args->old_inode->i_sb);
6647 struct ocfs2_dinode *di = (struct ocfs2_dinode *)args->old_bh->b_data;
6648 int inline_size = le16_to_cpu(di->i_xattr_inline_size);
6649 int header_off = osb->sb->s_blocksize - inline_size;
6650 struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)
6651 (args->old_bh->b_data + header_off);
6652 struct ocfs2_xattr_header *new_xh = (struct ocfs2_xattr_header *)
6653 (args->new_bh->b_data + header_off);
6654 struct ocfs2_alloc_context *meta_ac = NULL;
6655 struct ocfs2_inode_info *new_oi;
6656 struct ocfs2_dinode *new_di;
6657 struct ocfs2_xattr_value_buf vb = {
6658 .vb_bh = args->new_bh,
6659 .vb_access = ocfs2_journal_access_di,
6660 };
6661
6662 ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6663 &credits, &meta_ac);
6664 if (ret) {
6665 mlog_errno(ret);
6666 goto out;
6667 }
6668
6669 handle = ocfs2_start_trans(osb, credits);
6670 if (IS_ERR(handle)) {
6671 ret = PTR_ERR(handle);
6672 mlog_errno(ret);
6673 goto out;
6674 }
6675
6676 ret = ocfs2_journal_access_di(handle, INODE_CACHE(args->new_inode),
6677 args->new_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6678 if (ret) {
6679 mlog_errno(ret);
6680 goto out_commit;
6681 }
6682
6683 memcpy(args->new_bh->b_data + header_off,
6684 args->old_bh->b_data + header_off, inline_size);
6685
6686 new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6687 new_di->i_xattr_inline_size = cpu_to_le16(inline_size);
6688
6689 ret = ocfs2_reflink_xattr_header(handle, args, args->old_bh, xh,
6690 args->new_bh, new_xh, &vb, meta_ac,
6691 ocfs2_get_xattr_value_root, NULL);
6692 if (ret) {
6693 mlog_errno(ret);
6694 goto out_commit;
6695 }
6696
6697 new_oi = OCFS2_I(args->new_inode);
6698 spin_lock(&new_oi->ip_lock);
6699 new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL | OCFS2_INLINE_XATTR_FL;
6700 new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6701 spin_unlock(&new_oi->ip_lock);
6702
6703 ocfs2_journal_dirty(handle, args->new_bh);
6704
6705out_commit:
6706 ocfs2_commit_trans(osb, handle);
6707
6708out:
6709 if (meta_ac)
6710 ocfs2_free_alloc_context(meta_ac);
6711 return ret;
6712}
6713
6714static int ocfs2_create_empty_xattr_block(struct inode *inode,
6715 struct buffer_head *fe_bh,
6716 struct buffer_head **ret_bh,
6717 int indexed)
6718{
6719 int ret;
6720 handle_t *handle;
6721 struct ocfs2_alloc_context *meta_ac;
6722 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6723
6724 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
6725 if (ret < 0) {
6726 mlog_errno(ret);
6727 return ret;
6728 }
6729
6730 handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_CREATE_CREDITS);
6731 if (IS_ERR(handle)) {
6732 ret = PTR_ERR(handle);
6733 mlog_errno(ret);
6734 goto out;
6735 }
6736
6737 mlog(0, "create new xattr block for inode %llu, index = %d\n",
6738 (unsigned long long)fe_bh->b_blocknr, indexed);
6739 ret = ocfs2_create_xattr_block(handle, inode, fe_bh,
6740 meta_ac, ret_bh, indexed);
6741 if (ret)
6742 mlog_errno(ret);
6743
6744 ocfs2_commit_trans(osb, handle);
6745out:
6746 ocfs2_free_alloc_context(meta_ac);
6747 return ret;
6748}
6749
