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 if (rc)
2040 goto out;
2041 }
2042 }
2043
2044 ocfs2_xa_remove_entry(loc);
2045
2046 out:
2047 return rc;
2048 }
2049
2050 static void ocfs2_xa_install_value_root(struct ocfs2_xa_loc *loc)
2051 {
2052 int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
2053 char *nameval_buf;
2054
2055 nameval_buf = ocfs2_xa_offset_pointer(loc,
2056 le16_to_cpu(loc->xl_entry->xe_name_offset));
2057 memcpy(nameval_buf + name_size, &def_xv, OCFS2_XATTR_ROOT_SIZE);
2058 }
2059
2060 /*
2061 * Take an existing entry and make it ready for the new value. This
2062 * won't allocate space, but it may free space. It should be ready for
2063 * ocfs2_xa_prepare_entry() to finish the work.
2064 */
2065 static int ocfs2_xa_reuse_entry(struct ocfs2_xa_loc *loc,
2066 struct ocfs2_xattr_info *xi,
2067 struct ocfs2_xattr_set_ctxt *ctxt)
2068 {
2069 int rc = 0;
2070 int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2071 unsigned int orig_clusters;
2072 char *nameval_buf;
2073 int xe_local = ocfs2_xattr_is_local(loc->xl_entry);
2074 int xi_local = xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE;
2075
2076 BUG_ON(OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len) !=
2077 name_size);
2078
2079 nameval_buf = ocfs2_xa_offset_pointer(loc,
2080 le16_to_cpu(loc->xl_entry->xe_name_offset));
2081 if (xe_local) {
2082 memset(nameval_buf + name_size, 0,
2083 namevalue_size_xe(loc->xl_entry) - name_size);
2084 if (!xi_local)
2085 ocfs2_xa_install_value_root(loc);
2086 } else {
2087 orig_clusters = ocfs2_xa_value_clusters(loc);
2088 if (xi_local) {
2089 rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2090 if (rc < 0)
2091 mlog_errno(rc);
2092 else
2093 memset(nameval_buf + name_size, 0,
2094 namevalue_size_xe(loc->xl_entry) -
2095 name_size);
2096 } else if (le64_to_cpu(loc->xl_entry->xe_value_size) >
2097 xi->xi_value_len) {
2098 rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len,
2099 ctxt);
2100 if (rc < 0)
2101 mlog_errno(rc);
2102 }
2103
2104 if (rc) {
2105 ocfs2_xa_cleanup_value_truncate(loc, "reusing",
2106 orig_clusters);
2107 goto out;
2108 }
2109 }
2110
2111 loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
2112 ocfs2_xattr_set_local(loc->xl_entry, xi_local);
2113
2114 out:
2115 return rc;
2116 }
2117
2118 /*
2119 * Prepares loc->xl_entry to receive the new xattr. This includes
2120 * properly setting up the name+value pair region. If loc->xl_entry
2121 * already exists, it will take care of modifying it appropriately.
2122 *
2123 * Note that this modifies the data. You did journal_access already,
2124 * right?
2125 */
2126 static int ocfs2_xa_prepare_entry(struct ocfs2_xa_loc *loc,
2127 struct ocfs2_xattr_info *xi,
2128 u32 name_hash,
2129 struct ocfs2_xattr_set_ctxt *ctxt)
2130 {
2131 int rc = 0;
2132 unsigned int orig_clusters;
2133 __le64 orig_value_size = 0;
2134
2135 rc = ocfs2_xa_check_space(loc, xi);
2136 if (rc)
2137 goto out;
2138
2139 if (loc->xl_entry) {
2140 if (ocfs2_xa_can_reuse_entry(loc, xi)) {
2141 orig_value_size = loc->xl_entry->xe_value_size;
2142 rc = ocfs2_xa_reuse_entry(loc, xi, ctxt);
2143 if (rc)
2144 goto out;
2145 goto alloc_value;
2146 }
2147
2148 if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2149 orig_clusters = ocfs2_xa_value_clusters(loc);
2150 rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2151 if (rc) {
2152 mlog_errno(rc);
2153 ocfs2_xa_cleanup_value_truncate(loc,
2154 "overwriting",
2155 orig_clusters);
2156 goto out;
2157 }
2158 }
2159 ocfs2_xa_wipe_namevalue(loc);
2160 } else
2161 ocfs2_xa_add_entry(loc, name_hash);
2162
2163 /*
2164 * If we get here, we have a blank entry. Fill it. We grow our
2165 * name+value pair back from the end.
2166 */
2167 ocfs2_xa_add_namevalue(loc, xi);
2168 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
2169 ocfs2_xa_install_value_root(loc);
2170
2171 alloc_value:
2172 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2173 orig_clusters = ocfs2_xa_value_clusters(loc);
2174 rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len, ctxt);
2175 if (rc < 0) {
2176 ctxt->set_abort = 1;
2177 ocfs2_xa_cleanup_value_truncate(loc, "growing",
2178 orig_clusters);
2179 /*
2180 * If we were growing an existing value,
2181 * ocfs2_xa_cleanup_value_truncate() won't remove
2182 * the entry. We need to restore the original value
2183 * size.
2184 */
2185 if (loc->xl_entry) {
2186 BUG_ON(!orig_value_size);
2187 loc->xl_entry->xe_value_size = orig_value_size;
2188 }
2189 mlog_errno(rc);
2190 }
2191 }
2192
2193 out:
2194 return rc;
2195 }
2196
2197 /*
2198 * Store the value portion of the name+value pair. This will skip
2199 * values that are stored externally. Their tree roots were set up
2200 * by ocfs2_xa_prepare_entry().
2201 */
2202 static int ocfs2_xa_store_value(struct ocfs2_xa_loc *loc,
2203 struct ocfs2_xattr_info *xi,
2204 struct ocfs2_xattr_set_ctxt *ctxt)
2205 {
2206 int rc = 0;
2207 int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
2208 int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2209 char *nameval_buf;
2210 struct ocfs2_xattr_value_buf vb;
2211
2212 nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
2213 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2214 ocfs2_xa_fill_value_buf(loc, &vb);
2215 rc = __ocfs2_xattr_set_value_outside(loc->xl_inode,
2216 ctxt->handle, &vb,
2217 xi->xi_value,
2218 xi->xi_value_len);
2219 } else
2220 memcpy(nameval_buf + name_size, xi->xi_value, xi->xi_value_len);
2221
2222 return rc;
2223 }
2224
2225 static int ocfs2_xa_set(struct ocfs2_xa_loc *loc,
2226 struct ocfs2_xattr_info *xi,
2227 struct ocfs2_xattr_set_ctxt *ctxt)
2228 {
2229 int ret;
2230 u32 name_hash = ocfs2_xattr_name_hash(loc->xl_inode, xi->xi_name,
2231 xi->xi_name_len);
2232
2233 ret = ocfs2_xa_journal_access(ctxt->handle, loc,
2234 OCFS2_JOURNAL_ACCESS_WRITE);
2235 if (ret) {
2236 mlog_errno(ret);
2237 goto out;
2238 }
2239
2240 /*
2241 * From here on out, everything is going to modify the buffer a
2242 * little. Errors are going to leave the xattr header in a
2243 * sane state. Thus, even with errors we dirty the sucker.
2244 */
2245
2246 /* Don't worry, we are never called with !xi_value and !xl_entry */
2247 if (!xi->xi_value) {
2248 ret = ocfs2_xa_remove(loc, ctxt);
2249 goto out_dirty;
2250 }
2251
2252 ret = ocfs2_xa_prepare_entry(loc, xi, name_hash, ctxt);
2253 if (ret) {
2254 if (ret != -ENOSPC)
2255 mlog_errno(ret);
2256 goto out_dirty;
2257 }
2258
2259 ret = ocfs2_xa_store_value(loc, xi, ctxt);
2260 if (ret)
2261 mlog_errno(ret);
2262
2263 out_dirty:
2264 ocfs2_xa_journal_dirty(ctxt->handle, loc);
2265
2266 out:
2267 return ret;
2268 }
2269
2270 static void ocfs2_init_dinode_xa_loc(struct ocfs2_xa_loc *loc,
2271 struct inode *inode,
2272 struct buffer_head *bh,
2273 struct ocfs2_xattr_entry *entry)
2274 {
2275 struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2276
2277 BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_XATTR_FL));
2278
2279 loc->xl_inode = inode;
2280 loc->xl_ops = &ocfs2_xa_block_loc_ops;
2281 loc->xl_storage = bh;
2282 loc->xl_entry = entry;
2283 loc->xl_size = le16_to_cpu(di->i_xattr_inline_size);
2284 loc->xl_header =
2285 (struct ocfs2_xattr_header *)(bh->b_data + bh->b_size -
2286 loc->xl_size);
2287 }
2288
2289 static void ocfs2_init_xattr_block_xa_loc(struct ocfs2_xa_loc *loc,
2290 struct inode *inode,
2291 struct buffer_head *bh,
2292 struct ocfs2_xattr_entry *entry)
2293 {
2294 struct ocfs2_xattr_block *xb =
2295 (struct ocfs2_xattr_block *)bh->b_data;
2296
2297 BUG_ON(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED);
2298
2299 loc->xl_inode = inode;
2300 loc->xl_ops = &ocfs2_xa_block_loc_ops;
2301 loc->xl_storage = bh;
2302 loc->xl_header = &(xb->xb_attrs.xb_header);
2303 loc->xl_entry = entry;
2304 loc->xl_size = bh->b_size - offsetof(struct ocfs2_xattr_block,
2305 xb_attrs.xb_header);
2306 }
2307
2308 static void ocfs2_init_xattr_bucket_xa_loc(struct ocfs2_xa_loc *loc,
2309 struct ocfs2_xattr_bucket *bucket,
2310 struct ocfs2_xattr_entry *entry)
2311 {
2312 loc->xl_inode = bucket->bu_inode;
2313 loc->xl_ops = &ocfs2_xa_bucket_loc_ops;
2314 loc->xl_storage = bucket;
2315 loc->xl_header = bucket_xh(bucket);
2316 loc->xl_entry = entry;
2317 loc->xl_size = OCFS2_XATTR_BUCKET_SIZE;
2318 }
2319
2320 /*
2321 * In xattr remove, if it is stored outside and refcounted, we may have
2322 * the chance to split the refcount tree. So need the allocators.
2323 */
2324 static int ocfs2_lock_xattr_remove_allocators(struct inode *inode,
2325 struct ocfs2_xattr_value_root *xv,
2326 struct ocfs2_caching_info *ref_ci,
2327 struct buffer_head *ref_root_bh,
2328 struct ocfs2_alloc_context **meta_ac,
2329 int *ref_credits)
2330 {
2331 int ret, meta_add = 0;
2332 u32 p_cluster, num_clusters;
2333 unsigned int ext_flags;
2334
2335 *ref_credits = 0;
2336 ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
2337 &num_clusters,
2338 &xv->xr_list,
2339 &ext_flags);
2340 if (ret) {
2341 mlog_errno(ret);
2342 goto out;
2343 }
2344
2345 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
2346 goto out;
2347
2348 ret = ocfs2_refcounted_xattr_delete_need(inode, ref_ci,
2349 ref_root_bh, xv,
2350 &meta_add, ref_credits);
2351 if (ret) {
2352 mlog_errno(ret);
2353 goto out;
2354 }
2355
2356 ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
2357 meta_add, meta_ac);
2358 if (ret)
2359 mlog_errno(ret);
2360
2361 out:
2362 return ret;
2363 }
2364
2365 static int ocfs2_remove_value_outside(struct inode*inode,
2366 struct ocfs2_xattr_value_buf *vb,
2367 struct ocfs2_xattr_header *header,
2368 struct ocfs2_caching_info *ref_ci,
2369 struct buffer_head *ref_root_bh)
2370 {
2371 int ret = 0, i, ref_credits;
2372 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2373 struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2374 void *val;
2375
2376 ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
2377
2378 for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
2379 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
2380
2381 if (ocfs2_xattr_is_local(entry))
2382 continue;
2383
2384 val = (void *)header +
2385 le16_to_cpu(entry->xe_name_offset);
2386 vb->vb_xv = (struct ocfs2_xattr_value_root *)
2387 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
2388
2389 ret = ocfs2_lock_xattr_remove_allocators(inode, vb->vb_xv,
2390 ref_ci, ref_root_bh,
2391 &ctxt.meta_ac,
2392 &ref_credits);
2393
2394 ctxt.handle = ocfs2_start_trans(osb, ref_credits +
2395 ocfs2_remove_extent_credits(osb->sb));
2396 if (IS_ERR(ctxt.handle)) {
2397 ret = PTR_ERR(ctxt.handle);
2398 mlog_errno(ret);
2399 break;
2400 }
2401
2402 ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
2403
2404 ocfs2_commit_trans(osb, ctxt.handle);
2405 if (ctxt.meta_ac) {
2406 ocfs2_free_alloc_context(ctxt.meta_ac);
2407 ctxt.meta_ac = NULL;
2408 }
2409
2410 if (ret < 0) {
2411 mlog_errno(ret);
2412 break;
2413 }
2414
2415 }
2416
2417 if (ctxt.meta_ac)
2418 ocfs2_free_alloc_context(ctxt.meta_ac);
2419 ocfs2_schedule_truncate_log_flush(osb, 1);
2420 ocfs2_run_deallocs(osb, &ctxt.dealloc);
2421 return ret;
2422 }
2423
2424 static int ocfs2_xattr_ibody_remove(struct inode *inode,
2425 struct buffer_head *di_bh,
2426 struct ocfs2_caching_info *ref_ci,
2427 struct buffer_head *ref_root_bh)
2428 {
2429
2430 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2431 struct ocfs2_xattr_header *header;
2432 int ret;
2433 struct ocfs2_xattr_value_buf vb = {
2434 .vb_bh = di_bh,
2435 .vb_access = ocfs2_journal_access_di,
2436 };
2437
2438 header = (struct ocfs2_xattr_header *)
2439 ((void *)di + inode->i_sb->s_blocksize -
2440 le16_to_cpu(di->i_xattr_inline_size));
2441
2442 ret = ocfs2_remove_value_outside(inode, &vb, header,
2443 ref_ci, ref_root_bh);
2444
2445 return ret;
2446 }
2447
2448 struct ocfs2_rm_xattr_bucket_para {
2449 struct ocfs2_caching_info *ref_ci;
2450 struct buffer_head *ref_root_bh;
2451 };
2452
2453 static int ocfs2_xattr_block_remove(struct inode *inode,
2454 struct buffer_head *blk_bh,
2455 struct ocfs2_caching_info *ref_ci,
2456 struct buffer_head *ref_root_bh)
2457 {
2458 struct ocfs2_xattr_block *xb;
2459 int ret = 0;
2460 struct ocfs2_xattr_value_buf vb = {
2461 .vb_bh = blk_bh,
2462 .vb_access = ocfs2_journal_access_xb,
2463 };
2464 struct ocfs2_rm_xattr_bucket_para args = {
2465 .ref_ci = ref_ci,
2466 .ref_root_bh = ref_root_bh,
2467 };
2468
2469 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2470 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2471 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
2472 ret = ocfs2_remove_value_outside(inode, &vb, header,
2473 ref_ci, ref_root_bh);
2474 } else
2475 ret = ocfs2_iterate_xattr_index_block(inode,
2476 blk_bh,
2477 ocfs2_rm_xattr_cluster,
2478 &args);
2479
2480 return ret;
2481 }
2482
2483 static int ocfs2_xattr_free_block(struct inode *inode,
2484 u64 block,
2485 struct ocfs2_caching_info *ref_ci,
2486 struct buffer_head *ref_root_bh)
2487 {
2488 struct inode *xb_alloc_inode;
2489 struct buffer_head *xb_alloc_bh = NULL;
2490 struct buffer_head *blk_bh = NULL;
2491 struct ocfs2_xattr_block *xb;
2492 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2493 handle_t *handle;
2494 int ret = 0;
2495 u64 blk, bg_blkno;
2496 u16 bit;
2497
2498 ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
2499 if (ret < 0) {
2500 mlog_errno(ret);
2501 goto out;
2502 }
2503
2504 ret = ocfs2_xattr_block_remove(inode, blk_bh, ref_ci, ref_root_bh);
2505 if (ret < 0) {
2506 mlog_errno(ret);
2507 goto out;
2508 }
2509
2510 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2511 blk = le64_to_cpu(xb->xb_blkno);
2512 bit = le16_to_cpu(xb->xb_suballoc_bit);
2513 if (xb->xb_suballoc_loc)
2514 bg_blkno = le64_to_cpu(xb->xb_suballoc_loc);
2515 else
2516 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2517
2518 xb_alloc_inode = ocfs2_get_system_file_inode(osb,
2519 EXTENT_ALLOC_SYSTEM_INODE,
2520 le16_to_cpu(xb->xb_suballoc_slot));
2521 if (!xb_alloc_inode) {
2522 ret = -ENOMEM;
2523 mlog_errno(ret);
2524 goto out;
2525 }
2526 inode_lock(xb_alloc_inode);
2527
2528 ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
2529 if (ret < 0) {
2530 mlog_errno(ret);
2531 goto out_mutex;
2532 }
2533
2534 handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
2535 if (IS_ERR(handle)) {
2536 ret = PTR_ERR(handle);
2537 mlog_errno(ret);
2538 goto out_unlock;
2539 }
2540
2541 ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
2542 bit, bg_blkno, 1);
2543 if (ret < 0)
2544 mlog_errno(ret);
2545
2546 ocfs2_commit_trans(osb, handle);
2547 out_unlock:
2548 ocfs2_inode_unlock(xb_alloc_inode, 1);
2549 brelse(xb_alloc_bh);
2550 out_mutex:
2551 inode_unlock(xb_alloc_inode);
2552 iput(xb_alloc_inode);
2553 out:
2554 brelse(blk_bh);
2555 return ret;
2556 }
2557
2558 /*
2559 * ocfs2_xattr_remove()
2560 *
2561 * Free extended attribute resources associated with this inode.
2562 */
2563 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
2564 {
2565 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2566 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2567 struct ocfs2_refcount_tree *ref_tree = NULL;
2568 struct buffer_head *ref_root_bh = NULL;
2569 struct ocfs2_caching_info *ref_ci = NULL;
2570 handle_t *handle;
2571 int ret;
2572
2573 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2574 return 0;
2575
2576 if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
2577 return 0;
2578
2579 if (ocfs2_is_refcount_inode(inode)) {
2580 ret = ocfs2_lock_refcount_tree(OCFS2_SB(inode->i_sb),
2581 le64_to_cpu(di->i_refcount_loc),
2582 1, &ref_tree, &ref_root_bh);
2583 if (ret) {
2584 mlog_errno(ret);
2585 goto out;
2586 }
2587 ref_ci = &ref_tree->rf_ci;
2588
2589 }
2590
2591 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2592 ret = ocfs2_xattr_ibody_remove(inode, di_bh,
2593 ref_ci, ref_root_bh);
2594 if (ret < 0) {
2595 mlog_errno(ret);
2596 goto out;
2597 }
2598 }
2599
2600 if (di->i_xattr_loc) {
2601 ret = ocfs2_xattr_free_block(inode,
2602 le64_to_cpu(di->i_xattr_loc),
2603 ref_ci, ref_root_bh);
2604 if (ret < 0) {
2605 mlog_errno(ret);
2606 goto out;
2607 }
2608 }
2609
2610 handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
2611 OCFS2_INODE_UPDATE_CREDITS);
2612 if (IS_ERR(handle)) {
2613 ret = PTR_ERR(handle);
2614 mlog_errno(ret);
2615 goto out;
2616 }
2617 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2618 OCFS2_JOURNAL_ACCESS_WRITE);
2619 if (ret) {
2620 mlog_errno(ret);
2621 goto out_commit;
2622 }
2623
2624 di->i_xattr_loc = 0;
2625
2626 spin_lock(&oi->ip_lock);
2627 oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
2628 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2629 spin_unlock(&oi->ip_lock);
2630 ocfs2_update_inode_fsync_trans(handle, inode, 0);
2631
2632 ocfs2_journal_dirty(handle, di_bh);
2633 out_commit:
2634 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
2635 out:
2636 if (ref_tree)
2637 ocfs2_unlock_refcount_tree(OCFS2_SB(inode->i_sb), ref_tree, 1);
2638 brelse(ref_root_bh);
2639 return ret;
2640 }
2641
2642 static int ocfs2_xattr_has_space_inline(struct inode *inode,
2643 struct ocfs2_dinode *di)
2644 {
2645 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2646 unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
2647 int free;
2648
2649 if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
2650 return 0;
2651
2652 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2653 struct ocfs2_inline_data *idata = &di->id2.i_data;
2654 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
2655 } else if (ocfs2_inode_is_fast_symlink(inode)) {
2656 free = ocfs2_fast_symlink_chars(inode->i_sb) -
2657 le64_to_cpu(di->i_size);
2658 } else {
2659 struct ocfs2_extent_list *el = &di->id2.i_list;
2660 free = (le16_to_cpu(el->l_count) -
2661 le16_to_cpu(el->l_next_free_rec)) *
2662 sizeof(struct ocfs2_extent_rec);
2663 }
2664 if (free >= xattrsize)
2665 return 1;
2666
2667 return 0;
2668 }
2669
2670 /*
2671 * ocfs2_xattr_ibody_find()
2672 *
2673 * Find extended attribute in inode block and
2674 * fill search info into struct ocfs2_xattr_search.
