?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Btrfs: release trans metadata bytes before flushing delayed refs
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / btrfs / extent-tree.c
blobeb4fe56b08bbc51bb2880d25863c8a9bf31e9a89
1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
26 #include "compat.h"
27 #include "hash.h"
28 #include "ctree.h"
29 #include "disk-io.h"
30 #include "print-tree.h"
31 #include "transaction.h"
32 #include "volumes.h"
33 #include "locking.h"
34 #include "free-space-cache.h"
35
36 /* control flags for do_chunk_alloc's force field
37 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
38 * if we really need one.
39 *
40 * CHUNK_ALLOC_FORCE means it must try to allocate one
41 *
42 * CHUNK_ALLOC_LIMITED means to only try and allocate one
43 * if we have very few chunks already allocated. This is
44 * used as part of the clustering code to help make sure
45 * we have a good pool of storage to cluster in, without
46 * filling the FS with empty chunks
47 *
48 */
49 enum {
50 CHUNK_ALLOC_NO_FORCE = 0,
51 CHUNK_ALLOC_FORCE = 1,
52 CHUNK_ALLOC_LIMITED = 2,
53 };
54
55 /*
56 * Control how reservations are dealt with.
57 *
58 * RESERVE_FREE - freeing a reservation.
59 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
60 * ENOSPC accounting
61 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
62 * bytes_may_use as the ENOSPC accounting is done elsewhere
63 */
64 enum {
65 RESERVE_FREE = 0,
66 RESERVE_ALLOC = 1,
67 RESERVE_ALLOC_NO_ACCOUNT = 2,
68 };
69
70 static int update_block_group(struct btrfs_trans_handle *trans,
71 struct btrfs_root *root,
72 u64 bytenr, u64 num_bytes, int alloc);
73 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
74 struct btrfs_root *root,
75 u64 bytenr, u64 num_bytes, u64 parent,
76 u64 root_objectid, u64 owner_objectid,
77 u64 owner_offset, int refs_to_drop,
78 struct btrfs_delayed_extent_op *extra_op);
79 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
80 struct extent_buffer *leaf,
81 struct btrfs_extent_item *ei);
82 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
83 struct btrfs_root *root,
84 u64 parent, u64 root_objectid,
85 u64 flags, u64 owner, u64 offset,
86 struct btrfs_key *ins, int ref_mod);
87 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
88 struct btrfs_root *root,
89 u64 parent, u64 root_objectid,
90 u64 flags, struct btrfs_disk_key *key,
91 int level, struct btrfs_key *ins);
92 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
93 struct btrfs_root *extent_root, u64 alloc_bytes,
94 u64 flags, int force);
95 static int find_next_key(struct btrfs_path *path, int level,
96 struct btrfs_key *key);
97 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
98 int dump_block_groups);
99 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
100 u64 num_bytes, int reserve);
101
102 static noinline int
103 block_group_cache_done(struct btrfs_block_group_cache *cache)
104 {
105 smp_mb();
106 return cache->cached == BTRFS_CACHE_FINISHED;
107 }
108
109 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
110 {
111 return (cache->flags & bits) == bits;
112 }
113
114 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
115 {
116 atomic_inc(&cache->count);
117 }
118
119 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
120 {
121 if (atomic_dec_and_test(&cache->count)) {
122 WARN_ON(cache->pinned > 0);
123 WARN_ON(cache->reserved > 0);
124 kfree(cache->free_space_ctl);
125 kfree(cache);
126 }
127 }
128
129 /*
130 * this adds the block group to the fs_info rb tree for the block group
131 * cache
132 */
133 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
134 struct btrfs_block_group_cache *block_group)
135 {
136 struct rb_node **p;
137 struct rb_node *parent = NULL;
138 struct btrfs_block_group_cache *cache;
139
140 spin_lock(&info->block_group_cache_lock);
141 p = &info->block_group_cache_tree.rb_node;
142
143 while (*p) {
144 parent = *p;
145 cache = rb_entry(parent, struct btrfs_block_group_cache,
146 cache_node);
147 if (block_group->key.objectid < cache->key.objectid) {
148 p = &(*p)->rb_left;
149 } else if (block_group->key.objectid > cache->key.objectid) {
150 p = &(*p)->rb_right;
151 } else {
152 spin_unlock(&info->block_group_cache_lock);
153 return -EEXIST;
154 }
155 }
156
157 rb_link_node(&block_group->cache_node, parent, p);
158 rb_insert_color(&block_group->cache_node,
159 &info->block_group_cache_tree);
160 spin_unlock(&info->block_group_cache_lock);
161
162 return 0;
163 }
164
165 /*
166 * This will return the block group at or after bytenr if contains is 0, else
167 * it will return the block group that contains the bytenr
168 */
169 static struct btrfs_block_group_cache *
170 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
171 int contains)
172 {
173 struct btrfs_block_group_cache *cache, *ret = NULL;
174 struct rb_node *n;
175 u64 end, start;
176
177 spin_lock(&info->block_group_cache_lock);
178 n = info->block_group_cache_tree.rb_node;
179
180 while (n) {
181 cache = rb_entry(n, struct btrfs_block_group_cache,
182 cache_node);
183 end = cache->key.objectid + cache->key.offset - 1;
184 start = cache->key.objectid;
185
186 if (bytenr < start) {
187 if (!contains && (!ret || start < ret->key.objectid))
188 ret = cache;
189 n = n->rb_left;
190 } else if (bytenr > start) {
191 if (contains && bytenr <= end) {
192 ret = cache;
193 break;
194 }
195 n = n->rb_right;
196 } else {
197 ret = cache;
198 break;
199 }
200 }
201 if (ret)
202 btrfs_get_block_group(ret);
203 spin_unlock(&info->block_group_cache_lock);
204
205 return ret;
206 }
207
208 static int add_excluded_extent(struct btrfs_root *root,
209 u64 start, u64 num_bytes)
210 {
211 u64 end = start + num_bytes - 1;
212 set_extent_bits(&root->fs_info->freed_extents[0],
213 start, end, EXTENT_UPTODATE, GFP_NOFS);
214 set_extent_bits(&root->fs_info->freed_extents[1],
215 start, end, EXTENT_UPTODATE, GFP_NOFS);
216 return 0;
217 }
218
219 static void free_excluded_extents(struct btrfs_root *root,
220 struct btrfs_block_group_cache *cache)
221 {
222 u64 start, end;
223
224 start = cache->key.objectid;
225 end = start + cache->key.offset - 1;
226
227 clear_extent_bits(&root->fs_info->freed_extents[0],
228 start, end, EXTENT_UPTODATE, GFP_NOFS);
229 clear_extent_bits(&root->fs_info->freed_extents[1],
230 start, end, EXTENT_UPTODATE, GFP_NOFS);
231 }
232
233 static int exclude_super_stripes(struct btrfs_root *root,
234 struct btrfs_block_group_cache *cache)
235 {
236 u64 bytenr;
237 u64 *logical;
238 int stripe_len;
239 int i, nr, ret;
240
241 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
242 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
243 cache->bytes_super += stripe_len;
244 ret = add_excluded_extent(root, cache->key.objectid,
245 stripe_len);
246 BUG_ON(ret);
247 }
248
249 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
250 bytenr = btrfs_sb_offset(i);
251 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
252 cache->key.objectid, bytenr,
253 0, &logical, &nr, &stripe_len);
254 BUG_ON(ret);
255
256 while (nr--) {
257 cache->bytes_super += stripe_len;
258 ret = add_excluded_extent(root, logical[nr],
259 stripe_len);
260 BUG_ON(ret);
261 }
262
263 kfree(logical);
264 }
265 return 0;
266 }
267
268 static struct btrfs_caching_control *
269 get_caching_control(struct btrfs_block_group_cache *cache)
270 {
271 struct btrfs_caching_control *ctl;
272
273 spin_lock(&cache->lock);
274 if (cache->cached != BTRFS_CACHE_STARTED) {
275 spin_unlock(&cache->lock);
276 return NULL;
277 }
278
279 /* We're loading it the fast way, so we don't have a caching_ctl. */
280 if (!cache->caching_ctl) {
281 spin_unlock(&cache->lock);
282 return NULL;
283 }
284
285 ctl = cache->caching_ctl;
286 atomic_inc(&ctl->count);
287 spin_unlock(&cache->lock);
288 return ctl;
289 }
290
291 static void put_caching_control(struct btrfs_caching_control *ctl)
292 {
293 if (atomic_dec_and_test(&ctl->count))
294 kfree(ctl);
295 }
296
297 /*
298 * this is only called by cache_block_group, since we could have freed extents
299 * we need to check the pinned_extents for any extents that can't be used yet
300 * since their free space will be released as soon as the transaction commits.
301 */
302 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
303 struct btrfs_fs_info *info, u64 start, u64 end)
304 {
305 u64 extent_start, extent_end, size, total_added = 0;
306 int ret;
307
308 while (start < end) {
309 ret = find_first_extent_bit(info->pinned_extents, start,
310 &extent_start, &extent_end,
311 EXTENT_DIRTY | EXTENT_UPTODATE);
312 if (ret)
313 break;
314
315 if (extent_start <= start) {
316 start = extent_end + 1;
317 } else if (extent_start > start && extent_start < end) {
318 size = extent_start - start;
319 total_added += size;
320 ret = btrfs_add_free_space(block_group, start,
321 size);
322 BUG_ON(ret);
323 start = extent_end + 1;
324 } else {
325 break;
326 }
327 }
328
329 if (start < end) {
330 size = end - start;
331 total_added += size;
332 ret = btrfs_add_free_space(block_group, start, size);
333 BUG_ON(ret);
334 }
335
336 return total_added;
337 }
338
339 static noinline void caching_thread(struct btrfs_work *work)
340 {
341 struct btrfs_block_group_cache *block_group;
342 struct btrfs_fs_info *fs_info;
343 struct btrfs_caching_control *caching_ctl;
344 struct btrfs_root *extent_root;
345 struct btrfs_path *path;
346 struct extent_buffer *leaf;
347 struct btrfs_key key;
348 u64 total_found = 0;
349 u64 last = 0;
350 u32 nritems;
351 int ret = 0;
352
353 caching_ctl = container_of(work, struct btrfs_caching_control, work);
354 block_group = caching_ctl->block_group;
355 fs_info = block_group->fs_info;
356 extent_root = fs_info->extent_root;
357
358 path = btrfs_alloc_path();
359 if (!path)
360 goto out;
361
362 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
363
364 /*
365 * We don't want to deadlock with somebody trying to allocate a new
366 * extent for the extent root while also trying to search the extent
367 * root to add free space. So we skip locking and search the commit
368 * root, since its read-only
369 */
370 path->skip_locking = 1;
371 path->search_commit_root = 1;
372 path->reada = 1;
373
374 key.objectid = last;
375 key.offset = 0;
376 key.type = BTRFS_EXTENT_ITEM_KEY;
377 again:
378 mutex_lock(&caching_ctl->mutex);
379 /* need to make sure the commit_root doesn't disappear */
380 down_read(&fs_info->extent_commit_sem);
381
382 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
383 if (ret < 0)
384 goto err;
385
386 leaf = path->nodes[0];
387 nritems = btrfs_header_nritems(leaf);
388
389 while (1) {
390 if (btrfs_fs_closing(fs_info) > 1) {
391 last = (u64)-1;
392 break;
393 }
394
395 if (path->slots[0] < nritems) {
396 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
397 } else {
398 ret = find_next_key(path, 0, &key);
399 if (ret)
400 break;
401
402 if (need_resched() ||
403 btrfs_next_leaf(extent_root, path)) {
404 caching_ctl->progress = last;
405 btrfs_release_path(path);
406 up_read(&fs_info->extent_commit_sem);
407 mutex_unlock(&caching_ctl->mutex);
408 cond_resched();
409 goto again;
410 }
411 leaf = path->nodes[0];
412 nritems = btrfs_header_nritems(leaf);
413 continue;
414 }
415
416 if (key.objectid < block_group->key.objectid) {
417 path->slots[0]++;
418 continue;
419 }
420
421 if (key.objectid >= block_group->key.objectid +
422 block_group->key.offset)
423 break;
424
425 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
426 total_found += add_new_free_space(block_group,
427 fs_info, last,
428 key.objectid);
429 last = key.objectid + key.offset;
430
431 if (total_found > (1024 * 1024 * 2)) {
432 total_found = 0;
433 wake_up(&caching_ctl->wait);
434 }
435 }
436 path->slots[0]++;
437 }
438 ret = 0;
439
440 total_found += add_new_free_space(block_group, fs_info, last,
441 block_group->key.objectid +
442 block_group->key.offset);
443 caching_ctl->progress = (u64)-1;
444
445 spin_lock(&block_group->lock);
446 block_group->caching_ctl = NULL;
447 block_group->cached = BTRFS_CACHE_FINISHED;
448 spin_unlock(&block_group->lock);
449
450 err:
451 btrfs_free_path(path);
452 up_read(&fs_info->extent_commit_sem);
453
454 free_excluded_extents(extent_root, block_group);
455
456 mutex_unlock(&caching_ctl->mutex);
457 out:
458 wake_up(&caching_ctl->wait);
459
460 put_caching_control(caching_ctl);
461 btrfs_put_block_group(block_group);
462 }
463
464 static int cache_block_group(struct btrfs_block_group_cache *cache,
465 struct btrfs_trans_handle *trans,
466 struct btrfs_root *root,
467 int load_cache_only)
468 {
469 struct btrfs_fs_info *fs_info = cache->fs_info;
470 struct btrfs_caching_control *caching_ctl;
471 int ret = 0;
472
473 smp_mb();
474 if (cache->cached != BTRFS_CACHE_NO)
475 return 0;
476
477 /*
478 * We can't do the read from on-disk cache during a commit since we need
479 * to have the normal tree locking. Also if we are currently trying to
480 * allocate blocks for the tree root we can't do the fast caching since
481 * we likely hold important locks.
482 */
483 if (trans && (!trans->transaction->in_commit) &&
484 (root && root != root->fs_info->tree_root) &&
485 btrfs_test_opt(root, SPACE_CACHE)) {
486 spin_lock(&cache->lock);
487 if (cache->cached != BTRFS_CACHE_NO) {
488 spin_unlock(&cache->lock);
489 return 0;
490 }
491 cache->cached = BTRFS_CACHE_STARTED;
492 spin_unlock(&cache->lock);
493
494 ret = load_free_space_cache(fs_info, cache);
495
496 spin_lock(&cache->lock);
497 if (ret == 1) {
498 cache->cached = BTRFS_CACHE_FINISHED;
499 cache->last_byte_to_unpin = (u64)-1;
500 } else {
501 cache->cached = BTRFS_CACHE_NO;
502 }
503 spin_unlock(&cache->lock);
504 if (ret == 1) {
505 free_excluded_extents(fs_info->extent_root, cache);
506 return 0;
507 }
508 }
509
510 if (load_cache_only)
511 return 0;
512
513 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
514 BUG_ON(!caching_ctl);
515
516 INIT_LIST_HEAD(&caching_ctl->list);
517 mutex_init(&caching_ctl->mutex);
518 init_waitqueue_head(&caching_ctl->wait);
519 caching_ctl->block_group = cache;
520 caching_ctl->progress = cache->key.objectid;
521 /* one for caching kthread, one for caching block group list */
522 atomic_set(&caching_ctl->count, 2);
523 caching_ctl->work.func = caching_thread;
524
525 spin_lock(&cache->lock);
526 if (cache->cached != BTRFS_CACHE_NO) {
527 spin_unlock(&cache->lock);
528 kfree(caching_ctl);
529 return 0;
530 }
531 cache->caching_ctl = caching_ctl;
532 cache->cached = BTRFS_CACHE_STARTED;
533 spin_unlock(&cache->lock);
534
535 down_write(&fs_info->extent_commit_sem);
536 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
537 up_write(&fs_info->extent_commit_sem);
538
539 btrfs_get_block_group(cache);
540
541 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
542
543 return ret;
544 }
545
546 /*
547 * return the block group that starts at or after bytenr
548 */
549 static struct btrfs_block_group_cache *
550 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
551 {
552 struct btrfs_block_group_cache *cache;
553
554 cache = block_group_cache_tree_search(info, bytenr, 0);
555
556 return cache;
557 }
558
559 /*
560 * return the block group that contains the given bytenr
561 */
562 struct btrfs_block_group_cache *btrfs_lookup_block_group(
563 struct btrfs_fs_info *info,
564 u64 bytenr)
565 {
566 struct btrfs_block_group_cache *cache;
567
568 cache = block_group_cache_tree_search(info, bytenr, 1);
569
570 return cache;
571 }
572
573 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
574 u64 flags)
575 {
576 struct list_head *head = &info->space_info;
577 struct btrfs_space_info *found;
578
579 flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
580 BTRFS_BLOCK_GROUP_METADATA;
581
582 rcu_read_lock();
583 list_for_each_entry_rcu(found, head, list) {
584 if (found->flags & flags) {
585 rcu_read_unlock();
586 return found;
587 }
588 }
589 rcu_read_unlock();
590 return NULL;
591 }
592
593 /*
594 * after adding space to the filesystem, we need to clear the full flags
595 * on all the space infos.
596 */
597 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
598 {
599 struct list_head *head = &info->space_info;
600 struct btrfs_space_info *found;
601
602 rcu_read_lock();
603 list_for_each_entry_rcu(found, head, list)
604 found->full = 0;
605 rcu_read_unlock();
606 }
607
608 static u64 div_factor(u64 num, int factor)
609 {
610 if (factor == 10)
611 return num;
612 num *= factor;
613 do_div(num, 10);
614 return num;
615 }
616
617 static u64 div_factor_fine(u64 num, int factor)
618 {
619 if (factor == 100)
620 return num;
621 num *= factor;
622 do_div(num, 100);
623 return num;
624 }
625
626 u64 btrfs_find_block_group(struct btrfs_root *root,
627 u64 search_start, u64 search_hint, int owner)
628 {
629 struct btrfs_block_group_cache *cache;
630 u64 used;
631 u64 last = max(search_hint, search_start);
632 u64 group_start = 0;
633 int full_search = 0;
634 int factor = 9;
635 int wrapped = 0;
636 again:
637 while (1) {
638 cache = btrfs_lookup_first_block_group(root->fs_info, last);
639 if (!cache)
640 break;
641
642 spin_lock(&cache->lock);
643 last = cache->key.objectid + cache->key.offset;
644 used = btrfs_block_group_used(&cache->item);
645
646 if ((full_search || !cache->ro) &&
647 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
648 if (used + cache->pinned + cache->reserved <
649 div_factor(cache->key.offset, factor)) {
650 group_start = cache->key.objectid;
651 spin_unlock(&cache->lock);
652 btrfs_put_block_group(cache);
653 goto found;
654 }
655 }
656 spin_unlock(&cache->lock);
657 btrfs_put_block_group(cache);
658 cond_resched();
659 }
660 if (!wrapped) {
661 last = search_start;
662 wrapped = 1;
663 goto again;
664 }
665 if (!full_search && factor < 10) {
666 last = search_start;
667 full_search = 1;
668 factor = 10;
669 goto again;
670 }
671 found:
672 return group_start;
673 }
674
675 /* simple helper to search for an existing extent at a given offset */
676 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
677 {
678 int ret;
679 struct btrfs_key key;
680 struct btrfs_path *path;
681
682 path = btrfs_alloc_path();
683 if (!path)
684 return -ENOMEM;
685
686 key.objectid = start;
687 key.offset = len;
688 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
689 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
690 0, 0);
691 btrfs_free_path(path);
692 return ret;
693 }
694
695 /*
696 * helper function to lookup reference count and flags of extent.
697 *
698 * the head node for delayed ref is used to store the sum of all the
699 * reference count modifications queued up in the rbtree. the head
700 * node may also store the extent flags to set. This way you can check
701 * to see what the reference count and extent flags would be if all of
702 * the delayed refs are not processed.
703 */
704 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
705 struct btrfs_root *root, u64 bytenr,
706 u64 num_bytes, u64 *refs, u64 *flags)
707 {
708 struct btrfs_delayed_ref_head *head;
709 struct btrfs_delayed_ref_root *delayed_refs;
710 struct btrfs_path *path;
711 struct btrfs_extent_item *ei;
712 struct extent_buffer *leaf;
713 struct btrfs_key key;
714 u32 item_size;
715 u64 num_refs;
716 u64 extent_flags;
717 int ret;
718
719 path = btrfs_alloc_path();
720 if (!path)
721 return -ENOMEM;
722
723 key.objectid = bytenr;
724 key.type = BTRFS_EXTENT_ITEM_KEY;
725 key.offset = num_bytes;
726 if (!trans) {
727 path->skip_locking = 1;
728 path->search_commit_root = 1;
729 }
730 again:
731 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
732 &key, path, 0, 0);
733 if (ret < 0)
734 goto out_free;
735
736 if (ret == 0) {
737 leaf = path->nodes[0];
738 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
739 if (item_size >= sizeof(*ei)) {
740 ei = btrfs_item_ptr(leaf, path->slots[0],
741 struct btrfs_extent_item);
742 num_refs = btrfs_extent_refs(leaf, ei);
743 extent_flags = btrfs_extent_flags(leaf, ei);
744 } else {
745 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
746 struct btrfs_extent_item_v0 *ei0;
747 BUG_ON(item_size != sizeof(*ei0));
748 ei0 = btrfs_item_ptr(leaf, path->slots[0],
749 struct btrfs_extent_item_v0);
750 num_refs = btrfs_extent_refs_v0(leaf, ei0);
751 /* FIXME: this isn't correct for data */
752 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
753 #else
754 BUG();
755 #endif
756 }
757 BUG_ON(num_refs == 0);
758 } else {
759 num_refs = 0;
760 extent_flags = 0;
761 ret = 0;
762 }
763
764 if (!trans)
765 goto out;
766
767 delayed_refs = &trans->transaction->delayed_refs;
768 spin_lock(&delayed_refs->lock);
769 head = btrfs_find_delayed_ref_head(trans, bytenr);
770 if (head) {
771 if (!mutex_trylock(&head->mutex)) {
772 atomic_inc(&head->node.refs);
773 spin_unlock(&delayed_refs->lock);
774
775 btrfs_release_path(path);
776
777 /*
778 * Mutex was contended, block until it's released and try
779 * again
780 */
781 mutex_lock(&head->mutex);
782 mutex_unlock(&head->mutex);
783 btrfs_put_delayed_ref(&head->node);
784 goto again;
785 }
786 if (head->extent_op && head->extent_op->update_flags)
787 extent_flags |= head->extent_op->flags_to_set;
788 else
789 BUG_ON(num_refs == 0);
790
791 num_refs += head->node.ref_mod;
792 mutex_unlock(&head->mutex);
793 }
794 spin_unlock(&delayed_refs->lock);
795 out:
796 WARN_ON(num_refs == 0);
797 if (refs)
798 *refs = num_refs;
799 if (flags)
800 *flags = extent_flags;
801 out_free:
802 btrfs_free_path(path);
803 return ret;
804 }
805
806 /*
807 * Back reference rules. Back refs have three main goals:
808 *
809 * 1) differentiate between all holders of references to an extent so that
810 * when a reference is dropped we can make sure it was a valid reference
811 * before freeing the extent.
812 *
813 * 2) Provide enough information to quickly find the holders of an extent
814 * if we notice a given block is corrupted or bad.
815 *
816 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
817 * maintenance. This is actually the same as #2, but with a slightly
818 * different use case.
819 *
820 * There are two kinds of back refs. The implicit back refs is optimized
821 * for pointers in non-shared tree blocks. For a given pointer in a block,
822 * back refs of this kind provide information about the block's owner tree
823 * and the pointer's key. These information allow us to find the block by
824 * b-tree searching. The full back refs is for pointers in tree blocks not
825 * referenced by their owner trees. The location of tree block is recorded
826 * in the back refs. Actually the full back refs is generic, and can be
827 * used in all cases the implicit back refs is used. The major shortcoming
828 * of the full back refs is its overhead. Every time a tree block gets
829 * COWed, we have to update back refs entry for all pointers in it.
830 *
831 * For a newly allocated tree block, we use implicit back refs for
832 * pointers in it. This means most tree related operations only involve
833 * implicit back refs. For a tree block created in old transaction, the
834 * only way to drop a reference to it is COW it. So we can detect the
835 * event that tree block loses its owner tree's reference and do the
836 * back refs conversion.
837 *
838 * When a tree block is COW'd through a tree, there are four cases:
839 *
840 * The reference count of the block is one and the tree is the block's
841 * owner tree. Nothing to do in this case.
842 *
843 * The reference count of the block is one and the tree is not the
844 * block's owner tree. In this case, full back refs is used for pointers
845 * in the block. Remove these full back refs, add implicit back refs for
846 * every pointers in the new block.
847 *
848 * The reference count of the block is greater than one and the tree is
849 * the block's owner tree. In this case, implicit back refs is used for
850 * pointers in the block. Add full back refs for every pointers in the
851 * block, increase lower level extents' reference counts. The original
852 * implicit back refs are entailed to the new block.
853 *
854 * The reference count of the block is greater than one and the tree is
855 * not the block's owner tree. Add implicit back refs for every pointer in
856 * the new block, increase lower level extents' reference count.
857 *
858 * Back Reference Key composing:
859 *
860 * The key objectid corresponds to the first byte in the extent,
861 * The key type is used to differentiate between types of back refs.
862 * There are different meanings of the key offset for different types
863 * of back refs.
864 *
865 * File extents can be referenced by:
866 *
867 * - multiple snapshots, subvolumes, or different generations in one subvol
868 * - different files inside a single subvolume
869 * - different offsets inside a file (bookend extents in file.c)
870 *
871 * The extent ref structure for the implicit back refs has fields for:
872 *
873 * - Objectid of the subvolume root
874 * - objectid of the file holding the reference
875 * - original offset in the file
876 * - how many bookend extents
877 *
878 * The key offset for the implicit back refs is hash of the first
879 * three fields.
880 *
881 * The extent ref structure for the full back refs has field for:
882 *
883 * - number of pointers in the tree leaf
884 *
885 * The key offset for the implicit back refs is the first byte of
886 * the tree leaf
887 *
888 * When a file extent is allocated, The implicit back refs is used.
889 * the fields are filled in:
890 *
891 * (root_key.objectid, inode objectid, offset in file, 1)
892 *
893 * When a file extent is removed file truncation, we find the
894 * corresponding implicit back refs and check the following fields:
895 *
896 * (btrfs_header_owner(leaf), inode objectid, offset in file)
897 *
898 * Btree extents can be referenced by:
899 *
900 * - Different subvolumes
901 *
902 * Both the implicit back refs and the full back refs for tree blocks
903 * only consist of key. The key offset for the implicit back refs is
904 * objectid of block's owner tree. The key offset for the full back refs
905 * is the first byte of parent block.
906 *
907 * When implicit back refs is used, information about the lowest key and
908 * level of the tree block are required. These information are stored in
909 * tree block info structure.
