projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
btrfs: Bypass unrelated items before accessing its contents in scrub
[deliverable/linux.git] / fs / btrfs / extent-tree.c
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 <linux/ratelimit.h>
27 #include <linux/percpu_counter.h>
28 #include "hash.h"
29 #include "tree-log.h"
30 #include "disk-io.h"
31 #include "print-tree.h"
32 #include "volumes.h"
33 #include "raid56.h"
34 #include "locking.h"
35 #include "free-space-cache.h"
36 #include "math.h"
37 #include "sysfs.h"
38 #include "qgroup.h"
39
40 #undef SCRAMBLE_DELAYED_REFS
41
42 /*
43 * control flags for do_chunk_alloc's force field
44 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
45 * if we really need one.
46 *
47 * CHUNK_ALLOC_LIMITED means to only try and allocate one
48 * if we have very few chunks already allocated. This is
49 * used as part of the clustering code to help make sure
50 * we have a good pool of storage to cluster in, without
51 * filling the FS with empty chunks
52 *
53 * CHUNK_ALLOC_FORCE means it must try to allocate one
54 *
55 */
56 enum {
57 CHUNK_ALLOC_NO_FORCE = 0,
58 CHUNK_ALLOC_LIMITED = 1,
59 CHUNK_ALLOC_FORCE = 2,
60 };
61
62 /*
63 * Control how reservations are dealt with.
64 *
65 * RESERVE_FREE - freeing a reservation.
66 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
67 * ENOSPC accounting
68 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
69 * bytes_may_use as the ENOSPC accounting is done elsewhere
70 */
71 enum {
72 RESERVE_FREE = 0,
73 RESERVE_ALLOC = 1,
74 RESERVE_ALLOC_NO_ACCOUNT = 2,
75 };
76
77 static int update_block_group(struct btrfs_trans_handle *trans,
78 struct btrfs_root *root, u64 bytenr,
79 u64 num_bytes, int alloc);
80 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
81 struct btrfs_root *root,
82 struct btrfs_delayed_ref_node *node, u64 parent,
83 u64 root_objectid, u64 owner_objectid,
84 u64 owner_offset, int refs_to_drop,
85 struct btrfs_delayed_extent_op *extra_op);
86 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
87 struct extent_buffer *leaf,
88 struct btrfs_extent_item *ei);
89 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
90 struct btrfs_root *root,
91 u64 parent, u64 root_objectid,
92 u64 flags, u64 owner, u64 offset,
93 struct btrfs_key *ins, int ref_mod);
94 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
95 struct btrfs_root *root,
96 u64 parent, u64 root_objectid,
97 u64 flags, struct btrfs_disk_key *key,
98 int level, struct btrfs_key *ins,
99 int no_quota);
100 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
101 struct btrfs_root *extent_root, u64 flags,
102 int force);
103 static int find_next_key(struct btrfs_path *path, int level,
104 struct btrfs_key *key);
105 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
106 int dump_block_groups);
107 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
108 u64 num_bytes, int reserve,
109 int delalloc);
110 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
111 u64 num_bytes);
112 int btrfs_pin_extent(struct btrfs_root *root,
113 u64 bytenr, u64 num_bytes, int reserved);
114
115 static noinline int
116 block_group_cache_done(struct btrfs_block_group_cache *cache)
117 {
118 smp_mb();
119 return cache->cached == BTRFS_CACHE_FINISHED ||
120 cache->cached == BTRFS_CACHE_ERROR;
121 }
122
123 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
124 {
125 return (cache->flags & bits) == bits;
126 }
127
128 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
129 {
130 atomic_inc(&cache->count);
131 }
132
133 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
134 {
135 if (atomic_dec_and_test(&cache->count)) {
136 WARN_ON(cache->pinned > 0);
137 WARN_ON(cache->reserved > 0);
138 kfree(cache->free_space_ctl);
139 kfree(cache);
140 }
141 }
142
143 /*
144 * this adds the block group to the fs_info rb tree for the block group
145 * cache
146 */
147 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
148 struct btrfs_block_group_cache *block_group)
149 {
150 struct rb_node **p;
151 struct rb_node *parent = NULL;
152 struct btrfs_block_group_cache *cache;
153
154 spin_lock(&info->block_group_cache_lock);
155 p = &info->block_group_cache_tree.rb_node;
156
157 while (*p) {
158 parent = *p;
159 cache = rb_entry(parent, struct btrfs_block_group_cache,
160 cache_node);
161 if (block_group->key.objectid < cache->key.objectid) {
162 p = &(*p)->rb_left;
163 } else if (block_group->key.objectid > cache->key.objectid) {
164 p = &(*p)->rb_right;
165 } else {
166 spin_unlock(&info->block_group_cache_lock);
167 return -EEXIST;
168 }
169 }
170
171 rb_link_node(&block_group->cache_node, parent, p);
172 rb_insert_color(&block_group->cache_node,
173 &info->block_group_cache_tree);
174
175 if (info->first_logical_byte > block_group->key.objectid)
176 info->first_logical_byte = block_group->key.objectid;
177
178 spin_unlock(&info->block_group_cache_lock);
179
180 return 0;
181 }
182
183 /*
184 * This will return the block group at or after bytenr if contains is 0, else
185 * it will return the block group that contains the bytenr
186 */
187 static struct btrfs_block_group_cache *
188 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
189 int contains)
190 {
191 struct btrfs_block_group_cache *cache, *ret = NULL;
192 struct rb_node *n;
193 u64 end, start;
194
195 spin_lock(&info->block_group_cache_lock);
196 n = info->block_group_cache_tree.rb_node;
197
198 while (n) {
199 cache = rb_entry(n, struct btrfs_block_group_cache,
200 cache_node);
201 end = cache->key.objectid + cache->key.offset - 1;
202 start = cache->key.objectid;
203
204 if (bytenr < start) {
205 if (!contains && (!ret || start < ret->key.objectid))
206 ret = cache;
207 n = n->rb_left;
208 } else if (bytenr > start) {
209 if (contains && bytenr <= end) {
210 ret = cache;
211 break;
212 }
213 n = n->rb_right;
214 } else {
215 ret = cache;
216 break;
217 }
218 }
219 if (ret) {
220 btrfs_get_block_group(ret);
221 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
222 info->first_logical_byte = ret->key.objectid;
223 }
224 spin_unlock(&info->block_group_cache_lock);
225
226 return ret;
227 }
228
229 static int add_excluded_extent(struct btrfs_root *root,
230 u64 start, u64 num_bytes)
231 {
232 u64 end = start + num_bytes - 1;
233 set_extent_bits(&root->fs_info->freed_extents[0],
234 start, end, EXTENT_UPTODATE, GFP_NOFS);
235 set_extent_bits(&root->fs_info->freed_extents[1],
236 start, end, EXTENT_UPTODATE, GFP_NOFS);
237 return 0;
238 }
239
240 static void free_excluded_extents(struct btrfs_root *root,
241 struct btrfs_block_group_cache *cache)
242 {
243 u64 start, end;
244
245 start = cache->key.objectid;
246 end = start + cache->key.offset - 1;
247
248 clear_extent_bits(&root->fs_info->freed_extents[0],
249 start, end, EXTENT_UPTODATE, GFP_NOFS);
250 clear_extent_bits(&root->fs_info->freed_extents[1],
251 start, end, EXTENT_UPTODATE, GFP_NOFS);
252 }
253
254 static int exclude_super_stripes(struct btrfs_root *root,
255 struct btrfs_block_group_cache *cache)
256 {
257 u64 bytenr;
258 u64 *logical;
259 int stripe_len;
260 int i, nr, ret;
261
262 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
263 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
264 cache->bytes_super += stripe_len;
265 ret = add_excluded_extent(root, cache->key.objectid,
266 stripe_len);
267 if (ret)
268 return ret;
269 }
270
271 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
272 bytenr = btrfs_sb_offset(i);
273 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
274 cache->key.objectid, bytenr,
275 0, &logical, &nr, &stripe_len);
276 if (ret)
277 return ret;
278
279 while (nr--) {
280 u64 start, len;
281
282 if (logical[nr] > cache->key.objectid +
283 cache->key.offset)
284 continue;
285
286 if (logical[nr] + stripe_len <= cache->key.objectid)
287 continue;
288
289 start = logical[nr];
290 if (start < cache->key.objectid) {
291 start = cache->key.objectid;
292 len = (logical[nr] + stripe_len) - start;
293 } else {
294 len = min_t(u64, stripe_len,
295 cache->key.objectid +
296 cache->key.offset - start);
297 }
298
299 cache->bytes_super += len;
300 ret = add_excluded_extent(root, start, len);
301 if (ret) {
302 kfree(logical);
303 return ret;
304 }
305 }
306
307 kfree(logical);
308 }
309 return 0;
310 }
311
312 static struct btrfs_caching_control *
313 get_caching_control(struct btrfs_block_group_cache *cache)
314 {
315 struct btrfs_caching_control *ctl;
316
317 spin_lock(&cache->lock);
318 if (!cache->caching_ctl) {
319 spin_unlock(&cache->lock);
320 return NULL;
321 }
322
323 ctl = cache->caching_ctl;
324 atomic_inc(&ctl->count);
325 spin_unlock(&cache->lock);
326 return ctl;
327 }
328
329 static void put_caching_control(struct btrfs_caching_control *ctl)
330 {
331 if (atomic_dec_and_test(&ctl->count))
332 kfree(ctl);
333 }
334
335 /*
336 * this is only called by cache_block_group, since we could have freed extents
337 * we need to check the pinned_extents for any extents that can't be used yet
338 * since their free space will be released as soon as the transaction commits.
339 */
340 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
341 struct btrfs_fs_info *info, u64 start, u64 end)
342 {
343 u64 extent_start, extent_end, size, total_added = 0;
344 int ret;
345
346 while (start < end) {
347 ret = find_first_extent_bit(info->pinned_extents, start,
348 &extent_start, &extent_end,
349 EXTENT_DIRTY | EXTENT_UPTODATE,
350 NULL);
351 if (ret)
352 break;
353
354 if (extent_start <= start) {
355 start = extent_end + 1;
356 } else if (extent_start > start && extent_start < end) {
357 size = extent_start - start;
358 total_added += size;
359 ret = btrfs_add_free_space(block_group, start,
360 size);
361 BUG_ON(ret); /* -ENOMEM or logic error */
362 start = extent_end + 1;
363 } else {
364 break;
365 }
366 }
367
368 if (start < end) {
369 size = end - start;
370 total_added += size;
371 ret = btrfs_add_free_space(block_group, start, size);
372 BUG_ON(ret); /* -ENOMEM or logic error */
373 }
374
375 return total_added;
376 }
377
378 static noinline void caching_thread(struct btrfs_work *work)
379 {
380 struct btrfs_block_group_cache *block_group;
381 struct btrfs_fs_info *fs_info;
382 struct btrfs_caching_control *caching_ctl;
383 struct btrfs_root *extent_root;
384 struct btrfs_path *path;
385 struct extent_buffer *leaf;
386 struct btrfs_key key;
387 u64 total_found = 0;
388 u64 last = 0;
389 u32 nritems;
390 int ret = -ENOMEM;
391
392 caching_ctl = container_of(work, struct btrfs_caching_control, work);
393 block_group = caching_ctl->block_group;
394 fs_info = block_group->fs_info;
395 extent_root = fs_info->extent_root;
396
397 path = btrfs_alloc_path();
398 if (!path)
399 goto out;
400
401 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
402
403 /*
404 * We don't want to deadlock with somebody trying to allocate a new
405 * extent for the extent root while also trying to search the extent
406 * root to add free space. So we skip locking and search the commit
407 * root, since its read-only
408 */
409 path->skip_locking = 1;
410 path->search_commit_root = 1;
411 path->reada = 1;
412
413 key.objectid = last;
414 key.offset = 0;
415 key.type = BTRFS_EXTENT_ITEM_KEY;
416 again:
417 mutex_lock(&caching_ctl->mutex);
418 /* need to make sure the commit_root doesn't disappear */
419 down_read(&fs_info->commit_root_sem);
420
421 next:
422 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
423 if (ret < 0)
424 goto err;
425
426 leaf = path->nodes[0];
427 nritems = btrfs_header_nritems(leaf);
428
429 while (1) {
430 if (btrfs_fs_closing(fs_info) > 1) {
431 last = (u64)-1;
432 break;
433 }
434
435 if (path->slots[0] < nritems) {
436 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
437 } else {
438 ret = find_next_key(path, 0, &key);
439 if (ret)
440 break;
441
442 if (need_resched() ||
443 rwsem_is_contended(&fs_info->commit_root_sem)) {
444 caching_ctl->progress = last;
445 btrfs_release_path(path);
446 up_read(&fs_info->commit_root_sem);
447 mutex_unlock(&caching_ctl->mutex);
448 cond_resched();
449 goto again;
450 }
451
452 ret = btrfs_next_leaf(extent_root, path);
453 if (ret < 0)
454 goto err;
455 if (ret)
456 break;
457 leaf = path->nodes[0];
458 nritems = btrfs_header_nritems(leaf);
459 continue;
460 }
461
462 if (key.objectid < last) {
463 key.objectid = last;
464 key.offset = 0;
465 key.type = BTRFS_EXTENT_ITEM_KEY;
466
467 caching_ctl->progress = last;
468 btrfs_release_path(path);
469 goto next;
470 }
471
472 if (key.objectid < block_group->key.objectid) {
473 path->slots[0]++;
474 continue;
475 }
476
477 if (key.objectid >= block_group->key.objectid +
478 block_group->key.offset)
479 break;
480
481 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
482 key.type == BTRFS_METADATA_ITEM_KEY) {
483 total_found += add_new_free_space(block_group,
484 fs_info, last,
485 key.objectid);
486 if (key.type == BTRFS_METADATA_ITEM_KEY)
487 last = key.objectid +
488 fs_info->tree_root->nodesize;
489 else
490 last = key.objectid + key.offset;
491
492 if (total_found > (1024 * 1024 * 2)) {
493 total_found = 0;
494 wake_up(&caching_ctl->wait);
495 }
496 }
497 path->slots[0]++;
498 }
499 ret = 0;
500
501 total_found += add_new_free_space(block_group, fs_info, last,
502 block_group->key.objectid +
503 block_group->key.offset);
504 caching_ctl->progress = (u64)-1;
505
506 spin_lock(&block_group->lock);
507 block_group->caching_ctl = NULL;
508 block_group->cached = BTRFS_CACHE_FINISHED;
509 spin_unlock(&block_group->lock);
510
511 err:
512 btrfs_free_path(path);
513 up_read(&fs_info->commit_root_sem);
514
515 free_excluded_extents(extent_root, block_group);
516
517 mutex_unlock(&caching_ctl->mutex);
518 out:
519 if (ret) {
520 spin_lock(&block_group->lock);
521 block_group->caching_ctl = NULL;
522 block_group->cached = BTRFS_CACHE_ERROR;
523 spin_unlock(&block_group->lock);
524 }
525 wake_up(&caching_ctl->wait);
526
527 put_caching_control(caching_ctl);
528 btrfs_put_block_group(block_group);
529 }
530
531 static int cache_block_group(struct btrfs_block_group_cache *cache,
532 int load_cache_only)
533 {
534 DEFINE_WAIT(wait);
535 struct btrfs_fs_info *fs_info = cache->fs_info;
536 struct btrfs_caching_control *caching_ctl;
537 int ret = 0;
538
539 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
540 if (!caching_ctl)
541 return -ENOMEM;
542
543 INIT_LIST_HEAD(&caching_ctl->list);
544 mutex_init(&caching_ctl->mutex);
545 init_waitqueue_head(&caching_ctl->wait);
546 caching_ctl->block_group = cache;
547 caching_ctl->progress = cache->key.objectid;
548 atomic_set(&caching_ctl->count, 1);
549 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
550 caching_thread, NULL, NULL);
551
552 spin_lock(&cache->lock);
553 /*
554 * This should be a rare occasion, but this could happen I think in the
555 * case where one thread starts to load the space cache info, and then
556 * some other thread starts a transaction commit which tries to do an
557 * allocation while the other thread is still loading the space cache
558 * info. The previous loop should have kept us from choosing this block
559 * group, but if we've moved to the state where we will wait on caching
560 * block groups we need to first check if we're doing a fast load here,
561 * so we can wait for it to finish, otherwise we could end up allocating
562 * from a block group who's cache gets evicted for one reason or
563 * another.
564 */
565 while (cache->cached == BTRFS_CACHE_FAST) {
566 struct btrfs_caching_control *ctl;
567
568 ctl = cache->caching_ctl;
569 atomic_inc(&ctl->count);
570 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
571 spin_unlock(&cache->lock);
572
573 schedule();
574
575 finish_wait(&ctl->wait, &wait);
576 put_caching_control(ctl);
577 spin_lock(&cache->lock);
578 }
579
580 if (cache->cached != BTRFS_CACHE_NO) {
581 spin_unlock(&cache->lock);
582 kfree(caching_ctl);
583 return 0;
584 }
585 WARN_ON(cache->caching_ctl);
586 cache->caching_ctl = caching_ctl;
587 cache->cached = BTRFS_CACHE_FAST;
588 spin_unlock(&cache->lock);
589
590 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
591 mutex_lock(&caching_ctl->mutex);
592 ret = load_free_space_cache(fs_info, cache);
593
594 spin_lock(&cache->lock);
595 if (ret == 1) {
596 cache->caching_ctl = NULL;
597 cache->cached = BTRFS_CACHE_FINISHED;
598 cache->last_byte_to_unpin = (u64)-1;
599 caching_ctl->progress = (u64)-1;
600 } else {
601 if (load_cache_only) {
602 cache->caching_ctl = NULL;
603 cache->cached = BTRFS_CACHE_NO;
604 } else {
605 cache->cached = BTRFS_CACHE_STARTED;
606 cache->has_caching_ctl = 1;
607 }
608 }
609 spin_unlock(&cache->lock);
610 mutex_unlock(&caching_ctl->mutex);
611
612 wake_up(&caching_ctl->wait);
613 if (ret == 1) {
614 put_caching_control(caching_ctl);
615 free_excluded_extents(fs_info->extent_root, cache);
616 return 0;
617 }
618 } else {
619 /*
620 * We are not going to do the fast caching, set cached to the
621 * appropriate value and wakeup any waiters.
622 */
623 spin_lock(&cache->lock);
624 if (load_cache_only) {
625 cache->caching_ctl = NULL;
626 cache->cached = BTRFS_CACHE_NO;
627 } else {
628 cache->cached = BTRFS_CACHE_STARTED;
629 cache->has_caching_ctl = 1;
630 }
631 spin_unlock(&cache->lock);
632 wake_up(&caching_ctl->wait);
633 }
634
635 if (load_cache_only) {
636 put_caching_control(caching_ctl);
637 return 0;
638 }
639
640 down_write(&fs_info->commit_root_sem);
641 atomic_inc(&caching_ctl->count);
642 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
643 up_write(&fs_info->commit_root_sem);
644
645 btrfs_get_block_group(cache);
646
647 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
648
649 return ret;
650 }
651
652 /*
653 * return the block group that starts at or after bytenr
654 */
655 static struct btrfs_block_group_cache *
656 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
657 {
658 struct btrfs_block_group_cache *cache;
659
660 cache = block_group_cache_tree_search(info, bytenr, 0);
661
662 return cache;
663 }
664
665 /*
666 * return the block group that contains the given bytenr
667 */
668 struct btrfs_block_group_cache *btrfs_lookup_block_group(
669 struct btrfs_fs_info *info,
670 u64 bytenr)
671 {
672 struct btrfs_block_group_cache *cache;
673
674 cache = block_group_cache_tree_search(info, bytenr, 1);
675
676 return cache;
677 }
678
679 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
680 u64 flags)
681 {
682 struct list_head *head = &info->space_info;
683 struct btrfs_space_info *found;
684
685 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
686
687 rcu_read_lock();
688 list_for_each_entry_rcu(found, head, list) {
689 if (found->flags & flags) {
690 rcu_read_unlock();
691 return found;
692 }
693 }
694 rcu_read_unlock();
695 return NULL;
696 }
697
698 /*
699 * after adding space to the filesystem, we need to clear the full flags
700 * on all the space infos.
701 */
702 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
703 {
704 struct list_head *head = &info->space_info;
705 struct btrfs_space_info *found;
706
707 rcu_read_lock();
708 list_for_each_entry_rcu(found, head, list)
709 found->full = 0;
710 rcu_read_unlock();
711 }
712
713 /* simple helper to search for an existing data extent at a given offset */
714 int btrfs_lookup_data_extent(struct btrfs_root *root, u64 start, u64 len)
715 {
716 int ret;
717 struct btrfs_key key;
718 struct btrfs_path *path;
719
720 path = btrfs_alloc_path();
721 if (!path)
722 return -ENOMEM;
723
724 key.objectid = start;
725 key.offset = len;
726 key.type = BTRFS_EXTENT_ITEM_KEY;
727 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
728 0, 0);
729 btrfs_free_path(path);
730 return ret;
731 }
732
733 /*
734 * helper function to lookup reference count and flags of a tree block.
735 *
736 * the head node for delayed ref is used to store the sum of all the
737 * reference count modifications queued up in the rbtree. the head
738 * node may also store the extent flags to set. This way you can check
739 * to see what the reference count and extent flags would be if all of
740 * the delayed refs are not processed.
741 */
742 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
743 struct btrfs_root *root, u64 bytenr,
744 u64 offset, int metadata, u64 *refs, u64 *flags)
745 {
746 struct btrfs_delayed_ref_head *head;
747 struct btrfs_delayed_ref_root *delayed_refs;
748 struct btrfs_path *path;
749 struct btrfs_extent_item *ei;
750 struct extent_buffer *leaf;
751 struct btrfs_key key;
752 u32 item_size;
753 u64 num_refs;
754 u64 extent_flags;
755 int ret;
756
757 /*
758 * If we don't have skinny metadata, don't bother doing anything
759 * different
760 */
761 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
762 offset = root->nodesize;
763 metadata = 0;
764 }
765
766 path = btrfs_alloc_path();
767 if (!path)
768 return -ENOMEM;
769
770 if (!trans) {
771 path->skip_locking = 1;
772 path->search_commit_root = 1;
773 }
774
775 search_again:
776 key.objectid = bytenr;
777 key.offset = offset;
778 if (metadata)
779 key.type = BTRFS_METADATA_ITEM_KEY;
780 else
781 key.type = BTRFS_EXTENT_ITEM_KEY;
782
783 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
784 &key, path, 0, 0);
785 if (ret < 0)
786 goto out_free;
787
788 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
789 if (path->slots[0]) {
790 path->slots[0]--;
791 btrfs_item_key_to_cpu(path->nodes[0], &key,
792 path->slots[0]);
793 if (key.objectid == bytenr &&
794 key.type == BTRFS_EXTENT_ITEM_KEY &&
795 key.offset == root->nodesize)
796 ret = 0;
797 }
798 }
799
800 if (ret == 0) {
801 leaf = path->nodes[0];
802 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
803 if (item_size >= sizeof(*ei)) {
804 ei = btrfs_item_ptr(leaf, path->slots[0],
805 struct btrfs_extent_item);
806 num_refs = btrfs_extent_refs(leaf, ei);
807 extent_flags = btrfs_extent_flags(leaf, ei);
808 } else {
809 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
810 struct btrfs_extent_item_v0 *ei0;
811 BUG_ON(item_size != sizeof(*ei0));
812 ei0 = btrfs_item_ptr(leaf, path->slots[0],
813 struct btrfs_extent_item_v0);
814 num_refs = btrfs_extent_refs_v0(leaf, ei0);
815 /* FIXME: this isn't correct for data */
816 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
817 #else
818 BUG();
819 #endif
820 }
821 BUG_ON(num_refs == 0);
822 } else {
823 num_refs = 0;
824 extent_flags = 0;
825 ret = 0;
826 }
827
828 if (!trans)
829 goto out;
830
831 delayed_refs = &trans->transaction->delayed_refs;
832 spin_lock(&delayed_refs->lock);
833 head = btrfs_find_delayed_ref_head(trans, bytenr);
834 if (head) {
835 if (!mutex_trylock(&head->mutex)) {
836 atomic_inc(&head->node.refs);
837 spin_unlock(&delayed_refs->lock);
838
839 btrfs_release_path(path);
840
841 /*
842 * Mutex was contended, block until it's released and try
843 * again
844 */
845 mutex_lock(&head->mutex);
846 mutex_unlock(&head->mutex);
847 btrfs_put_delayed_ref(&head->node);
848 goto search_again;
849 }
850 spin_lock(&head->lock);
851 if (head->extent_op && head->extent_op->update_flags)
852 extent_flags |= head->extent_op->flags_to_set;
853 else
854 BUG_ON(num_refs == 0);
855
856 num_refs += head->node.ref_mod;
857 spin_unlock(&head->lock);
858 mutex_unlock(&head->mutex);
859 }
860 spin_unlock(&delayed_refs->lock);
861 out:
862 WARN_ON(num_refs == 0);
863 if (refs)
864 *refs = num_refs;
865 if (flags)
866 *flags = extent_flags;
867 out_free:
868 btrfs_free_path(path);
869 return ret;
870 }
871
872 /*
873 * Back reference rules. Back refs have three main goals:
874 *
875 * 1) differentiate between all holders of references to an extent so that
876 * when a reference is dropped we can make sure it was a valid reference
877 * before freeing the extent.
878 *
879 * 2) Provide enough information to quickly find the holders of an extent
880 * if we notice a given block is corrupted or bad.
881 *
882 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
883 * maintenance. This is actually the same as #2, but with a slightly
884 * different use case.
885 *
886 * There are two kinds of back refs. The implicit back refs is optimized
887 * for pointers in non-shared tree blocks. For a given pointer in a block,
888 * back refs of this kind provide information about the block's owner tree
889 * and the pointer's key. These information allow us to find the block by
890 * b-tree searching. The full back refs is for pointers in tree blocks not
891 * referenced by their owner trees. The location of tree block is recorded
892 * in the back refs. Actually the full back refs is generic, and can be
893 * used in all cases the implicit back refs is used. The major shortcoming
894 * of the full back refs is its overhead. Every time a tree block gets
895 * COWed, we have to update back refs entry for all pointers in it.
896 *
897 * For a newly allocated tree block, we use implicit back refs for
898 * pointers in it. This means most tree related operations only involve
899 * implicit back refs. For a tree block created in old transaction, the
900 * only way to drop a reference to it is COW it. So we can detect the
901 * event that tree block loses its owner tree's reference and do the
902 * back refs conversion.
903 *
904 * When a tree block is COW'd through a tree, there are four cases:
905 *
906 * The reference count of the block is one and the tree is the block's
907 * owner tree. Nothing to do in this case.
908 *
909 * The reference count of the block is one and the tree is not the
910 * block's owner tree. In this case, full back refs is used for pointers
911 * in the block. Remove these full back refs, add implicit back refs for
912 * every pointers in the new block.
913 *
914 * The reference count of the block is greater than one and the tree is
915 * the block's owner tree. In this case, implicit back refs is used for
916 * pointers in the block. Add full back refs for every pointers in the
917 * block, increase lower level extents' reference counts. The original
918 * implicit back refs are entailed to the new block.
919 *
920 * The reference count of the block is greater than one and the tree is
921 * not the block's owner tree. Add implicit back refs for every pointer in
922 * the new block, increase lower level extents' reference count.
923 *
924 * Back Reference Key composing:
925 *
926 * The key objectid corresponds to the first byte in the extent,
927 * The key type is used to differentiate between types of back refs.
928 * There are different meanings of the key offset for different types
929 * of back refs.
930 *
931 * File extents can be referenced by:
932 *
933 * - multiple snapshots, subvolumes, or different generations in one subvol
934 * - different files inside a single subvolume
935 * - different offsets inside a file (bookend extents in file.c)
936 *
937 * The extent ref structure for the implicit back refs has fields for:
938 *
939 * - Objectid of the subvolume root
940 * - objectid of the file holding the reference
941 * - original offset in the file
942 * - how many bookend extents
943 *
944 * The key offset for the implicit back refs is hash of the first
945 * three fields.
946 *
947 * The extent ref structure for the full back refs has field for:
948 *
949 * - number of pointers in the tree leaf
950 *
951 * The key offset for the implicit back refs is the first byte of
952 * the tree leaf
953 *
954 * When a file extent is allocated, The implicit back refs is used.
955 * the fields are filled in:
956 *
957 * (root_key.objectid, inode objectid, offset in file, 1)
958 *
959 * When a file extent is removed file truncation, we find the
960 * corresponding implicit back refs and check the following fields:
961 *
962 * (btrfs_header_owner(leaf), inode objectid, offset in file)
963 *
964 * Btree extents can be referenced by:
965 *
966 * - Different subvolumes
967 *
968 * Both the implicit back refs and the full back refs for tree blocks
969 * only consist of key. The key offset for the implicit back refs is
970 * objectid of block's owner tree. The key offset for the full back refs
971 * is the first byte of parent block.
972 *
973 * When implicit back refs is used, information about the lowest key and
974 * level of the tree block are required. These information are stored in
975 * tree block info structure.
976 */
977
978 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
979 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
980 struct btrfs_root *root,
981 struct btrfs_path *path,
982 u64 owner, u32 extra_size)
983 {
984 struct btrfs_extent_item *item;
985 struct btrfs_extent_item_v0 *ei0;
986 struct btrfs_extent_ref_v0 *ref0;
987 struct btrfs_tree_block_info *bi;
988 struct extent_buffer *leaf;
989 struct btrfs_key key;
990 struct btrfs_key found_key;
991 u32 new_size = sizeof(*item);
992 u64 refs;
993 int ret;
994
995 leaf = path->nodes[0];
996 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
997
998 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
999 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1000 struct btrfs_extent_item_v0);
1001 refs = btrfs_extent_refs_v0(leaf, ei0);
1002
1003 if (owner == (u64)-1) {
1004 while (1) {
1005 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1006 ret = btrfs_next_leaf(root, path);
1007 if (ret < 0)
1008 return ret;
1009 BUG_ON(ret > 0); /* Corruption */
1010 leaf = path->nodes[0];
1011 }
1012 btrfs_item_key_to_cpu(leaf, &found_key,
1013 path->slots[0]);
1014 BUG_ON(key.objectid != found_key.objectid);
1015 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1016 path->slots[0]++;
1017 continue;
1018 }
1019 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1020 struct btrfs_extent_ref_v0);
1021 owner = btrfs_ref_objectid_v0(leaf, ref0);
1022 break;
1023 }
1024 }
1025 btrfs_release_path(path);
1026
1027 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1028 new_size += sizeof(*bi);
1029
1030 new_size -= sizeof(*ei0);
1031 ret = btrfs_search_slot(trans, root, &key, path,
1032 new_size + extra_size, 1);
1033 if (ret < 0)
1034 return ret;
1035 BUG_ON(ret); /* Corruption */
1036
1037 btrfs_extend_item(root, path, new_size);
1038
1039 leaf = path->nodes[0];
1040 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1041 btrfs_set_extent_refs(leaf, item, refs);
1042 /* FIXME: get real generation */
1043 btrfs_set_extent_generation(leaf, item, 0);
1044 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1045 btrfs_set_extent_flags(leaf, item,
1046 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1047 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1048 bi = (struct btrfs_tree_block_info *)(item + 1);
1049 /* FIXME: get first key of the block */
1050 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1051 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1052 } else {
1053 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1054 }
1055 btrfs_mark_buffer_dirty(leaf);
1056 return 0;
1057 }
1058 #endif
1059
1060 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1061 {
1062 u32 high_crc = ~(u32)0;
1063 u32 low_crc = ~(u32)0;
1064 __le64 lenum;
1065
1066 lenum = cpu_to_le64(root_objectid);
1067 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1068 lenum = cpu_to_le64(owner);
1069 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1070 lenum = cpu_to_le64(offset);
1071 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1072
1073 return ((u64)high_crc << 31) ^ (u64)low_crc;
1074 }
1075
1076 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1077 struct btrfs_extent_data_ref *ref)
1078 {
1079 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1080 btrfs_extent_data_ref_objectid(leaf, ref),
1081 btrfs_extent_data_ref_offset(leaf, ref));
1082 }
1083
1084 static int match_extent_data_ref(struct extent_buffer *leaf,
1085 struct btrfs_extent_data_ref *ref,
1086 u64 root_objectid, u64 owner, u64 offset)
1087 {
1088 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1089 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1090 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1091 return 0;
1092 return 1;
1093 }
1094
1095 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1096 struct btrfs_root *root,
1097 struct btrfs_path *path,
1098 u64 bytenr, u64 parent,
1099 u64 root_objectid,
1100 u64 owner, u64 offset)
1101 {
1102 struct btrfs_key key;
1103 struct btrfs_extent_data_ref *ref;
1104 struct extent_buffer *leaf;
1105 u32 nritems;
1106 int ret;
1107 int recow;
1108 int err = -ENOENT;
1109
1110 key.objectid = bytenr;
1111 if (parent) {
1112 key.type = BTRFS_SHARED_DATA_REF_KEY;
1113 key.offset = parent;
1114 } else {
1115 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1116 key.offset = hash_extent_data_ref(root_objectid,
1117 owner, offset);
1118 }
1119 again:
1120 recow = 0;
1121 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1122 if (ret < 0) {
1123 err = ret;
1124 goto fail;
1125 }
1126
1127 if (parent) {
1128 if (!ret)
1129 return 0;
1130 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1131 key.type = BTRFS_EXTENT_REF_V0_KEY;
1132 btrfs_release_path(path);
1133 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1134 if (ret < 0) {
1135 err = ret;
1136 goto fail;
1137 }
1138 if (!ret)
1139 return 0;
1140 #endif
1141 goto fail;
1142 }
1143
1144 leaf = path->nodes[0];
1145 nritems = btrfs_header_nritems(leaf);
1146 while (1) {
1147 if (path->slots[0] >= nritems) {
1148 ret = btrfs_next_leaf(root, path);
1149 if (ret < 0)
1150 err = ret;
1151 if (ret)
1152 goto fail;
1153
1154 leaf = path->nodes[0];
1155 nritems = btrfs_header_nritems(leaf);
1156 recow = 1;
1157 }
1158
1159 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1160 if (key.objectid != bytenr ||
1161 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1162 goto fail;
1163
1164 ref = btrfs_item_ptr(leaf, path->slots[0],
1165 struct btrfs_extent_data_ref);
1166
1167 if (match_extent_data_ref(leaf, ref, root_objectid,
1168 owner, offset)) {
1169 if (recow) {
1170 btrfs_release_path(path);
1171 goto again;
1172 }
1173 err = 0;
1174 break;
1175 }
1176 path->slots[0]++;
1177 }
1178 fail:
1179 return err;
1180 }
1181
1182 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1183 struct btrfs_root *root,
1184 struct btrfs_path *path,
1185 u64 bytenr, u64 parent,
1186 u64 root_objectid, u64 owner,
1187 u64 offset, int refs_to_add)
1188 {
1189 struct btrfs_key key;
1190 struct extent_buffer *leaf;
1191 u32 size;
1192 u32 num_refs;
1193 int ret;
1194
1195 key.objectid = bytenr;
1196 if (parent) {
1197 key.type = BTRFS_SHARED_DATA_REF_KEY;
1198 key.offset = parent;
1199 size = sizeof(struct btrfs_shared_data_ref);
1200 } else {
1201 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1202 key.offset = hash_extent_data_ref(root_objectid,
1203 owner, offset);
1204 size = sizeof(struct btrfs_extent_data_ref);
1205 }
1206
1207 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1208 if (ret && ret != -EEXIST)
1209 goto fail;
1210
1211 leaf = path->nodes[0];
1212 if (parent) {
1213 struct btrfs_shared_data_ref *ref;
1214 ref = btrfs_item_ptr(leaf, path->slots[0],
1215 struct btrfs_shared_data_ref);
1216 if (ret == 0) {
1217 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1218 } else {
1219 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1220 num_refs += refs_to_add;
1221 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1222 }
1223 } else {
1224 struct btrfs_extent_data_ref *ref;
1225 while (ret == -EEXIST) {
1226 ref = btrfs_item_ptr(leaf, path->slots[0],
1227 struct btrfs_extent_data_ref);
1228 if (match_extent_data_ref(leaf, ref, root_objectid,
1229 owner, offset))
1230 break;
1231 btrfs_release_path(path);
1232 key.offset++;
1233 ret = btrfs_insert_empty_item(trans, root, path, &key,
1234 size);
1235 if (ret && ret != -EEXIST)
1236 goto fail;
1237
1238 leaf = path->nodes[0];
1239 }
1240 ref = btrfs_item_ptr(leaf, path->slots[0],
1241 struct btrfs_extent_data_ref);
1242 if (ret == 0) {
1243 btrfs_set_extent_data_ref_root(leaf, ref,
1244 root_objectid);
1245 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1246 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1247 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1248 } else {
1249 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1250 num_refs += refs_to_add;
1251 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1252 }
1253 }
1254 btrfs_mark_buffer_dirty(leaf);
1255 ret = 0;
1256 fail:
1257 btrfs_release_path(path);
1258 return ret;
1259 }
1260
1261 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1262 struct btrfs_root *root,
1263 struct btrfs_path *path,
1264 int refs_to_drop, int *last_ref)
1265 {
1266 struct btrfs_key key;
1267 struct btrfs_extent_data_ref *ref1 = NULL;
1268 struct btrfs_shared_data_ref *ref2 = NULL;
1269 struct extent_buffer *leaf;
1270 u32 num_refs = 0;
1271 int ret = 0;
1272
1273 leaf = path->nodes[0];
1274 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1275
1276 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1277 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1278 struct btrfs_extent_data_ref);
1279 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1280 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1281 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1282 struct btrfs_shared_data_ref);
1283 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1284 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1285 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1286 struct btrfs_extent_ref_v0 *ref0;
1287 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1288 struct btrfs_extent_ref_v0);
1289 num_refs = btrfs_ref_count_v0(leaf, ref0);
1290 #endif
1291 } else {
1292 BUG();
1293 }
1294
1295 BUG_ON(num_refs < refs_to_drop);
1296 num_refs -= refs_to_drop;
1297
1298 if (num_refs == 0) {
1299 ret = btrfs_del_item(trans, root, path);
1300 *last_ref = 1;
1301 } else {
1302 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1303 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1304 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1305 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1306 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1307 else {
1308 struct btrfs_extent_ref_v0 *ref0;
1309 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1310 struct btrfs_extent_ref_v0);
1311 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1312 }
1313 #endif
1314 btrfs_mark_buffer_dirty(leaf);
1315 }
1316 return ret;
1317 }
1318
1319 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1320 struct btrfs_path *path,
1321 struct btrfs_extent_inline_ref *iref)
1322 {
1323 struct btrfs_key key;
1324 struct extent_buffer *leaf;
1325 struct btrfs_extent_data_ref *ref1;
1326 struct btrfs_shared_data_ref *ref2;
1327 u32 num_refs = 0;
1328
1329 leaf = path->nodes[0];
1330 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1331 if (iref) {
1332 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1333 BTRFS_EXTENT_DATA_REF_KEY) {
1334 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1335 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1336 } else {
1337 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1338 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1339 }
1340 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1341 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1342 struct btrfs_extent_data_ref);
1343 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1344 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1345 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1346 struct btrfs_shared_data_ref);
1347 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1348 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1349 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1350 struct btrfs_extent_ref_v0 *ref0;
1351 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1352 struct btrfs_extent_ref_v0);
1353 num_refs = btrfs_ref_count_v0(leaf, ref0);
1354 #endif
1355 } else {
1356 WARN_ON(1);
1357 }
1358 return num_refs;
1359 }
1360
1361 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1362 struct btrfs_root *root,
1363 struct btrfs_path *path,
1364 u64 bytenr, u64 parent,
1365 u64 root_objectid)
1366 {
1367 struct btrfs_key key;
1368 int ret;
1369
1370 key.objectid = bytenr;
1371 if (parent) {
1372 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1373 key.offset = parent;
1374 } else {
1375 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1376 key.offset = root_objectid;
1377 }
1378
1379 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1380 if (ret > 0)
1381 ret = -ENOENT;
1382 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1383 if (ret == -ENOENT && parent) {
1384 btrfs_release_path(path);
1385 key.type = BTRFS_EXTENT_REF_V0_KEY;
1386 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1387 if (ret > 0)
1388 ret = -ENOENT;
1389 }
1390 #endif
1391 return ret;
1392 }
1393
1394 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1395 struct btrfs_root *root,
1396 struct btrfs_path *path,
1397 u64 bytenr, u64 parent,
1398 u64 root_objectid)
1399 {
1400 struct btrfs_key key;
1401 int ret;
1402
1403 key.objectid = bytenr;
1404 if (parent) {
1405 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1406 key.offset = parent;
1407 } else {
1408 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1409 key.offset = root_objectid;
1410 }
1411
1412 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1413 btrfs_release_path(path);
1414 return ret;
1415 }
1416
1417 static inline int extent_ref_type(u64 parent, u64 owner)
1418 {
1419 int type;
1420 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1421 if (parent > 0)
1422 type = BTRFS_SHARED_BLOCK_REF_KEY;
1423 else
1424 type = BTRFS_TREE_BLOCK_REF_KEY;
1425 } else {
1426 if (parent > 0)
1427 type = BTRFS_SHARED_DATA_REF_KEY;
1428 else
1429 type = BTRFS_EXTENT_DATA_REF_KEY;
1430 }
1431 return type;
1432 }
1433
1434 static int find_next_key(struct btrfs_path *path, int level,
1435 struct btrfs_key *key)
1436
1437 {
1438 for (; level < BTRFS_MAX_LEVEL; level++) {
1439 if (!path->nodes[level])
1440 break;
1441 if (path->slots[level] + 1 >=
1442 btrfs_header_nritems(path->nodes[level]))
1443 continue;
1444 if (level == 0)
1445 btrfs_item_key_to_cpu(path->nodes[level], key,
1446 path->slots[level] + 1);
1447 else
1448 btrfs_node_key_to_cpu(path->nodes[level], key,
1449 path->slots[level] + 1);
1450 return 0;
1451 }
1452 return 1;
1453 }
1454
1455 /*
1456 * look for inline back ref. if back ref is found, *ref_ret is set
1457 * to the address of inline back ref, and 0 is returned.
