?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
GUI: Fix Tomato RAF theme for all builds. Compilation typo.
[tomato.git] / release / src-rt-6.x.4708 / linux / linux-2.6.36 / fs / ocfs2 / refcounttree.c
blob5a6df222e9cc8bf76e882f5252f74fab0802e2a8
1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * refcounttree.c
5 *
6 * Copyright (C) 2009 Oracle. All rights reserved.
7 *
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public
10 * License version 2 as published by the Free Software Foundation.
11 *
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License for more details.
16 */
17
18 #include <linux/sort.h>
19 #define MLOG_MASK_PREFIX ML_REFCOUNT
20 #include <cluster/masklog.h>
21 #include "ocfs2.h"
22 #include "inode.h"
23 #include "alloc.h"
24 #include "suballoc.h"
25 #include "journal.h"
26 #include "uptodate.h"
27 #include "super.h"
28 #include "buffer_head_io.h"
29 #include "blockcheck.h"
30 #include "refcounttree.h"
31 #include "sysfile.h"
32 #include "dlmglue.h"
33 #include "extent_map.h"
34 #include "aops.h"
35 #include "xattr.h"
36 #include "namei.h"
37
38 #include <linux/bio.h>
39 #include <linux/blkdev.h>
40 #include <linux/slab.h>
41 #include <linux/writeback.h>
42 #include <linux/pagevec.h>
43 #include <linux/swap.h>
44 #include <linux/security.h>
45 #include <linux/fsnotify.h>
46 #include <linux/quotaops.h>
47 #include <linux/namei.h>
48 #include <linux/mount.h>
49
50 struct ocfs2_cow_context {
51 struct inode *inode;
52 u32 cow_start;
53 u32 cow_len;
54 struct ocfs2_extent_tree data_et;
55 struct ocfs2_refcount_tree *ref_tree;
56 struct buffer_head *ref_root_bh;
57 struct ocfs2_alloc_context *meta_ac;
58 struct ocfs2_alloc_context *data_ac;
59 struct ocfs2_cached_dealloc_ctxt dealloc;
60 void *cow_object;
61 struct ocfs2_post_refcount *post_refcount;
62 int extra_credits;
63 int (*get_clusters)(struct ocfs2_cow_context *context,
64 u32 v_cluster, u32 *p_cluster,
65 u32 *num_clusters,
66 unsigned int *extent_flags);
67 int (*cow_duplicate_clusters)(handle_t *handle,
68 struct ocfs2_cow_context *context,
69 u32 cpos, u32 old_cluster,
70 u32 new_cluster, u32 new_len);
71 };
72
73 static inline struct ocfs2_refcount_tree *
74 cache_info_to_refcount(struct ocfs2_caching_info *ci)
75 {
76 return container_of(ci, struct ocfs2_refcount_tree, rf_ci);
77 }
78
79 static int ocfs2_validate_refcount_block(struct super_block *sb,
80 struct buffer_head *bh)
81 {
82 int rc;
83 struct ocfs2_refcount_block *rb =
84 (struct ocfs2_refcount_block *)bh->b_data;
85
86 mlog(0, "Validating refcount block %llu\n",
87 (unsigned long long)bh->b_blocknr);
88
89 BUG_ON(!buffer_uptodate(bh));
90
91 /*
92 * If the ecc fails, we return the error but otherwise
93 * leave the filesystem running. We know any error is
94 * local to this block.
95 */
96 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &rb->rf_check);
97 if (rc) {
98 mlog(ML_ERROR, "Checksum failed for refcount block %llu\n",
99 (unsigned long long)bh->b_blocknr);
100 return rc;
101 }
102
103
104 if (!OCFS2_IS_VALID_REFCOUNT_BLOCK(rb)) {
105 ocfs2_error(sb,
106 "Refcount block #%llu has bad signature %.*s",
107 (unsigned long long)bh->b_blocknr, 7,
108 rb->rf_signature);
109 return -EINVAL;
110 }
111
112 if (le64_to_cpu(rb->rf_blkno) != bh->b_blocknr) {
113 ocfs2_error(sb,
114 "Refcount block #%llu has an invalid rf_blkno "
115 "of %llu",
116 (unsigned long long)bh->b_blocknr,
117 (unsigned long long)le64_to_cpu(rb->rf_blkno));
118 return -EINVAL;
119 }
120
121 if (le32_to_cpu(rb->rf_fs_generation) != OCFS2_SB(sb)->fs_generation) {
122 ocfs2_error(sb,
123 "Refcount block #%llu has an invalid "
124 "rf_fs_generation of #%u",
125 (unsigned long long)bh->b_blocknr,
126 le32_to_cpu(rb->rf_fs_generation));
127 return -EINVAL;
128 }
129
130 return 0;
131 }
132
133 static int ocfs2_read_refcount_block(struct ocfs2_caching_info *ci,
134 u64 rb_blkno,
135 struct buffer_head **bh)
136 {
137 int rc;
138 struct buffer_head *tmp = *bh;
139
140 rc = ocfs2_read_block(ci, rb_blkno, &tmp,
141 ocfs2_validate_refcount_block);
142
143 /* If ocfs2_read_block() got us a new bh, pass it up. */
144 if (!rc && !*bh)
145 *bh = tmp;
146
147 return rc;
148 }
149
150 static u64 ocfs2_refcount_cache_owner(struct ocfs2_caching_info *ci)
151 {
152 struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
153
154 return rf->rf_blkno;
155 }
156
157 static struct super_block *
158 ocfs2_refcount_cache_get_super(struct ocfs2_caching_info *ci)
159 {
160 struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
161
162 return rf->rf_sb;
163 }
164
165 static void ocfs2_refcount_cache_lock(struct ocfs2_caching_info *ci)
166 {
167 struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
168
169 spin_lock(&rf->rf_lock);
170 }
171
172 static void ocfs2_refcount_cache_unlock(struct ocfs2_caching_info *ci)
173 {
174 struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
175
176 spin_unlock(&rf->rf_lock);
177 }
178
179 static void ocfs2_refcount_cache_io_lock(struct ocfs2_caching_info *ci)
180 {
181 struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
182
183 mutex_lock(&rf->rf_io_mutex);
184 }
185
186 static void ocfs2_refcount_cache_io_unlock(struct ocfs2_caching_info *ci)
187 {
188 struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
189
190 mutex_unlock(&rf->rf_io_mutex);
191 }
192
193 static const struct ocfs2_caching_operations ocfs2_refcount_caching_ops = {
194 .co_owner = ocfs2_refcount_cache_owner,
195 .co_get_super = ocfs2_refcount_cache_get_super,
196 .co_cache_lock = ocfs2_refcount_cache_lock,
197 .co_cache_unlock = ocfs2_refcount_cache_unlock,
198 .co_io_lock = ocfs2_refcount_cache_io_lock,
199 .co_io_unlock = ocfs2_refcount_cache_io_unlock,
200 };
201
202 static struct ocfs2_refcount_tree *
203 ocfs2_find_refcount_tree(struct ocfs2_super *osb, u64 blkno)
204 {
205 struct rb_node *n = osb->osb_rf_lock_tree.rb_node;
206 struct ocfs2_refcount_tree *tree = NULL;
207
208 while (n) {
209 tree = rb_entry(n, struct ocfs2_refcount_tree, rf_node);
210
211 if (blkno < tree->rf_blkno)
212 n = n->rb_left;
213 else if (blkno > tree->rf_blkno)
214 n = n->rb_right;
215 else
216 return tree;
217 }
218
219 return NULL;
220 }
221
222 /* osb_lock is already locked. */
223 static void ocfs2_insert_refcount_tree(struct ocfs2_super *osb,
224 struct ocfs2_refcount_tree *new)
225 {
226 u64 rf_blkno = new->rf_blkno;
227 struct rb_node *parent = NULL;
228 struct rb_node **p = &osb->osb_rf_lock_tree.rb_node;
229 struct ocfs2_refcount_tree *tmp;
230
231 while (*p) {
232 parent = *p;
233
234 tmp = rb_entry(parent, struct ocfs2_refcount_tree,
235 rf_node);
236
237 if (rf_blkno < tmp->rf_blkno)
238 p = &(*p)->rb_left;
239 else if (rf_blkno > tmp->rf_blkno)
240 p = &(*p)->rb_right;
241 else {
242 /* This should never happen! */
243 mlog(ML_ERROR, "Duplicate refcount block %llu found!\n",
244 (unsigned long long)rf_blkno);
245 BUG();
246 }
247 }
248
249 rb_link_node(&new->rf_node, parent, p);
250 rb_insert_color(&new->rf_node, &osb->osb_rf_lock_tree);
251 }
252
253 static void ocfs2_free_refcount_tree(struct ocfs2_refcount_tree *tree)
254 {
255 ocfs2_metadata_cache_exit(&tree->rf_ci);
256 ocfs2_simple_drop_lockres(OCFS2_SB(tree->rf_sb), &tree->rf_lockres);
257 ocfs2_lock_res_free(&tree->rf_lockres);
258 kfree(tree);
259 }
260
261 static inline void
262 ocfs2_erase_refcount_tree_from_list_no_lock(struct ocfs2_super *osb,
263 struct ocfs2_refcount_tree *tree)
264 {
265 rb_erase(&tree->rf_node, &osb->osb_rf_lock_tree);
266 if (osb->osb_ref_tree_lru && osb->osb_ref_tree_lru == tree)
267 osb->osb_ref_tree_lru = NULL;
268 }
269
270 static void ocfs2_erase_refcount_tree_from_list(struct ocfs2_super *osb,
271 struct ocfs2_refcount_tree *tree)
272 {
273 spin_lock(&osb->osb_lock);
274 ocfs2_erase_refcount_tree_from_list_no_lock(osb, tree);
275 spin_unlock(&osb->osb_lock);
276 }
277
278 static void ocfs2_kref_remove_refcount_tree(struct kref *kref)
279 {
280 struct ocfs2_refcount_tree *tree =
281 container_of(kref, struct ocfs2_refcount_tree, rf_getcnt);
282
283 ocfs2_free_refcount_tree(tree);
284 }
285
286 static inline void
287 ocfs2_refcount_tree_get(struct ocfs2_refcount_tree *tree)
288 {
289 kref_get(&tree->rf_getcnt);
290 }
291
292 static inline void
293 ocfs2_refcount_tree_put(struct ocfs2_refcount_tree *tree)
294 {
295 kref_put(&tree->rf_getcnt, ocfs2_kref_remove_refcount_tree);
296 }
297
298 static inline void ocfs2_init_refcount_tree_ci(struct ocfs2_refcount_tree *new,
299 struct super_block *sb)
300 {
301 ocfs2_metadata_cache_init(&new->rf_ci, &ocfs2_refcount_caching_ops);
302 mutex_init(&new->rf_io_mutex);
303 new->rf_sb = sb;
304 spin_lock_init(&new->rf_lock);
305 }
306
307 static inline void ocfs2_init_refcount_tree_lock(struct ocfs2_super *osb,
308 struct ocfs2_refcount_tree *new,
309 u64 rf_blkno, u32 generation)
310 {
311 init_rwsem(&new->rf_sem);
312 ocfs2_refcount_lock_res_init(&new->rf_lockres, osb,
313 rf_blkno, generation);
314 }
315
316 static struct ocfs2_refcount_tree*
317 ocfs2_allocate_refcount_tree(struct ocfs2_super *osb, u64 rf_blkno)
318 {
319 struct ocfs2_refcount_tree *new;
320
321 new = kzalloc(sizeof(struct ocfs2_refcount_tree), GFP_NOFS);
322 if (!new)
323 return NULL;
324
325 new->rf_blkno = rf_blkno;
326 kref_init(&new->rf_getcnt);
327 ocfs2_init_refcount_tree_ci(new, osb->sb);
328
329 return new;
330 }
331
332 static int ocfs2_get_refcount_tree(struct ocfs2_super *osb, u64 rf_blkno,
333 struct ocfs2_refcount_tree **ret_tree)
334 {
335 int ret = 0;
336 struct ocfs2_refcount_tree *tree, *new = NULL;
337 struct buffer_head *ref_root_bh = NULL;
338 struct ocfs2_refcount_block *ref_rb;
339
340 spin_lock(&osb->osb_lock);
341 if (osb->osb_ref_tree_lru &&
342 osb->osb_ref_tree_lru->rf_blkno == rf_blkno)
343 tree = osb->osb_ref_tree_lru;
344 else
345 tree = ocfs2_find_refcount_tree(osb, rf_blkno);
346 if (tree)
347 goto out;
348
349 spin_unlock(&osb->osb_lock);
350
351 new = ocfs2_allocate_refcount_tree(osb, rf_blkno);
352 if (!new) {
353 ret = -ENOMEM;
354 mlog_errno(ret);
355 return ret;
356 }
357 /*
358 * We need the generation to create the refcount tree lock and since
359 * it isn't changed during the tree modification, we are safe here to
360 * read without protection.
361 * We also have to purge the cache after we create the lock since the
362 * refcount block may have the stale data. It can only be trusted when
363 * we hold the refcount lock.
364 */
365 ret = ocfs2_read_refcount_block(&new->rf_ci, rf_blkno, &ref_root_bh);
366 if (ret) {
367 mlog_errno(ret);
368 ocfs2_metadata_cache_exit(&new->rf_ci);
369 kfree(new);
370 return ret;
371 }
372
373 ref_rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
374 new->rf_generation = le32_to_cpu(ref_rb->rf_generation);
375 ocfs2_init_refcount_tree_lock(osb, new, rf_blkno,
376 new->rf_generation);
377 ocfs2_metadata_cache_purge(&new->rf_ci);
378
379 spin_lock(&osb->osb_lock);
380 tree = ocfs2_find_refcount_tree(osb, rf_blkno);
381 if (tree)
382 goto out;
383
384 ocfs2_insert_refcount_tree(osb, new);
385
386 tree = new;
387 new = NULL;
388
389 out:
390 *ret_tree = tree;
391
392 osb->osb_ref_tree_lru = tree;
393
394 spin_unlock(&osb->osb_lock);
395
396 if (new)
397 ocfs2_free_refcount_tree(new);
398
399 brelse(ref_root_bh);
400 return ret;
401 }
402
403 static int ocfs2_get_refcount_block(struct inode *inode, u64 *ref_blkno)
404 {
405 int ret;
406 struct buffer_head *di_bh = NULL;
407 struct ocfs2_dinode *di;
408
409 ret = ocfs2_read_inode_block(inode, &di_bh);
410 if (ret) {
411 mlog_errno(ret);
412 goto out;
413 }
414
415 BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
416
417 di = (struct ocfs2_dinode *)di_bh->b_data;
418 *ref_blkno = le64_to_cpu(di->i_refcount_loc);
419 brelse(di_bh);
420 out:
421 return ret;
422 }
423
424 static int __ocfs2_lock_refcount_tree(struct ocfs2_super *osb,
425 struct ocfs2_refcount_tree *tree, int rw)
426 {
427 int ret;
428
429 ret = ocfs2_refcount_lock(tree, rw);
430 if (ret) {
431 mlog_errno(ret);
432 goto out;
433 }
434
435 if (rw)
436 down_write(&tree->rf_sem);
437 else
438 down_read(&tree->rf_sem);
439
440 out:
441 return ret;
442 }
443
444 /*
445 * Lock the refcount tree pointed by ref_blkno and return the tree.
446 * In most case, we lock the tree and read the refcount block.
447 * So read it here if the caller really needs it.
448 *
449 * If the tree has been re-created by other node, it will free the
450 * old one and re-create it.
451 */
452 int ocfs2_lock_refcount_tree(struct ocfs2_super *osb,
453 u64 ref_blkno, int rw,
454 struct ocfs2_refcount_tree **ret_tree,
455 struct buffer_head **ref_bh)
456 {
457 int ret, delete_tree = 0;
458 struct ocfs2_refcount_tree *tree = NULL;
459 struct buffer_head *ref_root_bh = NULL;
460 struct ocfs2_refcount_block *rb;
461
462 again:
463 ret = ocfs2_get_refcount_tree(osb, ref_blkno, &tree);
464 if (ret) {
465 mlog_errno(ret);
466 return ret;
467 }
468
469 ocfs2_refcount_tree_get(tree);
470
471 ret = __ocfs2_lock_refcount_tree(osb, tree, rw);
472 if (ret) {
473 mlog_errno(ret);
474 ocfs2_refcount_tree_put(tree);
475 goto out;
476 }
477
478 ret = ocfs2_read_refcount_block(&tree->rf_ci, tree->rf_blkno,
479 &ref_root_bh);
480 if (ret) {
481 mlog_errno(ret);
482 ocfs2_unlock_refcount_tree(osb, tree, rw);
483 ocfs2_refcount_tree_put(tree);
484 goto out;
485 }
486
487 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
488 /*
489 * If the refcount block has been freed and re-created, we may need
490 * to recreate the refcount tree also.
491 *
492 * Here we just remove the tree from the rb-tree, and the last
493 * kref holder will unlock and delete this refcount_tree.
494 * Then we goto "again" and ocfs2_get_refcount_tree will create
495 * the new refcount tree for us.
496 */
497 if (tree->rf_generation != le32_to_cpu(rb->rf_generation)) {
498 if (!tree->rf_removed) {
499 ocfs2_erase_refcount_tree_from_list(osb, tree);
500 tree->rf_removed = 1;
501 delete_tree = 1;
502 }
503
504 ocfs2_unlock_refcount_tree(osb, tree, rw);
505 /*
506 * We get an extra reference when we create the refcount
507 * tree, so another put will destroy it.
508 */
509 if (delete_tree)
510 ocfs2_refcount_tree_put(tree);
511 brelse(ref_root_bh);
512 ref_root_bh = NULL;
513 goto again;
514 }
515
516 *ret_tree = tree;
517 if (ref_bh) {
518 *ref_bh = ref_root_bh;
519 ref_root_bh = NULL;
520 }
521 out:
522 brelse(ref_root_bh);
523 return ret;
524 }
525
526 void ocfs2_unlock_refcount_tree(struct ocfs2_super *osb,
527 struct ocfs2_refcount_tree *tree, int rw)
528 {
529 if (rw)
530 up_write(&tree->rf_sem);
531 else
532 up_read(&tree->rf_sem);
533
534 ocfs2_refcount_unlock(tree, rw);
535 ocfs2_refcount_tree_put(tree);
536 }
537
538 void ocfs2_purge_refcount_trees(struct ocfs2_super *osb)
539 {
540 struct rb_node *node;
541 struct ocfs2_refcount_tree *tree;
542 struct rb_root *root = &osb->osb_rf_lock_tree;
543
544 while ((node = rb_last(root)) != NULL) {
545 tree = rb_entry(node, struct ocfs2_refcount_tree, rf_node);
546
547 mlog(0, "Purge tree %llu\n",
548 (unsigned long long) tree->rf_blkno);
549
550 rb_erase(&tree->rf_node, root);
551 ocfs2_free_refcount_tree(tree);
552 }
553 }
554
555 /*
556 * Create a refcount tree for an inode.
557 * We take for granted that the inode is already locked.