2675 */
2676 static int ocfs2_xattr_ibody_find(struct inode *inode,
2677 int name_index,
2678 const char *name,
2679 struct ocfs2_xattr_search *xs)
2680 {
2681 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2682 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2683 int ret;
2684 int has_space = 0;
2685
2686 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2687 return 0;
2688
2689 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2690 down_read(&oi->ip_alloc_sem);
2691 has_space = ocfs2_xattr_has_space_inline(inode, di);
2692 up_read(&oi->ip_alloc_sem);
2693 if (!has_space)
2694 return 0;
2695 }
2696
2697 xs->xattr_bh = xs->inode_bh;
2698 xs->end = (void *)di + inode->i_sb->s_blocksize;
2699 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
2700 xs->header = (struct ocfs2_xattr_header *)
2701 (xs->end - le16_to_cpu(di->i_xattr_inline_size));
2702 else
2703 xs->header = (struct ocfs2_xattr_header *)
2704 (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
2705 xs->base = (void *)xs->header;
2706 xs->here = xs->header->xh_entries;
2707
2708 /* Find the named attribute. */
2709 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2710 ret = ocfs2_xattr_find_entry(inode, name_index, name, xs);
2711 if (ret && ret != -ENODATA)
2712 return ret;
2713 xs->not_found = ret;
2714 }
2715
2716 return 0;
2717 }
2718
2719 static int ocfs2_xattr_ibody_init(struct inode *inode,
2720 struct buffer_head *di_bh,
2721 struct ocfs2_xattr_set_ctxt *ctxt)
2722 {
2723 int ret;
2724 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2725 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2726 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2727 unsigned int xattrsize = osb->s_xattr_inline_size;
2728
2729 if (!ocfs2_xattr_has_space_inline(inode, di)) {
2730 ret = -ENOSPC;
2731 goto out;
2732 }
2733
2734 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode), di_bh,
2735 OCFS2_JOURNAL_ACCESS_WRITE);
2736 if (ret) {
2737 mlog_errno(ret);
2738 goto out;
2739 }
2740
2741 /*
2742 * Adjust extent record count or inline data size
2743 * to reserve space for extended attribute.
2744 */
2745 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2746 struct ocfs2_inline_data *idata = &di->id2.i_data;
2747 le16_add_cpu(&idata->id_count, -xattrsize);
2748 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
2749 struct ocfs2_extent_list *el = &di->id2.i_list;
2750 le16_add_cpu(&el->l_count, -(xattrsize /
2751 sizeof(struct ocfs2_extent_rec)));
2752 }
2753 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
2754
2755 spin_lock(&oi->ip_lock);
2756 oi->ip_dyn_features |= OCFS2_INLINE_XATTR_FL|OCFS2_HAS_XATTR_FL;
2757 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2758 spin_unlock(&oi->ip_lock);
2759
2760 ocfs2_journal_dirty(ctxt->handle, di_bh);
2761
2762 out:
2763 return ret;
2764 }
2765
2766 /*
2767 * ocfs2_xattr_ibody_set()
2768 *
2769 * Set, replace or remove an extended attribute into inode block.
2770 *
2771 */
2772 static int ocfs2_xattr_ibody_set(struct inode *inode,
2773 struct ocfs2_xattr_info *xi,
2774 struct ocfs2_xattr_search *xs,
2775 struct ocfs2_xattr_set_ctxt *ctxt)
2776 {
2777 int ret;
2778 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2779 struct ocfs2_xa_loc loc;
2780
2781 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2782 return -ENOSPC;
2783
2784 down_write(&oi->ip_alloc_sem);
2785 if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2786 ret = ocfs2_xattr_ibody_init(inode, xs->inode_bh, ctxt);
2787 if (ret) {
2788 if (ret != -ENOSPC)
2789 mlog_errno(ret);
2790 goto out;
2791 }
2792 }
2793
2794 ocfs2_init_dinode_xa_loc(&loc, inode, xs->inode_bh,
2795 xs->not_found ? NULL : xs->here);
2796 ret = ocfs2_xa_set(&loc, xi, ctxt);
2797 if (ret) {
2798 if (ret != -ENOSPC)
2799 mlog_errno(ret);
2800 goto out;
2801 }
2802 xs->here = loc.xl_entry;
2803
2804 out:
2805 up_write(&oi->ip_alloc_sem);
2806
2807 return ret;
2808 }
2809
2810 /*
2811 * ocfs2_xattr_block_find()
2812 *
2813 * Find extended attribute in external block and
2814 * fill search info into struct ocfs2_xattr_search.
2815 */
2816 static int ocfs2_xattr_block_find(struct inode *inode,
2817 int name_index,
2818 const char *name,
2819 struct ocfs2_xattr_search *xs)
2820 {
2821 struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2822 struct buffer_head *blk_bh = NULL;
2823 struct ocfs2_xattr_block *xb;
2824 int ret = 0;
2825
2826 if (!di->i_xattr_loc)
2827 return ret;
2828
2829 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2830 &blk_bh);
2831 if (ret < 0) {
2832 mlog_errno(ret);
2833 return ret;
2834 }
2835
2836 xs->xattr_bh = blk_bh;
2837 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2838
2839 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2840 xs->header = &xb->xb_attrs.xb_header;
2841 xs->base = (void *)xs->header;
2842 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2843 xs->here = xs->header->xh_entries;
2844
2845 ret = ocfs2_xattr_find_entry(inode, name_index, name, xs);
2846 } else
2847 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2848 name_index,
2849 name, xs);
2850
2851 if (ret && ret != -ENODATA) {
2852 xs->xattr_bh = NULL;
2853 goto cleanup;
2854 }
2855 xs->not_found = ret;
2856 return 0;
2857 cleanup:
2858 brelse(blk_bh);
2859
2860 return ret;
2861 }
2862
2863 static int ocfs2_create_xattr_block(struct inode *inode,
2864 struct buffer_head *inode_bh,
2865 struct ocfs2_xattr_set_ctxt *ctxt,
2866 int indexed,
2867 struct buffer_head **ret_bh)
2868 {
2869 int ret;
2870 u16 suballoc_bit_start;
2871 u32 num_got;
2872 u64 suballoc_loc, first_blkno;
2873 struct ocfs2_dinode *di = (struct ocfs2_dinode *)inode_bh->b_data;
2874 struct buffer_head *new_bh = NULL;
2875 struct ocfs2_xattr_block *xblk;
2876
2877 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
2878 inode_bh, OCFS2_JOURNAL_ACCESS_CREATE);
2879 if (ret < 0) {
2880 mlog_errno(ret);
2881 goto end;
2882 }
2883
2884 ret = ocfs2_claim_metadata(ctxt->handle, ctxt->meta_ac, 1,
2885 &suballoc_loc, &suballoc_bit_start,
2886 &num_got, &first_blkno);
2887 if (ret < 0) {
2888 mlog_errno(ret);
2889 goto end;
2890 }
2891
2892 new_bh = sb_getblk(inode->i_sb, first_blkno);
2893 if (!new_bh) {
2894 ret = -ENOMEM;
2895 mlog_errno(ret);
2896 goto end;
2897 }
2898
2899 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2900
2901 ret = ocfs2_journal_access_xb(ctxt->handle, INODE_CACHE(inode),
2902 new_bh,
2903 OCFS2_JOURNAL_ACCESS_CREATE);
2904 if (ret < 0) {
2905 mlog_errno(ret);
2906 goto end;
2907 }
2908
2909 /* Initialize ocfs2_xattr_block */
2910 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2911 memset(xblk, 0, inode->i_sb->s_blocksize);
2912 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2913 xblk->xb_suballoc_slot = cpu_to_le16(ctxt->meta_ac->ac_alloc_slot);
2914 xblk->xb_suballoc_loc = cpu_to_le64(suballoc_loc);
2915 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2916 xblk->xb_fs_generation =
2917 cpu_to_le32(OCFS2_SB(inode->i_sb)->fs_generation);
2918 xblk->xb_blkno = cpu_to_le64(first_blkno);
2919 if (indexed) {
2920 struct ocfs2_xattr_tree_root *xr = &xblk->xb_attrs.xb_root;
2921 xr->xt_clusters = cpu_to_le32(1);
2922 xr->xt_last_eb_blk = 0;
2923 xr->xt_list.l_tree_depth = 0;
2924 xr->xt_list.l_count = cpu_to_le16(
2925 ocfs2_xattr_recs_per_xb(inode->i_sb));
2926 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2927 xblk->xb_flags = cpu_to_le16(OCFS2_XATTR_INDEXED);
2928 }
2929 ocfs2_journal_dirty(ctxt->handle, new_bh);
2930
2931 /* Add it to the inode */
2932 di->i_xattr_loc = cpu_to_le64(first_blkno);
2933
2934 spin_lock(&OCFS2_I(inode)->ip_lock);
2935 OCFS2_I(inode)->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
2936 di->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
2937 spin_unlock(&OCFS2_I(inode)->ip_lock);
2938
2939 ocfs2_journal_dirty(ctxt->handle, inode_bh);
2940
2941 *ret_bh = new_bh;
2942 new_bh = NULL;
2943
2944 end:
2945 brelse(new_bh);
2946 return ret;
2947 }
2948
2949 /*
2950 * ocfs2_xattr_block_set()
2951 *
2952 * Set, replace or remove an extended attribute into external block.
2953 *
2954 */
2955 static int ocfs2_xattr_block_set(struct inode *inode,
2956 struct ocfs2_xattr_info *xi,
2957 struct ocfs2_xattr_search *xs,
2958 struct ocfs2_xattr_set_ctxt *ctxt)
2959 {
2960 struct buffer_head *new_bh = NULL;
2961 struct ocfs2_xattr_block *xblk = NULL;
2962 int ret;
2963 struct ocfs2_xa_loc loc;
2964
2965 if (!xs->xattr_bh) {
2966 ret = ocfs2_create_xattr_block(inode, xs->inode_bh, ctxt,
2967 0, &new_bh);
2968 if (ret) {
2969 mlog_errno(ret);
2970 goto end;
2971 }
2972
2973 xs->xattr_bh = new_bh;
2974 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2975 xs->header = &xblk->xb_attrs.xb_header;
2976 xs->base = (void *)xs->header;
2977 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2978 xs->here = xs->header->xh_entries;
2979 } else
2980 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2981
2982 if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2983 ocfs2_init_xattr_block_xa_loc(&loc, inode, xs->xattr_bh,
2984 xs->not_found ? NULL : xs->here);
2985
2986 ret = ocfs2_xa_set(&loc, xi, ctxt);
2987 if (!ret)
2988 xs->here = loc.xl_entry;
2989 else if ((ret != -ENOSPC) || ctxt->set_abort)
2990 goto end;
2991 else {
2992 ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2993 if (ret)
2994 goto end;
2995 }
2996 }
2997
2998 if (le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)
2999 ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
3000
3001 end:
3002 return ret;
3003 }
3004
3005 /* Check whether the new xattr can be inserted into the inode. */
3006 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
3007 struct ocfs2_xattr_info *xi,
3008 struct ocfs2_xattr_search *xs)
3009 {
3010 struct ocfs2_xattr_entry *last;
3011 int free, i;
3012 size_t min_offs = xs->end - xs->base;
3013
3014 if (!xs->header)
3015 return 0;
3016
3017 last = xs->header->xh_entries;
3018
3019 for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
3020 size_t offs = le16_to_cpu(last->xe_name_offset);
3021 if (offs < min_offs)
3022 min_offs = offs;
3023 last += 1;
3024 }
3025
3026 free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
3027 if (free < 0)
3028 return 0;
3029
3030 BUG_ON(!xs->not_found);
3031
3032 if (free >= (sizeof(struct ocfs2_xattr_entry) + namevalue_size_xi(xi)))
3033 return 1;
3034
3035 return 0;
3036 }
3037
3038 static int ocfs2_calc_xattr_set_need(struct inode *inode,
3039 struct ocfs2_dinode *di,
3040 struct ocfs2_xattr_info *xi,
3041 struct ocfs2_xattr_search *xis,
3042 struct ocfs2_xattr_search *xbs,
3043 int *clusters_need,
3044 int *meta_need,
3045 int *credits_need)
3046 {
3047 int ret = 0, old_in_xb = 0;
3048 int clusters_add = 0, meta_add = 0, credits = 0;
3049 struct buffer_head *bh = NULL;
3050 struct ocfs2_xattr_block *xb = NULL;
3051 struct ocfs2_xattr_entry *xe = NULL;
3052 struct ocfs2_xattr_value_root *xv = NULL;
3053 char *base = NULL;
3054 int name_offset, name_len = 0;
3055 u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
3056 xi->xi_value_len);
3057 u64 value_size;
3058
3059 /*
3060 * Calculate the clusters we need to write.
3061 * No matter whether we replace an old one or add a new one,
3062 * we need this for writing.
3063 */
3064 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
3065 credits += new_clusters *
3066 ocfs2_clusters_to_blocks(inode->i_sb, 1);
3067
3068 if (xis->not_found && xbs->not_found) {
3069 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3070
3071 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3072 clusters_add += new_clusters;
3073 credits += ocfs2_calc_extend_credits(inode->i_sb,
3074 &def_xv.xv.xr_list);
3075 }
3076
3077 goto meta_guess;
3078 }
3079
3080 if (!xis->not_found) {
3081 xe = xis->here;
3082 name_offset = le16_to_cpu(xe->xe_name_offset);
3083 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3084 base = xis->base;
3085 credits += OCFS2_INODE_UPDATE_CREDITS;
3086 } else {
3087 int i, block_off = 0;
3088 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3089 xe = xbs->here;
3090 name_offset = le16_to_cpu(xe->xe_name_offset);
3091 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3092 i = xbs->here - xbs->header->xh_entries;
3093 old_in_xb = 1;
3094
3095 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3096 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3097 bucket_xh(xbs->bucket),
3098 i, &block_off,
3099 &name_offset);
3100 base = bucket_block(xbs->bucket, block_off);
3101 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3102 } else {
3103 base = xbs->base;
3104 credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
3105 }
3106 }
3107
3108 /*
3109 * delete a xattr doesn't need metadata and cluster allocation.
3110 * so just calculate the credits and return.
3111 *
3112 * The credits for removing the value tree will be extended
3113 * by ocfs2_remove_extent itself.
3114 */
3115 if (!xi->xi_value) {
3116 if (!ocfs2_xattr_is_local(xe))
3117 credits += ocfs2_remove_extent_credits(inode->i_sb);
3118
3119 goto out;
3120 }
3121
3122 /* do cluster allocation guess first. */
3123 value_size = le64_to_cpu(xe->xe_value_size);
3124
3125 if (old_in_xb) {
3126 /*
3127 * In xattr set, we always try to set the xe in inode first,
3128 * so if it can be inserted into inode successfully, the old
3129 * one will be removed from the xattr block, and this xattr
3130 * will be inserted into inode as a new xattr in inode.
3131 */
3132 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
3133 clusters_add += new_clusters;
3134 credits += ocfs2_remove_extent_credits(inode->i_sb) +
3135 OCFS2_INODE_UPDATE_CREDITS;
3136 if (!ocfs2_xattr_is_local(xe))
3137 credits += ocfs2_calc_extend_credits(
3138 inode->i_sb,
3139 &def_xv.xv.xr_list);
3140 goto out;
3141 }
3142 }
3143
3144 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3145 /* the new values will be stored outside. */
3146 u32 old_clusters = 0;
3147
3148 if (!ocfs2_xattr_is_local(xe)) {
3149 old_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
3150 value_size);
3151 xv = (struct ocfs2_xattr_value_root *)
3152 (base + name_offset + name_len);
3153 value_size = OCFS2_XATTR_ROOT_SIZE;
3154 } else
3155 xv = &def_xv.xv;
3156
3157 if (old_clusters >= new_clusters) {
3158 credits += ocfs2_remove_extent_credits(inode->i_sb);
3159 goto out;
3160 } else {
3161 meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
3162 clusters_add += new_clusters - old_clusters;
3163 credits += ocfs2_calc_extend_credits(inode->i_sb,
3164 &xv->xr_list);
3165 if (value_size >= OCFS2_XATTR_ROOT_SIZE)
3166 goto out;
3167 }
3168 } else {
3169 /*
3170 * Now the new value will be stored inside. So if the new
3171 * value is smaller than the size of value root or the old
3172 * value, we don't need any allocation, otherwise we have
3173 * to guess metadata allocation.
3174 */
3175 if ((ocfs2_xattr_is_local(xe) &&
3176 (value_size >= xi->xi_value_len)) ||
3177 (!ocfs2_xattr_is_local(xe) &&
3178 OCFS2_XATTR_ROOT_SIZE >= xi->xi_value_len))
3179 goto out;
3180 }
3181
3182 meta_guess:
3183 /* calculate metadata allocation. */
3184 if (di->i_xattr_loc) {
3185 if (!xbs->xattr_bh) {
3186 ret = ocfs2_read_xattr_block(inode,
3187 le64_to_cpu(di->i_xattr_loc),
3188 &bh);
3189 if (ret) {
3190 mlog_errno(ret);
3191 goto out;
3192 }
3193
3194 xb = (struct ocfs2_xattr_block *)bh->b_data;
3195 } else
3196 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3197
3198 /*
3199 * If there is already an xattr tree, good, we can calculate
3200 * like other b-trees. Otherwise we may have the chance of
3201 * create a tree, the credit calculation is borrowed from
3202 * ocfs2_calc_extend_credits with root_el = NULL. And the
3203 * new tree will be cluster based, so no meta is needed.
3204 */
3205 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3206 struct ocfs2_extent_list *el =
3207 &xb->xb_attrs.xb_root.xt_list;
3208 meta_add += ocfs2_extend_meta_needed(el);
3209 credits += ocfs2_calc_extend_credits(inode->i_sb,
3210 el);
3211 } else
3212 credits += OCFS2_SUBALLOC_ALLOC + 1;
3213
3214 /*
3215 * This cluster will be used either for new bucket or for
3216 * new xattr block.
3217 * If the cluster size is the same as the bucket size, one
3218 * more is needed since we may need to extend the bucket
3219 * also.
3220 */
3221 clusters_add += 1;
3222 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3223 if (OCFS2_XATTR_BUCKET_SIZE ==
3224 OCFS2_SB(inode->i_sb)->s_clustersize) {
3225 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3226 clusters_add += 1;
3227 }
3228 } else {
3229 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
3230 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3231 struct ocfs2_extent_list *el = &def_xv.xv.xr_list;
3232 meta_add += ocfs2_extend_meta_needed(el);
3233 credits += ocfs2_calc_extend_credits(inode->i_sb,
3234 el);
3235 } else {
3236 meta_add += 1;
3237 }
3238 }
3239 out:
3240 if (clusters_need)
3241 *clusters_need = clusters_add;
3242 if (meta_need)
3243 *meta_need = meta_add;
3244 if (credits_need)
3245 *credits_need = credits;
3246 brelse(bh);
3247 return ret;
3248 }
3249
3250 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
3251 struct ocfs2_dinode *di,
3252 struct ocfs2_xattr_info *xi,
3253 struct ocfs2_xattr_search *xis,
3254 struct ocfs2_xattr_search *xbs,
3255 struct ocfs2_xattr_set_ctxt *ctxt,
3256 int extra_meta,
3257 int *credits)
3258 {
3259 int clusters_add, meta_add, ret;
3260 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3261
3262 memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
3263
3264 ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
3265
3266 ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
3267 &clusters_add, &meta_add, credits);
3268 if (ret) {
3269 mlog_errno(ret);
3270 return ret;
3271 }
3272
3273 meta_add += extra_meta;
3274 trace_ocfs2_init_xattr_set_ctxt(xi->xi_name, meta_add,
3275 clusters_add, *credits);
3276
3277 if (meta_add) {
3278 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
3279 &ctxt->meta_ac);
3280 if (ret) {
3281 mlog_errno(ret);
3282 goto out;
3283 }
3284 }
3285
3286 if (clusters_add) {
3287 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
3288 if (ret)
3289 mlog_errno(ret);
3290 }
3291 out:
3292 if (ret) {
3293 if (ctxt->meta_ac) {
3294 ocfs2_free_alloc_context(ctxt->meta_ac);
3295 ctxt->meta_ac = NULL;
3296 }
3297
3298 /*
3299 * We cannot have an error and a non null ctxt->data_ac.
3300 */
3301 }
3302
3303 return ret;
3304 }
3305
3306 static int __ocfs2_xattr_set_handle(struct inode *inode,
3307 struct ocfs2_dinode *di,
3308 struct ocfs2_xattr_info *xi,
3309 struct ocfs2_xattr_search *xis,
3310 struct ocfs2_xattr_search *xbs,
3311 struct ocfs2_xattr_set_ctxt *ctxt)
3312 {
3313 int ret = 0, credits, old_found;
3314
3315 if (!xi->xi_value) {
3316 /* Remove existing extended attribute */
3317 if (!xis->not_found)
3318 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3319 else if (!xbs->not_found)
3320 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3321 } else {
3322 /* We always try to set extended attribute into inode first*/
3323 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3324 if (!ret && !xbs->not_found) {
3325 /*
3326 * If succeed and that extended attribute existing in
3327 * external block, then we will remove it.
3328 */
3329 xi->xi_value = NULL;
3330 xi->xi_value_len = 0;
3331
3332 old_found = xis->not_found;
3333 xis->not_found = -ENODATA;
3334 ret = ocfs2_calc_xattr_set_need(inode,
3335 di,
3336 xi,
3337 xis,
3338 xbs,
3339 NULL,
3340 NULL,
3341 &credits);
3342 xis->not_found = old_found;
3343 if (ret) {
3344 mlog_errno(ret);
3345 goto out;
3346 }
3347
3348 ret = ocfs2_extend_trans(ctxt->handle, credits);
3349 if (ret) {
3350 mlog_errno(ret);
3351 goto out;
3352 }
3353 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3354 } else if ((ret == -ENOSPC) && !ctxt->set_abort) {
3355 if (di->i_xattr_loc && !xbs->xattr_bh) {
3356 ret = ocfs2_xattr_block_find(inode,
3357 xi->xi_name_index,
3358 xi->xi_name, xbs);
3359 if (ret)
3360 goto out;
3361
3362 old_found = xis->not_found;
3363 xis->not_found = -ENODATA;
3364 ret = ocfs2_calc_xattr_set_need(inode,
3365 di,
3366 xi,
3367 xis,
3368 xbs,
3369 NULL,
3370 NULL,
3371 &credits);
3372 xis->not_found = old_found;
3373 if (ret) {
3374 mlog_errno(ret);
3375 goto out;
3376 }
3377
3378 ret = ocfs2_extend_trans(ctxt->handle, credits);
3379 if (ret) {
3380 mlog_errno(ret);
3381 goto out;
3382 }
3383 }
3384 /*
3385 * If no space in inode, we will set extended attribute
3386 * into external block.
3387 */
3388 ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3389 if (ret)
3390 goto out;
3391 if (!xis->not_found) {
3392 /*
3393 * If succeed and that extended attribute
3394 * existing in inode, we will remove it.
3395 */
3396 xi->xi_value = NULL;
3397 xi->xi_value_len = 0;
3398 xbs->not_found = -ENODATA;
3399 ret = ocfs2_calc_xattr_set_need(inode,
3400 di,
3401 xi,
3402 xis,
3403 xbs,
3404 NULL,
3405 NULL,
3406 &credits);
3407 if (ret) {
3408 mlog_errno(ret);
3409 goto out;
3410 }
3411
3412 ret = ocfs2_extend_trans(ctxt->handle, credits);
3413 if (ret) {
3414 mlog_errno(ret);
3415 goto out;
3416 }
3417 ret = ocfs2_xattr_ibody_set(inode, xi,
3418 xis, ctxt);
3419 }
3420 }
3421 }
3422
3423 if (!ret) {
3424 /* Update inode ctime. */
3425 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
3426 xis->inode_bh,
3427 OCFS2_JOURNAL_ACCESS_WRITE);
3428 if (ret) {
3429 mlog_errno(ret);
3430 goto out;
3431 }
3432
3433 inode_set_ctime_current(inode);
3434 di->i_ctime = cpu_to_le64(inode_get_ctime_sec(inode));
3435 di->i_ctime_nsec = cpu_to_le32(inode_get_ctime_nsec(inode));
3436 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
3437 }
3438 out:
3439 return ret;
3440 }
3441
3442 /*
3443 * This function only called duing creating inode
3444 * for init security/acl xattrs of the new inode.
3445 * All transanction credits have been reserved in mknod.
3446 */
3447 int ocfs2_xattr_set_handle(handle_t *handle,
3448 struct inode *inode,
3449 struct buffer_head *di_bh,
3450 int name_index,
3451 const char *name,
3452 const void *value,
3453 size_t value_len,
3454 int flags,
3455 struct ocfs2_alloc_context *meta_ac,
3456 struct ocfs2_alloc_context *data_ac)
3457 {
3458 struct ocfs2_dinode *di;
3459 int ret;
3460
3461 struct ocfs2_xattr_info xi = {
3462 .xi_name_index = name_index,
3463 .xi_name = name,
3464 .xi_name_len = strlen(name),
3465 .xi_value = value,
3466 .xi_value_len = value_len,
3467 };
3468
3469 struct ocfs2_xattr_search xis = {
3470 .not_found = -ENODATA,
3471 };
3472
3473 struct ocfs2_xattr_search xbs = {
3474 .not_found = -ENODATA,
3475 };
3476
3477 struct ocfs2_xattr_set_ctxt ctxt = {
3478 .handle = handle,
3479 .meta_ac = meta_ac,
3480 .data_ac = data_ac,
3481 };
3482
3483 if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3484 return -EOPNOTSUPP;
3485
3486 /*
3487 * In extreme situation, may need xattr bucket when
3488 * block size is too small. And we have already reserved
3489 * the credits for bucket in mknod.
3490 */
3491 if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
3492 xbs.bucket = ocfs2_xattr_bucket_new(inode);
3493 if (!xbs.bucket) {
3494 mlog_errno(-ENOMEM);
3495 return -ENOMEM;
3496 }
3497 }
3498
3499 xis.inode_bh = xbs.inode_bh = di_bh;
3500 di = (struct ocfs2_dinode *)di_bh->b_data;
3501
3502 down_write(&OCFS2_I(inode)->ip_xattr_sem);
3503
3504 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3505 if (ret)
3506 goto cleanup;
3507 if (xis.not_found) {
3508 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3509 if (ret)
3510 goto cleanup;
3511 }
3512
3513 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3514
3515 cleanup:
3516 up_write(&OCFS2_I(inode)->ip_xattr_sem);
3517 brelse(xbs.xattr_bh);
3518 ocfs2_xattr_bucket_free(xbs.bucket);
3519
3520 return ret;
3521 }
3522
3523 /*
3524 * ocfs2_xattr_set()
3525 *
3526 * Set, replace or remove an extended attribute for this inode.
3527 * value is NULL to remove an existing extended attribute, else either
3528 * create or replace an extended attribute.
3529 */
3530 int ocfs2_xattr_set(struct inode *inode,
3531 int name_index,
3532 const char *name,
3533 const void *value,
3534 size_t value_len,
3535 int flags)
3536 {
3537 struct buffer_head *di_bh = NULL;
3538 struct ocfs2_dinode *di;
3539 int ret, credits, had_lock, ref_meta = 0, ref_credits = 0;
3540 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3541 struct inode *tl_inode = osb->osb_tl_inode;
3542 struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, NULL, };
3543 struct ocfs2_refcount_tree *ref_tree = NULL;
3544 struct ocfs2_lock_holder oh;
3545
3546 struct ocfs2_xattr_info xi = {
3547 .xi_name_index = name_index,
3548 .xi_name = name,
3549 .xi_name_len = strlen(name),
3550 .xi_value = value,
3551 .xi_value_len = value_len,
3552 };
3553
3554 struct ocfs2_xattr_search xis = {
3555 .not_found = -ENODATA,
3556 };
3557
3558 struct ocfs2_xattr_search xbs = {
3559 .not_found = -ENODATA,
3560 };
3561
3562 if (!ocfs2_supports_xattr(osb))
3563 return -EOPNOTSUPP;
3564
3565 /*
3566 * Only xbs will be used on indexed trees. xis doesn't need a
3567 * bucket.
3568 */
3569 xbs.bucket = ocfs2_xattr_bucket_new(inode);
3570 if (!xbs.bucket) {
3571 mlog_errno(-ENOMEM);
3572 return -ENOMEM;
3573 }
3574
3575 had_lock = ocfs2_inode_lock_tracker(inode, &di_bh, 1, &oh);
3576 if (had_lock < 0) {
3577 ret = had_lock;
3578 mlog_errno(ret);
3579 goto cleanup_nolock;
3580 }
3581 xis.inode_bh = xbs.inode_bh = di_bh;
3582 di = (struct ocfs2_dinode *)di_bh->b_data;
3583
3584 down_write(&OCFS2_I(inode)->ip_xattr_sem);
3585 /*
3586 * Scan inode and external block to find the same name
3587 * extended attribute and collect search information.
3588 */
3589 ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3590 if (ret)
3591 goto cleanup;
3592 if (xis.not_found) {
3593 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3594 if (ret)
3595 goto cleanup;
3596 }
3597
3598 if (xis.not_found && xbs.not_found) {
3599 ret = -ENODATA;
3600 if (flags & XATTR_REPLACE)
3601 goto cleanup;
3602 ret = 0;
3603 if (!value)
3604 goto cleanup;
3605 } else {
3606 ret = -EEXIST;
3607 if (flags & XATTR_CREATE)
3608 goto cleanup;
3609 }
3610
3611 /* Check whether the value is refcounted and do some preparation. */
3612 if (ocfs2_is_refcount_inode(inode) &&
3613 (!xis.not_found || !xbs.not_found)) {
3614 ret = ocfs2_prepare_refcount_xattr(inode, di, &xi,
3615 &xis, &xbs, &ref_tree,
3616 &ref_meta, &ref_credits);
3617 if (ret) {
3618 mlog_errno(ret);
3619 goto cleanup;
3620 }
3621 }
3622
3623 inode_lock(tl_inode);
3624
3625 if (ocfs2_truncate_log_needs_flush(osb)) {
3626 ret = __ocfs2_flush_truncate_log(osb);
3627 if (ret < 0) {
3628 inode_unlock(tl_inode);
3629 mlog_errno(ret);
3630 goto cleanup;
3631 }
3632 }
3633 inode_unlock(tl_inode);
3634
3635 ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
3636 &xbs, &ctxt, ref_meta, &credits);
3637 if (ret) {
3638 mlog_errno(ret);
3639 goto cleanup;
3640 }
3641
3642 /* we need to update inode's ctime field, so add credit for it. */
3643 credits += OCFS2_INODE_UPDATE_CREDITS;
3644 ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
3645 if (IS_ERR(ctxt.handle)) {
3646 ret = PTR_ERR(ctxt.handle);
3647 mlog_errno(ret);
3648 goto out_free_ac;
3649 }
3650
3651 ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3652 ocfs2_update_inode_fsync_trans(ctxt.handle, inode, 0);
3653
3654 ocfs2_commit_trans(osb, ctxt.handle);
3655
3656 out_free_ac:
3657 if (ctxt.data_ac)
3658 ocfs2_free_alloc_context(ctxt.data_ac);
3659 if (ctxt.meta_ac)
3660 ocfs2_free_alloc_context(ctxt.meta_ac);
3661 if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
3662 ocfs2_schedule_truncate_log_flush(osb, 1);
3663 ocfs2_run_deallocs(osb, &ctxt.dealloc);
3664
3665 cleanup:
3666 if (ref_tree)
3667 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3668 up_write(&OCFS2_I(inode)->ip_xattr_sem);
3669 if (!value && !ret) {
3670 ret = ocfs2_try_remove_refcount_tree(inode, di_bh);
3671 if (ret)
3672 mlog_errno(ret);
3673 }
3674 ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
3675 cleanup_nolock:
3676 brelse(di_bh);
3677 brelse(xbs.xattr_bh);
3678 ocfs2_xattr_bucket_free(xbs.bucket);
3679
3680 return ret;
3681 }
3682
3683 /*
3684 * Find the xattr extent rec which may contains name_hash.
3685 * e_cpos will be the first name hash of the xattr rec.
3686 * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
3687 */
3688 static int ocfs2_xattr_get_rec(struct inode *inode,
3689 u32 name_hash,
3690 u64 *p_blkno,
3691 u32 *e_cpos,
3692 u32 *num_clusters,
3693 struct ocfs2_extent_list *el)
3694 {
3695 int ret = 0, i;
3696 struct buffer_head *eb_bh = NULL;
3697 struct ocfs2_extent_block *eb;
3698 struct ocfs2_extent_rec *rec = NULL;
3699 u64 e_blkno = 0;
3700
3701 if (el->l_tree_depth) {
3702 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, name_hash,
3703 &eb_bh);
3704 if (ret) {
3705 mlog_errno(ret);
3706 goto out;
3707 }
3708
3709 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
3710 el = &eb->h_list;
3711
3712 if (el->l_tree_depth) {
3713 ret = ocfs2_error(inode->i_sb,
3714 "Inode %lu has non zero tree depth in xattr tree block %llu\n",
3715 inode->i_ino,
3716 (unsigned long long)eb_bh->b_blocknr);
3717 goto out;
3718 }
3719 }
3720
3721 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
3722 rec = &el->l_recs[i];
3723
3724 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
3725 e_blkno = le64_to_cpu(rec->e_blkno);
3726 break;
3727 }
3728 }
3729
3730 if (!e_blkno) {
3731 ret = ocfs2_error(inode->i_sb, "Inode %lu has bad extent record (%u, %u, 0) in xattr\n",
3732 inode->i_ino,
3733 le32_to_cpu(rec->e_cpos),
3734 ocfs2_rec_clusters(el, rec));
3735 goto out;
3736 }
3737
3738 *p_blkno = le64_to_cpu(rec->e_blkno);
3739 *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
3740 if (e_cpos)
3741 *e_cpos = le32_to_cpu(rec->e_cpos);
3742 out:
3743 brelse(eb_bh);
3744 return ret;
3745 }
3746
3747 typedef int (xattr_bucket_func)(struct inode *inode,
3748 struct ocfs2_xattr_bucket *bucket,
3749 void *para);
3750
3751 static int ocfs2_find_xe_in_bucket(struct inode *inode,
3752 struct ocfs2_xattr_bucket *bucket,
3753 int name_index,
3754 const char *name,
3755 u32 name_hash,
3756 u16 *xe_index,
3757 int *found)
3758 {
3759 int i, ret = 0, cmp = 1, block_off, new_offset;
3760 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3761 size_t name_len = strlen(name);
3762 struct ocfs2_xattr_entry *xe = NULL;
3763 char *xe_name;
3764
3765 /*
3766 * We don't use binary search in the bucket because there
3767 * may be multiple entries with the same name hash.
3768 */
3769 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3770 xe = &xh->xh_entries[i];
3771
3772 if (name_hash > le32_to_cpu(xe->xe_name_hash))
3773 continue;
3774 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
3775 break;
3776
3777 cmp = name_index - ocfs2_xattr_get_type(xe);
3778 if (!cmp)
3779 cmp = name_len - xe->xe_name_len;
3780 if (cmp)
3781 continue;
3782
3783 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3784 xh,
3785 i,
3786 &block_off,
3787 &new_offset);
3788 if (ret) {
3789 mlog_errno(ret);
3790 break;
3791 }
3792
3793
3794 xe_name = bucket_block(bucket, block_off) + new_offset;
3795 if (!memcmp(name, xe_name, name_len)) {
3796 *xe_index = i;
3797 *found = 1;
3798 ret = 0;
3799 break;
3800 }
3801 }
3802
3803 return ret;
3804 }
3805
3806 /*
3807 * Find the specified xattr entry in a series of buckets.
3808 * This series start from p_blkno and last for num_clusters.
3809 * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
3810 * the num of the valid buckets.
3811 *
3812 * Return the buffer_head this xattr should reside in. And if the xattr's
3813 * hash is in the gap of 2 buckets, return the lower bucket.
3814 */
3815 static int ocfs2_xattr_bucket_find(struct inode *inode,
3816 int name_index,
3817 const char *name,
3818 u32 name_hash,
3819 u64 p_blkno,
3820 u32 first_hash,
3821 u32 num_clusters,
3822 struct ocfs2_xattr_search *xs)
3823 {
3824 int ret, found = 0;
3825 struct ocfs2_xattr_header *xh = NULL;
3826 struct ocfs2_xattr_entry *xe = NULL;
3827 u16 index = 0;
3828 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3829 int low_bucket = 0, bucket, high_bucket;
3830 struct ocfs2_xattr_bucket *search;
3831 u64 blkno, lower_blkno = 0;
3832
3833 search = ocfs2_xattr_bucket_new(inode);
3834 if (!search) {
3835 ret = -ENOMEM;
3836 mlog_errno(ret);
3837 goto out;
3838 }
3839
3840 ret = ocfs2_read_xattr_bucket(search, p_blkno);
3841 if (ret) {
3842 mlog_errno(ret);
3843 goto out;
3844 }
3845
3846 xh = bucket_xh(search);
3847 high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
3848 while (low_bucket <= high_bucket) {
3849 ocfs2_xattr_bucket_relse(search);
3850
3851 bucket = (low_bucket + high_bucket) / 2;
3852 blkno = p_blkno + bucket * blk_per_bucket;
3853 ret = ocfs2_read_xattr_bucket(search, blkno);
3854 if (ret) {
3855 mlog_errno(ret);
3856 goto out;
3857 }
3858
3859 xh = bucket_xh(search);
3860 xe = &xh->xh_entries[0];
3861 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3862 high_bucket = bucket - 1;
3863 continue;
3864 }
3865
3866 /*
3867 * Check whether the hash of the last entry in our
3868 * bucket is larger than the search one. for an empty
3869 * bucket, the last one is also the first one.
3870 */
3871 if (xh->xh_count)
3872 xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3873
3874 /* record lower_blkno which may be the insert place. */
3875 lower_blkno = blkno;
3876
3877 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3878 low_bucket = bucket + 1;
3879 continue;
3880 }
3881
3882 /* the searched xattr should reside in this bucket if exists. */
3883 ret = ocfs2_find_xe_in_bucket(inode, search,
3884 name_index, name, name_hash,
3885 &index, &found);
3886 if (ret) {
3887 mlog_errno(ret);
3888 goto out;
3889 }
3890 break;
3891 }
3892
3893 /*
3894 * Record the bucket we have found.
3895 * When the xattr's hash value is in the gap of 2 buckets, we will
3896 * always set it to the previous bucket.
3897 */
3898 if (!lower_blkno)
3899 lower_blkno = p_blkno;
3900
3901 /* This should be in cache - we just read it during the search */
3902 ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3903 if (ret) {
3904 mlog_errno(ret);
3905 goto out;
3906 }
3907
3908 xs->header = bucket_xh(xs->bucket);
3909 xs->base = bucket_block(xs->bucket, 0);
3910 xs->end = xs->base + inode->i_sb->s_blocksize;
3911
3912 if (found) {
3913 xs->here = &xs->header->xh_entries[index];
3914 trace_ocfs2_xattr_bucket_find(OCFS2_I(inode)->ip_blkno,
3915 name, name_index, name_hash,
3916 (unsigned long long)bucket_blkno(xs->bucket),
3917 index);
3918 } else
3919 ret = -ENODATA;
3920
3921 out:
3922 ocfs2_xattr_bucket_free(search);
3923 return ret;
3924 }
3925
3926 static int ocfs2_xattr_index_block_find(struct inode *inode,
3927 struct buffer_head *root_bh,
3928 int name_index,
3929 const char *name,
3930 struct ocfs2_xattr_search *xs)
3931 {
3932 int ret;
3933 struct ocfs2_xattr_block *xb =
3934 (struct ocfs2_xattr_block *)root_bh->b_data;
3935 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3936 struct ocfs2_extent_list *el = &xb_root->xt_list;
3937 u64 p_blkno = 0;
3938 u32 first_hash, num_clusters = 0;
3939 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3940
3941 if (le16_to_cpu(el->l_next_free_rec) == 0)
3942 return -ENODATA;
3943
3944 trace_ocfs2_xattr_index_block_find(OCFS2_I(inode)->ip_blkno,
3945 name, name_index, name_hash,
3946 (unsigned long long)root_bh->b_blocknr,
3947 -1);
3948
3949 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3950 &num_clusters, el);
3951 if (ret) {
3952 mlog_errno(ret);
3953 goto out;
3954 }
3955
3956 BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3957
3958 trace_ocfs2_xattr_index_block_find_rec(OCFS2_I(inode)->ip_blkno,
3959 name, name_index, first_hash,
3960 (unsigned long long)p_blkno,
3961 num_clusters);
3962
3963 ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3964 p_blkno, first_hash, num_clusters, xs);
3965
3966 out:
3967 return ret;
3968 }
3969
3970 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3971 u64 blkno,
3972 u32 clusters,
3973 xattr_bucket_func *func,
3974 void *para)
3975 {
3976 int i, ret = 0;
3977 u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3978 u32 num_buckets = clusters * bpc;
3979 struct ocfs2_xattr_bucket *bucket;
3980
3981 bucket = ocfs2_xattr_bucket_new(inode);
3982 if (!bucket) {
3983 mlog_errno(-ENOMEM);
3984 return -ENOMEM;
3985 }
3986
3987 trace_ocfs2_iterate_xattr_buckets(
3988 (unsigned long long)OCFS2_I(inode)->ip_blkno,
3989 (unsigned long long)blkno, clusters);
3990
3991 for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3992 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3993 if (ret) {
3994 mlog_errno(ret);
3995 break;
3996 }
3997
3998 /*
3999 * The real bucket num in this series of blocks is stored
4000 * in the 1st bucket.
4001 */
4002 if (i == 0)
4003 num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
4004
4005 trace_ocfs2_iterate_xattr_bucket((unsigned long long)blkno,
4006 le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
4007 if (func) {
4008 ret = func(inode, bucket, para);
4009 if (ret && ret != -ERANGE)
4010 mlog_errno(ret);
4011 /* Fall through to bucket_relse() */
4012 }
4013
4014 ocfs2_xattr_bucket_relse(bucket);
4015 if (ret)
4016 break;
4017 }
4018
4019 ocfs2_xattr_bucket_free(bucket);
4020 return ret;
4021 }
4022
4023 struct ocfs2_xattr_tree_list {
4024 char *buffer;
4025 size_t buffer_size;
4026 size_t result;
4027 };
4028
4029 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
4030 struct ocfs2_xattr_header *xh,
4031 int index,
4032 int *block_off,
4033 int *new_offset)
4034 {
4035 u16 name_offset;
4036
4037 if (index < 0 || index >= le16_to_cpu(xh->xh_count))
4038 return -EINVAL;
4039
4040 name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
4041
4042 *block_off = name_offset >> sb->s_blocksize_bits;
4043 *new_offset = name_offset % sb->s_blocksize;
4044
4045 return 0;
4046 }
4047
4048 static int ocfs2_list_xattr_bucket(struct inode *inode,
4049 struct ocfs2_xattr_bucket *bucket,
4050 void *para)
4051 {
4052 int ret = 0, type;
4053 struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
4054 int i, block_off, new_offset;
4055 const char *name;
4056
4057 for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
4058 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
4059 type = ocfs2_xattr_get_type(entry);
4060
4061 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
4062 bucket_xh(bucket),
4063 i,
4064 &block_off,
4065 &new_offset);
4066 if (ret)
4067 break;
4068
4069 name = (const char *)bucket_block(bucket, block_off) +
4070 new_offset;
4071 ret = ocfs2_xattr_list_entry(inode->i_sb,
4072 xl->buffer,
4073 xl->buffer_size,
4074 &xl->result,
4075 type, name,
4076 entry->xe_name_len);
4077 if (ret)
4078 break;
4079 }
4080
4081 return ret;
4082 }
4083
4084 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
4085 struct buffer_head *blk_bh,
4086 xattr_tree_rec_func *rec_func,
4087 void *para)
4088 {
4089 struct ocfs2_xattr_block *xb =
4090 (struct ocfs2_xattr_block *)blk_bh->b_data;
4091 struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4092 int ret = 0;
4093 u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
4094 u64 p_blkno = 0;
4095
4096 if (!el->l_next_free_rec || !rec_func)
4097 return 0;
4098
4099 while (name_hash > 0) {
4100 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4101 &e_cpos, &num_clusters, el);
4102 if (ret) {
4103 mlog_errno(ret);
4104 break;
4105 }
4106
4107 ret = rec_func(inode, blk_bh, p_blkno, e_cpos,
4108 num_clusters, para);
4109 if (ret) {
4110 if (ret != -ERANGE)
4111 mlog_errno(ret);
4112 break;
4113 }
4114
4115 if (e_cpos == 0)
4116 break;
4117
4118 name_hash = e_cpos - 1;
4119 }
4120
4121 return ret;
4122
4123 }
4124
4125 static int ocfs2_list_xattr_tree_rec(struct inode *inode,
4126 struct buffer_head *root_bh,
4127 u64 blkno, u32 cpos, u32 len, void *para)
4128 {
4129 return ocfs2_iterate_xattr_buckets(inode, blkno, len,
4130 ocfs2_list_xattr_bucket, para);
4131 }
4132
4133 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
4134 struct buffer_head *blk_bh,
4135 char *buffer,
4136 size_t buffer_size)
4137 {
4138 int ret;
4139 struct ocfs2_xattr_tree_list xl = {
4140 .buffer = buffer,
4141 .buffer_size = buffer_size,
4142 .result = 0,
4143 };
4144
4145 ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
4146 ocfs2_list_xattr_tree_rec, &xl);
4147 if (ret) {
4148 mlog_errno(ret);
4149 goto out;
4150 }
4151
4152 ret = xl.result;
4153 out:
4154 return ret;
4155 }
4156
4157 static int cmp_xe(const void *a, const void *b)
4158 {
4159 const struct ocfs2_xattr_entry *l = a, *r = b;
4160 u32 l_hash = le32_to_cpu(l->xe_name_hash);
4161 u32 r_hash = le32_to_cpu(r->xe_name_hash);
4162
4163 if (l_hash > r_hash)
4164 return 1;
4165 if (l_hash < r_hash)
4166 return -1;
4167 return 0;
4168 }
4169
4170 static void swap_xe(void *a, void *b, int size)
4171 {
4172 struct ocfs2_xattr_entry *l = a, *r = b, tmp;
4173
4174 tmp = *l;
4175 memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
4176 memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
4177 }
4178
4179 /*
4180 * When the ocfs2_xattr_block is filled up, new bucket will be created
4181 * and all the xattr entries will be moved to the new bucket.
4182 * The header goes at the start of the bucket, and the names+values are
4183 * filled from the end. This is why *target starts as the last buffer.
4184 * Note: we need to sort the entries since they are not saved in order
4185 * in the ocfs2_xattr_block.
4186 */
4187 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
4188 struct buffer_head *xb_bh,
4189 struct ocfs2_xattr_bucket *bucket)
4190 {
4191 int i, blocksize = inode->i_sb->s_blocksize;
4192 int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4193 u16 offset, size, off_change;
4194 struct ocfs2_xattr_entry *xe;
4195 struct ocfs2_xattr_block *xb =
4196 (struct ocfs2_xattr_block *)xb_bh->b_data;
4197 struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
4198 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4199 u16 count = le16_to_cpu(xb_xh->xh_count);
4200 char *src = xb_bh->b_data;
4201 char *target = bucket_block(bucket, blks - 1);
4202
4203 trace_ocfs2_cp_xattr_block_to_bucket_begin(
4204 (unsigned long long)xb_bh->b_blocknr,
4205 (unsigned long long)bucket_blkno(bucket));
4206
4207 for (i = 0; i < blks; i++)
4208 memset(bucket_block(bucket, i), 0, blocksize);
4209
4210 /*
4211 * Since the xe_name_offset is based on ocfs2_xattr_header,
4212 * there is a offset change corresponding to the change of
4213 * ocfs2_xattr_header's position.
4214 */
4215 off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4216 xe = &xb_xh->xh_entries[count - 1];
4217 offset = le16_to_cpu(xe->xe_name_offset) + off_change;
4218 size = blocksize - offset;
4219
4220 /* copy all the names and values. */
4221 memcpy(target + offset, src + offset, size);
4222
4223 /* Init new header now. */
4224 xh->xh_count = xb_xh->xh_count;
4225 xh->xh_num_buckets = cpu_to_le16(1);
4226 xh->xh_name_value_len = cpu_to_le16(size);
4227 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
4228
4229 /* copy all the entries. */
4230 target = bucket_block(bucket, 0);
4231 offset = offsetof(struct ocfs2_xattr_header, xh_entries);
4232 size = count * sizeof(struct ocfs2_xattr_entry);
4233 memcpy(target + offset, (char *)xb_xh + offset, size);
4234
4235 /* Change the xe offset for all the xe because of the move. */
4236 off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
4237 offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4238 for (i = 0; i < count; i++)
4239 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
4240
4241 trace_ocfs2_cp_xattr_block_to_bucket_end(offset, size, off_change);
4242
4243 sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
4244 cmp_xe, swap_xe);
4245 }
4246
4247 /*
4248 * After we move xattr from block to index btree, we have to
4249 * update ocfs2_xattr_search to the new xe and base.
4250 *
4251 * When the entry is in xattr block, xattr_bh indicates the storage place.
4252 * While if the entry is in index b-tree, "bucket" indicates the
4253 * real place of the xattr.
4254 */
4255 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
4256 struct ocfs2_xattr_search *xs,
4257 struct buffer_head *old_bh)
4258 {
4259 char *buf = old_bh->b_data;
4260 struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
4261 struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
4262 int i;
4263
4264 xs->header = bucket_xh(xs->bucket);
4265 xs->base = bucket_block(xs->bucket, 0);
4266 xs->end = xs->base + inode->i_sb->s_blocksize;
4267
4268 if (xs->not_found)
4269 return;
4270
4271 i = xs->here - old_xh->xh_entries;
4272 xs->here = &xs->header->xh_entries[i];
4273 }
4274
4275 static int ocfs2_xattr_create_index_block(struct inode *inode,
4276 struct ocfs2_xattr_search *xs,
4277 struct ocfs2_xattr_set_ctxt *ctxt)
4278 {
4279 int ret;
4280 u32 bit_off, len;
4281 u64 blkno;
4282 handle_t *handle = ctxt->handle;
4283 struct ocfs2_inode_info *oi = OCFS2_I(inode);
4284 struct buffer_head *xb_bh = xs->xattr_bh;
4285 struct ocfs2_xattr_block *xb =
4286 (struct ocfs2_xattr_block *)xb_bh->b_data;
4287 struct ocfs2_xattr_tree_root *xr;
4288 u16 xb_flags = le16_to_cpu(xb->xb_flags);
4289
4290 trace_ocfs2_xattr_create_index_block_begin(
4291 (unsigned long long)xb_bh->b_blocknr);
4292
4293 BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
4294 BUG_ON(!xs->bucket);
4295
4296 /*
4297 * XXX:
4298 * We can use this lock for now, and maybe move to a dedicated mutex
4299 * if performance becomes a problem later.
4300 */
4301 down_write(&oi->ip_alloc_sem);
4302
4303 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), xb_bh,
4304 OCFS2_JOURNAL_ACCESS_WRITE);
4305 if (ret) {
4306 mlog_errno(ret);
4307 goto out;
4308 }
4309
4310 ret = __ocfs2_claim_clusters(handle, ctxt->data_ac,
4311 1, 1, &bit_off, &len);
4312 if (ret) {
4313 mlog_errno(ret);
4314 goto out;
4315 }
4316
4317 /*
4318 * The bucket may spread in many blocks, and
4319 * we will only touch the 1st block and the last block
4320 * in the whole bucket(one for entry and one for data).
4321 */
4322 blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
4323
4324 trace_ocfs2_xattr_create_index_block((unsigned long long)blkno);
4325
4326 ret = ocfs2_init_xattr_bucket(xs->bucket, blkno, 1);
4327 if (ret) {
4328 mlog_errno(ret);
4329 goto out;
4330 }
4331
4332 ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4333 OCFS2_JOURNAL_ACCESS_CREATE);
4334 if (ret) {
4335 mlog_errno(ret);
4336 goto out;
4337 }
4338
4339 ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
4340 ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4341
4342 ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
4343
4344 /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
4345 memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
4346 offsetof(struct ocfs2_xattr_block, xb_attrs));
4347
4348 xr = &xb->xb_attrs.xb_root;
4349 xr->xt_clusters = cpu_to_le32(1);
4350 xr->xt_last_eb_blk = 0;
4351 xr->xt_list.l_tree_depth = 0;
4352 xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
4353 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
4354
4355 xr->xt_list.l_recs[0].e_cpos = 0;
4356 xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
4357 xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
4358
4359 xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
4360
4361 ocfs2_journal_dirty(handle, xb_bh);
4362
4363 out:
4364 up_write(&oi->ip_alloc_sem);
4365
4366 return ret;
4367 }
4368
4369 static int cmp_xe_offset(const void *a, const void *b)
4370 {
4371 const struct ocfs2_xattr_entry *l = a, *r = b;
4372 u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
4373 u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
4374
4375 if (l_name_offset < r_name_offset)
4376 return 1;
4377 if (l_name_offset > r_name_offset)
4378 return -1;
4379 return 0;
4380 }
4381
4382 /*
4383 * defrag a xattr bucket if we find that the bucket has some
4384 * holes beteen name/value pairs.
4385 * We will move all the name/value pairs to the end of the bucket
4386 * so that we can spare some space for insertion.
4387 */
4388 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
4389 handle_t *handle,
4390 struct ocfs2_xattr_bucket *bucket)
4391 {
4392 int ret, i;
4393 size_t end, offset, len;
4394 struct ocfs2_xattr_header *xh;
4395 char *entries, *buf, *bucket_buf = NULL;
4396 u64 blkno = bucket_blkno(bucket);
4397 u16 xh_free_start;
4398 size_t blocksize = inode->i_sb->s_blocksize;
4399 struct ocfs2_xattr_entry *xe;
4400
4401 /*
4402 * In order to make the operation more efficient and generic,
4403 * we copy all the blocks into a contiguous memory and do the
4404 * defragment there, so if anything is error, we will not touch
4405 * the real block.
4406 */
4407 bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
4408 if (!bucket_buf) {
4409 ret = -EIO;
4410 goto out;
4411 }
4412
4413 buf = bucket_buf;
4414 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4415 memcpy(buf, bucket_block(bucket, i), blocksize);
4416
4417 ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
4418 OCFS2_JOURNAL_ACCESS_WRITE);
4419 if (ret < 0) {
4420 mlog_errno(ret);
4421 goto out;
4422 }
4423
4424 xh = (struct ocfs2_xattr_header *)bucket_buf;
4425 entries = (char *)xh->xh_entries;
4426 xh_free_start = le16_to_cpu(xh->xh_free_start);
4427
4428 trace_ocfs2_defrag_xattr_bucket(
4429 (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
4430 xh_free_start, le16_to_cpu(xh->xh_name_value_len));
4431
4432 /*
4433 * sort all the entries by their offset.
4434 * the largest will be the first, so that we can
4435 * move them to the end one by one.
4436 */
4437 sort(entries, le16_to_cpu(xh->xh_count),
4438 sizeof(struct ocfs2_xattr_entry),
4439 cmp_xe_offset, swap_xe);
4440
4441 /* Move all name/values to the end of the bucket. */
4442 xe = xh->xh_entries;
4443 end = OCFS2_XATTR_BUCKET_SIZE;
4444 for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
4445 offset = le16_to_cpu(xe->xe_name_offset);
4446 len = namevalue_size_xe(xe);
4447
4448 /*
4449 * We must make sure that the name/value pair
4450 * exist in the same block. So adjust end to
4451 * the previous block end if needed.
4452 */
4453 if (((end - len) / blocksize !=
4454 (end - 1) / blocksize))
4455 end = end - end % blocksize;
4456
4457 if (end > offset + len) {
4458 memmove(bucket_buf + end - len,
4459 bucket_buf + offset, len);
4460 xe->xe_name_offset = cpu_to_le16(end - len);
4461 }
4462
4463 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
4464 "bucket %llu\n", (unsigned long long)blkno);
4465
4466 end -= len;
4467 }
4468
4469 mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
4470 "bucket %llu\n", (unsigned long long)blkno);
4471
4472 if (xh_free_start == end)
4473 goto out;
4474
4475 memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
4476 xh->xh_free_start = cpu_to_le16(end);
4477
4478 /* sort the entries by their name_hash. */
4479 sort(entries, le16_to_cpu(xh->xh_count),
4480 sizeof(struct ocfs2_xattr_entry),
4481 cmp_xe, swap_xe);
4482
4483 buf = bucket_buf;
4484 for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4485 memcpy(bucket_block(bucket, i), buf, blocksize);
4486 ocfs2_xattr_bucket_journal_dirty(handle, bucket);
4487
4488 out:
4489 kfree(bucket_buf);
4490 return ret;
4491 }
4492
4493 /*
4494 * prev_blkno points to the start of an existing extent. new_blkno
4495 * points to a newly allocated extent. Because we know each of our
4496 * clusters contains more than bucket, we can easily split one cluster
4497 * at a bucket boundary. So we take the last cluster of the existing
4498 * extent and split it down the middle. We move the last half of the
4499 * buckets in the last cluster of the existing extent over to the new
4500 * extent.
4501 *
4502 * first_bh is the buffer at prev_blkno so we can update the existing
4503 * extent's bucket count. header_bh is the bucket were we were hoping
4504 * to insert our xattr. If the bucket move places the target in the new
4505 * extent, we'll update first_bh and header_bh after modifying the old
4506 * extent.
4507 *
4508 * first_hash will be set as the 1st xe's name_hash in the new extent.
4509 */
4510 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
4511 handle_t *handle,
4512 struct ocfs2_xattr_bucket *first,
4513 struct ocfs2_xattr_bucket *target,
4514 u64 new_blkno,
4515 u32 num_clusters,
4516 u32 *first_hash)
4517 {
4518 int ret;
4519 struct super_block *sb = inode->i_sb;
4520 int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
4521 int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
4522 int to_move = num_buckets / 2;
4523 u64 src_blkno;
4524 u64 last_cluster_blkno = bucket_blkno(first) +
4525 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
4526
4527 BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
4528 BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
4529
4530 trace_ocfs2_mv_xattr_bucket_cross_cluster(
4531 (unsigned long long)last_cluster_blkno,
4532 (unsigned long long)new_blkno);
4533
4534 ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
4535 last_cluster_blkno, new_blkno,
4536 to_move, first_hash);
4537 if (ret) {
4538 mlog_errno(ret);
4539 goto out;
4540 }
4541
4542 /* This is the first bucket that got moved */
4543 src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
4544
4545 /*
4546 * If the target bucket was part of the moved buckets, we need to
4547 * update first and target.
4548 */
4549 if (bucket_blkno(target) >= src_blkno) {
4550 /* Find the block for the new target bucket */
4551 src_blkno = new_blkno +
4552 (bucket_blkno(target) - src_blkno);
4553
4554 ocfs2_xattr_bucket_relse(first);
4555 ocfs2_xattr_bucket_relse(target);
4556
4557 /*
4558 * These shouldn't fail - the buffers are in the
4559 * journal from ocfs2_cp_xattr_bucket().
4560 */
4561 ret = ocfs2_read_xattr_bucket(first, new_blkno);
4562 if (ret) {
4563 mlog_errno(ret);
4564 goto out;
4565 }
4566 ret = ocfs2_read_xattr_bucket(target, src_blkno);
4567 if (ret)
4568 mlog_errno(ret);
4569
4570 }
4571
4572 out:
4573 return ret;
4574 }
4575
4576 /*
4577 * Find the suitable pos when we divide a bucket into 2.
4578 * We have to make sure the xattrs with the same hash value exist
4579 * in the same bucket.
4580 *
4581 * If this ocfs2_xattr_header covers more than one hash value, find a
4582 * place where the hash value changes. Try to find the most even split.
4583 * The most common case is that all entries have different hash values,
4584 * and the first check we make will find a place to split.
4585 */
4586 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
4587 {
4588 struct ocfs2_xattr_entry *entries = xh->xh_entries;
4589 int count = le16_to_cpu(xh->xh_count);
4590 int delta, middle = count / 2;
4591
4592 /*
4593 * We start at the middle. Each step gets farther away in both
4594 * directions. We therefore hit the change in hash value
4595 * nearest to the middle. Note that this loop does not execute for
4596 * count < 2.
4597 */
4598 for (delta = 0; delta < middle; delta++) {
4599 /* Let's check delta earlier than middle */
4600 if (cmp_xe(&entries[middle - delta - 1],
4601 &entries[middle - delta]))
4602 return middle - delta;
4603
4604 /* For even counts, don't walk off the end */
4605 if ((middle + delta + 1) == count)
4606 continue;
4607
4608 /* Now try delta past middle */
4609 if (cmp_xe(&entries[middle + delta],
4610 &entries[middle + delta + 1]))
4611 return middle + delta + 1;
4612 }
4613
4614 /* Every entry had the same hash */
4615 return count;
4616 }
4617
4618 /*
4619 * Move some xattrs in old bucket(blk) to new bucket(new_blk).
4620 * first_hash will record the 1st hash of the new bucket.
4621 *
4622 * Normally half of the xattrs will be moved. But we have to make
4623 * sure that the xattrs with the same hash value are stored in the
4624 * same bucket. If all the xattrs in this bucket have the same hash
4625 * value, the new bucket will be initialized as an empty one and the
4626 * first_hash will be initialized as (hash_value+1).
4627 */
4628 static int ocfs2_divide_xattr_bucket(struct inode *inode,
4629 handle_t *handle,
4630 u64 blk,
4631 u64 new_blk,
4632 u32 *first_hash,
4633 int new_bucket_head)
4634 {
4635 int ret, i;
4636 int count, start, len, name_value_len = 0, name_offset = 0;
4637 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4638 struct ocfs2_xattr_header *xh;
4639 struct ocfs2_xattr_entry *xe;
4640 int blocksize = inode->i_sb->s_blocksize;
4641
4642 trace_ocfs2_divide_xattr_bucket_begin((unsigned long long)blk,
4643 (unsigned long long)new_blk);
4644
4645 s_bucket = ocfs2_xattr_bucket_new(inode);
4646 t_bucket = ocfs2_xattr_bucket_new(inode);
4647 if (!s_bucket || !t_bucket) {
4648 ret = -ENOMEM;
4649 mlog_errno(ret);
4650 goto out;
4651 }
4652
4653 ret = ocfs2_read_xattr_bucket(s_bucket, blk);
4654 if (ret) {
4655 mlog_errno(ret);
4656 goto out;
4657 }
4658
4659 ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
4660 OCFS2_JOURNAL_ACCESS_WRITE);
4661 if (ret) {
4662 mlog_errno(ret);
4663 goto out;
4664 }
4665
4666 /*
4667 * Even if !new_bucket_head, we're overwriting t_bucket. Thus,
4668 * there's no need to read it.
4669 */
4670 ret = ocfs2_init_xattr_bucket(t_bucket, new_blk, new_bucket_head);
4671 if (ret) {
4672 mlog_errno(ret);
4673 goto out;
4674 }
4675
4676 /*
4677 * Hey, if we're overwriting t_bucket, what difference does
4678 * ACCESS_CREATE vs ACCESS_WRITE make? See the comment in the
4679 * same part of ocfs2_cp_xattr_bucket().
4680 */
4681 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4682 new_bucket_head ?
4683 OCFS2_JOURNAL_ACCESS_CREATE :
4684 OCFS2_JOURNAL_ACCESS_WRITE);
4685 if (ret) {
4686 mlog_errno(ret);
4687 goto out;
4688 }
4689
4690 xh = bucket_xh(s_bucket);
4691 count = le16_to_cpu(xh->xh_count);
4692 start = ocfs2_xattr_find_divide_pos(xh);
4693
4694 if (start == count) {
4695 xe = &xh->xh_entries[start-1];
4696
4697 /*
4698 * initialized a new empty bucket here.
4699 * The hash value is set as one larger than
4700 * that of the last entry in the previous bucket.
4701 */
4702 for (i = 0; i < t_bucket->bu_blocks; i++)
4703 memset(bucket_block(t_bucket, i), 0, blocksize);
4704
4705 xh = bucket_xh(t_bucket);
4706 xh->xh_free_start = cpu_to_le16(blocksize);
4707 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
4708 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
4709
4710 goto set_num_buckets;
4711 }
4712
4713 /* copy the whole bucket to the new first. */
4714 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4715
4716 /* update the new bucket. */
4717 xh = bucket_xh(t_bucket);
4718
4719 /*
4720 * Calculate the total name/value len and xh_free_start for
4721 * the old bucket first.
4722 */
4723 name_offset = OCFS2_XATTR_BUCKET_SIZE;
4724 name_value_len = 0;
4725 for (i = 0; i < start; i++) {
4726 xe = &xh->xh_entries[i];
4727 name_value_len += namevalue_size_xe(xe);
4728 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
4729 name_offset = le16_to_cpu(xe->xe_name_offset);
4730 }
4731
4732 /*
4733 * Now begin the modification to the new bucket.
4734 *
4735 * In the new bucket, We just move the xattr entry to the beginning
4736 * and don't touch the name/value. So there will be some holes in the
4737 * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
4738 * called.
4739 */
4740 xe = &xh->xh_entries[start];
4741 len = sizeof(struct ocfs2_xattr_entry) * (count - start);
4742 trace_ocfs2_divide_xattr_bucket_move(len,
4743 (int)((char *)xe - (char *)xh),
4744 (int)((char *)xh->xh_entries - (char *)xh));
4745 memmove((char *)xh->xh_entries, (char *)xe, len);
4746 xe = &xh->xh_entries[count - start];
4747 len = sizeof(struct ocfs2_xattr_entry) * start;
4748 memset((char *)xe, 0, len);
4749
4750 le16_add_cpu(&xh->xh_count, -start);
4751 le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
4752
4753 /* Calculate xh_free_start for the new bucket. */
4754 xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4755 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4756 xe = &xh->xh_entries[i];
4757 if (le16_to_cpu(xe->xe_name_offset) <
4758 le16_to_cpu(xh->xh_free_start))
4759 xh->xh_free_start = xe->xe_name_offset;
4760 }
4761
4762 set_num_buckets:
4763 /* set xh->xh_num_buckets for the new xh. */
4764 if (new_bucket_head)
4765 xh->xh_num_buckets = cpu_to_le16(1);
4766 else
4767 xh->xh_num_buckets = 0;
4768
4769 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4770
4771 /* store the first_hash of the new bucket. */
4772 if (first_hash)
4773 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
4774
4775 /*
4776 * Now only update the 1st block of the old bucket. If we
4777 * just added a new empty bucket, there is no need to modify
4778 * it.