1458 *
1459 * if back ref isn't found, *ref_ret is set to the address where it
1460 * should be inserted, and -ENOENT is returned.
1461 *
1462 * if insert is true and there are too many inline back refs, the path
1463 * points to the extent item, and -EAGAIN is returned.
1464 *
1465 * NOTE: inline back refs are ordered in the same way that back ref
1466 * items in the tree are ordered.
1467 */
1468 static noinline_for_stack
1469 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1470 struct btrfs_root *root,
1471 struct btrfs_path *path,
1472 struct btrfs_extent_inline_ref **ref_ret,
1473 u64 bytenr, u64 num_bytes,
1474 u64 parent, u64 root_objectid,
1475 u64 owner, u64 offset, int insert)
1476 {
1477 struct btrfs_key key;
1478 struct extent_buffer *leaf;
1479 struct btrfs_extent_item *ei;
1480 struct btrfs_extent_inline_ref *iref;
1481 u64 flags;
1482 u64 item_size;
1483 unsigned long ptr;
1484 unsigned long end;
1485 int extra_size;
1486 int type;
1487 int want;
1488 int ret;
1489 int err = 0;
1490 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1491 SKINNY_METADATA);
1492
1493 key.objectid = bytenr;
1494 key.type = BTRFS_EXTENT_ITEM_KEY;
1495 key.offset = num_bytes;
1496
1497 want = extent_ref_type(parent, owner);
1498 if (insert) {
1499 extra_size = btrfs_extent_inline_ref_size(want);
1500 path->keep_locks = 1;
1501 } else
1502 extra_size = -1;
1503
1504 /*
1505 * Owner is our parent level, so we can just add one to get the level
1506 * for the block we are interested in.
1507 */
1508 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1509 key.type = BTRFS_METADATA_ITEM_KEY;
1510 key.offset = owner;
1511 }
1512
1513 again:
1514 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1515 if (ret < 0) {
1516 err = ret;
1517 goto out;
1518 }
1519
1520 /*
1521 * We may be a newly converted file system which still has the old fat
1522 * extent entries for metadata, so try and see if we have one of those.
1523 */
1524 if (ret > 0 && skinny_metadata) {
1525 skinny_metadata = false;
1526 if (path->slots[0]) {
1527 path->slots[0]--;
1528 btrfs_item_key_to_cpu(path->nodes[0], &key,
1529 path->slots[0]);
1530 if (key.objectid == bytenr &&
1531 key.type == BTRFS_EXTENT_ITEM_KEY &&
1532 key.offset == num_bytes)
1533 ret = 0;
1534 }
1535 if (ret) {
1536 key.objectid = bytenr;
1537 key.type = BTRFS_EXTENT_ITEM_KEY;
1538 key.offset = num_bytes;
1539 btrfs_release_path(path);
1540 goto again;
1541 }
1542 }
1543
1544 if (ret && !insert) {
1545 err = -ENOENT;
1546 goto out;
1547 } else if (WARN_ON(ret)) {
1548 err = -EIO;
1549 goto out;
1550 }
1551
1552 leaf = path->nodes[0];
1553 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1554 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1555 if (item_size < sizeof(*ei)) {
1556 if (!insert) {
1557 err = -ENOENT;
1558 goto out;
1559 }
1560 ret = convert_extent_item_v0(trans, root, path, owner,
1561 extra_size);
1562 if (ret < 0) {
1563 err = ret;
1564 goto out;
1565 }
1566 leaf = path->nodes[0];
1567 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1568 }
1569 #endif
1570 BUG_ON(item_size < sizeof(*ei));
1571
1572 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1573 flags = btrfs_extent_flags(leaf, ei);
1574
1575 ptr = (unsigned long)(ei + 1);
1576 end = (unsigned long)ei + item_size;
1577
1578 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1579 ptr += sizeof(struct btrfs_tree_block_info);
1580 BUG_ON(ptr > end);
1581 }
1582
1583 err = -ENOENT;
1584 while (1) {
1585 if (ptr >= end) {
1586 WARN_ON(ptr > end);
1587 break;
1588 }
1589 iref = (struct btrfs_extent_inline_ref *)ptr;
1590 type = btrfs_extent_inline_ref_type(leaf, iref);
1591 if (want < type)
1592 break;
1593 if (want > type) {
1594 ptr += btrfs_extent_inline_ref_size(type);
1595 continue;
1596 }
1597
1598 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1599 struct btrfs_extent_data_ref *dref;
1600 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1601 if (match_extent_data_ref(leaf, dref, root_objectid,
1602 owner, offset)) {
1603 err = 0;
1604 break;
1605 }
1606 if (hash_extent_data_ref_item(leaf, dref) <
1607 hash_extent_data_ref(root_objectid, owner, offset))
1608 break;
1609 } else {
1610 u64 ref_offset;
1611 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1612 if (parent > 0) {
1613 if (parent == ref_offset) {
1614 err = 0;
1615 break;
1616 }
1617 if (ref_offset < parent)
1618 break;
1619 } else {
1620 if (root_objectid == ref_offset) {
1621 err = 0;
1622 break;
1623 }
1624 if (ref_offset < root_objectid)
1625 break;
1626 }
1627 }
1628 ptr += btrfs_extent_inline_ref_size(type);
1629 }
1630 if (err == -ENOENT && insert) {
1631 if (item_size + extra_size >=
1632 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1633 err = -EAGAIN;
1634 goto out;
1635 }
1636 /*
1637 * To add new inline back ref, we have to make sure
1638 * there is no corresponding back ref item.
1639 * For simplicity, we just do not add new inline back
1640 * ref if there is any kind of item for this block
1641 */
1642 if (find_next_key(path, 0, &key) == 0 &&
1643 key.objectid == bytenr &&
1644 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1645 err = -EAGAIN;
1646 goto out;
1647 }
1648 }
1649 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1650 out:
1651 if (insert) {
1652 path->keep_locks = 0;
1653 btrfs_unlock_up_safe(path, 1);
1654 }
1655 return err;
1656 }
1657
1658 /*
1659 * helper to add new inline back ref
1660 */
1661 static noinline_for_stack
1662 void setup_inline_extent_backref(struct btrfs_root *root,
1663 struct btrfs_path *path,
1664 struct btrfs_extent_inline_ref *iref,
1665 u64 parent, u64 root_objectid,
1666 u64 owner, u64 offset, int refs_to_add,
1667 struct btrfs_delayed_extent_op *extent_op)
1668 {
1669 struct extent_buffer *leaf;
1670 struct btrfs_extent_item *ei;
1671 unsigned long ptr;
1672 unsigned long end;
1673 unsigned long item_offset;
1674 u64 refs;
1675 int size;
1676 int type;
1677
1678 leaf = path->nodes[0];
1679 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1680 item_offset = (unsigned long)iref - (unsigned long)ei;
1681
1682 type = extent_ref_type(parent, owner);
1683 size = btrfs_extent_inline_ref_size(type);
1684
1685 btrfs_extend_item(root, path, size);
1686
1687 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1688 refs = btrfs_extent_refs(leaf, ei);
1689 refs += refs_to_add;
1690 btrfs_set_extent_refs(leaf, ei, refs);
1691 if (extent_op)
1692 __run_delayed_extent_op(extent_op, leaf, ei);
1693
1694 ptr = (unsigned long)ei + item_offset;
1695 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1696 if (ptr < end - size)
1697 memmove_extent_buffer(leaf, ptr + size, ptr,
1698 end - size - ptr);
1699
1700 iref = (struct btrfs_extent_inline_ref *)ptr;
1701 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1702 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1703 struct btrfs_extent_data_ref *dref;
1704 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1705 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1706 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1707 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1708 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1709 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1710 struct btrfs_shared_data_ref *sref;
1711 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1712 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1713 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1714 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1715 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1716 } else {
1717 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1718 }
1719 btrfs_mark_buffer_dirty(leaf);
1720 }
1721
1722 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1723 struct btrfs_root *root,
1724 struct btrfs_path *path,
1725 struct btrfs_extent_inline_ref **ref_ret,
1726 u64 bytenr, u64 num_bytes, u64 parent,
1727 u64 root_objectid, u64 owner, u64 offset)
1728 {
1729 int ret;
1730
1731 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1732 bytenr, num_bytes, parent,
1733 root_objectid, owner, offset, 0);
1734 if (ret != -ENOENT)
1735 return ret;
1736
1737 btrfs_release_path(path);
1738 *ref_ret = NULL;
1739
1740 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1741 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1742 root_objectid);
1743 } else {
1744 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1745 root_objectid, owner, offset);
1746 }
1747 return ret;
1748 }
1749
1750 /*
1751 * helper to update/remove inline back ref
1752 */
1753 static noinline_for_stack
1754 void update_inline_extent_backref(struct btrfs_root *root,
1755 struct btrfs_path *path,
1756 struct btrfs_extent_inline_ref *iref,
1757 int refs_to_mod,
1758 struct btrfs_delayed_extent_op *extent_op,
1759 int *last_ref)
1760 {
1761 struct extent_buffer *leaf;
1762 struct btrfs_extent_item *ei;
1763 struct btrfs_extent_data_ref *dref = NULL;
1764 struct btrfs_shared_data_ref *sref = NULL;
1765 unsigned long ptr;
1766 unsigned long end;
1767 u32 item_size;
1768 int size;
1769 int type;
1770 u64 refs;
1771
1772 leaf = path->nodes[0];
1773 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1774 refs = btrfs_extent_refs(leaf, ei);
1775 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1776 refs += refs_to_mod;
1777 btrfs_set_extent_refs(leaf, ei, refs);
1778 if (extent_op)
1779 __run_delayed_extent_op(extent_op, leaf, ei);
1780
1781 type = btrfs_extent_inline_ref_type(leaf, iref);
1782
1783 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1784 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1785 refs = btrfs_extent_data_ref_count(leaf, dref);
1786 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1787 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1788 refs = btrfs_shared_data_ref_count(leaf, sref);
1789 } else {
1790 refs = 1;
1791 BUG_ON(refs_to_mod != -1);
1792 }
1793
1794 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1795 refs += refs_to_mod;
1796
1797 if (refs > 0) {
1798 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1799 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1800 else
1801 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1802 } else {
1803 *last_ref = 1;
1804 size = btrfs_extent_inline_ref_size(type);
1805 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1806 ptr = (unsigned long)iref;
1807 end = (unsigned long)ei + item_size;
1808 if (ptr + size < end)
1809 memmove_extent_buffer(leaf, ptr, ptr + size,
1810 end - ptr - size);
1811 item_size -= size;
1812 btrfs_truncate_item(root, path, item_size, 1);
1813 }
1814 btrfs_mark_buffer_dirty(leaf);
1815 }
1816
1817 static noinline_for_stack
1818 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1819 struct btrfs_root *root,
1820 struct btrfs_path *path,
1821 u64 bytenr, u64 num_bytes, u64 parent,
1822 u64 root_objectid, u64 owner,
1823 u64 offset, int refs_to_add,
1824 struct btrfs_delayed_extent_op *extent_op)
1825 {
1826 struct btrfs_extent_inline_ref *iref;
1827 int ret;
1828
1829 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1830 bytenr, num_bytes, parent,
1831 root_objectid, owner, offset, 1);
1832 if (ret == 0) {
1833 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1834 update_inline_extent_backref(root, path, iref,
1835 refs_to_add, extent_op, NULL);
1836 } else if (ret == -ENOENT) {
1837 setup_inline_extent_backref(root, path, iref, parent,
1838 root_objectid, owner, offset,
1839 refs_to_add, extent_op);
1840 ret = 0;
1841 }
1842 return ret;
1843 }
1844
1845 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1846 struct btrfs_root *root,
1847 struct btrfs_path *path,
1848 u64 bytenr, u64 parent, u64 root_objectid,
1849 u64 owner, u64 offset, int refs_to_add)
1850 {
1851 int ret;
1852 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1853 BUG_ON(refs_to_add != 1);
1854 ret = insert_tree_block_ref(trans, root, path, bytenr,
1855 parent, root_objectid);
1856 } else {
1857 ret = insert_extent_data_ref(trans, root, path, bytenr,
1858 parent, root_objectid,
1859 owner, offset, refs_to_add);
1860 }
1861 return ret;
1862 }
1863
1864 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1865 struct btrfs_root *root,
1866 struct btrfs_path *path,
1867 struct btrfs_extent_inline_ref *iref,
1868 int refs_to_drop, int is_data, int *last_ref)
1869 {
1870 int ret = 0;
1871
1872 BUG_ON(!is_data && refs_to_drop != 1);
1873 if (iref) {
1874 update_inline_extent_backref(root, path, iref,
1875 -refs_to_drop, NULL, last_ref);
1876 } else if (is_data) {
1877 ret = remove_extent_data_ref(trans, root, path, refs_to_drop,
1878 last_ref);
1879 } else {
1880 *last_ref = 1;
1881 ret = btrfs_del_item(trans, root, path);
1882 }
1883 return ret;
1884 }
1885
1886 static int btrfs_issue_discard(struct block_device *bdev,
1887 u64 start, u64 len)
1888 {
1889 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1890 }
1891
1892 int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1893 u64 num_bytes, u64 *actual_bytes)
1894 {
1895 int ret;
1896 u64 discarded_bytes = 0;
1897 struct btrfs_bio *bbio = NULL;
1898
1899
1900 /* Tell the block device(s) that the sectors can be discarded */
1901 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1902 bytenr, &num_bytes, &bbio, 0);
1903 /* Error condition is -ENOMEM */
1904 if (!ret) {
1905 struct btrfs_bio_stripe *stripe = bbio->stripes;
1906 int i;
1907
1908
1909 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1910 if (!stripe->dev->can_discard)
1911 continue;
1912
1913 ret = btrfs_issue_discard(stripe->dev->bdev,
1914 stripe->physical,
1915 stripe->length);
1916 if (!ret)
1917 discarded_bytes += stripe->length;
1918 else if (ret != -EOPNOTSUPP)
1919 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1920
1921 /*
1922 * Just in case we get back EOPNOTSUPP for some reason,
1923 * just ignore the return value so we don't screw up
1924 * people calling discard_extent.
1925 */
1926 ret = 0;
1927 }
1928 btrfs_put_bbio(bbio);
1929 }
1930
1931 if (actual_bytes)
1932 *actual_bytes = discarded_bytes;
1933
1934
1935 if (ret == -EOPNOTSUPP)
1936 ret = 0;
1937 return ret;
1938 }
1939
1940 /* Can return -ENOMEM */
1941 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1942 struct btrfs_root *root,
1943 u64 bytenr, u64 num_bytes, u64 parent,
1944 u64 root_objectid, u64 owner, u64 offset,
1945 int no_quota)
1946 {
1947 int ret;
1948 struct btrfs_fs_info *fs_info = root->fs_info;
1949
1950 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1951 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1952
1953 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1954 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1955 num_bytes,
1956 parent, root_objectid, (int)owner,
1957 BTRFS_ADD_DELAYED_REF, NULL, no_quota);
1958 } else {
1959 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1960 num_bytes,
1961 parent, root_objectid, owner, offset,
1962 BTRFS_ADD_DELAYED_REF, NULL, no_quota);
1963 }
1964 return ret;
1965 }
1966
1967 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1968 struct btrfs_root *root,
1969 struct btrfs_delayed_ref_node *node,
1970 u64 parent, u64 root_objectid,
1971 u64 owner, u64 offset, int refs_to_add,
1972 struct btrfs_delayed_extent_op *extent_op)
1973 {
1974 struct btrfs_fs_info *fs_info = root->fs_info;
1975 struct btrfs_path *path;
1976 struct extent_buffer *leaf;
1977 struct btrfs_extent_item *item;
1978 struct btrfs_key key;
1979 u64 bytenr = node->bytenr;
1980 u64 num_bytes = node->num_bytes;
1981 u64 refs;
1982 int ret;
1983 int no_quota = node->no_quota;
1984
1985 path = btrfs_alloc_path();
1986 if (!path)
1987 return -ENOMEM;
1988
1989 if (!is_fstree(root_objectid) || !root->fs_info->quota_enabled)
1990 no_quota = 1;
1991
1992 path->reada = 1;
1993 path->leave_spinning = 1;
1994 /* this will setup the path even if it fails to insert the back ref */
1995 ret = insert_inline_extent_backref(trans, fs_info->extent_root, path,
1996 bytenr, num_bytes, parent,
1997 root_objectid, owner, offset,
1998 refs_to_add, extent_op);
1999 if ((ret < 0 && ret != -EAGAIN) || !ret)
2000 goto out;
2001
2002 /*
2003 * Ok we had -EAGAIN which means we didn't have space to insert and
2004 * inline extent ref, so just update the reference count and add a
2005 * normal backref.
2006 */
2007 leaf = path->nodes[0];
2008 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2009 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2010 refs = btrfs_extent_refs(leaf, item);
2011 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2012 if (extent_op)
2013 __run_delayed_extent_op(extent_op, leaf, item);
2014
2015 btrfs_mark_buffer_dirty(leaf);
2016 btrfs_release_path(path);
2017
2018 path->reada = 1;
2019 path->leave_spinning = 1;
2020 /* now insert the actual backref */
2021 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2022 path, bytenr, parent, root_objectid,
2023 owner, offset, refs_to_add);
2024 if (ret)
2025 btrfs_abort_transaction(trans, root, ret);
2026 out:
2027 btrfs_free_path(path);
2028 return ret;
2029 }
2030
2031 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2032 struct btrfs_root *root,
2033 struct btrfs_delayed_ref_node *node,
2034 struct btrfs_delayed_extent_op *extent_op,
2035 int insert_reserved)
2036 {
2037 int ret = 0;
2038 struct btrfs_delayed_data_ref *ref;
2039 struct btrfs_key ins;
2040 u64 parent = 0;
2041 u64 ref_root = 0;
2042 u64 flags = 0;
2043
2044 ins.objectid = node->bytenr;
2045 ins.offset = node->num_bytes;
2046 ins.type = BTRFS_EXTENT_ITEM_KEY;
2047
2048 ref = btrfs_delayed_node_to_data_ref(node);
2049 trace_run_delayed_data_ref(node, ref, node->action);
2050
2051 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2052 parent = ref->parent;
2053 ref_root = ref->root;
2054
2055 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2056 if (extent_op)
2057 flags |= extent_op->flags_to_set;
2058 ret = alloc_reserved_file_extent(trans, root,
2059 parent, ref_root, flags,
2060 ref->objectid, ref->offset,
2061 &ins, node->ref_mod);
2062 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2063 ret = __btrfs_inc_extent_ref(trans, root, node, parent,
2064 ref_root, ref->objectid,
2065 ref->offset, node->ref_mod,
2066 extent_op);
2067 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2068 ret = __btrfs_free_extent(trans, root, node, parent,
2069 ref_root, ref->objectid,
2070 ref->offset, node->ref_mod,
2071 extent_op);
2072 } else {
2073 BUG();
2074 }
2075 return ret;
2076 }
2077
2078 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2079 struct extent_buffer *leaf,
2080 struct btrfs_extent_item *ei)
2081 {
2082 u64 flags = btrfs_extent_flags(leaf, ei);
2083 if (extent_op->update_flags) {
2084 flags |= extent_op->flags_to_set;
2085 btrfs_set_extent_flags(leaf, ei, flags);
2086 }
2087
2088 if (extent_op->update_key) {
2089 struct btrfs_tree_block_info *bi;
2090 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2091 bi = (struct btrfs_tree_block_info *)(ei + 1);
2092 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2093 }
2094 }
2095
2096 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2097 struct btrfs_root *root,
2098 struct btrfs_delayed_ref_node *node,
2099 struct btrfs_delayed_extent_op *extent_op)
2100 {
2101 struct btrfs_key key;
2102 struct btrfs_path *path;
2103 struct btrfs_extent_item *ei;
2104 struct extent_buffer *leaf;
2105 u32 item_size;
2106 int ret;
2107 int err = 0;
2108 int metadata = !extent_op->is_data;
2109
2110 if (trans->aborted)
2111 return 0;
2112
2113 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2114 metadata = 0;
2115
2116 path = btrfs_alloc_path();
2117 if (!path)
2118 return -ENOMEM;
2119
2120 key.objectid = node->bytenr;
2121
2122 if (metadata) {
2123 key.type = BTRFS_METADATA_ITEM_KEY;
2124 key.offset = extent_op->level;
2125 } else {
2126 key.type = BTRFS_EXTENT_ITEM_KEY;
2127 key.offset = node->num_bytes;
2128 }
2129
2130 again:
2131 path->reada = 1;
2132 path->leave_spinning = 1;
2133 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2134 path, 0, 1);
2135 if (ret < 0) {
2136 err = ret;
2137 goto out;
2138 }
2139 if (ret > 0) {
2140 if (metadata) {
2141 if (path->slots[0] > 0) {
2142 path->slots[0]--;
2143 btrfs_item_key_to_cpu(path->nodes[0], &key,
2144 path->slots[0]);
2145 if (key.objectid == node->bytenr &&
2146 key.type == BTRFS_EXTENT_ITEM_KEY &&
2147 key.offset == node->num_bytes)
2148 ret = 0;
2149 }
2150 if (ret > 0) {
2151 btrfs_release_path(path);
2152 metadata = 0;
2153
2154 key.objectid = node->bytenr;
2155 key.offset = node->num_bytes;
2156 key.type = BTRFS_EXTENT_ITEM_KEY;
2157 goto again;
2158 }
2159 } else {
2160 err = -EIO;
2161 goto out;
2162 }
2163 }
2164
2165 leaf = path->nodes[0];
2166 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2167 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2168 if (item_size < sizeof(*ei)) {
2169 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2170 path, (u64)-1, 0);
2171 if (ret < 0) {
2172 err = ret;
2173 goto out;
2174 }
2175 leaf = path->nodes[0];
2176 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2177 }
2178 #endif
2179 BUG_ON(item_size < sizeof(*ei));
2180 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2181 __run_delayed_extent_op(extent_op, leaf, ei);
2182
2183 btrfs_mark_buffer_dirty(leaf);
2184 out:
2185 btrfs_free_path(path);
2186 return err;
2187 }
2188
2189 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2190 struct btrfs_root *root,
2191 struct btrfs_delayed_ref_node *node,
2192 struct btrfs_delayed_extent_op *extent_op,
2193 int insert_reserved)
2194 {
2195 int ret = 0;
2196 struct btrfs_delayed_tree_ref *ref;
2197 struct btrfs_key ins;
2198 u64 parent = 0;
2199 u64 ref_root = 0;
2200 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2201 SKINNY_METADATA);
2202
2203 ref = btrfs_delayed_node_to_tree_ref(node);
2204 trace_run_delayed_tree_ref(node, ref, node->action);
2205
2206 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2207 parent = ref->parent;
2208 ref_root = ref->root;
2209
2210 ins.objectid = node->bytenr;
2211 if (skinny_metadata) {
2212 ins.offset = ref->level;
2213 ins.type = BTRFS_METADATA_ITEM_KEY;
2214 } else {
2215 ins.offset = node->num_bytes;
2216 ins.type = BTRFS_EXTENT_ITEM_KEY;
2217 }
2218
2219 BUG_ON(node->ref_mod != 1);
2220 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2221 BUG_ON(!extent_op || !extent_op->update_flags);
2222 ret = alloc_reserved_tree_block(trans, root,
2223 parent, ref_root,
2224 extent_op->flags_to_set,
2225 &extent_op->key,
2226 ref->level, &ins,
2227 node->no_quota);
2228 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2229 ret = __btrfs_inc_extent_ref(trans, root, node,
2230 parent, ref_root,
2231 ref->level, 0, 1,
2232 extent_op);
2233 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2234 ret = __btrfs_free_extent(trans, root, node,
2235 parent, ref_root,
2236 ref->level, 0, 1, extent_op);
2237 } else {
2238 BUG();
2239 }
2240 return ret;
2241 }
2242
2243 /* helper function to actually process a single delayed ref entry */
2244 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2245 struct btrfs_root *root,
2246 struct btrfs_delayed_ref_node *node,
2247 struct btrfs_delayed_extent_op *extent_op,
2248 int insert_reserved)
2249 {
2250 int ret = 0;
2251
2252 if (trans->aborted) {
2253 if (insert_reserved)
2254 btrfs_pin_extent(root, node->bytenr,
2255 node->num_bytes, 1);
2256 return 0;
2257 }
2258
2259 if (btrfs_delayed_ref_is_head(node)) {
2260 struct btrfs_delayed_ref_head *head;
2261 /*
2262 * we've hit the end of the chain and we were supposed
2263 * to insert this extent into the tree. But, it got
2264 * deleted before we ever needed to insert it, so all
2265 * we have to do is clean up the accounting
2266 */
2267 BUG_ON(extent_op);
2268 head = btrfs_delayed_node_to_head(node);
2269 trace_run_delayed_ref_head(node, head, node->action);
2270
2271 if (insert_reserved) {
2272 btrfs_pin_extent(root, node->bytenr,
2273 node->num_bytes, 1);
2274 if (head->is_data) {
2275 ret = btrfs_del_csums(trans, root,
2276 node->bytenr,
2277 node->num_bytes);
2278 }
2279 }
2280 return ret;
2281 }
2282
2283 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2284 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2285 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2286 insert_reserved);
2287 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2288 node->type == BTRFS_SHARED_DATA_REF_KEY)
2289 ret = run_delayed_data_ref(trans, root, node, extent_op,
2290 insert_reserved);
2291 else
2292 BUG();
2293 return ret;
2294 }
2295
2296 static inline struct btrfs_delayed_ref_node *
2297 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2298 {
2299 struct btrfs_delayed_ref_node *ref;
2300
2301 if (list_empty(&head->ref_list))
2302 return NULL;
2303
2304 /*
2305 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2306 * This is to prevent a ref count from going down to zero, which deletes
2307 * the extent item from the extent tree, when there still are references
2308 * to add, which would fail because they would not find the extent item.
2309 */
2310 list_for_each_entry(ref, &head->ref_list, list) {
2311 if (ref->action == BTRFS_ADD_DELAYED_REF)
2312 return ref;
2313 }
2314
2315 return list_entry(head->ref_list.next, struct btrfs_delayed_ref_node,
2316 list);
2317 }
2318
2319 /*
2320 * Returns 0 on success or if called with an already aborted transaction.
2321 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2322 */
2323 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2324 struct btrfs_root *root,
2325 unsigned long nr)
2326 {
2327 struct btrfs_delayed_ref_root *delayed_refs;
2328 struct btrfs_delayed_ref_node *ref;
2329 struct btrfs_delayed_ref_head *locked_ref = NULL;
2330 struct btrfs_delayed_extent_op *extent_op;
2331 struct btrfs_fs_info *fs_info = root->fs_info;
2332 ktime_t start = ktime_get();
2333 int ret;
2334 unsigned long count = 0;
2335 unsigned long actual_count = 0;
2336 int must_insert_reserved = 0;
2337
2338 delayed_refs = &trans->transaction->delayed_refs;
2339 while (1) {
2340 if (!locked_ref) {
2341 if (count >= nr)
2342 break;
2343
2344 spin_lock(&delayed_refs->lock);
2345 locked_ref = btrfs_select_ref_head(trans);
2346 if (!locked_ref) {
2347 spin_unlock(&delayed_refs->lock);
2348 break;
2349 }
2350
2351 /* grab the lock that says we are going to process
2352 * all the refs for this head */
2353 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2354 spin_unlock(&delayed_refs->lock);
2355 /*
2356 * we may have dropped the spin lock to get the head
2357 * mutex lock, and that might have given someone else
2358 * time to free the head. If that's true, it has been
2359 * removed from our list and we can move on.
2360 */
2361 if (ret == -EAGAIN) {
2362 locked_ref = NULL;
2363 count++;
2364 continue;
2365 }
2366 }
2367
2368 spin_lock(&locked_ref->lock);
2369
2370 /*
2371 * locked_ref is the head node, so we have to go one
2372 * node back for any delayed ref updates
2373 */
2374 ref = select_delayed_ref(locked_ref);
2375
2376 if (ref && ref->seq &&
2377 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2378 spin_unlock(&locked_ref->lock);
2379 btrfs_delayed_ref_unlock(locked_ref);
2380 spin_lock(&delayed_refs->lock);
2381 locked_ref->processing = 0;
2382 delayed_refs->num_heads_ready++;
2383 spin_unlock(&delayed_refs->lock);
2384 locked_ref = NULL;
2385 cond_resched();
2386 count++;
2387 continue;
2388 }
2389
2390 /*
2391 * record the must insert reserved flag before we
2392 * drop the spin lock.
2393 */
2394 must_insert_reserved = locked_ref->must_insert_reserved;
2395 locked_ref->must_insert_reserved = 0;
2396
2397 extent_op = locked_ref->extent_op;
2398 locked_ref->extent_op = NULL;
2399
2400 if (!ref) {
2401
2402
2403 /* All delayed refs have been processed, Go ahead
2404 * and send the head node to run_one_delayed_ref,
2405 * so that any accounting fixes can happen
2406 */
2407 ref = &locked_ref->node;
2408
2409 if (extent_op && must_insert_reserved) {
2410 btrfs_free_delayed_extent_op(extent_op);
2411 extent_op = NULL;
2412 }
2413
2414 if (extent_op) {
2415 spin_unlock(&locked_ref->lock);
2416 ret = run_delayed_extent_op(trans, root,
2417 ref, extent_op);
2418 btrfs_free_delayed_extent_op(extent_op);
2419
2420 if (ret) {
2421 /*
2422 * Need to reset must_insert_reserved if
2423 * there was an error so the abort stuff
2424 * can cleanup the reserved space
2425 * properly.
2426 */
2427 if (must_insert_reserved)
2428 locked_ref->must_insert_reserved = 1;
2429 locked_ref->processing = 0;
2430 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2431 btrfs_delayed_ref_unlock(locked_ref);
2432 return ret;
2433 }
2434 continue;
2435 }
2436
2437 /*
2438 * Need to drop our head ref lock and re-aqcuire the
2439 * delayed ref lock and then re-check to make sure
2440 * nobody got added.
2441 */
2442 spin_unlock(&locked_ref->lock);
2443 spin_lock(&delayed_refs->lock);
2444 spin_lock(&locked_ref->lock);
2445 if (!list_empty(&locked_ref->ref_list) ||
2446 locked_ref->extent_op) {
2447 spin_unlock(&locked_ref->lock);
2448 spin_unlock(&delayed_refs->lock);
2449 continue;
2450 }
2451 ref->in_tree = 0;
2452 delayed_refs->num_heads--;
2453 rb_erase(&locked_ref->href_node,
2454 &delayed_refs->href_root);
2455 spin_unlock(&delayed_refs->lock);
2456 } else {
2457 actual_count++;
2458 ref->in_tree = 0;
2459 list_del(&ref->list);
2460 }
2461 atomic_dec(&delayed_refs->num_entries);
2462
2463 if (!btrfs_delayed_ref_is_head(ref)) {
2464 /*
2465 * when we play the delayed ref, also correct the
2466 * ref_mod on head
2467 */
2468 switch (ref->action) {
2469 case BTRFS_ADD_DELAYED_REF:
2470 case BTRFS_ADD_DELAYED_EXTENT:
2471 locked_ref->node.ref_mod -= ref->ref_mod;
2472 break;
2473 case BTRFS_DROP_DELAYED_REF:
2474 locked_ref->node.ref_mod += ref->ref_mod;
2475 break;
2476 default:
2477 WARN_ON(1);
2478 }
2479 }
2480 spin_unlock(&locked_ref->lock);
2481
2482 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2483 must_insert_reserved);
2484
2485 btrfs_free_delayed_extent_op(extent_op);
2486 if (ret) {
2487 locked_ref->processing = 0;
2488 btrfs_delayed_ref_unlock(locked_ref);
2489 btrfs_put_delayed_ref(ref);
2490 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2491 return ret;
2492 }
2493
2494 /*
2495 * If this node is a head, that means all the refs in this head
2496 * have been dealt with, and we will pick the next head to deal
2497 * with, so we must unlock the head and drop it from the cluster
2498 * list before we release it.
2499 */
2500 if (btrfs_delayed_ref_is_head(ref)) {
2501 if (locked_ref->is_data &&
2502 locked_ref->total_ref_mod < 0) {
2503 spin_lock(&delayed_refs->lock);
2504 delayed_refs->pending_csums -= ref->num_bytes;
2505 spin_unlock(&delayed_refs->lock);
2506 }
2507 btrfs_delayed_ref_unlock(locked_ref);
2508 locked_ref = NULL;
2509 }
2510 btrfs_put_delayed_ref(ref);
2511 count++;
2512 cond_resched();
2513 }
2514
2515 /*
2516 * We don't want to include ref heads since we can have empty ref heads
2517 * and those will drastically skew our runtime down since we just do
2518 * accounting, no actual extent tree updates.
2519 */
2520 if (actual_count > 0) {
2521 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2522 u64 avg;
2523
2524 /*
2525 * We weigh the current average higher than our current runtime
2526 * to avoid large swings in the average.
2527 */
2528 spin_lock(&delayed_refs->lock);
2529 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2530 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2531 spin_unlock(&delayed_refs->lock);
2532 }
2533 return 0;
2534 }
2535
2536 #ifdef SCRAMBLE_DELAYED_REFS
2537 /*
2538 * Normally delayed refs get processed in ascending bytenr order. This
2539 * correlates in most cases to the order added. To expose dependencies on this
2540 * order, we start to process the tree in the middle instead of the beginning
2541 */
2542 static u64 find_middle(struct rb_root *root)
2543 {
2544 struct rb_node *n = root->rb_node;
2545 struct btrfs_delayed_ref_node *entry;
2546 int alt = 1;
2547 u64 middle;
2548 u64 first = 0, last = 0;
2549
2550 n = rb_first(root);
2551 if (n) {
2552 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2553 first = entry->bytenr;
2554 }
2555 n = rb_last(root);
2556 if (n) {
2557 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2558 last = entry->bytenr;
2559 }
2560 n = root->rb_node;
2561
2562 while (n) {
2563 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2564 WARN_ON(!entry->in_tree);
2565
2566 middle = entry->bytenr;
2567
2568 if (alt)
2569 n = n->rb_left;
2570 else
2571 n = n->rb_right;
2572
2573 alt = 1 - alt;
2574 }
2575 return middle;
2576 }
2577 #endif
2578
2579 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2580 {
2581 u64 num_bytes;
2582
2583 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2584 sizeof(struct btrfs_extent_inline_ref));
2585 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2586 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2587
2588 /*
2589 * We don't ever fill up leaves all the way so multiply by 2 just to be
2590 * closer to what we're really going to want to ouse.
2591 */
2592 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2593 }
2594
2595 /*
2596 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2597 * would require to store the csums for that many bytes.
2598 */
2599 u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes)
2600 {
2601 u64 csum_size;
2602 u64 num_csums_per_leaf;
2603 u64 num_csums;
2604
2605 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
2606 num_csums_per_leaf = div64_u64(csum_size,
2607 (u64)btrfs_super_csum_size(root->fs_info->super_copy));
2608 num_csums = div64_u64(csum_bytes, root->sectorsize);
2609 num_csums += num_csums_per_leaf - 1;
2610 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2611 return num_csums;
2612 }
2613
2614 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2615 struct btrfs_root *root)
2616 {
2617 struct btrfs_block_rsv *global_rsv;
2618 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2619 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2620 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2621 u64 num_bytes, num_dirty_bgs_bytes;
2622 int ret = 0;
2623
2624 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2625 num_heads = heads_to_leaves(root, num_heads);
2626 if (num_heads > 1)
2627 num_bytes += (num_heads - 1) * root->nodesize;
2628 num_bytes <<= 1;
2629 num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize;
2630 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root,
2631 num_dirty_bgs);
2632 global_rsv = &root->fs_info->global_block_rsv;
2633
2634 /*
2635 * If we can't allocate any more chunks lets make sure we have _lots_ of
2636 * wiggle room since running delayed refs can create more delayed refs.