558 */
559 static int ocfs2_create_refcount_tree(struct inode *inode,
560 struct buffer_head *di_bh)
561 {
562 int ret;
563 handle_t *handle = NULL;
564 struct ocfs2_alloc_context *meta_ac = NULL;
565 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
566 struct ocfs2_inode_info *oi = OCFS2_I(inode);
567 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
568 struct buffer_head *new_bh = NULL;
569 struct ocfs2_refcount_block *rb;
570 struct ocfs2_refcount_tree *new_tree = NULL, *tree = NULL;
571 u16 suballoc_bit_start;
572 u32 num_got;
573 u64 suballoc_loc, first_blkno;
574
575 BUG_ON(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL);
576
577 mlog(0, "create tree for inode %lu\n", inode->i_ino);
578
579 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
580 if (ret) {
581 mlog_errno(ret);
582 goto out;
583 }
584
585 handle = ocfs2_start_trans(osb, OCFS2_REFCOUNT_TREE_CREATE_CREDITS);
586 if (IS_ERR(handle)) {
587 ret = PTR_ERR(handle);
588 mlog_errno(ret);
589 goto out;
590 }
591
592 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
593 OCFS2_JOURNAL_ACCESS_WRITE);
594 if (ret) {
595 mlog_errno(ret);
596 goto out_commit;
597 }
598
599 ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
600 &suballoc_bit_start, &num_got,
601 &first_blkno);
602 if (ret) {
603 mlog_errno(ret);
604 goto out_commit;
605 }
606
607 new_tree = ocfs2_allocate_refcount_tree(osb, first_blkno);
608 if (!new_tree) {
609 ret = -ENOMEM;
610 mlog_errno(ret);
611 goto out_commit;
612 }
613
614 new_bh = sb_getblk(inode->i_sb, first_blkno);
615 ocfs2_set_new_buffer_uptodate(&new_tree->rf_ci, new_bh);
616
617 ret = ocfs2_journal_access_rb(handle, &new_tree->rf_ci, new_bh,
618 OCFS2_JOURNAL_ACCESS_CREATE);
619 if (ret) {
620 mlog_errno(ret);
621 goto out_commit;
622 }
623
624 /* Initialize ocfs2_refcount_block. */
625 rb = (struct ocfs2_refcount_block *)new_bh->b_data;
626 memset(rb, 0, inode->i_sb->s_blocksize);
627 strcpy((void *)rb, OCFS2_REFCOUNT_BLOCK_SIGNATURE);
628 rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
629 rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
630 rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
631 rb->rf_fs_generation = cpu_to_le32(osb->fs_generation);
632 rb->rf_blkno = cpu_to_le64(first_blkno);
633 rb->rf_count = cpu_to_le32(1);
634 rb->rf_records.rl_count =
635 cpu_to_le16(ocfs2_refcount_recs_per_rb(osb->sb));
636 spin_lock(&osb->osb_lock);
637 rb->rf_generation = osb->s_next_generation++;
638 spin_unlock(&osb->osb_lock);
639
640 ocfs2_journal_dirty(handle, new_bh);
641
642 spin_lock(&oi->ip_lock);
643 oi->ip_dyn_features |= OCFS2_HAS_REFCOUNT_FL;
644 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
645 di->i_refcount_loc = cpu_to_le64(first_blkno);
646 spin_unlock(&oi->ip_lock);
647
648 mlog(0, "created tree for inode %lu, refblock %llu\n",
649 inode->i_ino, (unsigned long long)first_blkno);
650
651 ocfs2_journal_dirty(handle, di_bh);
652
653 /*
654 * We have to init the tree lock here since it will use
655 * the generation number to create it.
656 */
657 new_tree->rf_generation = le32_to_cpu(rb->rf_generation);
658 ocfs2_init_refcount_tree_lock(osb, new_tree, first_blkno,
659 new_tree->rf_generation);
660
661 spin_lock(&osb->osb_lock);
662 tree = ocfs2_find_refcount_tree(osb, first_blkno);
663
664 /*
665 * We've just created a new refcount tree in this block. If
666 * we found a refcount tree on the ocfs2_super, it must be
667 * one we just deleted. We free the old tree before
668 * inserting the new tree.
669 */
670 BUG_ON(tree && tree->rf_generation == new_tree->rf_generation);
671 if (tree)
672 ocfs2_erase_refcount_tree_from_list_no_lock(osb, tree);
673 ocfs2_insert_refcount_tree(osb, new_tree);
674 spin_unlock(&osb->osb_lock);
675 new_tree = NULL;
676 if (tree)
677 ocfs2_refcount_tree_put(tree);
678
679 out_commit:
680 ocfs2_commit_trans(osb, handle);
681
682 out:
683 if (new_tree) {
684 ocfs2_metadata_cache_exit(&new_tree->rf_ci);
685 kfree(new_tree);
686 }
687
688 brelse(new_bh);
689 if (meta_ac)
690 ocfs2_free_alloc_context(meta_ac);
691
692 return ret;
693 }
694
695 static int ocfs2_set_refcount_tree(struct inode *inode,
696 struct buffer_head *di_bh,
697 u64 refcount_loc)
698 {
699 int ret;
700 handle_t *handle = NULL;
701 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
702 struct ocfs2_inode_info *oi = OCFS2_I(inode);
703 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
704 struct buffer_head *ref_root_bh = NULL;
705 struct ocfs2_refcount_block *rb;
706 struct ocfs2_refcount_tree *ref_tree;
707
708 BUG_ON(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL);
709
710 ret = ocfs2_lock_refcount_tree(osb, refcount_loc, 1,
711 &ref_tree, &ref_root_bh);
712 if (ret) {
713 mlog_errno(ret);
714 return ret;
715 }
716
717 handle = ocfs2_start_trans(osb, OCFS2_REFCOUNT_TREE_SET_CREDITS);
718 if (IS_ERR(handle)) {
719 ret = PTR_ERR(handle);
720 mlog_errno(ret);
721 goto out;
722 }
723
724 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
725 OCFS2_JOURNAL_ACCESS_WRITE);
726 if (ret) {
727 mlog_errno(ret);
728 goto out_commit;
729 }
730
731 ret = ocfs2_journal_access_rb(handle, &ref_tree->rf_ci, ref_root_bh,
732 OCFS2_JOURNAL_ACCESS_WRITE);
733 if (ret) {
734 mlog_errno(ret);
735 goto out_commit;
736 }
737
738 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
739 le32_add_cpu(&rb->rf_count, 1);
740
741 ocfs2_journal_dirty(handle, ref_root_bh);
742
743 spin_lock(&oi->ip_lock);
744 oi->ip_dyn_features |= OCFS2_HAS_REFCOUNT_FL;
745 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
746 di->i_refcount_loc = cpu_to_le64(refcount_loc);
747 spin_unlock(&oi->ip_lock);
748 ocfs2_journal_dirty(handle, di_bh);
749
750 out_commit:
751 ocfs2_commit_trans(osb, handle);
752 out:
753 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
754 brelse(ref_root_bh);
755
756 return ret;
757 }
758
759 int ocfs2_remove_refcount_tree(struct inode *inode, struct buffer_head *di_bh)
760 {
761 int ret, delete_tree = 0;
762 handle_t *handle = NULL;
763 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
764 struct ocfs2_inode_info *oi = OCFS2_I(inode);
765 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
766 struct ocfs2_refcount_block *rb;
767 struct inode *alloc_inode = NULL;
768 struct buffer_head *alloc_bh = NULL;
769 struct buffer_head *blk_bh = NULL;
770 struct ocfs2_refcount_tree *ref_tree;
771 int credits = OCFS2_REFCOUNT_TREE_REMOVE_CREDITS;
772 u64 blk = 0, bg_blkno = 0, ref_blkno = le64_to_cpu(di->i_refcount_loc);
773 u16 bit = 0;
774
775 if (!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL))
776 return 0;
777
778 BUG_ON(!ref_blkno);
779 ret = ocfs2_lock_refcount_tree(osb, ref_blkno, 1, &ref_tree, &blk_bh);
780 if (ret) {
781 mlog_errno(ret);
782 return ret;
783 }
784
785 rb = (struct ocfs2_refcount_block *)blk_bh->b_data;
786
787 /*
788 * If we are the last user, we need to free the block.
789 * So lock the allocator ahead.
790 */
791 if (le32_to_cpu(rb->rf_count) == 1) {
792 blk = le64_to_cpu(rb->rf_blkno);
793 bit = le16_to_cpu(rb->rf_suballoc_bit);
794 if (rb->rf_suballoc_loc)
795 bg_blkno = le64_to_cpu(rb->rf_suballoc_loc);
796 else
797 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
798
799 alloc_inode = ocfs2_get_system_file_inode(osb,
800 EXTENT_ALLOC_SYSTEM_INODE,
801 le16_to_cpu(rb->rf_suballoc_slot));
802 if (!alloc_inode) {
803 ret = -ENOMEM;
804 mlog_errno(ret);
805 goto out;
806 }
807 mutex_lock(&alloc_inode->i_mutex);
808
809 ret = ocfs2_inode_lock(alloc_inode, &alloc_bh, 1);
810 if (ret) {
811 mlog_errno(ret);
812 goto out_mutex;
813 }
814
815 credits += OCFS2_SUBALLOC_FREE;
816 }
817
818 handle = ocfs2_start_trans(osb, credits);
819 if (IS_ERR(handle)) {
820 ret = PTR_ERR(handle);
821 mlog_errno(ret);
822 goto out_unlock;
823 }
824
825 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
826 OCFS2_JOURNAL_ACCESS_WRITE);
827 if (ret) {
828 mlog_errno(ret);
829 goto out_commit;
830 }
831
832 ret = ocfs2_journal_access_rb(handle, &ref_tree->rf_ci, blk_bh,
833 OCFS2_JOURNAL_ACCESS_WRITE);
834 if (ret) {
835 mlog_errno(ret);
836 goto out_commit;
837 }
838
839 spin_lock(&oi->ip_lock);
840 oi->ip_dyn_features &= ~OCFS2_HAS_REFCOUNT_FL;
841 di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
842 di->i_refcount_loc = 0;
843 spin_unlock(&oi->ip_lock);
844 ocfs2_journal_dirty(handle, di_bh);
845
846 le32_add_cpu(&rb->rf_count , -1);
847 ocfs2_journal_dirty(handle, blk_bh);
848
849 if (!rb->rf_count) {
850 delete_tree = 1;
851 ocfs2_erase_refcount_tree_from_list(osb, ref_tree);
852 ret = ocfs2_free_suballoc_bits(handle, alloc_inode,
853 alloc_bh, bit, bg_blkno, 1);
854 if (ret)
855 mlog_errno(ret);
856 }
857
858 out_commit:
859 ocfs2_commit_trans(osb, handle);
860 out_unlock:
861 if (alloc_inode) {
862 ocfs2_inode_unlock(alloc_inode, 1);
863 brelse(alloc_bh);
864 }
865 out_mutex:
866 if (alloc_inode) {
867 mutex_unlock(&alloc_inode->i_mutex);
868 iput(alloc_inode);
869 }
870 out:
871 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
872 if (delete_tree)
873 ocfs2_refcount_tree_put(ref_tree);
874 brelse(blk_bh);
875
876 return ret;
877 }
878
879 static void ocfs2_find_refcount_rec_in_rl(struct ocfs2_caching_info *ci,
880 struct buffer_head *ref_leaf_bh,
881 u64 cpos, unsigned int len,
882 struct ocfs2_refcount_rec *ret_rec,
883 int *index)
884 {
885 int i = 0;
886 struct ocfs2_refcount_block *rb =
887 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
888 struct ocfs2_refcount_rec *rec = NULL;
889
890 for (; i < le16_to_cpu(rb->rf_records.rl_used); i++) {
891 rec = &rb->rf_records.rl_recs[i];
892
893 if (le64_to_cpu(rec->r_cpos) +
894 le32_to_cpu(rec->r_clusters) <= cpos)
895 continue;
896 else if (le64_to_cpu(rec->r_cpos) > cpos)
897 break;
898
899 /* ok, cpos fail in this rec. Just return. */
900 if (ret_rec)
901 *ret_rec = *rec;
902 goto out;
903 }
904
905 if (ret_rec) {
906 /* We meet with a hole here, so fake the rec. */
907 ret_rec->r_cpos = cpu_to_le64(cpos);
908 ret_rec->r_refcount = 0;
909 if (i < le16_to_cpu(rb->rf_records.rl_used) &&
910 le64_to_cpu(rec->r_cpos) < cpos + len)
911 ret_rec->r_clusters =
912 cpu_to_le32(le64_to_cpu(rec->r_cpos) - cpos);
913 else
914 ret_rec->r_clusters = cpu_to_le32(len);
915 }
916
917 out:
918 *index = i;
919 }
920
921 /*
922 * Try to remove refcount tree. The mechanism is:
923 * 1) Check whether i_clusters == 0, if no, exit.
924 * 2) check whether we have i_xattr_loc in dinode. if yes, exit.
925 * 3) Check whether we have inline xattr stored outside, if yes, exit.
926 * 4) Remove the tree.
927 */
928 int ocfs2_try_remove_refcount_tree(struct inode *inode,
929 struct buffer_head *di_bh)
930 {
931 int ret;
932 struct ocfs2_inode_info *oi = OCFS2_I(inode);
933 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
934
935 down_write(&oi->ip_xattr_sem);
936 down_write(&oi->ip_alloc_sem);
937
938 if (oi->ip_clusters)
939 goto out;
940
941 if ((oi->ip_dyn_features & OCFS2_HAS_XATTR_FL) && di->i_xattr_loc)
942 goto out;
943
944 if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL &&
945 ocfs2_has_inline_xattr_value_outside(inode, di))
946 goto out;
947
948 ret = ocfs2_remove_refcount_tree(inode, di_bh);
949 if (ret)
950 mlog_errno(ret);
951 out:
952 up_write(&oi->ip_alloc_sem);
953 up_write(&oi->ip_xattr_sem);
954 return 0;
955 }
956
957 /*
958 * Find the end range for a leaf refcount block indicated by
959 * el->l_recs[index].e_blkno.
960 */
961 static int ocfs2_get_refcount_cpos_end(struct ocfs2_caching_info *ci,
962 struct buffer_head *ref_root_bh,
963 struct ocfs2_extent_block *eb,
964 struct ocfs2_extent_list *el,
965 int index, u32 *cpos_end)
966 {
967 int ret, i, subtree_root;
968 u32 cpos;
969 u64 blkno;
970 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
971 struct ocfs2_path *left_path = NULL, *right_path = NULL;
972 struct ocfs2_extent_tree et;
973 struct ocfs2_extent_list *tmp_el;
974
975 if (index < le16_to_cpu(el->l_next_free_rec) - 1) {
976 /*
977 * We have a extent rec after index, so just use the e_cpos
978 * of the next extent rec.
979 */
980 *cpos_end = le32_to_cpu(el->l_recs[index+1].e_cpos);
981 return 0;
982 }
983
984 if (!eb || (eb && !eb->h_next_leaf_blk)) {
985 /*
986 * We are the last extent rec, so any high cpos should
987 * be stored in this leaf refcount block.
988 */
989 *cpos_end = UINT_MAX;
990 return 0;
991 }
992
993 /*
994 * If the extent block isn't the last one, we have to find
995 * the subtree root between this extent block and the next
996 * leaf extent block and get the corresponding e_cpos from
997 * the subroot. Otherwise we may corrupt the b-tree.
998 */
999 ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
1000
1001 left_path = ocfs2_new_path_from_et(&et);
1002 if (!left_path) {
1003 ret = -ENOMEM;
1004 mlog_errno(ret);
1005 goto out;
1006 }
1007
1008 cpos = le32_to_cpu(eb->h_list.l_recs[index].e_cpos);
1009 ret = ocfs2_find_path(ci, left_path, cpos);
1010 if (ret) {
1011 mlog_errno(ret);
1012 goto out;
1013 }
1014
1015 right_path = ocfs2_new_path_from_path(left_path);
1016 if (!right_path) {
1017 ret = -ENOMEM;
1018 mlog_errno(ret);
1019 goto out;
1020 }
1021
1022 ret = ocfs2_find_cpos_for_right_leaf(sb, left_path, &cpos);
1023 if (ret) {
1024 mlog_errno(ret);
1025 goto out;
1026 }
1027
1028 ret = ocfs2_find_path(ci, right_path, cpos);
1029 if (ret) {
1030 mlog_errno(ret);
1031 goto out;
1032 }
1033
1034 subtree_root = ocfs2_find_subtree_root(&et, left_path,
1035 right_path);
1036
1037 tmp_el = left_path->p_node[subtree_root].el;
1038 blkno = left_path->p_node[subtree_root+1].bh->b_blocknr;
1039 for (i = 0; i < le32_to_cpu(tmp_el->l_next_free_rec); i++) {
1040 if (le64_to_cpu(tmp_el->l_recs[i].e_blkno) == blkno) {
1041 *cpos_end = le32_to_cpu(tmp_el->l_recs[i+1].e_cpos);
1042 break;
1043 }
1044 }
1045
1046 BUG_ON(i == le32_to_cpu(tmp_el->l_next_free_rec));
1047
1048 out:
1049 ocfs2_free_path(left_path);
1050 ocfs2_free_path(right_path);
1051 return ret;
1052 }
1053
1054 /*
1055 * Given a cpos and len, try to find the refcount record which contains cpos.
1056 * 1. If cpos can be found in one refcount record, return the record.
1057 * 2. If cpos can't be found, return a fake record which start from cpos
1058 * and end at a small value between cpos+len and start of the next record.
1059 * This fake record has r_refcount = 0.
1060 */
1061 static int ocfs2_get_refcount_rec(struct ocfs2_caching_info *ci,
1062 struct buffer_head *ref_root_bh,
1063 u64 cpos, unsigned int len,
1064 struct ocfs2_refcount_rec *ret_rec,
1065 int *index,
1066 struct buffer_head **ret_bh)
1067 {
1068 int ret = 0, i, found;
1069 u32 low_cpos, uninitialized_var(cpos_end);
1070 struct ocfs2_extent_list *el;
1071 struct ocfs2_extent_rec *rec = NULL;
1072 struct ocfs2_extent_block *eb = NULL;
1073 struct buffer_head *eb_bh = NULL, *ref_leaf_bh = NULL;
1074 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1075 struct ocfs2_refcount_block *rb =
1076 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1077
1078 if (!(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)) {
1079 ocfs2_find_refcount_rec_in_rl(ci, ref_root_bh, cpos, len,
1080 ret_rec, index);
1081 *ret_bh = ref_root_bh;
1082 get_bh(ref_root_bh);
1083 return 0;
1084 }
1085
1086 el = &rb->rf_list;
1087 low_cpos = cpos & OCFS2_32BIT_POS_MASK;
1088
1089 if (el->l_tree_depth) {
1090 ret = ocfs2_find_leaf(ci, el, low_cpos, &eb_bh);
1091 if (ret) {
1092 mlog_errno(ret);
1093 goto out;
1094 }
1095
1096 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
1097 el = &eb->h_list;
1098
1099 if (el->l_tree_depth) {
1100 ocfs2_error(sb,
1101 "refcount tree %llu has non zero tree "
1102 "depth in leaf btree tree block %llu\n",
1103 (unsigned long long)ocfs2_metadata_cache_owner(ci),
1104 (unsigned long long)eb_bh->b_blocknr);
1105 ret = -EROFS;
1106 goto out;
1107 }
1108 }
1109
1110 found = 0;
1111 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1112 rec = &el->l_recs[i];
1113
1114 if (le32_to_cpu(rec->e_cpos) <= low_cpos) {
1115 found = 1;
1116 break;
1117 }
1118 }
1119
1120 if (found) {
1121 ret = ocfs2_get_refcount_cpos_end(ci, ref_root_bh,
1122 eb, el, i, &cpos_end);
1123 if (ret) {
1124 mlog_errno(ret);
1125 goto out;
1126 }
1127
1128 if (cpos_end < low_cpos + len)
1129 len = cpos_end - low_cpos;
1130 }
1131
1132 ret = ocfs2_read_refcount_block(ci, le64_to_cpu(rec->e_blkno),
1133 &ref_leaf_bh);
1134 if (ret) {
1135 mlog_errno(ret);
1136 goto out;
1137 }
1138
1139 ocfs2_find_refcount_rec_in_rl(ci, ref_leaf_bh, cpos, len,
1140 ret_rec, index);
1141 *ret_bh = ref_leaf_bh;
1142 out:
1143 brelse(eb_bh);
1144 return ret;
1145 }
1146
1147 enum ocfs2_ref_rec_contig {
1148 REF_CONTIG_NONE = 0,
1149 REF_CONTIG_LEFT,
1150 REF_CONTIG_RIGHT,
1151 REF_CONTIG_LEFTRIGHT,
1152 };
1153
1154 static enum ocfs2_ref_rec_contig
1155 ocfs2_refcount_rec_adjacent(struct ocfs2_refcount_block *rb,
1156 int index)
1157 {
1158 if ((rb->rf_records.rl_recs[index].r_refcount ==
1159 rb->rf_records.rl_recs[index + 1].r_refcount) &&
1160 (le64_to_cpu(rb->rf_records.rl_recs[index].r_cpos) +
1161 le32_to_cpu(rb->rf_records.rl_recs[index].r_clusters) ==
1162 le64_to_cpu(rb->rf_records.rl_recs[index + 1].r_cpos)))
1163 return REF_CONTIG_RIGHT;
1164
1165 return REF_CONTIG_NONE;
1166 }
1167
1168 static enum ocfs2_ref_rec_contig
1169 ocfs2_refcount_rec_contig(struct ocfs2_refcount_block *rb,
1170 int index)
1171 {
1172 enum ocfs2_ref_rec_contig ret = REF_CONTIG_NONE;
1173
1174 if (index < le16_to_cpu(rb->rf_records.rl_used) - 1)
1175 ret = ocfs2_refcount_rec_adjacent(rb, index);
1176
1177 if (index > 0) {
1178 enum ocfs2_ref_rec_contig tmp;
1179
1180 tmp = ocfs2_refcount_rec_adjacent(rb, index - 1);
1181
1182 if (tmp == REF_CONTIG_RIGHT) {
1183 if (ret == REF_CONTIG_RIGHT)
1184 ret = REF_CONTIG_LEFTRIGHT;
1185 else
1186 ret = REF_CONTIG_LEFT;
1187 }
1188 }
1189
1190 return ret;
1191 }
1192
1193 static void ocfs2_rotate_refcount_rec_left(struct ocfs2_refcount_block *rb,
1194 int index)
1195 {
1196 BUG_ON(rb->rf_records.rl_recs[index].r_refcount !=
1197 rb->rf_records.rl_recs[index+1].r_refcount);
1198
1199 le32_add_cpu(&rb->rf_records.rl_recs[index].r_clusters,
1200 le32_to_cpu(rb->rf_records.rl_recs[index+1].r_clusters));
1201
1202 if (index < le16_to_cpu(rb->rf_records.rl_used) - 2)
1203 memmove(&rb->rf_records.rl_recs[index + 1],
1204 &rb->rf_records.rl_recs[index + 2],
1205 sizeof(struct ocfs2_refcount_rec) *
1206 (le16_to_cpu(rb->rf_records.rl_used) - index - 2));
1207
1208 memset(&rb->rf_records.rl_recs[le16_to_cpu(rb->rf_records.rl_used) - 1],
1209 0, sizeof(struct ocfs2_refcount_rec));
1210 le16_add_cpu(&rb->rf_records.rl_used, -1);
1211 }
1212
1213 /*
1214 * Merge the refcount rec if we are contiguous with the adjacent recs.
