?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
block: introduce REQ_FLUSH flag
[linux-2.6/cjktty.git] / fs / ocfs2 / refcounttree.c
blob3ac5aa733e9c8018090bc2493d819937593760a0
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 /*
1473 * XXX: Improvement later.
1474 * If we know all the high 32 bit cpos is the same, no need to sort.
1475 *
1476 * In order to make the whole process safe, we do:
1477 * 1. sort the entries by their low 32 bit cpos first so that we can
1478 * find the split cpos easily.
1479 * 2. call ocfs2_insert_extent to insert the new refcount block.
1480 * 3. move the refcount rec to the new block.
1481 * 4. sort the entries by their 64 bit cpos.
1482 * 5. dirty the new_rb and rb.
1483 */
1484 sort(&rl->rl_recs, le16_to_cpu(rl->rl_used),
1485 sizeof(struct ocfs2_refcount_rec),
1486 cmp_refcount_rec_by_low_cpos, swap_refcount_rec);
1487
1488 ret = ocfs2_find_refcount_split_pos(rl, &cpos, &split_index);
1489 if (ret) {
1490 mlog_errno(ret);
1491 return ret;
1492 }
1493
1494 new_rb->rf_cpos = cpu_to_le32(cpos);
1495
1496 /* move refcount records starting from split_index to the new block. */
1497 num_moved = le16_to_cpu(rl->rl_used) - split_index;
1498 memcpy(new_rl->rl_recs, &rl->rl_recs[split_index],
1499 num_moved * sizeof(struct ocfs2_refcount_rec));
1500
1501 /*ok, remove the entries we just moved over to the other block. */
1502 memset(&rl->rl_recs[split_index], 0,
1503 num_moved * sizeof(struct ocfs2_refcount_rec));
1504
1505 /* change old and new rl_used accordingly. */
1506 le16_add_cpu(&rl->rl_used, -num_moved);
1507 new_rl->rl_used = cpu_to_le16(num_moved);
1508
1509 sort(&rl->rl_recs, le16_to_cpu(rl->rl_used),
1510 sizeof(struct ocfs2_refcount_rec),
1511 cmp_refcount_rec_by_cpos, swap_refcount_rec);
1512
1513 sort(&new_rl->rl_recs, le16_to_cpu(new_rl->rl_used),
1514 sizeof(struct ocfs2_refcount_rec),
1515 cmp_refcount_rec_by_cpos, swap_refcount_rec);
1516
1517 *split_cpos = cpos;
1518 return 0;
1519 }
1520
1521 static int ocfs2_new_leaf_refcount_block(handle_t *handle,
1522 struct ocfs2_caching_info *ci,
1523 struct buffer_head *ref_root_bh,
1524 struct buffer_head *ref_leaf_bh,
1525 struct ocfs2_alloc_context *meta_ac)
1526 {
1527 int ret;
1528 u16 suballoc_bit_start;
1529 u32 num_got, new_cpos;
1530 u64 suballoc_loc, blkno;
1531 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1532 struct ocfs2_refcount_block *root_rb =
1533 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1534 struct buffer_head *new_bh = NULL;
1535 struct ocfs2_refcount_block *new_rb;
1536 struct ocfs2_extent_tree ref_et;
1537
1538 BUG_ON(!(le32_to_cpu(root_rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL));
1539
1540 ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
1541 OCFS2_JOURNAL_ACCESS_WRITE);
1542 if (ret) {
1543 mlog_errno(ret);
1544 goto out;
1545 }
1546
1547 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1548 OCFS2_JOURNAL_ACCESS_WRITE);
1549 if (ret) {
1550 mlog_errno(ret);
1551 goto out;
1552 }
1553
1554 ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
1555 &suballoc_bit_start, &num_got,
1556 &blkno);
1557 if (ret) {
1558 mlog_errno(ret);
1559 goto out;
1560 }
1561
1562 new_bh = sb_getblk(sb, blkno);
1563 if (new_bh == NULL) {
1564 ret = -EIO;
1565 mlog_errno(ret);
1566 goto out;
1567 }
1568 ocfs2_set_new_buffer_uptodate(ci, new_bh);
1569
1570 ret = ocfs2_journal_access_rb(handle, ci, new_bh,
1571 OCFS2_JOURNAL_ACCESS_CREATE);
1572 if (ret) {
1573 mlog_errno(ret);
1574 goto out;
1575 }
1576
1577 /* Initialize ocfs2_refcount_block. */
1578 new_rb = (struct ocfs2_refcount_block *)new_bh->b_data;
1579 memset(new_rb, 0, sb->s_blocksize);
1580 strcpy((void *)new_rb, OCFS2_REFCOUNT_BLOCK_SIGNATURE);
1581 new_rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
1582 new_rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
1583 new_rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1584 new_rb->rf_fs_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
1585 new_rb->rf_blkno = cpu_to_le64(blkno);
1586 new_rb->rf_parent = cpu_to_le64(ref_root_bh->b_blocknr);
1587 new_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_LEAF_FL);
1588 new_rb->rf_records.rl_count =
1589 cpu_to_le16(ocfs2_refcount_recs_per_rb(sb));
1590 new_rb->rf_generation = root_rb->rf_generation;
1591
1592 ret = ocfs2_divide_leaf_refcount_block(ref_leaf_bh, new_bh, &new_cpos);
1593 if (ret) {
1594 mlog_errno(ret);
1595 goto out;
1596 }
1597
1598 ocfs2_journal_dirty(handle, ref_leaf_bh);
1599 ocfs2_journal_dirty(handle, new_bh);
1600
1601 ocfs2_init_refcount_extent_tree(&ref_et, ci, ref_root_bh);
1602
1603 mlog(0, "insert new leaf block %llu at %u\n",
1604 (unsigned long long)new_bh->b_blocknr, new_cpos);
1605
1606 /* Insert the new leaf block with the specific offset cpos. */
1607 ret = ocfs2_insert_extent(handle, &ref_et, new_cpos, new_bh->b_blocknr,
1608 1, 0, meta_ac);
1609 if (ret)
1610 mlog_errno(ret);
1611
1612 out:
1613 brelse(new_bh);
1614 return ret;
1615 }
1616
1617 static int ocfs2_expand_refcount_tree(handle_t *handle,
1618 struct ocfs2_caching_info *ci,
1619 struct buffer_head *ref_root_bh,
1620 struct buffer_head *ref_leaf_bh,
1621 struct ocfs2_alloc_context *meta_ac)
1622 {
1623 int ret;
1624 struct buffer_head *expand_bh = NULL;
1625
1626 if (ref_root_bh == ref_leaf_bh) {
1627 /*
1628 * the old root bh hasn't been expanded to a b-tree,
1629 * so expand it first.
1630 */
1631 ret = ocfs2_expand_inline_ref_root(handle, ci, ref_root_bh,
1632 &expand_bh, meta_ac);
1633 if (ret) {
1634 mlog_errno(ret);
1635 goto out;
1636 }
1637 } else {
1638 expand_bh = ref_leaf_bh;
1639 get_bh(expand_bh);
1640 }
1641
1642
1643 /* Now add a new refcount block into the tree.*/
1644 ret = ocfs2_new_leaf_refcount_block(handle, ci, ref_root_bh,
1645 expand_bh, meta_ac);
1646 if (ret)
1647 mlog_errno(ret);
1648 out:
1649 brelse(expand_bh);
1650 return ret;
1651 }
1652
1653 /*
1654 * Adjust the extent rec in b-tree representing ref_leaf_bh.
1655 *
1656 * Only called when we have inserted a new refcount rec at index 0
1657 * which means ocfs2_extent_rec.e_cpos may need some change.
1658 */
1659 static int ocfs2_adjust_refcount_rec(handle_t *handle,
1660 struct ocfs2_caching_info *ci,
1661 struct buffer_head *ref_root_bh,
1662 struct buffer_head *ref_leaf_bh,
1663 struct ocfs2_refcount_rec *rec)
1664 {
1665 int ret = 0, i;
1666 u32 new_cpos, old_cpos;
1667 struct ocfs2_path *path = NULL;
1668 struct ocfs2_extent_tree et;
1669 struct ocfs2_refcount_block *rb =
1670 (struct ocfs2_refcount_block *)ref_root_bh->b_data;
1671 struct ocfs2_extent_list *el;
1672
1673 if (!(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL))
1674 goto out;
1675
1676 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1677 old_cpos = le32_to_cpu(rb->rf_cpos);
1678 new_cpos = le64_to_cpu(rec->r_cpos) & OCFS2_32BIT_POS_MASK;
1679 if (old_cpos <= new_cpos)
1680 goto out;
1681
1682 ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
1683
1684 path = ocfs2_new_path_from_et(&et);
1685 if (!path) {
1686 ret = -ENOMEM;
1687 mlog_errno(ret);
1688 goto out;
1689 }
1690
1691 ret = ocfs2_find_path(ci, path, old_cpos);
1692 if (ret) {
1693 mlog_errno(ret);
1694 goto out;
1695 }
1696
1697 /*
1698 * 2 more credits, one for the leaf refcount block, one for
1699 * the extent block contains the extent rec.
1700 */
1701 ret = ocfs2_extend_trans(handle, 2);
1702 if (ret < 0) {
1703 mlog_errno(ret);
1704 goto out;
1705 }
1706
1707 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1708 OCFS2_JOURNAL_ACCESS_WRITE);
1709 if (ret < 0) {
1710 mlog_errno(ret);
1711 goto out;
1712 }
1713
1714 ret = ocfs2_journal_access_eb(handle, ci, path_leaf_bh(path),
1715 OCFS2_JOURNAL_ACCESS_WRITE);
1716 if (ret < 0) {
1717 mlog_errno(ret);
1718 goto out;
1719 }
1720
1721 /* change the leaf extent block first. */
1722 el = path_leaf_el(path);
1723
1724 for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++)
1725 if (le32_to_cpu(el->l_recs[i].e_cpos) == old_cpos)
1726 break;
1727
1728 BUG_ON(i == le16_to_cpu(el->l_next_free_rec));
1729
1730 el->l_recs[i].e_cpos = cpu_to_le32(new_cpos);
1731
1732 /* change the r_cpos in the leaf block. */
1733 rb->rf_cpos = cpu_to_le32(new_cpos);
1734
1735 ocfs2_journal_dirty(handle, path_leaf_bh(path));
1736 ocfs2_journal_dirty(handle, ref_leaf_bh);
1737
1738 out:
1739 ocfs2_free_path(path);
1740 return ret;
1741 }
1742
1743 static int ocfs2_insert_refcount_rec(handle_t *handle,
1744 struct ocfs2_caching_info *ci,
1745 struct buffer_head *ref_root_bh,
1746 struct buffer_head *ref_leaf_bh,
1747 struct ocfs2_refcount_rec *rec,
1748 int index, int merge,
1749 struct ocfs2_alloc_context *meta_ac)
1750 {
1751 int ret;
1752 struct ocfs2_refcount_block *rb =
1753 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1754 struct ocfs2_refcount_list *rf_list = &rb->rf_records;
1755 struct buffer_head *new_bh = NULL;
1756
1757 BUG_ON(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL);
1758
1759 if (rf_list->rl_used == rf_list->rl_count) {
1760 u64 cpos = le64_to_cpu(rec->r_cpos);
1761 u32 len = le32_to_cpu(rec->r_clusters);
1762
1763 ret = ocfs2_expand_refcount_tree(handle, ci, ref_root_bh,
1764 ref_leaf_bh, meta_ac);
1765 if (ret) {
1766 mlog_errno(ret);
1767 goto out;
1768 }
1769
1770 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
1771 cpos, len, NULL, &index,
1772 &new_bh);
1773 if (ret) {
1774 mlog_errno(ret);
1775 goto out;
1776 }
1777
1778 ref_leaf_bh = new_bh;
1779 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1780 rf_list = &rb->rf_records;
1781 }
1782
1783 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1784 OCFS2_JOURNAL_ACCESS_WRITE);
1785 if (ret) {
1786 mlog_errno(ret);
1787 goto out;
1788 }
1789
1790 if (index < le16_to_cpu(rf_list->rl_used))
1791 memmove(&rf_list->rl_recs[index + 1],
1792 &rf_list->rl_recs[index],
1793 (le16_to_cpu(rf_list->rl_used) - index) *
1794 sizeof(struct ocfs2_refcount_rec));
1795
1796 mlog(0, "insert refcount record start %llu, len %u, count %u "
1797 "to leaf block %llu at index %d\n",
1798 (unsigned long long)le64_to_cpu(rec->r_cpos),
1799 le32_to_cpu(rec->r_clusters), le32_to_cpu(rec->r_refcount),
1800 (unsigned long long)ref_leaf_bh->b_blocknr, index);
1801
1802 rf_list->rl_recs[index] = *rec;
1803
1804 le16_add_cpu(&rf_list->rl_used, 1);
1805
1806 if (merge)
1807 ocfs2_refcount_rec_merge(rb, index);
1808
1809 ocfs2_journal_dirty(handle, ref_leaf_bh);
1810
1811 if (index == 0) {
1812 ret = ocfs2_adjust_refcount_rec(handle, ci,
1813 ref_root_bh,
1814 ref_leaf_bh, rec);
1815 if (ret)
1816 mlog_errno(ret);
1817 }
1818 out:
1819 brelse(new_bh);
1820 return ret;
1821 }
1822
1823 /*
1824 * Split the refcount_rec indexed by "index" in ref_leaf_bh.