2637 */
2638 if (global_rsv->space_info->full) {
2639 num_dirty_bgs_bytes <<= 1;
2640 num_bytes <<= 1;
2641 }
2642
2643 spin_lock(&global_rsv->lock);
2644 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2645 ret = 1;
2646 spin_unlock(&global_rsv->lock);
2647 return ret;
2648 }
2649
2650 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2651 struct btrfs_root *root)
2652 {
2653 struct btrfs_fs_info *fs_info = root->fs_info;
2654 u64 num_entries =
2655 atomic_read(&trans->transaction->delayed_refs.num_entries);
2656 u64 avg_runtime;
2657 u64 val;
2658
2659 smp_mb();
2660 avg_runtime = fs_info->avg_delayed_ref_runtime;
2661 val = num_entries * avg_runtime;
2662 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2663 return 1;
2664 if (val >= NSEC_PER_SEC / 2)
2665 return 2;
2666
2667 return btrfs_check_space_for_delayed_refs(trans, root);
2668 }
2669
2670 struct async_delayed_refs {
2671 struct btrfs_root *root;
2672 int count;
2673 int error;
2674 int sync;
2675 struct completion wait;
2676 struct btrfs_work work;
2677 };
2678
2679 static void delayed_ref_async_start(struct btrfs_work *work)
2680 {
2681 struct async_delayed_refs *async;
2682 struct btrfs_trans_handle *trans;
2683 int ret;
2684
2685 async = container_of(work, struct async_delayed_refs, work);
2686
2687 trans = btrfs_join_transaction(async->root);
2688 if (IS_ERR(trans)) {
2689 async->error = PTR_ERR(trans);
2690 goto done;
2691 }
2692
2693 /*
2694 * trans->sync means that when we call end_transaciton, we won't
2695 * wait on delayed refs
2696 */
2697 trans->sync = true;
2698 ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2699 if (ret)
2700 async->error = ret;
2701
2702 ret = btrfs_end_transaction(trans, async->root);
2703 if (ret && !async->error)
2704 async->error = ret;
2705 done:
2706 if (async->sync)
2707 complete(&async->wait);
2708 else
2709 kfree(async);
2710 }
2711
2712 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2713 unsigned long count, int wait)
2714 {
2715 struct async_delayed_refs *async;
2716 int ret;
2717
2718 async = kmalloc(sizeof(*async), GFP_NOFS);
2719 if (!async)
2720 return -ENOMEM;
2721
2722 async->root = root->fs_info->tree_root;
2723 async->count = count;
2724 async->error = 0;
2725 if (wait)
2726 async->sync = 1;
2727 else
2728 async->sync = 0;
2729 init_completion(&async->wait);
2730
2731 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2732 delayed_ref_async_start, NULL, NULL);
2733
2734 btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2735
2736 if (wait) {
2737 wait_for_completion(&async->wait);
2738 ret = async->error;
2739 kfree(async);
2740 return ret;
2741 }
2742 return 0;
2743 }
2744
2745 /*
2746 * this starts processing the delayed reference count updates and
2747 * extent insertions we have queued up so far. count can be
2748 * 0, which means to process everything in the tree at the start
2749 * of the run (but not newly added entries), or it can be some target
2750 * number you'd like to process.
2751 *
2752 * Returns 0 on success or if called with an aborted transaction
2753 * Returns <0 on error and aborts the transaction
2754 */
2755 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2756 struct btrfs_root *root, unsigned long count)
2757 {
2758 struct rb_node *node;
2759 struct btrfs_delayed_ref_root *delayed_refs;
2760 struct btrfs_delayed_ref_head *head;
2761 int ret;
2762 int run_all = count == (unsigned long)-1;
2763
2764 /* We'll clean this up in btrfs_cleanup_transaction */
2765 if (trans->aborted)
2766 return 0;
2767
2768 if (root == root->fs_info->extent_root)
2769 root = root->fs_info->tree_root;
2770
2771 delayed_refs = &trans->transaction->delayed_refs;
2772 if (count == 0)
2773 count = atomic_read(&delayed_refs->num_entries) * 2;
2774
2775 again:
2776 #ifdef SCRAMBLE_DELAYED_REFS
2777 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2778 #endif
2779 ret = __btrfs_run_delayed_refs(trans, root, count);
2780 if (ret < 0) {
2781 btrfs_abort_transaction(trans, root, ret);
2782 return ret;
2783 }
2784
2785 if (run_all) {
2786 if (!list_empty(&trans->new_bgs))
2787 btrfs_create_pending_block_groups(trans, root);
2788
2789 spin_lock(&delayed_refs->lock);
2790 node = rb_first(&delayed_refs->href_root);
2791 if (!node) {
2792 spin_unlock(&delayed_refs->lock);
2793 goto out;
2794 }
2795 count = (unsigned long)-1;
2796
2797 while (node) {
2798 head = rb_entry(node, struct btrfs_delayed_ref_head,
2799 href_node);
2800 if (btrfs_delayed_ref_is_head(&head->node)) {
2801 struct btrfs_delayed_ref_node *ref;
2802
2803 ref = &head->node;
2804 atomic_inc(&ref->refs);
2805
2806 spin_unlock(&delayed_refs->lock);
2807 /*
2808 * Mutex was contended, block until it's
2809 * released and try again
2810 */
2811 mutex_lock(&head->mutex);
2812 mutex_unlock(&head->mutex);
2813
2814 btrfs_put_delayed_ref(ref);
2815 cond_resched();
2816 goto again;
2817 } else {
2818 WARN_ON(1);
2819 }
2820 node = rb_next(node);
2821 }
2822 spin_unlock(&delayed_refs->lock);
2823 cond_resched();
2824 goto again;
2825 }
2826 out:
2827 assert_qgroups_uptodate(trans);
2828 return 0;
2829 }
2830
2831 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2832 struct btrfs_root *root,
2833 u64 bytenr, u64 num_bytes, u64 flags,
2834 int level, int is_data)
2835 {
2836 struct btrfs_delayed_extent_op *extent_op;
2837 int ret;
2838
2839 extent_op = btrfs_alloc_delayed_extent_op();
2840 if (!extent_op)
2841 return -ENOMEM;
2842
2843 extent_op->flags_to_set = flags;
2844 extent_op->update_flags = 1;
2845 extent_op->update_key = 0;
2846 extent_op->is_data = is_data ? 1 : 0;
2847 extent_op->level = level;
2848
2849 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2850 num_bytes, extent_op);
2851 if (ret)
2852 btrfs_free_delayed_extent_op(extent_op);
2853 return ret;
2854 }
2855
2856 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2857 struct btrfs_root *root,
2858 struct btrfs_path *path,
2859 u64 objectid, u64 offset, u64 bytenr)
2860 {
2861 struct btrfs_delayed_ref_head *head;
2862 struct btrfs_delayed_ref_node *ref;
2863 struct btrfs_delayed_data_ref *data_ref;
2864 struct btrfs_delayed_ref_root *delayed_refs;
2865 int ret = 0;
2866
2867 delayed_refs = &trans->transaction->delayed_refs;
2868 spin_lock(&delayed_refs->lock);
2869 head = btrfs_find_delayed_ref_head(trans, bytenr);
2870 if (!head) {
2871 spin_unlock(&delayed_refs->lock);
2872 return 0;
2873 }
2874
2875 if (!mutex_trylock(&head->mutex)) {
2876 atomic_inc(&head->node.refs);
2877 spin_unlock(&delayed_refs->lock);
2878
2879 btrfs_release_path(path);
2880
2881 /*
2882 * Mutex was contended, block until it's released and let
2883 * caller try again
2884 */
2885 mutex_lock(&head->mutex);
2886 mutex_unlock(&head->mutex);
2887 btrfs_put_delayed_ref(&head->node);
2888 return -EAGAIN;
2889 }
2890 spin_unlock(&delayed_refs->lock);
2891
2892 spin_lock(&head->lock);
2893 list_for_each_entry(ref, &head->ref_list, list) {
2894 /* If it's a shared ref we know a cross reference exists */
2895 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2896 ret = 1;
2897 break;
2898 }
2899
2900 data_ref = btrfs_delayed_node_to_data_ref(ref);
2901
2902 /*
2903 * If our ref doesn't match the one we're currently looking at
2904 * then we have a cross reference.
2905 */
2906 if (data_ref->root != root->root_key.objectid ||
2907 data_ref->objectid != objectid ||
2908 data_ref->offset != offset) {
2909 ret = 1;
2910 break;
2911 }
2912 }
2913 spin_unlock(&head->lock);
2914 mutex_unlock(&head->mutex);
2915 return ret;
2916 }
2917
2918 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2919 struct btrfs_root *root,
2920 struct btrfs_path *path,
2921 u64 objectid, u64 offset, u64 bytenr)
2922 {
2923 struct btrfs_root *extent_root = root->fs_info->extent_root;
2924 struct extent_buffer *leaf;
2925 struct btrfs_extent_data_ref *ref;
2926 struct btrfs_extent_inline_ref *iref;
2927 struct btrfs_extent_item *ei;
2928 struct btrfs_key key;
2929 u32 item_size;
2930 int ret;
2931
2932 key.objectid = bytenr;
2933 key.offset = (u64)-1;
2934 key.type = BTRFS_EXTENT_ITEM_KEY;
2935
2936 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2937 if (ret < 0)
2938 goto out;
2939 BUG_ON(ret == 0); /* Corruption */
2940
2941 ret = -ENOENT;
2942 if (path->slots[0] == 0)
2943 goto out;
2944
2945 path->slots[0]--;
2946 leaf = path->nodes[0];
2947 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2948
2949 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2950 goto out;
2951
2952 ret = 1;
2953 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2954 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2955 if (item_size < sizeof(*ei)) {
2956 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2957 goto out;
2958 }
2959 #endif
2960 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2961
2962 if (item_size != sizeof(*ei) +
2963 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2964 goto out;
2965
2966 if (btrfs_extent_generation(leaf, ei) <=
2967 btrfs_root_last_snapshot(&root->root_item))
2968 goto out;
2969
2970 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2971 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2972 BTRFS_EXTENT_DATA_REF_KEY)
2973 goto out;
2974
2975 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2976 if (btrfs_extent_refs(leaf, ei) !=
2977 btrfs_extent_data_ref_count(leaf, ref) ||
2978 btrfs_extent_data_ref_root(leaf, ref) !=
2979 root->root_key.objectid ||
2980 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2981 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2982 goto out;
2983
2984 ret = 0;
2985 out:
2986 return ret;
2987 }
2988
2989 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2990 struct btrfs_root *root,
2991 u64 objectid, u64 offset, u64 bytenr)
2992 {
2993 struct btrfs_path *path;
2994 int ret;
2995 int ret2;
2996
2997 path = btrfs_alloc_path();
2998 if (!path)
2999 return -ENOENT;
3000
3001 do {
3002 ret = check_committed_ref(trans, root, path, objectid,
3003 offset, bytenr);
3004 if (ret && ret != -ENOENT)
3005 goto out;
3006
3007 ret2 = check_delayed_ref(trans, root, path, objectid,
3008 offset, bytenr);
3009 } while (ret2 == -EAGAIN);
3010
3011 if (ret2 && ret2 != -ENOENT) {
3012 ret = ret2;
3013 goto out;
3014 }
3015
3016 if (ret != -ENOENT || ret2 != -ENOENT)
3017 ret = 0;
3018 out:
3019 btrfs_free_path(path);
3020 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3021 WARN_ON(ret > 0);
3022 return ret;
3023 }
3024
3025 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3026 struct btrfs_root *root,
3027 struct extent_buffer *buf,
3028 int full_backref, int inc)
3029 {
3030 u64 bytenr;
3031 u64 num_bytes;
3032 u64 parent;
3033 u64 ref_root;
3034 u32 nritems;
3035 struct btrfs_key key;
3036 struct btrfs_file_extent_item *fi;
3037 int i;
3038 int level;
3039 int ret = 0;
3040 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3041 u64, u64, u64, u64, u64, u64, int);
3042
3043
3044 if (btrfs_test_is_dummy_root(root))
3045 return 0;
3046
3047 ref_root = btrfs_header_owner(buf);
3048 nritems = btrfs_header_nritems(buf);
3049 level = btrfs_header_level(buf);
3050
3051 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3052 return 0;
3053
3054 if (inc)
3055 process_func = btrfs_inc_extent_ref;
3056 else
3057 process_func = btrfs_free_extent;
3058
3059 if (full_backref)
3060 parent = buf->start;
3061 else
3062 parent = 0;
3063
3064 for (i = 0; i < nritems; i++) {
3065 if (level == 0) {
3066 btrfs_item_key_to_cpu(buf, &key, i);
3067 if (key.type != BTRFS_EXTENT_DATA_KEY)
3068 continue;
3069 fi = btrfs_item_ptr(buf, i,
3070 struct btrfs_file_extent_item);
3071 if (btrfs_file_extent_type(buf, fi) ==
3072 BTRFS_FILE_EXTENT_INLINE)
3073 continue;
3074 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3075 if (bytenr == 0)
3076 continue;
3077
3078 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3079 key.offset -= btrfs_file_extent_offset(buf, fi);
3080 ret = process_func(trans, root, bytenr, num_bytes,
3081 parent, ref_root, key.objectid,
3082 key.offset, 1);
3083 if (ret)
3084 goto fail;
3085 } else {
3086 bytenr = btrfs_node_blockptr(buf, i);
3087 num_bytes = root->nodesize;
3088 ret = process_func(trans, root, bytenr, num_bytes,
3089 parent, ref_root, level - 1, 0,
3090 1);
3091 if (ret)
3092 goto fail;
3093 }
3094 }
3095 return 0;
3096 fail:
3097 return ret;
3098 }
3099
3100 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3101 struct extent_buffer *buf, int full_backref)
3102 {
3103 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3104 }
3105
3106 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3107 struct extent_buffer *buf, int full_backref)
3108 {
3109 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3110 }
3111
3112 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3113 struct btrfs_root *root,
3114 struct btrfs_path *path,
3115 struct btrfs_block_group_cache *cache)
3116 {
3117 int ret;
3118 struct btrfs_root *extent_root = root->fs_info->extent_root;
3119 unsigned long bi;
3120 struct extent_buffer *leaf;
3121
3122 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3123 if (ret) {
3124 if (ret > 0)
3125 ret = -ENOENT;
3126 goto fail;
3127 }
3128
3129 leaf = path->nodes[0];
3130 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3131 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3132 btrfs_mark_buffer_dirty(leaf);
3133 fail:
3134 btrfs_release_path(path);
3135 return ret;
3136
3137 }
3138
3139 static struct btrfs_block_group_cache *
3140 next_block_group(struct btrfs_root *root,
3141 struct btrfs_block_group_cache *cache)
3142 {
3143 struct rb_node *node;
3144
3145 spin_lock(&root->fs_info->block_group_cache_lock);
3146
3147 /* If our block group was removed, we need a full search. */
3148 if (RB_EMPTY_NODE(&cache->cache_node)) {
3149 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3150
3151 spin_unlock(&root->fs_info->block_group_cache_lock);
3152 btrfs_put_block_group(cache);
3153 cache = btrfs_lookup_first_block_group(root->fs_info,
3154 next_bytenr);
3155 return cache;
3156 }
3157 node = rb_next(&cache->cache_node);
3158 btrfs_put_block_group(cache);
3159 if (node) {
3160 cache = rb_entry(node, struct btrfs_block_group_cache,
3161 cache_node);
3162 btrfs_get_block_group(cache);
3163 } else
3164 cache = NULL;
3165 spin_unlock(&root->fs_info->block_group_cache_lock);
3166 return cache;
3167 }
3168
3169 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3170 struct btrfs_trans_handle *trans,
3171 struct btrfs_path *path)
3172 {
3173 struct btrfs_root *root = block_group->fs_info->tree_root;
3174 struct inode *inode = NULL;
3175 u64 alloc_hint = 0;
3176 int dcs = BTRFS_DC_ERROR;
3177 u64 num_pages = 0;
3178 int retries = 0;
3179 int ret = 0;
3180
3181 /*
3182 * If this block group is smaller than 100 megs don't bother caching the
3183 * block group.
3184 */
3185 if (block_group->key.offset < (100 * 1024 * 1024)) {
3186 spin_lock(&block_group->lock);
3187 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3188 spin_unlock(&block_group->lock);
3189 return 0;
3190 }
3191
3192 if (trans->aborted)
3193 return 0;
3194 again:
3195 inode = lookup_free_space_inode(root, block_group, path);
3196 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3197 ret = PTR_ERR(inode);
3198 btrfs_release_path(path);
3199 goto out;
3200 }
3201
3202 if (IS_ERR(inode)) {
3203 BUG_ON(retries);
3204 retries++;
3205
3206 if (block_group->ro)
3207 goto out_free;
3208
3209 ret = create_free_space_inode(root, trans, block_group, path);
3210 if (ret)
3211 goto out_free;
3212 goto again;
3213 }
3214
3215 /* We've already setup this transaction, go ahead and exit */
3216 if (block_group->cache_generation == trans->transid &&
3217 i_size_read(inode)) {
3218 dcs = BTRFS_DC_SETUP;
3219 goto out_put;
3220 }
3221
3222 /*
3223 * We want to set the generation to 0, that way if anything goes wrong
3224 * from here on out we know not to trust this cache when we load up next
3225 * time.
3226 */
3227 BTRFS_I(inode)->generation = 0;
3228 ret = btrfs_update_inode(trans, root, inode);
3229 if (ret) {
3230 /*
3231 * So theoretically we could recover from this, simply set the
3232 * super cache generation to 0 so we know to invalidate the
3233 * cache, but then we'd have to keep track of the block groups
3234 * that fail this way so we know we _have_ to reset this cache
3235 * before the next commit or risk reading stale cache. So to
3236 * limit our exposure to horrible edge cases lets just abort the
3237 * transaction, this only happens in really bad situations
3238 * anyway.
3239 */
3240 btrfs_abort_transaction(trans, root, ret);
3241 goto out_put;
3242 }
3243 WARN_ON(ret);
3244
3245 if (i_size_read(inode) > 0) {
3246 ret = btrfs_check_trunc_cache_free_space(root,
3247 &root->fs_info->global_block_rsv);
3248 if (ret)
3249 goto out_put;
3250
3251 ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode);
3252 if (ret)
3253 goto out_put;
3254 }
3255
3256 spin_lock(&block_group->lock);
3257 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3258 !btrfs_test_opt(root, SPACE_CACHE)) {
3259 /*
3260 * don't bother trying to write stuff out _if_
3261 * a) we're not cached,
3262 * b) we're with nospace_cache mount option.
3263 */
3264 dcs = BTRFS_DC_WRITTEN;
3265 spin_unlock(&block_group->lock);
3266 goto out_put;
3267 }
3268 spin_unlock(&block_group->lock);
3269
3270 /*
3271 * Try to preallocate enough space based on how big the block group is.
3272 * Keep in mind this has to include any pinned space which could end up
3273 * taking up quite a bit since it's not folded into the other space
3274 * cache.
3275 */
3276 num_pages = div_u64(block_group->key.offset, 256 * 1024 * 1024);
3277 if (!num_pages)
3278 num_pages = 1;
3279
3280 num_pages *= 16;
3281 num_pages *= PAGE_CACHE_SIZE;
3282
3283 ret = btrfs_check_data_free_space(inode, num_pages, num_pages);
3284 if (ret)
3285 goto out_put;
3286
3287 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3288 num_pages, num_pages,
3289 &alloc_hint);
3290 if (!ret)
3291 dcs = BTRFS_DC_SETUP;
3292 btrfs_free_reserved_data_space(inode, num_pages);
3293
3294 out_put:
3295 iput(inode);
3296 out_free:
3297 btrfs_release_path(path);
3298 out:
3299 spin_lock(&block_group->lock);
3300 if (!ret && dcs == BTRFS_DC_SETUP)
3301 block_group->cache_generation = trans->transid;
3302 block_group->disk_cache_state = dcs;
3303 spin_unlock(&block_group->lock);
3304
3305 return ret;
3306 }
3307
3308 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3309 struct btrfs_root *root)
3310 {
3311 struct btrfs_block_group_cache *cache, *tmp;
3312 struct btrfs_transaction *cur_trans = trans->transaction;
3313 struct btrfs_path *path;
3314
3315 if (list_empty(&cur_trans->dirty_bgs) ||
3316 !btrfs_test_opt(root, SPACE_CACHE))
3317 return 0;
3318
3319 path = btrfs_alloc_path();
3320 if (!path)
3321 return -ENOMEM;
3322
3323 /* Could add new block groups, use _safe just in case */
3324 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3325 dirty_list) {
3326 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3327 cache_save_setup(cache, trans, path);
3328 }
3329
3330 btrfs_free_path(path);
3331 return 0;
3332 }
3333
3334 /*
3335 * transaction commit does final block group cache writeback during a
3336 * critical section where nothing is allowed to change the FS. This is
3337 * required in order for the cache to actually match the block group,
3338 * but can introduce a lot of latency into the commit.
3339 *
3340 * So, btrfs_start_dirty_block_groups is here to kick off block group
3341 * cache IO. There's a chance we'll have to redo some of it if the
3342 * block group changes again during the commit, but it greatly reduces
3343 * the commit latency by getting rid of the easy block groups while
3344 * we're still allowing others to join the commit.
3345 */
3346 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3347 struct btrfs_root *root)
3348 {
3349 struct btrfs_block_group_cache *cache;
3350 struct btrfs_transaction *cur_trans = trans->transaction;
3351 int ret = 0;
3352 int should_put;
3353 struct btrfs_path *path = NULL;
3354 LIST_HEAD(dirty);
3355 struct list_head *io = &cur_trans->io_bgs;
3356 int num_started = 0;
3357 int loops = 0;
3358
3359 spin_lock(&cur_trans->dirty_bgs_lock);
3360 if (list_empty(&cur_trans->dirty_bgs)) {
3361 spin_unlock(&cur_trans->dirty_bgs_lock);
3362 return 0;
3363 }
3364 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3365 spin_unlock(&cur_trans->dirty_bgs_lock);
3366
3367 again:
3368 /*
3369 * make sure all the block groups on our dirty list actually
3370 * exist
3371 */
3372 btrfs_create_pending_block_groups(trans, root);
3373
3374 if (!path) {
3375 path = btrfs_alloc_path();
3376 if (!path)
3377 return -ENOMEM;
3378 }
3379
3380 /*
3381 * cache_write_mutex is here only to save us from balance or automatic
3382 * removal of empty block groups deleting this block group while we are
3383 * writing out the cache
3384 */
3385 mutex_lock(&trans->transaction->cache_write_mutex);
3386 while (!list_empty(&dirty)) {
3387 cache = list_first_entry(&dirty,
3388 struct btrfs_block_group_cache,
3389 dirty_list);
3390 /*
3391 * this can happen if something re-dirties a block
3392 * group that is already under IO. Just wait for it to
3393 * finish and then do it all again
3394 */
3395 if (!list_empty(&cache->io_list)) {
3396 list_del_init(&cache->io_list);
3397 btrfs_wait_cache_io(root, trans, cache,
3398 &cache->io_ctl, path,
3399 cache->key.objectid);
3400 btrfs_put_block_group(cache);
3401 }
3402
3403
3404 /*
3405 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3406 * if it should update the cache_state. Don't delete
3407 * until after we wait.
3408 *
3409 * Since we're not running in the commit critical section
3410 * we need the dirty_bgs_lock to protect from update_block_group
3411 */
3412 spin_lock(&cur_trans->dirty_bgs_lock);
3413 list_del_init(&cache->dirty_list);
3414 spin_unlock(&cur_trans->dirty_bgs_lock);
3415
3416 should_put = 1;
3417
3418 cache_save_setup(cache, trans, path);
3419
3420 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3421 cache->io_ctl.inode = NULL;
3422 ret = btrfs_write_out_cache(root, trans, cache, path);
3423 if (ret == 0 && cache->io_ctl.inode) {
3424 num_started++;
3425 should_put = 0;
3426
3427 /*
3428 * the cache_write_mutex is protecting
3429 * the io_list
3430 */
3431 list_add_tail(&cache->io_list, io);
3432 } else {
3433 /*
3434 * if we failed to write the cache, the
3435 * generation will be bad and life goes on
3436 */
3437 ret = 0;
3438 }
3439 }
3440 if (!ret) {
3441 ret = write_one_cache_group(trans, root, path, cache);
3442 /*
3443 * Our block group might still be attached to the list
3444 * of new block groups in the transaction handle of some
3445 * other task (struct btrfs_trans_handle->new_bgs). This
3446 * means its block group item isn't yet in the extent
3447 * tree. If this happens ignore the error, as we will
3448 * try again later in the critical section of the
3449 * transaction commit.
3450 */
3451 if (ret == -ENOENT) {
3452 ret = 0;
3453 spin_lock(&cur_trans->dirty_bgs_lock);
3454 if (list_empty(&cache->dirty_list)) {
3455 list_add_tail(&cache->dirty_list,
3456 &cur_trans->dirty_bgs);
3457 btrfs_get_block_group(cache);
3458 }
3459 spin_unlock(&cur_trans->dirty_bgs_lock);
3460 } else if (ret) {
3461 btrfs_abort_transaction(trans, root, ret);
3462 }
3463 }
3464
3465 /* if its not on the io list, we need to put the block group */
3466 if (should_put)
3467 btrfs_put_block_group(cache);
3468
3469 if (ret)
3470 break;
3471
3472 /*
3473 * Avoid blocking other tasks for too long. It might even save
3474 * us from writing caches for block groups that are going to be
3475 * removed.
3476 */
3477 mutex_unlock(&trans->transaction->cache_write_mutex);
3478 mutex_lock(&trans->transaction->cache_write_mutex);
3479 }
3480 mutex_unlock(&trans->transaction->cache_write_mutex);
3481
3482 /*
3483 * go through delayed refs for all the stuff we've just kicked off
3484 * and then loop back (just once)
3485 */
3486 ret = btrfs_run_delayed_refs(trans, root, 0);
3487 if (!ret && loops == 0) {
3488 loops++;
3489 spin_lock(&cur_trans->dirty_bgs_lock);
3490 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3491 /*
3492 * dirty_bgs_lock protects us from concurrent block group
3493 * deletes too (not just cache_write_mutex).
3494 */
3495 if (!list_empty(&dirty)) {
3496 spin_unlock(&cur_trans->dirty_bgs_lock);
3497 goto again;
3498 }
3499 spin_unlock(&cur_trans->dirty_bgs_lock);
3500 }
3501
3502 btrfs_free_path(path);
3503 return ret;
3504 }
3505
3506 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3507 struct btrfs_root *root)
3508 {
3509 struct btrfs_block_group_cache *cache;
3510 struct btrfs_transaction *cur_trans = trans->transaction;
3511 int ret = 0;
3512 int should_put;
3513 struct btrfs_path *path;
3514 struct list_head *io = &cur_trans->io_bgs;
3515 int num_started = 0;
3516
3517 path = btrfs_alloc_path();
3518 if (!path)
3519 return -ENOMEM;
3520
3521 /*
3522 * We don't need the lock here since we are protected by the transaction
3523 * commit. We want to do the cache_save_setup first and then run the
3524 * delayed refs to make sure we have the best chance at doing this all
3525 * in one shot.
3526 */
3527 while (!list_empty(&cur_trans->dirty_bgs)) {
3528 cache = list_first_entry(&cur_trans->dirty_bgs,
3529 struct btrfs_block_group_cache,
3530 dirty_list);
3531
3532 /*
3533 * this can happen if cache_save_setup re-dirties a block
3534 * group that is already under IO. Just wait for it to
3535 * finish and then do it all again
3536 */
3537 if (!list_empty(&cache->io_list)) {
3538 list_del_init(&cache->io_list);
3539 btrfs_wait_cache_io(root, trans, cache,
3540 &cache->io_ctl, path,
3541 cache->key.objectid);
3542 btrfs_put_block_group(cache);
3543 }
3544
3545 /*
3546 * don't remove from the dirty list until after we've waited
3547 * on any pending IO
3548 */
3549 list_del_init(&cache->dirty_list);
3550 should_put = 1;
3551
3552 cache_save_setup(cache, trans, path);
3553
3554 if (!ret)
3555 ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1);
3556
3557 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3558 cache->io_ctl.inode = NULL;
3559 ret = btrfs_write_out_cache(root, trans, cache, path);
3560 if (ret == 0 && cache->io_ctl.inode) {
3561 num_started++;
3562 should_put = 0;
3563 list_add_tail(&cache->io_list, io);
3564 } else {
3565 /*
3566 * if we failed to write the cache, the
3567 * generation will be bad and life goes on
3568 */
3569 ret = 0;
3570 }
3571 }
3572 if (!ret) {
3573 ret = write_one_cache_group(trans, root, path, cache);
3574 if (ret)
3575 btrfs_abort_transaction(trans, root, ret);
3576 }
3577
3578 /* if its not on the io list, we need to put the block group */
3579 if (should_put)
3580 btrfs_put_block_group(cache);
3581 }
3582
3583 while (!list_empty(io)) {
3584 cache = list_first_entry(io, struct btrfs_block_group_cache,
3585 io_list);
3586 list_del_init(&cache->io_list);
3587 btrfs_wait_cache_io(root, trans, cache,
3588 &cache->io_ctl, path, cache->key.objectid);
3589 btrfs_put_block_group(cache);
3590 }
3591
3592 btrfs_free_path(path);
3593 return ret;
3594 }
3595
3596 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3597 {
3598 struct btrfs_block_group_cache *block_group;
3599 int readonly = 0;
3600
3601 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3602 if (!block_group || block_group->ro)
3603 readonly = 1;
3604 if (block_group)
3605 btrfs_put_block_group(block_group);
3606 return readonly;
3607 }
3608
3609 static const char *alloc_name(u64 flags)
3610 {
3611 switch (flags) {
3612 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3613 return "mixed";
3614 case BTRFS_BLOCK_GROUP_METADATA:
3615 return "metadata";
3616 case BTRFS_BLOCK_GROUP_DATA:
3617 return "data";
3618 case BTRFS_BLOCK_GROUP_SYSTEM:
3619 return "system";
3620 default:
3621 WARN_ON(1);
3622 return "invalid-combination";
3623 };
3624 }
3625
3626 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3627 u64 total_bytes, u64 bytes_used,
3628 struct btrfs_space_info **space_info)
3629 {
3630 struct btrfs_space_info *found;
3631 int i;
3632 int factor;
3633 int ret;
3634
3635 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3636 BTRFS_BLOCK_GROUP_RAID10))
3637 factor = 2;
3638 else
3639 factor = 1;
3640
3641 found = __find_space_info(info, flags);
3642 if (found) {
3643 spin_lock(&found->lock);
3644 found->total_bytes += total_bytes;
3645 found->disk_total += total_bytes * factor;
3646 found->bytes_used += bytes_used;
3647 found->disk_used += bytes_used * factor;
3648 if (total_bytes > 0)
3649 found->full = 0;
3650 spin_unlock(&found->lock);
3651 *space_info = found;
3652 return 0;
3653 }
3654 found = kzalloc(sizeof(*found), GFP_NOFS);
3655 if (!found)
3656 return -ENOMEM;
3657
3658 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3659 if (ret) {
3660 kfree(found);
3661 return ret;
3662 }
3663
3664 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3665 INIT_LIST_HEAD(&found->block_groups[i]);
3666 init_rwsem(&found->groups_sem);
3667 spin_lock_init(&found->lock);
3668 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3669 found->total_bytes = total_bytes;
3670 found->disk_total = total_bytes * factor;
3671 found->bytes_used = bytes_used;
3672 found->disk_used = bytes_used * factor;
3673 found->bytes_pinned = 0;
3674 found->bytes_reserved = 0;
3675 found->bytes_readonly = 0;
3676 found->bytes_may_use = 0;
3677 if (total_bytes > 0)
3678 found->full = 0;
3679 else
3680 found->full = 1;
3681 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3682 found->chunk_alloc = 0;
3683 found->flush = 0;
3684 init_waitqueue_head(&found->wait);
3685 INIT_LIST_HEAD(&found->ro_bgs);
3686
3687 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3688 info->space_info_kobj, "%s",
3689 alloc_name(found->flags));
3690 if (ret) {
3691 kfree(found);
3692 return ret;
3693 }
3694
3695 *space_info = found;
3696 list_add_rcu(&found->list, &info->space_info);
3697 if (flags & BTRFS_BLOCK_GROUP_DATA)
3698 info->data_sinfo = found;
3699
3700 return ret;
3701 }
3702
3703 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3704 {
3705 u64 extra_flags = chunk_to_extended(flags) &
3706 BTRFS_EXTENDED_PROFILE_MASK;
3707
3708 write_seqlock(&fs_info->profiles_lock);
3709 if (flags & BTRFS_BLOCK_GROUP_DATA)
3710 fs_info->avail_data_alloc_bits |= extra_flags;
3711 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3712 fs_info->avail_metadata_alloc_bits |= extra_flags;
3713 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3714 fs_info->avail_system_alloc_bits |= extra_flags;
3715 write_sequnlock(&fs_info->profiles_lock);
3716 }
3717
3718 /*
3719 * returns target flags in extended format or 0 if restripe for this
3720 * chunk_type is not in progress
3721 *
3722 * should be called with either volume_mutex or balance_lock held
3723 */
3724 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3725 {
3726 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3727 u64 target = 0;
3728
3729 if (!bctl)
3730 return 0;
3731
3732 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3733 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3734 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3735 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3736 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3737 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3738 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3739 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3740 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3741 }
3742
3743 return target;
3744 }
3745
3746 /*
3747 * @flags: available profiles in extended format (see ctree.h)
3748 *
3749 * Returns reduced profile in chunk format. If profile changing is in
3750 * progress (either running or paused) picks the target profile (if it's
3751 * already available), otherwise falls back to plain reducing.
3752 */
3753 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3754 {
3755 u64 num_devices = root->fs_info->fs_devices->rw_devices;
3756 u64 target;
3757 u64 tmp;
3758
3759 /*
3760 * see if restripe for this chunk_type is in progress, if so
3761 * try to reduce to the target profile
3762 */
3763 spin_lock(&root->fs_info->balance_lock);
3764 target = get_restripe_target(root->fs_info, flags);
3765 if (target) {
3766 /* pick target profile only if it's already available */
3767 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3768 spin_unlock(&root->fs_info->balance_lock);
3769 return extended_to_chunk(target);
3770 }
3771 }
3772 spin_unlock(&root->fs_info->balance_lock);
3773
3774 /* First, mask out the RAID levels which aren't possible */
3775 if (num_devices == 1)
3776 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3777 BTRFS_BLOCK_GROUP_RAID5);
3778 if (num_devices < 3)
3779 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3780 if (num_devices < 4)
3781 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3782
3783 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3784 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3785 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3786 flags &= ~tmp;
3787
3788 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3789 tmp = BTRFS_BLOCK_GROUP_RAID6;
3790 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3791 tmp = BTRFS_BLOCK_GROUP_RAID5;
3792 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3793 tmp = BTRFS_BLOCK_GROUP_RAID10;
3794 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3795 tmp = BTRFS_BLOCK_GROUP_RAID1;
3796 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3797 tmp = BTRFS_BLOCK_GROUP_RAID0;
3798
3799 return extended_to_chunk(flags | tmp);
3800 }
3801
3802 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
3803 {
3804 unsigned seq;
3805 u64 flags;
3806
3807 do {
3808 flags = orig_flags;
3809 seq = read_seqbegin(&root->fs_info->profiles_lock);
3810
3811 if (flags & BTRFS_BLOCK_GROUP_DATA)
3812 flags |= root->fs_info->avail_data_alloc_bits;
3813 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3814 flags |= root->fs_info->avail_system_alloc_bits;
3815 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3816 flags |= root->fs_info->avail_metadata_alloc_bits;
3817 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3818
3819 return btrfs_reduce_alloc_profile(root, flags);
3820 }
3821
3822 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3823 {
3824 u64 flags;
3825 u64 ret;
3826
3827 if (data)
3828 flags = BTRFS_BLOCK_GROUP_DATA;
3829 else if (root == root->fs_info->chunk_root)
3830 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3831 else
3832 flags = BTRFS_BLOCK_GROUP_METADATA;
3833
3834 ret = get_alloc_profile(root, flags);
3835 return ret;
3836 }
3837
3838 /*
3839 * This will check the space that the inode allocates from to make sure we have
3840 * enough space for bytes.
3841 */
3842 int btrfs_check_data_free_space(struct inode *inode, u64 bytes, u64 write_bytes)
3843 {
3844 struct btrfs_space_info *data_sinfo;
3845 struct btrfs_root *root = BTRFS_I(inode)->root;
3846 struct btrfs_fs_info *fs_info = root->fs_info;
3847 u64 used;
3848 int ret = 0;
3849 int need_commit = 2;
3850 int have_pinned_space;
3851
3852 /* make sure bytes are sectorsize aligned */
3853 bytes = ALIGN(bytes, root->sectorsize);
3854
3855 if (btrfs_is_free_space_inode(inode)) {
3856 need_commit = 0;
3857 ASSERT(current->journal_info);
3858 }
3859
3860 data_sinfo = fs_info->data_sinfo;
3861 if (!data_sinfo)
3862 goto alloc;
3863
3864 again:
3865 /* make sure we have enough space to handle the data first */
3866 spin_lock(&data_sinfo->lock);
3867 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3868 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3869 data_sinfo->bytes_may_use;
3870
3871 if (used + bytes > data_sinfo->total_bytes) {
3872 struct btrfs_trans_handle *trans;
3873
3874 /*
3875 * if we don't have enough free bytes in this space then we need
3876 * to alloc a new chunk.
3877 */
3878 if (!data_sinfo->full) {
3879 u64 alloc_target;
3880
3881 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3882 spin_unlock(&data_sinfo->lock);
3883 alloc:
3884 alloc_target = btrfs_get_alloc_profile(root, 1);
3885 /*
3886 * It is ugly that we don't call nolock join
3887 * transaction for the free space inode case here.
3888 * But it is safe because we only do the data space
3889 * reservation for the free space cache in the
3890 * transaction context, the common join transaction
3891 * just increase the counter of the current transaction
3892 * handler, doesn't try to acquire the trans_lock of
3893 * the fs.
3894 */
3895 trans = btrfs_join_transaction(root);
3896 if (IS_ERR(trans))
3897 return PTR_ERR(trans);
3898
3899 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3900 alloc_target,
3901 CHUNK_ALLOC_NO_FORCE);
3902 btrfs_end_transaction(trans, root);
3903 if (ret < 0) {
3904 if (ret != -ENOSPC)
3905 return ret;
3906 else {
3907 have_pinned_space = 1;
3908 goto commit_trans;
3909 }
3910 }
3911
3912 if (!data_sinfo)
3913 data_sinfo = fs_info->data_sinfo;
3914
3915 goto again;
3916 }
3917
3918 /*
3919 * If we don't have enough pinned space to deal with this
3920 * allocation, and no removed chunk in current transaction,
3921 * don't bother committing the transaction.