910 */
911
912 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
913 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
914 struct btrfs_root *root,
915 struct btrfs_path *path,
916 u64 owner, u32 extra_size)
917 {
918 struct btrfs_extent_item *item;
919 struct btrfs_extent_item_v0 *ei0;
920 struct btrfs_extent_ref_v0 *ref0;
921 struct btrfs_tree_block_info *bi;
922 struct extent_buffer *leaf;
923 struct btrfs_key key;
924 struct btrfs_key found_key;
925 u32 new_size = sizeof(*item);
926 u64 refs;
927 int ret;
928
929 leaf = path->nodes[0];
930 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
931
932 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
933 ei0 = btrfs_item_ptr(leaf, path->slots[0],
934 struct btrfs_extent_item_v0);
935 refs = btrfs_extent_refs_v0(leaf, ei0);
936
937 if (owner == (u64)-1) {
938 while (1) {
939 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
940 ret = btrfs_next_leaf(root, path);
941 if (ret < 0)
942 return ret;
943 BUG_ON(ret > 0);
944 leaf = path->nodes[0];
945 }
946 btrfs_item_key_to_cpu(leaf, &found_key,
947 path->slots[0]);
948 BUG_ON(key.objectid != found_key.objectid);
949 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
950 path->slots[0]++;
951 continue;
952 }
953 ref0 = btrfs_item_ptr(leaf, path->slots[0],
954 struct btrfs_extent_ref_v0);
955 owner = btrfs_ref_objectid_v0(leaf, ref0);
956 break;
957 }
958 }
959 btrfs_release_path(path);
960
961 if (owner < BTRFS_FIRST_FREE_OBJECTID)
962 new_size += sizeof(*bi);
963
964 new_size -= sizeof(*ei0);
965 ret = btrfs_search_slot(trans, root, &key, path,
966 new_size + extra_size, 1);
967 if (ret < 0)
968 return ret;
969 BUG_ON(ret);
970
971 ret = btrfs_extend_item(trans, root, path, new_size);
972
973 leaf = path->nodes[0];
974 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
975 btrfs_set_extent_refs(leaf, item, refs);
976 /* FIXME: get real generation */
977 btrfs_set_extent_generation(leaf, item, 0);
978 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
979 btrfs_set_extent_flags(leaf, item,
980 BTRFS_EXTENT_FLAG_TREE_BLOCK |
981 BTRFS_BLOCK_FLAG_FULL_BACKREF);
982 bi = (struct btrfs_tree_block_info *)(item + 1);
983 /* FIXME: get first key of the block */
984 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
985 btrfs_set_tree_block_level(leaf, bi, (int)owner);
986 } else {
987 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
988 }
989 btrfs_mark_buffer_dirty(leaf);
990 return 0;
991 }
992 #endif
993
994 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
995 {
996 u32 high_crc = ~(u32)0;
997 u32 low_crc = ~(u32)0;
998 __le64 lenum;
999
1000 lenum = cpu_to_le64(root_objectid);
1001 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1002 lenum = cpu_to_le64(owner);
1003 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1004 lenum = cpu_to_le64(offset);
1005 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1006
1007 return ((u64)high_crc << 31) ^ (u64)low_crc;
1008 }
1009
1010 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1011 struct btrfs_extent_data_ref *ref)
1012 {
1013 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1014 btrfs_extent_data_ref_objectid(leaf, ref),
1015 btrfs_extent_data_ref_offset(leaf, ref));
1016 }
1017
1018 static int match_extent_data_ref(struct extent_buffer *leaf,
1019 struct btrfs_extent_data_ref *ref,
1020 u64 root_objectid, u64 owner, u64 offset)
1021 {
1022 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1023 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1024 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1025 return 0;
1026 return 1;
1027 }
1028
1029 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1030 struct btrfs_root *root,
1031 struct btrfs_path *path,
1032 u64 bytenr, u64 parent,
1033 u64 root_objectid,
1034 u64 owner, u64 offset)
1035 {
1036 struct btrfs_key key;
1037 struct btrfs_extent_data_ref *ref;
1038 struct extent_buffer *leaf;
1039 u32 nritems;
1040 int ret;
1041 int recow;
1042 int err = -ENOENT;
1043
1044 key.objectid = bytenr;
1045 if (parent) {
1046 key.type = BTRFS_SHARED_DATA_REF_KEY;
1047 key.offset = parent;
1048 } else {
1049 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1050 key.offset = hash_extent_data_ref(root_objectid,
1051 owner, offset);
1052 }
1053 again:
1054 recow = 0;
1055 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1056 if (ret < 0) {
1057 err = ret;
1058 goto fail;
1059 }
1060
1061 if (parent) {
1062 if (!ret)
1063 return 0;
1064 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1065 key.type = BTRFS_EXTENT_REF_V0_KEY;
1066 btrfs_release_path(path);
1067 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1068 if (ret < 0) {
1069 err = ret;
1070 goto fail;
1071 }
1072 if (!ret)
1073 return 0;
1074 #endif
1075 goto fail;
1076 }
1077
1078 leaf = path->nodes[0];
1079 nritems = btrfs_header_nritems(leaf);
1080 while (1) {
1081 if (path->slots[0] >= nritems) {
1082 ret = btrfs_next_leaf(root, path);
1083 if (ret < 0)
1084 err = ret;
1085 if (ret)
1086 goto fail;
1087
1088 leaf = path->nodes[0];
1089 nritems = btrfs_header_nritems(leaf);
1090 recow = 1;
1091 }
1092
1093 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1094 if (key.objectid != bytenr ||
1095 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1096 goto fail;
1097
1098 ref = btrfs_item_ptr(leaf, path->slots[0],
1099 struct btrfs_extent_data_ref);
1100
1101 if (match_extent_data_ref(leaf, ref, root_objectid,
1102 owner, offset)) {
1103 if (recow) {
1104 btrfs_release_path(path);
1105 goto again;
1106 }
1107 err = 0;
1108 break;
1109 }
1110 path->slots[0]++;
1111 }
1112 fail:
1113 return err;
1114 }
1115
1116 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1117 struct btrfs_root *root,
1118 struct btrfs_path *path,
1119 u64 bytenr, u64 parent,
1120 u64 root_objectid, u64 owner,
1121 u64 offset, int refs_to_add)
1122 {
1123 struct btrfs_key key;
1124 struct extent_buffer *leaf;
1125 u32 size;
1126 u32 num_refs;
1127 int ret;
1128
1129 key.objectid = bytenr;
1130 if (parent) {
1131 key.type = BTRFS_SHARED_DATA_REF_KEY;
1132 key.offset = parent;
1133 size = sizeof(struct btrfs_shared_data_ref);
1134 } else {
1135 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1136 key.offset = hash_extent_data_ref(root_objectid,
1137 owner, offset);
1138 size = sizeof(struct btrfs_extent_data_ref);
1139 }
1140
1141 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1142 if (ret && ret != -EEXIST)
1143 goto fail;
1144
1145 leaf = path->nodes[0];
1146 if (parent) {
1147 struct btrfs_shared_data_ref *ref;
1148 ref = btrfs_item_ptr(leaf, path->slots[0],
1149 struct btrfs_shared_data_ref);
1150 if (ret == 0) {
1151 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1152 } else {
1153 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1154 num_refs += refs_to_add;
1155 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1156 }
1157 } else {
1158 struct btrfs_extent_data_ref *ref;
1159 while (ret == -EEXIST) {
1160 ref = btrfs_item_ptr(leaf, path->slots[0],
1161 struct btrfs_extent_data_ref);
1162 if (match_extent_data_ref(leaf, ref, root_objectid,
1163 owner, offset))
1164 break;
1165 btrfs_release_path(path);
1166 key.offset++;
1167 ret = btrfs_insert_empty_item(trans, root, path, &key,
1168 size);
1169 if (ret && ret != -EEXIST)
1170 goto fail;
1171
1172 leaf = path->nodes[0];
1173 }
1174 ref = btrfs_item_ptr(leaf, path->slots[0],
1175 struct btrfs_extent_data_ref);
1176 if (ret == 0) {
1177 btrfs_set_extent_data_ref_root(leaf, ref,
1178 root_objectid);
1179 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1180 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1181 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1182 } else {
1183 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1184 num_refs += refs_to_add;
1185 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1186 }
1187 }
1188 btrfs_mark_buffer_dirty(leaf);
1189 ret = 0;
1190 fail:
1191 btrfs_release_path(path);
1192 return ret;
1193 }
1194
1195 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1196 struct btrfs_root *root,
1197 struct btrfs_path *path,
1198 int refs_to_drop)
1199 {
1200 struct btrfs_key key;
1201 struct btrfs_extent_data_ref *ref1 = NULL;
1202 struct btrfs_shared_data_ref *ref2 = NULL;
1203 struct extent_buffer *leaf;
1204 u32 num_refs = 0;
1205 int ret = 0;
1206
1207 leaf = path->nodes[0];
1208 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1209
1210 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1211 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1212 struct btrfs_extent_data_ref);
1213 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1214 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1215 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1216 struct btrfs_shared_data_ref);
1217 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1218 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1219 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1220 struct btrfs_extent_ref_v0 *ref0;
1221 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1222 struct btrfs_extent_ref_v0);
1223 num_refs = btrfs_ref_count_v0(leaf, ref0);
1224 #endif
1225 } else {
1226 BUG();
1227 }
1228
1229 BUG_ON(num_refs < refs_to_drop);
1230 num_refs -= refs_to_drop;
1231
1232 if (num_refs == 0) {
1233 ret = btrfs_del_item(trans, root, path);
1234 } else {
1235 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1236 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1237 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1238 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1239 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1240 else {
1241 struct btrfs_extent_ref_v0 *ref0;
1242 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1243 struct btrfs_extent_ref_v0);
1244 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1245 }
1246 #endif
1247 btrfs_mark_buffer_dirty(leaf);
1248 }
1249 return ret;
1250 }
1251
1252 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1253 struct btrfs_path *path,
1254 struct btrfs_extent_inline_ref *iref)
1255 {
1256 struct btrfs_key key;
1257 struct extent_buffer *leaf;
1258 struct btrfs_extent_data_ref *ref1;
1259 struct btrfs_shared_data_ref *ref2;
1260 u32 num_refs = 0;
1261
1262 leaf = path->nodes[0];
1263 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1264 if (iref) {
1265 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1266 BTRFS_EXTENT_DATA_REF_KEY) {
1267 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1268 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1269 } else {
1270 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1271 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1272 }
1273 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1274 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1275 struct btrfs_extent_data_ref);
1276 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1277 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1278 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1279 struct btrfs_shared_data_ref);
1280 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1281 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1282 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1283 struct btrfs_extent_ref_v0 *ref0;
1284 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1285 struct btrfs_extent_ref_v0);
1286 num_refs = btrfs_ref_count_v0(leaf, ref0);
1287 #endif
1288 } else {
1289 WARN_ON(1);
1290 }
1291 return num_refs;
1292 }
1293
1294 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1295 struct btrfs_root *root,
1296 struct btrfs_path *path,
1297 u64 bytenr, u64 parent,
1298 u64 root_objectid)
1299 {
1300 struct btrfs_key key;
1301 int ret;
1302
1303 key.objectid = bytenr;
1304 if (parent) {
1305 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1306 key.offset = parent;
1307 } else {
1308 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1309 key.offset = root_objectid;
1310 }
1311
1312 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1313 if (ret > 0)
1314 ret = -ENOENT;
1315 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1316 if (ret == -ENOENT && parent) {
1317 btrfs_release_path(path);
1318 key.type = BTRFS_EXTENT_REF_V0_KEY;
1319 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1320 if (ret > 0)
1321 ret = -ENOENT;
1322 }
1323 #endif
1324 return ret;
1325 }
1326
1327 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1328 struct btrfs_root *root,
1329 struct btrfs_path *path,
1330 u64 bytenr, u64 parent,
1331 u64 root_objectid)
1332 {
1333 struct btrfs_key key;
1334 int ret;
1335
1336 key.objectid = bytenr;
1337 if (parent) {
1338 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1339 key.offset = parent;
1340 } else {
1341 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1342 key.offset = root_objectid;
1343 }
1344
1345 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1346 btrfs_release_path(path);
1347 return ret;
1348 }
1349
1350 static inline int extent_ref_type(u64 parent, u64 owner)
1351 {
1352 int type;
1353 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1354 if (parent > 0)
1355 type = BTRFS_SHARED_BLOCK_REF_KEY;
1356 else
1357 type = BTRFS_TREE_BLOCK_REF_KEY;
1358 } else {
1359 if (parent > 0)
1360 type = BTRFS_SHARED_DATA_REF_KEY;
1361 else
1362 type = BTRFS_EXTENT_DATA_REF_KEY;
1363 }
1364 return type;
1365 }
1366
1367 static int find_next_key(struct btrfs_path *path, int level,
1368 struct btrfs_key *key)
1369
1370 {
1371 for (; level < BTRFS_MAX_LEVEL; level++) {
1372 if (!path->nodes[level])
1373 break;
1374 if (path->slots[level] + 1 >=
1375 btrfs_header_nritems(path->nodes[level]))
1376 continue;
1377 if (level == 0)
1378 btrfs_item_key_to_cpu(path->nodes[level], key,
1379 path->slots[level] + 1);
1380 else
1381 btrfs_node_key_to_cpu(path->nodes[level], key,
1382 path->slots[level] + 1);
1383 return 0;
1384 }
1385 return 1;
1386 }
1387
1388 /*
1389 * look for inline back ref. if back ref is found, *ref_ret is set
1390 * to the address of inline back ref, and 0 is returned.
1391 *
1392 * if back ref isn't found, *ref_ret is set to the address where it
1393 * should be inserted, and -ENOENT is returned.
1394 *
1395 * if insert is true and there are too many inline back refs, the path
1396 * points to the extent item, and -EAGAIN is returned.
1397 *
1398 * NOTE: inline back refs are ordered in the same way that back ref
1399 * items in the tree are ordered.
1400 */
1401 static noinline_for_stack
1402 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1403 struct btrfs_root *root,
1404 struct btrfs_path *path,
1405 struct btrfs_extent_inline_ref **ref_ret,
1406 u64 bytenr, u64 num_bytes,
1407 u64 parent, u64 root_objectid,
1408 u64 owner, u64 offset, int insert)
1409 {
1410 struct btrfs_key key;
1411 struct extent_buffer *leaf;
1412 struct btrfs_extent_item *ei;
1413 struct btrfs_extent_inline_ref *iref;
1414 u64 flags;
1415 u64 item_size;
1416 unsigned long ptr;
1417 unsigned long end;
1418 int extra_size;
1419 int type;
1420 int want;
1421 int ret;
1422 int err = 0;
1423
1424 key.objectid = bytenr;
1425 key.type = BTRFS_EXTENT_ITEM_KEY;
1426 key.offset = num_bytes;
1427
1428 want = extent_ref_type(parent, owner);
1429 if (insert) {
1430 extra_size = btrfs_extent_inline_ref_size(want);
1431 path->keep_locks = 1;
1432 } else
1433 extra_size = -1;
1434 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1435 if (ret < 0) {
1436 err = ret;
1437 goto out;
1438 }
1439 BUG_ON(ret);
1440
1441 leaf = path->nodes[0];
1442 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1443 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1444 if (item_size < sizeof(*ei)) {
1445 if (!insert) {
1446 err = -ENOENT;
1447 goto out;
1448 }
1449 ret = convert_extent_item_v0(trans, root, path, owner,
1450 extra_size);
1451 if (ret < 0) {
1452 err = ret;
1453 goto out;
1454 }
1455 leaf = path->nodes[0];
1456 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1457 }
1458 #endif
1459 BUG_ON(item_size < sizeof(*ei));
1460
1461 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1462 flags = btrfs_extent_flags(leaf, ei);
1463
1464 ptr = (unsigned long)(ei + 1);
1465 end = (unsigned long)ei + item_size;
1466
1467 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1468 ptr += sizeof(struct btrfs_tree_block_info);
1469 BUG_ON(ptr > end);
1470 } else {
1471 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1472 }
1473
1474 err = -ENOENT;
1475 while (1) {
1476 if (ptr >= end) {
1477 WARN_ON(ptr > end);
1478 break;
1479 }
1480 iref = (struct btrfs_extent_inline_ref *)ptr;
1481 type = btrfs_extent_inline_ref_type(leaf, iref);
1482 if (want < type)
1483 break;
1484 if (want > type) {
1485 ptr += btrfs_extent_inline_ref_size(type);
1486 continue;
1487 }
1488
1489 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1490 struct btrfs_extent_data_ref *dref;
1491 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1492 if (match_extent_data_ref(leaf, dref, root_objectid,
1493 owner, offset)) {
1494 err = 0;
1495 break;
1496 }
1497 if (hash_extent_data_ref_item(leaf, dref) <
1498 hash_extent_data_ref(root_objectid, owner, offset))
1499 break;
1500 } else {
1501 u64 ref_offset;
1502 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1503 if (parent > 0) {
1504 if (parent == ref_offset) {
1505 err = 0;
1506 break;
1507 }
1508 if (ref_offset < parent)
1509 break;
1510 } else {
1511 if (root_objectid == ref_offset) {
1512 err = 0;
1513 break;
1514 }
1515 if (ref_offset < root_objectid)
1516 break;
1517 }
1518 }
1519 ptr += btrfs_extent_inline_ref_size(type);
1520 }
1521 if (err == -ENOENT && insert) {
1522 if (item_size + extra_size >=
1523 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1524 err = -EAGAIN;
1525 goto out;
1526 }
1527 /*
1528 * To add new inline back ref, we have to make sure
1529 * there is no corresponding back ref item.
1530 * For simplicity, we just do not add new inline back
1531 * ref if there is any kind of item for this block
1532 */
1533 if (find_next_key(path, 0, &key) == 0 &&
1534 key.objectid == bytenr &&
1535 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1536 err = -EAGAIN;
1537 goto out;
1538 }
1539 }
1540 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1541 out:
1542 if (insert) {
1543 path->keep_locks = 0;
1544 btrfs_unlock_up_safe(path, 1);
1545 }
1546 return err;
1547 }
1548
1549 /*
1550 * helper to add new inline back ref
1551 */
1552 static noinline_for_stack
1553 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1554 struct btrfs_root *root,
1555 struct btrfs_path *path,
1556 struct btrfs_extent_inline_ref *iref,
1557 u64 parent, u64 root_objectid,
1558 u64 owner, u64 offset, int refs_to_add,
1559 struct btrfs_delayed_extent_op *extent_op)
1560 {
1561 struct extent_buffer *leaf;
1562 struct btrfs_extent_item *ei;
1563 unsigned long ptr;
1564 unsigned long end;
1565 unsigned long item_offset;
1566 u64 refs;
1567 int size;
1568 int type;
1569 int ret;
1570
1571 leaf = path->nodes[0];
1572 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1573 item_offset = (unsigned long)iref - (unsigned long)ei;
1574
1575 type = extent_ref_type(parent, owner);
1576 size = btrfs_extent_inline_ref_size(type);
1577
1578 ret = btrfs_extend_item(trans, root, path, size);
1579
1580 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1581 refs = btrfs_extent_refs(leaf, ei);
1582 refs += refs_to_add;
1583 btrfs_set_extent_refs(leaf, ei, refs);
1584 if (extent_op)
1585 __run_delayed_extent_op(extent_op, leaf, ei);
1586
1587 ptr = (unsigned long)ei + item_offset;
1588 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1589 if (ptr < end - size)
1590 memmove_extent_buffer(leaf, ptr + size, ptr,
1591 end - size - ptr);
1592
1593 iref = (struct btrfs_extent_inline_ref *)ptr;
1594 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1595 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1596 struct btrfs_extent_data_ref *dref;
1597 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1598 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1599 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1600 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1601 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1602 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1603 struct btrfs_shared_data_ref *sref;
1604 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1605 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1606 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1607 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1608 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1609 } else {
1610 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1611 }
1612 btrfs_mark_buffer_dirty(leaf);
1613 return 0;
1614 }
1615
1616 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1617 struct btrfs_root *root,
1618 struct btrfs_path *path,
1619 struct btrfs_extent_inline_ref **ref_ret,
1620 u64 bytenr, u64 num_bytes, u64 parent,
1621 u64 root_objectid, u64 owner, u64 offset)
1622 {
1623 int ret;
1624
1625 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1626 bytenr, num_bytes, parent,
1627 root_objectid, owner, offset, 0);
1628 if (ret != -ENOENT)
1629 return ret;
1630
1631 btrfs_release_path(path);
1632 *ref_ret = NULL;
1633
1634 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1635 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1636 root_objectid);
1637 } else {
1638 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1639 root_objectid, owner, offset);
1640 }
1641 return ret;
1642 }
1643
1644 /*
1645 * helper to update/remove inline back ref
1646 */
1647 static noinline_for_stack
1648 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1649 struct btrfs_root *root,
1650 struct btrfs_path *path,
1651 struct btrfs_extent_inline_ref *iref,
1652 int refs_to_mod,
1653 struct btrfs_delayed_extent_op *extent_op)
1654 {
1655 struct extent_buffer *leaf;
1656 struct btrfs_extent_item *ei;
1657 struct btrfs_extent_data_ref *dref = NULL;
1658 struct btrfs_shared_data_ref *sref = NULL;
1659 unsigned long ptr;
1660 unsigned long end;
1661 u32 item_size;
1662 int size;
1663 int type;
1664 int ret;
1665 u64 refs;
1666
1667 leaf = path->nodes[0];
1668 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1669 refs = btrfs_extent_refs(leaf, ei);
1670 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1671 refs += refs_to_mod;
1672 btrfs_set_extent_refs(leaf, ei, refs);
1673 if (extent_op)
1674 __run_delayed_extent_op(extent_op, leaf, ei);
1675
1676 type = btrfs_extent_inline_ref_type(leaf, iref);
1677
1678 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1679 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1680 refs = btrfs_extent_data_ref_count(leaf, dref);
1681 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1682 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1683 refs = btrfs_shared_data_ref_count(leaf, sref);
1684 } else {
1685 refs = 1;
1686 BUG_ON(refs_to_mod != -1);
1687 }
1688
1689 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1690 refs += refs_to_mod;
1691
1692 if (refs > 0) {
1693 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1694 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1695 else
1696 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1697 } else {
1698 size = btrfs_extent_inline_ref_size(type);
1699 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1700 ptr = (unsigned long)iref;
1701 end = (unsigned long)ei + item_size;
1702 if (ptr + size < end)
1703 memmove_extent_buffer(leaf, ptr, ptr + size,
1704 end - ptr - size);
1705 item_size -= size;
1706 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1707 }
1708 btrfs_mark_buffer_dirty(leaf);
1709 return 0;
1710 }
1711
1712 static noinline_for_stack
1713 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1714 struct btrfs_root *root,
1715 struct btrfs_path *path,
1716 u64 bytenr, u64 num_bytes, u64 parent,
1717 u64 root_objectid, u64 owner,
1718 u64 offset, int refs_to_add,
1719 struct btrfs_delayed_extent_op *extent_op)
1720 {
1721 struct btrfs_extent_inline_ref *iref;
1722 int ret;
1723
1724 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1725 bytenr, num_bytes, parent,
1726 root_objectid, owner, offset, 1);
1727 if (ret == 0) {
1728 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1729 ret = update_inline_extent_backref(trans, root, path, iref,
1730 refs_to_add, extent_op);
1731 } else if (ret == -ENOENT) {
1732 ret = setup_inline_extent_backref(trans, root, path, iref,
1733 parent, root_objectid,
1734 owner, offset, refs_to_add,
1735 extent_op);
1736 }
1737 return ret;
1738 }
1739
1740 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1741 struct btrfs_root *root,
1742 struct btrfs_path *path,
1743 u64 bytenr, u64 parent, u64 root_objectid,
1744 u64 owner, u64 offset, int refs_to_add)
1745 {
1746 int ret;
1747 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1748 BUG_ON(refs_to_add != 1);
1749 ret = insert_tree_block_ref(trans, root, path, bytenr,
1750 parent, root_objectid);
1751 } else {
1752 ret = insert_extent_data_ref(trans, root, path, bytenr,
1753 parent, root_objectid,
1754 owner, offset, refs_to_add);
1755 }
1756 return ret;
1757 }
1758
1759 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1760 struct btrfs_root *root,
1761 struct btrfs_path *path,
1762 struct btrfs_extent_inline_ref *iref,
1763 int refs_to_drop, int is_data)
1764 {
1765 int ret;
1766
1767 BUG_ON(!is_data && refs_to_drop != 1);
1768 if (iref) {
1769 ret = update_inline_extent_backref(trans, root, path, iref,
1770 -refs_to_drop, NULL);
1771 } else if (is_data) {
1772 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1773 } else {
1774 ret = btrfs_del_item(trans, root, path);
1775 }
1776 return ret;
1777 }
1778
1779 static int btrfs_issue_discard(struct block_device *bdev,
1780 u64 start, u64 len)
1781 {
1782 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1783 }
1784
1785 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1786 u64 num_bytes, u64 *actual_bytes)
1787 {
1788 int ret;
1789 u64 discarded_bytes = 0;
1790 struct btrfs_multi_bio *multi = NULL;
1791
1792
1793 /* Tell the block device(s) that the sectors can be discarded */
1794 ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD,
1795 bytenr, &num_bytes, &multi, 0);
1796 if (!ret) {
1797 struct btrfs_bio_stripe *stripe = multi->stripes;
1798 int i;
1799
1800
1801 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1802 if (!stripe->dev->can_discard)
1803 continue;
1804
1805 ret = btrfs_issue_discard(stripe->dev->bdev,
1806 stripe->physical,
1807 stripe->length);
1808 if (!ret)
1809 discarded_bytes += stripe->length;
1810 else if (ret != -EOPNOTSUPP)
1811 break;
1812
1813 /*
1814 * Just in case we get back EOPNOTSUPP for some reason,
1815 * just ignore the return value so we don't screw up
1816 * people calling discard_extent.