1215 */
1216 static void ocfs2_refcount_rec_merge(struct ocfs2_refcount_block *rb,
1217 int index)
1218 {
1219 enum ocfs2_ref_rec_contig contig =
1220 ocfs2_refcount_rec_contig(rb, index);
1221
1222 if (contig == REF_CONTIG_NONE)
1223 return;
1224
1225 if (contig == REF_CONTIG_LEFT || contig == REF_CONTIG_LEFTRIGHT) {
1226 BUG_ON(index == 0);
1227 index--;
1228 }
1229
1230 ocfs2_rotate_refcount_rec_left(rb, index);
1231
1232 if (contig == REF_CONTIG_LEFTRIGHT)
1233 ocfs2_rotate_refcount_rec_left(rb, index);
1234 }
1235
1236 /*
1237 * Change the refcount indexed by "index" in ref_bh.
1238 * If refcount reaches 0, remove it.
1239 */
1240 static int ocfs2_change_refcount_rec(handle_t *handle,
1241 struct ocfs2_caching_info *ci,
1242 struct buffer_head *ref_leaf_bh,
1243 int index, int merge, int change)
1244 {
1245 int ret;
1246 struct ocfs2_refcount_block *rb =
1247 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1248 struct ocfs2_refcount_list *rl = &rb->rf_records;
1249 struct ocfs2_refcount_rec *rec = &rl->rl_recs[index];
1250
1251 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1252 OCFS2_JOURNAL_ACCESS_WRITE);
1253 if (ret) {
1254 mlog_errno(ret);
1255 goto out;
1256 }
1257
1258 mlog(0, "change index %d, old count %u, change %d\n", index,
1259 le32_to_cpu(rec->r_refcount), change);
1260 le32_add_cpu(&rec->r_refcount, change);
1261
1262 if (!rec->r_refcount) {
1263 if (index != le16_to_cpu(rl->rl_used) - 1) {
1264 memmove(rec, rec + 1,
1265 (le16_to_cpu(rl->rl_used) - index - 1) *
1266 sizeof(struct ocfs2_refcount_rec));
1267 memset(&rl->rl_recs[le16_to_cpu(rl->rl_used) - 1],
1268 0, sizeof(struct ocfs2_refcount_rec));
1269 }
1270
1271 le16_add_cpu(&rl->rl_used, -1);
1272 } else if (merge)
1273 ocfs2_refcount_rec_merge(rb, index);
1274
1275 ocfs2_journal_dirty(handle, ref_leaf_bh);
1276 out:
1277 return ret;
1278 }
1279
1280 static int ocfs2_expand_inline_ref_root(handle_t *handle,
1281 struct ocfs2_caching_info *ci,
1282 struct buffer_head *ref_root_bh,
1283 struct buffer_head **ref_leaf_bh,
1284 struct ocfs2_alloc_context *meta_ac)
1285 {
1286 int ret;
1287 u16 suballoc_bit_start;
1288 u32 num_got;
1289 u64 suballoc_loc, blkno;
1290 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1291 struct buffer_head *new_bh = NULL;
1292 struct ocfs2_refcount_block *new_rb;
1293 struct ocfs2_refcount_block *root_rb =
1294 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1295
1296 ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
1297 OCFS2_JOURNAL_ACCESS_WRITE);
1298 if (ret) {
1299 mlog_errno(ret);
1300 goto out;
1301 }
1302
1303 ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
1304 &suballoc_bit_start, &num_got,
1305 &blkno);
1306 if (ret) {
1307 mlog_errno(ret);
1308 goto out;
1309 }
1310
1311 new_bh = sb_getblk(sb, blkno);
1312 if (new_bh == NULL) {
1313 ret = -EIO;
1314 mlog_errno(ret);
1315 goto out;
1316 }
1317 ocfs2_set_new_buffer_uptodate(ci, new_bh);
1318
1319 ret = ocfs2_journal_access_rb(handle, ci, new_bh,
1320 OCFS2_JOURNAL_ACCESS_CREATE);
1321 if (ret) {
1322 mlog_errno(ret);
1323 goto out;
1324 }
1325
1326 /*
1327 * Initialize ocfs2_refcount_block.
1328 * It should contain the same information as the old root.
1329 * so just memcpy it and change the corresponding field.
1330 */
1331 memcpy(new_bh->b_data, ref_root_bh->b_data, sb->s_blocksize);
1332
1333 new_rb = (struct ocfs2_refcount_block *)new_bh->b_data;
1334 new_rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
1335 new_rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
1336 new_rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1337 new_rb->rf_blkno = cpu_to_le64(blkno);
1338 new_rb->rf_cpos = cpu_to_le32(0);
1339 new_rb->rf_parent = cpu_to_le64(ref_root_bh->b_blocknr);
1340 new_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_LEAF_FL);
1341 ocfs2_journal_dirty(handle, new_bh);
1342
1343 /* Now change the root. */
1344 memset(&root_rb->rf_list, 0, sb->s_blocksize -
1345 offsetof(struct ocfs2_refcount_block, rf_list));
1346 root_rb->rf_list.l_count = cpu_to_le16(ocfs2_extent_recs_per_rb(sb));
1347 root_rb->rf_clusters = cpu_to_le32(1);
1348 root_rb->rf_list.l_next_free_rec = cpu_to_le16(1);
1349 root_rb->rf_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
1350 root_rb->rf_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
1351 root_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_TREE_FL);
1352
1353 ocfs2_journal_dirty(handle, ref_root_bh);
1354
1355 mlog(0, "new leaf block %llu, used %u\n", (unsigned long long)blkno,
1356 le16_to_cpu(new_rb->rf_records.rl_used));
1357
1358 *ref_leaf_bh = new_bh;
1359 new_bh = NULL;
1360 out:
1361 brelse(new_bh);
1362 return ret;
1363 }
1364
1365 static int ocfs2_refcount_rec_no_intersect(struct ocfs2_refcount_rec *prev,
1366 struct ocfs2_refcount_rec *next)
1367 {
1368 if (ocfs2_get_ref_rec_low_cpos(prev) + le32_to_cpu(prev->r_clusters) <=
1369 ocfs2_get_ref_rec_low_cpos(next))
1370 return 1;
1371
1372 return 0;
1373 }
1374
1375 static int cmp_refcount_rec_by_low_cpos(const void *a, const void *b)
1376 {
1377 const struct ocfs2_refcount_rec *l = a, *r = b;
1378 u32 l_cpos = ocfs2_get_ref_rec_low_cpos(l);
1379 u32 r_cpos = ocfs2_get_ref_rec_low_cpos(r);
1380
1381 if (l_cpos > r_cpos)
1382 return 1;
1383 if (l_cpos < r_cpos)
1384 return -1;
1385 return 0;
1386 }
1387
1388 static int cmp_refcount_rec_by_cpos(const void *a, const void *b)
1389 {
1390 const struct ocfs2_refcount_rec *l = a, *r = b;
1391 u64 l_cpos = le64_to_cpu(l->r_cpos);
1392 u64 r_cpos = le64_to_cpu(r->r_cpos);
1393
1394 if (l_cpos > r_cpos)
1395 return 1;
1396 if (l_cpos < r_cpos)
1397 return -1;
1398 return 0;
1399 }
1400
1401 static void swap_refcount_rec(void *a, void *b, int size)
1402 {
1403 struct ocfs2_refcount_rec *l = a, *r = b, tmp;
1404
1405 tmp = *(struct ocfs2_refcount_rec *)l;
1406 *(struct ocfs2_refcount_rec *)l =
1407 *(struct ocfs2_refcount_rec *)r;
1408 *(struct ocfs2_refcount_rec *)r = tmp;
1409 }
1410
1411 /*
1412 * The refcount cpos are ordered by their 64bit cpos,
1413 * But we will use the low 32 bit to be the e_cpos in the b-tree.
1414 * So we need to make sure that this pos isn't intersected with others.
1415 *
1416 * Note: The refcount block is already sorted by their low 32 bit cpos,
1417 * So just try the middle pos first, and we will exit when we find
1418 * the good position.
1419 */
1420 static int ocfs2_find_refcount_split_pos(struct ocfs2_refcount_list *rl,
1421 u32 *split_pos, int *split_index)
1422 {
1423 int num_used = le16_to_cpu(rl->rl_used);
1424 int delta, middle = num_used / 2;
1425
1426 for (delta = 0; delta < middle; delta++) {
1427 /* Let's check delta earlier than middle */
1428 if (ocfs2_refcount_rec_no_intersect(
1429 &rl->rl_recs[middle - delta - 1],
1430 &rl->rl_recs[middle - delta])) {
1431 *split_index = middle - delta;
1432 break;
1433 }
1434
1435 /* For even counts, don't walk off the end */
1436 if ((middle + delta + 1) == num_used)
1437 continue;
1438
1439 /* Now try delta past middle */
1440 if (ocfs2_refcount_rec_no_intersect(
1441 &rl->rl_recs[middle + delta],
1442 &rl->rl_recs[middle + delta + 1])) {
1443 *split_index = middle + delta + 1;
1444 break;
1445 }
1446 }
1447
1448 if (delta >= middle)
1449 return -ENOSPC;
1450
1451 *split_pos = ocfs2_get_ref_rec_low_cpos(&rl->rl_recs[*split_index]);
1452 return 0;
1453 }
1454
1455 static int ocfs2_divide_leaf_refcount_block(struct buffer_head *ref_leaf_bh,
1456 struct buffer_head *new_bh,
1457 u32 *split_cpos)
1458 {
1459 int split_index = 0, num_moved, ret;
1460 u32 cpos = 0;
1461 struct ocfs2_refcount_block *rb =
1462 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1463 struct ocfs2_refcount_list *rl = &rb->rf_records;
1464 struct ocfs2_refcount_block *new_rb =
1465 (struct ocfs2_refcount_block *)new_bh->b_data;
1466 struct ocfs2_refcount_list *new_rl = &new_rb->rf_records;
1467
1468 mlog(0, "split old leaf refcount block %llu, count = %u, used = %u\n",
1469 (unsigned long long)ref_leaf_bh->b_blocknr,
1470 le32_to_cpu(rl->rl_count), le32_to_cpu(rl->rl_used));
1471
1472 sort(&rl->rl_recs, le16_to_cpu(rl->rl_used),
1473 sizeof(struct ocfs2_refcount_rec),
1474 cmp_refcount_rec_by_low_cpos, swap_refcount_rec);
1475
1476 ret = ocfs2_find_refcount_split_pos(rl, &cpos, &split_index);
1477 if (ret) {
1478 mlog_errno(ret);
1479 return ret;
1480 }
1481
1482 new_rb->rf_cpos = cpu_to_le32(cpos);
1483
1484 /* move refcount records starting from split_index to the new block. */
1485 num_moved = le16_to_cpu(rl->rl_used) - split_index;
1486 memcpy(new_rl->rl_recs, &rl->rl_recs[split_index],
1487 num_moved * sizeof(struct ocfs2_refcount_rec));
1488
1489 /*ok, remove the entries we just moved over to the other block. */
1490 memset(&rl->rl_recs[split_index], 0,
1491 num_moved * sizeof(struct ocfs2_refcount_rec));
1492
1493 /* change old and new rl_used accordingly. */
1494 le16_add_cpu(&rl->rl_used, -num_moved);
1495 new_rl->rl_used = cpu_to_le16(num_moved);
1496
1497 sort(&rl->rl_recs, le16_to_cpu(rl->rl_used),
1498 sizeof(struct ocfs2_refcount_rec),
1499 cmp_refcount_rec_by_cpos, swap_refcount_rec);
1500
1501 sort(&new_rl->rl_recs, le16_to_cpu(new_rl->rl_used),
1502 sizeof(struct ocfs2_refcount_rec),
1503 cmp_refcount_rec_by_cpos, swap_refcount_rec);
1504
1505 *split_cpos = cpos;
1506 return 0;
1507 }
1508
1509 static int ocfs2_new_leaf_refcount_block(handle_t *handle,
1510 struct ocfs2_caching_info *ci,
1511 struct buffer_head *ref_root_bh,
1512 struct buffer_head *ref_leaf_bh,
1513 struct ocfs2_alloc_context *meta_ac)
1514 {
1515 int ret;
1516 u16 suballoc_bit_start;
1517 u32 num_got, new_cpos;
1518 u64 suballoc_loc, blkno;
1519 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1520 struct ocfs2_refcount_block *root_rb =
1521 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1522 struct buffer_head *new_bh = NULL;
1523 struct ocfs2_refcount_block *new_rb;
1524 struct ocfs2_extent_tree ref_et;
1525
1526 BUG_ON(!(le32_to_cpu(root_rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL));
1527
1528 ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
1529 OCFS2_JOURNAL_ACCESS_WRITE);
1530 if (ret) {
1531 mlog_errno(ret);
1532 goto out;
1533 }
1534
1535 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1536 OCFS2_JOURNAL_ACCESS_WRITE);
1537 if (ret) {
1538 mlog_errno(ret);
1539 goto out;
1540 }
1541
1542 ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
1543 &suballoc_bit_start, &num_got,
1544 &blkno);
1545 if (ret) {
1546 mlog_errno(ret);
1547 goto out;
1548 }
1549
1550 new_bh = sb_getblk(sb, blkno);
1551 if (new_bh == NULL) {
1552 ret = -EIO;
1553 mlog_errno(ret);
1554 goto out;
1555 }
1556 ocfs2_set_new_buffer_uptodate(ci, new_bh);
1557
1558 ret = ocfs2_journal_access_rb(handle, ci, new_bh,
1559 OCFS2_JOURNAL_ACCESS_CREATE);
1560 if (ret) {
1561 mlog_errno(ret);
1562 goto out;
1563 }
1564
1565 /* Initialize ocfs2_refcount_block. */
1566 new_rb = (struct ocfs2_refcount_block *)new_bh->b_data;
1567 memset(new_rb, 0, sb->s_blocksize);
1568 strcpy((void *)new_rb, OCFS2_REFCOUNT_BLOCK_SIGNATURE);
1569 new_rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
1570 new_rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
1571 new_rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1572 new_rb->rf_fs_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
1573 new_rb->rf_blkno = cpu_to_le64(blkno);
1574 new_rb->rf_parent = cpu_to_le64(ref_root_bh->b_blocknr);
1575 new_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_LEAF_FL);
1576 new_rb->rf_records.rl_count =
1577 cpu_to_le16(ocfs2_refcount_recs_per_rb(sb));
1578 new_rb->rf_generation = root_rb->rf_generation;
1579
1580 ret = ocfs2_divide_leaf_refcount_block(ref_leaf_bh, new_bh, &new_cpos);
1581 if (ret) {
1582 mlog_errno(ret);
1583 goto out;
1584 }
1585
1586 ocfs2_journal_dirty(handle, ref_leaf_bh);
1587 ocfs2_journal_dirty(handle, new_bh);
1588
1589 ocfs2_init_refcount_extent_tree(&ref_et, ci, ref_root_bh);
1590
1591 mlog(0, "insert new leaf block %llu at %u\n",
1592 (unsigned long long)new_bh->b_blocknr, new_cpos);
1593
1594 /* Insert the new leaf block with the specific offset cpos. */
1595 ret = ocfs2_insert_extent(handle, &ref_et, new_cpos, new_bh->b_blocknr,
1596 1, 0, meta_ac);
1597 if (ret)
1598 mlog_errno(ret);
1599
1600 out:
1601 brelse(new_bh);
1602 return ret;
1603 }
1604
1605 static int ocfs2_expand_refcount_tree(handle_t *handle,
1606 struct ocfs2_caching_info *ci,
1607 struct buffer_head *ref_root_bh,
1608 struct buffer_head *ref_leaf_bh,
1609 struct ocfs2_alloc_context *meta_ac)
1610 {
1611 int ret;
1612 struct buffer_head *expand_bh = NULL;
1613
1614 if (ref_root_bh == ref_leaf_bh) {
1615 /*
1616 * the old root bh hasn't been expanded to a b-tree,
1617 * so expand it first.
1618 */
1619 ret = ocfs2_expand_inline_ref_root(handle, ci, ref_root_bh,
1620 &expand_bh, meta_ac);
1621 if (ret) {
1622 mlog_errno(ret);
1623 goto out;
1624 }
1625 } else {
1626 expand_bh = ref_leaf_bh;
1627 get_bh(expand_bh);
1628 }
1629
1630
1631 /* Now add a new refcount block into the tree.*/
1632 ret = ocfs2_new_leaf_refcount_block(handle, ci, ref_root_bh,
1633 expand_bh, meta_ac);
1634 if (ret)
1635 mlog_errno(ret);
1636 out:
1637 brelse(expand_bh);
1638 return ret;
1639 }
1640
1641 /*
1642 * Adjust the extent rec in b-tree representing ref_leaf_bh.
1643 *
1644 * Only called when we have inserted a new refcount rec at index 0
1645 * which means ocfs2_extent_rec.e_cpos may need some change.