3922 */
3923 have_pinned_space = percpu_counter_compare(
3924 &data_sinfo->total_bytes_pinned,
3925 used + bytes - data_sinfo->total_bytes);
3926 spin_unlock(&data_sinfo->lock);
3927
3928 /* commit the current transaction and try again */
3929 commit_trans:
3930 if (need_commit &&
3931 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3932 need_commit--;
3933
3934 if (need_commit > 0)
3935 btrfs_wait_ordered_roots(fs_info, -1);
3936
3937 trans = btrfs_join_transaction(root);
3938 if (IS_ERR(trans))
3939 return PTR_ERR(trans);
3940 if (have_pinned_space >= 0 ||
3941 trans->transaction->have_free_bgs ||
3942 need_commit > 0) {
3943 ret = btrfs_commit_transaction(trans, root);
3944 if (ret)
3945 return ret;
3946 /*
3947 * make sure that all running delayed iput are
3948 * done
3949 */
3950 down_write(&root->fs_info->delayed_iput_sem);
3951 up_write(&root->fs_info->delayed_iput_sem);
3952 goto again;
3953 } else {
3954 btrfs_end_transaction(trans, root);
3955 }
3956 }
3957
3958 trace_btrfs_space_reservation(root->fs_info,
3959 "space_info:enospc",
3960 data_sinfo->flags, bytes, 1);
3961 return -ENOSPC;
3962 }
3963 ret = btrfs_qgroup_reserve(root, write_bytes);
3964 if (ret)
3965 goto out;
3966 data_sinfo->bytes_may_use += bytes;
3967 trace_btrfs_space_reservation(root->fs_info, "space_info",
3968 data_sinfo->flags, bytes, 1);
3969 out:
3970 spin_unlock(&data_sinfo->lock);
3971
3972 return ret;
3973 }
3974
3975 /*
3976 * Called if we need to clear a data reservation for this inode.
3977 */
3978 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3979 {
3980 struct btrfs_root *root = BTRFS_I(inode)->root;
3981 struct btrfs_space_info *data_sinfo;
3982
3983 /* make sure bytes are sectorsize aligned */
3984 bytes = ALIGN(bytes, root->sectorsize);
3985
3986 data_sinfo = root->fs_info->data_sinfo;
3987 spin_lock(&data_sinfo->lock);
3988 WARN_ON(data_sinfo->bytes_may_use < bytes);
3989 data_sinfo->bytes_may_use -= bytes;
3990 trace_btrfs_space_reservation(root->fs_info, "space_info",
3991 data_sinfo->flags, bytes, 0);
3992 spin_unlock(&data_sinfo->lock);
3993 }
3994
3995 static void force_metadata_allocation(struct btrfs_fs_info *info)
3996 {
3997 struct list_head *head = &info->space_info;
3998 struct btrfs_space_info *found;
3999
4000 rcu_read_lock();
4001 list_for_each_entry_rcu(found, head, list) {
4002 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4003 found->force_alloc = CHUNK_ALLOC_FORCE;
4004 }
4005 rcu_read_unlock();
4006 }
4007
4008 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4009 {
4010 return (global->size << 1);
4011 }
4012
4013 static int should_alloc_chunk(struct btrfs_root *root,
4014 struct btrfs_space_info *sinfo, int force)
4015 {
4016 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4017 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
4018 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
4019 u64 thresh;
4020
4021 if (force == CHUNK_ALLOC_FORCE)
4022 return 1;
4023
4024 /*
4025 * We need to take into account the global rsv because for all intents
4026 * and purposes it's used space. Don't worry about locking the
4027 * global_rsv, it doesn't change except when the transaction commits.
4028 */
4029 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4030 num_allocated += calc_global_rsv_need_space(global_rsv);
4031
4032 /*
4033 * in limited mode, we want to have some free space up to
4034 * about 1% of the FS size.
4035 */
4036 if (force == CHUNK_ALLOC_LIMITED) {
4037 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
4038 thresh = max_t(u64, 64 * 1024 * 1024,
4039 div_factor_fine(thresh, 1));
4040
4041 if (num_bytes - num_allocated < thresh)
4042 return 1;
4043 }
4044
4045 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
4046 return 0;
4047 return 1;
4048 }
4049
4050 static u64 get_profile_num_devs(struct btrfs_root *root, u64 type)
4051 {
4052 u64 num_dev;
4053
4054 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4055 BTRFS_BLOCK_GROUP_RAID0 |
4056 BTRFS_BLOCK_GROUP_RAID5 |
4057 BTRFS_BLOCK_GROUP_RAID6))
4058 num_dev = root->fs_info->fs_devices->rw_devices;
4059 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4060 num_dev = 2;
4061 else
4062 num_dev = 1; /* DUP or single */
4063
4064 return num_dev;
4065 }
4066
4067 /*
4068 * If @is_allocation is true, reserve space in the system space info necessary
4069 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4070 * removing a chunk.
4071 */
4072 void check_system_chunk(struct btrfs_trans_handle *trans,
4073 struct btrfs_root *root,
4074 u64 type)
4075 {
4076 struct btrfs_space_info *info;
4077 u64 left;
4078 u64 thresh;
4079 int ret = 0;
4080 u64 num_devs;
4081
4082 /*
4083 * Needed because we can end up allocating a system chunk and for an
4084 * atomic and race free space reservation in the chunk block reserve.
4085 */
4086 ASSERT(mutex_is_locked(&root->fs_info->chunk_mutex));
4087
4088 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4089 spin_lock(&info->lock);
4090 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
4091 info->bytes_reserved - info->bytes_readonly -
4092 info->bytes_may_use;
4093 spin_unlock(&info->lock);
4094
4095 num_devs = get_profile_num_devs(root, type);
4096
4097 /* num_devs device items to update and 1 chunk item to add or remove */
4098 thresh = btrfs_calc_trunc_metadata_size(root, num_devs) +
4099 btrfs_calc_trans_metadata_size(root, 1);
4100
4101 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
4102 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
4103 left, thresh, type);
4104 dump_space_info(info, 0, 0);
4105 }
4106
4107 if (left < thresh) {
4108 u64 flags;
4109
4110 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
4111 /*
4112 * Ignore failure to create system chunk. We might end up not
4113 * needing it, as we might not need to COW all nodes/leafs from
4114 * the paths we visit in the chunk tree (they were already COWed
4115 * or created in the current transaction for example).
4116 */
4117 ret = btrfs_alloc_chunk(trans, root, flags);
4118 }
4119
4120 if (!ret) {
4121 ret = btrfs_block_rsv_add(root->fs_info->chunk_root,
4122 &root->fs_info->chunk_block_rsv,
4123 thresh, BTRFS_RESERVE_NO_FLUSH);
4124 if (!ret)
4125 trans->chunk_bytes_reserved += thresh;
4126 }
4127 }
4128
4129 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4130 struct btrfs_root *extent_root, u64 flags, int force)
4131 {
4132 struct btrfs_space_info *space_info;
4133 struct btrfs_fs_info *fs_info = extent_root->fs_info;
4134 int wait_for_alloc = 0;
4135 int ret = 0;
4136
4137 /* Don't re-enter if we're already allocating a chunk */
4138 if (trans->allocating_chunk)
4139 return -ENOSPC;
4140
4141 space_info = __find_space_info(extent_root->fs_info, flags);
4142 if (!space_info) {
4143 ret = update_space_info(extent_root->fs_info, flags,
4144 0, 0, &space_info);
4145 BUG_ON(ret); /* -ENOMEM */
4146 }
4147 BUG_ON(!space_info); /* Logic error */
4148
4149 again:
4150 spin_lock(&space_info->lock);
4151 if (force < space_info->force_alloc)
4152 force = space_info->force_alloc;
4153 if (space_info->full) {
4154 if (should_alloc_chunk(extent_root, space_info, force))
4155 ret = -ENOSPC;
4156 else
4157 ret = 0;
4158 spin_unlock(&space_info->lock);
4159 return ret;
4160 }
4161
4162 if (!should_alloc_chunk(extent_root, space_info, force)) {
4163 spin_unlock(&space_info->lock);
4164 return 0;
4165 } else if (space_info->chunk_alloc) {
4166 wait_for_alloc = 1;
4167 } else {
4168 space_info->chunk_alloc = 1;
4169 }
4170
4171 spin_unlock(&space_info->lock);
4172
4173 mutex_lock(&fs_info->chunk_mutex);
4174
4175 /*
4176 * The chunk_mutex is held throughout the entirety of a chunk
4177 * allocation, so once we've acquired the chunk_mutex we know that the
4178 * other guy is done and we need to recheck and see if we should
4179 * allocate.
4180 */
4181 if (wait_for_alloc) {
4182 mutex_unlock(&fs_info->chunk_mutex);
4183 wait_for_alloc = 0;
4184 goto again;
4185 }
4186
4187 trans->allocating_chunk = true;
4188
4189 /*
4190 * If we have mixed data/metadata chunks we want to make sure we keep
4191 * allocating mixed chunks instead of individual chunks.
4192 */
4193 if (btrfs_mixed_space_info(space_info))
4194 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4195
4196 /*
4197 * if we're doing a data chunk, go ahead and make sure that
4198 * we keep a reasonable number of metadata chunks allocated in the
4199 * FS as well.
4200 */
4201 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4202 fs_info->data_chunk_allocations++;
4203 if (!(fs_info->data_chunk_allocations %
4204 fs_info->metadata_ratio))
4205 force_metadata_allocation(fs_info);
4206 }
4207
4208 /*
4209 * Check if we have enough space in SYSTEM chunk because we may need
4210 * to update devices.
4211 */
4212 check_system_chunk(trans, extent_root, flags);
4213
4214 ret = btrfs_alloc_chunk(trans, extent_root, flags);
4215 trans->allocating_chunk = false;
4216
4217 spin_lock(&space_info->lock);
4218 if (ret < 0 && ret != -ENOSPC)
4219 goto out;
4220 if (ret)
4221 space_info->full = 1;
4222 else
4223 ret = 1;
4224
4225 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4226 out:
4227 space_info->chunk_alloc = 0;
4228 spin_unlock(&space_info->lock);
4229 mutex_unlock(&fs_info->chunk_mutex);
4230 /*
4231 * When we allocate a new chunk we reserve space in the chunk block
4232 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4233 * add new nodes/leafs to it if we end up needing to do it when
4234 * inserting the chunk item and updating device items as part of the
4235 * second phase of chunk allocation, performed by
4236 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4237 * large number of new block groups to create in our transaction
4238 * handle's new_bgs list to avoid exhausting the chunk block reserve
4239 * in extreme cases - like having a single transaction create many new
4240 * block groups when starting to write out the free space caches of all
4241 * the block groups that were made dirty during the lifetime of the
4242 * transaction.
4243 */
4244 if (trans->chunk_bytes_reserved >= (2 * 1024 * 1024ull)) {
4245 btrfs_create_pending_block_groups(trans, trans->root);
4246 btrfs_trans_release_chunk_metadata(trans);
4247 }
4248 return ret;
4249 }
4250
4251 static int can_overcommit(struct btrfs_root *root,
4252 struct btrfs_space_info *space_info, u64 bytes,
4253 enum btrfs_reserve_flush_enum flush)
4254 {
4255 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4256 u64 profile = btrfs_get_alloc_profile(root, 0);
4257 u64 space_size;
4258 u64 avail;
4259 u64 used;
4260
4261 used = space_info->bytes_used + space_info->bytes_reserved +
4262 space_info->bytes_pinned + space_info->bytes_readonly;
4263
4264 /*
4265 * We only want to allow over committing if we have lots of actual space
4266 * free, but if we don't have enough space to handle the global reserve
4267 * space then we could end up having a real enospc problem when trying
4268 * to allocate a chunk or some other such important allocation.
4269 */
4270 spin_lock(&global_rsv->lock);
4271 space_size = calc_global_rsv_need_space(global_rsv);
4272 spin_unlock(&global_rsv->lock);
4273 if (used + space_size >= space_info->total_bytes)
4274 return 0;
4275
4276 used += space_info->bytes_may_use;
4277
4278 spin_lock(&root->fs_info->free_chunk_lock);
4279 avail = root->fs_info->free_chunk_space;
4280 spin_unlock(&root->fs_info->free_chunk_lock);
4281
4282 /*
4283 * If we have dup, raid1 or raid10 then only half of the free
4284 * space is actually useable. For raid56, the space info used
4285 * doesn't include the parity drive, so we don't have to
4286 * change the math
4287 */
4288 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4289 BTRFS_BLOCK_GROUP_RAID1 |
4290 BTRFS_BLOCK_GROUP_RAID10))
4291 avail >>= 1;
4292
4293 /*
4294 * If we aren't flushing all things, let us overcommit up to
4295 * 1/2th of the space. If we can flush, don't let us overcommit
4296 * too much, let it overcommit up to 1/8 of the space.
4297 */
4298 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4299 avail >>= 3;
4300 else
4301 avail >>= 1;
4302
4303 if (used + bytes < space_info->total_bytes + avail)
4304 return 1;
4305 return 0;
4306 }
4307
4308 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4309 unsigned long nr_pages, int nr_items)
4310 {
4311 struct super_block *sb = root->fs_info->sb;
4312
4313 if (down_read_trylock(&sb->s_umount)) {
4314 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4315 up_read(&sb->s_umount);
4316 } else {
4317 /*
4318 * We needn't worry the filesystem going from r/w to r/o though
4319 * we don't acquire ->s_umount mutex, because the filesystem
4320 * should guarantee the delalloc inodes list be empty after
4321 * the filesystem is readonly(all dirty pages are written to
4322 * the disk).
4323 */
4324 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4325 if (!current->journal_info)
4326 btrfs_wait_ordered_roots(root->fs_info, nr_items);
4327 }
4328 }
4329
4330 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4331 {
4332 u64 bytes;
4333 int nr;
4334
4335 bytes = btrfs_calc_trans_metadata_size(root, 1);
4336 nr = (int)div64_u64(to_reclaim, bytes);
4337 if (!nr)
4338 nr = 1;
4339 return nr;
4340 }
4341
4342 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4343
4344 /*
4345 * shrink metadata reservation for delalloc
4346 */
4347 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4348 bool wait_ordered)
4349 {
4350 struct btrfs_block_rsv *block_rsv;
4351 struct btrfs_space_info *space_info;
4352 struct btrfs_trans_handle *trans;
4353 u64 delalloc_bytes;
4354 u64 max_reclaim;
4355 long time_left;
4356 unsigned long nr_pages;
4357 int loops;
4358 int items;
4359 enum btrfs_reserve_flush_enum flush;
4360
4361 /* Calc the number of the pages we need flush for space reservation */
4362 items = calc_reclaim_items_nr(root, to_reclaim);
4363 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4364
4365 trans = (struct btrfs_trans_handle *)current->journal_info;
4366 block_rsv = &root->fs_info->delalloc_block_rsv;
4367 space_info = block_rsv->space_info;
4368
4369 delalloc_bytes = percpu_counter_sum_positive(
4370 &root->fs_info->delalloc_bytes);
4371 if (delalloc_bytes == 0) {
4372 if (trans)
4373 return;
4374 if (wait_ordered)
4375 btrfs_wait_ordered_roots(root->fs_info, items);
4376 return;
4377 }
4378
4379 loops = 0;
4380 while (delalloc_bytes && loops < 3) {
4381 max_reclaim = min(delalloc_bytes, to_reclaim);
4382 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4383 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4384 /*
4385 * We need to wait for the async pages to actually start before
4386 * we do anything.
4387 */
4388 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4389 if (!max_reclaim)
4390 goto skip_async;
4391
4392 if (max_reclaim <= nr_pages)
4393 max_reclaim = 0;
4394 else
4395 max_reclaim -= nr_pages;
4396
4397 wait_event(root->fs_info->async_submit_wait,
4398 atomic_read(&root->fs_info->async_delalloc_pages) <=
4399 (int)max_reclaim);
4400 skip_async:
4401 if (!trans)
4402 flush = BTRFS_RESERVE_FLUSH_ALL;
4403 else
4404 flush = BTRFS_RESERVE_NO_FLUSH;
4405 spin_lock(&space_info->lock);
4406 if (can_overcommit(root, space_info, orig, flush)) {
4407 spin_unlock(&space_info->lock);
4408 break;
4409 }
4410 spin_unlock(&space_info->lock);
4411
4412 loops++;
4413 if (wait_ordered && !trans) {
4414 btrfs_wait_ordered_roots(root->fs_info, items);
4415 } else {
4416 time_left = schedule_timeout_killable(1);
4417 if (time_left)
4418 break;
4419 }
4420 delalloc_bytes = percpu_counter_sum_positive(
4421 &root->fs_info->delalloc_bytes);
4422 }
4423 }
4424
4425 /**
4426 * maybe_commit_transaction - possibly commit the transaction if its ok to
4427 * @root - the root we're allocating for
4428 * @bytes - the number of bytes we want to reserve
4429 * @force - force the commit
4430 *
4431 * This will check to make sure that committing the transaction will actually
4432 * get us somewhere and then commit the transaction if it does. Otherwise it
4433 * will return -ENOSPC.
4434 */
4435 static int may_commit_transaction(struct btrfs_root *root,
4436 struct btrfs_space_info *space_info,
4437 u64 bytes, int force)
4438 {
4439 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4440 struct btrfs_trans_handle *trans;
4441
4442 trans = (struct btrfs_trans_handle *)current->journal_info;
4443 if (trans)
4444 return -EAGAIN;
4445
4446 if (force)
4447 goto commit;
4448
4449 /* See if there is enough pinned space to make this reservation */
4450 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4451 bytes) >= 0)
4452 goto commit;
4453
4454 /*
4455 * See if there is some space in the delayed insertion reservation for
4456 * this reservation.
4457 */
4458 if (space_info != delayed_rsv->space_info)
4459 return -ENOSPC;
4460
4461 spin_lock(&delayed_rsv->lock);
4462 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4463 bytes - delayed_rsv->size) >= 0) {
4464 spin_unlock(&delayed_rsv->lock);
4465 return -ENOSPC;
4466 }
4467 spin_unlock(&delayed_rsv->lock);
4468
4469 commit:
4470 trans = btrfs_join_transaction(root);
4471 if (IS_ERR(trans))
4472 return -ENOSPC;
4473
4474 return btrfs_commit_transaction(trans, root);
4475 }
4476
4477 enum flush_state {
4478 FLUSH_DELAYED_ITEMS_NR = 1,
4479 FLUSH_DELAYED_ITEMS = 2,
4480 FLUSH_DELALLOC = 3,
4481 FLUSH_DELALLOC_WAIT = 4,
4482 ALLOC_CHUNK = 5,
4483 COMMIT_TRANS = 6,
4484 };
4485
4486 static int flush_space(struct btrfs_root *root,
4487 struct btrfs_space_info *space_info, u64 num_bytes,
4488 u64 orig_bytes, int state)
4489 {
4490 struct btrfs_trans_handle *trans;
4491 int nr;
4492 int ret = 0;
4493
4494 switch (state) {
4495 case FLUSH_DELAYED_ITEMS_NR:
4496 case FLUSH_DELAYED_ITEMS:
4497 if (state == FLUSH_DELAYED_ITEMS_NR)
4498 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4499 else
4500 nr = -1;
4501
4502 trans = btrfs_join_transaction(root);
4503 if (IS_ERR(trans)) {
4504 ret = PTR_ERR(trans);
4505 break;
4506 }
4507 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4508 btrfs_end_transaction(trans, root);
4509 break;
4510 case FLUSH_DELALLOC:
4511 case FLUSH_DELALLOC_WAIT:
4512 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4513 state == FLUSH_DELALLOC_WAIT);
4514 break;
4515 case ALLOC_CHUNK:
4516 trans = btrfs_join_transaction(root);
4517 if (IS_ERR(trans)) {
4518 ret = PTR_ERR(trans);
4519 break;
4520 }
4521 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4522 btrfs_get_alloc_profile(root, 0),
4523 CHUNK_ALLOC_NO_FORCE);
4524 btrfs_end_transaction(trans, root);
4525 if (ret == -ENOSPC)
4526 ret = 0;
4527 break;
4528 case COMMIT_TRANS:
4529 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4530 break;
4531 default:
4532 ret = -ENOSPC;
4533 break;
4534 }
4535
4536 return ret;
4537 }
4538
4539 static inline u64
4540 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4541 struct btrfs_space_info *space_info)
4542 {
4543 u64 used;
4544 u64 expected;
4545 u64 to_reclaim;
4546
4547 to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024,
4548 16 * 1024 * 1024);
4549 spin_lock(&space_info->lock);
4550 if (can_overcommit(root, space_info, to_reclaim,
4551 BTRFS_RESERVE_FLUSH_ALL)) {
4552 to_reclaim = 0;
4553 goto out;
4554 }
4555
4556 used = space_info->bytes_used + space_info->bytes_reserved +
4557 space_info->bytes_pinned + space_info->bytes_readonly +
4558 space_info->bytes_may_use;
4559 if (can_overcommit(root, space_info, 1024 * 1024,
4560 BTRFS_RESERVE_FLUSH_ALL))
4561 expected = div_factor_fine(space_info->total_bytes, 95);
4562 else
4563 expected = div_factor_fine(space_info->total_bytes, 90);
4564
4565 if (used > expected)
4566 to_reclaim = used - expected;
4567 else
4568 to_reclaim = 0;
4569 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4570 space_info->bytes_reserved);
4571 out:
4572 spin_unlock(&space_info->lock);
4573
4574 return to_reclaim;
4575 }
4576
4577 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4578 struct btrfs_fs_info *fs_info, u64 used)
4579 {
4580 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4581
4582 /* If we're just plain full then async reclaim just slows us down. */
4583 if (space_info->bytes_used >= thresh)
4584 return 0;
4585
4586 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
4587 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4588 }
4589
4590 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4591 struct btrfs_fs_info *fs_info,
4592 int flush_state)
4593 {
4594 u64 used;
4595
4596 spin_lock(&space_info->lock);
4597 /*
4598 * We run out of space and have not got any free space via flush_space,
4599 * so don't bother doing async reclaim.
4600 */
4601 if (flush_state > COMMIT_TRANS && space_info->full) {
4602 spin_unlock(&space_info->lock);
4603 return 0;
4604 }
4605
4606 used = space_info->bytes_used + space_info->bytes_reserved +
4607 space_info->bytes_pinned + space_info->bytes_readonly +
4608 space_info->bytes_may_use;
4609 if (need_do_async_reclaim(space_info, fs_info, used)) {
4610 spin_unlock(&space_info->lock);
4611 return 1;
4612 }
4613 spin_unlock(&space_info->lock);
4614
4615 return 0;
4616 }
4617
4618 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4619 {
4620 struct btrfs_fs_info *fs_info;
4621 struct btrfs_space_info *space_info;
4622 u64 to_reclaim;
4623 int flush_state;
4624
4625 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4626 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4627
4628 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4629 space_info);
4630 if (!to_reclaim)
4631 return;
4632
4633 flush_state = FLUSH_DELAYED_ITEMS_NR;
4634 do {
4635 flush_space(fs_info->fs_root, space_info, to_reclaim,
4636 to_reclaim, flush_state);
4637 flush_state++;
4638 if (!btrfs_need_do_async_reclaim(space_info, fs_info,
4639 flush_state))
4640 return;
4641 } while (flush_state < COMMIT_TRANS);
4642 }
4643
4644 void btrfs_init_async_reclaim_work(struct work_struct *work)
4645 {
4646 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4647 }
4648
4649 /**
4650 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4651 * @root - the root we're allocating for
4652 * @block_rsv - the block_rsv we're allocating for
4653 * @orig_bytes - the number of bytes we want
4654 * @flush - whether or not we can flush to make our reservation
4655 *
4656 * This will reserve orgi_bytes number of bytes from the space info associated
4657 * with the block_rsv. If there is not enough space it will make an attempt to
4658 * flush out space to make room. It will do this by flushing delalloc if
4659 * possible or committing the transaction. If flush is 0 then no attempts to
4660 * regain reservations will be made and this will fail if there is not enough
4661 * space already.
4662 */
4663 static int reserve_metadata_bytes(struct btrfs_root *root,
4664 struct btrfs_block_rsv *block_rsv,
4665 u64 orig_bytes,
4666 enum btrfs_reserve_flush_enum flush)
4667 {
4668 struct btrfs_space_info *space_info = block_rsv->space_info;
4669 u64 used;
4670 u64 num_bytes = orig_bytes;
4671 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4672 int ret = 0;
4673 bool flushing = false;
4674
4675 again:
4676 ret = 0;
4677 spin_lock(&space_info->lock);
4678 /*
4679 * We only want to wait if somebody other than us is flushing and we
4680 * are actually allowed to flush all things.
4681 */
4682 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4683 space_info->flush) {
4684 spin_unlock(&space_info->lock);
4685 /*
4686 * If we have a trans handle we can't wait because the flusher
4687 * may have to commit the transaction, which would mean we would
4688 * deadlock since we are waiting for the flusher to finish, but
4689 * hold the current transaction open.
4690 */
4691 if (current->journal_info)
4692 return -EAGAIN;
4693 ret = wait_event_killable(space_info->wait, !space_info->flush);
4694 /* Must have been killed, return */
4695 if (ret)
4696 return -EINTR;
4697
4698 spin_lock(&space_info->lock);
4699 }
4700
4701 ret = -ENOSPC;
4702 used = space_info->bytes_used + space_info->bytes_reserved +
4703 space_info->bytes_pinned + space_info->bytes_readonly +
4704 space_info->bytes_may_use;
4705
4706 /*
4707 * The idea here is that we've not already over-reserved the block group
4708 * then we can go ahead and save our reservation first and then start
4709 * flushing if we need to. Otherwise if we've already overcommitted
4710 * lets start flushing stuff first and then come back and try to make
4711 * our reservation.
4712 */
4713 if (used <= space_info->total_bytes) {
4714 if (used + orig_bytes <= space_info->total_bytes) {
4715 space_info->bytes_may_use += orig_bytes;
4716 trace_btrfs_space_reservation(root->fs_info,
4717 "space_info", space_info->flags, orig_bytes, 1);
4718 ret = 0;
4719 } else {
4720 /*
4721 * Ok set num_bytes to orig_bytes since we aren't
4722 * overocmmitted, this way we only try and reclaim what
4723 * we need.
4724 */
4725 num_bytes = orig_bytes;
4726 }
4727 } else {
4728 /*
4729 * Ok we're over committed, set num_bytes to the overcommitted
4730 * amount plus the amount of bytes that we need for this
4731 * reservation.
4732 */
4733 num_bytes = used - space_info->total_bytes +
4734 (orig_bytes * 2);
4735 }
4736
4737 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4738 space_info->bytes_may_use += orig_bytes;
4739 trace_btrfs_space_reservation(root->fs_info, "space_info",
4740 space_info->flags, orig_bytes,
4741 1);
4742 ret = 0;
4743 }
4744
4745 /*
4746 * Couldn't make our reservation, save our place so while we're trying
4747 * to reclaim space we can actually use it instead of somebody else
4748 * stealing it from us.
4749 *
4750 * We make the other tasks wait for the flush only when we can flush
4751 * all things.
4752 */
4753 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4754 flushing = true;
4755 space_info->flush = 1;
4756 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
4757 used += orig_bytes;
4758 /*
4759 * We will do the space reservation dance during log replay,
4760 * which means we won't have fs_info->fs_root set, so don't do
4761 * the async reclaim as we will panic.
4762 */
4763 if (!root->fs_info->log_root_recovering &&
4764 need_do_async_reclaim(space_info, root->fs_info, used) &&
4765 !work_busy(&root->fs_info->async_reclaim_work))
4766 queue_work(system_unbound_wq,
4767 &root->fs_info->async_reclaim_work);
4768 }
4769 spin_unlock(&space_info->lock);
4770
4771 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4772 goto out;
4773
4774 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4775 flush_state);
4776 flush_state++;
4777
4778 /*
4779 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4780 * would happen. So skip delalloc flush.
4781 */
4782 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4783 (flush_state == FLUSH_DELALLOC ||
4784 flush_state == FLUSH_DELALLOC_WAIT))
4785 flush_state = ALLOC_CHUNK;
4786
4787 if (!ret)
4788 goto again;
4789 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4790 flush_state < COMMIT_TRANS)
4791 goto again;
4792 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4793 flush_state <= COMMIT_TRANS)
4794 goto again;
4795
4796 out:
4797 if (ret == -ENOSPC &&
4798 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4799 struct btrfs_block_rsv *global_rsv =
4800 &root->fs_info->global_block_rsv;
4801
4802 if (block_rsv != global_rsv &&
4803 !block_rsv_use_bytes(global_rsv, orig_bytes))
4804 ret = 0;
4805 }
4806 if (ret == -ENOSPC)
4807 trace_btrfs_space_reservation(root->fs_info,
4808 "space_info:enospc",
4809 space_info->flags, orig_bytes, 1);
4810 if (flushing) {
4811 spin_lock(&space_info->lock);
4812 space_info->flush = 0;
4813 wake_up_all(&space_info->wait);
4814 spin_unlock(&space_info->lock);
4815 }
4816 return ret;
4817 }
4818
4819 static struct btrfs_block_rsv *get_block_rsv(
4820 const struct btrfs_trans_handle *trans,
4821 const struct btrfs_root *root)
4822 {
4823 struct btrfs_block_rsv *block_rsv = NULL;
4824
4825 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4826 block_rsv = trans->block_rsv;
4827
4828 if (root == root->fs_info->csum_root && trans->adding_csums)
4829 block_rsv = trans->block_rsv;
4830
4831 if (root == root->fs_info->uuid_root)
4832 block_rsv = trans->block_rsv;
4833
4834 if (!block_rsv)
4835 block_rsv = root->block_rsv;
4836
4837 if (!block_rsv)
4838 block_rsv = &root->fs_info->empty_block_rsv;
4839
4840 return block_rsv;
4841 }
4842
4843 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4844 u64 num_bytes)
4845 {
4846 int ret = -ENOSPC;
4847 spin_lock(&block_rsv->lock);
4848 if (block_rsv->reserved >= num_bytes) {
4849 block_rsv->reserved -= num_bytes;
4850 if (block_rsv->reserved < block_rsv->size)
4851 block_rsv->full = 0;
4852 ret = 0;
4853 }
4854 spin_unlock(&block_rsv->lock);
4855 return ret;
4856 }
4857
4858 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4859 u64 num_bytes, int update_size)
4860 {
4861 spin_lock(&block_rsv->lock);
4862 block_rsv->reserved += num_bytes;
4863 if (update_size)
4864 block_rsv->size += num_bytes;
4865 else if (block_rsv->reserved >= block_rsv->size)
4866 block_rsv->full = 1;
4867 spin_unlock(&block_rsv->lock);
4868 }
4869
4870 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4871 struct btrfs_block_rsv *dest, u64 num_bytes,
4872 int min_factor)
4873 {
4874 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4875 u64 min_bytes;
4876
4877 if (global_rsv->space_info != dest->space_info)
4878 return -ENOSPC;
4879
4880 spin_lock(&global_rsv->lock);
4881 min_bytes = div_factor(global_rsv->size, min_factor);
4882 if (global_rsv->reserved < min_bytes + num_bytes) {
4883 spin_unlock(&global_rsv->lock);
4884 return -ENOSPC;
4885 }
4886 global_rsv->reserved -= num_bytes;
4887 if (global_rsv->reserved < global_rsv->size)
4888 global_rsv->full = 0;
4889 spin_unlock(&global_rsv->lock);
4890
4891 block_rsv_add_bytes(dest, num_bytes, 1);
4892 return 0;
4893 }
4894
4895 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4896 struct btrfs_block_rsv *block_rsv,
4897 struct btrfs_block_rsv *dest, u64 num_bytes)
4898 {
4899 struct btrfs_space_info *space_info = block_rsv->space_info;
4900
4901 spin_lock(&block_rsv->lock);
4902 if (num_bytes == (u64)-1)
4903 num_bytes = block_rsv->size;
4904 block_rsv->size -= num_bytes;
4905 if (block_rsv->reserved >= block_rsv->size) {
4906 num_bytes = block_rsv->reserved - block_rsv->size;
4907 block_rsv->reserved = block_rsv->size;
4908 block_rsv->full = 1;
4909 } else {
4910 num_bytes = 0;
4911 }
4912 spin_unlock(&block_rsv->lock);
4913
4914 if (num_bytes > 0) {
4915 if (dest) {
4916 spin_lock(&dest->lock);
4917 if (!dest->full) {
4918 u64 bytes_to_add;
4919
4920 bytes_to_add = dest->size - dest->reserved;
4921 bytes_to_add = min(num_bytes, bytes_to_add);
4922 dest->reserved += bytes_to_add;
4923 if (dest->reserved >= dest->size)
4924 dest->full = 1;
4925 num_bytes -= bytes_to_add;
4926 }
4927 spin_unlock(&dest->lock);
4928 }
4929 if (num_bytes) {
4930 spin_lock(&space_info->lock);
4931 space_info->bytes_may_use -= num_bytes;
4932 trace_btrfs_space_reservation(fs_info, "space_info",
4933 space_info->flags, num_bytes, 0);
4934 spin_unlock(&space_info->lock);
4935 }
4936 }
4937 }
4938
4939 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4940 struct btrfs_block_rsv *dst, u64 num_bytes)
4941 {
4942 int ret;
4943
4944 ret = block_rsv_use_bytes(src, num_bytes);
4945 if (ret)
4946 return ret;
4947
4948 block_rsv_add_bytes(dst, num_bytes, 1);
4949 return 0;
4950 }
4951
4952 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4953 {
4954 memset(rsv, 0, sizeof(*rsv));
4955 spin_lock_init(&rsv->lock);
4956 rsv->type = type;
4957 }
4958
4959 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4960 unsigned short type)
4961 {
4962 struct btrfs_block_rsv *block_rsv;
4963 struct btrfs_fs_info *fs_info = root->fs_info;
4964
4965 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4966 if (!block_rsv)
4967 return NULL;
4968
4969 btrfs_init_block_rsv(block_rsv, type);
4970 block_rsv->space_info = __find_space_info(fs_info,
4971 BTRFS_BLOCK_GROUP_METADATA);
4972 return block_rsv;
4973 }
4974
4975 void btrfs_free_block_rsv(struct btrfs_root *root,
4976 struct btrfs_block_rsv *rsv)
4977 {
4978 if (!rsv)
4979 return;
4980 btrfs_block_rsv_release(root, rsv, (u64)-1);
4981 kfree(rsv);
4982 }
4983
4984 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
4985 {
4986 kfree(rsv);
4987 }
4988
4989 int btrfs_block_rsv_add(struct btrfs_root *root,
4990 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4991 enum btrfs_reserve_flush_enum flush)
4992 {
4993 int ret;
4994
4995 if (num_bytes == 0)
4996 return 0;
4997
4998 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4999 if (!ret) {
5000 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5001 return 0;
5002 }
5003
5004 return ret;
5005 }
5006
5007 int btrfs_block_rsv_check(struct btrfs_root *root,
5008 struct btrfs_block_rsv *block_rsv, int min_factor)
5009 {
5010 u64 num_bytes = 0;
5011 int ret = -ENOSPC;
5012
5013 if (!block_rsv)
5014 return 0;
5015
5016 spin_lock(&block_rsv->lock);
5017 num_bytes = div_factor(block_rsv->size, min_factor);
5018 if (block_rsv->reserved >= num_bytes)
5019 ret = 0;
5020 spin_unlock(&block_rsv->lock);
5021
5022 return ret;
5023 }
5024
5025 int btrfs_block_rsv_refill(struct btrfs_root *root,
5026 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5027 enum btrfs_reserve_flush_enum flush)
5028 {
5029 u64 num_bytes = 0;
5030 int ret = -ENOSPC;
5031
5032 if (!block_rsv)
5033 return 0;
5034
5035 spin_lock(&block_rsv->lock);
5036 num_bytes = min_reserved;
5037 if (block_rsv->reserved >= num_bytes)
5038 ret = 0;
5039 else
5040 num_bytes -= block_rsv->reserved;
5041 spin_unlock(&block_rsv->lock);
5042
5043 if (!ret)
5044 return 0;
5045
5046 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5047 if (!ret) {
5048 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5049 return 0;
5050 }
5051
5052 return ret;
5053 }
5054
5055 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
5056 struct btrfs_block_rsv *dst_rsv,
5057 u64 num_bytes)
5058 {
5059 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5060 }
5061
5062 void btrfs_block_rsv_release(struct btrfs_root *root,
5063 struct btrfs_block_rsv *block_rsv,
5064 u64 num_bytes)
5065 {
5066 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5067 if (global_rsv == block_rsv ||
5068 block_rsv->space_info != global_rsv->space_info)
5069 global_rsv = NULL;
5070 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
5071 num_bytes);
5072 }
5073
5074 /*
5075 * helper to calculate size of global block reservation.
5076 * the desired value is sum of space used by extent tree,
5077 * checksum tree and root tree
5078 */
5079 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
5080 {
5081 struct btrfs_space_info *sinfo;
5082 u64 num_bytes;
5083 u64 meta_used;
5084 u64 data_used;
5085 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
5086
5087 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
5088 spin_lock(&sinfo->lock);
5089 data_used = sinfo->bytes_used;
5090 spin_unlock(&sinfo->lock);
5091
5092 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5093 spin_lock(&sinfo->lock);
5094 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
5095 data_used = 0;
5096 meta_used = sinfo->bytes_used;
5097 spin_unlock(&sinfo->lock);
5098
5099 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
5100 csum_size * 2;
5101 num_bytes += div_u64(data_used + meta_used, 50);
5102
5103 if (num_bytes * 3 > meta_used)
5104 num_bytes = div_u64(meta_used, 3);
5105
5106 return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
5107 }
5108
5109 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5110 {
5111 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5112 struct btrfs_space_info *sinfo = block_rsv->space_info;
5113 u64 num_bytes;
5114
5115 num_bytes = calc_global_metadata_size(fs_info);
5116
5117 spin_lock(&sinfo->lock);
5118 spin_lock(&block_rsv->lock);
5119
5120 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
5121
5122 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
5123 sinfo->bytes_reserved + sinfo->bytes_readonly +
5124 sinfo->bytes_may_use;
5125
5126 if (sinfo->total_bytes > num_bytes) {
5127 num_bytes = sinfo->total_bytes - num_bytes;
5128 block_rsv->reserved += num_bytes;
5129 sinfo->bytes_may_use += num_bytes;
5130 trace_btrfs_space_reservation(fs_info, "space_info",
5131 sinfo->flags, num_bytes, 1);
5132 }
5133
5134 if (block_rsv->reserved >= block_rsv->size) {
5135 num_bytes = block_rsv->reserved - block_rsv->size;
5136 sinfo->bytes_may_use -= num_bytes;
5137 trace_btrfs_space_reservation(fs_info, "space_info",
5138 sinfo->flags, num_bytes, 0);
5139 block_rsv->reserved = block_rsv->size;
5140 block_rsv->full = 1;
5141 }
5142
5143 spin_unlock(&block_rsv->lock);
5144 spin_unlock(&sinfo->lock);
5145 }
5146
5147 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5148 {
5149 struct btrfs_space_info *space_info;
5150
5151 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5152 fs_info->chunk_block_rsv.space_info = space_info;
5153
5154 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5155 fs_info->global_block_rsv.space_info = space_info;
5156 fs_info->delalloc_block_rsv.space_info = space_info;
5157 fs_info->trans_block_rsv.space_info = space_info;
5158 fs_info->empty_block_rsv.space_info = space_info;
5159 fs_info->delayed_block_rsv.space_info = space_info;
5160
5161 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5162 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5163 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5164 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5165 if (fs_info->quota_root)
5166 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5167 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5168
5169 update_global_block_rsv(fs_info);
5170 }
5171
5172 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5173 {
5174 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5175 (u64)-1);
5176 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5177 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5178 WARN_ON(fs_info->trans_block_rsv.size > 0);
5179 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5180 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5181 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5182 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5183 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5184 }
5185
5186 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5187 struct btrfs_root *root)
5188 {
5189 if (!trans->block_rsv)
5190 return;
5191
5192 if (!trans->bytes_reserved)
5193 return;
5194
5195 trace_btrfs_space_reservation(root->fs_info, "transaction",
5196 trans->transid, trans->bytes_reserved, 0);
5197 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
5198 trans->bytes_reserved = 0;
5199 }
5200
5201 /*
5202 * To be called after all the new block groups attached to the transaction
5203 * handle have been created (btrfs_create_pending_block_groups()).