1817 */
1818 ret = 0;
1819 }
1820 kfree(multi);
1821 }
1822
1823 if (actual_bytes)
1824 *actual_bytes = discarded_bytes;
1825
1826
1827 return ret;
1828 }
1829
1830 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1831 struct btrfs_root *root,
1832 u64 bytenr, u64 num_bytes, u64 parent,
1833 u64 root_objectid, u64 owner, u64 offset)
1834 {
1835 int ret;
1836 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1837 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1838
1839 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1840 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1841 parent, root_objectid, (int)owner,
1842 BTRFS_ADD_DELAYED_REF, NULL);
1843 } else {
1844 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1845 parent, root_objectid, owner, offset,
1846 BTRFS_ADD_DELAYED_REF, NULL);
1847 }
1848 return ret;
1849 }
1850
1851 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1852 struct btrfs_root *root,
1853 u64 bytenr, u64 num_bytes,
1854 u64 parent, u64 root_objectid,
1855 u64 owner, u64 offset, int refs_to_add,
1856 struct btrfs_delayed_extent_op *extent_op)
1857 {
1858 struct btrfs_path *path;
1859 struct extent_buffer *leaf;
1860 struct btrfs_extent_item *item;
1861 u64 refs;
1862 int ret;
1863 int err = 0;
1864
1865 path = btrfs_alloc_path();
1866 if (!path)
1867 return -ENOMEM;
1868
1869 path->reada = 1;
1870 path->leave_spinning = 1;
1871 /* this will setup the path even if it fails to insert the back ref */
1872 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1873 path, bytenr, num_bytes, parent,
1874 root_objectid, owner, offset,
1875 refs_to_add, extent_op);
1876 if (ret == 0)
1877 goto out;
1878
1879 if (ret != -EAGAIN) {
1880 err = ret;
1881 goto out;
1882 }
1883
1884 leaf = path->nodes[0];
1885 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1886 refs = btrfs_extent_refs(leaf, item);
1887 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1888 if (extent_op)
1889 __run_delayed_extent_op(extent_op, leaf, item);
1890
1891 btrfs_mark_buffer_dirty(leaf);
1892 btrfs_release_path(path);
1893
1894 path->reada = 1;
1895 path->leave_spinning = 1;
1896
1897 /* now insert the actual backref */
1898 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1899 path, bytenr, parent, root_objectid,
1900 owner, offset, refs_to_add);
1901 BUG_ON(ret);
1902 out:
1903 btrfs_free_path(path);
1904 return err;
1905 }
1906
1907 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1908 struct btrfs_root *root,
1909 struct btrfs_delayed_ref_node *node,
1910 struct btrfs_delayed_extent_op *extent_op,
1911 int insert_reserved)
1912 {
1913 int ret = 0;
1914 struct btrfs_delayed_data_ref *ref;
1915 struct btrfs_key ins;
1916 u64 parent = 0;
1917 u64 ref_root = 0;
1918 u64 flags = 0;
1919
1920 ins.objectid = node->bytenr;
1921 ins.offset = node->num_bytes;
1922 ins.type = BTRFS_EXTENT_ITEM_KEY;
1923
1924 ref = btrfs_delayed_node_to_data_ref(node);
1925 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1926 parent = ref->parent;
1927 else
1928 ref_root = ref->root;
1929
1930 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1931 if (extent_op) {
1932 BUG_ON(extent_op->update_key);
1933 flags |= extent_op->flags_to_set;
1934 }
1935 ret = alloc_reserved_file_extent(trans, root,
1936 parent, ref_root, flags,
1937 ref->objectid, ref->offset,
1938 &ins, node->ref_mod);
1939 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1940 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1941 node->num_bytes, parent,
1942 ref_root, ref->objectid,
1943 ref->offset, node->ref_mod,
1944 extent_op);
1945 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1946 ret = __btrfs_free_extent(trans, root, node->bytenr,
1947 node->num_bytes, parent,
1948 ref_root, ref->objectid,
1949 ref->offset, node->ref_mod,
1950 extent_op);
1951 } else {
1952 BUG();
1953 }
1954 return ret;
1955 }
1956
1957 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1958 struct extent_buffer *leaf,
1959 struct btrfs_extent_item *ei)
1960 {
1961 u64 flags = btrfs_extent_flags(leaf, ei);
1962 if (extent_op->update_flags) {
1963 flags |= extent_op->flags_to_set;
1964 btrfs_set_extent_flags(leaf, ei, flags);
1965 }
1966
1967 if (extent_op->update_key) {
1968 struct btrfs_tree_block_info *bi;
1969 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1970 bi = (struct btrfs_tree_block_info *)(ei + 1);
1971 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1972 }
1973 }
1974
1975 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1976 struct btrfs_root *root,
1977 struct btrfs_delayed_ref_node *node,
1978 struct btrfs_delayed_extent_op *extent_op)
1979 {
1980 struct btrfs_key key;
1981 struct btrfs_path *path;
1982 struct btrfs_extent_item *ei;
1983 struct extent_buffer *leaf;
1984 u32 item_size;
1985 int ret;
1986 int err = 0;
1987
1988 path = btrfs_alloc_path();
1989 if (!path)
1990 return -ENOMEM;
1991
1992 key.objectid = node->bytenr;
1993 key.type = BTRFS_EXTENT_ITEM_KEY;
1994 key.offset = node->num_bytes;
1995
1996 path->reada = 1;
1997 path->leave_spinning = 1;
1998 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1999 path, 0, 1);
2000 if (ret < 0) {
2001 err = ret;
2002 goto out;
2003 }
2004 if (ret > 0) {
2005 err = -EIO;
2006 goto out;
2007 }
2008
2009 leaf = path->nodes[0];
2010 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2011 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2012 if (item_size < sizeof(*ei)) {
2013 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2014 path, (u64)-1, 0);
2015 if (ret < 0) {
2016 err = ret;
2017 goto out;
2018 }
2019 leaf = path->nodes[0];
2020 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2021 }
2022 #endif
2023 BUG_ON(item_size < sizeof(*ei));
2024 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2025 __run_delayed_extent_op(extent_op, leaf, ei);
2026
2027 btrfs_mark_buffer_dirty(leaf);
2028 out:
2029 btrfs_free_path(path);
2030 return err;
2031 }
2032
2033 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2034 struct btrfs_root *root,
2035 struct btrfs_delayed_ref_node *node,
2036 struct btrfs_delayed_extent_op *extent_op,
2037 int insert_reserved)
2038 {
2039 int ret = 0;
2040 struct btrfs_delayed_tree_ref *ref;
2041 struct btrfs_key ins;
2042 u64 parent = 0;
2043 u64 ref_root = 0;
2044
2045 ins.objectid = node->bytenr;
2046 ins.offset = node->num_bytes;
2047 ins.type = BTRFS_EXTENT_ITEM_KEY;
2048
2049 ref = btrfs_delayed_node_to_tree_ref(node);
2050 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2051 parent = ref->parent;
2052 else
2053 ref_root = ref->root;
2054
2055 BUG_ON(node->ref_mod != 1);
2056 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2057 BUG_ON(!extent_op || !extent_op->update_flags ||
2058 !extent_op->update_key);
2059 ret = alloc_reserved_tree_block(trans, root,
2060 parent, ref_root,
2061 extent_op->flags_to_set,
2062 &extent_op->key,
2063 ref->level, &ins);
2064 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2065 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2066 node->num_bytes, parent, ref_root,
2067 ref->level, 0, 1, extent_op);
2068 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2069 ret = __btrfs_free_extent(trans, root, node->bytenr,
2070 node->num_bytes, parent, ref_root,
2071 ref->level, 0, 1, extent_op);
2072 } else {
2073 BUG();
2074 }
2075 return ret;
2076 }
2077
2078 /* helper function to actually process a single delayed ref entry */
2079 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2080 struct btrfs_root *root,
2081 struct btrfs_delayed_ref_node *node,
2082 struct btrfs_delayed_extent_op *extent_op,
2083 int insert_reserved)
2084 {
2085 int ret;
2086 if (btrfs_delayed_ref_is_head(node)) {
2087 struct btrfs_delayed_ref_head *head;
2088 /*
2089 * we've hit the end of the chain and we were supposed
2090 * to insert this extent into the tree. But, it got
2091 * deleted before we ever needed to insert it, so all
2092 * we have to do is clean up the accounting
2093 */
2094 BUG_ON(extent_op);
2095 head = btrfs_delayed_node_to_head(node);
2096 if (insert_reserved) {
2097 btrfs_pin_extent(root, node->bytenr,
2098 node->num_bytes, 1);
2099 if (head->is_data) {
2100 ret = btrfs_del_csums(trans, root,
2101 node->bytenr,
2102 node->num_bytes);
2103 BUG_ON(ret);
2104 }
2105 }
2106 mutex_unlock(&head->mutex);
2107 return 0;
2108 }
2109
2110 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2111 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2112 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2113 insert_reserved);
2114 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2115 node->type == BTRFS_SHARED_DATA_REF_KEY)
2116 ret = run_delayed_data_ref(trans, root, node, extent_op,
2117 insert_reserved);
2118 else
2119 BUG();
2120 return ret;
2121 }
2122
2123 static noinline struct btrfs_delayed_ref_node *
2124 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2125 {
2126 struct rb_node *node;
2127 struct btrfs_delayed_ref_node *ref;
2128 int action = BTRFS_ADD_DELAYED_REF;
2129 again:
2130 /*
2131 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2132 * this prevents ref count from going down to zero when
2133 * there still are pending delayed ref.
2134 */
2135 node = rb_prev(&head->node.rb_node);
2136 while (1) {
2137 if (!node)
2138 break;
2139 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2140 rb_node);
2141 if (ref->bytenr != head->node.bytenr)
2142 break;
2143 if (ref->action == action)
2144 return ref;
2145 node = rb_prev(node);
2146 }
2147 if (action == BTRFS_ADD_DELAYED_REF) {
2148 action = BTRFS_DROP_DELAYED_REF;
2149 goto again;
2150 }
2151 return NULL;
2152 }
2153
2154 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2155 struct btrfs_root *root,
2156 struct list_head *cluster)
2157 {
2158 struct btrfs_delayed_ref_root *delayed_refs;
2159 struct btrfs_delayed_ref_node *ref;
2160 struct btrfs_delayed_ref_head *locked_ref = NULL;
2161 struct btrfs_delayed_extent_op *extent_op;
2162 int ret;
2163 int count = 0;
2164 int must_insert_reserved = 0;
2165
2166 delayed_refs = &trans->transaction->delayed_refs;
2167 while (1) {
2168 if (!locked_ref) {
2169 /* pick a new head ref from the cluster list */
2170 if (list_empty(cluster))
2171 break;
2172
2173 locked_ref = list_entry(cluster->next,
2174 struct btrfs_delayed_ref_head, cluster);
2175
2176 /* grab the lock that says we are going to process
2177 * all the refs for this head */
2178 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2179
2180 /*
2181 * we may have dropped the spin lock to get the head
2182 * mutex lock, and that might have given someone else
2183 * time to free the head. If that's true, it has been
2184 * removed from our list and we can move on.
2185 */
2186 if (ret == -EAGAIN) {
2187 locked_ref = NULL;
2188 count++;
2189 continue;
2190 }
2191 }
2192
2193 /*
2194 * record the must insert reserved flag before we
2195 * drop the spin lock.
2196 */
2197 must_insert_reserved = locked_ref->must_insert_reserved;
2198 locked_ref->must_insert_reserved = 0;
2199
2200 extent_op = locked_ref->extent_op;
2201 locked_ref->extent_op = NULL;
2202
2203 /*
2204 * locked_ref is the head node, so we have to go one
2205 * node back for any delayed ref updates
2206 */
2207 ref = select_delayed_ref(locked_ref);
2208 if (!ref) {
2209 /* All delayed refs have been processed, Go ahead
2210 * and send the head node to run_one_delayed_ref,
2211 * so that any accounting fixes can happen
2212 */
2213 ref = &locked_ref->node;
2214
2215 if (extent_op && must_insert_reserved) {
2216 kfree(extent_op);
2217 extent_op = NULL;
2218 }
2219
2220 if (extent_op) {
2221 spin_unlock(&delayed_refs->lock);
2222
2223 ret = run_delayed_extent_op(trans, root,
2224 ref, extent_op);
2225 BUG_ON(ret);
2226 kfree(extent_op);
2227
2228 cond_resched();
2229 spin_lock(&delayed_refs->lock);
2230 continue;
2231 }
2232
2233 list_del_init(&locked_ref->cluster);
2234 locked_ref = NULL;
2235 }
2236
2237 ref->in_tree = 0;
2238 rb_erase(&ref->rb_node, &delayed_refs->root);
2239 delayed_refs->num_entries--;
2240
2241 spin_unlock(&delayed_refs->lock);
2242
2243 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2244 must_insert_reserved);
2245 BUG_ON(ret);
2246
2247 btrfs_put_delayed_ref(ref);
2248 kfree(extent_op);
2249 count++;
2250
2251 cond_resched();
2252 spin_lock(&delayed_refs->lock);
2253 }
2254 return count;
2255 }
2256
2257 /*
2258 * this starts processing the delayed reference count updates and
2259 * extent insertions we have queued up so far. count can be
2260 * 0, which means to process everything in the tree at the start
2261 * of the run (but not newly added entries), or it can be some target
2262 * number you'd like to process.
2263 */
2264 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2265 struct btrfs_root *root, unsigned long count)
2266 {
2267 struct rb_node *node;
2268 struct btrfs_delayed_ref_root *delayed_refs;
2269 struct btrfs_delayed_ref_node *ref;
2270 struct list_head cluster;
2271 int ret;
2272 int run_all = count == (unsigned long)-1;
2273 int run_most = 0;
2274
2275 if (root == root->fs_info->extent_root)
2276 root = root->fs_info->tree_root;
2277
2278 delayed_refs = &trans->transaction->delayed_refs;
2279 INIT_LIST_HEAD(&cluster);
2280 again:
2281 spin_lock(&delayed_refs->lock);
2282 if (count == 0) {
2283 count = delayed_refs->num_entries * 2;
2284 run_most = 1;
2285 }
2286 while (1) {
2287 if (!(run_all || run_most) &&
2288 delayed_refs->num_heads_ready < 64)
2289 break;
2290
2291 /*
2292 * go find something we can process in the rbtree. We start at
2293 * the beginning of the tree, and then build a cluster
2294 * of refs to process starting at the first one we are able to
2295 * lock
2296 */
2297 ret = btrfs_find_ref_cluster(trans, &cluster,
2298 delayed_refs->run_delayed_start);
2299 if (ret)
2300 break;
2301
2302 ret = run_clustered_refs(trans, root, &cluster);
2303 BUG_ON(ret < 0);
2304
2305 count -= min_t(unsigned long, ret, count);
2306
2307 if (count == 0)
2308 break;
2309 }
2310
2311 if (run_all) {
2312 node = rb_first(&delayed_refs->root);
2313 if (!node)
2314 goto out;
2315 count = (unsigned long)-1;
2316
2317 while (node) {
2318 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2319 rb_node);
2320 if (btrfs_delayed_ref_is_head(ref)) {
2321 struct btrfs_delayed_ref_head *head;
2322
2323 head = btrfs_delayed_node_to_head(ref);
2324 atomic_inc(&ref->refs);
2325
2326 spin_unlock(&delayed_refs->lock);
2327 /*
2328 * Mutex was contended, block until it's
2329 * released and try again
2330 */
2331 mutex_lock(&head->mutex);
2332 mutex_unlock(&head->mutex);
2333
2334 btrfs_put_delayed_ref(ref);
2335 cond_resched();
2336 goto again;
2337 }
2338 node = rb_next(node);
2339 }
2340 spin_unlock(&delayed_refs->lock);
2341 schedule_timeout(1);
2342 goto again;
2343 }
2344 out:
2345 spin_unlock(&delayed_refs->lock);
2346 return 0;
2347 }
2348
2349 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2350 struct btrfs_root *root,
2351 u64 bytenr, u64 num_bytes, u64 flags,
2352 int is_data)
2353 {
2354 struct btrfs_delayed_extent_op *extent_op;
2355 int ret;
2356
2357 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2358 if (!extent_op)
2359 return -ENOMEM;
2360
2361 extent_op->flags_to_set = flags;
2362 extent_op->update_flags = 1;
2363 extent_op->update_key = 0;
2364 extent_op->is_data = is_data ? 1 : 0;
2365
2366 ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2367 if (ret)
2368 kfree(extent_op);
2369 return ret;
2370 }
2371
2372 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2373 struct btrfs_root *root,
2374 struct btrfs_path *path,
2375 u64 objectid, u64 offset, u64 bytenr)
2376 {
2377 struct btrfs_delayed_ref_head *head;
2378 struct btrfs_delayed_ref_node *ref;
2379 struct btrfs_delayed_data_ref *data_ref;
2380 struct btrfs_delayed_ref_root *delayed_refs;
2381 struct rb_node *node;
2382 int ret = 0;
2383
2384 ret = -ENOENT;
2385 delayed_refs = &trans->transaction->delayed_refs;
2386 spin_lock(&delayed_refs->lock);
2387 head = btrfs_find_delayed_ref_head(trans, bytenr);
2388 if (!head)
2389 goto out;
2390
2391 if (!mutex_trylock(&head->mutex)) {
2392 atomic_inc(&head->node.refs);
2393 spin_unlock(&delayed_refs->lock);
2394
2395 btrfs_release_path(path);
2396
2397 /*
2398 * Mutex was contended, block until it's released and let
2399 * caller try again
2400 */
2401 mutex_lock(&head->mutex);
2402 mutex_unlock(&head->mutex);
2403 btrfs_put_delayed_ref(&head->node);
2404 return -EAGAIN;
2405 }
2406
2407 node = rb_prev(&head->node.rb_node);
2408 if (!node)
2409 goto out_unlock;
2410
2411 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2412
2413 if (ref->bytenr != bytenr)
2414 goto out_unlock;
2415
2416 ret = 1;
2417 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2418 goto out_unlock;
2419
2420 data_ref = btrfs_delayed_node_to_data_ref(ref);
2421
2422 node = rb_prev(node);
2423 if (node) {
2424 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2425 if (ref->bytenr == bytenr)
2426 goto out_unlock;
2427 }
2428
2429 if (data_ref->root != root->root_key.objectid ||
2430 data_ref->objectid != objectid || data_ref->offset != offset)
2431 goto out_unlock;
2432
2433 ret = 0;
2434 out_unlock:
2435 mutex_unlock(&head->mutex);
2436 out:
2437 spin_unlock(&delayed_refs->lock);
2438 return ret;
2439 }
2440
2441 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2442 struct btrfs_root *root,
2443 struct btrfs_path *path,
2444 u64 objectid, u64 offset, u64 bytenr)
2445 {
2446 struct btrfs_root *extent_root = root->fs_info->extent_root;
2447 struct extent_buffer *leaf;
2448 struct btrfs_extent_data_ref *ref;
2449 struct btrfs_extent_inline_ref *iref;
2450 struct btrfs_extent_item *ei;
2451 struct btrfs_key key;
2452 u32 item_size;
2453 int ret;
2454
2455 key.objectid = bytenr;
2456 key.offset = (u64)-1;
2457 key.type = BTRFS_EXTENT_ITEM_KEY;
2458
2459 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2460 if (ret < 0)
2461 goto out;
2462 BUG_ON(ret == 0);
2463
2464 ret = -ENOENT;
2465 if (path->slots[0] == 0)
2466 goto out;
2467
2468 path->slots[0]--;
2469 leaf = path->nodes[0];
2470 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2471
2472 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2473 goto out;
2474
2475 ret = 1;
2476 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2477 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2478 if (item_size < sizeof(*ei)) {
2479 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2480 goto out;
2481 }
2482 #endif
2483 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2484
2485 if (item_size != sizeof(*ei) +
2486 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2487 goto out;
2488
2489 if (btrfs_extent_generation(leaf, ei) <=
2490 btrfs_root_last_snapshot(&root->root_item))
2491 goto out;
2492
2493 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2494 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2495 BTRFS_EXTENT_DATA_REF_KEY)
2496 goto out;
2497
2498 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2499 if (btrfs_extent_refs(leaf, ei) !=
2500 btrfs_extent_data_ref_count(leaf, ref) ||
2501 btrfs_extent_data_ref_root(leaf, ref) !=
2502 root->root_key.objectid ||
2503 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2504 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2505 goto out;
2506
2507 ret = 0;
2508 out:
2509 return ret;
2510 }
2511
2512 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2513 struct btrfs_root *root,
2514 u64 objectid, u64 offset, u64 bytenr)
2515 {
2516 struct btrfs_path *path;
2517 int ret;
2518 int ret2;
2519
2520 path = btrfs_alloc_path();
2521 if (!path)
2522 return -ENOENT;
2523
2524 do {
2525 ret = check_committed_ref(trans, root, path, objectid,
2526 offset, bytenr);
2527 if (ret && ret != -ENOENT)
2528 goto out;
2529
2530 ret2 = check_delayed_ref(trans, root, path, objectid,
2531 offset, bytenr);
2532 } while (ret2 == -EAGAIN);
2533
2534 if (ret2 && ret2 != -ENOENT) {
2535 ret = ret2;
2536 goto out;
2537 }
2538
2539 if (ret != -ENOENT || ret2 != -ENOENT)
2540 ret = 0;
2541 out:
2542 btrfs_free_path(path);
2543 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2544 WARN_ON(ret > 0);
2545 return ret;
2546 }
2547
2548 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2549 struct btrfs_root *root,
2550 struct extent_buffer *buf,
2551 int full_backref, int inc)
2552 {
2553 u64 bytenr;
2554 u64 num_bytes;
2555 u64 parent;
2556 u64 ref_root;
2557 u32 nritems;
2558 struct btrfs_key key;
2559 struct btrfs_file_extent_item *fi;
2560 int i;
2561 int level;
2562 int ret = 0;
2563 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2564 u64, u64, u64, u64, u64, u64);
2565
2566 ref_root = btrfs_header_owner(buf);
2567 nritems = btrfs_header_nritems(buf);
2568 level = btrfs_header_level(buf);
2569
2570 if (!root->ref_cows && level == 0)
2571 return 0;
2572
2573 if (inc)
2574 process_func = btrfs_inc_extent_ref;
2575 else
2576 process_func = btrfs_free_extent;
2577
2578 if (full_backref)
2579 parent = buf->start;
2580 else
2581 parent = 0;
2582
2583 for (i = 0; i < nritems; i++) {
2584 if (level == 0) {
2585 btrfs_item_key_to_cpu(buf, &key, i);
2586 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2587 continue;
2588 fi = btrfs_item_ptr(buf, i,
2589 struct btrfs_file_extent_item);
2590 if (btrfs_file_extent_type(buf, fi) ==
2591 BTRFS_FILE_EXTENT_INLINE)
2592 continue;
2593 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2594 if (bytenr == 0)
2595 continue;
2596
2597 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2598 key.offset -= btrfs_file_extent_offset(buf, fi);
2599 ret = process_func(trans, root, bytenr, num_bytes,
2600 parent, ref_root, key.objectid,
2601 key.offset);
2602 if (ret)
2603 goto fail;
2604 } else {
2605 bytenr = btrfs_node_blockptr(buf, i);
2606 num_bytes = btrfs_level_size(root, level - 1);
2607 ret = process_func(trans, root, bytenr, num_bytes,
2608 parent, ref_root, level - 1, 0);
2609 if (ret)
2610 goto fail;
2611 }
2612 }
2613 return 0;
2614 fail:
2615 BUG();
2616 return ret;
2617 }
2618
2619 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2620 struct extent_buffer *buf, int full_backref)
2621 {
2622 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2623 }
2624
2625 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2626 struct extent_buffer *buf, int full_backref)
2627 {
2628 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2629 }
2630
2631 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2632 struct btrfs_root *root,
2633 struct btrfs_path *path,
2634 struct btrfs_block_group_cache *cache)
2635 {
2636 int ret;
2637 struct btrfs_root *extent_root = root->fs_info->extent_root;
2638 unsigned long bi;
2639 struct extent_buffer *leaf;
2640
2641 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2642 if (ret < 0)
2643 goto fail;
2644 BUG_ON(ret);
2645
2646 leaf = path->nodes[0];
2647 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2648 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2649 btrfs_mark_buffer_dirty(leaf);
2650 btrfs_release_path(path);
2651 fail:
2652 if (ret)
2653 return ret;
2654 return 0;
2655
2656 }
2657
2658 static struct btrfs_block_group_cache *
2659 next_block_group(struct btrfs_root *root,
2660 struct btrfs_block_group_cache *cache)
2661 {
2662 struct rb_node *node;
2663 spin_lock(&root->fs_info->block_group_cache_lock);
2664 node = rb_next(&cache->cache_node);
2665 btrfs_put_block_group(cache);
2666 if (node) {
2667 cache = rb_entry(node, struct btrfs_block_group_cache,
2668 cache_node);
2669 btrfs_get_block_group(cache);
2670 } else
2671 cache = NULL;
2672 spin_unlock(&root->fs_info->block_group_cache_lock);
2673 return cache;
2674 }
2675
2676 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2677 struct btrfs_trans_handle *trans,
2678 struct btrfs_path *path)
2679 {
2680 struct btrfs_root *root = block_group->fs_info->tree_root;
2681 struct inode *inode = NULL;
2682 u64 alloc_hint = 0;
2683 int dcs = BTRFS_DC_ERROR;
2684 int num_pages = 0;
2685 int retries = 0;
2686 int ret = 0;
2687
2688 /*
2689 * If this block group is smaller than 100 megs don't bother caching the
2690 * block group.
2691 */
2692 if (block_group->key.offset < (100 * 1024 * 1024)) {
2693 spin_lock(&block_group->lock);
2694 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2695 spin_unlock(&block_group->lock);
2696 return 0;
2697 }
2698
2699 again:
2700 inode = lookup_free_space_inode(root, block_group, path);
2701 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2702 ret = PTR_ERR(inode);
2703 btrfs_release_path(path);
2704 goto out;
2705 }
2706
2707 if (IS_ERR(inode)) {
2708 BUG_ON(retries);
2709 retries++;
2710
2711 if (block_group->ro)
2712 goto out_free;
2713
2714 ret = create_free_space_inode(root, trans, block_group, path);
2715 if (ret)
2716 goto out_free;
2717 goto again;
2718 }
2719
2720 /* We've already setup this transaction, go ahead and exit */
2721 if (block_group->cache_generation == trans->transid &&
2722 i_size_read(inode)) {
2723 dcs = BTRFS_DC_SETUP;
2724 goto out_put;
2725 }
2726
2727 /*
2728 * We want to set the generation to 0, that way if anything goes wrong
2729 * from here on out we know not to trust this cache when we load up next
2730 * time.
2731 */
2732 BTRFS_I(inode)->generation = 0;
2733 ret = btrfs_update_inode(trans, root, inode);
2734 WARN_ON(ret);
2735
2736 if (i_size_read(inode) > 0) {
2737 ret = btrfs_truncate_free_space_cache(root, trans, path,
2738 inode);
2739 if (ret)
2740 goto out_put;
2741 }
2742
2743 spin_lock(&block_group->lock);
2744 if (block_group->cached != BTRFS_CACHE_FINISHED) {
2745 /* We're not cached, don't bother trying to write stuff out */
2746 dcs = BTRFS_DC_WRITTEN;
2747 spin_unlock(&block_group->lock);
2748 goto out_put;
2749 }
2750 spin_unlock(&block_group->lock);
2751
2752 num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2753 if (!num_pages)
2754 num_pages = 1;
2755
2756 /*
2757 * Just to make absolutely sure we have enough space, we're going to
2758 * preallocate 12 pages worth of space for each block group. In
2759 * practice we ought to use at most 8, but we need extra space so we can
2760 * add our header and have a terminator between the extents and the
2761 * bitmaps.
2762 */
2763 num_pages *= 16;
2764 num_pages *= PAGE_CACHE_SIZE;
2765
2766 ret = btrfs_check_data_free_space(inode, num_pages);
2767 if (ret)
2768 goto out_put;
2769
2770 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2771 num_pages, num_pages,
2772 &alloc_hint);
2773 if (!ret)
2774 dcs = BTRFS_DC_SETUP;
2775 btrfs_free_reserved_data_space(inode, num_pages);
2776
2777 out_put:
2778 iput(inode);
2779 out_free:
2780 btrfs_release_path(path);
2781 out:
2782 spin_lock(&block_group->lock);
2783 if (!ret)
2784 block_group->cache_generation = trans->transid;
2785 block_group->disk_cache_state = dcs;
2786 spin_unlock(&block_group->lock);
2787
2788 return ret;
2789 }
2790
2791 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2792 struct btrfs_root *root)
2793 {
2794 struct btrfs_block_group_cache *cache;
2795 int err = 0;
2796 struct btrfs_path *path;
2797 u64 last = 0;
2798
2799 path = btrfs_alloc_path();
2800 if (!path)
2801 return -ENOMEM;
2802
2803 again:
2804 while (1) {
2805 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2806 while (cache) {
2807 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2808 break;
2809 cache = next_block_group(root, cache);
2810 }
2811 if (!cache) {
2812 if (last == 0)
2813 break;
2814 last = 0;
2815 continue;
2816 }
2817 err = cache_save_setup(cache, trans, path);
2818 last = cache->key.objectid + cache->key.offset;
2819 btrfs_put_block_group(cache);
2820 }
2821
2822 while (1) {
2823 if (last == 0) {
2824 err = btrfs_run_delayed_refs(trans, root,
2825 (unsigned long)-1);
2826 BUG_ON(err);
2827 }
2828
2829 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2830 while (cache) {
2831 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2832 btrfs_put_block_group(cache);
2833 goto again;
2834 }
2835
2836 if (cache->dirty)
2837 break;
2838 cache = next_block_group(root, cache);
2839 }
2840 if (!cache) {
2841 if (last == 0)
2842 break;
2843 last = 0;
2844 continue;
2845 }
2846
2847 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2848 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2849 cache->dirty = 0;
2850 last = cache->key.objectid + cache->key.offset;
2851
2852 err = write_one_cache_group(trans, root, path, cache);
2853 BUG_ON(err);
2854 btrfs_put_block_group(cache);
2855 }
2856
2857 while (1) {
2858 /*
2859 * I don't think this is needed since we're just marking our
2860 * preallocated extent as written, but just in case it can't
2861 * hurt.
2862 */
2863 if (last == 0) {
2864 err = btrfs_run_delayed_refs(trans, root,
2865 (unsigned long)-1);
2866 BUG_ON(err);
2867 }
2868
2869 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2870 while (cache) {
2871 /*
2872 * Really this shouldn't happen, but it could if we
2873 * couldn't write the entire preallocated extent and
2874 * splitting the extent resulted in a new block.
2875 */
2876 if (cache->dirty) {
2877 btrfs_put_block_group(cache);
2878 goto again;
2879 }
2880 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2881 break;
2882 cache = next_block_group(root, cache);
2883 }
2884 if (!cache) {
2885 if (last == 0)
2886 break;
2887 last = 0;
2888 continue;
2889 }
2890
2891 btrfs_write_out_cache(root, trans, cache, path);
2892
2893 /*
2894 * If we didn't have an error then the cache state is still
2895 * NEED_WRITE, so we can set it to WRITTEN.