1646 */
1647 static int ocfs2_adjust_refcount_rec(handle_t *handle,
1648 struct ocfs2_caching_info *ci,
1649 struct buffer_head *ref_root_bh,
1650 struct buffer_head *ref_leaf_bh,
1651 struct ocfs2_refcount_rec *rec)
1652 {
1653 int ret = 0, i;
1654 u32 new_cpos, old_cpos;
1655 struct ocfs2_path *path = NULL;
1656 struct ocfs2_extent_tree et;
1657 struct ocfs2_refcount_block *rb =
1658 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1659 struct ocfs2_extent_list *el;
1660
1661 if (!(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL))
1662 goto out;
1663
1664 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1665 old_cpos = le32_to_cpu(rb->rf_cpos);
1666 new_cpos = le64_to_cpu(rec->r_cpos) & OCFS2_32BIT_POS_MASK;
1667 if (old_cpos <= new_cpos)
1668 goto out;
1669
1670 ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
1671
1672 path = ocfs2_new_path_from_et(&et);
1673 if (!path) {
1674 ret = -ENOMEM;
1675 mlog_errno(ret);
1676 goto out;
1677 }
1678
1679 ret = ocfs2_find_path(ci, path, old_cpos);
1680 if (ret) {
1681 mlog_errno(ret);
1682 goto out;
1683 }
1684
1685 /*
1686 * 2 more credits, one for the leaf refcount block, one for
1687 * the extent block contains the extent rec.
1688 */
1689 ret = ocfs2_extend_trans(handle, 2);
1690 if (ret < 0) {
1691 mlog_errno(ret);
1692 goto out;
1693 }
1694
1695 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1696 OCFS2_JOURNAL_ACCESS_WRITE);
1697 if (ret < 0) {
1698 mlog_errno(ret);
1699 goto out;
1700 }
1701
1702 ret = ocfs2_journal_access_eb(handle, ci, path_leaf_bh(path),
1703 OCFS2_JOURNAL_ACCESS_WRITE);
1704 if (ret < 0) {
1705 mlog_errno(ret);
1706 goto out;
1707 }
1708
1709 /* change the leaf extent block first. */
1710 el = path_leaf_el(path);
1711
1712 for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++)
1713 if (le32_to_cpu(el->l_recs[i].e_cpos) == old_cpos)
1714 break;
1715
1716 BUG_ON(i == le16_to_cpu(el->l_next_free_rec));
1717
1718 el->l_recs[i].e_cpos = cpu_to_le32(new_cpos);
1719
1720 /* change the r_cpos in the leaf block. */
1721 rb->rf_cpos = cpu_to_le32(new_cpos);
1722
1723 ocfs2_journal_dirty(handle, path_leaf_bh(path));
1724 ocfs2_journal_dirty(handle, ref_leaf_bh);
1725
1726 out:
1727 ocfs2_free_path(path);
1728 return ret;
1729 }
1730
1731 static int ocfs2_insert_refcount_rec(handle_t *handle,
1732 struct ocfs2_caching_info *ci,
1733 struct buffer_head *ref_root_bh,
1734 struct buffer_head *ref_leaf_bh,
1735 struct ocfs2_refcount_rec *rec,
1736 int index, int merge,
1737 struct ocfs2_alloc_context *meta_ac)
1738 {
1739 int ret;
1740 struct ocfs2_refcount_block *rb =
1741 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1742 struct ocfs2_refcount_list *rf_list = &rb->rf_records;
1743 struct buffer_head *new_bh = NULL;
1744
1745 BUG_ON(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL);
1746
1747 if (rf_list->rl_used == rf_list->rl_count) {
1748 u64 cpos = le64_to_cpu(rec->r_cpos);
1749 u32 len = le32_to_cpu(rec->r_clusters);
1750
1751 ret = ocfs2_expand_refcount_tree(handle, ci, ref_root_bh,
1752 ref_leaf_bh, meta_ac);
1753 if (ret) {
1754 mlog_errno(ret);
1755 goto out;
1756 }
1757
1758 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
1759 cpos, len, NULL, &index,
1760 &new_bh);
1761 if (ret) {
1762 mlog_errno(ret);
1763 goto out;
1764 }
1765
1766 ref_leaf_bh = new_bh;
1767 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1768 rf_list = &rb->rf_records;
1769 }
1770
1771 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1772 OCFS2_JOURNAL_ACCESS_WRITE);
1773 if (ret) {
1774 mlog_errno(ret);
1775 goto out;
1776 }
1777
1778 if (index < le16_to_cpu(rf_list->rl_used))
1779 memmove(&rf_list->rl_recs[index + 1],
1780 &rf_list->rl_recs[index],
1781 (le16_to_cpu(rf_list->rl_used) - index) *
1782 sizeof(struct ocfs2_refcount_rec));
1783
1784 mlog(0, "insert refcount record start %llu, len %u, count %u "
1785 "to leaf block %llu at index %d\n",
1786 (unsigned long long)le64_to_cpu(rec->r_cpos),
1787 le32_to_cpu(rec->r_clusters), le32_to_cpu(rec->r_refcount),
1788 (unsigned long long)ref_leaf_bh->b_blocknr, index);
1789
1790 rf_list->rl_recs[index] = *rec;
1791
1792 le16_add_cpu(&rf_list->rl_used, 1);
1793
1794 if (merge)
1795 ocfs2_refcount_rec_merge(rb, index);
1796
1797 ocfs2_journal_dirty(handle, ref_leaf_bh);
1798
1799 if (index == 0) {
1800 ret = ocfs2_adjust_refcount_rec(handle, ci,
1801 ref_root_bh,
1802 ref_leaf_bh, rec);
1803 if (ret)
1804 mlog_errno(ret);
1805 }
1806 out:
1807 brelse(new_bh);
1808 return ret;
1809 }
1810
1811 /*
1812 * Split the refcount_rec indexed by "index" in ref_leaf_bh.
1813 * This is much simple than our b-tree code.
1814 * split_rec is the new refcount rec we want to insert.
1815 * If split_rec->r_refcount > 0, we are changing the refcount(in case we
1816 * increase refcount or decrease a refcount to non-zero).
1817 * If split_rec->r_refcount == 0, we are punching a hole in current refcount
1818 * rec( in case we decrease a refcount to zero).
1819 */
1820 static int ocfs2_split_refcount_rec(handle_t *handle,
1821 struct ocfs2_caching_info *ci,
1822 struct buffer_head *ref_root_bh,
1823 struct buffer_head *ref_leaf_bh,
1824 struct ocfs2_refcount_rec *split_rec,
1825 int index, int merge,
1826 struct ocfs2_alloc_context *meta_ac,
1827 struct ocfs2_cached_dealloc_ctxt *dealloc)
1828 {
1829 int ret, recs_need;
1830 u32 len;
1831 struct ocfs2_refcount_block *rb =
1832 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1833 struct ocfs2_refcount_list *rf_list = &rb->rf_records;
1834 struct ocfs2_refcount_rec *orig_rec = &rf_list->rl_recs[index];
1835 struct ocfs2_refcount_rec *tail_rec = NULL;
1836 struct buffer_head *new_bh = NULL;
1837
1838 BUG_ON(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL);
1839
1840 mlog(0, "original r_pos %llu, cluster %u, split %llu, cluster %u\n",
1841 le64_to_cpu(orig_rec->r_cpos), le32_to_cpu(orig_rec->r_clusters),
1842 le64_to_cpu(split_rec->r_cpos),
1843 le32_to_cpu(split_rec->r_clusters));
1844
1845 /*
1846 * If we just need to split the header or tail clusters,
1847 * no more recs are needed, just split is OK.
1848 * Otherwise we at least need one new recs.
1849 */
1850 if (!split_rec->r_refcount &&
1851 (split_rec->r_cpos == orig_rec->r_cpos ||
1852 le64_to_cpu(split_rec->r_cpos) +
1853 le32_to_cpu(split_rec->r_clusters) ==
1854 le64_to_cpu(orig_rec->r_cpos) + le32_to_cpu(orig_rec->r_clusters)))
1855 recs_need = 0;
1856 else
1857 recs_need = 1;
1858
1859 /*
1860 * We need one more rec if we split in the middle and the new rec have
1861 * some refcount in it.
1862 */
1863 if (split_rec->r_refcount &&
1864 (split_rec->r_cpos != orig_rec->r_cpos &&
1865 le64_to_cpu(split_rec->r_cpos) +
1866 le32_to_cpu(split_rec->r_clusters) !=
1867 le64_to_cpu(orig_rec->r_cpos) + le32_to_cpu(orig_rec->r_clusters)))
1868 recs_need++;
1869
1870 /* If the leaf block don't have enough record, expand it. */
1871 if (le16_to_cpu(rf_list->rl_used) + recs_need >
1872 le16_to_cpu(rf_list->rl_count)) {
1873 struct ocfs2_refcount_rec tmp_rec;
1874 u64 cpos = le64_to_cpu(orig_rec->r_cpos);
1875 len = le32_to_cpu(orig_rec->r_clusters);
1876 ret = ocfs2_expand_refcount_tree(handle, ci, ref_root_bh,
1877 ref_leaf_bh, meta_ac);
1878 if (ret) {
1879 mlog_errno(ret);
1880 goto out;
1881 }
1882
1883 /*
1884 * We have to re-get it since now cpos may be moved to
1885 * another leaf block.
1886 */
1887 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
1888 cpos, len, &tmp_rec, &index,
1889 &new_bh);
1890 if (ret) {
1891 mlog_errno(ret);
1892 goto out;
1893 }
1894
1895 ref_leaf_bh = new_bh;
1896 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1897 rf_list = &rb->rf_records;
1898 orig_rec = &rf_list->rl_recs[index];
1899 }
1900
1901 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1902 OCFS2_JOURNAL_ACCESS_WRITE);
1903 if (ret) {
1904 mlog_errno(ret);
1905 goto out;
1906 }
1907
1908 /*
1909 * We have calculated out how many new records we need and store
1910 * in recs_need, so spare enough space first by moving the records
1911 * after "index" to the end.
1912 */
1913 if (index != le16_to_cpu(rf_list->rl_used) - 1)
1914 memmove(&rf_list->rl_recs[index + 1 + recs_need],
1915 &rf_list->rl_recs[index + 1],
1916 (le16_to_cpu(rf_list->rl_used) - index - 1) *
1917 sizeof(struct ocfs2_refcount_rec));
1918
1919 len = (le64_to_cpu(orig_rec->r_cpos) +
1920 le32_to_cpu(orig_rec->r_clusters)) -
1921 (le64_to_cpu(split_rec->r_cpos) +
1922 le32_to_cpu(split_rec->r_clusters));
1923
1924 /*
1925 * If we have "len", the we will split in the tail and move it
1926 * to the end of the space we have just spared.
1927 */
1928 if (len) {
1929 tail_rec = &rf_list->rl_recs[index + recs_need];
1930
1931 memcpy(tail_rec, orig_rec, sizeof(struct ocfs2_refcount_rec));
1932 le64_add_cpu(&tail_rec->r_cpos,
1933 le32_to_cpu(tail_rec->r_clusters) - len);
1934 tail_rec->r_clusters = cpu_to_le32(len);
1935 }
1936
1937 /*
1938 * If the split pos isn't the same as the original one, we need to
1939 * split in the head.
1940 *
1941 * Note: We have the chance that split_rec.r_refcount = 0,
1942 * recs_need = 0 and len > 0, which means we just cut the head from
1943 * the orig_rec and in that case we have done some modification in
1944 * orig_rec above, so the check for r_cpos is faked.
1945 */
1946 if (split_rec->r_cpos != orig_rec->r_cpos && tail_rec != orig_rec) {
1947 len = le64_to_cpu(split_rec->r_cpos) -
1948 le64_to_cpu(orig_rec->r_cpos);
1949 orig_rec->r_clusters = cpu_to_le32(len);
1950 index++;
1951 }
1952
1953 le16_add_cpu(&rf_list->rl_used, recs_need);
1954
1955 if (split_rec->r_refcount) {
1956 rf_list->rl_recs[index] = *split_rec;
1957 mlog(0, "insert refcount record start %llu, len %u, count %u "
1958 "to leaf block %llu at index %d\n",
1959 (unsigned long long)le64_to_cpu(split_rec->r_cpos),
1960 le32_to_cpu(split_rec->r_clusters),
1961 le32_to_cpu(split_rec->r_refcount),
1962 (unsigned long long)ref_leaf_bh->b_blocknr, index);
1963
1964 if (merge)
1965 ocfs2_refcount_rec_merge(rb, index);
1966 }
1967
1968 ocfs2_journal_dirty(handle, ref_leaf_bh);
1969
1970 out:
1971 brelse(new_bh);
1972 return ret;
1973 }
1974
1975 static int __ocfs2_increase_refcount(handle_t *handle,
1976 struct ocfs2_caching_info *ci,
1977 struct buffer_head *ref_root_bh,
1978 u64 cpos, u32 len, int merge,
1979 struct ocfs2_alloc_context *meta_ac,
1980 struct ocfs2_cached_dealloc_ctxt *dealloc)
1981 {
1982 int ret = 0, index;
1983 struct buffer_head *ref_leaf_bh = NULL;
1984 struct ocfs2_refcount_rec rec;
1985 unsigned int set_len = 0;
1986
1987 mlog(0, "Tree owner %llu, add refcount start %llu, len %u\n",
1988 (unsigned long long)ocfs2_metadata_cache_owner(ci),
1989 (unsigned long long)cpos, len);
1990
1991 while (len) {
1992 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
1993 cpos, len, &rec, &index,
1994 &ref_leaf_bh);
1995 if (ret) {
1996 mlog_errno(ret);
1997 goto out;
1998 }
1999
2000 set_len = le32_to_cpu(rec.r_clusters);
2001
2002 /*
2003 * Here we may meet with 3 situations:
2004 *
2005 * 1. If we find an already existing record, and the length
2006 * is the same, cool, we just need to increase the r_refcount
2007 * and it is OK.
2008 * 2. If we find a hole, just insert it with r_refcount = 1.
2009 * 3. If we are in the middle of one extent record, split
2010 * it.
2011 */
2012 if (rec.r_refcount && le64_to_cpu(rec.r_cpos) == cpos &&
2013 set_len <= len) {
2014 mlog(0, "increase refcount rec, start %llu, len %u, "
2015 "count %u\n", (unsigned long long)cpos, set_len,
2016 le32_to_cpu(rec.r_refcount));
2017 ret = ocfs2_change_refcount_rec(handle, ci,
2018 ref_leaf_bh, index,
2019 merge, 1);
2020 if (ret) {
2021 mlog_errno(ret);
2022 goto out;
2023 }
2024 } else if (!rec.r_refcount) {
2025 rec.r_refcount = cpu_to_le32(1);
2026
2027 mlog(0, "insert refcount rec, start %llu, len %u\n",
2028 (unsigned long long)le64_to_cpu(rec.r_cpos),
2029 set_len);
2030 ret = ocfs2_insert_refcount_rec(handle, ci, ref_root_bh,
2031 ref_leaf_bh,
2032 &rec, index,
2033 merge, meta_ac);
2034 if (ret) {
2035 mlog_errno(ret);
2036 goto out;
2037 }
2038 } else {
2039 set_len = min((u64)(cpos + len),
2040 le64_to_cpu(rec.r_cpos) + set_len) - cpos;
2041 rec.r_cpos = cpu_to_le64(cpos);
2042 rec.r_clusters = cpu_to_le32(set_len);
2043 le32_add_cpu(&rec.r_refcount, 1);
2044
2045 mlog(0, "split refcount rec, start %llu, "
2046 "len %u, count %u\n",
2047 (unsigned long long)le64_to_cpu(rec.r_cpos),
2048 set_len, le32_to_cpu(rec.r_refcount));
2049 ret = ocfs2_split_refcount_rec(handle, ci,
2050 ref_root_bh, ref_leaf_bh,
2051 &rec, index, merge,
2052 meta_ac, dealloc);
2053 if (ret) {
2054 mlog_errno(ret);
2055 goto out;
2056 }
2057 }
2058
2059 cpos += set_len;
2060 len -= set_len;
2061 brelse(ref_leaf_bh);
2062 ref_leaf_bh = NULL;
2063 }
2064
2065 out:
2066 brelse(ref_leaf_bh);
2067 return ret;
2068 }
2069
2070 static int ocfs2_remove_refcount_extent(handle_t *handle,
2071 struct ocfs2_caching_info *ci,
2072 struct buffer_head *ref_root_bh,
2073 struct buffer_head *ref_leaf_bh,
2074 struct ocfs2_alloc_context *meta_ac,
2075 struct ocfs2_cached_dealloc_ctxt *dealloc)
2076 {
2077 int ret;
2078 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
2079 struct ocfs2_refcount_block *rb =
2080 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
2081 struct ocfs2_extent_tree et;
2082
2083 BUG_ON(rb->rf_records.rl_used);
2084
2085 ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
2086 ret = ocfs2_remove_extent(handle, &et, le32_to_cpu(rb->rf_cpos),
2087 1, meta_ac, dealloc);
2088 if (ret) {
2089 mlog_errno(ret);
2090 goto out;
2091 }
2092
2093 ocfs2_remove_from_cache(ci, ref_leaf_bh);
2094
2095 /*
2096 * add the freed block to the dealloc so that it will be freed
2097 * when we run dealloc.
2098 */
2099 ret = ocfs2_cache_block_dealloc(dealloc, EXTENT_ALLOC_SYSTEM_INODE,
2100 le16_to_cpu(rb->rf_suballoc_slot),
2101 le64_to_cpu(rb->rf_suballoc_loc),
2102 le64_to_cpu(rb->rf_blkno),
2103 le16_to_cpu(rb->rf_suballoc_bit));
2104 if (ret) {
2105 mlog_errno(ret);
2106 goto out;
2107 }
2108
2109 ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
2110 OCFS2_JOURNAL_ACCESS_WRITE);
2111 if (ret) {
2112 mlog_errno(ret);
2113 goto out;
2114 }
2115
2116 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
2117
2118 le32_add_cpu(&rb->rf_clusters, -1);
2119
2120 /*
2121 * check whether we need to restore the root refcount block if
2122 * there is no leaf extent block at atll.