5204 */
5205 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5206 {
5207 struct btrfs_fs_info *fs_info = trans->root->fs_info;
5208
5209 if (!trans->chunk_bytes_reserved)
5210 return;
5211
5212 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5213
5214 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5215 trans->chunk_bytes_reserved);
5216 trans->chunk_bytes_reserved = 0;
5217 }
5218
5219 /* Can only return 0 or -ENOSPC */
5220 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5221 struct inode *inode)
5222 {
5223 struct btrfs_root *root = BTRFS_I(inode)->root;
5224 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
5225 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5226
5227 /*
5228 * We need to hold space in order to delete our orphan item once we've
5229 * added it, so this takes the reservation so we can release it later
5230 * when we are truly done with the orphan item.
5231 */
5232 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5233 trace_btrfs_space_reservation(root->fs_info, "orphan",
5234 btrfs_ino(inode), num_bytes, 1);
5235 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5236 }
5237
5238 void btrfs_orphan_release_metadata(struct inode *inode)
5239 {
5240 struct btrfs_root *root = BTRFS_I(inode)->root;
5241 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5242 trace_btrfs_space_reservation(root->fs_info, "orphan",
5243 btrfs_ino(inode), num_bytes, 0);
5244 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
5245 }
5246
5247 /*
5248 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5249 * root: the root of the parent directory
5250 * rsv: block reservation
5251 * items: the number of items that we need do reservation
5252 * qgroup_reserved: used to return the reserved size in qgroup
5253 *
5254 * This function is used to reserve the space for snapshot/subvolume
5255 * creation and deletion. Those operations are different with the
5256 * common file/directory operations, they change two fs/file trees
5257 * and root tree, the number of items that the qgroup reserves is
5258 * different with the free space reservation. So we can not use
5259 * the space reseravtion mechanism in start_transaction().
5260 */
5261 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5262 struct btrfs_block_rsv *rsv,
5263 int items,
5264 u64 *qgroup_reserved,
5265 bool use_global_rsv)
5266 {
5267 u64 num_bytes;
5268 int ret;
5269 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5270
5271 if (root->fs_info->quota_enabled) {
5272 /* One for parent inode, two for dir entries */
5273 num_bytes = 3 * root->nodesize;
5274 ret = btrfs_qgroup_reserve(root, num_bytes);
5275 if (ret)
5276 return ret;
5277 } else {
5278 num_bytes = 0;
5279 }
5280
5281 *qgroup_reserved = num_bytes;
5282
5283 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5284 rsv->space_info = __find_space_info(root->fs_info,
5285 BTRFS_BLOCK_GROUP_METADATA);
5286 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5287 BTRFS_RESERVE_FLUSH_ALL);
5288
5289 if (ret == -ENOSPC && use_global_rsv)
5290 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
5291
5292 if (ret) {
5293 if (*qgroup_reserved)
5294 btrfs_qgroup_free(root, *qgroup_reserved);
5295 }
5296
5297 return ret;
5298 }
5299
5300 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5301 struct btrfs_block_rsv *rsv,
5302 u64 qgroup_reserved)
5303 {
5304 btrfs_block_rsv_release(root, rsv, (u64)-1);
5305 }
5306
5307 /**
5308 * drop_outstanding_extent - drop an outstanding extent
5309 * @inode: the inode we're dropping the extent for
5310 * @num_bytes: the number of bytes we're relaseing.
5311 *
5312 * This is called when we are freeing up an outstanding extent, either called
5313 * after an error or after an extent is written. This will return the number of
5314 * reserved extents that need to be freed. This must be called with
5315 * BTRFS_I(inode)->lock held.
5316 */
5317 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5318 {
5319 unsigned drop_inode_space = 0;
5320 unsigned dropped_extents = 0;
5321 unsigned num_extents = 0;
5322
5323 num_extents = (unsigned)div64_u64(num_bytes +
5324 BTRFS_MAX_EXTENT_SIZE - 1,
5325 BTRFS_MAX_EXTENT_SIZE);
5326 ASSERT(num_extents);
5327 ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5328 BTRFS_I(inode)->outstanding_extents -= num_extents;
5329
5330 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5331 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5332 &BTRFS_I(inode)->runtime_flags))
5333 drop_inode_space = 1;
5334
5335 /*
5336 * If we have more or the same amount of outsanding extents than we have
5337 * reserved then we need to leave the reserved extents count alone.
5338 */
5339 if (BTRFS_I(inode)->outstanding_extents >=
5340 BTRFS_I(inode)->reserved_extents)
5341 return drop_inode_space;
5342
5343 dropped_extents = BTRFS_I(inode)->reserved_extents -
5344 BTRFS_I(inode)->outstanding_extents;
5345 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5346 return dropped_extents + drop_inode_space;
5347 }
5348
5349 /**
5350 * calc_csum_metadata_size - return the amount of metada space that must be
5351 * reserved/free'd for the given bytes.
5352 * @inode: the inode we're manipulating
5353 * @num_bytes: the number of bytes in question
5354 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5355 *
5356 * This adjusts the number of csum_bytes in the inode and then returns the
5357 * correct amount of metadata that must either be reserved or freed. We
5358 * calculate how many checksums we can fit into one leaf and then divide the
5359 * number of bytes that will need to be checksumed by this value to figure out
5360 * how many checksums will be required. If we are adding bytes then the number
5361 * may go up and we will return the number of additional bytes that must be
5362 * reserved. If it is going down we will return the number of bytes that must
5363 * be freed.
5364 *
5365 * This must be called with BTRFS_I(inode)->lock held.
5366 */
5367 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5368 int reserve)
5369 {
5370 struct btrfs_root *root = BTRFS_I(inode)->root;
5371 u64 old_csums, num_csums;
5372
5373 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5374 BTRFS_I(inode)->csum_bytes == 0)
5375 return 0;
5376
5377 old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5378 if (reserve)
5379 BTRFS_I(inode)->csum_bytes += num_bytes;
5380 else
5381 BTRFS_I(inode)->csum_bytes -= num_bytes;
5382 num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5383
5384 /* No change, no need to reserve more */
5385 if (old_csums == num_csums)
5386 return 0;
5387
5388 if (reserve)
5389 return btrfs_calc_trans_metadata_size(root,
5390 num_csums - old_csums);
5391
5392 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5393 }
5394
5395 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5396 {
5397 struct btrfs_root *root = BTRFS_I(inode)->root;
5398 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5399 u64 to_reserve = 0;
5400 u64 csum_bytes;
5401 unsigned nr_extents = 0;
5402 int extra_reserve = 0;
5403 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5404 int ret = 0;
5405 bool delalloc_lock = true;
5406 u64 to_free = 0;
5407 unsigned dropped;
5408
5409 /* If we are a free space inode we need to not flush since we will be in
5410 * the middle of a transaction commit. We also don't need the delalloc
5411 * mutex since we won't race with anybody. We need this mostly to make
5412 * lockdep shut its filthy mouth.
5413 */
5414 if (btrfs_is_free_space_inode(inode)) {
5415 flush = BTRFS_RESERVE_NO_FLUSH;
5416 delalloc_lock = false;
5417 }
5418
5419 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5420 btrfs_transaction_in_commit(root->fs_info))
5421 schedule_timeout(1);
5422
5423 if (delalloc_lock)
5424 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5425
5426 num_bytes = ALIGN(num_bytes, root->sectorsize);
5427
5428 spin_lock(&BTRFS_I(inode)->lock);
5429 nr_extents = (unsigned)div64_u64(num_bytes +
5430 BTRFS_MAX_EXTENT_SIZE - 1,
5431 BTRFS_MAX_EXTENT_SIZE);
5432 BTRFS_I(inode)->outstanding_extents += nr_extents;
5433 nr_extents = 0;
5434
5435 if (BTRFS_I(inode)->outstanding_extents >
5436 BTRFS_I(inode)->reserved_extents)
5437 nr_extents = BTRFS_I(inode)->outstanding_extents -
5438 BTRFS_I(inode)->reserved_extents;
5439
5440 /*
5441 * Add an item to reserve for updating the inode when we complete the
5442 * delalloc io.
5443 */
5444 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5445 &BTRFS_I(inode)->runtime_flags)) {
5446 nr_extents++;
5447 extra_reserve = 1;
5448 }
5449
5450 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5451 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5452 csum_bytes = BTRFS_I(inode)->csum_bytes;
5453 spin_unlock(&BTRFS_I(inode)->lock);
5454
5455 if (root->fs_info->quota_enabled) {
5456 ret = btrfs_qgroup_reserve(root, nr_extents * root->nodesize);
5457 if (ret)
5458 goto out_fail;
5459 }
5460
5461 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5462 if (unlikely(ret)) {
5463 if (root->fs_info->quota_enabled)
5464 btrfs_qgroup_free(root, nr_extents * root->nodesize);
5465 goto out_fail;
5466 }
5467
5468 spin_lock(&BTRFS_I(inode)->lock);
5469 if (extra_reserve) {
5470 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5471 &BTRFS_I(inode)->runtime_flags);
5472 nr_extents--;
5473 }
5474 BTRFS_I(inode)->reserved_extents += nr_extents;
5475 spin_unlock(&BTRFS_I(inode)->lock);
5476
5477 if (delalloc_lock)
5478 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5479
5480 if (to_reserve)
5481 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5482 btrfs_ino(inode), to_reserve, 1);
5483 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5484
5485 return 0;
5486
5487 out_fail:
5488 spin_lock(&BTRFS_I(inode)->lock);
5489 dropped = drop_outstanding_extent(inode, num_bytes);
5490 /*
5491 * If the inodes csum_bytes is the same as the original
5492 * csum_bytes then we know we haven't raced with any free()ers
5493 * so we can just reduce our inodes csum bytes and carry on.
5494 */
5495 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5496 calc_csum_metadata_size(inode, num_bytes, 0);
5497 } else {
5498 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5499 u64 bytes;
5500
5501 /*
5502 * This is tricky, but first we need to figure out how much we
5503 * free'd from any free-ers that occured during this
5504 * reservation, so we reset ->csum_bytes to the csum_bytes
5505 * before we dropped our lock, and then call the free for the
5506 * number of bytes that were freed while we were trying our
5507 * reservation.
5508 */
5509 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5510 BTRFS_I(inode)->csum_bytes = csum_bytes;
5511 to_free = calc_csum_metadata_size(inode, bytes, 0);
5512
5513
5514 /*
5515 * Now we need to see how much we would have freed had we not
5516 * been making this reservation and our ->csum_bytes were not
5517 * artificially inflated.
5518 */
5519 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5520 bytes = csum_bytes - orig_csum_bytes;
5521 bytes = calc_csum_metadata_size(inode, bytes, 0);
5522
5523 /*
5524 * Now reset ->csum_bytes to what it should be. If bytes is
5525 * more than to_free then we would have free'd more space had we
5526 * not had an artificially high ->csum_bytes, so we need to free
5527 * the remainder. If bytes is the same or less then we don't
5528 * need to do anything, the other free-ers did the correct
5529 * thing.
5530 */
5531 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5532 if (bytes > to_free)
5533 to_free = bytes - to_free;
5534 else
5535 to_free = 0;
5536 }
5537 spin_unlock(&BTRFS_I(inode)->lock);
5538 if (dropped)
5539 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5540
5541 if (to_free) {
5542 btrfs_block_rsv_release(root, block_rsv, to_free);
5543 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5544 btrfs_ino(inode), to_free, 0);
5545 }
5546 if (delalloc_lock)
5547 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5548 return ret;
5549 }
5550
5551 /**
5552 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5553 * @inode: the inode to release the reservation for
5554 * @num_bytes: the number of bytes we're releasing
5555 *
5556 * This will release the metadata reservation for an inode. This can be called
5557 * once we complete IO for a given set of bytes to release their metadata
5558 * reservations.
5559 */
5560 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5561 {
5562 struct btrfs_root *root = BTRFS_I(inode)->root;
5563 u64 to_free = 0;
5564 unsigned dropped;
5565
5566 num_bytes = ALIGN(num_bytes, root->sectorsize);
5567 spin_lock(&BTRFS_I(inode)->lock);
5568 dropped = drop_outstanding_extent(inode, num_bytes);
5569
5570 if (num_bytes)
5571 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5572 spin_unlock(&BTRFS_I(inode)->lock);
5573 if (dropped > 0)
5574 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5575
5576 if (btrfs_test_is_dummy_root(root))
5577 return;
5578
5579 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5580 btrfs_ino(inode), to_free, 0);
5581
5582 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5583 to_free);
5584 }
5585
5586 /**
5587 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5588 * @inode: inode we're writing to
5589 * @num_bytes: the number of bytes we want to allocate
5590 *
5591 * This will do the following things
5592 *
5593 * o reserve space in the data space info for num_bytes
5594 * o reserve space in the metadata space info based on number of outstanding
5595 * extents and how much csums will be needed
5596 * o add to the inodes ->delalloc_bytes
5597 * o add it to the fs_info's delalloc inodes list.
5598 *
5599 * This will return 0 for success and -ENOSPC if there is no space left.
5600 */
5601 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5602 {
5603 int ret;
5604
5605 ret = btrfs_check_data_free_space(inode, num_bytes, num_bytes);
5606 if (ret)
5607 return ret;
5608
5609 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5610 if (ret) {
5611 btrfs_free_reserved_data_space(inode, num_bytes);
5612 return ret;
5613 }
5614
5615 return 0;
5616 }
5617
5618 /**
5619 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5620 * @inode: inode we're releasing space for
5621 * @num_bytes: the number of bytes we want to free up
5622 *
5623 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5624 * called in the case that we don't need the metadata AND data reservations
5625 * anymore. So if there is an error or we insert an inline extent.
5626 *
5627 * This function will release the metadata space that was not used and will
5628 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5629 * list if there are no delalloc bytes left.
5630 */
5631 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5632 {
5633 btrfs_delalloc_release_metadata(inode, num_bytes);
5634 btrfs_free_reserved_data_space(inode, num_bytes);
5635 }
5636
5637 static int update_block_group(struct btrfs_trans_handle *trans,
5638 struct btrfs_root *root, u64 bytenr,
5639 u64 num_bytes, int alloc)
5640 {
5641 struct btrfs_block_group_cache *cache = NULL;
5642 struct btrfs_fs_info *info = root->fs_info;
5643 u64 total = num_bytes;
5644 u64 old_val;
5645 u64 byte_in_group;
5646 int factor;
5647
5648 /* block accounting for super block */
5649 spin_lock(&info->delalloc_root_lock);
5650 old_val = btrfs_super_bytes_used(info->super_copy);
5651 if (alloc)
5652 old_val += num_bytes;
5653 else
5654 old_val -= num_bytes;
5655 btrfs_set_super_bytes_used(info->super_copy, old_val);
5656 spin_unlock(&info->delalloc_root_lock);
5657
5658 while (total) {
5659 cache = btrfs_lookup_block_group(info, bytenr);
5660 if (!cache)
5661 return -ENOENT;
5662 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5663 BTRFS_BLOCK_GROUP_RAID1 |
5664 BTRFS_BLOCK_GROUP_RAID10))
5665 factor = 2;
5666 else
5667 factor = 1;
5668 /*
5669 * If this block group has free space cache written out, we
5670 * need to make sure to load it if we are removing space. This
5671 * is because we need the unpinning stage to actually add the
5672 * space back to the block group, otherwise we will leak space.
5673 */
5674 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5675 cache_block_group(cache, 1);
5676
5677 byte_in_group = bytenr - cache->key.objectid;
5678 WARN_ON(byte_in_group > cache->key.offset);
5679
5680 spin_lock(&cache->space_info->lock);
5681 spin_lock(&cache->lock);
5682
5683 if (btrfs_test_opt(root, SPACE_CACHE) &&
5684 cache->disk_cache_state < BTRFS_DC_CLEAR)
5685 cache->disk_cache_state = BTRFS_DC_CLEAR;
5686
5687 old_val = btrfs_block_group_used(&cache->item);
5688 num_bytes = min(total, cache->key.offset - byte_in_group);
5689 if (alloc) {
5690 old_val += num_bytes;
5691 btrfs_set_block_group_used(&cache->item, old_val);
5692 cache->reserved -= num_bytes;
5693 cache->space_info->bytes_reserved -= num_bytes;
5694 cache->space_info->bytes_used += num_bytes;
5695 cache->space_info->disk_used += num_bytes * factor;
5696 spin_unlock(&cache->lock);
5697 spin_unlock(&cache->space_info->lock);
5698 } else {
5699 old_val -= num_bytes;
5700 btrfs_set_block_group_used(&cache->item, old_val);
5701 cache->pinned += num_bytes;
5702 cache->space_info->bytes_pinned += num_bytes;
5703 cache->space_info->bytes_used -= num_bytes;
5704 cache->space_info->disk_used -= num_bytes * factor;
5705 spin_unlock(&cache->lock);
5706 spin_unlock(&cache->space_info->lock);
5707
5708 set_extent_dirty(info->pinned_extents,
5709 bytenr, bytenr + num_bytes - 1,
5710 GFP_NOFS | __GFP_NOFAIL);
5711 /*
5712 * No longer have used bytes in this block group, queue
5713 * it for deletion.
5714 */
5715 if (old_val == 0) {
5716 spin_lock(&info->unused_bgs_lock);
5717 if (list_empty(&cache->bg_list)) {
5718 btrfs_get_block_group(cache);
5719 list_add_tail(&cache->bg_list,
5720 &info->unused_bgs);
5721 }
5722 spin_unlock(&info->unused_bgs_lock);
5723 }
5724 }
5725
5726 spin_lock(&trans->transaction->dirty_bgs_lock);
5727 if (list_empty(&cache->dirty_list)) {
5728 list_add_tail(&cache->dirty_list,
5729 &trans->transaction->dirty_bgs);
5730 trans->transaction->num_dirty_bgs++;
5731 btrfs_get_block_group(cache);
5732 }
5733 spin_unlock(&trans->transaction->dirty_bgs_lock);
5734
5735 btrfs_put_block_group(cache);
5736 total -= num_bytes;
5737 bytenr += num_bytes;
5738 }
5739 return 0;
5740 }
5741
5742 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5743 {
5744 struct btrfs_block_group_cache *cache;
5745 u64 bytenr;
5746
5747 spin_lock(&root->fs_info->block_group_cache_lock);
5748 bytenr = root->fs_info->first_logical_byte;
5749 spin_unlock(&root->fs_info->block_group_cache_lock);
5750
5751 if (bytenr < (u64)-1)
5752 return bytenr;
5753
5754 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5755 if (!cache)
5756 return 0;
5757
5758 bytenr = cache->key.objectid;
5759 btrfs_put_block_group(cache);
5760
5761 return bytenr;
5762 }
5763
5764 static int pin_down_extent(struct btrfs_root *root,
5765 struct btrfs_block_group_cache *cache,
5766 u64 bytenr, u64 num_bytes, int reserved)
5767 {
5768 spin_lock(&cache->space_info->lock);
5769 spin_lock(&cache->lock);
5770 cache->pinned += num_bytes;
5771 cache->space_info->bytes_pinned += num_bytes;
5772 if (reserved) {
5773 cache->reserved -= num_bytes;
5774 cache->space_info->bytes_reserved -= num_bytes;
5775 }
5776 spin_unlock(&cache->lock);
5777 spin_unlock(&cache->space_info->lock);
5778
5779 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5780 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5781 if (reserved)
5782 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5783 return 0;
5784 }
5785
5786 /*
5787 * this function must be called within transaction
5788 */
5789 int btrfs_pin_extent(struct btrfs_root *root,
5790 u64 bytenr, u64 num_bytes, int reserved)
5791 {
5792 struct btrfs_block_group_cache *cache;
5793
5794 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5795 BUG_ON(!cache); /* Logic error */
5796
5797 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5798
5799 btrfs_put_block_group(cache);
5800 return 0;
5801 }
5802
5803 /*
5804 * this function must be called within transaction
5805 */
5806 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5807 u64 bytenr, u64 num_bytes)
5808 {
5809 struct btrfs_block_group_cache *cache;
5810 int ret;
5811
5812 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5813 if (!cache)
5814 return -EINVAL;
5815
5816 /*
5817 * pull in the free space cache (if any) so that our pin
5818 * removes the free space from the cache. We have load_only set
5819 * to one because the slow code to read in the free extents does check
5820 * the pinned extents.
5821 */
5822 cache_block_group(cache, 1);
5823
5824 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5825
5826 /* remove us from the free space cache (if we're there at all) */
5827 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5828 btrfs_put_block_group(cache);
5829 return ret;
5830 }
5831
5832 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5833 {
5834 int ret;
5835 struct btrfs_block_group_cache *block_group;
5836 struct btrfs_caching_control *caching_ctl;
5837
5838 block_group = btrfs_lookup_block_group(root->fs_info, start);
5839 if (!block_group)
5840 return -EINVAL;
5841
5842 cache_block_group(block_group, 0);
5843 caching_ctl = get_caching_control(block_group);
5844
5845 if (!caching_ctl) {
5846 /* Logic error */
5847 BUG_ON(!block_group_cache_done(block_group));
5848 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5849 } else {
5850 mutex_lock(&caching_ctl->mutex);
5851
5852 if (start >= caching_ctl->progress) {
5853 ret = add_excluded_extent(root, start, num_bytes);
5854 } else if (start + num_bytes <= caching_ctl->progress) {
5855 ret = btrfs_remove_free_space(block_group,
5856 start, num_bytes);
5857 } else {
5858 num_bytes = caching_ctl->progress - start;
5859 ret = btrfs_remove_free_space(block_group,
5860 start, num_bytes);
5861 if (ret)
5862 goto out_lock;
5863
5864 num_bytes = (start + num_bytes) -
5865 caching_ctl->progress;
5866 start = caching_ctl->progress;
5867 ret = add_excluded_extent(root, start, num_bytes);
5868 }
5869 out_lock:
5870 mutex_unlock(&caching_ctl->mutex);
5871 put_caching_control(caching_ctl);
5872 }
5873 btrfs_put_block_group(block_group);
5874 return ret;
5875 }
5876
5877 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5878 struct extent_buffer *eb)
5879 {
5880 struct btrfs_file_extent_item *item;
5881 struct btrfs_key key;
5882 int found_type;
5883 int i;
5884
5885 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5886 return 0;
5887
5888 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5889 btrfs_item_key_to_cpu(eb, &key, i);
5890 if (key.type != BTRFS_EXTENT_DATA_KEY)
5891 continue;
5892 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5893 found_type = btrfs_file_extent_type(eb, item);
5894 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5895 continue;
5896 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5897 continue;
5898 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5899 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5900 __exclude_logged_extent(log, key.objectid, key.offset);
5901 }
5902
5903 return 0;
5904 }
5905
5906 /**
5907 * btrfs_update_reserved_bytes - update the block_group and space info counters
5908 * @cache: The cache we are manipulating
5909 * @num_bytes: The number of bytes in question
5910 * @reserve: One of the reservation enums
5911 * @delalloc: The blocks are allocated for the delalloc write
5912 *
5913 * This is called by the allocator when it reserves space, or by somebody who is
5914 * freeing space that was never actually used on disk. For example if you
5915 * reserve some space for a new leaf in transaction A and before transaction A
5916 * commits you free that leaf, you call this with reserve set to 0 in order to
5917 * clear the reservation.
5918 *
5919 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5920 * ENOSPC accounting. For data we handle the reservation through clearing the
5921 * delalloc bits in the io_tree. We have to do this since we could end up
5922 * allocating less disk space for the amount of data we have reserved in the
5923 * case of compression.
5924 *
5925 * If this is a reservation and the block group has become read only we cannot
5926 * make the reservation and return -EAGAIN, otherwise this function always
5927 * succeeds.
5928 */
5929 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5930 u64 num_bytes, int reserve, int delalloc)
5931 {
5932 struct btrfs_space_info *space_info = cache->space_info;
5933 int ret = 0;
5934
5935 spin_lock(&space_info->lock);
5936 spin_lock(&cache->lock);
5937 if (reserve != RESERVE_FREE) {
5938 if (cache->ro) {
5939 ret = -EAGAIN;
5940 } else {
5941 cache->reserved += num_bytes;
5942 space_info->bytes_reserved += num_bytes;
5943 if (reserve == RESERVE_ALLOC) {
5944 trace_btrfs_space_reservation(cache->fs_info,
5945 "space_info", space_info->flags,
5946 num_bytes, 0);
5947 space_info->bytes_may_use -= num_bytes;
5948 }
5949
5950 if (delalloc)
5951 cache->delalloc_bytes += num_bytes;
5952 }
5953 } else {
5954 if (cache->ro)
5955 space_info->bytes_readonly += num_bytes;
5956 cache->reserved -= num_bytes;
5957 space_info->bytes_reserved -= num_bytes;
5958
5959 if (delalloc)
5960 cache->delalloc_bytes -= num_bytes;
5961 }
5962 spin_unlock(&cache->lock);
5963 spin_unlock(&space_info->lock);
5964 return ret;
5965 }
5966
5967 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5968 struct btrfs_root *root)
5969 {
5970 struct btrfs_fs_info *fs_info = root->fs_info;
5971 struct btrfs_caching_control *next;
5972 struct btrfs_caching_control *caching_ctl;
5973 struct btrfs_block_group_cache *cache;
5974
5975 down_write(&fs_info->commit_root_sem);
5976
5977 list_for_each_entry_safe(caching_ctl, next,
5978 &fs_info->caching_block_groups, list) {
5979 cache = caching_ctl->block_group;
5980 if (block_group_cache_done(cache)) {
5981 cache->last_byte_to_unpin = (u64)-1;
5982 list_del_init(&caching_ctl->list);
5983 put_caching_control(caching_ctl);
5984 } else {
5985 cache->last_byte_to_unpin = caching_ctl->progress;
5986 }
5987 }
5988
5989 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5990 fs_info->pinned_extents = &fs_info->freed_extents[1];
5991 else
5992 fs_info->pinned_extents = &fs_info->freed_extents[0];
5993
5994 up_write(&fs_info->commit_root_sem);
5995
5996 update_global_block_rsv(fs_info);
5997 }
5998
5999 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
6000 const bool return_free_space)
6001 {
6002 struct btrfs_fs_info *fs_info = root->fs_info;
6003 struct btrfs_block_group_cache *cache = NULL;
6004 struct btrfs_space_info *space_info;
6005 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6006 u64 len;
6007 bool readonly;
6008
6009 while (start <= end) {
6010 readonly = false;
6011 if (!cache ||
6012 start >= cache->key.objectid + cache->key.offset) {
6013 if (cache)
6014 btrfs_put_block_group(cache);
6015 cache = btrfs_lookup_block_group(fs_info, start);
6016 BUG_ON(!cache); /* Logic error */
6017 }
6018
6019 len = cache->key.objectid + cache->key.offset - start;
6020 len = min(len, end + 1 - start);
6021
6022 if (start < cache->last_byte_to_unpin) {
6023 len = min(len, cache->last_byte_to_unpin - start);
6024 if (return_free_space)
6025 btrfs_add_free_space(cache, start, len);
6026 }
6027
6028 start += len;
6029 space_info = cache->space_info;
6030
6031 spin_lock(&space_info->lock);
6032 spin_lock(&cache->lock);
6033 cache->pinned -= len;
6034 space_info->bytes_pinned -= len;
6035 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6036 if (cache->ro) {
6037 space_info->bytes_readonly += len;
6038 readonly = true;
6039 }
6040 spin_unlock(&cache->lock);
6041 if (!readonly && global_rsv->space_info == space_info) {
6042 spin_lock(&global_rsv->lock);
6043 if (!global_rsv->full) {
6044 len = min(len, global_rsv->size -
6045 global_rsv->reserved);
6046 global_rsv->reserved += len;
6047 space_info->bytes_may_use += len;
6048 if (global_rsv->reserved >= global_rsv->size)
6049 global_rsv->full = 1;
6050 }
6051 spin_unlock(&global_rsv->lock);
6052 }
6053 spin_unlock(&space_info->lock);
6054 }
6055
6056 if (cache)
6057 btrfs_put_block_group(cache);
6058 return 0;
6059 }
6060
6061 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6062 struct btrfs_root *root)
6063 {
6064 struct btrfs_fs_info *fs_info = root->fs_info;
6065 struct extent_io_tree *unpin;
6066 u64 start;
6067 u64 end;
6068 int ret;
6069
6070 if (trans->aborted)
6071 return 0;
6072
6073 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6074 unpin = &fs_info->freed_extents[1];
6075 else
6076 unpin = &fs_info->freed_extents[0];
6077
6078 while (1) {
6079 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6080 ret = find_first_extent_bit(unpin, 0, &start, &end,
6081 EXTENT_DIRTY, NULL);
6082 if (ret) {
6083 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6084 break;
6085 }
6086
6087 if (btrfs_test_opt(root, DISCARD))
6088 ret = btrfs_discard_extent(root, start,
6089 end + 1 - start, NULL);
6090
6091 clear_extent_dirty(unpin, start, end, GFP_NOFS);
6092 unpin_extent_range(root, start, end, true);
6093 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6094 cond_resched();
6095 }
6096
6097 return 0;
6098 }
6099
6100 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
6101 u64 owner, u64 root_objectid)
6102 {
6103 struct btrfs_space_info *space_info;
6104 u64 flags;
6105
6106 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6107 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
6108 flags = BTRFS_BLOCK_GROUP_SYSTEM;
6109 else
6110 flags = BTRFS_BLOCK_GROUP_METADATA;
6111 } else {
6112 flags = BTRFS_BLOCK_GROUP_DATA;
6113 }
6114
6115 space_info = __find_space_info(fs_info, flags);
6116 BUG_ON(!space_info); /* Logic bug */
6117 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
6118 }
6119
6120
6121 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6122 struct btrfs_root *root,
6123 struct btrfs_delayed_ref_node *node, u64 parent,
6124 u64 root_objectid, u64 owner_objectid,
6125 u64 owner_offset, int refs_to_drop,
6126 struct btrfs_delayed_extent_op *extent_op)
6127 {
6128 struct btrfs_key key;
6129 struct btrfs_path *path;
6130 struct btrfs_fs_info *info = root->fs_info;
6131 struct btrfs_root *extent_root = info->extent_root;
6132 struct extent_buffer *leaf;
6133 struct btrfs_extent_item *ei;
6134 struct btrfs_extent_inline_ref *iref;
6135 int ret;
6136 int is_data;
6137 int extent_slot = 0;
6138 int found_extent = 0;
6139 int num_to_del = 1;
6140 int no_quota = node->no_quota;
6141 u32 item_size;
6142 u64 refs;
6143 u64 bytenr = node->bytenr;
6144 u64 num_bytes = node->num_bytes;
6145 int last_ref = 0;
6146 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6147 SKINNY_METADATA);
6148
6149 if (!info->quota_enabled || !is_fstree(root_objectid))
6150 no_quota = 1;
6151
6152 path = btrfs_alloc_path();
6153 if (!path)
6154 return -ENOMEM;
6155
6156 path->reada = 1;
6157 path->leave_spinning = 1;
6158
6159 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6160 BUG_ON(!is_data && refs_to_drop != 1);
6161
6162 if (is_data)
6163 skinny_metadata = 0;
6164
6165 ret = lookup_extent_backref(trans, extent_root, path, &iref,
6166 bytenr, num_bytes, parent,
6167 root_objectid, owner_objectid,
6168 owner_offset);
6169 if (ret == 0) {
6170 extent_slot = path->slots[0];
6171 while (extent_slot >= 0) {
6172 btrfs_item_key_to_cpu(path->nodes[0], &key,
6173 extent_slot);
6174 if (key.objectid != bytenr)
6175 break;
6176 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6177 key.offset == num_bytes) {
6178 found_extent = 1;
6179 break;
6180 }
6181 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6182 key.offset == owner_objectid) {
6183 found_extent = 1;
6184 break;
6185 }
6186 if (path->slots[0] - extent_slot > 5)
6187 break;
6188 extent_slot--;
6189 }
6190 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6191 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6192 if (found_extent && item_size < sizeof(*ei))
6193 found_extent = 0;
6194 #endif
6195 if (!found_extent) {
6196 BUG_ON(iref);
6197 ret = remove_extent_backref(trans, extent_root, path,
6198 NULL, refs_to_drop,
6199 is_data, &last_ref);
6200 if (ret) {
6201 btrfs_abort_transaction(trans, extent_root, ret);
6202 goto out;
6203 }
6204 btrfs_release_path(path);
6205 path->leave_spinning = 1;
6206
6207 key.objectid = bytenr;
6208 key.type = BTRFS_EXTENT_ITEM_KEY;
6209 key.offset = num_bytes;
6210
6211 if (!is_data && skinny_metadata) {
6212 key.type = BTRFS_METADATA_ITEM_KEY;
6213 key.offset = owner_objectid;
6214 }
6215
6216 ret = btrfs_search_slot(trans, extent_root,
6217 &key, path, -1, 1);
6218 if (ret > 0 && skinny_metadata && path->slots[0]) {
6219 /*
6220 * Couldn't find our skinny metadata item,
6221 * see if we have ye olde extent item.
6222 */
6223 path->slots[0]--;
6224 btrfs_item_key_to_cpu(path->nodes[0], &key,
6225 path->slots[0]);
6226 if (key.objectid == bytenr &&
6227 key.type == BTRFS_EXTENT_ITEM_KEY &&
6228 key.offset == num_bytes)
6229 ret = 0;
6230 }
6231
6232 if (ret > 0 && skinny_metadata) {
6233 skinny_metadata = false;
6234 key.objectid = bytenr;
6235 key.type = BTRFS_EXTENT_ITEM_KEY;
6236 key.offset = num_bytes;
6237 btrfs_release_path(path);
6238 ret = btrfs_search_slot(trans, extent_root,
6239 &key, path, -1, 1);
6240 }
6241
6242 if (ret) {
6243 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6244 ret, bytenr);
6245 if (ret > 0)
6246 btrfs_print_leaf(extent_root,
6247 path->nodes[0]);
6248 }
6249 if (ret < 0) {
6250 btrfs_abort_transaction(trans, extent_root, ret);
6251 goto out;
6252 }
6253 extent_slot = path->slots[0];
6254 }
6255 } else if (WARN_ON(ret == -ENOENT)) {
6256 btrfs_print_leaf(extent_root, path->nodes[0]);
6257 btrfs_err(info,
6258 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6259 bytenr, parent, root_objectid, owner_objectid,
6260 owner_offset);
6261 btrfs_abort_transaction(trans, extent_root, ret);
6262 goto out;
6263 } else {
6264 btrfs_abort_transaction(trans, extent_root, ret);
6265 goto out;
6266 }
6267
6268 leaf = path->nodes[0];
6269 item_size = btrfs_item_size_nr(leaf, extent_slot);
6270 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6271 if (item_size < sizeof(*ei)) {
6272 BUG_ON(found_extent || extent_slot != path->slots[0]);
6273 ret = convert_extent_item_v0(trans, extent_root, path,
6274 owner_objectid, 0);
6275 if (ret < 0) {
6276 btrfs_abort_transaction(trans, extent_root, ret);
6277 goto out;
6278 }
6279
6280 btrfs_release_path(path);
6281 path->leave_spinning = 1;
6282
6283 key.objectid = bytenr;
6284 key.type = BTRFS_EXTENT_ITEM_KEY;
6285 key.offset = num_bytes;
6286
6287 ret = btrfs_search_slot(trans, extent_root, &key, path,
6288 -1, 1);
6289 if (ret) {
6290 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6291 ret, bytenr);
6292 btrfs_print_leaf(extent_root, path->nodes[0]);
6293 }
6294 if (ret < 0) {
6295 btrfs_abort_transaction(trans, extent_root, ret);
6296 goto out;
6297 }
6298
6299 extent_slot = path->slots[0];
6300 leaf = path->nodes[0];
6301 item_size = btrfs_item_size_nr(leaf, extent_slot);
6302 }
6303 #endif
6304 BUG_ON(item_size < sizeof(*ei));
6305 ei = btrfs_item_ptr(leaf, extent_slot,
6306 struct btrfs_extent_item);
6307 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6308 key.type == BTRFS_EXTENT_ITEM_KEY) {
6309 struct btrfs_tree_block_info *bi;
6310 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6311 bi = (struct btrfs_tree_block_info *)(ei + 1);
6312 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6313 }
6314
6315 refs = btrfs_extent_refs(leaf, ei);
6316 if (refs < refs_to_drop) {
6317 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
6318 "for bytenr %Lu", refs_to_drop, refs, bytenr);
6319 ret = -EINVAL;
6320 btrfs_abort_transaction(trans, extent_root, ret);
6321 goto out;
6322 }
6323 refs -= refs_to_drop;
6324
6325 if (refs > 0) {
6326 if (extent_op)
6327 __run_delayed_extent_op(extent_op, leaf, ei);
6328 /*
6329 * In the case of inline back ref, reference count will
6330 * be updated by remove_extent_backref
6331 */
6332 if (iref) {
6333 BUG_ON(!found_extent);
6334 } else {
6335 btrfs_set_extent_refs(leaf, ei, refs);
6336 btrfs_mark_buffer_dirty(leaf);
6337 }
6338 if (found_extent) {
6339 ret = remove_extent_backref(trans, extent_root, path,
6340 iref, refs_to_drop,
6341 is_data, &last_ref);
6342 if (ret) {
6343 btrfs_abort_transaction(trans, extent_root, ret);
6344 goto out;
6345 }
6346 }
6347 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
6348 root_objectid);
6349 } else {
6350 if (found_extent) {
6351 BUG_ON(is_data && refs_to_drop !=
6352 extent_data_ref_count(root, path, iref));
6353 if (iref) {
6354 BUG_ON(path->slots[0] != extent_slot);
6355 } else {
6356 BUG_ON(path->slots[0] != extent_slot + 1);
6357 path->slots[0] = extent_slot;
6358 num_to_del = 2;
6359 }
6360 }
6361
6362 last_ref = 1;
6363 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6364 num_to_del);
6365 if (ret) {
6366 btrfs_abort_transaction(trans, extent_root, ret);
6367 goto out;
6368 }
6369 btrfs_release_path(path);
6370
6371 if (is_data) {
6372 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
6373 if (ret) {
6374 btrfs_abort_transaction(trans, extent_root, ret);
6375 goto out;
6376 }
6377 }
6378
6379 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
6380 if (ret) {
6381 btrfs_abort_transaction(trans, extent_root, ret);
6382 goto out;
6383 }
6384 }
6385 btrfs_release_path(path);
6386
6387 out:
6388 btrfs_free_path(path);
6389 return ret;
6390 }
6391
6392 /*
6393 * when we free an block, it is possible (and likely) that we free the last
6394 * delayed ref for that extent as well. This searches the delayed ref tree for
6395 * a given extent, and if there are no other delayed refs to be processed, it
6396 * removes it from the tree.