2896 */
2897 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2898 cache->disk_cache_state = BTRFS_DC_WRITTEN;
2899 last = cache->key.objectid + cache->key.offset;
2900 btrfs_put_block_group(cache);
2901 }
2902
2903 btrfs_free_path(path);
2904 return 0;
2905 }
2906
2907 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2908 {
2909 struct btrfs_block_group_cache *block_group;
2910 int readonly = 0;
2911
2912 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2913 if (!block_group || block_group->ro)
2914 readonly = 1;
2915 if (block_group)
2916 btrfs_put_block_group(block_group);
2917 return readonly;
2918 }
2919
2920 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2921 u64 total_bytes, u64 bytes_used,
2922 struct btrfs_space_info **space_info)
2923 {
2924 struct btrfs_space_info *found;
2925 int i;
2926 int factor;
2927
2928 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2929 BTRFS_BLOCK_GROUP_RAID10))
2930 factor = 2;
2931 else
2932 factor = 1;
2933
2934 found = __find_space_info(info, flags);
2935 if (found) {
2936 spin_lock(&found->lock);
2937 found->total_bytes += total_bytes;
2938 found->disk_total += total_bytes * factor;
2939 found->bytes_used += bytes_used;
2940 found->disk_used += bytes_used * factor;
2941 found->full = 0;
2942 spin_unlock(&found->lock);
2943 *space_info = found;
2944 return 0;
2945 }
2946 found = kzalloc(sizeof(*found), GFP_NOFS);
2947 if (!found)
2948 return -ENOMEM;
2949
2950 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
2951 INIT_LIST_HEAD(&found->block_groups[i]);
2952 init_rwsem(&found->groups_sem);
2953 spin_lock_init(&found->lock);
2954 found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
2955 BTRFS_BLOCK_GROUP_SYSTEM |
2956 BTRFS_BLOCK_GROUP_METADATA);
2957 found->total_bytes = total_bytes;
2958 found->disk_total = total_bytes * factor;
2959 found->bytes_used = bytes_used;
2960 found->disk_used = bytes_used * factor;
2961 found->bytes_pinned = 0;
2962 found->bytes_reserved = 0;
2963 found->bytes_readonly = 0;
2964 found->bytes_may_use = 0;
2965 found->full = 0;
2966 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
2967 found->chunk_alloc = 0;
2968 found->flush = 0;
2969 init_waitqueue_head(&found->wait);
2970 *space_info = found;
2971 list_add_rcu(&found->list, &info->space_info);
2972 return 0;
2973 }
2974
2975 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
2976 {
2977 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
2978 BTRFS_BLOCK_GROUP_RAID1 |
2979 BTRFS_BLOCK_GROUP_RAID10 |
2980 BTRFS_BLOCK_GROUP_DUP);
2981 if (extra_flags) {
2982 if (flags & BTRFS_BLOCK_GROUP_DATA)
2983 fs_info->avail_data_alloc_bits |= extra_flags;
2984 if (flags & BTRFS_BLOCK_GROUP_METADATA)
2985 fs_info->avail_metadata_alloc_bits |= extra_flags;
2986 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2987 fs_info->avail_system_alloc_bits |= extra_flags;
2988 }
2989 }
2990
2991 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
2992 {
2993 /*
2994 * we add in the count of missing devices because we want
2995 * to make sure that any RAID levels on a degraded FS
2996 * continue to be honored.
2997 */
2998 u64 num_devices = root->fs_info->fs_devices->rw_devices +
2999 root->fs_info->fs_devices->missing_devices;
3000
3001 if (num_devices == 1)
3002 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3003 if (num_devices < 4)
3004 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3005
3006 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3007 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3008 BTRFS_BLOCK_GROUP_RAID10))) {
3009 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3010 }
3011
3012 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3013 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3014 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3015 }
3016
3017 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3018 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3019 (flags & BTRFS_BLOCK_GROUP_RAID10) |
3020 (flags & BTRFS_BLOCK_GROUP_DUP)))
3021 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3022 return flags;
3023 }
3024
3025 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3026 {
3027 if (flags & BTRFS_BLOCK_GROUP_DATA)
3028 flags |= root->fs_info->avail_data_alloc_bits &
3029 root->fs_info->data_alloc_profile;
3030 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3031 flags |= root->fs_info->avail_system_alloc_bits &
3032 root->fs_info->system_alloc_profile;
3033 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3034 flags |= root->fs_info->avail_metadata_alloc_bits &
3035 root->fs_info->metadata_alloc_profile;
3036 return btrfs_reduce_alloc_profile(root, flags);
3037 }
3038
3039 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3040 {
3041 u64 flags;
3042
3043 if (data)
3044 flags = BTRFS_BLOCK_GROUP_DATA;
3045 else if (root == root->fs_info->chunk_root)
3046 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3047 else
3048 flags = BTRFS_BLOCK_GROUP_METADATA;
3049
3050 return get_alloc_profile(root, flags);
3051 }
3052
3053 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3054 {
3055 BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3056 BTRFS_BLOCK_GROUP_DATA);
3057 }
3058
3059 /*
3060 * This will check the space that the inode allocates from to make sure we have
3061 * enough space for bytes.
3062 */
3063 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3064 {
3065 struct btrfs_space_info *data_sinfo;
3066 struct btrfs_root *root = BTRFS_I(inode)->root;
3067 u64 used;
3068 int ret = 0, committed = 0, alloc_chunk = 1;
3069
3070 /* make sure bytes are sectorsize aligned */
3071 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3072
3073 if (root == root->fs_info->tree_root ||
3074 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3075 alloc_chunk = 0;
3076 committed = 1;
3077 }
3078
3079 data_sinfo = BTRFS_I(inode)->space_info;
3080 if (!data_sinfo)
3081 goto alloc;
3082
3083 again:
3084 /* make sure we have enough space to handle the data first */
3085 spin_lock(&data_sinfo->lock);
3086 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3087 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3088 data_sinfo->bytes_may_use;
3089
3090 if (used + bytes > data_sinfo->total_bytes) {
3091 struct btrfs_trans_handle *trans;
3092
3093 /*
3094 * if we don't have enough free bytes in this space then we need
3095 * to alloc a new chunk.
3096 */
3097 if (!data_sinfo->full && alloc_chunk) {
3098 u64 alloc_target;
3099
3100 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3101 spin_unlock(&data_sinfo->lock);
3102 alloc:
3103 alloc_target = btrfs_get_alloc_profile(root, 1);
3104 trans = btrfs_join_transaction(root);
3105 if (IS_ERR(trans))
3106 return PTR_ERR(trans);
3107
3108 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3109 bytes + 2 * 1024 * 1024,
3110 alloc_target,
3111 CHUNK_ALLOC_NO_FORCE);
3112 btrfs_end_transaction(trans, root);
3113 if (ret < 0) {
3114 if (ret != -ENOSPC)
3115 return ret;
3116 else
3117 goto commit_trans;
3118 }
3119
3120 if (!data_sinfo) {
3121 btrfs_set_inode_space_info(root, inode);
3122 data_sinfo = BTRFS_I(inode)->space_info;
3123 }
3124 goto again;
3125 }
3126
3127 /*
3128 * If we have less pinned bytes than we want to allocate then
3129 * don't bother committing the transaction, it won't help us.
3130 */
3131 if (data_sinfo->bytes_pinned < bytes)
3132 committed = 1;
3133 spin_unlock(&data_sinfo->lock);
3134
3135 /* commit the current transaction and try again */
3136 commit_trans:
3137 if (!committed &&
3138 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3139 committed = 1;
3140 trans = btrfs_join_transaction(root);
3141 if (IS_ERR(trans))
3142 return PTR_ERR(trans);
3143 ret = btrfs_commit_transaction(trans, root);
3144 if (ret)
3145 return ret;
3146 goto again;
3147 }
3148
3149 return -ENOSPC;
3150 }
3151 data_sinfo->bytes_may_use += bytes;
3152 spin_unlock(&data_sinfo->lock);
3153
3154 return 0;
3155 }
3156
3157 /*
3158 * Called if we need to clear a data reservation for this inode.
3159 */
3160 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3161 {
3162 struct btrfs_root *root = BTRFS_I(inode)->root;
3163 struct btrfs_space_info *data_sinfo;
3164
3165 /* make sure bytes are sectorsize aligned */
3166 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3167
3168 data_sinfo = BTRFS_I(inode)->space_info;
3169 spin_lock(&data_sinfo->lock);
3170 data_sinfo->bytes_may_use -= bytes;
3171 spin_unlock(&data_sinfo->lock);
3172 }
3173
3174 static void force_metadata_allocation(struct btrfs_fs_info *info)
3175 {
3176 struct list_head *head = &info->space_info;
3177 struct btrfs_space_info *found;
3178
3179 rcu_read_lock();
3180 list_for_each_entry_rcu(found, head, list) {
3181 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3182 found->force_alloc = CHUNK_ALLOC_FORCE;
3183 }
3184 rcu_read_unlock();
3185 }
3186
3187 static int should_alloc_chunk(struct btrfs_root *root,
3188 struct btrfs_space_info *sinfo, u64 alloc_bytes,
3189 int force)
3190 {
3191 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3192 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3193 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3194 u64 thresh;
3195
3196 if (force == CHUNK_ALLOC_FORCE)
3197 return 1;
3198
3199 /*
3200 * We need to take into account the global rsv because for all intents
3201 * and purposes it's used space. Don't worry about locking the
3202 * global_rsv, it doesn't change except when the transaction commits.
3203 */
3204 num_allocated += global_rsv->size;
3205
3206 /*
3207 * in limited mode, we want to have some free space up to
3208 * about 1% of the FS size.
3209 */
3210 if (force == CHUNK_ALLOC_LIMITED) {
3211 thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3212 thresh = max_t(u64, 64 * 1024 * 1024,
3213 div_factor_fine(thresh, 1));
3214
3215 if (num_bytes - num_allocated < thresh)
3216 return 1;
3217 }
3218
3219 /*
3220 * we have two similar checks here, one based on percentage
3221 * and once based on a hard number of 256MB. The idea
3222 * is that if we have a good amount of free
3223 * room, don't allocate a chunk. A good mount is
3224 * less than 80% utilized of the chunks we have allocated,
3225 * or more than 256MB free
3226 */
3227 if (num_allocated + alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3228 return 0;
3229
3230 if (num_allocated + alloc_bytes < div_factor(num_bytes, 8))
3231 return 0;
3232
3233 thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3234
3235 /* 256MB or 5% of the FS */
3236 thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5));
3237
3238 if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3))
3239 return 0;
3240 return 1;
3241 }
3242
3243 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3244 struct btrfs_root *extent_root, u64 alloc_bytes,
3245 u64 flags, int force)
3246 {
3247 struct btrfs_space_info *space_info;
3248 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3249 int wait_for_alloc = 0;
3250 int ret = 0;
3251
3252 flags = btrfs_reduce_alloc_profile(extent_root, flags);
3253
3254 space_info = __find_space_info(extent_root->fs_info, flags);
3255 if (!space_info) {
3256 ret = update_space_info(extent_root->fs_info, flags,
3257 0, 0, &space_info);
3258 BUG_ON(ret);
3259 }
3260 BUG_ON(!space_info);
3261
3262 again:
3263 spin_lock(&space_info->lock);
3264 if (space_info->force_alloc)
3265 force = space_info->force_alloc;
3266 if (space_info->full) {
3267 spin_unlock(&space_info->lock);
3268 return 0;
3269 }
3270
3271 if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
3272 spin_unlock(&space_info->lock);
3273 return 0;
3274 } else if (space_info->chunk_alloc) {
3275 wait_for_alloc = 1;
3276 } else {
3277 space_info->chunk_alloc = 1;
3278 }
3279
3280 spin_unlock(&space_info->lock);
3281
3282 mutex_lock(&fs_info->chunk_mutex);
3283
3284 /*
3285 * The chunk_mutex is held throughout the entirety of a chunk
3286 * allocation, so once we've acquired the chunk_mutex we know that the
3287 * other guy is done and we need to recheck and see if we should
3288 * allocate.
3289 */
3290 if (wait_for_alloc) {
3291 mutex_unlock(&fs_info->chunk_mutex);
3292 wait_for_alloc = 0;
3293 goto again;
3294 }
3295
3296 /*
3297 * If we have mixed data/metadata chunks we want to make sure we keep
3298 * allocating mixed chunks instead of individual chunks.
3299 */
3300 if (btrfs_mixed_space_info(space_info))
3301 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3302
3303 /*
3304 * if we're doing a data chunk, go ahead and make sure that
3305 * we keep a reasonable number of metadata chunks allocated in the
3306 * FS as well.
3307 */
3308 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3309 fs_info->data_chunk_allocations++;
3310 if (!(fs_info->data_chunk_allocations %
3311 fs_info->metadata_ratio))
3312 force_metadata_allocation(fs_info);
3313 }
3314
3315 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3316 if (ret < 0 && ret != -ENOSPC)
3317 goto out;
3318
3319 spin_lock(&space_info->lock);
3320 if (ret)
3321 space_info->full = 1;
3322 else
3323 ret = 1;
3324
3325 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3326 space_info->chunk_alloc = 0;
3327 spin_unlock(&space_info->lock);
3328 out:
3329 mutex_unlock(&extent_root->fs_info->chunk_mutex);
3330 return ret;
3331 }
3332
3333 /*
3334 * shrink metadata reservation for delalloc
3335 */
3336 static int shrink_delalloc(struct btrfs_trans_handle *trans,
3337 struct btrfs_root *root, u64 to_reclaim,
3338 bool wait_ordered)
3339 {
3340 struct btrfs_block_rsv *block_rsv;
3341 struct btrfs_space_info *space_info;
3342 u64 reserved;
3343 u64 max_reclaim;
3344 u64 reclaimed = 0;
3345 long time_left;
3346 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3347 int loops = 0;
3348 unsigned long progress;
3349
3350 block_rsv = &root->fs_info->delalloc_block_rsv;
3351 space_info = block_rsv->space_info;
3352
3353 smp_mb();
3354 reserved = space_info->bytes_may_use;
3355 progress = space_info->reservation_progress;
3356
3357 if (reserved == 0)
3358 return 0;
3359
3360 smp_mb();
3361 if (root->fs_info->delalloc_bytes == 0) {
3362 if (trans)
3363 return 0;
3364 btrfs_wait_ordered_extents(root, 0, 0);
3365 return 0;
3366 }
3367
3368 max_reclaim = min(reserved, to_reclaim);
3369 nr_pages = max_t(unsigned long, nr_pages,
3370 max_reclaim >> PAGE_CACHE_SHIFT);
3371 while (loops < 1024) {
3372 /* have the flusher threads jump in and do some IO */
3373 smp_mb();
3374 nr_pages = min_t(unsigned long, nr_pages,
3375 root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3376 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages);
3377
3378 spin_lock(&space_info->lock);
3379 if (reserved > space_info->bytes_may_use)
3380 reclaimed += reserved - space_info->bytes_may_use;
3381 reserved = space_info->bytes_may_use;
3382 spin_unlock(&space_info->lock);
3383
3384 loops++;
3385
3386 if (reserved == 0 || reclaimed >= max_reclaim)
3387 break;
3388
3389 if (trans && trans->transaction->blocked)
3390 return -EAGAIN;
3391
3392 if (wait_ordered && !trans) {
3393 btrfs_wait_ordered_extents(root, 0, 0);
3394 } else {
3395 time_left = schedule_timeout_interruptible(1);
3396
3397 /* We were interrupted, exit */
3398 if (time_left)
3399 break;
3400 }
3401
3402 /* we've kicked the IO a few times, if anything has been freed,
3403 * exit. There is no sense in looping here for a long time
3404 * when we really need to commit the transaction, or there are
3405 * just too many writers without enough free space
3406 */
3407
3408 if (loops > 3) {
3409 smp_mb();
3410 if (progress != space_info->reservation_progress)
3411 break;
3412 }
3413
3414 }
3415
3416 return reclaimed >= to_reclaim;
3417 }
3418
3419 /**
3420 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3421 * @root - the root we're allocating for
3422 * @block_rsv - the block_rsv we're allocating for
3423 * @orig_bytes - the number of bytes we want
3424 * @flush - wether or not we can flush to make our reservation
3425 * @check - wether this is just to check if we have enough space or not
3426 *
3427 * This will reserve orgi_bytes number of bytes from the space info associated
3428 * with the block_rsv. If there is not enough space it will make an attempt to
3429 * flush out space to make room. It will do this by flushing delalloc if
3430 * possible or committing the transaction. If flush is 0 then no attempts to
3431 * regain reservations will be made and this will fail if there is not enough
3432 * space already.
3433 */
3434 static int reserve_metadata_bytes(struct btrfs_root *root,
3435 struct btrfs_block_rsv *block_rsv,
3436 u64 orig_bytes, int flush, int check)
3437 {
3438 struct btrfs_space_info *space_info = block_rsv->space_info;
3439 struct btrfs_trans_handle *trans;
3440 u64 used;
3441 u64 num_bytes = orig_bytes;
3442 int retries = 0;
3443 int ret = 0;
3444 bool committed = false;
3445 bool flushing = false;
3446 bool wait_ordered = false;
3447
3448 trans = (struct btrfs_trans_handle *)current->journal_info;
3449 again:
3450 ret = 0;
3451 spin_lock(&space_info->lock);
3452 /*
3453 * We only want to wait if somebody other than us is flushing and we are
3454 * actually alloed to flush.
3455 */
3456 while (flush && !flushing && space_info->flush) {
3457 spin_unlock(&space_info->lock);
3458 /*
3459 * If we have a trans handle we can't wait because the flusher
3460 * may have to commit the transaction, which would mean we would
3461 * deadlock since we are waiting for the flusher to finish, but
3462 * hold the current transaction open.
3463 */
3464 if (trans)
3465 return -EAGAIN;
3466 ret = wait_event_interruptible(space_info->wait,
3467 !space_info->flush);
3468 /* Must have been interrupted, return */
3469 if (ret)
3470 return -EINTR;
3471
3472 spin_lock(&space_info->lock);
3473 }
3474
3475 ret = -ENOSPC;
3476 used = space_info->bytes_used + space_info->bytes_reserved +
3477 space_info->bytes_pinned + space_info->bytes_readonly +
3478 space_info->bytes_may_use;
3479
3480 /*
3481 * The idea here is that we've not already over-reserved the block group
3482 * then we can go ahead and save our reservation first and then start
3483 * flushing if we need to. Otherwise if we've already overcommitted
3484 * lets start flushing stuff first and then come back and try to make
3485 * our reservation.
3486 */
3487 if (used <= space_info->total_bytes) {
3488 if (used + orig_bytes <= space_info->total_bytes) {
3489 space_info->bytes_may_use += orig_bytes;
3490 ret = 0;
3491 } else {
3492 /*
3493 * Ok set num_bytes to orig_bytes since we aren't
3494 * overocmmitted, this way we only try and reclaim what
3495 * we need.
3496 */
3497 num_bytes = orig_bytes;
3498 }
3499 } else {
3500 /*
3501 * Ok we're over committed, set num_bytes to the overcommitted
3502 * amount plus the amount of bytes that we need for this
3503 * reservation.
3504 */
3505 wait_ordered = true;
3506 num_bytes = used - space_info->total_bytes +
3507 (orig_bytes * (retries + 1));
3508 }
3509
3510 if (ret && !check) {
3511 u64 profile = btrfs_get_alloc_profile(root, 0);
3512 u64 avail;
3513
3514 spin_lock(&root->fs_info->free_chunk_lock);
3515 avail = root->fs_info->free_chunk_space;
3516
3517 /*
3518 * If we have dup, raid1 or raid10 then only half of the free
3519 * space is actually useable.
3520 */
3521 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3522 BTRFS_BLOCK_GROUP_RAID1 |
3523 BTRFS_BLOCK_GROUP_RAID10))
3524 avail >>= 1;
3525
3526 /*
3527 * If we aren't flushing don't let us overcommit too much, say
3528 * 1/8th of the space. If we can flush, let it overcommit up to
3529 * 1/2 of the space.
3530 */
3531 if (flush)
3532 avail >>= 3;
3533 else
3534 avail >>= 1;
3535 spin_unlock(&root->fs_info->free_chunk_lock);
3536
3537 if (used + num_bytes < space_info->total_bytes + avail) {
3538 space_info->bytes_may_use += orig_bytes;
3539 ret = 0;
3540 } else {
3541 wait_ordered = true;
3542 }
3543 }
3544
3545 /*
3546 * Couldn't make our reservation, save our place so while we're trying
3547 * to reclaim space we can actually use it instead of somebody else
3548 * stealing it from us.
3549 */
3550 if (ret && flush) {
3551 flushing = true;
3552 space_info->flush = 1;
3553 }
3554
3555 spin_unlock(&space_info->lock);
3556
3557 if (!ret || !flush)
3558 goto out;
3559
3560 /*
3561 * We do synchronous shrinking since we don't actually unreserve
3562 * metadata until after the IO is completed.
3563 */
3564 ret = shrink_delalloc(trans, root, num_bytes, wait_ordered);
3565 if (ret < 0)
3566 goto out;
3567
3568 ret = 0;
3569
3570 /*
3571 * So if we were overcommitted it's possible that somebody else flushed
3572 * out enough space and we simply didn't have enough space to reclaim,
3573 * so go back around and try again.
3574 */
3575 if (retries < 2) {
3576 wait_ordered = true;
3577 retries++;
3578 goto again;
3579 }
3580
3581 ret = -EAGAIN;
3582 if (trans)
3583 goto out;
3584
3585 ret = -ENOSPC;
3586 if (committed)
3587 goto out;
3588
3589 trans = btrfs_join_transaction(root);
3590 if (IS_ERR(trans))
3591 goto out;
3592 ret = btrfs_commit_transaction(trans, root);
3593 if (!ret) {
3594 trans = NULL;
3595 committed = true;
3596 goto again;
3597 }
3598
3599 out:
3600 if (flushing) {
3601 spin_lock(&space_info->lock);
3602 space_info->flush = 0;
3603 wake_up_all(&space_info->wait);
3604 spin_unlock(&space_info->lock);
3605 }
3606 return ret;
3607 }
3608
3609 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3610 struct btrfs_root *root)
3611 {
3612 struct btrfs_block_rsv *block_rsv = NULL;
3613
3614 if (root->ref_cows || root == root->fs_info->csum_root)
3615 block_rsv = trans->block_rsv;
3616
3617 if (!block_rsv)
3618 block_rsv = root->block_rsv;
3619
3620 if (!block_rsv)
3621 block_rsv = &root->fs_info->empty_block_rsv;
3622
3623 return block_rsv;
3624 }
3625
3626 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3627 u64 num_bytes)
3628 {
3629 int ret = -ENOSPC;
3630 spin_lock(&block_rsv->lock);
3631 if (block_rsv->reserved >= num_bytes) {
3632 block_rsv->reserved -= num_bytes;
3633 if (block_rsv->reserved < block_rsv->size)
3634 block_rsv->full = 0;
3635 ret = 0;
3636 }
3637 spin_unlock(&block_rsv->lock);
3638 return ret;
3639 }
3640
3641 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3642 u64 num_bytes, int update_size)
3643 {
3644 spin_lock(&block_rsv->lock);
3645 block_rsv->reserved += num_bytes;
3646 if (update_size)
3647 block_rsv->size += num_bytes;
3648 else if (block_rsv->reserved >= block_rsv->size)
3649 block_rsv->full = 1;
3650 spin_unlock(&block_rsv->lock);
3651 }
3652
3653 static void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3654 struct btrfs_block_rsv *dest, u64 num_bytes)
3655 {
3656 struct btrfs_space_info *space_info = block_rsv->space_info;
3657
3658 spin_lock(&block_rsv->lock);
3659 if (num_bytes == (u64)-1)
3660 num_bytes = block_rsv->size;
3661 block_rsv->size -= num_bytes;
3662 if (block_rsv->reserved >= block_rsv->size) {
3663 num_bytes = block_rsv->reserved - block_rsv->size;
3664 block_rsv->reserved = block_rsv->size;
3665 block_rsv->full = 1;
3666 } else {
3667 num_bytes = 0;
3668 }
3669 spin_unlock(&block_rsv->lock);
3670
3671 if (num_bytes > 0) {
3672 if (dest) {
3673 spin_lock(&dest->lock);
3674 if (!dest->full) {
3675 u64 bytes_to_add;
3676
3677 bytes_to_add = dest->size - dest->reserved;
3678 bytes_to_add = min(num_bytes, bytes_to_add);
3679 dest->reserved += bytes_to_add;
3680 if (dest->reserved >= dest->size)
3681 dest->full = 1;
3682 num_bytes -= bytes_to_add;
3683 }
3684 spin_unlock(&dest->lock);
3685 }
3686 if (num_bytes) {
3687 spin_lock(&space_info->lock);
3688 space_info->bytes_may_use -= num_bytes;
3689 space_info->reservation_progress++;
3690 spin_unlock(&space_info->lock);
3691 }
3692 }
3693 }
3694
3695 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3696 struct btrfs_block_rsv *dst, u64 num_bytes)
3697 {
3698 int ret;
3699
3700 ret = block_rsv_use_bytes(src, num_bytes);
3701 if (ret)
3702 return ret;
3703
3704 block_rsv_add_bytes(dst, num_bytes, 1);
3705 return 0;
3706 }
3707
3708 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3709 {
3710 memset(rsv, 0, sizeof(*rsv));
3711 spin_lock_init(&rsv->lock);
3712 }
3713
3714 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3715 {
3716 struct btrfs_block_rsv *block_rsv;
3717 struct btrfs_fs_info *fs_info = root->fs_info;
3718
3719 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3720 if (!block_rsv)
3721 return NULL;
3722
3723 btrfs_init_block_rsv(block_rsv);
3724 block_rsv->space_info = __find_space_info(fs_info,
3725 BTRFS_BLOCK_GROUP_METADATA);
3726 return block_rsv;
3727 }
3728
3729 void btrfs_free_block_rsv(struct btrfs_root *root,
3730 struct btrfs_block_rsv *rsv)
3731 {
3732 btrfs_block_rsv_release(root, rsv, (u64)-1);
3733 kfree(rsv);
3734 }
3735
3736 int btrfs_block_rsv_add(struct btrfs_root *root,
3737 struct btrfs_block_rsv *block_rsv,
3738 u64 num_bytes)
3739 {
3740 int ret;
3741
3742 if (num_bytes == 0)
3743 return 0;
3744
3745 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, 1, 0);
3746 if (!ret) {
3747 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3748 return 0;
3749 }
3750
3751 return ret;
3752 }
3753
3754 int btrfs_block_rsv_check(struct btrfs_root *root,
3755 struct btrfs_block_rsv *block_rsv,
3756 u64 min_reserved, int min_factor, int flush)
3757 {
3758 u64 num_bytes = 0;
3759 int ret = -ENOSPC;
3760
3761 if (!block_rsv)
3762 return 0;
3763
3764 spin_lock(&block_rsv->lock);
3765 if (min_factor > 0)
3766 num_bytes = div_factor(block_rsv->size, min_factor);
3767 if (min_reserved > num_bytes)
3768 num_bytes = min_reserved;
3769
3770 if (block_rsv->reserved >= num_bytes)
3771 ret = 0;
3772 else
3773 num_bytes -= block_rsv->reserved;
3774 spin_unlock(&block_rsv->lock);
3775
3776 if (!ret)
3777 return 0;
3778
3779 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush, !flush);
3780 if (!ret) {
3781 block_rsv_add_bytes(block_rsv, num_bytes, 0);
3782 return 0;
3783 }
3784
3785 return ret;
3786 }
3787
3788 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3789 struct btrfs_block_rsv *dst_rsv,
3790 u64 num_bytes)
3791 {
3792 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3793 }
3794
3795 void btrfs_block_rsv_release(struct btrfs_root *root,
3796 struct btrfs_block_rsv *block_rsv,
3797 u64 num_bytes)
3798 {
3799 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3800 if (global_rsv->full || global_rsv == block_rsv ||
3801 block_rsv->space_info != global_rsv->space_info)
3802 global_rsv = NULL;
3803 block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3804 }
3805
3806 /*
3807 * helper to calculate size of global block reservation.