6750static int ocfs2_reflink_xattr_block(struct ocfs2_xattr_reflink *args,
6751 struct buffer_head *blk_bh,
6752 struct buffer_head *new_blk_bh)
6753{
6754 int ret = 0, credits = 0;
6755 handle_t *handle;
6756 struct ocfs2_inode_info *new_oi = OCFS2_I(args->new_inode);
6757 struct ocfs2_dinode *new_di;
6758 struct ocfs2_super *osb = OCFS2_SB(args->new_inode->i_sb);
6759 int header_off = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
6760 struct ocfs2_xattr_block *xb =
6761 (struct ocfs2_xattr_block *)blk_bh->b_data;
6762 struct ocfs2_xattr_header *xh = &xb->xb_attrs.xb_header;
6763 struct ocfs2_xattr_block *new_xb =
6764 (struct ocfs2_xattr_block *)new_blk_bh->b_data;
6765 struct ocfs2_xattr_header *new_xh = &new_xb->xb_attrs.xb_header;
6766 struct ocfs2_alloc_context *meta_ac;
6767 struct ocfs2_xattr_value_buf vb = {
6768 .vb_bh = new_blk_bh,
6769 .vb_access = ocfs2_journal_access_xb,
6770 };
6771
6772 ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6773 &credits, &meta_ac);
6774 if (ret) {
6775 mlog_errno(ret);
6776 return ret;
6777 }
6778
6779 /* One more credits in case we need to add xattr flags in new inode. */
6780 handle = ocfs2_start_trans(osb, credits + 1);
6781 if (IS_ERR(handle)) {
6782 ret = PTR_ERR(handle);
6783 mlog_errno(ret);
6784 goto out;
6785 }
6786
6787 if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6788 ret = ocfs2_journal_access_di(handle,
6789 INODE_CACHE(args->new_inode),
6790 args->new_bh,
6791 OCFS2_JOURNAL_ACCESS_WRITE);
6792 if (ret) {
6793 mlog_errno(ret);
6794 goto out_commit;
6795 }
6796 }
6797
6798 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(args->new_inode),
6799 new_blk_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6800 if (ret) {
6801 mlog_errno(ret);
6802 goto out_commit;
6803 }
6804
6805 memcpy(new_blk_bh->b_data + header_off, blk_bh->b_data + header_off,
6806 osb->sb->s_blocksize - header_off);
6807
6808 ret = ocfs2_reflink_xattr_header(handle, args, blk_bh, xh,
6809 new_blk_bh, new_xh, &vb, meta_ac,
6810 ocfs2_get_xattr_value_root, NULL);
6811 if (ret) {
6812 mlog_errno(ret);
6813 goto out_commit;
6814 }
6815
6816 ocfs2_journal_dirty(handle, new_blk_bh);
6817
6818 if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6819 new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6820 spin_lock(&new_oi->ip_lock);
6821 new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
6822 new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6823 spin_unlock(&new_oi->ip_lock);
6824
6825 ocfs2_journal_dirty(handle, args->new_bh);
6826 }
6827
6828out_commit:
6829 ocfs2_commit_trans(osb, handle);
6830
6831out:
6832 ocfs2_free_alloc_context(meta_ac);
6833 return ret;
6834}
6835
6836struct ocfs2_reflink_xattr_tree_args {
6837 struct ocfs2_xattr_reflink *reflink;
6838 struct buffer_head *old_blk_bh;
6839 struct buffer_head *new_blk_bh;
6840 struct ocfs2_xattr_bucket *old_bucket;
6841 struct ocfs2_xattr_bucket *new_bucket;
6842};
6843
6844/*
6845 * NOTE:
6846 * We have to handle the case that both old bucket and new bucket
6847 * will call this function to get the right ret_bh.
6848 * So The caller must give us the right bh.