6397 */
6398 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6399 struct btrfs_root *root, u64 bytenr)
6400 {
6401 struct btrfs_delayed_ref_head *head;
6402 struct btrfs_delayed_ref_root *delayed_refs;
6403 int ret = 0;
6404
6405 delayed_refs = &trans->transaction->delayed_refs;
6406 spin_lock(&delayed_refs->lock);
6407 head = btrfs_find_delayed_ref_head(trans, bytenr);
6408 if (!head)
6409 goto out_delayed_unlock;
6410
6411 spin_lock(&head->lock);
6412 if (!list_empty(&head->ref_list))
6413 goto out;
6414
6415 if (head->extent_op) {
6416 if (!head->must_insert_reserved)
6417 goto out;
6418 btrfs_free_delayed_extent_op(head->extent_op);
6419 head->extent_op = NULL;
6420 }
6421
6422 /*
6423 * waiting for the lock here would deadlock. If someone else has it
6424 * locked they are already in the process of dropping it anyway
6425 */
6426 if (!mutex_trylock(&head->mutex))
6427 goto out;
6428
6429 /*
6430 * at this point we have a head with no other entries. Go
6431 * ahead and process it.
6432 */
6433 head->node.in_tree = 0;
6434 rb_erase(&head->href_node, &delayed_refs->href_root);
6435
6436 atomic_dec(&delayed_refs->num_entries);
6437
6438 /*
6439 * we don't take a ref on the node because we're removing it from the
6440 * tree, so we just steal the ref the tree was holding.
6441 */
6442 delayed_refs->num_heads--;
6443 if (head->processing == 0)
6444 delayed_refs->num_heads_ready--;
6445 head->processing = 0;
6446 spin_unlock(&head->lock);
6447 spin_unlock(&delayed_refs->lock);
6448
6449 BUG_ON(head->extent_op);
6450 if (head->must_insert_reserved)
6451 ret = 1;
6452
6453 mutex_unlock(&head->mutex);
6454 btrfs_put_delayed_ref(&head->node);
6455 return ret;
6456 out:
6457 spin_unlock(&head->lock);
6458
6459 out_delayed_unlock:
6460 spin_unlock(&delayed_refs->lock);
6461 return 0;
6462 }
6463
6464 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6465 struct btrfs_root *root,
6466 struct extent_buffer *buf,
6467 u64 parent, int last_ref)
6468 {
6469 int pin = 1;
6470 int ret;
6471
6472 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6473 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6474 buf->start, buf->len,
6475 parent, root->root_key.objectid,
6476 btrfs_header_level(buf),
6477 BTRFS_DROP_DELAYED_REF, NULL, 0);
6478 BUG_ON(ret); /* -ENOMEM */
6479 }
6480
6481 if (!last_ref)
6482 return;
6483
6484 if (btrfs_header_generation(buf) == trans->transid) {
6485 struct btrfs_block_group_cache *cache;
6486
6487 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6488 ret = check_ref_cleanup(trans, root, buf->start);
6489 if (!ret)
6490 goto out;
6491 }
6492
6493 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6494
6495 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6496 pin_down_extent(root, cache, buf->start, buf->len, 1);
6497 btrfs_put_block_group(cache);
6498 goto out;
6499 }
6500
6501 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6502
6503 btrfs_add_free_space(cache, buf->start, buf->len);
6504 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
6505 btrfs_put_block_group(cache);
6506 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6507 pin = 0;
6508 }
6509 out:
6510 if (pin)
6511 add_pinned_bytes(root->fs_info, buf->len,
6512 btrfs_header_level(buf),
6513 root->root_key.objectid);
6514
6515 /*
6516 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6517 * anymore.
6518 */
6519 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6520 }
6521
6522 /* Can return -ENOMEM */
6523 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6524 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6525 u64 owner, u64 offset, int no_quota)
6526 {
6527 int ret;
6528 struct btrfs_fs_info *fs_info = root->fs_info;
6529
6530 if (btrfs_test_is_dummy_root(root))
6531 return 0;
6532
6533 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6534
6535 /*
6536 * tree log blocks never actually go into the extent allocation
6537 * tree, just update pinning info and exit early.
6538 */
6539 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6540 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6541 /* unlocks the pinned mutex */
6542 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6543 ret = 0;
6544 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6545 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6546 num_bytes,
6547 parent, root_objectid, (int)owner,
6548 BTRFS_DROP_DELAYED_REF, NULL, no_quota);
6549 } else {
6550 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6551 num_bytes,
6552 parent, root_objectid, owner,
6553 offset, BTRFS_DROP_DELAYED_REF,
6554 NULL, no_quota);
6555 }
6556 return ret;
6557 }
6558
6559 /*
6560 * when we wait for progress in the block group caching, its because
6561 * our allocation attempt failed at least once. So, we must sleep
6562 * and let some progress happen before we try again.
6563 *
6564 * This function will sleep at least once waiting for new free space to
6565 * show up, and then it will check the block group free space numbers
6566 * for our min num_bytes. Another option is to have it go ahead
6567 * and look in the rbtree for a free extent of a given size, but this
6568 * is a good start.
6569 *
6570 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6571 * any of the information in this block group.
6572 */
6573 static noinline void
6574 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6575 u64 num_bytes)
6576 {
6577 struct btrfs_caching_control *caching_ctl;
6578
6579 caching_ctl = get_caching_control(cache);
6580 if (!caching_ctl)
6581 return;
6582
6583 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6584 (cache->free_space_ctl->free_space >= num_bytes));
6585
6586 put_caching_control(caching_ctl);
6587 }
6588
6589 static noinline int
6590 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6591 {
6592 struct btrfs_caching_control *caching_ctl;
6593 int ret = 0;
6594
6595 caching_ctl = get_caching_control(cache);
6596 if (!caching_ctl)
6597 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6598
6599 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6600 if (cache->cached == BTRFS_CACHE_ERROR)
6601 ret = -EIO;
6602 put_caching_control(caching_ctl);
6603 return ret;
6604 }
6605
6606 int __get_raid_index(u64 flags)
6607 {
6608 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6609 return BTRFS_RAID_RAID10;
6610 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6611 return BTRFS_RAID_RAID1;
6612 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6613 return BTRFS_RAID_DUP;
6614 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6615 return BTRFS_RAID_RAID0;
6616 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6617 return BTRFS_RAID_RAID5;
6618 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6619 return BTRFS_RAID_RAID6;
6620
6621 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6622 }
6623
6624 int get_block_group_index(struct btrfs_block_group_cache *cache)
6625 {
6626 return __get_raid_index(cache->flags);
6627 }
6628
6629 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6630 [BTRFS_RAID_RAID10] = "raid10",
6631 [BTRFS_RAID_RAID1] = "raid1",
6632 [BTRFS_RAID_DUP] = "dup",
6633 [BTRFS_RAID_RAID0] = "raid0",
6634 [BTRFS_RAID_SINGLE] = "single",
6635 [BTRFS_RAID_RAID5] = "raid5",
6636 [BTRFS_RAID_RAID6] = "raid6",
6637 };
6638
6639 static const char *get_raid_name(enum btrfs_raid_types type)
6640 {
6641 if (type >= BTRFS_NR_RAID_TYPES)
6642 return NULL;
6643
6644 return btrfs_raid_type_names[type];
6645 }
6646
6647 enum btrfs_loop_type {
6648 LOOP_CACHING_NOWAIT = 0,
6649 LOOP_CACHING_WAIT = 1,
6650 LOOP_ALLOC_CHUNK = 2,
6651 LOOP_NO_EMPTY_SIZE = 3,
6652 };
6653
6654 static inline void
6655 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
6656 int delalloc)
6657 {
6658 if (delalloc)
6659 down_read(&cache->data_rwsem);
6660 }
6661
6662 static inline void
6663 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
6664 int delalloc)
6665 {
6666 btrfs_get_block_group(cache);
6667 if (delalloc)
6668 down_read(&cache->data_rwsem);
6669 }
6670
6671 static struct btrfs_block_group_cache *
6672 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
6673 struct btrfs_free_cluster *cluster,
6674 int delalloc)
6675 {
6676 struct btrfs_block_group_cache *used_bg;
6677 bool locked = false;
6678 again:
6679 spin_lock(&cluster->refill_lock);
6680 if (locked) {
6681 if (used_bg == cluster->block_group)
6682 return used_bg;
6683
6684 up_read(&used_bg->data_rwsem);
6685 btrfs_put_block_group(used_bg);
6686 }
6687
6688 used_bg = cluster->block_group;
6689 if (!used_bg)
6690 return NULL;
6691
6692 if (used_bg == block_group)
6693 return used_bg;
6694
6695 btrfs_get_block_group(used_bg);
6696
6697 if (!delalloc)
6698 return used_bg;
6699
6700 if (down_read_trylock(&used_bg->data_rwsem))
6701 return used_bg;
6702
6703 spin_unlock(&cluster->refill_lock);
6704 down_read(&used_bg->data_rwsem);
6705 locked = true;
6706 goto again;
6707 }
6708
6709 static inline void
6710 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
6711 int delalloc)
6712 {
6713 if (delalloc)
6714 up_read(&cache->data_rwsem);
6715 btrfs_put_block_group(cache);
6716 }
6717
6718 /*
6719 * walks the btree of allocated extents and find a hole of a given size.
6720 * The key ins is changed to record the hole:
6721 * ins->objectid == start position
6722 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6723 * ins->offset == the size of the hole.
6724 * Any available blocks before search_start are skipped.
6725 *
6726 * If there is no suitable free space, we will record the max size of
6727 * the free space extent currently.
6728 */
6729 static noinline int find_free_extent(struct btrfs_root *orig_root,
6730 u64 num_bytes, u64 empty_size,
6731 u64 hint_byte, struct btrfs_key *ins,
6732 u64 flags, int delalloc)
6733 {
6734 int ret = 0;
6735 struct btrfs_root *root = orig_root->fs_info->extent_root;
6736 struct btrfs_free_cluster *last_ptr = NULL;
6737 struct btrfs_block_group_cache *block_group = NULL;
6738 u64 search_start = 0;
6739 u64 max_extent_size = 0;
6740 int empty_cluster = 2 * 1024 * 1024;
6741 struct btrfs_space_info *space_info;
6742 int loop = 0;
6743 int index = __get_raid_index(flags);
6744 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6745 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6746 bool failed_cluster_refill = false;
6747 bool failed_alloc = false;
6748 bool use_cluster = true;
6749 bool have_caching_bg = false;
6750
6751 WARN_ON(num_bytes < root->sectorsize);
6752 ins->type = BTRFS_EXTENT_ITEM_KEY;
6753 ins->objectid = 0;
6754 ins->offset = 0;
6755
6756 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6757
6758 space_info = __find_space_info(root->fs_info, flags);
6759 if (!space_info) {
6760 btrfs_err(root->fs_info, "No space info for %llu", flags);
6761 return -ENOSPC;
6762 }
6763
6764 /*
6765 * If the space info is for both data and metadata it means we have a
6766 * small filesystem and we can't use the clustering stuff.
6767 */
6768 if (btrfs_mixed_space_info(space_info))
6769 use_cluster = false;
6770
6771 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6772 last_ptr = &root->fs_info->meta_alloc_cluster;
6773 if (!btrfs_test_opt(root, SSD))
6774 empty_cluster = 64 * 1024;
6775 }
6776
6777 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6778 btrfs_test_opt(root, SSD)) {
6779 last_ptr = &root->fs_info->data_alloc_cluster;
6780 }
6781
6782 if (last_ptr) {
6783 spin_lock(&last_ptr->lock);
6784 if (last_ptr->block_group)
6785 hint_byte = last_ptr->window_start;
6786 spin_unlock(&last_ptr->lock);
6787 }
6788
6789 search_start = max(search_start, first_logical_byte(root, 0));
6790 search_start = max(search_start, hint_byte);
6791
6792 if (!last_ptr)
6793 empty_cluster = 0;
6794
6795 if (search_start == hint_byte) {
6796 block_group = btrfs_lookup_block_group(root->fs_info,
6797 search_start);
6798 /*
6799 * we don't want to use the block group if it doesn't match our
6800 * allocation bits, or if its not cached.
6801 *
6802 * However if we are re-searching with an ideal block group
6803 * picked out then we don't care that the block group is cached.
6804 */
6805 if (block_group && block_group_bits(block_group, flags) &&
6806 block_group->cached != BTRFS_CACHE_NO) {
6807 down_read(&space_info->groups_sem);
6808 if (list_empty(&block_group->list) ||
6809 block_group->ro) {
6810 /*
6811 * someone is removing this block group,
6812 * we can't jump into the have_block_group
6813 * target because our list pointers are not
6814 * valid
6815 */
6816 btrfs_put_block_group(block_group);
6817 up_read(&space_info->groups_sem);
6818 } else {
6819 index = get_block_group_index(block_group);
6820 btrfs_lock_block_group(block_group, delalloc);
6821 goto have_block_group;
6822 }
6823 } else if (block_group) {
6824 btrfs_put_block_group(block_group);
6825 }
6826 }
6827 search:
6828 have_caching_bg = false;
6829 down_read(&space_info->groups_sem);
6830 list_for_each_entry(block_group, &space_info->block_groups[index],
6831 list) {
6832 u64 offset;
6833 int cached;
6834
6835 btrfs_grab_block_group(block_group, delalloc);
6836 search_start = block_group->key.objectid;
6837
6838 /*
6839 * this can happen if we end up cycling through all the
6840 * raid types, but we want to make sure we only allocate
6841 * for the proper type.
6842 */
6843 if (!block_group_bits(block_group, flags)) {
6844 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6845 BTRFS_BLOCK_GROUP_RAID1 |
6846 BTRFS_BLOCK_GROUP_RAID5 |
6847 BTRFS_BLOCK_GROUP_RAID6 |
6848 BTRFS_BLOCK_GROUP_RAID10;
6849
6850 /*
6851 * if they asked for extra copies and this block group
6852 * doesn't provide them, bail. This does allow us to
6853 * fill raid0 from raid1.
6854 */
6855 if ((flags & extra) && !(block_group->flags & extra))
6856 goto loop;
6857 }
6858
6859 have_block_group:
6860 cached = block_group_cache_done(block_group);
6861 if (unlikely(!cached)) {
6862 ret = cache_block_group(block_group, 0);
6863 BUG_ON(ret < 0);
6864 ret = 0;
6865 }
6866
6867 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6868 goto loop;
6869 if (unlikely(block_group->ro))
6870 goto loop;
6871
6872 /*
6873 * Ok we want to try and use the cluster allocator, so
6874 * lets look there
6875 */
6876 if (last_ptr) {
6877 struct btrfs_block_group_cache *used_block_group;
6878 unsigned long aligned_cluster;
6879 /*
6880 * the refill lock keeps out other
6881 * people trying to start a new cluster
6882 */
6883 used_block_group = btrfs_lock_cluster(block_group,
6884 last_ptr,
6885 delalloc);
6886 if (!used_block_group)
6887 goto refill_cluster;
6888
6889 if (used_block_group != block_group &&
6890 (used_block_group->ro ||
6891 !block_group_bits(used_block_group, flags)))
6892 goto release_cluster;
6893
6894 offset = btrfs_alloc_from_cluster(used_block_group,
6895 last_ptr,
6896 num_bytes,
6897 used_block_group->key.objectid,
6898 &max_extent_size);
6899 if (offset) {
6900 /* we have a block, we're done */
6901 spin_unlock(&last_ptr->refill_lock);
6902 trace_btrfs_reserve_extent_cluster(root,
6903 used_block_group,
6904 search_start, num_bytes);
6905 if (used_block_group != block_group) {
6906 btrfs_release_block_group(block_group,
6907 delalloc);
6908 block_group = used_block_group;
6909 }
6910 goto checks;
6911 }
6912
6913 WARN_ON(last_ptr->block_group != used_block_group);
6914 release_cluster:
6915 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6916 * set up a new clusters, so lets just skip it
6917 * and let the allocator find whatever block
6918 * it can find. If we reach this point, we
6919 * will have tried the cluster allocator
6920 * plenty of times and not have found
6921 * anything, so we are likely way too
6922 * fragmented for the clustering stuff to find
6923 * anything.
6924 *
6925 * However, if the cluster is taken from the
6926 * current block group, release the cluster
6927 * first, so that we stand a better chance of
6928 * succeeding in the unclustered
6929 * allocation. */
6930 if (loop >= LOOP_NO_EMPTY_SIZE &&
6931 used_block_group != block_group) {
6932 spin_unlock(&last_ptr->refill_lock);
6933 btrfs_release_block_group(used_block_group,
6934 delalloc);
6935 goto unclustered_alloc;
6936 }
6937
6938 /*
6939 * this cluster didn't work out, free it and
6940 * start over
6941 */
6942 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6943
6944 if (used_block_group != block_group)
6945 btrfs_release_block_group(used_block_group,
6946 delalloc);
6947 refill_cluster:
6948 if (loop >= LOOP_NO_EMPTY_SIZE) {
6949 spin_unlock(&last_ptr->refill_lock);
6950 goto unclustered_alloc;
6951 }
6952
6953 aligned_cluster = max_t(unsigned long,
6954 empty_cluster + empty_size,
6955 block_group->full_stripe_len);
6956
6957 /* allocate a cluster in this block group */
6958 ret = btrfs_find_space_cluster(root, block_group,
6959 last_ptr, search_start,
6960 num_bytes,
6961 aligned_cluster);
6962 if (ret == 0) {
6963 /*
6964 * now pull our allocation out of this
6965 * cluster
6966 */
6967 offset = btrfs_alloc_from_cluster(block_group,
6968 last_ptr,
6969 num_bytes,
6970 search_start,
6971 &max_extent_size);
6972 if (offset) {
6973 /* we found one, proceed */
6974 spin_unlock(&last_ptr->refill_lock);
6975 trace_btrfs_reserve_extent_cluster(root,
6976 block_group, search_start,
6977 num_bytes);
6978 goto checks;
6979 }
6980 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6981 && !failed_cluster_refill) {
6982 spin_unlock(&last_ptr->refill_lock);
6983
6984 failed_cluster_refill = true;
6985 wait_block_group_cache_progress(block_group,
6986 num_bytes + empty_cluster + empty_size);
6987 goto have_block_group;
6988 }
6989
6990 /*
6991 * at this point we either didn't find a cluster
6992 * or we weren't able to allocate a block from our
6993 * cluster. Free the cluster we've been trying
6994 * to use, and go to the next block group
6995 */
6996 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6997 spin_unlock(&last_ptr->refill_lock);
6998 goto loop;
6999 }
7000
7001 unclustered_alloc:
7002 spin_lock(&block_group->free_space_ctl->tree_lock);
7003 if (cached &&
7004 block_group->free_space_ctl->free_space <
7005 num_bytes + empty_cluster + empty_size) {
7006 if (block_group->free_space_ctl->free_space >
7007 max_extent_size)
7008 max_extent_size =
7009 block_group->free_space_ctl->free_space;
7010 spin_unlock(&block_group->free_space_ctl->tree_lock);
7011 goto loop;
7012 }
7013 spin_unlock(&block_group->free_space_ctl->tree_lock);
7014
7015 offset = btrfs_find_space_for_alloc(block_group, search_start,
7016 num_bytes, empty_size,
7017 &max_extent_size);
7018 /*
7019 * If we didn't find a chunk, and we haven't failed on this
7020 * block group before, and this block group is in the middle of
7021 * caching and we are ok with waiting, then go ahead and wait
7022 * for progress to be made, and set failed_alloc to true.
7023 *
7024 * If failed_alloc is true then we've already waited on this
7025 * block group once and should move on to the next block group.
7026 */
7027 if (!offset && !failed_alloc && !cached &&
7028 loop > LOOP_CACHING_NOWAIT) {
7029 wait_block_group_cache_progress(block_group,
7030 num_bytes + empty_size);
7031 failed_alloc = true;
7032 goto have_block_group;
7033 } else if (!offset) {
7034 if (!cached)
7035 have_caching_bg = true;
7036 goto loop;
7037 }
7038 checks:
7039 search_start = ALIGN(offset, root->stripesize);
7040
7041 /* move on to the next group */
7042 if (search_start + num_bytes >
7043 block_group->key.objectid + block_group->key.offset) {
7044 btrfs_add_free_space(block_group, offset, num_bytes);
7045 goto loop;
7046 }
7047
7048 if (offset < search_start)
7049 btrfs_add_free_space(block_group, offset,
7050 search_start - offset);
7051 BUG_ON(offset > search_start);
7052
7053 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
7054 alloc_type, delalloc);
7055 if (ret == -EAGAIN) {
7056 btrfs_add_free_space(block_group, offset, num_bytes);
7057 goto loop;
7058 }
7059
7060 /* we are all good, lets return */
7061 ins->objectid = search_start;
7062 ins->offset = num_bytes;
7063
7064 trace_btrfs_reserve_extent(orig_root, block_group,
7065 search_start, num_bytes);
7066 btrfs_release_block_group(block_group, delalloc);
7067 break;
7068 loop:
7069 failed_cluster_refill = false;
7070 failed_alloc = false;
7071 BUG_ON(index != get_block_group_index(block_group));
7072 btrfs_release_block_group(block_group, delalloc);
7073 }
7074 up_read(&space_info->groups_sem);
7075
7076 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7077 goto search;
7078
7079 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7080 goto search;
7081
7082 /*
7083 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7084 * caching kthreads as we move along
7085 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7086 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7087 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7088 * again
7089 */
7090 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7091 index = 0;
7092 loop++;
7093 if (loop == LOOP_ALLOC_CHUNK) {
7094 struct btrfs_trans_handle *trans;
7095 int exist = 0;
7096
7097 trans = current->journal_info;
7098 if (trans)
7099 exist = 1;
7100 else
7101 trans = btrfs_join_transaction(root);
7102
7103 if (IS_ERR(trans)) {
7104 ret = PTR_ERR(trans);
7105 goto out;
7106 }
7107
7108 ret = do_chunk_alloc(trans, root, flags,
7109 CHUNK_ALLOC_FORCE);
7110 /*
7111 * Do not bail out on ENOSPC since we
7112 * can do more things.
7113 */
7114 if (ret < 0 && ret != -ENOSPC)
7115 btrfs_abort_transaction(trans,
7116 root, ret);
7117 else
7118 ret = 0;
7119 if (!exist)
7120 btrfs_end_transaction(trans, root);
7121 if (ret)
7122 goto out;
7123 }
7124
7125 if (loop == LOOP_NO_EMPTY_SIZE) {
7126 empty_size = 0;
7127 empty_cluster = 0;
7128 }
7129
7130 goto search;
7131 } else if (!ins->objectid) {
7132 ret = -ENOSPC;
7133 } else if (ins->objectid) {
7134 ret = 0;
7135 }
7136 out:
7137 if (ret == -ENOSPC)
7138 ins->offset = max_extent_size;
7139 return ret;
7140 }
7141
7142 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
7143 int dump_block_groups)
7144 {
7145 struct btrfs_block_group_cache *cache;
7146 int index = 0;
7147
7148 spin_lock(&info->lock);
7149 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
7150 info->flags,
7151 info->total_bytes - info->bytes_used - info->bytes_pinned -
7152 info->bytes_reserved - info->bytes_readonly,
7153 (info->full) ? "" : "not ");
7154 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7155 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7156 info->total_bytes, info->bytes_used, info->bytes_pinned,
7157 info->bytes_reserved, info->bytes_may_use,
7158 info->bytes_readonly);
7159 spin_unlock(&info->lock);
7160
7161 if (!dump_block_groups)
7162 return;
7163
7164 down_read(&info->groups_sem);
7165 again:
7166 list_for_each_entry(cache, &info->block_groups[index], list) {
7167 spin_lock(&cache->lock);
7168 printk(KERN_INFO "BTRFS: "
7169 "block group %llu has %llu bytes, "
7170 "%llu used %llu pinned %llu reserved %s\n",
7171 cache->key.objectid, cache->key.offset,
7172 btrfs_block_group_used(&cache->item), cache->pinned,
7173 cache->reserved, cache->ro ? "[readonly]" : "");
7174 btrfs_dump_free_space(cache, bytes);
7175 spin_unlock(&cache->lock);
7176 }
7177 if (++index < BTRFS_NR_RAID_TYPES)
7178 goto again;
7179 up_read(&info->groups_sem);
7180 }
7181
7182 int btrfs_reserve_extent(struct btrfs_root *root,
7183 u64 num_bytes, u64 min_alloc_size,
7184 u64 empty_size, u64 hint_byte,
7185 struct btrfs_key *ins, int is_data, int delalloc)
7186 {
7187 bool final_tried = false;
7188 u64 flags;
7189 int ret;
7190
7191 flags = btrfs_get_alloc_profile(root, is_data);
7192 again:
7193 WARN_ON(num_bytes < root->sectorsize);
7194 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
7195 flags, delalloc);
7196
7197 if (ret == -ENOSPC) {
7198 if (!final_tried && ins->offset) {
7199 num_bytes = min(num_bytes >> 1, ins->offset);
7200 num_bytes = round_down(num_bytes, root->sectorsize);
7201 num_bytes = max(num_bytes, min_alloc_size);
7202 if (num_bytes == min_alloc_size)
7203 final_tried = true;
7204 goto again;
7205 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7206 struct btrfs_space_info *sinfo;
7207
7208 sinfo = __find_space_info(root->fs_info, flags);
7209 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
7210 flags, num_bytes);
7211 if (sinfo)
7212 dump_space_info(sinfo, num_bytes, 1);
7213 }
7214 }
7215
7216 return ret;
7217 }
7218
7219 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
7220 u64 start, u64 len,
7221 int pin, int delalloc)
7222 {
7223 struct btrfs_block_group_cache *cache;
7224 int ret = 0;
7225
7226 cache = btrfs_lookup_block_group(root->fs_info, start);
7227 if (!cache) {
7228 btrfs_err(root->fs_info, "Unable to find block group for %llu",
7229 start);
7230 return -ENOSPC;
7231 }
7232
7233 if (pin)
7234 pin_down_extent(root, cache, start, len, 1);
7235 else {
7236 if (btrfs_test_opt(root, DISCARD))
7237 ret = btrfs_discard_extent(root, start, len, NULL);
7238 btrfs_add_free_space(cache, start, len);
7239 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
7240 }
7241
7242 btrfs_put_block_group(cache);
7243
7244 trace_btrfs_reserved_extent_free(root, start, len);
7245
7246 return ret;
7247 }
7248
7249 int btrfs_free_reserved_extent(struct btrfs_root *root,
7250 u64 start, u64 len, int delalloc)
7251 {
7252 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
7253 }
7254
7255 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
7256 u64 start, u64 len)
7257 {
7258 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
7259 }
7260
7261 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7262 struct btrfs_root *root,
7263 u64 parent, u64 root_objectid,
7264 u64 flags, u64 owner, u64 offset,
7265 struct btrfs_key *ins, int ref_mod)
7266 {
7267 int ret;
7268 struct btrfs_fs_info *fs_info = root->fs_info;
7269 struct btrfs_extent_item *extent_item;
7270 struct btrfs_extent_inline_ref *iref;
7271 struct btrfs_path *path;
7272 struct extent_buffer *leaf;
7273 int type;
7274 u32 size;
7275
7276 if (parent > 0)
7277 type = BTRFS_SHARED_DATA_REF_KEY;
7278 else
7279 type = BTRFS_EXTENT_DATA_REF_KEY;
7280
7281 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7282
7283 path = btrfs_alloc_path();
7284 if (!path)
7285 return -ENOMEM;
7286
7287 path->leave_spinning = 1;
7288 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7289 ins, size);
7290 if (ret) {
7291 btrfs_free_path(path);
7292 return ret;
7293 }
7294
7295 leaf = path->nodes[0];
7296 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7297 struct btrfs_extent_item);
7298 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
7299 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7300 btrfs_set_extent_flags(leaf, extent_item,
7301 flags | BTRFS_EXTENT_FLAG_DATA);
7302
7303 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7304 btrfs_set_extent_inline_ref_type(leaf, iref, type);
7305 if (parent > 0) {
7306 struct btrfs_shared_data_ref *ref;
7307 ref = (struct btrfs_shared_data_ref *)(iref + 1);
7308 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7309 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
7310 } else {
7311 struct btrfs_extent_data_ref *ref;
7312 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
7313 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
7314 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
7315 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
7316 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
7317 }
7318
7319 btrfs_mark_buffer_dirty(path->nodes[0]);
7320 btrfs_free_path(path);
7321
7322 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
7323 if (ret) { /* -ENOENT, logic error */
7324 btrfs_err(fs_info, "update block group failed for %llu %llu",
7325 ins->objectid, ins->offset);
7326 BUG();
7327 }
7328 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
7329 return ret;
7330 }
7331
7332 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
7333 struct btrfs_root *root,
7334 u64 parent, u64 root_objectid,
7335 u64 flags, struct btrfs_disk_key *key,
7336 int level, struct btrfs_key *ins,
7337 int no_quota)
7338 {
7339 int ret;
7340 struct btrfs_fs_info *fs_info = root->fs_info;
7341 struct btrfs_extent_item *extent_item;
7342 struct btrfs_tree_block_info *block_info;
7343 struct btrfs_extent_inline_ref *iref;
7344 struct btrfs_path *path;
7345 struct extent_buffer *leaf;
7346 u32 size = sizeof(*extent_item) + sizeof(*iref);
7347 u64 num_bytes = ins->offset;
7348 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7349 SKINNY_METADATA);
7350
7351 if (!skinny_metadata)
7352 size += sizeof(*block_info);
7353
7354 path = btrfs_alloc_path();
7355 if (!path) {
7356 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7357 root->nodesize);
7358 return -ENOMEM;
7359 }
7360
7361 path->leave_spinning = 1;
7362 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7363 ins, size);
7364 if (ret) {
7365 btrfs_free_path(path);
7366 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7367 root->nodesize);
7368 return ret;
7369 }
7370
7371 leaf = path->nodes[0];
7372 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7373 struct btrfs_extent_item);
7374 btrfs_set_extent_refs(leaf, extent_item, 1);
7375 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7376 btrfs_set_extent_flags(leaf, extent_item,
7377 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
7378
7379 if (skinny_metadata) {
7380 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7381 num_bytes = root->nodesize;
7382 } else {
7383 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
7384 btrfs_set_tree_block_key(leaf, block_info, key);
7385 btrfs_set_tree_block_level(leaf, block_info, level);
7386 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
7387 }
7388
7389 if (parent > 0) {
7390 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
7391 btrfs_set_extent_inline_ref_type(leaf, iref,
7392 BTRFS_SHARED_BLOCK_REF_KEY);
7393 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7394 } else {
7395 btrfs_set_extent_inline_ref_type(leaf, iref,
7396 BTRFS_TREE_BLOCK_REF_KEY);
7397 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
7398 }
7399
7400 btrfs_mark_buffer_dirty(leaf);
7401 btrfs_free_path(path);
7402
7403 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
7404 1);
7405 if (ret) { /* -ENOENT, logic error */
7406 btrfs_err(fs_info, "update block group failed for %llu %llu",
7407 ins->objectid, ins->offset);
7408 BUG();
7409 }
7410
7411 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
7412 return ret;
7413 }
7414
7415 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7416 struct btrfs_root *root,
7417 u64 root_objectid, u64 owner,
7418 u64 offset, struct btrfs_key *ins)
7419 {
7420 int ret;
7421
7422 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
7423
7424 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
7425 ins->offset, 0,
7426 root_objectid, owner, offset,
7427 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
7428 return ret;
7429 }
7430
7431 /*
7432 * this is used by the tree logging recovery code. It records that
7433 * an extent has been allocated and makes sure to clear the free
7434 * space cache bits as well
7435 */
7436 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
7437 struct btrfs_root *root,
7438 u64 root_objectid, u64 owner, u64 offset,
7439 struct btrfs_key *ins)
7440 {
7441 int ret;
7442 struct btrfs_block_group_cache *block_group;
7443
7444 /*
7445 * Mixed block groups will exclude before processing the log so we only
7446 * need to do the exlude dance if this fs isn't mixed.
7447 */
7448 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
7449 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
7450 if (ret)
7451 return ret;
7452 }
7453
7454 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
7455 if (!block_group)
7456 return -EINVAL;
7457
7458 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
7459 RESERVE_ALLOC_NO_ACCOUNT, 0);
7460 BUG_ON(ret); /* logic error */
7461 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
7462 0, owner, offset, ins, 1);
7463 btrfs_put_block_group(block_group);
7464 return ret;
7465 }
7466
7467 static struct extent_buffer *
7468 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7469 u64 bytenr, int level)
7470 {
7471 struct extent_buffer *buf;
7472
7473 buf = btrfs_find_create_tree_block(root, bytenr);
7474 if (!buf)
7475 return ERR_PTR(-ENOMEM);
7476 btrfs_set_header_generation(buf, trans->transid);
7477 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
7478 btrfs_tree_lock(buf);
7479 clean_tree_block(trans, root->fs_info, buf);
7480 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
7481
7482 btrfs_set_lock_blocking(buf);
7483 btrfs_set_buffer_uptodate(buf);
7484
7485 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
7486 buf->log_index = root->log_transid % 2;
7487 /*
7488 * we allow two log transactions at a time, use different
7489 * EXENT bit to differentiate dirty pages.
7490 */
7491 if (buf->log_index == 0)
7492 set_extent_dirty(&root->dirty_log_pages, buf->start,
7493 buf->start + buf->len - 1, GFP_NOFS);
7494 else
7495 set_extent_new(&root->dirty_log_pages, buf->start,
7496 buf->start + buf->len - 1, GFP_NOFS);
7497 } else {
7498 buf->log_index = -1;
7499 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
7500 buf->start + buf->len - 1, GFP_NOFS);
7501 }
7502 trans->blocks_used++;
7503 /* this returns a buffer locked for blocking */
7504 return buf;
7505 }
7506
7507 static struct btrfs_block_rsv *
7508 use_block_rsv(struct btrfs_trans_handle *trans,
7509 struct btrfs_root *root, u32 blocksize)
7510 {
7511 struct btrfs_block_rsv *block_rsv;
7512 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
7513 int ret;
7514 bool global_updated = false;
7515
7516 block_rsv = get_block_rsv(trans, root);
7517
7518 if (unlikely(block_rsv->size == 0))
7519 goto try_reserve;
7520 again:
7521 ret = block_rsv_use_bytes(block_rsv, blocksize);
7522 if (!ret)
7523 return block_rsv;
7524
7525 if (block_rsv->failfast)
7526 return ERR_PTR(ret);
7527
7528 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
7529 global_updated = true;
7530 update_global_block_rsv(root->fs_info);
7531 goto again;
7532 }
7533
7534 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7535 static DEFINE_RATELIMIT_STATE(_rs,
7536 DEFAULT_RATELIMIT_INTERVAL * 10,
7537 /*DEFAULT_RATELIMIT_BURST*/ 1);
7538 if (__ratelimit(&_rs))
7539 WARN(1, KERN_DEBUG
7540 "BTRFS: block rsv returned %d\n", ret);
7541 }
7542 try_reserve:
7543 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
7544 BTRFS_RESERVE_NO_FLUSH);
7545 if (!ret)
7546 return block_rsv;
7547 /*
7548 * If we couldn't reserve metadata bytes try and use some from
7549 * the global reserve if its space type is the same as the global
7550 * reservation.
7551 */
7552 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
7553 block_rsv->space_info == global_rsv->space_info) {
7554 ret = block_rsv_use_bytes(global_rsv, blocksize);
7555 if (!ret)
7556 return global_rsv;
7557 }
7558 return ERR_PTR(ret);
7559 }
7560
7561 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
7562 struct btrfs_block_rsv *block_rsv, u32 blocksize)
7563 {
7564 block_rsv_add_bytes(block_rsv, blocksize, 0);
7565 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
7566 }
7567
7568 /*
7569 * finds a free extent and does all the dirty work required for allocation
7570 * returns the key for the extent through ins, and a tree buffer for
7571 * the first block of the extent through buf.
7572 *
7573 * returns the tree buffer or an ERR_PTR on error.
7574 */
7575 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
7576 struct btrfs_root *root,
7577 u64 parent, u64 root_objectid,
7578 struct btrfs_disk_key *key, int level,
7579 u64 hint, u64 empty_size)
7580 {
7581 struct btrfs_key ins;
7582 struct btrfs_block_rsv *block_rsv;
7583 struct extent_buffer *buf;
7584 struct btrfs_delayed_extent_op *extent_op;
7585 u64 flags = 0;
7586 int ret;
7587 u32 blocksize = root->nodesize;
7588 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7589 SKINNY_METADATA);
7590
7591 if (btrfs_test_is_dummy_root(root)) {
7592 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
7593 level);
7594 if (!IS_ERR(buf))
7595 root->alloc_bytenr += blocksize;
7596 return buf;
7597 }
7598
7599 block_rsv = use_block_rsv(trans, root, blocksize);
7600 if (IS_ERR(block_rsv))
7601 return ERR_CAST(block_rsv);
7602
7603 ret = btrfs_reserve_extent(root, blocksize, blocksize,
7604 empty_size, hint, &ins, 0, 0);
7605 if (ret)
7606 goto out_unuse;
7607
7608 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
7609 if (IS_ERR(buf)) {
7610 ret = PTR_ERR(buf);
7611 goto out_free_reserved;
7612 }
7613
7614 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7615 if (parent == 0)
7616 parent = ins.objectid;
7617 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7618 } else
7619 BUG_ON(parent > 0);
7620
7621 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7622 extent_op = btrfs_alloc_delayed_extent_op();
7623 if (!extent_op) {
7624 ret = -ENOMEM;
7625 goto out_free_buf;
7626 }
7627 if (key)
7628 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7629 else
7630 memset(&extent_op->key, 0, sizeof(extent_op->key));
7631 extent_op->flags_to_set = flags;
7632 if (skinny_metadata)
7633 extent_op->update_key = 0;
7634 else
7635 extent_op->update_key = 1;
7636 extent_op->update_flags = 1;
7637 extent_op->is_data = 0;
7638 extent_op->level = level;
7639
7640 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7641 ins.objectid, ins.offset,
7642 parent, root_objectid, level,
7643 BTRFS_ADD_DELAYED_EXTENT,
7644 extent_op, 0);
7645 if (ret)
7646 goto out_free_delayed;
7647 }
7648 return buf;
7649
7650 out_free_delayed:
7651 btrfs_free_delayed_extent_op(extent_op);
7652 out_free_buf:
7653 free_extent_buffer(buf);
7654 out_free_reserved:
7655 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0);
7656 out_unuse:
7657 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7658 return ERR_PTR(ret);
7659 }
7660
7661 struct walk_control {
7662 u64 refs[BTRFS_MAX_LEVEL];
7663 u64 flags[BTRFS_MAX_LEVEL];
7664 struct btrfs_key update_progress;
7665 int stage;
7666 int level;
7667 int shared_level;
7668 int update_ref;
7669 int keep_locks;
7670 int reada_slot;
7671 int reada_count;
7672 int for_reloc;
7673 };
7674
7675 #define DROP_REFERENCE 1
7676 #define UPDATE_BACKREF 2
7677
7678 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7679 struct btrfs_root *root,
7680 struct walk_control *wc,
7681 struct btrfs_path *path)
7682 {
7683 u64 bytenr;
7684 u64 generation;
7685 u64 refs;
7686 u64 flags;
7687 u32 nritems;
7688 u32 blocksize;
7689 struct btrfs_key key;
7690 struct extent_buffer *eb;
7691 int ret;
7692 int slot;
7693 int nread = 0;
7694
7695 if (path->slots[wc->level] < wc->reada_slot) {
7696 wc->reada_count = wc->reada_count * 2 / 3;
7697 wc->reada_count = max(wc->reada_count, 2);
7698 } else {
7699 wc->reada_count = wc->reada_count * 3 / 2;
7700 wc->reada_count = min_t(int, wc->reada_count,
7701 BTRFS_NODEPTRS_PER_BLOCK(root));
7702 }
7703
7704 eb = path->nodes[wc->level];
7705 nritems = btrfs_header_nritems(eb);
7706 blocksize = root->nodesize;
7707
7708 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7709 if (nread >= wc->reada_count)
7710 break;
7711
7712 cond_resched();
7713 bytenr = btrfs_node_blockptr(eb, slot);
7714 generation = btrfs_node_ptr_generation(eb, slot);
7715
7716 if (slot == path->slots[wc->level])
7717 goto reada;
7718
7719 if (wc->stage == UPDATE_BACKREF &&
7720 generation <= root->root_key.offset)
7721 continue;
7722
7723 /* We don't lock the tree block, it's OK to be racy here */
7724 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7725 wc->level - 1, 1, &refs,
7726 &flags);
7727 /* We don't care about errors in readahead. */
7728 if (ret < 0)
7729 continue;
7730 BUG_ON(refs == 0);
7731
7732 if (wc->stage == DROP_REFERENCE) {
7733 if (refs == 1)
7734 goto reada;
7735
7736 if (wc->level == 1 &&
7737 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7738 continue;
7739 if (!wc->update_ref ||
7740 generation <= root->root_key.offset)
7741 continue;
7742 btrfs_node_key_to_cpu(eb, &key, slot);
7743 ret = btrfs_comp_cpu_keys(&key,
7744 &wc->update_progress);
7745 if (ret < 0)
7746 continue;
7747 } else {
7748 if (wc->level == 1 &&
7749 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7750 continue;
7751 }
7752 reada:
7753 readahead_tree_block(root, bytenr);
7754 nread++;
7755 }
7756 wc->reada_slot = slot;
7757 }
7758
7759 /*
7760 * TODO: Modify related function to add related node/leaf to dirty_extent_root,
7761 * for later qgroup accounting.