3808 * the desired value is sum of space used by extent tree,
3809 * checksum tree and root tree
3810 */
3811 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3812 {
3813 struct btrfs_space_info *sinfo;
3814 u64 num_bytes;
3815 u64 meta_used;
3816 u64 data_used;
3817 int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
3818
3819 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3820 spin_lock(&sinfo->lock);
3821 data_used = sinfo->bytes_used;
3822 spin_unlock(&sinfo->lock);
3823
3824 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3825 spin_lock(&sinfo->lock);
3826 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3827 data_used = 0;
3828 meta_used = sinfo->bytes_used;
3829 spin_unlock(&sinfo->lock);
3830
3831 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3832 csum_size * 2;
3833 num_bytes += div64_u64(data_used + meta_used, 50);
3834
3835 if (num_bytes * 3 > meta_used)
3836 num_bytes = div64_u64(meta_used, 3);
3837
3838 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3839 }
3840
3841 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3842 {
3843 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3844 struct btrfs_space_info *sinfo = block_rsv->space_info;
3845 u64 num_bytes;
3846
3847 num_bytes = calc_global_metadata_size(fs_info);
3848
3849 spin_lock(&block_rsv->lock);
3850 spin_lock(&sinfo->lock);
3851
3852 block_rsv->size = num_bytes;
3853
3854 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3855 sinfo->bytes_reserved + sinfo->bytes_readonly +
3856 sinfo->bytes_may_use;
3857
3858 if (sinfo->total_bytes > num_bytes) {
3859 num_bytes = sinfo->total_bytes - num_bytes;
3860 block_rsv->reserved += num_bytes;
3861 sinfo->bytes_may_use += num_bytes;
3862 }
3863
3864 if (block_rsv->reserved >= block_rsv->size) {
3865 num_bytes = block_rsv->reserved - block_rsv->size;
3866 sinfo->bytes_may_use -= num_bytes;
3867 sinfo->reservation_progress++;
3868 block_rsv->reserved = block_rsv->size;
3869 block_rsv->full = 1;
3870 }
3871
3872 spin_unlock(&sinfo->lock);
3873 spin_unlock(&block_rsv->lock);
3874 }
3875
3876 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3877 {
3878 struct btrfs_space_info *space_info;
3879
3880 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3881 fs_info->chunk_block_rsv.space_info = space_info;
3882
3883 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3884 fs_info->global_block_rsv.space_info = space_info;
3885 fs_info->delalloc_block_rsv.space_info = space_info;
3886 fs_info->trans_block_rsv.space_info = space_info;
3887 fs_info->empty_block_rsv.space_info = space_info;
3888
3889 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3890 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3891 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3892 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
3893 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3894
3895 update_global_block_rsv(fs_info);
3896 }
3897
3898 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3899 {
3900 block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3901 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3902 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3903 WARN_ON(fs_info->trans_block_rsv.size > 0);
3904 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3905 WARN_ON(fs_info->chunk_block_rsv.size > 0);
3906 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
3907 }
3908
3909 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
3910 struct btrfs_root *root)
3911 {
3912 if (!trans->bytes_reserved)
3913 return;
3914
3915 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
3916 trans->bytes_reserved = 0;
3917 }
3918
3919 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
3920 struct inode *inode)
3921 {
3922 struct btrfs_root *root = BTRFS_I(inode)->root;
3923 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3924 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
3925
3926 /*
3927 * We need to hold space in order to delete our orphan item once we've
3928 * added it, so this takes the reservation so we can release it later
3929 * when we are truly done with the orphan item.
3930 */
3931 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
3932 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3933 }
3934
3935 void btrfs_orphan_release_metadata(struct inode *inode)
3936 {
3937 struct btrfs_root *root = BTRFS_I(inode)->root;
3938 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
3939 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
3940 }
3941
3942 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
3943 struct btrfs_pending_snapshot *pending)
3944 {
3945 struct btrfs_root *root = pending->root;
3946 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3947 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
3948 /*
3949 * two for root back/forward refs, two for directory entries
3950 * and one for root of the snapshot.
3951 */
3952 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3953 dst_rsv->space_info = src_rsv->space_info;
3954 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3955 }
3956
3957 /**
3958 * drop_outstanding_extent - drop an outstanding extent
3959 * @inode: the inode we're dropping the extent for
3960 *
3961 * This is called when we are freeing up an outstanding extent, either called
3962 * after an error or after an extent is written. This will return the number of
3963 * reserved extents that need to be freed. This must be called with
3964 * BTRFS_I(inode)->lock held.
3965 */
3966 static unsigned drop_outstanding_extent(struct inode *inode)
3967 {
3968 unsigned dropped_extents = 0;
3969
3970 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
3971 BTRFS_I(inode)->outstanding_extents--;
3972
3973 /*
3974 * If we have more or the same amount of outsanding extents than we have
3975 * reserved then we need to leave the reserved extents count alone.
3976 */
3977 if (BTRFS_I(inode)->outstanding_extents >=
3978 BTRFS_I(inode)->reserved_extents)
3979 return 0;
3980
3981 dropped_extents = BTRFS_I(inode)->reserved_extents -
3982 BTRFS_I(inode)->outstanding_extents;
3983 BTRFS_I(inode)->reserved_extents -= dropped_extents;
3984 return dropped_extents;
3985 }
3986
3987 /**
3988 * calc_csum_metadata_size - return the amount of metada space that must be
3989 * reserved/free'd for the given bytes.
3990 * @inode: the inode we're manipulating
3991 * @num_bytes: the number of bytes in question
3992 * @reserve: 1 if we are reserving space, 0 if we are freeing space
3993 *
3994 * This adjusts the number of csum_bytes in the inode and then returns the
3995 * correct amount of metadata that must either be reserved or freed. We
3996 * calculate how many checksums we can fit into one leaf and then divide the
3997 * number of bytes that will need to be checksumed by this value to figure out
3998 * how many checksums will be required. If we are adding bytes then the number
3999 * may go up and we will return the number of additional bytes that must be
4000 * reserved. If it is going down we will return the number of bytes that must
4001 * be freed.
4002 *
4003 * This must be called with BTRFS_I(inode)->lock held.
4004 */
4005 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4006 int reserve)
4007 {
4008 struct btrfs_root *root = BTRFS_I(inode)->root;
4009 u64 csum_size;
4010 int num_csums_per_leaf;
4011 int num_csums;
4012 int old_csums;
4013
4014 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4015 BTRFS_I(inode)->csum_bytes == 0)
4016 return 0;
4017
4018 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4019 if (reserve)
4020 BTRFS_I(inode)->csum_bytes += num_bytes;
4021 else
4022 BTRFS_I(inode)->csum_bytes -= num_bytes;
4023 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4024 num_csums_per_leaf = (int)div64_u64(csum_size,
4025 sizeof(struct btrfs_csum_item) +
4026 sizeof(struct btrfs_disk_key));
4027 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4028 num_csums = num_csums + num_csums_per_leaf - 1;
4029 num_csums = num_csums / num_csums_per_leaf;
4030
4031 old_csums = old_csums + num_csums_per_leaf - 1;
4032 old_csums = old_csums / num_csums_per_leaf;
4033
4034 /* No change, no need to reserve more */
4035 if (old_csums == num_csums)
4036 return 0;
4037
4038 if (reserve)
4039 return btrfs_calc_trans_metadata_size(root,
4040 num_csums - old_csums);
4041
4042 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4043 }
4044
4045 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4046 {
4047 struct btrfs_root *root = BTRFS_I(inode)->root;
4048 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4049 u64 to_reserve = 0;
4050 unsigned nr_extents = 0;
4051 int flush = 1;
4052 int ret;
4053
4054 if (btrfs_is_free_space_inode(root, inode))
4055 flush = 0;
4056
4057 if (flush && btrfs_transaction_in_commit(root->fs_info))
4058 schedule_timeout(1);
4059
4060 num_bytes = ALIGN(num_bytes, root->sectorsize);
4061
4062 spin_lock(&BTRFS_I(inode)->lock);
4063 BTRFS_I(inode)->outstanding_extents++;
4064
4065 if (BTRFS_I(inode)->outstanding_extents >
4066 BTRFS_I(inode)->reserved_extents) {
4067 nr_extents = BTRFS_I(inode)->outstanding_extents -
4068 BTRFS_I(inode)->reserved_extents;
4069 BTRFS_I(inode)->reserved_extents += nr_extents;
4070
4071 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4072 }
4073 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4074 spin_unlock(&BTRFS_I(inode)->lock);
4075
4076 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush, 0);
4077 if (ret) {
4078 u64 to_free = 0;
4079 unsigned dropped;
4080
4081 spin_lock(&BTRFS_I(inode)->lock);
4082 dropped = drop_outstanding_extent(inode);
4083 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4084 spin_unlock(&BTRFS_I(inode)->lock);
4085 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4086
4087 /*
4088 * Somebody could have come in and twiddled with the
4089 * reservation, so if we have to free more than we would have
4090 * reserved from this reservation go ahead and release those
4091 * bytes.
4092 */
4093 to_free -= to_reserve;
4094 if (to_free)
4095 btrfs_block_rsv_release(root, block_rsv, to_free);
4096 return ret;
4097 }
4098
4099 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4100
4101 return 0;
4102 }
4103
4104 /**
4105 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4106 * @inode: the inode to release the reservation for
4107 * @num_bytes: the number of bytes we're releasing
4108 *
4109 * This will release the metadata reservation for an inode. This can be called
4110 * once we complete IO for a given set of bytes to release their metadata
4111 * reservations.
4112 */
4113 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4114 {
4115 struct btrfs_root *root = BTRFS_I(inode)->root;
4116 u64 to_free = 0;
4117 unsigned dropped;
4118
4119 num_bytes = ALIGN(num_bytes, root->sectorsize);
4120 spin_lock(&BTRFS_I(inode)->lock);
4121 dropped = drop_outstanding_extent(inode);
4122
4123 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4124 spin_unlock(&BTRFS_I(inode)->lock);
4125 if (dropped > 0)
4126 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4127
4128 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4129 to_free);
4130 }
4131
4132 /**
4133 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4134 * @inode: inode we're writing to
4135 * @num_bytes: the number of bytes we want to allocate
4136 *
4137 * This will do the following things
4138 *
4139 * o reserve space in the data space info for num_bytes
4140 * o reserve space in the metadata space info based on number of outstanding
4141 * extents and how much csums will be needed
4142 * o add to the inodes ->delalloc_bytes
4143 * o add it to the fs_info's delalloc inodes list.
4144 *
4145 * This will return 0 for success and -ENOSPC if there is no space left.
4146 */
4147 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4148 {
4149 int ret;
4150
4151 ret = btrfs_check_data_free_space(inode, num_bytes);
4152 if (ret)
4153 return ret;
4154
4155 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4156 if (ret) {
4157 btrfs_free_reserved_data_space(inode, num_bytes);
4158 return ret;
4159 }
4160
4161 return 0;
4162 }
4163
4164 /**
4165 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4166 * @inode: inode we're releasing space for
4167 * @num_bytes: the number of bytes we want to free up
4168 *
4169 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4170 * called in the case that we don't need the metadata AND data reservations
4171 * anymore. So if there is an error or we insert an inline extent.
4172 *
4173 * This function will release the metadata space that was not used and will
4174 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4175 * list if there are no delalloc bytes left.
4176 */
4177 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4178 {
4179 btrfs_delalloc_release_metadata(inode, num_bytes);
4180 btrfs_free_reserved_data_space(inode, num_bytes);
4181 }
4182
4183 static int update_block_group(struct btrfs_trans_handle *trans,
4184 struct btrfs_root *root,
4185 u64 bytenr, u64 num_bytes, int alloc)
4186 {
4187 struct btrfs_block_group_cache *cache = NULL;
4188 struct btrfs_fs_info *info = root->fs_info;
4189 u64 total = num_bytes;
4190 u64 old_val;
4191 u64 byte_in_group;
4192 int factor;
4193
4194 /* block accounting for super block */
4195 spin_lock(&info->delalloc_lock);
4196 old_val = btrfs_super_bytes_used(&info->super_copy);
4197 if (alloc)
4198 old_val += num_bytes;
4199 else
4200 old_val -= num_bytes;
4201 btrfs_set_super_bytes_used(&info->super_copy, old_val);
4202 spin_unlock(&info->delalloc_lock);
4203
4204 while (total) {
4205 cache = btrfs_lookup_block_group(info, bytenr);
4206 if (!cache)
4207 return -1;
4208 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4209 BTRFS_BLOCK_GROUP_RAID1 |
4210 BTRFS_BLOCK_GROUP_RAID10))
4211 factor = 2;
4212 else
4213 factor = 1;
4214 /*
4215 * If this block group has free space cache written out, we
4216 * need to make sure to load it if we are removing space. This
4217 * is because we need the unpinning stage to actually add the
4218 * space back to the block group, otherwise we will leak space.
4219 */
4220 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4221 cache_block_group(cache, trans, NULL, 1);
4222
4223 byte_in_group = bytenr - cache->key.objectid;
4224 WARN_ON(byte_in_group > cache->key.offset);
4225
4226 spin_lock(&cache->space_info->lock);
4227 spin_lock(&cache->lock);
4228
4229 if (btrfs_test_opt(root, SPACE_CACHE) &&
4230 cache->disk_cache_state < BTRFS_DC_CLEAR)
4231 cache->disk_cache_state = BTRFS_DC_CLEAR;
4232
4233 cache->dirty = 1;
4234 old_val = btrfs_block_group_used(&cache->item);
4235 num_bytes = min(total, cache->key.offset - byte_in_group);
4236 if (alloc) {
4237 old_val += num_bytes;
4238 btrfs_set_block_group_used(&cache->item, old_val);
4239 cache->reserved -= num_bytes;
4240 cache->space_info->bytes_reserved -= num_bytes;
4241 cache->space_info->bytes_used += num_bytes;
4242 cache->space_info->disk_used += num_bytes * factor;
4243 spin_unlock(&cache->lock);
4244 spin_unlock(&cache->space_info->lock);
4245 } else {
4246 old_val -= num_bytes;
4247 btrfs_set_block_group_used(&cache->item, old_val);
4248 cache->pinned += num_bytes;
4249 cache->space_info->bytes_pinned += num_bytes;
4250 cache->space_info->bytes_used -= num_bytes;
4251 cache->space_info->disk_used -= num_bytes * factor;
4252 spin_unlock(&cache->lock);
4253 spin_unlock(&cache->space_info->lock);
4254
4255 set_extent_dirty(info->pinned_extents,
4256 bytenr, bytenr + num_bytes - 1,
4257 GFP_NOFS | __GFP_NOFAIL);
4258 }
4259 btrfs_put_block_group(cache);
4260 total -= num_bytes;
4261 bytenr += num_bytes;
4262 }
4263 return 0;
4264 }
4265
4266 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4267 {
4268 struct btrfs_block_group_cache *cache;
4269 u64 bytenr;
4270
4271 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4272 if (!cache)
4273 return 0;
4274
4275 bytenr = cache->key.objectid;
4276 btrfs_put_block_group(cache);
4277
4278 return bytenr;
4279 }
4280
4281 static int pin_down_extent(struct btrfs_root *root,
4282 struct btrfs_block_group_cache *cache,
4283 u64 bytenr, u64 num_bytes, int reserved)
4284 {
4285 spin_lock(&cache->space_info->lock);
4286 spin_lock(&cache->lock);
4287 cache->pinned += num_bytes;
4288 cache->space_info->bytes_pinned += num_bytes;
4289 if (reserved) {
4290 cache->reserved -= num_bytes;
4291 cache->space_info->bytes_reserved -= num_bytes;
4292 }
4293 spin_unlock(&cache->lock);
4294 spin_unlock(&cache->space_info->lock);
4295
4296 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4297 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4298 return 0;
4299 }
4300
4301 /*
4302 * this function must be called within transaction
4303 */
4304 int btrfs_pin_extent(struct btrfs_root *root,
4305 u64 bytenr, u64 num_bytes, int reserved)
4306 {
4307 struct btrfs_block_group_cache *cache;
4308
4309 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4310 BUG_ON(!cache);
4311
4312 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4313
4314 btrfs_put_block_group(cache);
4315 return 0;
4316 }
4317
4318 /**
4319 * btrfs_update_reserved_bytes - update the block_group and space info counters
4320 * @cache: The cache we are manipulating
4321 * @num_bytes: The number of bytes in question
4322 * @reserve: One of the reservation enums
4323 *
4324 * This is called by the allocator when it reserves space, or by somebody who is
4325 * freeing space that was never actually used on disk. For example if you
4326 * reserve some space for a new leaf in transaction A and before transaction A
4327 * commits you free that leaf, you call this with reserve set to 0 in order to
4328 * clear the reservation.
4329 *
4330 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4331 * ENOSPC accounting. For data we handle the reservation through clearing the
4332 * delalloc bits in the io_tree. We have to do this since we could end up
4333 * allocating less disk space for the amount of data we have reserved in the
4334 * case of compression.
4335 *
4336 * If this is a reservation and the block group has become read only we cannot
4337 * make the reservation and return -EAGAIN, otherwise this function always
4338 * succeeds.
4339 */
4340 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4341 u64 num_bytes, int reserve)
4342 {
4343 struct btrfs_space_info *space_info = cache->space_info;
4344 int ret = 0;
4345 spin_lock(&space_info->lock);
4346 spin_lock(&cache->lock);
4347 if (reserve != RESERVE_FREE) {
4348 if (cache->ro) {
4349 ret = -EAGAIN;
4350 } else {
4351 cache->reserved += num_bytes;
4352 space_info->bytes_reserved += num_bytes;
4353 if (reserve == RESERVE_ALLOC) {
4354 BUG_ON(space_info->bytes_may_use < num_bytes);
4355 space_info->bytes_may_use -= num_bytes;
4356 }
4357 }
4358 } else {
4359 if (cache->ro)
4360 space_info->bytes_readonly += num_bytes;
4361 cache->reserved -= num_bytes;
4362 space_info->bytes_reserved -= num_bytes;
4363 space_info->reservation_progress++;
4364 }
4365 spin_unlock(&cache->lock);
4366 spin_unlock(&space_info->lock);
4367 return ret;
4368 }
4369
4370 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4371 struct btrfs_root *root)
4372 {
4373 struct btrfs_fs_info *fs_info = root->fs_info;
4374 struct btrfs_caching_control *next;
4375 struct btrfs_caching_control *caching_ctl;
4376 struct btrfs_block_group_cache *cache;
4377
4378 down_write(&fs_info->extent_commit_sem);
4379
4380 list_for_each_entry_safe(caching_ctl, next,
4381 &fs_info->caching_block_groups, list) {
4382 cache = caching_ctl->block_group;
4383 if (block_group_cache_done(cache)) {
4384 cache->last_byte_to_unpin = (u64)-1;
4385 list_del_init(&caching_ctl->list);
4386 put_caching_control(caching_ctl);
4387 } else {
4388 cache->last_byte_to_unpin = caching_ctl->progress;
4389 }
4390 }
4391
4392 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4393 fs_info->pinned_extents = &fs_info->freed_extents[1];
4394 else
4395 fs_info->pinned_extents = &fs_info->freed_extents[0];
4396
4397 up_write(&fs_info->extent_commit_sem);
4398
4399 update_global_block_rsv(fs_info);
4400 return 0;
4401 }
4402
4403 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4404 {
4405 struct btrfs_fs_info *fs_info = root->fs_info;
4406 struct btrfs_block_group_cache *cache = NULL;
4407 u64 len;
4408
4409 while (start <= end) {
4410 if (!cache ||
4411 start >= cache->key.objectid + cache->key.offset) {
4412 if (cache)
4413 btrfs_put_block_group(cache);
4414 cache = btrfs_lookup_block_group(fs_info, start);
4415 BUG_ON(!cache);
4416 }
4417
4418 len = cache->key.objectid + cache->key.offset - start;
4419 len = min(len, end + 1 - start);
4420
4421 if (start < cache->last_byte_to_unpin) {
4422 len = min(len, cache->last_byte_to_unpin - start);
4423 btrfs_add_free_space(cache, start, len);
4424 }
4425
4426 start += len;
4427
4428 spin_lock(&cache->space_info->lock);
4429 spin_lock(&cache->lock);
4430 cache->pinned -= len;
4431 cache->space_info->bytes_pinned -= len;
4432 if (cache->ro)
4433 cache->space_info->bytes_readonly += len;
4434 spin_unlock(&cache->lock);
4435 spin_unlock(&cache->space_info->lock);
4436 }
4437
4438 if (cache)
4439 btrfs_put_block_group(cache);
4440 return 0;
4441 }
4442
4443 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4444 struct btrfs_root *root)
4445 {
4446 struct btrfs_fs_info *fs_info = root->fs_info;
4447 struct extent_io_tree *unpin;
4448 u64 start;
4449 u64 end;
4450 int ret;
4451
4452 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4453 unpin = &fs_info->freed_extents[1];
4454 else
4455 unpin = &fs_info->freed_extents[0];
4456
4457 while (1) {
4458 ret = find_first_extent_bit(unpin, 0, &start, &end,
4459 EXTENT_DIRTY);
4460 if (ret)
4461 break;
4462
4463 if (btrfs_test_opt(root, DISCARD))
4464 ret = btrfs_discard_extent(root, start,
4465 end + 1 - start, NULL);
4466
4467 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4468 unpin_extent_range(root, start, end);
4469 cond_resched();
4470 }
4471
4472 return 0;
4473 }
4474
4475 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4476 struct btrfs_root *root,
4477 u64 bytenr, u64 num_bytes, u64 parent,
4478 u64 root_objectid, u64 owner_objectid,
4479 u64 owner_offset, int refs_to_drop,
4480 struct btrfs_delayed_extent_op *extent_op)
4481 {
4482 struct btrfs_key key;
4483 struct btrfs_path *path;
4484 struct btrfs_fs_info *info = root->fs_info;
4485 struct btrfs_root *extent_root = info->extent_root;
4486 struct extent_buffer *leaf;
4487 struct btrfs_extent_item *ei;
4488 struct btrfs_extent_inline_ref *iref;
4489 int ret;
4490 int is_data;
4491 int extent_slot = 0;
4492 int found_extent = 0;
4493 int num_to_del = 1;
4494 u32 item_size;
4495 u64 refs;
4496
4497 path = btrfs_alloc_path();
4498 if (!path)
4499 return -ENOMEM;
4500
4501 path->reada = 1;
4502 path->leave_spinning = 1;
4503
4504 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4505 BUG_ON(!is_data && refs_to_drop != 1);
4506
4507 ret = lookup_extent_backref(trans, extent_root, path, &iref,
4508 bytenr, num_bytes, parent,
4509 root_objectid, owner_objectid,
4510 owner_offset);
4511 if (ret == 0) {
4512 extent_slot = path->slots[0];
4513 while (extent_slot >= 0) {
4514 btrfs_item_key_to_cpu(path->nodes[0], &key,
4515 extent_slot);
4516 if (key.objectid != bytenr)
4517 break;
4518 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4519 key.offset == num_bytes) {
4520 found_extent = 1;
4521 break;
4522 }
4523 if (path->slots[0] - extent_slot > 5)
4524 break;
4525 extent_slot--;
4526 }
4527 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4528 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4529 if (found_extent && item_size < sizeof(*ei))
4530 found_extent = 0;
4531 #endif
4532 if (!found_extent) {
4533 BUG_ON(iref);
4534 ret = remove_extent_backref(trans, extent_root, path,
4535 NULL, refs_to_drop,
4536 is_data);
4537 BUG_ON(ret);
4538 btrfs_release_path(path);
4539 path->leave_spinning = 1;
4540
4541 key.objectid = bytenr;
4542 key.type = BTRFS_EXTENT_ITEM_KEY;
4543 key.offset = num_bytes;
4544
4545 ret = btrfs_search_slot(trans, extent_root,
4546 &key, path, -1, 1);
4547 if (ret) {
4548 printk(KERN_ERR "umm, got %d back from search"
4549 ", was looking for %llu\n", ret,
4550 (unsigned long long)bytenr);
4551 if (ret > 0)
4552 btrfs_print_leaf(extent_root,
4553 path->nodes[0]);
4554 }
4555 BUG_ON(ret);
4556 extent_slot = path->slots[0];
4557 }
4558 } else {
4559 btrfs_print_leaf(extent_root, path->nodes[0]);
4560 WARN_ON(1);
4561 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4562 "parent %llu root %llu owner %llu offset %llu\n",
4563 (unsigned long long)bytenr,
4564 (unsigned long long)parent,
4565 (unsigned long long)root_objectid,
4566 (unsigned long long)owner_objectid,
4567 (unsigned long long)owner_offset);
4568 }
4569
4570 leaf = path->nodes[0];
4571 item_size = btrfs_item_size_nr(leaf, extent_slot);
4572 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4573 if (item_size < sizeof(*ei)) {
4574 BUG_ON(found_extent || extent_slot != path->slots[0]);
4575 ret = convert_extent_item_v0(trans, extent_root, path,
4576 owner_objectid, 0);
4577 BUG_ON(ret < 0);
4578
4579 btrfs_release_path(path);
4580 path->leave_spinning = 1;
4581
4582 key.objectid = bytenr;
4583 key.type = BTRFS_EXTENT_ITEM_KEY;
4584 key.offset = num_bytes;
4585
4586 ret = btrfs_search_slot(trans, extent_root, &key, path,
4587 -1, 1);
4588 if (ret) {
4589 printk(KERN_ERR "umm, got %d back from search"
4590 ", was looking for %llu\n", ret,
4591 (unsigned long long)bytenr);
4592 btrfs_print_leaf(extent_root, path->nodes[0]);
4593 }
4594 BUG_ON(ret);
4595 extent_slot = path->slots[0];
4596 leaf = path->nodes[0];
4597 item_size = btrfs_item_size_nr(leaf, extent_slot);
4598 }
4599 #endif
4600 BUG_ON(item_size < sizeof(*ei));
4601 ei = btrfs_item_ptr(leaf, extent_slot,
4602 struct btrfs_extent_item);
4603 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4604 struct btrfs_tree_block_info *bi;
4605 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4606 bi = (struct btrfs_tree_block_info *)(ei + 1);
4607 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4608 }
4609
4610 refs = btrfs_extent_refs(leaf, ei);
4611 BUG_ON(refs < refs_to_drop);
4612 refs -= refs_to_drop;
4613
4614 if (refs > 0) {
4615 if (extent_op)
4616 __run_delayed_extent_op(extent_op, leaf, ei);
4617 /*
4618 * In the case of inline back ref, reference count will
4619 * be updated by remove_extent_backref
4620 */
4621 if (iref) {
4622 BUG_ON(!found_extent);
4623 } else {
4624 btrfs_set_extent_refs(leaf, ei, refs);
4625 btrfs_mark_buffer_dirty(leaf);
4626 }
4627 if (found_extent) {
4628 ret = remove_extent_backref(trans, extent_root, path,
4629 iref, refs_to_drop,
4630 is_data);
4631 BUG_ON(ret);
4632 }
4633 } else {
4634 if (found_extent) {
4635 BUG_ON(is_data && refs_to_drop !=
4636 extent_data_ref_count(root, path, iref));
4637 if (iref) {
4638 BUG_ON(path->slots[0] != extent_slot);
4639 } else {
4640 BUG_ON(path->slots[0] != extent_slot + 1);
4641 path->slots[0] = extent_slot;
4642 num_to_del = 2;
4643 }
4644 }
4645
4646 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4647 num_to_del);
4648 BUG_ON(ret);
4649 btrfs_release_path(path);
4650
4651 if (is_data) {
4652 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4653 BUG_ON(ret);
4654 } else {
4655 invalidate_mapping_pages(info->btree_inode->i_mapping,
4656 bytenr >> PAGE_CACHE_SHIFT,
4657 (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4658 }
4659
4660 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4661 BUG_ON(ret);
4662 }
4663 btrfs_free_path(path);
4664 return ret;
4665 }
4666
4667 /*
4668 * when we free an block, it is possible (and likely) that we free the last
4669 * delayed ref for that extent as well. This searches the delayed ref tree for
4670 * a given extent, and if there are no other delayed refs to be processed, it
4671 * removes it from the tree.
4672 */
4673 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4674 struct btrfs_root *root, u64 bytenr)
4675 {
4676 struct btrfs_delayed_ref_head *head;
4677 struct btrfs_delayed_ref_root *delayed_refs;
4678 struct btrfs_delayed_ref_node *ref;
4679 struct rb_node *node;
4680 int ret = 0;
4681
4682 delayed_refs = &trans->transaction->delayed_refs;
4683 spin_lock(&delayed_refs->lock);
4684 head = btrfs_find_delayed_ref_head(trans, bytenr);
4685 if (!head)
4686 goto out;
4687
4688 node = rb_prev(&head->node.rb_node);
4689 if (!node)
4690 goto out;
4691
4692 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4693
4694 /* there are still entries for this ref, we can't drop it */
4695 if (ref->bytenr == bytenr)
4696 goto out;
4697
4698 if (head->extent_op) {
4699 if (!head->must_insert_reserved)
4700 goto out;
4701 kfree(head->extent_op);
4702 head->extent_op = NULL;
4703 }
4704
4705 /*
4706 * waiting for the lock here would deadlock. If someone else has it
4707 * locked they are already in the process of dropping it anyway
4708 */
4709 if (!mutex_trylock(&head->mutex))
4710 goto out;
4711
4712 /*
4713 * at this point we have a head with no other entries. Go
4714 * ahead and process it.