1825 * This is much simple than our b-tree code.
1826 * split_rec is the new refcount rec we want to insert.
1827 * If split_rec->r_refcount > 0, we are changing the refcount(in case we
1828 * increase refcount or decrease a refcount to non-zero).
1829 * If split_rec->r_refcount == 0, we are punching a hole in current refcount
1830 * rec( in case we decrease a refcount to zero).
1831 */
1832 static int ocfs2_split_refcount_rec(handle_t *handle,
1833 struct ocfs2_caching_info *ci,
1834 struct buffer_head *ref_root_bh,
1835 struct buffer_head *ref_leaf_bh,
1836 struct ocfs2_refcount_rec *split_rec,
1837 int index, int merge,
1838 struct ocfs2_alloc_context *meta_ac,
1839 struct ocfs2_cached_dealloc_ctxt *dealloc)
1840 {
1841 int ret, recs_need;
1842 u32 len;
1843 struct ocfs2_refcount_block *rb =
1844 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1845 struct ocfs2_refcount_list *rf_list = &rb->rf_records;
1846 struct ocfs2_refcount_rec *orig_rec = &rf_list->rl_recs[index];
1847 struct ocfs2_refcount_rec *tail_rec = NULL;
1848 struct buffer_head *new_bh = NULL;
1849
1850 BUG_ON(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL);
1851
1852 mlog(0, "original r_pos %llu, cluster %u, split %llu, cluster %u\n",
1853 le64_to_cpu(orig_rec->r_cpos), le32_to_cpu(orig_rec->r_clusters),
1854 le64_to_cpu(split_rec->r_cpos),
1855 le32_to_cpu(split_rec->r_clusters));
1856
1857 /*
1858 * If we just need to split the header or tail clusters,
1859 * no more recs are needed, just split is OK.
1860 * Otherwise we at least need one new recs.
1861 */
1862 if (!split_rec->r_refcount &&
1863 (split_rec->r_cpos == orig_rec->r_cpos ||
1864 le64_to_cpu(split_rec->r_cpos) +
1865 le32_to_cpu(split_rec->r_clusters) ==
1866 le64_to_cpu(orig_rec->r_cpos) + le32_to_cpu(orig_rec->r_clusters)))
1867 recs_need = 0;
1868 else
1869 recs_need = 1;
1870
1871 /*
1872 * We need one more rec if we split in the middle and the new rec have
1873 * some refcount in it.
1874 */
1875 if (split_rec->r_refcount &&
1876 (split_rec->r_cpos != orig_rec->r_cpos &&
1877 le64_to_cpu(split_rec->r_cpos) +
1878 le32_to_cpu(split_rec->r_clusters) !=
1879 le64_to_cpu(orig_rec->r_cpos) + le32_to_cpu(orig_rec->r_clusters)))
1880 recs_need++;
1881
1882 /* If the leaf block don't have enough record, expand it. */
1883 if (le16_to_cpu(rf_list->rl_used) + recs_need >
1884 le16_to_cpu(rf_list->rl_count)) {
1885 struct ocfs2_refcount_rec tmp_rec;
1886 u64 cpos = le64_to_cpu(orig_rec->r_cpos);
1887 len = le32_to_cpu(orig_rec->r_clusters);
1888 ret = ocfs2_expand_refcount_tree(handle, ci, ref_root_bh,
1889 ref_leaf_bh, meta_ac);
1890 if (ret) {
1891 mlog_errno(ret);
1892 goto out;
1893 }
1894
1895 /*
1896 * We have to re-get it since now cpos may be moved to
1897 * another leaf block.
1898 */
1899 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
1900 cpos, len, &tmp_rec, &index,
1901 &new_bh);
1902 if (ret) {
1903 mlog_errno(ret);
1904 goto out;
1905 }
1906
1907 ref_leaf_bh = new_bh;
1908 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1909 rf_list = &rb->rf_records;
1910 orig_rec = &rf_list->rl_recs[index];
1911 }
1912
1913 ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1914 OCFS2_JOURNAL_ACCESS_WRITE);
1915 if (ret) {
1916 mlog_errno(ret);
1917 goto out;
1918 }
1919
1920 /*
1921 * We have calculated out how many new records we need and store
1922 * in recs_need, so spare enough space first by moving the records
1923 * after "index" to the end.
1924 */
1925 if (index != le16_to_cpu(rf_list->rl_used) - 1)
1926 memmove(&rf_list->rl_recs[index + 1 + recs_need],
1927 &rf_list->rl_recs[index + 1],
1928 (le16_to_cpu(rf_list->rl_used) - index - 1) *
1929 sizeof(struct ocfs2_refcount_rec));
1930
1931 len = (le64_to_cpu(orig_rec->r_cpos) +
1932 le32_to_cpu(orig_rec->r_clusters)) -
1933 (le64_to_cpu(split_rec->r_cpos) +
1934 le32_to_cpu(split_rec->r_clusters));
1935
1936 /*
1937 * If we have "len", the we will split in the tail and move it
1938 * to the end of the space we have just spared.
1939 */
1940 if (len) {
1941 tail_rec = &rf_list->rl_recs[index + recs_need];
1942
1943 memcpy(tail_rec, orig_rec, sizeof(struct ocfs2_refcount_rec));
1944 le64_add_cpu(&tail_rec->r_cpos,
1945 le32_to_cpu(tail_rec->r_clusters) - len);
1946 tail_rec->r_clusters = cpu_to_le32(len);
1947 }
1948
1949 /*
1950 * If the split pos isn't the same as the original one, we need to
1951 * split in the head.
1952 *
1953 * Note: We have the chance that split_rec.r_refcount = 0,
1954 * recs_need = 0 and len > 0, which means we just cut the head from
1955 * the orig_rec and in that case we have done some modification in
1956 * orig_rec above, so the check for r_cpos is faked.
1957 */
1958 if (split_rec->r_cpos != orig_rec->r_cpos && tail_rec != orig_rec) {
1959 len = le64_to_cpu(split_rec->r_cpos) -
1960 le64_to_cpu(orig_rec->r_cpos);
1961 orig_rec->r_clusters = cpu_to_le32(len);
1962 index++;
1963 }
1964
1965 le16_add_cpu(&rf_list->rl_used, recs_need);
1966
1967 if (split_rec->r_refcount) {
1968 rf_list->rl_recs[index] = *split_rec;
1969 mlog(0, "insert refcount record start %llu, len %u, count %u "
1970 "to leaf block %llu at index %d\n",
1971 (unsigned long long)le64_to_cpu(split_rec->r_cpos),
1972 le32_to_cpu(split_rec->r_clusters),
1973 le32_to_cpu(split_rec->r_refcount),
1974 (unsigned long long)ref_leaf_bh->b_blocknr, index);
1975
1976 if (merge)
1977 ocfs2_refcount_rec_merge(rb, index);
1978 }
1979
1980 ocfs2_journal_dirty(handle, ref_leaf_bh);
1981
1982 out:
1983 brelse(new_bh);
1984 return ret;
1985 }
1986
1987 static int __ocfs2_increase_refcount(handle_t *handle,
1988 struct ocfs2_caching_info *ci,
1989 struct buffer_head *ref_root_bh,
1990 u64 cpos, u32 len, int merge,
1991 struct ocfs2_alloc_context *meta_ac,
1992 struct ocfs2_cached_dealloc_ctxt *dealloc)
1993 {
1994 int ret = 0, index;
1995 struct buffer_head *ref_leaf_bh = NULL;
1996 struct ocfs2_refcount_rec rec;
1997 unsigned int set_len = 0;
1998
1999 mlog(0, "Tree owner %llu, add refcount start %llu, len %u\n",
2000 (unsigned long long)ocfs2_metadata_cache_owner(ci),
2001 (unsigned long long)cpos, len);
2002
2003 while (len) {
2004 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2005 cpos, len, &rec, &index,
2006 &ref_leaf_bh);
2007 if (ret) {
2008 mlog_errno(ret);
2009 goto out;
2010 }
2011
2012 set_len = le32_to_cpu(rec.r_clusters);
2013
2014 /*
2015 * Here we may meet with 3 situations:
2016 *
2017 * 1. If we find an already existing record, and the length
2018 * is the same, cool, we just need to increase the r_refcount
2019 * and it is OK.
2020 * 2. If we find a hole, just insert it with r_refcount = 1.
2021 * 3. If we are in the middle of one extent record, split
2022 * it.
2023 */
2024 if (rec.r_refcount && le64_to_cpu(rec.r_cpos) == cpos &&
2025 set_len <= len) {
2026 mlog(0, "increase refcount rec, start %llu, len %u, "
2027 "count %u\n", (unsigned long long)cpos, set_len,
2028 le32_to_cpu(rec.r_refcount));
2029 ret = ocfs2_change_refcount_rec(handle, ci,
2030 ref_leaf_bh, index,
2031 merge, 1);
2032 if (ret) {
2033 mlog_errno(ret);
2034 goto out;
2035 }
2036 } else if (!rec.r_refcount) {
2037 rec.r_refcount = cpu_to_le32(1);
2038
2039 mlog(0, "insert refcount rec, start %llu, len %u\n",
2040 (unsigned long long)le64_to_cpu(rec.r_cpos),
2041 set_len);
2042 ret = ocfs2_insert_refcount_rec(handle, ci, ref_root_bh,
2043 ref_leaf_bh,
2044 &rec, index,
2045 merge, meta_ac);
2046 if (ret) {
2047 mlog_errno(ret);
2048 goto out;
2049 }
2050 } else {
2051 set_len = min((u64)(cpos + len),
2052 le64_to_cpu(rec.r_cpos) + set_len) - cpos;
2053 rec.r_cpos = cpu_to_le64(cpos);
2054 rec.r_clusters = cpu_to_le32(set_len);
2055 le32_add_cpu(&rec.r_refcount, 1);
2056
2057 mlog(0, "split refcount rec, start %llu, "
2058 "len %u, count %u\n",
2059 (unsigned long long)le64_to_cpu(rec.r_cpos),
2060 set_len, le32_to_cpu(rec.r_refcount));
2061 ret = ocfs2_split_refcount_rec(handle, ci,
2062 ref_root_bh, ref_leaf_bh,
2063 &rec, index, merge,
2064 meta_ac, dealloc);
2065 if (ret) {
2066 mlog_errno(ret);
2067 goto out;
2068 }
2069 }
2070
2071 cpos += set_len;
2072 len -= set_len;
2073 brelse(ref_leaf_bh);
2074 ref_leaf_bh = NULL;
2075 }
2076
2077 out:
2078 brelse(ref_leaf_bh);
2079 return ret;
2080 }
2081
2082 static int ocfs2_remove_refcount_extent(handle_t *handle,
2083 struct ocfs2_caching_info *ci,
2084 struct buffer_head *ref_root_bh,
2085 struct buffer_head *ref_leaf_bh,
2086 struct ocfs2_alloc_context *meta_ac,
2087 struct ocfs2_cached_dealloc_ctxt *dealloc)
2088 {
2089 int ret;
2090 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
2091 struct ocfs2_refcount_block *rb =
2092 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
2093 struct ocfs2_extent_tree et;
2094
2095 BUG_ON(rb->rf_records.rl_used);
2096
2097 ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
2098 ret = ocfs2_remove_extent(handle, &et, le32_to_cpu(rb->rf_cpos),
2099 1, meta_ac, dealloc);
2100 if (ret) {
2101 mlog_errno(ret);
2102 goto out;
2103 }
2104
2105 ocfs2_remove_from_cache(ci, ref_leaf_bh);
2106
2107 /*
2108 * add the freed block to the dealloc so that it will be freed
2109 * when we run dealloc.