7762 *
7763 * Current, this function does nothing.
7764 */
7765 static int account_leaf_items(struct btrfs_trans_handle *trans,
7766 struct btrfs_root *root,
7767 struct extent_buffer *eb)
7768 {
7769 int nr = btrfs_header_nritems(eb);
7770 int i, extent_type;
7771 struct btrfs_key key;
7772 struct btrfs_file_extent_item *fi;
7773 u64 bytenr, num_bytes;
7774
7775 for (i = 0; i < nr; i++) {
7776 btrfs_item_key_to_cpu(eb, &key, i);
7777
7778 if (key.type != BTRFS_EXTENT_DATA_KEY)
7779 continue;
7780
7781 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
7782 /* filter out non qgroup-accountable extents */
7783 extent_type = btrfs_file_extent_type(eb, fi);
7784
7785 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
7786 continue;
7787
7788 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
7789 if (!bytenr)
7790 continue;
7791
7792 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
7793 }
7794 return 0;
7795 }
7796
7797 /*
7798 * Walk up the tree from the bottom, freeing leaves and any interior
7799 * nodes which have had all slots visited. If a node (leaf or
7800 * interior) is freed, the node above it will have it's slot
7801 * incremented. The root node will never be freed.
7802 *
7803 * At the end of this function, we should have a path which has all
7804 * slots incremented to the next position for a search. If we need to
7805 * read a new node it will be NULL and the node above it will have the
7806 * correct slot selected for a later read.
7807 *
7808 * If we increment the root nodes slot counter past the number of
7809 * elements, 1 is returned to signal completion of the search.
7810 */
7811 static int adjust_slots_upwards(struct btrfs_root *root,
7812 struct btrfs_path *path, int root_level)
7813 {
7814 int level = 0;
7815 int nr, slot;
7816 struct extent_buffer *eb;
7817
7818 if (root_level == 0)
7819 return 1;
7820
7821 while (level <= root_level) {
7822 eb = path->nodes[level];
7823 nr = btrfs_header_nritems(eb);
7824 path->slots[level]++;
7825 slot = path->slots[level];
7826 if (slot >= nr || level == 0) {
7827 /*
7828 * Don't free the root - we will detect this
7829 * condition after our loop and return a
7830 * positive value for caller to stop walking the tree.
7831 */
7832 if (level != root_level) {
7833 btrfs_tree_unlock_rw(eb, path->locks[level]);
7834 path->locks[level] = 0;
7835
7836 free_extent_buffer(eb);
7837 path->nodes[level] = NULL;
7838 path->slots[level] = 0;
7839 }
7840 } else {
7841 /*
7842 * We have a valid slot to walk back down
7843 * from. Stop here so caller can process these
7844 * new nodes.
7845 */
7846 break;
7847 }
7848
7849 level++;
7850 }
7851
7852 eb = path->nodes[root_level];
7853 if (path->slots[root_level] >= btrfs_header_nritems(eb))
7854 return 1;
7855
7856 return 0;
7857 }
7858
7859 /*
7860 * root_eb is the subtree root and is locked before this function is called.
7861 * TODO: Modify this function to mark all (including complete shared node)
7862 * to dirty_extent_root to allow it get accounted in qgroup.
7863 */
7864 static int account_shared_subtree(struct btrfs_trans_handle *trans,
7865 struct btrfs_root *root,
7866 struct extent_buffer *root_eb,
7867 u64 root_gen,
7868 int root_level)
7869 {
7870 int ret = 0;
7871 int level;
7872 struct extent_buffer *eb = root_eb;
7873 struct btrfs_path *path = NULL;
7874
7875 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
7876 BUG_ON(root_eb == NULL);
7877
7878 if (!root->fs_info->quota_enabled)
7879 return 0;
7880
7881 if (!extent_buffer_uptodate(root_eb)) {
7882 ret = btrfs_read_buffer(root_eb, root_gen);
7883 if (ret)
7884 goto out;
7885 }
7886
7887 if (root_level == 0) {
7888 ret = account_leaf_items(trans, root, root_eb);
7889 goto out;
7890 }
7891
7892 path = btrfs_alloc_path();
7893 if (!path)
7894 return -ENOMEM;
7895
7896 /*
7897 * Walk down the tree. Missing extent blocks are filled in as
7898 * we go. Metadata is accounted every time we read a new
7899 * extent block.
7900 *
7901 * When we reach a leaf, we account for file extent items in it,
7902 * walk back up the tree (adjusting slot pointers as we go)
7903 * and restart the search process.
7904 */
7905 extent_buffer_get(root_eb); /* For path */
7906 path->nodes[root_level] = root_eb;
7907 path->slots[root_level] = 0;
7908 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
7909 walk_down:
7910 level = root_level;
7911 while (level >= 0) {
7912 if (path->nodes[level] == NULL) {
7913 int parent_slot;
7914 u64 child_gen;
7915 u64 child_bytenr;
7916
7917 /* We need to get child blockptr/gen from
7918 * parent before we can read it. */
7919 eb = path->nodes[level + 1];
7920 parent_slot = path->slots[level + 1];
7921 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
7922 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
7923
7924 eb = read_tree_block(root, child_bytenr, child_gen);
7925 if (IS_ERR(eb)) {
7926 ret = PTR_ERR(eb);
7927 goto out;
7928 } else if (!extent_buffer_uptodate(eb)) {
7929 free_extent_buffer(eb);
7930 ret = -EIO;
7931 goto out;
7932 }
7933
7934 path->nodes[level] = eb;
7935 path->slots[level] = 0;
7936
7937 btrfs_tree_read_lock(eb);
7938 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
7939 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
7940 }
7941
7942 if (level == 0) {
7943 ret = account_leaf_items(trans, root, path->nodes[level]);
7944 if (ret)
7945 goto out;
7946
7947 /* Nonzero return here means we completed our search */
7948 ret = adjust_slots_upwards(root, path, root_level);
7949 if (ret)
7950 break;
7951
7952 /* Restart search with new slots */
7953 goto walk_down;
7954 }
7955
7956 level--;
7957 }
7958
7959 ret = 0;
7960 out:
7961 btrfs_free_path(path);
7962
7963 return ret;
7964 }
7965
7966 /*
7967 * helper to process tree block while walking down the tree.
7968 *
7969 * when wc->stage == UPDATE_BACKREF, this function updates
7970 * back refs for pointers in the block.
7971 *
7972 * NOTE: return value 1 means we should stop walking down.
7973 */
7974 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7975 struct btrfs_root *root,
7976 struct btrfs_path *path,
7977 struct walk_control *wc, int lookup_info)
7978 {
7979 int level = wc->level;
7980 struct extent_buffer *eb = path->nodes[level];
7981 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7982 int ret;
7983
7984 if (wc->stage == UPDATE_BACKREF &&
7985 btrfs_header_owner(eb) != root->root_key.objectid)
7986 return 1;
7987
7988 /*
7989 * when reference count of tree block is 1, it won't increase
7990 * again. once full backref flag is set, we never clear it.
7991 */
7992 if (lookup_info &&
7993 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
7994 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
7995 BUG_ON(!path->locks[level]);
7996 ret = btrfs_lookup_extent_info(trans, root,
7997 eb->start, level, 1,
7998 &wc->refs[level],
7999 &wc->flags[level]);
8000 BUG_ON(ret == -ENOMEM);
8001 if (ret)
8002 return ret;
8003 BUG_ON(wc->refs[level] == 0);
8004 }
8005
8006 if (wc->stage == DROP_REFERENCE) {
8007 if (wc->refs[level] > 1)
8008 return 1;
8009
8010 if (path->locks[level] && !wc->keep_locks) {
8011 btrfs_tree_unlock_rw(eb, path->locks[level]);
8012 path->locks[level] = 0;
8013 }
8014 return 0;
8015 }
8016
8017 /* wc->stage == UPDATE_BACKREF */
8018 if (!(wc->flags[level] & flag)) {
8019 BUG_ON(!path->locks[level]);
8020 ret = btrfs_inc_ref(trans, root, eb, 1);
8021 BUG_ON(ret); /* -ENOMEM */
8022 ret = btrfs_dec_ref(trans, root, eb, 0);
8023 BUG_ON(ret); /* -ENOMEM */
8024 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
8025 eb->len, flag,
8026 btrfs_header_level(eb), 0);
8027 BUG_ON(ret); /* -ENOMEM */
8028 wc->flags[level] |= flag;
8029 }
8030
8031 /*
8032 * the block is shared by multiple trees, so it's not good to
8033 * keep the tree lock
8034 */
8035 if (path->locks[level] && level > 0) {
8036 btrfs_tree_unlock_rw(eb, path->locks[level]);
8037 path->locks[level] = 0;
8038 }
8039 return 0;
8040 }
8041
8042 /*
8043 * helper to process tree block pointer.
8044 *
8045 * when wc->stage == DROP_REFERENCE, this function checks
8046 * reference count of the block pointed to. if the block
8047 * is shared and we need update back refs for the subtree
8048 * rooted at the block, this function changes wc->stage to
8049 * UPDATE_BACKREF. if the block is shared and there is no
8050 * need to update back, this function drops the reference
8051 * to the block.
8052 *
8053 * NOTE: return value 1 means we should stop walking down.
8054 */
8055 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8056 struct btrfs_root *root,
8057 struct btrfs_path *path,
8058 struct walk_control *wc, int *lookup_info)
8059 {
8060 u64 bytenr;
8061 u64 generation;
8062 u64 parent;
8063 u32 blocksize;
8064 struct btrfs_key key;
8065 struct extent_buffer *next;
8066 int level = wc->level;
8067 int reada = 0;
8068 int ret = 0;
8069 bool need_account = false;
8070
8071 generation = btrfs_node_ptr_generation(path->nodes[level],
8072 path->slots[level]);
8073 /*
8074 * if the lower level block was created before the snapshot
8075 * was created, we know there is no need to update back refs
8076 * for the subtree
8077 */
8078 if (wc->stage == UPDATE_BACKREF &&
8079 generation <= root->root_key.offset) {
8080 *lookup_info = 1;
8081 return 1;
8082 }
8083
8084 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8085 blocksize = root->nodesize;
8086
8087 next = btrfs_find_tree_block(root->fs_info, bytenr);
8088 if (!next) {
8089 next = btrfs_find_create_tree_block(root, bytenr);
8090 if (!next)
8091 return -ENOMEM;
8092 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8093 level - 1);
8094 reada = 1;
8095 }
8096 btrfs_tree_lock(next);
8097 btrfs_set_lock_blocking(next);
8098
8099 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
8100 &wc->refs[level - 1],
8101 &wc->flags[level - 1]);
8102 if (ret < 0) {
8103 btrfs_tree_unlock(next);
8104 return ret;
8105 }
8106
8107 if (unlikely(wc->refs[level - 1] == 0)) {
8108 btrfs_err(root->fs_info, "Missing references.");
8109 BUG();
8110 }
8111 *lookup_info = 0;
8112
8113 if (wc->stage == DROP_REFERENCE) {
8114 if (wc->refs[level - 1] > 1) {
8115 need_account = true;
8116 if (level == 1 &&
8117 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8118 goto skip;
8119
8120 if (!wc->update_ref ||
8121 generation <= root->root_key.offset)
8122 goto skip;
8123
8124 btrfs_node_key_to_cpu(path->nodes[level], &key,
8125 path->slots[level]);
8126 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8127 if (ret < 0)
8128 goto skip;
8129
8130 wc->stage = UPDATE_BACKREF;
8131 wc->shared_level = level - 1;
8132 }
8133 } else {
8134 if (level == 1 &&
8135 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8136 goto skip;
8137 }
8138
8139 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8140 btrfs_tree_unlock(next);
8141 free_extent_buffer(next);
8142 next = NULL;
8143 *lookup_info = 1;
8144 }
8145
8146 if (!next) {
8147 if (reada && level == 1)
8148 reada_walk_down(trans, root, wc, path);
8149 next = read_tree_block(root, bytenr, generation);
8150 if (IS_ERR(next)) {
8151 return PTR_ERR(next);
8152 } else if (!extent_buffer_uptodate(next)) {
8153 free_extent_buffer(next);
8154 return -EIO;
8155 }
8156 btrfs_tree_lock(next);
8157 btrfs_set_lock_blocking(next);
8158 }
8159
8160 level--;
8161 BUG_ON(level != btrfs_header_level(next));
8162 path->nodes[level] = next;
8163 path->slots[level] = 0;
8164 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8165 wc->level = level;
8166 if (wc->level == 1)
8167 wc->reada_slot = 0;
8168 return 0;
8169 skip:
8170 wc->refs[level - 1] = 0;
8171 wc->flags[level - 1] = 0;
8172 if (wc->stage == DROP_REFERENCE) {
8173 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8174 parent = path->nodes[level]->start;
8175 } else {
8176 BUG_ON(root->root_key.objectid !=
8177 btrfs_header_owner(path->nodes[level]));
8178 parent = 0;
8179 }
8180
8181 if (need_account) {
8182 ret = account_shared_subtree(trans, root, next,
8183 generation, level - 1);
8184 if (ret) {
8185 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
8186 "%d accounting shared subtree. Quota "
8187 "is out of sync, rescan required.\n",
8188 root->fs_info->sb->s_id, ret);
8189 }
8190 }
8191 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
8192 root->root_key.objectid, level - 1, 0, 0);
8193 BUG_ON(ret); /* -ENOMEM */
8194 }
8195 btrfs_tree_unlock(next);
8196 free_extent_buffer(next);
8197 *lookup_info = 1;
8198 return 1;
8199 }
8200
8201 /*
8202 * helper to process tree block while walking up the tree.
8203 *
8204 * when wc->stage == DROP_REFERENCE, this function drops
8205 * reference count on the block.
8206 *
8207 * when wc->stage == UPDATE_BACKREF, this function changes
8208 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8209 * to UPDATE_BACKREF previously while processing the block.
8210 *
8211 * NOTE: return value 1 means we should stop walking up.
8212 */
8213 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8214 struct btrfs_root *root,
8215 struct btrfs_path *path,
8216 struct walk_control *wc)
8217 {
8218 int ret;
8219 int level = wc->level;
8220 struct extent_buffer *eb = path->nodes[level];
8221 u64 parent = 0;
8222
8223 if (wc->stage == UPDATE_BACKREF) {
8224 BUG_ON(wc->shared_level < level);
8225 if (level < wc->shared_level)
8226 goto out;
8227
8228 ret = find_next_key(path, level + 1, &wc->update_progress);
8229 if (ret > 0)
8230 wc->update_ref = 0;
8231
8232 wc->stage = DROP_REFERENCE;
8233 wc->shared_level = -1;
8234 path->slots[level] = 0;
8235
8236 /*
8237 * check reference count again if the block isn't locked.
8238 * we should start walking down the tree again if reference
8239 * count is one.
8240 */
8241 if (!path->locks[level]) {
8242 BUG_ON(level == 0);
8243 btrfs_tree_lock(eb);
8244 btrfs_set_lock_blocking(eb);
8245 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8246
8247 ret = btrfs_lookup_extent_info(trans, root,
8248 eb->start, level, 1,
8249 &wc->refs[level],
8250 &wc->flags[level]);
8251 if (ret < 0) {
8252 btrfs_tree_unlock_rw(eb, path->locks[level]);
8253 path->locks[level] = 0;
8254 return ret;
8255 }
8256 BUG_ON(wc->refs[level] == 0);
8257 if (wc->refs[level] == 1) {
8258 btrfs_tree_unlock_rw(eb, path->locks[level]);
8259 path->locks[level] = 0;
8260 return 1;
8261 }
8262 }
8263 }
8264
8265 /* wc->stage == DROP_REFERENCE */
8266 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8267
8268 if (wc->refs[level] == 1) {
8269 if (level == 0) {
8270 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8271 ret = btrfs_dec_ref(trans, root, eb, 1);
8272 else
8273 ret = btrfs_dec_ref(trans, root, eb, 0);
8274 BUG_ON(ret); /* -ENOMEM */
8275 ret = account_leaf_items(trans, root, eb);
8276 if (ret) {
8277 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
8278 "%d accounting leaf items. Quota "
8279 "is out of sync, rescan required.\n",
8280 root->fs_info->sb->s_id, ret);
8281 }
8282 }
8283 /* make block locked assertion in clean_tree_block happy */
8284 if (!path->locks[level] &&
8285 btrfs_header_generation(eb) == trans->transid) {
8286 btrfs_tree_lock(eb);
8287 btrfs_set_lock_blocking(eb);
8288 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8289 }
8290 clean_tree_block(trans, root->fs_info, eb);
8291 }
8292
8293 if (eb == root->node) {
8294 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8295 parent = eb->start;
8296 else
8297 BUG_ON(root->root_key.objectid !=
8298 btrfs_header_owner(eb));
8299 } else {
8300 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8301 parent = path->nodes[level + 1]->start;
8302 else
8303 BUG_ON(root->root_key.objectid !=
8304 btrfs_header_owner(path->nodes[level + 1]));
8305 }
8306
8307 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8308 out:
8309 wc->refs[level] = 0;
8310 wc->flags[level] = 0;
8311 return 0;
8312 }
8313
8314 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8315 struct btrfs_root *root,
8316 struct btrfs_path *path,
8317 struct walk_control *wc)
8318 {
8319 int level = wc->level;
8320 int lookup_info = 1;
8321 int ret;
8322
8323 while (level >= 0) {
8324 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8325 if (ret > 0)
8326 break;
8327
8328 if (level == 0)
8329 break;
8330
8331 if (path->slots[level] >=
8332 btrfs_header_nritems(path->nodes[level]))
8333 break;
8334
8335 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8336 if (ret > 0) {
8337 path->slots[level]++;
8338 continue;
8339 } else if (ret < 0)
8340 return ret;
8341 level = wc->level;
8342 }
8343 return 0;
8344 }
8345
8346 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8347 struct btrfs_root *root,
8348 struct btrfs_path *path,
8349 struct walk_control *wc, int max_level)
8350 {
8351 int level = wc->level;
8352 int ret;
8353
8354 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8355 while (level < max_level && path->nodes[level]) {
8356 wc->level = level;
8357 if (path->slots[level] + 1 <
8358 btrfs_header_nritems(path->nodes[level])) {
8359 path->slots[level]++;
8360 return 0;
8361 } else {
8362 ret = walk_up_proc(trans, root, path, wc);
8363 if (ret > 0)
8364 return 0;
8365
8366 if (path->locks[level]) {
8367 btrfs_tree_unlock_rw(path->nodes[level],
8368 path->locks[level]);
8369 path->locks[level] = 0;
8370 }
8371 free_extent_buffer(path->nodes[level]);
8372 path->nodes[level] = NULL;
8373 level++;
8374 }
8375 }
8376 return 1;
8377 }
8378
8379 /*
8380 * drop a subvolume tree.
8381 *
8382 * this function traverses the tree freeing any blocks that only
8383 * referenced by the tree.
8384 *
8385 * when a shared tree block is found. this function decreases its
8386 * reference count by one. if update_ref is true, this function
8387 * also make sure backrefs for the shared block and all lower level
8388 * blocks are properly updated.
8389 *
8390 * If called with for_reloc == 0, may exit early with -EAGAIN
8391 */
8392 int btrfs_drop_snapshot(struct btrfs_root *root,
8393 struct btrfs_block_rsv *block_rsv, int update_ref,
8394 int for_reloc)
8395 {
8396 struct btrfs_path *path;
8397 struct btrfs_trans_handle *trans;
8398 struct btrfs_root *tree_root = root->fs_info->tree_root;
8399 struct btrfs_root_item *root_item = &root->root_item;
8400 struct walk_control *wc;
8401 struct btrfs_key key;
8402 int err = 0;
8403 int ret;
8404 int level;
8405 bool root_dropped = false;
8406
8407 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
8408
8409 path = btrfs_alloc_path();
8410 if (!path) {
8411 err = -ENOMEM;
8412 goto out;
8413 }
8414
8415 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8416 if (!wc) {
8417 btrfs_free_path(path);
8418 err = -ENOMEM;
8419 goto out;
8420 }
8421
8422 trans = btrfs_start_transaction(tree_root, 0);
8423 if (IS_ERR(trans)) {
8424 err = PTR_ERR(trans);
8425 goto out_free;
8426 }
8427
8428 if (block_rsv)
8429 trans->block_rsv = block_rsv;
8430
8431 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
8432 level = btrfs_header_level(root->node);
8433 path->nodes[level] = btrfs_lock_root_node(root);
8434 btrfs_set_lock_blocking(path->nodes[level]);
8435 path->slots[level] = 0;
8436 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8437 memset(&wc->update_progress, 0,
8438 sizeof(wc->update_progress));
8439 } else {
8440 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
8441 memcpy(&wc->update_progress, &key,
8442 sizeof(wc->update_progress));
8443
8444 level = root_item->drop_level;
8445 BUG_ON(level == 0);
8446 path->lowest_level = level;
8447 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8448 path->lowest_level = 0;
8449 if (ret < 0) {
8450 err = ret;
8451 goto out_end_trans;
8452 }
8453 WARN_ON(ret > 0);
8454
8455 /*
8456 * unlock our path, this is safe because only this
8457 * function is allowed to delete this snapshot
8458 */
8459 btrfs_unlock_up_safe(path, 0);
8460
8461 level = btrfs_header_level(root->node);
8462 while (1) {
8463 btrfs_tree_lock(path->nodes[level]);
8464 btrfs_set_lock_blocking(path->nodes[level]);
8465 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8466
8467 ret = btrfs_lookup_extent_info(trans, root,
8468 path->nodes[level]->start,
8469 level, 1, &wc->refs[level],
8470 &wc->flags[level]);
8471 if (ret < 0) {
8472 err = ret;
8473 goto out_end_trans;
8474 }
8475 BUG_ON(wc->refs[level] == 0);
8476
8477 if (level == root_item->drop_level)
8478 break;
8479
8480 btrfs_tree_unlock(path->nodes[level]);
8481 path->locks[level] = 0;
8482 WARN_ON(wc->refs[level] != 1);
8483 level--;
8484 }
8485 }
8486
8487 wc->level = level;
8488 wc->shared_level = -1;
8489 wc->stage = DROP_REFERENCE;
8490 wc->update_ref = update_ref;
8491 wc->keep_locks = 0;
8492 wc->for_reloc = for_reloc;
8493 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8494
8495 while (1) {
8496
8497 ret = walk_down_tree(trans, root, path, wc);
8498 if (ret < 0) {
8499 err = ret;
8500 break;
8501 }
8502
8503 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
8504 if (ret < 0) {
8505 err = ret;
8506 break;
8507 }
8508
8509 if (ret > 0) {
8510 BUG_ON(wc->stage != DROP_REFERENCE);
8511 break;
8512 }
8513
8514 if (wc->stage == DROP_REFERENCE) {
8515 level = wc->level;
8516 btrfs_node_key(path->nodes[level],
8517 &root_item->drop_progress,
8518 path->slots[level]);
8519 root_item->drop_level = level;
8520 }
8521
8522 BUG_ON(wc->level == 0);
8523 if (btrfs_should_end_transaction(trans, tree_root) ||
8524 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
8525 ret = btrfs_update_root(trans, tree_root,
8526 &root->root_key,
8527 root_item);
8528 if (ret) {
8529 btrfs_abort_transaction(trans, tree_root, ret);
8530 err = ret;
8531 goto out_end_trans;
8532 }
8533
8534 btrfs_end_transaction_throttle(trans, tree_root);
8535 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
8536 pr_debug("BTRFS: drop snapshot early exit\n");
8537 err = -EAGAIN;
8538 goto out_free;
8539 }
8540
8541 trans = btrfs_start_transaction(tree_root, 0);
8542 if (IS_ERR(trans)) {
8543 err = PTR_ERR(trans);
8544 goto out_free;
8545 }
8546 if (block_rsv)
8547 trans->block_rsv = block_rsv;
8548 }
8549 }
8550 btrfs_release_path(path);
8551 if (err)
8552 goto out_end_trans;
8553
8554 ret = btrfs_del_root(trans, tree_root, &root->root_key);
8555 if (ret) {
8556 btrfs_abort_transaction(trans, tree_root, ret);
8557 goto out_end_trans;
8558 }
8559
8560 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
8561 ret = btrfs_find_root(tree_root, &root->root_key, path,
8562 NULL, NULL);
8563 if (ret < 0) {
8564 btrfs_abort_transaction(trans, tree_root, ret);
8565 err = ret;
8566 goto out_end_trans;
8567 } else if (ret > 0) {
8568 /* if we fail to delete the orphan item this time
8569 * around, it'll get picked up the next time.
8570 *
8571 * The most common failure here is just -ENOENT.
8572 */
8573 btrfs_del_orphan_item(trans, tree_root,
8574 root->root_key.objectid);
8575 }
8576 }
8577
8578 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
8579 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
8580 } else {
8581 free_extent_buffer(root->node);
8582 free_extent_buffer(root->commit_root);
8583 btrfs_put_fs_root(root);
8584 }
8585 root_dropped = true;
8586 out_end_trans:
8587 btrfs_end_transaction_throttle(trans, tree_root);
8588 out_free:
8589 kfree(wc);
8590 btrfs_free_path(path);
8591 out:
8592 /*
8593 * So if we need to stop dropping the snapshot for whatever reason we
8594 * need to make sure to add it back to the dead root list so that we
8595 * keep trying to do the work later. This also cleans up roots if we
8596 * don't have it in the radix (like when we recover after a power fail
8597 * or unmount) so we don't leak memory.
8598 */
8599 if (!for_reloc && root_dropped == false)
8600 btrfs_add_dead_root(root);
8601 if (err && err != -EAGAIN)
8602 btrfs_std_error(root->fs_info, err);
8603 return err;
8604 }
8605
8606 /*
8607 * drop subtree rooted at tree block 'node'.
8608 *
8609 * NOTE: this function will unlock and release tree block 'node'
8610 * only used by relocation code
8611 */
8612 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
8613 struct btrfs_root *root,
8614 struct extent_buffer *node,
8615 struct extent_buffer *parent)
8616 {
8617 struct btrfs_path *path;
8618 struct walk_control *wc;
8619 int level;
8620 int parent_level;
8621 int ret = 0;
8622 int wret;
8623
8624 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
8625
8626 path = btrfs_alloc_path();
8627 if (!path)
8628 return -ENOMEM;
8629
8630 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8631 if (!wc) {
8632 btrfs_free_path(path);
8633 return -ENOMEM;
8634 }
8635
8636 btrfs_assert_tree_locked(parent);
8637 parent_level = btrfs_header_level(parent);
8638 extent_buffer_get(parent);
8639 path->nodes[parent_level] = parent;
8640 path->slots[parent_level] = btrfs_header_nritems(parent);
8641
8642 btrfs_assert_tree_locked(node);
8643 level = btrfs_header_level(node);
8644 path->nodes[level] = node;
8645 path->slots[level] = 0;
8646 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8647
8648 wc->refs[parent_level] = 1;
8649 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8650 wc->level = level;
8651 wc->shared_level = -1;
8652 wc->stage = DROP_REFERENCE;
8653 wc->update_ref = 0;
8654 wc->keep_locks = 1;
8655 wc->for_reloc = 1;
8656 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8657
8658 while (1) {
8659 wret = walk_down_tree(trans, root, path, wc);
8660 if (wret < 0) {
8661 ret = wret;
8662 break;
8663 }
8664
8665 wret = walk_up_tree(trans, root, path, wc, parent_level);
8666 if (wret < 0)
8667 ret = wret;
8668 if (wret != 0)
8669 break;
8670 }
8671
8672 kfree(wc);
8673 btrfs_free_path(path);
8674 return ret;
8675 }
8676
8677 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
8678 {
8679 u64 num_devices;
8680 u64 stripped;
8681
8682 /*
8683 * if restripe for this chunk_type is on pick target profile and
8684 * return, otherwise do the usual balance
8685 */
8686 stripped = get_restripe_target(root->fs_info, flags);
8687 if (stripped)
8688 return extended_to_chunk(stripped);
8689
8690 num_devices = root->fs_info->fs_devices->rw_devices;
8691
8692 stripped = BTRFS_BLOCK_GROUP_RAID0 |
8693 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
8694 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
8695
8696 if (num_devices == 1) {
8697 stripped |= BTRFS_BLOCK_GROUP_DUP;
8698 stripped = flags & ~stripped;
8699
8700 /* turn raid0 into single device chunks */
8701 if (flags & BTRFS_BLOCK_GROUP_RAID0)
8702 return stripped;
8703
8704 /* turn mirroring into duplication */
8705 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
8706 BTRFS_BLOCK_GROUP_RAID10))
8707 return stripped | BTRFS_BLOCK_GROUP_DUP;
8708 } else {
8709 /* they already had raid on here, just return */
8710 if (flags & stripped)
8711 return flags;
8712
8713 stripped |= BTRFS_BLOCK_GROUP_DUP;
8714 stripped = flags & ~stripped;
8715
8716 /* switch duplicated blocks with raid1 */
8717 if (flags & BTRFS_BLOCK_GROUP_DUP)
8718 return stripped | BTRFS_BLOCK_GROUP_RAID1;
8719
8720 /* this is drive concat, leave it alone */
8721 }
8722
8723 return flags;
8724 }
8725
8726 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
8727 {
8728 struct btrfs_space_info *sinfo = cache->space_info;
8729 u64 num_bytes;
8730 u64 min_allocable_bytes;
8731 int ret = -ENOSPC;
8732
8733
8734 /*
8735 * We need some metadata space and system metadata space for
8736 * allocating chunks in some corner cases until we force to set
8737 * it to be readonly.
8738 */
8739 if ((sinfo->flags &
8740 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
8741 !force)
8742 min_allocable_bytes = 1 * 1024 * 1024;
8743 else
8744 min_allocable_bytes = 0;
8745
8746 spin_lock(&sinfo->lock);
8747 spin_lock(&cache->lock);
8748
8749 if (cache->ro) {
8750 ret = 0;
8751 goto out;
8752 }
8753
8754 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8755 cache->bytes_super - btrfs_block_group_used(&cache->item);
8756
8757 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8758 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
8759 min_allocable_bytes <= sinfo->total_bytes) {
8760 sinfo->bytes_readonly += num_bytes;
8761 cache->ro = 1;
8762 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
8763 ret = 0;
8764 }
8765 out:
8766 spin_unlock(&cache->lock);
8767 spin_unlock(&sinfo->lock);
8768 return ret;
8769 }
8770
8771 int btrfs_set_block_group_ro(struct btrfs_root *root,
8772 struct btrfs_block_group_cache *cache)
8773
8774 {
8775 struct btrfs_trans_handle *trans;
8776 u64 alloc_flags;
8777 int ret;
8778
8779 BUG_ON(cache->ro);
8780
8781 again:
8782 trans = btrfs_join_transaction(root);
8783 if (IS_ERR(trans))
8784 return PTR_ERR(trans);
8785
8786 /*
8787 * we're not allowed to set block groups readonly after the dirty
8788 * block groups cache has started writing. If it already started,
8789 * back off and let this transaction commit
8790 */
8791 mutex_lock(&root->fs_info->ro_block_group_mutex);
8792 if (trans->transaction->dirty_bg_run) {
8793 u64 transid = trans->transid;
8794
8795 mutex_unlock(&root->fs_info->ro_block_group_mutex);
8796 btrfs_end_transaction(trans, root);
8797
8798 ret = btrfs_wait_for_commit(root, transid);
8799 if (ret)
8800 return ret;
8801 goto again;
8802 }
8803
8804 /*
8805 * if we are changing raid levels, try to allocate a corresponding
8806 * block group with the new raid level.
8807 */
8808 alloc_flags = update_block_group_flags(root, cache->flags);
8809 if (alloc_flags != cache->flags) {
8810 ret = do_chunk_alloc(trans, root, alloc_flags,
8811 CHUNK_ALLOC_FORCE);
8812 /*
8813 * ENOSPC is allowed here, we may have enough space
8814 * already allocated at the new raid level to
8815 * carry on
8816 */
8817 if (ret == -ENOSPC)
8818 ret = 0;
8819 if (ret < 0)
8820 goto out;
8821 }
8822
8823 ret = set_block_group_ro(cache, 0);
8824 if (!ret)
8825 goto out;
8826 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8827 ret = do_chunk_alloc(trans, root, alloc_flags,
8828 CHUNK_ALLOC_FORCE);
8829 if (ret < 0)
8830 goto out;
8831 ret = set_block_group_ro(cache, 0);
8832 out:
8833 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
8834 alloc_flags = update_block_group_flags(root, cache->flags);
8835 lock_chunks(root->fs_info->chunk_root);
8836 check_system_chunk(trans, root, alloc_flags);
8837 unlock_chunks(root->fs_info->chunk_root);
8838 }
8839 mutex_unlock(&root->fs_info->ro_block_group_mutex);
8840
8841 btrfs_end_transaction(trans, root);
8842 return ret;
8843 }
8844
8845 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8846 struct btrfs_root *root, u64 type)
8847 {
8848 u64 alloc_flags = get_alloc_profile(root, type);
8849 return do_chunk_alloc(trans, root, alloc_flags,
8850 CHUNK_ALLOC_FORCE);
8851 }
8852
8853 /*
8854 * helper to account the unused space of all the readonly block group in the
8855 * space_info. takes mirrors into account.
8856 */
8857 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8858 {
8859 struct btrfs_block_group_cache *block_group;
8860 u64 free_bytes = 0;
8861 int factor;
8862
8863 /* It's df, we don't care if it's racey */
8864 if (list_empty(&sinfo->ro_bgs))
8865 return 0;
8866
8867 spin_lock(&sinfo->lock);
8868 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
8869 spin_lock(&block_group->lock);
8870
8871 if (!block_group->ro) {
8872 spin_unlock(&block_group->lock);
8873 continue;
8874 }
8875
8876 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8877 BTRFS_BLOCK_GROUP_RAID10 |
8878 BTRFS_BLOCK_GROUP_DUP))
8879 factor = 2;
8880 else
8881 factor = 1;
8882
8883 free_bytes += (block_group->key.offset -
8884 btrfs_block_group_used(&block_group->item)) *
8885 factor;
8886
8887 spin_unlock(&block_group->lock);
8888 }
8889 spin_unlock(&sinfo->lock);
8890
8891 return free_bytes;
8892 }
8893
8894 void btrfs_set_block_group_rw(struct btrfs_root *root,
8895 struct btrfs_block_group_cache *cache)
8896 {
8897 struct btrfs_space_info *sinfo = cache->space_info;
8898 u64 num_bytes;
8899
8900 BUG_ON(!cache->ro);
8901
8902 spin_lock(&sinfo->lock);
8903 spin_lock(&cache->lock);
8904 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8905 cache->bytes_super - btrfs_block_group_used(&cache->item);
8906 sinfo->bytes_readonly -= num_bytes;
8907 cache->ro = 0;
8908 list_del_init(&cache->ro_list);
8909 spin_unlock(&cache->lock);
8910 spin_unlock(&sinfo->lock);
8911 }
8912
8913 /*
8914 * checks to see if its even possible to relocate this block group.
8915 *
8916 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8917 * ok to go ahead and try.
8918 */
8919 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8920 {
8921 struct btrfs_block_group_cache *block_group;
8922 struct btrfs_space_info *space_info;
8923 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8924 struct btrfs_device *device;
8925 struct btrfs_trans_handle *trans;
8926 u64 min_free;
8927 u64 dev_min = 1;
8928 u64 dev_nr = 0;
8929 u64 target;
8930 int index;
8931 int full = 0;
8932 int ret = 0;
8933
8934 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8935
8936 /* odd, couldn't find the block group, leave it alone */
8937 if (!block_group)
8938 return -1;
8939
8940 min_free = btrfs_block_group_used(&block_group->item);
8941
8942 /* no bytes used, we're good */
8943 if (!min_free)
8944 goto out;
8945
8946 space_info = block_group->space_info;
8947 spin_lock(&space_info->lock);
8948
8949 full = space_info->full;
8950
8951 /*
8952 * if this is the last block group we have in this space, we can't
8953 * relocate it unless we're able to allocate a new chunk below.
8954 *
8955 * Otherwise, we need to make sure we have room in the space to handle
8956 * all of the extents from this block group. If we can, we're good
8957 */
8958 if ((space_info->total_bytes != block_group->key.offset) &&
8959 (space_info->bytes_used + space_info->bytes_reserved +
8960 space_info->bytes_pinned + space_info->bytes_readonly +
8961 min_free < space_info->total_bytes)) {
8962 spin_unlock(&space_info->lock);
8963 goto out;
8964 }
8965 spin_unlock(&space_info->lock);
8966
8967 /*
8968 * ok we don't have enough space, but maybe we have free space on our
8969 * devices to allocate new chunks for relocation, so loop through our
8970 * alloc devices and guess if we have enough space. if this block
8971 * group is going to be restriped, run checks against the target
8972 * profile instead of the current one.