4715 */
4716 head->node.in_tree = 0;
4717 rb_erase(&head->node.rb_node, &delayed_refs->root);
4718
4719 delayed_refs->num_entries--;
4720
4721 /*
4722 * we don't take a ref on the node because we're removing it from the
4723 * tree, so we just steal the ref the tree was holding.
4724 */
4725 delayed_refs->num_heads--;
4726 if (list_empty(&head->cluster))
4727 delayed_refs->num_heads_ready--;
4728
4729 list_del_init(&head->cluster);
4730 spin_unlock(&delayed_refs->lock);
4731
4732 BUG_ON(head->extent_op);
4733 if (head->must_insert_reserved)
4734 ret = 1;
4735
4736 mutex_unlock(&head->mutex);
4737 btrfs_put_delayed_ref(&head->node);
4738 return ret;
4739 out:
4740 spin_unlock(&delayed_refs->lock);
4741 return 0;
4742 }
4743
4744 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4745 struct btrfs_root *root,
4746 struct extent_buffer *buf,
4747 u64 parent, int last_ref)
4748 {
4749 struct btrfs_block_group_cache *cache = NULL;
4750 int ret;
4751
4752 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4753 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4754 parent, root->root_key.objectid,
4755 btrfs_header_level(buf),
4756 BTRFS_DROP_DELAYED_REF, NULL);
4757 BUG_ON(ret);
4758 }
4759
4760 if (!last_ref)
4761 return;
4762
4763 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4764
4765 if (btrfs_header_generation(buf) == trans->transid) {
4766 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4767 ret = check_ref_cleanup(trans, root, buf->start);
4768 if (!ret)
4769 goto out;
4770 }
4771
4772 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4773 pin_down_extent(root, cache, buf->start, buf->len, 1);
4774 goto out;
4775 }
4776
4777 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4778
4779 btrfs_add_free_space(cache, buf->start, buf->len);
4780 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
4781 }
4782 out:
4783 /*
4784 * Deleting the buffer, clear the corrupt flag since it doesn't matter
4785 * anymore.
4786 */
4787 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
4788 btrfs_put_block_group(cache);
4789 }
4790
4791 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4792 struct btrfs_root *root,
4793 u64 bytenr, u64 num_bytes, u64 parent,
4794 u64 root_objectid, u64 owner, u64 offset)
4795 {
4796 int ret;
4797
4798 /*
4799 * tree log blocks never actually go into the extent allocation
4800 * tree, just update pinning info and exit early.
4801 */
4802 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4803 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4804 /* unlocks the pinned mutex */
4805 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4806 ret = 0;
4807 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4808 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4809 parent, root_objectid, (int)owner,
4810 BTRFS_DROP_DELAYED_REF, NULL);
4811 BUG_ON(ret);
4812 } else {
4813 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4814 parent, root_objectid, owner,
4815 offset, BTRFS_DROP_DELAYED_REF, NULL);
4816 BUG_ON(ret);
4817 }
4818 return ret;
4819 }
4820
4821 static u64 stripe_align(struct btrfs_root *root, u64 val)
4822 {
4823 u64 mask = ((u64)root->stripesize - 1);
4824 u64 ret = (val + mask) & ~mask;
4825 return ret;
4826 }
4827
4828 /*
4829 * when we wait for progress in the block group caching, its because
4830 * our allocation attempt failed at least once. So, we must sleep
4831 * and let some progress happen before we try again.
4832 *
4833 * This function will sleep at least once waiting for new free space to
4834 * show up, and then it will check the block group free space numbers
4835 * for our min num_bytes. Another option is to have it go ahead
4836 * and look in the rbtree for a free extent of a given size, but this
4837 * is a good start.
4838 */
4839 static noinline int
4840 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4841 u64 num_bytes)
4842 {
4843 struct btrfs_caching_control *caching_ctl;
4844 DEFINE_WAIT(wait);
4845
4846 caching_ctl = get_caching_control(cache);
4847 if (!caching_ctl)
4848 return 0;
4849
4850 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4851 (cache->free_space_ctl->free_space >= num_bytes));
4852
4853 put_caching_control(caching_ctl);
4854 return 0;
4855 }
4856
4857 static noinline int
4858 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4859 {
4860 struct btrfs_caching_control *caching_ctl;
4861 DEFINE_WAIT(wait);
4862
4863 caching_ctl = get_caching_control(cache);
4864 if (!caching_ctl)
4865 return 0;
4866
4867 wait_event(caching_ctl->wait, block_group_cache_done(cache));
4868
4869 put_caching_control(caching_ctl);
4870 return 0;
4871 }
4872
4873 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4874 {
4875 int index;
4876 if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4877 index = 0;
4878 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4879 index = 1;
4880 else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4881 index = 2;
4882 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4883 index = 3;
4884 else
4885 index = 4;
4886 return index;
4887 }
4888
4889 enum btrfs_loop_type {
4890 LOOP_FIND_IDEAL = 0,
4891 LOOP_CACHING_NOWAIT = 1,
4892 LOOP_CACHING_WAIT = 2,
4893 LOOP_ALLOC_CHUNK = 3,
4894 LOOP_NO_EMPTY_SIZE = 4,
4895 };
4896
4897 /*
4898 * walks the btree of allocated extents and find a hole of a given size.
4899 * The key ins is changed to record the hole:
4900 * ins->objectid == block start
4901 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4902 * ins->offset == number of blocks
4903 * Any available blocks before search_start are skipped.
4904 */
4905 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4906 struct btrfs_root *orig_root,
4907 u64 num_bytes, u64 empty_size,
4908 u64 search_start, u64 search_end,
4909 u64 hint_byte, struct btrfs_key *ins,
4910 u64 data)
4911 {
4912 int ret = 0;
4913 struct btrfs_root *root = orig_root->fs_info->extent_root;
4914 struct btrfs_free_cluster *last_ptr = NULL;
4915 struct btrfs_block_group_cache *block_group = NULL;
4916 int empty_cluster = 2 * 1024 * 1024;
4917 int allowed_chunk_alloc = 0;
4918 int done_chunk_alloc = 0;
4919 struct btrfs_space_info *space_info;
4920 int last_ptr_loop = 0;
4921 int loop = 0;
4922 int index = 0;
4923 int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
4924 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
4925 bool found_uncached_bg = false;
4926 bool failed_cluster_refill = false;
4927 bool failed_alloc = false;
4928 bool use_cluster = true;
4929 u64 ideal_cache_percent = 0;
4930 u64 ideal_cache_offset = 0;
4931
4932 WARN_ON(num_bytes < root->sectorsize);
4933 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4934 ins->objectid = 0;
4935 ins->offset = 0;
4936
4937 space_info = __find_space_info(root->fs_info, data);
4938 if (!space_info) {
4939 printk(KERN_ERR "No space info for %llu\n", data);
4940 return -ENOSPC;
4941 }
4942
4943 /*
4944 * If the space info is for both data and metadata it means we have a
4945 * small filesystem and we can't use the clustering stuff.
4946 */
4947 if (btrfs_mixed_space_info(space_info))
4948 use_cluster = false;
4949
4950 if (orig_root->ref_cows || empty_size)
4951 allowed_chunk_alloc = 1;
4952
4953 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
4954 last_ptr = &root->fs_info->meta_alloc_cluster;
4955 if (!btrfs_test_opt(root, SSD))
4956 empty_cluster = 64 * 1024;
4957 }
4958
4959 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
4960 btrfs_test_opt(root, SSD)) {
4961 last_ptr = &root->fs_info->data_alloc_cluster;
4962 }
4963
4964 if (last_ptr) {
4965 spin_lock(&last_ptr->lock);
4966 if (last_ptr->block_group)
4967 hint_byte = last_ptr->window_start;
4968 spin_unlock(&last_ptr->lock);
4969 }
4970
4971 search_start = max(search_start, first_logical_byte(root, 0));
4972 search_start = max(search_start, hint_byte);
4973
4974 if (!last_ptr)
4975 empty_cluster = 0;
4976
4977 if (search_start == hint_byte) {
4978 ideal_cache:
4979 block_group = btrfs_lookup_block_group(root->fs_info,
4980 search_start);
4981 /*
4982 * we don't want to use the block group if it doesn't match our
4983 * allocation bits, or if its not cached.
4984 *
4985 * However if we are re-searching with an ideal block group
4986 * picked out then we don't care that the block group is cached.
4987 */
4988 if (block_group && block_group_bits(block_group, data) &&
4989 (block_group->cached != BTRFS_CACHE_NO ||
4990 search_start == ideal_cache_offset)) {
4991 down_read(&space_info->groups_sem);
4992 if (list_empty(&block_group->list) ||
4993 block_group->ro) {
4994 /*
4995 * someone is removing this block group,
4996 * we can't jump into the have_block_group
4997 * target because our list pointers are not
4998 * valid
4999 */
5000 btrfs_put_block_group(block_group);
5001 up_read(&space_info->groups_sem);
5002 } else {
5003 index = get_block_group_index(block_group);
5004 goto have_block_group;
5005 }
5006 } else if (block_group) {
5007 btrfs_put_block_group(block_group);
5008 }
5009 }
5010 search:
5011 down_read(&space_info->groups_sem);
5012 list_for_each_entry(block_group, &space_info->block_groups[index],
5013 list) {
5014 u64 offset;
5015 int cached;
5016
5017 btrfs_get_block_group(block_group);
5018 search_start = block_group->key.objectid;
5019
5020 /*
5021 * this can happen if we end up cycling through all the
5022 * raid types, but we want to make sure we only allocate
5023 * for the proper type.
5024 */
5025 if (!block_group_bits(block_group, data)) {
5026 u64 extra = BTRFS_BLOCK_GROUP_DUP |
5027 BTRFS_BLOCK_GROUP_RAID1 |
5028 BTRFS_BLOCK_GROUP_RAID10;
5029
5030 /*
5031 * if they asked for extra copies and this block group
5032 * doesn't provide them, bail. This does allow us to
5033 * fill raid0 from raid1.
5034 */
5035 if ((data & extra) && !(block_group->flags & extra))
5036 goto loop;
5037 }
5038
5039 have_block_group:
5040 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
5041 u64 free_percent;
5042
5043 ret = cache_block_group(block_group, trans,
5044 orig_root, 1);
5045 if (block_group->cached == BTRFS_CACHE_FINISHED)
5046 goto have_block_group;
5047
5048 free_percent = btrfs_block_group_used(&block_group->item);
5049 free_percent *= 100;
5050 free_percent = div64_u64(free_percent,
5051 block_group->key.offset);
5052 free_percent = 100 - free_percent;
5053 if (free_percent > ideal_cache_percent &&
5054 likely(!block_group->ro)) {
5055 ideal_cache_offset = block_group->key.objectid;
5056 ideal_cache_percent = free_percent;
5057 }
5058
5059 /*
5060 * The caching workers are limited to 2 threads, so we
5061 * can queue as much work as we care to.
5062 */
5063 if (loop > LOOP_FIND_IDEAL) {
5064 ret = cache_block_group(block_group, trans,
5065 orig_root, 0);
5066 BUG_ON(ret);
5067 }
5068 found_uncached_bg = true;
5069
5070 /*
5071 * If loop is set for cached only, try the next block
5072 * group.
5073 */
5074 if (loop == LOOP_FIND_IDEAL)
5075 goto loop;
5076 }
5077
5078 cached = block_group_cache_done(block_group);
5079 if (unlikely(!cached))
5080 found_uncached_bg = true;
5081
5082 if (unlikely(block_group->ro))
5083 goto loop;
5084
5085 spin_lock(&block_group->free_space_ctl->tree_lock);
5086 if (cached &&
5087 block_group->free_space_ctl->free_space <
5088 num_bytes + empty_size) {
5089 spin_unlock(&block_group->free_space_ctl->tree_lock);
5090 goto loop;
5091 }
5092 spin_unlock(&block_group->free_space_ctl->tree_lock);
5093
5094 /*
5095 * Ok we want to try and use the cluster allocator, so lets look
5096 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5097 * have tried the cluster allocator plenty of times at this
5098 * point and not have found anything, so we are likely way too
5099 * fragmented for the clustering stuff to find anything, so lets
5100 * just skip it and let the allocator find whatever block it can
5101 * find
5102 */
5103 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
5104 /*
5105 * the refill lock keeps out other
5106 * people trying to start a new cluster
5107 */
5108 spin_lock(&last_ptr->refill_lock);
5109 if (last_ptr->block_group &&
5110 (last_ptr->block_group->ro ||
5111 !block_group_bits(last_ptr->block_group, data))) {
5112 offset = 0;
5113 goto refill_cluster;
5114 }
5115
5116 offset = btrfs_alloc_from_cluster(block_group, last_ptr,
5117 num_bytes, search_start);
5118 if (offset) {
5119 /* we have a block, we're done */
5120 spin_unlock(&last_ptr->refill_lock);
5121 goto checks;
5122 }
5123
5124 spin_lock(&last_ptr->lock);
5125 /*
5126 * whoops, this cluster doesn't actually point to
5127 * this block group. Get a ref on the block
5128 * group is does point to and try again
5129 */
5130 if (!last_ptr_loop && last_ptr->block_group &&
5131 last_ptr->block_group != block_group &&
5132 index <=
5133 get_block_group_index(last_ptr->block_group)) {
5134
5135 btrfs_put_block_group(block_group);
5136 block_group = last_ptr->block_group;
5137 btrfs_get_block_group(block_group);
5138 spin_unlock(&last_ptr->lock);
5139 spin_unlock(&last_ptr->refill_lock);
5140
5141 last_ptr_loop = 1;
5142 search_start = block_group->key.objectid;
5143 /*
5144 * we know this block group is properly
5145 * in the list because
5146 * btrfs_remove_block_group, drops the
5147 * cluster before it removes the block
5148 * group from the list
5149 */
5150 goto have_block_group;
5151 }
5152 spin_unlock(&last_ptr->lock);
5153 refill_cluster:
5154 /*
5155 * this cluster didn't work out, free it and
5156 * start over
5157 */
5158 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5159
5160 last_ptr_loop = 0;
5161
5162 /* allocate a cluster in this block group */
5163 ret = btrfs_find_space_cluster(trans, root,
5164 block_group, last_ptr,
5165 offset, num_bytes,
5166 empty_cluster + empty_size);
5167 if (ret == 0) {
5168 /*
5169 * now pull our allocation out of this
5170 * cluster
5171 */
5172 offset = btrfs_alloc_from_cluster(block_group,
5173 last_ptr, num_bytes,
5174 search_start);
5175 if (offset) {
5176 /* we found one, proceed */
5177 spin_unlock(&last_ptr->refill_lock);
5178 goto checks;
5179 }
5180 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5181 && !failed_cluster_refill) {
5182 spin_unlock(&last_ptr->refill_lock);
5183
5184 failed_cluster_refill = true;
5185 wait_block_group_cache_progress(block_group,
5186 num_bytes + empty_cluster + empty_size);
5187 goto have_block_group;
5188 }
5189
5190 /*
5191 * at this point we either didn't find a cluster
5192 * or we weren't able to allocate a block from our
5193 * cluster. Free the cluster we've been trying
5194 * to use, and go to the next block group
5195 */
5196 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5197 spin_unlock(&last_ptr->refill_lock);
5198 goto loop;
5199 }
5200
5201 offset = btrfs_find_space_for_alloc(block_group, search_start,
5202 num_bytes, empty_size);
5203 /*
5204 * If we didn't find a chunk, and we haven't failed on this
5205 * block group before, and this block group is in the middle of
5206 * caching and we are ok with waiting, then go ahead and wait
5207 * for progress to be made, and set failed_alloc to true.
5208 *
5209 * If failed_alloc is true then we've already waited on this
5210 * block group once and should move on to the next block group.
5211 */
5212 if (!offset && !failed_alloc && !cached &&
5213 loop > LOOP_CACHING_NOWAIT) {
5214 wait_block_group_cache_progress(block_group,
5215 num_bytes + empty_size);
5216 failed_alloc = true;
5217 goto have_block_group;
5218 } else if (!offset) {
5219 goto loop;
5220 }
5221 checks:
5222 search_start = stripe_align(root, offset);
5223 /* move on to the next group */
5224 if (search_start + num_bytes >= search_end) {
5225 btrfs_add_free_space(block_group, offset, num_bytes);
5226 goto loop;
5227 }
5228
5229 /* move on to the next group */
5230 if (search_start + num_bytes >
5231 block_group->key.objectid + block_group->key.offset) {
5232 btrfs_add_free_space(block_group, offset, num_bytes);
5233 goto loop;
5234 }
5235
5236 ins->objectid = search_start;
5237 ins->offset = num_bytes;
5238
5239 if (offset < search_start)
5240 btrfs_add_free_space(block_group, offset,
5241 search_start - offset);
5242 BUG_ON(offset > search_start);
5243
5244 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
5245 alloc_type);
5246 if (ret == -EAGAIN) {
5247 btrfs_add_free_space(block_group, offset, num_bytes);
5248 goto loop;
5249 }
5250
5251 /* we are all good, lets return */
5252 ins->objectid = search_start;
5253 ins->offset = num_bytes;
5254
5255 if (offset < search_start)
5256 btrfs_add_free_space(block_group, offset,
5257 search_start - offset);
5258 BUG_ON(offset > search_start);
5259 btrfs_put_block_group(block_group);
5260 break;
5261 loop:
5262 failed_cluster_refill = false;
5263 failed_alloc = false;
5264 BUG_ON(index != get_block_group_index(block_group));
5265 btrfs_put_block_group(block_group);
5266 }
5267 up_read(&space_info->groups_sem);
5268
5269 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5270 goto search;
5271
5272 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5273 * for them to make caching progress. Also
5274 * determine the best possible bg to cache
5275 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5276 * caching kthreads as we move along
5277 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5278 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5279 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5280 * again
5281 */
5282 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5283 index = 0;
5284 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5285 found_uncached_bg = false;
5286 loop++;
5287 if (!ideal_cache_percent)
5288 goto search;
5289
5290 /*
5291 * 1 of the following 2 things have happened so far
5292 *
5293 * 1) We found an ideal block group for caching that
5294 * is mostly full and will cache quickly, so we might
5295 * as well wait for it.
5296 *
5297 * 2) We searched for cached only and we didn't find
5298 * anything, and we didn't start any caching kthreads
5299 * either, so chances are we will loop through and
5300 * start a couple caching kthreads, and then come back
5301 * around and just wait for them. This will be slower
5302 * because we will have 2 caching kthreads reading at
5303 * the same time when we could have just started one
5304 * and waited for it to get far enough to give us an
5305 * allocation, so go ahead and go to the wait caching
5306 * loop.
5307 */
5308 loop = LOOP_CACHING_WAIT;
5309 search_start = ideal_cache_offset;
5310 ideal_cache_percent = 0;
5311 goto ideal_cache;
5312 } else if (loop == LOOP_FIND_IDEAL) {
5313 /*
5314 * Didn't find a uncached bg, wait on anything we find
5315 * next.
5316 */
5317 loop = LOOP_CACHING_WAIT;
5318 goto search;
5319 }
5320
5321 loop++;
5322
5323 if (loop == LOOP_ALLOC_CHUNK) {
5324 if (allowed_chunk_alloc) {
5325 ret = do_chunk_alloc(trans, root, num_bytes +
5326 2 * 1024 * 1024, data,
5327 CHUNK_ALLOC_LIMITED);
5328 allowed_chunk_alloc = 0;
5329 if (ret == 1)
5330 done_chunk_alloc = 1;
5331 } else if (!done_chunk_alloc &&
5332 space_info->force_alloc ==
5333 CHUNK_ALLOC_NO_FORCE) {
5334 space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5335 }
5336
5337 /*
5338 * We didn't allocate a chunk, go ahead and drop the
5339 * empty size and loop again.
5340 */
5341 if (!done_chunk_alloc)
5342 loop = LOOP_NO_EMPTY_SIZE;
5343 }
5344
5345 if (loop == LOOP_NO_EMPTY_SIZE) {
5346 empty_size = 0;
5347 empty_cluster = 0;
5348 }
5349
5350 goto search;
5351 } else if (!ins->objectid) {
5352 ret = -ENOSPC;
5353 } else if (ins->objectid) {
5354 ret = 0;
5355 }
5356
5357 return ret;
5358 }
5359
5360 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5361 int dump_block_groups)
5362 {
5363 struct btrfs_block_group_cache *cache;
5364 int index = 0;
5365
5366 spin_lock(&info->lock);
5367 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5368 (unsigned long long)info->flags,
5369 (unsigned long long)(info->total_bytes - info->bytes_used -
5370 info->bytes_pinned - info->bytes_reserved -
5371 info->bytes_readonly),
5372 (info->full) ? "" : "not ");
5373 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5374 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5375 (unsigned long long)info->total_bytes,
5376 (unsigned long long)info->bytes_used,
5377 (unsigned long long)info->bytes_pinned,
5378 (unsigned long long)info->bytes_reserved,
5379 (unsigned long long)info->bytes_may_use,
5380 (unsigned long long)info->bytes_readonly);
5381 spin_unlock(&info->lock);
5382
5383 if (!dump_block_groups)
5384 return;
5385
5386 down_read(&info->groups_sem);
5387 again:
5388 list_for_each_entry(cache, &info->block_groups[index], list) {
5389 spin_lock(&cache->lock);
5390 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5391 "%llu pinned %llu reserved\n",
5392 (unsigned long long)cache->key.objectid,
5393 (unsigned long long)cache->key.offset,
5394 (unsigned long long)btrfs_block_group_used(&cache->item),
5395 (unsigned long long)cache->pinned,
5396 (unsigned long long)cache->reserved);
5397 btrfs_dump_free_space(cache, bytes);
5398 spin_unlock(&cache->lock);
5399 }
5400 if (++index < BTRFS_NR_RAID_TYPES)
5401 goto again;
5402 up_read(&info->groups_sem);
5403 }
5404
5405 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5406 struct btrfs_root *root,
5407 u64 num_bytes, u64 min_alloc_size,
5408 u64 empty_size, u64 hint_byte,
5409 u64 search_end, struct btrfs_key *ins,
5410 u64 data)
5411 {
5412 int ret;
5413 u64 search_start = 0;
5414
5415 data = btrfs_get_alloc_profile(root, data);
5416 again:
5417 /*
5418 * the only place that sets empty_size is btrfs_realloc_node, which
5419 * is not called recursively on allocations
5420 */
5421 if (empty_size || root->ref_cows)
5422 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5423 num_bytes + 2 * 1024 * 1024, data,
5424 CHUNK_ALLOC_NO_FORCE);
5425
5426 WARN_ON(num_bytes < root->sectorsize);
5427 ret = find_free_extent(trans, root, num_bytes, empty_size,
5428 search_start, search_end, hint_byte,
5429 ins, data);
5430
5431 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5432 num_bytes = num_bytes >> 1;
5433 num_bytes = num_bytes & ~(root->sectorsize - 1);
5434 num_bytes = max(num_bytes, min_alloc_size);
5435 do_chunk_alloc(trans, root->fs_info->extent_root,
5436 num_bytes, data, CHUNK_ALLOC_FORCE);
5437 goto again;
5438 }
5439 if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) {
5440 struct btrfs_space_info *sinfo;
5441
5442 sinfo = __find_space_info(root->fs_info, data);
5443 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5444 "wanted %llu\n", (unsigned long long)data,
5445 (unsigned long long)num_bytes);
5446 dump_space_info(sinfo, num_bytes, 1);
5447 }
5448
5449 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5450
5451 return ret;
5452 }
5453
5454 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
5455 {
5456 struct btrfs_block_group_cache *cache;
5457 int ret = 0;
5458
5459 cache = btrfs_lookup_block_group(root->fs_info, start);
5460 if (!cache) {
5461 printk(KERN_ERR "Unable to find block group for %llu\n",
5462 (unsigned long long)start);
5463 return -ENOSPC;
5464 }
5465
5466 if (btrfs_test_opt(root, DISCARD))
5467 ret = btrfs_discard_extent(root, start, len, NULL);
5468
5469 btrfs_add_free_space(cache, start, len);
5470 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
5471 btrfs_put_block_group(cache);
5472
5473 trace_btrfs_reserved_extent_free(root, start, len);
5474
5475 return ret;
5476 }
5477
5478 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5479 struct btrfs_root *root,
5480 u64 parent, u64 root_objectid,
5481 u64 flags, u64 owner, u64 offset,
5482 struct btrfs_key *ins, int ref_mod)
5483 {
5484 int ret;
5485 struct btrfs_fs_info *fs_info = root->fs_info;
5486 struct btrfs_extent_item *extent_item;
5487 struct btrfs_extent_inline_ref *iref;
5488 struct btrfs_path *path;
5489 struct extent_buffer *leaf;
5490 int type;
5491 u32 size;
5492
5493 if (parent > 0)
5494 type = BTRFS_SHARED_DATA_REF_KEY;
5495 else
5496 type = BTRFS_EXTENT_DATA_REF_KEY;
5497
5498 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5499
5500 path = btrfs_alloc_path();
5501 if (!path)
5502 return -ENOMEM;
5503
5504 path->leave_spinning = 1;
5505 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5506 ins, size);
5507 BUG_ON(ret);
5508
5509 leaf = path->nodes[0];
5510 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5511 struct btrfs_extent_item);
5512 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5513 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5514 btrfs_set_extent_flags(leaf, extent_item,
5515 flags | BTRFS_EXTENT_FLAG_DATA);
5516
5517 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5518 btrfs_set_extent_inline_ref_type(leaf, iref, type);
5519 if (parent > 0) {
5520 struct btrfs_shared_data_ref *ref;
5521 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5522 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5523 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5524 } else {
5525 struct btrfs_extent_data_ref *ref;
5526 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5527 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5528 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5529 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5530 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5531 }
5532
5533 btrfs_mark_buffer_dirty(path->nodes[0]);
5534 btrfs_free_path(path);
5535
5536 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5537 if (ret) {
5538 printk(KERN_ERR "btrfs update block group failed for %llu "
5539 "%llu\n", (unsigned long long)ins->objectid,
5540 (unsigned long long)ins->offset);
5541 BUG();
5542 }
5543 return ret;
5544 }
5545
5546 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5547 struct btrfs_root *root,
5548 u64 parent, u64 root_objectid,
5549 u64 flags, struct btrfs_disk_key *key,
5550 int level, struct btrfs_key *ins)
5551 {
5552 int ret;
5553 struct btrfs_fs_info *fs_info = root->fs_info;
5554 struct btrfs_extent_item *extent_item;
5555 struct btrfs_tree_block_info *block_info;
5556 struct btrfs_extent_inline_ref *iref;
5557 struct btrfs_path *path;
5558 struct extent_buffer *leaf;
5559 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5560
5561 path = btrfs_alloc_path();
5562 if (!path)
5563 return -ENOMEM;
5564
5565 path->leave_spinning = 1;
5566 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5567 ins, size);
5568 BUG_ON(ret);
5569
5570 leaf = path->nodes[0];
5571 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5572 struct btrfs_extent_item);
5573 btrfs_set_extent_refs(leaf, extent_item, 1);
5574 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5575 btrfs_set_extent_flags(leaf, extent_item,
5576 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5577 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5578
5579 btrfs_set_tree_block_key(leaf, block_info, key);
5580 btrfs_set_tree_block_level(leaf, block_info, level);
5581
5582 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5583 if (parent > 0) {
5584 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5585 btrfs_set_extent_inline_ref_type(leaf, iref,
5586 BTRFS_SHARED_BLOCK_REF_KEY);
5587 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5588 } else {
5589 btrfs_set_extent_inline_ref_type(leaf, iref,
5590 BTRFS_TREE_BLOCK_REF_KEY);
5591 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5592 }
5593
5594 btrfs_mark_buffer_dirty(leaf);
5595 btrfs_free_path(path);
5596
5597 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5598 if (ret) {
5599 printk(KERN_ERR "btrfs update block group failed for %llu "
5600 "%llu\n", (unsigned long long)ins->objectid,
5601 (unsigned long long)ins->offset);
5602 BUG();
5603 }
5604 return ret;
5605 }
5606
5607 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5608 struct btrfs_root *root,
5609 u64 root_objectid, u64 owner,
5610 u64 offset, struct btrfs_key *ins)
5611 {
5612 int ret;
5613
5614 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5615
5616 ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5617 0, root_objectid, owner, offset,
5618 BTRFS_ADD_DELAYED_EXTENT, NULL);
5619 return ret;
5620 }
5621
5622 /*
5623 * this is used by the tree logging recovery code. It records that
5624 * an extent has been allocated and makes sure to clear the free
5625 * space cache bits as well
5626 */
5627 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5628 struct btrfs_root *root,
5629 u64 root_objectid, u64 owner, u64 offset,
5630 struct btrfs_key *ins)
5631 {
5632 int ret;
5633 struct btrfs_block_group_cache *block_group;
5634 struct btrfs_caching_control *caching_ctl;
5635 u64 start = ins->objectid;
5636 u64 num_bytes = ins->offset;
5637
5638 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5639 cache_block_group(block_group, trans, NULL, 0);
5640 caching_ctl = get_caching_control(block_group);
5641
5642 if (!caching_ctl) {
5643 BUG_ON(!block_group_cache_done(block_group));
5644 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5645 BUG_ON(ret);
5646 } else {
5647 mutex_lock(&caching_ctl->mutex);
5648
5649 if (start >= caching_ctl->progress) {
5650 ret = add_excluded_extent(root, start, num_bytes);
5651 BUG_ON(ret);
5652 } else if (start + num_bytes <= caching_ctl->progress) {
5653 ret = btrfs_remove_free_space(block_group,
5654 start, num_bytes);
5655 BUG_ON(ret);
5656 } else {
5657 num_bytes = caching_ctl->progress - start;
5658 ret = btrfs_remove_free_space(block_group,
5659 start, num_bytes);
5660 BUG_ON(ret);
5661
5662 start = caching_ctl->progress;
5663 num_bytes = ins->objectid + ins->offset -
5664 caching_ctl->progress;
5665 ret = add_excluded_extent(root, start, num_bytes);
5666 BUG_ON(ret);
5667 }
5668
5669 mutex_unlock(&caching_ctl->mutex);
5670 put_caching_control(caching_ctl);
5671 }
5672
5673 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
5674 RESERVE_ALLOC_NO_ACCOUNT);
5675 BUG_ON(ret);
5676 btrfs_put_block_group(block_group);
5677 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5678 0, owner, offset, ins, 1);
5679 return ret;
5680 }
5681
5682 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5683 struct btrfs_root *root,
5684 u64 bytenr, u32 blocksize,
5685 int level)
5686 {
5687 struct extent_buffer *buf;
5688
5689 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5690 if (!buf)
5691 return ERR_PTR(-ENOMEM);
5692 btrfs_set_header_generation(buf, trans->transid);
5693 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
5694 btrfs_tree_lock(buf);
5695 clean_tree_block(trans, root, buf);
5696
5697 btrfs_set_lock_blocking(buf);
5698 btrfs_set_buffer_uptodate(buf);
5699
5700 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5701 /*
5702 * we allow two log transactions at a time, use different
5703 * EXENT bit to differentiate dirty pages.