4779 */
4780 if (start == count)
4781 goto out;
4782
4783 xh = bucket_xh(s_bucket);
4784 memset(&xh->xh_entries[start], 0,
4785 sizeof(struct ocfs2_xattr_entry) * (count - start));
4786 xh->xh_count = cpu_to_le16(start);
4787 xh->xh_free_start = cpu_to_le16(name_offset);
4788 xh->xh_name_value_len = cpu_to_le16(name_value_len);
4789
4790 ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
4791
4792 out:
4793 ocfs2_xattr_bucket_free(s_bucket);
4794 ocfs2_xattr_bucket_free(t_bucket);
4795
4796 return ret;
4797 }
4798
4799 /*
4800 * Copy xattr from one bucket to another bucket.
4801 *
4802 * The caller must make sure that the journal transaction
4803 * has enough space for journaling.
4804 */
4805 static int ocfs2_cp_xattr_bucket(struct inode *inode,
4806 handle_t *handle,
4807 u64 s_blkno,
4808 u64 t_blkno,
4809 int t_is_new)
4810 {
4811 int ret;
4812 struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4813
4814 BUG_ON(s_blkno == t_blkno);
4815
4816 trace_ocfs2_cp_xattr_bucket((unsigned long long)s_blkno,
4817 (unsigned long long)t_blkno,
4818 t_is_new);
4819
4820 s_bucket = ocfs2_xattr_bucket_new(inode);
4821 t_bucket = ocfs2_xattr_bucket_new(inode);
4822 if (!s_bucket || !t_bucket) {
4823 ret = -ENOMEM;
4824 mlog_errno(ret);
4825 goto out;
4826 }
4827
4828 ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
4829 if (ret)
4830 goto out;
4831
4832 /*
4833 * Even if !t_is_new, we're overwriting t_bucket. Thus,
4834 * there's no need to read it.
4835 */
4836 ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno, t_is_new);
4837 if (ret)
4838 goto out;
4839
4840 /*
4841 * Hey, if we're overwriting t_bucket, what difference does
4842 * ACCESS_CREATE vs ACCESS_WRITE make? Well, if we allocated a new
4843 * cluster to fill, we came here from
4844 * ocfs2_mv_xattr_buckets(), and it is really new -
4845 * ACCESS_CREATE is required. But we also might have moved data
4846 * out of t_bucket before extending back into it.
4847 * ocfs2_add_new_xattr_bucket() can do this - its call to
4848 * ocfs2_add_new_xattr_cluster() may have created a new extent
4849 * and copied out the end of the old extent. Then it re-extends
4850 * the old extent back to create space for new xattrs. That's
4851 * how we get here, and the bucket isn't really new.
4852 */
4853 ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4854 t_is_new ?
4855 OCFS2_JOURNAL_ACCESS_CREATE :
4856 OCFS2_JOURNAL_ACCESS_WRITE);
4857 if (ret)
4858 goto out;
4859
4860 ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4861 ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4862
4863 out:
4864 ocfs2_xattr_bucket_free(t_bucket);
4865 ocfs2_xattr_bucket_free(s_bucket);
4866
4867 return ret;
4868 }
4869
4870 /*
4871 * src_blk points to the start of an existing extent. last_blk points to
4872 * last cluster in that extent. to_blk points to a newly allocated
4873 * extent. We copy the buckets from the cluster at last_blk to the new
4874 * extent. If start_bucket is non-zero, we skip that many buckets before
4875 * we start copying. The new extent's xh_num_buckets gets set to the
4876 * number of buckets we copied. The old extent's xh_num_buckets shrinks
4877 * by the same amount.
4878 */
4879 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4880 u64 src_blk, u64 last_blk, u64 to_blk,
4881 unsigned int start_bucket,
4882 u32 *first_hash)
4883 {
4884 int i, ret, credits;
4885 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4886 int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4887 int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4888 struct ocfs2_xattr_bucket *old_first, *new_first;
4889
4890 trace_ocfs2_mv_xattr_buckets((unsigned long long)last_blk,
4891 (unsigned long long)to_blk);
4892
4893 BUG_ON(start_bucket >= num_buckets);
4894 if (start_bucket) {
4895 num_buckets -= start_bucket;
4896 last_blk += (start_bucket * blks_per_bucket);
4897 }
4898
4899 /* The first bucket of the original extent */
4900 old_first = ocfs2_xattr_bucket_new(inode);
4901 /* The first bucket of the new extent */
4902 new_first = ocfs2_xattr_bucket_new(inode);
4903 if (!old_first || !new_first) {
4904 ret = -ENOMEM;
4905 mlog_errno(ret);
4906 goto out;
4907 }
4908
4909 ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4910 if (ret) {
4911 mlog_errno(ret);
4912 goto out;
4913 }
4914
4915 /*
4916 * We need to update the first bucket of the old extent and all
4917 * the buckets going to the new extent.
4918 */
4919 credits = ((num_buckets + 1) * blks_per_bucket);
4920 ret = ocfs2_extend_trans(handle, credits);
4921 if (ret) {
4922 mlog_errno(ret);
4923 goto out;
4924 }
4925
4926 ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4927 OCFS2_JOURNAL_ACCESS_WRITE);
4928 if (ret) {
4929 mlog_errno(ret);
4930 goto out;
4931 }
4932
4933 for (i = 0; i < num_buckets; i++) {
4934 ret = ocfs2_cp_xattr_bucket(inode, handle,
4935 last_blk + (i * blks_per_bucket),
4936 to_blk + (i * blks_per_bucket),
4937 1);
4938 if (ret) {
4939 mlog_errno(ret);
4940 goto out;
4941 }
4942 }
4943
4944 /*
4945 * Get the new bucket ready before we dirty anything
4946 * (This actually shouldn't fail, because we already dirtied
4947 * it once in ocfs2_cp_xattr_bucket()).
4948 */
4949 ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4950 if (ret) {
4951 mlog_errno(ret);
4952 goto out;
4953 }
4954 ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4955 OCFS2_JOURNAL_ACCESS_WRITE);
4956 if (ret) {
4957 mlog_errno(ret);
4958 goto out;
4959 }
4960
4961 /* Now update the headers */
4962 le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4963 ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4964
4965 bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4966 ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4967
4968 if (first_hash)
4969 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4970
4971 out:
4972 ocfs2_xattr_bucket_free(new_first);
4973 ocfs2_xattr_bucket_free(old_first);
4974 return ret;
4975 }
4976
4977 /*
4978 * Move some xattrs in this cluster to the new cluster.
4979 * This function should only be called when bucket size == cluster size.
4980 * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4981 */
4982 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4983 handle_t *handle,
4984 u64 prev_blk,
4985 u64 new_blk,
4986 u32 *first_hash)
4987 {
4988 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4989 int ret, credits = 2 * blk_per_bucket;
4990
4991 BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4992
4993 ret = ocfs2_extend_trans(handle, credits);
4994 if (ret) {
4995 mlog_errno(ret);
4996 return ret;
4997 }
4998
4999 /* Move half of the xattr in start_blk to the next bucket. */
5000 return ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
5001 new_blk, first_hash, 1);
5002 }
5003
5004 /*
5005 * Move some xattrs from the old cluster to the new one since they are not
5006 * contiguous in ocfs2 xattr tree.
5007 *
5008 * new_blk starts a new separate cluster, and we will move some xattrs from
5009 * prev_blk to it. v_start will be set as the first name hash value in this
5010 * new cluster so that it can be used as e_cpos during tree insertion and
5011 * don't collide with our original b-tree operations. first_bh and header_bh
5012 * will also be updated since they will be used in ocfs2_extend_xattr_bucket
5013 * to extend the insert bucket.
5014 *
5015 * The problem is how much xattr should we move to the new one and when should
5016 * we update first_bh and header_bh?
5017 * 1. If cluster size > bucket size, that means the previous cluster has more
5018 * than 1 bucket, so just move half nums of bucket into the new cluster and
5019 * update the first_bh and header_bh if the insert bucket has been moved
5020 * to the new cluster.
5021 * 2. If cluster_size == bucket_size:
5022 * a) If the previous extent rec has more than one cluster and the insert
5023 * place isn't in the last cluster, copy the entire last cluster to the
5024 * new one. This time, we don't need to upate the first_bh and header_bh
5025 * since they will not be moved into the new cluster.
5026 * b) Otherwise, move the bottom half of the xattrs in the last cluster into
5027 * the new one. And we set the extend flag to zero if the insert place is
5028 * moved into the new allocated cluster since no extend is needed.
5029 */
5030 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
5031 handle_t *handle,
5032 struct ocfs2_xattr_bucket *first,
5033 struct ocfs2_xattr_bucket *target,
5034 u64 new_blk,
5035 u32 prev_clusters,
5036 u32 *v_start,
5037 int *extend)
5038 {
5039 int ret;
5040
5041 trace_ocfs2_adjust_xattr_cross_cluster(
5042 (unsigned long long)bucket_blkno(first),
5043 (unsigned long long)new_blk, prev_clusters);
5044
5045 if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
5046 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
5047 handle,
5048 first, target,
5049 new_blk,
5050 prev_clusters,
5051 v_start);
5052 if (ret)
5053 mlog_errno(ret);
5054 } else {
5055 /* The start of the last cluster in the first extent */
5056 u64 last_blk = bucket_blkno(first) +
5057 ((prev_clusters - 1) *
5058 ocfs2_clusters_to_blocks(inode->i_sb, 1));
5059
5060 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
5061 ret = ocfs2_mv_xattr_buckets(inode, handle,
5062 bucket_blkno(first),
5063 last_blk, new_blk, 0,
5064 v_start);
5065 if (ret)
5066 mlog_errno(ret);
5067 } else {
5068 ret = ocfs2_divide_xattr_cluster(inode, handle,
5069 last_blk, new_blk,
5070 v_start);
5071 if (ret)
5072 mlog_errno(ret);
5073
5074 if ((bucket_blkno(target) == last_blk) && extend)
5075 *extend = 0;
5076 }
5077 }
5078
5079 return ret;
5080 }
5081
5082 /*
5083 * Add a new cluster for xattr storage.
5084 *
5085 * If the new cluster is contiguous with the previous one, it will be
5086 * appended to the same extent record, and num_clusters will be updated.
5087 * If not, we will insert a new extent for it and move some xattrs in
5088 * the last cluster into the new allocated one.
5089 * We also need to limit the maximum size of a btree leaf, otherwise we'll
5090 * lose the benefits of hashing because we'll have to search large leaves.
5091 * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
5092 * if it's bigger).
5093 *
5094 * first_bh is the first block of the previous extent rec and header_bh
5095 * indicates the bucket we will insert the new xattrs. They will be updated
5096 * when the header_bh is moved into the new cluster.
5097 */
5098 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
5099 struct buffer_head *root_bh,
5100 struct ocfs2_xattr_bucket *first,
5101 struct ocfs2_xattr_bucket *target,
5102 u32 *num_clusters,
5103 u32 prev_cpos,
5104 int *extend,
5105 struct ocfs2_xattr_set_ctxt *ctxt)
5106 {
5107 int ret;
5108 u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
5109 u32 prev_clusters = *num_clusters;
5110 u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
5111 u64 block;
5112 handle_t *handle = ctxt->handle;
5113 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5114 struct ocfs2_extent_tree et;
5115
5116 trace_ocfs2_add_new_xattr_cluster_begin(
5117 (unsigned long long)OCFS2_I(inode)->ip_blkno,
5118 (unsigned long long)bucket_blkno(first),
5119 prev_cpos, prev_clusters);
5120
5121 ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5122
5123 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5124 OCFS2_JOURNAL_ACCESS_WRITE);
5125 if (ret < 0) {
5126 mlog_errno(ret);
5127 goto leave;
5128 }
5129
5130 ret = __ocfs2_claim_clusters(handle, ctxt->data_ac, 1,
5131 clusters_to_add, &bit_off, &num_bits);
5132 if (ret < 0) {
5133 if (ret != -ENOSPC)
5134 mlog_errno(ret);
5135 goto leave;
5136 }
5137
5138 BUG_ON(num_bits > clusters_to_add);
5139
5140 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
5141 trace_ocfs2_add_new_xattr_cluster((unsigned long long)block, num_bits);
5142
5143 if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
5144 (prev_clusters + num_bits) << osb->s_clustersize_bits <=
5145 OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
5146 /*
5147 * If this cluster is contiguous with the old one and
5148 * adding this new cluster, we don't surpass the limit of
5149 * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
5150 * initialized and used like other buckets in the previous
5151 * cluster.
5152 * So add it as a contiguous one. The caller will handle
5153 * its init process.
5154 */
5155 v_start = prev_cpos + prev_clusters;
5156 *num_clusters = prev_clusters + num_bits;
5157 } else {
5158 ret = ocfs2_adjust_xattr_cross_cluster(inode,
5159 handle,
5160 first,
5161 target,
5162 block,
5163 prev_clusters,
5164 &v_start,
5165 extend);
5166 if (ret) {
5167 mlog_errno(ret);
5168 goto leave;
5169 }
5170 }
5171
5172 trace_ocfs2_add_new_xattr_cluster_insert((unsigned long long)block,
5173 v_start, num_bits);
5174 ret = ocfs2_insert_extent(handle, &et, v_start, block,
5175 num_bits, 0, ctxt->meta_ac);
5176 if (ret < 0) {
5177 mlog_errno(ret);
5178 goto leave;
5179 }
5180
5181 ocfs2_journal_dirty(handle, root_bh);
5182
5183 leave:
5184 return ret;
5185 }
5186
5187 /*
5188 * We are given an extent. 'first' is the bucket at the very front of
5189 * the extent. The extent has space for an additional bucket past
5190 * bucket_xh(first)->xh_num_buckets. 'target_blkno' is the block number
5191 * of the target bucket. We wish to shift every bucket past the target
5192 * down one, filling in that additional space. When we get back to the
5193 * target, we split the target between itself and the now-empty bucket
5194 * at target+1 (aka, target_blkno + blks_per_bucket).
5195 */
5196 static int ocfs2_extend_xattr_bucket(struct inode *inode,
5197 handle_t *handle,
5198 struct ocfs2_xattr_bucket *first,
5199 u64 target_blk,
5200 u32 num_clusters)
5201 {
5202 int ret, credits;
5203 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5204 u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5205 u64 end_blk;
5206 u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
5207
5208 trace_ocfs2_extend_xattr_bucket((unsigned long long)target_blk,
5209 (unsigned long long)bucket_blkno(first),
5210 num_clusters, new_bucket);
5211
5212 /* The extent must have room for an additional bucket */
5213 BUG_ON(new_bucket >=
5214 (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
5215
5216 /* end_blk points to the last existing bucket */
5217 end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
5218
5219 /*
5220 * end_blk is the start of the last existing bucket.
5221 * Thus, (end_blk - target_blk) covers the target bucket and
5222 * every bucket after it up to, but not including, the last
5223 * existing bucket. Then we add the last existing bucket, the
5224 * new bucket, and the first bucket (3 * blk_per_bucket).
5225 */
5226 credits = (end_blk - target_blk) + (3 * blk_per_bucket);
5227 ret = ocfs2_extend_trans(handle, credits);
5228 if (ret) {
5229 mlog_errno(ret);
5230 goto out;
5231 }
5232
5233 ret = ocfs2_xattr_bucket_journal_access(handle, first,
5234 OCFS2_JOURNAL_ACCESS_WRITE);
5235 if (ret) {
5236 mlog_errno(ret);
5237 goto out;
5238 }
5239
5240 while (end_blk != target_blk) {
5241 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
5242 end_blk + blk_per_bucket, 0);
5243 if (ret)
5244 goto out;
5245 end_blk -= blk_per_bucket;
5246 }
5247
5248 /* Move half of the xattr in target_blkno to the next bucket. */
5249 ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
5250 target_blk + blk_per_bucket, NULL, 0);
5251
5252 le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
5253 ocfs2_xattr_bucket_journal_dirty(handle, first);
5254
5255 out:
5256 return ret;
5257 }
5258
5259 /*
5260 * Add new xattr bucket in an extent record and adjust the buckets
5261 * accordingly. xb_bh is the ocfs2_xattr_block, and target is the
5262 * bucket we want to insert into.
5263 *
5264 * In the easy case, we will move all the buckets after target down by
5265 * one. Half of target's xattrs will be moved to the next bucket.
5266 *
5267 * If current cluster is full, we'll allocate a new one. This may not
5268 * be contiguous. The underlying calls will make sure that there is
5269 * space for the insert, shifting buckets around if necessary.
5270 * 'target' may be moved by those calls.
5271 */
5272 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
5273 struct buffer_head *xb_bh,
5274 struct ocfs2_xattr_bucket *target,
5275 struct ocfs2_xattr_set_ctxt *ctxt)
5276 {
5277 struct ocfs2_xattr_block *xb =
5278 (struct ocfs2_xattr_block *)xb_bh->b_data;
5279 struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
5280 struct ocfs2_extent_list *el = &xb_root->xt_list;
5281 u32 name_hash =
5282 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
5283 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5284 int ret, num_buckets, extend = 1;
5285 u64 p_blkno;
5286 u32 e_cpos, num_clusters;
5287 /* The bucket at the front of the extent */
5288 struct ocfs2_xattr_bucket *first;
5289
5290 trace_ocfs2_add_new_xattr_bucket(
5291 (unsigned long long)bucket_blkno(target));
5292
5293 /* The first bucket of the original extent */
5294 first = ocfs2_xattr_bucket_new(inode);
5295 if (!first) {
5296 ret = -ENOMEM;
5297 mlog_errno(ret);
5298 goto out;
5299 }
5300
5301 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
5302 &num_clusters, el);
5303 if (ret) {
5304 mlog_errno(ret);
5305 goto out;
5306 }
5307
5308 ret = ocfs2_read_xattr_bucket(first, p_blkno);
5309 if (ret) {
5310 mlog_errno(ret);
5311 goto out;
5312 }
5313
5314 num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
5315 if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
5316 /*
5317 * This can move first+target if the target bucket moves
5318 * to the new extent.
5319 */
5320 ret = ocfs2_add_new_xattr_cluster(inode,
5321 xb_bh,
5322 first,
5323 target,
5324 &num_clusters,
5325 e_cpos,
5326 &extend,
5327 ctxt);
5328 if (ret) {
5329 mlog_errno(ret);
5330 goto out;
5331 }
5332 }
5333
5334 if (extend) {
5335 ret = ocfs2_extend_xattr_bucket(inode,
5336 ctxt->handle,
5337 first,
5338 bucket_blkno(target),
5339 num_clusters);
5340 if (ret)
5341 mlog_errno(ret);
5342 }
5343
5344 out:
5345 ocfs2_xattr_bucket_free(first);
5346
5347 return ret;
5348 }
5349
5350 /*
5351 * Truncate the specified xe_off entry in xattr bucket.
5352 * bucket is indicated by header_bh and len is the new length.
5353 * Both the ocfs2_xattr_value_root and the entry will be updated here.
5354 *
5355 * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
5356 */
5357 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
5358 struct ocfs2_xattr_bucket *bucket,
5359 int xe_off,
5360 int len,
5361 struct ocfs2_xattr_set_ctxt *ctxt)
5362 {
5363 int ret, offset;
5364 u64 value_blk;
5365 struct ocfs2_xattr_entry *xe;
5366 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5367 size_t blocksize = inode->i_sb->s_blocksize;
5368 struct ocfs2_xattr_value_buf vb = {
5369 .vb_access = ocfs2_journal_access,
5370 };
5371
5372 xe = &xh->xh_entries[xe_off];
5373
5374 BUG_ON(!xe || ocfs2_xattr_is_local(xe));
5375
5376 offset = le16_to_cpu(xe->xe_name_offset) +
5377 OCFS2_XATTR_SIZE(xe->xe_name_len);
5378
5379 value_blk = offset / blocksize;
5380
5381 /* We don't allow ocfs2_xattr_value to be stored in different block. */
5382 BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
5383
5384 vb.vb_bh = bucket->bu_bhs[value_blk];
5385 BUG_ON(!vb.vb_bh);
5386
5387 vb.vb_xv = (struct ocfs2_xattr_value_root *)
5388 (vb.vb_bh->b_data + offset % blocksize);
5389
5390 /*
5391 * From here on out we have to dirty the bucket. The generic
5392 * value calls only modify one of the bucket's bhs, but we need
5393 * to send the bucket at once. So if they error, they *could* have
5394 * modified something. We have to assume they did, and dirty
5395 * the whole bucket. This leaves us in a consistent state.
5396 */
5397 trace_ocfs2_xattr_bucket_value_truncate(
5398 (unsigned long long)bucket_blkno(bucket), xe_off, len);
5399 ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
5400 if (ret) {
5401 mlog_errno(ret);
5402 goto out;
5403 }
5404
5405 ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
5406 OCFS2_JOURNAL_ACCESS_WRITE);
5407 if (ret) {
5408 mlog_errno(ret);
5409 goto out;
5410 }
5411
5412 xe->xe_value_size = cpu_to_le64(len);
5413
5414 ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
5415
5416 out:
5417 return ret;
5418 }
5419
5420 static int ocfs2_rm_xattr_cluster(struct inode *inode,
5421 struct buffer_head *root_bh,
5422 u64 blkno,
5423 u32 cpos,
5424 u32 len,
5425 void *para)
5426 {
5427 int ret;
5428 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5429 struct inode *tl_inode = osb->osb_tl_inode;
5430 handle_t *handle;
5431 struct ocfs2_xattr_block *xb =
5432 (struct ocfs2_xattr_block *)root_bh->b_data;
5433 struct ocfs2_alloc_context *meta_ac = NULL;
5434 struct ocfs2_cached_dealloc_ctxt dealloc;
5435 struct ocfs2_extent_tree et;
5436
5437 ret = ocfs2_iterate_xattr_buckets(inode, blkno, len,
5438 ocfs2_delete_xattr_in_bucket, para);
5439 if (ret) {
5440 mlog_errno(ret);
5441 return ret;
5442 }
5443
5444 ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5445
5446 ocfs2_init_dealloc_ctxt(&dealloc);
5447
5448 trace_ocfs2_rm_xattr_cluster(
5449 (unsigned long long)OCFS2_I(inode)->ip_blkno,
5450 (unsigned long long)blkno, cpos, len);
5451
5452 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode), blkno,
5453 len);
5454
5455 ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
5456 if (ret) {
5457 mlog_errno(ret);
5458 return ret;
5459 }
5460
5461 inode_lock(tl_inode);
5462
5463 if (ocfs2_truncate_log_needs_flush(osb)) {
5464 ret = __ocfs2_flush_truncate_log(osb);
5465 if (ret < 0) {
5466 mlog_errno(ret);
5467 goto out;
5468 }
5469 }
5470
5471 handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
5472 if (IS_ERR(handle)) {
5473 ret = -ENOMEM;
5474 mlog_errno(ret);
5475 goto out;
5476 }
5477
5478 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5479 OCFS2_JOURNAL_ACCESS_WRITE);
5480 if (ret) {
5481 mlog_errno(ret);
5482 goto out_commit;
5483 }
5484
5485 ret = ocfs2_remove_extent(handle, &et, cpos, len, meta_ac,
5486 &dealloc);
5487 if (ret) {
5488 mlog_errno(ret);
5489 goto out_commit;
5490 }
5491
5492 le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
5493 ocfs2_journal_dirty(handle, root_bh);
5494
5495 ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
5496 if (ret)
5497 mlog_errno(ret);
5498 ocfs2_update_inode_fsync_trans(handle, inode, 0);
5499
5500 out_commit:
5501 ocfs2_commit_trans(osb, handle);
5502 out:
5503 ocfs2_schedule_truncate_log_flush(osb, 1);
5504
5505 inode_unlock(tl_inode);
5506
5507 if (meta_ac)
5508 ocfs2_free_alloc_context(meta_ac);
5509
5510 ocfs2_run_deallocs(osb, &dealloc);
5511
5512 return ret;
5513 }
5514
5515 /*
5516 * check whether the xattr bucket is filled up with the same hash value.