8973 */
8974 ret = -1;
8975
8976 /*
8977 * index:
8978 * 0: raid10
8979 * 1: raid1
8980 * 2: dup
8981 * 3: raid0
8982 * 4: single
8983 */
8984 target = get_restripe_target(root->fs_info, block_group->flags);
8985 if (target) {
8986 index = __get_raid_index(extended_to_chunk(target));
8987 } else {
8988 /*
8989 * this is just a balance, so if we were marked as full
8990 * we know there is no space for a new chunk
8991 */
8992 if (full)
8993 goto out;
8994
8995 index = get_block_group_index(block_group);
8996 }
8997
8998 if (index == BTRFS_RAID_RAID10) {
8999 dev_min = 4;
9000 /* Divide by 2 */
9001 min_free >>= 1;
9002 } else if (index == BTRFS_RAID_RAID1) {
9003 dev_min = 2;
9004 } else if (index == BTRFS_RAID_DUP) {
9005 /* Multiply by 2 */
9006 min_free <<= 1;
9007 } else if (index == BTRFS_RAID_RAID0) {
9008 dev_min = fs_devices->rw_devices;
9009 min_free = div64_u64(min_free, dev_min);
9010 }
9011
9012 /* We need to do this so that we can look at pending chunks */
9013 trans = btrfs_join_transaction(root);
9014 if (IS_ERR(trans)) {
9015 ret = PTR_ERR(trans);
9016 goto out;
9017 }
9018
9019 mutex_lock(&root->fs_info->chunk_mutex);
9020 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9021 u64 dev_offset;
9022
9023 /*
9024 * check to make sure we can actually find a chunk with enough
9025 * space to fit our block group in.
9026 */
9027 if (device->total_bytes > device->bytes_used + min_free &&
9028 !device->is_tgtdev_for_dev_replace) {
9029 ret = find_free_dev_extent(trans, device, min_free,
9030 &dev_offset, NULL);
9031 if (!ret)
9032 dev_nr++;
9033
9034 if (dev_nr >= dev_min)
9035 break;
9036
9037 ret = -1;
9038 }
9039 }
9040 mutex_unlock(&root->fs_info->chunk_mutex);
9041 btrfs_end_transaction(trans, root);
9042 out:
9043 btrfs_put_block_group(block_group);
9044 return ret;
9045 }
9046
9047 static int find_first_block_group(struct btrfs_root *root,
9048 struct btrfs_path *path, struct btrfs_key *key)
9049 {
9050 int ret = 0;
9051 struct btrfs_key found_key;
9052 struct extent_buffer *leaf;
9053 int slot;
9054
9055 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9056 if (ret < 0)
9057 goto out;
9058
9059 while (1) {
9060 slot = path->slots[0];
9061 leaf = path->nodes[0];
9062 if (slot >= btrfs_header_nritems(leaf)) {
9063 ret = btrfs_next_leaf(root, path);
9064 if (ret == 0)
9065 continue;
9066 if (ret < 0)
9067 goto out;
9068 break;
9069 }
9070 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9071
9072 if (found_key.objectid >= key->objectid &&
9073 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9074 ret = 0;
9075 goto out;
9076 }
9077 path->slots[0]++;
9078 }
9079 out:
9080 return ret;
9081 }
9082
9083 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9084 {
9085 struct btrfs_block_group_cache *block_group;
9086 u64 last = 0;
9087
9088 while (1) {
9089 struct inode *inode;
9090
9091 block_group = btrfs_lookup_first_block_group(info, last);
9092 while (block_group) {
9093 spin_lock(&block_group->lock);
9094 if (block_group->iref)
9095 break;
9096 spin_unlock(&block_group->lock);
9097 block_group = next_block_group(info->tree_root,
9098 block_group);
9099 }
9100 if (!block_group) {
9101 if (last == 0)
9102 break;
9103 last = 0;
9104 continue;
9105 }
9106
9107 inode = block_group->inode;
9108 block_group->iref = 0;
9109 block_group->inode = NULL;
9110 spin_unlock(&block_group->lock);
9111 iput(inode);
9112 last = block_group->key.objectid + block_group->key.offset;
9113 btrfs_put_block_group(block_group);
9114 }
9115 }
9116
9117 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9118 {
9119 struct btrfs_block_group_cache *block_group;
9120 struct btrfs_space_info *space_info;
9121 struct btrfs_caching_control *caching_ctl;
9122 struct rb_node *n;
9123
9124 down_write(&info->commit_root_sem);
9125 while (!list_empty(&info->caching_block_groups)) {
9126 caching_ctl = list_entry(info->caching_block_groups.next,
9127 struct btrfs_caching_control, list);
9128 list_del(&caching_ctl->list);
9129 put_caching_control(caching_ctl);
9130 }
9131 up_write(&info->commit_root_sem);
9132
9133 spin_lock(&info->unused_bgs_lock);
9134 while (!list_empty(&info->unused_bgs)) {
9135 block_group = list_first_entry(&info->unused_bgs,
9136 struct btrfs_block_group_cache,
9137 bg_list);
9138 list_del_init(&block_group->bg_list);
9139 btrfs_put_block_group(block_group);
9140 }
9141 spin_unlock(&info->unused_bgs_lock);
9142
9143 spin_lock(&info->block_group_cache_lock);
9144 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9145 block_group = rb_entry(n, struct btrfs_block_group_cache,
9146 cache_node);
9147 rb_erase(&block_group->cache_node,
9148 &info->block_group_cache_tree);
9149 RB_CLEAR_NODE(&block_group->cache_node);
9150 spin_unlock(&info->block_group_cache_lock);
9151
9152 down_write(&block_group->space_info->groups_sem);
9153 list_del(&block_group->list);
9154 up_write(&block_group->space_info->groups_sem);
9155
9156 if (block_group->cached == BTRFS_CACHE_STARTED)
9157 wait_block_group_cache_done(block_group);
9158
9159 /*
9160 * We haven't cached this block group, which means we could
9161 * possibly have excluded extents on this block group.
9162 */
9163 if (block_group->cached == BTRFS_CACHE_NO ||
9164 block_group->cached == BTRFS_CACHE_ERROR)
9165 free_excluded_extents(info->extent_root, block_group);
9166
9167 btrfs_remove_free_space_cache(block_group);
9168 btrfs_put_block_group(block_group);
9169
9170 spin_lock(&info->block_group_cache_lock);
9171 }
9172 spin_unlock(&info->block_group_cache_lock);
9173
9174 /* now that all the block groups are freed, go through and
9175 * free all the space_info structs. This is only called during
9176 * the final stages of unmount, and so we know nobody is
9177 * using them. We call synchronize_rcu() once before we start,
9178 * just to be on the safe side.
9179 */
9180 synchronize_rcu();
9181
9182 release_global_block_rsv(info);
9183
9184 while (!list_empty(&info->space_info)) {
9185 int i;
9186
9187 space_info = list_entry(info->space_info.next,
9188 struct btrfs_space_info,
9189 list);
9190 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
9191 if (WARN_ON(space_info->bytes_pinned > 0 ||
9192 space_info->bytes_reserved > 0 ||
9193 space_info->bytes_may_use > 0)) {
9194 dump_space_info(space_info, 0, 0);
9195 }
9196 }
9197 list_del(&space_info->list);
9198 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9199 struct kobject *kobj;
9200 kobj = space_info->block_group_kobjs[i];
9201 space_info->block_group_kobjs[i] = NULL;
9202 if (kobj) {
9203 kobject_del(kobj);
9204 kobject_put(kobj);
9205 }
9206 }
9207 kobject_del(&space_info->kobj);
9208 kobject_put(&space_info->kobj);
9209 }
9210 return 0;
9211 }
9212
9213 static void __link_block_group(struct btrfs_space_info *space_info,
9214 struct btrfs_block_group_cache *cache)
9215 {
9216 int index = get_block_group_index(cache);
9217 bool first = false;
9218
9219 down_write(&space_info->groups_sem);
9220 if (list_empty(&space_info->block_groups[index]))
9221 first = true;
9222 list_add_tail(&cache->list, &space_info->block_groups[index]);
9223 up_write(&space_info->groups_sem);
9224
9225 if (first) {
9226 struct raid_kobject *rkobj;
9227 int ret;
9228
9229 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9230 if (!rkobj)
9231 goto out_err;
9232 rkobj->raid_type = index;
9233 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9234 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9235 "%s", get_raid_name(index));
9236 if (ret) {
9237 kobject_put(&rkobj->kobj);
9238 goto out_err;
9239 }
9240 space_info->block_group_kobjs[index] = &rkobj->kobj;
9241 }
9242
9243 return;
9244 out_err:
9245 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
9246 }
9247
9248 static struct btrfs_block_group_cache *
9249 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
9250 {
9251 struct btrfs_block_group_cache *cache;
9252
9253 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9254 if (!cache)
9255 return NULL;
9256
9257 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9258 GFP_NOFS);
9259 if (!cache->free_space_ctl) {
9260 kfree(cache);
9261 return NULL;
9262 }
9263
9264 cache->key.objectid = start;
9265 cache->key.offset = size;
9266 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9267
9268 cache->sectorsize = root->sectorsize;
9269 cache->fs_info = root->fs_info;
9270 cache->full_stripe_len = btrfs_full_stripe_len(root,
9271 &root->fs_info->mapping_tree,
9272 start);
9273 atomic_set(&cache->count, 1);
9274 spin_lock_init(&cache->lock);
9275 init_rwsem(&cache->data_rwsem);
9276 INIT_LIST_HEAD(&cache->list);
9277 INIT_LIST_HEAD(&cache->cluster_list);
9278 INIT_LIST_HEAD(&cache->bg_list);
9279 INIT_LIST_HEAD(&cache->ro_list);
9280 INIT_LIST_HEAD(&cache->dirty_list);
9281 INIT_LIST_HEAD(&cache->io_list);
9282 btrfs_init_free_space_ctl(cache);
9283 atomic_set(&cache->trimming, 0);
9284
9285 return cache;
9286 }
9287
9288 int btrfs_read_block_groups(struct btrfs_root *root)
9289 {
9290 struct btrfs_path *path;
9291 int ret;
9292 struct btrfs_block_group_cache *cache;
9293 struct btrfs_fs_info *info = root->fs_info;
9294 struct btrfs_space_info *space_info;
9295 struct btrfs_key key;
9296 struct btrfs_key found_key;
9297 struct extent_buffer *leaf;
9298 int need_clear = 0;
9299 u64 cache_gen;
9300
9301 root = info->extent_root;
9302 key.objectid = 0;
9303 key.offset = 0;
9304 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9305 path = btrfs_alloc_path();
9306 if (!path)
9307 return -ENOMEM;
9308 path->reada = 1;
9309
9310 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
9311 if (btrfs_test_opt(root, SPACE_CACHE) &&
9312 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
9313 need_clear = 1;
9314 if (btrfs_test_opt(root, CLEAR_CACHE))
9315 need_clear = 1;
9316
9317 while (1) {
9318 ret = find_first_block_group(root, path, &key);
9319 if (ret > 0)
9320 break;
9321 if (ret != 0)
9322 goto error;
9323
9324 leaf = path->nodes[0];
9325 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9326
9327 cache = btrfs_create_block_group_cache(root, found_key.objectid,
9328 found_key.offset);
9329 if (!cache) {
9330 ret = -ENOMEM;
9331 goto error;
9332 }
9333
9334 if (need_clear) {
9335 /*
9336 * When we mount with old space cache, we need to
9337 * set BTRFS_DC_CLEAR and set dirty flag.
9338 *
9339 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9340 * truncate the old free space cache inode and
9341 * setup a new one.
9342 * b) Setting 'dirty flag' makes sure that we flush
9343 * the new space cache info onto disk.
9344 */
9345 if (btrfs_test_opt(root, SPACE_CACHE))
9346 cache->disk_cache_state = BTRFS_DC_CLEAR;
9347 }
9348
9349 read_extent_buffer(leaf, &cache->item,
9350 btrfs_item_ptr_offset(leaf, path->slots[0]),
9351 sizeof(cache->item));
9352 cache->flags = btrfs_block_group_flags(&cache->item);
9353
9354 key.objectid = found_key.objectid + found_key.offset;
9355 btrfs_release_path(path);
9356
9357 /*
9358 * We need to exclude the super stripes now so that the space
9359 * info has super bytes accounted for, otherwise we'll think
9360 * we have more space than we actually do.
9361 */
9362 ret = exclude_super_stripes(root, cache);
9363 if (ret) {
9364 /*
9365 * We may have excluded something, so call this just in
9366 * case.
9367 */
9368 free_excluded_extents(root, cache);
9369 btrfs_put_block_group(cache);
9370 goto error;
9371 }
9372
9373 /*
9374 * check for two cases, either we are full, and therefore
9375 * don't need to bother with the caching work since we won't
9376 * find any space, or we are empty, and we can just add all
9377 * the space in and be done with it. This saves us _alot_ of
9378 * time, particularly in the full case.
9379 */
9380 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
9381 cache->last_byte_to_unpin = (u64)-1;
9382 cache->cached = BTRFS_CACHE_FINISHED;
9383 free_excluded_extents(root, cache);
9384 } else if (btrfs_block_group_used(&cache->item) == 0) {
9385 cache->last_byte_to_unpin = (u64)-1;
9386 cache->cached = BTRFS_CACHE_FINISHED;
9387 add_new_free_space(cache, root->fs_info,
9388 found_key.objectid,
9389 found_key.objectid +
9390 found_key.offset);
9391 free_excluded_extents(root, cache);
9392 }
9393
9394 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9395 if (ret) {
9396 btrfs_remove_free_space_cache(cache);
9397 btrfs_put_block_group(cache);
9398 goto error;
9399 }
9400
9401 ret = update_space_info(info, cache->flags, found_key.offset,
9402 btrfs_block_group_used(&cache->item),
9403 &space_info);
9404 if (ret) {
9405 btrfs_remove_free_space_cache(cache);
9406 spin_lock(&info->block_group_cache_lock);
9407 rb_erase(&cache->cache_node,
9408 &info->block_group_cache_tree);
9409 RB_CLEAR_NODE(&cache->cache_node);
9410 spin_unlock(&info->block_group_cache_lock);
9411 btrfs_put_block_group(cache);
9412 goto error;
9413 }
9414
9415 cache->space_info = space_info;
9416 spin_lock(&cache->space_info->lock);
9417 cache->space_info->bytes_readonly += cache->bytes_super;
9418 spin_unlock(&cache->space_info->lock);
9419
9420 __link_block_group(space_info, cache);
9421
9422 set_avail_alloc_bits(root->fs_info, cache->flags);
9423 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
9424 set_block_group_ro(cache, 1);
9425 } else if (btrfs_block_group_used(&cache->item) == 0) {
9426 spin_lock(&info->unused_bgs_lock);
9427 /* Should always be true but just in case. */
9428 if (list_empty(&cache->bg_list)) {
9429 btrfs_get_block_group(cache);
9430 list_add_tail(&cache->bg_list,
9431 &info->unused_bgs);
9432 }
9433 spin_unlock(&info->unused_bgs_lock);
9434 }
9435 }
9436
9437 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
9438 if (!(get_alloc_profile(root, space_info->flags) &
9439 (BTRFS_BLOCK_GROUP_RAID10 |
9440 BTRFS_BLOCK_GROUP_RAID1 |
9441 BTRFS_BLOCK_GROUP_RAID5 |
9442 BTRFS_BLOCK_GROUP_RAID6 |
9443 BTRFS_BLOCK_GROUP_DUP)))
9444 continue;
9445 /*
9446 * avoid allocating from un-mirrored block group if there are
9447 * mirrored block groups.
9448 */
9449 list_for_each_entry(cache,
9450 &space_info->block_groups[BTRFS_RAID_RAID0],
9451 list)
9452 set_block_group_ro(cache, 1);
9453 list_for_each_entry(cache,
9454 &space_info->block_groups[BTRFS_RAID_SINGLE],
9455 list)
9456 set_block_group_ro(cache, 1);
9457 }
9458
9459 init_global_block_rsv(info);
9460 ret = 0;
9461 error:
9462 btrfs_free_path(path);
9463 return ret;
9464 }
9465
9466 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
9467 struct btrfs_root *root)
9468 {
9469 struct btrfs_block_group_cache *block_group, *tmp;
9470 struct btrfs_root *extent_root = root->fs_info->extent_root;
9471 struct btrfs_block_group_item item;
9472 struct btrfs_key key;
9473 int ret = 0;
9474
9475 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
9476 if (ret)
9477 goto next;
9478
9479 spin_lock(&block_group->lock);
9480 memcpy(&item, &block_group->item, sizeof(item));
9481 memcpy(&key, &block_group->key, sizeof(key));
9482 spin_unlock(&block_group->lock);
9483
9484 ret = btrfs_insert_item(trans, extent_root, &key, &item,
9485 sizeof(item));
9486 if (ret)
9487 btrfs_abort_transaction(trans, extent_root, ret);
9488 ret = btrfs_finish_chunk_alloc(trans, extent_root,
9489 key.objectid, key.offset);
9490 if (ret)
9491 btrfs_abort_transaction(trans, extent_root, ret);
9492 next:
9493 list_del_init(&block_group->bg_list);
9494 }
9495 }
9496
9497 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
9498 struct btrfs_root *root, u64 bytes_used,
9499 u64 type, u64 chunk_objectid, u64 chunk_offset,
9500 u64 size)
9501 {
9502 int ret;
9503 struct btrfs_root *extent_root;
9504 struct btrfs_block_group_cache *cache;
9505
9506 extent_root = root->fs_info->extent_root;
9507
9508 btrfs_set_log_full_commit(root->fs_info, trans);
9509
9510 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
9511 if (!cache)
9512 return -ENOMEM;
9513
9514 btrfs_set_block_group_used(&cache->item, bytes_used);
9515 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
9516 btrfs_set_block_group_flags(&cache->item, type);
9517
9518 cache->flags = type;
9519 cache->last_byte_to_unpin = (u64)-1;
9520 cache->cached = BTRFS_CACHE_FINISHED;
9521 ret = exclude_super_stripes(root, cache);
9522 if (ret) {
9523 /*
9524 * We may have excluded something, so call this just in
9525 * case.
9526 */
9527 free_excluded_extents(root, cache);
9528 btrfs_put_block_group(cache);
9529 return ret;
9530 }
9531
9532 add_new_free_space(cache, root->fs_info, chunk_offset,
9533 chunk_offset + size);
9534
9535 free_excluded_extents(root, cache);
9536
9537 /*
9538 * Call to ensure the corresponding space_info object is created and
9539 * assigned to our block group, but don't update its counters just yet.
9540 * We want our bg to be added to the rbtree with its ->space_info set.
9541 */
9542 ret = update_space_info(root->fs_info, cache->flags, 0, 0,
9543 &cache->space_info);
9544 if (ret) {
9545 btrfs_remove_free_space_cache(cache);
9546 btrfs_put_block_group(cache);
9547 return ret;
9548 }
9549
9550 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9551 if (ret) {
9552 btrfs_remove_free_space_cache(cache);
9553 btrfs_put_block_group(cache);
9554 return ret;
9555 }
9556
9557 /*
9558 * Now that our block group has its ->space_info set and is inserted in
9559 * the rbtree, update the space info's counters.
9560 */
9561 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
9562 &cache->space_info);
9563 if (ret) {
9564 btrfs_remove_free_space_cache(cache);
9565 spin_lock(&root->fs_info->block_group_cache_lock);
9566 rb_erase(&cache->cache_node,
9567 &root->fs_info->block_group_cache_tree);
9568 RB_CLEAR_NODE(&cache->cache_node);
9569 spin_unlock(&root->fs_info->block_group_cache_lock);
9570 btrfs_put_block_group(cache);
9571 return ret;
9572 }
9573 update_global_block_rsv(root->fs_info);
9574
9575 spin_lock(&cache->space_info->lock);
9576 cache->space_info->bytes_readonly += cache->bytes_super;
9577 spin_unlock(&cache->space_info->lock);
9578
9579 __link_block_group(cache->space_info, cache);
9580
9581 list_add_tail(&cache->bg_list, &trans->new_bgs);
9582
9583 set_avail_alloc_bits(extent_root->fs_info, type);
9584
9585 return 0;
9586 }
9587
9588 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
9589 {
9590 u64 extra_flags = chunk_to_extended(flags) &
9591 BTRFS_EXTENDED_PROFILE_MASK;
9592
9593 write_seqlock(&fs_info->profiles_lock);
9594 if (flags & BTRFS_BLOCK_GROUP_DATA)
9595 fs_info->avail_data_alloc_bits &= ~extra_flags;
9596 if (flags & BTRFS_BLOCK_GROUP_METADATA)
9597 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
9598 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
9599 fs_info->avail_system_alloc_bits &= ~extra_flags;
9600 write_sequnlock(&fs_info->profiles_lock);
9601 }
9602
9603 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
9604 struct btrfs_root *root, u64 group_start,
9605 struct extent_map *em)
9606 {
9607 struct btrfs_path *path;
9608 struct btrfs_block_group_cache *block_group;
9609 struct btrfs_free_cluster *cluster;
9610 struct btrfs_root *tree_root = root->fs_info->tree_root;
9611 struct btrfs_key key;
9612 struct inode *inode;
9613 struct kobject *kobj = NULL;
9614 int ret;
9615 int index;
9616 int factor;
9617 struct btrfs_caching_control *caching_ctl = NULL;
9618 bool remove_em;
9619
9620 root = root->fs_info->extent_root;
9621
9622 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
9623 BUG_ON(!block_group);
9624 BUG_ON(!block_group->ro);
9625
9626 /*
9627 * Free the reserved super bytes from this block group before
9628 * remove it.
9629 */
9630 free_excluded_extents(root, block_group);
9631
9632 memcpy(&key, &block_group->key, sizeof(key));
9633 index = get_block_group_index(block_group);
9634 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
9635 BTRFS_BLOCK_GROUP_RAID1 |
9636 BTRFS_BLOCK_GROUP_RAID10))
9637 factor = 2;
9638 else
9639 factor = 1;
9640
9641 /* make sure this block group isn't part of an allocation cluster */
9642 cluster = &root->fs_info->data_alloc_cluster;
9643 spin_lock(&cluster->refill_lock);
9644 btrfs_return_cluster_to_free_space(block_group, cluster);
9645 spin_unlock(&cluster->refill_lock);
9646
9647 /*
9648 * make sure this block group isn't part of a metadata
9649 * allocation cluster
9650 */
9651 cluster = &root->fs_info->meta_alloc_cluster;
9652 spin_lock(&cluster->refill_lock);
9653 btrfs_return_cluster_to_free_space(block_group, cluster);
9654 spin_unlock(&cluster->refill_lock);
9655
9656 path = btrfs_alloc_path();
9657 if (!path) {
9658 ret = -ENOMEM;
9659 goto out;
9660 }
9661
9662 /*
9663 * get the inode first so any iput calls done for the io_list
9664 * aren't the final iput (no unlinks allowed now)
9665 */
9666 inode = lookup_free_space_inode(tree_root, block_group, path);
9667
9668 mutex_lock(&trans->transaction->cache_write_mutex);
9669 /*
9670 * make sure our free spache cache IO is done before remove the
9671 * free space inode
9672 */
9673 spin_lock(&trans->transaction->dirty_bgs_lock);
9674 if (!list_empty(&block_group->io_list)) {
9675 list_del_init(&block_group->io_list);
9676
9677 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
9678
9679 spin_unlock(&trans->transaction->dirty_bgs_lock);
9680 btrfs_wait_cache_io(root, trans, block_group,
9681 &block_group->io_ctl, path,
9682 block_group->key.objectid);
9683 btrfs_put_block_group(block_group);
9684 spin_lock(&trans->transaction->dirty_bgs_lock);
9685 }
9686
9687 if (!list_empty(&block_group->dirty_list)) {
9688 list_del_init(&block_group->dirty_list);
9689 btrfs_put_block_group(block_group);
9690 }
9691 spin_unlock(&trans->transaction->dirty_bgs_lock);
9692 mutex_unlock(&trans->transaction->cache_write_mutex);
9693
9694 if (!IS_ERR(inode)) {
9695 ret = btrfs_orphan_add(trans, inode);
9696 if (ret) {
9697 btrfs_add_delayed_iput(inode);
9698 goto out;
9699 }
9700 clear_nlink(inode);
9701 /* One for the block groups ref */
9702 spin_lock(&block_group->lock);
9703 if (block_group->iref) {
9704 block_group->iref = 0;
9705 block_group->inode = NULL;
9706 spin_unlock(&block_group->lock);
9707 iput(inode);
9708 } else {
9709 spin_unlock(&block_group->lock);
9710 }
9711 /* One for our lookup ref */
9712 btrfs_add_delayed_iput(inode);
9713 }
9714
9715 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
9716 key.offset = block_group->key.objectid;
9717 key.type = 0;
9718
9719 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
9720 if (ret < 0)
9721 goto out;
9722 if (ret > 0)
9723 btrfs_release_path(path);
9724 if (ret == 0) {
9725 ret = btrfs_del_item(trans, tree_root, path);
9726 if (ret)
9727 goto out;
9728 btrfs_release_path(path);
9729 }
9730
9731 spin_lock(&root->fs_info->block_group_cache_lock);
9732 rb_erase(&block_group->cache_node,
9733 &root->fs_info->block_group_cache_tree);
9734 RB_CLEAR_NODE(&block_group->cache_node);
9735
9736 if (root->fs_info->first_logical_byte == block_group->key.objectid)
9737 root->fs_info->first_logical_byte = (u64)-1;
9738 spin_unlock(&root->fs_info->block_group_cache_lock);
9739
9740 down_write(&block_group->space_info->groups_sem);
9741 /*
9742 * we must use list_del_init so people can check to see if they
9743 * are still on the list after taking the semaphore
9744 */
9745 list_del_init(&block_group->list);
9746 if (list_empty(&block_group->space_info->block_groups[index])) {
9747 kobj = block_group->space_info->block_group_kobjs[index];
9748 block_group->space_info->block_group_kobjs[index] = NULL;
9749 clear_avail_alloc_bits(root->fs_info, block_group->flags);
9750 }
9751 up_write(&block_group->space_info->groups_sem);
9752 if (kobj) {
9753 kobject_del(kobj);
9754 kobject_put(kobj);
9755 }
9756
9757 if (block_group->has_caching_ctl)
9758 caching_ctl = get_caching_control(block_group);
9759 if (block_group->cached == BTRFS_CACHE_STARTED)
9760 wait_block_group_cache_done(block_group);
9761 if (block_group->has_caching_ctl) {
9762 down_write(&root->fs_info->commit_root_sem);
9763 if (!caching_ctl) {
9764 struct btrfs_caching_control *ctl;
9765
9766 list_for_each_entry(ctl,
9767 &root->fs_info->caching_block_groups, list)
9768 if (ctl->block_group == block_group) {
9769 caching_ctl = ctl;
9770 atomic_inc(&caching_ctl->count);
9771 break;
9772 }
9773 }
9774 if (caching_ctl)
9775 list_del_init(&caching_ctl->list);
9776 up_write(&root->fs_info->commit_root_sem);
9777 if (caching_ctl) {
9778 /* Once for the caching bgs list and once for us. */
9779 put_caching_control(caching_ctl);
9780 put_caching_control(caching_ctl);
9781 }
9782 }
9783
9784 spin_lock(&trans->transaction->dirty_bgs_lock);
9785 if (!list_empty(&block_group->dirty_list)) {
9786 WARN_ON(1);
9787 }
9788 if (!list_empty(&block_group->io_list)) {
9789 WARN_ON(1);
9790 }
9791 spin_unlock(&trans->transaction->dirty_bgs_lock);
9792 btrfs_remove_free_space_cache(block_group);
9793
9794 spin_lock(&block_group->space_info->lock);
9795 list_del_init(&block_group->ro_list);
9796
9797 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
9798 WARN_ON(block_group->space_info->total_bytes
9799 < block_group->key.offset);
9800 WARN_ON(block_group->space_info->bytes_readonly
9801 < block_group->key.offset);
9802 WARN_ON(block_group->space_info->disk_total
9803 < block_group->key.offset * factor);
9804 }
9805 block_group->space_info->total_bytes -= block_group->key.offset;
9806 block_group->space_info->bytes_readonly -= block_group->key.offset;
9807 block_group->space_info->disk_total -= block_group->key.offset * factor;
9808
9809 spin_unlock(&block_group->space_info->lock);
9810
9811 memcpy(&key, &block_group->key, sizeof(key));
9812
9813 lock_chunks(root);
9814 if (!list_empty(&em->list)) {
9815 /* We're in the transaction->pending_chunks list. */
9816 free_extent_map(em);
9817 }
9818 spin_lock(&block_group->lock);
9819 block_group->removed = 1;
9820 /*
9821 * At this point trimming can't start on this block group, because we
9822 * removed the block group from the tree fs_info->block_group_cache_tree
9823 * so no one can't find it anymore and even if someone already got this
9824 * block group before we removed it from the rbtree, they have already
9825 * incremented block_group->trimming - if they didn't, they won't find
9826 * any free space entries because we already removed them all when we
9827 * called btrfs_remove_free_space_cache().
9828 *
9829 * And we must not remove the extent map from the fs_info->mapping_tree
9830 * to prevent the same logical address range and physical device space
9831 * ranges from being reused for a new block group. This is because our
9832 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
9833 * completely transactionless, so while it is trimming a range the
9834 * currently running transaction might finish and a new one start,
9835 * allowing for new block groups to be created that can reuse the same
9836 * physical device locations unless we take this special care.
9837 */
9838 remove_em = (atomic_read(&block_group->trimming) == 0);
9839 /*
9840 * Make sure a trimmer task always sees the em in the pinned_chunks list
9841 * if it sees block_group->removed == 1 (needs to lock block_group->lock
9842 * before checking block_group->removed).
9843 */
9844 if (!remove_em) {
9845 /*
9846 * Our em might be in trans->transaction->pending_chunks which
9847 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
9848 * and so is the fs_info->pinned_chunks list.
9849 *
9850 * So at this point we must be holding the chunk_mutex to avoid
9851 * any races with chunk allocation (more specifically at
9852 * volumes.c:contains_pending_extent()), to ensure it always
9853 * sees the em, either in the pending_chunks list or in the
9854 * pinned_chunks list.
9855 */
9856 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
9857 }
9858 spin_unlock(&block_group->lock);
9859
9860 if (remove_em) {
9861 struct extent_map_tree *em_tree;
9862
9863 em_tree = &root->fs_info->mapping_tree.map_tree;
9864 write_lock(&em_tree->lock);
9865 /*
9866 * The em might be in the pending_chunks list, so make sure the
9867 * chunk mutex is locked, since remove_extent_mapping() will
9868 * delete us from that list.
9869 */
9870 remove_extent_mapping(em_tree, em);
9871 write_unlock(&em_tree->lock);
9872 /* once for the tree */
9873 free_extent_map(em);
9874 }
9875
9876 unlock_chunks(root);
9877
9878 btrfs_put_block_group(block_group);
9879 btrfs_put_block_group(block_group);
9880
9881 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
9882 if (ret > 0)
9883 ret = -EIO;
9884 if (ret < 0)
9885 goto out;
9886
9887 ret = btrfs_del_item(trans, root, path);
9888 out:
9889 btrfs_free_path(path);
9890 return ret;
9891 }
9892
9893 /*
9894 * Process the unused_bgs list and remove any that don't have any allocated
9895 * space inside of them.
9896 */
9897 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
9898 {
9899 struct btrfs_block_group_cache *block_group;
9900 struct btrfs_space_info *space_info;
9901 struct btrfs_root *root = fs_info->extent_root;
9902 struct btrfs_trans_handle *trans;
9903 int ret = 0;
9904
9905 if (!fs_info->open)
9906 return;
9907
9908 spin_lock(&fs_info->unused_bgs_lock);
9909 while (!list_empty(&fs_info->unused_bgs)) {
9910 u64 start, end;
9911
9912 block_group = list_first_entry(&fs_info->unused_bgs,
9913 struct btrfs_block_group_cache,
9914 bg_list);
9915 space_info = block_group->space_info;
9916 list_del_init(&block_group->bg_list);
9917 if (ret || btrfs_mixed_space_info(space_info)) {
9918 btrfs_put_block_group(block_group);
9919 continue;
9920 }
9921 spin_unlock(&fs_info->unused_bgs_lock);
9922
9923 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
9924
9925 /* Don't want to race with allocators so take the groups_sem */
9926 down_write(&space_info->groups_sem);
9927 spin_lock(&block_group->lock);
9928 if (block_group->reserved ||
9929 btrfs_block_group_used(&block_group->item) ||
9930 block_group->ro) {
9931 /*
9932 * We want to bail if we made new allocations or have
9933 * outstanding allocations in this block group. We do
9934 * the ro check in case balance is currently acting on
9935 * this block group.
9936 */
9937 spin_unlock(&block_group->lock);
9938 up_write(&space_info->groups_sem);
9939 goto next;
9940 }
9941 spin_unlock(&block_group->lock);
9942
9943 /* We don't want to force the issue, only flip if it's ok. */
9944 ret = set_block_group_ro(block_group, 0);
9945 up_write(&space_info->groups_sem);
9946 if (ret < 0) {
9947 ret = 0;
9948 goto next;
9949 }
9950
9951 /*
9952 * Want to do this before we do anything else so we can recover
9953 * properly if we fail to join the transaction.
9954 */
9955 /* 1 for btrfs_orphan_reserve_metadata() */
9956 trans = btrfs_start_transaction(root, 1);
9957 if (IS_ERR(trans)) {
9958 btrfs_set_block_group_rw(root, block_group);
9959 ret = PTR_ERR(trans);
9960 goto next;
9961 }
9962
9963 /*
9964 * We could have pending pinned extents for this block group,
9965 * just delete them, we don't care about them anymore.
9966 */
9967 start = block_group->key.objectid;
9968 end = start + block_group->key.offset - 1;
9969 /*
9970 * Hold the unused_bg_unpin_mutex lock to avoid racing with
9971 * btrfs_finish_extent_commit(). If we are at transaction N,
9972 * another task might be running finish_extent_commit() for the
9973 * previous transaction N - 1, and have seen a range belonging
9974 * to the block group in freed_extents[] before we were able to
9975 * clear the whole block group range from freed_extents[]. This
9976 * means that task can lookup for the block group after we
9977 * unpinned it from freed_extents[] and removed it, leading to
9978 * a BUG_ON() at btrfs_unpin_extent_range().
9979 */
9980 mutex_lock(&fs_info->unused_bg_unpin_mutex);
9981 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
9982 EXTENT_DIRTY, GFP_NOFS);
9983 if (ret) {
9984 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
9985 btrfs_set_block_group_rw(root, block_group);
9986 goto end_trans;
9987 }
9988 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
9989 EXTENT_DIRTY, GFP_NOFS);
9990 if (ret) {
9991 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
9992 btrfs_set_block_group_rw(root, block_group);
9993 goto end_trans;
9994 }
9995 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
9996
9997 /* Reset pinned so btrfs_put_block_group doesn't complain */
9998 spin_lock(&space_info->lock);
9999 spin_lock(&block_group->lock);
10000
10001 space_info->bytes_pinned -= block_group->pinned;
10002 space_info->bytes_readonly += block_group->pinned;
10003 percpu_counter_add(&space_info->total_bytes_pinned,
10004 -block_group->pinned);
10005 block_group->pinned = 0;
10006
10007 spin_unlock(&block_group->lock);
10008 spin_unlock(&space_info->lock);
10009
10010 /*
10011 * Btrfs_remove_chunk will abort the transaction if things go
10012 * horribly wrong.
10013 */
10014 ret = btrfs_remove_chunk(trans, root,
10015 block_group->key.objectid);
10016 end_trans:
10017 btrfs_end_transaction(trans, root);
10018 next:
10019 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
10020 btrfs_put_block_group(block_group);
10021 spin_lock(&fs_info->unused_bgs_lock);
10022 }
10023 spin_unlock(&fs_info->unused_bgs_lock);
10024 }
10025
10026 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10027 {
10028 struct btrfs_space_info *space_info;
10029 struct btrfs_super_block *disk_super;
10030 u64 features;
10031 u64 flags;
10032 int mixed = 0;
10033 int ret;
10034
10035 disk_super = fs_info->super_copy;
10036 if (!btrfs_super_root(disk_super))
10037 return 1;
10038
10039 features = btrfs_super_incompat_flags(disk_super);
10040 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10041 mixed = 1;
10042
10043 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10044 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10045 if (ret)
10046 goto out;
10047
10048 if (mixed) {
10049 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10050 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10051 } else {
10052 flags = BTRFS_BLOCK_GROUP_METADATA;
10053 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10054 if (ret)
10055 goto out;
10056
10057 flags = BTRFS_BLOCK_GROUP_DATA;
10058 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10059 }
10060 out:
10061 return ret;
10062 }
10063
10064 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
10065 {
10066 return unpin_extent_range(root, start, end, false);
10067 }
10068
10069 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
10070 {
10071 struct btrfs_fs_info *fs_info = root->fs_info;
10072 struct btrfs_block_group_cache *cache = NULL;
10073 u64 group_trimmed;
10074 u64 start;
10075 u64 end;
10076 u64 trimmed = 0;
10077 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10078 int ret = 0;
10079
10080 /*
10081 * try to trim all FS space, our block group may start from non-zero.
10082 */
10083 if (range->len == total_bytes)
10084 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10085 else
10086 cache = btrfs_lookup_block_group(fs_info, range->start);
10087
10088 while (cache) {
10089 if (cache->key.objectid >= (range->start + range->len)) {
10090 btrfs_put_block_group(cache);
10091 break;
10092 }
10093
10094 start = max(range->start, cache->key.objectid);
10095 end = min(range->start + range->len,
10096 cache->key.objectid + cache->key.offset);
10097
10098 if (end - start >= range->minlen) {
10099 if (!block_group_cache_done(cache)) {
10100 ret = cache_block_group(cache, 0);
10101 if (ret) {
10102 btrfs_put_block_group(cache);
10103 break;
10104 }
10105 ret = wait_block_group_cache_done(cache);
10106 if (ret) {
10107 btrfs_put_block_group(cache);
10108 break;
10109 }
10110 }
10111 ret = btrfs_trim_block_group(cache,
10112 &group_trimmed,
10113 start,
10114 end,
10115 range->minlen);
10116
10117 trimmed += group_trimmed;
10118 if (ret) {
10119 btrfs_put_block_group(cache);
10120 break;
10121 }
10122 }
10123
10124 cache = next_block_group(fs_info->tree_root, cache);
10125 }
10126
10127 range->len = trimmed;
10128 return ret;
10129 }
10130
10131 /*
10132 * btrfs_{start,end}_write_no_snapshoting() are similar to
10133 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10134 * data into the page cache through nocow before the subvolume is snapshoted,
10135 * but flush the data into disk after the snapshot creation, or to prevent
10136 * operations while snapshoting is ongoing and that cause the snapshot to be
10137 * inconsistent (writes followed by expanding truncates for example).
10138 */
10139 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
10140 {
10141 percpu_counter_dec(&root->subv_writers->counter);
10142 /*
10143 * Make sure counter is updated before we wake up
10144 * waiters.
10145 */
10146 smp_mb();
10147 if (waitqueue_active(&root->subv_writers->wait))
10148 wake_up(&root->subv_writers->wait);
10149 }
10150
10151 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
10152 {
10153 if (atomic_read(&root->will_be_snapshoted))
10154 return 0;
10155
10156 percpu_counter_inc(&root->subv_writers->counter);
10157 /*
10158 * Make sure counter is updated before we check for snapshot creation.
10159 */
10160 smp_mb();
10161 if (atomic_read(&root->will_be_snapshoted)) {
10162 btrfs_end_write_no_snapshoting(root);
10163 return 0;
10164 }
10165 return 1;
10166 }
Linux kernel - Deliverables
This page took 0.868347 seconds and 6 git commands to generate.