2110 */
2111 ret = ocfs2_cache_block_dealloc(dealloc, EXTENT_ALLOC_SYSTEM_INODE,
2112 le16_to_cpu(rb->rf_suballoc_slot),
2113 le64_to_cpu(rb->rf_suballoc_loc),
2114 le64_to_cpu(rb->rf_blkno),
2115 le16_to_cpu(rb->rf_suballoc_bit));
2116 if (ret) {
2117 mlog_errno(ret);
2118 goto out;
2119 }
2120
2121 ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
2122 OCFS2_JOURNAL_ACCESS_WRITE);
2123 if (ret) {
2124 mlog_errno(ret);
2125 goto out;
2126 }
2127
2128 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
2129
2130 le32_add_cpu(&rb->rf_clusters, -1);
2131
2132 /*
2133 * check whether we need to restore the root refcount block if
2134 * there is no leaf extent block at atll.
2135 */
2136 if (!rb->rf_list.l_next_free_rec) {
2137 BUG_ON(rb->rf_clusters);
2138
2139 mlog(0, "reset refcount tree root %llu to be a record block.\n",
2140 (unsigned long long)ref_root_bh->b_blocknr);
2141
2142 rb->rf_flags = 0;
2143 rb->rf_parent = 0;
2144 rb->rf_cpos = 0;
2145 memset(&rb->rf_records, 0, sb->s_blocksize -
2146 offsetof(struct ocfs2_refcount_block, rf_records));
2147 rb->rf_records.rl_count =
2148 cpu_to_le16(ocfs2_refcount_recs_per_rb(sb));
2149 }
2150
2151 ocfs2_journal_dirty(handle, ref_root_bh);
2152
2153 out:
2154 return ret;
2155 }
2156
2157 int ocfs2_increase_refcount(handle_t *handle,
2158 struct ocfs2_caching_info *ci,
2159 struct buffer_head *ref_root_bh,
2160 u64 cpos, u32 len,
2161 struct ocfs2_alloc_context *meta_ac,
2162 struct ocfs2_cached_dealloc_ctxt *dealloc)
2163 {
2164 return __ocfs2_increase_refcount(handle, ci, ref_root_bh,
2165 cpos, len, 1,
2166 meta_ac, dealloc);
2167 }
2168
2169 static int ocfs2_decrease_refcount_rec(handle_t *handle,
2170 struct ocfs2_caching_info *ci,
2171 struct buffer_head *ref_root_bh,
2172 struct buffer_head *ref_leaf_bh,
2173 int index, u64 cpos, unsigned int len,
2174 struct ocfs2_alloc_context *meta_ac,
2175 struct ocfs2_cached_dealloc_ctxt *dealloc)
2176 {
2177 int ret;
2178 struct ocfs2_refcount_block *rb =
2179 (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
2180 struct ocfs2_refcount_rec *rec = &rb->rf_records.rl_recs[index];
2181
2182 BUG_ON(cpos < le64_to_cpu(rec->r_cpos));
2183 BUG_ON(cpos + len >
2184 le64_to_cpu(rec->r_cpos) + le32_to_cpu(rec->r_clusters));
2185
2186 if (cpos == le64_to_cpu(rec->r_cpos) &&
2187 len == le32_to_cpu(rec->r_clusters))
2188 ret = ocfs2_change_refcount_rec(handle, ci,
2189 ref_leaf_bh, index, 1, -1);
2190 else {
2191 struct ocfs2_refcount_rec split = *rec;
2192 split.r_cpos = cpu_to_le64(cpos);
2193 split.r_clusters = cpu_to_le32(len);
2194
2195 le32_add_cpu(&split.r_refcount, -1);
2196
2197 mlog(0, "split refcount rec, start %llu, "
2198 "len %u, count %u, original start %llu, len %u\n",
2199 (unsigned long long)le64_to_cpu(split.r_cpos),
2200 len, le32_to_cpu(split.r_refcount),
2201 (unsigned long long)le64_to_cpu(rec->r_cpos),
2202 le32_to_cpu(rec->r_clusters));
2203 ret = ocfs2_split_refcount_rec(handle, ci,
2204 ref_root_bh, ref_leaf_bh,
2205 &split, index, 1,
2206 meta_ac, dealloc);
2207 }
2208
2209 if (ret) {
2210 mlog_errno(ret);
2211 goto out;
2212 }
2213
2214 /* Remove the leaf refcount block if it contains no refcount record. */
2215 if (!rb->rf_records.rl_used && ref_leaf_bh != ref_root_bh) {
2216 ret = ocfs2_remove_refcount_extent(handle, ci, ref_root_bh,
2217 ref_leaf_bh, meta_ac,
2218 dealloc);
2219 if (ret)
2220 mlog_errno(ret);
2221 }
2222
2223 out:
2224 return ret;
2225 }
2226
2227 static int __ocfs2_decrease_refcount(handle_t *handle,
2228 struct ocfs2_caching_info *ci,
2229 struct buffer_head *ref_root_bh,
2230 u64 cpos, u32 len,
2231 struct ocfs2_alloc_context *meta_ac,
2232 struct ocfs2_cached_dealloc_ctxt *dealloc,
2233 int delete)
2234 {
2235 int ret = 0, index = 0;
2236 struct ocfs2_refcount_rec rec;
2237 unsigned int r_count = 0, r_len;
2238 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
2239 struct buffer_head *ref_leaf_bh = NULL;
2240
2241 mlog(0, "Tree owner %llu, decrease refcount start %llu, "
2242 "len %u, delete %u\n",
2243 (unsigned long long)ocfs2_metadata_cache_owner(ci),
2244 (unsigned long long)cpos, len, delete);
2245
2246 while (len) {
2247 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2248 cpos, len, &rec, &index,
2249 &ref_leaf_bh);
2250 if (ret) {
2251 mlog_errno(ret);
2252 goto out;
2253 }
2254
2255 r_count = le32_to_cpu(rec.r_refcount);
2256 BUG_ON(r_count == 0);
2257 if (!delete)
2258 BUG_ON(r_count > 1);
2259
2260 r_len = min((u64)(cpos + len), le64_to_cpu(rec.r_cpos) +
2261 le32_to_cpu(rec.r_clusters)) - cpos;
2262
2263 ret = ocfs2_decrease_refcount_rec(handle, ci, ref_root_bh,
2264 ref_leaf_bh, index,
2265 cpos, r_len,
2266 meta_ac, dealloc);
2267 if (ret) {
2268 mlog_errno(ret);
2269 goto out;
2270 }
2271
2272 if (le32_to_cpu(rec.r_refcount) == 1 && delete) {
2273 ret = ocfs2_cache_cluster_dealloc(dealloc,
2274 ocfs2_clusters_to_blocks(sb, cpos),
2275 r_len);
2276 if (ret) {
2277 mlog_errno(ret);
2278 goto out;
2279 }
2280 }
2281
2282 cpos += r_len;
2283 len -= r_len;
2284 brelse(ref_leaf_bh);
2285 ref_leaf_bh = NULL;
2286 }
2287
2288 out:
2289 brelse(ref_leaf_bh);
2290 return ret;
2291 }
2292
2293 /* Caller must hold refcount tree lock. */
2294 int ocfs2_decrease_refcount(struct inode *inode,
2295 handle_t *handle, u32 cpos, u32 len,
2296 struct ocfs2_alloc_context *meta_ac,
2297 struct ocfs2_cached_dealloc_ctxt *dealloc,
2298 int delete)
2299 {
2300 int ret;
2301 u64 ref_blkno;
2302 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2303 struct buffer_head *ref_root_bh = NULL;
2304 struct ocfs2_refcount_tree *tree;
2305
2306 BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
2307
2308 ret = ocfs2_get_refcount_block(inode, &ref_blkno);
2309 if (ret) {
2310 mlog_errno(ret);
2311 goto out;
2312 }
2313
2314 ret = ocfs2_get_refcount_tree(OCFS2_SB(inode->i_sb), ref_blkno, &tree);
2315 if (ret) {
2316 mlog_errno(ret);
2317 goto out;
2318 }
2319
2320 ret = ocfs2_read_refcount_block(&tree->rf_ci, tree->rf_blkno,
2321 &ref_root_bh);
2322 if (ret) {
2323 mlog_errno(ret);
2324 goto out;
2325 }
2326
2327 ret = __ocfs2_decrease_refcount(handle, &tree->rf_ci, ref_root_bh,
2328 cpos, len, meta_ac, dealloc, delete);
2329 if (ret)
2330 mlog_errno(ret);
2331 out:
2332 brelse(ref_root_bh);
2333 return ret;
2334 }
2335
2336 /*
2337 * Mark the already-existing extent at cpos as refcounted for len clusters.
2338 * This adds the refcount extent flag.
2339 *
2340 * If the existing extent is larger than the request, initiate a
2341 * split. An attempt will be made at merging with adjacent extents.
2342 *
2343 * The caller is responsible for passing down meta_ac if we'll need it.
2344 */
2345 static int ocfs2_mark_extent_refcounted(struct inode *inode,
2346 struct ocfs2_extent_tree *et,
2347 handle_t *handle, u32 cpos,
2348 u32 len, u32 phys,
2349 struct ocfs2_alloc_context *meta_ac,
2350 struct ocfs2_cached_dealloc_ctxt *dealloc)
2351 {
2352 int ret;
2353
2354 mlog(0, "Inode %lu refcount tree cpos %u, len %u, phys cluster %u\n",
2355 inode->i_ino, cpos, len, phys);
2356
2357 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
2358 ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
2359 "tree, but the feature bit is not set in the "
2360 "super block.", inode->i_ino);
2361 ret = -EROFS;
2362 goto out;
2363 }
2364
2365 ret = ocfs2_change_extent_flag(handle, et, cpos,
2366 len, phys, meta_ac, dealloc,
2367 OCFS2_EXT_REFCOUNTED, 0);
2368 if (ret)
2369 mlog_errno(ret);
2370
2371 out:
2372 return ret;
2373 }
2374
2375 /*
2376 * Given some contiguous physical clusters, calculate what we need
2377 * for modifying their refcount.
2378 */
2379 static int ocfs2_calc_refcount_meta_credits(struct super_block *sb,
2380 struct ocfs2_caching_info *ci,
2381 struct buffer_head *ref_root_bh,
2382 u64 start_cpos,
2383 u32 clusters,
2384 int *meta_add,
2385 int *credits)
2386 {
2387 int ret = 0, index, ref_blocks = 0, recs_add = 0;
2388 u64 cpos = start_cpos;
2389 struct ocfs2_refcount_block *rb;
2390 struct ocfs2_refcount_rec rec;
2391 struct buffer_head *ref_leaf_bh = NULL, *prev_bh = NULL;
2392 u32 len;
2393
2394 mlog(0, "start_cpos %llu, clusters %u\n",
2395 (unsigned long long)start_cpos, clusters);
2396 while (clusters) {
2397 ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2398 cpos, clusters, &rec,
2399 &index, &ref_leaf_bh);
2400 if (ret) {
2401 mlog_errno(ret);
2402 goto out;
2403 }
2404
2405 if (ref_leaf_bh != prev_bh) {
2406 /*
2407 * Now we encounter a new leaf block, so calculate
2408 * whether we need to extend the old leaf.
2409 */
2410 if (prev_bh) {
2411 rb = (struct ocfs2_refcount_block *)
2412 prev_bh->b_data;
2413
2414 if (le64_to_cpu(rb->rf_records.rl_used) +
2415 recs_add >
2416 le16_to_cpu(rb->rf_records.rl_count))
2417 ref_blocks++;
2418 }
2419
2420 recs_add = 0;
2421 *credits += 1;
2422 brelse(prev_bh);
2423 prev_bh = ref_leaf_bh;
2424 get_bh(prev_bh);
2425 }
2426
2427 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
2428
2429 mlog(0, "recs_add %d,cpos %llu, clusters %u, rec->r_cpos %llu,"
2430 "rec->r_clusters %u, rec->r_refcount %u, index %d\n",
2431 recs_add, (unsigned long long)cpos, clusters,
2432 (unsigned long long)le64_to_cpu(rec.r_cpos),
2433 le32_to_cpu(rec.r_clusters),
2434 le32_to_cpu(rec.r_refcount), index);
2435
2436 len = min((u64)cpos + clusters, le64_to_cpu(rec.r_cpos) +
2437 le32_to_cpu(rec.r_clusters)) - cpos;
2438 /*
2439 * If the refcount rec already exist, cool. We just need
2440 * to check whether there is a split. Otherwise we just need
2441 * to increase the refcount.