5704 */
5705 if (root->log_transid % 2 == 0)
5706 set_extent_dirty(&root->dirty_log_pages, buf->start,
5707 buf->start + buf->len - 1, GFP_NOFS);
5708 else
5709 set_extent_new(&root->dirty_log_pages, buf->start,
5710 buf->start + buf->len - 1, GFP_NOFS);
5711 } else {
5712 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5713 buf->start + buf->len - 1, GFP_NOFS);
5714 }
5715 trans->blocks_used++;
5716 /* this returns a buffer locked for blocking */
5717 return buf;
5718 }
5719
5720 static struct btrfs_block_rsv *
5721 use_block_rsv(struct btrfs_trans_handle *trans,
5722 struct btrfs_root *root, u32 blocksize)
5723 {
5724 struct btrfs_block_rsv *block_rsv;
5725 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5726 int ret;
5727
5728 block_rsv = get_block_rsv(trans, root);
5729
5730 if (block_rsv->size == 0) {
5731 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0, 0);
5732 /*
5733 * If we couldn't reserve metadata bytes try and use some from
5734 * the global reserve.
5735 */
5736 if (ret && block_rsv != global_rsv) {
5737 ret = block_rsv_use_bytes(global_rsv, blocksize);
5738 if (!ret)
5739 return global_rsv;
5740 return ERR_PTR(ret);
5741 } else if (ret) {
5742 return ERR_PTR(ret);
5743 }
5744 return block_rsv;
5745 }
5746
5747 ret = block_rsv_use_bytes(block_rsv, blocksize);
5748 if (!ret)
5749 return block_rsv;
5750 if (ret) {
5751 WARN_ON(1);
5752 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0, 0);
5753 if (!ret) {
5754 return block_rsv;
5755 } else if (ret && block_rsv != global_rsv) {
5756 ret = block_rsv_use_bytes(global_rsv, blocksize);
5757 if (!ret)
5758 return global_rsv;
5759 }
5760 }
5761
5762 return ERR_PTR(-ENOSPC);
5763 }
5764
5765 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5766 {
5767 block_rsv_add_bytes(block_rsv, blocksize, 0);
5768 block_rsv_release_bytes(block_rsv, NULL, 0);
5769 }
5770
5771 /*
5772 * finds a free extent and does all the dirty work required for allocation
5773 * returns the key for the extent through ins, and a tree buffer for
5774 * the first block of the extent through buf.
5775 *
5776 * returns the tree buffer or NULL.
5777 */
5778 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5779 struct btrfs_root *root, u32 blocksize,
5780 u64 parent, u64 root_objectid,
5781 struct btrfs_disk_key *key, int level,
5782 u64 hint, u64 empty_size)
5783 {
5784 struct btrfs_key ins;
5785 struct btrfs_block_rsv *block_rsv;
5786 struct extent_buffer *buf;
5787 u64 flags = 0;
5788 int ret;
5789
5790
5791 block_rsv = use_block_rsv(trans, root, blocksize);
5792 if (IS_ERR(block_rsv))
5793 return ERR_CAST(block_rsv);
5794
5795 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5796 empty_size, hint, (u64)-1, &ins, 0);
5797 if (ret) {
5798 unuse_block_rsv(block_rsv, blocksize);
5799 return ERR_PTR(ret);
5800 }
5801
5802 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5803 blocksize, level);
5804 BUG_ON(IS_ERR(buf));
5805
5806 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5807 if (parent == 0)
5808 parent = ins.objectid;
5809 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5810 } else
5811 BUG_ON(parent > 0);
5812
5813 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5814 struct btrfs_delayed_extent_op *extent_op;
5815 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5816 BUG_ON(!extent_op);
5817 if (key)
5818 memcpy(&extent_op->key, key, sizeof(extent_op->key));
5819 else
5820 memset(&extent_op->key, 0, sizeof(extent_op->key));
5821 extent_op->flags_to_set = flags;
5822 extent_op->update_key = 1;
5823 extent_op->update_flags = 1;
5824 extent_op->is_data = 0;
5825
5826 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5827 ins.offset, parent, root_objectid,
5828 level, BTRFS_ADD_DELAYED_EXTENT,
5829 extent_op);
5830 BUG_ON(ret);
5831 }
5832 return buf;
5833 }
5834
5835 struct walk_control {
5836 u64 refs[BTRFS_MAX_LEVEL];
5837 u64 flags[BTRFS_MAX_LEVEL];
5838 struct btrfs_key update_progress;
5839 int stage;
5840 int level;
5841 int shared_level;
5842 int update_ref;
5843 int keep_locks;
5844 int reada_slot;
5845 int reada_count;
5846 };
5847
5848 #define DROP_REFERENCE 1
5849 #define UPDATE_BACKREF 2
5850
5851 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5852 struct btrfs_root *root,
5853 struct walk_control *wc,
5854 struct btrfs_path *path)
5855 {
5856 u64 bytenr;
5857 u64 generation;
5858 u64 refs;
5859 u64 flags;
5860 u32 nritems;
5861 u32 blocksize;
5862 struct btrfs_key key;
5863 struct extent_buffer *eb;
5864 int ret;
5865 int slot;
5866 int nread = 0;
5867
5868 if (path->slots[wc->level] < wc->reada_slot) {
5869 wc->reada_count = wc->reada_count * 2 / 3;
5870 wc->reada_count = max(wc->reada_count, 2);
5871 } else {
5872 wc->reada_count = wc->reada_count * 3 / 2;
5873 wc->reada_count = min_t(int, wc->reada_count,
5874 BTRFS_NODEPTRS_PER_BLOCK(root));
5875 }
5876
5877 eb = path->nodes[wc->level];
5878 nritems = btrfs_header_nritems(eb);
5879 blocksize = btrfs_level_size(root, wc->level - 1);
5880
5881 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5882 if (nread >= wc->reada_count)
5883 break;
5884
5885 cond_resched();
5886 bytenr = btrfs_node_blockptr(eb, slot);
5887 generation = btrfs_node_ptr_generation(eb, slot);
5888
5889 if (slot == path->slots[wc->level])
5890 goto reada;
5891
5892 if (wc->stage == UPDATE_BACKREF &&
5893 generation <= root->root_key.offset)
5894 continue;
5895
5896 /* We don't lock the tree block, it's OK to be racy here */
5897 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5898 &refs, &flags);
5899 BUG_ON(ret);
5900 BUG_ON(refs == 0);
5901
5902 if (wc->stage == DROP_REFERENCE) {
5903 if (refs == 1)
5904 goto reada;
5905
5906 if (wc->level == 1 &&
5907 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5908 continue;
5909 if (!wc->update_ref ||
5910 generation <= root->root_key.offset)
5911 continue;
5912 btrfs_node_key_to_cpu(eb, &key, slot);
5913 ret = btrfs_comp_cpu_keys(&key,
5914 &wc->update_progress);
5915 if (ret < 0)
5916 continue;
5917 } else {
5918 if (wc->level == 1 &&
5919 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5920 continue;
5921 }
5922 reada:
5923 ret = readahead_tree_block(root, bytenr, blocksize,
5924 generation);
5925 if (ret)
5926 break;
5927 nread++;
5928 }
5929 wc->reada_slot = slot;
5930 }
5931
5932 /*
5933 * hepler to process tree block while walking down the tree.
5934 *
5935 * when wc->stage == UPDATE_BACKREF, this function updates
5936 * back refs for pointers in the block.
5937 *
5938 * NOTE: return value 1 means we should stop walking down.
5939 */
5940 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5941 struct btrfs_root *root,
5942 struct btrfs_path *path,
5943 struct walk_control *wc, int lookup_info)
5944 {
5945 int level = wc->level;
5946 struct extent_buffer *eb = path->nodes[level];
5947 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5948 int ret;
5949
5950 if (wc->stage == UPDATE_BACKREF &&
5951 btrfs_header_owner(eb) != root->root_key.objectid)
5952 return 1;
5953
5954 /*
5955 * when reference count of tree block is 1, it won't increase
5956 * again. once full backref flag is set, we never clear it.
5957 */
5958 if (lookup_info &&
5959 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5960 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5961 BUG_ON(!path->locks[level]);
5962 ret = btrfs_lookup_extent_info(trans, root,
5963 eb->start, eb->len,
5964 &wc->refs[level],
5965 &wc->flags[level]);
5966 BUG_ON(ret);
5967 BUG_ON(wc->refs[level] == 0);
5968 }
5969
5970 if (wc->stage == DROP_REFERENCE) {
5971 if (wc->refs[level] > 1)
5972 return 1;
5973
5974 if (path->locks[level] && !wc->keep_locks) {
5975 btrfs_tree_unlock_rw(eb, path->locks[level]);
5976 path->locks[level] = 0;
5977 }
5978 return 0;
5979 }
5980
5981 /* wc->stage == UPDATE_BACKREF */
5982 if (!(wc->flags[level] & flag)) {
5983 BUG_ON(!path->locks[level]);
5984 ret = btrfs_inc_ref(trans, root, eb, 1);
5985 BUG_ON(ret);
5986 ret = btrfs_dec_ref(trans, root, eb, 0);
5987 BUG_ON(ret);
5988 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5989 eb->len, flag, 0);
5990 BUG_ON(ret);
5991 wc->flags[level] |= flag;
5992 }
5993
5994 /*
5995 * the block is shared by multiple trees, so it's not good to
5996 * keep the tree lock
5997 */
5998 if (path->locks[level] && level > 0) {
5999 btrfs_tree_unlock_rw(eb, path->locks[level]);
6000 path->locks[level] = 0;
6001 }
6002 return 0;
6003 }
6004
6005 /*
6006 * hepler to process tree block pointer.
6007 *
6008 * when wc->stage == DROP_REFERENCE, this function checks
6009 * reference count of the block pointed to. if the block
6010 * is shared and we need update back refs for the subtree
6011 * rooted at the block, this function changes wc->stage to
6012 * UPDATE_BACKREF. if the block is shared and there is no
6013 * need to update back, this function drops the reference
6014 * to the block.
6015 *
6016 * NOTE: return value 1 means we should stop walking down.
6017 */
6018 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6019 struct btrfs_root *root,
6020 struct btrfs_path *path,
6021 struct walk_control *wc, int *lookup_info)
6022 {
6023 u64 bytenr;
6024 u64 generation;
6025 u64 parent;
6026 u32 blocksize;
6027 struct btrfs_key key;
6028 struct extent_buffer *next;
6029 int level = wc->level;
6030 int reada = 0;
6031 int ret = 0;
6032
6033 generation = btrfs_node_ptr_generation(path->nodes[level],
6034 path->slots[level]);
6035 /*
6036 * if the lower level block was created before the snapshot
6037 * was created, we know there is no need to update back refs
6038 * for the subtree
6039 */
6040 if (wc->stage == UPDATE_BACKREF &&
6041 generation <= root->root_key.offset) {
6042 *lookup_info = 1;
6043 return 1;
6044 }
6045
6046 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6047 blocksize = btrfs_level_size(root, level - 1);
6048
6049 next = btrfs_find_tree_block(root, bytenr, blocksize);
6050 if (!next) {
6051 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6052 if (!next)
6053 return -ENOMEM;
6054 reada = 1;
6055 }
6056 btrfs_tree_lock(next);
6057 btrfs_set_lock_blocking(next);
6058
6059 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6060 &wc->refs[level - 1],
6061 &wc->flags[level - 1]);
6062 BUG_ON(ret);
6063 BUG_ON(wc->refs[level - 1] == 0);
6064 *lookup_info = 0;
6065
6066 if (wc->stage == DROP_REFERENCE) {
6067 if (wc->refs[level - 1] > 1) {
6068 if (level == 1 &&
6069 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6070 goto skip;
6071
6072 if (!wc->update_ref ||
6073 generation <= root->root_key.offset)
6074 goto skip;
6075
6076 btrfs_node_key_to_cpu(path->nodes[level], &key,
6077 path->slots[level]);
6078 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6079 if (ret < 0)
6080 goto skip;
6081
6082 wc->stage = UPDATE_BACKREF;
6083 wc->shared_level = level - 1;
6084 }
6085 } else {
6086 if (level == 1 &&
6087 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6088 goto skip;
6089 }
6090
6091 if (!btrfs_buffer_uptodate(next, generation)) {
6092 btrfs_tree_unlock(next);
6093 free_extent_buffer(next);
6094 next = NULL;
6095 *lookup_info = 1;
6096 }
6097
6098 if (!next) {
6099 if (reada && level == 1)
6100 reada_walk_down(trans, root, wc, path);
6101 next = read_tree_block(root, bytenr, blocksize, generation);
6102 if (!next)
6103 return -EIO;
6104 btrfs_tree_lock(next);
6105 btrfs_set_lock_blocking(next);
6106 }
6107
6108 level--;
6109 BUG_ON(level != btrfs_header_level(next));
6110 path->nodes[level] = next;
6111 path->slots[level] = 0;
6112 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6113 wc->level = level;
6114 if (wc->level == 1)
6115 wc->reada_slot = 0;
6116 return 0;
6117 skip:
6118 wc->refs[level - 1] = 0;
6119 wc->flags[level - 1] = 0;
6120 if (wc->stage == DROP_REFERENCE) {
6121 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6122 parent = path->nodes[level]->start;
6123 } else {
6124 BUG_ON(root->root_key.objectid !=
6125 btrfs_header_owner(path->nodes[level]));
6126 parent = 0;
6127 }
6128
6129 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6130 root->root_key.objectid, level - 1, 0);
6131 BUG_ON(ret);
6132 }
6133 btrfs_tree_unlock(next);
6134 free_extent_buffer(next);
6135 *lookup_info = 1;
6136 return 1;
6137 }
6138
6139 /*
6140 * hepler to process tree block while walking up the tree.
6141 *
6142 * when wc->stage == DROP_REFERENCE, this function drops
6143 * reference count on the block.
6144 *
6145 * when wc->stage == UPDATE_BACKREF, this function changes
6146 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6147 * to UPDATE_BACKREF previously while processing the block.
6148 *
6149 * NOTE: return value 1 means we should stop walking up.
6150 */
6151 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6152 struct btrfs_root *root,
6153 struct btrfs_path *path,
6154 struct walk_control *wc)
6155 {
6156 int ret;
6157 int level = wc->level;
6158 struct extent_buffer *eb = path->nodes[level];
6159 u64 parent = 0;
6160
6161 if (wc->stage == UPDATE_BACKREF) {
6162 BUG_ON(wc->shared_level < level);
6163 if (level < wc->shared_level)
6164 goto out;
6165
6166 ret = find_next_key(path, level + 1, &wc->update_progress);
6167 if (ret > 0)
6168 wc->update_ref = 0;
6169
6170 wc->stage = DROP_REFERENCE;
6171 wc->shared_level = -1;
6172 path->slots[level] = 0;
6173
6174 /*
6175 * check reference count again if the block isn't locked.
6176 * we should start walking down the tree again if reference
6177 * count is one.
6178 */
6179 if (!path->locks[level]) {
6180 BUG_ON(level == 0);
6181 btrfs_tree_lock(eb);
6182 btrfs_set_lock_blocking(eb);
6183 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6184
6185 ret = btrfs_lookup_extent_info(trans, root,
6186 eb->start, eb->len,
6187 &wc->refs[level],
6188 &wc->flags[level]);
6189 BUG_ON(ret);
6190 BUG_ON(wc->refs[level] == 0);
6191 if (wc->refs[level] == 1) {
6192 btrfs_tree_unlock_rw(eb, path->locks[level]);
6193 return 1;
6194 }
6195 }
6196 }
6197
6198 /* wc->stage == DROP_REFERENCE */
6199 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6200
6201 if (wc->refs[level] == 1) {
6202 if (level == 0) {
6203 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6204 ret = btrfs_dec_ref(trans, root, eb, 1);
6205 else
6206 ret = btrfs_dec_ref(trans, root, eb, 0);
6207 BUG_ON(ret);
6208 }
6209 /* make block locked assertion in clean_tree_block happy */
6210 if (!path->locks[level] &&
6211 btrfs_header_generation(eb) == trans->transid) {
6212 btrfs_tree_lock(eb);
6213 btrfs_set_lock_blocking(eb);
6214 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6215 }
6216 clean_tree_block(trans, root, eb);
6217 }
6218
6219 if (eb == root->node) {
6220 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6221 parent = eb->start;
6222 else
6223 BUG_ON(root->root_key.objectid !=
6224 btrfs_header_owner(eb));
6225 } else {
6226 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6227 parent = path->nodes[level + 1]->start;
6228 else
6229 BUG_ON(root->root_key.objectid !=
6230 btrfs_header_owner(path->nodes[level + 1]));
6231 }
6232
6233 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6234 out:
6235 wc->refs[level] = 0;
6236 wc->flags[level] = 0;
6237 return 0;
6238 }
6239
6240 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6241 struct btrfs_root *root,
6242 struct btrfs_path *path,
6243 struct walk_control *wc)
6244 {
6245 int level = wc->level;
6246 int lookup_info = 1;
6247 int ret;
6248
6249 while (level >= 0) {
6250 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6251 if (ret > 0)
6252 break;
6253
6254 if (level == 0)
6255 break;
6256
6257 if (path->slots[level] >=
6258 btrfs_header_nritems(path->nodes[level]))
6259 break;
6260
6261 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6262 if (ret > 0) {
6263 path->slots[level]++;
6264 continue;
6265 } else if (ret < 0)
6266 return ret;
6267 level = wc->level;
6268 }
6269 return 0;
6270 }
6271
6272 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6273 struct btrfs_root *root,
6274 struct btrfs_path *path,
6275 struct walk_control *wc, int max_level)
6276 {
6277 int level = wc->level;
6278 int ret;
6279
6280 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6281 while (level < max_level && path->nodes[level]) {
6282 wc->level = level;
6283 if (path->slots[level] + 1 <
6284 btrfs_header_nritems(path->nodes[level])) {
6285 path->slots[level]++;
6286 return 0;
6287 } else {
6288 ret = walk_up_proc(trans, root, path, wc);
6289 if (ret > 0)
6290 return 0;
6291
6292 if (path->locks[level]) {
6293 btrfs_tree_unlock_rw(path->nodes[level],
6294 path->locks[level]);
6295 path->locks[level] = 0;
6296 }
6297 free_extent_buffer(path->nodes[level]);
6298 path->nodes[level] = NULL;
6299 level++;
6300 }
6301 }
6302 return 1;
6303 }
6304
6305 /*
6306 * drop a subvolume tree.
6307 *
6308 * this function traverses the tree freeing any blocks that only
6309 * referenced by the tree.
6310 *
6311 * when a shared tree block is found. this function decreases its
6312 * reference count by one. if update_ref is true, this function
6313 * also make sure backrefs for the shared block and all lower level
6314 * blocks are properly updated.
6315 */
6316 void btrfs_drop_snapshot(struct btrfs_root *root,
6317 struct btrfs_block_rsv *block_rsv, int update_ref)
6318 {
6319 struct btrfs_path *path;
6320 struct btrfs_trans_handle *trans;
6321 struct btrfs_root *tree_root = root->fs_info->tree_root;
6322 struct btrfs_root_item *root_item = &root->root_item;
6323 struct walk_control *wc;
6324 struct btrfs_key key;
6325 int err = 0;
6326 int ret;
6327 int level;
6328
6329 path = btrfs_alloc_path();
6330 if (!path) {
6331 err = -ENOMEM;
6332 goto out;
6333 }
6334
6335 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6336 if (!wc) {
6337 btrfs_free_path(path);
6338 err = -ENOMEM;
6339 goto out;
6340 }
6341
6342 trans = btrfs_start_transaction(tree_root, 0);
6343 BUG_ON(IS_ERR(trans));
6344
6345 if (block_rsv)
6346 trans->block_rsv = block_rsv;
6347
6348 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6349 level = btrfs_header_level(root->node);
6350 path->nodes[level] = btrfs_lock_root_node(root);
6351 btrfs_set_lock_blocking(path->nodes[level]);
6352 path->slots[level] = 0;
6353 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6354 memset(&wc->update_progress, 0,
6355 sizeof(wc->update_progress));
6356 } else {
6357 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6358 memcpy(&wc->update_progress, &key,
6359 sizeof(wc->update_progress));
6360
6361 level = root_item->drop_level;
6362 BUG_ON(level == 0);
6363 path->lowest_level = level;
6364 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6365 path->lowest_level = 0;
6366 if (ret < 0) {
6367 err = ret;
6368 goto out_free;
6369 }
6370 WARN_ON(ret > 0);
6371
6372 /*
6373 * unlock our path, this is safe because only this
6374 * function is allowed to delete this snapshot
6375 */
6376 btrfs_unlock_up_safe(path, 0);
6377
6378 level = btrfs_header_level(root->node);
6379 while (1) {
6380 btrfs_tree_lock(path->nodes[level]);
6381 btrfs_set_lock_blocking(path->nodes[level]);
6382
6383 ret = btrfs_lookup_extent_info(trans, root,
6384 path->nodes[level]->start,
6385 path->nodes[level]->len,
6386 &wc->refs[level],
6387 &wc->flags[level]);
6388 BUG_ON(ret);
6389 BUG_ON(wc->refs[level] == 0);
6390
6391 if (level == root_item->drop_level)
6392 break;
6393
6394 btrfs_tree_unlock(path->nodes[level]);
6395 WARN_ON(wc->refs[level] != 1);
6396 level--;
6397 }
6398 }
6399
6400 wc->level = level;
6401 wc->shared_level = -1;
6402 wc->stage = DROP_REFERENCE;
6403 wc->update_ref = update_ref;
6404 wc->keep_locks = 0;
6405 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6406
6407 while (1) {
6408 ret = walk_down_tree(trans, root, path, wc);
6409 if (ret < 0) {
6410 err = ret;
6411 break;
6412 }
6413
6414 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6415 if (ret < 0) {
6416 err = ret;
6417 break;
6418 }
6419
6420 if (ret > 0) {
6421 BUG_ON(wc->stage != DROP_REFERENCE);
6422 break;
6423 }
6424
6425 if (wc->stage == DROP_REFERENCE) {
6426 level = wc->level;
6427 btrfs_node_key(path->nodes[level],
6428 &root_item->drop_progress,
6429 path->slots[level]);
6430 root_item->drop_level = level;
6431 }
6432
6433 BUG_ON(wc->level == 0);
6434 if (btrfs_should_end_transaction(trans, tree_root)) {
6435 ret = btrfs_update_root(trans, tree_root,
6436 &root->root_key,
6437 root_item);
6438 BUG_ON(ret);
6439
6440 btrfs_end_transaction_throttle(trans, tree_root);
6441 trans = btrfs_start_transaction(tree_root, 0);
6442 BUG_ON(IS_ERR(trans));
6443 if (block_rsv)
6444 trans->block_rsv = block_rsv;
6445 }
6446 }
6447 btrfs_release_path(path);
6448 BUG_ON(err);
6449
6450 ret = btrfs_del_root(trans, tree_root, &root->root_key);
6451 BUG_ON(ret);
6452
6453 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6454 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6455 NULL, NULL);
6456 BUG_ON(ret < 0);
6457 if (ret > 0) {
6458 /* if we fail to delete the orphan item this time
6459 * around, it'll get picked up the next time.
6460 *
6461 * The most common failure here is just -ENOENT.
6462 */
6463 btrfs_del_orphan_item(trans, tree_root,
6464 root->root_key.objectid);
6465 }
6466 }
6467
6468 if (root->in_radix) {
6469 btrfs_free_fs_root(tree_root->fs_info, root);
6470 } else {
6471 free_extent_buffer(root->node);
6472 free_extent_buffer(root->commit_root);
6473 kfree(root);
6474 }
6475 out_free:
6476 btrfs_end_transaction_throttle(trans, tree_root);
6477 kfree(wc);
6478 btrfs_free_path(path);
6479 out:
6480 if (err)
6481 btrfs_std_error(root->fs_info, err);
6482 return;
6483 }
6484
6485 /*
6486 * drop subtree rooted at tree block 'node'.
6487 *
6488 * NOTE: this function will unlock and release tree block 'node'
6489 */
6490 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6491 struct btrfs_root *root,
6492 struct extent_buffer *node,
6493 struct extent_buffer *parent)
6494 {
6495 struct btrfs_path *path;
6496 struct walk_control *wc;
6497 int level;
6498 int parent_level;
6499 int ret = 0;
6500 int wret;
6501
6502 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6503
6504 path = btrfs_alloc_path();
6505 if (!path)
6506 return -ENOMEM;
6507
6508 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6509 if (!wc) {
6510 btrfs_free_path(path);
6511 return -ENOMEM;
6512 }
6513
6514 btrfs_assert_tree_locked(parent);
6515 parent_level = btrfs_header_level(parent);
6516 extent_buffer_get(parent);
6517 path->nodes[parent_level] = parent;
6518 path->slots[parent_level] = btrfs_header_nritems(parent);
6519
6520 btrfs_assert_tree_locked(node);
6521 level = btrfs_header_level(node);
6522 path->nodes[level] = node;
6523 path->slots[level] = 0;
6524 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6525
6526 wc->refs[parent_level] = 1;
6527 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6528 wc->level = level;
6529 wc->shared_level = -1;
6530 wc->stage = DROP_REFERENCE;
6531 wc->update_ref = 0;
6532 wc->keep_locks = 1;
6533 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6534
6535 while (1) {
6536 wret = walk_down_tree(trans, root, path, wc);
6537 if (wret < 0) {
6538 ret = wret;
6539 break;
6540 }
6541
6542 wret = walk_up_tree(trans, root, path, wc, parent_level);
6543 if (wret < 0)
6544 ret = wret;
6545 if (wret != 0)
6546 break;
6547 }
6548
6549 kfree(wc);
6550 btrfs_free_path(path);
6551 return ret;
6552 }
6553
6554 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
6555 {
6556 u64 num_devices;
6557 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
6558 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
6559
6560 /*
6561 * we add in the count of missing devices because we want
6562 * to make sure that any RAID levels on a degraded FS
6563 * continue to be honored.