5517 * If we want to insert the xattr with the same hash, return -ENOSPC.
5518 * If we want to insert a xattr with different hash value, go ahead
5519 * and ocfs2_divide_xattr_bucket will handle this.
5520 */
5521 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
5522 struct ocfs2_xattr_bucket *bucket,
5523 const char *name)
5524 {
5525 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5526 u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5527
5528 if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5529 return 0;
5530
5531 if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5532 xh->xh_entries[0].xe_name_hash) {
5533 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5534 "hash = %u\n",
5535 (unsigned long long)bucket_blkno(bucket),
5536 le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5537 return -ENOSPC;
5538 }
5539
5540 return 0;
5541 }
5542
5543 /*
5544 * Try to set the entry in the current bucket. If we fail, the caller
5545 * will handle getting us another bucket.
5546 */
5547 static int ocfs2_xattr_set_entry_bucket(struct inode *inode,
5548 struct ocfs2_xattr_info *xi,
5549 struct ocfs2_xattr_search *xs,
5550 struct ocfs2_xattr_set_ctxt *ctxt)
5551 {
5552 int ret;
5553 struct ocfs2_xa_loc loc;
5554
5555 trace_ocfs2_xattr_set_entry_bucket(xi->xi_name);
5556
5557 ocfs2_init_xattr_bucket_xa_loc(&loc, xs->bucket,
5558 xs->not_found ? NULL : xs->here);
5559 ret = ocfs2_xa_set(&loc, xi, ctxt);
5560 if (!ret) {
5561 xs->here = loc.xl_entry;
5562 goto out;
5563 }
5564 if (ret != -ENOSPC) {
5565 mlog_errno(ret);
5566 goto out;
5567 }
5568
5569 /* Ok, we need space. Let's try defragmenting the bucket. */
5570 ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5571 xs->bucket);
5572 if (ret) {
5573 mlog_errno(ret);
5574 goto out;
5575 }
5576
5577 ret = ocfs2_xa_set(&loc, xi, ctxt);
5578 if (!ret) {
5579 xs->here = loc.xl_entry;
5580 goto out;
5581 }
5582 if (ret != -ENOSPC)
5583 mlog_errno(ret);
5584
5585
5586 out:
5587 return ret;
5588 }
5589
5590 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5591 struct ocfs2_xattr_info *xi,
5592 struct ocfs2_xattr_search *xs,
5593 struct ocfs2_xattr_set_ctxt *ctxt)
5594 {
5595 int ret;
5596
5597 trace_ocfs2_xattr_set_entry_index_block(xi->xi_name);
5598
5599 ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5600 if (!ret)
5601 goto out;
5602 if (ret != -ENOSPC) {
5603 mlog_errno(ret);
5604 goto out;
5605 }
5606
5607 /* Ack, need more space. Let's try to get another bucket! */
5608
5609 /*
5610 * We do not allow for overlapping ranges between buckets. And
5611 * the maximum number of collisions we will allow for then is
5612 * one bucket's worth, so check it here whether we need to
5613 * add a new bucket for the insert.
5614 */
5615 ret = ocfs2_check_xattr_bucket_collision(inode,
5616 xs->bucket,
5617 xi->xi_name);
5618 if (ret) {
5619 mlog_errno(ret);
5620 goto out;
5621 }
5622
5623 ret = ocfs2_add_new_xattr_bucket(inode,
5624 xs->xattr_bh,
5625 xs->bucket,
5626 ctxt);
5627 if (ret) {
5628 mlog_errno(ret);
5629 goto out;
5630 }
5631
5632 /*
5633 * ocfs2_add_new_xattr_bucket() will have updated
5634 * xs->bucket if it moved, but it will not have updated
5635 * any of the other search fields. Thus, we drop it and
5636 * re-search. Everything should be cached, so it'll be
5637 * quick.
5638 */
5639 ocfs2_xattr_bucket_relse(xs->bucket);
5640 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5641 xi->xi_name_index,
5642 xi->xi_name, xs);
5643 if (ret && ret != -ENODATA)
5644 goto out;
5645 xs->not_found = ret;
5646
5647 /* Ok, we have a new bucket, let's try again */
5648 ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5649 if (ret && (ret != -ENOSPC))
5650 mlog_errno(ret);
5651
5652 out:
5653 return ret;
5654 }
5655
5656 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5657 struct ocfs2_xattr_bucket *bucket,
5658 void *para)
5659 {
5660 int ret = 0, ref_credits;
5661 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5662 u16 i;
5663 struct ocfs2_xattr_entry *xe;
5664 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5665 struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5666 int credits = ocfs2_remove_extent_credits(osb->sb) +
5667 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5668 struct ocfs2_xattr_value_root *xv;
5669 struct ocfs2_rm_xattr_bucket_para *args =
5670 (struct ocfs2_rm_xattr_bucket_para *)para;
5671
5672 ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5673
5674 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5675 xe = &xh->xh_entries[i];
5676 if (ocfs2_xattr_is_local(xe))
5677 continue;
5678
5679 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket,
5680 i, &xv, NULL);
5681 if (ret) {
5682 mlog_errno(ret);
5683 break;
5684 }
5685
5686 ret = ocfs2_lock_xattr_remove_allocators(inode, xv,
5687 args->ref_ci,
5688 args->ref_root_bh,
5689 &ctxt.meta_ac,
5690 &ref_credits);
5691
5692 ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
5693 if (IS_ERR(ctxt.handle)) {
5694 ret = PTR_ERR(ctxt.handle);
5695 mlog_errno(ret);
5696 break;
5697 }
5698
5699 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5700 i, 0, &ctxt);
5701
5702 ocfs2_commit_trans(osb, ctxt.handle);
5703 if (ctxt.meta_ac) {
5704 ocfs2_free_alloc_context(ctxt.meta_ac);
5705 ctxt.meta_ac = NULL;
5706 }
5707 if (ret) {
5708 mlog_errno(ret);
5709 break;
5710 }
5711 }
5712
5713 if (ctxt.meta_ac)
5714 ocfs2_free_alloc_context(ctxt.meta_ac);
5715 ocfs2_schedule_truncate_log_flush(osb, 1);
5716 ocfs2_run_deallocs(osb, &ctxt.dealloc);
5717 return ret;
5718 }
5719
5720 /*
5721 * Whenever we modify a xattr value root in the bucket(e.g, CoW
5722 * or change the extent record flag), we need to recalculate
5723 * the metaecc for the whole bucket. So it is done here.
5724 *
5725 * Note:
5726 * We have to give the extra credits for the caller.
5727 */
5728 static int ocfs2_xattr_bucket_post_refcount(struct inode *inode,
5729 handle_t *handle,
5730 void *para)
5731 {
5732 int ret;
5733 struct ocfs2_xattr_bucket *bucket =
5734 (struct ocfs2_xattr_bucket *)para;
5735
5736 ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
5737 OCFS2_JOURNAL_ACCESS_WRITE);
5738 if (ret) {
5739 mlog_errno(ret);
5740 return ret;
5741 }
5742
5743 ocfs2_xattr_bucket_journal_dirty(handle, bucket);
5744
5745 return 0;
5746 }
5747
5748 /*
5749 * Special action we need if the xattr value is refcounted.
5750 *
5751 * 1. If the xattr is refcounted, lock the tree.
5752 * 2. CoW the xattr if we are setting the new value and the value
5753 * will be stored outside.
5754 * 3. In other case, decrease_refcount will work for us, so just
5755 * lock the refcount tree, calculate the meta and credits is OK.
5756 *
5757 * We have to do CoW before ocfs2_init_xattr_set_ctxt since
5758 * currently CoW is a completed transaction, while this function
5759 * will also lock the allocators and let us deadlock. So we will
5760 * CoW the whole xattr value.
5761 */
5762 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
5763 struct ocfs2_dinode *di,
5764 struct ocfs2_xattr_info *xi,
5765 struct ocfs2_xattr_search *xis,
5766 struct ocfs2_xattr_search *xbs,
5767 struct ocfs2_refcount_tree **ref_tree,
5768 int *meta_add,
5769 int *credits)
5770 {
5771 int ret = 0;
5772 struct ocfs2_xattr_block *xb;
5773 struct ocfs2_xattr_entry *xe;
5774 char *base;
5775 u32 p_cluster, num_clusters;
5776 unsigned int ext_flags;
5777 int name_offset, name_len;
5778 struct ocfs2_xattr_value_buf vb;
5779 struct ocfs2_xattr_bucket *bucket = NULL;
5780 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5781 struct ocfs2_post_refcount refcount;
5782 struct ocfs2_post_refcount *p = NULL;
5783 struct buffer_head *ref_root_bh = NULL;
5784
5785 if (!xis->not_found) {
5786 xe = xis->here;
5787 name_offset = le16_to_cpu(xe->xe_name_offset);
5788 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5789 base = xis->base;
5790 vb.vb_bh = xis->inode_bh;
5791 vb.vb_access = ocfs2_journal_access_di;
5792 } else {
5793 int i, block_off = 0;
5794 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
5795 xe = xbs->here;
5796 name_offset = le16_to_cpu(xe->xe_name_offset);
5797 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5798 i = xbs->here - xbs->header->xh_entries;
5799
5800 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
5801 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
5802 bucket_xh(xbs->bucket),
5803 i, &block_off,
5804 &name_offset);
5805 if (ret) {
5806 mlog_errno(ret);
5807 goto out;
5808 }
5809 base = bucket_block(xbs->bucket, block_off);
5810 vb.vb_bh = xbs->bucket->bu_bhs[block_off];
5811 vb.vb_access = ocfs2_journal_access;
5812
5813 if (ocfs2_meta_ecc(osb)) {
5814 /*create parameters for ocfs2_post_refcount. */
5815 bucket = xbs->bucket;
5816 refcount.credits = bucket->bu_blocks;
5817 refcount.para = bucket;
5818 refcount.func =
5819 ocfs2_xattr_bucket_post_refcount;
5820 p = &refcount;
5821 }
5822 } else {
5823 base = xbs->base;
5824 vb.vb_bh = xbs->xattr_bh;
5825 vb.vb_access = ocfs2_journal_access_xb;
5826 }
5827 }
5828
5829 if (ocfs2_xattr_is_local(xe))
5830 goto out;
5831
5832 vb.vb_xv = (struct ocfs2_xattr_value_root *)
5833 (base + name_offset + name_len);
5834
5835 ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
5836 &num_clusters, &vb.vb_xv->xr_list,
5837 &ext_flags);
5838 if (ret) {
5839 mlog_errno(ret);
5840 goto out;
5841 }
5842
5843 /*
5844 * We just need to check the 1st extent record, since we always
5845 * CoW the whole xattr. So there shouldn't be a xattr with
5846 * some REFCOUNT extent recs after the 1st one.
5847 */
5848 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
5849 goto out;
5850
5851 ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
5852 1, ref_tree, &ref_root_bh);
5853 if (ret) {
5854 mlog_errno(ret);
5855 goto out;
5856 }
5857
5858 /*
5859 * If we are deleting the xattr or the new size will be stored inside,
5860 * cool, leave it there, the xattr truncate process will remove them
5861 * for us(it still needs the refcount tree lock and the meta, credits).
5862 * And the worse case is that every cluster truncate will split the
5863 * refcount tree, and make the original extent become 3. So we will need
5864 * 2 * cluster more extent recs at most.
5865 */
5866 if (!xi->xi_value || xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE) {
5867
5868 ret = ocfs2_refcounted_xattr_delete_need(inode,
5869 &(*ref_tree)->rf_ci,
5870 ref_root_bh, vb.vb_xv,
5871 meta_add, credits);
5872 if (ret)
5873 mlog_errno(ret);
5874 goto out;
5875 }
5876
5877 ret = ocfs2_refcount_cow_xattr(inode, di, &vb,
5878 *ref_tree, ref_root_bh, 0,
5879 le32_to_cpu(vb.vb_xv->xr_clusters), p);
5880 if (ret)
5881 mlog_errno(ret);
5882
5883 out:
5884 brelse(ref_root_bh);
5885 return ret;
5886 }
5887
5888 /*
5889 * Add the REFCOUNTED flags for all the extent rec in ocfs2_xattr_value_root.
5890 * The physical clusters will be added to refcount tree.
5891 */
5892 static int ocfs2_xattr_value_attach_refcount(struct inode *inode,
5893 struct ocfs2_xattr_value_root *xv,
5894 struct ocfs2_extent_tree *value_et,
5895 struct ocfs2_caching_info *ref_ci,
5896 struct buffer_head *ref_root_bh,
5897 struct ocfs2_cached_dealloc_ctxt *dealloc,
5898 struct ocfs2_post_refcount *refcount)
5899 {
5900 int ret = 0;
5901 u32 clusters = le32_to_cpu(xv->xr_clusters);
5902 u32 cpos, p_cluster, num_clusters;
5903 struct ocfs2_extent_list *el = &xv->xr_list;
5904 unsigned int ext_flags;
5905
5906 cpos = 0;
5907 while (cpos < clusters) {
5908 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
5909 &num_clusters, el, &ext_flags);
5910 if (ret) {
5911 mlog_errno(ret);
5912 break;
5913 }
5914
5915 cpos += num_clusters;
5916 if ((ext_flags & OCFS2_EXT_REFCOUNTED))
5917 continue;
5918
5919 BUG_ON(!p_cluster);
5920
5921 ret = ocfs2_add_refcount_flag(inode, value_et,
5922 ref_ci, ref_root_bh,
5923 cpos - num_clusters,
5924 p_cluster, num_clusters,
5925 dealloc, refcount);
5926 if (ret) {
5927 mlog_errno(ret);
5928 break;
5929 }
5930 }
5931
5932 return ret;
5933 }
5934
5935 /*
5936 * Given a normal ocfs2_xattr_header, refcount all the entries which
5937 * have value stored outside.
5938 * Used for xattrs stored in inode and ocfs2_xattr_block.
5939 */
5940 static int ocfs2_xattr_attach_refcount_normal(struct inode *inode,
5941 struct ocfs2_xattr_value_buf *vb,
5942 struct ocfs2_xattr_header *header,
5943 struct ocfs2_caching_info *ref_ci,
5944 struct buffer_head *ref_root_bh,
5945 struct ocfs2_cached_dealloc_ctxt *dealloc)
5946 {
5947
5948 struct ocfs2_xattr_entry *xe;
5949 struct ocfs2_xattr_value_root *xv;
5950 struct ocfs2_extent_tree et;
5951 int i, ret = 0;
5952
5953 for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
5954 xe = &header->xh_entries[i];
5955
5956 if (ocfs2_xattr_is_local(xe))
5957 continue;
5958
5959 xv = (struct ocfs2_xattr_value_root *)((void *)header +
5960 le16_to_cpu(xe->xe_name_offset) +
5961 OCFS2_XATTR_SIZE(xe->xe_name_len));
5962
5963 vb->vb_xv = xv;
5964 ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
5965
5966 ret = ocfs2_xattr_value_attach_refcount(inode, xv, &et,
5967 ref_ci, ref_root_bh,
5968 dealloc, NULL);
5969 if (ret) {
5970 mlog_errno(ret);
5971 break;
5972 }
5973 }
5974
5975 return ret;
5976 }
5977
5978 static int ocfs2_xattr_inline_attach_refcount(struct inode *inode,
5979 struct buffer_head *fe_bh,
5980 struct ocfs2_caching_info *ref_ci,
5981 struct buffer_head *ref_root_bh,
5982 struct ocfs2_cached_dealloc_ctxt *dealloc)
5983 {
5984 struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
5985 struct ocfs2_xattr_header *header = (struct ocfs2_xattr_header *)
5986 (fe_bh->b_data + inode->i_sb->s_blocksize -
5987 le16_to_cpu(di->i_xattr_inline_size));
5988 struct ocfs2_xattr_value_buf vb = {
5989 .vb_bh = fe_bh,
5990 .vb_access = ocfs2_journal_access_di,
5991 };
5992
5993 return ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
5994 ref_ci, ref_root_bh, dealloc);
5995 }
5996
5997 struct ocfs2_xattr_tree_value_refcount_para {
5998 struct ocfs2_caching_info *ref_ci;
5999 struct buffer_head *ref_root_bh;
6000 struct ocfs2_cached_dealloc_ctxt *dealloc;
6001 };
6002
6003 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
6004 struct ocfs2_xattr_bucket *bucket,
6005 int offset,
6006 struct ocfs2_xattr_value_root **xv,
6007 struct buffer_head **bh)
6008 {
6009 int ret, block_off, name_offset;
6010 struct ocfs2_xattr_header *xh = bucket_xh(bucket);
6011 struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
6012 void *base;
6013
6014 ret = ocfs2_xattr_bucket_get_name_value(sb,
6015 bucket_xh(bucket),
6016 offset,
6017 &block_off,
6018 &name_offset);
6019 if (ret) {
6020 mlog_errno(ret);
6021 goto out;
6022 }
6023
6024 base = bucket_block(bucket, block_off);
6025
6026 *xv = (struct ocfs2_xattr_value_root *)(base + name_offset +
6027 OCFS2_XATTR_SIZE(xe->xe_name_len));
6028
6029 if (bh)
6030 *bh = bucket->bu_bhs[block_off];
6031 out:
6032 return ret;
6033 }
6034
6035 /*
6036 * For a given xattr bucket, refcount all the entries which
6037 * have value stored outside.
6038 */
6039 static int ocfs2_xattr_bucket_value_refcount(struct inode *inode,
6040 struct ocfs2_xattr_bucket *bucket,
6041 void *para)
6042 {
6043 int i, ret = 0;
6044 struct ocfs2_extent_tree et;
6045 struct ocfs2_xattr_tree_value_refcount_para *ref =
6046 (struct ocfs2_xattr_tree_value_refcount_para *)para;
6047 struct ocfs2_xattr_header *xh =
6048 (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6049 struct ocfs2_xattr_entry *xe;
6050 struct ocfs2_xattr_value_buf vb = {
6051 .vb_access = ocfs2_journal_access,
6052 };
6053 struct ocfs2_post_refcount refcount = {
6054 .credits = bucket->bu_blocks,
6055 .para = bucket,
6056 .func = ocfs2_xattr_bucket_post_refcount,
6057 };
6058 struct ocfs2_post_refcount *p = NULL;
6059
6060 /* We only need post_refcount if we support metaecc. */
6061 if (ocfs2_meta_ecc(OCFS2_SB(inode->i_sb)))
6062 p = &refcount;
6063
6064 trace_ocfs2_xattr_bucket_value_refcount(
6065 (unsigned long long)bucket_blkno(bucket),
6066 le16_to_cpu(xh->xh_count));
6067 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6068 xe = &xh->xh_entries[i];
6069
6070 if (ocfs2_xattr_is_local(xe))
6071 continue;
6072
6073 ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket, i,
6074 &vb.vb_xv, &vb.vb_bh);
6075 if (ret) {
6076 mlog_errno(ret);
6077 break;
6078 }
6079
6080 ocfs2_init_xattr_value_extent_tree(&et,
6081 INODE_CACHE(inode), &vb);
6082
6083 ret = ocfs2_xattr_value_attach_refcount(inode, vb.vb_xv,
6084 &et, ref->ref_ci,
6085 ref->ref_root_bh,
6086 ref->dealloc, p);
6087 if (ret) {
6088 mlog_errno(ret);
6089 break;
6090 }
6091 }
6092
6093 return ret;
6094
6095 }
6096
6097 static int ocfs2_refcount_xattr_tree_rec(struct inode *inode,
6098 struct buffer_head *root_bh,
6099 u64 blkno, u32 cpos, u32 len, void *para)
6100 {
6101 return ocfs2_iterate_xattr_buckets(inode, blkno, len,
6102 ocfs2_xattr_bucket_value_refcount,
6103 para);
6104 }
6105
6106 static int ocfs2_xattr_block_attach_refcount(struct inode *inode,
6107 struct buffer_head *blk_bh,
6108 struct ocfs2_caching_info *ref_ci,
6109 struct buffer_head *ref_root_bh,
6110 struct ocfs2_cached_dealloc_ctxt *dealloc)
6111 {
6112 int ret = 0;
6113 struct ocfs2_xattr_block *xb =
6114 (struct ocfs2_xattr_block *)blk_bh->b_data;
6115
6116 if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
6117 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
6118 struct ocfs2_xattr_value_buf vb = {
6119 .vb_bh = blk_bh,
6120 .vb_access = ocfs2_journal_access_xb,
6121 };
6122
6123 ret = ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
6124 ref_ci, ref_root_bh,
6125 dealloc);
6126 } else {
6127 struct ocfs2_xattr_tree_value_refcount_para para = {
6128 .ref_ci = ref_ci,
6129 .ref_root_bh = ref_root_bh,
6130 .dealloc = dealloc,
6131 };
6132
6133 ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
6134 ocfs2_refcount_xattr_tree_rec,
6135 ¶);
6136 }
6137
6138 return ret;
6139 }
6140
6141 int ocfs2_xattr_attach_refcount_tree(struct inode *inode,
6142 struct buffer_head *fe_bh,
6143 struct ocfs2_caching_info *ref_ci,
6144 struct buffer_head *ref_root_bh,
6145 struct ocfs2_cached_dealloc_ctxt *dealloc)
6146 {
6147 int ret = 0;
6148 struct ocfs2_inode_info *oi = OCFS2_I(inode);
6149 struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
6150 struct buffer_head *blk_bh = NULL;
6151
6152 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
6153 ret = ocfs2_xattr_inline_attach_refcount(inode, fe_bh,
6154 ref_ci, ref_root_bh,
6155 dealloc);
6156 if (ret) {
6157 mlog_errno(ret);
6158 goto out;
6159 }
6160 }
6161
6162 if (!di->i_xattr_loc)
6163 goto out;
6164
6165 ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
6166 &blk_bh);
6167 if (ret < 0) {
6168 mlog_errno(ret);
6169 goto out;
6170 }
6171
6172 ret = ocfs2_xattr_block_attach_refcount(inode, blk_bh, ref_ci,
6173 ref_root_bh, dealloc);
6174 if (ret)
6175 mlog_errno(ret);
6176
6177 brelse(blk_bh);
6178 out:
6179
6180 return ret;
6181 }
6182
6183 typedef int (should_xattr_reflinked)(struct ocfs2_xattr_entry *xe);
6184 /*
6185 * Store the information we need in xattr reflink.
6186 * old_bh and new_bh are inode bh for the old and new inode.
6187 */
6188 struct ocfs2_xattr_reflink {
6189 struct inode *old_inode;
6190 struct inode *new_inode;
6191 struct buffer_head *old_bh;
6192 struct buffer_head *new_bh;
6193 struct ocfs2_caching_info *ref_ci;
6194 struct buffer_head *ref_root_bh;
6195 struct ocfs2_cached_dealloc_ctxt *dealloc;
6196 should_xattr_reflinked *xattr_reflinked;
6197 };
6198
6199 /*
6200 * Given a xattr header and xe offset,
6201 * return the proper xv and the corresponding bh.
6202 * xattr in inode, block and xattr tree have different implementaions.
6203 */
6204 typedef int (get_xattr_value_root)(struct super_block *sb,
6205 struct buffer_head *bh,
6206 struct ocfs2_xattr_header *xh,
6207 int offset,
6208 struct ocfs2_xattr_value_root **xv,
6209 struct buffer_head **ret_bh,
6210 void *para);
6211
6212 /*
6213 * Calculate all the xattr value root metadata stored in this xattr header and
6214 * credits we need if we create them from the scratch.
6215 * We use get_xattr_value_root so that all types of xattr container can use it.
6216 */
6217 static int ocfs2_value_metas_in_xattr_header(struct super_block *sb,
6218 struct buffer_head *bh,
6219 struct ocfs2_xattr_header *xh,
6220 int *metas, int *credits,
6221 int *num_recs,
6222 get_xattr_value_root *func,
6223 void *para)
6224 {
6225 int i, ret = 0;
6226 struct ocfs2_xattr_value_root *xv;
6227 struct ocfs2_xattr_entry *xe;
6228
6229 for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6230 xe = &xh->xh_entries[i];
6231 if (ocfs2_xattr_is_local(xe))
6232 continue;
6233
6234 ret = func(sb, bh, xh, i, &xv, NULL, para);
6235 if (ret) {
6236 mlog_errno(ret);
6237 break;
6238 }
6239
6240 *metas += le16_to_cpu(xv->xr_list.l_tree_depth) *
6241 le16_to_cpu(xv->xr_list.l_next_free_rec);
6242
6243 *credits += ocfs2_calc_extend_credits(sb,
6244 &def_xv.xv.xr_list);
6245
6246 /*
6247 * If the value is a tree with depth > 1, We don't go deep
6248 * to the extent block, so just calculate a maximum record num.
6249 */
6250 if (!xv->xr_list.l_tree_depth)
6251 *num_recs += le16_to_cpu(xv->xr_list.l_next_free_rec);
6252 else
6253 *num_recs += ocfs2_clusters_for_bytes(sb,
6254 XATTR_SIZE_MAX);
6255 }
6256
6257 return ret;
6258 }
6259
6260 /* Used by xattr inode and block to return the right xv and buffer_head. */
6261 static int ocfs2_get_xattr_value_root(struct super_block *sb,
6262 struct buffer_head *bh,
6263 struct ocfs2_xattr_header *xh,
6264 int offset,
6265 struct ocfs2_xattr_value_root **xv,
6266 struct buffer_head **ret_bh,
6267 void *para)
6268 {
6269 struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
6270
6271 *xv = (struct ocfs2_xattr_value_root *)((void *)xh +
6272 le16_to_cpu(xe->xe_name_offset) +
6273 OCFS2_XATTR_SIZE(xe->xe_name_len));
6274
6275 if (ret_bh)
6276 *ret_bh = bh;
6277
6278 return 0;
6279 }
6280
6281 /*
6282 * Lock the meta_ac and caculate how much credits we need for reflink xattrs.
6283 * It is only used for inline xattr and xattr block.
6284 */
6285 static int ocfs2_reflink_lock_xattr_allocators(struct ocfs2_super *osb,
6286 struct ocfs2_xattr_header *xh,
6287 struct buffer_head *ref_root_bh,
6288 int *credits,
6289 struct ocfs2_alloc_context **meta_ac)
6290 {
6291 int ret, meta_add = 0, num_recs = 0;
6292 struct ocfs2_refcount_block *rb =
6293 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
6294
6295 *credits = 0;
6296
6297 ret = ocfs2_value_metas_in_xattr_header(osb->sb, NULL, xh,
6298 &meta_add, credits, &num_recs,
6299 ocfs2_get_xattr_value_root,
6300 NULL);
6301 if (ret) {
6302 mlog_errno(ret);
6303 goto out;
6304 }
6305
6306 /*
6307 * We need to add/modify num_recs in refcount tree, so just calculate
6308 * an approximate number we need for refcount tree change.
6309 * Sometimes we need to split the tree, and after split, half recs
6310 * will be moved to the new block, and a new block can only provide
6311 * half number of recs. So we multiple new blocks by 2.
6312 */
6313 num_recs = num_recs / ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6314 meta_add += num_recs;
6315 *credits += num_recs + num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6316 if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6317 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6318 le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6319 else
6320 *credits += 1;
6321
6322 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add, meta_ac);
6323 if (ret)
6324 mlog_errno(ret);
6325
6326 out:
6327 return ret;
6328 }
6329
6330 /*
6331 * Given a xattr header, reflink all the xattrs in this container.
6332 * It can be used for inode, block and bucket.
6333 *
6334 * NOTE:
6335 * Before we call this function, the caller has memcpy the xattr in
6336 * old_xh to the new_xh.
6337 *
6338 * If args.xattr_reflinked is set, call it to decide whether the xe should
6339 * be reflinked or not. If not, remove it from the new xattr header.
6340 */
6341 static int ocfs2_reflink_xattr_header(handle_t *handle,
6342 struct ocfs2_xattr_reflink *args,
6343 struct buffer_head *old_bh,
6344 struct ocfs2_xattr_header *xh,
6345 struct buffer_head *new_bh,
6346 struct ocfs2_xattr_header *new_xh,
6347 struct ocfs2_xattr_value_buf *vb,
6348 struct ocfs2_alloc_context *meta_ac,
6349 get_xattr_value_root *func,
6350 void *para)
6351 {
6352 int ret = 0, i, j;
6353 struct super_block *sb = args->old_inode->i_sb;
6354 struct buffer_head *value_bh;
6355 struct ocfs2_xattr_entry *xe, *last;
6356 struct ocfs2_xattr_value_root *xv, *new_xv;
6357 struct ocfs2_extent_tree data_et;
6358 u32 clusters, cpos, p_cluster, num_clusters;
6359 unsigned int ext_flags = 0;
6360
6361 trace_ocfs2_reflink_xattr_header((unsigned long long)old_bh->b_blocknr,
6362 le16_to_cpu(xh->xh_count));
6363
6364 last = &new_xh->xh_entries[le16_to_cpu(new_xh->xh_count)];
6365 for (i = 0, j = 0; i < le16_to_cpu(xh->xh_count); i++, j++) {
6366 xe = &xh->xh_entries[i];
6367
6368 if (args->xattr_reflinked && !args->xattr_reflinked(xe)) {
6369 xe = &new_xh->xh_entries[j];
6370
6371 le16_add_cpu(&new_xh->xh_count, -1);
6372 if (new_xh->xh_count) {
6373 memmove(xe, xe + 1,
6374 (void *)last - (void *)xe);
6375 memset(last, 0,
6376 sizeof(struct ocfs2_xattr_entry));
6377 }
6378
6379 /*
6380 * We don't want j to increase in the next round since
6381 * it is already moved ahead.
6382 */
6383 j--;
6384 continue;
6385 }
6386
6387 if (ocfs2_xattr_is_local(xe))
6388 continue;
6389
6390 ret = func(sb, old_bh, xh, i, &xv, NULL, para);
6391 if (ret) {
6392 mlog_errno(ret);
6393 break;
6394 }
6395
6396 ret = func(sb, new_bh, new_xh, j, &new_xv, &value_bh, para);
6397 if (ret) {
6398 mlog_errno(ret);
6399 break;
6400 }
6401
6402 /*
6403 * For the xattr which has l_tree_depth = 0, all the extent
6404 * recs have already be copied to the new xh with the
6405 * propriate OCFS2_EXT_REFCOUNTED flag we just need to
6406 * increase the refount count int the refcount tree.
6407 *
6408 * For the xattr which has l_tree_depth > 0, we need
6409 * to initialize it to the empty default value root,
6410 * and then insert the extents one by one.
6411 */
6412 if (xv->xr_list.l_tree_depth) {
6413 memcpy(new_xv, &def_xv, OCFS2_XATTR_ROOT_SIZE);
6414 vb->vb_xv = new_xv;
6415 vb->vb_bh = value_bh;
6416 ocfs2_init_xattr_value_extent_tree(&data_et,
6417 INODE_CACHE(args->new_inode), vb);
6418 }
6419
6420 clusters = le32_to_cpu(xv->xr_clusters);
6421 cpos = 0;
6422 while (cpos < clusters) {
6423 ret = ocfs2_xattr_get_clusters(args->old_inode,
6424 cpos,
6425 &p_cluster,
6426 &num_clusters,
6427 &xv->xr_list,
6428 &ext_flags);
6429 if (ret) {
6430 mlog_errno(ret);
6431 goto out;
6432 }
6433
6434 BUG_ON(!p_cluster);
6435
6436 if (xv->xr_list.l_tree_depth) {
6437 ret = ocfs2_insert_extent(handle,
6438 &data_et, cpos,
6439 ocfs2_clusters_to_blocks(
6440 args->old_inode->i_sb,
6441 p_cluster),
6442 num_clusters, ext_flags,
6443 meta_ac);
6444 if (ret) {
6445 mlog_errno(ret);
6446 goto out;
6447 }
6448 }
6449
6450 ret = ocfs2_increase_refcount(handle, args->ref_ci,
6451 args->ref_root_bh,
6452 p_cluster, num_clusters,
6453 meta_ac, args->dealloc);
6454 if (ret) {
6455 mlog_errno(ret);
6456 goto out;
6457 }
6458
6459 cpos += num_clusters;
6460 }
6461 }
6462
6463 out:
6464 return ret;
6465 }
6466
6467 static int ocfs2_reflink_xattr_inline(struct ocfs2_xattr_reflink *args)
6468 {
6469 int ret = 0, credits = 0;
6470 handle_t *handle;
6471 struct ocfs2_super *osb = OCFS2_SB(args->old_inode->i_sb);
6472 struct ocfs2_dinode *di = (struct ocfs2_dinode *)args->old_bh->b_data;
6473 int inline_size = le16_to_cpu(di->i_xattr_inline_size);
6474 int header_off = osb->sb->s_blocksize - inline_size;
6475 struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)
6476 (args->old_bh->b_data + header_off);
6477 struct ocfs2_xattr_header *new_xh = (struct ocfs2_xattr_header *)
6478 (args->new_bh->b_data + header_off);
6479 struct ocfs2_alloc_context *meta_ac = NULL;
6480 struct ocfs2_inode_info *new_oi;
6481 struct ocfs2_dinode *new_di;
6482 struct ocfs2_xattr_value_buf vb = {
6483 .vb_bh = args->new_bh,
6484 .vb_access = ocfs2_journal_access_di,
6485 };
6486
6487 ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6488 &credits, &meta_ac);
6489 if (ret) {
6490 mlog_errno(ret);
6491 goto out;
6492 }
6493
6494 handle = ocfs2_start_trans(osb, credits);
6495 if (IS_ERR(handle)) {
6496 ret = PTR_ERR(handle);
6497 mlog_errno(ret);
6498 goto out;
6499 }
6500
6501 ret = ocfs2_journal_access_di(handle, INODE_CACHE(args->new_inode),
6502 args->new_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6503 if (ret) {
6504 mlog_errno(ret);
6505 goto out_commit;
6506 }
6507
6508 memcpy(args->new_bh->b_data + header_off,
6509 args->old_bh->b_data + header_off, inline_size);
6510
6511 new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6512 new_di->i_xattr_inline_size = cpu_to_le16(inline_size);
6513
6514 ret = ocfs2_reflink_xattr_header(handle, args, args->old_bh, xh,
6515 args->new_bh, new_xh, &vb, meta_ac,
6516 ocfs2_get_xattr_value_root, NULL);
6517 if (ret) {
6518 mlog_errno(ret);
6519 goto out_commit;
6520 }
6521
6522 new_oi = OCFS2_I(args->new_inode);
6523
6524 spin_lock(&new_oi->ip_lock);
6525 new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL | OCFS2_INLINE_XATTR_FL;
6526 new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6527 spin_unlock(&new_oi->ip_lock);
6528
6529 ocfs2_journal_dirty(handle, args->new_bh);
6530
6531 out_commit:
6532 ocfs2_commit_trans(osb, handle);
6533
6534 out:
6535 if (meta_ac)
6536 ocfs2_free_alloc_context(meta_ac);
6537 return ret;
6538 }
6539
6540 static int ocfs2_create_empty_xattr_block(struct inode *inode,
6541 struct buffer_head *fe_bh,
6542 struct buffer_head **ret_bh,
6543 int indexed)
6544 {
6545 int ret;
6546 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6547 struct ocfs2_xattr_set_ctxt ctxt;
6548
6549 memset(&ctxt, 0, sizeof(ctxt));
6550 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &ctxt.meta_ac);
6551 if (ret < 0) {
6552 mlog_errno(ret);
6553 return ret;
6554 }
6555
6556 ctxt.handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_CREATE_CREDITS);
6557 if (IS_ERR(ctxt.handle)) {
6558 ret = PTR_ERR(ctxt.handle);
6559 mlog_errno(ret);
6560 goto out;
6561 }
6562
6563 trace_ocfs2_create_empty_xattr_block(
6564 (unsigned long long)fe_bh->b_blocknr, indexed);
6565 ret = ocfs2_create_xattr_block(inode, fe_bh, &ctxt, indexed,
6566 ret_bh);
6567 if (ret)
6568 mlog_errno(ret);
6569
6570 ocfs2_commit_trans(osb, ctxt.handle);
6571 out:
6572 ocfs2_free_alloc_context(ctxt.meta_ac);
6573 return ret;
6574 }
6575
6576 static int ocfs2_reflink_xattr_block(struct ocfs2_xattr_reflink *args,
6577 struct buffer_head *blk_bh,
6578 struct buffer_head *new_blk_bh)
6579 {
6580 int ret = 0, credits = 0;
6581 handle_t *handle;
6582 struct ocfs2_inode_info *new_oi = OCFS2_I(args->new_inode);
6583 struct ocfs2_dinode *new_di;
6584 struct ocfs2_super *osb = OCFS2_SB(args->new_inode->i_sb);
6585 int header_off = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
6586 struct ocfs2_xattr_block *xb =
6587 (struct ocfs2_xattr_block *)blk_bh->b_data;
6588 struct ocfs2_xattr_header *xh = &xb->xb_attrs.xb_header;
6589 struct ocfs2_xattr_block *new_xb =
6590 (struct ocfs2_xattr_block *)new_blk_bh->b_data;
6591 struct ocfs2_xattr_header *new_xh = &new_xb->xb_attrs.xb_header;
6592 struct ocfs2_alloc_context *meta_ac;
6593 struct ocfs2_xattr_value_buf vb = {
6594 .vb_bh = new_blk_bh,
6595 .vb_access = ocfs2_journal_access_xb,
6596 };
6597
6598 ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6599 &credits, &meta_ac);
6600 if (ret) {
6601 mlog_errno(ret);
6602 return ret;
6603 }
6604
6605 /* One more credits in case we need to add xattr flags in new inode. */
6606 handle = ocfs2_start_trans(osb, credits + 1);
6607 if (IS_ERR(handle)) {
6608 ret = PTR_ERR(handle);
6609 mlog_errno(ret);
6610 goto out;
6611 }
6612
6613 if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6614 ret = ocfs2_journal_access_di(handle,
6615 INODE_CACHE(args->new_inode),
6616 args->new_bh,
6617 OCFS2_JOURNAL_ACCESS_WRITE);
6618 if (ret) {
6619 mlog_errno(ret);
6620 goto out_commit;
6621 }
6622 }
6623
6624 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(args->new_inode),
6625 new_blk_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6626 if (ret) {
6627 mlog_errno(ret);
6628 goto out_commit;
6629 }
6630
6631 memcpy(new_blk_bh->b_data + header_off, blk_bh->b_data + header_off,
6632 osb->sb->s_blocksize - header_off);
6633
6634 ret = ocfs2_reflink_xattr_header(handle, args, blk_bh, xh,
6635 new_blk_bh, new_xh, &vb, meta_ac,
6636 ocfs2_get_xattr_value_root, NULL);
6637 if (ret) {
6638 mlog_errno(ret);
6639 goto out_commit;
6640 }
6641
6642 ocfs2_journal_dirty(handle, new_blk_bh);
6643
6644 if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6645 new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6646 spin_lock(&new_oi->ip_lock);
6647 new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
6648 new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6649 spin_unlock(&new_oi->ip_lock);
6650
6651 ocfs2_journal_dirty(handle, args->new_bh);
6652 }
6653
6654 out_commit:
6655 ocfs2_commit_trans(osb, handle);
6656
6657 out:
6658 ocfs2_free_alloc_context(meta_ac);
6659 return ret;
6660 }
6661
6662 struct ocfs2_reflink_xattr_tree_args {
6663 struct ocfs2_xattr_reflink *reflink;
6664 struct buffer_head *old_blk_bh;
6665 struct buffer_head *new_blk_bh;
6666 struct ocfs2_xattr_bucket *old_bucket;
6667 struct ocfs2_xattr_bucket *new_bucket;
6668 };
6669
6670 /*
6671 * NOTE:
6672 * We have to handle the case that both old bucket and new bucket
6673 * will call this function to get the right ret_bh.
6674 * So The caller must give us the right bh.
6675 */
6676 static int ocfs2_get_reflink_xattr_value_root(struct super_block *sb,
6677 struct buffer_head *bh,
6678 struct ocfs2_xattr_header *xh,
6679 int offset,
6680 struct ocfs2_xattr_value_root **xv,
6681 struct buffer_head **ret_bh,
6682 void *para)
6683 {
6684 struct ocfs2_reflink_xattr_tree_args *args =
6685 (struct ocfs2_reflink_xattr_tree_args *)para;
6686 struct ocfs2_xattr_bucket *bucket;
6687
6688 if (bh == args->old_bucket->bu_bhs[0])
6689 bucket = args->old_bucket;
6690 else
6691 bucket = args->new_bucket;
6692
6693 return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6694 xv, ret_bh);
6695 }
6696
6697 struct ocfs2_value_tree_metas {
6698 int num_metas;
6699 int credits;
6700 int num_recs;
6701 };
6702
6703 static int ocfs2_value_tree_metas_in_bucket(struct super_block *sb,
6704 struct buffer_head *bh,
6705 struct ocfs2_xattr_header *xh,
6706 int offset,
6707 struct ocfs2_xattr_value_root **xv,
6708 struct buffer_head **ret_bh,
6709 void *para)
6710 {
6711 struct ocfs2_xattr_bucket *bucket =
6712 (struct ocfs2_xattr_bucket *)para;
6713
6714 return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6715 xv, ret_bh);
6716 }
6717
6718 static int ocfs2_calc_value_tree_metas(struct inode *inode,
6719 struct ocfs2_xattr_bucket *bucket,
6720 void *para)
6721 {
6722 struct ocfs2_value_tree_metas *metas =
6723 (struct ocfs2_value_tree_metas *)para;
6724 struct ocfs2_xattr_header *xh =
6725 (struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6726
6727 /* Add the credits for this bucket first. */
6728 metas->credits += bucket->bu_blocks;
6729 return ocfs2_value_metas_in_xattr_header(inode->i_sb, bucket->bu_bhs[0],
6730 xh, &metas->num_metas,
6731 &metas->credits, &metas->num_recs,
6732 ocfs2_value_tree_metas_in_bucket,
6733 bucket);
6734 }
6735
6736 /*
6737 * Given a xattr extent rec starting from blkno and having len clusters,
6738 * iterate all the buckets calculate how much metadata we need for reflinking
6739 * all the ocfs2_xattr_value_root and lock the allocators accordingly.