2442 * If we will insert one, increases recs_add.
2443 *
2444 * We record all the records which will be inserted to the
2445 * same refcount block, so that we can tell exactly whether
2446 * we need a new refcount block or not.
2447 */
2448 if (rec.r_refcount) {
2449 /* Check whether we need a split at the beginning. */
2450 if (cpos == start_cpos &&
2451 cpos != le64_to_cpu(rec.r_cpos))
2452 recs_add++;
2453
2454 /* Check whether we need a split in the end. */
2455 if (cpos + clusters < le64_to_cpu(rec.r_cpos) +
2456 le32_to_cpu(rec.r_clusters))
2457 recs_add++;
2458 } else
2459 recs_add++;
2460
2461 brelse(ref_leaf_bh);
2462 ref_leaf_bh = NULL;
2463 clusters -= len;
2464 cpos += len;
2465 }
2466
2467 if (prev_bh) {
2468 rb = (struct ocfs2_refcount_block *)prev_bh->b_data;
2469
2470 if (le64_to_cpu(rb->rf_records.rl_used) + recs_add >
2471 le16_to_cpu(rb->rf_records.rl_count))
2472 ref_blocks++;
2473
2474 *credits += 1;
2475 }
2476
2477 if (!ref_blocks)
2478 goto out;
2479
2480 mlog(0, "we need ref_blocks %d\n", ref_blocks);
2481 *meta_add += ref_blocks;
2482 *credits += ref_blocks;
2483
2484 /*
2485 * So we may need ref_blocks to insert into the tree.
2486 * That also means we need to change the b-tree and add that number
2487 * of records since we never merge them.
2488 * We need one more block for expansion since the new created leaf
2489 * block is also full and needs split.
2490 */
2491 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
2492 if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL) {
2493 struct ocfs2_extent_tree et;
2494
2495 ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
2496 *meta_add += ocfs2_extend_meta_needed(et.et_root_el);
2497 *credits += ocfs2_calc_extend_credits(sb,
2498 et.et_root_el,
2499 ref_blocks);
2500 } else {
2501 *credits += OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
2502 *meta_add += 1;
2503 }
2504
2505 out:
2506 brelse(ref_leaf_bh);
2507 brelse(prev_bh);
2508 return ret;
2509 }
2510
2511 /*
2512 * For refcount tree, we will decrease some contiguous clusters
2513 * refcount count, so just go through it to see how many blocks
2514 * we gonna touch and whether we need to create new blocks.
2515 *
2516 * Normally the refcount blocks store these refcount should be
2517 * contiguous also, so that we can get the number easily.
2518 * We will at most add split 2 refcount records and 2 more
2519 * refcount blocks, so just check it in a rough way.
2520 *
2521 * Caller must hold refcount tree lock.
2522 */
2523 int ocfs2_prepare_refcount_change_for_del(struct inode *inode,
2524 u64 refcount_loc,
2525 u64 phys_blkno,
2526 u32 clusters,
2527 int *credits,
2528 int *ref_blocks)
2529 {
2530 int ret;
2531 struct ocfs2_inode_info *oi = OCFS2_I(inode);
2532 struct buffer_head *ref_root_bh = NULL;
2533 struct ocfs2_refcount_tree *tree;
2534 u64 start_cpos = ocfs2_blocks_to_clusters(inode->i_sb, phys_blkno);
2535
2536 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
2537 ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
2538 "tree, but the feature bit is not set in the "
2539 "super block.", inode->i_ino);
2540 ret = -EROFS;
2541 goto out;
2542 }
2543
2544 BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
2545
2546 ret = ocfs2_get_refcount_tree(OCFS2_SB(inode->i_sb),
2547 refcount_loc, &tree);
2548 if (ret) {
2549 mlog_errno(ret);
2550 goto out;
2551 }
2552
2553 ret = ocfs2_read_refcount_block(&tree->rf_ci, refcount_loc,
2554 &ref_root_bh);
2555 if (ret) {
2556 mlog_errno(ret);
2557 goto out;
2558 }
2559
2560 ret = ocfs2_calc_refcount_meta_credits(inode->i_sb,
2561 &tree->rf_ci,
2562 ref_root_bh,
2563 start_cpos, clusters,
2564 ref_blocks, credits);
2565 if (ret) {
2566 mlog_errno(ret);
2567 goto out;
2568 }
2569
2570 mlog(0, "reserve new metadata %d blocks, credits = %d\n",
2571 *ref_blocks, *credits);
2572
2573 out:
2574 brelse(ref_root_bh);
2575 return ret;
2576 }
2577
2578 #define MAX_CONTIG_BYTES 1048576
2579
2580 static inline unsigned int ocfs2_cow_contig_clusters(struct super_block *sb)
2581 {
2582 return ocfs2_clusters_for_bytes(sb, MAX_CONTIG_BYTES);
2583 }
2584
2585 static inline unsigned int ocfs2_cow_contig_mask(struct super_block *sb)
2586 {
2587 return ~(ocfs2_cow_contig_clusters(sb) - 1);
2588 }
2589
2590 /*
2591 * Given an extent that starts at 'start' and an I/O that starts at 'cpos',
2592 * find an offset (start + (n * contig_clusters)) that is closest to cpos
2593 * while still being less than or equal to it.
2594 *
2595 * The goal is to break the extent at a multiple of contig_clusters.
2596 */
2597 static inline unsigned int ocfs2_cow_align_start(struct super_block *sb,
2598 unsigned int start,
2599 unsigned int cpos)
2600 {
2601 BUG_ON(start > cpos);
2602
2603 return start + ((cpos - start) & ocfs2_cow_contig_mask(sb));
2604 }
2605
2606 /*
2607 * Given a cluster count of len, pad it out so that it is a multiple
2608 * of contig_clusters.
2609 */
2610 static inline unsigned int ocfs2_cow_align_length(struct super_block *sb,
2611 unsigned int len)
2612 {
2613 unsigned int padded =
2614 (len + (ocfs2_cow_contig_clusters(sb) - 1)) &
2615 ocfs2_cow_contig_mask(sb);
2616
2617 /* Did we wrap? */
2618 if (padded < len)
2619 padded = UINT_MAX;
2620
2621 return padded;
2622 }
2623
2624 /*
2625 * Calculate out the start and number of virtual clusters we need to to CoW.
2626 *
2627 * cpos is vitual start cluster position we want to do CoW in a
2628 * file and write_len is the cluster length.
2629 * max_cpos is the place where we want to stop CoW intentionally.
2630 *
2631 * Normal we will start CoW from the beginning of extent record cotaining cpos.
2632 * We try to break up extents on boundaries of MAX_CONTIG_BYTES so that we
2633 * get good I/O from the resulting extent tree.
2634 */
2635 static int ocfs2_refcount_cal_cow_clusters(struct inode *inode,
2636 struct ocfs2_extent_list *el,
2637 u32 cpos,
2638 u32 write_len,
2639 u32 max_cpos,
2640 u32 *cow_start,
2641 u32 *cow_len)
2642 {
2643 int ret = 0;
2644 int tree_height = le16_to_cpu(el->l_tree_depth), i;
2645 struct buffer_head *eb_bh = NULL;
2646 struct ocfs2_extent_block *eb = NULL;
2647 struct ocfs2_extent_rec *rec;
2648 unsigned int want_clusters, rec_end = 0;
2649 int contig_clusters = ocfs2_cow_contig_clusters(inode->i_sb);
2650 int leaf_clusters;
2651
2652 BUG_ON(cpos + write_len > max_cpos);
2653
2654 if (tree_height > 0) {
2655 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, cpos, &eb_bh);
2656 if (ret) {
2657 mlog_errno(ret);
2658 goto out;
2659 }
2660
2661 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2662 el = &eb->h_list;
2663
2664 if (el->l_tree_depth) {
2665 ocfs2_error(inode->i_sb,
2666 "Inode %lu has non zero tree depth in "
2667 "leaf block %llu\n", inode->i_ino,
2668 (unsigned long long)eb_bh->b_blocknr);
2669 ret = -EROFS;
2670 goto out;
2671 }
2672 }
2673
2674 *cow_len = 0;
2675 for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
2676 rec = &el->l_recs[i];
2677
2678 if (ocfs2_is_empty_extent(rec)) {
2679 mlog_bug_on_msg(i != 0, "Inode %lu has empty record in "
2680 "index %d\n", inode->i_ino, i);
2681 continue;
2682 }
2683
2684 if (le32_to_cpu(rec->e_cpos) +
2685 le16_to_cpu(rec->e_leaf_clusters) <= cpos)
2686 continue;
2687
2688 if (*cow_len == 0) {
2689 /*
2690 * We should find a refcounted record in the
2691 * first pass.
2692 */
2693 BUG_ON(!(rec->e_flags & OCFS2_EXT_REFCOUNTED));
2694 *cow_start = le32_to_cpu(rec->e_cpos);
2695 }
2696
2697 /*
2698 * If we encounter a hole, a non-refcounted record or
2699 * pass the max_cpos, stop the search.
2700 */
2701 if ((!(rec->e_flags & OCFS2_EXT_REFCOUNTED)) ||
2702 (*cow_len && rec_end != le32_to_cpu(rec->e_cpos)) ||
2703 (max_cpos <= le32_to_cpu(rec->e_cpos)))
2704 break;
2705
2706 leaf_clusters = le16_to_cpu(rec->e_leaf_clusters);
2707 rec_end = le32_to_cpu(rec->e_cpos) + leaf_clusters;
2708 if (rec_end > max_cpos) {
2709 rec_end = max_cpos;
2710 leaf_clusters = rec_end - le32_to_cpu(rec->e_cpos);
2711 }
2712
2713 /*
2714 * How many clusters do we actually need from
2715 * this extent? First we see how many we actually
2716 * need to complete the write. If that's smaller
2717 * than contig_clusters, we try for contig_clusters.
2718 */
2719 if (!*cow_len)
2720 want_clusters = write_len;
2721 else
2722 want_clusters = (cpos + write_len) -
2723 (*cow_start + *cow_len);
2724 if (want_clusters < contig_clusters)
2725 want_clusters = contig_clusters;
2726
2727 /*
2728 * If the write does not cover the whole extent, we
2729 * need to calculate how we're going to split the extent.
2730 * We try to do it on contig_clusters boundaries.
2731 *
2732 * Any extent smaller than contig_clusters will be
2733 * CoWed in its entirety.
2734 */
2735 if (leaf_clusters <= contig_clusters)
2736 *cow_len += leaf_clusters;
2737 else if (*cow_len || (*cow_start == cpos)) {
2738 /*
2739 * This extent needs to be CoW'd from its
2740 * beginning, so all we have to do is compute
2741 * how many clusters to grab. We align
2742 * want_clusters to the edge of contig_clusters
2743 * to get better I/O.
2744 */
2745 want_clusters = ocfs2_cow_align_length(inode->i_sb,
2746 want_clusters);
2747
2748 if (leaf_clusters < want_clusters)
2749 *cow_len += leaf_clusters;
2750 else
2751 *cow_len += want_clusters;
2752 } else if ((*cow_start + contig_clusters) >=
2753 (cpos + write_len)) {
2754 /*
2755 * Breaking off contig_clusters at the front
2756 * of the extent will cover our write. That's
2757 * easy.
2758 */
2759 *cow_len = contig_clusters;
2760 } else if ((rec_end - cpos) <= contig_clusters) {
2761 /*
2762 * Breaking off contig_clusters at the tail of
2763 * this extent will cover cpos.
2764 */
2765 *cow_start = rec_end - contig_clusters;
2766 *cow_len = contig_clusters;
2767 } else if ((rec_end - cpos) <= want_clusters) {
2768 /*
2769 * While we can't fit the entire write in this
2770 * extent, we know that the write goes from cpos
2771 * to the end of the extent. Break that off.
2772 * We try to break it at some multiple of
2773 * contig_clusters from the front of the extent.
2774 * Failing that (ie, cpos is within
2775 * contig_clusters of the front), we'll CoW the
2776 * entire extent.
2777 */
2778 *cow_start = ocfs2_cow_align_start(inode->i_sb,
2779 *cow_start, cpos);
2780 *cow_len = rec_end - *cow_start;
2781 } else {
2782 /*
2783 * Ok, the entire write lives in the middle of
2784 * this extent. Let's try to slice the extent up
2785 * nicely. Optimally, our CoW region starts at
2786 * m*contig_clusters from the beginning of the
2787 * extent and goes for n*contig_clusters,
2788 * covering the entire write.
