?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm
[linux-stable.git] / fs / ocfs2 / xattr.c
blob73a6f6fd8a8ef578390986081bacc6a83c2b68d9
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * xattr.c
4 *
5 * Copyright (C) 2004, 2008 Oracle. All rights reserved.
6 *
7 * CREDITS:
8 * Lots of code in this file is copy from linux/fs/ext3/xattr.c.
9 * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
10 */
11
12 #include <linux/capability.h>
13 #include <linux/fs.h>
14 #include <linux/types.h>
15 #include <linux/slab.h>
16 #include <linux/highmem.h>
17 #include <linux/pagemap.h>
18 #include <linux/uio.h>
19 #include <linux/sched.h>
20 #include <linux/splice.h>
21 #include <linux/mount.h>
22 #include <linux/writeback.h>
23 #include <linux/falloc.h>
24 #include <linux/sort.h>
25 #include <linux/init.h>
26 #include <linux/module.h>
27 #include <linux/string.h>
28 #include <linux/security.h>
29
30 #include <cluster/masklog.h>
31
32 #include "ocfs2.h"
33 #include "alloc.h"
34 #include "blockcheck.h"
35 #include "dlmglue.h"
36 #include "file.h"
37 #include "symlink.h"
38 #include "sysfile.h"
39 #include "inode.h"
40 #include "journal.h"
41 #include "ocfs2_fs.h"
42 #include "suballoc.h"
43 #include "uptodate.h"
44 #include "buffer_head_io.h"
45 #include "super.h"
46 #include "xattr.h"
47 #include "refcounttree.h"
48 #include "acl.h"
49 #include "ocfs2_trace.h"
50
51 struct ocfs2_xattr_def_value_root {
52 struct ocfs2_xattr_value_root xv;
53 struct ocfs2_extent_rec er;
54 };
55
56 struct ocfs2_xattr_bucket {
57 /* The inode these xattrs are associated with */
58 struct inode *bu_inode;
59
60 /* The actual buffers that make up the bucket */
61 struct buffer_head *bu_bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
62
63 /* How many blocks make up one bucket for this filesystem */
64 int bu_blocks;
65 };
66
67 struct ocfs2_xattr_set_ctxt {
68 handle_t *handle;
69 struct ocfs2_alloc_context *meta_ac;
70 struct ocfs2_alloc_context *data_ac;
71 struct ocfs2_cached_dealloc_ctxt dealloc;
72 int set_abort;
73 };
74
75 #define OCFS2_XATTR_ROOT_SIZE (sizeof(struct ocfs2_xattr_def_value_root))
76 #define OCFS2_XATTR_INLINE_SIZE 80
77 #define OCFS2_XATTR_HEADER_GAP 4
78 #define OCFS2_XATTR_FREE_IN_IBODY (OCFS2_MIN_XATTR_INLINE_SIZE \
79 - sizeof(struct ocfs2_xattr_header) \
80 - OCFS2_XATTR_HEADER_GAP)
81 #define OCFS2_XATTR_FREE_IN_BLOCK(ptr) ((ptr)->i_sb->s_blocksize \
82 - sizeof(struct ocfs2_xattr_block) \
83 - sizeof(struct ocfs2_xattr_header) \
84 - OCFS2_XATTR_HEADER_GAP)
85
86 static struct ocfs2_xattr_def_value_root def_xv = {
87 .xv.xr_list.l_count = cpu_to_le16(1),
88 };
89
90 const struct xattr_handler * const ocfs2_xattr_handlers[] = {
91 &ocfs2_xattr_user_handler,
92 &ocfs2_xattr_trusted_handler,
93 &ocfs2_xattr_security_handler,
94 NULL
95 };
96
97 static const struct xattr_handler * const ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
98 [OCFS2_XATTR_INDEX_USER] = &ocfs2_xattr_user_handler,
99 [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS] = &nop_posix_acl_access,
100 [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT] = &nop_posix_acl_default,
101 [OCFS2_XATTR_INDEX_TRUSTED] = &ocfs2_xattr_trusted_handler,
102 [OCFS2_XATTR_INDEX_SECURITY] = &ocfs2_xattr_security_handler,
103 };
104
105 struct ocfs2_xattr_info {
106 int xi_name_index;
107 const char *xi_name;
108 int xi_name_len;
109 const void *xi_value;
110 size_t xi_value_len;
111 };
112
113 struct ocfs2_xattr_search {
114 struct buffer_head *inode_bh;
115 /*
116 * xattr_bh point to the block buffer head which has extended attribute
117 * when extended attribute in inode, xattr_bh is equal to inode_bh.
118 */
119 struct buffer_head *xattr_bh;
120 struct ocfs2_xattr_header *header;
121 struct ocfs2_xattr_bucket *bucket;
122 void *base;
123 void *end;
124 struct ocfs2_xattr_entry *here;
125 int not_found;
126 };
127
128 /* Operations on struct ocfs2_xa_entry */
129 struct ocfs2_xa_loc;
130 struct ocfs2_xa_loc_operations {
131 /*
132 * Journal functions
133 */
134 int (*xlo_journal_access)(handle_t *handle, struct ocfs2_xa_loc *loc,
135 int type);
136 void (*xlo_journal_dirty)(handle_t *handle, struct ocfs2_xa_loc *loc);
137
138 /*
139 * Return a pointer to the appropriate buffer in loc->xl_storage
140 * at the given offset from loc->xl_header.
141 */
142 void *(*xlo_offset_pointer)(struct ocfs2_xa_loc *loc, int offset);
143
144 /* Can we reuse the existing entry for the new value? */
145 int (*xlo_can_reuse)(struct ocfs2_xa_loc *loc,
146 struct ocfs2_xattr_info *xi);
147
148 /* How much space is needed for the new value? */
149 int (*xlo_check_space)(struct ocfs2_xa_loc *loc,
150 struct ocfs2_xattr_info *xi);
151
152 /*
153 * Return the offset of the first name+value pair. This is
154 * the start of our downward-filling free space.
155 */
156 int (*xlo_get_free_start)(struct ocfs2_xa_loc *loc);
157
158 /*
159 * Remove the name+value at this location. Do whatever is
160 * appropriate with the remaining name+value pairs.
161 */
162 void (*xlo_wipe_namevalue)(struct ocfs2_xa_loc *loc);
163
164 /* Fill xl_entry with a new entry */
165 void (*xlo_add_entry)(struct ocfs2_xa_loc *loc, u32 name_hash);
166
167 /* Add name+value storage to an entry */
168 void (*xlo_add_namevalue)(struct ocfs2_xa_loc *loc, int size);
169
170 /*
171 * Initialize the value buf's access and bh fields for this entry.
172 * ocfs2_xa_fill_value_buf() will handle the xv pointer.
173 */
174 void (*xlo_fill_value_buf)(struct ocfs2_xa_loc *loc,
175 struct ocfs2_xattr_value_buf *vb);
176 };
177
178 /*
179 * Describes an xattr entry location. This is a memory structure
180 * tracking the on-disk structure.
181 */
182 struct ocfs2_xa_loc {
183 /* This xattr belongs to this inode */
184 struct inode *xl_inode;
185
186 /* The ocfs2_xattr_header inside the on-disk storage. Not NULL. */
187 struct ocfs2_xattr_header *xl_header;
188
189 /* Bytes from xl_header to the end of the storage */
190 int xl_size;
191
192 /*
193 * The ocfs2_xattr_entry this location describes. If this is
194 * NULL, this location describes the on-disk structure where it
195 * would have been.
196 */
197 struct ocfs2_xattr_entry *xl_entry;
198
199 /*
200 * Internal housekeeping
201 */
202
203 /* Buffer(s) containing this entry */
204 void *xl_storage;
205
206 /* Operations on the storage backing this location */
207 const struct ocfs2_xa_loc_operations *xl_ops;
208 };
209
210 /*
211 * Convenience functions to calculate how much space is needed for a
212 * given name+value pair
213 */
214 static int namevalue_size(int name_len, uint64_t value_len)
215 {
216 if (value_len > OCFS2_XATTR_INLINE_SIZE)
217 return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
218 else
219 return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
220 }
221
222 static int namevalue_size_xi(struct ocfs2_xattr_info *xi)
223 {
224 return namevalue_size(xi->xi_name_len, xi->xi_value_len);
225 }
226
227 static int namevalue_size_xe(struct ocfs2_xattr_entry *xe)
228 {
229 u64 value_len = le64_to_cpu(xe->xe_value_size);
230
231 BUG_ON((value_len > OCFS2_XATTR_INLINE_SIZE) &&
232 ocfs2_xattr_is_local(xe));
233 return namevalue_size(xe->xe_name_len, value_len);
234 }
235
236
237 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
238 struct ocfs2_xattr_header *xh,
239 int index,
240 int *block_off,
241 int *new_offset);
242
243 static int ocfs2_xattr_block_find(struct inode *inode,
244 int name_index,
245 const char *name,
246 struct ocfs2_xattr_search *xs);
247 static int ocfs2_xattr_index_block_find(struct inode *inode,
248 struct buffer_head *root_bh,
249 int name_index,
250 const char *name,
251 struct ocfs2_xattr_search *xs);
252
253 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
254 struct buffer_head *blk_bh,
255 char *buffer,
256 size_t buffer_size);
257
258 static int ocfs2_xattr_create_index_block(struct inode *inode,
259 struct ocfs2_xattr_search *xs,
260 struct ocfs2_xattr_set_ctxt *ctxt);
261
262 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
263 struct ocfs2_xattr_info *xi,
264 struct ocfs2_xattr_search *xs,
265 struct ocfs2_xattr_set_ctxt *ctxt);
266
267 typedef int (xattr_tree_rec_func)(struct inode *inode,
268 struct buffer_head *root_bh,
269 u64 blkno, u32 cpos, u32 len, void *para);
270 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
271 struct buffer_head *root_bh,
272 xattr_tree_rec_func *rec_func,
273 void *para);
274 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
275 struct ocfs2_xattr_bucket *bucket,
276 void *para);
277 static int ocfs2_rm_xattr_cluster(struct inode *inode,
278 struct buffer_head *root_bh,
279 u64 blkno,
280 u32 cpos,
281 u32 len,
282 void *para);
283
284 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
285 u64 src_blk, u64 last_blk, u64 to_blk,
286 unsigned int start_bucket,
287 u32 *first_hash);
288 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
289 struct ocfs2_dinode *di,
290 struct ocfs2_xattr_info *xi,
291 struct ocfs2_xattr_search *xis,
292 struct ocfs2_xattr_search *xbs,
293 struct ocfs2_refcount_tree **ref_tree,
294 int *meta_need,
295 int *credits);
296 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
297 struct ocfs2_xattr_bucket *bucket,
298 int offset,
299 struct ocfs2_xattr_value_root **xv,
300 struct buffer_head **bh);
301
302 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
303 {
304 return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
305 }
306
307 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
308 {
309 return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
310 }
311
312 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
313 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
314 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
315
316 static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
317 {
318 struct ocfs2_xattr_bucket *bucket;
319 int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
320
321 BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
322
323 bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
324 if (bucket) {
325 bucket->bu_inode = inode;
326 bucket->bu_blocks = blks;
327 }
328
329 return bucket;
330 }
331
332 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
333 {
334 int i;
335
336 for (i = 0; i < bucket->bu_blocks; i++) {
337 brelse(bucket->bu_bhs[i]);
338 bucket->bu_bhs[i] = NULL;
339 }
340 }
341
342 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
343 {
344 if (bucket) {
345 ocfs2_xattr_bucket_relse(bucket);
346 bucket->bu_inode = NULL;
347 kfree(bucket);
348 }
349 }
350
351 /*
352 * A bucket that has never been written to disk doesn't need to be
353 * read. We just need the buffer_heads. Don't call this for
354 * buckets that are already on disk. ocfs2_read_xattr_bucket() initializes
355 * them fully.
356 */
357 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
358 u64 xb_blkno, int new)
359 {
360 int i, rc = 0;
361
362 for (i = 0; i < bucket->bu_blocks; i++) {
363 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
364 xb_blkno + i);
365 if (!bucket->bu_bhs[i]) {
366 rc = -ENOMEM;
367 mlog_errno(rc);
368 break;
369 }
370
371 if (!ocfs2_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
372 bucket->bu_bhs[i])) {
373 if (new)
374 ocfs2_set_new_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
375 bucket->bu_bhs[i]);
376 else {
377 set_buffer_uptodate(bucket->bu_bhs[i]);
378 ocfs2_set_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
379 bucket->bu_bhs[i]);
380 }
381 }
382 }
383
384 if (rc)
385 ocfs2_xattr_bucket_relse(bucket);
386 return rc;
387 }
388
389 /* Read the xattr bucket at xb_blkno */
390 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
391 u64 xb_blkno)
392 {
393 int rc;
394
395 rc = ocfs2_read_blocks(INODE_CACHE(bucket->bu_inode), xb_blkno,
396 bucket->bu_blocks, bucket->bu_bhs, 0,
397 NULL);
398 if (!rc) {
399 spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
400 rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
401 bucket->bu_bhs,
402 bucket->bu_blocks,
403 &bucket_xh(bucket)->xh_check);
404 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
405 if (rc)
406 mlog_errno(rc);
407 }
408
409 if (rc)
410 ocfs2_xattr_bucket_relse(bucket);
411 return rc;
412 }
413
414 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
415 struct ocfs2_xattr_bucket *bucket,
416 int type)
417 {
418 int i, rc = 0;
419
420 for (i = 0; i < bucket->bu_blocks; i++) {
421 rc = ocfs2_journal_access(handle,
422 INODE_CACHE(bucket->bu_inode),
423 bucket->bu_bhs[i], type);
424 if (rc) {
425 mlog_errno(rc);
426 break;
427 }
428 }
429
430 return rc;
431 }
432
433 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
434 struct ocfs2_xattr_bucket *bucket)
435 {
436 int i;
437
438 spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
439 ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
440 bucket->bu_bhs, bucket->bu_blocks,
441 &bucket_xh(bucket)->xh_check);
442 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
443
444 for (i = 0; i < bucket->bu_blocks; i++)
445 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
446 }
447
448 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
449 struct ocfs2_xattr_bucket *src)
450 {
451 int i;
452 int blocksize = src->bu_inode->i_sb->s_blocksize;
453
454 BUG_ON(dest->bu_blocks != src->bu_blocks);
455 BUG_ON(dest->bu_inode != src->bu_inode);
456
457 for (i = 0; i < src->bu_blocks; i++) {
458 memcpy(bucket_block(dest, i), bucket_block(src, i),
459 blocksize);
460 }
461 }
462
463 static int ocfs2_validate_xattr_block(struct super_block *sb,
464 struct buffer_head *bh)
465 {
466 int rc;
467 struct ocfs2_xattr_block *xb =
468 (struct ocfs2_xattr_block *)bh->b_data;
469
470 trace_ocfs2_validate_xattr_block((unsigned long long)bh->b_blocknr);
471
472 BUG_ON(!buffer_uptodate(bh));
473
474 /*
475 * If the ecc fails, we return the error but otherwise
476 * leave the filesystem running. We know any error is
477 * local to this block.
478 */
479 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
480 if (rc)
481 return rc;
482
483 /*
484 * Errors after here are fatal
485 */
486
487 if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
488 return ocfs2_error(sb,
489 "Extended attribute block #%llu has bad signature %.*s\n",
490 (unsigned long long)bh->b_blocknr, 7,
491 xb->xb_signature);
492 }
493
494 if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
495 return ocfs2_error(sb,
496 "Extended attribute block #%llu has an invalid xb_blkno of %llu\n",
497 (unsigned long long)bh->b_blocknr,
498 (unsigned long long)le64_to_cpu(xb->xb_blkno));
499 }
500
501 if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
502 return ocfs2_error(sb,
503 "Extended attribute block #%llu has an invalid xb_fs_generation of #%u\n",
504 (unsigned long long)bh->b_blocknr,
505 le32_to_cpu(xb->xb_fs_generation));
506 }
507
508 return 0;
509 }
510
511 static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
512 struct buffer_head **bh)
513 {
514 int rc;
515 struct buffer_head *tmp = *bh;
516
517 rc = ocfs2_read_block(INODE_CACHE(inode), xb_blkno, &tmp,
518 ocfs2_validate_xattr_block);
519
520 /* If ocfs2_read_block() got us a new bh, pass it up. */
521 if (!rc && !*bh)
522 *bh = tmp;
523
524 return rc;
525 }
526
527 static inline const char *ocfs2_xattr_prefix(int name_index)
528 {
529 const struct xattr_handler *handler = NULL;
530
531 if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
532 handler = ocfs2_xattr_handler_map[name_index];
533 return handler ? xattr_prefix(handler) : NULL;
534 }
535
536 static u32 ocfs2_xattr_name_hash(struct inode *inode,
537 const char *name,
538 int name_len)
539 {
540 /* Get hash value of uuid from super block */
541 u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
542 int i;
543
544 /* hash extended attribute name */
545 for (i = 0; i < name_len; i++) {
546 hash = (hash << OCFS2_HASH_SHIFT) ^
547 (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
548 *name++;
549 }
550
551 return hash;
552 }
553
554 static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
555 {
556 return namevalue_size(name_len, value_len) +
557 sizeof(struct ocfs2_xattr_entry);
558 }
559
560 static int ocfs2_xi_entry_usage(struct ocfs2_xattr_info *xi)
561 {
562 return namevalue_size_xi(xi) +
563 sizeof(struct ocfs2_xattr_entry);
564 }
565
566 static int ocfs2_xe_entry_usage(struct ocfs2_xattr_entry *xe)
567 {
568 return namevalue_size_xe(xe) +
569 sizeof(struct ocfs2_xattr_entry);
570 }
571
572 int ocfs2_calc_security_init(struct inode *dir,
573 struct ocfs2_security_xattr_info *si,
574 int *want_clusters,
575 int *xattr_credits,
576 struct ocfs2_alloc_context **xattr_ac)
577 {
578 int ret = 0;
579 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
580 int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
581 si->value_len);
582
583 /*
584 * The max space of security xattr taken inline is
585 * 256(name) + 80(value) + 16(entry) = 352 bytes,
586 * So reserve one metadata block for it is ok.
587 */
588 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
589 s_size > OCFS2_XATTR_FREE_IN_IBODY) {
590 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
591 if (ret) {
592 mlog_errno(ret);
593 return ret;
594 }
595 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
596 }
597
598 /* reserve clusters for xattr value which will be set in B tree*/
599 if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
600 int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
601 si->value_len);
602
603 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
604 new_clusters);
605 *want_clusters += new_clusters;
606 }
607 return ret;
608 }
609
610 int ocfs2_calc_xattr_init(struct inode *dir,
611 struct buffer_head *dir_bh,
612 umode_t mode,
613 struct ocfs2_security_xattr_info *si,
614 int *want_clusters,
615 int *xattr_credits,
616 int *want_meta)
617 {
618 int ret = 0;
619 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
620 int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
621
622 if (si->enable)
623 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
624 si->value_len);
625
626 if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
627 down_read(&OCFS2_I(dir)->ip_xattr_sem);
628 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
629 OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
630 "", NULL, 0);
631 up_read(&OCFS2_I(dir)->ip_xattr_sem);
632 if (acl_len > 0) {
633 a_size = ocfs2_xattr_entry_real_size(0, acl_len);
634 if (S_ISDIR(mode))
635 a_size <<= 1;
636 } else if (acl_len != 0 && acl_len != -ENODATA) {
637 ret = acl_len;
638 mlog_errno(ret);
639 return ret;
640 }
641 }
642
643 if (!(s_size + a_size))
644 return ret;
645
646 /*
647 * The max space of security xattr taken inline is
648 * 256(name) + 80(value) + 16(entry) = 352 bytes,
649 * The max space of acl xattr taken inline is
650 * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
651 * when blocksize = 512, may reserve one more cluser for
652 * xattr bucket, otherwise reserve one metadata block
653 * for them is ok.
654 * If this is a new directory with inline data,
655 * we choose to reserve the entire inline area for
656 * directory contents and force an external xattr block.
657 */
658 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
659 (S_ISDIR(mode) && ocfs2_supports_inline_data(osb)) ||
660 (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
661 *want_meta = *want_meta + 1;
662 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
663 }
664
665 if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
666 (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
667 *want_clusters += 1;
668 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
669 }
670
671 /*
672 * reserve credits and clusters for xattrs which has large value
673 * and have to be set outside
674 */
675 if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
676 new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
677 si->value_len);
678 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
679 new_clusters);
680 *want_clusters += new_clusters;
681 }
682 if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
683 acl_len > OCFS2_XATTR_INLINE_SIZE) {
684 /* for directory, it has DEFAULT and ACCESS two types of acls */
685 new_clusters = (S_ISDIR(mode) ? 2 : 1) *
686 ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
687 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
688 new_clusters);
689 *want_clusters += new_clusters;
690 }
691
692 return ret;
693 }
694
695 static int ocfs2_xattr_extend_allocation(struct inode *inode,
696 u32 clusters_to_add,
697 struct ocfs2_xattr_value_buf *vb,
698 struct ocfs2_xattr_set_ctxt *ctxt)
699 {
700 int status = 0, credits;
701 handle_t *handle = ctxt->handle;
702 enum ocfs2_alloc_restarted why;
703 u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
704 struct ocfs2_extent_tree et;
705
706 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
707
708 while (clusters_to_add) {
709 trace_ocfs2_xattr_extend_allocation(clusters_to_add);
710
711 status = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
712 OCFS2_JOURNAL_ACCESS_WRITE);
713 if (status < 0) {
714 mlog_errno(status);
715 break;
716 }
717
718 prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
719 status = ocfs2_add_clusters_in_btree(handle,
720 &et,
721 &logical_start,
722 clusters_to_add,
723 0,
724 ctxt->data_ac,
725 ctxt->meta_ac,
726 &why);
727 if ((status < 0) && (status != -EAGAIN)) {
728 if (status != -ENOSPC)
729 mlog_errno(status);
730 break;
731 }
732
733 ocfs2_journal_dirty(handle, vb->vb_bh);
734
735 clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) -
736 prev_clusters;
737
738 if (why != RESTART_NONE && clusters_to_add) {
739 /*
740 * We can only fail in case the alloc file doesn't give
741 * up enough clusters.