6740 */
6741 static int ocfs2_lock_reflink_xattr_rec_allocators(
6742 struct ocfs2_reflink_xattr_tree_args *args,
6743 struct ocfs2_extent_tree *xt_et,
6744 u64 blkno, u32 len, int *credits,
6745 struct ocfs2_alloc_context **meta_ac,
6746 struct ocfs2_alloc_context **data_ac)
6747 {
6748 int ret, num_free_extents;
6749 struct ocfs2_value_tree_metas metas;
6750 struct ocfs2_super *osb = OCFS2_SB(args->reflink->old_inode->i_sb);
6751 struct ocfs2_refcount_block *rb;
6752
6753 memset(&metas, 0, sizeof(metas));
6754
6755 ret = ocfs2_iterate_xattr_buckets(args->reflink->old_inode, blkno, len,
6756 ocfs2_calc_value_tree_metas, &metas);
6757 if (ret) {
6758 mlog_errno(ret);
6759 goto out;
6760 }
6761
6762 *credits = metas.credits;
6763
6764 /*
6765 * Calculate we need for refcount tree change.
6766 *
6767 * We need to add/modify num_recs in refcount tree, so just calculate
6768 * an approximate number we need for refcount tree change.
6769 * Sometimes we need to split the tree, and after split, half recs
6770 * will be moved to the new block, and a new block can only provide
6771 * half number of recs. So we multiple new blocks by 2.
6772 * In the end, we have to add credits for modifying the already
6773 * existed refcount block.
6774 */
6775 rb = (struct ocfs2_refcount_block *)args->reflink->ref_root_bh->b_data;
6776 metas.num_recs =
6777 (metas.num_recs + ocfs2_refcount_recs_per_rb(osb->sb) - 1) /
6778 ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6779 metas.num_metas += metas.num_recs;
6780 *credits += metas.num_recs +
6781 metas.num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6782 if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6783 *credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6784 le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6785 else
6786 *credits += 1;
6787
6788 /* count in the xattr tree change. */
6789 num_free_extents = ocfs2_num_free_extents(xt_et);
6790 if (num_free_extents < 0) {
6791 ret = num_free_extents;
6792 mlog_errno(ret);
6793 goto out;
6794 }
6795
6796 if (num_free_extents < len)
6797 metas.num_metas += ocfs2_extend_meta_needed(xt_et->et_root_el);
6798
6799 *credits += ocfs2_calc_extend_credits(osb->sb,
6800 xt_et->et_root_el);
6801
6802 if (metas.num_metas) {
6803 ret = ocfs2_reserve_new_metadata_blocks(osb, metas.num_metas,
6804 meta_ac);
6805 if (ret) {
6806 mlog_errno(ret);
6807 goto out;
6808 }
6809 }
6810
6811 if (len) {
6812 ret = ocfs2_reserve_clusters(osb, len, data_ac);
6813 if (ret)
6814 mlog_errno(ret);
6815 }
6816 out:
6817 if (ret) {
6818 if (*meta_ac) {
6819 ocfs2_free_alloc_context(*meta_ac);
6820 *meta_ac = NULL;
6821 }
6822 }
6823
6824 return ret;
6825 }
6826
6827 static int ocfs2_reflink_xattr_bucket(handle_t *handle,
6828 u64 blkno, u64 new_blkno, u32 clusters,
6829 u32 *cpos, int num_buckets,
6830 struct ocfs2_alloc_context *meta_ac,
6831 struct ocfs2_alloc_context *data_ac,
6832 struct ocfs2_reflink_xattr_tree_args *args)
6833 {
6834 int i, j, ret = 0;
6835 struct super_block *sb = args->reflink->old_inode->i_sb;
6836 int bpb = args->old_bucket->bu_blocks;
6837 struct ocfs2_xattr_value_buf vb = {
6838 .vb_access = ocfs2_journal_access,
6839 };
6840
6841 for (i = 0; i < num_buckets; i++, blkno += bpb, new_blkno += bpb) {
6842 ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6843 if (ret) {
6844 mlog_errno(ret);
6845 break;
6846 }
6847
6848 ret = ocfs2_init_xattr_bucket(args->new_bucket, new_blkno, 1);
6849 if (ret) {
6850 mlog_errno(ret);
6851 break;
6852 }
6853
6854 ret = ocfs2_xattr_bucket_journal_access(handle,
6855 args->new_bucket,
6856 OCFS2_JOURNAL_ACCESS_CREATE);
6857 if (ret) {
6858 mlog_errno(ret);
6859 break;
6860 }
6861
6862 for (j = 0; j < bpb; j++)
6863 memcpy(bucket_block(args->new_bucket, j),
6864 bucket_block(args->old_bucket, j),
6865 sb->s_blocksize);
6866
6867 /*
6868 * Record the start cpos so that we can use it to initialize
6869 * our xattr tree we also set the xh_num_bucket for the new
6870 * bucket.
6871 */
6872 if (i == 0) {
6873 *cpos = le32_to_cpu(bucket_xh(args->new_bucket)->
6874 xh_entries[0].xe_name_hash);
6875 bucket_xh(args->new_bucket)->xh_num_buckets =
6876 cpu_to_le16(num_buckets);
6877 }
6878
6879 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6880
6881 ret = ocfs2_reflink_xattr_header(handle, args->reflink,
6882 args->old_bucket->bu_bhs[0],
6883 bucket_xh(args->old_bucket),
6884 args->new_bucket->bu_bhs[0],
6885 bucket_xh(args->new_bucket),
6886 &vb, meta_ac,
6887 ocfs2_get_reflink_xattr_value_root,
6888 args);
6889 if (ret) {
6890 mlog_errno(ret);
6891 break;
6892 }
6893
6894 /*
6895 * Re-access and dirty the bucket to calculate metaecc.
6896 * Because we may extend the transaction in reflink_xattr_header
6897 * which will let the already accessed block gone.
6898 */
6899 ret = ocfs2_xattr_bucket_journal_access(handle,
6900 args->new_bucket,
6901 OCFS2_JOURNAL_ACCESS_WRITE);
6902 if (ret) {
6903 mlog_errno(ret);
6904 break;
6905 }
6906
6907 ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6908
6909 ocfs2_xattr_bucket_relse(args->old_bucket);
6910 ocfs2_xattr_bucket_relse(args->new_bucket);
6911 }
6912
6913 ocfs2_xattr_bucket_relse(args->old_bucket);
6914 ocfs2_xattr_bucket_relse(args->new_bucket);
6915 return ret;
6916 }
6917
6918 static int ocfs2_reflink_xattr_buckets(handle_t *handle,
6919 struct inode *inode,
6920 struct ocfs2_reflink_xattr_tree_args *args,
6921 struct ocfs2_extent_tree *et,
6922 struct ocfs2_alloc_context *meta_ac,
6923 struct ocfs2_alloc_context *data_ac,
6924 u64 blkno, u32 cpos, u32 len)
6925 {
6926 int ret, first_inserted = 0;
6927 u32 p_cluster, num_clusters, reflink_cpos = 0;
6928 u64 new_blkno;
6929 unsigned int num_buckets, reflink_buckets;
6930 unsigned int bpc =
6931 ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
6932
6933 ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6934 if (ret) {
6935 mlog_errno(ret);
6936 goto out;
6937 }
6938 num_buckets = le16_to_cpu(bucket_xh(args->old_bucket)->xh_num_buckets);
6939 ocfs2_xattr_bucket_relse(args->old_bucket);
6940
6941 while (len && num_buckets) {
6942 ret = ocfs2_claim_clusters(handle, data_ac,
6943 1, &p_cluster, &num_clusters);
6944 if (ret) {
6945 mlog_errno(ret);
6946 goto out;
6947 }
6948
6949 new_blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
6950 reflink_buckets = min(num_buckets, bpc * num_clusters);
6951
6952 ret = ocfs2_reflink_xattr_bucket(handle, blkno,
6953 new_blkno, num_clusters,
6954 &reflink_cpos, reflink_buckets,
6955 meta_ac, data_ac, args);
6956 if (ret) {
6957 mlog_errno(ret);
6958 goto out;
6959 }
6960
6961 /*
6962 * For the 1st allocated cluster, we make it use the same cpos
6963 * so that the xattr tree looks the same as the original one
6964 * in the most case.
6965 */
6966 if (!first_inserted) {
6967 reflink_cpos = cpos;
6968 first_inserted = 1;
6969 }
6970 ret = ocfs2_insert_extent(handle, et, reflink_cpos, new_blkno,
6971 num_clusters, 0, meta_ac);
6972 if (ret)
6973 mlog_errno(ret);
6974
6975 trace_ocfs2_reflink_xattr_buckets((unsigned long long)new_blkno,
6976 num_clusters, reflink_cpos);
6977
6978 len -= num_clusters;
6979 blkno += ocfs2_clusters_to_blocks(inode->i_sb, num_clusters);
6980 num_buckets -= reflink_buckets;
6981 }
6982 out:
6983 return ret;
6984 }
6985
6986 /*
6987 * Create the same xattr extent record in the new inode's xattr tree.
6988 */
6989 static int ocfs2_reflink_xattr_rec(struct inode *inode,
6990 struct buffer_head *root_bh,
6991 u64 blkno,
6992 u32 cpos,
6993 u32 len,
6994 void *para)
6995 {
6996 int ret, credits = 0;
6997 handle_t *handle;
6998 struct ocfs2_reflink_xattr_tree_args *args =
6999 (struct ocfs2_reflink_xattr_tree_args *)para;
7000 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7001 struct ocfs2_alloc_context *meta_ac = NULL;
7002 struct ocfs2_alloc_context *data_ac = NULL;
7003 struct ocfs2_extent_tree et;
7004
7005 trace_ocfs2_reflink_xattr_rec((unsigned long long)blkno, len);
7006
7007 ocfs2_init_xattr_tree_extent_tree(&et,
7008 INODE_CACHE(args->reflink->new_inode),
7009 args->new_blk_bh);
7010
7011 ret = ocfs2_lock_reflink_xattr_rec_allocators(args, &et, blkno,
7012 len, &credits,
7013 &meta_ac, &data_ac);
7014 if (ret) {
7015 mlog_errno(ret);
7016 goto out;
7017 }
7018
7019 handle = ocfs2_start_trans(osb, credits);
7020 if (IS_ERR(handle)) {
7021 ret = PTR_ERR(handle);
7022 mlog_errno(ret);
7023 goto out;
7024 }
7025
7026 ret = ocfs2_reflink_xattr_buckets(handle, inode, args, &et,
7027 meta_ac, data_ac,
7028 blkno, cpos, len);
7029 if (ret)
7030 mlog_errno(ret);
7031
7032 ocfs2_commit_trans(osb, handle);
7033
7034 out:
7035 if (meta_ac)
7036 ocfs2_free_alloc_context(meta_ac);
7037 if (data_ac)
7038 ocfs2_free_alloc_context(data_ac);
7039 return ret;
7040 }
7041
7042 /*
7043 * Create reflinked xattr buckets.
7044 * We will add bucket one by one, and refcount all the xattrs in the bucket
7045 * if they are stored outside.
7046 */
7047 static int ocfs2_reflink_xattr_tree(struct ocfs2_xattr_reflink *args,
7048 struct buffer_head *blk_bh,
7049 struct buffer_head *new_blk_bh)
7050 {
7051 int ret;
7052 struct ocfs2_reflink_xattr_tree_args para;
7053
7054 memset(¶, 0, sizeof(para));
7055 para.reflink = args;
7056 para.old_blk_bh = blk_bh;
7057 para.new_blk_bh = new_blk_bh;
7058
7059 para.old_bucket = ocfs2_xattr_bucket_new(args->old_inode);
7060 if (!para.old_bucket) {
7061 mlog_errno(-ENOMEM);
7062 return -ENOMEM;
7063 }
7064
7065 para.new_bucket = ocfs2_xattr_bucket_new(args->new_inode);
7066 if (!para.new_bucket) {
7067 ret = -ENOMEM;
7068 mlog_errno(ret);
7069 goto out;
7070 }
7071
7072 ret = ocfs2_iterate_xattr_index_block(args->old_inode, blk_bh,
7073 ocfs2_reflink_xattr_rec,
7074 ¶);
7075 if (ret)
7076 mlog_errno(ret);
7077
7078 out:
7079 ocfs2_xattr_bucket_free(para.old_bucket);
7080 ocfs2_xattr_bucket_free(para.new_bucket);
7081 return ret;
7082 }
7083
7084 static int ocfs2_reflink_xattr_in_block(struct ocfs2_xattr_reflink *args,
7085 struct buffer_head *blk_bh)
7086 {
7087 int ret, indexed = 0;
7088 struct buffer_head *new_blk_bh = NULL;
7089 struct ocfs2_xattr_block *xb =
7090 (struct ocfs2_xattr_block *)blk_bh->b_data;
7091
7092
7093 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)
7094 indexed = 1;
7095
7096 ret = ocfs2_create_empty_xattr_block(args->new_inode, args->new_bh,
7097 &new_blk_bh, indexed);
7098 if (ret) {
7099 mlog_errno(ret);
7100 goto out;
7101 }
7102
7103 if (!indexed)
7104 ret = ocfs2_reflink_xattr_block(args, blk_bh, new_blk_bh);
7105 else
7106 ret = ocfs2_reflink_xattr_tree(args, blk_bh, new_blk_bh);
7107 if (ret)
7108 mlog_errno(ret);
7109
7110 out:
7111 brelse(new_blk_bh);
7112 return ret;
7113 }
7114
7115 static int ocfs2_reflink_xattr_no_security(struct ocfs2_xattr_entry *xe)
7116 {
7117 int type = ocfs2_xattr_get_type(xe);
7118
7119 return type != OCFS2_XATTR_INDEX_SECURITY &&
7120 type != OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS &&
7121 type != OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT;
7122 }
7123
7124 int ocfs2_reflink_xattrs(struct inode *old_inode,
7125 struct buffer_head *old_bh,
7126 struct inode *new_inode,
7127 struct buffer_head *new_bh,
7128 bool preserve_security)
7129 {
7130 int ret;
7131 struct ocfs2_xattr_reflink args;
7132 struct ocfs2_inode_info *oi = OCFS2_I(old_inode);
7133 struct ocfs2_dinode *di = (struct ocfs2_dinode *)old_bh->b_data;
7134 struct buffer_head *blk_bh = NULL;
7135 struct ocfs2_cached_dealloc_ctxt dealloc;
7136 struct ocfs2_refcount_tree *ref_tree;
7137 struct buffer_head *ref_root_bh = NULL;
7138
7139 ret = ocfs2_lock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7140 le64_to_cpu(di->i_refcount_loc),
7141 1, &ref_tree, &ref_root_bh);
7142 if (ret) {
7143 mlog_errno(ret);
7144 goto out;
7145 }
7146
7147 ocfs2_init_dealloc_ctxt(&dealloc);
7148
7149 args.old_inode = old_inode;
7150 args.new_inode = new_inode;
7151 args.old_bh = old_bh;
7152 args.new_bh = new_bh;
7153 args.ref_ci = &ref_tree->rf_ci;
7154 args.ref_root_bh = ref_root_bh;
7155 args.dealloc = &dealloc;
7156 if (preserve_security)
7157 args.xattr_reflinked = NULL;
7158 else
7159 args.xattr_reflinked = ocfs2_reflink_xattr_no_security;
7160
7161 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
7162 ret = ocfs2_reflink_xattr_inline(&args);
7163 if (ret) {
7164 mlog_errno(ret);
7165 goto out_unlock;
7166 }
7167 }
7168
7169 if (!di->i_xattr_loc)
7170 goto out_unlock;
7171
7172 ret = ocfs2_read_xattr_block(old_inode, le64_to_cpu(di->i_xattr_loc),
7173 &blk_bh);
7174 if (ret < 0) {
7175 mlog_errno(ret);
7176 goto out_unlock;
7177 }
7178
7179 ret = ocfs2_reflink_xattr_in_block(&args, blk_bh);
7180 if (ret)
7181 mlog_errno(ret);
7182
7183 brelse(blk_bh);
7184
7185 out_unlock:
7186 ocfs2_unlock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7187 ref_tree, 1);
7188 brelse(ref_root_bh);
7189
7190 if (ocfs2_dealloc_has_cluster(&dealloc)) {
7191 ocfs2_schedule_truncate_log_flush(OCFS2_SB(old_inode->i_sb), 1);
7192 ocfs2_run_deallocs(OCFS2_SB(old_inode->i_sb), &dealloc);
7193 }
7194
7195 out:
7196 return ret;
7197 }
7198
7199 /*
7200 * Initialize security and acl for a already created inode.
7201 * Used for reflink a non-preserve-security file.
7202 *
7203 * It uses common api like ocfs2_xattr_set, so the caller
7204 * must not hold any lock expect i_rwsem.
7205 */
7206 int ocfs2_init_security_and_acl(struct inode *dir,
7207 struct inode *inode,
7208 const struct qstr *qstr)
7209 {
7210 int ret = 0;
7211 struct buffer_head *dir_bh = NULL;
7212
7213 ret = ocfs2_init_security_get(inode, dir, qstr, NULL);
7214 if (ret) {
7215 mlog_errno(ret);
7216 goto leave;
7217 }
7218
7219 ret = ocfs2_inode_lock(dir, &dir_bh, 0);
7220 if (ret) {
7221 mlog_errno(ret);
7222 goto leave;
7223 }
7224 ret = ocfs2_init_acl(NULL, inode, dir, NULL, dir_bh, NULL, NULL);
7225 if (ret)
7226 mlog_errno(ret);
7227
7228 ocfs2_inode_unlock(dir, 0);
7229 brelse(dir_bh);
7230 leave:
7231 return ret;
7232 }
7233
7234 /*
7235 * 'security' attributes support
7236 */
7237 static int ocfs2_xattr_security_get(const struct xattr_handler *handler,
7238 struct dentry *unused, struct inode *inode,
7239 const char *name, void *buffer, size_t size)
7240 {
7241 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_SECURITY,
7242 name, buffer, size);
7243 }
7244
7245 static int ocfs2_xattr_security_set(const struct xattr_handler *handler,
7246 struct mnt_idmap *idmap,
7247 struct dentry *unused, struct inode *inode,
7248 const char *name, const void *value,
7249 size_t size, int flags)
7250 {
7251 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY,
7252 name, value, size, flags);
7253 }
7254
7255 static int ocfs2_initxattrs(struct inode *inode, const struct xattr *xattr_array,
7256 void *fs_info)
7257 {
7258 struct ocfs2_security_xattr_info *si = fs_info;
7259 const struct xattr *xattr;
7260 int err = 0;
7261
7262 if (si) {
7263 si->value = kmemdup(xattr_array->value, xattr_array->value_len,
7264 GFP_KERNEL);
7265 if (!si->value)
7266 return -ENOMEM;
7267
7268 si->name = xattr_array->name;
7269 si->value_len = xattr_array->value_len;
7270 return 0;
7271 }
7272
7273 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
7274 err = ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY,
7275 xattr->name, xattr->value,
7276 xattr->value_len, XATTR_CREATE);
7277 if (err)
7278 break;
7279 }
7280 return err;
7281 }
7282
7283 int ocfs2_init_security_get(struct inode *inode,
7284 struct inode *dir,
7285 const struct qstr *qstr,
7286 struct ocfs2_security_xattr_info *si)
7287 {
7288 int ret;
7289
7290 /* check whether ocfs2 support feature xattr */
7291 if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
7292 return -EOPNOTSUPP;
7293 if (si) {
7294 ret = security_inode_init_security(inode, dir, qstr,
7295 &ocfs2_initxattrs, si);
7296 /*
7297 * security_inode_init_security() does not return -EOPNOTSUPP,
7298 * we have to check the xattr ourselves.
7299 */
7300 if (!ret && !si->name)
7301 si->enable = 0;
7302
7303 return ret;
7304 }
7305
7306 return security_inode_init_security(inode, dir, qstr,
7307 &ocfs2_initxattrs, NULL);
7308 }
7309
7310 int ocfs2_init_security_set(handle_t *handle,
7311 struct inode *inode,
7312 struct buffer_head *di_bh,
7313 struct ocfs2_security_xattr_info *si,
7314 struct ocfs2_alloc_context *xattr_ac,
7315 struct ocfs2_alloc_context *data_ac)
7316 {
7317 return ocfs2_xattr_set_handle(handle, inode, di_bh,
7318 OCFS2_XATTR_INDEX_SECURITY,
7319 si->name, si->value, si->value_len, 0,
7320 xattr_ac, data_ac);
7321 }
7322
7323 const struct xattr_handler ocfs2_xattr_security_handler = {
7324 .prefix = XATTR_SECURITY_PREFIX,
7325 .get = ocfs2_xattr_security_get,
7326 .set = ocfs2_xattr_security_set,
7327 };
7328
7329 /*
7330 * 'trusted' attributes support
7331 */
7332 static int ocfs2_xattr_trusted_get(const struct xattr_handler *handler,
7333 struct dentry *unused, struct inode *inode,
7334 const char *name, void *buffer, size_t size)
7335 {
7336 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED,
7337 name, buffer, size);
7338 }
7339
7340 static int ocfs2_xattr_trusted_set(const struct xattr_handler *handler,
7341 struct mnt_idmap *idmap,
7342 struct dentry *unused, struct inode *inode,
7343 const char *name, const void *value,
7344 size_t size, int flags)
7345 {
7346 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED,
7347 name, value, size, flags);
7348 }
7349
7350 const struct xattr_handler ocfs2_xattr_trusted_handler = {
7351 .prefix = XATTR_TRUSTED_PREFIX,
7352 .get = ocfs2_xattr_trusted_get,
7353 .set = ocfs2_xattr_trusted_set,
7354 };
7355
7356 /*
7357 * 'user' attributes support
7358 */
7359 static int ocfs2_xattr_user_get(const struct xattr_handler *handler,
7360 struct dentry *unused, struct inode *inode,
7361 const char *name, void *buffer, size_t size)
7362 {
7363 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7364
7365 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7366 return -EOPNOTSUPP;
7367 return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
7368 buffer, size);
7369 }
7370
7371 static int ocfs2_xattr_user_set(const struct xattr_handler *handler,
7372 struct mnt_idmap *idmap,
7373 struct dentry *unused, struct inode *inode,
7374 const char *name, const void *value,
7375 size_t size, int flags)
7376 {
7377 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7378
7379 if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7380 return -EOPNOTSUPP;
7381
7382 return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER,
7383 name, value, size, flags);
7384 }
7385
7386 const struct xattr_handler ocfs2_xattr_user_handler = {
7387 .prefix = XATTR_USER_PREFIX,
7388 .get = ocfs2_xattr_user_get,
7389 .set = ocfs2_xattr_user_set,
7390 };
7391
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.