2789 */
2790 *cow_start = ocfs2_cow_align_start(inode->i_sb,
2791 *cow_start, cpos);
2792
2793 want_clusters = (cpos + write_len) - *cow_start;
2794 want_clusters = ocfs2_cow_align_length(inode->i_sb,
2795 want_clusters);
2796 if (*cow_start + want_clusters <= rec_end)
2797 *cow_len = want_clusters;
2798 else
2799 *cow_len = rec_end - *cow_start;
2800 }
2801
2802 /* Have we covered our entire write yet? */
2803 if ((*cow_start + *cow_len) >= (cpos + write_len))
2804 break;
2805
2806 /*
2807 * If we reach the end of the extent block and don't get enough
2808 * clusters, continue with the next extent block if possible.
2809 */
2810 if (i + 1 == le16_to_cpu(el->l_next_free_rec) &&
2811 eb && eb->h_next_leaf_blk) {
2812 brelse(eb_bh);
2813 eb_bh = NULL;
2814
2815 ret = ocfs2_read_extent_block(INODE_CACHE(inode),
2816 le64_to_cpu(eb->h_next_leaf_blk),
2817 &eb_bh);
2818 if (ret) {
2819 mlog_errno(ret);
2820 goto out;
2821 }
2822
2823 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2824 el = &eb->h_list;
2825 i = -1;
2826 }
2827 }
2828
2829 out:
2830 brelse(eb_bh);
2831 return ret;
2832 }
2833
2834 /*
2835 * Prepare meta_ac, data_ac and calculate credits when we want to add some
2836 * num_clusters in data_tree "et" and change the refcount for the old
2837 * clusters(starting form p_cluster) in the refcount tree.
2838 *
2839 * Note:
2840 * 1. since we may split the old tree, so we at most will need num_clusters + 2
2841 * more new leaf records.
2842 * 2. In some case, we may not need to reserve new clusters(e.g, reflink), so
2843 * just give data_ac = NULL.
2844 */
2845 static int ocfs2_lock_refcount_allocators(struct super_block *sb,
2846 u32 p_cluster, u32 num_clusters,
2847 struct ocfs2_extent_tree *et,
2848 struct ocfs2_caching_info *ref_ci,
2849 struct buffer_head *ref_root_bh,
2850 struct ocfs2_alloc_context **meta_ac,
2851 struct ocfs2_alloc_context **data_ac,
2852 int *credits)
2853 {
2854 int ret = 0, meta_add = 0;
2855 int num_free_extents = ocfs2_num_free_extents(OCFS2_SB(sb), et);
2856
2857 if (num_free_extents < 0) {
2858 ret = num_free_extents;
2859 mlog_errno(ret);
2860 goto out;
2861 }
2862
2863 if (num_free_extents < num_clusters + 2)
2864 meta_add =
2865 ocfs2_extend_meta_needed(et->et_root_el);
2866
2867 *credits += ocfs2_calc_extend_credits(sb, et->et_root_el,
2868 num_clusters + 2);
2869
2870 ret = ocfs2_calc_refcount_meta_credits(sb, ref_ci, ref_root_bh,
2871 p_cluster, num_clusters,
2872 &meta_add, credits);
2873 if (ret) {
2874 mlog_errno(ret);
2875 goto out;
2876 }
2877
2878 mlog(0, "reserve new metadata %d, clusters %u, credits = %d\n",
2879 meta_add, num_clusters, *credits);
2880 ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(sb), meta_add,
2881 meta_ac);
2882 if (ret) {
2883 mlog_errno(ret);
2884 goto out;
2885 }
2886
2887 if (data_ac) {
2888 ret = ocfs2_reserve_clusters(OCFS2_SB(sb), num_clusters,
2889 data_ac);
2890 if (ret)
2891 mlog_errno(ret);
2892 }
2893
2894 out:
2895 if (ret) {
2896 if (*meta_ac) {
2897 ocfs2_free_alloc_context(*meta_ac);
2898 *meta_ac = NULL;
2899 }
2900 }
2901
2902 return ret;
2903 }
2904
2905 static int ocfs2_clear_cow_buffer(handle_t *handle, struct buffer_head *bh)
2906 {
2907 BUG_ON(buffer_dirty(bh));
2908
2909 clear_buffer_mapped(bh);
2910
2911 return 0;
2912 }
2913
2914 static int ocfs2_duplicate_clusters_by_page(handle_t *handle,
2915 struct ocfs2_cow_context *context,
2916 u32 cpos, u32 old_cluster,
2917 u32 new_cluster, u32 new_len)
2918 {
2919 int ret = 0, partial;
2920 struct ocfs2_caching_info *ci = context->data_et.et_ci;
2921 struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
2922 u64 new_block = ocfs2_clusters_to_blocks(sb, new_cluster);
2923 struct page *page;
2924 pgoff_t page_index;
2925 unsigned int from, to;
2926 loff_t offset, end, map_end;
2927 struct address_space *mapping = context->inode->i_mapping;
2928
2929 mlog(0, "old_cluster %u, new %u, len %u at offset %u\n", old_cluster,
2930 new_cluster, new_len, cpos);
2931
2932 offset = ((loff_t)cpos) << OCFS2_SB(sb)->s_clustersize_bits;
2933 end = offset + (new_len << OCFS2_SB(sb)->s_clustersize_bits);
2934 /*
2935 * We only duplicate pages until we reach the page contains i_size - 1.
2936 * So trim 'end' to i_size.
2937 */
2938 if (end > i_size_read(context->inode))
2939 end = i_size_read(context->inode);
2940
2941 while (offset < end) {
2942 page_index = offset >> PAGE_CACHE_SHIFT;
2943 map_end = ((loff_t)page_index + 1) << PAGE_CACHE_SHIFT;
2944 if (map_end > end)
2945 map_end = end;
2946
2947 /* from, to is the offset within the page. */
2948 from = offset & (PAGE_CACHE_SIZE - 1);
2949 to = PAGE_CACHE_SIZE;
2950 if (map_end & (PAGE_CACHE_SIZE - 1))
2951 to = map_end & (PAGE_CACHE_SIZE - 1);
2952
2953 page = grab_cache_page(mapping, page_index);
2954
2955 /*
2956 * In case PAGE_CACHE_SIZE <= CLUSTER_SIZE, This page
2957 * can't be dirtied before we CoW it out.
2958 */
2959 if (PAGE_CACHE_SIZE <= OCFS2_SB(sb)->s_clustersize)
2960 BUG_ON(PageDirty(page));
2961
2962 if (!PageUptodate(page)) {
2963 ret = block_read_full_page(page, ocfs2_get_block);
2964 if (ret) {
2965 mlog_errno(ret);
2966 goto unlock;
2967 }
2968 lock_page(page);
2969 }
2970
2971 if (page_has_buffers(page)) {
2972 ret = walk_page_buffers(handle, page_buffers(page),
2973 from, to, &partial,
2974 ocfs2_clear_cow_buffer);
2975 if (ret) {
2976 mlog_errno(ret);
2977 goto unlock;
2978 }
2979 }
2980
2981 ocfs2_map_and_dirty_page(context->inode,
2982 handle, from, to,
2983 page, 0, &new_block);
2984 mark_page_accessed(page);
2985 unlock:
2986 unlock_page(page);
2987 page_cache_release(page);
2988 page = NULL;
2989 offset = map_end;
2990 if (ret)
2991 break;
2992 }
2993
2994 return ret;
2995 }
2996
2997 static int ocfs2_duplicate_clusters_by_jbd(handle_t *handle,
2998 struct ocfs2_cow_context *context,
2999 u32 cpos, u32 old_cluster,
3000 u32 new_cluster, u32 new_len)
3001 {
3002 int ret = 0;
3003 struct super_block *sb = context->inode->i_sb;
3004 struct ocfs2_caching_info *ci = context->data_et.et_ci;
3005 int i, blocks = ocfs2_clusters_to_blocks(sb, new_len);
3006 u64 old_block = ocfs2_clusters_to_blocks(sb, old_cluster);
3007 u64 new_block = ocfs2_clusters_to_blocks(sb, new_cluster);
3008 struct ocfs2_super *osb = OCFS2_SB(sb);
3009 struct buffer_head *old_bh = NULL;
3010 struct buffer_head *new_bh = NULL;
3011
3012 mlog(0, "old_cluster %u, new %u, len %u\n", old_cluster,
3013 new_cluster, new_len);
3014
3015 for (i = 0; i < blocks; i++, old_block++, new_block++) {
3016 new_bh = sb_getblk(osb->sb, new_block);
3017 if (new_bh == NULL) {
3018 ret = -EIO;
3019 mlog_errno(ret);
3020 break;
3021 }
3022
3023 ocfs2_set_new_buffer_uptodate(ci, new_bh);
3024
3025 ret = ocfs2_read_block(ci, old_block, &old_bh, NULL);
3026 if (ret) {
3027 mlog_errno(ret);
3028 break;
3029 }
3030
3031 ret = ocfs2_journal_access(handle, ci, new_bh,
3032 OCFS2_JOURNAL_ACCESS_CREATE);
3033 if (ret) {
3034 mlog_errno(ret);
3035 break;
3036 }
3037
3038 memcpy(new_bh->b_data, old_bh->b_data, sb->s_blocksize);
3039 ocfs2_journal_dirty(handle, new_bh);
3040
3041 brelse(new_bh);
3042 brelse(old_bh);
3043 new_bh = NULL;
3044 old_bh = NULL;
3045 }
3046
3047 brelse(new_bh);
3048 brelse(old_bh);
3049 return ret;
3050 }
3051
3052 static int ocfs2_clear_ext_refcount(handle_t *handle,
3053 struct ocfs2_extent_tree *et,
3054 u32 cpos, u32 p_cluster, u32 len,
3055 unsigned int ext_flags,
3056 struct ocfs2_alloc_context *meta_ac,
3057 struct ocfs2_cached_dealloc_ctxt *dealloc)
3058 {
3059 int ret, index;
3060 struct ocfs2_extent_rec replace_rec;
3061 struct ocfs2_path *path = NULL;
3062 struct ocfs2_extent_list *el;
3063 struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
3064 u64 ino = ocfs2_metadata_cache_owner(et->et_ci);
3065
3066 mlog(0, "inode %llu cpos %u, len %u, p_cluster %u, ext_flags %u\n",
3067 (unsigned long long)ino, cpos, len, p_cluster, ext_flags);
3068
3069 memset(&replace_rec, 0, sizeof(replace_rec));
3070 replace_rec.e_cpos = cpu_to_le32(cpos);
3071 replace_rec.e_leaf_clusters = cpu_to_le16(len);
3072 replace_rec.e_blkno = cpu_to_le64(ocfs2_clusters_to_blocks(sb,
3073 p_cluster));
3074 replace_rec.e_flags = ext_flags;
3075 replace_rec.e_flags &= ~OCFS2_EXT_REFCOUNTED;
3076
3077 path = ocfs2_new_path_from_et(et);
3078 if (!path) {
3079 ret = -ENOMEM;
3080 mlog_errno(ret);
3081 goto out;
3082 }
3083
3084 ret = ocfs2_find_path(et->et_ci, path, cpos);
3085 if (ret) {
3086 mlog_errno(ret);
3087 goto out;
3088 }
3089
3090 el = path_leaf_el(path);
3091
3092 index = ocfs2_search_extent_list(el, cpos);
3093 if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) {
3094 ocfs2_error(sb,
3095 "Inode %llu has an extent at cpos %u which can no "
3096 "longer be found.\n",
3097 (unsigned long long)ino, cpos);
3098 ret = -EROFS;
3099 goto out;
3100 }
3101
3102 ret = ocfs2_split_extent(handle, et, path, index,
3103 &replace_rec, meta_ac, dealloc);
3104 if (ret)
3105 mlog_errno(ret);
3106
3107 out:
3108 ocfs2_free_path(path);
3109 return ret;
3110 }
3111
3112 static int ocfs2_replace_clusters(handle_t *handle,
3113 struct ocfs2_cow_context *context,
3114 u32 cpos, u32 old,
3115 u32 new, u32 len,
3116 unsigned int ext_flags)
3117 {
3118 int ret;
3119 struct ocfs2_caching_info *ci = context->data_et.et_ci;
3120 u64 ino = ocfs2_metadata_cache_owner(ci);
3121
3122 mlog(0, "inode %llu, cpos %u, old %u, new %u, len %u, ext_flags %u\n",
3123 (unsigned long long)ino, cpos, old, new, len, ext_flags);
3124
3125 /*If the old clusters is unwritten, no need to duplicate. */
3126 if (!(ext_flags & OCFS2_EXT_UNWRITTEN)) {
3127 ret = context->cow_duplicate_clusters(handle, context, cpos,
3128 old, new, len);
3129 if (ret) {
3130 mlog_errno(ret);
3131 goto out;
3132 }
3133 }
3134
3135 ret = ocfs2_clear_ext_refcount(handle, &context->data_et,
3136 cpos, new, len, ext_flags,
3137 context->meta_ac, &context->dealloc);
3138 if (ret)
3139 mlog_errno(ret);
3140 out:
3141 return ret;
3142 }
3143
3144 static int ocfs2_cow_sync_writeback(struct super_block *sb,
3145 struct ocfs2_cow_context *context,
3146 u32 cpos, u32 num_clusters)
3147 {
3148 int ret = 0;
3149 loff_t offset, end, map_end;
3150 pgoff_t page_index;
3151 struct page *page;
3152
3153 if (ocfs2_should_order_data(context->inode))
3154 return 0;
3155
3156 offset = ((loff_t)cpos) << OCFS2_SB(sb)->s_clustersize_bits;
3157 end = offset + (num_clusters << OCFS2_SB(sb)->s_clustersize_bits);
3158
3159 ret = filemap_fdatawrite_range(context->inode->i_mapping,
3160 offset, end - 1);
3161 if (ret < 0) {
3162 mlog_errno(ret);
3163 return ret;
3164 }
3165
3166 while (offset < end) {
3167 page_index = offset >> PAGE_CACHE_SHIFT;
3168 map_end = ((loff_t)page_index + 1) << PAGE_CACHE_SHIFT;
3169 if (map_end > end)
3170 map_end = end;
3171
3172 page = grab_cache_page(context->inode->i_mapping, page_index);
3173 BUG_ON(!page);
3174
3175 wait_on_page_writeback(page);
3176 if (PageError(page)) {
3177 ret = -EIO;
3178 mlog_errno(ret);
3179 } else
3180 mark_page_accessed(page);
3181
3182 unlock_page(page);
3183 page_cache_release(page);
3184 page = NULL;
3185 offset = map_end;
3186 if (ret)
3187 break;
3188 }
3189
3190 return ret;
3191 }
3192
3193 static int ocfs2_di_get_clusters(struct ocfs2_cow_context *context,
3194 u32 v_cluster, u32 *p_cluster,
3195 u32 *num_clusters,
3196 unsigned int *extent_flags)
3197 {
3198 return ocfs2_get_clusters(context->inode, v_cluster, p_cluster,
3199 num_clusters, extent_flags);
3200 }
3201
3202 static int ocfs2_make_clusters_writable(struct super_block *sb,
3203 struct ocfs2_cow_context *context,
3204 u32 cpos, u32 p_cluster,
3205 u32 num_clusters, unsigned int e_flags)
3206 {
3207 int ret, delete, index, credits = 0;
3208 u32 new_bit, new_len;
3209 unsigned int set_len;
3210 struct ocfs2_super *osb = OCFS2_SB(sb);
3211 handle_t *handle;
3212 struct buffer_head *ref_leaf_bh = NULL;
3213 struct ocfs2_caching_info *ref_ci = &context->ref_tree->rf_ci;
3214 struct ocfs2_refcount_rec rec;
3215
3216 mlog(0, "cpos %u, p_cluster %u, num_clusters %u, e_flags %u\n",
3217 cpos, p_cluster, num_clusters, e_flags);
3218
3219 ret = ocfs2_lock_refcount_allocators(sb, p_cluster, num_clusters,
3220 &context->data_et,
3221 ref_ci,
3222 context->ref_root_bh,
3223 &context->meta_ac,
3224 &context->data_ac, &credits);
3225 if (ret) {
3226 mlog_errno(ret);
3227 return ret;
3228 }
3229
3230 if (context->post_refcount)
3231 credits += context->post_refcount->credits;
3232
3233 credits += context->extra_credits;
3234 handle = ocfs2_start_trans(osb, credits);
3235 if (IS_ERR(handle)) {
3236 ret = PTR_ERR(handle);
3237 mlog_errno(ret);
3238 goto out;
3239 }
3240
3241 while (num_clusters) {
3242 ret = ocfs2_get_refcount_rec(ref_ci, context->ref_root_bh,
3243 p_cluster, num_clusters,
3244 &rec, &index, &ref_leaf_bh);
3245 if (ret) {
3246 mlog_errno(ret);
3247 goto out_commit;
3248 }
3249
3250 BUG_ON(!rec.r_refcount);
3251 set_len = min((u64)p_cluster + num_clusters,
3252 le64_to_cpu(rec.r_cpos) +
3253 le32_to_cpu(rec.r_clusters)) - p_cluster;
3254
3255 /*
3256 * There are many different situation here.
3257 * 1. If refcount == 1, remove the flag and don't COW.
3258 * 2. If refcount > 1, allocate clusters.
3259 * Here we may not allocate r_len once at a time, so continue
3260 * until we reach num_clusters.
3261 */
3262 if (le32_to_cpu(rec.r_refcount) == 1) {
3263 delete = 0;
3264 ret = ocfs2_clear_ext_refcount(handle,
3265 &context->data_et,
3266 cpos, p_cluster,
3267 set_len, e_flags,
3268 context->meta_ac,
3269 &context->dealloc);
3270 if (ret) {
3271 mlog_errno(ret);
3272 goto out_commit;
3273 }
3274 } else {
3275 delete = 1;
3276
3277 ret = __ocfs2_claim_clusters(handle,
3278 context->data_ac,
3279 1, set_len,
3280 &new_bit, &new_len);
3281 if (ret) {
3282 mlog_errno(ret);
3283 goto out_commit;
3284 }
3285
3286 ret = ocfs2_replace_clusters(handle, context,
3287 cpos, p_cluster, new_bit,
3288 new_len, e_flags);
3289 if (ret) {
3290 mlog_errno(ret);
3291 goto out_commit;
3292 }
3293 set_len = new_len;
3294 }
3295
3296 ret = __ocfs2_decrease_refcount(handle, ref_ci,
3297 context->ref_root_bh,
3298 p_cluster, set_len,
3299 context->meta_ac,
3300 &context->dealloc, delete);
3301 if (ret) {
3302 mlog_errno(ret);
3303 goto out_commit;
3304 }
3305
3306 cpos += set_len;
3307 p_cluster += set_len;
3308 num_clusters -= set_len;
3309 brelse(ref_leaf_bh);
3310 ref_leaf_bh = NULL;
3311 }
3312
3313 /* handle any post_cow action. */
3314 if (context->post_refcount && context->post_refcount->func) {
3315 ret = context->post_refcount->func(context->inode, handle,
3316 context->post_refcount->para);
3317 if (ret) {
3318 mlog_errno(ret);
3319 goto out_commit;
3320 }
3321 }
3322
3323 /*
3324 * Here we should write the new page out first if we are
3325 * in write-back mode.
3326 */
3327 if (context->get_clusters == ocfs2_di_get_clusters) {
3328 ret = ocfs2_cow_sync_writeback(sb, context, cpos, num_clusters);
3329 if (ret)
3330 mlog_errno(ret);
3331 }
3332
3333 out_commit:
3334 ocfs2_commit_trans(osb, handle);
3335
3336 out:
3337 if (context->data_ac) {
3338 ocfs2_free_alloc_context(context->data_ac);
3339 context->data_ac = NULL;
3340 }
3341 if (context->meta_ac) {
3342 ocfs2_free_alloc_context(context->meta_ac);
3343 context->meta_ac = NULL;
3344 }
3345 brelse(ref_leaf_bh);
3346
3347 return ret;
3348 }
3349
3350 static int ocfs2_replace_cow(struct ocfs2_cow_context *context)
3351 {
3352 int ret = 0;
3353 struct inode *inode = context->inode;
3354 u32 cow_start = context->cow_start, cow_len = context->cow_len;
3355 u32 p_cluster, num_clusters;
3356 unsigned int ext_flags;
3357 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3358
3359 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
3360 ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
3361 "tree, but the feature bit is not set in the "
3362 "super block.", inode->i_ino);
3363 return -EROFS;
3364 }
3365
3366 ocfs2_init_dealloc_ctxt(&context->dealloc);
3367
3368 while (cow_len) {
3369 ret = context->get_clusters(context, cow_start, &p_cluster,
3370 &num_clusters, &ext_flags);
3371 if (ret) {
3372 mlog_errno(ret);
3373 break;
3374 }
3375
3376 BUG_ON(!(ext_flags & OCFS2_EXT_REFCOUNTED));
3377
3378 if (cow_len < num_clusters)
3379 num_clusters = cow_len;
3380
3381 ret = ocfs2_make_clusters_writable(inode->i_sb, context,
3382 cow_start, p_cluster,
3383 num_clusters, ext_flags);
3384 if (ret) {
3385 mlog_errno(ret);
3386 break;
3387 }
3388
3389 cow_len -= num_clusters;
3390 cow_start += num_clusters;
3391 }
3392
3393 if (ocfs2_dealloc_has_cluster(&context->dealloc)) {
3394 ocfs2_schedule_truncate_log_flush(osb, 1);
3395 ocfs2_run_deallocs(osb, &context->dealloc);
3396 }
3397
3398 return ret;
3399 }
3400
3401 /*
3402 * Starting at cpos, try to CoW write_len clusters. Don't CoW
3403 * past max_cpos. This will stop when it runs into a hole or an
3404 * unrefcounted extent.
3405 */
3406 static int ocfs2_refcount_cow_hunk(struct inode *inode,
3407 struct buffer_head *di_bh,
3408 u32 cpos, u32 write_len, u32 max_cpos)
3409 {
3410 int ret;
3411 u32 cow_start = 0, cow_len = 0;
3412 struct ocfs2_inode_info *oi = OCFS2_I(inode);
3413 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3414 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3415 struct buffer_head *ref_root_bh = NULL;
3416 struct ocfs2_refcount_tree *ref_tree;
3417 struct ocfs2_cow_context *context = NULL;
3418
3419 BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
3420
3421 ret = ocfs2_refcount_cal_cow_clusters(inode, &di->id2.i_list,
3422 cpos, write_len, max_cpos,
3423 &cow_start, &cow_len);
3424 if (ret) {
3425 mlog_errno(ret);
3426 goto out;
3427 }
3428
3429 mlog(0, "CoW inode %lu, cpos %u, write_len %u, cow_start %u, "
3430 "cow_len %u\n", inode->i_ino,
3431 cpos, write_len, cow_start, cow_len);
3432
3433 BUG_ON(cow_len == 0);
3434
3435 context = kzalloc(sizeof(struct ocfs2_cow_context), GFP_NOFS);
3436 if (!context) {
3437 ret = -ENOMEM;
3438 mlog_errno(ret);
3439 goto out;
3440 }
3441
3442 ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
3443 1, &ref_tree, &ref_root_bh);
3444 if (ret) {
3445 mlog_errno(ret);
3446 goto out;
3447 }
3448
3449 context->inode = inode;
3450 context->cow_start = cow_start;
3451 context->cow_len = cow_len;
3452 context->ref_tree = ref_tree;
3453 context->ref_root_bh = ref_root_bh;
3454 context->cow_duplicate_clusters = ocfs2_duplicate_clusters_by_page;
3455 context->get_clusters = ocfs2_di_get_clusters;
3456
3457 ocfs2_init_dinode_extent_tree(&context->data_et,
3458 INODE_CACHE(inode), di_bh);
3459
3460 ret = ocfs2_replace_cow(context);
3461 if (ret)
3462 mlog_errno(ret);
3463
3464 /*
3465 * truncate the extent map here since no matter whether we meet with
3466 * any error during the action, we shouldn't trust cached extent map
3467 * any more.
3468 */
3469 ocfs2_extent_map_trunc(inode, cow_start);
3470
3471 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3472 brelse(ref_root_bh);
3473 out:
3474 kfree(context);
3475 return ret;
3476 }
3477
3478 /*
3479 * CoW any and all clusters between cpos and cpos+write_len.
3480 * Don't CoW past max_cpos. If this returns successfully, all
3481 * clusters between cpos and cpos+write_len are safe to modify.