2123 */
2124 if (!rb->rf_list.l_next_free_rec) {
2125 BUG_ON(rb->rf_clusters);
2126
2127 mlog(0, "reset refcount tree root %llu to be a record block.\n",
2128 (unsigned long long)ref_root_bh->b_blocknr);
2129
2130 rb->rf_flags = 0;
2131 rb->rf_parent = 0;
2132 rb->rf_cpos = 0;
2133 memset(&rb->rf_records, 0, sb->s_blocksize -
2134 offsetof(struct ocfs2_refcount_block, rf_records));
2135 rb->rf_records.rl_count =
2136 cpu_to_le16(ocfs2_refcount_recs_per_rb(sb));
2137 }
2138
2139 ocfs2_journal_dirty(handle, ref_root_bh);
2140
2141 out:
2142 return ret;
2143 }
2144
2145 int ocfs2_increase_refcount(handle_t *handle,
2146 struct ocfs2_caching_info *ci,
2147 struct buffer_head *ref_root_bh,
2148 u64 cpos, u32 len,
2149 struct ocfs2_alloc_context *meta_ac,
2150 struct ocfs2_cached_dealloc_ctxt *dealloc)
2151 {
2152 return __ocfs2_increase_refcount(handle, ci, ref_root_bh,
2153 cpos, len, 1,
2154 meta_ac, dealloc);
2155 }
2156
2157 static int ocfs2_decrease_refcount_rec(handle_t *handle,
2158 struct ocfs2_caching_info *ci,
2159 struct buffer_head *ref_root_bh,
2160 struct buffer_head *ref_leaf_bh,
2161 int index, u64 cpos, unsigned int len,
2162 struct ocfs2_alloc_context *meta_ac,
2163 struct ocfs2_cached_dealloc_ctxt *dealloc)
2164 {
2165 int ret;
2166 struct ocfs2_refcount_block *rb =
2167 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
2168 struct ocfs2_refcount_rec *rec = &rb->rf_records.rl_recs[index];
2169
2170 BUG_ON(cpos < le64_to_cpu(rec->r_cpos));
2171 BUG_ON(cpos + len >
2172 le64_to_cpu(rec->r_cpos) + le32_to_cpu(rec->r_clusters));
2173
2174 if (cpos == le64_to_cpu(rec->r_cpos) &&
2175 len == le32_to_cpu(rec->r_clusters))
2176 ret = ocfs2_change_refcount_rec(handle, ci,
2177 ref_leaf_bh, index, 1, -1);
2178 else {
2179 struct ocfs2_refcount_rec split = *rec;
2180 split.r_cpos = cpu_to_le64(cpos);
2181 split.r_clusters = cpu_to_le32(len);
2182
2183 le32_add_cpu(&split.r_refcount, -1);
2184
2185 mlog(0, "split refcount rec, start %llu, "
2186 "len %u, count %u, original start %llu, len %u\n",
2187 (unsigned long long)le64_to_cpu(split.r_cpos),
2188 len, le32_to_cpu(split.r_refcount),
2189 (unsigned long long)le64_to_cpu(rec->r_cpos),
2190 le32_to_cpu(rec->r_clusters));
2191 ret = ocfs2_split_refcount_rec(handle, ci,
2192 ref_root_bh, ref_leaf_bh,
2193 &split, index, 1,
2194 meta_ac, dealloc);
2195 }
2196
2197 if (ret) {
2198 mlog_errno(ret);
2199 goto out;
2200 }
2201
2202 /* Remove the leaf refcount block if it contains no refcount record. */
2203 if (!rb->rf_records.rl_used && ref_leaf_bh != ref_root_bh) {
2204 ret = ocfs2_remove_refcount_extent(handle, ci, ref_root_bh,
2205 ref_leaf_bh, meta_ac,
2206 dealloc);
2207 if (ret)
2208 mlog_errno(ret);
2209 }
2210
2211 out:
2212 return ret;
2213 }
2214
2215 static int __ocfs2_decrease_refcount(handle_t *handle,
2216 struct ocfs2_caching_info *ci,
2217 struct buffer_head *ref_root_bh,
2218 u64 cpos, u32 len,
2219 struct ocfs2_alloc_context *meta_ac,
2220 struct ocfs2_cached_dealloc_ctxt *dealloc,
2221 int delete)
2222 {
2223 int ret = 0, index = 0;
2224 struct ocfs2_refcount_rec rec;
2225 unsigned int r_count = 0, r_len;
2226 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
2227 struct buffer_head *ref_leaf_bh = NULL;
2228
2229 mlog(0, "Tree owner %llu, decrease refcount start %llu, "
2230 "len %u, delete %u\n",
2231 (unsigned long long)ocfs2_metadata_cache_owner(ci),
2232 (unsigned long long)cpos, len, delete);
2233
2234 while (len) {
2235 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2236 cpos, len, &rec, &index,
2237 &ref_leaf_bh);
2238 if (ret) {
2239 mlog_errno(ret);
2240 goto out;
2241 }
2242
2243 r_count = le32_to_cpu(rec.r_refcount);
2244 BUG_ON(r_count == 0);
2245 if (!delete)
2246 BUG_ON(r_count > 1);
2247
2248 r_len = min((u64)(cpos + len), le64_to_cpu(rec.r_cpos) +
2249 le32_to_cpu(rec.r_clusters)) - cpos;
2250
2251 ret = ocfs2_decrease_refcount_rec(handle, ci, ref_root_bh,
2252 ref_leaf_bh, index,
2253 cpos, r_len,
2254 meta_ac, dealloc);
2255 if (ret) {
2256 mlog_errno(ret);
2257 goto out;
2258 }
2259
2260 if (le32_to_cpu(rec.r_refcount) == 1 && delete) {
2261 ret = ocfs2_cache_cluster_dealloc(dealloc,
2262 ocfs2_clusters_to_blocks(sb, cpos),
2263 r_len);
2264 if (ret) {
2265 mlog_errno(ret);
2266 goto out;
2267 }
2268 }
2269
2270 cpos += r_len;
2271 len -= r_len;
2272 brelse(ref_leaf_bh);
2273 ref_leaf_bh = NULL;
2274 }
2275
2276 out:
2277 brelse(ref_leaf_bh);
2278 return ret;
2279 }
2280
2281 /* Caller must hold refcount tree lock. */
2282 int ocfs2_decrease_refcount(struct inode *inode,
2283 handle_t *handle, u32 cpos, u32 len,
2284 struct ocfs2_alloc_context *meta_ac,
2285 struct ocfs2_cached_dealloc_ctxt *dealloc,
2286 int delete)
2287 {
2288 int ret;
2289 u64 ref_blkno;
2290 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2291 struct buffer_head *ref_root_bh = NULL;
2292 struct ocfs2_refcount_tree *tree;
2293
2294 BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
2295
2296 ret = ocfs2_get_refcount_block(inode, &ref_blkno);
2297 if (ret) {
2298 mlog_errno(ret);
2299 goto out;
2300 }
2301
2302 ret = ocfs2_get_refcount_tree(OCFS2_SB(inode->i_sb), ref_blkno, &tree);
2303 if (ret) {
2304 mlog_errno(ret);
2305 goto out;
2306 }
2307
2308 ret = ocfs2_read_refcount_block(&tree->rf_ci, tree->rf_blkno,
2309 &ref_root_bh);
2310 if (ret) {
2311 mlog_errno(ret);
2312 goto out;
2313 }
2314
2315 ret = __ocfs2_decrease_refcount(handle, &tree->rf_ci, ref_root_bh,
2316 cpos, len, meta_ac, dealloc, delete);
2317 if (ret)
2318 mlog_errno(ret);
2319 out:
2320 brelse(ref_root_bh);
2321 return ret;
2322 }
2323
2324 /*
2325 * Mark the already-existing extent at cpos as refcounted for len clusters.
2326 * This adds the refcount extent flag.
2327 *
2328 * If the existing extent is larger than the request, initiate a
2329 * split. An attempt will be made at merging with adjacent extents.
2330 *
2331 * The caller is responsible for passing down meta_ac if we'll need it.
2332 */
2333 static int ocfs2_mark_extent_refcounted(struct inode *inode,
2334 struct ocfs2_extent_tree *et,
2335 handle_t *handle, u32 cpos,
2336 u32 len, u32 phys,
2337 struct ocfs2_alloc_context *meta_ac,
2338 struct ocfs2_cached_dealloc_ctxt *dealloc)
2339 {
2340 int ret;
2341
2342 mlog(0, "Inode %lu refcount tree cpos %u, len %u, phys cluster %u\n",
2343 inode->i_ino, cpos, len, phys);
2344
2345 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
2346 ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
2347 "tree, but the feature bit is not set in the "
2348 "super block.", inode->i_ino);
2349 ret = -EROFS;
2350 goto out;
2351 }
2352
2353 ret = ocfs2_change_extent_flag(handle, et, cpos,
2354 len, phys, meta_ac, dealloc,
2355 OCFS2_EXT_REFCOUNTED, 0);
2356 if (ret)
2357 mlog_errno(ret);
2358
2359 out:
2360 return ret;
2361 }
2362
2363 /*
2364 * Given some contiguous physical clusters, calculate what we need
2365 * for modifying their refcount.
2366 */
2367 static int ocfs2_calc_refcount_meta_credits(struct super_block *sb,
2368 struct ocfs2_caching_info *ci,
2369 struct buffer_head *ref_root_bh,
2370 u64 start_cpos,
2371 u32 clusters,
2372 int *meta_add,
2373 int *credits)
2374 {
2375 int ret = 0, index, ref_blocks = 0, recs_add = 0;
2376 u64 cpos = start_cpos;
2377 struct ocfs2_refcount_block *rb;
2378 struct ocfs2_refcount_rec rec;
2379 struct buffer_head *ref_leaf_bh = NULL, *prev_bh = NULL;
2380 u32 len;
2381
2382 mlog(0, "start_cpos %llu, clusters %u\n",
2383 (unsigned long long)start_cpos, clusters);
2384 while (clusters) {
2385 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2386 cpos, clusters, &rec,
2387 &index, &ref_leaf_bh);
2388 if (ret) {
2389 mlog_errno(ret);
2390 goto out;
2391 }
2392
2393 if (ref_leaf_bh != prev_bh) {
2394 /*
2395 * Now we encounter a new leaf block, so calculate
2396 * whether we need to extend the old leaf.
2397 */
2398 if (prev_bh) {
2399 rb = (struct ocfs2_refcount_block *)
2400 prev_bh->b_data;
2401
2402 if (le64_to_cpu(rb->rf_records.rl_used) +
2403 recs_add >
2404 le16_to_cpu(rb->rf_records.rl_count))
2405 ref_blocks++;
2406 }
2407
2408 recs_add = 0;
2409 *credits += 1;
2410 brelse(prev_bh);
2411 prev_bh = ref_leaf_bh;
2412 get_bh(prev_bh);
2413 }
2414
2415 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
2416
2417 mlog(0, "recs_add %d,cpos %llu, clusters %u, rec->r_cpos %llu,"
2418 "rec->r_clusters %u, rec->r_refcount %u, index %d\n",
2419 recs_add, (unsigned long long)cpos, clusters,
2420 (unsigned long long)le64_to_cpu(rec.r_cpos),
2421 le32_to_cpu(rec.r_clusters),
2422 le32_to_cpu(rec.r_refcount), index);
2423
2424 len = min((u64)cpos + clusters, le64_to_cpu(rec.r_cpos) +
2425 le32_to_cpu(rec.r_clusters)) - cpos;
2426 /*
2427 * We record all the records which will be inserted to the
2428 * same refcount block, so that we can tell exactly whether
2429 * we need a new refcount block or not.
2430 *
2431 * If we will insert a new one, this is easy and only happens
2432 * during adding refcounted flag to the extent, so we don't
2433 * have a chance of spliting. We just need one record.
2434 *
2435 * If the refcount rec already exists, that would be a little
2436 * complicated. we may have to:
2437 * 1) split at the beginning if the start pos isn't aligned.
2438 * we need 1 more record in this case.
2439 * 2) split int the end if the end pos isn't aligned.
2440 * we need 1 more record in this case.
2441 * 3) split in the middle because of file system fragmentation.
2442 * we need 2 more records in this case(we can't detect this
2443 * beforehand, so always think of the worst case).
2444 */
2445 if (rec.r_refcount) {
2446 recs_add += 2;
2447 /* Check whether we need a split at the beginning. */
2448 if (cpos == start_cpos &&
2449 cpos != le64_to_cpu(rec.r_cpos))
2450 recs_add++;
2451
2452 /* Check whether we need a split in the end. */
2453 if (cpos + clusters < le64_to_cpu(rec.r_cpos) +
2454 le32_to_cpu(rec.r_clusters))
2455 recs_add++;
2456 } else
2457 recs_add++;
2458
2459 brelse(ref_leaf_bh);
2460 ref_leaf_bh = NULL;
2461 clusters -= len;
2462 cpos += len;
2463 }
2464
2465 if (prev_bh) {
2466 rb = (struct ocfs2_refcount_block *)prev_bh->b_data;
2467
2468 if (le64_to_cpu(rb->rf_records.rl_used) + recs_add >
2469 le16_to_cpu(rb->rf_records.rl_count))
2470 ref_blocks++;
2471
2472 *credits += 1;
2473 }
2474
2475 if (!ref_blocks)
2476 goto out;
2477
2478 mlog(0, "we need ref_blocks %d\n", ref_blocks);
2479 *meta_add += ref_blocks;
2480 *credits += ref_blocks;
2481
2482 /*
2483 * So we may need ref_blocks to insert into the tree.
2484 * That also means we need to change the b-tree and add that number
2485 * of records since we never merge them.
2486 * We need one more block for expansion since the new created leaf
2487 * block is also full and needs split.
2488 */
2489 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
2490 if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL) {
2491 struct ocfs2_extent_tree et;
2492
2493 ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
2494 *meta_add += ocfs2_extend_meta_needed(et.et_root_el);
2495 *credits += ocfs2_calc_extend_credits(sb,
2496 et.et_root_el,
2497 ref_blocks);
2498 } else {
2499 *credits += OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
2500 *meta_add += 1;
2501 }
2502
2503 out:
2504 brelse(ref_leaf_bh);
2505 brelse(prev_bh);
2506 return ret;
2507 }
2508
2509 /*
2510 * For refcount tree, we will decrease some contiguous clusters
2511 * refcount count, so just go through it to see how many blocks
2512 * we gonna touch and whether we need to create new blocks.
2513 *
2514 * Normally the refcount blocks store these refcount should be
2515 * contiguous also, so that we can get the number easily.
2516 * We will at most add split 2 refcount records and 2 more
2517 * refcount blocks, so just check it in a rough way.
2518 *
2519 * Caller must hold refcount tree lock.
2520 */
2521 int ocfs2_prepare_refcount_change_for_del(struct inode *inode,
2522 u64 refcount_loc,
2523 u64 phys_blkno,
2524 u32 clusters,
2525 int *credits,
2526 int *ref_blocks)
2527 {
2528 int ret;
2529 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2530 struct buffer_head *ref_root_bh = NULL;
2531 struct ocfs2_refcount_tree *tree;
2532 u64 start_cpos = ocfs2_blocks_to_clusters(inode->i_sb, phys_blkno);
2533
2534 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
2535 ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
2536 "tree, but the feature bit is not set in the "
2537 "super block.", inode->i_ino);
2538 ret = -EROFS;
2539 goto out;
2540 }
2541
2542 BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
2543
2544 ret = ocfs2_get_refcount_tree(OCFS2_SB(inode->i_sb),
2545 refcount_loc, &tree);
2546 if (ret) {
2547 mlog_errno(ret);
2548 goto out;
2549 }
2550
2551 ret = ocfs2_read_refcount_block(&tree->rf_ci, refcount_loc,
2552 &ref_root_bh);
2553 if (ret) {
2554 mlog_errno(ret);
2555 goto out;
2556 }
2557
2558 ret = ocfs2_calc_refcount_meta_credits(inode->i_sb,
2559 &tree->rf_ci,
2560 ref_root_bh,
2561 start_cpos, clusters,
2562 ref_blocks, credits);
2563 if (ret) {
2564 mlog_errno(ret);
2565 goto out;
2566 }
2567
2568 mlog(0, "reserve new metadata %d blocks, credits = %d\n",
2569 *ref_blocks, *credits);
2570
2571 out:
2572 brelse(ref_root_bh);
2573 return ret;
2574 }
2575
2576 #define MAX_CONTIG_BYTES 1048576
2577
2578 static inline unsigned int ocfs2_cow_contig_clusters(struct super_block *sb)
2579 {
2580 return ocfs2_clusters_for_bytes(sb, MAX_CONTIG_BYTES);
2581 }
2582
2583 static inline unsigned int ocfs2_cow_contig_mask(struct super_block *sb)
2584 {
2585 return ~(ocfs2_cow_contig_clusters(sb) - 1);
2586 }
2587
2588 /*
2589 * Given an extent that starts at 'start' and an I/O that starts at 'cpos',
2590 * find an offset (start + (n * contig_clusters)) that is closest to cpos
2591 * while still being less than or equal to it.
2592 *
2593 * The goal is to break the extent at a multiple of contig_clusters.
2594 */
2595 static inline unsigned int ocfs2_cow_align_start(struct super_block *sb,
2596 unsigned int start,
2597 unsigned int cpos)
2598 {
2599 BUG_ON(start > cpos);
2600
2601 return start + ((cpos - start) & ocfs2_cow_contig_mask(sb));
2602 }
2603
2604 /*
2605 * Given a cluster count of len, pad it out so that it is a multiple
2606 * of contig_clusters.
2607 */
2608 static inline unsigned int ocfs2_cow_align_length(struct super_block *sb,
2609 unsigned int len)
2610 {
2611 unsigned int padded =
2612 (len + (ocfs2_cow_contig_clusters(sb) - 1)) &
2613 ocfs2_cow_contig_mask(sb);
2614
2615 /* Did we wrap? */
2616 if (padded < len)
2617 padded = UINT_MAX;
2618
2619 return padded;
2620 }
2621
2622 /*
2623 * Calculate out the start and number of virtual clusters we need to to CoW.
2624 *
2625 * cpos is vitual start cluster position we want to do CoW in a
2626 * file and write_len is the cluster length.
2627 * max_cpos is the place where we want to stop CoW intentionally.
2628 *
2629 * Normal we will start CoW from the beginning of extent record cotaining cpos.
2630 * We try to break up extents on boundaries of MAX_CONTIG_BYTES so that we
2631 * get good I/O from the resulting extent tree.
2632 */
2633 static int ocfs2_refcount_cal_cow_clusters(struct inode *inode,
2634 struct ocfs2_extent_list *el,
2635 u32 cpos,
2636 u32 write_len,
2637 u32 max_cpos,
2638 u32 *cow_start,
2639 u32 *cow_len)
2640 {
2641 int ret = 0;
2642 int tree_height = le16_to_cpu(el->l_tree_depth), i;
2643 struct buffer_head *eb_bh = NULL;
2644 struct ocfs2_extent_block *eb = NULL;
2645 struct ocfs2_extent_rec *rec;
2646 unsigned int want_clusters, rec_end = 0;
2647 int contig_clusters = ocfs2_cow_contig_clusters(inode->i_sb);
2648 int leaf_clusters;
2649
2650 BUG_ON(cpos + write_len > max_cpos);
2651
2652 if (tree_height > 0) {
2653 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, cpos, &eb_bh);
2654 if (ret) {
2655 mlog_errno(ret);
2656 goto out;
2657 }
2658
2659 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2660 el = &eb->h_list;
2661
2662 if (el->l_tree_depth) {
2663 ocfs2_error(inode->i_sb,
2664 "Inode %lu has non zero tree depth in "
2665 "leaf block %llu\n", inode->i_ino,
2666 (unsigned long long)eb_bh->b_blocknr);
2667 ret = -EROFS;
2668 goto out;
2669 }
2670 }
2671
2672 *cow_len = 0;
2673 for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
2674 rec = &el->l_recs[i];
2675
2676 if (ocfs2_is_empty_extent(rec)) {
2677 mlog_bug_on_msg(i != 0, "Inode %lu has empty record in "
2678 "index %d\n", inode->i_ino, i);
2679 continue;
2680 }
2681
2682 if (le32_to_cpu(rec->e_cpos) +
2683 le16_to_cpu(rec->e_leaf_clusters) <= cpos)
2684 continue;
2685
2686 if (*cow_len == 0) {
2687 /*
2688 * We should find a refcounted record in the
2689 * first pass.
2690 */
2691 BUG_ON(!(rec->e_flags & OCFS2_EXT_REFCOUNTED));
2692 *cow_start = le32_to_cpu(rec->e_cpos);
2693 }
2694
2695 /*
2696 * If we encounter a hole, a non-refcounted record or
2697 * pass the max_cpos, stop the search.
2698 */
2699 if ((!(rec->e_flags & OCFS2_EXT_REFCOUNTED)) ||
2700 (*cow_len && rec_end != le32_to_cpu(rec->e_cpos)) ||
2701 (max_cpos <= le32_to_cpu(rec->e_cpos)))
2702 break;
2703
2704 leaf_clusters = le16_to_cpu(rec->e_leaf_clusters);
2705 rec_end = le32_to_cpu(rec->e_cpos) + leaf_clusters;
2706 if (rec_end > max_cpos) {
2707 rec_end = max_cpos;
2708 leaf_clusters = rec_end - le32_to_cpu(rec->e_cpos);
2709 }
2710
2711 /*
2712 * How many clusters do we actually need from
2713 * this extent? First we see how many we actually
2714 * need to complete the write. If that's smaller
2715 * than contig_clusters, we try for contig_clusters.
2716 */
2717 if (!*cow_len)
2718 want_clusters = write_len;
2719 else
2720 want_clusters = (cpos + write_len) -
2721 (*cow_start + *cow_len);
2722 if (want_clusters < contig_clusters)
2723 want_clusters = contig_clusters;
2724
2725 /*
2726 * If the write does not cover the whole extent, we
2727 * need to calculate how we're going to split the extent.
2728 * We try to do it on contig_clusters boundaries.
2729 *
2730 * Any extent smaller than contig_clusters will be
2731 * CoWed in its entirety.