742 */
743 BUG_ON(why == RESTART_META);
744
745 credits = ocfs2_calc_extend_credits(inode->i_sb,
746 &vb->vb_xv->xr_list);
747 status = ocfs2_extend_trans(handle, credits);
748 if (status < 0) {
749 status = -ENOMEM;
750 mlog_errno(status);
751 break;
752 }
753 }
754 }
755
756 return status;
757 }
758
759 static int __ocfs2_remove_xattr_range(struct inode *inode,
760 struct ocfs2_xattr_value_buf *vb,
761 u32 cpos, u32 phys_cpos, u32 len,
762 unsigned int ext_flags,
763 struct ocfs2_xattr_set_ctxt *ctxt)
764 {
765 int ret;
766 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
767 handle_t *handle = ctxt->handle;
768 struct ocfs2_extent_tree et;
769
770 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
771
772 ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
773 OCFS2_JOURNAL_ACCESS_WRITE);
774 if (ret) {
775 mlog_errno(ret);
776 goto out;
777 }
778
779 ret = ocfs2_remove_extent(handle, &et, cpos, len, ctxt->meta_ac,
780 &ctxt->dealloc);
781 if (ret) {
782 mlog_errno(ret);
783 goto out;
784 }
785
786 le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
787 ocfs2_journal_dirty(handle, vb->vb_bh);
788
789 if (ext_flags & OCFS2_EXT_REFCOUNTED)
790 ret = ocfs2_decrease_refcount(inode, handle,
791 ocfs2_blocks_to_clusters(inode->i_sb,
792 phys_blkno),
793 len, ctxt->meta_ac, &ctxt->dealloc, 1);
794 else
795 ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc,
796 phys_blkno, len);
797 if (ret)
798 mlog_errno(ret);
799
800 out:
801 return ret;
802 }
803
804 static int ocfs2_xattr_shrink_size(struct inode *inode,
805 u32 old_clusters,
806 u32 new_clusters,
807 struct ocfs2_xattr_value_buf *vb,
808 struct ocfs2_xattr_set_ctxt *ctxt)
809 {
810 int ret = 0;
811 unsigned int ext_flags;
812 u32 trunc_len, cpos, phys_cpos, alloc_size;
813 u64 block;
814
815 if (old_clusters <= new_clusters)
816 return 0;
817
818 cpos = new_clusters;
819 trunc_len = old_clusters - new_clusters;
820 while (trunc_len) {
821 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
822 &alloc_size,
823 &vb->vb_xv->xr_list, &ext_flags);
824 if (ret) {
825 mlog_errno(ret);
826 goto out;
827 }
828
829 if (alloc_size > trunc_len)
830 alloc_size = trunc_len;
831
832 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
833 phys_cpos, alloc_size,
834 ext_flags, ctxt);
835 if (ret) {
836 mlog_errno(ret);
837 goto out;
838 }
839
840 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
841 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode),
842 block, alloc_size);
843 cpos += alloc_size;
844 trunc_len -= alloc_size;
845 }
846
847 out:
848 return ret;
849 }
850
851 static int ocfs2_xattr_value_truncate(struct inode *inode,
852 struct ocfs2_xattr_value_buf *vb,
853 int len,
854 struct ocfs2_xattr_set_ctxt *ctxt)
855 {
856 int ret;
857 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
858 u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
859
860 if (new_clusters == old_clusters)
861 return 0;
862
863 if (new_clusters > old_clusters)
864 ret = ocfs2_xattr_extend_allocation(inode,
865 new_clusters - old_clusters,
866 vb, ctxt);
867 else
868 ret = ocfs2_xattr_shrink_size(inode,
869 old_clusters, new_clusters,
870 vb, ctxt);
871
872 return ret;
873 }
874
875 static int ocfs2_xattr_list_entry(struct super_block *sb,
876 char *buffer, size_t size,
877 size_t *result, int type,
878 const char *name, int name_len)
879 {
880 char *p = buffer + *result;
881 const char *prefix;
882 int prefix_len;
883 int total_len;
884
885 switch(type) {
886 case OCFS2_XATTR_INDEX_USER:
887 if (OCFS2_SB(sb)->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
888 return 0;
889 break;
890
891 case OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS:
892 case OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT:
893 if (!(sb->s_flags & SB_POSIXACL))
894 return 0;
895 break;
896
897 case OCFS2_XATTR_INDEX_TRUSTED:
898 if (!capable(CAP_SYS_ADMIN))
899 return 0;
900 break;
901 }
902
903 prefix = ocfs2_xattr_prefix(type);
904 if (!prefix)
905 return 0;
906 prefix_len = strlen(prefix);
907 total_len = prefix_len + name_len + 1;
908 *result += total_len;
909
910 /* we are just looking for how big our buffer needs to be */
911 if (!size)
912 return 0;
913
914 if (*result > size)
915 return -ERANGE;
916
917 memcpy(p, prefix, prefix_len);
918 memcpy(p + prefix_len, name, name_len);
919 p[prefix_len + name_len] = '\0';
920
921 return 0;
922 }
923
924 static int ocfs2_xattr_list_entries(struct inode *inode,
925 struct ocfs2_xattr_header *header,
926 char *buffer, size_t buffer_size)
927 {
928 size_t result = 0;
929 int i, type, ret;
930 const char *name;
931
932 for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
933 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
934 type = ocfs2_xattr_get_type(entry);
935 name = (const char *)header +
936 le16_to_cpu(entry->xe_name_offset);
937
938 ret = ocfs2_xattr_list_entry(inode->i_sb,
939 buffer, buffer_size,
940 &result, type, name,
941 entry->xe_name_len);
942 if (ret)
943 return ret;
944 }
945
946 return result;
947 }
948
949 int ocfs2_has_inline_xattr_value_outside(struct inode *inode,
950 struct ocfs2_dinode *di)
951 {
952 struct ocfs2_xattr_header *xh;
953 int i;
954
955 xh = (struct ocfs2_xattr_header *)
956 ((void *)di + inode->i_sb->s_blocksize -
957 le16_to_cpu(di->i_xattr_inline_size));
958
959 for (i = 0; i < le16_to_cpu(xh->xh_count); i++)
960 if (!ocfs2_xattr_is_local(&xh->xh_entries[i]))
961 return 1;
962
963 return 0;
964 }
965
966 static int ocfs2_xattr_ibody_list(struct inode *inode,
967 struct ocfs2_dinode *di,
968 char *buffer,
969 size_t buffer_size)
970 {
971 struct ocfs2_xattr_header *header = NULL;
972 struct ocfs2_inode_info *oi = OCFS2_I(inode);
973 int ret = 0;
974
975 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
976 return ret;
977
978 header = (struct ocfs2_xattr_header *)
979 ((void *)di + inode->i_sb->s_blocksize -
980 le16_to_cpu(di->i_xattr_inline_size));
981
982 ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
983
984 return ret;
985 }
986
987 static int ocfs2_xattr_block_list(struct inode *inode,
988 struct ocfs2_dinode *di,
989 char *buffer,
990 size_t buffer_size)
991 {
992 struct buffer_head *blk_bh = NULL;
993 struct ocfs2_xattr_block *xb;
994 int ret = 0;
995
996 if (!di->i_xattr_loc)
997 return ret;
998
999 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
1000 &blk_bh);
1001 if (ret < 0) {
1002 mlog_errno(ret);
1003 return ret;
1004 }
1005
1006 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1007 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1008 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
1009 ret = ocfs2_xattr_list_entries(inode, header,
1010 buffer, buffer_size);
1011 } else
1012 ret = ocfs2_xattr_tree_list_index_block(inode, blk_bh,
1013 buffer, buffer_size);
1014
1015 brelse(blk_bh);
1016
1017 return ret;
1018 }
1019
1020 ssize_t ocfs2_listxattr(struct dentry *dentry,
1021 char *buffer,
1022 size_t size)
1023 {
1024 int ret = 0, i_ret = 0, b_ret = 0;
1025 struct buffer_head *di_bh = NULL;
1026 struct ocfs2_dinode *di = NULL;
1027 struct ocfs2_inode_info *oi = OCFS2_I(d_inode(dentry));
1028
1029 if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
1030 return -EOPNOTSUPP;
1031
1032 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1033 return ret;
1034
1035 ret = ocfs2_inode_lock(d_inode(dentry), &di_bh, 0);
1036 if (ret < 0) {
1037 mlog_errno(ret);
1038 return ret;
1039 }
1040
1041 di = (struct ocfs2_dinode *)di_bh->b_data;
1042
1043 down_read(&oi->ip_xattr_sem);
1044 i_ret = ocfs2_xattr_ibody_list(d_inode(dentry), di, buffer, size);
1045 if (i_ret < 0)
1046 b_ret = 0;
1047 else {
1048 if (buffer) {
1049 buffer += i_ret;
1050 size -= i_ret;
1051 }
1052 b_ret = ocfs2_xattr_block_list(d_inode(dentry), di,
1053 buffer, size);
1054 if (b_ret < 0)
1055 i_ret = 0;
1056 }
1057 up_read(&oi->ip_xattr_sem);
1058 ocfs2_inode_unlock(d_inode(dentry), 0);
1059
1060 brelse(di_bh);
1061
1062 return i_ret + b_ret;
1063 }
1064
1065 static int ocfs2_xattr_find_entry(struct inode *inode, int name_index,
1066 const char *name,
1067 struct ocfs2_xattr_search *xs)
1068 {
1069 struct ocfs2_xattr_entry *entry;
1070 size_t name_len;
1071 int i, name_offset, cmp = 1;
1072
1073 if (name == NULL)
1074 return -EINVAL;
1075
1076 name_len = strlen(name);
1077 entry = xs->here;
1078 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
1079 if ((void *)entry >= xs->end) {
1080 ocfs2_error(inode->i_sb, "corrupted xattr entries");
1081 return -EFSCORRUPTED;
1082 }
1083 cmp = name_index - ocfs2_xattr_get_type(entry);
1084 if (!cmp)
1085 cmp = name_len - entry->xe_name_len;
1086 if (!cmp) {
1087 name_offset = le16_to_cpu(entry->xe_name_offset);
1088 if ((xs->base + name_offset + name_len) > xs->end) {
1089 ocfs2_error(inode->i_sb,
1090 "corrupted xattr entries");
1091 return -EFSCORRUPTED;
1092 }
1093 cmp = memcmp(name, (xs->base + name_offset), name_len);
1094 }
1095 if (cmp == 0)
1096 break;
1097 entry += 1;
1098 }
1099 xs->here = entry;
1100
1101 return cmp ? -ENODATA : 0;
1102 }
1103
1104 static int ocfs2_xattr_get_value_outside(struct inode *inode,
1105 struct ocfs2_xattr_value_root *xv,
1106 void *buffer,
1107 size_t len)
1108 {
1109 u32 cpos, p_cluster, num_clusters, bpc, clusters;
1110 u64 blkno;
1111 int i, ret = 0;
1112 size_t cplen, blocksize;
1113 struct buffer_head *bh = NULL;
1114 struct ocfs2_extent_list *el;
1115
1116 el = &xv->xr_list;
1117 clusters = le32_to_cpu(xv->xr_clusters);
1118 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1119 blocksize = inode->i_sb->s_blocksize;
1120
1121 cpos = 0;
1122 while (cpos < clusters) {
1123 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1124 &num_clusters, el, NULL);
1125 if (ret) {
1126 mlog_errno(ret);
1127 goto out;
1128 }
1129
1130 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1131 /* Copy ocfs2_xattr_value */
1132 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1133 ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1134 &bh, NULL);
1135 if (ret) {
1136 mlog_errno(ret);
1137 goto out;
1138 }
1139
1140 cplen = len >= blocksize ? blocksize : len;
1141 memcpy(buffer, bh->b_data, cplen);
1142 len -= cplen;
1143 buffer += cplen;
1144
1145 brelse(bh);
1146 bh = NULL;
1147 if (len == 0)
1148 break;
1149 }
1150 cpos += num_clusters;
1151 }
1152 out:
1153 return ret;
1154 }
1155
1156 static int ocfs2_xattr_ibody_get(struct inode *inode,
1157 int name_index,
1158 const char *name,
1159 void *buffer,
1160 size_t buffer_size,
1161 struct ocfs2_xattr_search *xs)
1162 {
1163 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1164 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1165 struct ocfs2_xattr_value_root *xv;
1166 size_t size;
1167 int ret = 0;
1168
1169 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1170 return -ENODATA;
1171
1172 xs->end = (void *)di + inode->i_sb->s_blocksize;
1173 xs->header = (struct ocfs2_xattr_header *)
1174 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1175 xs->base = (void *)xs->header;
1176 xs->here = xs->header->xh_entries;
1177
1178 ret = ocfs2_xattr_find_entry(inode, name_index, name, xs);
1179 if (ret)
1180 return ret;
1181 size = le64_to_cpu(xs->here->xe_value_size);
1182 if (buffer) {
1183 if (size > buffer_size)
1184 return -ERANGE;
1185 if (ocfs2_xattr_is_local(xs->here)) {
1186 memcpy(buffer, (void *)xs->base +
1187 le16_to_cpu(xs->here->xe_name_offset) +
1188 OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1189 } else {
1190 xv = (struct ocfs2_xattr_value_root *)
1191 (xs->base + le16_to_cpu(
1192 xs->here->xe_name_offset) +
1193 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1194 ret = ocfs2_xattr_get_value_outside(inode, xv,
1195 buffer, size);
1196 if (ret < 0) {
1197 mlog_errno(ret);
1198 return ret;
1199 }
1200 }
1201 }
1202
1203 return size;
1204 }
1205
1206 static int ocfs2_xattr_block_get(struct inode *inode,
1207 int name_index,
1208 const char *name,
1209 void *buffer,
1210 size_t buffer_size,
1211 struct ocfs2_xattr_search *xs)
1212 {
1213 struct ocfs2_xattr_block *xb;
1214 struct ocfs2_xattr_value_root *xv;
1215 size_t size;
1216 int ret = -ENODATA, name_offset, name_len, i;
1217 int block_off;
1218
1219 xs->bucket = ocfs2_xattr_bucket_new(inode);
1220 if (!xs->bucket) {
1221 ret = -ENOMEM;
1222 mlog_errno(ret);
1223 goto cleanup;
1224 }
1225
1226 ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1227 if (ret) {
1228 mlog_errno(ret);
1229 goto cleanup;
1230 }
1231
1232 if (xs->not_found) {
1233 ret = -ENODATA;
1234 goto cleanup;
1235 }
1236
1237 xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1238 size = le64_to_cpu(xs->here->xe_value_size);
1239 if (buffer) {
1240 ret = -ERANGE;
1241 if (size > buffer_size)
1242 goto cleanup;
1243
1244 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1245 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1246 i = xs->here - xs->header->xh_entries;
1247
1248 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1249 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
1250 bucket_xh(xs->bucket),
1251 i,
1252 &block_off,
1253 &name_offset);
1254 if (ret) {
1255 mlog_errno(ret);
1256 goto cleanup;
1257 }
1258 xs->base = bucket_block(xs->bucket, block_off);
1259 }
1260 if (ocfs2_xattr_is_local(xs->here)) {
1261 memcpy(buffer, (void *)xs->base +
1262 name_offset + name_len, size);
1263 } else {
1264 xv = (struct ocfs2_xattr_value_root *)
1265 (xs->base + name_offset + name_len);
1266 ret = ocfs2_xattr_get_value_outside(inode, xv,
1267 buffer, size);
1268 if (ret < 0) {
1269 mlog_errno(ret);
1270 goto cleanup;
1271 }
1272 }
1273 }
1274 ret = size;
1275 cleanup:
1276 ocfs2_xattr_bucket_free(xs->bucket);
1277
1278 brelse(xs->xattr_bh);
1279 xs->xattr_bh = NULL;
1280 return ret;
1281 }
1282
1283 int ocfs2_xattr_get_nolock(struct inode *inode,
1284 struct buffer_head *di_bh,
1285 int name_index,
1286 const char *name,
1287 void *buffer,
1288 size_t buffer_size)
1289 {
1290 int ret;
1291 struct ocfs2_dinode *di = NULL;
1292 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1293 struct ocfs2_xattr_search xis = {
1294 .not_found = -ENODATA,
1295 };
1296 struct ocfs2_xattr_search xbs = {
1297 .not_found = -ENODATA,
1298 };
1299
1300 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1301 return -EOPNOTSUPP;
1302
1303 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1304 return -ENODATA;
1305
1306 xis.inode_bh = xbs.inode_bh = di_bh;
1307 di = (struct ocfs2_dinode *)di_bh->b_data;
1308
1309 ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1310 buffer_size, &xis);
1311 if (ret == -ENODATA && di->i_xattr_loc)
1312 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1313 buffer_size, &xbs);
1314
1315 return ret;
1316 }
1317
1318 /* ocfs2_xattr_get()
1319 *
1320 * Copy an extended attribute into the buffer provided.
1321 * Buffer is NULL to compute the size of buffer required.
1322 */
1323 static int ocfs2_xattr_get(struct inode *inode,
1324 int name_index,
1325 const char *name,
1326 void *buffer,
1327 size_t buffer_size)
1328 {
1329 int ret, had_lock;
1330 struct buffer_head *di_bh = NULL;
1331 struct ocfs2_lock_holder oh;
1332
1333 had_lock = ocfs2_inode_lock_tracker(inode, &di_bh, 0, &oh);
1334 if (had_lock < 0) {
1335 mlog_errno(had_lock);
1336 return had_lock;
1337 }
1338 down_read(&OCFS2_I(inode)->ip_xattr_sem);
1339 ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1340 name, buffer, buffer_size);
1341 up_read(&OCFS2_I(inode)->ip_xattr_sem);
1342
1343 ocfs2_inode_unlock_tracker(inode, 0, &oh, had_lock);
1344
1345 brelse(di_bh);
1346
1347 return ret;
1348 }
1349
1350 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1351 handle_t *handle,
1352 struct ocfs2_xattr_value_buf *vb,
1353 const void *value,
1354 int value_len)
1355 {
1356 int ret = 0, i, cp_len;
1357 u16 blocksize = inode->i_sb->s_blocksize;
1358 u32 p_cluster, num_clusters;
1359 u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1360 u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1361 u64 blkno;
1362 struct buffer_head *bh = NULL;
1363 unsigned int ext_flags;
1364 struct ocfs2_xattr_value_root *xv = vb->vb_xv;
1365
1366 BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1367
1368 while (cpos < clusters) {
1369 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1370 &num_clusters, &xv->xr_list,
1371 &ext_flags);
1372 if (ret) {
1373 mlog_errno(ret);
1374 goto out;
1375 }
1376
1377 BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1378
1379 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1380
1381 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1382 ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1383 &bh, NULL);
1384 if (ret) {
1385 mlog_errno(ret);
1386 goto out;
1387 }
1388
1389 ret = ocfs2_journal_access(handle,
1390 INODE_CACHE(inode),
1391 bh,
1392 OCFS2_JOURNAL_ACCESS_WRITE);
1393 if (ret < 0) {
1394 mlog_errno(ret);
1395 goto out;
1396 }
1397
1398 cp_len = value_len > blocksize ? blocksize : value_len;
1399 memcpy(bh->b_data, value, cp_len);
1400 value_len -= cp_len;
1401 value += cp_len;
1402 if (cp_len < blocksize)
1403 memset(bh->b_data + cp_len, 0,
1404 blocksize - cp_len);
1405
1406 ocfs2_journal_dirty(handle, bh);
1407 brelse(bh);
1408 bh = NULL;
1409
1410 /*
1411 * XXX: do we need to empty all the following
1412 * blocks in this cluster?
1413 */
1414 if (!value_len)
1415 break;
1416 }
1417 cpos += num_clusters;
1418 }
1419 out:
1420 brelse(bh);
1421
1422 return ret;
1423 }
1424
1425 static int ocfs2_xa_check_space_helper(int needed_space, int free_start,
1426 int num_entries)
1427 {
1428 int free_space;
1429
1430 if (!needed_space)
1431 return 0;
1432
1433 free_space = free_start -
1434 sizeof(struct ocfs2_xattr_header) -
1435 (num_entries * sizeof(struct ocfs2_xattr_entry)) -
1436 OCFS2_XATTR_HEADER_GAP;
1437 if (free_space < 0)
1438 return -EIO;
1439 if (free_space < needed_space)
1440 return -ENOSPC;
1441
1442 return 0;
1443 }
1444
1445 static int ocfs2_xa_journal_access(handle_t *handle, struct ocfs2_xa_loc *loc,
1446 int type)
1447 {
1448 return loc->xl_ops->xlo_journal_access(handle, loc, type);
1449 }
1450
1451 static void ocfs2_xa_journal_dirty(handle_t *handle, struct ocfs2_xa_loc *loc)
1452 {
1453 loc->xl_ops->xlo_journal_dirty(handle, loc);
1454 }
1455
1456 /* Give a pointer into the storage for the given offset */
1457 static void *ocfs2_xa_offset_pointer(struct ocfs2_xa_loc *loc, int offset)
1458 {
1459 BUG_ON(offset >= loc->xl_size);
1460 return loc->xl_ops->xlo_offset_pointer(loc, offset);
1461 }
1462
1463 /*
1464 * Wipe the name+value pair and allow the storage to reclaim it. This
1465 * must be followed by either removal of the entry or a call to
1466 * ocfs2_xa_add_namevalue().
1467 */
1468 static void ocfs2_xa_wipe_namevalue(struct ocfs2_xa_loc *loc)
1469 {
1470 loc->xl_ops->xlo_wipe_namevalue(loc);
1471 }
1472
1473 /*
1474 * Find lowest offset to a name+value pair. This is the start of our
1475 * downward-growing free space.
1476 */
1477 static int ocfs2_xa_get_free_start(struct ocfs2_xa_loc *loc)
1478 {
1479 return loc->xl_ops->xlo_get_free_start(loc);
1480 }
1481
1482 /* Can we reuse loc->xl_entry for xi? */
1483 static int ocfs2_xa_can_reuse_entry(struct ocfs2_xa_loc *loc,
1484 struct ocfs2_xattr_info *xi)
1485 {
1486 return loc->xl_ops->xlo_can_reuse(loc, xi);
1487 }
1488
1489 /* How much free space is needed to set the new value */
1490 static int ocfs2_xa_check_space(struct ocfs2_xa_loc *loc,
1491 struct ocfs2_xattr_info *xi)
1492 {
1493 return loc->xl_ops->xlo_check_space(loc, xi);
1494 }
1495
1496 static void ocfs2_xa_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1497 {
1498 loc->xl_ops->xlo_add_entry(loc, name_hash);
1499 loc->xl_entry->xe_name_hash = cpu_to_le32(name_hash);
1500 /*
1501 * We can't leave the new entry's xe_name_offset at zero or
1502 * add_namevalue() will go nuts. We set it to the size of our
1503 * storage so that it can never be less than any other entry.
1504 */
1505 loc->xl_entry->xe_name_offset = cpu_to_le16(loc->xl_size);
1506 }
1507
1508 static void ocfs2_xa_add_namevalue(struct ocfs2_xa_loc *loc,
1509 struct ocfs2_xattr_info *xi)
1510 {
1511 int size = namevalue_size_xi(xi);
1512 int nameval_offset;
1513 char *nameval_buf;
1514
1515 loc->xl_ops->xlo_add_namevalue(loc, size);
1516 loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
1517 loc->xl_entry->xe_name_len = xi->xi_name_len;
1518 ocfs2_xattr_set_type(loc->xl_entry, xi->xi_name_index);
1519 ocfs2_xattr_set_local(loc->xl_entry,
1520 xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE);
1521
1522 nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1523 nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
1524 memset(nameval_buf, 0, size);
1525 memcpy(nameval_buf, xi->xi_name, xi->xi_name_len);
1526 }
1527
1528 static void ocfs2_xa_fill_value_buf(struct ocfs2_xa_loc *loc,
1529 struct ocfs2_xattr_value_buf *vb)
1530 {
1531 int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1532 int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
1533
1534 /* Value bufs are for value trees */
1535 BUG_ON(ocfs2_xattr_is_local(loc->xl_entry));
1536 BUG_ON(namevalue_size_xe(loc->xl_entry) !=
1537 (name_size + OCFS2_XATTR_ROOT_SIZE));
1538
1539 loc->xl_ops->xlo_fill_value_buf(loc, vb);
1540 vb->vb_xv =
1541 (struct ocfs2_xattr_value_root *)ocfs2_xa_offset_pointer(loc,
1542 nameval_offset +
1543 name_size);
1544 }
1545
1546 static int ocfs2_xa_block_journal_access(handle_t *handle,
1547 struct ocfs2_xa_loc *loc, int type)
1548 {
1549 struct buffer_head *bh = loc->xl_storage;
1550 ocfs2_journal_access_func access;
1551
1552 if (loc->xl_size == (bh->b_size -
1553 offsetof(struct ocfs2_xattr_block,
1554 xb_attrs.xb_header)))
1555 access = ocfs2_journal_access_xb;
1556 else
1557 access = ocfs2_journal_access_di;
1558 return access(handle, INODE_CACHE(loc->xl_inode), bh, type);
1559 }
1560
1561 static void ocfs2_xa_block_journal_dirty(handle_t *handle,
1562 struct ocfs2_xa_loc *loc)
1563 {
1564 struct buffer_head *bh = loc->xl_storage;
1565
1566 ocfs2_journal_dirty(handle, bh);
1567 }
1568
1569 static void *ocfs2_xa_block_offset_pointer(struct ocfs2_xa_loc *loc,
1570 int offset)
1571 {
1572 return (char *)loc->xl_header + offset;
1573 }
1574
1575 static int ocfs2_xa_block_can_reuse(struct ocfs2_xa_loc *loc,
1576 struct ocfs2_xattr_info *xi)
1577 {
1578 /*
1579 * Block storage is strict. If the sizes aren't exact, we will
1580 * remove the old one and reinsert the new.
1581 */
1582 return namevalue_size_xe(loc->xl_entry) ==
1583 namevalue_size_xi(xi);
1584 }
1585
1586 static int ocfs2_xa_block_get_free_start(struct ocfs2_xa_loc *loc)
1587 {
1588 struct ocfs2_xattr_header *xh = loc->xl_header;
1589 int i, count = le16_to_cpu(xh->xh_count);
1590 int offset, free_start = loc->xl_size;
1591
1592 for (i = 0; i < count; i++) {
1593 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1594 if (offset < free_start)
1595 free_start = offset;
1596 }
1597
1598 return free_start;
1599 }
1600
1601 static int ocfs2_xa_block_check_space(struct ocfs2_xa_loc *loc,
1602 struct ocfs2_xattr_info *xi)
1603 {
1604 int count = le16_to_cpu(loc->xl_header->xh_count);
1605 int free_start = ocfs2_xa_get_free_start(loc);
1606 int needed_space = ocfs2_xi_entry_usage(xi);
1607
1608 /*
1609 * Block storage will reclaim the original entry before inserting
1610 * the new value, so we only need the difference. If the new
1611 * entry is smaller than the old one, we don't need anything.
1612 */
1613 if (loc->xl_entry) {
1614 /* Don't need space if we're reusing! */
1615 if (ocfs2_xa_can_reuse_entry(loc, xi))
1616 needed_space = 0;
1617 else
1618 needed_space -= ocfs2_xe_entry_usage(loc->xl_entry);
1619 }
1620 if (needed_space < 0)
1621 needed_space = 0;
1622 return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1623 }
1624
1625 /*
1626 * Block storage for xattrs keeps the name+value pairs compacted. When
1627 * we remove one, we have to shift any that preceded it towards the end.
1628 */
1629 static void ocfs2_xa_block_wipe_namevalue(struct ocfs2_xa_loc *loc)
1630 {
1631 int i, offset;
1632 int namevalue_offset, first_namevalue_offset, namevalue_size;
1633 struct ocfs2_xattr_entry *entry = loc->xl_entry;
1634 struct ocfs2_xattr_header *xh = loc->xl_header;
1635 int count = le16_to_cpu(xh->xh_count);
1636
1637 namevalue_offset = le16_to_cpu(entry->xe_name_offset);
1638 namevalue_size = namevalue_size_xe(entry);
1639 first_namevalue_offset = ocfs2_xa_get_free_start(loc);
1640
1641 /* Shift the name+value pairs */
1642 memmove((char *)xh + first_namevalue_offset + namevalue_size,
1643 (char *)xh + first_namevalue_offset,
1644 namevalue_offset - first_namevalue_offset);
1645 memset((char *)xh + first_namevalue_offset, 0, namevalue_size);
1646
1647 /* Now tell xh->xh_entries about it */
1648 for (i = 0; i < count; i++) {
1649 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1650 if (offset <= namevalue_offset)
1651 le16_add_cpu(&xh->xh_entries[i].xe_name_offset,
1652 namevalue_size);
1653 }
1654
1655 /*
1656 * Note that we don't update xh_free_start or xh_name_value_len
1657 * because they're not used in block-stored xattrs.
1658 */
1659 }
1660
1661 static void ocfs2_xa_block_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1662 {
1663 int count = le16_to_cpu(loc->xl_header->xh_count);
1664 loc->xl_entry = &(loc->xl_header->xh_entries[count]);
1665 le16_add_cpu(&loc->xl_header->xh_count, 1);
1666 memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1667 }
1668
1669 static void ocfs2_xa_block_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1670 {
1671 int free_start = ocfs2_xa_get_free_start(loc);
1672
1673 loc->xl_entry->xe_name_offset = cpu_to_le16(free_start - size);
1674 }
1675
1676 static void ocfs2_xa_block_fill_value_buf(struct ocfs2_xa_loc *loc,
1677 struct ocfs2_xattr_value_buf *vb)
1678 {
1679 struct buffer_head *bh = loc->xl_storage;
1680
1681 if (loc->xl_size == (bh->b_size -
1682 offsetof(struct ocfs2_xattr_block,
1683 xb_attrs.xb_header)))
1684 vb->vb_access = ocfs2_journal_access_xb;
1685 else
1686 vb->vb_access = ocfs2_journal_access_di;
1687 vb->vb_bh = bh;
1688 }
1689
1690 /*
1691 * Operations for xattrs stored in blocks. This includes inline inode
1692 * storage and unindexed ocfs2_xattr_blocks.
1693 */
1694 static const struct ocfs2_xa_loc_operations ocfs2_xa_block_loc_ops = {
1695 .xlo_journal_access = ocfs2_xa_block_journal_access,
1696 .xlo_journal_dirty = ocfs2_xa_block_journal_dirty,
1697 .xlo_offset_pointer = ocfs2_xa_block_offset_pointer,
1698 .xlo_check_space = ocfs2_xa_block_check_space,
1699 .xlo_can_reuse = ocfs2_xa_block_can_reuse,
1700 .xlo_get_free_start = ocfs2_xa_block_get_free_start,
1701 .xlo_wipe_namevalue = ocfs2_xa_block_wipe_namevalue,
1702 .xlo_add_entry = ocfs2_xa_block_add_entry,
1703 .xlo_add_namevalue = ocfs2_xa_block_add_namevalue,
1704 .xlo_fill_value_buf = ocfs2_xa_block_fill_value_buf,
1705 };
1706
1707 static int ocfs2_xa_bucket_journal_access(handle_t *handle,
1708 struct ocfs2_xa_loc *loc, int type)
1709 {
1710 struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1711
1712 return ocfs2_xattr_bucket_journal_access(handle, bucket, type);
1713 }
1714
1715 static void ocfs2_xa_bucket_journal_dirty(handle_t *handle,
1716 struct ocfs2_xa_loc *loc)
1717 {
1718 struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1719
1720 ocfs2_xattr_bucket_journal_dirty(handle, bucket);
1721 }
1722
1723 static void *ocfs2_xa_bucket_offset_pointer(struct ocfs2_xa_loc *loc,
1724 int offset)
1725 {
1726 struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1727 int block, block_offset;
1728
1729 /* The header is at the front of the bucket */
1730 block = offset >> loc->xl_inode->i_sb->s_blocksize_bits;
1731 block_offset = offset % loc->xl_inode->i_sb->s_blocksize;
1732
1733 return bucket_block(bucket, block) + block_offset;
1734 }
1735
1736 static int ocfs2_xa_bucket_can_reuse(struct ocfs2_xa_loc *loc,
1737 struct ocfs2_xattr_info *xi)
1738 {
1739 return namevalue_size_xe(loc->xl_entry) >=
1740 namevalue_size_xi(xi);
1741 }
1742
1743 static int ocfs2_xa_bucket_get_free_start(struct ocfs2_xa_loc *loc)
1744 {
1745 struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1746 return le16_to_cpu(bucket_xh(bucket)->xh_free_start);
1747 }
1748
1749 static int ocfs2_bucket_align_free_start(struct super_block *sb,
1750 int free_start, int size)
1751 {
1752 /*
1753 * We need to make sure that the name+value pair fits within
1754 * one block.
1755 */
1756 if (((free_start - size) >> sb->s_blocksize_bits) !=
1757 ((free_start - 1) >> sb->s_blocksize_bits))
1758 free_start -= free_start % sb->s_blocksize;
1759
1760 return free_start;
1761 }
1762
1763 static int ocfs2_xa_bucket_check_space(struct ocfs2_xa_loc *loc,
1764 struct ocfs2_xattr_info *xi)
1765 {
1766 int rc;
1767 int count = le16_to_cpu(loc->xl_header->xh_count);
1768 int free_start = ocfs2_xa_get_free_start(loc);
1769 int needed_space = ocfs2_xi_entry_usage(xi);
1770 int size = namevalue_size_xi(xi);
1771 struct super_block *sb = loc->xl_inode->i_sb;
1772
1773 /*
1774 * Bucket storage does not reclaim name+value pairs it cannot
1775 * reuse. They live as holes until the bucket fills, and then
1776 * the bucket is defragmented. However, the bucket can reclaim
1777 * the ocfs2_xattr_entry.
1778 */
1779 if (loc->xl_entry) {
1780 /* Don't need space if we're reusing! */
1781 if (ocfs2_xa_can_reuse_entry(loc, xi))
1782 needed_space = 0;
1783 else
1784 needed_space -= sizeof(struct ocfs2_xattr_entry);
1785 }
1786 BUG_ON(needed_space < 0);
1787
1788 if (free_start < size) {
1789 if (needed_space)
1790 return -ENOSPC;
1791 } else {
1792 /*
1793 * First we check if it would fit in the first place.
1794 * Below, we align the free start to a block. This may
1795 * slide us below the minimum gap. By checking unaligned
1796 * first, we avoid that error.
1797 */
1798 rc = ocfs2_xa_check_space_helper(needed_space, free_start,
1799 count);
1800 if (rc)
1801 return rc;
1802 free_start = ocfs2_bucket_align_free_start(sb, free_start,
1803 size);
1804 }
1805 return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1806 }
1807
1808 static void ocfs2_xa_bucket_wipe_namevalue(struct ocfs2_xa_loc *loc)
1809 {
1810 le16_add_cpu(&loc->xl_header->xh_name_value_len,
1811 -namevalue_size_xe(loc->xl_entry));
1812 }
1813
1814 static void ocfs2_xa_bucket_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1815 {
1816 struct ocfs2_xattr_header *xh = loc->xl_header;
1817 int count = le16_to_cpu(xh->xh_count);
1818 int low = 0, high = count - 1, tmp;
1819 struct ocfs2_xattr_entry *tmp_xe;
1820
1821 /*
1822 * We keep buckets sorted by name_hash, so we need to find
1823 * our insert place.
1824 */
1825 while (low <= high && count) {
1826 tmp = (low + high) / 2;
1827 tmp_xe = &xh->xh_entries[tmp];
1828
1829 if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
1830 low = tmp + 1;
1831 else if (name_hash < le32_to_cpu(tmp_xe->xe_name_hash))
1832 high = tmp - 1;
1833 else {
1834 low = tmp;
1835 break;
1836 }
1837 }
1838
1839 if (low != count)
1840 memmove(&xh->xh_entries[low + 1],
1841 &xh->xh_entries[low],
1842 ((count - low) * sizeof(struct ocfs2_xattr_entry)));
1843
1844 le16_add_cpu(&xh->xh_count, 1);
1845 loc->xl_entry = &xh->xh_entries[low];
1846 memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1847 }
1848
1849 static void ocfs2_xa_bucket_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1850 {
1851 int free_start = ocfs2_xa_get_free_start(loc);
1852 struct ocfs2_xattr_header *xh = loc->xl_header;
1853 struct super_block *sb = loc->xl_inode->i_sb;
1854 int nameval_offset;
1855
1856 free_start = ocfs2_bucket_align_free_start(sb, free_start, size);
1857 nameval_offset = free_start - size;
1858 loc->xl_entry->xe_name_offset = cpu_to_le16(nameval_offset);
1859 xh->xh_free_start = cpu_to_le16(nameval_offset);
1860 le16_add_cpu(&xh->xh_name_value_len, size);
1861
1862 }
1863
1864 static void ocfs2_xa_bucket_fill_value_buf(struct ocfs2_xa_loc *loc,
1865 struct ocfs2_xattr_value_buf *vb)
1866 {
1867 struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1868 struct super_block *sb = loc->xl_inode->i_sb;
1869 int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1870 int size = namevalue_size_xe(loc->xl_entry);
1871 int block_offset = nameval_offset >> sb->s_blocksize_bits;
1872
1873 /* Values are not allowed to straddle block boundaries */
1874 BUG_ON(block_offset !=
1875 ((nameval_offset + size - 1) >> sb->s_blocksize_bits));
1876 /* We expect the bucket to be filled in */
1877 BUG_ON(!bucket->bu_bhs[block_offset]);
1878
1879 vb->vb_access = ocfs2_journal_access;
1880 vb->vb_bh = bucket->bu_bhs[block_offset];
1881 }
1882
1883 /* Operations for xattrs stored in buckets. */
1884 static const struct ocfs2_xa_loc_operations ocfs2_xa_bucket_loc_ops = {
1885 .xlo_journal_access = ocfs2_xa_bucket_journal_access,
1886 .xlo_journal_dirty = ocfs2_xa_bucket_journal_dirty,
1887 .xlo_offset_pointer = ocfs2_xa_bucket_offset_pointer,
1888 .xlo_check_space = ocfs2_xa_bucket_check_space,
1889 .xlo_can_reuse = ocfs2_xa_bucket_can_reuse,
1890 .xlo_get_free_start = ocfs2_xa_bucket_get_free_start,
1891 .xlo_wipe_namevalue = ocfs2_xa_bucket_wipe_namevalue,
1892 .xlo_add_entry = ocfs2_xa_bucket_add_entry,
1893 .xlo_add_namevalue = ocfs2_xa_bucket_add_namevalue,
1894 .xlo_fill_value_buf = ocfs2_xa_bucket_fill_value_buf,
1895 };
1896
1897 static unsigned int ocfs2_xa_value_clusters(struct ocfs2_xa_loc *loc)
1898 {
1899 struct ocfs2_xattr_value_buf vb;
1900
1901 if (ocfs2_xattr_is_local(loc->xl_entry))
1902 return 0;
1903
1904 ocfs2_xa_fill_value_buf(loc, &vb);
1905 return le32_to_cpu(vb.vb_xv->xr_clusters);
1906 }
1907
1908 static int ocfs2_xa_value_truncate(struct ocfs2_xa_loc *loc, u64 bytes,
1909 struct ocfs2_xattr_set_ctxt *ctxt)
1910 {
1911 int trunc_rc, access_rc;
1912 struct ocfs2_xattr_value_buf vb;
1913
1914 ocfs2_xa_fill_value_buf(loc, &vb);
1915 trunc_rc = ocfs2_xattr_value_truncate(loc->xl_inode, &vb, bytes,
1916 ctxt);
1917
1918 /*
1919 * The caller of ocfs2_xa_value_truncate() has already called
1920 * ocfs2_xa_journal_access on the loc. However, The truncate code
1921 * calls ocfs2_extend_trans(). This may commit the previous
1922 * transaction and open a new one. If this is a bucket, truncate
1923 * could leave only vb->vb_bh set up for journaling. Meanwhile,
1924 * the caller is expecting to dirty the entire bucket. So we must
1925 * reset the journal work. We do this even if truncate has failed,
1926 * as it could have failed after committing the extend.
1927 */
1928 access_rc = ocfs2_xa_journal_access(ctxt->handle, loc,
1929 OCFS2_JOURNAL_ACCESS_WRITE);
1930
1931 /* Errors in truncate take precedence */
1932 return trunc_rc ? trunc_rc : access_rc;
1933 }
1934
1935 static void ocfs2_xa_remove_entry(struct ocfs2_xa_loc *loc)
1936 {
1937 int index, count;
1938 struct ocfs2_xattr_header *xh = loc->xl_header;
1939 struct ocfs2_xattr_entry *entry = loc->xl_entry;
1940
1941 ocfs2_xa_wipe_namevalue(loc);
1942 loc->xl_entry = NULL;
1943
1944 le16_add_cpu(&xh->xh_count, -1);
1945 count = le16_to_cpu(xh->xh_count);
1946
1947 /*
1948 * Only zero out the entry if there are more remaining. This is
1949 * important for an empty bucket, as it keeps track of the
1950 * bucket's hash value. It doesn't hurt empty block storage.
1951 */
1952 if (count) {
1953 index = ((char *)entry - (char *)&xh->xh_entries) /
1954 sizeof(struct ocfs2_xattr_entry);
1955 memmove(&xh->xh_entries[index], &xh->xh_entries[index + 1],
1956 (count - index) * sizeof(struct ocfs2_xattr_entry));
1957 memset(&xh->xh_entries[count], 0,
1958 sizeof(struct ocfs2_xattr_entry));
1959 }
1960 }
1961
1962 /*
1963 * If we have a problem adjusting the size of an external value during
1964 * ocfs2_xa_prepare_entry() or ocfs2_xa_remove(), we may have an xattr
1965 * in an intermediate state. For example, the value may be partially
1966 * truncated.
1967 *
1968 * If the value tree hasn't changed, the extend/truncate went nowhere.
1969 * We have nothing to do. The caller can treat it as a straight error.
1970 *
1971 * If the value tree got partially truncated, we now have a corrupted
1972 * extended attribute. We're going to wipe its entry and leak the
1973 * clusters. Better to leak some storage than leave a corrupt entry.
1974 *
1975 * If the value tree grew, it obviously didn't grow enough for the
1976 * new entry. We're not going to try and reclaim those clusters either.
1977 * If there was already an external value there (orig_clusters != 0),
1978 * the new clusters are attached safely and we can just leave the old
1979 * value in place. If there was no external value there, we remove
1980 * the entry.
1981 *
1982 * This way, the xattr block we store in the journal will be consistent.
1983 * If the size change broke because of the journal, no changes will hit
1984 * disk anyway.
1985 */
1986 static void ocfs2_xa_cleanup_value_truncate(struct ocfs2_xa_loc *loc,
1987 const char *what,
1988 unsigned int orig_clusters)
1989 {
1990 unsigned int new_clusters = ocfs2_xa_value_clusters(loc);
1991 char *nameval_buf = ocfs2_xa_offset_pointer(loc,
1992 le16_to_cpu(loc->xl_entry->xe_name_offset));
1993
1994 if (new_clusters < orig_clusters) {
1995 mlog(ML_ERROR,
1996 "Partial truncate while %s xattr %.*s. Leaking "
1997 "%u clusters and removing the entry\n",
1998 what, loc->xl_entry->xe_name_len, nameval_buf,
1999 orig_clusters - new_clusters);
2000 ocfs2_xa_remove_entry(loc);
2001 } else if (!orig_clusters) {
2002 mlog(ML_ERROR,
2003 "Unable to allocate an external value for xattr "
2004 "%.*s safely. Leaking %u clusters and removing the "
2005 "entry\n",
2006 loc->xl_entry->xe_name_len, nameval_buf,
2007 new_clusters - orig_clusters);
2008 ocfs2_xa_remove_entry(loc);
2009 } else if (new_clusters > orig_clusters)
2010 mlog(ML_ERROR,
2011 "Unable to grow xattr %.*s safely. %u new clusters "
2012 "have been added, but the value will not be "
2013 "modified\n",
2014 loc->xl_entry->xe_name_len, nameval_buf,
2015 new_clusters - orig_clusters);
2016 }
2017
2018 static int ocfs2_xa_remove(struct ocfs2_xa_loc *loc,
2019 struct ocfs2_xattr_set_ctxt *ctxt)
2020 {
2021 int rc = 0;
2022 unsigned int orig_clusters;
2023
2024 if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2025 orig_clusters = ocfs2_xa_value_clusters(loc);
2026 rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2027 if (rc) {
2028 mlog_errno(rc);
2029 /*
2030 * Since this is remove, we can return 0 if
2031 * ocfs2_xa_cleanup_value_truncate() is going to
2032 * wipe the entry anyway. So we check the
2033 * cluster count as well.
2034 */
2035 if (orig_clusters != ocfs2_xa_value_clusters(loc))
2036 rc = 0;
2037 ocfs2_xa_cleanup_value_truncate(loc, "removing",
2038 orig_clusters);
2039 goto out;
2040 }
2041 }
2042
2043 ocfs2_xa_remove_entry(loc);
2044
2045 out:
2046 return rc;
2047 }
2048
2049 static void ocfs2_xa_install_value_root(struct ocfs2_xa_loc *loc)
2050 {
2051 int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
2052 char *nameval_buf;
2053
2054 nameval_buf = ocfs2_xa_offset_pointer(loc,
2055 le16_to_cpu(loc->xl_entry->xe_name_offset));
2056 memcpy(nameval_buf + name_size, &def_xv, OCFS2_XATTR_ROOT_SIZE);
2057 }
2058
2059 /*
2060 * Take an existing entry and make it ready for the new value. This
2061 * won't allocate space, but it may free space. It should be ready for
2062 * ocfs2_xa_prepare_entry() to finish the work.
2063 */
2064 static int ocfs2_xa_reuse_entry(struct ocfs2_xa_loc *loc,
2065 struct ocfs2_xattr_info *xi,
2066 struct ocfs2_xattr_set_ctxt *ctxt)
2067 {
2068 int rc = 0;
2069 int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2070 unsigned int orig_clusters;
2071 char *nameval_buf;
2072 int xe_local = ocfs2_xattr_is_local(loc->xl_entry);
2073 int xi_local = xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE;
2074
2075 BUG_ON(OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len) !=
2076 name_size);
2077
2078 nameval_buf = ocfs2_xa_offset_pointer(loc,
2079 le16_to_cpu(loc->xl_entry->xe_name_offset));
2080 if (xe_local) {
2081 memset(nameval_buf + name_size, 0,
2082 namevalue_size_xe(loc->xl_entry) - name_size);
2083 if (!xi_local)
2084 ocfs2_xa_install_value_root(loc);
2085 } else {
2086 orig_clusters = ocfs2_xa_value_clusters(loc);
2087 if (xi_local) {
2088 rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2089 if (rc < 0)
2090 mlog_errno(rc);
2091 else
2092 memset(nameval_buf + name_size, 0,
2093 namevalue_size_xe(loc->xl_entry) -
2094 name_size);
2095 } else if (le64_to_cpu(loc->xl_entry->xe_value_size) >
2096 xi->xi_value_len) {
2097 rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len,
2098 ctxt);
2099 if (rc < 0)
2100 mlog_errno(rc);
2101 }
2102
2103 if (rc) {
2104 ocfs2_xa_cleanup_value_truncate(loc, "reusing",
2105 orig_clusters);
2106 goto out;
2107 }
2108 }
2109
2110 loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
2111 ocfs2_xattr_set_local(loc->xl_entry, xi_local);
2112
2113 out:
2114 return rc;
2115 }
2116
2117 /*
2118 * Prepares loc->xl_entry to receive the new xattr. This includes
2119 * properly setting up the name+value pair region. If loc->xl_entry
2120 * already exists, it will take care of modifying it appropriately.
2121 *
2122 * Note that this modifies the data. You did journal_access already,
2123 * right?
2124 */
2125 static int ocfs2_xa_prepare_entry(struct ocfs2_xa_loc *loc,
2126 struct ocfs2_xattr_info *xi,
2127 u32 name_hash,
2128 struct ocfs2_xattr_set_ctxt *ctxt)
2129 {
2130 int rc = 0;
2131 unsigned int orig_clusters;
2132 __le64 orig_value_size = 0;
2133
2134 rc = ocfs2_xa_check_space(loc, xi);
2135 if (rc)
2136 goto out;
2137
2138 if (loc->xl_entry) {
2139 if (ocfs2_xa_can_reuse_entry(loc, xi)) {
2140 orig_value_size = loc->xl_entry->xe_value_size;
2141 rc = ocfs2_xa_reuse_entry(loc, xi, ctxt);
2142 if (rc)
2143 goto out;
2144 goto alloc_value;
2145 }
2146
2147 if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2148 orig_clusters = ocfs2_xa_value_clusters(loc);
2149 rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2150 if (rc) {
2151 mlog_errno(rc);
2152 ocfs2_xa_cleanup_value_truncate(loc,
2153 "overwriting",
2154 orig_clusters);
2155 goto out;
2156 }
2157 }
2158 ocfs2_xa_wipe_namevalue(loc);
2159 } else
2160 ocfs2_xa_add_entry(loc, name_hash);
2161
2162 /*
2163 * If we get here, we have a blank entry. Fill it. We grow our
2164 * name+value pair back from the end.
2165 */
2166 ocfs2_xa_add_namevalue(loc, xi);
2167 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
2168 ocfs2_xa_install_value_root(loc);
2169
2170 alloc_value:
2171 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2172 orig_clusters = ocfs2_xa_value_clusters(loc);
2173 rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len, ctxt);
2174 if (rc < 0) {
2175 ctxt->set_abort = 1;
2176 ocfs2_xa_cleanup_value_truncate(loc, "growing",
2177 orig_clusters);
2178 /*
2179 * If we were growing an existing value,
2180 * ocfs2_xa_cleanup_value_truncate() won't remove
2181 * the entry. We need to restore the original value
2182 * size.
2183 */
2184 if (loc->xl_entry) {
2185 BUG_ON(!orig_value_size);
2186 loc->xl_entry->xe_value_size = orig_value_size;
2187 }
2188 mlog_errno(rc);
2189 }
2190 }
2191
2192 out:
2193 return rc;
2194 }
2195
2196 /*
2197 * Store the value portion of the name+value pair. This will skip
2198 * values that are stored externally. Their tree roots were set up
2199 * by ocfs2_xa_prepare_entry().
2200 */
2201 static int ocfs2_xa_store_value(struct ocfs2_xa_loc *loc,
2202 struct ocfs2_xattr_info *xi,
2203 struct ocfs2_xattr_set_ctxt *ctxt)
2204 {
2205 int rc = 0;
2206 int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
2207 int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2208 char *nameval_buf;
2209 struct ocfs2_xattr_value_buf vb;
2210
2211 nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
2212 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2213 ocfs2_xa_fill_value_buf(loc, &vb);
2214 rc = __ocfs2_xattr_set_value_outside(loc->xl_inode,
2215 ctxt->handle, &vb,
2216 xi->xi_value,
2217 xi->xi_value_len);
2218 } else
2219 memcpy(nameval_buf + name_size, xi->xi_value, xi->xi_value_len);
2220
2221 return rc;
2222 }
2223
2224 static int ocfs2_xa_set(struct ocfs2_xa_loc *loc,
2225 struct ocfs2_xattr_info *xi,
2226 struct ocfs2_xattr_set_ctxt *ctxt)
2227 {
2228 int ret;
2229 u32 name_hash = ocfs2_xattr_name_hash(loc->xl_inode, xi->xi_name,
2230 xi->xi_name_len);
2231
2232 ret = ocfs2_xa_journal_access(ctxt->handle, loc,
2233 OCFS2_JOURNAL_ACCESS_WRITE);
2234 if (ret) {
2235 mlog_errno(ret);
2236 goto out;
2237 }
2238
2239 /*
2240 * From here on out, everything is going to modify the buffer a
2241 * little. Errors are going to leave the xattr header in a
2242 * sane state. Thus, even with errors we dirty the sucker.
2243 */
2244
2245 /* Don't worry, we are never called with !xi_value and !xl_entry */
2246 if (!xi->xi_value) {
2247 ret = ocfs2_xa_remove(loc, ctxt);
2248 goto out_dirty;
2249 }
2250
2251 ret = ocfs2_xa_prepare_entry(loc, xi, name_hash, ctxt);
2252 if (ret) {
2253 if (ret != -ENOSPC)
2254 mlog_errno(ret);
2255 goto out_dirty;
2256 }
2257
2258 ret = ocfs2_xa_store_value(loc, xi, ctxt);
2259 if (ret)
2260 mlog_errno(ret);
2261
2262 out_dirty:
2263 ocfs2_xa_journal_dirty(ctxt->handle, loc);
2264
2265 out:
2266 return ret;
2267 }
2268
2269 static void ocfs2_init_dinode_xa_loc(struct ocfs2_xa_loc *loc,
2270 struct inode *inode,
2271 struct buffer_head *bh,
2272 struct ocfs2_xattr_entry *entry)
2273 {
2274 struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2275
2276 BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_XATTR_FL));
2277
2278 loc->xl_inode = inode;
2279 loc->xl_ops = &ocfs2_xa_block_loc_ops;
2280 loc->xl_storage = bh;
2281 loc->xl_entry = entry;
2282 loc->xl_size = le16_to_cpu(di->i_xattr_inline_size);
2283 loc->xl_header =
2284 (struct ocfs2_xattr_header *)(bh->b_data + bh->b_size -
2285 loc->xl_size);
2286 }
2287
2288 static void ocfs2_init_xattr_block_xa_loc(struct ocfs2_xa_loc *loc,
2289 struct inode *inode,
2290 struct buffer_head *bh,
2291 struct ocfs2_xattr_entry *entry)
2292 {
2293 struct ocfs2_xattr_block *xb =
2294 (struct ocfs2_xattr_block *)bh->b_data;
2295
2296 BUG_ON(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED);
2297
2298 loc->xl_inode = inode;
2299 loc->xl_ops = &ocfs2_xa_block_loc_ops;
2300 loc->xl_storage = bh;
2301 loc->xl_header = &(xb->xb_attrs.xb_header);
2302 loc->xl_entry = entry;
2303 loc->xl_size = bh->b_size - offsetof(struct ocfs2_xattr_block,
2304 xb_attrs.xb_header);
2305 }
2306
2307 static void ocfs2_init_xattr_bucket_xa_loc(struct ocfs2_xa_loc *loc,
2308 struct ocfs2_xattr_bucket *bucket,
2309 struct ocfs2_xattr_entry *entry)
2310 {
2311 loc->xl_inode = bucket->bu_inode;
2312 loc->xl_ops = &ocfs2_xa_bucket_loc_ops;
2313 loc->xl_storage = bucket;
2314 loc->xl_header = bucket_xh(bucket);
2315 loc->xl_entry = entry;
2316 loc->xl_size = OCFS2_XATTR_BUCKET_SIZE;
2317 }
2318
2319 /*
2320 * In xattr remove, if it is stored outside and refcounted, we may have
2321 * the chance to split the refcount tree. So need the allocators.
2322 */
2323 static int ocfs2_lock_xattr_remove_allocators(struct inode *inode,
2324 struct ocfs2_xattr_value_root *xv,
2325 struct ocfs2_caching_info *ref_ci,
2326 struct buffer_head *ref_root_bh,
2327 struct ocfs2_alloc_context **meta_ac,
2328 int *ref_credits)
2329 {
2330 int ret, meta_add = 0;
2331 u32 p_cluster, num_clusters;
2332 unsigned int ext_flags;
2333
2334 *ref_credits = 0;
2335 ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
2336 &num_clusters,
2337 &xv->xr_list,
2338 &ext_flags);
2339 if (ret) {
2340 mlog_errno(ret);
2341 goto out;
2342 }
2343
2344 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
2345 goto out;
2346
2347 ret = ocfs2_refcounted_xattr_delete_need(inode, ref_ci,
2348 ref_root_bh, xv,
2349 &meta_add, ref_credits);
2350 if (ret) {
2351 mlog_errno(ret);
2352 goto out;
2353 }
2354
2355 ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
2356 meta_add, meta_ac);
2357 if (ret)
2358 mlog_errno(ret);
2359
2360 out:
2361 return ret;
2362 }
2363
2364 static int ocfs2_remove_value_outside(struct inode*inode,
2365 struct ocfs2_xattr_value_buf *vb,
2366 struct ocfs2_xattr_header *header,
2367 struct ocfs2_caching_info *ref_ci,
2368 struct buffer_head *ref_root_bh)
2369 {
2370 int ret = 0, i, ref_credits;
2371 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2372 struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2373 void *val;
2374
2375 ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
2376
2377 for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
2378 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
2379
2380 if (ocfs2_xattr_is_local(entry))
2381 continue;
2382
2383 val = (void *)header +
2384 le16_to_cpu(entry->xe_name_offset);
2385 vb->vb_xv = (struct ocfs2_xattr_value_root *)
2386 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
2387
2388 ret = ocfs2_lock_xattr_remove_allocators(inode, vb->vb_xv,
2389 ref_ci, ref_root_bh,
2390 &ctxt.meta_ac,
2391 &ref_credits);
2392
2393 ctxt.handle = ocfs2_start_trans(osb, ref_credits +
2394 ocfs2_remove_extent_credits(osb->sb));
2395 if (IS_ERR(ctxt.handle)) {
2396 ret = PTR_ERR(ctxt.handle);
2397 mlog_errno(ret);
2398 break;
2399 }
2400
2401 ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
2402
2403 ocfs2_commit_trans(osb, ctxt.handle);
2404 if (ctxt.meta_ac) {
2405 ocfs2_free_alloc_context(ctxt.meta_ac);
2406 ctxt.meta_ac = NULL;
2407 }
2408
2409 if (ret < 0) {
2410 mlog_errno(ret);
2411 break;
2412 }
2413
2414 }
2415
2416 if (ctxt.meta_ac)
2417 ocfs2_free_alloc_context(ctxt.meta_ac);
2418 ocfs2_schedule_truncate_log_flush(osb, 1);
2419 ocfs2_run_deallocs(osb, &ctxt.dealloc);
2420 return ret;
2421 }
2422
2423 static int ocfs2_xattr_ibody_remove(struct inode *inode,
2424 struct buffer_head *di_bh,
2425 struct ocfs2_caching_info *ref_ci,
2426 struct buffer_head *ref_root_bh)
2427 {
2428
2429 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2430 struct ocfs2_xattr_header *header;
2431 int ret;
2432 struct ocfs2_xattr_value_buf vb = {
2433 .vb_bh = di_bh,
2434 .vb_access = ocfs2_journal_access_di,
2435 };
2436
2437 header = (struct ocfs2_xattr_header *)
2438 ((void *)di + inode->i_sb->s_blocksize -
2439 le16_to_cpu(di->i_xattr_inline_size));
2440
2441 ret = ocfs2_remove_value_outside(inode, &vb, header,
2442 ref_ci, ref_root_bh);
2443
2444 return ret;
2445 }
2446
2447 struct ocfs2_rm_xattr_bucket_para {
2448 struct ocfs2_caching_info *ref_ci;
2449 struct buffer_head *ref_root_bh;
2450 };
2451
2452 static int ocfs2_xattr_block_remove(struct inode *inode,
2453 struct buffer_head *blk_bh,
2454 struct ocfs2_caching_info *ref_ci,
2455 struct buffer_head *ref_root_bh)
2456 {
2457 struct ocfs2_xattr_block *xb;
2458 int ret = 0;
2459 struct ocfs2_xattr_value_buf vb = {
2460 .vb_bh = blk_bh,
2461 .vb_access = ocfs2_journal_access_xb,
2462 };
2463 struct ocfs2_rm_xattr_bucket_para args = {
2464 .ref_ci = ref_ci,
2465 .ref_root_bh = ref_root_bh,
2466 };
2467
2468 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2469 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2470 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
2471 ret = ocfs2_remove_value_outside(inode, &vb, header,
2472 ref_ci, ref_root_bh);
2473 } else
2474 ret = ocfs2_iterate_xattr_index_block(inode,
2475 blk_bh,
2476 ocfs2_rm_xattr_cluster,
2477 &args);
2478
2479 return ret;
2480 }
2481
2482 static int ocfs2_xattr_free_block(struct inode *inode,
2483 u64 block,
2484 struct ocfs2_caching_info *ref_ci,
2485 struct buffer_head *ref_root_bh)
2486 {
2487 struct inode *xb_alloc_inode;
2488 struct buffer_head *xb_alloc_bh = NULL;
2489 struct buffer_head *blk_bh = NULL;
2490 struct ocfs2_xattr_block *xb;
2491 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2492 handle_t *handle;
2493 int ret = 0;
2494 u64 blk, bg_blkno;
2495 u16 bit;
2496
2497 ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
2498 if (ret < 0) {
2499 mlog_errno(ret);
2500 goto out;
2501 }
2502
2503 ret = ocfs2_xattr_block_remove(inode, blk_bh, ref_ci, ref_root_bh);
2504 if (ret < 0) {
2505 mlog_errno(ret);
2506 goto out;
2507 }
2508
2509 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2510 blk = le64_to_cpu(xb->xb_blkno);
2511 bit = le16_to_cpu(xb->xb_suballoc_bit);
2512 if (xb->xb_suballoc_loc)
2513 bg_blkno = le64_to_cpu(xb->xb_suballoc_loc);
2514 else
2515 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2516
2517 xb_alloc_inode = ocfs2_get_system_file_inode(osb,
2518 EXTENT_ALLOC_SYSTEM_INODE,
2519 le16_to_cpu(xb->xb_suballoc_slot));
2520 if (!xb_alloc_inode) {
2521 ret = -ENOMEM;
2522 mlog_errno(ret);
2523 goto out;
2524 }
2525 inode_lock(xb_alloc_inode);
2526
2527 ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
2528 if (ret < 0) {
2529 mlog_errno(ret);
2530 goto out_mutex;
2531 }
2532
2533 handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
2534 if (IS_ERR(handle)) {
2535 ret = PTR_ERR(handle);
2536 mlog_errno(ret);
2537 goto out_unlock;
2538 }
2539
2540 ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
2541 bit, bg_blkno, 1);
2542 if (ret < 0)
2543 mlog_errno(ret);
2544
2545 ocfs2_commit_trans(osb, handle);
2546 out_unlock:
2547 ocfs2_inode_unlock(xb_alloc_inode, 1);
2548 brelse(xb_alloc_bh);
2549 out_mutex:
2550 inode_unlock(xb_alloc_inode);
2551 iput(xb_alloc_inode);
2552 out:
2553 brelse(blk_bh);
2554 return ret;
2555 }
2556
2557 /*
2558 * ocfs2_xattr_remove()
2559 *
2560 * Free extended attribute resources associated with this inode.
2561 */
2562 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
2563 {
2564 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2565 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2566 struct ocfs2_refcount_tree *ref_tree = NULL;
2567 struct buffer_head *ref_root_bh = NULL;
2568 struct ocfs2_caching_info *ref_ci = NULL;
2569 handle_t *handle;
2570 int ret;
2571
2572 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2573 return 0;
2574
2575 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
2576 return 0;
2577
2578 if (ocfs2_is_refcount_inode(inode)) {
2579 ret = ocfs2_lock_refcount_tree(OCFS2_SB(inode->i_sb),
2580 le64_to_cpu(di->i_refcount_loc),
2581 1, &ref_tree, &ref_root_bh);
2582 if (ret) {
2583 mlog_errno(ret);
2584 goto out;
2585 }
2586 ref_ci = &ref_tree->rf_ci;
2587
2588 }
2589
2590 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2591 ret = ocfs2_xattr_ibody_remove(inode, di_bh,
2592 ref_ci, ref_root_bh);
2593 if (ret < 0) {
2594 mlog_errno(ret);
2595 goto out;
2596 }
2597 }
2598
2599 if (di->i_xattr_loc) {
2600 ret = ocfs2_xattr_free_block(inode,
2601 le64_to_cpu(di->i_xattr_loc),
2602 ref_ci, ref_root_bh);
2603 if (ret < 0) {
2604 mlog_errno(ret);
2605 goto out;
2606 }
2607 }
2608
2609 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
2610 OCFS2_INODE_UPDATE_CREDITS);
2611 if (IS_ERR(handle)) {
2612 ret = PTR_ERR(handle);
2613 mlog_errno(ret);
2614 goto out;
2615 }
2616 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2617 OCFS2_JOURNAL_ACCESS_WRITE);
2618 if (ret) {
2619 mlog_errno(ret);
2620 goto out_commit;
2621 }
2622
2623 di->i_xattr_loc = 0;
2624
2625 spin_lock(&oi->ip_lock);
2626 oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
2627 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2628 spin_unlock(&oi->ip_lock);
2629 ocfs2_update_inode_fsync_trans(handle, inode, 0);
2630
2631 ocfs2_journal_dirty(handle, di_bh);
2632 out_commit:
2633 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
2634 out:
2635 if (ref_tree)
2636 ocfs2_unlock_refcount_tree(OCFS2_SB(inode->i_sb), ref_tree, 1);
2637 brelse(ref_root_bh);
2638 return ret;
2639 }
2640
2641 static int ocfs2_xattr_has_space_inline(struct inode *inode,
2642 struct ocfs2_dinode *di)
2643 {
2644 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2645 unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
2646 int free;
2647
2648 if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
2649 return 0;
2650
2651 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2652 struct ocfs2_inline_data *idata = &di->id2.i_data;
2653 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
2654 } else if (ocfs2_inode_is_fast_symlink(inode)) {
2655 free = ocfs2_fast_symlink_chars(inode->i_sb) -
2656 le64_to_cpu(di->i_size);
2657 } else {
2658 struct ocfs2_extent_list *el = &di->id2.i_list;
2659 free = (le16_to_cpu(el->l_count) -
2660 le16_to_cpu(el->l_next_free_rec)) *
2661 sizeof(struct ocfs2_extent_rec);
2662 }
2663 if (free >= xattrsize)
2664 return 1;
2665
2666 return 0;
2667 }
2668
2669 /*
2670 * ocfs2_xattr_ibody_find()
2671 *
2672 * Find extended attribute in inode block and
2673 * fill search info into struct ocfs2_xattr_search.
2674 */
2675 static int ocfs2_xattr_ibody_find(struct inode *inode,
2676 int name_index,
2677 const char *name,
2678 struct ocfs2_xattr_search *xs)
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 int has_space = 0;
2684
2685 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2686 return 0;
2687
2688 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2689 down_read(&oi->ip_alloc_sem);
2690 has_space = ocfs2_xattr_has_space_inline(inode, di);
2691 up_read(&oi->ip_alloc_sem);
2692 if (!has_space)
2693 return 0;
2694 }
2695
2696 xs->xattr_bh = xs->inode_bh;
2697 xs->end = (void *)di + inode->i_sb->s_blocksize;
2698 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
2699 xs->header = (struct ocfs2_xattr_header *)
2700 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
2701 else
2702 xs->header = (struct ocfs2_xattr_header *)
2703 (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
2704 xs->base = (void *)xs->header;
2705 xs->here = xs->header->xh_entries;
2706
2707 /* Find the named attribute. */
2708 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2709 ret = ocfs2_xattr_find_entry(inode, name_index, name, xs);
2710 if (ret && ret != -ENODATA)
2711 return ret;
2712 xs->not_found = ret;
2713 }
2714
2715 return 0;
2716 }
2717
2718 static int ocfs2_xattr_ibody_init(struct inode *inode,
2719 struct buffer_head *di_bh,
2720 struct ocfs2_xattr_set_ctxt *ctxt)
2721 {
2722 int ret;
2723 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2724 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2725 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2726 unsigned int xattrsize = osb->s_xattr_inline_size;
2727
2728 if (!ocfs2_xattr_has_space_inline(inode, di)) {
2729 ret = -ENOSPC;
2730 goto out;
2731 }
2732
2733 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode), di_bh,
2734 OCFS2_JOURNAL_ACCESS_WRITE);
2735 if (ret) {
2736 mlog_errno(ret);
2737 goto out;
2738 }
2739
2740 /*
2741 * Adjust extent record count or inline data size
2742 * to reserve space for extended attribute.
2743 */
2744 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2745 struct ocfs2_inline_data *idata = &di->id2.i_data;
2746 le16_add_cpu(&idata->id_count, -xattrsize);
2747 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
2748 struct ocfs2_extent_list *el = &di->id2.i_list;
2749 le16_add_cpu(&el->l_count, -(xattrsize /
2750 sizeof(struct ocfs2_extent_rec)));
2751 }
2752 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
2753
2754 spin_lock(&oi->ip_lock);
2755 oi->ip_dyn_features |= OCFS2_INLINE_XATTR_FL|OCFS2_HAS_XATTR_FL;
2756 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2757 spin_unlock(&oi->ip_lock);
2758
2759 ocfs2_journal_dirty(ctxt->handle, di_bh);
2760
2761 out:
2762 return ret;
2763 }
2764
2765 /*
2766 * ocfs2_xattr_ibody_set()
2767 *
2768 * Set, replace or remove an extended attribute into inode block.
2769 *
2770 */
2771 static int ocfs2_xattr_ibody_set(struct inode *inode,
2772 struct ocfs2_xattr_info *xi,
2773 struct ocfs2_xattr_search *xs,
2774 struct ocfs2_xattr_set_ctxt *ctxt)
2775 {
2776 int ret;
2777 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2778 struct ocfs2_xa_loc loc;
2779
2780 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2781 return -ENOSPC;
2782
2783 down_write(&oi->ip_alloc_sem);
2784 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2785 ret = ocfs2_xattr_ibody_init(inode, xs->inode_bh, ctxt);
2786 if (ret) {
2787 if (ret != -ENOSPC)
2788 mlog_errno(ret);
2789 goto out;
2790 }
2791 }
2792
2793 ocfs2_init_dinode_xa_loc(&loc, inode, xs->inode_bh,
2794 xs->not_found ? NULL : xs->here);
2795 ret = ocfs2_xa_set(&loc, xi, ctxt);
2796 if (ret) {
2797 if (ret != -ENOSPC)
2798 mlog_errno(ret);
2799 goto out;
2800 }
2801 xs->here = loc.xl_entry;
2802
2803 out:
2804 up_write(&oi->ip_alloc_sem);
2805
2806 return ret;
2807 }
2808
2809 /*
2810 * ocfs2_xattr_block_find()
2811 *
2812 * Find extended attribute in external block and
2813 * fill search info into struct ocfs2_xattr_search.
2814 */
2815 static int ocfs2_xattr_block_find(struct inode *inode,
2816 int name_index,
2817 const char *name,
2818 struct ocfs2_xattr_search *xs)
2819 {
2820 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2821 struct buffer_head *blk_bh = NULL;
2822 struct ocfs2_xattr_block *xb;
2823 int ret = 0;
2824
2825 if (!di->i_xattr_loc)
2826 return ret;
2827
2828 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2829 &blk_bh);
2830 if (ret < 0) {
2831 mlog_errno(ret);
2832 return ret;
2833 }
2834
2835 xs->xattr_bh = blk_bh;
2836 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2837
2838 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2839 xs->header = &xb->xb_attrs.xb_header;
2840 xs->base = (void *)xs->header;
2841 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2842 xs->here = xs->header->xh_entries;
2843
2844 ret = ocfs2_xattr_find_entry(inode, name_index, name, xs);
2845 } else
2846 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2847 name_index,
2848 name, xs);
2849
2850 if (ret && ret != -ENODATA) {
2851 xs->xattr_bh = NULL;
2852 goto cleanup;
2853 }
2854 xs->not_found = ret;
2855 return 0;
2856 cleanup:
2857 brelse(blk_bh);
2858
2859 return ret;
2860 }
2861
2862 static int ocfs2_create_xattr_block(struct inode *inode,
2863 struct buffer_head *inode_bh,
2864 struct ocfs2_xattr_set_ctxt *ctxt,
2865 int indexed,
2866 struct buffer_head **ret_bh)
2867 {
2868 int ret;
2869 u16 suballoc_bit_start;
2870 u32 num_got;
2871 u64 suballoc_loc, first_blkno;
2872 struct ocfs2_dinode *di = (struct ocfs2_dinode *)inode_bh->b_data;
2873 struct buffer_head *new_bh = NULL;
2874 struct ocfs2_xattr_block *xblk;
2875
2876 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
2877 inode_bh, OCFS2_JOURNAL_ACCESS_CREATE);
2878 if (ret < 0) {
2879 mlog_errno(ret);
2880 goto end;
2881 }
2882
2883 ret = ocfs2_claim_metadata(ctxt->handle, ctxt->meta_ac, 1,
2884 &suballoc_loc, &suballoc_bit_start,
2885 &num_got, &first_blkno);
2886 if (ret < 0) {
2887 mlog_errno(ret);
2888 goto end;
2889 }
2890
2891 new_bh = sb_getblk(inode->i_sb, first_blkno);
2892 if (!new_bh) {
2893 ret = -ENOMEM;
2894 mlog_errno(ret);
2895 goto end;
2896 }
2897
2898 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2899
2900 ret = ocfs2_journal_access_xb(ctxt->handle, INODE_CACHE(inode),
2901 new_bh,
2902 OCFS2_JOURNAL_ACCESS_CREATE);
2903 if (ret < 0) {
2904 mlog_errno(ret);
2905 goto end;
2906 }
2907
2908 /* Initialize ocfs2_xattr_block */
2909 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2910 memset(xblk, 0, inode->i_sb->s_blocksize);
2911 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2912 xblk->xb_suballoc_slot = cpu_to_le16(ctxt->meta_ac->ac_alloc_slot);
2913 xblk->xb_suballoc_loc = cpu_to_le64(suballoc_loc);
2914 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2915 xblk->xb_fs_generation =
2916 cpu_to_le32(OCFS2_SB(inode->i_sb)->fs_generation);
2917 xblk->xb_blkno = cpu_to_le64(first_blkno);
2918 if (indexed) {
2919 struct ocfs2_xattr_tree_root *xr = &xblk->xb_attrs.xb_root;
2920 xr->xt_clusters = cpu_to_le32(1);
2921 xr->xt_last_eb_blk = 0;
2922 xr->xt_list.l_tree_depth = 0;
2923 xr->xt_list.l_count = cpu_to_le16(
2924 ocfs2_xattr_recs_per_xb(inode->i_sb));
2925 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2926 xblk->xb_flags = cpu_to_le16(OCFS2_XATTR_INDEXED);
2927 }
2928 ocfs2_journal_dirty(ctxt->handle, new_bh);
2929
2930 /* Add it to the inode */
2931 di->i_xattr_loc = cpu_to_le64(first_blkno);
2932
2933 spin_lock(&OCFS2_I(inode)->ip_lock);
2934 OCFS2_I(inode)->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
2935 di->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
2936 spin_unlock(&OCFS2_I(inode)->ip_lock);
2937
2938 ocfs2_journal_dirty(ctxt->handle, inode_bh);
2939
2940 *ret_bh = new_bh;
2941 new_bh = NULL;
2942
2943 end:
2944 brelse(new_bh);
2945 return ret;
2946 }
2947
2948 /*
2949 * ocfs2_xattr_block_set()
2950 *
2951 * Set, replace or remove an extended attribute into external block.
2952 *
2953 */
2954 static int ocfs2_xattr_block_set(struct inode *inode,
2955 struct ocfs2_xattr_info *xi,
2956 struct ocfs2_xattr_search *xs,
2957 struct ocfs2_xattr_set_ctxt *ctxt)
2958 {
2959 struct buffer_head *new_bh = NULL;
2960 struct ocfs2_xattr_block *xblk = NULL;
2961 int ret;
2962 struct ocfs2_xa_loc loc;
2963
2964 if (!xs->xattr_bh) {
2965 ret = ocfs2_create_xattr_block(inode, xs->inode_bh, ctxt,
2966 0, &new_bh);
2967 if (ret) {
2968 mlog_errno(ret);
2969 goto end;
2970 }
2971
2972 xs->xattr_bh = new_bh;
2973 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2974 xs->header = &xblk->xb_attrs.xb_header;
2975 xs->base = (void *)xs->header;
2976 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2977 xs->here = xs->header->xh_entries;
2978 } else
2979 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2980
2981 if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2982 ocfs2_init_xattr_block_xa_loc(&loc, inode, xs->xattr_bh,
2983 xs->not_found ? NULL : xs->here);
2984
2985 ret = ocfs2_xa_set(&loc, xi, ctxt);
2986 if (!ret)
2987 xs->here = loc.xl_entry;
2988 else if ((ret != -ENOSPC) || ctxt->set_abort)
2989 goto end;
2990 else {
2991 ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2992 if (ret)
2993 goto end;
2994 }
2995 }
2996
2997 if (le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)
2998 ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2999
3000 end:
3001 return ret;
3002 }
3003
3004 /* Check whether the new xattr can be inserted into the inode. */
3005 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
3006 struct ocfs2_xattr_info *xi,
3007 struct ocfs2_xattr_search *xs)
3008 {
3009 struct ocfs2_xattr_entry *last;
3010 int free, i;
3011 size_t min_offs = xs->end - xs->base;
3012
3013 if (!xs->header)
3014 return 0;
3015
3016 last = xs->header->xh_entries;
3017
3018 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
3019 size_t offs = le16_to_cpu(last->xe_name_offset);
3020 if (offs < min_offs)
3021 min_offs = offs;
3022 last += 1;
3023 }
3024
3025 free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
3026 if (free < 0)
3027 return 0;
3028
3029 BUG_ON(!xs->not_found);
3030
3031 if (free >= (sizeof(struct ocfs2_xattr_entry) + namevalue_size_xi(xi)))
3032 return 1;
3033
3034 return 0;
3035 }
3036
3037 static int ocfs2_calc_xattr_set_need(struct inode *inode,
3038 struct ocfs2_dinode *di,
3039 struct ocfs2_xattr_info *xi,
3040 struct ocfs2_xattr_search *xis,
3041 struct ocfs2_xattr_search *xbs,
3042 int *clusters_need,
3043 int *meta_need,
3044 int *credits_need)
3045 {
3046 int ret = 0, old_in_xb = 0;
3047 int clusters_add = 0, meta_add = 0, credits = 0;
3048 struct buffer_head *bh = NULL;
3049 struct ocfs2_xattr_block *xb = NULL;
3050 struct ocfs2_xattr_entry *xe = NULL;
3051 struct ocfs2_xattr_value_root *xv = NULL;
3052 char *base = NULL;
3053 int name_offset, name_len = 0;
3054 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
3055 xi->xi_value_len);
3056 u64 value_size;
3057
3058 /*
3059 * Calculate the clusters we need to write.
3060 * No matter whether we replace an old one or add a new one,
3061 * we need this for writing.
3062 */
3063 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
3064 credits += new_clusters *
3065 ocfs2_clusters_to_blocks(inode->i_sb, 1);
3066
3067 if (xis->not_found && xbs->not_found) {
3068 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3069
3070 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3071 clusters_add += new_clusters;
3072 credits += ocfs2_calc_extend_credits(inode->i_sb,
3073 &def_xv.xv.xr_list);
3074 }
3075
3076 goto meta_guess;
3077 }
3078
3079 if (!xis->not_found) {
3080 xe = xis->here;
3081 name_offset = le16_to_cpu(xe->xe_name_offset);
3082 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3083 base = xis->base;
3084 credits += OCFS2_INODE_UPDATE_CREDITS;
3085 } else {
3086 int i, block_off = 0;
3087 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3088 xe = xbs->here;
3089 name_offset = le16_to_cpu(xe->xe_name_offset);
3090 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3091 i = xbs->here - xbs->header->xh_entries;
3092 old_in_xb = 1;
3093
3094 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3095 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3096 bucket_xh(xbs->bucket),
3097 i, &block_off,
3098 &name_offset);
3099 base = bucket_block(xbs->bucket, block_off);
3100 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3101 } else {
3102 base = xbs->base;
3103 credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
3104 }
3105 }
3106
3107 /*
3108 * delete a xattr doesn't need metadata and cluster allocation.
3109 * so just calculate the credits and return.
3110 *
3111 * The credits for removing the value tree will be extended
3112 * by ocfs2_remove_extent itself.
3113 */
3114 if (!xi->xi_value) {
3115 if (!ocfs2_xattr_is_local(xe))
3116 credits += ocfs2_remove_extent_credits(inode->i_sb);
3117
3118 goto out;
3119 }
3120
3121 /* do cluster allocation guess first. */
3122 value_size = le64_to_cpu(xe->xe_value_size);
3123
3124 if (old_in_xb) {
3125 /*
3126 * In xattr set, we always try to set the xe in inode first,
3127 * so if it can be inserted into inode successfully, the old
3128 * one will be removed from the xattr block, and this xattr
3129 * will be inserted into inode as a new xattr in inode.
3130 */
3131 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
3132 clusters_add += new_clusters;
3133 credits += ocfs2_remove_extent_credits(inode->i_sb) +
3134 OCFS2_INODE_UPDATE_CREDITS;
3135 if (!ocfs2_xattr_is_local(xe))
3136 credits += ocfs2_calc_extend_credits(
3137 inode->i_sb,
3138 &def_xv.xv.xr_list);
3139 goto out;
3140 }
3141 }
3142
3143 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3144 /* the new values will be stored outside. */
3145 u32 old_clusters = 0;
3146
3147 if (!ocfs2_xattr_is_local(xe)) {
3148 old_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
3149 value_size);
3150 xv = (struct ocfs2_xattr_value_root *)
3151 (base + name_offset + name_len);
3152 value_size = OCFS2_XATTR_ROOT_SIZE;
3153 } else
3154 xv = &def_xv.xv;
3155
3156 if (old_clusters >= new_clusters) {
3157 credits += ocfs2_remove_extent_credits(inode->i_sb);
3158 goto out;
3159 } else {
3160 meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
3161 clusters_add += new_clusters - old_clusters;
3162 credits += ocfs2_calc_extend_credits(inode->i_sb,
3163 &xv->xr_list);
3164 if (value_size >= OCFS2_XATTR_ROOT_SIZE)
3165 goto out;
3166 }
3167 } else {
3168 /*
3169 * Now the new value will be stored inside. So if the new
3170 * value is smaller than the size of value root or the old
3171 * value, we don't need any allocation, otherwise we have
3172 * to guess metadata allocation.
3173 */
3174 if ((ocfs2_xattr_is_local(xe) &&
3175 (value_size >= xi->xi_value_len)) ||
3176 (!ocfs2_xattr_is_local(xe) &&
3177 OCFS2_XATTR_ROOT_SIZE >= xi->xi_value_len))
3178 goto out;
3179 }
3180
3181 meta_guess:
3182 /* calculate metadata allocation. */
3183 if (di->i_xattr_loc) {
3184 if (!xbs->xattr_bh) {
3185 ret = ocfs2_read_xattr_block(inode,
3186 le64_to_cpu(di->i_xattr_loc),
3187 &bh);
3188 if (ret) {
3189 mlog_errno(ret);
3190 goto out;
3191 }
3192
3193 xb = (struct ocfs2_xattr_block *)bh->b_data;
3194 } else
3195 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3196
3197 /*
3198 * If there is already an xattr tree, good, we can calculate
3199 * like other b-trees. Otherwise we may have the chance of
3200 * create a tree, the credit calculation is borrowed from
3201 * ocfs2_calc_extend_credits with root_el = NULL. And the
3202 * new tree will be cluster based, so no meta is needed.
3203 */
3204 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3205 struct ocfs2_extent_list *el =
3206 &xb->xb_attrs.xb_root.xt_list;
3207 meta_add += ocfs2_extend_meta_needed(el);
3208 credits += ocfs2_calc_extend_credits(inode->i_sb,
3209 el);
3210 } else
3211 credits += OCFS2_SUBALLOC_ALLOC + 1;
3212
3213 /*
3214 * This cluster will be used either for new bucket or for
3215 * new xattr block.
3216 * If the cluster size is the same as the bucket size, one
3217 * more is needed since we may need to extend the bucket
3218 * also.
3219 */
3220 clusters_add += 1;
3221 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3222 if (OCFS2_XATTR_BUCKET_SIZE ==
3223 OCFS2_SB(inode->i_sb)->s_clustersize) {
3224 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3225 clusters_add += 1;
3226 }
3227 } else {
3228 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
3229 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3230 struct ocfs2_extent_list *el = &def_xv.xv.xr_list;
3231 meta_add += ocfs2_extend_meta_needed(el);
3232 credits += ocfs2_calc_extend_credits(inode->i_sb,
3233 el);
3234 } else {
3235 meta_add += 1;
3236 }
3237 }
3238 out:
3239 if (clusters_need)
3240 *clusters_need = clusters_add;
3241 if (meta_need)
3242 *meta_need = meta_add;
3243 if (credits_need)
3244 *credits_need = credits;
3245 brelse(bh);
3246 return ret;
3247 }
3248
3249 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
3250 struct ocfs2_dinode *di,
3251 struct ocfs2_xattr_info *xi,
3252 struct ocfs2_xattr_search *xis,
3253 struct ocfs2_xattr_search *xbs,
3254 struct ocfs2_xattr_set_ctxt *ctxt,
3255 int extra_meta,
3256 int *credits)
3257 {
3258 int clusters_add, meta_add, ret;
3259 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3260
3261 memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
3262
3263 ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
3264
3265 ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
3266 &clusters_add, &meta_add, credits);
3267 if (ret) {
3268 mlog_errno(ret);
3269 return ret;
3270 }
3271
3272 meta_add += extra_meta;
3273 trace_ocfs2_init_xattr_set_ctxt(xi->xi_name, meta_add,
3274 clusters_add, *credits);
3275
3276 if (meta_add) {
3277 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
3278 &ctxt->meta_ac);
3279 if (ret) {
3280 mlog_errno(ret);
3281 goto out;
3282 }
3283 }
3284
3285 if (clusters_add) {
3286 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
3287 if (ret)
3288 mlog_errno(ret);
3289 }
3290 out:
3291 if (ret) {
3292 if (ctxt->meta_ac) {
3293 ocfs2_free_alloc_context(ctxt->meta_ac);
3294 ctxt->meta_ac = NULL;
3295 }
3296
3297 /*
3298 * We cannot have an error and a non null ctxt->data_ac.
3299 */
3300 }
3301
3302 return ret;
3303 }
3304
3305 static int __ocfs2_xattr_set_handle(struct inode *inode,
3306 struct ocfs2_dinode *di,
3307 struct ocfs2_xattr_info *xi,
3308 struct ocfs2_xattr_search *xis,
3309 struct ocfs2_xattr_search *xbs,
3310 struct ocfs2_xattr_set_ctxt *ctxt)
3311 {
3312 int ret = 0, credits, old_found;
3313
3314 if (!xi->xi_value) {
3315 /* Remove existing extended attribute */
3316 if (!xis->not_found)
3317 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3318 else if (!xbs->not_found)
3319 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3320 } else {
3321 /* We always try to set extended attribute into inode first*/
3322 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3323 if (!ret && !xbs->not_found) {
3324 /*
3325 * If succeed and that extended attribute existing in
3326 * external block, then we will remove it.
3327 */
3328 xi->xi_value = NULL;
3329 xi->xi_value_len = 0;
3330
3331 old_found = xis->not_found;
3332 xis->not_found = -ENODATA;
3333 ret = ocfs2_calc_xattr_set_need(inode,
3334 di,
3335 xi,
3336 xis,
3337 xbs,
3338 NULL,
3339 NULL,
3340 &credits);
3341 xis->not_found = old_found;
3342 if (ret) {
3343 mlog_errno(ret);
3344 goto out;
3345 }
3346
3347 ret = ocfs2_extend_trans(ctxt->handle, credits);
3348 if (ret) {
3349 mlog_errno(ret);
3350 goto out;
3351 }
3352 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3353 } else if ((ret == -ENOSPC) && !ctxt->set_abort) {
3354 if (di->i_xattr_loc && !xbs->xattr_bh) {
3355 ret = ocfs2_xattr_block_find(inode,
3356 xi->xi_name_index,
3357 xi->xi_name, xbs);
3358 if (ret)
3359 goto out;
3360
3361 old_found = xis->not_found;
3362 xis->not_found = -ENODATA;
3363 ret = ocfs2_calc_xattr_set_need(inode,
3364 di,
3365 xi,
3366 xis,
3367 xbs,
3368 NULL,
3369 NULL,
3370 &credits);
3371 xis->not_found = old_found;
3372 if (ret) {
3373 mlog_errno(ret);
3374 goto out;
3375 }
3376
3377 ret = ocfs2_extend_trans(ctxt->handle, credits);
3378 if (ret) {
3379 mlog_errno(ret);
3380 goto out;
3381 }
3382 }
3383 /*
3384 * If no space in inode, we will set extended attribute
3385 * into external block.
3386 */
3387 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3388 if (ret)
3389 goto out;
3390 if (!xis->not_found) {
3391 /*
3392 * If succeed and that extended attribute
3393 * existing in inode, we will remove it.
3394 */
3395 xi->xi_value = NULL;
3396 xi->xi_value_len = 0;
3397 xbs->not_found = -ENODATA;
3398 ret = ocfs2_calc_xattr_set_need(inode,
3399 di,
3400 xi,
3401 xis,
3402 xbs,
3403 NULL,
3404 NULL,
3405 &credits);
3406 if (ret) {
3407 mlog_errno(ret);
3408 goto out;
3409 }
3410
3411 ret = ocfs2_extend_trans(ctxt->handle, credits);
3412 if (ret) {
3413 mlog_errno(ret);
3414 goto out;
3415 }
3416 ret = ocfs2_xattr_ibody_set(inode, xi,
3417 xis, ctxt);
3418 }
3419 }
3420 }
3421
3422 if (!ret) {
3423 /* Update inode ctime. */
3424 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
3425 xis->inode_bh,
3426 OCFS2_JOURNAL_ACCESS_WRITE);
3427 if (ret) {
3428 mlog_errno(ret);
3429 goto out;
3430 }
3431
3432 inode_set_ctime_current(inode);
3433 di->i_ctime = cpu_to_le64(inode_get_ctime_sec(inode));
3434 di->i_ctime_nsec = cpu_to_le32(inode_get_ctime_nsec(inode));
3435 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
3436 }
3437 out:
3438 return ret;
3439 }
3440
3441 /*
3442 * This function only called duing creating inode
3443 * for init security/acl xattrs of the new inode.
3444 * All transanction credits have been reserved in mknod.
3445 */
3446 int ocfs2_xattr_set_handle(handle_t *handle,
3447 struct inode *inode,
3448 struct buffer_head *di_bh,
3449 int name_index,
3450 const char *name,
3451 const void *value,
3452 size_t value_len,
3453 int flags,
3454 struct ocfs2_alloc_context *meta_ac,
3455 struct ocfs2_alloc_context *data_ac)
3456 {
3457 struct ocfs2_dinode *di;
3458 int ret;
3459
3460 struct ocfs2_xattr_info xi = {
3461 .xi_name_index = name_index,
3462 .xi_name = name,
3463 .xi_name_len = strlen(name),
3464 .xi_value = value,
3465 .xi_value_len = value_len,
3466 };
3467
3468 struct ocfs2_xattr_search xis = {
3469 .not_found = -ENODATA,
3470 };
3471
3472 struct ocfs2_xattr_search xbs = {
3473 .not_found = -ENODATA,
3474 };
3475
3476 struct ocfs2_xattr_set_ctxt ctxt = {
3477 .handle = handle,
3478 .meta_ac = meta_ac,
3479 .data_ac = data_ac,
3480 };
3481
3482 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3483 return -EOPNOTSUPP;
3484
3485 /*
3486 * In extreme situation, may need xattr bucket when
3487 * block size is too small. And we have already reserved
3488 * the credits for bucket in mknod.
3489 */
3490 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
3491 xbs.bucket = ocfs2_xattr_bucket_new(inode);
3492 if (!xbs.bucket) {
3493 mlog_errno(-ENOMEM);
3494 return -ENOMEM;
3495 }
3496 }
3497
3498 xis.inode_bh = xbs.inode_bh = di_bh;
3499 di = (struct ocfs2_dinode *)di_bh->b_data;
3500
3501 down_write(&OCFS2_I(inode)->ip_xattr_sem);
3502
3503 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3504 if (ret)
3505 goto cleanup;
3506 if (xis.not_found) {
3507 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3508 if (ret)
3509 goto cleanup;
3510 }
3511
3512 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3513
3514 cleanup:
3515 up_write(&OCFS2_I(inode)->ip_xattr_sem);
3516 brelse(xbs.xattr_bh);
3517 ocfs2_xattr_bucket_free(xbs.bucket);
3518
3519 return ret;
3520 }
3521
3522 /*
3523 * ocfs2_xattr_set()
3524 *
3525 * Set, replace or remove an extended attribute for this inode.
3526 * value is NULL to remove an existing extended attribute, else either
3527 * create or replace an extended attribute.
3528 */
3529 int ocfs2_xattr_set(struct inode *inode,
3530 int name_index,
3531 const char *name,
3532 const void *value,
3533 size_t value_len,
3534 int flags)
3535 {
3536 struct buffer_head *di_bh = NULL;
3537 struct ocfs2_dinode *di;
3538 int ret, credits, had_lock, ref_meta = 0, ref_credits = 0;
3539 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3540 struct inode *tl_inode = osb->osb_tl_inode;
3541 struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, NULL, };
3542 struct ocfs2_refcount_tree *ref_tree = NULL;
3543 struct ocfs2_lock_holder oh;
3544
3545 struct ocfs2_xattr_info xi = {
3546 .xi_name_index = name_index,
3547 .xi_name = name,
3548 .xi_name_len = strlen(name),
3549 .xi_value = value,
3550 .xi_value_len = value_len,
3551 };
3552
3553 struct ocfs2_xattr_search xis = {
3554 .not_found = -ENODATA,
3555 };
3556
3557 struct ocfs2_xattr_search xbs = {
3558 .not_found = -ENODATA,
3559 };
3560
3561 if (!ocfs2_supports_xattr(osb))
3562 return -EOPNOTSUPP;
3563
3564 /*
3565 * Only xbs will be used on indexed trees. xis doesn't need a
3566 * bucket.
3567 */
3568 xbs.bucket = ocfs2_xattr_bucket_new(inode);
3569 if (!xbs.bucket) {
3570 mlog_errno(-ENOMEM);
3571 return -ENOMEM;
3572 }
3573
3574 had_lock = ocfs2_inode_lock_tracker(inode, &di_bh, 1, &oh);
3575 if (had_lock < 0) {
3576 ret = had_lock;
3577 mlog_errno(ret);
3578 goto cleanup_nolock;
3579 }
3580 xis.inode_bh = xbs.inode_bh = di_bh;
3581 di = (struct ocfs2_dinode *)di_bh->b_data;
3582
3583 down_write(&OCFS2_I(inode)->ip_xattr_sem);
3584 /*
3585 * Scan inode and external block to find the same name
3586 * extended attribute and collect search information.
3587 */
3588 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3589 if (ret)
3590 goto cleanup;
3591 if (xis.not_found) {
3592 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3593 if (ret)
3594 goto cleanup;
3595 }
3596
3597 if (xis.not_found && xbs.not_found) {
3598 ret = -ENODATA;
3599 if (flags & XATTR_REPLACE)
3600 goto cleanup;
3601 ret = 0;
3602 if (!value)
3603 goto cleanup;
3604 } else {
3605 ret = -EEXIST;
3606 if (flags & XATTR_CREATE)
3607 goto cleanup;
3608 }
3609
3610 /* Check whether the value is refcounted and do some preparation. */
3611 if (ocfs2_is_refcount_inode(inode) &&
3612 (!xis.not_found || !xbs.not_found)) {
3613 ret = ocfs2_prepare_refcount_xattr(inode, di, &xi,
3614 &xis, &xbs, &ref_tree,
3615 &ref_meta, &ref_credits);
3616 if (ret) {
3617 mlog_errno(ret);
3618 goto cleanup;
3619 }
3620 }
3621
3622 inode_lock(tl_inode);
3623
3624 if (ocfs2_truncate_log_needs_flush(osb)) {
3625 ret = __ocfs2_flush_truncate_log(osb);
3626 if (ret < 0) {
3627 inode_unlock(tl_inode);
3628 mlog_errno(ret);
3629 goto cleanup;
3630 }
3631 }
3632 inode_unlock(tl_inode);
3633
3634 ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
3635 &xbs, &ctxt, ref_meta, &credits);
3636 if (ret) {
3637 mlog_errno(ret);
3638 goto cleanup;
3639 }
3640
3641 /* we need to update inode's ctime field, so add credit for it. */
3642 credits += OCFS2_INODE_UPDATE_CREDITS;
3643 ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
3644 if (IS_ERR(ctxt.handle)) {
3645 ret = PTR_ERR(ctxt.handle);
3646 mlog_errno(ret);
3647 goto out_free_ac;
3648 }
3649
3650 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3651 ocfs2_update_inode_fsync_trans(ctxt.handle, inode, 0);
3652
3653 ocfs2_commit_trans(osb, ctxt.handle);
3654
3655 out_free_ac:
3656 if (ctxt.data_ac)
3657 ocfs2_free_alloc_context(ctxt.data_ac);
3658 if (ctxt.meta_ac)
3659 ocfs2_free_alloc_context(ctxt.meta_ac);
3660 if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
3661 ocfs2_schedule_truncate_log_flush(osb, 1);
3662 ocfs2_run_deallocs(osb, &ctxt.dealloc);
3663
3664 cleanup:
3665 if (ref_tree)
3666 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3667 up_write(&OCFS2_I(inode)->ip_xattr_sem);
3668 if (!value && !ret) {
3669 ret = ocfs2_try_remove_refcount_tree(inode, di_bh);
3670 if (ret)
3671 mlog_errno(ret);
3672 }
3673 ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
3674 cleanup_nolock:
3675 brelse(di_bh);
3676 brelse(xbs.xattr_bh);
3677 ocfs2_xattr_bucket_free(xbs.bucket);
3678
3679 return ret;
3680 }
3681
3682 /*
3683 * Find the xattr extent rec which may contains name_hash.
3684 * e_cpos will be the first name hash of the xattr rec.
3685 * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
3686 */
3687 static int ocfs2_xattr_get_rec(struct inode *inode,
3688 u32 name_hash,
3689 u64 *p_blkno,
3690 u32 *e_cpos,
3691 u32 *num_clusters,
3692 struct ocfs2_extent_list *el)
3693 {
3694 int ret = 0, i;
3695 struct buffer_head *eb_bh = NULL;
3696 struct ocfs2_extent_block *eb;
3697 struct ocfs2_extent_rec *rec = NULL;
3698 u64 e_blkno = 0;
3699
3700 if (el->l_tree_depth) {
3701 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, name_hash,
3702 &eb_bh);
3703 if (ret) {
3704 mlog_errno(ret);
3705 goto out;
3706 }
3707
3708 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
3709 el = &eb->h_list;
3710
3711 if (el->l_tree_depth) {
3712 ret = ocfs2_error(inode->i_sb,
3713 "Inode %lu has non zero tree depth in xattr tree block %llu\n",
3714 inode->i_ino,
3715 (unsigned long long)eb_bh->b_blocknr);
3716 goto out;
3717 }
3718 }
3719
3720 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
3721 rec = &el->l_recs[i];
3722
3723 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
3724 e_blkno = le64_to_cpu(rec->e_blkno);
3725 break;
3726 }
3727 }
3728
3729 if (!e_blkno) {
3730 ret = ocfs2_error(inode->i_sb, "Inode %lu has bad extent record (%u, %u, 0) in xattr\n",
3731 inode->i_ino,
3732 le32_to_cpu(rec->e_cpos),
3733 ocfs2_rec_clusters(el, rec));
3734 goto out;
3735 }
3736
3737 *p_blkno = le64_to_cpu(rec->e_blkno);
3738 *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
3739 if (e_cpos)
3740 *e_cpos = le32_to_cpu(rec->e_cpos);
3741 out:
3742 brelse(eb_bh);
3743 return ret;
3744 }
3745
3746 typedef int (xattr_bucket_func)(struct inode *inode,
3747 struct ocfs2_xattr_bucket *bucket,
3748 void *para);
3749
3750 static int ocfs2_find_xe_in_bucket(struct inode *inode,
3751 struct ocfs2_xattr_bucket *bucket,
3752 int name_index,
3753 const char *name,
3754 u32 name_hash,
3755 u16 *xe_index,
3756 int *found)
3757 {
3758 int i, ret = 0, cmp = 1, block_off, new_offset;
3759 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3760 size_t name_len = strlen(name);
3761 struct ocfs2_xattr_entry *xe = NULL;
3762 char *xe_name;
3763
3764 /*
3765 * We don't use binary search in the bucket because there
3766 * may be multiple entries with the same name hash.
3767 */
3768 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3769 xe = &xh->xh_entries[i];
3770
3771 if (name_hash > le32_to_cpu(xe->xe_name_hash))
3772 continue;
3773 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
3774 break;
3775
3776 cmp = name_index - ocfs2_xattr_get_type(xe);
3777 if (!cmp)
3778 cmp = name_len - xe->xe_name_len;
3779 if (cmp)
3780 continue;
3781
3782 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3783 xh,
3784 i,
3785 &block_off,
3786 &new_offset);
3787 if (ret) {
3788 mlog_errno(ret);
3789 break;
3790 }
3791
3792
3793 xe_name = bucket_block(bucket, block_off) + new_offset;
3794 if (!memcmp(name, xe_name, name_len)) {
3795 *xe_index = i;
3796 *found = 1;
3797 ret = 0;
3798 break;
3799 }
3800 }
3801
3802 return ret;
3803 }
3804
3805 /*
3806 * Find the specified xattr entry in a series of buckets.
3807 * This series start from p_blkno and last for num_clusters.
3808 * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
3809 * the num of the valid buckets.
3810 *
3811 * Return the buffer_head this xattr should reside in. And if the xattr's
3812 * hash is in the gap of 2 buckets, return the lower bucket.
3813 */
3814 static int ocfs2_xattr_bucket_find(struct inode *inode,
3815 int name_index,
3816 const char *name,
3817 u32 name_hash,
3818 u64 p_blkno,
3819 u32 first_hash,
3820 u32 num_clusters,
3821 struct ocfs2_xattr_search *xs)
3822 {
3823 int ret, found = 0;
3824 struct ocfs2_xattr_header *xh = NULL;
3825 struct ocfs2_xattr_entry *xe = NULL;
3826 u16 index = 0;
3827 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3828 int low_bucket = 0, bucket, high_bucket;
3829 struct ocfs2_xattr_bucket *search;
3830 u64 blkno, lower_blkno = 0;
3831
3832 search = ocfs2_xattr_bucket_new(inode);
3833 if (!search) {
3834 ret = -ENOMEM;
3835 mlog_errno(ret);
3836 goto out;
3837 }
3838
3839 ret = ocfs2_read_xattr_bucket(search, p_blkno);
3840 if (ret) {
3841 mlog_errno(ret);
3842 goto out;
3843 }
3844
3845 xh = bucket_xh(search);
3846 high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
3847 while (low_bucket <= high_bucket) {
3848 ocfs2_xattr_bucket_relse(search);
3849
3850 bucket = (low_bucket + high_bucket) / 2;
3851 blkno = p_blkno + bucket * blk_per_bucket;
3852 ret = ocfs2_read_xattr_bucket(search, blkno);
3853 if (ret) {
3854 mlog_errno(ret);
3855 goto out;
3856 }
3857
3858 xh = bucket_xh(search);
3859 xe = &xh->xh_entries[0];
3860 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3861 high_bucket = bucket - 1;
3862 continue;
3863 }
3864
3865 /*
3866 * Check whether the hash of the last entry in our
3867 * bucket is larger than the search one. for an empty
3868 * bucket, the last one is also the first one.
3869 */
3870 if (xh->xh_count)
3871 xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3872
3873 /* record lower_blkno which may be the insert place. */
3874 lower_blkno = blkno;
3875
3876 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3877 low_bucket = bucket + 1;
3878 continue;
3879 }
3880
3881 /* the searched xattr should reside in this bucket if exists. */
3882 ret = ocfs2_find_xe_in_bucket(inode, search,
3883 name_index, name, name_hash,
3884 &index, &found);
3885 if (ret) {
3886 mlog_errno(ret);
3887 goto out;
3888 }
3889 break;
3890 }
3891
3892 /*
3893 * Record the bucket we have found.
3894 * When the xattr's hash value is in the gap of 2 buckets, we will
3895 * always set it to the previous bucket.
3896 */
3897 if (!lower_blkno)
3898 lower_blkno = p_blkno;
3899
3900 /* This should be in cache - we just read it during the search */
3901 ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3902 if (ret) {
3903 mlog_errno(ret);
3904 goto out;
3905 }
3906
3907 xs->header = bucket_xh(xs->bucket);
3908 xs->base = bucket_block(xs->bucket, 0);
3909 xs->end = xs->base + inode->i_sb->s_blocksize;
3910
3911 if (found) {
3912 xs->here = &xs->header->xh_entries[index];
3913 trace_ocfs2_xattr_bucket_find(OCFS2_I(inode)->ip_blkno,
3914 name, name_index, name_hash,
3915 (unsigned long long)bucket_blkno(xs->bucket),
3916 index);
3917 } else
3918 ret = -ENODATA;
3919
3920 out:
3921 ocfs2_xattr_bucket_free(search);
3922 return ret;
3923 }
3924
3925 static int ocfs2_xattr_index_block_find(struct inode *inode,
3926 struct buffer_head *root_bh,
3927 int name_index,
3928 const char *name,
3929 struct ocfs2_xattr_search *xs)
3930 {
3931 int ret;
3932 struct ocfs2_xattr_block *xb =
3933 (struct ocfs2_xattr_block *)root_bh->b_data;
3934 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3935 struct ocfs2_extent_list *el = &xb_root->xt_list;
3936 u64 p_blkno = 0;
3937 u32 first_hash, num_clusters = 0;
3938 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3939
3940 if (le16_to_cpu(el->l_next_free_rec) == 0)
3941 return -ENODATA;
3942
3943 trace_ocfs2_xattr_index_block_find(OCFS2_I(inode)->ip_blkno,
3944 name, name_index, name_hash,
3945 (unsigned long long)root_bh->b_blocknr,
3946 -1);
3947
3948 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3949 &num_clusters, el);
3950 if (ret) {
3951 mlog_errno(ret);
3952 goto out;
3953 }
3954
3955 BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3956
3957 trace_ocfs2_xattr_index_block_find_rec(OCFS2_I(inode)->ip_blkno,
3958 name, name_index, first_hash,
3959 (unsigned long long)p_blkno,
3960 num_clusters);
3961
3962 ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3963 p_blkno, first_hash, num_clusters, xs);
3964
3965 out:
3966 return ret;
3967 }
3968
3969 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3970 u64 blkno,
3971 u32 clusters,
3972 xattr_bucket_func *func,
3973 void *para)
3974 {
3975 int i, ret = 0;
3976 u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3977 u32 num_buckets = clusters * bpc;
3978 struct ocfs2_xattr_bucket *bucket;
3979
3980 bucket = ocfs2_xattr_bucket_new(inode);
3981 if (!bucket) {
3982 mlog_errno(-ENOMEM);
3983 return -ENOMEM;
3984 }
3985
3986 trace_ocfs2_iterate_xattr_buckets(
3987 (unsigned long long)OCFS2_I(inode)->ip_blkno,
3988 (unsigned long long)blkno, clusters);
3989
3990 for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3991 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3992 if (ret) {
3993 mlog_errno(ret);
3994 break;
3995 }
3996
3997 /*
3998 * The real bucket num in this series of blocks is stored
3999 * in the 1st bucket.
4000 */
4001 if (i == 0)
4002 num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
4003
4004 trace_ocfs2_iterate_xattr_bucket((unsigned long long)blkno,
4005 le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
4006 if (func) {
4007 ret = func(inode, bucket, para);
4008 if (ret && ret != -ERANGE)
4009 mlog_errno(ret);
4010 /* Fall through to bucket_relse() */
4011 }
4012
4013 ocfs2_xattr_bucket_relse(bucket);
4014 if (ret)
4015 break;
4016 }
4017
4018 ocfs2_xattr_bucket_free(bucket);
4019 return ret;
4020 }
4021
4022 struct ocfs2_xattr_tree_list {
4023 char *buffer;
4024 size_t buffer_size;
4025 size_t result;
4026 };
4027
4028 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
4029 struct ocfs2_xattr_header *xh,
4030 int index,
4031 int *block_off,
4032 int *new_offset)
4033 {
4034 u16 name_offset;
4035
4036 if (index < 0 || index >= le16_to_cpu(xh->xh_count))
4037 return -EINVAL;
4038
4039 name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
4040
4041 *block_off = name_offset >> sb->s_blocksize_bits;
4042 *new_offset = name_offset % sb->s_blocksize;
4043
4044 return 0;
4045 }
4046
4047 static int ocfs2_list_xattr_bucket(struct inode *inode,
4048 struct ocfs2_xattr_bucket *bucket,
4049 void *para)
4050 {
4051 int ret = 0, type;
4052 struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
4053 int i, block_off, new_offset;
4054 const char *name;
4055
4056 for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
4057 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
4058 type = ocfs2_xattr_get_type(entry);
4059
4060 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
4061 bucket_xh(bucket),
4062 i,
4063 &block_off,
4064 &new_offset);
4065 if (ret)
4066 break;
4067
4068 name = (const char *)bucket_block(bucket, block_off) +
4069 new_offset;
4070 ret = ocfs2_xattr_list_entry(inode->i_sb,
4071 xl->buffer,
4072 xl->buffer_size,
4073 &xl->result,
4074 type, name,
4075 entry->xe_name_len);
4076 if (ret)
4077 break;
4078 }
4079
4080 return ret;
4081 }
4082
4083 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
4084 struct buffer_head *blk_bh,
4085 xattr_tree_rec_func *rec_func,
4086 void *para)
4087 {
4088 struct ocfs2_xattr_block *xb =
4089 (struct ocfs2_xattr_block *)blk_bh->b_data;
4090 struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4091 int ret = 0;
4092 u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
4093 u64 p_blkno = 0;
4094
4095 if (!el->l_next_free_rec || !rec_func)
4096 return 0;
4097
4098 while (name_hash > 0) {
4099 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4100 &e_cpos, &num_clusters, el);
4101 if (ret) {
4102 mlog_errno(ret);
4103 break;
4104 }
4105
4106 ret = rec_func(inode, blk_bh, p_blkno, e_cpos,
4107 num_clusters, para);
4108 if (ret) {
4109 if (ret != -ERANGE)
4110 mlog_errno(ret);
4111 break;
4112 }
4113
4114 if (e_cpos == 0)
4115 break;
4116
4117 name_hash = e_cpos - 1;
4118 }
4119
4120 return ret;
4121
4122 }
4123
4124 static int ocfs2_list_xattr_tree_rec(struct inode *inode,
4125 struct buffer_head *root_bh,
4126 u64 blkno, u32 cpos, u32 len, void *para)
4127 {
4128 return ocfs2_iterate_xattr_buckets(inode, blkno, len,
4129 ocfs2_list_xattr_bucket, para);
4130 }
4131
4132 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
4133 struct buffer_head *blk_bh,
4134 char *buffer,
4135 size_t buffer_size)
4136 {
4137 int ret;
4138 struct ocfs2_xattr_tree_list xl = {
4139 .buffer = buffer,
4140 .buffer_size = buffer_size,
4141 .result = 0,
4142 };
4143
4144 ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
4145 ocfs2_list_xattr_tree_rec, &xl);
4146 if (ret) {
4147 mlog_errno(ret);
4148 goto out;
4149 }
4150
4151 ret = xl.result;
4152 out:
4153 return ret;
4154 }
4155
4156 static int cmp_xe(const void *a, const void *b)
4157 {
4158 const struct ocfs2_xattr_entry *l = a, *r = b;
4159 u32 l_hash = le32_to_cpu(l->xe_name_hash);
4160 u32 r_hash = le32_to_cpu(r->xe_name_hash);
4161
4162 if (l_hash > r_hash)
4163 return 1;
4164 if (l_hash < r_hash)
4165 return -1;
4166 return 0;
4167 }
4168
4169 /*
4170 * When the ocfs2_xattr_block is filled up, new bucket will be created
4171 * and all the xattr entries will be moved to the new bucket.
4172 * The header goes at the start of the bucket, and the names+values are
4173 * filled from the end. This is why *target starts as the last buffer.
4174 * Note: we need to sort the entries since they are not saved in order
4175 * in the ocfs2_xattr_block.
4176 */
4177 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
4178 struct buffer_head *xb_bh,
4179 struct ocfs2_xattr_bucket *bucket)
4180 {
4181 int i, blocksize = inode->i_sb->s_blocksize;
4182 int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4183 u16 offset, size, off_change;
4184 struct ocfs2_xattr_entry *xe;
4185 struct ocfs2_xattr_block *xb =
4186 (struct ocfs2_xattr_block *)xb_bh->b_data;
4187 struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
4188 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4189 u16 count = le16_to_cpu(xb_xh->xh_count);
4190 char *src = xb_bh->b_data;
4191 char *target = bucket_block(bucket, blks - 1);
4192
4193 trace_ocfs2_cp_xattr_block_to_bucket_begin(
4194 (unsigned long long)xb_bh->b_blocknr,
4195 (unsigned long long)bucket_blkno(bucket));
4196
4197 for (i = 0; i < blks; i++)
4198 memset(bucket_block(bucket, i), 0, blocksize);
4199
4200 /*
4201 * Since the xe_name_offset is based on ocfs2_xattr_header,
4202 * there is a offset change corresponding to the change of
4203 * ocfs2_xattr_header's position.
4204 */
4205 off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4206 xe = &xb_xh->xh_entries[count - 1];
4207 offset = le16_to_cpu(xe->xe_name_offset) + off_change;
4208 size = blocksize - offset;
4209
4210 /* copy all the names and values. */
4211 memcpy(target + offset, src + offset, size);
4212
4213 /* Init new header now. */
4214 xh->xh_count = xb_xh->xh_count;
4215 xh->xh_num_buckets = cpu_to_le16(1);
4216 xh->xh_name_value_len = cpu_to_le16(size);
4217 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
4218
4219 /* copy all the entries. */
4220 target = bucket_block(bucket, 0);
4221 offset = offsetof(struct ocfs2_xattr_header, xh_entries);
4222 size = count * sizeof(struct ocfs2_xattr_entry);
4223 memcpy(target + offset, (char *)xb_xh + offset, size);
4224
4225 /* Change the xe offset for all the xe because of the move. */
4226 off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
4227 offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4228 for (i = 0; i < count; i++)
4229 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
4230
4231 trace_ocfs2_cp_xattr_block_to_bucket_end(offset, size, off_change);
4232
4233 sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
4234 cmp_xe, NULL);
4235 }
4236
4237 /*
4238 * After we move xattr from block to index btree, we have to
4239 * update ocfs2_xattr_search to the new xe and base.
4240 *
4241 * When the entry is in xattr block, xattr_bh indicates the storage place.
4242 * While if the entry is in index b-tree, "bucket" indicates the
4243 * real place of the xattr.
4244 */
4245 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
4246 struct ocfs2_xattr_search *xs,
4247 struct buffer_head *old_bh)
4248 {
4249 char *buf = old_bh->b_data;
4250 struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
4251 struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
4252 int i;
4253
4254 xs->header = bucket_xh(xs->bucket);
4255 xs->base = bucket_block(xs->bucket, 0);
4256 xs->end = xs->base + inode->i_sb->s_blocksize;
4257
4258 if (xs->not_found)
4259 return;
4260
4261 i = xs->here - old_xh->xh_entries;
4262 xs->here = &xs->header->xh_entries[i];
4263 }
4264
4265 static int ocfs2_xattr_create_index_block(struct inode *inode,
4266 struct ocfs2_xattr_search *xs,
4267 struct ocfs2_xattr_set_ctxt *ctxt)
4268 {
4269 int ret;
4270 u32 bit_off, len;
4271 u64 blkno;
4272 handle_t *handle = ctxt->handle;
4273 struct ocfs2_inode_info *oi = OCFS2_I(inode);
4274 struct buffer_head *xb_bh = xs->xattr_bh;
4275 struct ocfs2_xattr_block *xb =
4276 (struct ocfs2_xattr_block *)xb_bh->b_data;
4277 struct ocfs2_xattr_tree_root *xr;
4278 u16 xb_flags = le16_to_cpu(xb->xb_flags);
4279
4280 trace_ocfs2_xattr_create_index_block_begin(
4281 (unsigned long long)xb_bh->b_blocknr);
4282
4283 BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
4284 BUG_ON(!xs->bucket);
4285
4286 /*
4287 * XXX:
4288 * We can use this lock for now, and maybe move to a dedicated mutex
4289 * if performance becomes a problem later.
4290 */
4291 down_write(&oi->ip_alloc_sem);
4292
4293 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), xb_bh,
4294 OCFS2_JOURNAL_ACCESS_WRITE);
4295 if (ret) {
4296 mlog_errno(ret);
4297 goto out;
4298 }
4299
4300 ret = __ocfs2_claim_clusters(handle, ctxt->data_ac,
4301 1, 1, &bit_off, &len);
4302 if (ret) {
4303 mlog_errno(ret);
4304 goto out;
4305 }
4306
4307 /*
4308 * The bucket may spread in many blocks, and
4309 * we will only touch the 1st block and the last block
4310 * in the whole bucket(one for entry and one for data).
4311 */
4312 blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
4313
4314 trace_ocfs2_xattr_create_index_block((unsigned long long)blkno);
4315
4316 ret = ocfs2_init_xattr_bucket(xs->bucket, blkno, 1);
4317 if (ret) {
4318 mlog_errno(ret);
4319 goto out;
4320 }
4321
4322 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4323 OCFS2_JOURNAL_ACCESS_CREATE);
4324 if (ret) {
4325 mlog_errno(ret);
4326 goto out;
4327 }
4328
4329 ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
4330 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4331
4332 ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
4333
4334 /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
4335 memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
4336 offsetof(struct ocfs2_xattr_block, xb_attrs));
4337
4338 xr = &xb->xb_attrs.xb_root;
4339 xr->xt_clusters = cpu_to_le32(1);
4340 xr->xt_last_eb_blk = 0;
4341 xr->xt_list.l_tree_depth = 0;
4342 xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
4343 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
4344
4345 xr->xt_list.l_recs[0].e_cpos = 0;
4346 xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
4347 xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
4348
4349 xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
4350
4351 ocfs2_journal_dirty(handle, xb_bh);
4352
4353 out:
4354 up_write(&oi->ip_alloc_sem);
4355
4356 return ret;
4357 }
4358
4359 static int cmp_xe_offset(const void *a, const void *b)
4360 {
4361 const struct ocfs2_xattr_entry *l = a, *r = b;
4362 u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
4363 u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
4364
4365 if (l_name_offset < r_name_offset)
4366 return 1;
4367 if (l_name_offset > r_name_offset)
4368 return -1;
4369 return 0;
4370 }
4371
4372 /*
4373 * defrag a xattr bucket if we find that the bucket has some
4374 * holes beteen name/value pairs.
4375 * We will move all the name/value pairs to the end of the bucket
4376 * so that we can spare some space for insertion.
4377 */
4378 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
4379 handle_t *handle,
4380 struct ocfs2_xattr_bucket *bucket)
4381 {
4382 int ret, i;
4383 size_t end, offset, len;
4384 struct ocfs2_xattr_header *xh;
4385 char *entries, *buf, *bucket_buf = NULL;
4386 u64 blkno = bucket_blkno(bucket);
4387 u16 xh_free_start;
4388 size_t blocksize = inode->i_sb->s_blocksize;
4389 struct ocfs2_xattr_entry *xe;
4390
4391 /*
4392 * In order to make the operation more efficient and generic,
4393 * we copy all the blocks into a contiguous memory and do the
4394 * defragment there, so if anything is error, we will not touch
4395 * the real block.
4396 */
4397 bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
4398 if (!bucket_buf) {
4399 ret = -EIO;
4400 goto out;
4401 }
4402
4403 buf = bucket_buf;
4404 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4405 memcpy(buf, bucket_block(bucket, i), blocksize);
4406
4407 ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
4408 OCFS2_JOURNAL_ACCESS_WRITE);
4409 if (ret < 0) {
4410 mlog_errno(ret);
4411 goto out;
4412 }
4413
4414 xh = (struct ocfs2_xattr_header *)bucket_buf;
4415 entries = (char *)xh->xh_entries;
4416 xh_free_start = le16_to_cpu(xh->xh_free_start);
4417
4418 trace_ocfs2_defrag_xattr_bucket(
4419 (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
4420 xh_free_start, le16_to_cpu(xh->xh_name_value_len));
4421
4422 /*
4423 * sort all the entries by their offset.
4424 * the largest will be the first, so that we can
4425 * move them to the end one by one.
4426 */
4427 sort(entries, le16_to_cpu(xh->xh_count),
4428 sizeof(struct ocfs2_xattr_entry),
4429 cmp_xe_offset, NULL);
4430
4431 /* Move all name/values to the end of the bucket. */
4432 xe = xh->xh_entries;
4433 end = OCFS2_XATTR_BUCKET_SIZE;
4434 for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
4435 offset = le16_to_cpu(xe->xe_name_offset);
4436 len = namevalue_size_xe(xe);
4437
4438 /*
4439 * We must make sure that the name/value pair
4440 * exist in the same block. So adjust end to
4441 * the previous block end if needed.
4442 */
4443 if (((end - len) / blocksize !=
4444 (end - 1) / blocksize))
4445 end = end - end % blocksize;
4446
4447 if (end > offset + len) {
4448 memmove(bucket_buf + end - len,
4449 bucket_buf + offset, len);
4450 xe->xe_name_offset = cpu_to_le16(end - len);
4451 }
4452
4453 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
4454 "bucket %llu\n", (unsigned long long)blkno);
4455
4456 end -= len;
4457 }
4458
4459 mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
4460 "bucket %llu\n", (unsigned long long)blkno);
4461
4462 if (xh_free_start == end)
4463 goto out;
4464
4465 memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
4466 xh->xh_free_start = cpu_to_le16(end);
4467
4468 /* sort the entries by their name_hash. */
4469 sort(entries, le16_to_cpu(xh->xh_count),
4470 sizeof(struct ocfs2_xattr_entry),
4471 cmp_xe, NULL);
4472
4473 buf = bucket_buf;
4474 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4475 memcpy(bucket_block(bucket, i), buf, blocksize);
4476 ocfs2_xattr_bucket_journal_dirty(handle, bucket);
4477
4478 out:
4479 kfree(bucket_buf);
4480 return ret;
4481 }
4482
4483 /*
4484 * prev_blkno points to the start of an existing extent. new_blkno
4485 * points to a newly allocated extent. Because we know each of our
4486 * clusters contains more than bucket, we can easily split one cluster
4487 * at a bucket boundary. So we take the last cluster of the existing
4488 * extent and split it down the middle. We move the last half of the
4489 * buckets in the last cluster of the existing extent over to the new
4490 * extent.
4491 *
4492 * first_bh is the buffer at prev_blkno so we can update the existing
4493 * extent's bucket count. header_bh is the bucket were we were hoping
4494 * to insert our xattr. If the bucket move places the target in the new
4495 * extent, we'll update first_bh and header_bh after modifying the old
4496 * extent.
4497 *
4498 * first_hash will be set as the 1st xe's name_hash in the new extent.
4499 */
4500 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
4501 handle_t *handle,
4502 struct ocfs2_xattr_bucket *first,
4503 struct ocfs2_xattr_bucket *target,
4504 u64 new_blkno,
4505 u32 num_clusters,
4506 u32 *first_hash)
4507 {
4508 int ret;
4509 struct super_block *sb = inode->i_sb;
4510 int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
4511 int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
4512 int to_move = num_buckets / 2;
4513 u64 src_blkno;
4514 u64 last_cluster_blkno = bucket_blkno(first) +
4515 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
4516
4517 BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
4518 BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
4519
4520 trace_ocfs2_mv_xattr_bucket_cross_cluster(
4521 (unsigned long long)last_cluster_blkno,
4522 (unsigned long long)new_blkno);
4523
4524 ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
4525 last_cluster_blkno, new_blkno,
4526 to_move, first_hash);
4527 if (ret) {
4528 mlog_errno(ret);
4529 goto out;
4530 }
4531
4532 /* This is the first bucket that got moved */
4533 src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
4534
4535 /*
4536 * If the target bucket was part of the moved buckets, we need to
4537 * update first and target.
4538 */
4539 if (bucket_blkno(target) >= src_blkno) {
4540 /* Find the block for the new target bucket */
4541 src_blkno = new_blkno +
4542 (bucket_blkno(target) - src_blkno);
4543
4544 ocfs2_xattr_bucket_relse(first);
4545 ocfs2_xattr_bucket_relse(target);
4546
4547 /*
4548 * These shouldn't fail - the buffers are in the
4549 * journal from ocfs2_cp_xattr_bucket().
4550 */
4551 ret = ocfs2_read_xattr_bucket(first, new_blkno);
4552 if (ret) {
4553 mlog_errno(ret);
4554 goto out;
4555 }
4556 ret = ocfs2_read_xattr_bucket(target, src_blkno);
4557 if (ret)
4558 mlog_errno(ret);
4559
4560 }
4561
4562 out:
4563 return ret;
4564 }
4565
4566 /*
4567 * Find the suitable pos when we divide a bucket into 2.
4568 * We have to make sure the xattrs with the same hash value exist
4569 * in the same bucket.
4570 *
4571 * If this ocfs2_xattr_header covers more than one hash value, find a
4572 * place where the hash value changes. Try to find the most even split.
4573 * The most common case is that all entries have different hash values,
4574 * and the first check we make will find a place to split.
4575 */
4576 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
4577 {
4578 struct ocfs2_xattr_entry *entries = xh->xh_entries;
4579 int count = le16_to_cpu(xh->xh_count);
4580 int delta, middle = count / 2;
4581
4582 /*
4583 * We start at the middle. Each step gets farther away in both
4584 * directions. We therefore hit the change in hash value
4585 * nearest to the middle. Note that this loop does not execute for
4586 * count < 2.
4587 */
4588 for (delta = 0; delta < middle; delta++) {
4589 /* Let's check delta earlier than middle */
4590 if (cmp_xe(&entries[middle - delta - 1],
4591 &entries[middle - delta]))
4592 return middle - delta;
4593
4594 /* For even counts, don't walk off the end */
4595 if ((middle + delta + 1) == count)
4596 continue;
4597
4598 /* Now try delta past middle */
4599 if (cmp_xe(&entries[middle + delta],
4600 &entries[middle + delta + 1]))
4601 return middle + delta + 1;
4602 }
4603
4604 /* Every entry had the same hash */
4605 return count;
4606 }
4607
4608 /*
4609 * Move some xattrs in old bucket(blk) to new bucket(new_blk).
4610 * first_hash will record the 1st hash of the new bucket.
4611 *
4612 * Normally half of the xattrs will be moved. But we have to make
4613 * sure that the xattrs with the same hash value are stored in the
4614 * same bucket. If all the xattrs in this bucket have the same hash
4615 * value, the new bucket will be initialized as an empty one and the
4616 * first_hash will be initialized as (hash_value+1).
4617 */
4618 static int ocfs2_divide_xattr_bucket(struct inode *inode,
4619 handle_t *handle,
4620 u64 blk,
4621 u64 new_blk,
4622 u32 *first_hash,
4623 int new_bucket_head)
4624 {
4625 int ret, i;
4626 int count, start, len, name_value_len = 0, name_offset = 0;
4627 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4628 struct ocfs2_xattr_header *xh;
4629 struct ocfs2_xattr_entry *xe;
4630 int blocksize = inode->i_sb->s_blocksize;
4631
4632 trace_ocfs2_divide_xattr_bucket_begin((unsigned long long)blk,
4633 (unsigned long long)new_blk);
4634
4635 s_bucket = ocfs2_xattr_bucket_new(inode);
4636 t_bucket = ocfs2_xattr_bucket_new(inode);
4637 if (!s_bucket || !t_bucket) {
4638 ret = -ENOMEM;
4639 mlog_errno(ret);
4640 goto out;
4641 }
4642
4643 ret = ocfs2_read_xattr_bucket(s_bucket, blk);
4644 if (ret) {
4645 mlog_errno(ret);
4646 goto out;
4647 }
4648
4649 ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
4650 OCFS2_JOURNAL_ACCESS_WRITE);
4651 if (ret) {
4652 mlog_errno(ret);
4653 goto out;
4654 }
4655
4656 /*
4657 * Even if !new_bucket_head, we're overwriting t_bucket. Thus,
4658 * there's no need to read it.
4659 */
4660 ret = ocfs2_init_xattr_bucket(t_bucket, new_blk, new_bucket_head);
4661 if (ret) {
4662 mlog_errno(ret);
4663 goto out;
4664 }
4665
4666 /*
4667 * Hey, if we're overwriting t_bucket, what difference does
4668 * ACCESS_CREATE vs ACCESS_WRITE make? See the comment in the
4669 * same part of ocfs2_cp_xattr_bucket().
4670 */
4671 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4672 new_bucket_head ?
4673 OCFS2_JOURNAL_ACCESS_CREATE :
4674 OCFS2_JOURNAL_ACCESS_WRITE);
4675 if (ret) {
4676 mlog_errno(ret);
4677 goto out;
4678 }
4679
4680 xh = bucket_xh(s_bucket);
4681 count = le16_to_cpu(xh->xh_count);
4682 start = ocfs2_xattr_find_divide_pos(xh);
4683
4684 if (start == count) {
4685 xe = &xh->xh_entries[start-1];
4686
4687 /*
4688 * initialized a new empty bucket here.
4689 * The hash value is set as one larger than
4690 * that of the last entry in the previous bucket.
4691 */
4692 for (i = 0; i < t_bucket->bu_blocks; i++)
4693 memset(bucket_block(t_bucket, i), 0, blocksize);
4694
4695 xh = bucket_xh(t_bucket);
4696 xh->xh_free_start = cpu_to_le16(blocksize);
4697 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
4698 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
4699
4700 goto set_num_buckets;
4701 }
4702
4703 /* copy the whole bucket to the new first. */
4704 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4705
4706 /* update the new bucket. */
4707 xh = bucket_xh(t_bucket);
4708
4709 /*
4710 * Calculate the total name/value len and xh_free_start for
4711 * the old bucket first.
4712 */
4713 name_offset = OCFS2_XATTR_BUCKET_SIZE;
4714 name_value_len = 0;
4715 for (i = 0; i < start; i++) {
4716 xe = &xh->xh_entries[i];
4717 name_value_len += namevalue_size_xe(xe);
4718 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
4719 name_offset = le16_to_cpu(xe->xe_name_offset);
4720 }
4721
4722 /*
4723 * Now begin the modification to the new bucket.
4724 *
4725 * In the new bucket, We just move the xattr entry to the beginning
4726 * and don't touch the name/value. So there will be some holes in the
4727 * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
4728 * called.
4729 */
4730 xe = &xh->xh_entries[start];
4731 len = sizeof(struct ocfs2_xattr_entry) * (count - start);
4732 trace_ocfs2_divide_xattr_bucket_move(len,
4733 (int)((char *)xe - (char *)xh),
4734 (int)((char *)xh->xh_entries - (char *)xh));
4735 memmove((char *)xh->xh_entries, (char *)xe, len);
4736 xe = &xh->xh_entries[count - start];
4737 len = sizeof(struct ocfs2_xattr_entry) * start;
4738 memset((char *)xe, 0, len);
4739
4740 le16_add_cpu(&xh->xh_count, -start);
4741 le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
4742
4743 /* Calculate xh_free_start for the new bucket. */
4744 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4745 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4746 xe = &xh->xh_entries[i];
4747 if (le16_to_cpu(xe->xe_name_offset) <
4748 le16_to_cpu(xh->xh_free_start))
4749 xh->xh_free_start = xe->xe_name_offset;
4750 }
4751
4752 set_num_buckets:
4753 /* set xh->xh_num_buckets for the new xh. */
4754 if (new_bucket_head)
4755 xh->xh_num_buckets = cpu_to_le16(1);
4756 else
4757 xh->xh_num_buckets = 0;
4758
4759 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4760
4761 /* store the first_hash of the new bucket. */
4762 if (first_hash)
4763 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
4764
4765 /*
4766 * Now only update the 1st block of the old bucket. If we
4767 * just added a new empty bucket, there is no need to modify
4768 * it.
4769 */
4770 if (start == count)
4771 goto out;
4772
4773 xh = bucket_xh(s_bucket);
4774 memset(&xh->xh_entries[start], 0,
4775 sizeof(struct ocfs2_xattr_entry) * (count - start));
4776 xh->xh_count = cpu_to_le16(start);
4777 xh->xh_free_start = cpu_to_le16(name_offset);
4778 xh->xh_name_value_len = cpu_to_le16(name_value_len);
4779
4780 ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
4781
4782 out:
4783 ocfs2_xattr_bucket_free(s_bucket);
4784 ocfs2_xattr_bucket_free(t_bucket);
4785
4786 return ret;
4787 }
4788
4789 /*
4790 * Copy xattr from one bucket to another bucket.
4791 *
4792 * The caller must make sure that the journal transaction
4793 * has enough space for journaling.
4794 */
4795 static int ocfs2_cp_xattr_bucket(struct inode *inode,
4796 handle_t *handle,
4797 u64 s_blkno,
4798 u64 t_blkno,
4799 int t_is_new)
4800 {
4801 int ret;
4802 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4803
4804 BUG_ON(s_blkno == t_blkno);
4805
4806 trace_ocfs2_cp_xattr_bucket((unsigned long long)s_blkno,
4807 (unsigned long long)t_blkno,
4808 t_is_new);
4809
4810 s_bucket = ocfs2_xattr_bucket_new(inode);
4811 t_bucket = ocfs2_xattr_bucket_new(inode);
4812 if (!s_bucket || !t_bucket) {
4813 ret = -ENOMEM;
4814 mlog_errno(ret);
4815 goto out;
4816 }
4817
4818 ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
4819 if (ret)
4820 goto out;
4821
4822 /*
4823 * Even if !t_is_new, we're overwriting t_bucket. Thus,
4824 * there's no need to read it.
4825 */
4826 ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno, t_is_new);
4827 if (ret)
4828 goto out;
4829
4830 /*
4831 * Hey, if we're overwriting t_bucket, what difference does
4832 * ACCESS_CREATE vs ACCESS_WRITE make? Well, if we allocated a new
4833 * cluster to fill, we came here from
4834 * ocfs2_mv_xattr_buckets(), and it is really new -
4835 * ACCESS_CREATE is required. But we also might have moved data
4836 * out of t_bucket before extending back into it.
4837 * ocfs2_add_new_xattr_bucket() can do this - its call to
4838 * ocfs2_add_new_xattr_cluster() may have created a new extent
4839 * and copied out the end of the old extent. Then it re-extends
4840 * the old extent back to create space for new xattrs. That's
4841 * how we get here, and the bucket isn't really new.
4842 */
4843 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4844 t_is_new ?
4845 OCFS2_JOURNAL_ACCESS_CREATE :
4846 OCFS2_JOURNAL_ACCESS_WRITE);
4847 if (ret)
4848 goto out;
4849
4850 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4851 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4852
4853 out:
4854 ocfs2_xattr_bucket_free(t_bucket);
4855 ocfs2_xattr_bucket_free(s_bucket);
4856
4857 return ret;
4858 }
4859
4860 /*
4861 * src_blk points to the start of an existing extent. last_blk points to
4862 * last cluster in that extent. to_blk points to a newly allocated
4863 * extent. We copy the buckets from the cluster at last_blk to the new
4864 * extent. If start_bucket is non-zero, we skip that many buckets before
4865 * we start copying. The new extent's xh_num_buckets gets set to the
4866 * number of buckets we copied. The old extent's xh_num_buckets shrinks
4867 * by the same amount.
4868 */
4869 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4870 u64 src_blk, u64 last_blk, u64 to_blk,
4871 unsigned int start_bucket,
4872 u32 *first_hash)
4873 {
4874 int i, ret, credits;
4875 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4876 int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4877 int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4878 struct ocfs2_xattr_bucket *old_first, *new_first;
4879
4880 trace_ocfs2_mv_xattr_buckets((unsigned long long)last_blk,
4881 (unsigned long long)to_blk);
4882
4883 BUG_ON(start_bucket >= num_buckets);
4884 if (start_bucket) {
4885 num_buckets -= start_bucket;
4886 last_blk += (start_bucket * blks_per_bucket);
4887 }
4888
4889 /* The first bucket of the original extent */
4890 old_first = ocfs2_xattr_bucket_new(inode);
4891 /* The first bucket of the new extent */
4892 new_first = ocfs2_xattr_bucket_new(inode);
4893 if (!old_first || !new_first) {
4894 ret = -ENOMEM;
4895 mlog_errno(ret);
4896 goto out;
4897 }
4898
4899 ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4900 if (ret) {
4901 mlog_errno(ret);
4902 goto out;
4903 }
4904
4905 /*
4906 * We need to update the first bucket of the old extent and all
4907 * the buckets going to the new extent.
4908 */
4909 credits = ((num_buckets + 1) * blks_per_bucket);
4910 ret = ocfs2_extend_trans(handle, credits);
4911 if (ret) {
4912 mlog_errno(ret);
4913 goto out;
4914 }
4915
4916 ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4917 OCFS2_JOURNAL_ACCESS_WRITE);
4918 if (ret) {
4919 mlog_errno(ret);
4920 goto out;
4921 }
4922
4923 for (i = 0; i < num_buckets; i++) {
4924 ret = ocfs2_cp_xattr_bucket(inode, handle,
4925 last_blk + (i * blks_per_bucket),
4926 to_blk + (i * blks_per_bucket),
4927 1);
4928 if (ret) {
4929 mlog_errno(ret);
4930 goto out;
4931 }
4932 }
4933
4934 /*
4935 * Get the new bucket ready before we dirty anything
4936 * (This actually shouldn't fail, because we already dirtied
4937 * it once in ocfs2_cp_xattr_bucket()).
4938 */
4939 ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4940 if (ret) {
4941 mlog_errno(ret);
4942 goto out;
4943 }
4944 ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4945 OCFS2_JOURNAL_ACCESS_WRITE);
4946 if (ret) {
4947 mlog_errno(ret);
4948 goto out;
4949 }
4950
4951 /* Now update the headers */
4952 le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4953 ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4954
4955 bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4956 ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4957
4958 if (first_hash)
4959 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4960
4961 out:
4962 ocfs2_xattr_bucket_free(new_first);
4963 ocfs2_xattr_bucket_free(old_first);
4964 return ret;
4965 }
4966
4967 /*
4968 * Move some xattrs in this cluster to the new cluster.
4969 * This function should only be called when bucket size == cluster size.
4970 * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4971 */
4972 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4973 handle_t *handle,
4974 u64 prev_blk,
4975 u64 new_blk,
4976 u32 *first_hash)
4977 {
4978 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4979 int ret, credits = 2 * blk_per_bucket;
4980
4981 BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4982
4983 ret = ocfs2_extend_trans(handle, credits);
4984 if (ret) {
4985 mlog_errno(ret);
4986 return ret;
4987 }
4988
4989 /* Move half of the xattr in start_blk to the next bucket. */
4990 return ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4991 new_blk, first_hash, 1);
4992 }
4993
4994 /*
4995 * Move some xattrs from the old cluster to the new one since they are not
4996 * contiguous in ocfs2 xattr tree.
4997 *
4998 * new_blk starts a new separate cluster, and we will move some xattrs from
4999 * prev_blk to it. v_start will be set as the first name hash value in this
5000 * new cluster so that it can be used as e_cpos during tree insertion and
5001 * don't collide with our original b-tree operations. first_bh and header_bh
5002 * will also be updated since they will be used in ocfs2_extend_xattr_bucket
5003 * to extend the insert bucket.
5004 *
5005 * The problem is how much xattr should we move to the new one and when should
5006 * we update first_bh and header_bh?
5007 * 1. If cluster size > bucket size, that means the previous cluster has more
5008 * than 1 bucket, so just move half nums of bucket into the new cluster and
5009 * update the first_bh and header_bh if the insert bucket has been moved
5010 * to the new cluster.
5011 * 2. If cluster_size == bucket_size:
5012 * a) If the previous extent rec has more than one cluster and the insert
5013 * place isn't in the last cluster, copy the entire last cluster to the
5014 * new one. This time, we don't need to upate the first_bh and header_bh
5015 * since they will not be moved into the new cluster.
5016 * b) Otherwise, move the bottom half of the xattrs in the last cluster into
5017 * the new one. And we set the extend flag to zero if the insert place is
5018 * moved into the new allocated cluster since no extend is needed.
5019 */
5020 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
5021 handle_t *handle,
5022 struct ocfs2_xattr_bucket *first,
5023 struct ocfs2_xattr_bucket *target,
5024 u64 new_blk,
5025 u32 prev_clusters,
5026 u32 *v_start,
5027 int *extend)
5028 {
5029 int ret;
5030
5031 trace_ocfs2_adjust_xattr_cross_cluster(
5032 (unsigned long long)bucket_blkno(first),
5033 (unsigned long long)new_blk, prev_clusters);
5034
5035 if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
5036 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
5037 handle,
5038 first, target,
5039 new_blk,
5040 prev_clusters,
5041 v_start);
5042 if (ret)
5043 mlog_errno(ret);
5044 } else {
5045 /* The start of the last cluster in the first extent */
5046 u64 last_blk = bucket_blkno(first) +
5047 ((prev_clusters - 1) *
5048 ocfs2_clusters_to_blocks(inode->i_sb, 1));
5049
5050 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
5051 ret = ocfs2_mv_xattr_buckets(inode, handle,
5052 bucket_blkno(first),
5053 last_blk, new_blk, 0,
5054 v_start);
5055 if (ret)
5056 mlog_errno(ret);
5057 } else {
5058 ret = ocfs2_divide_xattr_cluster(inode, handle,
5059 last_blk, new_blk,
5060 v_start);
5061 if (ret)
5062 mlog_errno(ret);
5063
5064 if ((bucket_blkno(target) == last_blk) && extend)
5065 *extend = 0;
5066 }
5067 }
5068
5069 return ret;
5070 }
5071
5072 /*
5073 * Add a new cluster for xattr storage.
5074 *
5075 * If the new cluster is contiguous with the previous one, it will be
5076 * appended to the same extent record, and num_clusters will be updated.
5077 * If not, we will insert a new extent for it and move some xattrs in
5078 * the last cluster into the new allocated one.
5079 * We also need to limit the maximum size of a btree leaf, otherwise we'll
5080 * lose the benefits of hashing because we'll have to search large leaves.
5081 * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
5082 * if it's bigger).
5083 *
5084 * first_bh is the first block of the previous extent rec and header_bh
5085 * indicates the bucket we will insert the new xattrs. They will be updated
5086 * when the header_bh is moved into the new cluster.
5087 */
5088 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
5089 struct buffer_head *root_bh,
5090 struct ocfs2_xattr_bucket *first,
5091 struct ocfs2_xattr_bucket *target,
5092 u32 *num_clusters,
5093 u32 prev_cpos,
5094 int *extend,
5095 struct ocfs2_xattr_set_ctxt *ctxt)
5096 {
5097 int ret;
5098 u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
5099 u32 prev_clusters = *num_clusters;
5100 u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
5101 u64 block;
5102 handle_t *handle = ctxt->handle;
5103 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5104 struct ocfs2_extent_tree et;
5105
5106 trace_ocfs2_add_new_xattr_cluster_begin(
5107 (unsigned long long)OCFS2_I(inode)->ip_blkno,
5108 (unsigned long long)bucket_blkno(first),
5109 prev_cpos, prev_clusters);
5110
5111 ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5112
5113 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5114 OCFS2_JOURNAL_ACCESS_WRITE);
5115 if (ret < 0) {
5116 mlog_errno(ret);
5117 goto leave;
5118 }
5119
5120 ret = __ocfs2_claim_clusters(handle, ctxt->data_ac, 1,
5121 clusters_to_add, &bit_off, &num_bits);
5122 if (ret < 0) {
5123 if (ret != -ENOSPC)
5124 mlog_errno(ret);
5125 goto leave;
5126 }
5127
5128 BUG_ON(num_bits > clusters_to_add);
5129
5130 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
5131 trace_ocfs2_add_new_xattr_cluster((unsigned long long)block, num_bits);
5132
5133 if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
5134 (prev_clusters + num_bits) << osb->s_clustersize_bits <=
5135 OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
5136 /*
5137 * If this cluster is contiguous with the old one and
5138 * adding this new cluster, we don't surpass the limit of
5139 * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
5140 * initialized and used like other buckets in the previous
5141 * cluster.
5142 * So add it as a contiguous one. The caller will handle
5143 * its init process.
5144 */
5145 v_start = prev_cpos + prev_clusters;
5146 *num_clusters = prev_clusters + num_bits;
5147 } else {
5148 ret = ocfs2_adjust_xattr_cross_cluster(inode,
5149 handle,
5150 first,
5151 target,
5152 block,
5153 prev_clusters,
5154 &v_start,
5155 extend);
5156 if (ret) {
5157 mlog_errno(ret);
5158 goto leave;
5159 }
5160 }
5161
5162 trace_ocfs2_add_new_xattr_cluster_insert((unsigned long long)block,
5163 v_start, num_bits);
5164 ret = ocfs2_insert_extent(handle, &et, v_start, block,
5165 num_bits, 0, ctxt->meta_ac);
5166 if (ret < 0) {
5167 mlog_errno(ret);
5168 goto leave;
5169 }
5170
5171 ocfs2_journal_dirty(handle, root_bh);
5172
5173 leave:
5174 return ret;
5175 }
5176
5177 /*
5178 * We are given an extent. 'first' is the bucket at the very front of
5179 * the extent. The extent has space for an additional bucket past
5180 * bucket_xh(first)->xh_num_buckets. 'target_blkno' is the block number
5181 * of the target bucket. We wish to shift every bucket past the target
5182 * down one, filling in that additional space. When we get back to the
5183 * target, we split the target between itself and the now-empty bucket
5184 * at target+1 (aka, target_blkno + blks_per_bucket).
5185 */
5186 static int ocfs2_extend_xattr_bucket(struct inode *inode,
5187 handle_t *handle,
5188 struct ocfs2_xattr_bucket *first,
5189 u64 target_blk,
5190 u32 num_clusters)
5191 {
5192 int ret, credits;
5193 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5194 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5195 u64 end_blk;
5196 u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
5197
5198 trace_ocfs2_extend_xattr_bucket((unsigned long long)target_blk,
5199 (unsigned long long)bucket_blkno(first),
5200 num_clusters, new_bucket);
5201
5202 /* The extent must have room for an additional bucket */
5203 BUG_ON(new_bucket >=
5204 (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
5205
5206 /* end_blk points to the last existing bucket */
5207 end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
5208
5209 /*
5210 * end_blk is the start of the last existing bucket.
5211 * Thus, (end_blk - target_blk) covers the target bucket and
5212 * every bucket after it up to, but not including, the last
5213 * existing bucket. Then we add the last existing bucket, the
5214 * new bucket, and the first bucket (3 * blk_per_bucket).
5215 */
5216 credits = (end_blk - target_blk) + (3 * blk_per_bucket);
5217 ret = ocfs2_extend_trans(handle, credits);
5218 if (ret) {
5219 mlog_errno(ret);
5220 goto out;
5221 }
5222
5223 ret = ocfs2_xattr_bucket_journal_access(handle, first,
5224 OCFS2_JOURNAL_ACCESS_WRITE);
5225 if (ret) {
5226 mlog_errno(ret);
5227 goto out;
5228 }
5229
5230 while (end_blk != target_blk) {
5231 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
5232 end_blk + blk_per_bucket, 0);
5233 if (ret)
5234 goto out;
5235 end_blk -= blk_per_bucket;
5236 }
5237
5238 /* Move half of the xattr in target_blkno to the next bucket. */
5239 ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
5240 target_blk + blk_per_bucket, NULL, 0);
5241
5242 le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
5243 ocfs2_xattr_bucket_journal_dirty(handle, first);
5244
5245 out:
5246 return ret;
5247 }
5248
5249 /*
5250 * Add new xattr bucket in an extent record and adjust the buckets
5251 * accordingly. xb_bh is the ocfs2_xattr_block, and target is the
5252 * bucket we want to insert into.
5253 *
5254 * In the easy case, we will move all the buckets after target down by
5255 * one. Half of target's xattrs will be moved to the next bucket.
5256 *
5257 * If current cluster is full, we'll allocate a new one. This may not
5258 * be contiguous. The underlying calls will make sure that there is
5259 * space for the insert, shifting buckets around if necessary.
5260 * 'target' may be moved by those calls.
5261 */
5262 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
5263 struct buffer_head *xb_bh,
5264 struct ocfs2_xattr_bucket *target,
5265 struct ocfs2_xattr_set_ctxt *ctxt)
5266 {
5267 struct ocfs2_xattr_block *xb =
5268 (struct ocfs2_xattr_block *)xb_bh->b_data;
5269 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
5270 struct ocfs2_extent_list *el = &xb_root->xt_list;
5271 u32 name_hash =
5272 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
5273 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5274 int ret, num_buckets, extend = 1;
5275 u64 p_blkno;
5276 u32 e_cpos, num_clusters;
5277 /* The bucket at the front of the extent */
5278 struct ocfs2_xattr_bucket *first;
5279
5280 trace_ocfs2_add_new_xattr_bucket(
5281 (unsigned long long)bucket_blkno(target));
5282
5283 /* The first bucket of the original extent */
5284 first = ocfs2_xattr_bucket_new(inode);
5285 if (!first) {
5286 ret = -ENOMEM;
5287 mlog_errno(ret);
5288 goto out;
5289 }
5290
5291 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
5292 &num_clusters, el);
5293 if (ret) {
5294 mlog_errno(ret);
5295 goto out;
5296 }
5297
5298 ret = ocfs2_read_xattr_bucket(first, p_blkno);
5299 if (ret) {
5300 mlog_errno(ret);
5301 goto out;
5302 }
5303
5304 num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
5305 if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
5306 /*
5307 * This can move first+target if the target bucket moves
5308 * to the new extent.
5309 */
5310 ret = ocfs2_add_new_xattr_cluster(inode,
5311 xb_bh,
5312 first,
5313 target,
5314 &num_clusters,
5315 e_cpos,
5316 &extend,
5317 ctxt);
5318 if (ret) {
5319 mlog_errno(ret);
5320 goto out;
5321 }
5322 }
5323
5324 if (extend) {
5325 ret = ocfs2_extend_xattr_bucket(inode,
5326 ctxt->handle,
5327 first,
5328 bucket_blkno(target),
5329 num_clusters);
5330 if (ret)
5331 mlog_errno(ret);
5332 }
5333
5334 out:
5335 ocfs2_xattr_bucket_free(first);
5336
5337 return ret;
5338 }
5339
5340 /*
5341 * Truncate the specified xe_off entry in xattr bucket.
5342 * bucket is indicated by header_bh and len is the new length.
5343 * Both the ocfs2_xattr_value_root and the entry will be updated here.
5344 *
5345 * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
5346 */
5347 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
5348 struct ocfs2_xattr_bucket *bucket,
5349 int xe_off,
5350 int len,
5351 struct ocfs2_xattr_set_ctxt *ctxt)
5352 {
5353 int ret, offset;
5354 u64 value_blk;
5355 struct ocfs2_xattr_entry *xe;
5356 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5357 size_t blocksize = inode->i_sb->s_blocksize;
5358 struct ocfs2_xattr_value_buf vb = {
5359 .vb_access = ocfs2_journal_access,
5360 };
5361
5362 xe = &xh->xh_entries[xe_off];
5363
5364 BUG_ON(!xe || ocfs2_xattr_is_local(xe));
5365
5366 offset = le16_to_cpu(xe->xe_name_offset) +
5367 OCFS2_XATTR_SIZE(xe->xe_name_len);
5368
5369 value_blk = offset / blocksize;
5370
5371 /* We don't allow ocfs2_xattr_value to be stored in different block. */
5372 BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
5373
5374 vb.vb_bh = bucket->bu_bhs[value_blk];
5375 BUG_ON(!vb.vb_bh);
5376
5377 vb.vb_xv = (struct ocfs2_xattr_value_root *)
5378 (vb.vb_bh->b_data + offset % blocksize);
5379
5380 /*
5381 * From here on out we have to dirty the bucket. The generic
5382 * value calls only modify one of the bucket's bhs, but we need
5383 * to send the bucket at once. So if they error, they *could* have
5384 * modified something. We have to assume they did, and dirty
5385 * the whole bucket. This leaves us in a consistent state.
5386 */
5387 trace_ocfs2_xattr_bucket_value_truncate(
5388 (unsigned long long)bucket_blkno(bucket), xe_off, len);
5389 ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
5390 if (ret) {
5391 mlog_errno(ret);
5392 goto out;
5393 }
5394
5395 ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
5396 OCFS2_JOURNAL_ACCESS_WRITE);
5397 if (ret) {
5398 mlog_errno(ret);
5399 goto out;
5400 }
5401
5402 xe->xe_value_size = cpu_to_le64(len);
5403
5404 ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
5405
5406 out:
5407 return ret;
5408 }
5409
5410 static int ocfs2_rm_xattr_cluster(struct inode *inode,
5411 struct buffer_head *root_bh,
5412 u64 blkno,
5413 u32 cpos,
5414 u32 len,
5415 void *para)
5416 {
5417 int ret;
5418 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5419 struct inode *tl_inode = osb->osb_tl_inode;
5420 handle_t *handle;
5421 struct ocfs2_xattr_block *xb =
5422 (struct ocfs2_xattr_block *)root_bh->b_data;
5423 struct ocfs2_alloc_context *meta_ac = NULL;
5424 struct ocfs2_cached_dealloc_ctxt dealloc;
5425 struct ocfs2_extent_tree et;
5426
5427 ret = ocfs2_iterate_xattr_buckets(inode, blkno, len,
5428 ocfs2_delete_xattr_in_bucket, para);
5429 if (ret) {
5430 mlog_errno(ret);
5431 return ret;
5432 }
5433
5434 ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5435
5436 ocfs2_init_dealloc_ctxt(&dealloc);
5437
5438 trace_ocfs2_rm_xattr_cluster(
5439 (unsigned long long)OCFS2_I(inode)->ip_blkno,
5440 (unsigned long long)blkno, cpos, len);
5441
5442 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode), blkno,
5443 len);
5444
5445 ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
5446 if (ret) {
5447 mlog_errno(ret);
5448 return ret;
5449 }
5450
5451 inode_lock(tl_inode);
5452
5453 if (ocfs2_truncate_log_needs_flush(osb)) {
5454 ret = __ocfs2_flush_truncate_log(osb);
5455 if (ret < 0) {
5456 mlog_errno(ret);
5457 goto out;
5458 }
5459 }
5460
5461 handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
5462 if (IS_ERR(handle)) {
5463 ret = -ENOMEM;
5464 mlog_errno(ret);
5465 goto out;
5466 }
5467
5468 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5469 OCFS2_JOURNAL_ACCESS_WRITE);
5470 if (ret) {
5471 mlog_errno(ret);
5472 goto out_commit;
5473 }
5474
5475 ret = ocfs2_remove_extent(handle, &et, cpos, len, meta_ac,
5476 &dealloc);
5477 if (ret) {
5478 mlog_errno(ret);
5479 goto out_commit;
5480 }
5481
5482 le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
5483 ocfs2_journal_dirty(handle, root_bh);
5484
5485 ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
5486 if (ret)
5487 mlog_errno(ret);
5488 ocfs2_update_inode_fsync_trans(handle, inode, 0);
5489
5490 out_commit:
5491 ocfs2_commit_trans(osb, handle);
5492 out:
5493 ocfs2_schedule_truncate_log_flush(osb, 1);
5494
5495 inode_unlock(tl_inode);
5496
5497 if (meta_ac)
5498 ocfs2_free_alloc_context(meta_ac);
5499
5500 ocfs2_run_deallocs(osb, &dealloc);
5501
5502 return ret;
5503 }
5504
5505 /*
5506 * check whether the xattr bucket is filled up with the same hash value.
5507 * If we want to insert the xattr with the same hash, return -ENOSPC.
5508 * If we want to insert a xattr with different hash value, go ahead
5509 * and ocfs2_divide_xattr_bucket will handle this.
5510 */
5511 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
5512 struct ocfs2_xattr_bucket *bucket,
5513 const char *name)
5514 {
5515 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5516 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5517
5518 if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5519 return 0;
5520
5521 if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5522 xh->xh_entries[0].xe_name_hash) {
5523 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5524 "hash = %u\n",
5525 (unsigned long long)bucket_blkno(bucket),
5526 le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5527 return -ENOSPC;
5528 }
5529
5530 return 0;
5531 }
5532
5533 /*
5534 * Try to set the entry in the current bucket. If we fail, the caller
5535 * will handle getting us another bucket.
5536 */
5537 static int ocfs2_xattr_set_entry_bucket(struct inode *inode,
5538 struct ocfs2_xattr_info *xi,
5539 struct ocfs2_xattr_search *xs,
5540 struct ocfs2_xattr_set_ctxt *ctxt)
5541 {
5542 int ret;
5543 struct ocfs2_xa_loc loc;
5544
5545 trace_ocfs2_xattr_set_entry_bucket(xi->xi_name);
5546
5547 ocfs2_init_xattr_bucket_xa_loc(&loc, xs->bucket,
5548 xs->not_found ? NULL : xs->here);
5549 ret = ocfs2_xa_set(&loc, xi, ctxt);
5550 if (!ret) {
5551 xs->here = loc.xl_entry;
5552 goto out;
5553 }
5554 if (ret != -ENOSPC) {
5555 mlog_errno(ret);
5556 goto out;
5557 }
5558
5559 /* Ok, we need space. Let's try defragmenting the bucket. */
5560 ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5561 xs->bucket);
5562 if (ret) {
5563 mlog_errno(ret);
5564 goto out;
5565 }
5566
5567 ret = ocfs2_xa_set(&loc, xi, ctxt);
5568 if (!ret) {
5569 xs->here = loc.xl_entry;
5570 goto out;
5571 }
5572 if (ret != -ENOSPC)
5573 mlog_errno(ret);
5574
5575
5576 out:
5577 return ret;
5578 }
5579
5580 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5581 struct ocfs2_xattr_info *xi,
5582 struct ocfs2_xattr_search *xs,
5583 struct ocfs2_xattr_set_ctxt *ctxt)
5584 {
5585 int ret;
5586
5587 trace_ocfs2_xattr_set_entry_index_block(xi->xi_name);
5588
5589 ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5590 if (!ret)
5591 goto out;
5592 if (ret != -ENOSPC) {
5593 mlog_errno(ret);
5594 goto out;
5595 }
5596
5597 /* Ack, need more space. Let's try to get another bucket! */
5598
5599 /*
5600 * We do not allow for overlapping ranges between buckets. And
5601 * the maximum number of collisions we will allow for then is
5602 * one bucket's worth, so check it here whether we need to
5603 * add a new bucket for the insert.
5604 */
5605 ret = ocfs2_check_xattr_bucket_collision(inode,
5606 xs->bucket,
5607 xi->xi_name);
5608 if (ret) {
5609 mlog_errno(ret);
5610 goto out;
5611 }
5612
5613 ret = ocfs2_add_new_xattr_bucket(inode,
5614 xs->xattr_bh,
5615 xs->bucket,
5616 ctxt);
5617 if (ret) {
5618 mlog_errno(ret);
5619 goto out;
5620 }
5621
5622 /*
5623 * ocfs2_add_new_xattr_bucket() will have updated
5624 * xs->bucket if it moved, but it will not have updated
5625 * any of the other search fields. Thus, we drop it and
5626 * re-search. Everything should be cached, so it'll be
5627 * quick.
5628 */
5629 ocfs2_xattr_bucket_relse(xs->bucket);
5630 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5631 xi->xi_name_index,
5632 xi->xi_name, xs);
5633 if (ret && ret != -ENODATA)
5634 goto out;
5635 xs->not_found = ret;
5636
5637 /* Ok, we have a new bucket, let's try again */
5638 ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5639 if (ret && (ret != -ENOSPC))
5640 mlog_errno(ret);
5641
5642 out:
5643 return ret;
5644 }
5645
5646 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5647 struct ocfs2_xattr_bucket *bucket,
5648 void *para)
5649 {
5650 int ret = 0, ref_credits;
5651 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5652 u16 i;
5653 struct ocfs2_xattr_entry *xe;
5654 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5655 struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5656 int credits = ocfs2_remove_extent_credits(osb->sb) +
5657 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5658 struct ocfs2_xattr_value_root *xv;
5659 struct ocfs2_rm_xattr_bucket_para *args =
5660 (struct ocfs2_rm_xattr_bucket_para *)para;
5661
5662 ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5663
5664 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5665 xe = &xh->xh_entries[i];
5666 if (ocfs2_xattr_is_local(xe))
5667 continue;
5668
5669 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket,
5670 i, &xv, NULL);
5671 if (ret) {
5672 mlog_errno(ret);
5673 break;
5674 }
5675
5676 ret = ocfs2_lock_xattr_remove_allocators(inode, xv,
5677 args->ref_ci,
5678 args->ref_root_bh,
5679 &ctxt.meta_ac,
5680 &ref_credits);
5681
5682 ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
5683 if (IS_ERR(ctxt.handle)) {
5684 ret = PTR_ERR(ctxt.handle);
5685 mlog_errno(ret);
5686 break;
5687 }
5688
5689 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5690 i, 0, &ctxt);
5691
5692 ocfs2_commit_trans(osb, ctxt.handle);
5693 if (ctxt.meta_ac) {
5694 ocfs2_free_alloc_context(ctxt.meta_ac);
5695 ctxt.meta_ac = NULL;
5696 }
5697 if (ret) {
5698 mlog_errno(ret);
5699 break;
5700 }
5701 }
5702
5703 if (ctxt.meta_ac)
5704 ocfs2_free_alloc_context(ctxt.meta_ac);
5705 ocfs2_schedule_truncate_log_flush(osb, 1);
5706 ocfs2_run_deallocs(osb, &ctxt.dealloc);
5707 return ret;
5708 }
5709
5710 /*
5711 * Whenever we modify a xattr value root in the bucket(e.g, CoW
5712 * or change the extent record flag), we need to recalculate
5713 * the metaecc for the whole bucket. So it is done here.
5714 *
5715 * Note:
5716 * We have to give the extra credits for the caller.
5717 */
5718 static int ocfs2_xattr_bucket_post_refcount(struct inode *inode,
5719 handle_t *handle,
5720 void *para)
5721 {
5722 int ret;
5723 struct ocfs2_xattr_bucket *bucket =
5724 (struct ocfs2_xattr_bucket *)para;
5725
5726 ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
5727 OCFS2_JOURNAL_ACCESS_WRITE);
5728 if (ret) {
5729 mlog_errno(ret);
5730 return ret;
5731 }
5732
5733 ocfs2_xattr_bucket_journal_dirty(handle, bucket);
5734
5735 return 0;
5736 }
5737
5738 /*
5739 * Special action we need if the xattr value is refcounted.
5740 *
5741 * 1. If the xattr is refcounted, lock the tree.
5742 * 2. CoW the xattr if we are setting the new value and the value
5743 * will be stored outside.
5744 * 3. In other case, decrease_refcount will work for us, so just
5745 * lock the refcount tree, calculate the meta and credits is OK.
5746 *
5747 * We have to do CoW before ocfs2_init_xattr_set_ctxt since
5748 * currently CoW is a completed transaction, while this function
5749 * will also lock the allocators and let us deadlock. So we will
5750 * CoW the whole xattr value.
5751 */
5752 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
5753 struct ocfs2_dinode *di,
5754 struct ocfs2_xattr_info *xi,
5755 struct ocfs2_xattr_search *xis,
5756 struct ocfs2_xattr_search *xbs,
5757 struct ocfs2_refcount_tree **ref_tree,
5758 int *meta_add,
5759 int *credits)
5760 {
5761 int ret = 0;
5762 struct ocfs2_xattr_block *xb;
5763 struct ocfs2_xattr_entry *xe;
5764 char *base;
5765 u32 p_cluster, num_clusters;
5766 unsigned int ext_flags;
5767 int name_offset, name_len;
5768 struct ocfs2_xattr_value_buf vb;
5769 struct ocfs2_xattr_bucket *bucket = NULL;
5770 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5771 struct ocfs2_post_refcount refcount;
5772 struct ocfs2_post_refcount *p = NULL;
5773 struct buffer_head *ref_root_bh = NULL;
5774
5775 if (!xis->not_found) {
5776 xe = xis->here;
5777 name_offset = le16_to_cpu(xe->xe_name_offset);
5778 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5779 base = xis->base;
5780 vb.vb_bh = xis->inode_bh;
5781 vb.vb_access = ocfs2_journal_access_di;
5782 } else {
5783 int i, block_off = 0;
5784 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
5785 xe = xbs->here;
5786 name_offset = le16_to_cpu(xe->xe_name_offset);
5787 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5788 i = xbs->here - xbs->header->xh_entries;
5789
5790 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
5791 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
5792 bucket_xh(xbs->bucket),
5793 i, &block_off,
5794 &name_offset);
5795 if (ret) {
5796 mlog_errno(ret);
5797 goto out;
5798 }
5799 base = bucket_block(xbs->bucket, block_off);
5800 vb.vb_bh = xbs->bucket->bu_bhs[block_off];
5801 vb.vb_access = ocfs2_journal_access;
5802
5803 if (ocfs2_meta_ecc(osb)) {
5804 /*create parameters for ocfs2_post_refcount. */
5805 bucket = xbs->bucket;
5806 refcount.credits = bucket->bu_blocks;
5807 refcount.para = bucket;
5808 refcount.func =
5809 ocfs2_xattr_bucket_post_refcount;
5810 p = &refcount;
5811 }
5812 } else {
5813 base = xbs->base;
5814 vb.vb_bh = xbs->xattr_bh;
5815 vb.vb_access = ocfs2_journal_access_xb;
5816 }
5817 }
5818
5819 if (ocfs2_xattr_is_local(xe))
5820 goto out;
5821
5822 vb.vb_xv = (struct ocfs2_xattr_value_root *)
5823 (base + name_offset + name_len);
5824
5825 ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
5826 &num_clusters, &vb.vb_xv->xr_list,
5827 &ext_flags);
5828 if (ret) {
5829 mlog_errno(ret);
5830 goto out;
5831 }
5832
5833 /*
5834 * We just need to check the 1st extent record, since we always
5835 * CoW the whole xattr. So there shouldn't be a xattr with
5836 * some REFCOUNT extent recs after the 1st one.
5837 */
5838 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
5839 goto out;
5840
5841 ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
5842 1, ref_tree, &ref_root_bh);
5843 if (ret) {
5844 mlog_errno(ret);
5845 goto out;
5846 }
5847
5848 /*
5849 * If we are deleting the xattr or the new size will be stored inside,
5850 * cool, leave it there, the xattr truncate process will remove them
5851 * for us(it still needs the refcount tree lock and the meta, credits).
5852 * And the worse case is that every cluster truncate will split the
5853 * refcount tree, and make the original extent become 3. So we will need
5854 * 2 * cluster more extent recs at most.
5855 */
5856 if (!xi->xi_value || xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE) {
5857
5858 ret = ocfs2_refcounted_xattr_delete_need(inode,
5859 &(*ref_tree)->rf_ci,
5860 ref_root_bh, vb.vb_xv,
5861 meta_add, credits);
5862 if (ret)
5863 mlog_errno(ret);
5864 goto out;
5865 }
5866
5867 ret = ocfs2_refcount_cow_xattr(inode, di, &vb,
5868 *ref_tree, ref_root_bh, 0,
5869 le32_to_cpu(vb.vb_xv->xr_clusters), p);
5870 if (ret)
5871 mlog_errno(ret);
5872
5873 out:
5874 brelse(ref_root_bh);
5875 return ret;
5876 }
5877
5878 /*
5879 * Add the REFCOUNTED flags for all the extent rec in ocfs2_xattr_value_root.
5880 * The physical clusters will be added to refcount tree.
5881 */
5882 static int ocfs2_xattr_value_attach_refcount(struct inode *inode,
5883 struct ocfs2_xattr_value_root *xv,
5884 struct ocfs2_extent_tree *value_et,
5885 struct ocfs2_caching_info *ref_ci,
5886 struct buffer_head *ref_root_bh,
5887 struct ocfs2_cached_dealloc_ctxt *dealloc,
5888 struct ocfs2_post_refcount *refcount)
5889 {
5890 int ret = 0;
5891 u32 clusters = le32_to_cpu(xv->xr_clusters);
5892 u32 cpos, p_cluster, num_clusters;
5893 struct ocfs2_extent_list *el = &xv->xr_list;
5894 unsigned int ext_flags;
5895
5896 cpos = 0;
5897 while (cpos < clusters) {
5898 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
5899 &num_clusters, el, &ext_flags);
5900 if (ret) {
5901 mlog_errno(ret);
5902 break;
5903 }
5904
5905 cpos += num_clusters;
5906 if ((ext_flags & OCFS2_EXT_REFCOUNTED))
5907 continue;
5908
5909 BUG_ON(!p_cluster);
5910
5911 ret = ocfs2_add_refcount_flag(inode, value_et,
5912 ref_ci, ref_root_bh,
5913 cpos - num_clusters,
5914 p_cluster, num_clusters,
5915 dealloc, refcount);
5916 if (ret) {
5917 mlog_errno(ret);
5918 break;
5919 }
5920 }
5921
5922 return ret;
5923 }
5924
5925 /*
5926 * Given a normal ocfs2_xattr_header, refcount all the entries which
5927 * have value stored outside.
5928 * Used for xattrs stored in inode and ocfs2_xattr_block.
5929 */
5930 static int ocfs2_xattr_attach_refcount_normal(struct inode *inode,
5931 struct ocfs2_xattr_value_buf *vb,
5932 struct ocfs2_xattr_header *header,
5933 struct ocfs2_caching_info *ref_ci,
5934 struct buffer_head *ref_root_bh,
5935 struct ocfs2_cached_dealloc_ctxt *dealloc)
5936 {
5937
5938 struct ocfs2_xattr_entry *xe;
5939 struct ocfs2_xattr_value_root *xv;
5940 struct ocfs2_extent_tree et;
5941 int i, ret = 0;
5942
5943 for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
5944 xe = &header->xh_entries[i];
5945
5946 if (ocfs2_xattr_is_local(xe))
5947 continue;
5948
5949 xv = (struct ocfs2_xattr_value_root *)((void *)header +
5950 le16_to_cpu(xe->xe_name_offset) +
5951 OCFS2_XATTR_SIZE(xe->xe_name_len));
5952
5953 vb->vb_xv = xv;
5954 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
5955
5956 ret = ocfs2_xattr_value_attach_refcount(inode, xv, &et,
5957 ref_ci, ref_root_bh,
5958 dealloc, NULL);
5959 if (ret) {
5960 mlog_errno(ret);
5961 break;
5962 }
5963 }
5964
5965 return ret;
5966 }
5967
5968 static int ocfs2_xattr_inline_attach_refcount(struct inode *inode,
5969 struct buffer_head *fe_bh,
5970 struct ocfs2_caching_info *ref_ci,
5971 struct buffer_head *ref_root_bh,
5972 struct ocfs2_cached_dealloc_ctxt *dealloc)
5973 {
5974 struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
5975 struct ocfs2_xattr_header *header = (struct ocfs2_xattr_header *)
5976 (fe_bh->b_data + inode->i_sb->s_blocksize -
5977 le16_to_cpu(di->i_xattr_inline_size));
5978 struct ocfs2_xattr_value_buf vb = {
5979 .vb_bh = fe_bh,
5980 .vb_access = ocfs2_journal_access_di,
5981 };
5982
5983 return ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
5984 ref_ci, ref_root_bh, dealloc);
5985 }
5986
5987 struct ocfs2_xattr_tree_value_refcount_para {
5988 struct ocfs2_caching_info *ref_ci;
5989 struct buffer_head *ref_root_bh;
5990 struct ocfs2_cached_dealloc_ctxt *dealloc;
5991 };
5992
5993 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
5994 struct ocfs2_xattr_bucket *bucket,
5995 int offset,
5996 struct ocfs2_xattr_value_root **xv,
5997 struct buffer_head **bh)
5998 {
5999 int ret, block_off, name_offset;
6000 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
6001 struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
6002 void *base;
6003
6004 ret = ocfs2_xattr_bucket_get_name_value(sb,
6005 bucket_xh(bucket),
6006 offset,
6007 &block_off,
6008 &name_offset);
6009 if (ret) {
6010 mlog_errno(ret);
6011 goto out;
6012 }
6013
6014 base = bucket_block(bucket, block_off);
6015
6016 *xv = (struct ocfs2_xattr_value_root *)(base + name_offset +
6017 OCFS2_XATTR_SIZE(xe->xe_name_len));
6018
6019 if (bh)
6020 *bh = bucket->bu_bhs[block_off];
6021 out:
6022 return ret;
6023 }
6024
6025 /*
6026 * For a given xattr bucket, refcount all the entries which
6027 * have value stored outside.
6028 */
6029 static int ocfs2_xattr_bucket_value_refcount(struct inode *inode,
6030 struct ocfs2_xattr_bucket *bucket,
6031 void *para)
6032 {
6033 int i, ret = 0;
6034 struct ocfs2_extent_tree et;
6035 struct ocfs2_xattr_tree_value_refcount_para *ref =
6036 (struct ocfs2_xattr_tree_value_refcount_para *)para;
6037 struct ocfs2_xattr_header *xh =
6038 (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6039 struct ocfs2_xattr_entry *xe;
6040 struct ocfs2_xattr_value_buf vb = {
6041 .vb_access = ocfs2_journal_access,
6042 };
6043 struct ocfs2_post_refcount refcount = {
6044 .credits = bucket->bu_blocks,
6045 .para = bucket,
6046 .func = ocfs2_xattr_bucket_post_refcount,
6047 };
6048 struct ocfs2_post_refcount *p = NULL;
6049
6050 /* We only need post_refcount if we support metaecc. */
6051 if (ocfs2_meta_ecc(OCFS2_SB(inode->i_sb)))
6052 p = &refcount;
6053
6054 trace_ocfs2_xattr_bucket_value_refcount(
6055 (unsigned long long)bucket_blkno(bucket),
6056 le16_to_cpu(xh->xh_count));
6057 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6058 xe = &xh->xh_entries[i];
6059
6060 if (ocfs2_xattr_is_local(xe))
6061 continue;
6062
6063 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket, i,
6064 &vb.vb_xv, &vb.vb_bh);
6065 if (ret) {
6066 mlog_errno(ret);
6067 break;
6068 }
6069
6070 ocfs2_init_xattr_value_extent_tree(&et,
6071 INODE_CACHE(inode), &vb);
6072
6073 ret = ocfs2_xattr_value_attach_refcount(inode, vb.vb_xv,
6074 &et, ref->ref_ci,
6075 ref->ref_root_bh,
6076 ref->dealloc, p);
6077 if (ret) {
6078 mlog_errno(ret);
6079 break;
6080 }
6081 }
6082
6083 return ret;
6084
6085 }
6086
6087 static int ocfs2_refcount_xattr_tree_rec(struct inode *inode,
6088 struct buffer_head *root_bh,
6089 u64 blkno, u32 cpos, u32 len, void *para)
6090 {
6091 return ocfs2_iterate_xattr_buckets(inode, blkno, len,
6092 ocfs2_xattr_bucket_value_refcount,
6093 para);
6094 }
6095
6096 static int ocfs2_xattr_block_attach_refcount(struct inode *inode,
6097 struct buffer_head *blk_bh,
6098 struct ocfs2_caching_info *ref_ci,
6099 struct buffer_head *ref_root_bh,
6100 struct ocfs2_cached_dealloc_ctxt *dealloc)
6101 {
6102 int ret = 0;
6103 struct ocfs2_xattr_block *xb =
6104 (struct ocfs2_xattr_block *)blk_bh->b_data;
6105
6106 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
6107 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
6108 struct ocfs2_xattr_value_buf vb = {
6109 .vb_bh = blk_bh,
6110 .vb_access = ocfs2_journal_access_xb,
6111 };
6112
6113 ret = ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
6114 ref_ci, ref_root_bh,
6115 dealloc);
6116 } else {
6117 struct ocfs2_xattr_tree_value_refcount_para para = {
6118 .ref_ci = ref_ci,
6119 .ref_root_bh = ref_root_bh,
6120 .dealloc = dealloc,
6121 };
6122
6123 ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
6124 ocfs2_refcount_xattr_tree_rec,
6125 &para);
6126 }
6127
6128 return ret;
6129 }
6130
6131 int ocfs2_xattr_attach_refcount_tree(struct inode *inode,
6132 struct buffer_head *fe_bh,
6133 struct ocfs2_caching_info *ref_ci,
6134 struct buffer_head *ref_root_bh,
6135 struct ocfs2_cached_dealloc_ctxt *dealloc)
6136 {
6137 int ret = 0;
6138 struct ocfs2_inode_info *oi = OCFS2_I(inode);
6139 struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
6140 struct buffer_head *blk_bh = NULL;
6141
6142 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
6143 ret = ocfs2_xattr_inline_attach_refcount(inode, fe_bh,
6144 ref_ci, ref_root_bh,
6145 dealloc);
6146 if (ret) {
6147 mlog_errno(ret);
6148 goto out;
6149 }
6150 }
6151
6152 if (!di->i_xattr_loc)
6153 goto out;
6154
6155 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
6156 &blk_bh);
6157 if (ret < 0) {
6158 mlog_errno(ret);
6159 goto out;
6160 }
6161
6162 ret = ocfs2_xattr_block_attach_refcount(inode, blk_bh, ref_ci,
6163 ref_root_bh, dealloc);
6164 if (ret)
6165 mlog_errno(ret);
6166
6167 brelse(blk_bh);
6168 out:
6169
6170 return ret;
6171 }
6172
6173 typedef int (should_xattr_reflinked)(struct ocfs2_xattr_entry *xe);
6174 /*
6175 * Store the information we need in xattr reflink.
6176 * old_bh and new_bh are inode bh for the old and new inode.
6177 */
6178 struct ocfs2_xattr_reflink {
6179 struct inode *old_inode;
6180 struct inode *new_inode;
6181 struct buffer_head *old_bh;
6182 struct buffer_head *new_bh;
6183 struct ocfs2_caching_info *ref_ci;
6184 struct buffer_head *ref_root_bh;
6185 struct ocfs2_cached_dealloc_ctxt *dealloc;
6186 should_xattr_reflinked *xattr_reflinked;
6187 };
6188
6189 /*
6190 * Given a xattr header and xe offset,
6191 * return the proper xv and the corresponding bh.
6192 * xattr in inode, block and xattr tree have different implementaions.
6193 */
6194 typedef int (get_xattr_value_root)(struct super_block *sb,
6195 struct buffer_head *bh,
6196 struct ocfs2_xattr_header *xh,
6197 int offset,
6198 struct ocfs2_xattr_value_root **xv,
6199 struct buffer_head **ret_bh,
6200 void *para);
6201
6202 /*
6203 * Calculate all the xattr value root metadata stored in this xattr header and
6204 * credits we need if we create them from the scratch.
6205 * We use get_xattr_value_root so that all types of xattr container can use it.
6206 */
6207 static int ocfs2_value_metas_in_xattr_header(struct super_block *sb,
6208 struct buffer_head *bh,
6209 struct ocfs2_xattr_header *xh,
6210 int *metas, int *credits,
6211 int *num_recs,
6212 get_xattr_value_root *func,
6213 void *para)
6214 {
6215 int i, ret = 0;
6216 struct ocfs2_xattr_value_root *xv;
6217 struct ocfs2_xattr_entry *xe;
6218
6219 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6220 xe = &xh->xh_entries[i];
6221 if (ocfs2_xattr_is_local(xe))
6222 continue;
6223
6224 ret = func(sb, bh, xh, i, &xv, NULL, para);
6225 if (ret) {
6226 mlog_errno(ret);
6227 break;
6228 }
6229
6230 *metas += le16_to_cpu(xv->xr_list.l_tree_depth) *
6231 le16_to_cpu(xv->xr_list.l_next_free_rec);
6232
6233 *credits += ocfs2_calc_extend_credits(sb,
6234 &def_xv.xv.xr_list);
6235
6236 /*
6237 * If the value is a tree with depth > 1, We don't go deep
6238 * to the extent block, so just calculate a maximum record num.
6239 */
6240 if (!xv->xr_list.l_tree_depth)
6241 *num_recs += le16_to_cpu(xv->xr_list.l_next_free_rec);
6242 else
6243 *num_recs += ocfs2_clusters_for_bytes(sb,
6244 XATTR_SIZE_MAX);
6245 }
6246
6247 return ret;
6248 }
6249
6250 /* Used by xattr inode and block to return the right xv and buffer_head. */
6251 static int ocfs2_get_xattr_value_root(struct super_block *sb,
6252 struct buffer_head *bh,
6253 struct ocfs2_xattr_header *xh,
6254 int offset,
6255 struct ocfs2_xattr_value_root **xv,
6256 struct buffer_head **ret_bh,
6257 void *para)
6258 {
6259 struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
6260
6261 *xv = (struct ocfs2_xattr_value_root *)((void *)xh +
6262 le16_to_cpu(xe->xe_name_offset) +
6263 OCFS2_XATTR_SIZE(xe->xe_name_len));
6264
6265 if (ret_bh)
6266 *ret_bh = bh;
6267
6268 return 0;
6269 }
6270
6271 /*
6272 * Lock the meta_ac and caculate how much credits we need for reflink xattrs.
6273 * It is only used for inline xattr and xattr block.
6274 */
6275 static int ocfs2_reflink_lock_xattr_allocators(struct ocfs2_super *osb,
6276 struct ocfs2_xattr_header *xh,
6277 struct buffer_head *ref_root_bh,
6278 int *credits,
6279 struct ocfs2_alloc_context **meta_ac)
6280 {
6281 int ret, meta_add = 0, num_recs = 0;
6282 struct ocfs2_refcount_block *rb =
6283 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
6284
6285 *credits = 0;
6286
6287 ret = ocfs2_value_metas_in_xattr_header(osb->sb, NULL, xh,
6288 &meta_add, credits, &num_recs,
6289 ocfs2_get_xattr_value_root,
6290 NULL);
6291 if (ret) {
6292 mlog_errno(ret);
6293 goto out;
6294 }
6295
6296 /*
6297 * We need to add/modify num_recs in refcount tree, so just calculate
6298 * an approximate number we need for refcount tree change.
6299 * Sometimes we need to split the tree, and after split, half recs
6300 * will be moved to the new block, and a new block can only provide
6301 * half number of recs. So we multiple new blocks by 2.
6302 */
6303 num_recs = num_recs / ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6304 meta_add += num_recs;
6305 *credits += num_recs + num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6306 if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6307 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6308 le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6309 else
6310 *credits += 1;
6311
6312 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add, meta_ac);
6313 if (ret)
6314 mlog_errno(ret);
6315
6316 out:
6317 return ret;
6318 }
6319
6320 /*
6321 * Given a xattr header, reflink all the xattrs in this container.
6322 * It can be used for inode, block and bucket.
6323 *
6324 * NOTE:
6325 * Before we call this function, the caller has memcpy the xattr in
6326 * old_xh to the new_xh.
6327 *
6328 * If args.xattr_reflinked is set, call it to decide whether the xe should
6329 * be reflinked or not. If not, remove it from the new xattr header.
6330 */
6331 static int ocfs2_reflink_xattr_header(handle_t *handle,
6332 struct ocfs2_xattr_reflink *args,
6333 struct buffer_head *old_bh,
6334 struct ocfs2_xattr_header *xh,
6335 struct buffer_head *new_bh,
6336 struct ocfs2_xattr_header *new_xh,
6337 struct ocfs2_xattr_value_buf *vb,
6338 struct ocfs2_alloc_context *meta_ac,
6339 get_xattr_value_root *func,
6340 void *para)
6341 {
6342 int ret = 0, i, j;
6343 struct super_block *sb = args->old_inode->i_sb;
6344 struct buffer_head *value_bh;
6345 struct ocfs2_xattr_entry *xe, *last;
6346 struct ocfs2_xattr_value_root *xv, *new_xv;
6347 struct ocfs2_extent_tree data_et;
6348 u32 clusters, cpos, p_cluster, num_clusters;
6349 unsigned int ext_flags = 0;
6350
6351 trace_ocfs2_reflink_xattr_header((unsigned long long)old_bh->b_blocknr,
6352 le16_to_cpu(xh->xh_count));
6353
6354 last = &new_xh->xh_entries[le16_to_cpu(new_xh->xh_count)];
6355 for (i = 0, j = 0; i < le16_to_cpu(xh->xh_count); i++, j++) {
6356 xe = &xh->xh_entries[i];
6357
6358 if (args->xattr_reflinked && !args->xattr_reflinked(xe)) {
6359 xe = &new_xh->xh_entries[j];
6360
6361 le16_add_cpu(&new_xh->xh_count, -1);
6362 if (new_xh->xh_count) {
6363 memmove(xe, xe + 1,
6364 (void *)last - (void *)xe);
6365 memset(last, 0,
6366 sizeof(struct ocfs2_xattr_entry));
6367 }
6368
6369 /*
6370 * We don't want j to increase in the next round since
6371 * it is already moved ahead.
6372 */
6373 j--;
6374 continue;
6375 }
6376
6377 if (ocfs2_xattr_is_local(xe))
6378 continue;
6379
6380 ret = func(sb, old_bh, xh, i, &xv, NULL, para);
6381 if (ret) {
6382 mlog_errno(ret);
6383 break;
6384 }
6385
6386 ret = func(sb, new_bh, new_xh, j, &new_xv, &value_bh, para);
6387 if (ret) {
6388 mlog_errno(ret);
6389 break;
6390 }
6391
6392 /*
6393 * For the xattr which has l_tree_depth = 0, all the extent
6394 * recs have already be copied to the new xh with the
6395 * propriate OCFS2_EXT_REFCOUNTED flag we just need to
6396 * increase the refount count int the refcount tree.
6397 *
6398 * For the xattr which has l_tree_depth > 0, we need
6399 * to initialize it to the empty default value root,
6400 * and then insert the extents one by one.
6401 */
6402 if (xv->xr_list.l_tree_depth) {
6403 memcpy(new_xv, &def_xv, OCFS2_XATTR_ROOT_SIZE);
6404 vb->vb_xv = new_xv;
6405 vb->vb_bh = value_bh;
6406 ocfs2_init_xattr_value_extent_tree(&data_et,
6407 INODE_CACHE(args->new_inode), vb);
6408 }
6409
6410 clusters = le32_to_cpu(xv->xr_clusters);
6411 cpos = 0;
6412 while (cpos < clusters) {
6413 ret = ocfs2_xattr_get_clusters(args->old_inode,
6414 cpos,
6415 &p_cluster,
6416 &num_clusters,
6417 &xv->xr_list,
6418 &ext_flags);
6419 if (ret) {
6420 mlog_errno(ret);
6421 goto out;
6422 }
6423
6424 BUG_ON(!p_cluster);
6425
6426 if (xv->xr_list.l_tree_depth) {
6427 ret = ocfs2_insert_extent(handle,
6428 &data_et, cpos,
6429 ocfs2_clusters_to_blocks(
6430 args->old_inode->i_sb,
6431 p_cluster),
6432 num_clusters, ext_flags,
6433 meta_ac);
6434 if (ret) {
6435 mlog_errno(ret);
6436 goto out;
6437 }
6438 }
6439
6440 ret = ocfs2_increase_refcount(handle, args->ref_ci,
6441 args->ref_root_bh,
6442 p_cluster, num_clusters,
6443 meta_ac, args->dealloc);
6444 if (ret) {
6445 mlog_errno(ret);
6446 goto out;
6447 }
6448
6449 cpos += num_clusters;
6450 }
6451 }
6452
6453 out:
6454 return ret;
6455 }
6456
6457 static int ocfs2_reflink_xattr_inline(struct ocfs2_xattr_reflink *args)
6458 {
6459 int ret = 0, credits = 0;
6460 handle_t *handle;
6461 struct ocfs2_super *osb = OCFS2_SB(args->old_inode->i_sb);
6462 struct ocfs2_dinode *di = (struct ocfs2_dinode *)args->old_bh->b_data;
6463 int inline_size = le16_to_cpu(di->i_xattr_inline_size);
6464 int header_off = osb->sb->s_blocksize - inline_size;
6465 struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)
6466 (args->old_bh->b_data + header_off);
6467 struct ocfs2_xattr_header *new_xh = (struct ocfs2_xattr_header *)
6468 (args->new_bh->b_data + header_off);
6469 struct ocfs2_alloc_context *meta_ac = NULL;
6470 struct ocfs2_inode_info *new_oi;
6471 struct ocfs2_dinode *new_di;
6472 struct ocfs2_xattr_value_buf vb = {
6473 .vb_bh = args->new_bh,
6474 .vb_access = ocfs2_journal_access_di,
6475 };
6476
6477 ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6478 &credits, &meta_ac);
6479 if (ret) {
6480 mlog_errno(ret);
6481 goto out;
6482 }
6483
6484 handle = ocfs2_start_trans(osb, credits);
6485 if (IS_ERR(handle)) {
6486 ret = PTR_ERR(handle);
6487 mlog_errno(ret);
6488 goto out;
6489 }
6490
6491 ret = ocfs2_journal_access_di(handle, INODE_CACHE(args->new_inode),
6492 args->new_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6493 if (ret) {
6494 mlog_errno(ret);
6495 goto out_commit;
6496 }
6497
6498 memcpy(args->new_bh->b_data + header_off,
6499 args->old_bh->b_data + header_off, inline_size);
6500
6501 new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6502 new_di->i_xattr_inline_size = cpu_to_le16(inline_size);
6503
6504 ret = ocfs2_reflink_xattr_header(handle, args, args->old_bh, xh,
6505 args->new_bh, new_xh, &vb, meta_ac,
6506 ocfs2_get_xattr_value_root, NULL);
6507 if (ret) {
6508 mlog_errno(ret);
6509 goto out_commit;
6510 }
6511
6512 new_oi = OCFS2_I(args->new_inode);
6513
6514 spin_lock(&new_oi->ip_lock);
6515 new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL | OCFS2_INLINE_XATTR_FL;
6516 new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6517 spin_unlock(&new_oi->ip_lock);
6518
6519 ocfs2_journal_dirty(handle, args->new_bh);
6520
6521 out_commit:
6522 ocfs2_commit_trans(osb, handle);
6523
6524 out:
6525 if (meta_ac)
6526 ocfs2_free_alloc_context(meta_ac);
6527 return ret;
6528 }
6529
6530 static int ocfs2_create_empty_xattr_block(struct inode *inode,
6531 struct buffer_head *fe_bh,
6532 struct buffer_head **ret_bh,
6533 int indexed)
6534 {
6535 int ret;
6536 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6537 struct ocfs2_xattr_set_ctxt ctxt;
6538
6539 memset(&ctxt, 0, sizeof(ctxt));
6540 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &ctxt.meta_ac);
6541 if (ret < 0) {
6542 mlog_errno(ret);
6543 return ret;
6544 }
6545
6546 ctxt.handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_CREATE_CREDITS);
6547 if (IS_ERR(ctxt.handle)) {
6548 ret = PTR_ERR(ctxt.handle);
6549 mlog_errno(ret);
6550 goto out;
6551 }
6552
6553 trace_ocfs2_create_empty_xattr_block(
6554 (unsigned long long)fe_bh->b_blocknr, indexed);
6555 ret = ocfs2_create_xattr_block(inode, fe_bh, &ctxt, indexed,
6556 ret_bh);
6557 if (ret)
6558 mlog_errno(ret);
6559
6560 ocfs2_commit_trans(osb, ctxt.handle);
6561 out:
6562 ocfs2_free_alloc_context(ctxt.meta_ac);
6563 return ret;
6564 }
6565
6566 static int ocfs2_reflink_xattr_block(struct ocfs2_xattr_reflink *args,
6567 struct buffer_head *blk_bh,
6568 struct buffer_head *new_blk_bh)
6569 {
6570 int ret = 0, credits = 0;
6571 handle_t *handle;
6572 struct ocfs2_inode_info *new_oi = OCFS2_I(args->new_inode);
6573 struct ocfs2_dinode *new_di;
6574 struct ocfs2_super *osb = OCFS2_SB(args->new_inode->i_sb);
6575 int header_off = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
6576 struct ocfs2_xattr_block *xb =
6577 (struct ocfs2_xattr_block *)blk_bh->b_data;
6578 struct ocfs2_xattr_header *xh = &xb->xb_attrs.xb_header;
6579 struct ocfs2_xattr_block *new_xb =
6580 (struct ocfs2_xattr_block *)new_blk_bh->b_data;
6581 struct ocfs2_xattr_header *new_xh = &new_xb->xb_attrs.xb_header;
6582 struct ocfs2_alloc_context *meta_ac;
6583 struct ocfs2_xattr_value_buf vb = {
6584 .vb_bh = new_blk_bh,
6585 .vb_access = ocfs2_journal_access_xb,
6586 };
6587
6588 ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6589 &credits, &meta_ac);
6590 if (ret) {
6591 mlog_errno(ret);
6592 return ret;
6593 }
6594
6595 /* One more credits in case we need to add xattr flags in new inode. */
6596 handle = ocfs2_start_trans(osb, credits + 1);
6597 if (IS_ERR(handle)) {
6598 ret = PTR_ERR(handle);
6599 mlog_errno(ret);
6600 goto out;
6601 }
6602
6603 if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6604 ret = ocfs2_journal_access_di(handle,
6605 INODE_CACHE(args->new_inode),
6606 args->new_bh,
6607 OCFS2_JOURNAL_ACCESS_WRITE);
6608 if (ret) {
6609 mlog_errno(ret);
6610 goto out_commit;
6611 }
6612 }
6613
6614 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(args->new_inode),
6615 new_blk_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6616 if (ret) {
6617 mlog_errno(ret);
6618 goto out_commit;
6619 }
6620
6621 memcpy(new_blk_bh->b_data + header_off, blk_bh->b_data + header_off,
6622 osb->sb->s_blocksize - header_off);
6623
6624 ret = ocfs2_reflink_xattr_header(handle, args, blk_bh, xh,
6625 new_blk_bh, new_xh, &vb, meta_ac,
6626 ocfs2_get_xattr_value_root, NULL);
6627 if (ret) {
6628 mlog_errno(ret);
6629 goto out_commit;
6630 }
6631
6632 ocfs2_journal_dirty(handle, new_blk_bh);
6633
6634 if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6635 new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6636 spin_lock(&new_oi->ip_lock);
6637 new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
6638 new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6639 spin_unlock(&new_oi->ip_lock);
6640
6641 ocfs2_journal_dirty(handle, args->new_bh);
6642 }
6643
6644 out_commit:
6645 ocfs2_commit_trans(osb, handle);
6646
6647 out:
6648 ocfs2_free_alloc_context(meta_ac);
6649 return ret;
6650 }
6651
6652 struct ocfs2_reflink_xattr_tree_args {
6653 struct ocfs2_xattr_reflink *reflink;
6654 struct buffer_head *old_blk_bh;
6655 struct buffer_head *new_blk_bh;
6656 struct ocfs2_xattr_bucket *old_bucket;
6657 struct ocfs2_xattr_bucket *new_bucket;
6658 };
6659
6660 /*
6661 * NOTE:
6662 * We have to handle the case that both old bucket and new bucket
6663 * will call this function to get the right ret_bh.
6664 * So The caller must give us the right bh.
6665 */
6666 static int ocfs2_get_reflink_xattr_value_root(struct super_block *sb,
6667 struct buffer_head *bh,
6668 struct ocfs2_xattr_header *xh,
6669 int offset,
6670 struct ocfs2_xattr_value_root **xv,
6671 struct buffer_head **ret_bh,
6672 void *para)
6673 {
6674 struct ocfs2_reflink_xattr_tree_args *args =
6675 (struct ocfs2_reflink_xattr_tree_args *)para;
6676 struct ocfs2_xattr_bucket *bucket;
6677
6678 if (bh == args->old_bucket->bu_bhs[0])
6679 bucket = args->old_bucket;
6680 else
6681 bucket = args->new_bucket;
6682
6683 return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6684 xv, ret_bh);
6685 }
6686
6687 struct ocfs2_value_tree_metas {
6688 int num_metas;
6689 int credits;
6690 int num_recs;
6691 };
6692
6693 static int ocfs2_value_tree_metas_in_bucket(struct super_block *sb,
6694 struct buffer_head *bh,
6695 struct ocfs2_xattr_header *xh,
6696 int offset,
6697 struct ocfs2_xattr_value_root **xv,
6698 struct buffer_head **ret_bh,
6699 void *para)
6700 {
6701 struct ocfs2_xattr_bucket *bucket =
6702 (struct ocfs2_xattr_bucket *)para;
6703
6704 return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6705 xv, ret_bh);
6706 }
6707
6708 static int ocfs2_calc_value_tree_metas(struct inode *inode,
6709 struct ocfs2_xattr_bucket *bucket,
6710 void *para)
6711 {
6712 struct ocfs2_value_tree_metas *metas =
6713 (struct ocfs2_value_tree_metas *)para;
6714 struct ocfs2_xattr_header *xh =
6715 (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6716
6717 /* Add the credits for this bucket first. */
6718 metas->credits += bucket->bu_blocks;
6719 return ocfs2_value_metas_in_xattr_header(inode->i_sb, bucket->bu_bhs[0],
6720 xh, &metas->num_metas,
6721 &metas->credits, &metas->num_recs,
6722 ocfs2_value_tree_metas_in_bucket,
6723 bucket);
6724 }
6725
6726 /*
6727 * Given a xattr extent rec starting from blkno and having len clusters,
6728 * iterate all the buckets calculate how much metadata we need for reflinking
6729 * all the ocfs2_xattr_value_root and lock the allocators accordingly.
6730 */
6731 static int ocfs2_lock_reflink_xattr_rec_allocators(
6732 struct ocfs2_reflink_xattr_tree_args *args,
6733 struct ocfs2_extent_tree *xt_et,
6734 u64 blkno, u32 len, int *credits,
6735 struct ocfs2_alloc_context **meta_ac,
6736 struct ocfs2_alloc_context **data_ac)
6737 {
6738 int ret, num_free_extents;
6739 struct ocfs2_value_tree_metas metas;
6740 struct ocfs2_super *osb = OCFS2_SB(args->reflink->old_inode->i_sb);
6741 struct ocfs2_refcount_block *rb;
6742
6743 memset(&metas, 0, sizeof(metas));
6744
6745 ret = ocfs2_iterate_xattr_buckets(args->reflink->old_inode, blkno, len,
6746 ocfs2_calc_value_tree_metas, &metas);
6747 if (ret) {
6748 mlog_errno(ret);
6749 goto out;
6750 }
6751
6752 *credits = metas.credits;
6753
6754 /*
6755 * Calculate we need for refcount tree change.
6756 *
6757 * We need to add/modify num_recs in refcount tree, so just calculate
6758 * an approximate number we need for refcount tree change.
6759 * Sometimes we need to split the tree, and after split, half recs
6760 * will be moved to the new block, and a new block can only provide
6761 * half number of recs. So we multiple new blocks by 2.
6762 * In the end, we have to add credits for modifying the already
6763 * existed refcount block.
6764 */
6765 rb = (struct ocfs2_refcount_block *)args->reflink->ref_root_bh->b_data;
6766 metas.num_recs =
6767 (metas.num_recs + ocfs2_refcount_recs_per_rb(osb->sb) - 1) /
6768 ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6769 metas.num_metas += metas.num_recs;
6770 *credits += metas.num_recs +
6771 metas.num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6772 if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6773 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6774 le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6775 else
6776 *credits += 1;
6777
6778 /* count in the xattr tree change. */
6779 num_free_extents = ocfs2_num_free_extents(xt_et);
6780 if (num_free_extents < 0) {
6781 ret = num_free_extents;
6782 mlog_errno(ret);
6783 goto out;
6784 }
6785
6786 if (num_free_extents < len)
6787 metas.num_metas += ocfs2_extend_meta_needed(xt_et->et_root_el);
6788
6789 *credits += ocfs2_calc_extend_credits(osb->sb,
6790 xt_et->et_root_el);
6791
6792 if (metas.num_metas) {
6793 ret = ocfs2_reserve_new_metadata_blocks(osb, metas.num_metas,
6794 meta_ac);
6795 if (ret) {
6796 mlog_errno(ret);
6797 goto out;
6798 }
6799 }
6800
6801 if (len) {
6802 ret = ocfs2_reserve_clusters(osb, len, data_ac);
6803 if (ret)
6804 mlog_errno(ret);
6805 }
6806 out:
6807 if (ret) {
6808 if (*meta_ac) {
6809 ocfs2_free_alloc_context(*meta_ac);
6810 *meta_ac = NULL;
6811 }
6812 }
6813
6814 return ret;
6815 }
6816
6817 static int ocfs2_reflink_xattr_bucket(handle_t *handle,
6818 u64 blkno, u64 new_blkno, u32 clusters,
6819 u32 *cpos, int num_buckets,
6820 struct ocfs2_alloc_context *meta_ac,
6821 struct ocfs2_alloc_context *data_ac,
6822 struct ocfs2_reflink_xattr_tree_args *args)
6823 {
6824 int i, j, ret = 0;
6825 struct super_block *sb = args->reflink->old_inode->i_sb;
6826 int bpb = args->old_bucket->bu_blocks;
6827 struct ocfs2_xattr_value_buf vb = {
6828 .vb_access = ocfs2_journal_access,
6829 };
6830
6831 for (i = 0; i < num_buckets; i++, blkno += bpb, new_blkno += bpb) {
6832 ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6833 if (ret) {
6834 mlog_errno(ret);
6835 break;
6836 }
6837
6838 ret = ocfs2_init_xattr_bucket(args->new_bucket, new_blkno, 1);
6839 if (ret) {
6840 mlog_errno(ret);
6841 break;
6842 }
6843
6844 ret = ocfs2_xattr_bucket_journal_access(handle,
6845 args->new_bucket,
6846 OCFS2_JOURNAL_ACCESS_CREATE);
6847 if (ret) {
6848 mlog_errno(ret);
6849 break;
6850 }
6851
6852 for (j = 0; j < bpb; j++)
6853 memcpy(bucket_block(args->new_bucket, j),
6854 bucket_block(args->old_bucket, j),
6855 sb->s_blocksize);
6856
6857 /*
6858 * Record the start cpos so that we can use it to initialize
6859 * our xattr tree we also set the xh_num_bucket for the new
6860 * bucket.
6861 */
6862 if (i == 0) {
6863 *cpos = le32_to_cpu(bucket_xh(args->new_bucket)->
6864 xh_entries[0].xe_name_hash);
6865 bucket_xh(args->new_bucket)->xh_num_buckets =
6866 cpu_to_le16(num_buckets);
6867 }
6868
6869 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6870
6871 ret = ocfs2_reflink_xattr_header(handle, args->reflink,
6872 args->old_bucket->bu_bhs[0],
6873 bucket_xh(args->old_bucket),
6874 args->new_bucket->bu_bhs[0],
6875 bucket_xh(args->new_bucket),
6876 &vb, meta_ac,
6877 ocfs2_get_reflink_xattr_value_root,
6878 args);
6879 if (ret) {
6880 mlog_errno(ret);
6881 break;
6882 }
6883
6884 /*
6885 * Re-access and dirty the bucket to calculate metaecc.
6886 * Because we may extend the transaction in reflink_xattr_header
6887 * which will let the already accessed block gone.
6888 */
6889 ret = ocfs2_xattr_bucket_journal_access(handle,
6890 args->new_bucket,
6891 OCFS2_JOURNAL_ACCESS_WRITE);
6892 if (ret) {
6893 mlog_errno(ret);
6894 break;
6895 }
6896
6897 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6898
6899 ocfs2_xattr_bucket_relse(args->old_bucket);
6900 ocfs2_xattr_bucket_relse(args->new_bucket);
6901 }
6902
6903 ocfs2_xattr_bucket_relse(args->old_bucket);
6904 ocfs2_xattr_bucket_relse(args->new_bucket);
6905 return ret;
6906 }
6907
6908 static int ocfs2_reflink_xattr_buckets(handle_t *handle,
6909 struct inode *inode,
6910 struct ocfs2_reflink_xattr_tree_args *args,
6911 struct ocfs2_extent_tree *et,
6912 struct ocfs2_alloc_context *meta_ac,
6913 struct ocfs2_alloc_context *data_ac,
6914 u64 blkno, u32 cpos, u32 len)
6915 {
6916 int ret, first_inserted = 0;
6917 u32 p_cluster, num_clusters, reflink_cpos = 0;
6918 u64 new_blkno;
6919 unsigned int num_buckets, reflink_buckets;
6920 unsigned int bpc =
6921 ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
6922
6923 ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6924 if (ret) {
6925 mlog_errno(ret);
6926 goto out;
6927 }
6928 num_buckets = le16_to_cpu(bucket_xh(args->old_bucket)->xh_num_buckets);
6929 ocfs2_xattr_bucket_relse(args->old_bucket);
6930
6931 while (len && num_buckets) {
6932 ret = ocfs2_claim_clusters(handle, data_ac,
6933 1, &p_cluster, &num_clusters);
6934 if (ret) {
6935 mlog_errno(ret);
6936 goto out;
6937 }
6938
6939 new_blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
6940 reflink_buckets = min(num_buckets, bpc * num_clusters);
6941
6942 ret = ocfs2_reflink_xattr_bucket(handle, blkno,
6943 new_blkno, num_clusters,
6944 &reflink_cpos, reflink_buckets,
6945 meta_ac, data_ac, args);
6946 if (ret) {
6947 mlog_errno(ret);
6948 goto out;
6949 }
6950
6951 /*
6952 * For the 1st allocated cluster, we make it use the same cpos
6953 * so that the xattr tree looks the same as the original one
6954 * in the most case.
6955 */
6956 if (!first_inserted) {
6957 reflink_cpos = cpos;
6958 first_inserted = 1;
6959 }
6960 ret = ocfs2_insert_extent(handle, et, reflink_cpos, new_blkno,
6961 num_clusters, 0, meta_ac);
6962 if (ret)
6963 mlog_errno(ret);
6964
6965 trace_ocfs2_reflink_xattr_buckets((unsigned long long)new_blkno,
6966 num_clusters, reflink_cpos);
6967
6968 len -= num_clusters;
6969 blkno += ocfs2_clusters_to_blocks(inode->i_sb, num_clusters);
6970 num_buckets -= reflink_buckets;
6971 }
6972 out:
6973 return ret;
6974 }
6975
6976 /*
6977 * Create the same xattr extent record in the new inode's xattr tree.
6978 */
6979 static int ocfs2_reflink_xattr_rec(struct inode *inode,
6980 struct buffer_head *root_bh,
6981 u64 blkno,
6982 u32 cpos,
6983 u32 len,
6984 void *para)
6985 {
6986 int ret, credits = 0;
6987 handle_t *handle;
6988 struct ocfs2_reflink_xattr_tree_args *args =
6989 (struct ocfs2_reflink_xattr_tree_args *)para;
6990 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6991 struct ocfs2_alloc_context *meta_ac = NULL;
6992 struct ocfs2_alloc_context *data_ac = NULL;
6993 struct ocfs2_extent_tree et;
6994
6995 trace_ocfs2_reflink_xattr_rec((unsigned long long)blkno, len);
6996
6997 ocfs2_init_xattr_tree_extent_tree(&et,
6998 INODE_CACHE(args->reflink->new_inode),
6999 args->new_blk_bh);
7000
7001 ret = ocfs2_lock_reflink_xattr_rec_allocators(args, &et, blkno,
7002 len, &credits,
7003 &meta_ac, &data_ac);
7004 if (ret) {
7005 mlog_errno(ret);
7006 goto out;
7007 }
7008
7009 handle = ocfs2_start_trans(osb, credits);
7010 if (IS_ERR(handle)) {
7011 ret = PTR_ERR(handle);
7012 mlog_errno(ret);
7013 goto out;
7014 }
7015
7016 ret = ocfs2_reflink_xattr_buckets(handle, inode, args, &et,
7017 meta_ac, data_ac,
7018 blkno, cpos, len);
7019 if (ret)
7020 mlog_errno(ret);
7021
7022 ocfs2_commit_trans(osb, handle);
7023
7024 out:
7025 if (meta_ac)
7026 ocfs2_free_alloc_context(meta_ac);
7027 if (data_ac)
7028 ocfs2_free_alloc_context(data_ac);
7029 return ret;
7030 }
7031
7032 /*
7033 * Create reflinked xattr buckets.
7034 * We will add bucket one by one, and refcount all the xattrs in the bucket
7035 * if they are stored outside.
7036 */
7037 static int ocfs2_reflink_xattr_tree(struct ocfs2_xattr_reflink *args,
7038 struct buffer_head *blk_bh,
7039 struct buffer_head *new_blk_bh)
7040 {
7041 int ret;
7042 struct ocfs2_reflink_xattr_tree_args para;
7043
7044 memset(&para, 0, sizeof(para));
7045 para.reflink = args;
7046 para.old_blk_bh = blk_bh;
7047 para.new_blk_bh = new_blk_bh;
7048
7049 para.old_bucket = ocfs2_xattr_bucket_new(args->old_inode);
7050 if (!para.old_bucket) {
7051 mlog_errno(-ENOMEM);
7052 return -ENOMEM;
7053 }
7054
7055 para.new_bucket = ocfs2_xattr_bucket_new(args->new_inode);
7056 if (!para.new_bucket) {
7057 ret = -ENOMEM;
7058 mlog_errno(ret);
7059 goto out;
7060 }
7061
7062 ret = ocfs2_iterate_xattr_index_block(args->old_inode, blk_bh,
7063 ocfs2_reflink_xattr_rec,
7064 &para);
7065 if (ret)
7066 mlog_errno(ret);
7067
7068 out:
7069 ocfs2_xattr_bucket_free(para.old_bucket);
7070 ocfs2_xattr_bucket_free(para.new_bucket);
7071 return ret;
7072 }
7073
7074 static int ocfs2_reflink_xattr_in_block(struct ocfs2_xattr_reflink *args,
7075 struct buffer_head *blk_bh)
7076 {
7077 int ret, indexed = 0;
7078 struct buffer_head *new_blk_bh = NULL;
7079 struct ocfs2_xattr_block *xb =
7080 (struct ocfs2_xattr_block *)blk_bh->b_data;
7081
7082
7083 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)
7084 indexed = 1;
7085
7086 ret = ocfs2_create_empty_xattr_block(args->new_inode, args->new_bh,
7087 &new_blk_bh, indexed);
7088 if (ret) {
7089 mlog_errno(ret);
7090 goto out;
7091 }
7092
7093 if (!indexed)
7094 ret = ocfs2_reflink_xattr_block(args, blk_bh, new_blk_bh);
7095 else
7096 ret = ocfs2_reflink_xattr_tree(args, blk_bh, new_blk_bh);
7097 if (ret)
7098 mlog_errno(ret);
7099
7100 out:
7101 brelse(new_blk_bh);
7102 return ret;
7103 }
7104
7105 static int ocfs2_reflink_xattr_no_security(struct ocfs2_xattr_entry *xe)
7106 {
7107 int type = ocfs2_xattr_get_type(xe);
7108
7109 return type != OCFS2_XATTR_INDEX_SECURITY &&
7110 type != OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS &&
7111 type != OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT;
7112 }
7113
7114 int ocfs2_reflink_xattrs(struct inode *old_inode,
7115 struct buffer_head *old_bh,
7116 struct inode *new_inode,
7117 struct buffer_head *new_bh,
7118 bool preserve_security)
7119 {
7120 int ret;
7121 struct ocfs2_xattr_reflink args;
7122 struct ocfs2_inode_info *oi = OCFS2_I(old_inode);
7123 struct ocfs2_dinode *di = (struct ocfs2_dinode *)old_bh->b_data;
7124 struct buffer_head *blk_bh = NULL;
7125 struct ocfs2_cached_dealloc_ctxt dealloc;
7126 struct ocfs2_refcount_tree *ref_tree;
7127 struct buffer_head *ref_root_bh = NULL;
7128
7129 ret = ocfs2_lock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7130 le64_to_cpu(di->i_refcount_loc),
7131 1, &ref_tree, &ref_root_bh);
7132 if (ret) {
7133 mlog_errno(ret);
7134 goto out;
7135 }
7136
7137 ocfs2_init_dealloc_ctxt(&dealloc);
7138
7139 args.old_inode = old_inode;
7140 args.new_inode = new_inode;
7141 args.old_bh = old_bh;
7142 args.new_bh = new_bh;
7143 args.ref_ci = &ref_tree->rf_ci;
7144 args.ref_root_bh = ref_root_bh;
7145 args.dealloc = &dealloc;
7146 if (preserve_security)
7147 args.xattr_reflinked = NULL;
7148 else
7149 args.xattr_reflinked = ocfs2_reflink_xattr_no_security;
7150
7151 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
7152 ret = ocfs2_reflink_xattr_inline(&args);
7153 if (ret) {
7154 mlog_errno(ret);
7155 goto out_unlock;
7156 }
7157 }
7158
7159 if (!di->i_xattr_loc)
7160 goto out_unlock;
7161
7162 ret = ocfs2_read_xattr_block(old_inode, le64_to_cpu(di->i_xattr_loc),
7163 &blk_bh);
7164 if (ret < 0) {
7165 mlog_errno(ret);
7166 goto out_unlock;
7167 }
7168
7169 ret = ocfs2_reflink_xattr_in_block(&args, blk_bh);
7170 if (ret)
7171 mlog_errno(ret);
7172
7173 brelse(blk_bh);
7174
7175 out_unlock:
7176 ocfs2_unlock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7177 ref_tree, 1);
7178 brelse(ref_root_bh);
7179
7180 if (ocfs2_dealloc_has_cluster(&dealloc)) {
7181 ocfs2_schedule_truncate_log_flush(OCFS2_SB(old_inode->i_sb), 1);
7182 ocfs2_run_deallocs(OCFS2_SB(old_inode->i_sb), &dealloc);
7183 }
7184
7185 out:
7186 return ret;
7187 }
7188
7189 /*
7190 * Initialize security and acl for a already created inode.
7191 * Used for reflink a non-preserve-security file.
7192 *
7193 * It uses common api like ocfs2_xattr_set, so the caller
7194 * must not hold any lock expect i_rwsem.
7195 */
7196 int ocfs2_init_security_and_acl(struct inode *dir,
7197 struct inode *inode,
7198 const struct qstr *qstr)
7199 {
7200 int ret = 0;
7201 struct buffer_head *dir_bh = NULL;
7202
7203 ret = ocfs2_init_security_get(inode, dir, qstr, NULL);
7204 if (ret) {
7205 mlog_errno(ret);
7206 goto leave;
7207 }
7208
7209 ret = ocfs2_inode_lock(dir, &dir_bh, 0);
7210 if (ret) {
7211 mlog_errno(ret);
7212 goto leave;
7213 }
7214 ret = ocfs2_init_acl(NULL, inode, dir, NULL, dir_bh, NULL, NULL);
7215 if (ret)
7216 mlog_errno(ret);
7217
7218 ocfs2_inode_unlock(dir, 0);
7219 brelse(dir_bh);
7220 leave:
7221 return ret;
7222 }
7223
7224 /*
7225 * 'security' attributes support
7226 */
7227 static int ocfs2_xattr_security_get(const struct xattr_handler *handler,
7228 struct dentry *unused, struct inode *inode,
7229 const char *name, void *buffer, size_t size)
7230 {
7231 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_SECURITY,
7232 name, buffer, size);
7233 }
7234
7235 static int ocfs2_xattr_security_set(const struct xattr_handler *handler,
7236 struct mnt_idmap *idmap,
7237 struct dentry *unused, struct inode *inode,
7238 const char *name, const void *value,
7239 size_t size, int flags)
7240 {
7241 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY,
7242 name, value, size, flags);
7243 }
7244
7245 static int ocfs2_initxattrs(struct inode *inode, const struct xattr *xattr_array,
7246 void *fs_info)
7247 {
7248 struct ocfs2_security_xattr_info *si = fs_info;
7249 const struct xattr *xattr;
7250 int err = 0;
7251
7252 if (si) {
7253 si->value = kmemdup(xattr_array->value, xattr_array->value_len,
7254 GFP_KERNEL);
7255 if (!si->value)
7256 return -ENOMEM;
7257
7258 si->name = xattr_array->name;
7259 si->value_len = xattr_array->value_len;
7260 return 0;
7261 }
7262
7263 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
7264 err = ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY,
7265 xattr->name, xattr->value,
7266 xattr->value_len, XATTR_CREATE);
7267 if (err)
7268 break;
7269 }
7270 return err;
7271 }
7272
7273 int ocfs2_init_security_get(struct inode *inode,
7274 struct inode *dir,
7275 const struct qstr *qstr,
7276 struct ocfs2_security_xattr_info *si)
7277 {
7278 int ret;
7279
7280 /* check whether ocfs2 support feature xattr */
7281 if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
7282 return -EOPNOTSUPP;
7283 if (si) {
7284 ret = security_inode_init_security(inode, dir, qstr,
7285 &ocfs2_initxattrs, si);
7286 /*
7287 * security_inode_init_security() does not return -EOPNOTSUPP,
7288 * we have to check the xattr ourselves.
7289 */
7290 if (!ret && !si->name)
7291 si->enable = 0;
7292
7293 return ret;
7294 }
7295
7296 return security_inode_init_security(inode, dir, qstr,
7297 &ocfs2_initxattrs, NULL);
7298 }
7299
7300 int ocfs2_init_security_set(handle_t *handle,
7301 struct inode *inode,
7302 struct buffer_head *di_bh,
7303 struct ocfs2_security_xattr_info *si,
7304 struct ocfs2_alloc_context *xattr_ac,
7305 struct ocfs2_alloc_context *data_ac)
7306 {
7307 return ocfs2_xattr_set_handle(handle, inode, di_bh,
7308 OCFS2_XATTR_INDEX_SECURITY,
7309 si->name, si->value, si->value_len, 0,
7310 xattr_ac, data_ac);
7311 }
7312
7313 const struct xattr_handler ocfs2_xattr_security_handler = {
7314 .prefix = XATTR_SECURITY_PREFIX,
7315 .get = ocfs2_xattr_security_get,
7316 .set = ocfs2_xattr_security_set,
7317 };
7318
7319 /*
7320 * 'trusted' attributes support
7321 */
7322 static int ocfs2_xattr_trusted_get(const struct xattr_handler *handler,
7323 struct dentry *unused, struct inode *inode,
7324 const char *name, void *buffer, size_t size)
7325 {
7326 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED,
7327 name, buffer, size);
7328 }
7329
7330 static int ocfs2_xattr_trusted_set(const struct xattr_handler *handler,
7331 struct mnt_idmap *idmap,
7332 struct dentry *unused, struct inode *inode,
7333 const char *name, const void *value,
7334 size_t size, int flags)
7335 {
7336 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED,
7337 name, value, size, flags);
7338 }
7339
7340 const struct xattr_handler ocfs2_xattr_trusted_handler = {
7341 .prefix = XATTR_TRUSTED_PREFIX,
7342 .get = ocfs2_xattr_trusted_get,
7343 .set = ocfs2_xattr_trusted_set,
7344 };
7345
7346 /*
7347 * 'user' attributes support
7348 */
7349 static int ocfs2_xattr_user_get(const struct xattr_handler *handler,
7350 struct dentry *unused, struct inode *inode,
7351 const char *name, void *buffer, size_t size)
7352 {
7353 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7354
7355 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7356 return -EOPNOTSUPP;
7357 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
7358 buffer, size);
7359 }
7360
7361 static int ocfs2_xattr_user_set(const struct xattr_handler *handler,
7362 struct mnt_idmap *idmap,
7363 struct dentry *unused, struct inode *inode,
7364 const char *name, const void *value,
7365 size_t size, int flags)
7366 {
7367 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7368
7369 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7370 return -EOPNOTSUPP;
7371
7372 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER,
7373 name, value, size, flags);
7374 }
7375
7376 const struct xattr_handler ocfs2_xattr_user_handler = {
7377 .prefix = XATTR_USER_PREFIX,
7378 .get = ocfs2_xattr_user_get,
7379 .set = ocfs2_xattr_user_set,
7380 };
Linux kernel stable tree