3482 */
3483 int ocfs2_refcount_cow(struct inode *inode,
3484 struct buffer_head *di_bh,
3485 u32 cpos, u32 write_len, u32 max_cpos)
3486 {
3487 int ret = 0;
3488 u32 p_cluster, num_clusters;
3489 unsigned int ext_flags;
3490
3491 while (write_len) {
3492 ret = ocfs2_get_clusters(inode, cpos, &p_cluster,
3493 &num_clusters, &ext_flags);
3494 if (ret) {
3495 mlog_errno(ret);
3496 break;
3497 }
3498
3499 if (write_len < num_clusters)
3500 num_clusters = write_len;
3501
3502 if (ext_flags & OCFS2_EXT_REFCOUNTED) {
3503 ret = ocfs2_refcount_cow_hunk(inode, di_bh, cpos,
3504 num_clusters, max_cpos);
3505 if (ret) {
3506 mlog_errno(ret);
3507 break;
3508 }
3509 }
3510
3511 write_len -= num_clusters;
3512 cpos += num_clusters;
3513 }
3514
3515 return ret;
3516 }
3517
3518 static int ocfs2_xattr_value_get_clusters(struct ocfs2_cow_context *context,
3519 u32 v_cluster, u32 *p_cluster,
3520 u32 *num_clusters,
3521 unsigned int *extent_flags)
3522 {
3523 struct inode *inode = context->inode;
3524 struct ocfs2_xattr_value_root *xv = context->cow_object;
3525
3526 return ocfs2_xattr_get_clusters(inode, v_cluster, p_cluster,
3527 num_clusters, &xv->xr_list,
3528 extent_flags);
3529 }
3530
3531 /*
3532 * Given a xattr value root, calculate the most meta/credits we need for
3533 * refcount tree change if we truncate it to 0.
3534 */
3535 int ocfs2_refcounted_xattr_delete_need(struct inode *inode,
3536 struct ocfs2_caching_info *ref_ci,
3537 struct buffer_head *ref_root_bh,
3538 struct ocfs2_xattr_value_root *xv,
3539 int *meta_add, int *credits)
3540 {
3541 int ret = 0, index, ref_blocks = 0;
3542 u32 p_cluster, num_clusters;
3543 u32 cpos = 0, clusters = le32_to_cpu(xv->xr_clusters);
3544 struct ocfs2_refcount_block *rb;
3545 struct ocfs2_refcount_rec rec;
3546 struct buffer_head *ref_leaf_bh = NULL;
3547
3548 while (cpos < clusters) {
3549 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
3550 &num_clusters, &xv->xr_list,
3551 NULL);
3552 if (ret) {
3553 mlog_errno(ret);
3554 goto out;
3555 }
3556
3557 cpos += num_clusters;
3558
3559 while (num_clusters) {
3560 ret = ocfs2_get_refcount_rec(ref_ci, ref_root_bh,
3561 p_cluster, num_clusters,
3562 &rec, &index,
3563 &ref_leaf_bh);
3564 if (ret) {
3565 mlog_errno(ret);
3566 goto out;
3567 }
3568
3569 BUG_ON(!rec.r_refcount);
3570
3571 rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
3572
3573 /*
3574 * We really don't know whether the other clusters is in
3575 * this refcount block or not, so just take the worst
3576 * case that all the clusters are in this block and each
3577 * one will split a refcount rec, so totally we need
3578 * clusters * 2 new refcount rec.
3579 */
3580 if (le64_to_cpu(rb->rf_records.rl_used) + clusters * 2 >
3581 le16_to_cpu(rb->rf_records.rl_count))
3582 ref_blocks++;
3583
3584 *credits += 1;
3585 brelse(ref_leaf_bh);
3586 ref_leaf_bh = NULL;
3587
3588 if (num_clusters <= le32_to_cpu(rec.r_clusters))
3589 break;
3590 else
3591 num_clusters -= le32_to_cpu(rec.r_clusters);
3592 p_cluster += num_clusters;
3593 }
3594 }
3595
3596 *meta_add += ref_blocks;
3597 if (!ref_blocks)
3598 goto out;
3599
3600 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
3601 if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
3602 *credits += OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
3603 else {
3604 struct ocfs2_extent_tree et;
3605
3606 ocfs2_init_refcount_extent_tree(&et, ref_ci, ref_root_bh);
3607 *credits += ocfs2_calc_extend_credits(inode->i_sb,
3608 et.et_root_el,
3609 ref_blocks);
3610 }
3611
3612 out:
3613 brelse(ref_leaf_bh);
3614 return ret;
3615 }
3616
3617 /*
3618 * Do CoW for xattr.
3619 */
3620 int ocfs2_refcount_cow_xattr(struct inode *inode,
3621 struct ocfs2_dinode *di,
3622 struct ocfs2_xattr_value_buf *vb,
3623 struct ocfs2_refcount_tree *ref_tree,
3624 struct buffer_head *ref_root_bh,
3625 u32 cpos, u32 write_len,
3626 struct ocfs2_post_refcount *post)
3627 {
3628 int ret;
3629 struct ocfs2_xattr_value_root *xv = vb->vb_xv;
3630 struct ocfs2_inode_info *oi = OCFS2_I(inode);
3631 struct ocfs2_cow_context *context = NULL;
3632 u32 cow_start, cow_len;
3633
3634 BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
3635
3636 ret = ocfs2_refcount_cal_cow_clusters(inode, &xv->xr_list,
3637 cpos, write_len, UINT_MAX,
3638 &cow_start, &cow_len);
3639 if (ret) {
3640 mlog_errno(ret);
3641 goto out;
3642 }
3643
3644 BUG_ON(cow_len == 0);
3645
3646 context = kzalloc(sizeof(struct ocfs2_cow_context), GFP_NOFS);
3647 if (!context) {
3648 ret = -ENOMEM;
3649 mlog_errno(ret);
3650 goto out;
3651 }
3652
3653 context->inode = inode;
3654 context->cow_start = cow_start;
3655 context->cow_len = cow_len;
3656 context->ref_tree = ref_tree;
3657 context->ref_root_bh = ref_root_bh;;
3658 context->cow_object = xv;
3659
3660 context->cow_duplicate_clusters = ocfs2_duplicate_clusters_by_jbd;
3661 /* We need the extra credits for duplicate_clusters by jbd. */
3662 context->extra_credits =
3663 ocfs2_clusters_to_blocks(inode->i_sb, 1) * cow_len;
3664 context->get_clusters = ocfs2_xattr_value_get_clusters;
3665 context->post_refcount = post;
3666
3667 ocfs2_init_xattr_value_extent_tree(&context->data_et,
3668 INODE_CACHE(inode), vb);
3669
3670 ret = ocfs2_replace_cow(context);
3671 if (ret)
3672 mlog_errno(ret);
3673
3674 out:
3675 kfree(context);
3676 return ret;
3677 }
3678
3679 /*
3680 * Insert a new extent into refcount tree and mark a extent rec
3681 * as refcounted in the dinode tree.
3682 */
3683 int ocfs2_add_refcount_flag(struct inode *inode,
3684 struct ocfs2_extent_tree *data_et,
3685 struct ocfs2_caching_info *ref_ci,
3686 struct buffer_head *ref_root_bh,
3687 u32 cpos, u32 p_cluster, u32 num_clusters,
3688 struct ocfs2_cached_dealloc_ctxt *dealloc,
3689 struct ocfs2_post_refcount *post)
3690 {
3691 int ret;
3692 handle_t *handle;
3693 int credits = 1, ref_blocks = 0;
3694 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3695 struct ocfs2_alloc_context *meta_ac = NULL;
3696
3697 ret = ocfs2_calc_refcount_meta_credits(inode->i_sb,
3698 ref_ci, ref_root_bh,
3699 p_cluster, num_clusters,
3700 &ref_blocks, &credits);
3701 if (ret) {
3702 mlog_errno(ret);
3703 goto out;
3704 }
3705
3706 mlog(0, "reserve new metadata %d, credits = %d\n",
3707 ref_blocks, credits);
3708
3709 if (ref_blocks) {
3710 ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
3711 ref_blocks, &meta_ac);
3712 if (ret) {
3713 mlog_errno(ret);
3714 goto out;
3715 }
3716 }
3717
3718 if (post)
3719 credits += post->credits;
3720
3721 handle = ocfs2_start_trans(osb, credits);
3722 if (IS_ERR(handle)) {
3723 ret = PTR_ERR(handle);
3724 mlog_errno(ret);
3725 goto out;
3726 }
3727
3728 ret = ocfs2_mark_extent_refcounted(inode, data_et, handle,
3729 cpos, num_clusters, p_cluster,
3730 meta_ac, dealloc);
3731 if (ret) {
3732 mlog_errno(ret);
3733 goto out_commit;
3734 }
3735
3736 ret = __ocfs2_increase_refcount(handle, ref_ci, ref_root_bh,
3737 p_cluster, num_clusters, 0,
3738 meta_ac, dealloc);
3739 if (ret) {
3740 mlog_errno(ret);
3741 goto out_commit;
3742 }
3743
3744 if (post && post->func) {
3745 ret = post->func(inode, handle, post->para);
3746 if (ret)
3747 mlog_errno(ret);
3748 }
3749
3750 out_commit:
3751 ocfs2_commit_trans(osb, handle);
3752 out:
3753 if (meta_ac)
3754 ocfs2_free_alloc_context(meta_ac);
3755 return ret;
3756 }
3757
3758 static int ocfs2_change_ctime(struct inode *inode,
3759 struct buffer_head *di_bh)
3760 {
3761 int ret;
3762 handle_t *handle;
3763 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3764
3765 handle = ocfs2_start_trans(OCFS2_SB(inode->i_sb),
3766 OCFS2_INODE_UPDATE_CREDITS);
3767 if (IS_ERR(handle)) {
3768 ret = PTR_ERR(handle);
3769 mlog_errno(ret);
3770 goto out;
3771 }
3772
3773 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
3774 OCFS2_JOURNAL_ACCESS_WRITE);
3775 if (ret) {
3776 mlog_errno(ret);
3777 goto out_commit;
3778 }
3779
3780 inode->i_ctime = CURRENT_TIME;
3781 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
3782 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
3783
3784 ocfs2_journal_dirty(handle, di_bh);
3785
3786 out_commit:
3787 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
3788 out:
3789 return ret;
3790 }
3791
3792 static int ocfs2_attach_refcount_tree(struct inode *inode,
3793 struct buffer_head *di_bh)
3794 {
3795 int ret, data_changed = 0;
3796 struct buffer_head *ref_root_bh = NULL;
3797 struct ocfs2_inode_info *oi = OCFS2_I(inode);
3798 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3799 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3800 struct ocfs2_refcount_tree *ref_tree;
3801 unsigned int ext_flags;
3802 loff_t size;
3803 u32 cpos, num_clusters, clusters, p_cluster;
3804 struct ocfs2_cached_dealloc_ctxt dealloc;
3805 struct ocfs2_extent_tree di_et;
3806
3807 ocfs2_init_dealloc_ctxt(&dealloc);
3808
3809 if (!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL)) {
3810 ret = ocfs2_create_refcount_tree(inode, di_bh);
3811 if (ret) {
3812 mlog_errno(ret);
3813 goto out;
3814 }
3815 }
3816
3817 BUG_ON(!di->i_refcount_loc);
3818 ret = ocfs2_lock_refcount_tree(osb,
3819 le64_to_cpu(di->i_refcount_loc), 1,
3820 &ref_tree, &ref_root_bh);
3821 if (ret) {
3822 mlog_errno(ret);
3823 goto out;
3824 }
3825
3826 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
3827 goto attach_xattr;
3828
3829 ocfs2_init_dinode_extent_tree(&di_et, INODE_CACHE(inode), di_bh);
3830
3831 size = i_size_read(inode);
3832 clusters = ocfs2_clusters_for_bytes(inode->i_sb, size);
3833
3834 cpos = 0;
3835 while (cpos < clusters) {
3836 ret = ocfs2_get_clusters(inode, cpos, &p_cluster,
3837 &num_clusters, &ext_flags);
3838
3839 if (p_cluster && !(ext_flags & OCFS2_EXT_REFCOUNTED)) {
3840 ret = ocfs2_add_refcount_flag(inode, &di_et,
3841 &ref_tree->rf_ci,
3842 ref_root_bh, cpos,
3843 p_cluster, num_clusters,
3844 &dealloc, NULL);
3845 if (ret) {
3846 mlog_errno(ret);
3847 goto unlock;
3848 }
3849
3850 data_changed = 1;
3851 }
3852 cpos += num_clusters;
3853 }
3854
3855 attach_xattr:
3856 if (oi->ip_dyn_features & OCFS2_HAS_XATTR_FL) {
3857 ret = ocfs2_xattr_attach_refcount_tree(inode, di_bh,
3858 &ref_tree->rf_ci,
3859 ref_root_bh,
3860 &dealloc);
3861 if (ret) {
3862 mlog_errno(ret);
3863 goto unlock;
3864 }
3865 }
3866
3867 if (data_changed) {
3868 ret = ocfs2_change_ctime(inode, di_bh);
3869 if (ret)
3870 mlog_errno(ret);
3871 }
3872
3873 unlock:
3874 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3875 brelse(ref_root_bh);
3876
3877 if (!ret && ocfs2_dealloc_has_cluster(&dealloc)) {
3878 ocfs2_schedule_truncate_log_flush(osb, 1);
3879 ocfs2_run_deallocs(osb, &dealloc);
3880 }
3881 out:
3882 /*
3883 * Empty the extent map so that we may get the right extent
3884 * record from the disk.
3885 */
3886 ocfs2_extent_map_trunc(inode, 0);
3887
3888 return ret;
3889 }
3890
3891 static int ocfs2_add_refcounted_extent(struct inode *inode,
3892 struct ocfs2_extent_tree *et,
3893 struct ocfs2_caching_info *ref_ci,
3894 struct buffer_head *ref_root_bh,
3895 u32 cpos, u32 p_cluster, u32 num_clusters,
3896 unsigned int ext_flags,
3897 struct ocfs2_cached_dealloc_ctxt *dealloc)
3898 {
3899 int ret;
3900 handle_t *handle;
3901 int credits = 0;
3902 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3903 struct ocfs2_alloc_context *meta_ac = NULL;
3904
3905 ret = ocfs2_lock_refcount_allocators(inode->i_sb,
3906 p_cluster, num_clusters,
3907 et, ref_ci,
3908 ref_root_bh, &meta_ac,
3909 NULL, &credits);
3910 if (ret) {
3911 mlog_errno(ret);
3912 goto out;
3913 }
3914
3915 handle = ocfs2_start_trans(osb, credits);
3916 if (IS_ERR(handle)) {
3917 ret = PTR_ERR(handle);
3918 mlog_errno(ret);
3919 goto out;
3920 }
3921
3922 ret = ocfs2_insert_extent(handle, et, cpos,
3923 ocfs2_clusters_to_blocks(inode->i_sb, p_cluster),
3924 num_clusters, ext_flags, meta_ac);
3925 if (ret) {
3926 mlog_errno(ret);
3927 goto out_commit;
3928 }
3929
3930 ret = ocfs2_increase_refcount(handle, ref_ci, ref_root_bh,
3931 p_cluster, num_clusters,
3932 meta_ac, dealloc);
3933 if (ret)
3934 mlog_errno(ret);
3935
3936 out_commit:
3937 ocfs2_commit_trans(osb, handle);
3938 out:
3939 if (meta_ac)
3940 ocfs2_free_alloc_context(meta_ac);
3941 return ret;
3942 }
3943
3944 static int ocfs2_duplicate_inline_data(struct inode *s_inode,
3945 struct buffer_head *s_bh,
3946 struct inode *t_inode,
3947 struct buffer_head *t_bh)
3948 {
3949 int ret;
3950 handle_t *handle;
3951 struct ocfs2_super *osb = OCFS2_SB(s_inode->i_sb);
3952 struct ocfs2_dinode *s_di = (struct ocfs2_dinode *)s_bh->b_data;
3953 struct ocfs2_dinode *t_di = (struct ocfs2_dinode *)t_bh->b_data;
3954
3955 BUG_ON(!(OCFS2_I(s_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL));
3956
3957 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
3958 if (IS_ERR(handle)) {
3959 ret = PTR_ERR(handle);
3960 mlog_errno(ret);
3961 goto out;
3962 }
3963
3964 ret = ocfs2_journal_access_di(handle, INODE_CACHE(t_inode), t_bh,
3965 OCFS2_JOURNAL_ACCESS_WRITE);
3966 if (ret) {
3967 mlog_errno(ret);
3968 goto out_commit;
3969 }
3970
3971 t_di->id2.i_data.id_count = s_di->id2.i_data.id_count;
3972 memcpy(t_di->id2.i_data.id_data, s_di->id2.i_data.id_data,
3973 le16_to_cpu(s_di->id2.i_data.id_count));
3974 spin_lock(&OCFS2_I(t_inode)->ip_lock);
3975 OCFS2_I(t_inode)->ip_dyn_features |= OCFS2_INLINE_DATA_FL;
3976 t_di->i_dyn_features = cpu_to_le16(OCFS2_I(t_inode)->ip_dyn_features);
3977 spin_unlock(&OCFS2_I(t_inode)->ip_lock);
3978
3979 ocfs2_journal_dirty(handle, t_bh);
3980
3981 out_commit:
3982 ocfs2_commit_trans(osb, handle);
3983 out:
3984 return ret;
3985 }
3986
3987 static int ocfs2_duplicate_extent_list(struct inode *s_inode,
3988 struct inode *t_inode,
3989 struct buffer_head *t_bh,
3990 struct ocfs2_caching_info *ref_ci,
3991 struct buffer_head *ref_root_bh,
3992 struct ocfs2_cached_dealloc_ctxt *dealloc)
3993 {
3994 int ret = 0;
3995 u32 p_cluster, num_clusters, clusters, cpos;
3996 loff_t size;
3997 unsigned int ext_flags;
3998 struct ocfs2_extent_tree et;
3999
4000 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(t_inode), t_bh);
4001
4002 size = i_size_read(s_inode);
4003 clusters = ocfs2_clusters_for_bytes(s_inode->i_sb, size);
4004
4005 cpos = 0;
4006 while (cpos < clusters) {
4007 ret = ocfs2_get_clusters(s_inode, cpos, &p_cluster,
4008 &num_clusters, &ext_flags);
4009
4010 if (p_cluster) {
4011 ret = ocfs2_add_refcounted_extent(t_inode, &et,
4012 ref_ci, ref_root_bh,
4013 cpos, p_cluster,
4014 num_clusters,
4015 ext_flags,
4016 dealloc);
4017 if (ret) {
4018 mlog_errno(ret);
4019 goto out;
4020 }
4021 }
4022
4023 cpos += num_clusters;
4024 }
4025
4026 out:
4027 return ret;
4028 }
4029
4030 /*
4031 * change the new file's attributes to the src.
4032 *
4033 * reflink creates a snapshot of a file, that means the attributes
4034 * must be identical except for three exceptions - nlink, ino, and ctime.
4035 */
4036 static int ocfs2_complete_reflink(struct inode *s_inode,
4037 struct buffer_head *s_bh,
4038 struct inode *t_inode,
4039 struct buffer_head *t_bh,
4040 bool preserve)
4041 {
4042 int ret;
4043 handle_t *handle;
4044 struct ocfs2_dinode *s_di = (struct ocfs2_dinode *)s_bh->b_data;
4045 struct ocfs2_dinode *di = (struct ocfs2_dinode *)t_bh->b_data;
4046 loff_t size = i_size_read(s_inode);
4047
4048 handle = ocfs2_start_trans(OCFS2_SB(t_inode->i_sb),
4049 OCFS2_INODE_UPDATE_CREDITS);
4050 if (IS_ERR(handle)) {
4051 ret = PTR_ERR(handle);
4052 mlog_errno(ret);
4053 return ret;
4054 }
4055
4056 ret = ocfs2_journal_access_di(handle, INODE_CACHE(t_inode), t_bh,
4057 OCFS2_JOURNAL_ACCESS_WRITE);
4058 if (ret) {
4059 mlog_errno(ret);
4060 goto out_commit;
4061 }
4062
4063 spin_lock(&OCFS2_I(t_inode)->ip_lock);
4064 OCFS2_I(t_inode)->ip_clusters = OCFS2_I(s_inode)->ip_clusters;
4065 OCFS2_I(t_inode)->ip_attr = OCFS2_I(s_inode)->ip_attr;
4066 OCFS2_I(t_inode)->ip_dyn_features = OCFS2_I(s_inode)->ip_dyn_features;
4067 spin_unlock(&OCFS2_I(t_inode)->ip_lock);
4068 i_size_write(t_inode, size);
4069 t_inode->i_blocks = s_inode->i_blocks;
4070
4071 di->i_xattr_inline_size = s_di->i_xattr_inline_size;
4072 di->i_clusters = s_di->i_clusters;
4073 di->i_size = s_di->i_size;
4074 di->i_dyn_features = s_di->i_dyn_features;
4075 di->i_attr = s_di->i_attr;
4076
4077 if (preserve) {
4078 t_inode->i_uid = s_inode->i_uid;
4079 t_inode->i_gid = s_inode->i_gid;
4080 t_inode->i_mode = s_inode->i_mode;
4081 di->i_uid = s_di->i_uid;
4082 di->i_gid = s_di->i_gid;
4083 di->i_mode = s_di->i_mode;
4084
4085 /*
4086 * update time.
4087 * we want mtime to appear identical to the source and
4088 * update ctime.
4089 */
4090 t_inode->i_ctime = CURRENT_TIME;
4091
4092 di->i_ctime = cpu_to_le64(t_inode->i_ctime.tv_sec);
4093 di->i_ctime_nsec = cpu_to_le32(t_inode->i_ctime.tv_nsec);
4094
4095 t_inode->i_mtime = s_inode->i_mtime;
4096 di->i_mtime = s_di->i_mtime;
4097 di->i_mtime_nsec = s_di->i_mtime_nsec;
4098 }
4099
4100 ocfs2_journal_dirty(handle, t_bh);
4101
4102 out_commit:
4103 ocfs2_commit_trans(OCFS2_SB(t_inode->i_sb), handle);
4104 return ret;
4105 }
4106
4107 static int ocfs2_create_reflink_node(struct inode *s_inode,
4108 struct buffer_head *s_bh,
4109 struct inode *t_inode,
4110 struct buffer_head *t_bh,
4111 bool preserve)
4112 {
4113 int ret;
4114 struct buffer_head *ref_root_bh = NULL;
4115 struct ocfs2_cached_dealloc_ctxt dealloc;
4116 struct ocfs2_super *osb = OCFS2_SB(s_inode->i_sb);
4117 struct ocfs2_refcount_block *rb;
4118 struct ocfs2_dinode *di = (struct ocfs2_dinode *)s_bh->b_data;
4119 struct ocfs2_refcount_tree *ref_tree;
4120
4121 ocfs2_init_dealloc_ctxt(&dealloc);
4122
4123 ret = ocfs2_set_refcount_tree(t_inode, t_bh,
4124 le64_to_cpu(di->i_refcount_loc));
4125 if (ret) {
4126 mlog_errno(ret);
4127 goto out;
4128 }
4129
4130 if (OCFS2_I(s_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
4131 ret = ocfs2_duplicate_inline_data(s_inode, s_bh,
4132 t_inode, t_bh);
4133 if (ret)
4134 mlog_errno(ret);
4135 goto out;
4136 }
4137
4138 ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
4139 1, &ref_tree, &ref_root_bh);
4140 if (ret) {
4141 mlog_errno(ret);
4142 goto out;
4143 }
4144 rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
4145
4146 ret = ocfs2_duplicate_extent_list(s_inode, t_inode, t_bh,
4147 &ref_tree->rf_ci, ref_root_bh,
4148 &dealloc);
4149 if (ret) {
4150 mlog_errno(ret);
4151 goto out_unlock_refcount;
4152 }
4153
4154 out_unlock_refcount:
4155 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
4156 brelse(ref_root_bh);
4157 out:
4158 if (ocfs2_dealloc_has_cluster(&dealloc)) {
4159 ocfs2_schedule_truncate_log_flush(osb, 1);
4160 ocfs2_run_deallocs(osb, &dealloc);
4161 }
4162
4163 return ret;
4164 }
4165
4166 static int __ocfs2_reflink(struct dentry *old_dentry,
4167 struct buffer_head *old_bh,
4168 struct inode *new_inode,
4169 bool preserve)
4170 {
4171 int ret;
4172 struct inode *inode = old_dentry->d_inode;
4173 struct buffer_head *new_bh = NULL;
4174
4175 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
4176 ret = -EINVAL;
4177 mlog_errno(ret);
4178 goto out;
4179 }
4180
4181 ret = filemap_fdatawrite(inode->i_mapping);
4182 if (ret) {
4183 mlog_errno(ret);
4184 goto out;
4185 }
4186
4187 ret = ocfs2_attach_refcount_tree(inode, old_bh);
4188 if (ret) {
4189 mlog_errno(ret);
4190 goto out;
4191 }
4192
4193 mutex_lock(&new_inode->i_mutex);
4194 ret = ocfs2_inode_lock(new_inode, &new_bh, 1);
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 }
CJK support for tty framebuffer vt