6849 */
6850static int ocfs2_get_reflink_xattr_value_root(struct super_block *sb,
6851 struct buffer_head *bh,
6852 struct ocfs2_xattr_header *xh,
6853 int offset,
6854 struct ocfs2_xattr_value_root **xv,
6855 struct buffer_head **ret_bh,
6856 void *para)
6857{
6858 struct ocfs2_reflink_xattr_tree_args *args =
6859 (struct ocfs2_reflink_xattr_tree_args *)para;
6860 struct ocfs2_xattr_bucket *bucket;
6861
6862 if (bh == args->old_bucket->bu_bhs[0])
6863 bucket = args->old_bucket;
6864 else
6865 bucket = args->new_bucket;
6866
6867 return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6868 xv, ret_bh);
6869}
6870
6871struct ocfs2_value_tree_metas {
6872 int num_metas;
6873 int credits;
6874 int num_recs;
6875};
6876
6877static int ocfs2_value_tree_metas_in_bucket(struct super_block *sb,
6878 struct buffer_head *bh,
6879 struct ocfs2_xattr_header *xh,
6880 int offset,
6881 struct ocfs2_xattr_value_root **xv,
6882 struct buffer_head **ret_bh,
6883 void *para)
6884{
6885 struct ocfs2_xattr_bucket *bucket =
6886 (struct ocfs2_xattr_bucket *)para;
6887
6888 return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6889 xv, ret_bh);
6890}
6891
6892static int ocfs2_calc_value_tree_metas(struct inode *inode,
6893 struct ocfs2_xattr_bucket *bucket,
6894 void *para)
6895{
6896 struct ocfs2_value_tree_metas *metas =
6897 (struct ocfs2_value_tree_metas *)para;
6898 struct ocfs2_xattr_header *xh =
6899 (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6900
6901 /* Add the credits for this bucket first. */
6902 metas->credits += bucket->bu_blocks;
6903 return ocfs2_value_metas_in_xattr_header(inode->i_sb, bucket->bu_bhs[0],
6904 xh, &metas->num_metas,
6905 &metas->credits, &metas->num_recs,
6906 ocfs2_value_tree_metas_in_bucket,
6907 bucket);
6908}
6909
6910/*
6911 * Given a xattr extent rec starting from blkno and having len clusters,
6912 * iterate all the buckets calculate how much metadata we need for reflinking
6913 * all the ocfs2_xattr_value_root and lock the allocators accordingly.
6914 */
6915static int ocfs2_lock_reflink_xattr_rec_allocators(
6916 struct ocfs2_reflink_xattr_tree_args *args,
6917 struct ocfs2_extent_tree *xt_et,
6918 u64 blkno, u32 len, int *credits,
6919 struct ocfs2_alloc_context **meta_ac,
6920 struct ocfs2_alloc_context **data_ac)
6921{
6922 int ret, num_free_extents;
6923 struct ocfs2_value_tree_metas metas;
6924 struct ocfs2_super *osb = OCFS2_SB(args->reflink->old_inode->i_sb);
6925 struct ocfs2_refcount_block *rb;
6926
6927 memset(&metas, 0, sizeof(metas));
6928
6929 ret = ocfs2_iterate_xattr_buckets(args->reflink->old_inode, blkno, len,
6930 ocfs2_calc_value_tree_metas, &metas);
6931 if (ret) {
6932 mlog_errno(ret);
6933 goto out;
6934 }
6935
6936 *credits = metas.credits;
6937
6938 /*
6939 * Calculate we need for refcount tree change.
6940 *
6941 * We need to add/modify num_recs in refcount tree, so just calculate
6942 * an approximate number we need for refcount tree change.
6943 * Sometimes we need to split the tree, and after split, half recs
6944 * will be moved to the new block, and a new block can only provide
6945 * half number of recs. So we multiple new blocks by 2.
6946 * In the end, we have to add credits for modifying the already
6947 * existed refcount block.
6948 */
6949 rb = (struct ocfs2_refcount_block *)args->reflink->ref_root_bh->b_data;
6950 metas.num_recs =
6951 (metas.num_recs + ocfs2_refcount_recs_per_rb(osb->sb) - 1) /
6952 ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6953 metas.num_metas += metas.num_recs;
6954 *credits += metas.num_recs +
6955 metas.num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6956 if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6957 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6958 le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6959 else
6960 *credits += 1;
6961
6962 /* count in the xattr tree change. */
6963 num_free_extents = ocfs2_num_free_extents(osb, xt_et);
6964 if (num_free_extents < 0) {
6965 ret = num_free_extents;
6966 mlog_errno(ret);
6967 goto out;
6968 }
6969
6970 if (num_free_extents < len)
6971 metas.num_metas += ocfs2_extend_meta_needed(xt_et->et_root_el);
6972
6973 *credits += ocfs2_calc_extend_credits(osb->sb,
6974 xt_et->et_root_el, len);
6975
6976 if (metas.num_metas) {
6977 ret = ocfs2_reserve_new_metadata_blocks(osb, metas.num_metas,
6978 meta_ac);
6979 if (ret) {
6980 mlog_errno(ret);
6981 goto out;
6982 }
6983 }
6984
6985 if (len) {
6986 ret = ocfs2_reserve_clusters(osb, len, data_ac);
6987 if (ret)
6988 mlog_errno(ret);
6989 }
6990out:
6991 if (ret) {
6992 if (*meta_ac) {
6993 ocfs2_free_alloc_context(*meta_ac);
6994 meta_ac = NULL;
6995 }
6996 }
6997
6998 return ret;
6999}
7000
7001static int ocfs2_reflink_xattr_buckets(handle_t *handle,
7002 u64 blkno, u64 new_blkno, u32 clusters,
7003 struct ocfs2_alloc_context *meta_ac,
7004 struct ocfs2_alloc_context *data_ac,
7005 struct ocfs2_reflink_xattr_tree_args *args)
7006{
7007 int i, j, ret = 0;
7008 struct super_block *sb = args->reflink->old_inode->i_sb;
7009 u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
7010 u32 num_buckets = clusters * bpc;
7011 int bpb = args->old_bucket->bu_blocks;
7012 struct ocfs2_xattr_value_buf vb = {
7013 .vb_access = ocfs2_journal_access,
7014 };
7015
7016 for (i = 0; i < num_buckets; i++, blkno += bpb, new_blkno += bpb) {
7017 ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
7018 if (ret) {
7019 mlog_errno(ret);
7020 break;
7021 }
7022
7023 ret = ocfs2_init_xattr_bucket(args->new_bucket, new_blkno);
7024 if (ret) {
7025 mlog_errno(ret);
7026 break;
7027 }
7028
7029 /*
7030 * The real bucket num in this series of blocks is stored
7031 * in the 1st bucket.
7032 */
7033 if (i == 0)
7034 num_buckets = le16_to_cpu(
7035 bucket_xh(args->old_bucket)->xh_num_buckets);
7036
7037 ret = ocfs2_xattr_bucket_journal_access(handle,
7038 args->new_bucket,
7039 OCFS2_JOURNAL_ACCESS_CREATE);
7040 if (ret) {
7041 mlog_errno(ret);
7042 break;
7043 }
7044
7045 for (j = 0; j < bpb; j++)
7046 memcpy(bucket_block(args->new_bucket, j),
7047 bucket_block(args->old_bucket, j),
7048 sb->s_blocksize);
7049
7050 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
7051
7052 ret = ocfs2_reflink_xattr_header(handle, args->reflink,
7053 args->old_bucket->bu_bhs[0],
7054 bucket_xh(args->old_bucket),
7055 args->new_bucket->bu_bhs[0],
7056 bucket_xh(args->new_bucket),
7057 &vb, meta_ac,
7058 ocfs2_get_reflink_xattr_value_root,
7059 args);
7060 if (ret) {
7061 mlog_errno(ret);
7062 break;
7063 }
7064
7065 /*
7066 * Re-access and dirty the bucket to calculate metaecc.
7067 * Because we may extend the transaction in reflink_xattr_header
7068 * which will let the already accessed block gone.
7069 */
7070 ret = ocfs2_xattr_bucket_journal_access(handle,
7071 args->new_bucket,
7072 OCFS2_JOURNAL_ACCESS_WRITE);
7073 if (ret) {
7074 mlog_errno(ret);
7075 break;
7076 }
7077
7078 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
7079 ocfs2_xattr_bucket_relse(args->old_bucket);
7080 ocfs2_xattr_bucket_relse(args->new_bucket);
7081 }
7082
7083 ocfs2_xattr_bucket_relse(args->old_bucket);
7084 ocfs2_xattr_bucket_relse(args->new_bucket);
7085 return ret;
7086}
7087/*
7088 * Create the same xattr extent record in the new inode's xattr tree.
7089 */
7090static int ocfs2_reflink_xattr_rec(struct inode *inode,
7091 struct buffer_head *root_bh,
7092 u64 blkno,
7093 u32 cpos,
7094 u32 len,
7095 void *para)
7096{
7097 int ret, credits = 0;
7098 u32 p_cluster, num_clusters;
7099 u64 new_blkno;
7100 handle_t *handle;
7101 struct ocfs2_reflink_xattr_tree_args *args =
7102 (struct ocfs2_reflink_xattr_tree_args *)para;
7103 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7104 struct ocfs2_alloc_context *meta_ac = NULL;
7105 struct ocfs2_alloc_context *data_ac = NULL;
7106 struct ocfs2_extent_tree et;
7107
7108 ocfs2_init_xattr_tree_extent_tree(&et,
7109 INODE_CACHE(args->reflink->new_inode),
7110 args->new_blk_bh);
7111
7112 ret = ocfs2_lock_reflink_xattr_rec_allocators(args, &et, blkno,
7113 len, &credits,
7114 &meta_ac, &data_ac);
7115 if (ret) {
7116 mlog_errno(ret);
7117 goto out;
7118 }
7119
7120 handle = ocfs2_start_trans(osb, credits);
7121 if (IS_ERR(handle)) {
7122 ret = PTR_ERR(handle);
7123 mlog_errno(ret);
7124 goto out;
7125 }
7126
7127 ret = ocfs2_claim_clusters(osb, handle, data_ac,
7128 len, &p_cluster, &num_clusters);
7129 if (ret) {
7130 mlog_errno(ret);
7131 goto out_commit;
7132 }
7133
7134 new_blkno = ocfs2_clusters_to_blocks(osb->sb, p_cluster);
7135
7136 mlog(0, "reflink xattr buckets %llu to %llu, len %u\n",
7137 (unsigned long long)blkno, (unsigned long long)new_blkno, len);
7138 ret = ocfs2_reflink_xattr_buckets(handle, blkno, new_blkno, len,
7139 meta_ac, data_ac, args);
7140 if (ret) {
7141 mlog_errno(ret);
7142 goto out_commit;
7143 }
7144
7145 mlog(0, "insert new xattr extent rec start %llu len %u to %u\n",
7146 (unsigned long long)new_blkno, len, cpos);
7147 ret = ocfs2_insert_extent(handle, &et, cpos, new_blkno,
7148 len, 0, meta_ac);
7149 if (ret)
7150 mlog_errno(ret);
7151
7152out_commit:
7153 ocfs2_commit_trans(osb, handle);
7154
7155out:
7156 if (meta_ac)
7157 ocfs2_free_alloc_context(meta_ac);
7158 if (data_ac)
7159 ocfs2_free_alloc_context(data_ac);
7160 return ret;
7161}
7162
7163/*
7164 * Create reflinked xattr buckets.
7165 * We will add bucket one by one, and refcount all the xattrs in the bucket
7166 * if they are stored outside.
7167 */
7168static int ocfs2_reflink_xattr_tree(struct ocfs2_xattr_reflink *args,
7169 struct buffer_head *blk_bh,
7170 struct buffer_head *new_blk_bh)
7171{
7172 int ret;
7173 struct ocfs2_reflink_xattr_tree_args para;
7174
7175 memset(&para, 0, sizeof(para));
7176 para.reflink = args;
7177 para.old_blk_bh = blk_bh;
7178 para.new_blk_bh = new_blk_bh;
7179
7180 para.old_bucket = ocfs2_xattr_bucket_new(args->old_inode);
7181 if (!para.old_bucket) {
7182 mlog_errno(-ENOMEM);
7183 return -ENOMEM;
7184 }
7185
7186 para.new_bucket = ocfs2_xattr_bucket_new(args->new_inode);
7187 if (!para.new_bucket) {
7188 ret = -ENOMEM;
7189 mlog_errno(ret);
7190 goto out;
7191 }
7192
7193 ret = ocfs2_iterate_xattr_index_block(args->old_inode, blk_bh,
7194 ocfs2_reflink_xattr_rec,
7195 &para);
7196 if (ret)
7197 mlog_errno(ret);
7198
7199out:
7200 ocfs2_xattr_bucket_free(para.old_bucket);
7201 ocfs2_xattr_bucket_free(para.new_bucket);
7202 return ret;
7203}
7204
7205static int ocfs2_reflink_xattr_in_block(struct ocfs2_xattr_reflink *args,
7206 struct buffer_head *blk_bh)
7207{
7208 int ret, indexed = 0;
7209 struct buffer_head *new_blk_bh = NULL;
7210 struct ocfs2_xattr_block *xb =
7211 (struct ocfs2_xattr_block *)blk_bh->b_data;
7212
7213
7214 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)
7215 indexed = 1;
7216
7217 ret = ocfs2_create_empty_xattr_block(args->new_inode, args->new_bh,
7218 &new_blk_bh, indexed);
7219 if (ret) {
7220 mlog_errno(ret);
7221 goto out;
7222 }
7223
7224 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED))
7225 ret = ocfs2_reflink_xattr_block(args, blk_bh, new_blk_bh);
7226 else
7227 ret = ocfs2_reflink_xattr_tree(args, blk_bh, new_blk_bh);
7228 if (ret)
7229 mlog_errno(ret);
7230
7231out:
7232 brelse(new_blk_bh);
7233 return ret;
7234}
7235
0fe9b66c
TM		 7236static int ocfs2_reflink_xattr_no_security(struct ocfs2_xattr_entry *xe)
7237{
7238 int type = ocfs2_xattr_get_type(xe);
7239
7240 return type != OCFS2_XATTR_INDEX_SECURITY &&
7241 type != OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS &&
7242 type != OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT;
7243}
7244
2999d12f
TM		 7245int ocfs2_reflink_xattrs(struct inode *old_inode,
7246 struct buffer_head *old_bh,
7247 struct inode *new_inode,
0fe9b66c
TM		 7248 struct buffer_head *new_bh,
7249 bool preserve_security)
2999d12f
TM		 7250{
7251 int ret;
7252 struct ocfs2_xattr_reflink args;
7253 struct ocfs2_inode_info *oi = OCFS2_I(old_inode);
7254 struct ocfs2_dinode *di = (struct ocfs2_dinode *)old_bh->b_data;
7255 struct buffer_head *blk_bh = NULL;
7256 struct ocfs2_cached_dealloc_ctxt dealloc;
7257 struct ocfs2_refcount_tree *ref_tree;
7258 struct buffer_head *ref_root_bh = NULL;
7259
7260 ret = ocfs2_lock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7261 le64_to_cpu(di->i_refcount_loc),
7262 1, &ref_tree, &ref_root_bh);
7263 if (ret) {
7264 mlog_errno(ret);
7265 goto out;
7266 }
7267
7268 ocfs2_init_dealloc_ctxt(&dealloc);
7269
7270 args.old_inode = old_inode;
7271 args.new_inode = new_inode;
7272 args.old_bh = old_bh;
7273 args.new_bh = new_bh;
7274 args.ref_ci = &ref_tree->rf_ci;
7275 args.ref_root_bh = ref_root_bh;
7276 args.dealloc = &dealloc;
0fe9b66c
TM		 7277 if (preserve_security)
7278 args.xattr_reflinked = NULL;
7279 else
7280 args.xattr_reflinked = ocfs2_reflink_xattr_no_security;
2999d12f
TM		 7281
7282 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
7283 ret = ocfs2_reflink_xattr_inline(&args);
7284 if (ret) {
7285 mlog_errno(ret);
7286 goto out_unlock;
7287 }
7288 }
7289
7290 if (!di->i_xattr_loc)
7291 goto out_unlock;
7292
7293 ret = ocfs2_read_xattr_block(old_inode, le64_to_cpu(di->i_xattr_loc),
7294 &blk_bh);
7295 if (ret < 0) {
7296 mlog_errno(ret);
7297 goto out_unlock;
7298 }
7299
7300 ret = ocfs2_reflink_xattr_in_block(&args, blk_bh);
7301 if (ret)
7302 mlog_errno(ret);
7303
7304 brelse(blk_bh);
7305
7306out_unlock:
7307 ocfs2_unlock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7308 ref_tree, 1);
7309 brelse(ref_root_bh);
7310
7311 if (ocfs2_dealloc_has_cluster(&dealloc)) {
7312 ocfs2_schedule_truncate_log_flush(OCFS2_SB(old_inode->i_sb), 1);
7313 ocfs2_run_deallocs(OCFS2_SB(old_inode->i_sb), &dealloc);
7314 }
7315
7316out:
7317 return ret;
7318}
7319
0fe9b66c
TM		 7320/*
7321 * Initialize security and acl for a already created inode.
7322 * Used for reflink a non-preserve-security file.
7323 *
7324 * It uses common api like ocfs2_xattr_set, so the caller
7325 * must not hold any lock expect i_mutex.
7326 */
7327int ocfs2_init_security_and_acl(struct inode *dir,
7328 struct inode *inode)
7329{
7330 int ret = 0;
7331 struct buffer_head *dir_bh = NULL;
7332 struct ocfs2_security_xattr_info si = {
7333 .enable = 1,
7334 };
7335
7336 ret = ocfs2_init_security_get(inode, dir, &si);
7337 if (!ret) {
431547b3
CH		 7338 ret = ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY,
7339 si.name, si.value, si.value_len,
7340 XATTR_CREATE);
0fe9b66c
TM		 7341 if (ret) {
7342 mlog_errno(ret);
7343 goto leave;
7344 }
7345 } else if (ret != -EOPNOTSUPP) {
7346 mlog_errno(ret);
7347 goto leave;
7348 }
7349
7350 ret = ocfs2_inode_lock(dir, &dir_bh, 0);
7351 if (ret) {
7352 mlog_errno(ret);
7353 goto leave;
7354 }
7355
7356 ret = ocfs2_init_acl(NULL, inode, dir, NULL, dir_bh, NULL, NULL);
7357 if (ret)
7358 mlog_errno(ret);
7359
7360 ocfs2_inode_unlock(dir, 0);
7361 brelse(dir_bh);
7362leave:
7363 return ret;
7364}
923f7f31
TY		 7365/*
7366 * 'security' attributes support
7367 */
431547b3 7368static size_t ocfs2_xattr_security_list(struct dentry *dentry, char *list,
923f7f31 7369 size_t list_size, const char *name,
431547b3 7370 size_t name_len, int type)
923f7f31
TY		 7371{
7372 const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
7373 const size_t total_len = prefix_len + name_len + 1;
7374
7375 if (list && total_len <= list_size) {
7376 memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
7377 memcpy(list + prefix_len, name, name_len);
7378 list[prefix_len + name_len] = '\0';
7379 }
7380 return total_len;
7381}
7382
431547b3
CH		 7383static int ocfs2_xattr_security_get(struct dentry *dentry, const char *name,
7384 void *buffer, size_t size, int type)
923f7f31
TY		 7385{
7386 if (strcmp(name, "") == 0)
7387 return -EINVAL;
431547b3
CH		 7388 return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_SECURITY,
7389 name, buffer, size);
923f7f31
TY		 7390}
7391
431547b3
CH		 7392static int ocfs2_xattr_security_set(struct dentry *dentry, const char *name,
7393 const void *value, size_t size, int flags, int type)
923f7f31
TY		 7394{
7395 if (strcmp(name, "") == 0)
7396 return -EINVAL;
7397
431547b3
CH		 7398 return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_SECURITY,
7399 name, value, size, flags);
923f7f31
TY		 7400}
7401
534eaddd
TY		 7402int ocfs2_init_security_get(struct inode *inode,
7403 struct inode *dir,
7404 struct ocfs2_security_xattr_info *si)
7405{
38d59ef6
TY		 7406 /* check whether ocfs2 support feature xattr */
7407 if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
7408 return -EOPNOTSUPP;
534eaddd
TY		 7409 return security_inode_init_security(inode, dir, &si->name, &si->value,
7410 &si->value_len);
7411}
7412
7413int ocfs2_init_security_set(handle_t *handle,
7414 struct inode *inode,
7415 struct buffer_head *di_bh,
7416 struct ocfs2_security_xattr_info *si,
7417 struct ocfs2_alloc_context *xattr_ac,
7418 struct ocfs2_alloc_context *data_ac)
7419{
7420 return ocfs2_xattr_set_handle(handle, inode, di_bh,
7421 OCFS2_XATTR_INDEX_SECURITY,
7422 si->name, si->value, si->value_len, 0,
7423 xattr_ac, data_ac);
7424}
7425
923f7f31
TY		 7426struct xattr_handler ocfs2_xattr_security_handler = {
7427 .prefix = XATTR_SECURITY_PREFIX,
7428 .list = ocfs2_xattr_security_list,
7429 .get = ocfs2_xattr_security_get,
7430 .set = ocfs2_xattr_security_set,
7431};
7432
99219aea
MF		 7433/*
7434 * 'trusted' attributes support
7435 */
431547b3 7436static size_t ocfs2_xattr_trusted_list(struct dentry *dentry, char *list,
99219aea 7437 size_t list_size, const char *name,
431547b3 7438 size_t name_len, int type)
99219aea 7439{
ceb1eba3 7440 const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
99219aea
MF		 7441 const size_t total_len = prefix_len + name_len + 1;
7442
7443 if (list && total_len <= list_size) {
7444 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
7445 memcpy(list + prefix_len, name, name_len);
7446 list[prefix_len + name_len] = '\0';
7447 }
7448 return total_len;
7449}
7450
431547b3
CH		 7451static int ocfs2_xattr_trusted_get(struct dentry *dentry, const char *name,
7452 void *buffer, size_t size, int type)
99219aea
MF		 7453{
7454 if (strcmp(name, "") == 0)
7455 return -EINVAL;
431547b3
CH		 7456 return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_TRUSTED,
7457 name, buffer, size);
99219aea
MF		 7458}
7459
431547b3
CH		 7460static int ocfs2_xattr_trusted_set(struct dentry *dentry, const char *name,
7461 const void *value, size_t size, int flags, int type)
99219aea
MF		 7462{
7463 if (strcmp(name, "") == 0)
7464 return -EINVAL;
7465
431547b3
CH		 7466 return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_TRUSTED,
7467 name, value, size, flags);
99219aea
MF		 7468}
7469
7470struct xattr_handler ocfs2_xattr_trusted_handler = {
7471 .prefix = XATTR_TRUSTED_PREFIX,
7472 .list = ocfs2_xattr_trusted_list,
7473 .get = ocfs2_xattr_trusted_get,
7474 .set = ocfs2_xattr_trusted_set,
7475};
7476
99219aea
MF		 7477/*
7478 * 'user' attributes support
7479 */
431547b3 7480static size_t ocfs2_xattr_user_list(struct dentry *dentry, char *list,
99219aea 7481 size_t list_size, const char *name,
431547b3 7482 size_t name_len, int type)
99219aea 7483{
ceb1eba3 7484 const size_t prefix_len = XATTR_USER_PREFIX_LEN;
99219aea 7485 const size_t total_len = prefix_len + name_len + 1;
431547b3 7486 struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
99219aea
MF		 7487
7488 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7489 return 0;
7490
7491 if (list && total_len <= list_size) {
7492 memcpy(list, XATTR_USER_PREFIX, prefix_len);
7493 memcpy(list + prefix_len, name, name_len);
7494 list[prefix_len + name_len] = '\0';
7495 }
7496 return total_len;
7497}
7498
431547b3
CH		 7499static int ocfs2_xattr_user_get(struct dentry *dentry, const char *name,
7500 void *buffer, size_t size, int type)
99219aea 7501{
431547b3 7502 struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
99219aea
MF		 7503
7504 if (strcmp(name, "") == 0)
7505 return -EINVAL;
7506 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7507 return -EOPNOTSUPP;
431547b3 7508 return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_USER, name,
99219aea
MF		 7509 buffer, size);
7510}
7511
431547b3
CH		 7512static int ocfs2_xattr_user_set(struct dentry *dentry, const char *name,
7513 const void *value, size_t size, int flags, int type)
99219aea 7514{
431547b3 7515 struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
99219aea
MF		 7516
7517 if (strcmp(name, "") == 0)
7518 return -EINVAL;
7519 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7520 return -EOPNOTSUPP;
7521
431547b3
CH		 7522 return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_USER,
7523 name, value, size, flags);
99219aea
MF		 7524}
7525
7526struct xattr_handler ocfs2_xattr_user_handler = {
7527 .prefix = XATTR_USER_PREFIX,
7528 .list = ocfs2_xattr_user_list,
7529 .get = ocfs2_xattr_user_get,
7530 .set = ocfs2_xattr_user_set,
7531};
Clone of davem's net-next-2.6 tree