2732 */
2733 if (leaf_clusters <= contig_clusters)
2734 *cow_len += leaf_clusters;
2735 else if (*cow_len || (*cow_start == cpos)) {
2736 /*
2737 * This extent needs to be CoW'd from its
2738 * beginning, so all we have to do is compute
2739 * how many clusters to grab. We align
2740 * want_clusters to the edge of contig_clusters
2741 * to get better I/O.
2742 */
2743 want_clusters = ocfs2_cow_align_length(inode->i_sb,
2744 want_clusters);
2745
2746 if (leaf_clusters < want_clusters)
2747 *cow_len += leaf_clusters;
2748 else
2749 *cow_len += want_clusters;
2750 } else if ((*cow_start + contig_clusters) >=
2751 (cpos + write_len)) {
2752 /*
2753 * Breaking off contig_clusters at the front
2754 * of the extent will cover our write. That's
2755 * easy.
2756 */
2757 *cow_len = contig_clusters;
2758 } else if ((rec_end - cpos) <= contig_clusters) {
2759 /*
2760 * Breaking off contig_clusters at the tail of
2761 * this extent will cover cpos.
2762 */
2763 *cow_start = rec_end - contig_clusters;
2764 *cow_len = contig_clusters;
2765 } else if ((rec_end - cpos) <= want_clusters) {
2766 /*
2767 * While we can't fit the entire write in this
2768 * extent, we know that the write goes from cpos
2769 * to the end of the extent. Break that off.
2770 * We try to break it at some multiple of
2771 * contig_clusters from the front of the extent.
2772 * Failing that (ie, cpos is within
2773 * contig_clusters of the front), we'll CoW the
2774 * entire extent.
2775 */
2776 *cow_start = ocfs2_cow_align_start(inode->i_sb,
2777 *cow_start, cpos);
2778 *cow_len = rec_end - *cow_start;
2779 } else {
2780 /*
2781 * Ok, the entire write lives in the middle of
2782 * this extent. Let's try to slice the extent up
2783 * nicely. Optimally, our CoW region starts at
2784 * m*contig_clusters from the beginning of the
2785 * extent and goes for n*contig_clusters,
2786 * covering the entire write.
2787 */
2788 *cow_start = ocfs2_cow_align_start(inode->i_sb,
2789 *cow_start, cpos);
2790
2791 want_clusters = (cpos + write_len) - *cow_start;
2792 want_clusters = ocfs2_cow_align_length(inode->i_sb,
2793 want_clusters);
2794 if (*cow_start + want_clusters <= rec_end)
2795 *cow_len = want_clusters;
2796 else
2797 *cow_len = rec_end - *cow_start;
2798 }
2799
2800 /* Have we covered our entire write yet? */
2801 if ((*cow_start + *cow_len) >= (cpos + write_len))
2802 break;
2803
2804 /*
2805 * If we reach the end of the extent block and don't get enough
2806 * clusters, continue with the next extent block if possible.
2807 */
2808 if (i + 1 == le16_to_cpu(el->l_next_free_rec) &&
2809 eb && eb->h_next_leaf_blk) {
2810 brelse(eb_bh);
2811 eb_bh = NULL;
2812
2813 ret = ocfs2_read_extent_block(INODE_CACHE(inode),
2814 le64_to_cpu(eb->h_next_leaf_blk),
2815 &eb_bh);
2816 if (ret) {
2817 mlog_errno(ret);
2818 goto out;
2819 }
2820
2821 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2822 el = &eb->h_list;
2823 i = -1;
2824 }
2825 }
2826
2827 out:
2828 brelse(eb_bh);
2829 return ret;
2830 }
2831
2832 /*
2833 * Prepare meta_ac, data_ac and calculate credits when we want to add some
2834 * num_clusters in data_tree "et" and change the refcount for the old
2835 * clusters(starting form p_cluster) in the refcount tree.
2836 *
2837 * Note:
2838 * 1. since we may split the old tree, so we at most will need num_clusters + 2
2839 * more new leaf records.
2840 * 2. In some case, we may not need to reserve new clusters(e.g, reflink), so
2841 * just give data_ac = NULL.
2842 */
2843 static int ocfs2_lock_refcount_allocators(struct super_block *sb,
2844 u32 p_cluster, u32 num_clusters,
2845 struct ocfs2_extent_tree *et,
2846 struct ocfs2_caching_info *ref_ci,
2847 struct buffer_head *ref_root_bh,
2848 struct ocfs2_alloc_context **meta_ac,
2849 struct ocfs2_alloc_context **data_ac,
2850 int *credits)
2851 {
2852 int ret = 0, meta_add = 0;
2853 int num_free_extents = ocfs2_num_free_extents(OCFS2_SB(sb), et);
2854
2855 if (num_free_extents < 0) {
2856 ret = num_free_extents;
2857 mlog_errno(ret);
2858 goto out;
2859 }
2860
2861 if (num_free_extents < num_clusters + 2)
2862 meta_add =
2863 ocfs2_extend_meta_needed(et->et_root_el);
2864
2865 *credits += ocfs2_calc_extend_credits(sb, et->et_root_el,
2866 num_clusters + 2);
2867
2868 ret = ocfs2_calc_refcount_meta_credits(sb, ref_ci, ref_root_bh,
2869 p_cluster, num_clusters,
2870 &meta_add, credits);
2871 if (ret) {
2872 mlog_errno(ret);
2873 goto out;
2874 }
2875
2876 mlog(0, "reserve new metadata %d, clusters %u, credits = %d\n",
2877 meta_add, num_clusters, *credits);
2878 ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(sb), meta_add,
2879 meta_ac);
2880 if (ret) {
2881 mlog_errno(ret);
2882 goto out;
2883 }
2884
2885 if (data_ac) {
2886 ret = ocfs2_reserve_clusters(OCFS2_SB(sb), num_clusters,
2887 data_ac);
2888 if (ret)
2889 mlog_errno(ret);
2890 }
2891
2892 out:
2893 if (ret) {
2894 if (*meta_ac) {
2895 ocfs2_free_alloc_context(*meta_ac);
2896 *meta_ac = NULL;
2897 }
2898 }
2899
2900 return ret;
2901 }
2902
2903 static int ocfs2_clear_cow_buffer(handle_t *handle, struct buffer_head *bh)
2904 {
2905 BUG_ON(buffer_dirty(bh));
2906
2907 clear_buffer_mapped(bh);
2908
2909 return 0;
2910 }
2911
2912 static int ocfs2_duplicate_clusters_by_page(handle_t *handle,
2913 struct ocfs2_cow_context *context,
2914 u32 cpos, u32 old_cluster,
2915 u32 new_cluster, u32 new_len)
2916 {
2917 int ret = 0, partial;
2918 struct ocfs2_caching_info *ci = context->data_et.et_ci;
2919 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
2920 u64 new_block = ocfs2_clusters_to_blocks(sb, new_cluster);
2921 struct page *page;
2922 pgoff_t page_index;
2923 unsigned int from, to;
2924 loff_t offset, end, map_end;
2925 struct address_space *mapping = context->inode->i_mapping;
2926
2927 mlog(0, "old_cluster %u, new %u, len %u at offset %u\n", old_cluster,
2928 new_cluster, new_len, cpos);
2929
2930 offset = ((loff_t)cpos) << OCFS2_SB(sb)->s_clustersize_bits;
2931 end = offset + (new_len << OCFS2_SB(sb)->s_clustersize_bits);
2932 /*
2933 * We only duplicate pages until we reach the page contains i_size - 1.
2934 * So trim 'end' to i_size.
2935 */
2936 if (end > i_size_read(context->inode))
2937 end = i_size_read(context->inode);
2938
2939 while (offset < end) {
2940 page_index = offset >> PAGE_CACHE_SHIFT;
2941 map_end = ((loff_t)page_index + 1) << PAGE_CACHE_SHIFT;
2942 if (map_end > end)
2943 map_end = end;
2944
2945 /* from, to is the offset within the page. */
2946 from = offset & (PAGE_CACHE_SIZE - 1);
2947 to = PAGE_CACHE_SIZE;
2948 if (map_end & (PAGE_CACHE_SIZE - 1))
2949 to = map_end & (PAGE_CACHE_SIZE - 1);
2950
2951 page = find_or_create_page(mapping, page_index, GFP_NOFS);
2952
2953 /*
2954 * In case PAGE_CACHE_SIZE <= CLUSTER_SIZE, This page
2955 * can't be dirtied before we CoW it out.
2956 */
2957 if (PAGE_CACHE_SIZE <= OCFS2_SB(sb)->s_clustersize)
2958 BUG_ON(PageDirty(page));
2959
2960 if (!PageUptodate(page)) {
2961 ret = block_read_full_page(page, ocfs2_get_block);
2962 if (ret) {
2963 mlog_errno(ret);
2964 goto unlock;
2965 }
2966 lock_page(page);
2967 }
2968
2969 if (page_has_buffers(page)) {
2970 ret = walk_page_buffers(handle, page_buffers(page),
2971 from, to, &partial,
2972 ocfs2_clear_cow_buffer);
2973 if (ret) {
2974 mlog_errno(ret);
2975 goto unlock;
2976 }
2977 }
2978
2979 ocfs2_map_and_dirty_page(context->inode,
2980 handle, from, to,
2981 page, 0, &new_block);
2982 mark_page_accessed(page);
2983 unlock:
2984 unlock_page(page);
2985 page_cache_release(page);
2986 page = NULL;
2987 offset = map_end;
2988 if (ret)
2989 break;
2990 }
2991
2992 return ret;
2993 }
2994
2995 static int ocfs2_duplicate_clusters_by_jbd(handle_t *handle,
2996 struct ocfs2_cow_context *context,
2997 u32 cpos, u32 old_cluster,
2998 u32 new_cluster, u32 new_len)
2999 {
3000 int ret = 0;
3001 struct super_block *sb = context->inode->i_sb;
3002 struct ocfs2_caching_info *ci = context->data_et.et_ci;
3003 int i, blocks = ocfs2_clusters_to_blocks(sb, new_len);
3004 u64 old_block = ocfs2_clusters_to_blocks(sb, old_cluster);
3005 u64 new_block = ocfs2_clusters_to_blocks(sb, new_cluster);
3006 struct ocfs2_super *osb = OCFS2_SB(sb);
3007 struct buffer_head *old_bh = NULL;
3008 struct buffer_head *new_bh = NULL;
3009
3010 mlog(0, "old_cluster %u, new %u, len %u\n", old_cluster,
3011 new_cluster, new_len);
3012
3013 for (i = 0; i < blocks; i++, old_block++, new_block++) {
3014 new_bh = sb_getblk(osb->sb, new_block);
3015 if (new_bh == NULL) {
3016 ret = -EIO;
3017 mlog_errno(ret);
3018 break;
3019 }
3020
3021 ocfs2_set_new_buffer_uptodate(ci, new_bh);
3022
3023 ret = ocfs2_read_block(ci, old_block, &old_bh, NULL);
3024 if (ret) {
3025 mlog_errno(ret);
3026 break;
3027 }
3028
3029 ret = ocfs2_journal_access(handle, ci, new_bh,
3030 OCFS2_JOURNAL_ACCESS_CREATE);
3031 if (ret) {
3032 mlog_errno(ret);
3033 break;
3034 }
3035
3036 memcpy(new_bh->b_data, old_bh->b_data, sb->s_blocksize);
3037 ocfs2_journal_dirty(handle, new_bh);
3038
3039 brelse(new_bh);
3040 brelse(old_bh);
3041 new_bh = NULL;
3042 old_bh = NULL;
3043 }
3044
3045 brelse(new_bh);
3046 brelse(old_bh);
3047 return ret;
3048 }
3049
3050 static int ocfs2_clear_ext_refcount(handle_t *handle,
3051 struct ocfs2_extent_tree *et,
3052 u32 cpos, u32 p_cluster, u32 len,
3053 unsigned int ext_flags,
3054 struct ocfs2_alloc_context *meta_ac,
3055 struct ocfs2_cached_dealloc_ctxt *dealloc)
3056 {
3057 int ret, index;
3058 struct ocfs2_extent_rec replace_rec;
3059 struct ocfs2_path *path = NULL;
3060 struct ocfs2_extent_list *el;
3061 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
3062 u64 ino = ocfs2_metadata_cache_owner(et->et_ci);
3063
3064 mlog(0, "inode %llu cpos %u, len %u, p_cluster %u, ext_flags %u\n",
3065 (unsigned long long)ino, cpos, len, p_cluster, ext_flags);
3066
3067 memset(&replace_rec, 0, sizeof(replace_rec));
3068 replace_rec.e_cpos = cpu_to_le32(cpos);
3069 replace_rec.e_leaf_clusters = cpu_to_le16(len);
3070 replace_rec.e_blkno = cpu_to_le64(ocfs2_clusters_to_blocks(sb,
3071 p_cluster));
3072 replace_rec.e_flags = ext_flags;
3073 replace_rec.e_flags &= ~OCFS2_EXT_REFCOUNTED;
3074
3075 path = ocfs2_new_path_from_et(et);
3076 if (!path) {
3077 ret = -ENOMEM;
3078 mlog_errno(ret);
3079 goto out;
3080 }
3081
3082 ret = ocfs2_find_path(et->et_ci, path, cpos);
3083 if (ret) {
3084 mlog_errno(ret);
3085 goto out;
3086 }
3087
3088 el = path_leaf_el(path);
3089
3090 index = ocfs2_search_extent_list(el, cpos);
3091 if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) {
3092 ocfs2_error(sb,
3093 "Inode %llu has an extent at cpos %u which can no "
3094 "longer be found.\n",
3095 (unsigned long long)ino, cpos);
3096 ret = -EROFS;
3097 goto out;
3098 }
3099
3100 ret = ocfs2_split_extent(handle, et, path, index,
3101 &replace_rec, meta_ac, dealloc);
3102 if (ret)
3103 mlog_errno(ret);
3104
3105 out:
3106 ocfs2_free_path(path);
3107 return ret;
3108 }
3109
3110 static int ocfs2_replace_clusters(handle_t *handle,
3111 struct ocfs2_cow_context *context,
3112 u32 cpos, u32 old,
3113 u32 new, u32 len,
3114 unsigned int ext_flags)
3115 {
3116 int ret;
3117 struct ocfs2_caching_info *ci = context->data_et.et_ci;
3118 u64 ino = ocfs2_metadata_cache_owner(ci);
3119
3120 mlog(0, "inode %llu, cpos %u, old %u, new %u, len %u, ext_flags %u\n",
3121 (unsigned long long)ino, cpos, old, new, len, ext_flags);
3122
3123 /*If the old clusters is unwritten, no need to duplicate. */
3124 if (!(ext_flags & OCFS2_EXT_UNWRITTEN)) {
3125 ret = context->cow_duplicate_clusters(handle, context, cpos,
3126 old, new, len);
3127 if (ret) {
3128 mlog_errno(ret);
3129 goto out;
3130 }
3131 }
3132
3133 ret = ocfs2_clear_ext_refcount(handle, &context->data_et,
3134 cpos, new, len, ext_flags,
3135 context->meta_ac, &context->dealloc);
3136 if (ret)
3137 mlog_errno(ret);
3138 out:
3139 return ret;
3140 }
3141
3142 static int ocfs2_cow_sync_writeback(struct super_block *sb,
3143 struct ocfs2_cow_context *context,
3144 u32 cpos, u32 num_clusters)
3145 {
3146 int ret = 0;
3147 loff_t offset, end, map_end;
3148 pgoff_t page_index;
3149 struct page *page;
3150
3151 if (ocfs2_should_order_data(context->inode))
3152 return 0;
3153
3154 offset = ((loff_t)cpos) << OCFS2_SB(sb)->s_clustersize_bits;
3155 end = offset + (num_clusters << OCFS2_SB(sb)->s_clustersize_bits);
3156
3157 ret = filemap_fdatawrite_range(context->inode->i_mapping,
3158 offset, end - 1);
3159 if (ret < 0) {
3160 mlog_errno(ret);
3161 return ret;
3162 }
3163
3164 while (offset < end) {
3165 page_index = offset >> PAGE_CACHE_SHIFT;
3166 map_end = ((loff_t)page_index + 1) << PAGE_CACHE_SHIFT;
3167 if (map_end > end)
3168 map_end = end;
3169
3170 page = find_or_create_page(context->inode->i_mapping,
3171 page_index, GFP_NOFS);
3172 BUG_ON(!page);
3173
3174 wait_on_page_writeback(page);
3175 if (PageError(page)) {
3176 ret = -EIO;
3177 mlog_errno(ret);
3178 } else
3179 mark_page_accessed(page);
3180
3181 unlock_page(page);
3182 page_cache_release(page);
3183 page = NULL;
3184 offset = map_end;
3185 if (ret)
3186 break;
3187 }
3188
3189 return ret;
3190 }
3191
3192 static int ocfs2_di_get_clusters(struct ocfs2_cow_context *context,
3193 u32 v_cluster, u32 *p_cluster,
3194 u32 *num_clusters,
3195 unsigned int *extent_flags)
3196 {
3197 return ocfs2_get_clusters(context->inode, v_cluster, p_cluster,
3198 num_clusters, extent_flags);
3199 }
3200
3201 static int ocfs2_make_clusters_writable(struct super_block *sb,
3202 struct ocfs2_cow_context *context,
3203 u32 cpos, u32 p_cluster,
3204 u32 num_clusters, unsigned int e_flags)
3205 {
3206 int ret, delete, index, credits = 0;
3207 u32 new_bit, new_len;
3208 unsigned int set_len;
3209 struct ocfs2_super *osb = OCFS2_SB(sb);
3210 handle_t *handle;
3211 struct buffer_head *ref_leaf_bh = NULL;
3212 struct ocfs2_caching_info *ref_ci = &context->ref_tree->rf_ci;
3213 struct ocfs2_refcount_rec rec;
3214
3215 mlog(0, "cpos %u, p_cluster %u, num_clusters %u, e_flags %u\n",
3216 cpos, p_cluster, num_clusters, e_flags);
3217
3218 ret = ocfs2_lock_refcount_allocators(sb, p_cluster, num_clusters,
3219 &context->data_et,
3220 ref_ci,
3221 context->ref_root_bh,
3222 &context->meta_ac,
3223 &context->data_ac, &credits);
3224 if (ret) {
3225 mlog_errno(ret);
3226 return ret;
3227 }
3228
3229 if (context->post_refcount)
3230 credits += context->post_refcount->credits;
3231
3232 credits += context->extra_credits;
3233 handle = ocfs2_start_trans(osb, credits);
3234 if (IS_ERR(handle)) {
3235 ret = PTR_ERR(handle);
3236 mlog_errno(ret);
3237 goto out;
3238 }
3239
3240 while (num_clusters) {
3241 ret = ocfs2_get_refcount_rec(ref_ci, context->ref_root_bh,
3242 p_cluster, num_clusters,
3243 &rec, &index, &ref_leaf_bh);
3244 if (ret) {
3245 mlog_errno(ret);
3246 goto out_commit;
3247 }
3248
3249 BUG_ON(!rec.r_refcount);
3250 set_len = min((u64)p_cluster + num_clusters,
3251 le64_to_cpu(rec.r_cpos) +
3252 le32_to_cpu(rec.r_clusters)) - p_cluster;
3253
3254 /*
3255 * There are many different situation here.
3256 * 1. If refcount == 1, remove the flag and don't COW.
3257 * 2. If refcount > 1, allocate clusters.
3258 * Here we may not allocate r_len once at a time, so continue
3259 * until we reach num_clusters.
3260 */
3261 if (le32_to_cpu(rec.r_refcount) == 1) {
3262 delete = 0;
3263 ret = ocfs2_clear_ext_refcount(handle,
3264 &context->data_et,
3265 cpos, p_cluster,
3266 set_len, e_flags,
3267 context->meta_ac,
3268 &context->dealloc);
3269 if (ret) {
3270 mlog_errno(ret);
3271 goto out_commit;
3272 }
3273 } else {
3274 delete = 1;
3275
3276 ret = __ocfs2_claim_clusters(handle,
3277 context->data_ac,
3278 1, set_len,
3279 &new_bit, &new_len);
3280 if (ret) {
3281 mlog_errno(ret);
3282 goto out_commit;
3283 }
3284
3285 ret = ocfs2_replace_clusters(handle, context,
3286 cpos, p_cluster, new_bit,
3287 new_len, e_flags);
3288 if (ret) {
3289 mlog_errno(ret);
3290 goto out_commit;
3291 }
3292 set_len = new_len;
3293 }
3294
3295 ret = __ocfs2_decrease_refcount(handle, ref_ci,
3296 context->ref_root_bh,
3297 p_cluster, set_len,
3298 context->meta_ac,
3299 &context->dealloc, delete);
3300 if (ret) {
3301 mlog_errno(ret);
3302 goto out_commit;
3303 }
3304
3305 cpos += set_len;
3306 p_cluster += set_len;
3307 num_clusters -= set_len;
3308 brelse(ref_leaf_bh);
3309 ref_leaf_bh = NULL;
3310 }
3311
3312 /* handle any post_cow action. */
3313 if (context->post_refcount && context->post_refcount->func) {
3314 ret = context->post_refcount->func(context->inode, handle,
3315 context->post_refcount->para);
3316 if (ret) {
3317 mlog_errno(ret);
3318 goto out_commit;
3319 }
3320 }
3321
3322 /*
3323 * Here we should write the new page out first if we are
3324 * in write-back mode.
3325 */
3326 if (context->get_clusters == ocfs2_di_get_clusters) {
3327 ret = ocfs2_cow_sync_writeback(sb, context, cpos, num_clusters);
3328 if (ret)
3329 mlog_errno(ret);
3330 }
3331
3332 out_commit:
3333 ocfs2_commit_trans(osb, handle);
3334
3335 out:
3336 if (context->data_ac) {
3337 ocfs2_free_alloc_context(context->data_ac);
3338 context->data_ac = NULL;
3339 }
3340 if (context->meta_ac) {
3341 ocfs2_free_alloc_context(context->meta_ac);
3342 context->meta_ac = NULL;
3343 }
3344 brelse(ref_leaf_bh);
3345
3346 return ret;
3347 }
3348
3349 static int ocfs2_replace_cow(struct ocfs2_cow_context *context)
3350 {
3351 int ret = 0;
3352 struct inode *inode = context->inode;
3353 u32 cow_start = context->cow_start, cow_len = context->cow_len;
3354 u32 p_cluster, num_clusters;
3355 unsigned int ext_flags;
3356 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3357
3358 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
3359 ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
3360 "tree, but the feature bit is not set in the "
3361 "super block.", inode->i_ino);
3362 return -EROFS;
3363 }
3364
3365 ocfs2_init_dealloc_ctxt(&context->dealloc);
3366
3367 while (cow_len) {
3368 ret = context->get_clusters(context, cow_start, &p_cluster,
3369 &num_clusters, &ext_flags);
3370 if (ret) {
3371 mlog_errno(ret);
3372 break;
3373 }
3374
3375 BUG_ON(!(ext_flags & OCFS2_EXT_REFCOUNTED));
3376
3377 if (cow_len < num_clusters)
3378 num_clusters = cow_len;
3379
3380 ret = ocfs2_make_clusters_writable(inode->i_sb, context,
3381 cow_start, p_cluster,
3382 num_clusters, ext_flags);
3383 if (ret) {
3384 mlog_errno(ret);
3385 break;
3386 }
3387
3388 cow_len -= num_clusters;
3389 cow_start += num_clusters;
3390 }
3391
3392 if (ocfs2_dealloc_has_cluster(&context->dealloc)) {
3393 ocfs2_schedule_truncate_log_flush(osb, 1);
3394 ocfs2_run_deallocs(osb, &context->dealloc);
3395 }
3396
3397 return ret;
3398 }
3399
3400 /*
3401 * Starting at cpos, try to CoW write_len clusters. Don't CoW
3402 * past max_cpos. This will stop when it runs into a hole or an
3403 * unrefcounted extent.
3404 */
3405 static int ocfs2_refcount_cow_hunk(struct inode *inode,
3406 struct buffer_head *di_bh,
3407 u32 cpos, u32 write_len, u32 max_cpos)
3408 {
3409 int ret;
3410 u32 cow_start = 0, cow_len = 0;
3411 struct ocfs2_inode_info *oi = OCFS2_I(inode);
3412 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3413 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3414 struct buffer_head *ref_root_bh = NULL;
3415 struct ocfs2_refcount_tree *ref_tree;
3416 struct ocfs2_cow_context *context = NULL;
3417
3418 BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
3419
3420 ret = ocfs2_refcount_cal_cow_clusters(inode, &di->id2.i_list,
3421 cpos, write_len, max_cpos,
3422 &cow_start, &cow_len);
3423 if (ret) {
3424 mlog_errno(ret);
3425 goto out;
3426 }
3427
3428 mlog(0, "CoW inode %lu, cpos %u, write_len %u, cow_start %u, "
3429 "cow_len %u\n", inode->i_ino,
3430 cpos, write_len, cow_start, cow_len);
3431
3432 BUG_ON(cow_len == 0);
3433
3434 context = kzalloc(sizeof(struct ocfs2_cow_context), GFP_NOFS);
3435 if (!context) {
3436 ret = -ENOMEM;
3437 mlog_errno(ret);
3438 goto out;
3439 }
3440
3441 ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
3442 1, &ref_tree, &ref_root_bh);
3443 if (ret) {
3444 mlog_errno(ret);
3445 goto out;
3446 }
3447
3448 context->inode = inode;
3449 context->cow_start = cow_start;
3450 context->cow_len = cow_len;
3451 context->ref_tree = ref_tree;
3452 context->ref_root_bh = ref_root_bh;
3453 context->cow_duplicate_clusters = ocfs2_duplicate_clusters_by_page;
3454 context->get_clusters = ocfs2_di_get_clusters;
3455
3456 ocfs2_init_dinode_extent_tree(&context->data_et,
3457 INODE_CACHE(inode), di_bh);
3458
3459 ret = ocfs2_replace_cow(context);
3460 if (ret)
3461 mlog_errno(ret);
3462
3463 /*
3464 * truncate the extent map here since no matter whether we meet with
3465 * any error during the action, we shouldn't trust cached extent map
3466 * any more.
3467 */
3468 ocfs2_extent_map_trunc(inode, cow_start);
3469
3470 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3471 brelse(ref_root_bh);
3472 out:
3473 kfree(context);
3474 return ret;
3475 }
3476
3477 /*
3478 * CoW any and all clusters between cpos and cpos+write_len.
3479 * Don't CoW past max_cpos. If this returns successfully, all
3480 * clusters between cpos and cpos+write_len are safe to modify.
3481 */
3482 int ocfs2_refcount_cow(struct inode *inode,
3483 struct buffer_head *di_bh,
3484 u32 cpos, u32 write_len, u32 max_cpos)
3485 {
3486 int ret = 0;
3487 u32 p_cluster, num_clusters;
3488 unsigned int ext_flags;
3489
3490 while (write_len) {
3491 ret = ocfs2_get_clusters(inode, cpos, &p_cluster,
3492 &num_clusters, &ext_flags);
3493 if (ret) {
3494 mlog_errno(ret);
3495 break;
3496 }
3497
3498 if (write_len < num_clusters)
3499 num_clusters = write_len;
3500
3501 if (ext_flags & OCFS2_EXT_REFCOUNTED) {
3502 ret = ocfs2_refcount_cow_hunk(inode, di_bh, cpos,
3503 num_clusters, max_cpos);
3504 if (ret) {
3505 mlog_errno(ret);
3506 break;
3507 }
3508 }
3509
3510 write_len -= num_clusters;
3511 cpos += num_clusters;
3512 }
3513
3514 return ret;
3515 }
3516
3517 static int ocfs2_xattr_value_get_clusters(struct ocfs2_cow_context *context,
3518 u32 v_cluster, u32 *p_cluster,
3519 u32 *num_clusters,
3520 unsigned int *extent_flags)
3521 {
3522 struct inode *inode = context->inode;
3523 struct ocfs2_xattr_value_root *xv = context->cow_object;
3524
3525 return ocfs2_xattr_get_clusters(inode, v_cluster, p_cluster,
3526 num_clusters, &xv->xr_list,
3527 extent_flags);
3528 }
3529
3530 /*
3531 * Given a xattr value root, calculate the most meta/credits we need for
3532 * refcount tree change if we truncate it to 0.
3533 */
3534 int ocfs2_refcounted_xattr_delete_need(struct inode *inode,
3535 struct ocfs2_caching_info *ref_ci,
3536 struct buffer_head *ref_root_bh,
3537 struct ocfs2_xattr_value_root *xv,
3538 int *meta_add, int *credits)
3539 {
3540 int ret = 0, index, ref_blocks = 0;
3541 u32 p_cluster, num_clusters;
3542 u32 cpos = 0, clusters = le32_to_cpu(xv->xr_clusters);
3543 struct ocfs2_refcount_block *rb;
3544 struct ocfs2_refcount_rec rec;
3545 struct buffer_head *ref_leaf_bh = NULL;
3546
3547 while (cpos < clusters) {
3548 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
3549 &num_clusters, &xv->xr_list,
3550 NULL);
3551 if (ret) {
3552 mlog_errno(ret);
3553 goto out;
3554 }
3555
3556 cpos += num_clusters;
3557
3558 while (num_clusters) {
3559 ret = ocfs2_get_refcount_rec(ref_ci, ref_root_bh,
3560 p_cluster, num_clusters,
3561 &rec, &index,
3562 &ref_leaf_bh);
3563 if (ret) {
3564 mlog_errno(ret);
3565 goto out;
3566 }
3567
3568 BUG_ON(!rec.r_refcount);
3569
3570 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
3571
3572 /*
3573 * We really don't know whether the other clusters is in
3574 * this refcount block or not, so just take the worst
3575 * case that all the clusters are in this block and each
3576 * one will split a refcount rec, so totally we need
3577 * clusters * 2 new refcount rec.
3578 */
3579 if (le64_to_cpu(rb->rf_records.rl_used) + clusters * 2 >
3580 le16_to_cpu(rb->rf_records.rl_count))
3581 ref_blocks++;
3582
3583 *credits += 1;
3584 brelse(ref_leaf_bh);
3585 ref_leaf_bh = NULL;
3586
3587 if (num_clusters <= le32_to_cpu(rec.r_clusters))
3588 break;
3589 else
3590 num_clusters -= le32_to_cpu(rec.r_clusters);
3591 p_cluster += num_clusters;
3592 }
3593 }
3594
3595 *meta_add += ref_blocks;
3596 if (!ref_blocks)
3597 goto out;
3598
3599 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
3600 if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
3601 *credits += OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
3602 else {
3603 struct ocfs2_extent_tree et;
3604
3605 ocfs2_init_refcount_extent_tree(&et, ref_ci, ref_root_bh);
3606 *credits += ocfs2_calc_extend_credits(inode->i_sb,
3607 et.et_root_el,
3608 ref_blocks);
3609 }
3610
3611 out:
3612 brelse(ref_leaf_bh);
3613 return ret;
3614 }
3615
3616 /*
3617 * Do CoW for xattr.
3618 */
3619 int ocfs2_refcount_cow_xattr(struct inode *inode,
3620 struct ocfs2_dinode *di,
3621 struct ocfs2_xattr_value_buf *vb,
3622 struct ocfs2_refcount_tree *ref_tree,
3623 struct buffer_head *ref_root_bh,
3624 u32 cpos, u32 write_len,
3625 struct ocfs2_post_refcount *post)
3626 {
3627 int ret;
3628 struct ocfs2_xattr_value_root *xv = vb->vb_xv;
3629 struct ocfs2_inode_info *oi = OCFS2_I(inode);
3630 struct ocfs2_cow_context *context = NULL;
3631 u32 cow_start, cow_len;
3632
3633 BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
3634
3635 ret = ocfs2_refcount_cal_cow_clusters(inode, &xv->xr_list,
3636 cpos, write_len, UINT_MAX,
3637 &cow_start, &cow_len);
3638 if (ret) {
3639 mlog_errno(ret);
3640 goto out;
3641 }
3642
3643 BUG_ON(cow_len == 0);
3644
3645 context = kzalloc(sizeof(struct ocfs2_cow_context), GFP_NOFS);
3646 if (!context) {
3647 ret = -ENOMEM;
3648 mlog_errno(ret);
3649 goto out;
3650 }
3651
3652 context->inode = inode;
3653 context->cow_start = cow_start;
3654 context->cow_len = cow_len;
3655 context->ref_tree = ref_tree;
3656 context->ref_root_bh = ref_root_bh;;
3657 context->cow_object = xv;
3658
3659 context->cow_duplicate_clusters = ocfs2_duplicate_clusters_by_jbd;
3660 /* We need the extra credits for duplicate_clusters by jbd. */
3661 context->extra_credits =
3662 ocfs2_clusters_to_blocks(inode->i_sb, 1) * cow_len;
3663 context->get_clusters = ocfs2_xattr_value_get_clusters;
3664 context->post_refcount = post;
3665
3666 ocfs2_init_xattr_value_extent_tree(&context->data_et,
3667 INODE_CACHE(inode), vb);
3668
3669 ret = ocfs2_replace_cow(context);
3670 if (ret)
3671 mlog_errno(ret);
3672
3673 out:
3674 kfree(context);
3675 return ret;
3676 }
3677
3678 /*
3679 * Insert a new extent into refcount tree and mark a extent rec
3680 * as refcounted in the dinode tree.
3681 */
3682 int ocfs2_add_refcount_flag(struct inode *inode,
3683 struct ocfs2_extent_tree *data_et,
3684 struct ocfs2_caching_info *ref_ci,
3685 struct buffer_head *ref_root_bh,
3686 u32 cpos, u32 p_cluster, u32 num_clusters,
3687 struct ocfs2_cached_dealloc_ctxt *dealloc,
3688 struct ocfs2_post_refcount *post)
3689 {
3690 int ret;
3691 handle_t *handle;
3692 int credits = 1, ref_blocks = 0;
3693 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3694 struct ocfs2_alloc_context *meta_ac = NULL;
3695
3696 ret = ocfs2_calc_refcount_meta_credits(inode->i_sb,
3697 ref_ci, ref_root_bh,
3698 p_cluster, num_clusters,
3699 &ref_blocks, &credits);
3700 if (ret) {
3701 mlog_errno(ret);
3702 goto out;
3703 }
3704
3705 mlog(0, "reserve new metadata %d, credits = %d\n",
3706 ref_blocks, credits);
3707
3708 if (ref_blocks) {
3709 ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
3710 ref_blocks, &meta_ac);
3711 if (ret) {
3712 mlog_errno(ret);
3713 goto out;
3714 }
3715 }
3716
3717 if (post)
3718 credits += post->credits;
3719
3720 handle = ocfs2_start_trans(osb, credits);
3721 if (IS_ERR(handle)) {
3722 ret = PTR_ERR(handle);
3723 mlog_errno(ret);
3724 goto out;
3725 }
3726
3727 ret = ocfs2_mark_extent_refcounted(inode, data_et, handle,
3728 cpos, num_clusters, p_cluster,
3729 meta_ac, dealloc);
3730 if (ret) {
3731 mlog_errno(ret);
3732 goto out_commit;
3733 }
3734
3735 ret = __ocfs2_increase_refcount(handle, ref_ci, ref_root_bh,
3736 p_cluster, num_clusters, 0,
3737 meta_ac, dealloc);
3738 if (ret) {
3739 mlog_errno(ret);
3740 goto out_commit;
3741 }
3742
3743 if (post && post->func) {
3744 ret = post->func(inode, handle, post->para);
3745 if (ret)
3746 mlog_errno(ret);
3747 }
3748
3749 out_commit:
3750 ocfs2_commit_trans(osb, handle);
3751 out:
3752 if (meta_ac)
3753 ocfs2_free_alloc_context(meta_ac);
3754 return ret;
3755 }
3756
3757 static int ocfs2_change_ctime(struct inode *inode,
3758 struct buffer_head *di_bh)
3759 {
3760 int ret;
3761 handle_t *handle;
3762 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3763
3764 handle = ocfs2_start_trans(OCFS2_SB(inode->i_sb),
3765 OCFS2_INODE_UPDATE_CREDITS);
3766 if (IS_ERR(handle)) {
3767 ret = PTR_ERR(handle);
3768 mlog_errno(ret);
3769 goto out;
3770 }
3771
3772 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
3773 OCFS2_JOURNAL_ACCESS_WRITE);
3774 if (ret) {
3775 mlog_errno(ret);
3776 goto out_commit;
3777 }
3778
3779 inode->i_ctime = CURRENT_TIME;
3780 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
3781 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
3782
3783 ocfs2_journal_dirty(handle, di_bh);
3784
3785 out_commit:
3786 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
3787 out:
3788 return ret;
3789 }
3790
3791 static int ocfs2_attach_refcount_tree(struct inode *inode,
3792 struct buffer_head *di_bh)
3793 {
3794 int ret, data_changed = 0;
3795 struct buffer_head *ref_root_bh = NULL;
3796 struct ocfs2_inode_info *oi = OCFS2_I(inode);
3797 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3798 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3799 struct ocfs2_refcount_tree *ref_tree;
3800 unsigned int ext_flags;
3801 loff_t size;
3802 u32 cpos, num_clusters, clusters, p_cluster;
3803 struct ocfs2_cached_dealloc_ctxt dealloc;
3804 struct ocfs2_extent_tree di_et;
3805
3806 ocfs2_init_dealloc_ctxt(&dealloc);
3807
3808 if (!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL)) {
3809 ret = ocfs2_create_refcount_tree(inode, di_bh);
3810 if (ret) {
3811 mlog_errno(ret);
3812 goto out;
3813 }
3814 }
3815
3816 BUG_ON(!di->i_refcount_loc);
3817 ret = ocfs2_lock_refcount_tree(osb,
3818 le64_to_cpu(di->i_refcount_loc), 1,
3819 &ref_tree, &ref_root_bh);
3820 if (ret) {
3821 mlog_errno(ret);
3822 goto out;
3823 }
3824
3825 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
3826 goto attach_xattr;
3827
3828 ocfs2_init_dinode_extent_tree(&di_et, INODE_CACHE(inode), di_bh);
3829
3830 size = i_size_read(inode);
3831 clusters = ocfs2_clusters_for_bytes(inode->i_sb, size);
3832
3833 cpos = 0;
3834 while (cpos < clusters) {
3835 ret = ocfs2_get_clusters(inode, cpos, &p_cluster,
3836 &num_clusters, &ext_flags);
3837
3838 if (p_cluster && !(ext_flags & OCFS2_EXT_REFCOUNTED)) {
3839 ret = ocfs2_add_refcount_flag(inode, &di_et,
3840 &ref_tree->rf_ci,
3841 ref_root_bh, cpos,
3842 p_cluster, num_clusters,
3843 &dealloc, NULL);
3844 if (ret) {
3845 mlog_errno(ret);
3846 goto unlock;
3847 }
3848
3849 data_changed = 1;
3850 }
3851 cpos += num_clusters;
3852 }
3853
3854 attach_xattr:
3855 if (oi->ip_dyn_features & OCFS2_HAS_XATTR_FL) {
3856 ret = ocfs2_xattr_attach_refcount_tree(inode, di_bh,
3857 &ref_tree->rf_ci,
3858 ref_root_bh,
3859 &dealloc);
3860 if (ret) {
3861 mlog_errno(ret);
3862 goto unlock;
3863 }
3864 }
3865
3866 if (data_changed) {
3867 ret = ocfs2_change_ctime(inode, di_bh);
3868 if (ret)
3869 mlog_errno(ret);
3870 }
3871
3872 unlock:
3873 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3874 brelse(ref_root_bh);
3875
3876 if (!ret && ocfs2_dealloc_has_cluster(&dealloc)) {
3877 ocfs2_schedule_truncate_log_flush(osb, 1);
3878 ocfs2_run_deallocs(osb, &dealloc);
3879 }
3880 out:
3881 /*
3882 * Empty the extent map so that we may get the right extent
3883 * record from the disk.
3884 */
3885 ocfs2_extent_map_trunc(inode, 0);
3886
3887 return ret;
3888 }
3889
3890 static int ocfs2_add_refcounted_extent(struct inode *inode,
3891 struct ocfs2_extent_tree *et,
3892 struct ocfs2_caching_info *ref_ci,
3893 struct buffer_head *ref_root_bh,
3894 u32 cpos, u32 p_cluster, u32 num_clusters,
3895 unsigned int ext_flags,
3896 struct ocfs2_cached_dealloc_ctxt *dealloc)
3897 {
3898 int ret;
3899 handle_t *handle;
3900 int credits = 0;
3901 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3902 struct ocfs2_alloc_context *meta_ac = NULL;
3903
3904 ret = ocfs2_lock_refcount_allocators(inode->i_sb,
3905 p_cluster, num_clusters,
3906 et, ref_ci,
3907 ref_root_bh, &meta_ac,
3908 NULL, &credits);
3909 if (ret) {
3910 mlog_errno(ret);
3911 goto out;
3912 }
3913
3914 handle = ocfs2_start_trans(osb, credits);
3915 if (IS_ERR(handle)) {
3916 ret = PTR_ERR(handle);
3917 mlog_errno(ret);
3918 goto out;
3919 }
3920
3921 ret = ocfs2_insert_extent(handle, et, cpos,
3922 ocfs2_clusters_to_blocks(inode->i_sb, p_cluster),
3923 num_clusters, ext_flags, meta_ac);
3924 if (ret) {
3925 mlog_errno(ret);
3926 goto out_commit;
3927 }
3928
3929 ret = ocfs2_increase_refcount(handle, ref_ci, ref_root_bh,
3930 p_cluster, num_clusters,
3931 meta_ac, dealloc);
3932 if (ret)
3933 mlog_errno(ret);
3934
3935 out_commit:
3936 ocfs2_commit_trans(osb, handle);
3937 out:
3938 if (meta_ac)
3939 ocfs2_free_alloc_context(meta_ac);
3940 return ret;
3941 }
3942
3943 static int ocfs2_duplicate_inline_data(struct inode *s_inode,
3944 struct buffer_head *s_bh,
3945 struct inode *t_inode,
3946 struct buffer_head *t_bh)
3947 {
3948 int ret;
3949 handle_t *handle;
3950 struct ocfs2_super *osb = OCFS2_SB(s_inode->i_sb);
3951 struct ocfs2_dinode *s_di = (struct ocfs2_dinode *)s_bh->b_data;
3952 struct ocfs2_dinode *t_di = (struct ocfs2_dinode *)t_bh->b_data;
3953
3954 BUG_ON(!(OCFS2_I(s_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL));
3955
3956 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
3957 if (IS_ERR(handle)) {
3958 ret = PTR_ERR(handle);
3959 mlog_errno(ret);
3960 goto out;
3961 }
3962
3963 ret = ocfs2_journal_access_di(handle, INODE_CACHE(t_inode), t_bh,
3964 OCFS2_JOURNAL_ACCESS_WRITE);
3965 if (ret) {
3966 mlog_errno(ret);
3967 goto out_commit;
3968 }
3969
3970 t_di->id2.i_data.id_count = s_di->id2.i_data.id_count;
3971 memcpy(t_di->id2.i_data.id_data, s_di->id2.i_data.id_data,
3972 le16_to_cpu(s_di->id2.i_data.id_count));
3973 spin_lock(&OCFS2_I(t_inode)->ip_lock);
3974 OCFS2_I(t_inode)->ip_dyn_features |= OCFS2_INLINE_DATA_FL;
3975 t_di->i_dyn_features = cpu_to_le16(OCFS2_I(t_inode)->ip_dyn_features);
3976 spin_unlock(&OCFS2_I(t_inode)->ip_lock);
3977
3978 ocfs2_journal_dirty(handle, t_bh);
3979
3980 out_commit:
3981 ocfs2_commit_trans(osb, handle);
3982 out:
3983 return ret;
3984 }
3985
3986 static int ocfs2_duplicate_extent_list(struct inode *s_inode,
3987 struct inode *t_inode,
3988 struct buffer_head *t_bh,
3989 struct ocfs2_caching_info *ref_ci,
3990 struct buffer_head *ref_root_bh,
3991 struct ocfs2_cached_dealloc_ctxt *dealloc)
3992 {
3993 int ret = 0;
3994 u32 p_cluster, num_clusters, clusters, cpos;
3995 loff_t size;
3996 unsigned int ext_flags;
3997 struct ocfs2_extent_tree et;
3998
3999 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(t_inode), t_bh);
4000
4001 size = i_size_read(s_inode);
4002 clusters = ocfs2_clusters_for_bytes(s_inode->i_sb, size);
4003
4004 cpos = 0;
4005 while (cpos < clusters) {
4006 ret = ocfs2_get_clusters(s_inode, cpos, &p_cluster,
4007 &num_clusters, &ext_flags);
4008
4009 if (p_cluster) {
4010 ret = ocfs2_add_refcounted_extent(t_inode, &et,
4011 ref_ci, ref_root_bh,
4012 cpos, p_cluster,
4013 num_clusters,
4014 ext_flags,
4015 dealloc);
4016 if (ret) {
4017 mlog_errno(ret);
4018 goto out;
4019 }
4020 }
4021
4022 cpos += num_clusters;
4023 }
4024
4025 out:
4026 return ret;
4027 }
4028
4029 /*
4030 * change the new file's attributes to the src.
4031 *
4032 * reflink creates a snapshot of a file, that means the attributes
4033 * must be identical except for three exceptions - nlink, ino, and ctime.
4034 */
4035 static int ocfs2_complete_reflink(struct inode *s_inode,
4036 struct buffer_head *s_bh,
4037 struct inode *t_inode,
4038 struct buffer_head *t_bh,
4039 bool preserve)
4040 {
4041 int ret;
4042 handle_t *handle;
4043 struct ocfs2_dinode *s_di = (struct ocfs2_dinode *)s_bh->b_data;
4044 struct ocfs2_dinode *di = (struct ocfs2_dinode *)t_bh->b_data;
4045 loff_t size = i_size_read(s_inode);
4046
4047 handle = ocfs2_start_trans(OCFS2_SB(t_inode->i_sb),
4048 OCFS2_INODE_UPDATE_CREDITS);
4049 if (IS_ERR(handle)) {
4050 ret = PTR_ERR(handle);
4051 mlog_errno(ret);
4052 return ret;
4053 }
4054
4055 ret = ocfs2_journal_access_di(handle, INODE_CACHE(t_inode), t_bh,
4056 OCFS2_JOURNAL_ACCESS_WRITE);
4057 if (ret) {
4058 mlog_errno(ret);
4059 goto out_commit;
4060 }
4061
4062 spin_lock(&OCFS2_I(t_inode)->ip_lock);
4063 OCFS2_I(t_inode)->ip_clusters = OCFS2_I(s_inode)->ip_clusters;
4064 OCFS2_I(t_inode)->ip_attr = OCFS2_I(s_inode)->ip_attr;
4065 OCFS2_I(t_inode)->ip_dyn_features = OCFS2_I(s_inode)->ip_dyn_features;
4066 spin_unlock(&OCFS2_I(t_inode)->ip_lock);
4067 i_size_write(t_inode, size);
4068 t_inode->i_blocks = s_inode->i_blocks;
4069
4070 di->i_xattr_inline_size = s_di->i_xattr_inline_size;
4071 di->i_clusters = s_di->i_clusters;
4072 di->i_size = s_di->i_size;
4073 di->i_dyn_features = s_di->i_dyn_features;
4074 di->i_attr = s_di->i_attr;
4075
4076 if (preserve) {
4077 t_inode->i_uid = s_inode->i_uid;
4078 t_inode->i_gid = s_inode->i_gid;
4079 t_inode->i_mode = s_inode->i_mode;
4080 di->i_uid = s_di->i_uid;
4081 di->i_gid = s_di->i_gid;
4082 di->i_mode = s_di->i_mode;
4083
4084 /*
4085 * update time.
4086 * we want mtime to appear identical to the source and
4087 * update ctime.
4088 */
4089 t_inode->i_ctime = CURRENT_TIME;
4090
4091 di->i_ctime = cpu_to_le64(t_inode->i_ctime.tv_sec);
4092 di->i_ctime_nsec = cpu_to_le32(t_inode->i_ctime.tv_nsec);
4093
4094 t_inode->i_mtime = s_inode->i_mtime;
4095 di->i_mtime = s_di->i_mtime;
4096 di->i_mtime_nsec = s_di->i_mtime_nsec;
4097 }
4098
4099 ocfs2_journal_dirty(handle, t_bh);
4100
4101 out_commit:
4102 ocfs2_commit_trans(OCFS2_SB(t_inode->i_sb), handle);
4103 return ret;
4104 }
4105
4106 static int ocfs2_create_reflink_node(struct inode *s_inode,
4107 struct buffer_head *s_bh,
4108 struct inode *t_inode,
4109 struct buffer_head *t_bh,
4110 bool preserve)
4111 {
4112 int ret;
4113 struct buffer_head *ref_root_bh = NULL;
4114 struct ocfs2_cached_dealloc_ctxt dealloc;
4115 struct ocfs2_super *osb = OCFS2_SB(s_inode->i_sb);
4116 struct ocfs2_refcount_block *rb;
4117 struct ocfs2_dinode *di = (struct ocfs2_dinode *)s_bh->b_data;
4118 struct ocfs2_refcount_tree *ref_tree;
4119
4120 ocfs2_init_dealloc_ctxt(&dealloc);
4121
4122 ret = ocfs2_set_refcount_tree(t_inode, t_bh,
4123 le64_to_cpu(di->i_refcount_loc));
4124 if (ret) {
4125 mlog_errno(ret);
4126 goto out;
4127 }
4128
4129 if (OCFS2_I(s_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
4130 ret = ocfs2_duplicate_inline_data(s_inode, s_bh,
4131 t_inode, t_bh);
4132 if (ret)
4133 mlog_errno(ret);
4134 goto out;
4135 }
4136
4137 ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
4138 1, &ref_tree, &ref_root_bh);
4139 if (ret) {
4140 mlog_errno(ret);
4141 goto out;
4142 }
4143 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
4144
4145 ret = ocfs2_duplicate_extent_list(s_inode, t_inode, t_bh,
4146 &ref_tree->rf_ci, ref_root_bh,
4147 &dealloc);
4148 if (ret) {
4149 mlog_errno(ret);
4150 goto out_unlock_refcount;
4151 }
4152
4153 out_unlock_refcount:
4154 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
4155 brelse(ref_root_bh);
4156 out:
4157 if (ocfs2_dealloc_has_cluster(&dealloc)) {
4158 ocfs2_schedule_truncate_log_flush(osb, 1);
4159 ocfs2_run_deallocs(osb, &dealloc);
4160 }
4161
4162 return ret;
4163 }
4164
4165 static int __ocfs2_reflink(struct dentry *old_dentry,
4166 struct buffer_head *old_bh,
4167 struct inode *new_inode,
4168 bool preserve)
4169 {
4170 int ret;
4171 struct inode *inode = old_dentry->d_inode;
4172 struct buffer_head *new_bh = NULL;
4173
4174 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
4175 ret = -EINVAL;
4176 mlog_errno(ret);
4177 goto out;
4178 }
4179
4180 ret = filemap_fdatawrite(inode->i_mapping);
4181 if (ret) {
4182 mlog_errno(ret);
4183 goto out;
4184 }
4185
4186 ret = ocfs2_attach_refcount_tree(inode, old_bh);
4187 if (ret) {
4188 mlog_errno(ret);
4189 goto out;
4190 }
4191
4192 mutex_lock_nested(&new_inode->i_mutex, I_MUTEX_CHILD);
4193 ret = ocfs2_inode_lock_nested(new_inode, &new_bh, 1,
4194 OI_LS_REFLINK_TARGET);
4195 if (ret) {
4196 mlog_errno(ret);
4197 goto out_unlock;
4198 }
4199
4200 ret = ocfs2_create_reflink_node(inode, old_bh,
4201 new_inode, new_bh, preserve);
4202 if (ret) {
4203 mlog_errno(ret);
4204 goto inode_unlock;
4205 }
4206
4207 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_XATTR_FL) {
4208 ret = ocfs2_reflink_xattrs(inode, old_bh,
4209 new_inode, new_bh,
4210 preserve);
4211 if (ret) {
4212 mlog_errno(ret);
4213 goto inode_unlock;
4214 }
4215 }
4216
4217 ret = ocfs2_complete_reflink(inode, old_bh,
4218 new_inode, new_bh, preserve);
4219 if (ret)
4220 mlog_errno(ret);
4221
4222 inode_unlock:
4223 ocfs2_inode_unlock(new_inode, 1);
4224 brelse(new_bh);
4225 out_unlock:
4226 mutex_unlock(&new_inode->i_mutex);
4227 out:
4228 if (!ret) {
4229 ret = filemap_fdatawait(inode->i_mapping);
4230 if (ret)
4231 mlog_errno(ret);
4232 }
4233 return ret;
4234 }
4235
4236 static int ocfs2_reflink(struct dentry *old_dentry, struct inode *dir,
4237 struct dentry *new_dentry, bool preserve)
4238 {
4239 int error;
4240 struct inode *inode = old_dentry->d_inode;
4241 struct buffer_head *old_bh = NULL;
4242 struct inode *new_orphan_inode = NULL;
4243
4244 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)))
4245 return -EOPNOTSUPP;
4246
4247 error = ocfs2_create_inode_in_orphan(dir, inode->i_mode,
4248 &new_orphan_inode);
4249 if (error) {
4250 mlog_errno(error);
4251 goto out;
4252 }
4253
4254 error = ocfs2_inode_lock(inode, &old_bh, 1);
4255 if (error) {
4256 mlog_errno(error);
4257 goto out;
4258 }
4259
4260 down_write(&OCFS2_I(inode)->ip_xattr_sem);
4261 down_write(&OCFS2_I(inode)->ip_alloc_sem);
4262 error = __ocfs2_reflink(old_dentry, old_bh,
4263 new_orphan_inode, preserve);
4264 up_write(&OCFS2_I(inode)->ip_alloc_sem);
4265 up_write(&OCFS2_I(inode)->ip_xattr_sem);
4266
4267 ocfs2_inode_unlock(inode, 1);
4268 brelse(old_bh);
4269
4270 if (error) {
4271 mlog_errno(error);
4272 goto out;
4273 }
4274
4275 /* If the security isn't preserved, we need to re-initialize them. */
4276 if (!preserve) {
4277 error = ocfs2_init_security_and_acl(dir, new_orphan_inode);
4278 if (error)
4279 mlog_errno(error);
4280 }
4281 out:
4282 if (!error) {
4283 error = ocfs2_mv_orphaned_inode_to_new(dir, new_orphan_inode,
4284 new_dentry);
4285 if (error)
4286 mlog_errno(error);
4287 }
4288
4289 if (new_orphan_inode) {
4290 /*
4291 * We need to open_unlock the inode no matter whether we
4292 * succeed or not, so that other nodes can delete it later.
4293 */
4294 ocfs2_open_unlock(new_orphan_inode);
4295 if (error)
4296 iput(new_orphan_inode);
4297 }
4298
4299 return error;
4300 }
4301
4302 /*
4303 * Below here are the bits used by OCFS2_IOC_REFLINK() to fake
4304 * sys_reflink(). This will go away when vfs_reflink() exists in
4305 * fs/namei.c.
4306 */
4307
4308 /* copied from may_create in VFS. */
4309 static inline int ocfs2_may_create(struct inode *dir, struct dentry *child)
4310 {
4311 if (child->d_inode)
4312 return -EEXIST;
4313 if (IS_DEADDIR(dir))
4314 return -ENOENT;
4315 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
4316 }
4317
4318 /* copied from user_path_parent. */
4319 static int ocfs2_user_path_parent(const char __user *path,
4320 struct nameidata *nd, char **name)
4321 {
4322 char *s = getname(path);
4323 int error;
4324
4325 if (IS_ERR(s))
4326 return PTR_ERR(s);
4327
4328 error = path_lookup(s, LOOKUP_PARENT, nd);
4329 if (error)
4330 putname(s);
4331 else
4332 *name = s;
4333
4334 return error;
4335 }
4336
4337 /**
4338 * ocfs2_vfs_reflink - Create a reference-counted link
4339 *
4340 * @old_dentry: source dentry + inode
4341 * @dir: directory to create the target
4342 * @new_dentry: target dentry
4343 * @preserve: if true, preserve all file attributes
4344 */
4345 static int ocfs2_vfs_reflink(struct dentry *old_dentry, struct inode *dir,
4346 struct dentry *new_dentry, bool preserve)
4347 {
4348 struct inode *inode = old_dentry->d_inode;
4349 int error;
4350
4351 if (!inode)
4352 return -ENOENT;
4353
4354 error = ocfs2_may_create(dir, new_dentry);
4355 if (error)
4356 return error;
4357
4358 if (dir->i_sb != inode->i_sb)
4359 return -EXDEV;
4360
4361 /*
4362 * A reflink to an append-only or immutable file cannot be created.
4363 */
4364 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4365 return -EPERM;
4366
4367 /* Only regular files can be reflinked. */
4368 if (!S_ISREG(inode->i_mode))
4369 return -EPERM;
4370
4371 /*
4372 * If the caller wants to preserve ownership, they require the
4373 * rights to do so.
4374 */
4375 if (preserve) {
4376 if ((current_fsuid() != inode->i_uid) && !capable(CAP_CHOWN))
4377 return -EPERM;
4378 if (!in_group_p(inode->i_gid) && !capable(CAP_CHOWN))
4379 return -EPERM;
4380 }
4381
4382 /*
4383 * If the caller is modifying any aspect of the attributes, they
4384 * are not creating a snapshot. They need read permission on the
4385 * file.
4386 */
4387 if (!preserve) {
4388 error = inode_permission(inode, MAY_READ);
4389 if (error)
4390 return error;
4391 }
4392
4393 mutex_lock(&inode->i_mutex);
4394 dquot_initialize(dir);
4395 error = ocfs2_reflink(old_dentry, dir, new_dentry, preserve);
4396 mutex_unlock(&inode->i_mutex);
4397 if (!error)
4398 fsnotify_create(dir, new_dentry);
4399 return error;
4400 }
4401 /*
4402 * Most codes are copied from sys_linkat.
4403 */
4404 int ocfs2_reflink_ioctl(struct inode *inode,
4405 const char __user *oldname,
4406 const char __user *newname,
4407 bool preserve)
4408 {
4409 struct dentry *new_dentry;
4410 struct nameidata nd;
4411 struct path old_path;
4412 int error;
4413 char *to = NULL;
4414
4415 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)))
4416 return -EOPNOTSUPP;
4417
4418 error = user_path_at(AT_FDCWD, oldname, 0, &old_path);
4419 if (error) {
4420 mlog_errno(error);
4421 return error;
4422 }
4423
4424 error = ocfs2_user_path_parent(newname, &nd, &to);
4425 if (error) {
4426 mlog_errno(error);
4427 goto out;
4428 }
4429
4430 error = -EXDEV;
4431 if (old_path.mnt != nd.path.mnt)
4432 goto out_release;
4433 new_dentry = lookup_create(&nd, 0);
4434 error = PTR_ERR(new_dentry);
4435 if (IS_ERR(new_dentry)) {
4436 mlog_errno(error);
4437 goto out_unlock;
4438 }
4439
4440 error = mnt_want_write(nd.path.mnt);
4441 if (error) {
4442 mlog_errno(error);
4443 goto out_dput;
4444 }
4445
4446 error = ocfs2_vfs_reflink(old_path.dentry,
4447 nd.path.dentry->d_inode,
4448 new_dentry, preserve);
4449 mnt_drop_write(nd.path.mnt);
4450 out_dput:
4451 dput(new_dentry);
4452 out_unlock:
4453 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
4454 out_release:
4455 path_put(&nd.path);
4456 putname(to);
4457 out:
4458 path_put(&old_path);
4459
4460 return error;
4461 }
Tomato firmware and modifications.