6564 */
6565 num_devices = root->fs_info->fs_devices->rw_devices +
6566 root->fs_info->fs_devices->missing_devices;
6567
6568 if (num_devices == 1) {
6569 stripped |= BTRFS_BLOCK_GROUP_DUP;
6570 stripped = flags & ~stripped;
6571
6572 /* turn raid0 into single device chunks */
6573 if (flags & BTRFS_BLOCK_GROUP_RAID0)
6574 return stripped;
6575
6576 /* turn mirroring into duplication */
6577 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
6578 BTRFS_BLOCK_GROUP_RAID10))
6579 return stripped | BTRFS_BLOCK_GROUP_DUP;
6580 return flags;
6581 } else {
6582 /* they already had raid on here, just return */
6583 if (flags & stripped)
6584 return flags;
6585
6586 stripped |= BTRFS_BLOCK_GROUP_DUP;
6587 stripped = flags & ~stripped;
6588
6589 /* switch duplicated blocks with raid1 */
6590 if (flags & BTRFS_BLOCK_GROUP_DUP)
6591 return stripped | BTRFS_BLOCK_GROUP_RAID1;
6592
6593 /* turn single device chunks into raid0 */
6594 return stripped | BTRFS_BLOCK_GROUP_RAID0;
6595 }
6596 return flags;
6597 }
6598
6599 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
6600 {
6601 struct btrfs_space_info *sinfo = cache->space_info;
6602 u64 num_bytes;
6603 u64 min_allocable_bytes;
6604 int ret = -ENOSPC;
6605
6606
6607 /*
6608 * We need some metadata space and system metadata space for
6609 * allocating chunks in some corner cases until we force to set
6610 * it to be readonly.
6611 */
6612 if ((sinfo->flags &
6613 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
6614 !force)
6615 min_allocable_bytes = 1 * 1024 * 1024;
6616 else
6617 min_allocable_bytes = 0;
6618
6619 spin_lock(&sinfo->lock);
6620 spin_lock(&cache->lock);
6621
6622 if (cache->ro) {
6623 ret = 0;
6624 goto out;
6625 }
6626
6627 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
6628 cache->bytes_super - btrfs_block_group_used(&cache->item);
6629
6630 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
6631 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
6632 min_allocable_bytes <= sinfo->total_bytes) {
6633 sinfo->bytes_readonly += num_bytes;
6634 cache->ro = 1;
6635 ret = 0;
6636 }
6637 out:
6638 spin_unlock(&cache->lock);
6639 spin_unlock(&sinfo->lock);
6640 return ret;
6641 }
6642
6643 int btrfs_set_block_group_ro(struct btrfs_root *root,
6644 struct btrfs_block_group_cache *cache)
6645
6646 {
6647 struct btrfs_trans_handle *trans;
6648 u64 alloc_flags;
6649 int ret;
6650
6651 BUG_ON(cache->ro);
6652
6653 trans = btrfs_join_transaction(root);
6654 BUG_ON(IS_ERR(trans));
6655
6656 alloc_flags = update_block_group_flags(root, cache->flags);
6657 if (alloc_flags != cache->flags)
6658 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6659 CHUNK_ALLOC_FORCE);
6660
6661 ret = set_block_group_ro(cache, 0);
6662 if (!ret)
6663 goto out;
6664 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
6665 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6666 CHUNK_ALLOC_FORCE);
6667 if (ret < 0)
6668 goto out;
6669 ret = set_block_group_ro(cache, 0);
6670 out:
6671 btrfs_end_transaction(trans, root);
6672 return ret;
6673 }
6674
6675 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
6676 struct btrfs_root *root, u64 type)
6677 {
6678 u64 alloc_flags = get_alloc_profile(root, type);
6679 return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6680 CHUNK_ALLOC_FORCE);
6681 }
6682
6683 /*
6684 * helper to account the unused space of all the readonly block group in the
6685 * list. takes mirrors into account.
6686 */
6687 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
6688 {
6689 struct btrfs_block_group_cache *block_group;
6690 u64 free_bytes = 0;
6691 int factor;
6692
6693 list_for_each_entry(block_group, groups_list, list) {
6694 spin_lock(&block_group->lock);
6695
6696 if (!block_group->ro) {
6697 spin_unlock(&block_group->lock);
6698 continue;
6699 }
6700
6701 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
6702 BTRFS_BLOCK_GROUP_RAID10 |
6703 BTRFS_BLOCK_GROUP_DUP))
6704 factor = 2;
6705 else
6706 factor = 1;
6707
6708 free_bytes += (block_group->key.offset -
6709 btrfs_block_group_used(&block_group->item)) *
6710 factor;
6711
6712 spin_unlock(&block_group->lock);
6713 }
6714
6715 return free_bytes;
6716 }
6717
6718 /*
6719 * helper to account the unused space of all the readonly block group in the
6720 * space_info. takes mirrors into account.
6721 */
6722 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
6723 {
6724 int i;
6725 u64 free_bytes = 0;
6726
6727 spin_lock(&sinfo->lock);
6728
6729 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
6730 if (!list_empty(&sinfo->block_groups[i]))
6731 free_bytes += __btrfs_get_ro_block_group_free_space(
6732 &sinfo->block_groups[i]);
6733
6734 spin_unlock(&sinfo->lock);
6735
6736 return free_bytes;
6737 }
6738
6739 int btrfs_set_block_group_rw(struct btrfs_root *root,
6740 struct btrfs_block_group_cache *cache)
6741 {
6742 struct btrfs_space_info *sinfo = cache->space_info;
6743 u64 num_bytes;
6744
6745 BUG_ON(!cache->ro);
6746
6747 spin_lock(&sinfo->lock);
6748 spin_lock(&cache->lock);
6749 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
6750 cache->bytes_super - btrfs_block_group_used(&cache->item);
6751 sinfo->bytes_readonly -= num_bytes;
6752 cache->ro = 0;
6753 spin_unlock(&cache->lock);
6754 spin_unlock(&sinfo->lock);
6755 return 0;
6756 }
6757
6758 /*
6759 * checks to see if its even possible to relocate this block group.
6760 *
6761 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
6762 * ok to go ahead and try.
6763 */
6764 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
6765 {
6766 struct btrfs_block_group_cache *block_group;
6767 struct btrfs_space_info *space_info;
6768 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6769 struct btrfs_device *device;
6770 u64 min_free;
6771 u64 dev_min = 1;
6772 u64 dev_nr = 0;
6773 int index;
6774 int full = 0;
6775 int ret = 0;
6776
6777 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
6778
6779 /* odd, couldn't find the block group, leave it alone */
6780 if (!block_group)
6781 return -1;
6782
6783 min_free = btrfs_block_group_used(&block_group->item);
6784
6785 /* no bytes used, we're good */
6786 if (!min_free)
6787 goto out;
6788
6789 space_info = block_group->space_info;
6790 spin_lock(&space_info->lock);
6791
6792 full = space_info->full;
6793
6794 /*
6795 * if this is the last block group we have in this space, we can't
6796 * relocate it unless we're able to allocate a new chunk below.
6797 *
6798 * Otherwise, we need to make sure we have room in the space to handle
6799 * all of the extents from this block group. If we can, we're good
6800 */
6801 if ((space_info->total_bytes != block_group->key.offset) &&
6802 (space_info->bytes_used + space_info->bytes_reserved +
6803 space_info->bytes_pinned + space_info->bytes_readonly +
6804 min_free < space_info->total_bytes)) {
6805 spin_unlock(&space_info->lock);
6806 goto out;
6807 }
6808 spin_unlock(&space_info->lock);
6809
6810 /*
6811 * ok we don't have enough space, but maybe we have free space on our
6812 * devices to allocate new chunks for relocation, so loop through our
6813 * alloc devices and guess if we have enough space. However, if we
6814 * were marked as full, then we know there aren't enough chunks, and we
6815 * can just return.
6816 */
6817 ret = -1;
6818 if (full)
6819 goto out;
6820
6821 /*
6822 * index:
6823 * 0: raid10
6824 * 1: raid1
6825 * 2: dup
6826 * 3: raid0
6827 * 4: single
6828 */
6829 index = get_block_group_index(block_group);
6830 if (index == 0) {
6831 dev_min = 4;
6832 /* Divide by 2 */
6833 min_free >>= 1;
6834 } else if (index == 1) {
6835 dev_min = 2;
6836 } else if (index == 2) {
6837 /* Multiply by 2 */
6838 min_free <<= 1;
6839 } else if (index == 3) {
6840 dev_min = fs_devices->rw_devices;
6841 do_div(min_free, dev_min);
6842 }
6843
6844 mutex_lock(&root->fs_info->chunk_mutex);
6845 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
6846 u64 dev_offset;
6847
6848 /*
6849 * check to make sure we can actually find a chunk with enough
6850 * space to fit our block group in.
6851 */
6852 if (device->total_bytes > device->bytes_used + min_free) {
6853 ret = find_free_dev_extent(NULL, device, min_free,
6854 &dev_offset, NULL);
6855 if (!ret)
6856 dev_nr++;
6857
6858 if (dev_nr >= dev_min)
6859 break;
6860
6861 ret = -1;
6862 }
6863 }
6864 mutex_unlock(&root->fs_info->chunk_mutex);
6865 out:
6866 btrfs_put_block_group(block_group);
6867 return ret;
6868 }
6869
6870 static int find_first_block_group(struct btrfs_root *root,
6871 struct btrfs_path *path, struct btrfs_key *key)
6872 {
6873 int ret = 0;
6874 struct btrfs_key found_key;
6875 struct extent_buffer *leaf;
6876 int slot;
6877
6878 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
6879 if (ret < 0)
6880 goto out;
6881
6882 while (1) {
6883 slot = path->slots[0];
6884 leaf = path->nodes[0];
6885 if (slot >= btrfs_header_nritems(leaf)) {
6886 ret = btrfs_next_leaf(root, path);
6887 if (ret == 0)
6888 continue;
6889 if (ret < 0)
6890 goto out;
6891 break;
6892 }
6893 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6894
6895 if (found_key.objectid >= key->objectid &&
6896 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
6897 ret = 0;
6898 goto out;
6899 }
6900 path->slots[0]++;
6901 }
6902 out:
6903 return ret;
6904 }
6905
6906 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
6907 {
6908 struct btrfs_block_group_cache *block_group;
6909 u64 last = 0;
6910
6911 while (1) {
6912 struct inode *inode;
6913
6914 block_group = btrfs_lookup_first_block_group(info, last);
6915 while (block_group) {
6916 spin_lock(&block_group->lock);
6917 if (block_group->iref)
6918 break;
6919 spin_unlock(&block_group->lock);
6920 block_group = next_block_group(info->tree_root,
6921 block_group);
6922 }
6923 if (!block_group) {
6924 if (last == 0)
6925 break;
6926 last = 0;
6927 continue;
6928 }
6929
6930 inode = block_group->inode;
6931 block_group->iref = 0;
6932 block_group->inode = NULL;
6933 spin_unlock(&block_group->lock);
6934 iput(inode);
6935 last = block_group->key.objectid + block_group->key.offset;
6936 btrfs_put_block_group(block_group);
6937 }
6938 }
6939
6940 int btrfs_free_block_groups(struct btrfs_fs_info *info)
6941 {
6942 struct btrfs_block_group_cache *block_group;
6943 struct btrfs_space_info *space_info;
6944 struct btrfs_caching_control *caching_ctl;
6945 struct rb_node *n;
6946
6947 down_write(&info->extent_commit_sem);
6948 while (!list_empty(&info->caching_block_groups)) {
6949 caching_ctl = list_entry(info->caching_block_groups.next,
6950 struct btrfs_caching_control, list);
6951 list_del(&caching_ctl->list);
6952 put_caching_control(caching_ctl);
6953 }
6954 up_write(&info->extent_commit_sem);
6955
6956 spin_lock(&info->block_group_cache_lock);
6957 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
6958 block_group = rb_entry(n, struct btrfs_block_group_cache,
6959 cache_node);
6960 rb_erase(&block_group->cache_node,
6961 &info->block_group_cache_tree);
6962 spin_unlock(&info->block_group_cache_lock);
6963
6964 down_write(&block_group->space_info->groups_sem);
6965 list_del(&block_group->list);
6966 up_write(&block_group->space_info->groups_sem);
6967
6968 if (block_group->cached == BTRFS_CACHE_STARTED)
6969 wait_block_group_cache_done(block_group);
6970
6971 /*
6972 * We haven't cached this block group, which means we could
6973 * possibly have excluded extents on this block group.
6974 */
6975 if (block_group->cached == BTRFS_CACHE_NO)
6976 free_excluded_extents(info->extent_root, block_group);
6977
6978 btrfs_remove_free_space_cache(block_group);
6979 btrfs_put_block_group(block_group);
6980
6981 spin_lock(&info->block_group_cache_lock);
6982 }
6983 spin_unlock(&info->block_group_cache_lock);
6984
6985 /* now that all the block groups are freed, go through and
6986 * free all the space_info structs. This is only called during
6987 * the final stages of unmount, and so we know nobody is
6988 * using them. We call synchronize_rcu() once before we start,
6989 * just to be on the safe side.
6990 */
6991 synchronize_rcu();
6992
6993 release_global_block_rsv(info);
6994
6995 while(!list_empty(&info->space_info)) {
6996 space_info = list_entry(info->space_info.next,
6997 struct btrfs_space_info,
6998 list);
6999 if (space_info->bytes_pinned > 0 ||
7000 space_info->bytes_reserved > 0 ||
7001 space_info->bytes_may_use > 0) {
7002 WARN_ON(1);
7003 dump_space_info(space_info, 0, 0);
7004 }
7005 list_del(&space_info->list);
7006 kfree(space_info);
7007 }
7008 return 0;
7009 }
7010
7011 static void __link_block_group(struct btrfs_space_info *space_info,
7012 struct btrfs_block_group_cache *cache)
7013 {
7014 int index = get_block_group_index(cache);
7015
7016 down_write(&space_info->groups_sem);
7017 list_add_tail(&cache->list, &space_info->block_groups[index]);
7018 up_write(&space_info->groups_sem);
7019 }
7020
7021 int btrfs_read_block_groups(struct btrfs_root *root)
7022 {
7023 struct btrfs_path *path;
7024 int ret;
7025 struct btrfs_block_group_cache *cache;
7026 struct btrfs_fs_info *info = root->fs_info;
7027 struct btrfs_space_info *space_info;
7028 struct btrfs_key key;
7029 struct btrfs_key found_key;
7030 struct extent_buffer *leaf;
7031 int need_clear = 0;
7032 u64 cache_gen;
7033
7034 root = info->extent_root;
7035 key.objectid = 0;
7036 key.offset = 0;
7037 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7038 path = btrfs_alloc_path();
7039 if (!path)
7040 return -ENOMEM;
7041 path->reada = 1;
7042
7043 cache_gen = btrfs_super_cache_generation(&root->fs_info->super_copy);
7044 if (btrfs_test_opt(root, SPACE_CACHE) &&
7045 btrfs_super_generation(&root->fs_info->super_copy) != cache_gen)
7046 need_clear = 1;
7047 if (btrfs_test_opt(root, CLEAR_CACHE))
7048 need_clear = 1;
7049
7050 while (1) {
7051 ret = find_first_block_group(root, path, &key);
7052 if (ret > 0)
7053 break;
7054 if (ret != 0)
7055 goto error;
7056 leaf = path->nodes[0];
7057 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7058 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7059 if (!cache) {
7060 ret = -ENOMEM;
7061 goto error;
7062 }
7063 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7064 GFP_NOFS);
7065 if (!cache->free_space_ctl) {
7066 kfree(cache);
7067 ret = -ENOMEM;
7068 goto error;
7069 }
7070
7071 atomic_set(&cache->count, 1);
7072 spin_lock_init(&cache->lock);
7073 cache->fs_info = info;
7074 INIT_LIST_HEAD(&cache->list);
7075 INIT_LIST_HEAD(&cache->cluster_list);
7076
7077 if (need_clear)
7078 cache->disk_cache_state = BTRFS_DC_CLEAR;
7079
7080 read_extent_buffer(leaf, &cache->item,
7081 btrfs_item_ptr_offset(leaf, path->slots[0]),
7082 sizeof(cache->item));
7083 memcpy(&cache->key, &found_key, sizeof(found_key));
7084
7085 key.objectid = found_key.objectid + found_key.offset;
7086 btrfs_release_path(path);
7087 cache->flags = btrfs_block_group_flags(&cache->item);
7088 cache->sectorsize = root->sectorsize;
7089
7090 btrfs_init_free_space_ctl(cache);
7091
7092 /*
7093 * We need to exclude the super stripes now so that the space
7094 * info has super bytes accounted for, otherwise we'll think
7095 * we have more space than we actually do.
7096 */
7097 exclude_super_stripes(root, cache);
7098
7099 /*
7100 * check for two cases, either we are full, and therefore
7101 * don't need to bother with the caching work since we won't
7102 * find any space, or we are empty, and we can just add all
7103 * the space in and be done with it. This saves us _alot_ of
7104 * time, particularly in the full case.
7105 */
7106 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7107 cache->last_byte_to_unpin = (u64)-1;
7108 cache->cached = BTRFS_CACHE_FINISHED;
7109 free_excluded_extents(root, cache);
7110 } else if (btrfs_block_group_used(&cache->item) == 0) {
7111 cache->last_byte_to_unpin = (u64)-1;
7112 cache->cached = BTRFS_CACHE_FINISHED;
7113 add_new_free_space(cache, root->fs_info,
7114 found_key.objectid,
7115 found_key.objectid +
7116 found_key.offset);
7117 free_excluded_extents(root, cache);
7118 }
7119
7120 ret = update_space_info(info, cache->flags, found_key.offset,
7121 btrfs_block_group_used(&cache->item),
7122 &space_info);
7123 BUG_ON(ret);
7124 cache->space_info = space_info;
7125 spin_lock(&cache->space_info->lock);
7126 cache->space_info->bytes_readonly += cache->bytes_super;
7127 spin_unlock(&cache->space_info->lock);
7128
7129 __link_block_group(space_info, cache);
7130
7131 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7132 BUG_ON(ret);
7133
7134 set_avail_alloc_bits(root->fs_info, cache->flags);
7135 if (btrfs_chunk_readonly(root, cache->key.objectid))
7136 set_block_group_ro(cache, 1);
7137 }
7138
7139 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7140 if (!(get_alloc_profile(root, space_info->flags) &
7141 (BTRFS_BLOCK_GROUP_RAID10 |
7142 BTRFS_BLOCK_GROUP_RAID1 |
7143 BTRFS_BLOCK_GROUP_DUP)))
7144 continue;
7145 /*
7146 * avoid allocating from un-mirrored block group if there are
7147 * mirrored block groups.
7148 */
7149 list_for_each_entry(cache, &space_info->block_groups[3], list)
7150 set_block_group_ro(cache, 1);
7151 list_for_each_entry(cache, &space_info->block_groups[4], list)
7152 set_block_group_ro(cache, 1);
7153 }
7154
7155 init_global_block_rsv(info);
7156 ret = 0;
7157 error:
7158 btrfs_free_path(path);
7159 return ret;
7160 }
7161
7162 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7163 struct btrfs_root *root, u64 bytes_used,
7164 u64 type, u64 chunk_objectid, u64 chunk_offset,
7165 u64 size)
7166 {
7167 int ret;
7168 struct btrfs_root *extent_root;
7169 struct btrfs_block_group_cache *cache;
7170
7171 extent_root = root->fs_info->extent_root;
7172
7173 root->fs_info->last_trans_log_full_commit = trans->transid;
7174
7175 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7176 if (!cache)
7177 return -ENOMEM;
7178 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7179 GFP_NOFS);
7180 if (!cache->free_space_ctl) {
7181 kfree(cache);
7182 return -ENOMEM;
7183 }
7184
7185 cache->key.objectid = chunk_offset;
7186 cache->key.offset = size;
7187 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7188 cache->sectorsize = root->sectorsize;
7189 cache->fs_info = root->fs_info;
7190
7191 atomic_set(&cache->count, 1);
7192 spin_lock_init(&cache->lock);
7193 INIT_LIST_HEAD(&cache->list);
7194 INIT_LIST_HEAD(&cache->cluster_list);
7195
7196 btrfs_init_free_space_ctl(cache);
7197
7198 btrfs_set_block_group_used(&cache->item, bytes_used);
7199 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7200 cache->flags = type;
7201 btrfs_set_block_group_flags(&cache->item, type);
7202
7203 cache->last_byte_to_unpin = (u64)-1;
7204 cache->cached = BTRFS_CACHE_FINISHED;
7205 exclude_super_stripes(root, cache);
7206
7207 add_new_free_space(cache, root->fs_info, chunk_offset,
7208 chunk_offset + size);
7209
7210 free_excluded_extents(root, cache);
7211
7212 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7213 &cache->space_info);
7214 BUG_ON(ret);
7215
7216 spin_lock(&cache->space_info->lock);
7217 cache->space_info->bytes_readonly += cache->bytes_super;
7218 spin_unlock(&cache->space_info->lock);
7219
7220 __link_block_group(cache->space_info, cache);
7221
7222 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7223 BUG_ON(ret);
7224
7225 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7226 sizeof(cache->item));
7227 BUG_ON(ret);
7228
7229 set_avail_alloc_bits(extent_root->fs_info, type);
7230
7231 return 0;
7232 }
7233
7234 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7235 struct btrfs_root *root, u64 group_start)
7236 {
7237 struct btrfs_path *path;
7238 struct btrfs_block_group_cache *block_group;
7239 struct btrfs_free_cluster *cluster;
7240 struct btrfs_root *tree_root = root->fs_info->tree_root;
7241 struct btrfs_key key;
7242 struct inode *inode;
7243 int ret;
7244 int factor;
7245
7246 root = root->fs_info->extent_root;
7247
7248 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7249 BUG_ON(!block_group);
7250 BUG_ON(!block_group->ro);
7251
7252 /*
7253 * Free the reserved super bytes from this block group before
7254 * remove it.
7255 */
7256 free_excluded_extents(root, block_group);
7257
7258 memcpy(&key, &block_group->key, sizeof(key));
7259 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7260 BTRFS_BLOCK_GROUP_RAID1 |
7261 BTRFS_BLOCK_GROUP_RAID10))
7262 factor = 2;
7263 else
7264 factor = 1;
7265
7266 /* make sure this block group isn't part of an allocation cluster */
7267 cluster = &root->fs_info->data_alloc_cluster;
7268 spin_lock(&cluster->refill_lock);
7269 btrfs_return_cluster_to_free_space(block_group, cluster);
7270 spin_unlock(&cluster->refill_lock);
7271
7272 /*
7273 * make sure this block group isn't part of a metadata
7274 * allocation cluster
7275 */
7276 cluster = &root->fs_info->meta_alloc_cluster;
7277 spin_lock(&cluster->refill_lock);
7278 btrfs_return_cluster_to_free_space(block_group, cluster);
7279 spin_unlock(&cluster->refill_lock);
7280
7281 path = btrfs_alloc_path();
7282 if (!path) {
7283 ret = -ENOMEM;
7284 goto out;
7285 }
7286
7287 inode = lookup_free_space_inode(root, block_group, path);
7288 if (!IS_ERR(inode)) {
7289 ret = btrfs_orphan_add(trans, inode);
7290 BUG_ON(ret);
7291 clear_nlink(inode);
7292 /* One for the block groups ref */
7293 spin_lock(&block_group->lock);
7294 if (block_group->iref) {
7295 block_group->iref = 0;
7296 block_group->inode = NULL;
7297 spin_unlock(&block_group->lock);
7298 iput(inode);
7299 } else {
7300 spin_unlock(&block_group->lock);
7301 }
7302 /* One for our lookup ref */
7303 btrfs_add_delayed_iput(inode);
7304 }
7305
7306 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
7307 key.offset = block_group->key.objectid;
7308 key.type = 0;
7309
7310 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
7311 if (ret < 0)
7312 goto out;
7313 if (ret > 0)
7314 btrfs_release_path(path);
7315 if (ret == 0) {
7316 ret = btrfs_del_item(trans, tree_root, path);
7317 if (ret)
7318 goto out;
7319 btrfs_release_path(path);
7320 }
7321
7322 spin_lock(&root->fs_info->block_group_cache_lock);
7323 rb_erase(&block_group->cache_node,
7324 &root->fs_info->block_group_cache_tree);
7325 spin_unlock(&root->fs_info->block_group_cache_lock);
7326
7327 down_write(&block_group->space_info->groups_sem);
7328 /*
7329 * we must use list_del_init so people can check to see if they
7330 * are still on the list after taking the semaphore
7331 */
7332 list_del_init(&block_group->list);
7333 up_write(&block_group->space_info->groups_sem);
7334
7335 if (block_group->cached == BTRFS_CACHE_STARTED)
7336 wait_block_group_cache_done(block_group);
7337
7338 btrfs_remove_free_space_cache(block_group);
7339
7340 spin_lock(&block_group->space_info->lock);
7341 block_group->space_info->total_bytes -= block_group->key.offset;
7342 block_group->space_info->bytes_readonly -= block_group->key.offset;
7343 block_group->space_info->disk_total -= block_group->key.offset * factor;
7344 spin_unlock(&block_group->space_info->lock);
7345
7346 memcpy(&key, &block_group->key, sizeof(key));
7347
7348 btrfs_clear_space_info_full(root->fs_info);
7349
7350 btrfs_put_block_group(block_group);
7351 btrfs_put_block_group(block_group);
7352
7353 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7354 if (ret > 0)
7355 ret = -EIO;
7356 if (ret < 0)
7357 goto out;
7358
7359 ret = btrfs_del_item(trans, root, path);
7360 out:
7361 btrfs_free_path(path);
7362 return ret;
7363 }
7364
7365 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
7366 {
7367 struct btrfs_space_info *space_info;
7368 struct btrfs_super_block *disk_super;
7369 u64 features;
7370 u64 flags;
7371 int mixed = 0;
7372 int ret;
7373
7374 disk_super = &fs_info->super_copy;
7375 if (!btrfs_super_root(disk_super))
7376 return 1;
7377
7378 features = btrfs_super_incompat_flags(disk_super);
7379 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
7380 mixed = 1;
7381
7382 flags = BTRFS_BLOCK_GROUP_SYSTEM;
7383 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7384 if (ret)
7385 goto out;
7386
7387 if (mixed) {
7388 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
7389 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7390 } else {
7391 flags = BTRFS_BLOCK_GROUP_METADATA;
7392 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7393 if (ret)
7394 goto out;
7395
7396 flags = BTRFS_BLOCK_GROUP_DATA;
7397 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7398 }
7399 out:
7400 return ret;
7401 }
7402
7403 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
7404 {
7405 return unpin_extent_range(root, start, end);
7406 }
7407
7408 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
7409 u64 num_bytes, u64 *actual_bytes)
7410 {
7411 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
7412 }
7413
7414 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
7415 {
7416 struct btrfs_fs_info *fs_info = root->fs_info;
7417 struct btrfs_block_group_cache *cache = NULL;
7418 u64 group_trimmed;
7419 u64 start;
7420 u64 end;
7421 u64 trimmed = 0;
7422 int ret = 0;
7423
7424 cache = btrfs_lookup_block_group(fs_info, range->start);
7425
7426 while (cache) {
7427 if (cache->key.objectid >= (range->start + range->len)) {
7428 btrfs_put_block_group(cache);
7429 break;
7430 }
7431
7432 start = max(range->start, cache->key.objectid);
7433 end = min(range->start + range->len,
7434 cache->key.objectid + cache->key.offset);
7435
7436 if (end - start >= range->minlen) {
7437 if (!block_group_cache_done(cache)) {
7438 ret = cache_block_group(cache, NULL, root, 0);
7439 if (!ret)
7440 wait_block_group_cache_done(cache);
7441 }
7442 ret = btrfs_trim_block_group(cache,
7443 &group_trimmed,
7444 start,
7445 end,
7446 range->minlen);
7447
7448 trimmed += group_trimmed;
7449 if (ret) {
7450 btrfs_put_block_group(cache);
7451 break;
7452 }
7453 }
7454
7455 cache = next_block_group(fs_info->tree_root, cache);
7456 }
7457
7458 range->len = trimmed;
7459 return ret;
7460 }
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree