projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge remote-tracking branch 'iommu/next'
[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 "free-space-tree.h"
37 #include "math.h"
38 #include "sysfs.h"
39 #include "qgroup.h"
40
41 #undef SCRAMBLE_DELAYED_REFS
42
43 /*
44 * control flags for do_chunk_alloc's force field
45 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
46 * if we really need one.
47 *
48 * CHUNK_ALLOC_LIMITED means to only try and allocate one
49 * if we have very few chunks already allocated. This is
50 * used as part of the clustering code to help make sure
51 * we have a good pool of storage to cluster in, without
52 * filling the FS with empty chunks
53 *
54 * CHUNK_ALLOC_FORCE means it must try to allocate one
55 *
56 */
57 enum {
58 CHUNK_ALLOC_NO_FORCE = 0,
59 CHUNK_ALLOC_LIMITED = 1,
60 CHUNK_ALLOC_FORCE = 2,
61 };
62
63 static int update_block_group(struct btrfs_trans_handle *trans,
64 struct btrfs_root *root, u64 bytenr,
65 u64 num_bytes, int alloc);
66 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
67 struct btrfs_root *root,
68 struct btrfs_delayed_ref_node *node, u64 parent,
69 u64 root_objectid, u64 owner_objectid,
70 u64 owner_offset, int refs_to_drop,
71 struct btrfs_delayed_extent_op *extra_op);
72 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
73 struct extent_buffer *leaf,
74 struct btrfs_extent_item *ei);
75 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
76 struct btrfs_root *root,
77 u64 parent, u64 root_objectid,
78 u64 flags, u64 owner, u64 offset,
79 struct btrfs_key *ins, int ref_mod);
80 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
81 struct btrfs_root *root,
82 u64 parent, u64 root_objectid,
83 u64 flags, struct btrfs_disk_key *key,
84 int level, struct btrfs_key *ins);
85 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
86 struct btrfs_root *extent_root, u64 flags,
87 int force);
88 static int find_next_key(struct btrfs_path *path, int level,
89 struct btrfs_key *key);
90 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
91 int dump_block_groups);
92 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
93 u64 ram_bytes, u64 num_bytes, int delalloc);
94 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
95 u64 num_bytes, int delalloc);
96 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
97 u64 num_bytes);
98 int btrfs_pin_extent(struct btrfs_root *root,
99 u64 bytenr, u64 num_bytes, int reserved);
100 static int __reserve_metadata_bytes(struct btrfs_root *root,
101 struct btrfs_space_info *space_info,
102 u64 orig_bytes,
103 enum btrfs_reserve_flush_enum flush);
104 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
105 struct btrfs_space_info *space_info,
106 u64 num_bytes);
107 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
108 struct btrfs_space_info *space_info,
109 u64 num_bytes);
110
111 static noinline int
112 block_group_cache_done(struct btrfs_block_group_cache *cache)
113 {
114 smp_mb();
115 return cache->cached == BTRFS_CACHE_FINISHED ||
116 cache->cached == BTRFS_CACHE_ERROR;
117 }
118
119 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
120 {
121 return (cache->flags & bits) == bits;
122 }
123
124 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
125 {
126 atomic_inc(&cache->count);
127 }
128
129 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
130 {
131 if (atomic_dec_and_test(&cache->count)) {
132 WARN_ON(cache->pinned > 0);
133 WARN_ON(cache->reserved > 0);
134 kfree(cache->free_space_ctl);
135 kfree(cache);
136 }
137 }
138
139 /*
140 * this adds the block group to the fs_info rb tree for the block group
141 * cache
142 */
143 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
144 struct btrfs_block_group_cache *block_group)
145 {
146 struct rb_node **p;
147 struct rb_node *parent = NULL;
148 struct btrfs_block_group_cache *cache;
149
150 spin_lock(&info->block_group_cache_lock);
151 p = &info->block_group_cache_tree.rb_node;
152
153 while (*p) {
154 parent = *p;
155 cache = rb_entry(parent, struct btrfs_block_group_cache,
156 cache_node);
157 if (block_group->key.objectid < cache->key.objectid) {
158 p = &(*p)->rb_left;
159 } else if (block_group->key.objectid > cache->key.objectid) {
160 p = &(*p)->rb_right;
161 } else {
162 spin_unlock(&info->block_group_cache_lock);
163 return -EEXIST;
164 }
165 }
166
167 rb_link_node(&block_group->cache_node, parent, p);
168 rb_insert_color(&block_group->cache_node,
169 &info->block_group_cache_tree);
170
171 if (info->first_logical_byte > block_group->key.objectid)
172 info->first_logical_byte = block_group->key.objectid;
173
174 spin_unlock(&info->block_group_cache_lock);
175
176 return 0;
177 }
178
179 /*
180 * This will return the block group at or after bytenr if contains is 0, else
181 * it will return the block group that contains the bytenr
182 */
183 static struct btrfs_block_group_cache *
184 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
185 int contains)
186 {
187 struct btrfs_block_group_cache *cache, *ret = NULL;
188 struct rb_node *n;
189 u64 end, start;
190
191 spin_lock(&info->block_group_cache_lock);
192 n = info->block_group_cache_tree.rb_node;
193
194 while (n) {
195 cache = rb_entry(n, struct btrfs_block_group_cache,
196 cache_node);
197 end = cache->key.objectid + cache->key.offset - 1;
198 start = cache->key.objectid;
199
200 if (bytenr < start) {
201 if (!contains && (!ret || start < ret->key.objectid))
202 ret = cache;
203 n = n->rb_left;
204 } else if (bytenr > start) {
205 if (contains && bytenr <= end) {
206 ret = cache;
207 break;
208 }
209 n = n->rb_right;
210 } else {
211 ret = cache;
212 break;
213 }
214 }
215 if (ret) {
216 btrfs_get_block_group(ret);
217 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
218 info->first_logical_byte = ret->key.objectid;
219 }
220 spin_unlock(&info->block_group_cache_lock);
221
222 return ret;
223 }
224
225 static int add_excluded_extent(struct btrfs_root *root,
226 u64 start, u64 num_bytes)
227 {
228 u64 end = start + num_bytes - 1;
229 set_extent_bits(&root->fs_info->freed_extents[0],
230 start, end, EXTENT_UPTODATE);
231 set_extent_bits(&root->fs_info->freed_extents[1],
232 start, end, EXTENT_UPTODATE);
233 return 0;
234 }
235
236 static void free_excluded_extents(struct btrfs_root *root,
237 struct btrfs_block_group_cache *cache)
238 {
239 u64 start, end;
240
241 start = cache->key.objectid;
242 end = start + cache->key.offset - 1;
243
244 clear_extent_bits(&root->fs_info->freed_extents[0],
245 start, end, EXTENT_UPTODATE);
246 clear_extent_bits(&root->fs_info->freed_extents[1],
247 start, end, EXTENT_UPTODATE);
248 }
249
250 static int exclude_super_stripes(struct btrfs_root *root,
251 struct btrfs_block_group_cache *cache)
252 {
253 u64 bytenr;
254 u64 *logical;
255 int stripe_len;
256 int i, nr, ret;
257
258 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
259 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
260 cache->bytes_super += stripe_len;
261 ret = add_excluded_extent(root, cache->key.objectid,
262 stripe_len);
263 if (ret)
264 return ret;
265 }
266
267 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
268 bytenr = btrfs_sb_offset(i);
269 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
270 cache->key.objectid, bytenr,
271 0, &logical, &nr, &stripe_len);
272 if (ret)
273 return ret;
274
275 while (nr--) {
276 u64 start, len;
277
278 if (logical[nr] > cache->key.objectid +
279 cache->key.offset)
280 continue;
281
282 if (logical[nr] + stripe_len <= cache->key.objectid)
283 continue;
284
285 start = logical[nr];
286 if (start < cache->key.objectid) {
287 start = cache->key.objectid;
288 len = (logical[nr] + stripe_len) - start;
289 } else {
290 len = min_t(u64, stripe_len,
291 cache->key.objectid +
292 cache->key.offset - start);
293 }
294
295 cache->bytes_super += len;
296 ret = add_excluded_extent(root, start, len);
297 if (ret) {
298 kfree(logical);
299 return ret;
300 }
301 }
302
303 kfree(logical);
304 }
305 return 0;
306 }
307
308 static struct btrfs_caching_control *
309 get_caching_control(struct btrfs_block_group_cache *cache)
310 {
311 struct btrfs_caching_control *ctl;
312
313 spin_lock(&cache->lock);
314 if (!cache->caching_ctl) {
315 spin_unlock(&cache->lock);
316 return NULL;
317 }
318
319 ctl = cache->caching_ctl;
320 atomic_inc(&ctl->count);
321 spin_unlock(&cache->lock);
322 return ctl;
323 }
324
325 static void put_caching_control(struct btrfs_caching_control *ctl)
326 {
327 if (atomic_dec_and_test(&ctl->count))
328 kfree(ctl);
329 }
330
331 #ifdef CONFIG_BTRFS_DEBUG
332 static void fragment_free_space(struct btrfs_root *root,
333 struct btrfs_block_group_cache *block_group)
334 {
335 u64 start = block_group->key.objectid;
336 u64 len = block_group->key.offset;
337 u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ?
338 root->nodesize : root->sectorsize;
339 u64 step = chunk << 1;
340
341 while (len > chunk) {
342 btrfs_remove_free_space(block_group, start, chunk);
343 start += step;
344 if (len < step)
345 len = 0;
346 else
347 len -= step;
348 }
349 }
350 #endif
351
352 /*
353 * this is only called by cache_block_group, since we could have freed extents
354 * we need to check the pinned_extents for any extents that can't be used yet
355 * since their free space will be released as soon as the transaction commits.
356 */
357 u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
358 struct btrfs_fs_info *info, u64 start, u64 end)
359 {
360 u64 extent_start, extent_end, size, total_added = 0;
361 int ret;
362
363 while (start < end) {
364 ret = find_first_extent_bit(info->pinned_extents, start,
365 &extent_start, &extent_end,
366 EXTENT_DIRTY | EXTENT_UPTODATE,
367 NULL);
368 if (ret)
369 break;
370
371 if (extent_start <= start) {
372 start = extent_end + 1;
373 } else if (extent_start > start && extent_start < end) {
374 size = extent_start - start;
375 total_added += size;
376 ret = btrfs_add_free_space(block_group, start,
377 size);
378 BUG_ON(ret); /* -ENOMEM or logic error */
379 start = extent_end + 1;
380 } else {
381 break;
382 }
383 }
384
385 if (start < end) {
386 size = end - start;
387 total_added += size;
388 ret = btrfs_add_free_space(block_group, start, size);
389 BUG_ON(ret); /* -ENOMEM or logic error */
390 }
391
392 return total_added;
393 }
394
395 static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl)
396 {
397 struct btrfs_block_group_cache *block_group;
398 struct btrfs_fs_info *fs_info;
399 struct btrfs_root *extent_root;
400 struct btrfs_path *path;
401 struct extent_buffer *leaf;
402 struct btrfs_key key;
403 u64 total_found = 0;
404 u64 last = 0;
405 u32 nritems;
406 int ret;
407 bool wakeup = true;
408
409 block_group = caching_ctl->block_group;
410 fs_info = block_group->fs_info;
411 extent_root = fs_info->extent_root;
412
413 path = btrfs_alloc_path();
414 if (!path)
415 return -ENOMEM;
416
417 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
418
419 #ifdef CONFIG_BTRFS_DEBUG
420 /*
421 * If we're fragmenting we don't want to make anybody think we can
422 * allocate from this block group until we've had a chance to fragment
423 * the free space.
424 */
425 if (btrfs_should_fragment_free_space(extent_root, block_group))
426 wakeup = false;
427 #endif
428 /*
429 * We don't want to deadlock with somebody trying to allocate a new
430 * extent for the extent root while also trying to search the extent
431 * root to add free space. So we skip locking and search the commit
432 * root, since its read-only
433 */
434 path->skip_locking = 1;
435 path->search_commit_root = 1;
436 path->reada = READA_FORWARD;
437
438 key.objectid = last;
439 key.offset = 0;
440 key.type = BTRFS_EXTENT_ITEM_KEY;
441
442 next:
443 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
444 if (ret < 0)
445 goto out;
446
447 leaf = path->nodes[0];
448 nritems = btrfs_header_nritems(leaf);
449
450 while (1) {
451 if (btrfs_fs_closing(fs_info) > 1) {
452 last = (u64)-1;
453 break;
454 }
455
456 if (path->slots[0] < nritems) {
457 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
458 } else {
459 ret = find_next_key(path, 0, &key);
460 if (ret)
461 break;
462
463 if (need_resched() ||
464 rwsem_is_contended(&fs_info->commit_root_sem)) {
465 if (wakeup)
466 caching_ctl->progress = last;
467 btrfs_release_path(path);
468 up_read(&fs_info->commit_root_sem);
469 mutex_unlock(&caching_ctl->mutex);
470 cond_resched();
471 mutex_lock(&caching_ctl->mutex);
472 down_read(&fs_info->commit_root_sem);
473 goto next;
474 }
475
476 ret = btrfs_next_leaf(extent_root, path);
477 if (ret < 0)
478 goto out;
479 if (ret)
480 break;
481 leaf = path->nodes[0];
482 nritems = btrfs_header_nritems(leaf);
483 continue;
484 }
485
486 if (key.objectid < last) {
487 key.objectid = last;
488 key.offset = 0;
489 key.type = BTRFS_EXTENT_ITEM_KEY;
490
491 if (wakeup)
492 caching_ctl->progress = last;
493 btrfs_release_path(path);
494 goto next;
495 }
496
497 if (key.objectid < block_group->key.objectid) {
498 path->slots[0]++;
499 continue;
500 }
501
502 if (key.objectid >= block_group->key.objectid +
503 block_group->key.offset)
504 break;
505
506 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
507 key.type == BTRFS_METADATA_ITEM_KEY) {
508 total_found += add_new_free_space(block_group,
509 fs_info, last,
510 key.objectid);
511 if (key.type == BTRFS_METADATA_ITEM_KEY)
512 last = key.objectid +
513 fs_info->tree_root->nodesize;
514 else
515 last = key.objectid + key.offset;
516
517 if (total_found > CACHING_CTL_WAKE_UP) {
518 total_found = 0;
519 if (wakeup)
520 wake_up(&caching_ctl->wait);
521 }
522 }
523 path->slots[0]++;
524 }
525 ret = 0;
526
527 total_found += add_new_free_space(block_group, fs_info, last,
528 block_group->key.objectid +
529 block_group->key.offset);
530 caching_ctl->progress = (u64)-1;
531
532 out:
533 btrfs_free_path(path);
534 return ret;
535 }
536
537 static noinline void caching_thread(struct btrfs_work *work)
538 {
539 struct btrfs_block_group_cache *block_group;
540 struct btrfs_fs_info *fs_info;
541 struct btrfs_caching_control *caching_ctl;
542 struct btrfs_root *extent_root;
543 int ret;
544
545 caching_ctl = container_of(work, struct btrfs_caching_control, work);
546 block_group = caching_ctl->block_group;
547 fs_info = block_group->fs_info;
548 extent_root = fs_info->extent_root;
549
550 mutex_lock(&caching_ctl->mutex);
551 down_read(&fs_info->commit_root_sem);
552
553 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
554 ret = load_free_space_tree(caching_ctl);
555 else
556 ret = load_extent_tree_free(caching_ctl);
557
558 spin_lock(&block_group->lock);
559 block_group->caching_ctl = NULL;
560 block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED;
561 spin_unlock(&block_group->lock);
562
563 #ifdef CONFIG_BTRFS_DEBUG
564 if (btrfs_should_fragment_free_space(extent_root, block_group)) {
565 u64 bytes_used;
566
567 spin_lock(&block_group->space_info->lock);
568 spin_lock(&block_group->lock);
569 bytes_used = block_group->key.offset -
570 btrfs_block_group_used(&block_group->item);
571 block_group->space_info->bytes_used += bytes_used >> 1;
572 spin_unlock(&block_group->lock);
573 spin_unlock(&block_group->space_info->lock);
574 fragment_free_space(extent_root, block_group);
575 }
576 #endif
577
578 caching_ctl->progress = (u64)-1;
579
580 up_read(&fs_info->commit_root_sem);
581 free_excluded_extents(fs_info->extent_root, block_group);
582 mutex_unlock(&caching_ctl->mutex);
583
584 wake_up(&caching_ctl->wait);
585
586 put_caching_control(caching_ctl);
587 btrfs_put_block_group(block_group);
588 }
589
590 static int cache_block_group(struct btrfs_block_group_cache *cache,
591 int load_cache_only)
592 {
593 DEFINE_WAIT(wait);
594 struct btrfs_fs_info *fs_info = cache->fs_info;
595 struct btrfs_caching_control *caching_ctl;
596 int ret = 0;
597
598 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
599 if (!caching_ctl)
600 return -ENOMEM;
601
602 INIT_LIST_HEAD(&caching_ctl->list);
603 mutex_init(&caching_ctl->mutex);
604 init_waitqueue_head(&caching_ctl->wait);
605 caching_ctl->block_group = cache;
606 caching_ctl->progress = cache->key.objectid;
607 atomic_set(&caching_ctl->count, 1);
608 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
609 caching_thread, NULL, NULL);
610
611 spin_lock(&cache->lock);
612 /*
613 * This should be a rare occasion, but this could happen I think in the
614 * case where one thread starts to load the space cache info, and then
615 * some other thread starts a transaction commit which tries to do an
616 * allocation while the other thread is still loading the space cache
617 * info. The previous loop should have kept us from choosing this block
618 * group, but if we've moved to the state where we will wait on caching
619 * block groups we need to first check if we're doing a fast load here,
620 * so we can wait for it to finish, otherwise we could end up allocating
621 * from a block group who's cache gets evicted for one reason or
622 * another.
623 */
624 while (cache->cached == BTRFS_CACHE_FAST) {
625 struct btrfs_caching_control *ctl;
626
627 ctl = cache->caching_ctl;
628 atomic_inc(&ctl->count);
629 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
630 spin_unlock(&cache->lock);
631
632 schedule();
633
634 finish_wait(&ctl->wait, &wait);
635 put_caching_control(ctl);
636 spin_lock(&cache->lock);
637 }
638
639 if (cache->cached != BTRFS_CACHE_NO) {
640 spin_unlock(&cache->lock);
641 kfree(caching_ctl);
642 return 0;
643 }
644 WARN_ON(cache->caching_ctl);
645 cache->caching_ctl = caching_ctl;
646 cache->cached = BTRFS_CACHE_FAST;
647 spin_unlock(&cache->lock);
648
649 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
650 mutex_lock(&caching_ctl->mutex);
651 ret = load_free_space_cache(fs_info, cache);
652
653 spin_lock(&cache->lock);
654 if (ret == 1) {
655 cache->caching_ctl = NULL;
656 cache->cached = BTRFS_CACHE_FINISHED;
657 cache->last_byte_to_unpin = (u64)-1;
658 caching_ctl->progress = (u64)-1;
659 } else {
660 if (load_cache_only) {
661 cache->caching_ctl = NULL;
662 cache->cached = BTRFS_CACHE_NO;
663 } else {
664 cache->cached = BTRFS_CACHE_STARTED;
665 cache->has_caching_ctl = 1;
666 }
667 }
668 spin_unlock(&cache->lock);
669 #ifdef CONFIG_BTRFS_DEBUG
670 if (ret == 1 &&
671 btrfs_should_fragment_free_space(fs_info->extent_root,
672 cache)) {
673 u64 bytes_used;
674
675 spin_lock(&cache->space_info->lock);
676 spin_lock(&cache->lock);
677 bytes_used = cache->key.offset -
678 btrfs_block_group_used(&cache->item);
679 cache->space_info->bytes_used += bytes_used >> 1;
680 spin_unlock(&cache->lock);
681 spin_unlock(&cache->space_info->lock);
682 fragment_free_space(fs_info->extent_root, cache);
683 }
684 #endif
685 mutex_unlock(&caching_ctl->mutex);
686
687 wake_up(&caching_ctl->wait);
688 if (ret == 1) {
689 put_caching_control(caching_ctl);
690 free_excluded_extents(fs_info->extent_root, cache);
691 return 0;
692 }
693 } else {
694 /*
695 * We're either using the free space tree or no caching at all.
696 * Set cached to the appropriate value and wakeup any waiters.
697 */
698 spin_lock(&cache->lock);
699 if (load_cache_only) {
700 cache->caching_ctl = NULL;
701 cache->cached = BTRFS_CACHE_NO;
702 } else {
703 cache->cached = BTRFS_CACHE_STARTED;
704 cache->has_caching_ctl = 1;
705 }
706 spin_unlock(&cache->lock);
707 wake_up(&caching_ctl->wait);
708 }
709
710 if (load_cache_only) {
711 put_caching_control(caching_ctl);
712 return 0;
713 }
714
715 down_write(&fs_info->commit_root_sem);
716 atomic_inc(&caching_ctl->count);
717 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
718 up_write(&fs_info->commit_root_sem);
719
720 btrfs_get_block_group(cache);
721
722 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
723
724 return ret;
725 }
726
727 /*
728 * return the block group that starts at or after bytenr
729 */
730 static struct btrfs_block_group_cache *
731 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
732 {
733 struct btrfs_block_group_cache *cache;
734
735 cache = block_group_cache_tree_search(info, bytenr, 0);
736
737 return cache;
738 }
739
740 /*
741 * return the block group that contains the given bytenr
742 */
743 struct btrfs_block_group_cache *btrfs_lookup_block_group(
744 struct btrfs_fs_info *info,
745 u64 bytenr)
746 {
747 struct btrfs_block_group_cache *cache;
748
749 cache = block_group_cache_tree_search(info, bytenr, 1);
750
751 return cache;
752 }
753
754 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
755 u64 flags)
756 {
757 struct list_head *head = &info->space_info;
758 struct btrfs_space_info *found;
759
760 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
761
762 rcu_read_lock();
763 list_for_each_entry_rcu(found, head, list) {
764 if (found->flags & flags) {
765 rcu_read_unlock();
766 return found;
767 }
768 }
769 rcu_read_unlock();
770 return NULL;
771 }
772
773 /*
774 * after adding space to the filesystem, we need to clear the full flags
775 * on all the space infos.
776 */
777 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
778 {
779 struct list_head *head = &info->space_info;
780 struct btrfs_space_info *found;
781
782 rcu_read_lock();
783 list_for_each_entry_rcu(found, head, list)
784 found->full = 0;
785 rcu_read_unlock();
786 }
787
788 /* simple helper to search for an existing data extent at a given offset */
789 int btrfs_lookup_data_extent(struct btrfs_root *root, u64 start, u64 len)
790 {
791 int ret;
792 struct btrfs_key key;
793 struct btrfs_path *path;
794
795 path = btrfs_alloc_path();
796 if (!path)
797 return -ENOMEM;
798
799 key.objectid = start;
800 key.offset = len;
801 key.type = BTRFS_EXTENT_ITEM_KEY;
802 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
803 0, 0);
804 btrfs_free_path(path);
805 return ret;
806 }
807
808 /*
809 * helper function to lookup reference count and flags of a tree block.
810 *
811 * the head node for delayed ref is used to store the sum of all the
812 * reference count modifications queued up in the rbtree. the head
813 * node may also store the extent flags to set. This way you can check
814 * to see what the reference count and extent flags would be if all of
815 * the delayed refs are not processed.
816 */
817 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
818 struct btrfs_root *root, u64 bytenr,
819 u64 offset, int metadata, u64 *refs, u64 *flags)
820 {
821 struct btrfs_delayed_ref_head *head;
822 struct btrfs_delayed_ref_root *delayed_refs;
823 struct btrfs_path *path;
824 struct btrfs_extent_item *ei;
825 struct extent_buffer *leaf;
826 struct btrfs_key key;
827 u32 item_size;
828 u64 num_refs;
829 u64 extent_flags;
830 int ret;
831
832 /*
833 * If we don't have skinny metadata, don't bother doing anything
834 * different
835 */
836 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
837 offset = root->nodesize;
838 metadata = 0;
839 }
840
841 path = btrfs_alloc_path();
842 if (!path)
843 return -ENOMEM;
844
845 if (!trans) {
846 path->skip_locking = 1;
847 path->search_commit_root = 1;
848 }
849
850 search_again:
851 key.objectid = bytenr;
852 key.offset = offset;
853 if (metadata)
854 key.type = BTRFS_METADATA_ITEM_KEY;
855 else
856 key.type = BTRFS_EXTENT_ITEM_KEY;
857
858 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
859 &key, path, 0, 0);
860 if (ret < 0)
861 goto out_free;
862
863 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
864 if (path->slots[0]) {
865 path->slots[0]--;
866 btrfs_item_key_to_cpu(path->nodes[0], &key,
867 path->slots[0]);
868 if (key.objectid == bytenr &&
869 key.type == BTRFS_EXTENT_ITEM_KEY &&
870 key.offset == root->nodesize)
871 ret = 0;
872 }
873 }
874
875 if (ret == 0) {
876 leaf = path->nodes[0];
877 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
878 if (item_size >= sizeof(*ei)) {
879 ei = btrfs_item_ptr(leaf, path->slots[0],
880 struct btrfs_extent_item);
881 num_refs = btrfs_extent_refs(leaf, ei);
882 extent_flags = btrfs_extent_flags(leaf, ei);
883 } else {
884 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
885 struct btrfs_extent_item_v0 *ei0;
886 BUG_ON(item_size != sizeof(*ei0));
887 ei0 = btrfs_item_ptr(leaf, path->slots[0],
888 struct btrfs_extent_item_v0);
889 num_refs = btrfs_extent_refs_v0(leaf, ei0);
890 /* FIXME: this isn't correct for data */
891 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
892 #else
893 BUG();
894 #endif
895 }
896 BUG_ON(num_refs == 0);
897 } else {
898 num_refs = 0;
899 extent_flags = 0;
900 ret = 0;
901 }
902
903 if (!trans)
904 goto out;
905
906 delayed_refs = &trans->transaction->delayed_refs;
907 spin_lock(&delayed_refs->lock);
908 head = btrfs_find_delayed_ref_head(trans, bytenr);
909 if (head) {
910 if (!mutex_trylock(&head->mutex)) {
911 atomic_inc(&head->node.refs);
912 spin_unlock(&delayed_refs->lock);
913
914 btrfs_release_path(path);
915
916 /*
917 * Mutex was contended, block until it's released and try
918 * again
919 */
920 mutex_lock(&head->mutex);
921 mutex_unlock(&head->mutex);
922 btrfs_put_delayed_ref(&head->node);
923 goto search_again;
924 }
925 spin_lock(&head->lock);
926 if (head->extent_op && head->extent_op->update_flags)
927 extent_flags |= head->extent_op->flags_to_set;
928 else
929 BUG_ON(num_refs == 0);
930
931 num_refs += head->node.ref_mod;
932 spin_unlock(&head->lock);
933 mutex_unlock(&head->mutex);
934 }
935 spin_unlock(&delayed_refs->lock);
936 out:
937 WARN_ON(num_refs == 0);
938 if (refs)
939 *refs = num_refs;
940 if (flags)
941 *flags = extent_flags;
942 out_free:
943 btrfs_free_path(path);
944 return ret;
945 }
946
947 /*
948 * Back reference rules. Back refs have three main goals:
949 *
950 * 1) differentiate between all holders of references to an extent so that
951 * when a reference is dropped we can make sure it was a valid reference
952 * before freeing the extent.
953 *
954 * 2) Provide enough information to quickly find the holders of an extent
955 * if we notice a given block is corrupted or bad.
956 *
957 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
958 * maintenance. This is actually the same as #2, but with a slightly
959 * different use case.
960 *
961 * There are two kinds of back refs. The implicit back refs is optimized
962 * for pointers in non-shared tree blocks. For a given pointer in a block,
963 * back refs of this kind provide information about the block's owner tree
964 * and the pointer's key. These information allow us to find the block by
965 * b-tree searching. The full back refs is for pointers in tree blocks not
966 * referenced by their owner trees. The location of tree block is recorded
967 * in the back refs. Actually the full back refs is generic, and can be
968 * used in all cases the implicit back refs is used. The major shortcoming
969 * of the full back refs is its overhead. Every time a tree block gets
970 * COWed, we have to update back refs entry for all pointers in it.
971 *
972 * For a newly allocated tree block, we use implicit back refs for
973 * pointers in it. This means most tree related operations only involve
974 * implicit back refs. For a tree block created in old transaction, the
975 * only way to drop a reference to it is COW it. So we can detect the
976 * event that tree block loses its owner tree's reference and do the
977 * back refs conversion.
978 *
979 * When a tree block is COWed through a tree, there are four cases:
980 *
981 * The reference count of the block is one and the tree is the block's
982 * owner tree. Nothing to do in this case.
983 *
984 * The reference count of the block is one and the tree is not the
985 * block's owner tree. In this case, full back refs is used for pointers
986 * in the block. Remove these full back refs, add implicit back refs for
987 * every pointers in the new block.
988 *
989 * The reference count of the block is greater than one and the tree is
990 * the block's owner tree. In this case, implicit back refs is used for
991 * pointers in the block. Add full back refs for every pointers in the
992 * block, increase lower level extents' reference counts. The original
993 * implicit back refs are entailed to the new block.
994 *
995 * The reference count of the block is greater than one and the tree is
996 * not the block's owner tree. Add implicit back refs for every pointer in
997 * the new block, increase lower level extents' reference count.
998 *
999 * Back Reference Key composing:
1000 *
1001 * The key objectid corresponds to the first byte in the extent,
1002 * The key type is used to differentiate between types of back refs.
1003 * There are different meanings of the key offset for different types
1004 * of back refs.
1005 *
1006 * File extents can be referenced by:
1007 *
1008 * - multiple snapshots, subvolumes, or different generations in one subvol
1009 * - different files inside a single subvolume
1010 * - different offsets inside a file (bookend extents in file.c)
1011 *
1012 * The extent ref structure for the implicit back refs has fields for:
1013 *
1014 * - Objectid of the subvolume root
1015 * - objectid of the file holding the reference
1016 * - original offset in the file
1017 * - how many bookend extents
1018 *
1019 * The key offset for the implicit back refs is hash of the first
1020 * three fields.
1021 *
1022 * The extent ref structure for the full back refs has field for:
1023 *
1024 * - number of pointers in the tree leaf
1025 *
1026 * The key offset for the implicit back refs is the first byte of
1027 * the tree leaf
1028 *
1029 * When a file extent is allocated, The implicit back refs is used.
1030 * the fields are filled in:
1031 *
1032 * (root_key.objectid, inode objectid, offset in file, 1)
1033 *
1034 * When a file extent is removed file truncation, we find the
1035 * corresponding implicit back refs and check the following fields:
1036 *
1037 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1038 *
1039 * Btree extents can be referenced by:
1040 *
1041 * - Different subvolumes
1042 *
1043 * Both the implicit back refs and the full back refs for tree blocks
1044 * only consist of key. The key offset for the implicit back refs is
1045 * objectid of block's owner tree. The key offset for the full back refs
1046 * is the first byte of parent block.
1047 *
1048 * When implicit back refs is used, information about the lowest key and
1049 * level of the tree block are required. These information are stored in
1050 * tree block info structure.
1051 */
1052
1053 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1054 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
1055 struct btrfs_root *root,
1056 struct btrfs_path *path,
1057 u64 owner, u32 extra_size)
1058 {
1059 struct btrfs_extent_item *item;
1060 struct btrfs_extent_item_v0 *ei0;
1061 struct btrfs_extent_ref_v0 *ref0;
1062 struct btrfs_tree_block_info *bi;
1063 struct extent_buffer *leaf;
1064 struct btrfs_key key;
1065 struct btrfs_key found_key;
1066 u32 new_size = sizeof(*item);
1067 u64 refs;
1068 int ret;
1069
1070 leaf = path->nodes[0];
1071 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1072
1073 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1074 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1075 struct btrfs_extent_item_v0);
1076 refs = btrfs_extent_refs_v0(leaf, ei0);
1077
1078 if (owner == (u64)-1) {
1079 while (1) {
1080 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1081 ret = btrfs_next_leaf(root, path);
1082 if (ret < 0)
1083 return ret;
1084 BUG_ON(ret > 0); /* Corruption */
1085 leaf = path->nodes[0];
1086 }
1087 btrfs_item_key_to_cpu(leaf, &found_key,
1088 path->slots[0]);
1089 BUG_ON(key.objectid != found_key.objectid);
1090 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1091 path->slots[0]++;
1092 continue;
1093 }
1094 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1095 struct btrfs_extent_ref_v0);
1096 owner = btrfs_ref_objectid_v0(leaf, ref0);
1097 break;
1098 }
1099 }
1100 btrfs_release_path(path);
1101
1102 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1103 new_size += sizeof(*bi);
1104
1105 new_size -= sizeof(*ei0);
1106 ret = btrfs_search_slot(trans, root, &key, path,
1107 new_size + extra_size, 1);
1108 if (ret < 0)
1109 return ret;
1110 BUG_ON(ret); /* Corruption */
1111
1112 btrfs_extend_item(root, path, new_size);
1113
1114 leaf = path->nodes[0];
1115 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1116 btrfs_set_extent_refs(leaf, item, refs);
1117 /* FIXME: get real generation */
1118 btrfs_set_extent_generation(leaf, item, 0);
1119 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1120 btrfs_set_extent_flags(leaf, item,
1121 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1122 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1123 bi = (struct btrfs_tree_block_info *)(item + 1);
1124 /* FIXME: get first key of the block */
1125 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1126 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1127 } else {
1128 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1129 }
1130 btrfs_mark_buffer_dirty(leaf);
1131 return 0;
1132 }
1133 #endif
1134
1135 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1136 {
1137 u32 high_crc = ~(u32)0;
1138 u32 low_crc = ~(u32)0;
1139 __le64 lenum;
1140
1141 lenum = cpu_to_le64(root_objectid);
1142 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1143 lenum = cpu_to_le64(owner);
1144 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1145 lenum = cpu_to_le64(offset);
1146 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1147
1148 return ((u64)high_crc << 31) ^ (u64)low_crc;
1149 }
1150
1151 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1152 struct btrfs_extent_data_ref *ref)
1153 {
1154 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1155 btrfs_extent_data_ref_objectid(leaf, ref),
1156 btrfs_extent_data_ref_offset(leaf, ref));
1157 }
1158
1159 static int match_extent_data_ref(struct extent_buffer *leaf,
1160 struct btrfs_extent_data_ref *ref,
1161 u64 root_objectid, u64 owner, u64 offset)
1162 {
1163 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1164 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1165 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1166 return 0;
1167 return 1;
1168 }
1169
1170 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1171 struct btrfs_root *root,
1172 struct btrfs_path *path,
1173 u64 bytenr, u64 parent,
1174 u64 root_objectid,
1175 u64 owner, u64 offset)
1176 {
1177 struct btrfs_key key;
1178 struct btrfs_extent_data_ref *ref;
1179 struct extent_buffer *leaf;
1180 u32 nritems;
1181 int ret;
1182 int recow;
1183 int err = -ENOENT;
1184
1185 key.objectid = bytenr;
1186 if (parent) {
1187 key.type = BTRFS_SHARED_DATA_REF_KEY;
1188 key.offset = parent;
1189 } else {
1190 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1191 key.offset = hash_extent_data_ref(root_objectid,
1192 owner, offset);
1193 }
1194 again:
1195 recow = 0;
1196 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1197 if (ret < 0) {
1198 err = ret;
1199 goto fail;
1200 }
1201
1202 if (parent) {
1203 if (!ret)
1204 return 0;
1205 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1206 key.type = BTRFS_EXTENT_REF_V0_KEY;
1207 btrfs_release_path(path);
1208 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1209 if (ret < 0) {
1210 err = ret;
1211 goto fail;
1212 }
1213 if (!ret)
1214 return 0;
1215 #endif
1216 goto fail;
1217 }
1218
1219 leaf = path->nodes[0];
1220 nritems = btrfs_header_nritems(leaf);
1221 while (1) {
1222 if (path->slots[0] >= nritems) {
1223 ret = btrfs_next_leaf(root, path);
1224 if (ret < 0)
1225 err = ret;
1226 if (ret)
1227 goto fail;
1228
1229 leaf = path->nodes[0];
1230 nritems = btrfs_header_nritems(leaf);
1231 recow = 1;
1232 }
1233
1234 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1235 if (key.objectid != bytenr ||
1236 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1237 goto fail;
1238
1239 ref = btrfs_item_ptr(leaf, path->slots[0],
1240 struct btrfs_extent_data_ref);
1241
1242 if (match_extent_data_ref(leaf, ref, root_objectid,
1243 owner, offset)) {
1244 if (recow) {
1245 btrfs_release_path(path);
1246 goto again;
1247 }
1248 err = 0;
1249 break;
1250 }
1251 path->slots[0]++;
1252 }
1253 fail:
1254 return err;
1255 }
1256
1257 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1258 struct btrfs_root *root,
1259 struct btrfs_path *path,
1260 u64 bytenr, u64 parent,
1261 u64 root_objectid, u64 owner,
1262 u64 offset, int refs_to_add)
1263 {
1264 struct btrfs_key key;
1265 struct extent_buffer *leaf;
1266 u32 size;
1267 u32 num_refs;
1268 int ret;
1269
1270 key.objectid = bytenr;
1271 if (parent) {
1272 key.type = BTRFS_SHARED_DATA_REF_KEY;
1273 key.offset = parent;
1274 size = sizeof(struct btrfs_shared_data_ref);
1275 } else {
1276 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1277 key.offset = hash_extent_data_ref(root_objectid,
1278 owner, offset);
1279 size = sizeof(struct btrfs_extent_data_ref);
1280 }
1281
1282 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1283 if (ret && ret != -EEXIST)
1284 goto fail;
1285
1286 leaf = path->nodes[0];
1287 if (parent) {
1288 struct btrfs_shared_data_ref *ref;
1289 ref = btrfs_item_ptr(leaf, path->slots[0],
1290 struct btrfs_shared_data_ref);
1291 if (ret == 0) {
1292 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1293 } else {
1294 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1295 num_refs += refs_to_add;
1296 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1297 }
1298 } else {
1299 struct btrfs_extent_data_ref *ref;
1300 while (ret == -EEXIST) {
1301 ref = btrfs_item_ptr(leaf, path->slots[0],
1302 struct btrfs_extent_data_ref);
1303 if (match_extent_data_ref(leaf, ref, root_objectid,
1304 owner, offset))
1305 break;
1306 btrfs_release_path(path);
1307 key.offset++;
1308 ret = btrfs_insert_empty_item(trans, root, path, &key,
1309 size);
1310 if (ret && ret != -EEXIST)
1311 goto fail;
1312
1313 leaf = path->nodes[0];
1314 }
1315 ref = btrfs_item_ptr(leaf, path->slots[0],
1316 struct btrfs_extent_data_ref);
1317 if (ret == 0) {
1318 btrfs_set_extent_data_ref_root(leaf, ref,
1319 root_objectid);
1320 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1321 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1322 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1323 } else {
1324 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1325 num_refs += refs_to_add;
1326 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1327 }
1328 }
1329 btrfs_mark_buffer_dirty(leaf);
1330 ret = 0;
1331 fail:
1332 btrfs_release_path(path);
1333 return ret;
1334 }
1335
1336 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1337 struct btrfs_root *root,
1338 struct btrfs_path *path,
1339 int refs_to_drop, int *last_ref)
1340 {
1341 struct btrfs_key key;
1342 struct btrfs_extent_data_ref *ref1 = NULL;
1343 struct btrfs_shared_data_ref *ref2 = NULL;
1344 struct extent_buffer *leaf;
1345 u32 num_refs = 0;
1346 int ret = 0;
1347
1348 leaf = path->nodes[0];
1349 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1350
1351 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1352 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1353 struct btrfs_extent_data_ref);
1354 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1355 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1356 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1357 struct btrfs_shared_data_ref);
1358 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1359 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1360 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1361 struct btrfs_extent_ref_v0 *ref0;
1362 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1363 struct btrfs_extent_ref_v0);
1364 num_refs = btrfs_ref_count_v0(leaf, ref0);
1365 #endif
1366 } else {
1367 BUG();
1368 }
1369
1370 BUG_ON(num_refs < refs_to_drop);
1371 num_refs -= refs_to_drop;
1372
1373 if (num_refs == 0) {
1374 ret = btrfs_del_item(trans, root, path);
1375 *last_ref = 1;
1376 } else {
1377 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1378 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1379 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1380 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1381 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1382 else {
1383 struct btrfs_extent_ref_v0 *ref0;
1384 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1385 struct btrfs_extent_ref_v0);
1386 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1387 }
1388 #endif
1389 btrfs_mark_buffer_dirty(leaf);
1390 }
1391 return ret;
1392 }
1393
1394 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
1395 struct btrfs_extent_inline_ref *iref)
1396 {
1397 struct btrfs_key key;
1398 struct extent_buffer *leaf;
1399 struct btrfs_extent_data_ref *ref1;
1400 struct btrfs_shared_data_ref *ref2;
1401 u32 num_refs = 0;
1402
1403 leaf = path->nodes[0];
1404 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1405 if (iref) {
1406 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1407 BTRFS_EXTENT_DATA_REF_KEY) {
1408 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1409 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1410 } else {
1411 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1412 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1413 }
1414 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1415 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1416 struct btrfs_extent_data_ref);
1417 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1418 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1419 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1420 struct btrfs_shared_data_ref);
1421 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1422 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1423 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1424 struct btrfs_extent_ref_v0 *ref0;
1425 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1426 struct btrfs_extent_ref_v0);
1427 num_refs = btrfs_ref_count_v0(leaf, ref0);
1428 #endif
1429 } else {
1430 WARN_ON(1);
1431 }
1432 return num_refs;
1433 }
1434
1435 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1436 struct btrfs_root *root,
1437 struct btrfs_path *path,
1438 u64 bytenr, u64 parent,
1439 u64 root_objectid)
1440 {
1441 struct btrfs_key key;
1442 int ret;
1443
1444 key.objectid = bytenr;
1445 if (parent) {
1446 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1447 key.offset = parent;
1448 } else {
1449 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1450 key.offset = root_objectid;
1451 }
1452
1453 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1454 if (ret > 0)
1455 ret = -ENOENT;
1456 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1457 if (ret == -ENOENT && parent) {
1458 btrfs_release_path(path);
1459 key.type = BTRFS_EXTENT_REF_V0_KEY;
1460 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1461 if (ret > 0)
1462 ret = -ENOENT;
1463 }
1464 #endif
1465 return ret;
1466 }
1467
1468 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1469 struct btrfs_root *root,
1470 struct btrfs_path *path,
1471 u64 bytenr, u64 parent,
1472 u64 root_objectid)
1473 {
1474 struct btrfs_key key;
1475 int ret;
1476
1477 key.objectid = bytenr;
1478 if (parent) {
1479 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1480 key.offset = parent;
1481 } else {
1482 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1483 key.offset = root_objectid;
1484 }
1485
1486 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1487 btrfs_release_path(path);
1488 return ret;
1489 }
1490
1491 static inline int extent_ref_type(u64 parent, u64 owner)
1492 {
1493 int type;
1494 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1495 if (parent > 0)
1496 type = BTRFS_SHARED_BLOCK_REF_KEY;
1497 else
1498 type = BTRFS_TREE_BLOCK_REF_KEY;
1499 } else {
1500 if (parent > 0)
1501 type = BTRFS_SHARED_DATA_REF_KEY;
1502 else
1503 type = BTRFS_EXTENT_DATA_REF_KEY;
1504 }
1505 return type;
1506 }
1507
1508 static int find_next_key(struct btrfs_path *path, int level,
1509 struct btrfs_key *key)
1510
1511 {
1512 for (; level < BTRFS_MAX_LEVEL; level++) {
1513 if (!path->nodes[level])
1514 break;
1515 if (path->slots[level] + 1 >=
1516 btrfs_header_nritems(path->nodes[level]))
1517 continue;
1518 if (level == 0)
1519 btrfs_item_key_to_cpu(path->nodes[level], key,
1520 path->slots[level] + 1);
1521 else
1522 btrfs_node_key_to_cpu(path->nodes[level], key,
1523 path->slots[level] + 1);
1524 return 0;
1525 }
1526 return 1;
1527 }
1528
1529 /*
1530 * look for inline back ref. if back ref is found, *ref_ret is set
1531 * to the address of inline back ref, and 0 is returned.
1532 *
1533 * if back ref isn't found, *ref_ret is set to the address where it
1534 * should be inserted, and -ENOENT is returned.
1535 *
1536 * if insert is true and there are too many inline back refs, the path
1537 * points to the extent item, and -EAGAIN is returned.
1538 *
1539 * NOTE: inline back refs are ordered in the same way that back ref
1540 * items in the tree are ordered.
1541 */
1542 static noinline_for_stack
1543 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1544 struct btrfs_root *root,
1545 struct btrfs_path *path,
1546 struct btrfs_extent_inline_ref **ref_ret,
1547 u64 bytenr, u64 num_bytes,
1548 u64 parent, u64 root_objectid,
1549 u64 owner, u64 offset, int insert)
1550 {
1551 struct btrfs_key key;
1552 struct extent_buffer *leaf;
1553 struct btrfs_extent_item *ei;
1554 struct btrfs_extent_inline_ref *iref;
1555 u64 flags;
1556 u64 item_size;
1557 unsigned long ptr;
1558 unsigned long end;
1559 int extra_size;
1560 int type;
1561 int want;
1562 int ret;
1563 int err = 0;
1564 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1565 SKINNY_METADATA);
1566
1567 key.objectid = bytenr;
1568 key.type = BTRFS_EXTENT_ITEM_KEY;
1569 key.offset = num_bytes;
1570
1571 want = extent_ref_type(parent, owner);
1572 if (insert) {
1573 extra_size = btrfs_extent_inline_ref_size(want);
1574 path->keep_locks = 1;
1575 } else
1576 extra_size = -1;
1577
1578 /*
1579 * Owner is our parent level, so we can just add one to get the level
1580 * for the block we are interested in.
1581 */
1582 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1583 key.type = BTRFS_METADATA_ITEM_KEY;
1584 key.offset = owner;
1585 }
1586
1587 again:
1588 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1589 if (ret < 0) {
1590 err = ret;
1591 goto out;
1592 }
1593
1594 /*
1595 * We may be a newly converted file system which still has the old fat
1596 * extent entries for metadata, so try and see if we have one of those.
1597 */
1598 if (ret > 0 && skinny_metadata) {
1599 skinny_metadata = false;
1600 if (path->slots[0]) {
1601 path->slots[0]--;
1602 btrfs_item_key_to_cpu(path->nodes[0], &key,
1603 path->slots[0]);
1604 if (key.objectid == bytenr &&
1605 key.type == BTRFS_EXTENT_ITEM_KEY &&
1606 key.offset == num_bytes)
1607 ret = 0;
1608 }
1609 if (ret) {
1610 key.objectid = bytenr;
1611 key.type = BTRFS_EXTENT_ITEM_KEY;
1612 key.offset = num_bytes;
1613 btrfs_release_path(path);
1614 goto again;
1615 }
1616 }
1617
1618 if (ret && !insert) {
1619 err = -ENOENT;
1620 goto out;
1621 } else if (WARN_ON(ret)) {
1622 err = -EIO;
1623 goto out;
1624 }
1625
1626 leaf = path->nodes[0];
1627 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1628 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1629 if (item_size < sizeof(*ei)) {
1630 if (!insert) {
1631 err = -ENOENT;
1632 goto out;
1633 }
1634 ret = convert_extent_item_v0(trans, root, path, owner,
1635 extra_size);
1636 if (ret < 0) {
1637 err = ret;
1638 goto out;
1639 }
1640 leaf = path->nodes[0];
1641 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1642 }
1643 #endif
1644 BUG_ON(item_size < sizeof(*ei));
1645
1646 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1647 flags = btrfs_extent_flags(leaf, ei);
1648
1649 ptr = (unsigned long)(ei + 1);
1650 end = (unsigned long)ei + item_size;
1651
1652 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1653 ptr += sizeof(struct btrfs_tree_block_info);
1654 BUG_ON(ptr > end);
1655 }
1656
1657 err = -ENOENT;
1658 while (1) {
1659 if (ptr >= end) {
1660 WARN_ON(ptr > end);
1661 break;
1662 }
1663 iref = (struct btrfs_extent_inline_ref *)ptr;
1664 type = btrfs_extent_inline_ref_type(leaf, iref);
1665 if (want < type)
1666 break;
1667 if (want > type) {
1668 ptr += btrfs_extent_inline_ref_size(type);
1669 continue;
1670 }
1671
1672 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1673 struct btrfs_extent_data_ref *dref;
1674 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1675 if (match_extent_data_ref(leaf, dref, root_objectid,
1676 owner, offset)) {
1677 err = 0;
1678 break;
1679 }
1680 if (hash_extent_data_ref_item(leaf, dref) <
1681 hash_extent_data_ref(root_objectid, owner, offset))
1682 break;
1683 } else {
1684 u64 ref_offset;
1685 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1686 if (parent > 0) {
1687 if (parent == ref_offset) {
1688 err = 0;
1689 break;
1690 }
1691 if (ref_offset < parent)
1692 break;
1693 } else {
1694 if (root_objectid == ref_offset) {
1695 err = 0;
1696 break;
1697 }
1698 if (ref_offset < root_objectid)
1699 break;
1700 }
1701 }
1702 ptr += btrfs_extent_inline_ref_size(type);
1703 }
1704 if (err == -ENOENT && insert) {
1705 if (item_size + extra_size >=
1706 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1707 err = -EAGAIN;
1708 goto out;
1709 }
1710 /*
1711 * To add new inline back ref, we have to make sure
1712 * there is no corresponding back ref item.
1713 * For simplicity, we just do not add new inline back
1714 * ref if there is any kind of item for this block
1715 */
1716 if (find_next_key(path, 0, &key) == 0 &&
1717 key.objectid == bytenr &&
1718 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1719 err = -EAGAIN;
1720 goto out;
1721 }
1722 }
1723 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1724 out:
1725 if (insert) {
1726 path->keep_locks = 0;
1727 btrfs_unlock_up_safe(path, 1);
1728 }
1729 return err;
1730 }
1731
1732 /*
1733 * helper to add new inline back ref
1734 */
1735 static noinline_for_stack
1736 void setup_inline_extent_backref(struct btrfs_root *root,
1737 struct btrfs_path *path,
1738 struct btrfs_extent_inline_ref *iref,
1739 u64 parent, u64 root_objectid,
1740 u64 owner, u64 offset, int refs_to_add,
1741 struct btrfs_delayed_extent_op *extent_op)
1742 {
1743 struct extent_buffer *leaf;
1744 struct btrfs_extent_item *ei;
1745 unsigned long ptr;
1746 unsigned long end;
1747 unsigned long item_offset;
1748 u64 refs;
1749 int size;
1750 int type;
1751
1752 leaf = path->nodes[0];
1753 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1754 item_offset = (unsigned long)iref - (unsigned long)ei;
1755
1756 type = extent_ref_type(parent, owner);
1757 size = btrfs_extent_inline_ref_size(type);
1758
1759 btrfs_extend_item(root, path, size);
1760
1761 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1762 refs = btrfs_extent_refs(leaf, ei);
1763 refs += refs_to_add;
1764 btrfs_set_extent_refs(leaf, ei, refs);
1765 if (extent_op)
1766 __run_delayed_extent_op(extent_op, leaf, ei);
1767
1768 ptr = (unsigned long)ei + item_offset;
1769 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1770 if (ptr < end - size)
1771 memmove_extent_buffer(leaf, ptr + size, ptr,
1772 end - size - ptr);
1773
1774 iref = (struct btrfs_extent_inline_ref *)ptr;
1775 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1776 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1777 struct btrfs_extent_data_ref *dref;
1778 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1779 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1780 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1781 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1782 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1783 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1784 struct btrfs_shared_data_ref *sref;
1785 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1786 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1787 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1788 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1789 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1790 } else {
1791 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1792 }
1793 btrfs_mark_buffer_dirty(leaf);
1794 }
1795
1796 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1797 struct btrfs_root *root,
1798 struct btrfs_path *path,
1799 struct btrfs_extent_inline_ref **ref_ret,
1800 u64 bytenr, u64 num_bytes, u64 parent,
1801 u64 root_objectid, u64 owner, u64 offset)
1802 {
1803 int ret;
1804
1805 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1806 bytenr, num_bytes, parent,
1807 root_objectid, owner, offset, 0);
1808 if (ret != -ENOENT)
1809 return ret;
1810
1811 btrfs_release_path(path);
1812 *ref_ret = NULL;
1813
1814 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1815 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1816 root_objectid);
1817 } else {
1818 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1819 root_objectid, owner, offset);
1820 }
1821 return ret;
1822 }
1823
1824 /*
1825 * helper to update/remove inline back ref
1826 */
1827 static noinline_for_stack
1828 void update_inline_extent_backref(struct btrfs_root *root,
1829 struct btrfs_path *path,
1830 struct btrfs_extent_inline_ref *iref,
1831 int refs_to_mod,
1832 struct btrfs_delayed_extent_op *extent_op,
1833 int *last_ref)
1834 {
1835 struct extent_buffer *leaf;
1836 struct btrfs_extent_item *ei;
1837 struct btrfs_extent_data_ref *dref = NULL;
1838 struct btrfs_shared_data_ref *sref = NULL;
1839 unsigned long ptr;
1840 unsigned long end;
1841 u32 item_size;
1842 int size;
1843 int type;
1844 u64 refs;
1845
1846 leaf = path->nodes[0];
1847 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1848 refs = btrfs_extent_refs(leaf, ei);
1849 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1850 refs += refs_to_mod;
1851 btrfs_set_extent_refs(leaf, ei, refs);
1852 if (extent_op)
1853 __run_delayed_extent_op(extent_op, leaf, ei);
1854
1855 type = btrfs_extent_inline_ref_type(leaf, iref);
1856
1857 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1858 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1859 refs = btrfs_extent_data_ref_count(leaf, dref);
1860 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1861 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1862 refs = btrfs_shared_data_ref_count(leaf, sref);
1863 } else {
1864 refs = 1;
1865 BUG_ON(refs_to_mod != -1);
1866 }
1867
1868 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1869 refs += refs_to_mod;
1870
1871 if (refs > 0) {
1872 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1873 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1874 else
1875 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1876 } else {
1877 *last_ref = 1;
1878 size = btrfs_extent_inline_ref_size(type);
1879 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1880 ptr = (unsigned long)iref;
1881 end = (unsigned long)ei + item_size;
1882 if (ptr + size < end)
1883 memmove_extent_buffer(leaf, ptr, ptr + size,
1884 end - ptr - size);
1885 item_size -= size;
1886 btrfs_truncate_item(root, path, item_size, 1);
1887 }
1888 btrfs_mark_buffer_dirty(leaf);
1889 }
1890
1891 static noinline_for_stack
1892 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1893 struct btrfs_root *root,
1894 struct btrfs_path *path,
1895 u64 bytenr, u64 num_bytes, u64 parent,
1896 u64 root_objectid, u64 owner,
1897 u64 offset, int refs_to_add,
1898 struct btrfs_delayed_extent_op *extent_op)
1899 {
1900 struct btrfs_extent_inline_ref *iref;
1901 int ret;
1902
1903 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1904 bytenr, num_bytes, parent,
1905 root_objectid, owner, offset, 1);
1906 if (ret == 0) {
1907 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1908 update_inline_extent_backref(root, path, iref,
1909 refs_to_add, extent_op, NULL);
1910 } else if (ret == -ENOENT) {
1911 setup_inline_extent_backref(root, path, iref, parent,
1912 root_objectid, owner, offset,
1913 refs_to_add, extent_op);
1914 ret = 0;
1915 }
1916 return ret;
1917 }
1918
1919 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1920 struct btrfs_root *root,
1921 struct btrfs_path *path,
1922 u64 bytenr, u64 parent, u64 root_objectid,
1923 u64 owner, u64 offset, int refs_to_add)
1924 {
1925 int ret;
1926 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1927 BUG_ON(refs_to_add != 1);
1928 ret = insert_tree_block_ref(trans, root, path, bytenr,
1929 parent, root_objectid);
1930 } else {
1931 ret = insert_extent_data_ref(trans, root, path, bytenr,
1932 parent, root_objectid,
1933 owner, offset, refs_to_add);
1934 }
1935 return ret;
1936 }
1937
1938 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1939 struct btrfs_root *root,
1940 struct btrfs_path *path,
1941 struct btrfs_extent_inline_ref *iref,
1942 int refs_to_drop, int is_data, int *last_ref)
1943 {
1944 int ret = 0;
1945
1946 BUG_ON(!is_data && refs_to_drop != 1);
1947 if (iref) {
1948 update_inline_extent_backref(root, path, iref,
1949 -refs_to_drop, NULL, last_ref);
1950 } else if (is_data) {
1951 ret = remove_extent_data_ref(trans, root, path, refs_to_drop,
1952 last_ref);
1953 } else {
1954 *last_ref = 1;
1955 ret = btrfs_del_item(trans, root, path);
1956 }
1957 return ret;
1958 }
1959
1960 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
1961 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1962 u64 *discarded_bytes)
1963 {
1964 int j, ret = 0;
1965 u64 bytes_left, end;
1966 u64 aligned_start = ALIGN(start, 1 << 9);
1967
1968 if (WARN_ON(start != aligned_start)) {
1969 len -= aligned_start - start;
1970 len = round_down(len, 1 << 9);
1971 start = aligned_start;
1972 }
1973
1974 *discarded_bytes = 0;
1975
1976 if (!len)
1977 return 0;
1978
1979 end = start + len;
1980 bytes_left = len;
1981
1982 /* Skip any superblocks on this device. */
1983 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1984 u64 sb_start = btrfs_sb_offset(j);
1985 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
1986 u64 size = sb_start - start;
1987
1988 if (!in_range(sb_start, start, bytes_left) &&
1989 !in_range(sb_end, start, bytes_left) &&
1990 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
1991 continue;
1992
1993 /*
1994 * Superblock spans beginning of range. Adjust start and
1995 * try again.
1996 */
1997 if (sb_start <= start) {
1998 start += sb_end - start;
1999 if (start > end) {
2000 bytes_left = 0;
2001 break;
2002 }
2003 bytes_left = end - start;
2004 continue;
2005 }
2006
2007 if (size) {
2008 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
2009 GFP_NOFS, 0);
2010 if (!ret)
2011 *discarded_bytes += size;
2012 else if (ret != -EOPNOTSUPP)
2013 return ret;
2014 }
2015
2016 start = sb_end;
2017 if (start > end) {
2018 bytes_left = 0;
2019 break;
2020 }
2021 bytes_left = end - start;
2022 }
2023
2024 if (bytes_left) {
2025 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
2026 GFP_NOFS, 0);
2027 if (!ret)
2028 *discarded_bytes += bytes_left;
2029 }
2030 return ret;
2031 }
2032
2033 int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
2034 u64 num_bytes, u64 *actual_bytes)
2035 {
2036 int ret;
2037 u64 discarded_bytes = 0;
2038 struct btrfs_bio *bbio = NULL;
2039
2040
2041 /*
2042 * Avoid races with device replace and make sure our bbio has devices
2043 * associated to its stripes that don't go away while we are discarding.
2044 */
2045 btrfs_bio_counter_inc_blocked(root->fs_info);
2046 /* Tell the block device(s) that the sectors can be discarded */
2047 ret = btrfs_map_block(root->fs_info, REQ_OP_DISCARD,
2048 bytenr, &num_bytes, &bbio, 0);
2049 /* Error condition is -ENOMEM */
2050 if (!ret) {
2051 struct btrfs_bio_stripe *stripe = bbio->stripes;
2052 int i;
2053
2054
2055 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
2056 u64 bytes;
2057 if (!stripe->dev->can_discard)
2058 continue;
2059
2060 ret = btrfs_issue_discard(stripe->dev->bdev,
2061 stripe->physical,
2062 stripe->length,
2063 &bytes);
2064 if (!ret)
2065 discarded_bytes += bytes;
2066 else if (ret != -EOPNOTSUPP)
2067 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2068
2069 /*
2070 * Just in case we get back EOPNOTSUPP for some reason,
2071 * just ignore the return value so we don't screw up
2072 * people calling discard_extent.
2073 */
2074 ret = 0;
2075 }
2076 btrfs_put_bbio(bbio);
2077 }
2078 btrfs_bio_counter_dec(root->fs_info);
2079
2080 if (actual_bytes)
2081 *actual_bytes = discarded_bytes;
2082
2083
2084 if (ret == -EOPNOTSUPP)
2085 ret = 0;
2086 return ret;
2087 }
2088
2089 /* Can return -ENOMEM */
2090 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2091 struct btrfs_root *root,
2092 u64 bytenr, u64 num_bytes, u64 parent,
2093 u64 root_objectid, u64 owner, u64 offset)
2094 {
2095 int ret;
2096 struct btrfs_fs_info *fs_info = root->fs_info;
2097
2098 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2099 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2100
2101 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2102 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
2103 num_bytes,
2104 parent, root_objectid, (int)owner,
2105 BTRFS_ADD_DELAYED_REF, NULL);
2106 } else {
2107 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
2108 num_bytes, parent, root_objectid,
2109 owner, offset, 0,
2110 BTRFS_ADD_DELAYED_REF, NULL);
2111 }
2112 return ret;
2113 }
2114
2115 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2116 struct btrfs_root *root,
2117 struct btrfs_delayed_ref_node *node,
2118 u64 parent, u64 root_objectid,
2119 u64 owner, u64 offset, int refs_to_add,
2120 struct btrfs_delayed_extent_op *extent_op)
2121 {
2122 struct btrfs_fs_info *fs_info = root->fs_info;
2123 struct btrfs_path *path;
2124 struct extent_buffer *leaf;
2125 struct btrfs_extent_item *item;
2126 struct btrfs_key key;
2127 u64 bytenr = node->bytenr;
2128 u64 num_bytes = node->num_bytes;
2129 u64 refs;
2130 int ret;
2131
2132 path = btrfs_alloc_path();
2133 if (!path)
2134 return -ENOMEM;
2135
2136 path->reada = READA_FORWARD;
2137 path->leave_spinning = 1;
2138 /* this will setup the path even if it fails to insert the back ref */
2139 ret = insert_inline_extent_backref(trans, fs_info->extent_root, path,
2140 bytenr, num_bytes, parent,
2141 root_objectid, owner, offset,
2142 refs_to_add, extent_op);
2143 if ((ret < 0 && ret != -EAGAIN) || !ret)
2144 goto out;
2145
2146 /*
2147 * Ok we had -EAGAIN which means we didn't have space to insert and
2148 * inline extent ref, so just update the reference count and add a
2149 * normal backref.
2150 */
2151 leaf = path->nodes[0];
2152 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2153 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2154 refs = btrfs_extent_refs(leaf, item);
2155 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2156 if (extent_op)
2157 __run_delayed_extent_op(extent_op, leaf, item);
2158
2159 btrfs_mark_buffer_dirty(leaf);
2160 btrfs_release_path(path);
2161
2162 path->reada = READA_FORWARD;
2163 path->leave_spinning = 1;
2164 /* now insert the actual backref */
2165 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2166 path, bytenr, parent, root_objectid,
2167 owner, offset, refs_to_add);
2168 if (ret)
2169 btrfs_abort_transaction(trans, ret);
2170 out:
2171 btrfs_free_path(path);
2172 return ret;
2173 }
2174
2175 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2176 struct btrfs_root *root,
2177 struct btrfs_delayed_ref_node *node,
2178 struct btrfs_delayed_extent_op *extent_op,
2179 int insert_reserved)
2180 {
2181 int ret = 0;
2182 struct btrfs_delayed_data_ref *ref;
2183 struct btrfs_key ins;
2184 u64 parent = 0;
2185 u64 ref_root = 0;
2186 u64 flags = 0;
2187
2188 ins.objectid = node->bytenr;
2189 ins.offset = node->num_bytes;
2190 ins.type = BTRFS_EXTENT_ITEM_KEY;
2191
2192 ref = btrfs_delayed_node_to_data_ref(node);
2193 trace_run_delayed_data_ref(root->fs_info, node, ref, node->action);
2194
2195 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2196 parent = ref->parent;
2197 ref_root = ref->root;
2198
2199 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2200 if (extent_op)
2201 flags |= extent_op->flags_to_set;
2202 ret = alloc_reserved_file_extent(trans, root,
2203 parent, ref_root, flags,
2204 ref->objectid, ref->offset,
2205 &ins, node->ref_mod);
2206 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2207 ret = __btrfs_inc_extent_ref(trans, root, node, parent,
2208 ref_root, ref->objectid,
2209 ref->offset, node->ref_mod,
2210 extent_op);
2211 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2212 ret = __btrfs_free_extent(trans, root, node, parent,
2213 ref_root, ref->objectid,
2214 ref->offset, node->ref_mod,
2215 extent_op);
2216 } else {
2217 BUG();
2218 }
2219 return ret;
2220 }
2221
2222 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2223 struct extent_buffer *leaf,
2224 struct btrfs_extent_item *ei)
2225 {
2226 u64 flags = btrfs_extent_flags(leaf, ei);
2227 if (extent_op->update_flags) {
2228 flags |= extent_op->flags_to_set;
2229 btrfs_set_extent_flags(leaf, ei, flags);
2230 }
2231
2232 if (extent_op->update_key) {
2233 struct btrfs_tree_block_info *bi;
2234 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2235 bi = (struct btrfs_tree_block_info *)(ei + 1);
2236 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2237 }
2238 }
2239
2240 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2241 struct btrfs_root *root,
2242 struct btrfs_delayed_ref_node *node,
2243 struct btrfs_delayed_extent_op *extent_op)
2244 {
2245 struct btrfs_key key;
2246 struct btrfs_path *path;
2247 struct btrfs_extent_item *ei;
2248 struct extent_buffer *leaf;
2249 u32 item_size;
2250 int ret;
2251 int err = 0;
2252 int metadata = !extent_op->is_data;
2253
2254 if (trans->aborted)
2255 return 0;
2256
2257 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2258 metadata = 0;
2259
2260 path = btrfs_alloc_path();
2261 if (!path)
2262 return -ENOMEM;
2263
2264 key.objectid = node->bytenr;
2265
2266 if (metadata) {
2267 key.type = BTRFS_METADATA_ITEM_KEY;
2268 key.offset = extent_op->level;
2269 } else {
2270 key.type = BTRFS_EXTENT_ITEM_KEY;
2271 key.offset = node->num_bytes;
2272 }
2273
2274 again:
2275 path->reada = READA_FORWARD;
2276 path->leave_spinning = 1;
2277 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2278 path, 0, 1);
2279 if (ret < 0) {
2280 err = ret;
2281 goto out;
2282 }
2283 if (ret > 0) {
2284 if (metadata) {
2285 if (path->slots[0] > 0) {
2286 path->slots[0]--;
2287 btrfs_item_key_to_cpu(path->nodes[0], &key,
2288 path->slots[0]);
2289 if (key.objectid == node->bytenr &&
2290 key.type == BTRFS_EXTENT_ITEM_KEY &&
2291 key.offset == node->num_bytes)
2292 ret = 0;
2293 }
2294 if (ret > 0) {
2295 btrfs_release_path(path);
2296 metadata = 0;
2297
2298 key.objectid = node->bytenr;
2299 key.offset = node->num_bytes;
2300 key.type = BTRFS_EXTENT_ITEM_KEY;
2301 goto again;
2302 }
2303 } else {
2304 err = -EIO;
2305 goto out;
2306 }
2307 }
2308
2309 leaf = path->nodes[0];
2310 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2311 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2312 if (item_size < sizeof(*ei)) {
2313 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2314 path, (u64)-1, 0);
2315 if (ret < 0) {
2316 err = ret;
2317 goto out;
2318 }
2319 leaf = path->nodes[0];
2320 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2321 }
2322 #endif
2323 BUG_ON(item_size < sizeof(*ei));
2324 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2325 __run_delayed_extent_op(extent_op, leaf, ei);
2326
2327 btrfs_mark_buffer_dirty(leaf);
2328 out:
2329 btrfs_free_path(path);
2330 return err;
2331 }
2332
2333 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2334 struct btrfs_root *root,
2335 struct btrfs_delayed_ref_node *node,
2336 struct btrfs_delayed_extent_op *extent_op,
2337 int insert_reserved)
2338 {
2339 int ret = 0;
2340 struct btrfs_delayed_tree_ref *ref;
2341 struct btrfs_key ins;
2342 u64 parent = 0;
2343 u64 ref_root = 0;
2344 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2345 SKINNY_METADATA);
2346
2347 ref = btrfs_delayed_node_to_tree_ref(node);
2348 trace_run_delayed_tree_ref(root->fs_info, node, ref, node->action);
2349
2350 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2351 parent = ref->parent;
2352 ref_root = ref->root;
2353
2354 ins.objectid = node->bytenr;
2355 if (skinny_metadata) {
2356 ins.offset = ref->level;
2357 ins.type = BTRFS_METADATA_ITEM_KEY;
2358 } else {
2359 ins.offset = node->num_bytes;
2360 ins.type = BTRFS_EXTENT_ITEM_KEY;
2361 }
2362
2363 BUG_ON(node->ref_mod != 1);
2364 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2365 BUG_ON(!extent_op || !extent_op->update_flags);
2366 ret = alloc_reserved_tree_block(trans, root,
2367 parent, ref_root,
2368 extent_op->flags_to_set,
2369 &extent_op->key,
2370 ref->level, &ins);
2371 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2372 ret = __btrfs_inc_extent_ref(trans, root, node,
2373 parent, ref_root,
2374 ref->level, 0, 1,
2375 extent_op);
2376 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2377 ret = __btrfs_free_extent(trans, root, node,
2378 parent, ref_root,
2379 ref->level, 0, 1, extent_op);
2380 } else {
2381 BUG();
2382 }
2383 return ret;
2384 }
2385
2386 /* helper function to actually process a single delayed ref entry */
2387 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2388 struct btrfs_root *root,
2389 struct btrfs_delayed_ref_node *node,
2390 struct btrfs_delayed_extent_op *extent_op,
2391 int insert_reserved)
2392 {
2393 int ret = 0;
2394
2395 if (trans->aborted) {
2396 if (insert_reserved)
2397 btrfs_pin_extent(root, node->bytenr,
2398 node->num_bytes, 1);
2399 return 0;
2400 }
2401
2402 if (btrfs_delayed_ref_is_head(node)) {
2403 struct btrfs_delayed_ref_head *head;
2404 /*
2405 * we've hit the end of the chain and we were supposed
2406 * to insert this extent into the tree. But, it got
2407 * deleted before we ever needed to insert it, so all
2408 * we have to do is clean up the accounting
2409 */
2410 BUG_ON(extent_op);
2411 head = btrfs_delayed_node_to_head(node);
2412 trace_run_delayed_ref_head(root->fs_info, node, head,
2413 node->action);
2414
2415 if (insert_reserved) {
2416 btrfs_pin_extent(root, node->bytenr,
2417 node->num_bytes, 1);
2418 if (head->is_data) {
2419 ret = btrfs_del_csums(trans, root,
2420 node->bytenr,
2421 node->num_bytes);
2422 }
2423 }
2424
2425 /* Also free its reserved qgroup space */
2426 btrfs_qgroup_free_delayed_ref(root->fs_info,
2427 head->qgroup_ref_root,
2428 head->qgroup_reserved);
2429 return ret;
2430 }
2431
2432 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2433 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2434 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2435 insert_reserved);
2436 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2437 node->type == BTRFS_SHARED_DATA_REF_KEY)
2438 ret = run_delayed_data_ref(trans, root, node, extent_op,
2439 insert_reserved);
2440 else
2441 BUG();
2442 return ret;
2443 }
2444
2445 static inline struct btrfs_delayed_ref_node *
2446 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2447 {
2448 struct btrfs_delayed_ref_node *ref;
2449
2450 if (list_empty(&head->ref_list))
2451 return NULL;
2452
2453 /*
2454 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2455 * This is to prevent a ref count from going down to zero, which deletes
2456 * the extent item from the extent tree, when there still are references
2457 * to add, which would fail because they would not find the extent item.
2458 */
2459 list_for_each_entry(ref, &head->ref_list, list) {
2460 if (ref->action == BTRFS_ADD_DELAYED_REF)
2461 return ref;
2462 }
2463
2464 return list_entry(head->ref_list.next, struct btrfs_delayed_ref_node,
2465 list);
2466 }
2467
2468 /*
2469 * Returns 0 on success or if called with an already aborted transaction.
2470 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2471 */
2472 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2473 struct btrfs_root *root,
2474 unsigned long nr)
2475 {
2476 struct btrfs_delayed_ref_root *delayed_refs;
2477 struct btrfs_delayed_ref_node *ref;
2478 struct btrfs_delayed_ref_head *locked_ref = NULL;
2479 struct btrfs_delayed_extent_op *extent_op;
2480 struct btrfs_fs_info *fs_info = root->fs_info;
2481 ktime_t start = ktime_get();
2482 int ret;
2483 unsigned long count = 0;
2484 unsigned long actual_count = 0;
2485 int must_insert_reserved = 0;
2486
2487 delayed_refs = &trans->transaction->delayed_refs;
2488 while (1) {
2489 if (!locked_ref) {
2490 if (count >= nr)
2491 break;
2492
2493 spin_lock(&delayed_refs->lock);
2494 locked_ref = btrfs_select_ref_head(trans);
2495 if (!locked_ref) {
2496 spin_unlock(&delayed_refs->lock);
2497 break;
2498 }
2499
2500 /* grab the lock that says we are going to process
2501 * all the refs for this head */
2502 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2503 spin_unlock(&delayed_refs->lock);
2504 /*
2505 * we may have dropped the spin lock to get the head
2506 * mutex lock, and that might have given someone else
2507 * time to free the head. If that's true, it has been
2508 * removed from our list and we can move on.
2509 */
2510 if (ret == -EAGAIN) {
2511 locked_ref = NULL;
2512 count++;
2513 continue;
2514 }
2515 }
2516
2517 /*
2518 * We need to try and merge add/drops of the same ref since we
2519 * can run into issues with relocate dropping the implicit ref
2520 * and then it being added back again before the drop can
2521 * finish. If we merged anything we need to re-loop so we can
2522 * get a good ref.
2523 * Or we can get node references of the same type that weren't
2524 * merged when created due to bumps in the tree mod seq, and
2525 * we need to merge them to prevent adding an inline extent
2526 * backref before dropping it (triggering a BUG_ON at
2527 * insert_inline_extent_backref()).
2528 */
2529 spin_lock(&locked_ref->lock);
2530 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2531 locked_ref);
2532
2533 /*
2534 * locked_ref is the head node, so we have to go one
2535 * node back for any delayed ref updates
2536 */
2537 ref = select_delayed_ref(locked_ref);
2538
2539 if (ref && ref->seq &&
2540 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2541 spin_unlock(&locked_ref->lock);
2542 btrfs_delayed_ref_unlock(locked_ref);
2543 spin_lock(&delayed_refs->lock);
2544 locked_ref->processing = 0;
2545 delayed_refs->num_heads_ready++;
2546 spin_unlock(&delayed_refs->lock);
2547 locked_ref = NULL;
2548 cond_resched();
2549 count++;
2550 continue;
2551 }
2552
2553 /*
2554 * record the must insert reserved flag before we
2555 * drop the spin lock.
2556 */
2557 must_insert_reserved = locked_ref->must_insert_reserved;
2558 locked_ref->must_insert_reserved = 0;
2559
2560 extent_op = locked_ref->extent_op;
2561 locked_ref->extent_op = NULL;
2562
2563 if (!ref) {
2564
2565
2566 /* All delayed refs have been processed, Go ahead
2567 * and send the head node to run_one_delayed_ref,
2568 * so that any accounting fixes can happen
2569 */
2570 ref = &locked_ref->node;
2571
2572 if (extent_op && must_insert_reserved) {
2573 btrfs_free_delayed_extent_op(extent_op);
2574 extent_op = NULL;
2575 }
2576
2577 if (extent_op) {
2578 spin_unlock(&locked_ref->lock);
2579 ret = run_delayed_extent_op(trans, root,
2580 ref, extent_op);
2581 btrfs_free_delayed_extent_op(extent_op);
2582
2583 if (ret) {
2584 /*
2585 * Need to reset must_insert_reserved if
2586 * there was an error so the abort stuff
2587 * can cleanup the reserved space
2588 * properly.
2589 */
2590 if (must_insert_reserved)
2591 locked_ref->must_insert_reserved = 1;
2592 locked_ref->processing = 0;
2593 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2594 btrfs_delayed_ref_unlock(locked_ref);
2595 return ret;
2596 }
2597 continue;
2598 }
2599
2600 /*
2601 * Need to drop our head ref lock and re-acquire the
2602 * delayed ref lock and then re-check to make sure
2603 * nobody got added.
2604 */
2605 spin_unlock(&locked_ref->lock);
2606 spin_lock(&delayed_refs->lock);
2607 spin_lock(&locked_ref->lock);
2608 if (!list_empty(&locked_ref->ref_list) ||
2609 locked_ref->extent_op) {
2610 spin_unlock(&locked_ref->lock);
2611 spin_unlock(&delayed_refs->lock);
2612 continue;
2613 }
2614 ref->in_tree = 0;
2615 delayed_refs->num_heads--;
2616 rb_erase(&locked_ref->href_node,
2617 &delayed_refs->href_root);
2618 spin_unlock(&delayed_refs->lock);
2619 } else {
2620 actual_count++;
2621 ref->in_tree = 0;
2622 list_del(&ref->list);
2623 }
2624 atomic_dec(&delayed_refs->num_entries);
2625
2626 if (!btrfs_delayed_ref_is_head(ref)) {
2627 /*
2628 * when we play the delayed ref, also correct the
2629 * ref_mod on head
2630 */
2631 switch (ref->action) {
2632 case BTRFS_ADD_DELAYED_REF:
2633 case BTRFS_ADD_DELAYED_EXTENT:
2634 locked_ref->node.ref_mod -= ref->ref_mod;
2635 break;
2636 case BTRFS_DROP_DELAYED_REF:
2637 locked_ref->node.ref_mod += ref->ref_mod;
2638 break;
2639 default:
2640 WARN_ON(1);
2641 }
2642 }
2643 spin_unlock(&locked_ref->lock);
2644
2645 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2646 must_insert_reserved);
2647
2648 btrfs_free_delayed_extent_op(extent_op);
2649 if (ret) {
2650 spin_lock(&delayed_refs->lock);
2651 locked_ref->processing = 0;
2652 delayed_refs->num_heads_ready++;
2653 spin_unlock(&delayed_refs->lock);
2654 btrfs_delayed_ref_unlock(locked_ref);
2655 btrfs_put_delayed_ref(ref);
2656 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2657 return ret;
2658 }
2659
2660 /*
2661 * If this node is a head, that means all the refs in this head
2662 * have been dealt with, and we will pick the next head to deal
2663 * with, so we must unlock the head and drop it from the cluster
2664 * list before we release it.
2665 */
2666 if (btrfs_delayed_ref_is_head(ref)) {
2667 if (locked_ref->is_data &&
2668 locked_ref->total_ref_mod < 0) {
2669 spin_lock(&delayed_refs->lock);
2670 delayed_refs->pending_csums -= ref->num_bytes;
2671 spin_unlock(&delayed_refs->lock);
2672 }
2673 btrfs_delayed_ref_unlock(locked_ref);
2674 locked_ref = NULL;
2675 }
2676 btrfs_put_delayed_ref(ref);
2677 count++;
2678 cond_resched();
2679 }
2680
2681 /*
2682 * We don't want to include ref heads since we can have empty ref heads
2683 * and those will drastically skew our runtime down since we just do
2684 * accounting, no actual extent tree updates.
2685 */
2686 if (actual_count > 0) {
2687 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2688 u64 avg;
2689
2690 /*
2691 * We weigh the current average higher than our current runtime
2692 * to avoid large swings in the average.
2693 */
2694 spin_lock(&delayed_refs->lock);
2695 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2696 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2697 spin_unlock(&delayed_refs->lock);
2698 }
2699 return 0;
2700 }
2701
2702 #ifdef SCRAMBLE_DELAYED_REFS
2703 /*
2704 * Normally delayed refs get processed in ascending bytenr order. This
2705 * correlates in most cases to the order added. To expose dependencies on this
2706 * order, we start to process the tree in the middle instead of the beginning
2707 */
2708 static u64 find_middle(struct rb_root *root)
2709 {
2710 struct rb_node *n = root->rb_node;
2711 struct btrfs_delayed_ref_node *entry;
2712 int alt = 1;
2713 u64 middle;
2714 u64 first = 0, last = 0;
2715
2716 n = rb_first(root);
2717 if (n) {
2718 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2719 first = entry->bytenr;
2720 }
2721 n = rb_last(root);
2722 if (n) {
2723 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2724 last = entry->bytenr;
2725 }
2726 n = root->rb_node;
2727
2728 while (n) {
2729 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2730 WARN_ON(!entry->in_tree);
2731
2732 middle = entry->bytenr;
2733
2734 if (alt)
2735 n = n->rb_left;
2736 else
2737 n = n->rb_right;
2738
2739 alt = 1 - alt;
2740 }
2741 return middle;
2742 }
2743 #endif
2744
2745 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2746 {
2747 u64 num_bytes;
2748
2749 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2750 sizeof(struct btrfs_extent_inline_ref));
2751 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2752 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2753
2754 /*
2755 * We don't ever fill up leaves all the way so multiply by 2 just to be
2756 * closer to what we're really going to want to use.
2757 */
2758 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2759 }
2760
2761 /*
2762 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2763 * would require to store the csums for that many bytes.
2764 */
2765 u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes)
2766 {
2767 u64 csum_size;
2768 u64 num_csums_per_leaf;
2769 u64 num_csums;
2770
2771 csum_size = BTRFS_MAX_ITEM_SIZE(root);
2772 num_csums_per_leaf = div64_u64(csum_size,
2773 (u64)btrfs_super_csum_size(root->fs_info->super_copy));
2774 num_csums = div64_u64(csum_bytes, root->sectorsize);
2775 num_csums += num_csums_per_leaf - 1;
2776 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2777 return num_csums;
2778 }
2779
2780 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2781 struct btrfs_root *root)
2782 {
2783 struct btrfs_block_rsv *global_rsv;
2784 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2785 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2786 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2787 u64 num_bytes, num_dirty_bgs_bytes;
2788 int ret = 0;
2789
2790 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2791 num_heads = heads_to_leaves(root, num_heads);
2792 if (num_heads > 1)
2793 num_bytes += (num_heads - 1) * root->nodesize;
2794 num_bytes <<= 1;
2795 num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize;
2796 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root,
2797 num_dirty_bgs);
2798 global_rsv = &root->fs_info->global_block_rsv;
2799
2800 /*
2801 * If we can't allocate any more chunks lets make sure we have _lots_ of
2802 * wiggle room since running delayed refs can create more delayed refs.
2803 */
2804 if (global_rsv->space_info->full) {
2805 num_dirty_bgs_bytes <<= 1;
2806 num_bytes <<= 1;
2807 }
2808
2809 spin_lock(&global_rsv->lock);
2810 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2811 ret = 1;
2812 spin_unlock(&global_rsv->lock);
2813 return ret;
2814 }
2815
2816 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2817 struct btrfs_root *root)
2818 {
2819 struct btrfs_fs_info *fs_info = root->fs_info;
2820 u64 num_entries =
2821 atomic_read(&trans->transaction->delayed_refs.num_entries);
2822 u64 avg_runtime;
2823 u64 val;
2824
2825 smp_mb();
2826 avg_runtime = fs_info->avg_delayed_ref_runtime;
2827 val = num_entries * avg_runtime;
2828 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2829 return 1;
2830 if (val >= NSEC_PER_SEC / 2)
2831 return 2;
2832
2833 return btrfs_check_space_for_delayed_refs(trans, root);
2834 }
2835
2836 struct async_delayed_refs {
2837 struct btrfs_root *root;
2838 u64 transid;
2839 int count;
2840 int error;
2841 int sync;
2842 struct completion wait;
2843 struct btrfs_work work;
2844 };
2845
2846 static void delayed_ref_async_start(struct btrfs_work *work)
2847 {
2848 struct async_delayed_refs *async;
2849 struct btrfs_trans_handle *trans;
2850 int ret;
2851
2852 async = container_of(work, struct async_delayed_refs, work);
2853
2854 /* if the commit is already started, we don't need to wait here */
2855 if (btrfs_transaction_blocked(async->root->fs_info))
2856 goto done;
2857
2858 trans = btrfs_join_transaction(async->root);
2859 if (IS_ERR(trans)) {
2860 async->error = PTR_ERR(trans);
2861 goto done;
2862 }
2863
2864 /*
2865 * trans->sync means that when we call end_transaction, we won't
2866 * wait on delayed refs
2867 */
2868 trans->sync = true;
2869
2870 /* Don't bother flushing if we got into a different transaction */
2871 if (trans->transid > async->transid)
2872 goto end;
2873
2874 ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2875 if (ret)
2876 async->error = ret;
2877 end:
2878 ret = btrfs_end_transaction(trans, async->root);
2879 if (ret && !async->error)
2880 async->error = ret;
2881 done:
2882 if (async->sync)
2883 complete(&async->wait);
2884 else
2885 kfree(async);
2886 }
2887
2888 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2889 unsigned long count, u64 transid, int wait)
2890 {
2891 struct async_delayed_refs *async;
2892 int ret;
2893
2894 async = kmalloc(sizeof(*async), GFP_NOFS);
2895 if (!async)
2896 return -ENOMEM;
2897
2898 async->root = root->fs_info->tree_root;
2899 async->count = count;
2900 async->error = 0;
2901 async->transid = transid;
2902 if (wait)
2903 async->sync = 1;
2904 else
2905 async->sync = 0;
2906 init_completion(&async->wait);
2907
2908 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2909 delayed_ref_async_start, NULL, NULL);
2910
2911 btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2912
2913 if (wait) {
2914 wait_for_completion(&async->wait);
2915 ret = async->error;
2916 kfree(async);
2917 return ret;
2918 }
2919 return 0;
2920 }
2921
2922 /*
2923 * this starts processing the delayed reference count updates and
2924 * extent insertions we have queued up so far. count can be
2925 * 0, which means to process everything in the tree at the start
2926 * of the run (but not newly added entries), or it can be some target
2927 * number you'd like to process.
2928 *
2929 * Returns 0 on success or if called with an aborted transaction
2930 * Returns <0 on error and aborts the transaction
2931 */
2932 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2933 struct btrfs_root *root, unsigned long count)
2934 {
2935 struct rb_node *node;
2936 struct btrfs_delayed_ref_root *delayed_refs;
2937 struct btrfs_delayed_ref_head *head;
2938 int ret;
2939 int run_all = count == (unsigned long)-1;
2940 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
2941
2942 /* We'll clean this up in btrfs_cleanup_transaction */
2943 if (trans->aborted)
2944 return 0;
2945
2946 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &root->fs_info->flags))
2947 return 0;
2948
2949 if (root == root->fs_info->extent_root)
2950 root = root->fs_info->tree_root;
2951
2952 delayed_refs = &trans->transaction->delayed_refs;
2953 if (count == 0)
2954 count = atomic_read(&delayed_refs->num_entries) * 2;
2955
2956 again:
2957 #ifdef SCRAMBLE_DELAYED_REFS
2958 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2959 #endif
2960 trans->can_flush_pending_bgs = false;
2961 ret = __btrfs_run_delayed_refs(trans, root, count);
2962 if (ret < 0) {
2963 btrfs_abort_transaction(trans, ret);
2964 return ret;
2965 }
2966
2967 if (run_all) {
2968 if (!list_empty(&trans->new_bgs))
2969 btrfs_create_pending_block_groups(trans, root);
2970
2971 spin_lock(&delayed_refs->lock);
2972 node = rb_first(&delayed_refs->href_root);
2973 if (!node) {
2974 spin_unlock(&delayed_refs->lock);
2975 goto out;
2976 }
2977 count = (unsigned long)-1;
2978
2979 while (node) {
2980 head = rb_entry(node, struct btrfs_delayed_ref_head,
2981 href_node);
2982 if (btrfs_delayed_ref_is_head(&head->node)) {
2983 struct btrfs_delayed_ref_node *ref;
2984
2985 ref = &head->node;
2986 atomic_inc(&ref->refs);
2987
2988 spin_unlock(&delayed_refs->lock);
2989 /*
2990 * Mutex was contended, block until it's
2991 * released and try again
2992 */
2993 mutex_lock(&head->mutex);
2994 mutex_unlock(&head->mutex);
2995
2996 btrfs_put_delayed_ref(ref);
2997 cond_resched();
2998 goto again;
2999 } else {
3000 WARN_ON(1);
3001 }
3002 node = rb_next(node);
3003 }
3004 spin_unlock(&delayed_refs->lock);
3005 cond_resched();
3006 goto again;
3007 }
3008 out:
3009 assert_qgroups_uptodate(trans);
3010 trans->can_flush_pending_bgs = can_flush_pending_bgs;
3011 return 0;
3012 }
3013
3014 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
3015 struct btrfs_root *root,
3016 u64 bytenr, u64 num_bytes, u64 flags,
3017 int level, int is_data)
3018 {
3019 struct btrfs_delayed_extent_op *extent_op;
3020 int ret;
3021
3022 extent_op = btrfs_alloc_delayed_extent_op();
3023 if (!extent_op)
3024 return -ENOMEM;
3025
3026 extent_op->flags_to_set = flags;
3027 extent_op->update_flags = true;
3028 extent_op->update_key = false;
3029 extent_op->is_data = is_data ? true : false;
3030 extent_op->level = level;
3031
3032 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
3033 num_bytes, extent_op);
3034 if (ret)
3035 btrfs_free_delayed_extent_op(extent_op);
3036 return ret;
3037 }
3038
3039 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
3040 struct btrfs_root *root,
3041 struct btrfs_path *path,
3042 u64 objectid, u64 offset, u64 bytenr)
3043 {
3044 struct btrfs_delayed_ref_head *head;
3045 struct btrfs_delayed_ref_node *ref;
3046 struct btrfs_delayed_data_ref *data_ref;
3047 struct btrfs_delayed_ref_root *delayed_refs;
3048 int ret = 0;
3049
3050 delayed_refs = &trans->transaction->delayed_refs;
3051 spin_lock(&delayed_refs->lock);
3052 head = btrfs_find_delayed_ref_head(trans, bytenr);
3053 if (!head) {
3054 spin_unlock(&delayed_refs->lock);
3055 return 0;
3056 }
3057
3058 if (!mutex_trylock(&head->mutex)) {
3059 atomic_inc(&head->node.refs);
3060 spin_unlock(&delayed_refs->lock);
3061
3062 btrfs_release_path(path);
3063
3064 /*
3065 * Mutex was contended, block until it's released and let
3066 * caller try again
3067 */
3068 mutex_lock(&head->mutex);
3069 mutex_unlock(&head->mutex);
3070 btrfs_put_delayed_ref(&head->node);
3071 return -EAGAIN;
3072 }
3073 spin_unlock(&delayed_refs->lock);
3074
3075 spin_lock(&head->lock);
3076 list_for_each_entry(ref, &head->ref_list, list) {
3077 /* If it's a shared ref we know a cross reference exists */
3078 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3079 ret = 1;
3080 break;
3081 }
3082
3083 data_ref = btrfs_delayed_node_to_data_ref(ref);
3084
3085 /*
3086 * If our ref doesn't match the one we're currently looking at
3087 * then we have a cross reference.
3088 */
3089 if (data_ref->root != root->root_key.objectid ||
3090 data_ref->objectid != objectid ||
3091 data_ref->offset != offset) {
3092 ret = 1;
3093 break;
3094 }
3095 }
3096 spin_unlock(&head->lock);
3097 mutex_unlock(&head->mutex);
3098 return ret;
3099 }
3100
3101 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
3102 struct btrfs_root *root,
3103 struct btrfs_path *path,
3104 u64 objectid, u64 offset, u64 bytenr)
3105 {
3106 struct btrfs_root *extent_root = root->fs_info->extent_root;
3107 struct extent_buffer *leaf;
3108 struct btrfs_extent_data_ref *ref;
3109 struct btrfs_extent_inline_ref *iref;
3110 struct btrfs_extent_item *ei;
3111 struct btrfs_key key;
3112 u32 item_size;
3113 int ret;
3114
3115 key.objectid = bytenr;
3116 key.offset = (u64)-1;
3117 key.type = BTRFS_EXTENT_ITEM_KEY;
3118
3119 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3120 if (ret < 0)
3121 goto out;
3122 BUG_ON(ret == 0); /* Corruption */
3123
3124 ret = -ENOENT;
3125 if (path->slots[0] == 0)
3126 goto out;
3127
3128 path->slots[0]--;
3129 leaf = path->nodes[0];
3130 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3131
3132 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3133 goto out;
3134
3135 ret = 1;
3136 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3137 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3138 if (item_size < sizeof(*ei)) {
3139 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3140 goto out;
3141 }
3142 #endif
3143 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3144
3145 if (item_size != sizeof(*ei) +
3146 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3147 goto out;
3148
3149 if (btrfs_extent_generation(leaf, ei) <=
3150 btrfs_root_last_snapshot(&root->root_item))
3151 goto out;
3152
3153 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3154 if (btrfs_extent_inline_ref_type(leaf, iref) !=
3155 BTRFS_EXTENT_DATA_REF_KEY)
3156 goto out;
3157
3158 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3159 if (btrfs_extent_refs(leaf, ei) !=
3160 btrfs_extent_data_ref_count(leaf, ref) ||
3161 btrfs_extent_data_ref_root(leaf, ref) !=
3162 root->root_key.objectid ||
3163 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3164 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3165 goto out;
3166
3167 ret = 0;
3168 out:
3169 return ret;
3170 }
3171
3172 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
3173 struct btrfs_root *root,
3174 u64 objectid, u64 offset, u64 bytenr)
3175 {
3176 struct btrfs_path *path;
3177 int ret;
3178 int ret2;
3179
3180 path = btrfs_alloc_path();
3181 if (!path)
3182 return -ENOENT;
3183
3184 do {
3185 ret = check_committed_ref(trans, root, path, objectid,
3186 offset, bytenr);
3187 if (ret && ret != -ENOENT)
3188 goto out;
3189
3190 ret2 = check_delayed_ref(trans, root, path, objectid,
3191 offset, bytenr);
3192 } while (ret2 == -EAGAIN);
3193
3194 if (ret2 && ret2 != -ENOENT) {
3195 ret = ret2;
3196 goto out;
3197 }
3198
3199 if (ret != -ENOENT || ret2 != -ENOENT)
3200 ret = 0;
3201 out:
3202 btrfs_free_path(path);
3203 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3204 WARN_ON(ret > 0);
3205 return ret;
3206 }
3207
3208 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3209 struct btrfs_root *root,
3210 struct extent_buffer *buf,
3211 int full_backref, int inc)
3212 {
3213 u64 bytenr;
3214 u64 num_bytes;
3215 u64 parent;
3216 u64 ref_root;
3217 u32 nritems;
3218 struct btrfs_key key;
3219 struct btrfs_file_extent_item *fi;
3220 int i;
3221 int level;
3222 int ret = 0;
3223 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3224 u64, u64, u64, u64, u64, u64);
3225
3226
3227 if (btrfs_is_testing(root->fs_info))
3228 return 0;
3229
3230 ref_root = btrfs_header_owner(buf);
3231 nritems = btrfs_header_nritems(buf);
3232 level = btrfs_header_level(buf);
3233
3234 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3235 return 0;
3236
3237 if (inc)
3238 process_func = btrfs_inc_extent_ref;
3239 else
3240 process_func = btrfs_free_extent;
3241
3242 if (full_backref)
3243 parent = buf->start;
3244 else
3245 parent = 0;
3246
3247 for (i = 0; i < nritems; i++) {
3248 if (level == 0) {
3249 btrfs_item_key_to_cpu(buf, &key, i);
3250 if (key.type != BTRFS_EXTENT_DATA_KEY)
3251 continue;
3252 fi = btrfs_item_ptr(buf, i,
3253 struct btrfs_file_extent_item);
3254 if (btrfs_file_extent_type(buf, fi) ==
3255 BTRFS_FILE_EXTENT_INLINE)
3256 continue;
3257 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3258 if (bytenr == 0)
3259 continue;
3260
3261 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3262 key.offset -= btrfs_file_extent_offset(buf, fi);
3263 ret = process_func(trans, root, bytenr, num_bytes,
3264 parent, ref_root, key.objectid,
3265 key.offset);
3266 if (ret)
3267 goto fail;
3268 } else {
3269 bytenr = btrfs_node_blockptr(buf, i);
3270 num_bytes = root->nodesize;
3271 ret = process_func(trans, root, bytenr, num_bytes,
3272 parent, ref_root, level - 1, 0);
3273 if (ret)
3274 goto fail;
3275 }
3276 }
3277 return 0;
3278 fail:
3279 return ret;
3280 }
3281
3282 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3283 struct extent_buffer *buf, int full_backref)
3284 {
3285 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3286 }
3287
3288 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3289 struct extent_buffer *buf, int full_backref)
3290 {
3291 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3292 }
3293
3294 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3295 struct btrfs_root *root,
3296 struct btrfs_path *path,
3297 struct btrfs_block_group_cache *cache)
3298 {
3299 int ret;
3300 struct btrfs_root *extent_root = root->fs_info->extent_root;
3301 unsigned long bi;
3302 struct extent_buffer *leaf;
3303
3304 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3305 if (ret) {
3306 if (ret > 0)
3307 ret = -ENOENT;
3308 goto fail;
3309 }
3310
3311 leaf = path->nodes[0];
3312 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3313 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3314 btrfs_mark_buffer_dirty(leaf);
3315 fail:
3316 btrfs_release_path(path);
3317 return ret;
3318
3319 }
3320
3321 static struct btrfs_block_group_cache *
3322 next_block_group(struct btrfs_root *root,
3323 struct btrfs_block_group_cache *cache)
3324 {
3325 struct rb_node *node;
3326
3327 spin_lock(&root->fs_info->block_group_cache_lock);
3328
3329 /* If our block group was removed, we need a full search. */
3330 if (RB_EMPTY_NODE(&cache->cache_node)) {
3331 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3332
3333 spin_unlock(&root->fs_info->block_group_cache_lock);
3334 btrfs_put_block_group(cache);
3335 cache = btrfs_lookup_first_block_group(root->fs_info,
3336 next_bytenr);
3337 return cache;
3338 }
3339 node = rb_next(&cache->cache_node);
3340 btrfs_put_block_group(cache);
3341 if (node) {
3342 cache = rb_entry(node, struct btrfs_block_group_cache,
3343 cache_node);
3344 btrfs_get_block_group(cache);
3345 } else
3346 cache = NULL;
3347 spin_unlock(&root->fs_info->block_group_cache_lock);
3348 return cache;
3349 }
3350
3351 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3352 struct btrfs_trans_handle *trans,
3353 struct btrfs_path *path)
3354 {
3355 struct btrfs_root *root = block_group->fs_info->tree_root;
3356 struct inode *inode = NULL;
3357 u64 alloc_hint = 0;
3358 int dcs = BTRFS_DC_ERROR;
3359 u64 num_pages = 0;
3360 int retries = 0;
3361 int ret = 0;
3362
3363 /*
3364 * If this block group is smaller than 100 megs don't bother caching the
3365 * block group.
3366 */
3367 if (block_group->key.offset < (100 * SZ_1M)) {
3368 spin_lock(&block_group->lock);
3369 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3370 spin_unlock(&block_group->lock);
3371 return 0;
3372 }
3373
3374 if (trans->aborted)
3375 return 0;
3376 again:
3377 inode = lookup_free_space_inode(root, block_group, path);
3378 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3379 ret = PTR_ERR(inode);
3380 btrfs_release_path(path);
3381 goto out;
3382 }
3383
3384 if (IS_ERR(inode)) {
3385 BUG_ON(retries);
3386 retries++;
3387
3388 if (block_group->ro)
3389 goto out_free;
3390
3391 ret = create_free_space_inode(root, trans, block_group, path);
3392 if (ret)
3393 goto out_free;
3394 goto again;
3395 }
3396
3397 /* We've already setup this transaction, go ahead and exit */
3398 if (block_group->cache_generation == trans->transid &&
3399 i_size_read(inode)) {
3400 dcs = BTRFS_DC_SETUP;
3401 goto out_put;
3402 }
3403
3404 /*
3405 * We want to set the generation to 0, that way if anything goes wrong
3406 * from here on out we know not to trust this cache when we load up next
3407 * time.
3408 */
3409 BTRFS_I(inode)->generation = 0;
3410 ret = btrfs_update_inode(trans, root, inode);
3411 if (ret) {
3412 /*
3413 * So theoretically we could recover from this, simply set the
3414 * super cache generation to 0 so we know to invalidate the
3415 * cache, but then we'd have to keep track of the block groups
3416 * that fail this way so we know we _have_ to reset this cache
3417 * before the next commit or risk reading stale cache. So to
3418 * limit our exposure to horrible edge cases lets just abort the
3419 * transaction, this only happens in really bad situations
3420 * anyway.
3421 */
3422 btrfs_abort_transaction(trans, ret);
3423 goto out_put;
3424 }
3425 WARN_ON(ret);
3426
3427 if (i_size_read(inode) > 0) {
3428 ret = btrfs_check_trunc_cache_free_space(root,
3429 &root->fs_info->global_block_rsv);
3430 if (ret)
3431 goto out_put;
3432
3433 ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode);
3434 if (ret)
3435 goto out_put;
3436 }
3437
3438 spin_lock(&block_group->lock);
3439 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3440 !btrfs_test_opt(root->fs_info, SPACE_CACHE)) {
3441 /*
3442 * don't bother trying to write stuff out _if_
3443 * a) we're not cached,
3444 * b) we're with nospace_cache mount option.
3445 */
3446 dcs = BTRFS_DC_WRITTEN;
3447 spin_unlock(&block_group->lock);
3448 goto out_put;
3449 }
3450 spin_unlock(&block_group->lock);
3451
3452 /*
3453 * We hit an ENOSPC when setting up the cache in this transaction, just
3454 * skip doing the setup, we've already cleared the cache so we're safe.
3455 */
3456 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3457 ret = -ENOSPC;
3458 goto out_put;
3459 }
3460
3461 /*
3462 * Try to preallocate enough space based on how big the block group is.
3463 * Keep in mind this has to include any pinned space which could end up
3464 * taking up quite a bit since it's not folded into the other space
3465 * cache.
3466 */
3467 num_pages = div_u64(block_group->key.offset, SZ_256M);
3468 if (!num_pages)
3469 num_pages = 1;
3470
3471 num_pages *= 16;
3472 num_pages *= PAGE_SIZE;
3473
3474 ret = btrfs_check_data_free_space(inode, 0, num_pages);
3475 if (ret)
3476 goto out_put;
3477
3478 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3479 num_pages, num_pages,
3480 &alloc_hint);
3481 /*
3482 * Our cache requires contiguous chunks so that we don't modify a bunch
3483 * of metadata or split extents when writing the cache out, which means
3484 * we can enospc if we are heavily fragmented in addition to just normal
3485 * out of space conditions. So if we hit this just skip setting up any
3486 * other block groups for this transaction, maybe we'll unpin enough
3487 * space the next time around.
3488 */
3489 if (!ret)
3490 dcs = BTRFS_DC_SETUP;
3491 else if (ret == -ENOSPC)
3492 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3493
3494 out_put:
3495 iput(inode);
3496 out_free:
3497 btrfs_release_path(path);
3498 out:
3499 spin_lock(&block_group->lock);
3500 if (!ret && dcs == BTRFS_DC_SETUP)
3501 block_group->cache_generation = trans->transid;
3502 block_group->disk_cache_state = dcs;
3503 spin_unlock(&block_group->lock);
3504
3505 return ret;
3506 }
3507
3508 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3509 struct btrfs_root *root)
3510 {
3511 struct btrfs_block_group_cache *cache, *tmp;
3512 struct btrfs_transaction *cur_trans = trans->transaction;
3513 struct btrfs_path *path;
3514
3515 if (list_empty(&cur_trans->dirty_bgs) ||
3516 !btrfs_test_opt(root->fs_info, SPACE_CACHE))
3517 return 0;
3518
3519 path = btrfs_alloc_path();
3520 if (!path)
3521 return -ENOMEM;
3522
3523 /* Could add new block groups, use _safe just in case */
3524 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3525 dirty_list) {
3526 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3527 cache_save_setup(cache, trans, path);
3528 }
3529
3530 btrfs_free_path(path);
3531 return 0;
3532 }
3533
3534 /*
3535 * transaction commit does final block group cache writeback during a
3536 * critical section where nothing is allowed to change the FS. This is
3537 * required in order for the cache to actually match the block group,
3538 * but can introduce a lot of latency into the commit.
3539 *
3540 * So, btrfs_start_dirty_block_groups is here to kick off block group
3541 * cache IO. There's a chance we'll have to redo some of it if the
3542 * block group changes again during the commit, but it greatly reduces
3543 * the commit latency by getting rid of the easy block groups while
3544 * we're still allowing others to join the commit.
3545 */
3546 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3547 struct btrfs_root *root)
3548 {
3549 struct btrfs_block_group_cache *cache;
3550 struct btrfs_transaction *cur_trans = trans->transaction;
3551 int ret = 0;
3552 int should_put;
3553 struct btrfs_path *path = NULL;
3554 LIST_HEAD(dirty);
3555 struct list_head *io = &cur_trans->io_bgs;
3556 int num_started = 0;
3557 int loops = 0;
3558
3559 spin_lock(&cur_trans->dirty_bgs_lock);
3560 if (list_empty(&cur_trans->dirty_bgs)) {
3561 spin_unlock(&cur_trans->dirty_bgs_lock);
3562 return 0;
3563 }
3564 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3565 spin_unlock(&cur_trans->dirty_bgs_lock);
3566
3567 again:
3568 /*
3569 * make sure all the block groups on our dirty list actually
3570 * exist
3571 */
3572 btrfs_create_pending_block_groups(trans, root);
3573
3574 if (!path) {
3575 path = btrfs_alloc_path();
3576 if (!path)
3577 return -ENOMEM;
3578 }
3579
3580 /*
3581 * cache_write_mutex is here only to save us from balance or automatic
3582 * removal of empty block groups deleting this block group while we are
3583 * writing out the cache
3584 */
3585 mutex_lock(&trans->transaction->cache_write_mutex);
3586 while (!list_empty(&dirty)) {
3587 cache = list_first_entry(&dirty,
3588 struct btrfs_block_group_cache,
3589 dirty_list);
3590 /*
3591 * this can happen if something re-dirties a block
3592 * group that is already under IO. Just wait for it to
3593 * finish and then do it all again
3594 */
3595 if (!list_empty(&cache->io_list)) {
3596 list_del_init(&cache->io_list);
3597 btrfs_wait_cache_io(root, trans, cache,
3598 &cache->io_ctl, path,
3599 cache->key.objectid);
3600 btrfs_put_block_group(cache);
3601 }
3602
3603
3604 /*
3605 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3606 * if it should update the cache_state. Don't delete
3607 * until after we wait.
3608 *
3609 * Since we're not running in the commit critical section
3610 * we need the dirty_bgs_lock to protect from update_block_group
3611 */
3612 spin_lock(&cur_trans->dirty_bgs_lock);
3613 list_del_init(&cache->dirty_list);
3614 spin_unlock(&cur_trans->dirty_bgs_lock);
3615
3616 should_put = 1;
3617
3618 cache_save_setup(cache, trans, path);
3619
3620 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3621 cache->io_ctl.inode = NULL;
3622 ret = btrfs_write_out_cache(root, trans, cache, path);
3623 if (ret == 0 && cache->io_ctl.inode) {
3624 num_started++;
3625 should_put = 0;
3626
3627 /*
3628 * the cache_write_mutex is protecting
3629 * the io_list
3630 */
3631 list_add_tail(&cache->io_list, io);
3632 } else {
3633 /*
3634 * if we failed to write the cache, the
3635 * generation will be bad and life goes on
3636 */
3637 ret = 0;
3638 }
3639 }
3640 if (!ret) {
3641 ret = write_one_cache_group(trans, root, path, cache);
3642 /*
3643 * Our block group might still be attached to the list
3644 * of new block groups in the transaction handle of some
3645 * other task (struct btrfs_trans_handle->new_bgs). This
3646 * means its block group item isn't yet in the extent
3647 * tree. If this happens ignore the error, as we will
3648 * try again later in the critical section of the
3649 * transaction commit.
3650 */
3651 if (ret == -ENOENT) {
3652 ret = 0;
3653 spin_lock(&cur_trans->dirty_bgs_lock);
3654 if (list_empty(&cache->dirty_list)) {
3655 list_add_tail(&cache->dirty_list,
3656 &cur_trans->dirty_bgs);
3657 btrfs_get_block_group(cache);
3658 }
3659 spin_unlock(&cur_trans->dirty_bgs_lock);
3660 } else if (ret) {
3661 btrfs_abort_transaction(trans, ret);
3662 }
3663 }
3664
3665 /* if its not on the io list, we need to put the block group */
3666 if (should_put)
3667 btrfs_put_block_group(cache);
3668
3669 if (ret)
3670 break;
3671
3672 /*
3673 * Avoid blocking other tasks for too long. It might even save
3674 * us from writing caches for block groups that are going to be
3675 * removed.
3676 */
3677 mutex_unlock(&trans->transaction->cache_write_mutex);
3678 mutex_lock(&trans->transaction->cache_write_mutex);
3679 }
3680 mutex_unlock(&trans->transaction->cache_write_mutex);
3681
3682 /*
3683 * go through delayed refs for all the stuff we've just kicked off
3684 * and then loop back (just once)
3685 */
3686 ret = btrfs_run_delayed_refs(trans, root, 0);
3687 if (!ret && loops == 0) {
3688 loops++;
3689 spin_lock(&cur_trans->dirty_bgs_lock);
3690 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3691 /*
3692 * dirty_bgs_lock protects us from concurrent block group
3693 * deletes too (not just cache_write_mutex).
3694 */
3695 if (!list_empty(&dirty)) {
3696 spin_unlock(&cur_trans->dirty_bgs_lock);
3697 goto again;
3698 }
3699 spin_unlock(&cur_trans->dirty_bgs_lock);
3700 } else if (ret < 0) {
3701 btrfs_cleanup_dirty_bgs(cur_trans, root);
3702 }
3703
3704 btrfs_free_path(path);
3705 return ret;
3706 }
3707
3708 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3709 struct btrfs_root *root)
3710 {
3711 struct btrfs_block_group_cache *cache;
3712 struct btrfs_transaction *cur_trans = trans->transaction;
3713 int ret = 0;
3714 int should_put;
3715 struct btrfs_path *path;
3716 struct list_head *io = &cur_trans->io_bgs;
3717 int num_started = 0;
3718
3719 path = btrfs_alloc_path();
3720 if (!path)
3721 return -ENOMEM;
3722
3723 /*
3724 * Even though we are in the critical section of the transaction commit,
3725 * we can still have concurrent tasks adding elements to this
3726 * transaction's list of dirty block groups. These tasks correspond to
3727 * endio free space workers started when writeback finishes for a
3728 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3729 * allocate new block groups as a result of COWing nodes of the root
3730 * tree when updating the free space inode. The writeback for the space
3731 * caches is triggered by an earlier call to
3732 * btrfs_start_dirty_block_groups() and iterations of the following
3733 * loop.
3734 * Also we want to do the cache_save_setup first and then run the
3735 * delayed refs to make sure we have the best chance at doing this all
3736 * in one shot.
3737 */
3738 spin_lock(&cur_trans->dirty_bgs_lock);
3739 while (!list_empty(&cur_trans->dirty_bgs)) {
3740 cache = list_first_entry(&cur_trans->dirty_bgs,
3741 struct btrfs_block_group_cache,
3742 dirty_list);
3743
3744 /*
3745 * this can happen if cache_save_setup re-dirties a block
3746 * group that is already under IO. Just wait for it to
3747 * finish and then do it all again
3748 */
3749 if (!list_empty(&cache->io_list)) {
3750 spin_unlock(&cur_trans->dirty_bgs_lock);
3751 list_del_init(&cache->io_list);
3752 btrfs_wait_cache_io(root, trans, cache,
3753 &cache->io_ctl, path,
3754 cache->key.objectid);
3755 btrfs_put_block_group(cache);
3756 spin_lock(&cur_trans->dirty_bgs_lock);
3757 }
3758
3759 /*
3760 * don't remove from the dirty list until after we've waited
3761 * on any pending IO
3762 */
3763 list_del_init(&cache->dirty_list);
3764 spin_unlock(&cur_trans->dirty_bgs_lock);
3765 should_put = 1;
3766
3767 cache_save_setup(cache, trans, path);
3768
3769 if (!ret)
3770 ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1);
3771
3772 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3773 cache->io_ctl.inode = NULL;
3774 ret = btrfs_write_out_cache(root, trans, cache, path);
3775 if (ret == 0 && cache->io_ctl.inode) {
3776 num_started++;
3777 should_put = 0;
3778 list_add_tail(&cache->io_list, io);
3779 } else {
3780 /*
3781 * if we failed to write the cache, the
3782 * generation will be bad and life goes on
3783 */
3784 ret = 0;
3785 }
3786 }
3787 if (!ret) {
3788 ret = write_one_cache_group(trans, root, path, cache);
3789 /*
3790 * One of the free space endio workers might have
3791 * created a new block group while updating a free space
3792 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3793 * and hasn't released its transaction handle yet, in
3794 * which case the new block group is still attached to
3795 * its transaction handle and its creation has not
3796 * finished yet (no block group item in the extent tree
3797 * yet, etc). If this is the case, wait for all free
3798 * space endio workers to finish and retry. This is a
3799 * a very rare case so no need for a more efficient and
3800 * complex approach.
3801 */
3802 if (ret == -ENOENT) {
3803 wait_event(cur_trans->writer_wait,
3804 atomic_read(&cur_trans->num_writers) == 1);
3805 ret = write_one_cache_group(trans, root, path,
3806 cache);
3807 }
3808 if (ret)
3809 btrfs_abort_transaction(trans, ret);
3810 }
3811
3812 /* if its not on the io list, we need to put the block group */
3813 if (should_put)
3814 btrfs_put_block_group(cache);
3815 spin_lock(&cur_trans->dirty_bgs_lock);
3816 }
3817 spin_unlock(&cur_trans->dirty_bgs_lock);
3818
3819 while (!list_empty(io)) {
3820 cache = list_first_entry(io, struct btrfs_block_group_cache,
3821 io_list);
3822 list_del_init(&cache->io_list);
3823 btrfs_wait_cache_io(root, trans, cache,
3824 &cache->io_ctl, path, cache->key.objectid);
3825 btrfs_put_block_group(cache);
3826 }
3827
3828 btrfs_free_path(path);
3829 return ret;
3830 }
3831
3832 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3833 {
3834 struct btrfs_block_group_cache *block_group;
3835 int readonly = 0;
3836
3837 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3838 if (!block_group || block_group->ro)
3839 readonly = 1;
3840 if (block_group)
3841 btrfs_put_block_group(block_group);
3842 return readonly;
3843 }
3844
3845 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3846 {
3847 struct btrfs_block_group_cache *bg;
3848 bool ret = true;
3849
3850 bg = btrfs_lookup_block_group(fs_info, bytenr);
3851 if (!bg)
3852 return false;
3853
3854 spin_lock(&bg->lock);
3855 if (bg->ro)
3856 ret = false;
3857 else
3858 atomic_inc(&bg->nocow_writers);
3859 spin_unlock(&bg->lock);
3860
3861 /* no put on block group, done by btrfs_dec_nocow_writers */
3862 if (!ret)
3863 btrfs_put_block_group(bg);
3864
3865 return ret;
3866
3867 }
3868
3869 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3870 {
3871 struct btrfs_block_group_cache *bg;
3872
3873 bg = btrfs_lookup_block_group(fs_info, bytenr);
3874 ASSERT(bg);
3875 if (atomic_dec_and_test(&bg->nocow_writers))
3876 wake_up_atomic_t(&bg->nocow_writers);
3877 /*
3878 * Once for our lookup and once for the lookup done by a previous call
3879 * to btrfs_inc_nocow_writers()
3880 */
3881 btrfs_put_block_group(bg);
3882 btrfs_put_block_group(bg);
3883 }
3884
3885 static int btrfs_wait_nocow_writers_atomic_t(atomic_t *a)
3886 {
3887 schedule();
3888 return 0;
3889 }
3890
3891 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
3892 {
3893 wait_on_atomic_t(&bg->nocow_writers,
3894 btrfs_wait_nocow_writers_atomic_t,
3895 TASK_UNINTERRUPTIBLE);
3896 }
3897
3898 static const char *alloc_name(u64 flags)
3899 {
3900 switch (flags) {
3901 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3902 return "mixed";
3903 case BTRFS_BLOCK_GROUP_METADATA:
3904 return "metadata";
3905 case BTRFS_BLOCK_GROUP_DATA:
3906 return "data";
3907 case BTRFS_BLOCK_GROUP_SYSTEM:
3908 return "system";
3909 default:
3910 WARN_ON(1);
3911 return "invalid-combination";
3912 };
3913 }
3914
3915 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3916 u64 total_bytes, u64 bytes_used,
3917 u64 bytes_readonly,
3918 struct btrfs_space_info **space_info)
3919 {
3920 struct btrfs_space_info *found;
3921 int i;
3922 int factor;
3923 int ret;
3924
3925 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3926 BTRFS_BLOCK_GROUP_RAID10))
3927 factor = 2;
3928 else
3929 factor = 1;
3930
3931 found = __find_space_info(info, flags);
3932 if (found) {
3933 spin_lock(&found->lock);
3934 found->total_bytes += total_bytes;
3935 found->disk_total += total_bytes * factor;
3936 found->bytes_used += bytes_used;
3937 found->disk_used += bytes_used * factor;
3938 found->bytes_readonly += bytes_readonly;
3939 if (total_bytes > 0)
3940 found->full = 0;
3941 space_info_add_new_bytes(info, found, total_bytes -
3942 bytes_used - bytes_readonly);
3943 spin_unlock(&found->lock);
3944 *space_info = found;
3945 return 0;
3946 }
3947 found = kzalloc(sizeof(*found), GFP_NOFS);
3948 if (!found)
3949 return -ENOMEM;
3950
3951 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3952 if (ret) {
3953 kfree(found);
3954 return ret;
3955 }
3956
3957 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3958 INIT_LIST_HEAD(&found->block_groups[i]);
3959 init_rwsem(&found->groups_sem);
3960 spin_lock_init(&found->lock);
3961 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3962 found->total_bytes = total_bytes;
3963 found->disk_total = total_bytes * factor;
3964 found->bytes_used = bytes_used;
3965 found->disk_used = bytes_used * factor;
3966 found->bytes_pinned = 0;
3967 found->bytes_reserved = 0;
3968 found->bytes_readonly = bytes_readonly;
3969 found->bytes_may_use = 0;
3970 found->full = 0;
3971 found->max_extent_size = 0;
3972 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3973 found->chunk_alloc = 0;
3974 found->flush = 0;
3975 init_waitqueue_head(&found->wait);
3976 INIT_LIST_HEAD(&found->ro_bgs);
3977 INIT_LIST_HEAD(&found->tickets);
3978 INIT_LIST_HEAD(&found->priority_tickets);
3979
3980 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3981 info->space_info_kobj, "%s",
3982 alloc_name(found->flags));
3983 if (ret) {
3984 kfree(found);
3985 return ret;
3986 }
3987
3988 *space_info = found;
3989 list_add_rcu(&found->list, &info->space_info);
3990 if (flags & BTRFS_BLOCK_GROUP_DATA)
3991 info->data_sinfo = found;
3992
3993 return ret;
3994 }
3995
3996 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3997 {
3998 u64 extra_flags = chunk_to_extended(flags) &
3999 BTRFS_EXTENDED_PROFILE_MASK;
4000
4001 write_seqlock(&fs_info->profiles_lock);
4002 if (flags & BTRFS_BLOCK_GROUP_DATA)
4003 fs_info->avail_data_alloc_bits |= extra_flags;
4004 if (flags & BTRFS_BLOCK_GROUP_METADATA)
4005 fs_info->avail_metadata_alloc_bits |= extra_flags;
4006 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4007 fs_info->avail_system_alloc_bits |= extra_flags;
4008 write_sequnlock(&fs_info->profiles_lock);
4009 }
4010
4011 /*
4012 * returns target flags in extended format or 0 if restripe for this
4013 * chunk_type is not in progress
4014 *
4015 * should be called with either volume_mutex or balance_lock held
4016 */
4017 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
4018 {
4019 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4020 u64 target = 0;
4021
4022 if (!bctl)
4023 return 0;
4024
4025 if (flags & BTRFS_BLOCK_GROUP_DATA &&
4026 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4027 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
4028 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
4029 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4030 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
4031 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
4032 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4033 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
4034 }
4035
4036 return target;
4037 }
4038
4039 /*
4040 * @flags: available profiles in extended format (see ctree.h)
4041 *
4042 * Returns reduced profile in chunk format. If profile changing is in
4043 * progress (either running or paused) picks the target profile (if it's
4044 * already available), otherwise falls back to plain reducing.
4045 */
4046 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
4047 {
4048 u64 num_devices = root->fs_info->fs_devices->rw_devices;
4049 u64 target;
4050 u64 raid_type;
4051 u64 allowed = 0;
4052
4053 /*
4054 * see if restripe for this chunk_type is in progress, if so
4055 * try to reduce to the target profile
4056 */
4057 spin_lock(&root->fs_info->balance_lock);
4058 target = get_restripe_target(root->fs_info, flags);
4059 if (target) {
4060 /* pick target profile only if it's already available */
4061 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
4062 spin_unlock(&root->fs_info->balance_lock);
4063 return extended_to_chunk(target);
4064 }
4065 }
4066 spin_unlock(&root->fs_info->balance_lock);
4067
4068 /* First, mask out the RAID levels which aren't possible */
4069 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4070 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
4071 allowed |= btrfs_raid_group[raid_type];
4072 }
4073 allowed &= flags;
4074
4075 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
4076 allowed = BTRFS_BLOCK_GROUP_RAID6;
4077 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4078 allowed = BTRFS_BLOCK_GROUP_RAID5;
4079 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4080 allowed = BTRFS_BLOCK_GROUP_RAID10;
4081 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4082 allowed = BTRFS_BLOCK_GROUP_RAID1;
4083 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4084 allowed = BTRFS_BLOCK_GROUP_RAID0;
4085
4086 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4087
4088 return extended_to_chunk(flags | allowed);
4089 }
4090
4091 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
4092 {
4093 unsigned seq;
4094 u64 flags;
4095
4096 do {
4097 flags = orig_flags;
4098 seq = read_seqbegin(&root->fs_info->profiles_lock);
4099
4100 if (flags & BTRFS_BLOCK_GROUP_DATA)
4101 flags |= root->fs_info->avail_data_alloc_bits;
4102 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4103 flags |= root->fs_info->avail_system_alloc_bits;
4104 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4105 flags |= root->fs_info->avail_metadata_alloc_bits;
4106 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
4107
4108 return btrfs_reduce_alloc_profile(root, flags);
4109 }
4110
4111 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
4112 {
4113 u64 flags;
4114 u64 ret;
4115
4116 if (data)
4117 flags = BTRFS_BLOCK_GROUP_DATA;
4118 else if (root == root->fs_info->chunk_root)
4119 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4120 else
4121 flags = BTRFS_BLOCK_GROUP_METADATA;
4122
4123 ret = get_alloc_profile(root, flags);
4124 return ret;
4125 }
4126
4127 int btrfs_alloc_data_chunk_ondemand(struct inode *inode, u64 bytes)
4128 {
4129 struct btrfs_space_info *data_sinfo;
4130 struct btrfs_root *root = BTRFS_I(inode)->root;
4131 struct btrfs_fs_info *fs_info = root->fs_info;
4132 u64 used;
4133 int ret = 0;
4134 int need_commit = 2;
4135 int have_pinned_space;
4136 int have_bg_delete_sem = 0;
4137 bool free_space_inode = btrfs_is_free_space_inode(inode);
4138
4139 /* make sure bytes are sectorsize aligned */
4140 bytes = ALIGN(bytes, root->sectorsize);
4141
4142 if (free_space_inode) {
4143 need_commit = 0;
4144 ASSERT(current->journal_info);
4145 }
4146
4147 /*
4148 * Here we shouldn't call down_read(bg_delete_sem) for free space inode,
4149 * there is lock order between bg_delete_sem and "wait current trans
4150 * finished". Meanwhile because we only do the data space reservation
4151 * for free space cache in the transaction context,
4152 * btrfs_delete_unused_bgs() will either have finished its job, or start
4153 * a new transaction waiting current transaction to complete, there will
4154 * be no unused block groups to be deleted, so it's safe to not call
4155 * down_read(bg_delete_sem).
4156 */
4157 data_sinfo = fs_info->data_sinfo;
4158 if (!data_sinfo) {
4159 if (!free_space_inode) {
4160 down_read(&root->fs_info->bg_delete_sem);
4161 have_bg_delete_sem = 1;
4162 }
4163 goto alloc;
4164 }
4165
4166 again:
4167 /* make sure we have enough space to handle the data first */
4168 spin_lock(&data_sinfo->lock);
4169 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
4170 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
4171 data_sinfo->bytes_may_use;
4172
4173 if (used + bytes > data_sinfo->total_bytes) {
4174 struct btrfs_trans_handle *trans;
4175
4176 /*
4177 * We may need to allocate new chunk, so we should block
4178 * btrfs_delete_unused_bgs()
4179 */
4180 if (!have_bg_delete_sem && !free_space_inode) {
4181 spin_unlock(&data_sinfo->lock);
4182 down_read(&root->fs_info->bg_delete_sem);
4183 have_bg_delete_sem = 1;
4184 goto again;
4185 }
4186
4187 /*
4188 * if we don't have enough free bytes in this space then we need
4189 * to alloc a new chunk.
4190 */
4191 if (!data_sinfo->full) {
4192 u64 alloc_target;
4193
4194 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4195 spin_unlock(&data_sinfo->lock);
4196 alloc:
4197 alloc_target = btrfs_get_alloc_profile(root, 1);
4198 /*
4199 * It is ugly that we don't call nolock join
4200 * transaction for the free space inode case here.
4201 * But it is safe because we only do the data space
4202 * reservation for the free space cache in the
4203 * transaction context, the common join transaction
4204 * just increase the counter of the current transaction
4205 * handler, doesn't try to acquire the trans_lock of
4206 * the fs.
4207 */
4208 trans = btrfs_join_transaction(root);
4209 if (IS_ERR(trans)) {
4210 ret = PTR_ERR(trans);
4211 goto out;
4212 }
4213
4214 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4215 alloc_target,
4216 CHUNK_ALLOC_NO_FORCE);
4217 btrfs_end_transaction(trans, root);
4218 if (ret < 0) {
4219 if (ret != -ENOSPC)
4220 goto out;
4221 else {
4222 have_pinned_space = 1;
4223 goto commit_trans;
4224 }
4225 }
4226
4227 if (!data_sinfo)
4228 data_sinfo = fs_info->data_sinfo;
4229
4230 goto again;
4231 }
4232
4233 /*
4234 * If we don't have enough pinned space to deal with this
4235 * allocation, and no removed chunk in current transaction,
4236 * don't bother committing the transaction.
4237 */
4238 have_pinned_space = percpu_counter_compare(
4239 &data_sinfo->total_bytes_pinned,
4240 used + bytes - data_sinfo->total_bytes);
4241 spin_unlock(&data_sinfo->lock);
4242
4243 /* commit the current transaction and try again */
4244 commit_trans:
4245 if (need_commit &&
4246 !atomic_read(&root->fs_info->open_ioctl_trans)) {
4247 need_commit--;
4248
4249 if (need_commit > 0) {
4250 btrfs_start_delalloc_roots(fs_info, 0, -1);
4251 btrfs_wait_ordered_roots(fs_info, -1, 0, (u64)-1);
4252 }
4253
4254 trans = btrfs_join_transaction(root);
4255 if (IS_ERR(trans)) {
4256 ret = PTR_ERR(trans);
4257 goto out;
4258 }
4259 if (have_pinned_space >= 0 ||
4260 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4261 &trans->transaction->flags) ||
4262 need_commit > 0) {
4263 ret = btrfs_commit_transaction(trans, root);
4264 if (ret)
4265 goto out;
4266 /*
4267 * The cleaner kthread might still be doing iput
4268 * operations. Wait for it to finish so that
4269 * more space is released.
4270 */
4271 mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex);
4272 mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex);
4273 goto again;
4274 } else {
4275 btrfs_end_transaction(trans, root);
4276 }
4277 }
4278
4279 trace_btrfs_space_reservation(root->fs_info,
4280 "space_info:enospc",
4281 data_sinfo->flags, bytes, 1);
4282 ret = -ENOSPC;
4283 goto out;
4284 }
4285 data_sinfo->bytes_may_use += bytes;
4286 trace_btrfs_space_reservation(root->fs_info, "space_info",
4287 data_sinfo->flags, bytes, 1);
4288 spin_unlock(&data_sinfo->lock);
4289
4290 out:
4291 if (have_bg_delete_sem && !free_space_inode)
4292 up_read(&root->fs_info->bg_delete_sem);
4293
4294 return ret;
4295 }
4296
4297 /*
4298 * New check_data_free_space() with ability for precious data reservation
4299 * Will replace old btrfs_check_data_free_space(), but for patch split,
4300 * add a new function first and then replace it.
4301 */
4302 int btrfs_check_data_free_space(struct inode *inode, u64 start, u64 len)
4303 {
4304 struct btrfs_root *root = BTRFS_I(inode)->root;
4305 int ret;
4306
4307 /* align the range */
4308 len = round_up(start + len, root->sectorsize) -
4309 round_down(start, root->sectorsize);
4310 start = round_down(start, root->sectorsize);
4311
4312 ret = btrfs_alloc_data_chunk_ondemand(inode, len);
4313 if (ret < 0)
4314 return ret;
4315
4316 /*
4317 * Use new btrfs_qgroup_reserve_data to reserve precious data space
4318 *
4319 * TODO: Find a good method to avoid reserve data space for NOCOW
4320 * range, but don't impact performance on quota disable case.
4321 */
4322 ret = btrfs_qgroup_reserve_data(inode, start, len);
4323 if (ret < 0)
4324 /* Qgroup reserve failed, need to cleanup reserved data space */
4325 btrfs_free_reserved_data_space(inode, start, len);
4326 return ret;
4327 }
4328
4329 /*
4330 * Called if we need to clear a data reservation for this inode
4331 * Normally in a error case.
4332 *
4333 * This one will *NOT* use accurate qgroup reserved space API, just for case
4334 * which we can't sleep and is sure it won't affect qgroup reserved space.
4335 * Like clear_bit_hook().
4336 */
4337 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4338 u64 len)
4339 {
4340 struct btrfs_root *root = BTRFS_I(inode)->root;
4341 struct btrfs_space_info *data_sinfo;
4342
4343 /* Make sure the range is aligned to sectorsize */
4344 len = round_up(start + len, root->sectorsize) -
4345 round_down(start, root->sectorsize);
4346 start = round_down(start, root->sectorsize);
4347
4348 data_sinfo = root->fs_info->data_sinfo;
4349 spin_lock(&data_sinfo->lock);
4350 if (WARN_ON(data_sinfo->bytes_may_use < len))
4351 data_sinfo->bytes_may_use = 0;
4352 else
4353 data_sinfo->bytes_may_use -= len;
4354 trace_btrfs_space_reservation(root->fs_info, "space_info",
4355 data_sinfo->flags, len, 0);
4356 spin_unlock(&data_sinfo->lock);
4357 }
4358
4359 /*
4360 * Called if we need to clear a data reservation for this inode
4361 * Normally in a error case.
4362 *
4363 * This one will handle the per-inode data rsv map for accurate reserved
4364 * space framework.
4365 */
4366 void btrfs_free_reserved_data_space(struct inode *inode, u64 start, u64 len)
4367 {
4368 btrfs_free_reserved_data_space_noquota(inode, start, len);
4369 btrfs_qgroup_free_data(inode, start, len);
4370 }
4371
4372 static void force_metadata_allocation(struct btrfs_fs_info *info)
4373 {
4374 struct list_head *head = &info->space_info;
4375 struct btrfs_space_info *found;
4376
4377 rcu_read_lock();
4378 list_for_each_entry_rcu(found, head, list) {
4379 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4380 found->force_alloc = CHUNK_ALLOC_FORCE;
4381 }
4382 rcu_read_unlock();
4383 }
4384
4385 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4386 {
4387 return (global->size << 1);
4388 }
4389
4390 static int should_alloc_chunk(struct btrfs_root *root,
4391 struct btrfs_space_info *sinfo, int force)
4392 {
4393 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4394 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
4395 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
4396 u64 thresh;
4397
4398 if (force == CHUNK_ALLOC_FORCE)
4399 return 1;
4400
4401 /*
4402 * We need to take into account the global rsv because for all intents
4403 * and purposes it's used space. Don't worry about locking the
4404 * global_rsv, it doesn't change except when the transaction commits.
4405 */
4406 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4407 num_allocated += calc_global_rsv_need_space(global_rsv);
4408
4409 /*
4410 * in limited mode, we want to have some free space up to
4411 * about 1% of the FS size.
4412 */
4413 if (force == CHUNK_ALLOC_LIMITED) {
4414 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
4415 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4416
4417 if (num_bytes - num_allocated < thresh)
4418 return 1;
4419 }
4420
4421 if (num_allocated + SZ_2M < div_factor(num_bytes, 8))
4422 return 0;
4423 return 1;
4424 }
4425
4426 static u64 get_profile_num_devs(struct btrfs_root *root, u64 type)
4427 {
4428 u64 num_dev;
4429
4430 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4431 BTRFS_BLOCK_GROUP_RAID0 |
4432 BTRFS_BLOCK_GROUP_RAID5 |
4433 BTRFS_BLOCK_GROUP_RAID6))
4434 num_dev = root->fs_info->fs_devices->rw_devices;
4435 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4436 num_dev = 2;
4437 else
4438 num_dev = 1; /* DUP or single */
4439
4440 return num_dev;
4441 }
4442
4443 /*
4444 * If @is_allocation is true, reserve space in the system space info necessary
4445 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4446 * removing a chunk.
4447 */
4448 void check_system_chunk(struct btrfs_trans_handle *trans,
4449 struct btrfs_root *root,
4450 u64 type)
4451 {
4452 struct btrfs_space_info *info;
4453 u64 left;
4454 u64 thresh;
4455 int ret = 0;
4456 u64 num_devs;
4457
4458 /*
4459 * Needed because we can end up allocating a system chunk and for an
4460 * atomic and race free space reservation in the chunk block reserve.
4461 */
4462 ASSERT(mutex_is_locked(&root->fs_info->chunk_mutex));
4463
4464 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4465 spin_lock(&info->lock);
4466 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
4467 info->bytes_reserved - info->bytes_readonly -
4468 info->bytes_may_use;
4469 spin_unlock(&info->lock);
4470
4471 num_devs = get_profile_num_devs(root, type);
4472
4473 /* num_devs device items to update and 1 chunk item to add or remove */
4474 thresh = btrfs_calc_trunc_metadata_size(root, num_devs) +
4475 btrfs_calc_trans_metadata_size(root, 1);
4476
4477 if (left < thresh && btrfs_test_opt(root->fs_info, ENOSPC_DEBUG)) {
4478 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
4479 left, thresh, type);
4480 dump_space_info(info, 0, 0);
4481 }
4482
4483 if (left < thresh) {
4484 u64 flags;
4485
4486 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
4487 /*
4488 * Ignore failure to create system chunk. We might end up not
4489 * needing it, as we might not need to COW all nodes/leafs from
4490 * the paths we visit in the chunk tree (they were already COWed
4491 * or created in the current transaction for example).
4492 */
4493 ret = btrfs_alloc_chunk(trans, root, flags);
4494 }
4495
4496 if (!ret) {
4497 ret = btrfs_block_rsv_add(root->fs_info->chunk_root,
4498 &root->fs_info->chunk_block_rsv,
4499 thresh, BTRFS_RESERVE_NO_FLUSH);
4500 if (!ret)
4501 trans->chunk_bytes_reserved += thresh;
4502 }
4503 }
4504
4505 /*
4506 * If force is CHUNK_ALLOC_FORCE:
4507 * - return 1 if it successfully allocates a chunk,
4508 * - return errors including -ENOSPC otherwise.
4509 * If force is NOT CHUNK_ALLOC_FORCE:
4510 * - return 0 if it doesn't need to allocate a new chunk,
4511 * - return 1 if it successfully allocates a chunk,
4512 * - return errors including -ENOSPC otherwise.
4513 */
4514 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4515 struct btrfs_root *extent_root, u64 flags, int force)
4516 {
4517 struct btrfs_space_info *space_info;
4518 struct btrfs_fs_info *fs_info = extent_root->fs_info;
4519 int wait_for_alloc = 0;
4520 int ret = 0;
4521
4522 /* Don't re-enter if we're already allocating a chunk */
4523 if (trans->allocating_chunk)
4524 return -ENOSPC;
4525
4526 space_info = __find_space_info(extent_root->fs_info, flags);
4527 if (!space_info) {
4528 ret = update_space_info(extent_root->fs_info, flags,
4529 0, 0, 0, &space_info);
4530 BUG_ON(ret); /* -ENOMEM */
4531 }
4532 BUG_ON(!space_info); /* Logic error */
4533
4534 again:
4535 spin_lock(&space_info->lock);
4536 if (force < space_info->force_alloc)
4537 force = space_info->force_alloc;
4538 if (space_info->full) {
4539 if (should_alloc_chunk(extent_root, space_info, force))
4540 ret = -ENOSPC;
4541 else
4542 ret = 0;
4543 spin_unlock(&space_info->lock);
4544 return ret;
4545 }
4546
4547 if (!should_alloc_chunk(extent_root, space_info, force)) {
4548 spin_unlock(&space_info->lock);
4549 return 0;
4550 } else if (space_info->chunk_alloc) {
4551 wait_for_alloc = 1;
4552 } else {
4553 space_info->chunk_alloc = 1;
4554 }
4555
4556 spin_unlock(&space_info->lock);
4557
4558 mutex_lock(&fs_info->chunk_mutex);
4559
4560 /*
4561 * The chunk_mutex is held throughout the entirety of a chunk
4562 * allocation, so once we've acquired the chunk_mutex we know that the
4563 * other guy is done and we need to recheck and see if we should
4564 * allocate.
4565 */
4566 if (wait_for_alloc) {
4567 mutex_unlock(&fs_info->chunk_mutex);
4568 wait_for_alloc = 0;
4569 goto again;
4570 }
4571
4572 trans->allocating_chunk = true;
4573
4574 /*
4575 * If we have mixed data/metadata chunks we want to make sure we keep
4576 * allocating mixed chunks instead of individual chunks.
4577 */
4578 if (btrfs_mixed_space_info(space_info))
4579 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4580
4581 /*
4582 * if we're doing a data chunk, go ahead and make sure that
4583 * we keep a reasonable number of metadata chunks allocated in the
4584 * FS as well.
4585 */
4586 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4587 fs_info->data_chunk_allocations++;
4588 if (!(fs_info->data_chunk_allocations %
4589 fs_info->metadata_ratio))
4590 force_metadata_allocation(fs_info);
4591 }
4592
4593 /*
4594 * Check if we have enough space in SYSTEM chunk because we may need
4595 * to update devices.
4596 */
4597 check_system_chunk(trans, extent_root, flags);
4598
4599 ret = btrfs_alloc_chunk(trans, extent_root, flags);
4600 trans->allocating_chunk = false;
4601
4602 spin_lock(&space_info->lock);
4603 if (ret < 0 && ret != -ENOSPC)
4604 goto out;
4605 if (ret)
4606 space_info->full = 1;
4607 else
4608 ret = 1;
4609
4610 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4611 out:
4612 space_info->chunk_alloc = 0;
4613 spin_unlock(&space_info->lock);
4614 mutex_unlock(&fs_info->chunk_mutex);
4615 /*
4616 * When we allocate a new chunk we reserve space in the chunk block
4617 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4618 * add new nodes/leafs to it if we end up needing to do it when
4619 * inserting the chunk item and updating device items as part of the
4620 * second phase of chunk allocation, performed by
4621 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4622 * large number of new block groups to create in our transaction
4623 * handle's new_bgs list to avoid exhausting the chunk block reserve
4624 * in extreme cases - like having a single transaction create many new
4625 * block groups when starting to write out the free space caches of all
4626 * the block groups that were made dirty during the lifetime of the
4627 * transaction.
4628 */
4629 if (trans->can_flush_pending_bgs &&
4630 trans->chunk_bytes_reserved >= (u64)SZ_2M) {
4631 btrfs_create_pending_block_groups(trans, extent_root);
4632 btrfs_trans_release_chunk_metadata(trans);
4633 }
4634 return ret;
4635 }
4636
4637 static int can_overcommit(struct btrfs_root *root,
4638 struct btrfs_space_info *space_info, u64 bytes,
4639 enum btrfs_reserve_flush_enum flush)
4640 {
4641 struct btrfs_block_rsv *global_rsv;
4642 u64 profile;
4643 u64 space_size;
4644 u64 avail;
4645 u64 used;
4646
4647 /* Don't overcommit when in mixed mode. */
4648 if (space_info->flags & BTRFS_BLOCK_GROUP_DATA)
4649 return 0;
4650
4651 BUG_ON(root->fs_info == NULL);
4652 global_rsv = &root->fs_info->global_block_rsv;
4653 profile = btrfs_get_alloc_profile(root, 0);
4654 used = space_info->bytes_used + space_info->bytes_reserved +
4655 space_info->bytes_pinned + space_info->bytes_readonly;
4656
4657 /*
4658 * We only want to allow over committing if we have lots of actual space
4659 * free, but if we don't have enough space to handle the global reserve
4660 * space then we could end up having a real enospc problem when trying
4661 * to allocate a chunk or some other such important allocation.
4662 */
4663 spin_lock(&global_rsv->lock);
4664 space_size = calc_global_rsv_need_space(global_rsv);
4665 spin_unlock(&global_rsv->lock);
4666 if (used + space_size >= space_info->total_bytes)
4667 return 0;
4668
4669 used += space_info->bytes_may_use;
4670
4671 spin_lock(&root->fs_info->free_chunk_lock);
4672 avail = root->fs_info->free_chunk_space;
4673 spin_unlock(&root->fs_info->free_chunk_lock);
4674
4675 /*
4676 * If we have dup, raid1 or raid10 then only half of the free
4677 * space is actually useable. For raid56, the space info used
4678 * doesn't include the parity drive, so we don't have to
4679 * change the math
4680 */
4681 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4682 BTRFS_BLOCK_GROUP_RAID1 |
4683 BTRFS_BLOCK_GROUP_RAID10))
4684 avail >>= 1;
4685
4686 /*
4687 * If we aren't flushing all things, let us overcommit up to
4688 * 1/2th of the space. If we can flush, don't let us overcommit
4689 * too much, let it overcommit up to 1/8 of the space.
4690 */
4691 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4692 avail >>= 3;
4693 else
4694 avail >>= 1;
4695
4696 if (used + bytes < space_info->total_bytes + avail)
4697 return 1;
4698 return 0;
4699 }
4700
4701 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4702 unsigned long nr_pages, int nr_items)
4703 {
4704 struct super_block *sb = root->fs_info->sb;
4705
4706 if (down_read_trylock(&sb->s_umount)) {
4707 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4708 up_read(&sb->s_umount);
4709 } else {
4710 /*
4711 * We needn't worry the filesystem going from r/w to r/o though
4712 * we don't acquire ->s_umount mutex, because the filesystem
4713 * should guarantee the delalloc inodes list be empty after
4714 * the filesystem is readonly(all dirty pages are written to
4715 * the disk).
4716 */
4717 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4718 if (!current->journal_info)
4719 btrfs_wait_ordered_roots(root->fs_info, nr_items,
4720 0, (u64)-1);
4721 }
4722 }
4723
4724 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4725 {
4726 u64 bytes;
4727 int nr;
4728
4729 bytes = btrfs_calc_trans_metadata_size(root, 1);
4730 nr = (int)div64_u64(to_reclaim, bytes);
4731 if (!nr)
4732 nr = 1;
4733 return nr;
4734 }
4735
4736 #define EXTENT_SIZE_PER_ITEM SZ_256K
4737
4738 /*
4739 * shrink metadata reservation for delalloc
4740 */
4741 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4742 bool wait_ordered)
4743 {
4744 struct btrfs_block_rsv *block_rsv;
4745 struct btrfs_space_info *space_info;
4746 struct btrfs_trans_handle *trans;
4747 u64 delalloc_bytes;
4748 u64 max_reclaim;
4749 long time_left;
4750 unsigned long nr_pages;
4751 int loops;
4752 int items;
4753 enum btrfs_reserve_flush_enum flush;
4754
4755 /* Calc the number of the pages we need flush for space reservation */
4756 items = calc_reclaim_items_nr(root, to_reclaim);
4757 to_reclaim = (u64)items * EXTENT_SIZE_PER_ITEM;
4758
4759 trans = (struct btrfs_trans_handle *)current->journal_info;
4760 block_rsv = &root->fs_info->delalloc_block_rsv;
4761 space_info = block_rsv->space_info;
4762
4763 delalloc_bytes = percpu_counter_sum_positive(
4764 &root->fs_info->delalloc_bytes);
4765 if (delalloc_bytes == 0) {
4766 if (trans)
4767 return;
4768 if (wait_ordered)
4769 btrfs_wait_ordered_roots(root->fs_info, items,
4770 0, (u64)-1);
4771 return;
4772 }
4773
4774 loops = 0;
4775 while (delalloc_bytes && loops < 3) {
4776 max_reclaim = min(delalloc_bytes, to_reclaim);
4777 nr_pages = max_reclaim >> PAGE_SHIFT;
4778 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4779 /*
4780 * We need to wait for the async pages to actually start before
4781 * we do anything.
4782 */
4783 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4784 if (!max_reclaim)
4785 goto skip_async;
4786
4787 if (max_reclaim <= nr_pages)
4788 max_reclaim = 0;
4789 else
4790 max_reclaim -= nr_pages;
4791
4792 wait_event(root->fs_info->async_submit_wait,
4793 atomic_read(&root->fs_info->async_delalloc_pages) <=
4794 (int)max_reclaim);
4795 skip_async:
4796 if (!trans)
4797 flush = BTRFS_RESERVE_FLUSH_ALL;
4798 else
4799 flush = BTRFS_RESERVE_NO_FLUSH;
4800 spin_lock(&space_info->lock);
4801 if (can_overcommit(root, space_info, orig, flush)) {
4802 spin_unlock(&space_info->lock);
4803 break;
4804 }
4805 if (list_empty(&space_info->tickets) &&
4806 list_empty(&space_info->priority_tickets)) {
4807 spin_unlock(&space_info->lock);
4808 break;
4809 }
4810 spin_unlock(&space_info->lock);
4811
4812 loops++;
4813 if (wait_ordered && !trans) {
4814 btrfs_wait_ordered_roots(root->fs_info, items,
4815 0, (u64)-1);
4816 } else {
4817 time_left = schedule_timeout_killable(1);
4818 if (time_left)
4819 break;
4820 }
4821 delalloc_bytes = percpu_counter_sum_positive(
4822 &root->fs_info->delalloc_bytes);
4823 }
4824 }
4825
4826 /**
4827 * maybe_commit_transaction - possibly commit the transaction if its ok to
4828 * @root - the root we're allocating for
4829 * @bytes - the number of bytes we want to reserve
4830 * @force - force the commit
4831 *
4832 * This will check to make sure that committing the transaction will actually
4833 * get us somewhere and then commit the transaction if it does. Otherwise it
4834 * will return -ENOSPC.
4835 */
4836 static int may_commit_transaction(struct btrfs_root *root,
4837 struct btrfs_space_info *space_info,
4838 u64 bytes, int force)
4839 {
4840 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4841 struct btrfs_trans_handle *trans;
4842
4843 trans = (struct btrfs_trans_handle *)current->journal_info;
4844 if (trans)
4845 return -EAGAIN;
4846
4847 if (force)
4848 goto commit;
4849
4850 /* See if there is enough pinned space to make this reservation */
4851 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4852 bytes) >= 0)
4853 goto commit;
4854
4855 /*
4856 * See if there is some space in the delayed insertion reservation for
4857 * this reservation.
4858 */
4859 if (space_info != delayed_rsv->space_info)
4860 return -ENOSPC;
4861
4862 spin_lock(&delayed_rsv->lock);
4863 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4864 bytes - delayed_rsv->size) >= 0) {
4865 spin_unlock(&delayed_rsv->lock);
4866 return -ENOSPC;
4867 }
4868 spin_unlock(&delayed_rsv->lock);
4869
4870 commit:
4871 trans = btrfs_join_transaction(root);
4872 if (IS_ERR(trans))
4873 return -ENOSPC;
4874
4875 return btrfs_commit_transaction(trans, root);
4876 }
4877
4878 struct reserve_ticket {
4879 u64 bytes;
4880 int error;
4881 struct list_head list;
4882 wait_queue_head_t wait;
4883 };
4884
4885 static int flush_space(struct btrfs_root *root,
4886 struct btrfs_space_info *space_info, u64 num_bytes,
4887 u64 orig_bytes, int state)
4888 {
4889 struct btrfs_trans_handle *trans;
4890 int nr;
4891 int ret = 0;
4892
4893 switch (state) {
4894 case FLUSH_DELAYED_ITEMS_NR:
4895 case FLUSH_DELAYED_ITEMS:
4896 if (state == FLUSH_DELAYED_ITEMS_NR)
4897 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4898 else
4899 nr = -1;
4900
4901 trans = btrfs_join_transaction(root);
4902 if (IS_ERR(trans)) {
4903 ret = PTR_ERR(trans);
4904 break;
4905 }
4906 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4907 btrfs_end_transaction(trans, root);
4908 break;
4909 case FLUSH_DELALLOC:
4910 case FLUSH_DELALLOC_WAIT:
4911 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4912 state == FLUSH_DELALLOC_WAIT);
4913 break;
4914 case ALLOC_CHUNK:
4915 trans = btrfs_join_transaction(root);
4916 if (IS_ERR(trans)) {
4917 ret = PTR_ERR(trans);
4918 break;
4919 }
4920 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4921 btrfs_get_alloc_profile(root, 0),
4922 CHUNK_ALLOC_NO_FORCE);
4923 btrfs_end_transaction(trans, root);
4924 if (ret > 0 || ret == -ENOSPC)
4925 ret = 0;
4926 break;
4927 case COMMIT_TRANS:
4928 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4929 break;
4930 default:
4931 ret = -ENOSPC;
4932 break;
4933 }
4934
4935 trace_btrfs_flush_space(root->fs_info, space_info->flags, num_bytes,
4936 orig_bytes, state, ret);
4937 return ret;
4938 }
4939
4940 static inline u64
4941 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4942 struct btrfs_space_info *space_info)
4943 {
4944 struct reserve_ticket *ticket;
4945 u64 used;
4946 u64 expected;
4947 u64 to_reclaim = 0;
4948
4949 list_for_each_entry(ticket, &space_info->tickets, list)
4950 to_reclaim += ticket->bytes;
4951 list_for_each_entry(ticket, &space_info->priority_tickets, list)
4952 to_reclaim += ticket->bytes;
4953 if (to_reclaim)
4954 return to_reclaim;
4955
4956 to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
4957 if (can_overcommit(root, space_info, to_reclaim,
4958 BTRFS_RESERVE_FLUSH_ALL))
4959 return 0;
4960
4961 used = space_info->bytes_used + space_info->bytes_reserved +
4962 space_info->bytes_pinned + space_info->bytes_readonly +
4963 space_info->bytes_may_use;
4964 if (can_overcommit(root, space_info, SZ_1M, BTRFS_RESERVE_FLUSH_ALL))
4965 expected = div_factor_fine(space_info->total_bytes, 95);
4966 else
4967 expected = div_factor_fine(space_info->total_bytes, 90);
4968
4969 if (used > expected)
4970 to_reclaim = used - expected;
4971 else
4972 to_reclaim = 0;
4973 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4974 space_info->bytes_reserved);
4975 return to_reclaim;
4976 }
4977
4978 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4979 struct btrfs_root *root, u64 used)
4980 {
4981 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4982
4983 /* If we're just plain full then async reclaim just slows us down. */
4984 if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
4985 return 0;
4986
4987 if (!btrfs_calc_reclaim_metadata_size(root, space_info))
4988 return 0;
4989
4990 return (used >= thresh && !btrfs_fs_closing(root->fs_info) &&
4991 !test_bit(BTRFS_FS_STATE_REMOUNTING,
4992 &root->fs_info->fs_state));
4993 }
4994
4995 static void wake_all_tickets(struct list_head *head)
4996 {
4997 struct reserve_ticket *ticket;
4998
4999 while (!list_empty(head)) {
5000 ticket = list_first_entry(head, struct reserve_ticket, list);
5001 list_del_init(&ticket->list);
5002 ticket->error = -ENOSPC;
5003 wake_up(&ticket->wait);
5004 }
5005 }
5006
5007 /*
5008 * This is for normal flushers, we can wait all goddamned day if we want to. We
5009 * will loop and continuously try to flush as long as we are making progress.
5010 * We count progress as clearing off tickets each time we have to loop.
5011 */
5012 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
5013 {
5014 struct btrfs_fs_info *fs_info;
5015 struct btrfs_space_info *space_info;
5016 u64 to_reclaim;
5017 int flush_state;
5018 int commit_cycles = 0;
5019 u64 last_tickets_id;
5020
5021 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
5022 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5023
5024 spin_lock(&space_info->lock);
5025 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
5026 space_info);
5027 if (!to_reclaim) {
5028 space_info->flush = 0;
5029 spin_unlock(&space_info->lock);
5030 return;
5031 }
5032 last_tickets_id = space_info->tickets_id;
5033 spin_unlock(&space_info->lock);
5034
5035 flush_state = FLUSH_DELAYED_ITEMS_NR;
5036 do {
5037 struct reserve_ticket *ticket;
5038 int ret;
5039
5040 ret = flush_space(fs_info->fs_root, space_info, to_reclaim,
5041 to_reclaim, flush_state);
5042 spin_lock(&space_info->lock);
5043 if (list_empty(&space_info->tickets)) {
5044 space_info->flush = 0;
5045 spin_unlock(&space_info->lock);
5046 return;
5047 }
5048 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
5049 space_info);
5050 ticket = list_first_entry(&space_info->tickets,
5051 struct reserve_ticket, list);
5052 if (last_tickets_id == space_info->tickets_id) {
5053 flush_state++;
5054 } else {
5055 last_tickets_id = space_info->tickets_id;
5056 flush_state = FLUSH_DELAYED_ITEMS_NR;
5057 if (commit_cycles)
5058 commit_cycles--;
5059 }
5060
5061 if (flush_state > COMMIT_TRANS) {
5062 commit_cycles++;
5063 if (commit_cycles > 2) {
5064 wake_all_tickets(&space_info->tickets);
5065 space_info->flush = 0;
5066 } else {
5067 flush_state = FLUSH_DELAYED_ITEMS_NR;
5068 }
5069 }
5070 spin_unlock(&space_info->lock);
5071 } while (flush_state <= COMMIT_TRANS);
5072 }
5073
5074 void btrfs_init_async_reclaim_work(struct work_struct *work)
5075 {
5076 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
5077 }
5078
5079 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
5080 struct btrfs_space_info *space_info,
5081 struct reserve_ticket *ticket)
5082 {
5083 u64 to_reclaim;
5084 int flush_state = FLUSH_DELAYED_ITEMS_NR;
5085
5086 spin_lock(&space_info->lock);
5087 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
5088 space_info);
5089 if (!to_reclaim) {
5090 spin_unlock(&space_info->lock);
5091 return;
5092 }
5093 spin_unlock(&space_info->lock);
5094
5095 do {
5096 flush_space(fs_info->fs_root, space_info, to_reclaim,
5097 to_reclaim, flush_state);
5098 flush_state++;
5099 spin_lock(&space_info->lock);
5100 if (ticket->bytes == 0) {
5101 spin_unlock(&space_info->lock);
5102 return;
5103 }
5104 spin_unlock(&space_info->lock);
5105
5106 /*
5107 * Priority flushers can't wait on delalloc without
5108 * deadlocking.
5109 */
5110 if (flush_state == FLUSH_DELALLOC ||
5111 flush_state == FLUSH_DELALLOC_WAIT)
5112 flush_state = ALLOC_CHUNK;
5113 } while (flush_state < COMMIT_TRANS);
5114 }
5115
5116 static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
5117 struct btrfs_space_info *space_info,
5118 struct reserve_ticket *ticket, u64 orig_bytes)
5119
5120 {
5121 DEFINE_WAIT(wait);
5122 int ret = 0;
5123
5124 spin_lock(&space_info->lock);
5125 while (ticket->bytes > 0 && ticket->error == 0) {
5126 ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE);
5127 if (ret) {
5128 ret = -EINTR;
5129 break;
5130 }
5131 spin_unlock(&space_info->lock);
5132
5133 schedule();
5134
5135 finish_wait(&ticket->wait, &wait);
5136 spin_lock(&space_info->lock);
5137 }
5138 if (!ret)
5139 ret = ticket->error;
5140 if (!list_empty(&ticket->list))
5141 list_del_init(&ticket->list);
5142 if (ticket->bytes && ticket->bytes < orig_bytes) {
5143 u64 num_bytes = orig_bytes - ticket->bytes;
5144 space_info->bytes_may_use -= num_bytes;
5145 trace_btrfs_space_reservation(fs_info, "space_info",
5146 space_info->flags, num_bytes, 0);
5147 }
5148 spin_unlock(&space_info->lock);
5149
5150 return ret;
5151 }
5152
5153 /**
5154 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5155 * @root - the root we're allocating for
5156 * @space_info - the space info we want to allocate from
5157 * @orig_bytes - the number of bytes we want
5158 * @flush - whether or not we can flush to make our reservation
5159 *
5160 * This will reserve orig_bytes number of bytes from the space info associated
5161 * with the block_rsv. If there is not enough space it will make an attempt to
5162 * flush out space to make room. It will do this by flushing delalloc if
5163 * possible or committing the transaction. If flush is 0 then no attempts to
5164 * regain reservations will be made and this will fail if there is not enough
5165 * space already.
5166 */
5167 static int __reserve_metadata_bytes(struct btrfs_root *root,
5168 struct btrfs_space_info *space_info,
5169 u64 orig_bytes,
5170 enum btrfs_reserve_flush_enum flush)
5171 {
5172 struct reserve_ticket ticket;
5173 u64 used;
5174 int ret = 0;
5175
5176 ASSERT(orig_bytes);
5177 ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL);
5178
5179 spin_lock(&space_info->lock);
5180 ret = -ENOSPC;
5181 used = space_info->bytes_used + space_info->bytes_reserved +
5182 space_info->bytes_pinned + space_info->bytes_readonly +
5183 space_info->bytes_may_use;
5184
5185 /*
5186 * If we have enough space then hooray, make our reservation and carry
5187 * on. If not see if we can overcommit, and if we can, hooray carry on.
5188 * If not things get more complicated.
5189 */
5190 if (used + orig_bytes <= space_info->total_bytes) {
5191 space_info->bytes_may_use += orig_bytes;
5192 trace_btrfs_space_reservation(root->fs_info, "space_info",
5193 space_info->flags, orig_bytes,
5194 1);
5195 ret = 0;
5196 } else if (can_overcommit(root, space_info, orig_bytes, flush)) {
5197 space_info->bytes_may_use += orig_bytes;
5198 trace_btrfs_space_reservation(root->fs_info, "space_info",
5199 space_info->flags, orig_bytes,
5200 1);
5201 ret = 0;
5202 }
5203
5204 /*
5205 * If we couldn't make a reservation then setup our reservation ticket
5206 * and kick the async worker if it's not already running.
5207 *
5208 * If we are a priority flusher then we just need to add our ticket to
5209 * the list and we will do our own flushing further down.
5210 */
5211 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5212 ticket.bytes = orig_bytes;
5213 ticket.error = 0;
5214 init_waitqueue_head(&ticket.wait);
5215 if (flush == BTRFS_RESERVE_FLUSH_ALL) {
5216 list_add_tail(&ticket.list, &space_info->tickets);
5217 if (!space_info->flush) {
5218 space_info->flush = 1;
5219 trace_btrfs_trigger_flush(root->fs_info,
5220 space_info->flags,
5221 orig_bytes, flush,
5222 "enospc");
5223 queue_work(system_unbound_wq,
5224 &root->fs_info->async_reclaim_work);
5225 }
5226 } else {
5227 list_add_tail(&ticket.list,
5228 &space_info->priority_tickets);
5229 }
5230 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5231 used += orig_bytes;
5232 /*
5233 * We will do the space reservation dance during log replay,
5234 * which means we won't have fs_info->fs_root set, so don't do
5235 * the async reclaim as we will panic.
5236 */
5237 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags) &&
5238 need_do_async_reclaim(space_info, root, used) &&
5239 !work_busy(&root->fs_info->async_reclaim_work)) {
5240 trace_btrfs_trigger_flush(root->fs_info,
5241 space_info->flags,
5242 orig_bytes, flush,
5243 "preempt");
5244 queue_work(system_unbound_wq,
5245 &root->fs_info->async_reclaim_work);
5246 }
5247 }
5248 spin_unlock(&space_info->lock);
5249 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5250 return ret;
5251
5252 if (flush == BTRFS_RESERVE_FLUSH_ALL)
5253 return wait_reserve_ticket(root->fs_info, space_info, &ticket,
5254 orig_bytes);
5255
5256 ret = 0;
5257 priority_reclaim_metadata_space(root->fs_info, space_info, &ticket);
5258 spin_lock(&space_info->lock);
5259 if (ticket.bytes) {
5260 if (ticket.bytes < orig_bytes) {
5261 u64 num_bytes = orig_bytes - ticket.bytes;
5262 space_info->bytes_may_use -= num_bytes;
5263 trace_btrfs_space_reservation(root->fs_info,
5264 "space_info", space_info->flags,
5265 num_bytes, 0);
5266
5267 }
5268 list_del_init(&ticket.list);
5269 ret = -ENOSPC;
5270 }
5271 spin_unlock(&space_info->lock);
5272 ASSERT(list_empty(&ticket.list));
5273 return ret;
5274 }
5275
5276 /**
5277 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5278 * @root - the root we're allocating for
5279 * @block_rsv - the block_rsv we're allocating for
5280 * @orig_bytes - the number of bytes we want
5281 * @flush - whether or not we can flush to make our reservation
5282 *
5283 * This will reserve orgi_bytes number of bytes from the space info associated
5284 * with the block_rsv. If there is not enough space it will make an attempt to
5285 * flush out space to make room. It will do this by flushing delalloc if
5286 * possible or committing the transaction. If flush is 0 then no attempts to
5287 * regain reservations will be made and this will fail if there is not enough
5288 * space already.
5289 */
5290 static int reserve_metadata_bytes(struct btrfs_root *root,
5291 struct btrfs_block_rsv *block_rsv,
5292 u64 orig_bytes,
5293 enum btrfs_reserve_flush_enum flush)
5294 {
5295 int ret;
5296
5297 ret = __reserve_metadata_bytes(root, block_rsv->space_info, orig_bytes,
5298 flush);
5299 if (ret == -ENOSPC &&
5300 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5301 struct btrfs_block_rsv *global_rsv =
5302 &root->fs_info->global_block_rsv;
5303
5304 if (block_rsv != global_rsv &&
5305 !block_rsv_use_bytes(global_rsv, orig_bytes))
5306 ret = 0;
5307 }
5308 if (ret == -ENOSPC)
5309 trace_btrfs_space_reservation(root->fs_info,
5310 "space_info:enospc",
5311 block_rsv->space_info->flags,
5312 orig_bytes, 1);
5313 return ret;
5314 }
5315
5316 static struct btrfs_block_rsv *get_block_rsv(
5317 const struct btrfs_trans_handle *trans,
5318 const struct btrfs_root *root)
5319 {
5320 struct btrfs_block_rsv *block_rsv = NULL;
5321
5322 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5323 (root == root->fs_info->csum_root && trans->adding_csums) ||
5324 (root == root->fs_info->uuid_root))
5325 block_rsv = trans->block_rsv;
5326
5327 if (!block_rsv)
5328 block_rsv = root->block_rsv;
5329
5330 if (!block_rsv)
5331 block_rsv = &root->fs_info->empty_block_rsv;
5332
5333 return block_rsv;
5334 }
5335
5336 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5337 u64 num_bytes)
5338 {
5339 int ret = -ENOSPC;
5340 spin_lock(&block_rsv->lock);
5341 if (block_rsv->reserved >= num_bytes) {
5342 block_rsv->reserved -= num_bytes;
5343 if (block_rsv->reserved < block_rsv->size)
5344 block_rsv->full = 0;
5345 ret = 0;
5346 }
5347 spin_unlock(&block_rsv->lock);
5348 return ret;
5349 }
5350
5351 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5352 u64 num_bytes, int update_size)
5353 {
5354 spin_lock(&block_rsv->lock);
5355 block_rsv->reserved += num_bytes;
5356 if (update_size)
5357 block_rsv->size += num_bytes;
5358 else if (block_rsv->reserved >= block_rsv->size)
5359 block_rsv->full = 1;
5360 spin_unlock(&block_rsv->lock);
5361 }
5362
5363 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5364 struct btrfs_block_rsv *dest, u64 num_bytes,
5365 int min_factor)
5366 {
5367 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5368 u64 min_bytes;
5369
5370 if (global_rsv->space_info != dest->space_info)
5371 return -ENOSPC;
5372
5373 spin_lock(&global_rsv->lock);
5374 min_bytes = div_factor(global_rsv->size, min_factor);
5375 if (global_rsv->reserved < min_bytes + num_bytes) {
5376 spin_unlock(&global_rsv->lock);
5377 return -ENOSPC;
5378 }
5379 global_rsv->reserved -= num_bytes;
5380 if (global_rsv->reserved < global_rsv->size)
5381 global_rsv->full = 0;
5382 spin_unlock(&global_rsv->lock);
5383
5384 block_rsv_add_bytes(dest, num_bytes, 1);
5385 return 0;
5386 }
5387
5388 /*
5389 * This is for space we already have accounted in space_info->bytes_may_use, so
5390 * basically when we're returning space from block_rsv's.
5391 */
5392 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
5393 struct btrfs_space_info *space_info,
5394 u64 num_bytes)
5395 {
5396 struct reserve_ticket *ticket;
5397 struct list_head *head;
5398 u64 used;
5399 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH;
5400 bool check_overcommit = false;
5401
5402 spin_lock(&space_info->lock);
5403 head = &space_info->priority_tickets;
5404
5405 /*
5406 * If we are over our limit then we need to check and see if we can
5407 * overcommit, and if we can't then we just need to free up our space
5408 * and not satisfy any requests.
5409 */
5410 used = space_info->bytes_used + space_info->bytes_reserved +
5411 space_info->bytes_pinned + space_info->bytes_readonly +
5412 space_info->bytes_may_use;
5413 if (used - num_bytes >= space_info->total_bytes)
5414 check_overcommit = true;
5415 again:
5416 while (!list_empty(head) && num_bytes) {
5417 ticket = list_first_entry(head, struct reserve_ticket,
5418 list);
5419 /*
5420 * We use 0 bytes because this space is already reserved, so
5421 * adding the ticket space would be a double count.
5422 */
5423 if (check_overcommit &&
5424 !can_overcommit(fs_info->extent_root, space_info, 0,
5425 flush))
5426 break;
5427 if (num_bytes >= ticket->bytes) {
5428 list_del_init(&ticket->list);
5429 num_bytes -= ticket->bytes;
5430 ticket->bytes = 0;
5431 space_info->tickets_id++;
5432 wake_up(&ticket->wait);
5433 } else {
5434 ticket->bytes -= num_bytes;
5435 num_bytes = 0;
5436 }
5437 }
5438
5439 if (num_bytes && head == &space_info->priority_tickets) {
5440 head = &space_info->tickets;
5441 flush = BTRFS_RESERVE_FLUSH_ALL;
5442 goto again;
5443 }
5444 space_info->bytes_may_use -= num_bytes;
5445 trace_btrfs_space_reservation(fs_info, "space_info",
5446 space_info->flags, num_bytes, 0);
5447 spin_unlock(&space_info->lock);
5448 }
5449
5450 /*
5451 * This is for newly allocated space that isn't accounted in
5452 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5453 * we use this helper.
5454 */
5455 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
5456 struct btrfs_space_info *space_info,
5457 u64 num_bytes)
5458 {
5459 struct reserve_ticket *ticket;
5460 struct list_head *head = &space_info->priority_tickets;
5461
5462 again:
5463 while (!list_empty(head) && num_bytes) {
5464 ticket = list_first_entry(head, struct reserve_ticket,
5465 list);
5466 if (num_bytes >= ticket->bytes) {
5467 trace_btrfs_space_reservation(fs_info, "space_info",
5468 space_info->flags,
5469 ticket->bytes, 1);
5470 list_del_init(&ticket->list);
5471 num_bytes -= ticket->bytes;
5472 space_info->bytes_may_use += ticket->bytes;
5473 ticket->bytes = 0;
5474 space_info->tickets_id++;
5475 wake_up(&ticket->wait);
5476 } else {
5477 trace_btrfs_space_reservation(fs_info, "space_info",
5478 space_info->flags,
5479 num_bytes, 1);
5480 space_info->bytes_may_use += num_bytes;
5481 ticket->bytes -= num_bytes;
5482 num_bytes = 0;
5483 }
5484 }
5485
5486 if (num_bytes && head == &space_info->priority_tickets) {
5487 head = &space_info->tickets;
5488 goto again;
5489 }
5490 }
5491
5492 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5493 struct btrfs_block_rsv *block_rsv,
5494 struct btrfs_block_rsv *dest, u64 num_bytes)
5495 {
5496 struct btrfs_space_info *space_info = block_rsv->space_info;
5497
5498 spin_lock(&block_rsv->lock);
5499 if (num_bytes == (u64)-1)
5500 num_bytes = block_rsv->size;
5501 block_rsv->size -= num_bytes;
5502 if (block_rsv->reserved >= block_rsv->size) {
5503 num_bytes = block_rsv->reserved - block_rsv->size;
5504 block_rsv->reserved = block_rsv->size;
5505 block_rsv->full = 1;
5506 } else {
5507 num_bytes = 0;
5508 }
5509 spin_unlock(&block_rsv->lock);
5510
5511 if (num_bytes > 0) {
5512 if (dest) {
5513 spin_lock(&dest->lock);
5514 if (!dest->full) {
5515 u64 bytes_to_add;
5516
5517 bytes_to_add = dest->size - dest->reserved;
5518 bytes_to_add = min(num_bytes, bytes_to_add);
5519 dest->reserved += bytes_to_add;
5520 if (dest->reserved >= dest->size)
5521 dest->full = 1;
5522 num_bytes -= bytes_to_add;
5523 }
5524 spin_unlock(&dest->lock);
5525 }
5526 if (num_bytes)
5527 space_info_add_old_bytes(fs_info, space_info,
5528 num_bytes);
5529 }
5530 }
5531
5532 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
5533 struct btrfs_block_rsv *dst, u64 num_bytes,
5534 int update_size)
5535 {
5536 int ret;
5537
5538 ret = block_rsv_use_bytes(src, num_bytes);
5539 if (ret)
5540 return ret;
5541
5542 block_rsv_add_bytes(dst, num_bytes, update_size);
5543 return 0;
5544 }
5545
5546 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5547 {
5548 memset(rsv, 0, sizeof(*rsv));
5549 spin_lock_init(&rsv->lock);
5550 rsv->type = type;
5551 }
5552
5553 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
5554 unsigned short type)
5555 {
5556 struct btrfs_block_rsv *block_rsv;
5557 struct btrfs_fs_info *fs_info = root->fs_info;
5558
5559 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5560 if (!block_rsv)
5561 return NULL;
5562
5563 btrfs_init_block_rsv(block_rsv, type);
5564 block_rsv->space_info = __find_space_info(fs_info,
5565 BTRFS_BLOCK_GROUP_METADATA);
5566 return block_rsv;
5567 }
5568
5569 void btrfs_free_block_rsv(struct btrfs_root *root,
5570 struct btrfs_block_rsv *rsv)
5571 {
5572 if (!rsv)
5573 return;
5574 btrfs_block_rsv_release(root, rsv, (u64)-1);
5575 kfree(rsv);
5576 }
5577
5578 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5579 {
5580 kfree(rsv);
5581 }
5582
5583 int btrfs_block_rsv_add(struct btrfs_root *root,
5584 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5585 enum btrfs_reserve_flush_enum flush)
5586 {
5587 int ret;
5588
5589 if (num_bytes == 0)
5590 return 0;
5591
5592 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5593 if (!ret) {
5594 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5595 return 0;
5596 }
5597
5598 return ret;
5599 }
5600
5601 int btrfs_block_rsv_check(struct btrfs_root *root,
5602 struct btrfs_block_rsv *block_rsv, int min_factor)
5603 {
5604 u64 num_bytes = 0;
5605 int ret = -ENOSPC;
5606
5607 if (!block_rsv)
5608 return 0;
5609
5610 spin_lock(&block_rsv->lock);
5611 num_bytes = div_factor(block_rsv->size, min_factor);
5612 if (block_rsv->reserved >= num_bytes)
5613 ret = 0;
5614 spin_unlock(&block_rsv->lock);
5615
5616 return ret;
5617 }
5618
5619 int btrfs_block_rsv_refill(struct btrfs_root *root,
5620 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5621 enum btrfs_reserve_flush_enum flush)
5622 {
5623 u64 num_bytes = 0;
5624 int ret = -ENOSPC;
5625
5626 if (!block_rsv)
5627 return 0;
5628
5629 spin_lock(&block_rsv->lock);
5630 num_bytes = min_reserved;
5631 if (block_rsv->reserved >= num_bytes)
5632 ret = 0;
5633 else
5634 num_bytes -= block_rsv->reserved;
5635 spin_unlock(&block_rsv->lock);
5636
5637 if (!ret)
5638 return 0;
5639
5640 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5641 if (!ret) {
5642 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5643 return 0;
5644 }
5645
5646 return ret;
5647 }
5648
5649 void btrfs_block_rsv_release(struct btrfs_root *root,
5650 struct btrfs_block_rsv *block_rsv,
5651 u64 num_bytes)
5652 {
5653 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5654 if (global_rsv == block_rsv ||
5655 block_rsv->space_info != global_rsv->space_info)
5656 global_rsv = NULL;
5657 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
5658 num_bytes);
5659 }
5660
5661 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5662 {
5663 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5664 struct btrfs_space_info *sinfo = block_rsv->space_info;
5665 u64 num_bytes;
5666
5667 /*
5668 * The global block rsv is based on the size of the extent tree, the
5669 * checksum tree and the root tree. If the fs is empty we want to set
5670 * it to a minimal amount for safety.
5671 */
5672 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
5673 btrfs_root_used(&fs_info->csum_root->root_item) +
5674 btrfs_root_used(&fs_info->tree_root->root_item);
5675 num_bytes = max_t(u64, num_bytes, SZ_16M);
5676
5677 spin_lock(&sinfo->lock);
5678 spin_lock(&block_rsv->lock);
5679
5680 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5681
5682 if (block_rsv->reserved < block_rsv->size) {
5683 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
5684 sinfo->bytes_reserved + sinfo->bytes_readonly +
5685 sinfo->bytes_may_use;
5686 if (sinfo->total_bytes > num_bytes) {
5687 num_bytes = sinfo->total_bytes - num_bytes;
5688 num_bytes = min(num_bytes,
5689 block_rsv->size - block_rsv->reserved);
5690 block_rsv->reserved += num_bytes;
5691 sinfo->bytes_may_use += num_bytes;
5692 trace_btrfs_space_reservation(fs_info, "space_info",
5693 sinfo->flags, num_bytes,
5694 1);
5695 }
5696 } else if (block_rsv->reserved > block_rsv->size) {
5697 num_bytes = block_rsv->reserved - block_rsv->size;
5698 sinfo->bytes_may_use -= num_bytes;
5699 trace_btrfs_space_reservation(fs_info, "space_info",
5700 sinfo->flags, num_bytes, 0);
5701 block_rsv->reserved = block_rsv->size;
5702 }
5703
5704 if (block_rsv->reserved == block_rsv->size)
5705 block_rsv->full = 1;
5706 else
5707 block_rsv->full = 0;
5708
5709 spin_unlock(&block_rsv->lock);
5710 spin_unlock(&sinfo->lock);
5711 }
5712
5713 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5714 {
5715 struct btrfs_space_info *space_info;
5716
5717 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5718 fs_info->chunk_block_rsv.space_info = space_info;
5719
5720 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5721 fs_info->global_block_rsv.space_info = space_info;
5722 fs_info->delalloc_block_rsv.space_info = space_info;
5723 fs_info->trans_block_rsv.space_info = space_info;
5724 fs_info->empty_block_rsv.space_info = space_info;
5725 fs_info->delayed_block_rsv.space_info = space_info;
5726
5727 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5728 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5729 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5730 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5731 if (fs_info->quota_root)
5732 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5733 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5734
5735 update_global_block_rsv(fs_info);
5736 }
5737
5738 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5739 {
5740 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5741 (u64)-1);
5742 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5743 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5744 WARN_ON(fs_info->trans_block_rsv.size > 0);
5745 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5746 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5747 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5748 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5749 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5750 }
5751
5752 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5753 struct btrfs_root *root)
5754 {
5755 if (!trans->block_rsv)
5756 return;
5757
5758 if (!trans->bytes_reserved)
5759 return;
5760
5761 trace_btrfs_space_reservation(root->fs_info, "transaction",
5762 trans->transid, trans->bytes_reserved, 0);
5763 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
5764 trans->bytes_reserved = 0;
5765 }
5766
5767 /*
5768 * To be called after all the new block groups attached to the transaction
5769 * handle have been created (btrfs_create_pending_block_groups()).
5770 */
5771 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5772 {
5773 struct btrfs_fs_info *fs_info = trans->fs_info;
5774
5775 if (!trans->chunk_bytes_reserved)
5776 return;
5777
5778 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5779
5780 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5781 trans->chunk_bytes_reserved);
5782 trans->chunk_bytes_reserved = 0;
5783 }
5784
5785 /* Can only return 0 or -ENOSPC */
5786 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5787 struct inode *inode)
5788 {
5789 struct btrfs_root *root = BTRFS_I(inode)->root;
5790 /*
5791 * We always use trans->block_rsv here as we will have reserved space
5792 * for our orphan when starting the transaction, using get_block_rsv()
5793 * here will sometimes make us choose the wrong block rsv as we could be
5794 * doing a reloc inode for a non refcounted root.
5795 */
5796 struct btrfs_block_rsv *src_rsv = trans->block_rsv;
5797 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5798
5799 /*
5800 * We need to hold space in order to delete our orphan item once we've
5801 * added it, so this takes the reservation so we can release it later
5802 * when we are truly done with the orphan item.
5803 */
5804 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5805 trace_btrfs_space_reservation(root->fs_info, "orphan",
5806 btrfs_ino(inode), num_bytes, 1);
5807 return btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
5808 }
5809
5810 void btrfs_orphan_release_metadata(struct inode *inode)
5811 {
5812 struct btrfs_root *root = BTRFS_I(inode)->root;
5813 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5814 trace_btrfs_space_reservation(root->fs_info, "orphan",
5815 btrfs_ino(inode), num_bytes, 0);
5816 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
5817 }
5818
5819 /*
5820 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5821 * root: the root of the parent directory
5822 * rsv: block reservation
5823 * items: the number of items that we need do reservation
5824 * qgroup_reserved: used to return the reserved size in qgroup
5825 *
5826 * This function is used to reserve the space for snapshot/subvolume
5827 * creation and deletion. Those operations are different with the
5828 * common file/directory operations, they change two fs/file trees
5829 * and root tree, the number of items that the qgroup reserves is
5830 * different with the free space reservation. So we can not use
5831 * the space reservation mechanism in start_transaction().
5832 */
5833 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5834 struct btrfs_block_rsv *rsv,
5835 int items,
5836 u64 *qgroup_reserved,
5837 bool use_global_rsv)
5838 {
5839 u64 num_bytes;
5840 int ret;
5841 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5842
5843 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags)) {
5844 /* One for parent inode, two for dir entries */
5845 num_bytes = 3 * root->nodesize;
5846 ret = btrfs_qgroup_reserve_meta(root, num_bytes);
5847 if (ret)
5848 return ret;
5849 } else {
5850 num_bytes = 0;
5851 }
5852
5853 *qgroup_reserved = num_bytes;
5854
5855 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5856 rsv->space_info = __find_space_info(root->fs_info,
5857 BTRFS_BLOCK_GROUP_METADATA);
5858 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5859 BTRFS_RESERVE_FLUSH_ALL);
5860
5861 if (ret == -ENOSPC && use_global_rsv)
5862 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1);
5863
5864 if (ret && *qgroup_reserved)
5865 btrfs_qgroup_free_meta(root, *qgroup_reserved);
5866
5867 return ret;
5868 }
5869
5870 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5871 struct btrfs_block_rsv *rsv,
5872 u64 qgroup_reserved)
5873 {
5874 btrfs_block_rsv_release(root, rsv, (u64)-1);
5875 }
5876
5877 /**
5878 * drop_outstanding_extent - drop an outstanding extent
5879 * @inode: the inode we're dropping the extent for
5880 * @num_bytes: the number of bytes we're releasing.
5881 *
5882 * This is called when we are freeing up an outstanding extent, either called
5883 * after an error or after an extent is written. This will return the number of
5884 * reserved extents that need to be freed. This must be called with
5885 * BTRFS_I(inode)->lock held.
5886 */
5887 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5888 {
5889 unsigned drop_inode_space = 0;
5890 unsigned dropped_extents = 0;
5891 unsigned num_extents = 0;
5892
5893 num_extents = (unsigned)div64_u64(num_bytes +
5894 BTRFS_MAX_EXTENT_SIZE - 1,
5895 BTRFS_MAX_EXTENT_SIZE);
5896 ASSERT(num_extents);
5897 ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5898 BTRFS_I(inode)->outstanding_extents -= num_extents;
5899
5900 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5901 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5902 &BTRFS_I(inode)->runtime_flags))
5903 drop_inode_space = 1;
5904
5905 /*
5906 * If we have more or the same amount of outstanding extents than we have
5907 * reserved then we need to leave the reserved extents count alone.
5908 */
5909 if (BTRFS_I(inode)->outstanding_extents >=
5910 BTRFS_I(inode)->reserved_extents)
5911 return drop_inode_space;
5912
5913 dropped_extents = BTRFS_I(inode)->reserved_extents -
5914 BTRFS_I(inode)->outstanding_extents;
5915 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5916 return dropped_extents + drop_inode_space;
5917 }
5918
5919 /**
5920 * calc_csum_metadata_size - return the amount of metadata space that must be
5921 * reserved/freed for the given bytes.
5922 * @inode: the inode we're manipulating
5923 * @num_bytes: the number of bytes in question
5924 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5925 *
5926 * This adjusts the number of csum_bytes in the inode and then returns the
5927 * correct amount of metadata that must either be reserved or freed. We
5928 * calculate how many checksums we can fit into one leaf and then divide the
5929 * number of bytes that will need to be checksumed by this value to figure out
5930 * how many checksums will be required. If we are adding bytes then the number
5931 * may go up and we will return the number of additional bytes that must be
5932 * reserved. If it is going down we will return the number of bytes that must
5933 * be freed.
5934 *
5935 * This must be called with BTRFS_I(inode)->lock held.
5936 */
5937 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5938 int reserve)
5939 {
5940 struct btrfs_root *root = BTRFS_I(inode)->root;
5941 u64 old_csums, num_csums;
5942
5943 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5944 BTRFS_I(inode)->csum_bytes == 0)
5945 return 0;
5946
5947 old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5948 if (reserve)
5949 BTRFS_I(inode)->csum_bytes += num_bytes;
5950 else
5951 BTRFS_I(inode)->csum_bytes -= num_bytes;
5952 num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5953
5954 /* No change, no need to reserve more */
5955 if (old_csums == num_csums)
5956 return 0;
5957
5958 if (reserve)
5959 return btrfs_calc_trans_metadata_size(root,
5960 num_csums - old_csums);
5961
5962 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5963 }
5964
5965 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5966 {
5967 struct btrfs_root *root = BTRFS_I(inode)->root;
5968 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5969 u64 to_reserve = 0;
5970 u64 csum_bytes;
5971 unsigned nr_extents = 0;
5972 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5973 int ret = 0;
5974 bool delalloc_lock = true;
5975 u64 to_free = 0;
5976 unsigned dropped;
5977 bool release_extra = false;
5978
5979 /* If we are a free space inode we need to not flush since we will be in
5980 * the middle of a transaction commit. We also don't need the delalloc
5981 * mutex since we won't race with anybody. We need this mostly to make
5982 * lockdep shut its filthy mouth.
5983 *
5984 * If we have a transaction open (can happen if we call truncate_block
5985 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
5986 */
5987 if (btrfs_is_free_space_inode(inode)) {
5988 flush = BTRFS_RESERVE_NO_FLUSH;
5989 delalloc_lock = false;
5990 } else if (current->journal_info) {
5991 flush = BTRFS_RESERVE_FLUSH_LIMIT;
5992 }
5993
5994 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5995 btrfs_transaction_in_commit(root->fs_info))
5996 schedule_timeout(1);
5997
5998 if (delalloc_lock)
5999 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
6000
6001 num_bytes = ALIGN(num_bytes, root->sectorsize);
6002
6003 spin_lock(&BTRFS_I(inode)->lock);
6004 nr_extents = (unsigned)div64_u64(num_bytes +
6005 BTRFS_MAX_EXTENT_SIZE - 1,
6006 BTRFS_MAX_EXTENT_SIZE);
6007 BTRFS_I(inode)->outstanding_extents += nr_extents;
6008
6009 nr_extents = 0;
6010 if (BTRFS_I(inode)->outstanding_extents >
6011 BTRFS_I(inode)->reserved_extents)
6012 nr_extents += BTRFS_I(inode)->outstanding_extents -
6013 BTRFS_I(inode)->reserved_extents;
6014
6015 /* We always want to reserve a slot for updating the inode. */
6016 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents + 1);
6017 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
6018 csum_bytes = BTRFS_I(inode)->csum_bytes;
6019 spin_unlock(&BTRFS_I(inode)->lock);
6020
6021 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags)) {
6022 ret = btrfs_qgroup_reserve_meta(root,
6023 nr_extents * root->nodesize);
6024 if (ret)
6025 goto out_fail;
6026 }
6027
6028 ret = btrfs_block_rsv_add(root, block_rsv, to_reserve, flush);
6029 if (unlikely(ret)) {
6030 btrfs_qgroup_free_meta(root, nr_extents * root->nodesize);
6031 goto out_fail;
6032 }
6033
6034 spin_lock(&BTRFS_I(inode)->lock);
6035 if (test_and_set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
6036 &BTRFS_I(inode)->runtime_flags)) {
6037 to_reserve -= btrfs_calc_trans_metadata_size(root, 1);
6038 release_extra = true;
6039 }
6040 BTRFS_I(inode)->reserved_extents += nr_extents;
6041 spin_unlock(&BTRFS_I(inode)->lock);
6042
6043 if (delalloc_lock)
6044 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
6045
6046 if (to_reserve)
6047 trace_btrfs_space_reservation(root->fs_info, "delalloc",
6048 btrfs_ino(inode), to_reserve, 1);
6049 if (release_extra)
6050 btrfs_block_rsv_release(root, block_rsv,
6051 btrfs_calc_trans_metadata_size(root,
6052 1));
6053 return 0;
6054
6055 out_fail:
6056 spin_lock(&BTRFS_I(inode)->lock);
6057 dropped = drop_outstanding_extent(inode, num_bytes);
6058 /*
6059 * If the inodes csum_bytes is the same as the original
6060 * csum_bytes then we know we haven't raced with any free()ers
6061 * so we can just reduce our inodes csum bytes and carry on.
6062 */
6063 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
6064 calc_csum_metadata_size(inode, num_bytes, 0);
6065 } else {
6066 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
6067 u64 bytes;
6068
6069 /*
6070 * This is tricky, but first we need to figure out how much we
6071 * freed from any free-ers that occurred during this
6072 * reservation, so we reset ->csum_bytes to the csum_bytes
6073 * before we dropped our lock, and then call the free for the
6074 * number of bytes that were freed while we were trying our
6075 * reservation.
6076 */
6077 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
6078 BTRFS_I(inode)->csum_bytes = csum_bytes;
6079 to_free = calc_csum_metadata_size(inode, bytes, 0);
6080
6081
6082 /*
6083 * Now we need to see how much we would have freed had we not
6084 * been making this reservation and our ->csum_bytes were not
6085 * artificially inflated.
6086 */
6087 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
6088 bytes = csum_bytes - orig_csum_bytes;
6089 bytes = calc_csum_metadata_size(inode, bytes, 0);
6090
6091 /*
6092 * Now reset ->csum_bytes to what it should be. If bytes is
6093 * more than to_free then we would have freed more space had we
6094 * not had an artificially high ->csum_bytes, so we need to free
6095 * the remainder. If bytes is the same or less then we don't
6096 * need to do anything, the other free-ers did the correct
6097 * thing.
6098 */
6099 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
6100 if (bytes > to_free)
6101 to_free = bytes - to_free;
6102 else
6103 to_free = 0;
6104 }
6105 spin_unlock(&BTRFS_I(inode)->lock);
6106 if (dropped)
6107 to_free += btrfs_calc_trans_metadata_size(root, dropped);
6108
6109 if (to_free) {
6110 btrfs_block_rsv_release(root, block_rsv, to_free);
6111 trace_btrfs_space_reservation(root->fs_info, "delalloc",
6112 btrfs_ino(inode), to_free, 0);
6113 }
6114 if (delalloc_lock)
6115 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
6116 return ret;
6117 }
6118
6119 /**
6120 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6121 * @inode: the inode to release the reservation for
6122 * @num_bytes: the number of bytes we're releasing
6123 *
6124 * This will release the metadata reservation for an inode. This can be called
6125 * once we complete IO for a given set of bytes to release their metadata
6126 * reservations.
6127 */
6128 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
6129 {
6130 struct btrfs_root *root = BTRFS_I(inode)->root;
6131 u64 to_free = 0;
6132 unsigned dropped;
6133
6134 num_bytes = ALIGN(num_bytes, root->sectorsize);
6135 spin_lock(&BTRFS_I(inode)->lock);
6136 dropped = drop_outstanding_extent(inode, num_bytes);
6137
6138 if (num_bytes)
6139 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
6140 spin_unlock(&BTRFS_I(inode)->lock);
6141 if (dropped > 0)
6142 to_free += btrfs_calc_trans_metadata_size(root, dropped);
6143
6144 if (btrfs_is_testing(root->fs_info))
6145 return;
6146
6147 trace_btrfs_space_reservation(root->fs_info, "delalloc",
6148 btrfs_ino(inode), to_free, 0);
6149
6150 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
6151 to_free);
6152 }
6153
6154 /**
6155 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6156 * delalloc
6157 * @inode: inode we're writing to
6158 * @start: start range we are writing to
6159 * @len: how long the range we are writing to
6160 *
6161 * TODO: This function will finally replace old btrfs_delalloc_reserve_space()
6162 *
6163 * This will do the following things
6164 *
6165 * o reserve space in data space info for num bytes
6166 * and reserve precious corresponding qgroup space
6167 * (Done in check_data_free_space)
6168 *
6169 * o reserve space for metadata space, based on the number of outstanding
6170 * extents and how much csums will be needed
6171 * also reserve metadata space in a per root over-reserve method.
6172 * o add to the inodes->delalloc_bytes
6173 * o add it to the fs_info's delalloc inodes list.
6174 * (Above 3 all done in delalloc_reserve_metadata)
6175 *
6176 * Return 0 for success
6177 * Return <0 for error(-ENOSPC or -EQUOT)
6178 */
6179 int btrfs_delalloc_reserve_space(struct inode *inode, u64 start, u64 len)
6180 {
6181 int ret;
6182
6183 ret = btrfs_check_data_free_space(inode, start, len);
6184 if (ret < 0)
6185 return ret;
6186 ret = btrfs_delalloc_reserve_metadata(inode, len);
6187 if (ret < 0)
6188 btrfs_free_reserved_data_space(inode, start, len);
6189 return ret;
6190 }
6191
6192 /**
6193 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6194 * @inode: inode we're releasing space for
6195 * @start: start position of the space already reserved
6196 * @len: the len of the space already reserved
6197 *
6198 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
6199 * called in the case that we don't need the metadata AND data reservations
6200 * anymore. So if there is an error or we insert an inline extent.
6201 *
6202 * This function will release the metadata space that was not used and will
6203 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6204 * list if there are no delalloc bytes left.
6205 * Also it will handle the qgroup reserved space.
6206 */
6207 void btrfs_delalloc_release_space(struct inode *inode, u64 start, u64 len)
6208 {
6209 btrfs_delalloc_release_metadata(inode, len);
6210 btrfs_free_reserved_data_space(inode, start, len);
6211 }
6212
6213 static int update_block_group(struct btrfs_trans_handle *trans,
6214 struct btrfs_root *root, u64 bytenr,
6215 u64 num_bytes, int alloc)
6216 {
6217 struct btrfs_block_group_cache *cache = NULL;
6218 struct btrfs_fs_info *info = root->fs_info;
6219 u64 total = num_bytes;
6220 u64 old_val;
6221 u64 byte_in_group;
6222 int factor;
6223
6224 /* block accounting for super block */
6225 spin_lock(&info->delalloc_root_lock);
6226 old_val = btrfs_super_bytes_used(info->super_copy);
6227 if (alloc)
6228 old_val += num_bytes;
6229 else
6230 old_val -= num_bytes;
6231 btrfs_set_super_bytes_used(info->super_copy, old_val);
6232 spin_unlock(&info->delalloc_root_lock);
6233
6234 while (total) {
6235 cache = btrfs_lookup_block_group(info, bytenr);
6236 if (!cache)
6237 return -ENOENT;
6238 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
6239 BTRFS_BLOCK_GROUP_RAID1 |
6240 BTRFS_BLOCK_GROUP_RAID10))
6241 factor = 2;
6242 else
6243 factor = 1;
6244 /*
6245 * If this block group has free space cache written out, we
6246 * need to make sure to load it if we are removing space. This
6247 * is because we need the unpinning stage to actually add the
6248 * space back to the block group, otherwise we will leak space.
6249 */
6250 if (!alloc && cache->cached == BTRFS_CACHE_NO)
6251 cache_block_group(cache, 1);
6252
6253 byte_in_group = bytenr - cache->key.objectid;
6254 WARN_ON(byte_in_group > cache->key.offset);
6255
6256 spin_lock(&cache->space_info->lock);
6257 spin_lock(&cache->lock);
6258
6259 if (btrfs_test_opt(root->fs_info, SPACE_CACHE) &&
6260 cache->disk_cache_state < BTRFS_DC_CLEAR)
6261 cache->disk_cache_state = BTRFS_DC_CLEAR;
6262
6263 old_val = btrfs_block_group_used(&cache->item);
6264 num_bytes = min(total, cache->key.offset - byte_in_group);
6265 if (alloc) {
6266 old_val += num_bytes;
6267 btrfs_set_block_group_used(&cache->item, old_val);
6268 cache->reserved -= num_bytes;
6269 cache->space_info->bytes_reserved -= num_bytes;
6270 cache->space_info->bytes_used += num_bytes;
6271 cache->space_info->disk_used += num_bytes * factor;
6272 spin_unlock(&cache->lock);
6273 spin_unlock(&cache->space_info->lock);
6274 } else {
6275 old_val -= num_bytes;
6276 btrfs_set_block_group_used(&cache->item, old_val);
6277 cache->pinned += num_bytes;
6278 cache->space_info->bytes_pinned += num_bytes;
6279 cache->space_info->bytes_used -= num_bytes;
6280 cache->space_info->disk_used -= num_bytes * factor;
6281 spin_unlock(&cache->lock);
6282 spin_unlock(&cache->space_info->lock);
6283
6284 trace_btrfs_space_reservation(root->fs_info, "pinned",
6285 cache->space_info->flags,
6286 num_bytes, 1);
6287 set_extent_dirty(info->pinned_extents,
6288 bytenr, bytenr + num_bytes - 1,
6289 GFP_NOFS | __GFP_NOFAIL);
6290 }
6291
6292 spin_lock(&trans->transaction->dirty_bgs_lock);
6293 if (list_empty(&cache->dirty_list)) {
6294 list_add_tail(&cache->dirty_list,
6295 &trans->transaction->dirty_bgs);
6296 trans->transaction->num_dirty_bgs++;
6297 btrfs_get_block_group(cache);
6298 }
6299 spin_unlock(&trans->transaction->dirty_bgs_lock);
6300
6301 /*
6302 * No longer have used bytes in this block group, queue it for
6303 * deletion. We do this after adding the block group to the
6304 * dirty list to avoid races between cleaner kthread and space
6305 * cache writeout.
6306 */
6307 if (!alloc && old_val == 0) {
6308 spin_lock(&info->unused_bgs_lock);
6309 if (list_empty(&cache->bg_list)) {
6310 btrfs_get_block_group(cache);
6311 list_add_tail(&cache->bg_list,
6312 &info->unused_bgs);
6313 }
6314 spin_unlock(&info->unused_bgs_lock);
6315 }
6316
6317 btrfs_put_block_group(cache);
6318 total -= num_bytes;
6319 bytenr += num_bytes;
6320 }
6321 return 0;
6322 }
6323
6324 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
6325 {
6326 struct btrfs_block_group_cache *cache;
6327 u64 bytenr;
6328
6329 spin_lock(&root->fs_info->block_group_cache_lock);
6330 bytenr = root->fs_info->first_logical_byte;
6331 spin_unlock(&root->fs_info->block_group_cache_lock);
6332
6333 if (bytenr < (u64)-1)
6334 return bytenr;
6335
6336 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
6337 if (!cache)
6338 return 0;
6339
6340 bytenr = cache->key.objectid;
6341 btrfs_put_block_group(cache);
6342
6343 return bytenr;
6344 }
6345
6346 static int pin_down_extent(struct btrfs_root *root,
6347 struct btrfs_block_group_cache *cache,
6348 u64 bytenr, u64 num_bytes, int reserved)
6349 {
6350 spin_lock(&cache->space_info->lock);
6351 spin_lock(&cache->lock);
6352 cache->pinned += num_bytes;
6353 cache->space_info->bytes_pinned += num_bytes;
6354 if (reserved) {
6355 cache->reserved -= num_bytes;
6356 cache->space_info->bytes_reserved -= num_bytes;
6357 }
6358 spin_unlock(&cache->lock);
6359 spin_unlock(&cache->space_info->lock);
6360
6361 trace_btrfs_space_reservation(root->fs_info, "pinned",
6362 cache->space_info->flags, num_bytes, 1);
6363 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
6364 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6365 return 0;
6366 }
6367
6368 /*
6369 * this function must be called within transaction
6370 */
6371 int btrfs_pin_extent(struct btrfs_root *root,
6372 u64 bytenr, u64 num_bytes, int reserved)
6373 {
6374 struct btrfs_block_group_cache *cache;
6375
6376 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
6377 BUG_ON(!cache); /* Logic error */
6378
6379 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
6380
6381 btrfs_put_block_group(cache);
6382 return 0;
6383 }
6384
6385 /*
6386 * this function must be called within transaction
6387 */
6388 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
6389 u64 bytenr, u64 num_bytes)
6390 {
6391 struct btrfs_block_group_cache *cache;
6392 int ret;
6393
6394 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
6395 if (!cache)
6396 return -EINVAL;
6397
6398 /*
6399 * pull in the free space cache (if any) so that our pin
6400 * removes the free space from the cache. We have load_only set
6401 * to one because the slow code to read in the free extents does check
6402 * the pinned extents.
6403 */
6404 cache_block_group(cache, 1);
6405
6406 pin_down_extent(root, cache, bytenr, num_bytes, 0);
6407
6408 /* remove us from the free space cache (if we're there at all) */
6409 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6410 btrfs_put_block_group(cache);
6411 return ret;
6412 }
6413
6414 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
6415 {
6416 int ret;
6417 struct btrfs_block_group_cache *block_group;
6418 struct btrfs_caching_control *caching_ctl;
6419
6420 block_group = btrfs_lookup_block_group(root->fs_info, start);
6421 if (!block_group)
6422 return -EINVAL;
6423
6424 cache_block_group(block_group, 0);
6425 caching_ctl = get_caching_control(block_group);
6426
6427 if (!caching_ctl) {
6428 /* Logic error */
6429 BUG_ON(!block_group_cache_done(block_group));
6430 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6431 } else {
6432 mutex_lock(&caching_ctl->mutex);
6433
6434 if (start >= caching_ctl->progress) {
6435 ret = add_excluded_extent(root, start, num_bytes);
6436 } else if (start + num_bytes <= caching_ctl->progress) {
6437 ret = btrfs_remove_free_space(block_group,
6438 start, num_bytes);
6439 } else {
6440 num_bytes = caching_ctl->progress - start;
6441 ret = btrfs_remove_free_space(block_group,
6442 start, num_bytes);
6443 if (ret)
6444 goto out_lock;
6445
6446 num_bytes = (start + num_bytes) -
6447 caching_ctl->progress;
6448 start = caching_ctl->progress;
6449 ret = add_excluded_extent(root, start, num_bytes);
6450 }
6451 out_lock:
6452 mutex_unlock(&caching_ctl->mutex);
6453 put_caching_control(caching_ctl);
6454 }
6455 btrfs_put_block_group(block_group);
6456 return ret;
6457 }
6458
6459 int btrfs_exclude_logged_extents(struct btrfs_root *log,
6460 struct extent_buffer *eb)
6461 {
6462 struct btrfs_file_extent_item *item;
6463 struct btrfs_key key;
6464 int found_type;
6465 int i;
6466
6467 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
6468 return 0;
6469
6470 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6471 btrfs_item_key_to_cpu(eb, &key, i);
6472 if (key.type != BTRFS_EXTENT_DATA_KEY)
6473 continue;
6474 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6475 found_type = btrfs_file_extent_type(eb, item);
6476 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6477 continue;
6478 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6479 continue;
6480 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6481 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6482 __exclude_logged_extent(log, key.objectid, key.offset);
6483 }
6484
6485 return 0;
6486 }
6487
6488 static void
6489 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
6490 {
6491 atomic_inc(&bg->reservations);
6492 }
6493
6494 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
6495 const u64 start)
6496 {
6497 struct btrfs_block_group_cache *bg;
6498
6499 bg = btrfs_lookup_block_group(fs_info, start);
6500 ASSERT(bg);
6501 if (atomic_dec_and_test(&bg->reservations))
6502 wake_up_atomic_t(&bg->reservations);
6503 btrfs_put_block_group(bg);
6504 }
6505
6506 static int btrfs_wait_bg_reservations_atomic_t(atomic_t *a)
6507 {
6508 schedule();
6509 return 0;
6510 }
6511
6512 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg)
6513 {
6514 struct btrfs_space_info *space_info = bg->space_info;
6515
6516 ASSERT(bg->ro);
6517
6518 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
6519 return;
6520
6521 /*
6522 * Our block group is read only but before we set it to read only,
6523 * some task might have had allocated an extent from it already, but it
6524 * has not yet created a respective ordered extent (and added it to a
6525 * root's list of ordered extents).
6526 * Therefore wait for any task currently allocating extents, since the
6527 * block group's reservations counter is incremented while a read lock
6528 * on the groups' semaphore is held and decremented after releasing
6529 * the read access on that semaphore and creating the ordered extent.
6530 */
6531 down_write(&space_info->groups_sem);
6532 up_write(&space_info->groups_sem);
6533
6534 wait_on_atomic_t(&bg->reservations,
6535 btrfs_wait_bg_reservations_atomic_t,
6536 TASK_UNINTERRUPTIBLE);
6537 }
6538
6539 /**
6540 * btrfs_add_reserved_bytes - update the block_group and space info counters
6541 * @cache: The cache we are manipulating
6542 * @ram_bytes: The number of bytes of file content, and will be same to
6543 * @num_bytes except for the compress path.
6544 * @num_bytes: The number of bytes in question
6545 * @delalloc: The blocks are allocated for the delalloc write
6546 *
6547 * This is called by the allocator when it reserves space. Metadata
6548 * reservations should be called with RESERVE_ALLOC so we do the proper
6549 * ENOSPC accounting. For data we handle the reservation through clearing the
6550 * delalloc bits in the io_tree. We have to do this since we could end up
6551 * allocating less disk space for the amount of data we have reserved in the
6552 * case of compression.
6553 *
6554 * If this is a reservation and the block group has become read only we cannot
6555 * make the reservation and return -EAGAIN, otherwise this function always
6556 * succeeds.
6557 */
6558 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
6559 u64 ram_bytes, u64 num_bytes, int delalloc)
6560 {
6561 struct btrfs_space_info *space_info = cache->space_info;
6562 int ret = 0;
6563
6564 spin_lock(&space_info->lock);
6565 spin_lock(&cache->lock);
6566 if (cache->ro) {
6567 ret = -EAGAIN;
6568 } else {
6569 cache->reserved += num_bytes;
6570 space_info->bytes_reserved += num_bytes;
6571
6572 trace_btrfs_space_reservation(cache->fs_info,
6573 "space_info", space_info->flags,
6574 ram_bytes, 0);
6575 space_info->bytes_may_use -= ram_bytes;
6576 if (delalloc)
6577 cache->delalloc_bytes += num_bytes;
6578 }
6579 spin_unlock(&cache->lock);
6580 spin_unlock(&space_info->lock);
6581 return ret;
6582 }
6583
6584 /**
6585 * btrfs_free_reserved_bytes - update the block_group and space info counters
6586 * @cache: The cache we are manipulating
6587 * @num_bytes: The number of bytes in question
6588 * @delalloc: The blocks are allocated for the delalloc write
6589 *
6590 * This is called by somebody who is freeing space that was never actually used
6591 * on disk. For example if you reserve some space for a new leaf in transaction
6592 * A and before transaction A commits you free that leaf, you call this with
6593 * reserve set to 0 in order to clear the reservation.
6594 */
6595
6596 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
6597 u64 num_bytes, int delalloc)
6598 {
6599 struct btrfs_space_info *space_info = cache->space_info;
6600 int ret = 0;
6601
6602 spin_lock(&space_info->lock);
6603 spin_lock(&cache->lock);
6604 if (cache->ro)
6605 space_info->bytes_readonly += num_bytes;
6606 cache->reserved -= num_bytes;
6607 space_info->bytes_reserved -= num_bytes;
6608
6609 if (delalloc)
6610 cache->delalloc_bytes -= num_bytes;
6611 spin_unlock(&cache->lock);
6612 spin_unlock(&space_info->lock);
6613 return ret;
6614 }
6615 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
6616 struct btrfs_root *root)
6617 {
6618 struct btrfs_fs_info *fs_info = root->fs_info;
6619 struct btrfs_caching_control *next;
6620 struct btrfs_caching_control *caching_ctl;
6621 struct btrfs_block_group_cache *cache;
6622
6623 down_write(&fs_info->commit_root_sem);
6624
6625 list_for_each_entry_safe(caching_ctl, next,
6626 &fs_info->caching_block_groups, list) {
6627 cache = caching_ctl->block_group;
6628 if (block_group_cache_done(cache)) {
6629 cache->last_byte_to_unpin = (u64)-1;
6630 list_del_init(&caching_ctl->list);
6631 put_caching_control(caching_ctl);
6632 } else {
6633 cache->last_byte_to_unpin = caching_ctl->progress;
6634 }
6635 }
6636
6637 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6638 fs_info->pinned_extents = &fs_info->freed_extents[1];
6639 else
6640 fs_info->pinned_extents = &fs_info->freed_extents[0];
6641
6642 up_write(&fs_info->commit_root_sem);
6643
6644 update_global_block_rsv(fs_info);
6645 }
6646
6647 /*
6648 * Returns the free cluster for the given space info and sets empty_cluster to
6649 * what it should be based on the mount options.
6650 */
6651 static struct btrfs_free_cluster *
6652 fetch_cluster_info(struct btrfs_root *root, struct btrfs_space_info *space_info,
6653 u64 *empty_cluster)
6654 {
6655 struct btrfs_free_cluster *ret = NULL;
6656 bool ssd = btrfs_test_opt(root->fs_info, SSD);
6657
6658 *empty_cluster = 0;
6659 if (btrfs_mixed_space_info(space_info))
6660 return ret;
6661
6662 if (ssd)
6663 *empty_cluster = SZ_2M;
6664 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6665 ret = &root->fs_info->meta_alloc_cluster;
6666 if (!ssd)
6667 *empty_cluster = SZ_64K;
6668 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && ssd) {
6669 ret = &root->fs_info->data_alloc_cluster;
6670 }
6671
6672 return ret;
6673 }
6674
6675 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
6676 const bool return_free_space)
6677 {
6678 struct btrfs_fs_info *fs_info = root->fs_info;
6679 struct btrfs_block_group_cache *cache = NULL;
6680 struct btrfs_space_info *space_info;
6681 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6682 struct btrfs_free_cluster *cluster = NULL;
6683 u64 len;
6684 u64 total_unpinned = 0;
6685 u64 empty_cluster = 0;
6686 bool readonly;
6687
6688 while (start <= end) {
6689 readonly = false;
6690 if (!cache ||
6691 start >= cache->key.objectid + cache->key.offset) {
6692 if (cache)
6693 btrfs_put_block_group(cache);
6694 total_unpinned = 0;
6695 cache = btrfs_lookup_block_group(fs_info, start);
6696 BUG_ON(!cache); /* Logic error */
6697
6698 cluster = fetch_cluster_info(root,
6699 cache->space_info,
6700 &empty_cluster);
6701 empty_cluster <<= 1;
6702 }
6703
6704 len = cache->key.objectid + cache->key.offset - start;
6705 len = min(len, end + 1 - start);
6706
6707 if (start < cache->last_byte_to_unpin) {
6708 len = min(len, cache->last_byte_to_unpin - start);
6709 if (return_free_space)
6710 btrfs_add_free_space(cache, start, len);
6711 }
6712
6713 start += len;
6714 total_unpinned += len;
6715 space_info = cache->space_info;
6716
6717 /*
6718 * If this space cluster has been marked as fragmented and we've
6719 * unpinned enough in this block group to potentially allow a
6720 * cluster to be created inside of it go ahead and clear the
6721 * fragmented check.
6722 */
6723 if (cluster && cluster->fragmented &&
6724 total_unpinned > empty_cluster) {
6725 spin_lock(&cluster->lock);
6726 cluster->fragmented = 0;
6727 spin_unlock(&cluster->lock);
6728 }
6729
6730 spin_lock(&space_info->lock);
6731 spin_lock(&cache->lock);
6732 cache->pinned -= len;
6733 space_info->bytes_pinned -= len;
6734
6735 trace_btrfs_space_reservation(fs_info, "pinned",
6736 space_info->flags, len, 0);
6737 space_info->max_extent_size = 0;
6738 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6739 if (cache->ro) {
6740 space_info->bytes_readonly += len;
6741 readonly = true;
6742 }
6743 spin_unlock(&cache->lock);
6744 if (!readonly && return_free_space &&
6745 global_rsv->space_info == space_info) {
6746 u64 to_add = len;
6747 WARN_ON(!return_free_space);
6748 spin_lock(&global_rsv->lock);
6749 if (!global_rsv->full) {
6750 to_add = min(len, global_rsv->size -
6751 global_rsv->reserved);
6752 global_rsv->reserved += to_add;
6753 space_info->bytes_may_use += to_add;
6754 if (global_rsv->reserved >= global_rsv->size)
6755 global_rsv->full = 1;
6756 trace_btrfs_space_reservation(fs_info,
6757 "space_info",
6758 space_info->flags,
6759 to_add, 1);
6760 len -= to_add;
6761 }
6762 spin_unlock(&global_rsv->lock);
6763 /* Add to any tickets we may have */
6764 if (len)
6765 space_info_add_new_bytes(fs_info, space_info,
6766 len);
6767 }
6768 spin_unlock(&space_info->lock);
6769 }
6770
6771 if (cache)
6772 btrfs_put_block_group(cache);
6773 return 0;
6774 }
6775
6776 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6777 struct btrfs_root *root)
6778 {
6779 struct btrfs_fs_info *fs_info = root->fs_info;
6780 struct btrfs_block_group_cache *block_group, *tmp;
6781 struct list_head *deleted_bgs;
6782 struct extent_io_tree *unpin;
6783 u64 start;
6784 u64 end;
6785 int ret;
6786
6787 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6788 unpin = &fs_info->freed_extents[1];
6789 else
6790 unpin = &fs_info->freed_extents[0];
6791
6792 while (!trans->aborted) {
6793 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6794 ret = find_first_extent_bit(unpin, 0, &start, &end,
6795 EXTENT_DIRTY, NULL);
6796 if (ret) {
6797 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6798 break;
6799 }
6800
6801 if (btrfs_test_opt(root->fs_info, DISCARD))
6802 ret = btrfs_discard_extent(root, start,
6803 end + 1 - start, NULL);
6804
6805 clear_extent_dirty(unpin, start, end);
6806 unpin_extent_range(root, start, end, true);
6807 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6808 cond_resched();
6809 }
6810
6811 /*
6812 * Transaction is finished. We don't need the lock anymore. We
6813 * do need to clean up the block groups in case of a transaction
6814 * abort.
6815 */
6816 deleted_bgs = &trans->transaction->deleted_bgs;
6817 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6818 u64 trimmed = 0;
6819
6820 ret = -EROFS;
6821 if (!trans->aborted)
6822 ret = btrfs_discard_extent(root,
6823 block_group->key.objectid,
6824 block_group->key.offset,
6825 &trimmed);
6826
6827 list_del_init(&block_group->bg_list);
6828 btrfs_put_block_group_trimming(block_group);
6829 btrfs_put_block_group(block_group);
6830
6831 if (ret) {
6832 const char *errstr = btrfs_decode_error(ret);
6833 btrfs_warn(fs_info,
6834 "Discard failed while removing blockgroup: errno=%d %s\n",
6835 ret, errstr);
6836 }
6837 }
6838
6839 return 0;
6840 }
6841
6842 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
6843 u64 owner, u64 root_objectid)
6844 {
6845 struct btrfs_space_info *space_info;
6846 u64 flags;
6847
6848 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6849 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
6850 flags = BTRFS_BLOCK_GROUP_SYSTEM;
6851 else
6852 flags = BTRFS_BLOCK_GROUP_METADATA;
6853 } else {
6854 flags = BTRFS_BLOCK_GROUP_DATA;
6855 }
6856
6857 space_info = __find_space_info(fs_info, flags);
6858 BUG_ON(!space_info); /* Logic bug */
6859 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
6860 }
6861
6862
6863 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6864 struct btrfs_root *root,
6865 struct btrfs_delayed_ref_node *node, u64 parent,
6866 u64 root_objectid, u64 owner_objectid,
6867 u64 owner_offset, int refs_to_drop,
6868 struct btrfs_delayed_extent_op *extent_op)
6869 {
6870 struct btrfs_key key;
6871 struct btrfs_path *path;
6872 struct btrfs_fs_info *info = root->fs_info;
6873 struct btrfs_root *extent_root = info->extent_root;
6874 struct extent_buffer *leaf;
6875 struct btrfs_extent_item *ei;
6876 struct btrfs_extent_inline_ref *iref;
6877 int ret;
6878 int is_data;
6879 int extent_slot = 0;
6880 int found_extent = 0;
6881 int num_to_del = 1;
6882 u32 item_size;
6883 u64 refs;
6884 u64 bytenr = node->bytenr;
6885 u64 num_bytes = node->num_bytes;
6886 int last_ref = 0;
6887 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6888 SKINNY_METADATA);
6889
6890 path = btrfs_alloc_path();
6891 if (!path)
6892 return -ENOMEM;
6893
6894 path->reada = READA_FORWARD;
6895 path->leave_spinning = 1;
6896
6897 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6898 BUG_ON(!is_data && refs_to_drop != 1);
6899
6900 if (is_data)
6901 skinny_metadata = 0;
6902
6903 ret = lookup_extent_backref(trans, extent_root, path, &iref,
6904 bytenr, num_bytes, parent,
6905 root_objectid, owner_objectid,
6906 owner_offset);
6907 if (ret == 0) {
6908 extent_slot = path->slots[0];
6909 while (extent_slot >= 0) {
6910 btrfs_item_key_to_cpu(path->nodes[0], &key,
6911 extent_slot);
6912 if (key.objectid != bytenr)
6913 break;
6914 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6915 key.offset == num_bytes) {
6916 found_extent = 1;
6917 break;
6918 }
6919 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6920 key.offset == owner_objectid) {
6921 found_extent = 1;
6922 break;
6923 }
6924 if (path->slots[0] - extent_slot > 5)
6925 break;
6926 extent_slot--;
6927 }
6928 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6929 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6930 if (found_extent && item_size < sizeof(*ei))
6931 found_extent = 0;
6932 #endif
6933 if (!found_extent) {
6934 BUG_ON(iref);
6935 ret = remove_extent_backref(trans, extent_root, path,
6936 NULL, refs_to_drop,
6937 is_data, &last_ref);
6938 if (ret) {
6939 btrfs_abort_transaction(trans, ret);
6940 goto out;
6941 }
6942 btrfs_release_path(path);
6943 path->leave_spinning = 1;
6944
6945 key.objectid = bytenr;
6946 key.type = BTRFS_EXTENT_ITEM_KEY;
6947 key.offset = num_bytes;
6948
6949 if (!is_data && skinny_metadata) {
6950 key.type = BTRFS_METADATA_ITEM_KEY;
6951 key.offset = owner_objectid;
6952 }
6953
6954 ret = btrfs_search_slot(trans, extent_root,
6955 &key, path, -1, 1);
6956 if (ret > 0 && skinny_metadata && path->slots[0]) {
6957 /*
6958 * Couldn't find our skinny metadata item,
6959 * see if we have ye olde extent item.
6960 */
6961 path->slots[0]--;
6962 btrfs_item_key_to_cpu(path->nodes[0], &key,
6963 path->slots[0]);
6964 if (key.objectid == bytenr &&
6965 key.type == BTRFS_EXTENT_ITEM_KEY &&
6966 key.offset == num_bytes)
6967 ret = 0;
6968 }
6969
6970 if (ret > 0 && skinny_metadata) {
6971 skinny_metadata = false;
6972 key.objectid = bytenr;
6973 key.type = BTRFS_EXTENT_ITEM_KEY;
6974 key.offset = num_bytes;
6975 btrfs_release_path(path);
6976 ret = btrfs_search_slot(trans, extent_root,
6977 &key, path, -1, 1);
6978 }
6979
6980 if (ret) {
6981 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6982 ret, bytenr);
6983 if (ret > 0)
6984 btrfs_print_leaf(extent_root,
6985 path->nodes[0]);
6986 }
6987 if (ret < 0) {
6988 btrfs_abort_transaction(trans, ret);
6989 goto out;
6990 }
6991 extent_slot = path->slots[0];
6992 }
6993 } else if (WARN_ON(ret == -ENOENT)) {
6994 btrfs_print_leaf(extent_root, path->nodes[0]);
6995 btrfs_err(info,
6996 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6997 bytenr, parent, root_objectid, owner_objectid,
6998 owner_offset);
6999 btrfs_abort_transaction(trans, ret);
7000 goto out;
7001 } else {
7002 btrfs_abort_transaction(trans, ret);
7003 goto out;
7004 }
7005
7006 leaf = path->nodes[0];
7007 item_size = btrfs_item_size_nr(leaf, extent_slot);
7008 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
7009 if (item_size < sizeof(*ei)) {
7010 BUG_ON(found_extent || extent_slot != path->slots[0]);
7011 ret = convert_extent_item_v0(trans, extent_root, path,
7012 owner_objectid, 0);
7013 if (ret < 0) {
7014 btrfs_abort_transaction(trans, ret);
7015 goto out;
7016 }
7017
7018 btrfs_release_path(path);
7019 path->leave_spinning = 1;
7020
7021 key.objectid = bytenr;
7022 key.type = BTRFS_EXTENT_ITEM_KEY;
7023 key.offset = num_bytes;
7024
7025 ret = btrfs_search_slot(trans, extent_root, &key, path,
7026 -1, 1);
7027 if (ret) {
7028 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
7029 ret, bytenr);
7030 btrfs_print_leaf(extent_root, path->nodes[0]);
7031 }
7032 if (ret < 0) {
7033 btrfs_abort_transaction(trans, ret);
7034 goto out;
7035 }
7036
7037 extent_slot = path->slots[0];
7038 leaf = path->nodes[0];
7039 item_size = btrfs_item_size_nr(leaf, extent_slot);
7040 }
7041 #endif
7042 BUG_ON(item_size < sizeof(*ei));
7043 ei = btrfs_item_ptr(leaf, extent_slot,
7044 struct btrfs_extent_item);
7045 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
7046 key.type == BTRFS_EXTENT_ITEM_KEY) {
7047 struct btrfs_tree_block_info *bi;
7048 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
7049 bi = (struct btrfs_tree_block_info *)(ei + 1);
7050 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
7051 }
7052
7053 refs = btrfs_extent_refs(leaf, ei);
7054 if (refs < refs_to_drop) {
7055 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
7056 "for bytenr %Lu", refs_to_drop, refs, bytenr);
7057 ret = -EINVAL;
7058 btrfs_abort_transaction(trans, ret);
7059 goto out;
7060 }
7061 refs -= refs_to_drop;
7062
7063 if (refs > 0) {
7064 if (extent_op)
7065 __run_delayed_extent_op(extent_op, leaf, ei);
7066 /*
7067 * In the case of inline back ref, reference count will
7068 * be updated by remove_extent_backref
7069 */
7070 if (iref) {
7071 BUG_ON(!found_extent);
7072 } else {
7073 btrfs_set_extent_refs(leaf, ei, refs);
7074 btrfs_mark_buffer_dirty(leaf);
7075 }
7076 if (found_extent) {
7077 ret = remove_extent_backref(trans, extent_root, path,
7078 iref, refs_to_drop,
7079 is_data, &last_ref);
7080 if (ret) {
7081 btrfs_abort_transaction(trans, ret);
7082 goto out;
7083 }
7084 }
7085 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
7086 root_objectid);
7087 } else {
7088 if (found_extent) {
7089 BUG_ON(is_data && refs_to_drop !=
7090 extent_data_ref_count(path, iref));
7091 if (iref) {
7092 BUG_ON(path->slots[0] != extent_slot);
7093 } else {
7094 BUG_ON(path->slots[0] != extent_slot + 1);
7095 path->slots[0] = extent_slot;
7096 num_to_del = 2;
7097 }
7098 }
7099
7100 last_ref = 1;
7101 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
7102 num_to_del);
7103 if (ret) {
7104 btrfs_abort_transaction(trans, ret);
7105 goto out;
7106 }
7107 btrfs_release_path(path);
7108
7109 if (is_data) {
7110 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
7111 if (ret) {
7112 btrfs_abort_transaction(trans, ret);
7113 goto out;
7114 }
7115 }
7116
7117 ret = add_to_free_space_tree(trans, root->fs_info, bytenr,
7118 num_bytes);
7119 if (ret) {
7120 btrfs_abort_transaction(trans, ret);
7121 goto out;
7122 }
7123
7124 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
7125 if (ret) {
7126 btrfs_abort_transaction(trans, ret);
7127 goto out;
7128 }
7129 }
7130 btrfs_release_path(path);
7131
7132 out:
7133 btrfs_free_path(path);
7134 return ret;
7135 }
7136
7137 /*
7138 * when we free an block, it is possible (and likely) that we free the last
7139 * delayed ref for that extent as well. This searches the delayed ref tree for
7140 * a given extent, and if there are no other delayed refs to be processed, it
7141 * removes it from the tree.
7142 */
7143 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
7144 struct btrfs_root *root, u64 bytenr)
7145 {
7146 struct btrfs_delayed_ref_head *head;
7147 struct btrfs_delayed_ref_root *delayed_refs;
7148 int ret = 0;
7149
7150 delayed_refs = &trans->transaction->delayed_refs;
7151 spin_lock(&delayed_refs->lock);
7152 head = btrfs_find_delayed_ref_head(trans, bytenr);
7153 if (!head)
7154 goto out_delayed_unlock;
7155
7156 spin_lock(&head->lock);
7157 if (!list_empty(&head->ref_list))
7158 goto out;
7159
7160 if (head->extent_op) {
7161 if (!head->must_insert_reserved)
7162 goto out;
7163 btrfs_free_delayed_extent_op(head->extent_op);
7164 head->extent_op = NULL;
7165 }
7166
7167 /*
7168 * waiting for the lock here would deadlock. If someone else has it
7169 * locked they are already in the process of dropping it anyway
7170 */
7171 if (!mutex_trylock(&head->mutex))
7172 goto out;
7173
7174 /*
7175 * at this point we have a head with no other entries. Go
7176 * ahead and process it.
7177 */
7178 head->node.in_tree = 0;
7179 rb_erase(&head->href_node, &delayed_refs->href_root);
7180
7181 atomic_dec(&delayed_refs->num_entries);
7182
7183 /*
7184 * we don't take a ref on the node because we're removing it from the
7185 * tree, so we just steal the ref the tree was holding.
7186 */
7187 delayed_refs->num_heads--;
7188 if (head->processing == 0)
7189 delayed_refs->num_heads_ready--;
7190 head->processing = 0;
7191 spin_unlock(&head->lock);
7192 spin_unlock(&delayed_refs->lock);
7193
7194 BUG_ON(head->extent_op);
7195 if (head->must_insert_reserved)
7196 ret = 1;
7197
7198 mutex_unlock(&head->mutex);
7199 btrfs_put_delayed_ref(&head->node);
7200 return ret;
7201 out:
7202 spin_unlock(&head->lock);
7203
7204 out_delayed_unlock:
7205 spin_unlock(&delayed_refs->lock);
7206 return 0;
7207 }
7208
7209 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
7210 struct btrfs_root *root,
7211 struct extent_buffer *buf,
7212 u64 parent, int last_ref)
7213 {
7214 int pin = 1;
7215 int ret;
7216
7217 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7218 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7219 buf->start, buf->len,
7220 parent, root->root_key.objectid,
7221 btrfs_header_level(buf),
7222 BTRFS_DROP_DELAYED_REF, NULL);
7223 BUG_ON(ret); /* -ENOMEM */
7224 }
7225
7226 if (!last_ref)
7227 return;
7228
7229 if (btrfs_header_generation(buf) == trans->transid) {
7230 struct btrfs_block_group_cache *cache;
7231
7232 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7233 ret = check_ref_cleanup(trans, root, buf->start);
7234 if (!ret)
7235 goto out;
7236 }
7237
7238 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
7239
7240 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
7241 pin_down_extent(root, cache, buf->start, buf->len, 1);
7242 btrfs_put_block_group(cache);
7243 goto out;
7244 }
7245
7246 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
7247
7248 btrfs_add_free_space(cache, buf->start, buf->len);
7249 btrfs_free_reserved_bytes(cache, buf->len, 0);
7250 btrfs_put_block_group(cache);
7251 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
7252 pin = 0;
7253 }
7254 out:
7255 if (pin)
7256 add_pinned_bytes(root->fs_info, buf->len,
7257 btrfs_header_level(buf),
7258 root->root_key.objectid);
7259
7260 /*
7261 * Deleting the buffer, clear the corrupt flag since it doesn't matter
7262 * anymore.
7263 */
7264 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
7265 }
7266
7267 /* Can return -ENOMEM */
7268 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7269 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
7270 u64 owner, u64 offset)
7271 {
7272 int ret;
7273 struct btrfs_fs_info *fs_info = root->fs_info;
7274
7275 if (btrfs_is_testing(fs_info))
7276 return 0;
7277
7278 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
7279
7280 /*
7281 * tree log blocks never actually go into the extent allocation
7282 * tree, just update pinning info and exit early.
7283 */
7284 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
7285 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
7286 /* unlocks the pinned mutex */
7287 btrfs_pin_extent(root, bytenr, num_bytes, 1);
7288 ret = 0;
7289 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
7290 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
7291 num_bytes,
7292 parent, root_objectid, (int)owner,
7293 BTRFS_DROP_DELAYED_REF, NULL);
7294 } else {
7295 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
7296 num_bytes,
7297 parent, root_objectid, owner,
7298 offset, 0,
7299 BTRFS_DROP_DELAYED_REF, NULL);
7300 }
7301 return ret;
7302 }
7303
7304 /*
7305 * when we wait for progress in the block group caching, its because
7306 * our allocation attempt failed at least once. So, we must sleep
7307 * and let some progress happen before we try again.
7308 *
7309 * This function will sleep at least once waiting for new free space to
7310 * show up, and then it will check the block group free space numbers
7311 * for our min num_bytes. Another option is to have it go ahead
7312 * and look in the rbtree for a free extent of a given size, but this
7313 * is a good start.
7314 *
7315 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7316 * any of the information in this block group.
7317 */
7318 static noinline void
7319 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
7320 u64 num_bytes)
7321 {
7322 struct btrfs_caching_control *caching_ctl;
7323
7324 caching_ctl = get_caching_control(cache);
7325 if (!caching_ctl)
7326 return;
7327
7328 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
7329 (cache->free_space_ctl->free_space >= num_bytes));
7330
7331 put_caching_control(caching_ctl);
7332 }
7333
7334 static noinline int
7335 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
7336 {
7337 struct btrfs_caching_control *caching_ctl;
7338 int ret = 0;
7339
7340 caching_ctl = get_caching_control(cache);
7341 if (!caching_ctl)
7342 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
7343
7344 wait_event(caching_ctl->wait, block_group_cache_done(cache));
7345 if (cache->cached == BTRFS_CACHE_ERROR)
7346 ret = -EIO;
7347 put_caching_control(caching_ctl);
7348 return ret;
7349 }
7350
7351 int __get_raid_index(u64 flags)
7352 {
7353 if (flags & BTRFS_BLOCK_GROUP_RAID10)
7354 return BTRFS_RAID_RAID10;
7355 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
7356 return BTRFS_RAID_RAID1;
7357 else if (flags & BTRFS_BLOCK_GROUP_DUP)
7358 return BTRFS_RAID_DUP;
7359 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
7360 return BTRFS_RAID_RAID0;
7361 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
7362 return BTRFS_RAID_RAID5;
7363 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
7364 return BTRFS_RAID_RAID6;
7365
7366 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
7367 }
7368
7369 int get_block_group_index(struct btrfs_block_group_cache *cache)
7370 {
7371 return __get_raid_index(cache->flags);
7372 }
7373
7374 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
7375 [BTRFS_RAID_RAID10] = "raid10",
7376 [BTRFS_RAID_RAID1] = "raid1",
7377 [BTRFS_RAID_DUP] = "dup",
7378 [BTRFS_RAID_RAID0] = "raid0",
7379 [BTRFS_RAID_SINGLE] = "single",
7380 [BTRFS_RAID_RAID5] = "raid5",
7381 [BTRFS_RAID_RAID6] = "raid6",
7382 };
7383
7384 static const char *get_raid_name(enum btrfs_raid_types type)
7385 {
7386 if (type >= BTRFS_NR_RAID_TYPES)
7387 return NULL;
7388
7389 return btrfs_raid_type_names[type];
7390 }
7391
7392 enum btrfs_loop_type {
7393 LOOP_CACHING_NOWAIT = 0,
7394 LOOP_CACHING_WAIT = 1,
7395 LOOP_ALLOC_CHUNK = 2,
7396 LOOP_NO_EMPTY_SIZE = 3,
7397 };
7398
7399 static inline void
7400 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7401 int delalloc)
7402 {
7403 if (delalloc)
7404 down_read(&cache->data_rwsem);
7405 }
7406
7407 static inline void
7408 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7409 int delalloc)
7410 {
7411 btrfs_get_block_group(cache);
7412 if (delalloc)
7413 down_read(&cache->data_rwsem);
7414 }
7415
7416 static struct btrfs_block_group_cache *
7417 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7418 struct btrfs_free_cluster *cluster,
7419 int delalloc)
7420 {
7421 struct btrfs_block_group_cache *used_bg = NULL;
7422
7423 spin_lock(&cluster->refill_lock);
7424 while (1) {
7425 used_bg = cluster->block_group;
7426 if (!used_bg)
7427 return NULL;
7428
7429 if (used_bg == block_group)
7430 return used_bg;
7431
7432 btrfs_get_block_group(used_bg);
7433
7434 if (!delalloc)
7435 return used_bg;
7436
7437 if (down_read_trylock(&used_bg->data_rwsem))
7438 return used_bg;
7439
7440 spin_unlock(&cluster->refill_lock);
7441
7442 down_read(&used_bg->data_rwsem);
7443
7444 spin_lock(&cluster->refill_lock);
7445 if (used_bg == cluster->block_group)
7446 return used_bg;
7447
7448 up_read(&used_bg->data_rwsem);
7449 btrfs_put_block_group(used_bg);
7450 }
7451 }
7452
7453 static inline void
7454 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7455 int delalloc)
7456 {
7457 if (delalloc)
7458 up_read(&cache->data_rwsem);
7459 btrfs_put_block_group(cache);
7460 }
7461
7462 /*
7463 * walks the btree of allocated extents and find a hole of a given size.
7464 * The key ins is changed to record the hole:
7465 * ins->objectid == start position
7466 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7467 * ins->offset == the size of the hole.
7468 * Any available blocks before search_start are skipped.
7469 *
7470 * If there is no suitable free space, we will record the max size of
7471 * the free space extent currently.
7472 */
7473 static noinline int find_free_extent(struct btrfs_root *orig_root,
7474 u64 ram_bytes, u64 num_bytes, u64 empty_size,
7475 u64 hint_byte, struct btrfs_key *ins,
7476 u64 flags, int delalloc)
7477 {
7478 int ret = 0;
7479 struct btrfs_root *root = orig_root->fs_info->extent_root;
7480 struct btrfs_free_cluster *last_ptr = NULL;
7481 struct btrfs_block_group_cache *block_group = NULL;
7482 u64 search_start = 0;
7483 u64 max_extent_size = 0;
7484 u64 empty_cluster = 0;
7485 struct btrfs_space_info *space_info;
7486 int loop = 0;
7487 int index = __get_raid_index(flags);
7488 bool failed_cluster_refill = false;
7489 bool failed_alloc = false;
7490 bool use_cluster = true;
7491 bool have_caching_bg = false;
7492 bool orig_have_caching_bg = false;
7493 bool full_search = false;
7494
7495 WARN_ON(num_bytes < root->sectorsize);
7496 ins->type = BTRFS_EXTENT_ITEM_KEY;
7497 ins->objectid = 0;
7498 ins->offset = 0;
7499
7500 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
7501
7502 space_info = __find_space_info(root->fs_info, flags);
7503 if (!space_info) {
7504 btrfs_err(root->fs_info, "No space info for %llu", flags);
7505 return -ENOSPC;
7506 }
7507
7508 /*
7509 * If our free space is heavily fragmented we may not be able to make
7510 * big contiguous allocations, so instead of doing the expensive search
7511 * for free space, simply return ENOSPC with our max_extent_size so we
7512 * can go ahead and search for a more manageable chunk.
7513 *
7514 * If our max_extent_size is large enough for our allocation simply
7515 * disable clustering since we will likely not be able to find enough
7516 * space to create a cluster and induce latency trying.
7517 */
7518 if (unlikely(space_info->max_extent_size)) {
7519 spin_lock(&space_info->lock);
7520 if (space_info->max_extent_size &&
7521 num_bytes > space_info->max_extent_size) {
7522 ins->offset = space_info->max_extent_size;
7523 spin_unlock(&space_info->lock);
7524 return -ENOSPC;
7525 } else if (space_info->max_extent_size) {
7526 use_cluster = false;
7527 }
7528 spin_unlock(&space_info->lock);
7529 }
7530
7531 last_ptr = fetch_cluster_info(orig_root, space_info, &empty_cluster);
7532 if (last_ptr) {
7533 spin_lock(&last_ptr->lock);
7534 if (last_ptr->block_group)
7535 hint_byte = last_ptr->window_start;
7536 if (last_ptr->fragmented) {
7537 /*
7538 * We still set window_start so we can keep track of the
7539 * last place we found an allocation to try and save
7540 * some time.
7541 */
7542 hint_byte = last_ptr->window_start;
7543 use_cluster = false;
7544 }
7545 spin_unlock(&last_ptr->lock);
7546 }
7547
7548 search_start = max(search_start, first_logical_byte(root, 0));
7549 search_start = max(search_start, hint_byte);
7550 if (search_start == hint_byte) {
7551 block_group = btrfs_lookup_block_group(root->fs_info,
7552 search_start);
7553 /*
7554 * we don't want to use the block group if it doesn't match our
7555 * allocation bits, or if its not cached.
7556 *
7557 * However if we are re-searching with an ideal block group
7558 * picked out then we don't care that the block group is cached.
7559 */
7560 if (block_group && block_group_bits(block_group, flags) &&
7561 block_group->cached != BTRFS_CACHE_NO) {
7562 down_read(&space_info->groups_sem);
7563 if (list_empty(&block_group->list) ||
7564 block_group->ro) {
7565 /*
7566 * someone is removing this block group,
7567 * we can't jump into the have_block_group
7568 * target because our list pointers are not
7569 * valid
7570 */
7571 btrfs_put_block_group(block_group);
7572 up_read(&space_info->groups_sem);
7573 } else {
7574 index = get_block_group_index(block_group);
7575 btrfs_lock_block_group(block_group, delalloc);
7576 goto have_block_group;
7577 }
7578 } else if (block_group) {
7579 btrfs_put_block_group(block_group);
7580 }
7581 }
7582 search:
7583 have_caching_bg = false;
7584 if (index == 0 || index == __get_raid_index(flags))
7585 full_search = true;
7586 down_read(&space_info->groups_sem);
7587 list_for_each_entry(block_group, &space_info->block_groups[index],
7588 list) {
7589 u64 offset;
7590 int cached;
7591
7592 btrfs_grab_block_group(block_group, delalloc);
7593 search_start = block_group->key.objectid;
7594
7595 /*
7596 * this can happen if we end up cycling through all the
7597 * raid types, but we want to make sure we only allocate
7598 * for the proper type.
7599 */
7600 if (!block_group_bits(block_group, flags)) {
7601 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7602 BTRFS_BLOCK_GROUP_RAID1 |
7603 BTRFS_BLOCK_GROUP_RAID5 |
7604 BTRFS_BLOCK_GROUP_RAID6 |
7605 BTRFS_BLOCK_GROUP_RAID10;
7606
7607 /*
7608 * if they asked for extra copies and this block group
7609 * doesn't provide them, bail. This does allow us to
7610 * fill raid0 from raid1.
7611 */
7612 if ((flags & extra) && !(block_group->flags & extra))
7613 goto loop;
7614 }
7615
7616 have_block_group:
7617 cached = block_group_cache_done(block_group);
7618 if (unlikely(!cached)) {
7619 have_caching_bg = true;
7620 ret = cache_block_group(block_group, 0);
7621 BUG_ON(ret < 0);
7622 ret = 0;
7623 }
7624
7625 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7626 goto loop;
7627 if (unlikely(block_group->ro))
7628 goto loop;
7629
7630 /*
7631 * Ok we want to try and use the cluster allocator, so
7632 * lets look there
7633 */
7634 if (last_ptr && use_cluster) {
7635 struct btrfs_block_group_cache *used_block_group;
7636 unsigned long aligned_cluster;
7637 /*
7638 * the refill lock keeps out other
7639 * people trying to start a new cluster
7640 */
7641 used_block_group = btrfs_lock_cluster(block_group,
7642 last_ptr,
7643 delalloc);
7644 if (!used_block_group)
7645 goto refill_cluster;
7646
7647 if (used_block_group != block_group &&
7648 (used_block_group->ro ||
7649 !block_group_bits(used_block_group, flags)))
7650 goto release_cluster;
7651
7652 offset = btrfs_alloc_from_cluster(used_block_group,
7653 last_ptr,
7654 num_bytes,
7655 used_block_group->key.objectid,
7656 &max_extent_size);
7657 if (offset) {
7658 /* we have a block, we're done */
7659 spin_unlock(&last_ptr->refill_lock);
7660 trace_btrfs_reserve_extent_cluster(root,
7661 used_block_group,
7662 search_start, num_bytes);
7663 if (used_block_group != block_group) {
7664 btrfs_release_block_group(block_group,
7665 delalloc);
7666 block_group = used_block_group;
7667 }
7668 goto checks;
7669 }
7670
7671 WARN_ON(last_ptr->block_group != used_block_group);
7672 release_cluster:
7673 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7674 * set up a new clusters, so lets just skip it
7675 * and let the allocator find whatever block
7676 * it can find. If we reach this point, we
7677 * will have tried the cluster allocator
7678 * plenty of times and not have found
7679 * anything, so we are likely way too
7680 * fragmented for the clustering stuff to find
7681 * anything.
7682 *
7683 * However, if the cluster is taken from the
7684 * current block group, release the cluster
7685 * first, so that we stand a better chance of
7686 * succeeding in the unclustered
7687 * allocation. */
7688 if (loop >= LOOP_NO_EMPTY_SIZE &&
7689 used_block_group != block_group) {
7690 spin_unlock(&last_ptr->refill_lock);
7691 btrfs_release_block_group(used_block_group,
7692 delalloc);
7693 goto unclustered_alloc;
7694 }
7695
7696 /*
7697 * this cluster didn't work out, free it and
7698 * start over
7699 */
7700 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7701
7702 if (used_block_group != block_group)
7703 btrfs_release_block_group(used_block_group,
7704 delalloc);
7705 refill_cluster:
7706 if (loop >= LOOP_NO_EMPTY_SIZE) {
7707 spin_unlock(&last_ptr->refill_lock);
7708 goto unclustered_alloc;
7709 }
7710
7711 aligned_cluster = max_t(unsigned long,
7712 empty_cluster + empty_size,
7713 block_group->full_stripe_len);
7714
7715 /* allocate a cluster in this block group */
7716 ret = btrfs_find_space_cluster(root, block_group,
7717 last_ptr, search_start,
7718 num_bytes,
7719 aligned_cluster);
7720 if (ret == 0) {
7721 /*
7722 * now pull our allocation out of this
7723 * cluster
7724 */
7725 offset = btrfs_alloc_from_cluster(block_group,
7726 last_ptr,
7727 num_bytes,
7728 search_start,
7729 &max_extent_size);
7730 if (offset) {
7731 /* we found one, proceed */
7732 spin_unlock(&last_ptr->refill_lock);
7733 trace_btrfs_reserve_extent_cluster(root,
7734 block_group, search_start,
7735 num_bytes);
7736 goto checks;
7737 }
7738 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7739 && !failed_cluster_refill) {
7740 spin_unlock(&last_ptr->refill_lock);
7741
7742 failed_cluster_refill = true;
7743 wait_block_group_cache_progress(block_group,
7744 num_bytes + empty_cluster + empty_size);
7745 goto have_block_group;
7746 }
7747
7748 /*
7749 * at this point we either didn't find a cluster
7750 * or we weren't able to allocate a block from our
7751 * cluster. Free the cluster we've been trying
7752 * to use, and go to the next block group
7753 */
7754 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7755 spin_unlock(&last_ptr->refill_lock);
7756 goto loop;
7757 }
7758
7759 unclustered_alloc:
7760 /*
7761 * We are doing an unclustered alloc, set the fragmented flag so
7762 * we don't bother trying to setup a cluster again until we get
7763 * more space.
7764 */
7765 if (unlikely(last_ptr)) {
7766 spin_lock(&last_ptr->lock);
7767 last_ptr->fragmented = 1;
7768 spin_unlock(&last_ptr->lock);
7769 }
7770 spin_lock(&block_group->free_space_ctl->tree_lock);
7771 if (cached &&
7772 block_group->free_space_ctl->free_space <
7773 num_bytes + empty_cluster + empty_size) {
7774 if (block_group->free_space_ctl->free_space >
7775 max_extent_size)
7776 max_extent_size =
7777 block_group->free_space_ctl->free_space;
7778 spin_unlock(&block_group->free_space_ctl->tree_lock);
7779 goto loop;
7780 }
7781 spin_unlock(&block_group->free_space_ctl->tree_lock);
7782
7783 offset = btrfs_find_space_for_alloc(block_group, search_start,
7784 num_bytes, empty_size,
7785 &max_extent_size);
7786 /*
7787 * If we didn't find a chunk, and we haven't failed on this
7788 * block group before, and this block group is in the middle of
7789 * caching and we are ok with waiting, then go ahead and wait
7790 * for progress to be made, and set failed_alloc to true.
7791 *
7792 * If failed_alloc is true then we've already waited on this
7793 * block group once and should move on to the next block group.
7794 */
7795 if (!offset && !failed_alloc && !cached &&
7796 loop > LOOP_CACHING_NOWAIT) {
7797 wait_block_group_cache_progress(block_group,
7798 num_bytes + empty_size);
7799 failed_alloc = true;
7800 goto have_block_group;
7801 } else if (!offset) {
7802 goto loop;
7803 }
7804 checks:
7805 search_start = ALIGN(offset, root->stripesize);
7806
7807 /* move on to the next group */
7808 if (search_start + num_bytes >
7809 block_group->key.objectid + block_group->key.offset) {
7810 btrfs_add_free_space(block_group, offset, num_bytes);
7811 goto loop;
7812 }
7813
7814 if (offset < search_start)
7815 btrfs_add_free_space(block_group, offset,
7816 search_start - offset);
7817 BUG_ON(offset > search_start);
7818
7819 ret = btrfs_add_reserved_bytes(block_group, ram_bytes,
7820 num_bytes, delalloc);
7821 if (ret == -EAGAIN) {
7822 btrfs_add_free_space(block_group, offset, num_bytes);
7823 goto loop;
7824 }
7825 btrfs_inc_block_group_reservations(block_group);
7826
7827 /* we are all good, lets return */
7828 ins->objectid = search_start;
7829 ins->offset = num_bytes;
7830
7831 trace_btrfs_reserve_extent(orig_root, block_group,
7832 search_start, num_bytes);
7833 btrfs_release_block_group(block_group, delalloc);
7834 break;
7835 loop:
7836 failed_cluster_refill = false;
7837 failed_alloc = false;
7838 BUG_ON(index != get_block_group_index(block_group));
7839 btrfs_release_block_group(block_group, delalloc);
7840 }
7841 up_read(&space_info->groups_sem);
7842
7843 if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg
7844 && !orig_have_caching_bg)
7845 orig_have_caching_bg = true;
7846
7847 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7848 goto search;
7849
7850 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7851 goto search;
7852
7853 /*
7854 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7855 * caching kthreads as we move along
7856 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7857 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7858 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7859 * again
7860 */
7861 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7862 index = 0;
7863 if (loop == LOOP_CACHING_NOWAIT) {
7864 /*
7865 * We want to skip the LOOP_CACHING_WAIT step if we
7866 * don't have any uncached bgs and we've already done a
7867 * full search through.
7868 */
7869 if (orig_have_caching_bg || !full_search)
7870 loop = LOOP_CACHING_WAIT;
7871 else
7872 loop = LOOP_ALLOC_CHUNK;
7873 } else {
7874 loop++;
7875 }
7876
7877 if (loop == LOOP_ALLOC_CHUNK) {
7878 struct btrfs_trans_handle *trans;
7879 int exist = 0;
7880
7881 trans = current->journal_info;
7882 if (trans)
7883 exist = 1;
7884 else
7885 trans = btrfs_join_transaction(root);
7886
7887 if (IS_ERR(trans)) {
7888 ret = PTR_ERR(trans);
7889 goto out;
7890 }
7891
7892 ret = do_chunk_alloc(trans, root, flags,
7893 CHUNK_ALLOC_FORCE);
7894
7895 /*
7896 * If we can't allocate a new chunk we've already looped
7897 * through at least once, move on to the NO_EMPTY_SIZE
7898 * case.
7899 */
7900 if (ret == -ENOSPC)
7901 loop = LOOP_NO_EMPTY_SIZE;
7902
7903 /*
7904 * Do not bail out on ENOSPC since we
7905 * can do more things.
7906 */
7907 if (ret < 0 && ret != -ENOSPC)
7908 btrfs_abort_transaction(trans, ret);
7909 else
7910 ret = 0;
7911 if (!exist)
7912 btrfs_end_transaction(trans, root);
7913 if (ret)
7914 goto out;
7915 }
7916
7917 if (loop == LOOP_NO_EMPTY_SIZE) {
7918 /*
7919 * Don't loop again if we already have no empty_size and
7920 * no empty_cluster.
7921 */
7922 if (empty_size == 0 &&
7923 empty_cluster == 0) {
7924 ret = -ENOSPC;
7925 goto out;
7926 }
7927 empty_size = 0;
7928 empty_cluster = 0;
7929 }
7930
7931 goto search;
7932 } else if (!ins->objectid) {
7933 ret = -ENOSPC;
7934 } else if (ins->objectid) {
7935 if (!use_cluster && last_ptr) {
7936 spin_lock(&last_ptr->lock);
7937 last_ptr->window_start = ins->objectid;
7938 spin_unlock(&last_ptr->lock);
7939 }
7940 ret = 0;
7941 }
7942 out:
7943 if (ret == -ENOSPC) {
7944 spin_lock(&space_info->lock);
7945 space_info->max_extent_size = max_extent_size;
7946 spin_unlock(&space_info->lock);
7947 ins->offset = max_extent_size;
7948 }
7949 return ret;
7950 }
7951
7952 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
7953 int dump_block_groups)
7954 {
7955 struct btrfs_block_group_cache *cache;
7956 int index = 0;
7957
7958 spin_lock(&info->lock);
7959 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
7960 info->flags,
7961 info->total_bytes - info->bytes_used - info->bytes_pinned -
7962 info->bytes_reserved - info->bytes_readonly -
7963 info->bytes_may_use, (info->full) ? "" : "not ");
7964 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7965 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7966 info->total_bytes, info->bytes_used, info->bytes_pinned,
7967 info->bytes_reserved, info->bytes_may_use,
7968 info->bytes_readonly);
7969 spin_unlock(&info->lock);
7970
7971 if (!dump_block_groups)
7972 return;
7973
7974 down_read(&info->groups_sem);
7975 again:
7976 list_for_each_entry(cache, &info->block_groups[index], list) {
7977 spin_lock(&cache->lock);
7978 printk(KERN_INFO "BTRFS: "
7979 "block group %llu has %llu bytes, "
7980 "%llu used %llu pinned %llu reserved %s\n",
7981 cache->key.objectid, cache->key.offset,
7982 btrfs_block_group_used(&cache->item), cache->pinned,
7983 cache->reserved, cache->ro ? "[readonly]" : "");
7984 btrfs_dump_free_space(cache, bytes);
7985 spin_unlock(&cache->lock);
7986 }
7987 if (++index < BTRFS_NR_RAID_TYPES)
7988 goto again;
7989 up_read(&info->groups_sem);
7990 }
7991
7992 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
7993 u64 num_bytes, u64 min_alloc_size,
7994 u64 empty_size, u64 hint_byte,
7995 struct btrfs_key *ins, int is_data, int delalloc)
7996 {
7997 bool final_tried = num_bytes == min_alloc_size;
7998 u64 flags;
7999 int ret;
8000
8001 flags = btrfs_get_alloc_profile(root, is_data);
8002 again:
8003 WARN_ON(num_bytes < root->sectorsize);
8004 ret = find_free_extent(root, ram_bytes, num_bytes, empty_size,
8005 hint_byte, ins, flags, delalloc);
8006 if (!ret && !is_data) {
8007 btrfs_dec_block_group_reservations(root->fs_info,
8008 ins->objectid);
8009 } else if (ret == -ENOSPC) {
8010 if (!final_tried && ins->offset) {
8011 num_bytes = min(num_bytes >> 1, ins->offset);
8012 num_bytes = round_down(num_bytes, root->sectorsize);
8013 num_bytes = max(num_bytes, min_alloc_size);
8014 ram_bytes = num_bytes;
8015 if (num_bytes == min_alloc_size)
8016 final_tried = true;
8017 goto again;
8018 } else if (btrfs_test_opt(root->fs_info, ENOSPC_DEBUG)) {
8019 struct btrfs_space_info *sinfo;
8020
8021 sinfo = __find_space_info(root->fs_info, flags);
8022 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
8023 flags, num_bytes);
8024 if (sinfo)
8025 dump_space_info(sinfo, num_bytes, 1);
8026 }
8027 }
8028
8029 return ret;
8030 }
8031
8032 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
8033 u64 start, u64 len,
8034 int pin, int delalloc)
8035 {
8036 struct btrfs_block_group_cache *cache;
8037 int ret = 0;
8038
8039 cache = btrfs_lookup_block_group(root->fs_info, start);
8040 if (!cache) {
8041 btrfs_err(root->fs_info, "Unable to find block group for %llu",
8042 start);
8043 return -ENOSPC;
8044 }
8045
8046 if (pin)
8047 pin_down_extent(root, cache, start, len, 1);
8048 else {
8049 if (btrfs_test_opt(root->fs_info, DISCARD))
8050 ret = btrfs_discard_extent(root, start, len, NULL);
8051 btrfs_add_free_space(cache, start, len);
8052 btrfs_free_reserved_bytes(cache, len, delalloc);
8053 trace_btrfs_reserved_extent_free(root, start, len);
8054 }
8055
8056 btrfs_put_block_group(cache);
8057 return ret;
8058 }
8059
8060 int btrfs_free_reserved_extent(struct btrfs_root *root,
8061 u64 start, u64 len, int delalloc)
8062 {
8063 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
8064 }
8065
8066 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
8067 u64 start, u64 len)
8068 {
8069 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
8070 }
8071
8072 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8073 struct btrfs_root *root,
8074 u64 parent, u64 root_objectid,
8075 u64 flags, u64 owner, u64 offset,
8076 struct btrfs_key *ins, int ref_mod)
8077 {
8078 int ret;
8079 struct btrfs_fs_info *fs_info = root->fs_info;
8080 struct btrfs_extent_item *extent_item;
8081 struct btrfs_extent_inline_ref *iref;
8082 struct btrfs_path *path;
8083 struct extent_buffer *leaf;
8084 int type;
8085 u32 size;
8086
8087 if (parent > 0)
8088 type = BTRFS_SHARED_DATA_REF_KEY;
8089 else
8090 type = BTRFS_EXTENT_DATA_REF_KEY;
8091
8092 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
8093
8094 path = btrfs_alloc_path();
8095 if (!path)
8096 return -ENOMEM;
8097
8098 path->leave_spinning = 1;
8099 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8100 ins, size);
8101 if (ret) {
8102 btrfs_free_path(path);
8103 return ret;
8104 }
8105
8106 leaf = path->nodes[0];
8107 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8108 struct btrfs_extent_item);
8109 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
8110 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8111 btrfs_set_extent_flags(leaf, extent_item,
8112 flags | BTRFS_EXTENT_FLAG_DATA);
8113
8114 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8115 btrfs_set_extent_inline_ref_type(leaf, iref, type);
8116 if (parent > 0) {
8117 struct btrfs_shared_data_ref *ref;
8118 ref = (struct btrfs_shared_data_ref *)(iref + 1);
8119 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8120 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
8121 } else {
8122 struct btrfs_extent_data_ref *ref;
8123 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
8124 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
8125 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
8126 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
8127 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
8128 }
8129
8130 btrfs_mark_buffer_dirty(path->nodes[0]);
8131 btrfs_free_path(path);
8132
8133 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8134 ins->offset);
8135 if (ret)
8136 return ret;
8137
8138 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
8139 if (ret) { /* -ENOENT, logic error */
8140 btrfs_err(fs_info, "update block group failed for %llu %llu",
8141 ins->objectid, ins->offset);
8142 BUG();
8143 }
8144 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
8145 return ret;
8146 }
8147
8148 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
8149 struct btrfs_root *root,
8150 u64 parent, u64 root_objectid,
8151 u64 flags, struct btrfs_disk_key *key,
8152 int level, struct btrfs_key *ins)
8153 {
8154 int ret;
8155 struct btrfs_fs_info *fs_info = root->fs_info;
8156 struct btrfs_extent_item *extent_item;
8157 struct btrfs_tree_block_info *block_info;
8158 struct btrfs_extent_inline_ref *iref;
8159 struct btrfs_path *path;
8160 struct extent_buffer *leaf;
8161 u32 size = sizeof(*extent_item) + sizeof(*iref);
8162 u64 num_bytes = ins->offset;
8163 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
8164 SKINNY_METADATA);
8165
8166 if (!skinny_metadata)
8167 size += sizeof(*block_info);
8168
8169 path = btrfs_alloc_path();
8170 if (!path) {
8171 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
8172 root->nodesize);
8173 return -ENOMEM;
8174 }
8175
8176 path->leave_spinning = 1;
8177 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8178 ins, size);
8179 if (ret) {
8180 btrfs_free_path(path);
8181 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
8182 root->nodesize);
8183 return ret;
8184 }
8185
8186 leaf = path->nodes[0];
8187 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8188 struct btrfs_extent_item);
8189 btrfs_set_extent_refs(leaf, extent_item, 1);
8190 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8191 btrfs_set_extent_flags(leaf, extent_item,
8192 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
8193
8194 if (skinny_metadata) {
8195 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8196 num_bytes = root->nodesize;
8197 } else {
8198 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
8199 btrfs_set_tree_block_key(leaf, block_info, key);
8200 btrfs_set_tree_block_level(leaf, block_info, level);
8201 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
8202 }
8203
8204 if (parent > 0) {
8205 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
8206 btrfs_set_extent_inline_ref_type(leaf, iref,
8207 BTRFS_SHARED_BLOCK_REF_KEY);
8208 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8209 } else {
8210 btrfs_set_extent_inline_ref_type(leaf, iref,
8211 BTRFS_TREE_BLOCK_REF_KEY);
8212 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
8213 }
8214
8215 btrfs_mark_buffer_dirty(leaf);
8216 btrfs_free_path(path);
8217
8218 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8219 num_bytes);
8220 if (ret)
8221 return ret;
8222
8223 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
8224 1);
8225 if (ret) { /* -ENOENT, logic error */
8226 btrfs_err(fs_info, "update block group failed for %llu %llu",
8227 ins->objectid, ins->offset);
8228 BUG();
8229 }
8230
8231 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
8232 return ret;
8233 }
8234
8235 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8236 struct btrfs_root *root,
8237 u64 root_objectid, u64 owner,
8238 u64 offset, u64 ram_bytes,
8239 struct btrfs_key *ins)
8240 {
8241 int ret;
8242
8243 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
8244
8245 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
8246 ins->offset, 0,
8247 root_objectid, owner, offset,
8248 ram_bytes, BTRFS_ADD_DELAYED_EXTENT,
8249 NULL);
8250 return ret;
8251 }
8252
8253 /*
8254 * this is used by the tree logging recovery code. It records that
8255 * an extent has been allocated and makes sure to clear the free
8256 * space cache bits as well
8257 */
8258 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
8259 struct btrfs_root *root,
8260 u64 root_objectid, u64 owner, u64 offset,
8261 struct btrfs_key *ins)
8262 {
8263 int ret;
8264 struct btrfs_block_group_cache *block_group;
8265 struct btrfs_space_info *space_info;
8266
8267 /*
8268 * Mixed block groups will exclude before processing the log so we only
8269 * need to do the exclude dance if this fs isn't mixed.
8270 */
8271 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
8272 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
8273 if (ret)
8274 return ret;
8275 }
8276
8277 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
8278 if (!block_group)
8279 return -EINVAL;
8280
8281 space_info = block_group->space_info;
8282 spin_lock(&space_info->lock);
8283 spin_lock(&block_group->lock);
8284 space_info->bytes_reserved += ins->offset;
8285 block_group->reserved += ins->offset;
8286 spin_unlock(&block_group->lock);
8287 spin_unlock(&space_info->lock);
8288
8289 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
8290 0, owner, offset, ins, 1);
8291 btrfs_put_block_group(block_group);
8292 return ret;
8293 }
8294
8295 static struct extent_buffer *
8296 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
8297 u64 bytenr, int level)
8298 {
8299 struct extent_buffer *buf;
8300
8301 buf = btrfs_find_create_tree_block(root, bytenr);
8302 if (IS_ERR(buf))
8303 return buf;
8304
8305 btrfs_set_header_generation(buf, trans->transid);
8306 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
8307 btrfs_tree_lock(buf);
8308 clean_tree_block(trans, root->fs_info, buf);
8309 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
8310
8311 btrfs_set_lock_blocking(buf);
8312 set_extent_buffer_uptodate(buf);
8313
8314 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8315 buf->log_index = root->log_transid % 2;
8316 /*
8317 * we allow two log transactions at a time, use different
8318 * EXENT bit to differentiate dirty pages.
8319 */
8320 if (buf->log_index == 0)
8321 set_extent_dirty(&root->dirty_log_pages, buf->start,
8322 buf->start + buf->len - 1, GFP_NOFS);
8323 else
8324 set_extent_new(&root->dirty_log_pages, buf->start,
8325 buf->start + buf->len - 1);
8326 } else {
8327 buf->log_index = -1;
8328 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
8329 buf->start + buf->len - 1, GFP_NOFS);
8330 }
8331 trans->dirty = true;
8332 /* this returns a buffer locked for blocking */
8333 return buf;
8334 }
8335
8336 static struct btrfs_block_rsv *
8337 use_block_rsv(struct btrfs_trans_handle *trans,
8338 struct btrfs_root *root, u32 blocksize)
8339 {
8340 struct btrfs_block_rsv *block_rsv;
8341 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
8342 int ret;
8343 bool global_updated = false;
8344
8345 block_rsv = get_block_rsv(trans, root);
8346
8347 if (unlikely(block_rsv->size == 0))
8348 goto try_reserve;
8349 again:
8350 ret = block_rsv_use_bytes(block_rsv, blocksize);
8351 if (!ret)
8352 return block_rsv;
8353
8354 if (block_rsv->failfast)
8355 return ERR_PTR(ret);
8356
8357 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
8358 global_updated = true;
8359 update_global_block_rsv(root->fs_info);
8360 goto again;
8361 }
8362
8363 if (btrfs_test_opt(root->fs_info, ENOSPC_DEBUG)) {
8364 static DEFINE_RATELIMIT_STATE(_rs,
8365 DEFAULT_RATELIMIT_INTERVAL * 10,
8366 /*DEFAULT_RATELIMIT_BURST*/ 1);
8367 if (__ratelimit(&_rs))
8368 WARN(1, KERN_DEBUG
8369 "BTRFS: block rsv returned %d\n", ret);
8370 }
8371 try_reserve:
8372 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
8373 BTRFS_RESERVE_NO_FLUSH);
8374 if (!ret)
8375 return block_rsv;
8376 /*
8377 * If we couldn't reserve metadata bytes try and use some from
8378 * the global reserve if its space type is the same as the global
8379 * reservation.
8380 */
8381 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
8382 block_rsv->space_info == global_rsv->space_info) {
8383 ret = block_rsv_use_bytes(global_rsv, blocksize);
8384 if (!ret)
8385 return global_rsv;
8386 }
8387 return ERR_PTR(ret);
8388 }
8389
8390 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
8391 struct btrfs_block_rsv *block_rsv, u32 blocksize)
8392 {
8393 block_rsv_add_bytes(block_rsv, blocksize, 0);
8394 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
8395 }
8396
8397 /*
8398 * finds a free extent and does all the dirty work required for allocation
8399 * returns the tree buffer or an ERR_PTR on error.
8400 */
8401 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8402 struct btrfs_root *root,
8403 u64 parent, u64 root_objectid,
8404 struct btrfs_disk_key *key, int level,
8405 u64 hint, u64 empty_size)
8406 {
8407 struct btrfs_key ins;
8408 struct btrfs_block_rsv *block_rsv;
8409 struct extent_buffer *buf;
8410 struct btrfs_delayed_extent_op *extent_op;
8411 u64 flags = 0;
8412 int ret;
8413 u32 blocksize = root->nodesize;
8414 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
8415 SKINNY_METADATA);
8416
8417 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8418 if (btrfs_is_testing(root->fs_info)) {
8419 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8420 level);
8421 if (!IS_ERR(buf))
8422 root->alloc_bytenr += blocksize;
8423 return buf;
8424 }
8425 #endif
8426
8427 block_rsv = use_block_rsv(trans, root, blocksize);
8428 if (IS_ERR(block_rsv))
8429 return ERR_CAST(block_rsv);
8430
8431 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
8432 empty_size, hint, &ins, 0, 0);
8433 if (ret)
8434 goto out_unuse;
8435
8436 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
8437 if (IS_ERR(buf)) {
8438 ret = PTR_ERR(buf);
8439 goto out_free_reserved;
8440 }
8441
8442 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8443 if (parent == 0)
8444 parent = ins.objectid;
8445 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8446 } else
8447 BUG_ON(parent > 0);
8448
8449 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8450 extent_op = btrfs_alloc_delayed_extent_op();
8451 if (!extent_op) {
8452 ret = -ENOMEM;
8453 goto out_free_buf;
8454 }
8455 if (key)
8456 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8457 else
8458 memset(&extent_op->key, 0, sizeof(extent_op->key));
8459 extent_op->flags_to_set = flags;
8460 extent_op->update_key = skinny_metadata ? false : true;
8461 extent_op->update_flags = true;
8462 extent_op->is_data = false;
8463 extent_op->level = level;
8464
8465 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
8466 ins.objectid, ins.offset,
8467 parent, root_objectid, level,
8468 BTRFS_ADD_DELAYED_EXTENT,
8469 extent_op);
8470 if (ret)
8471 goto out_free_delayed;
8472 }
8473 return buf;
8474
8475 out_free_delayed:
8476 btrfs_free_delayed_extent_op(extent_op);
8477 out_free_buf:
8478 free_extent_buffer(buf);
8479 out_free_reserved:
8480 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0);
8481 out_unuse:
8482 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
8483 return ERR_PTR(ret);
8484 }
8485
8486 struct walk_control {
8487 u64 refs[BTRFS_MAX_LEVEL];
8488 u64 flags[BTRFS_MAX_LEVEL];
8489 struct btrfs_key update_progress;
8490 int stage;
8491 int level;
8492 int shared_level;
8493 int update_ref;
8494 int keep_locks;
8495 int reada_slot;
8496 int reada_count;
8497 int for_reloc;
8498 };
8499
8500 #define DROP_REFERENCE 1
8501 #define UPDATE_BACKREF 2
8502
8503 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8504 struct btrfs_root *root,
8505 struct walk_control *wc,
8506 struct btrfs_path *path)
8507 {
8508 u64 bytenr;
8509 u64 generation;
8510 u64 refs;
8511 u64 flags;
8512 u32 nritems;
8513 struct btrfs_key key;
8514 struct extent_buffer *eb;
8515 int ret;
8516 int slot;
8517 int nread = 0;
8518
8519 if (path->slots[wc->level] < wc->reada_slot) {
8520 wc->reada_count = wc->reada_count * 2 / 3;
8521 wc->reada_count = max(wc->reada_count, 2);
8522 } else {
8523 wc->reada_count = wc->reada_count * 3 / 2;
8524 wc->reada_count = min_t(int, wc->reada_count,
8525 BTRFS_NODEPTRS_PER_BLOCK(root));
8526 }
8527
8528 eb = path->nodes[wc->level];
8529 nritems = btrfs_header_nritems(eb);
8530
8531 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8532 if (nread >= wc->reada_count)
8533 break;
8534
8535 cond_resched();
8536 bytenr = btrfs_node_blockptr(eb, slot);
8537 generation = btrfs_node_ptr_generation(eb, slot);
8538
8539 if (slot == path->slots[wc->level])
8540 goto reada;
8541
8542 if (wc->stage == UPDATE_BACKREF &&
8543 generation <= root->root_key.offset)
8544 continue;
8545
8546 /* We don't lock the tree block, it's OK to be racy here */
8547 ret = btrfs_lookup_extent_info(trans, root, bytenr,
8548 wc->level - 1, 1, &refs,
8549 &flags);
8550 /* We don't care about errors in readahead. */
8551 if (ret < 0)
8552 continue;
8553 BUG_ON(refs == 0);
8554
8555 if (wc->stage == DROP_REFERENCE) {
8556 if (refs == 1)
8557 goto reada;
8558
8559 if (wc->level == 1 &&
8560 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8561 continue;
8562 if (!wc->update_ref ||
8563 generation <= root->root_key.offset)
8564 continue;
8565 btrfs_node_key_to_cpu(eb, &key, slot);
8566 ret = btrfs_comp_cpu_keys(&key,
8567 &wc->update_progress);
8568 if (ret < 0)
8569 continue;
8570 } else {
8571 if (wc->level == 1 &&
8572 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8573 continue;
8574 }
8575 reada:
8576 readahead_tree_block(root, bytenr);
8577 nread++;
8578 }
8579 wc->reada_slot = slot;
8580 }
8581
8582 static int account_leaf_items(struct btrfs_trans_handle *trans,
8583 struct btrfs_root *root,
8584 struct extent_buffer *eb)
8585 {
8586 int nr = btrfs_header_nritems(eb);
8587 int i, extent_type, ret;
8588 struct btrfs_key key;
8589 struct btrfs_file_extent_item *fi;
8590 u64 bytenr, num_bytes;
8591
8592 /* We can be called directly from walk_up_proc() */
8593 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags))
8594 return 0;
8595
8596 for (i = 0; i < nr; i++) {
8597 btrfs_item_key_to_cpu(eb, &key, i);
8598
8599 if (key.type != BTRFS_EXTENT_DATA_KEY)
8600 continue;
8601
8602 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
8603 /* filter out non qgroup-accountable extents */
8604 extent_type = btrfs_file_extent_type(eb, fi);
8605
8606 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
8607 continue;
8608
8609 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
8610 if (!bytenr)
8611 continue;
8612
8613 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
8614
8615 ret = btrfs_qgroup_insert_dirty_extent(trans, root->fs_info,
8616 bytenr, num_bytes, GFP_NOFS);
8617 if (ret)
8618 return ret;
8619 }
8620 return 0;
8621 }
8622
8623 /*
8624 * Walk up the tree from the bottom, freeing leaves and any interior
8625 * nodes which have had all slots visited. If a node (leaf or
8626 * interior) is freed, the node above it will have it's slot
8627 * incremented. The root node will never be freed.
8628 *
8629 * At the end of this function, we should have a path which has all
8630 * slots incremented to the next position for a search. If we need to
8631 * read a new node it will be NULL and the node above it will have the
8632 * correct slot selected for a later read.
8633 *
8634 * If we increment the root nodes slot counter past the number of
8635 * elements, 1 is returned to signal completion of the search.
8636 */
8637 static int adjust_slots_upwards(struct btrfs_root *root,
8638 struct btrfs_path *path, int root_level)
8639 {
8640 int level = 0;
8641 int nr, slot;
8642 struct extent_buffer *eb;
8643
8644 if (root_level == 0)
8645 return 1;
8646
8647 while (level <= root_level) {
8648 eb = path->nodes[level];
8649 nr = btrfs_header_nritems(eb);
8650 path->slots[level]++;
8651 slot = path->slots[level];
8652 if (slot >= nr || level == 0) {
8653 /*
8654 * Don't free the root - we will detect this
8655 * condition after our loop and return a
8656 * positive value for caller to stop walking the tree.
8657 */
8658 if (level != root_level) {
8659 btrfs_tree_unlock_rw(eb, path->locks[level]);
8660 path->locks[level] = 0;
8661
8662 free_extent_buffer(eb);
8663 path->nodes[level] = NULL;
8664 path->slots[level] = 0;
8665 }
8666 } else {
8667 /*
8668 * We have a valid slot to walk back down
8669 * from. Stop here so caller can process these
8670 * new nodes.
8671 */
8672 break;
8673 }
8674
8675 level++;
8676 }
8677
8678 eb = path->nodes[root_level];
8679 if (path->slots[root_level] >= btrfs_header_nritems(eb))
8680 return 1;
8681
8682 return 0;
8683 }
8684
8685 /*
8686 * root_eb is the subtree root and is locked before this function is called.
8687 */
8688 static int account_shared_subtree(struct btrfs_trans_handle *trans,
8689 struct btrfs_root *root,
8690 struct extent_buffer *root_eb,
8691 u64 root_gen,
8692 int root_level)
8693 {
8694 int ret = 0;
8695 int level;
8696 struct extent_buffer *eb = root_eb;
8697 struct btrfs_path *path = NULL;
8698
8699 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
8700 BUG_ON(root_eb == NULL);
8701
8702 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags))
8703 return 0;
8704
8705 if (!extent_buffer_uptodate(root_eb)) {
8706 ret = btrfs_read_buffer(root_eb, root_gen);
8707 if (ret)
8708 goto out;
8709 }
8710
8711 if (root_level == 0) {
8712 ret = account_leaf_items(trans, root, root_eb);
8713 goto out;
8714 }
8715
8716 path = btrfs_alloc_path();
8717 if (!path)
8718 return -ENOMEM;
8719
8720 /*
8721 * Walk down the tree. Missing extent blocks are filled in as
8722 * we go. Metadata is accounted every time we read a new
8723 * extent block.
8724 *
8725 * When we reach a leaf, we account for file extent items in it,
8726 * walk back up the tree (adjusting slot pointers as we go)
8727 * and restart the search process.
8728 */
8729 extent_buffer_get(root_eb); /* For path */
8730 path->nodes[root_level] = root_eb;
8731 path->slots[root_level] = 0;
8732 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
8733 walk_down:
8734 level = root_level;
8735 while (level >= 0) {
8736 if (path->nodes[level] == NULL) {
8737 int parent_slot;
8738 u64 child_gen;
8739 u64 child_bytenr;
8740
8741 /* We need to get child blockptr/gen from
8742 * parent before we can read it. */
8743 eb = path->nodes[level + 1];
8744 parent_slot = path->slots[level + 1];
8745 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
8746 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
8747
8748 eb = read_tree_block(root, child_bytenr, child_gen);
8749 if (IS_ERR(eb)) {
8750 ret = PTR_ERR(eb);
8751 goto out;
8752 } else if (!extent_buffer_uptodate(eb)) {
8753 free_extent_buffer(eb);
8754 ret = -EIO;
8755 goto out;
8756 }
8757
8758 path->nodes[level] = eb;
8759 path->slots[level] = 0;
8760
8761 btrfs_tree_read_lock(eb);
8762 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
8763 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
8764
8765 ret = btrfs_qgroup_insert_dirty_extent(trans,
8766 root->fs_info, child_bytenr,
8767 root->nodesize, GFP_NOFS);
8768 if (ret)
8769 goto out;
8770 }
8771
8772 if (level == 0) {
8773 ret = account_leaf_items(trans, root, path->nodes[level]);
8774 if (ret)
8775 goto out;
8776
8777 /* Nonzero return here means we completed our search */
8778 ret = adjust_slots_upwards(root, path, root_level);
8779 if (ret)
8780 break;
8781
8782 /* Restart search with new slots */
8783 goto walk_down;
8784 }
8785
8786 level--;
8787 }
8788
8789 ret = 0;
8790 out:
8791 btrfs_free_path(path);
8792
8793 return ret;
8794 }
8795
8796 /*
8797 * helper to process tree block while walking down the tree.
8798 *
8799 * when wc->stage == UPDATE_BACKREF, this function updates
8800 * back refs for pointers in the block.
8801 *
8802 * NOTE: return value 1 means we should stop walking down.
8803 */
8804 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8805 struct btrfs_root *root,
8806 struct btrfs_path *path,
8807 struct walk_control *wc, int lookup_info)
8808 {
8809 int level = wc->level;
8810 struct extent_buffer *eb = path->nodes[level];
8811 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8812 int ret;
8813
8814 if (wc->stage == UPDATE_BACKREF &&
8815 btrfs_header_owner(eb) != root->root_key.objectid)
8816 return 1;
8817
8818 /*
8819 * when reference count of tree block is 1, it won't increase
8820 * again. once full backref flag is set, we never clear it.
8821 */
8822 if (lookup_info &&
8823 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8824 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8825 BUG_ON(!path->locks[level]);
8826 ret = btrfs_lookup_extent_info(trans, root,
8827 eb->start, level, 1,
8828 &wc->refs[level],
8829 &wc->flags[level]);
8830 BUG_ON(ret == -ENOMEM);
8831 if (ret)
8832 return ret;
8833 BUG_ON(wc->refs[level] == 0);
8834 }
8835
8836 if (wc->stage == DROP_REFERENCE) {
8837 if (wc->refs[level] > 1)
8838 return 1;
8839
8840 if (path->locks[level] && !wc->keep_locks) {
8841 btrfs_tree_unlock_rw(eb, path->locks[level]);
8842 path->locks[level] = 0;
8843 }
8844 return 0;
8845 }
8846
8847 /* wc->stage == UPDATE_BACKREF */
8848 if (!(wc->flags[level] & flag)) {
8849 BUG_ON(!path->locks[level]);
8850 ret = btrfs_inc_ref(trans, root, eb, 1);
8851 BUG_ON(ret); /* -ENOMEM */
8852 ret = btrfs_dec_ref(trans, root, eb, 0);
8853 BUG_ON(ret); /* -ENOMEM */
8854 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
8855 eb->len, flag,
8856 btrfs_header_level(eb), 0);
8857 BUG_ON(ret); /* -ENOMEM */
8858 wc->flags[level] |= flag;
8859 }
8860
8861 /*
8862 * the block is shared by multiple trees, so it's not good to
8863 * keep the tree lock
8864 */
8865 if (path->locks[level] && level > 0) {
8866 btrfs_tree_unlock_rw(eb, path->locks[level]);
8867 path->locks[level] = 0;
8868 }
8869 return 0;
8870 }
8871
8872 /*
8873 * helper to process tree block pointer.
8874 *
8875 * when wc->stage == DROP_REFERENCE, this function checks
8876 * reference count of the block pointed to. if the block
8877 * is shared and we need update back refs for the subtree
8878 * rooted at the block, this function changes wc->stage to
8879 * UPDATE_BACKREF. if the block is shared and there is no
8880 * need to update back, this function drops the reference
8881 * to the block.
8882 *
8883 * NOTE: return value 1 means we should stop walking down.
8884 */
8885 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8886 struct btrfs_root *root,
8887 struct btrfs_path *path,
8888 struct walk_control *wc, int *lookup_info)
8889 {
8890 u64 bytenr;
8891 u64 generation;
8892 u64 parent;
8893 u32 blocksize;
8894 struct btrfs_key key;
8895 struct extent_buffer *next;
8896 int level = wc->level;
8897 int reada = 0;
8898 int ret = 0;
8899 bool need_account = false;
8900
8901 generation = btrfs_node_ptr_generation(path->nodes[level],
8902 path->slots[level]);
8903 /*
8904 * if the lower level block was created before the snapshot
8905 * was created, we know there is no need to update back refs
8906 * for the subtree
8907 */
8908 if (wc->stage == UPDATE_BACKREF &&
8909 generation <= root->root_key.offset) {
8910 *lookup_info = 1;
8911 return 1;
8912 }
8913
8914 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8915 blocksize = root->nodesize;
8916
8917 next = btrfs_find_tree_block(root->fs_info, bytenr);
8918 if (!next) {
8919 next = btrfs_find_create_tree_block(root, bytenr);
8920 if (IS_ERR(next))
8921 return PTR_ERR(next);
8922
8923 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8924 level - 1);
8925 reada = 1;
8926 }
8927 btrfs_tree_lock(next);
8928 btrfs_set_lock_blocking(next);
8929
8930 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
8931 &wc->refs[level - 1],
8932 &wc->flags[level - 1]);
8933 if (ret < 0) {
8934 btrfs_tree_unlock(next);
8935 return ret;
8936 }
8937
8938 if (unlikely(wc->refs[level - 1] == 0)) {
8939 btrfs_err(root->fs_info, "Missing references.");
8940 BUG();
8941 }
8942 *lookup_info = 0;
8943
8944 if (wc->stage == DROP_REFERENCE) {
8945 if (wc->refs[level - 1] > 1) {
8946 need_account = true;
8947 if (level == 1 &&
8948 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8949 goto skip;
8950
8951 if (!wc->update_ref ||
8952 generation <= root->root_key.offset)
8953 goto skip;
8954
8955 btrfs_node_key_to_cpu(path->nodes[level], &key,
8956 path->slots[level]);
8957 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8958 if (ret < 0)
8959 goto skip;
8960
8961 wc->stage = UPDATE_BACKREF;
8962 wc->shared_level = level - 1;
8963 }
8964 } else {
8965 if (level == 1 &&
8966 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8967 goto skip;
8968 }
8969
8970 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8971 btrfs_tree_unlock(next);
8972 free_extent_buffer(next);
8973 next = NULL;
8974 *lookup_info = 1;
8975 }
8976
8977 if (!next) {
8978 if (reada && level == 1)
8979 reada_walk_down(trans, root, wc, path);
8980 next = read_tree_block(root, bytenr, generation);
8981 if (IS_ERR(next)) {
8982 return PTR_ERR(next);
8983 } else if (!extent_buffer_uptodate(next)) {
8984 free_extent_buffer(next);
8985 return -EIO;
8986 }
8987 btrfs_tree_lock(next);
8988 btrfs_set_lock_blocking(next);
8989 }
8990
8991 level--;
8992 BUG_ON(level != btrfs_header_level(next));
8993 path->nodes[level] = next;
8994 path->slots[level] = 0;
8995 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8996 wc->level = level;
8997 if (wc->level == 1)
8998 wc->reada_slot = 0;
8999 return 0;
9000 skip:
9001 wc->refs[level - 1] = 0;
9002 wc->flags[level - 1] = 0;
9003 if (wc->stage == DROP_REFERENCE) {
9004 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
9005 parent = path->nodes[level]->start;
9006 } else {
9007 BUG_ON(root->root_key.objectid !=
9008 btrfs_header_owner(path->nodes[level]));
9009 parent = 0;
9010 }
9011
9012 if (need_account) {
9013 ret = account_shared_subtree(trans, root, next,
9014 generation, level - 1);
9015 if (ret) {
9016 btrfs_err_rl(root->fs_info,
9017 "Error "
9018 "%d accounting shared subtree. Quota "
9019 "is out of sync, rescan required.",
9020 ret);
9021 }
9022 }
9023 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
9024 root->root_key.objectid, level - 1, 0);
9025 BUG_ON(ret); /* -ENOMEM */
9026 }
9027 btrfs_tree_unlock(next);
9028 free_extent_buffer(next);
9029 *lookup_info = 1;
9030 return 1;
9031 }
9032
9033 /*
9034 * helper to process tree block while walking up the tree.
9035 *
9036 * when wc->stage == DROP_REFERENCE, this function drops
9037 * reference count on the block.
9038 *
9039 * when wc->stage == UPDATE_BACKREF, this function changes
9040 * wc->stage back to DROP_REFERENCE if we changed wc->stage
9041 * to UPDATE_BACKREF previously while processing the block.
9042 *
9043 * NOTE: return value 1 means we should stop walking up.
9044 */
9045 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
9046 struct btrfs_root *root,
9047 struct btrfs_path *path,
9048 struct walk_control *wc)
9049 {
9050 int ret;
9051 int level = wc->level;
9052 struct extent_buffer *eb = path->nodes[level];
9053 u64 parent = 0;
9054
9055 if (wc->stage == UPDATE_BACKREF) {
9056 BUG_ON(wc->shared_level < level);
9057 if (level < wc->shared_level)
9058 goto out;
9059
9060 ret = find_next_key(path, level + 1, &wc->update_progress);
9061 if (ret > 0)
9062 wc->update_ref = 0;
9063
9064 wc->stage = DROP_REFERENCE;
9065 wc->shared_level = -1;
9066 path->slots[level] = 0;
9067
9068 /*
9069 * check reference count again if the block isn't locked.
9070 * we should start walking down the tree again if reference
9071 * count is one.
9072 */
9073 if (!path->locks[level]) {
9074 BUG_ON(level == 0);
9075 btrfs_tree_lock(eb);
9076 btrfs_set_lock_blocking(eb);
9077 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9078
9079 ret = btrfs_lookup_extent_info(trans, root,
9080 eb->start, level, 1,
9081 &wc->refs[level],
9082 &wc->flags[level]);
9083 if (ret < 0) {
9084 btrfs_tree_unlock_rw(eb, path->locks[level]);
9085 path->locks[level] = 0;
9086 return ret;
9087 }
9088 BUG_ON(wc->refs[level] == 0);
9089 if (wc->refs[level] == 1) {
9090 btrfs_tree_unlock_rw(eb, path->locks[level]);
9091 path->locks[level] = 0;
9092 return 1;
9093 }
9094 }
9095 }
9096
9097 /* wc->stage == DROP_REFERENCE */
9098 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
9099
9100 if (wc->refs[level] == 1) {
9101 if (level == 0) {
9102 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9103 ret = btrfs_dec_ref(trans, root, eb, 1);
9104 else
9105 ret = btrfs_dec_ref(trans, root, eb, 0);
9106 BUG_ON(ret); /* -ENOMEM */
9107 ret = account_leaf_items(trans, root, eb);
9108 if (ret) {
9109 btrfs_err_rl(root->fs_info,
9110 "error "
9111 "%d accounting leaf items. Quota "
9112 "is out of sync, rescan required.",
9113 ret);
9114 }
9115 }
9116 /* make block locked assertion in clean_tree_block happy */
9117 if (!path->locks[level] &&
9118 btrfs_header_generation(eb) == trans->transid) {
9119 btrfs_tree_lock(eb);
9120 btrfs_set_lock_blocking(eb);
9121 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9122 }
9123 clean_tree_block(trans, root->fs_info, eb);
9124 }
9125
9126 if (eb == root->node) {
9127 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9128 parent = eb->start;
9129 else
9130 BUG_ON(root->root_key.objectid !=
9131 btrfs_header_owner(eb));
9132 } else {
9133 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9134 parent = path->nodes[level + 1]->start;
9135 else
9136 BUG_ON(root->root_key.objectid !=
9137 btrfs_header_owner(path->nodes[level + 1]));
9138 }
9139
9140 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
9141 out:
9142 wc->refs[level] = 0;
9143 wc->flags[level] = 0;
9144 return 0;
9145 }
9146
9147 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
9148 struct btrfs_root *root,
9149 struct btrfs_path *path,
9150 struct walk_control *wc)
9151 {
9152 int level = wc->level;
9153 int lookup_info = 1;
9154 int ret;
9155
9156 while (level >= 0) {
9157 ret = walk_down_proc(trans, root, path, wc, lookup_info);
9158 if (ret > 0)
9159 break;
9160
9161 if (level == 0)
9162 break;
9163
9164 if (path->slots[level] >=
9165 btrfs_header_nritems(path->nodes[level]))
9166 break;
9167
9168 ret = do_walk_down(trans, root, path, wc, &lookup_info);
9169 if (ret > 0) {
9170 path->slots[level]++;
9171 continue;
9172 } else if (ret < 0)
9173 return ret;
9174 level = wc->level;
9175 }
9176 return 0;
9177 }
9178
9179 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
9180 struct btrfs_root *root,
9181 struct btrfs_path *path,
9182 struct walk_control *wc, int max_level)
9183 {
9184 int level = wc->level;
9185 int ret;
9186
9187 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
9188 while (level < max_level && path->nodes[level]) {
9189 wc->level = level;
9190 if (path->slots[level] + 1 <
9191 btrfs_header_nritems(path->nodes[level])) {
9192 path->slots[level]++;
9193 return 0;
9194 } else {
9195 ret = walk_up_proc(trans, root, path, wc);
9196 if (ret > 0)
9197 return 0;
9198
9199 if (path->locks[level]) {
9200 btrfs_tree_unlock_rw(path->nodes[level],
9201 path->locks[level]);
9202 path->locks[level] = 0;
9203 }
9204 free_extent_buffer(path->nodes[level]);
9205 path->nodes[level] = NULL;
9206 level++;
9207 }
9208 }
9209 return 1;
9210 }
9211
9212 /*
9213 * drop a subvolume tree.
9214 *
9215 * this function traverses the tree freeing any blocks that only
9216 * referenced by the tree.
9217 *
9218 * when a shared tree block is found. this function decreases its
9219 * reference count by one. if update_ref is true, this function
9220 * also make sure backrefs for the shared block and all lower level
9221 * blocks are properly updated.
9222 *
9223 * If called with for_reloc == 0, may exit early with -EAGAIN
9224 */
9225 int btrfs_drop_snapshot(struct btrfs_root *root,
9226 struct btrfs_block_rsv *block_rsv, int update_ref,
9227 int for_reloc)
9228 {
9229 struct btrfs_path *path;
9230 struct btrfs_trans_handle *trans;
9231 struct btrfs_root *tree_root = root->fs_info->tree_root;
9232 struct btrfs_root_item *root_item = &root->root_item;
9233 struct walk_control *wc;
9234 struct btrfs_key key;
9235 int err = 0;
9236 int ret;
9237 int level;
9238 bool root_dropped = false;
9239
9240 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
9241
9242 path = btrfs_alloc_path();
9243 if (!path) {
9244 err = -ENOMEM;
9245 goto out;
9246 }
9247
9248 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9249 if (!wc) {
9250 btrfs_free_path(path);
9251 err = -ENOMEM;
9252 goto out;
9253 }
9254
9255 trans = btrfs_start_transaction(tree_root, 0);
9256 if (IS_ERR(trans)) {
9257 err = PTR_ERR(trans);
9258 goto out_free;
9259 }
9260
9261 if (block_rsv)
9262 trans->block_rsv = block_rsv;
9263
9264 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
9265 level = btrfs_header_level(root->node);
9266 path->nodes[level] = btrfs_lock_root_node(root);
9267 btrfs_set_lock_blocking(path->nodes[level]);
9268 path->slots[level] = 0;
9269 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9270 memset(&wc->update_progress, 0,
9271 sizeof(wc->update_progress));
9272 } else {
9273 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
9274 memcpy(&wc->update_progress, &key,
9275 sizeof(wc->update_progress));
9276
9277 level = root_item->drop_level;
9278 BUG_ON(level == 0);
9279 path->lowest_level = level;
9280 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
9281 path->lowest_level = 0;
9282 if (ret < 0) {
9283 err = ret;
9284 goto out_end_trans;
9285 }
9286 WARN_ON(ret > 0);
9287
9288 /*
9289 * unlock our path, this is safe because only this
9290 * function is allowed to delete this snapshot
9291 */
9292 btrfs_unlock_up_safe(path, 0);
9293
9294 level = btrfs_header_level(root->node);
9295 while (1) {
9296 btrfs_tree_lock(path->nodes[level]);
9297 btrfs_set_lock_blocking(path->nodes[level]);
9298 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9299
9300 ret = btrfs_lookup_extent_info(trans, root,
9301 path->nodes[level]->start,
9302 level, 1, &wc->refs[level],
9303 &wc->flags[level]);
9304 if (ret < 0) {
9305 err = ret;
9306 goto out_end_trans;
9307 }
9308 BUG_ON(wc->refs[level] == 0);
9309
9310 if (level == root_item->drop_level)
9311 break;
9312
9313 btrfs_tree_unlock(path->nodes[level]);
9314 path->locks[level] = 0;
9315 WARN_ON(wc->refs[level] != 1);
9316 level--;
9317 }
9318 }
9319
9320 wc->level = level;
9321 wc->shared_level = -1;
9322 wc->stage = DROP_REFERENCE;
9323 wc->update_ref = update_ref;
9324 wc->keep_locks = 0;
9325 wc->for_reloc = for_reloc;
9326 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
9327
9328 while (1) {
9329
9330 ret = walk_down_tree(trans, root, path, wc);
9331 if (ret < 0) {
9332 err = ret;
9333 break;
9334 }
9335
9336 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
9337 if (ret < 0) {
9338 err = ret;
9339 break;
9340 }
9341
9342 if (ret > 0) {
9343 BUG_ON(wc->stage != DROP_REFERENCE);
9344 break;
9345 }
9346
9347 if (wc->stage == DROP_REFERENCE) {
9348 level = wc->level;
9349 btrfs_node_key(path->nodes[level],
9350 &root_item->drop_progress,
9351 path->slots[level]);
9352 root_item->drop_level = level;
9353 }
9354
9355 BUG_ON(wc->level == 0);
9356 if (btrfs_should_end_transaction(trans, tree_root) ||
9357 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
9358 ret = btrfs_update_root(trans, tree_root,
9359 &root->root_key,
9360 root_item);
9361 if (ret) {
9362 btrfs_abort_transaction(trans, ret);
9363 err = ret;
9364 goto out_end_trans;
9365 }
9366
9367 btrfs_end_transaction_throttle(trans, tree_root);
9368 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
9369 pr_debug("BTRFS: drop snapshot early exit\n");
9370 err = -EAGAIN;
9371 goto out_free;
9372 }
9373
9374 trans = btrfs_start_transaction(tree_root, 0);
9375 if (IS_ERR(trans)) {
9376 err = PTR_ERR(trans);
9377 goto out_free;
9378 }
9379 if (block_rsv)
9380 trans->block_rsv = block_rsv;
9381 }
9382 }
9383 btrfs_release_path(path);
9384 if (err)
9385 goto out_end_trans;
9386
9387 ret = btrfs_del_root(trans, tree_root, &root->root_key);
9388 if (ret) {
9389 btrfs_abort_transaction(trans, ret);
9390 goto out_end_trans;
9391 }
9392
9393 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9394 ret = btrfs_find_root(tree_root, &root->root_key, path,
9395 NULL, NULL);
9396 if (ret < 0) {
9397 btrfs_abort_transaction(trans, ret);
9398 err = ret;
9399 goto out_end_trans;
9400 } else if (ret > 0) {
9401 /* if we fail to delete the orphan item this time
9402 * around, it'll get picked up the next time.
9403 *
9404 * The most common failure here is just -ENOENT.
9405 */
9406 btrfs_del_orphan_item(trans, tree_root,
9407 root->root_key.objectid);
9408 }
9409 }
9410
9411 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9412 btrfs_add_dropped_root(trans, root);
9413 } else {
9414 free_extent_buffer(root->node);
9415 free_extent_buffer(root->commit_root);
9416 btrfs_put_fs_root(root);
9417 }
9418 root_dropped = true;
9419 out_end_trans:
9420 btrfs_end_transaction_throttle(trans, tree_root);
9421 out_free:
9422 kfree(wc);
9423 btrfs_free_path(path);
9424 out:
9425 /*
9426 * So if we need to stop dropping the snapshot for whatever reason we
9427 * need to make sure to add it back to the dead root list so that we
9428 * keep trying to do the work later. This also cleans up roots if we
9429 * don't have it in the radix (like when we recover after a power fail
9430 * or unmount) so we don't leak memory.
9431 */
9432 if (!for_reloc && root_dropped == false)
9433 btrfs_add_dead_root(root);
9434 if (err && err != -EAGAIN)
9435 btrfs_handle_fs_error(root->fs_info, err, NULL);
9436 return err;
9437 }
9438
9439 /*
9440 * drop subtree rooted at tree block 'node'.
9441 *
9442 * NOTE: this function will unlock and release tree block 'node'
9443 * only used by relocation code
9444 */
9445 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9446 struct btrfs_root *root,
9447 struct extent_buffer *node,
9448 struct extent_buffer *parent)
9449 {
9450 struct btrfs_path *path;
9451 struct walk_control *wc;
9452 int level;
9453 int parent_level;
9454 int ret = 0;
9455 int wret;
9456
9457 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9458
9459 path = btrfs_alloc_path();
9460 if (!path)
9461 return -ENOMEM;
9462
9463 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9464 if (!wc) {
9465 btrfs_free_path(path);
9466 return -ENOMEM;
9467 }
9468
9469 btrfs_assert_tree_locked(parent);
9470 parent_level = btrfs_header_level(parent);
9471 extent_buffer_get(parent);
9472 path->nodes[parent_level] = parent;
9473 path->slots[parent_level] = btrfs_header_nritems(parent);
9474
9475 btrfs_assert_tree_locked(node);
9476 level = btrfs_header_level(node);
9477 path->nodes[level] = node;
9478 path->slots[level] = 0;
9479 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9480
9481 wc->refs[parent_level] = 1;
9482 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9483 wc->level = level;
9484 wc->shared_level = -1;
9485 wc->stage = DROP_REFERENCE;
9486 wc->update_ref = 0;
9487 wc->keep_locks = 1;
9488 wc->for_reloc = 1;
9489 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
9490
9491 while (1) {
9492 wret = walk_down_tree(trans, root, path, wc);
9493 if (wret < 0) {
9494 ret = wret;
9495 break;
9496 }
9497
9498 wret = walk_up_tree(trans, root, path, wc, parent_level);
9499 if (wret < 0)
9500 ret = wret;
9501 if (wret != 0)
9502 break;
9503 }
9504
9505 kfree(wc);
9506 btrfs_free_path(path);
9507 return ret;
9508 }
9509
9510 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
9511 {
9512 u64 num_devices;
9513 u64 stripped;
9514
9515 /*
9516 * if restripe for this chunk_type is on pick target profile and
9517 * return, otherwise do the usual balance
9518 */
9519 stripped = get_restripe_target(root->fs_info, flags);
9520 if (stripped)
9521 return extended_to_chunk(stripped);
9522
9523 num_devices = root->fs_info->fs_devices->rw_devices;
9524
9525 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9526 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9527 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9528
9529 if (num_devices == 1) {
9530 stripped |= BTRFS_BLOCK_GROUP_DUP;
9531 stripped = flags & ~stripped;
9532
9533 /* turn raid0 into single device chunks */
9534 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9535 return stripped;
9536
9537 /* turn mirroring into duplication */
9538 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9539 BTRFS_BLOCK_GROUP_RAID10))
9540 return stripped | BTRFS_BLOCK_GROUP_DUP;
9541 } else {
9542 /* they already had raid on here, just return */
9543 if (flags & stripped)
9544 return flags;
9545
9546 stripped |= BTRFS_BLOCK_GROUP_DUP;
9547 stripped = flags & ~stripped;
9548
9549 /* switch duplicated blocks with raid1 */
9550 if (flags & BTRFS_BLOCK_GROUP_DUP)
9551 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9552
9553 /* this is drive concat, leave it alone */
9554 }
9555
9556 return flags;
9557 }
9558
9559 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9560 {
9561 struct btrfs_space_info *sinfo = cache->space_info;
9562 u64 num_bytes;
9563 u64 min_allocable_bytes;
9564 int ret = -ENOSPC;
9565
9566 /*
9567 * We need some metadata space and system metadata space for
9568 * allocating chunks in some corner cases until we force to set
9569 * it to be readonly.
9570 */
9571 if ((sinfo->flags &
9572 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9573 !force)
9574 min_allocable_bytes = SZ_1M;
9575 else
9576 min_allocable_bytes = 0;
9577
9578 spin_lock(&sinfo->lock);
9579 spin_lock(&cache->lock);
9580
9581 if (cache->ro) {
9582 cache->ro++;
9583 ret = 0;
9584 goto out;
9585 }
9586
9587 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9588 cache->bytes_super - btrfs_block_group_used(&cache->item);
9589
9590 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
9591 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
9592 min_allocable_bytes <= sinfo->total_bytes) {
9593 sinfo->bytes_readonly += num_bytes;
9594 cache->ro++;
9595 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9596 ret = 0;
9597 }
9598 out:
9599 spin_unlock(&cache->lock);
9600 spin_unlock(&sinfo->lock);
9601 return ret;
9602 }
9603
9604 int btrfs_inc_block_group_ro(struct btrfs_root *root,
9605 struct btrfs_block_group_cache *cache)
9606
9607 {
9608 struct btrfs_trans_handle *trans;
9609 u64 alloc_flags;
9610 int ret;
9611
9612 again:
9613 trans = btrfs_join_transaction(root);
9614 if (IS_ERR(trans))
9615 return PTR_ERR(trans);
9616
9617 /*
9618 * we're not allowed to set block groups readonly after the dirty
9619 * block groups cache has started writing. If it already started,
9620 * back off and let this transaction commit
9621 */
9622 mutex_lock(&root->fs_info->ro_block_group_mutex);
9623 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9624 u64 transid = trans->transid;
9625
9626 mutex_unlock(&root->fs_info->ro_block_group_mutex);
9627 btrfs_end_transaction(trans, root);
9628
9629 ret = btrfs_wait_for_commit(root, transid);
9630 if (ret)
9631 return ret;
9632 goto again;
9633 }
9634
9635 /*
9636 * if we are changing raid levels, try to allocate a corresponding
9637 * block group with the new raid level.
9638 */
9639 alloc_flags = update_block_group_flags(root, cache->flags);
9640 if (alloc_flags != cache->flags) {
9641 ret = do_chunk_alloc(trans, root, alloc_flags,
9642 CHUNK_ALLOC_FORCE);
9643 /*
9644 * ENOSPC is allowed here, we may have enough space
9645 * already allocated at the new raid level to
9646 * carry on
9647 */
9648 if (ret == -ENOSPC)
9649 ret = 0;
9650 if (ret < 0)
9651 goto out;
9652 }
9653
9654 ret = inc_block_group_ro(cache, 0);
9655 if (!ret)
9656 goto out;
9657 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
9658 ret = do_chunk_alloc(trans, root, alloc_flags,
9659 CHUNK_ALLOC_FORCE);
9660 if (ret < 0)
9661 goto out;
9662 ret = inc_block_group_ro(cache, 0);
9663 out:
9664 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9665 alloc_flags = update_block_group_flags(root, cache->flags);
9666 lock_chunks(root->fs_info->chunk_root);
9667 check_system_chunk(trans, root, alloc_flags);
9668 unlock_chunks(root->fs_info->chunk_root);
9669 }
9670 mutex_unlock(&root->fs_info->ro_block_group_mutex);
9671
9672 btrfs_end_transaction(trans, root);
9673 return ret;
9674 }
9675
9676 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
9677 struct btrfs_root *root, u64 type)
9678 {
9679 u64 alloc_flags = get_alloc_profile(root, type);
9680 return do_chunk_alloc(trans, root, alloc_flags,
9681 CHUNK_ALLOC_FORCE);
9682 }
9683
9684 /*
9685 * helper to account the unused space of all the readonly block group in the
9686 * space_info. takes mirrors into account.
9687 */
9688 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9689 {
9690 struct btrfs_block_group_cache *block_group;
9691 u64 free_bytes = 0;
9692 int factor;
9693
9694 /* It's df, we don't care if it's racy */
9695 if (list_empty(&sinfo->ro_bgs))
9696 return 0;
9697
9698 spin_lock(&sinfo->lock);
9699 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9700 spin_lock(&block_group->lock);
9701
9702 if (!block_group->ro) {
9703 spin_unlock(&block_group->lock);
9704 continue;
9705 }
9706
9707 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
9708 BTRFS_BLOCK_GROUP_RAID10 |
9709 BTRFS_BLOCK_GROUP_DUP))
9710 factor = 2;
9711 else
9712 factor = 1;
9713
9714 free_bytes += (block_group->key.offset -
9715 btrfs_block_group_used(&block_group->item)) *
9716 factor;
9717
9718 spin_unlock(&block_group->lock);
9719 }
9720 spin_unlock(&sinfo->lock);
9721
9722 return free_bytes;
9723 }
9724
9725 void btrfs_dec_block_group_ro(struct btrfs_root *root,
9726 struct btrfs_block_group_cache *cache)
9727 {
9728 struct btrfs_space_info *sinfo = cache->space_info;
9729 u64 num_bytes;
9730
9731 BUG_ON(!cache->ro);
9732
9733 spin_lock(&sinfo->lock);
9734 spin_lock(&cache->lock);
9735 if (!--cache->ro) {
9736 num_bytes = cache->key.offset - cache->reserved -
9737 cache->pinned - cache->bytes_super -
9738 btrfs_block_group_used(&cache->item);
9739 sinfo->bytes_readonly -= num_bytes;
9740 list_del_init(&cache->ro_list);
9741 }
9742 spin_unlock(&cache->lock);
9743 spin_unlock(&sinfo->lock);
9744 }
9745
9746 /*
9747 * checks to see if its even possible to relocate this block group.
9748 *
9749 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9750 * ok to go ahead and try.
9751 */
9752 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
9753 {
9754 struct btrfs_block_group_cache *block_group;
9755 struct btrfs_space_info *space_info;
9756 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
9757 struct btrfs_device *device;
9758 struct btrfs_trans_handle *trans;
9759 u64 min_free;
9760 u64 dev_min = 1;
9761 u64 dev_nr = 0;
9762 u64 target;
9763 int debug;
9764 int index;
9765 int full = 0;
9766 int ret = 0;
9767
9768 debug = btrfs_test_opt(root->fs_info, ENOSPC_DEBUG);
9769
9770 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
9771
9772 /* odd, couldn't find the block group, leave it alone */
9773 if (!block_group) {
9774 if (debug)
9775 btrfs_warn(root->fs_info,
9776 "can't find block group for bytenr %llu",
9777 bytenr);
9778 return -1;
9779 }
9780
9781 min_free = btrfs_block_group_used(&block_group->item);
9782
9783 /* no bytes used, we're good */
9784 if (!min_free)
9785 goto out;
9786
9787 space_info = block_group->space_info;
9788 spin_lock(&space_info->lock);
9789
9790 full = space_info->full;
9791
9792 /*
9793 * if this is the last block group we have in this space, we can't
9794 * relocate it unless we're able to allocate a new chunk below.
9795 *
9796 * Otherwise, we need to make sure we have room in the space to handle
9797 * all of the extents from this block group. If we can, we're good
9798 */
9799 if ((space_info->total_bytes != block_group->key.offset) &&
9800 (space_info->bytes_used + space_info->bytes_reserved +
9801 space_info->bytes_pinned + space_info->bytes_readonly +
9802 min_free < space_info->total_bytes)) {
9803 spin_unlock(&space_info->lock);
9804 goto out;
9805 }
9806 spin_unlock(&space_info->lock);
9807
9808 /*
9809 * ok we don't have enough space, but maybe we have free space on our
9810 * devices to allocate new chunks for relocation, so loop through our
9811 * alloc devices and guess if we have enough space. if this block
9812 * group is going to be restriped, run checks against the target
9813 * profile instead of the current one.
9814 */
9815 ret = -1;
9816
9817 /*
9818 * index:
9819 * 0: raid10
9820 * 1: raid1
9821 * 2: dup
9822 * 3: raid0
9823 * 4: single
9824 */
9825 target = get_restripe_target(root->fs_info, block_group->flags);
9826 if (target) {
9827 index = __get_raid_index(extended_to_chunk(target));
9828 } else {
9829 /*
9830 * this is just a balance, so if we were marked as full
9831 * we know there is no space for a new chunk
9832 */
9833 if (full) {
9834 if (debug)
9835 btrfs_warn(root->fs_info,
9836 "no space to alloc new chunk for block group %llu",
9837 block_group->key.objectid);
9838 goto out;
9839 }
9840
9841 index = get_block_group_index(block_group);
9842 }
9843
9844 if (index == BTRFS_RAID_RAID10) {
9845 dev_min = 4;
9846 /* Divide by 2 */
9847 min_free >>= 1;
9848 } else if (index == BTRFS_RAID_RAID1) {
9849 dev_min = 2;
9850 } else if (index == BTRFS_RAID_DUP) {
9851 /* Multiply by 2 */
9852 min_free <<= 1;
9853 } else if (index == BTRFS_RAID_RAID0) {
9854 dev_min = fs_devices->rw_devices;
9855 min_free = div64_u64(min_free, dev_min);
9856 }
9857
9858 /* We need to do this so that we can look at pending chunks */
9859 trans = btrfs_join_transaction(root);
9860 if (IS_ERR(trans)) {
9861 ret = PTR_ERR(trans);
9862 goto out;
9863 }
9864
9865 mutex_lock(&root->fs_info->chunk_mutex);
9866 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9867 u64 dev_offset;
9868
9869 /*
9870 * check to make sure we can actually find a chunk with enough
9871 * space to fit our block group in.
9872 */
9873 if (device->total_bytes > device->bytes_used + min_free &&
9874 !device->is_tgtdev_for_dev_replace) {
9875 ret = find_free_dev_extent(trans, device, min_free,
9876 &dev_offset, NULL);
9877 if (!ret)
9878 dev_nr++;
9879
9880 if (dev_nr >= dev_min)
9881 break;
9882
9883 ret = -1;
9884 }
9885 }
9886 if (debug && ret == -1)
9887 btrfs_warn(root->fs_info,
9888 "no space to allocate a new chunk for block group %llu",
9889 block_group->key.objectid);
9890 mutex_unlock(&root->fs_info->chunk_mutex);
9891 btrfs_end_transaction(trans, root);
9892 out:
9893 btrfs_put_block_group(block_group);
9894 return ret;
9895 }
9896
9897 static int find_first_block_group(struct btrfs_root *root,
9898 struct btrfs_path *path, struct btrfs_key *key)
9899 {
9900 int ret = 0;
9901 struct btrfs_key found_key;
9902 struct extent_buffer *leaf;
9903 int slot;
9904
9905 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9906 if (ret < 0)
9907 goto out;
9908
9909 while (1) {
9910 slot = path->slots[0];
9911 leaf = path->nodes[0];
9912 if (slot >= btrfs_header_nritems(leaf)) {
9913 ret = btrfs_next_leaf(root, path);
9914 if (ret == 0)
9915 continue;
9916 if (ret < 0)
9917 goto out;
9918 break;
9919 }
9920 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9921
9922 if (found_key.objectid >= key->objectid &&
9923 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9924 struct extent_map_tree *em_tree;
9925 struct extent_map *em;
9926
9927 em_tree = &root->fs_info->mapping_tree.map_tree;
9928 read_lock(&em_tree->lock);
9929 em = lookup_extent_mapping(em_tree, found_key.objectid,
9930 found_key.offset);
9931 read_unlock(&em_tree->lock);
9932 if (!em) {
9933 btrfs_err(root->fs_info,
9934 "logical %llu len %llu found bg but no related chunk",
9935 found_key.objectid, found_key.offset);
9936 ret = -ENOENT;
9937 } else {
9938 ret = 0;
9939 }
9940 free_extent_map(em);
9941 goto out;
9942 }
9943 path->slots[0]++;
9944 }
9945 out:
9946 return ret;
9947 }
9948
9949 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9950 {
9951 struct btrfs_block_group_cache *block_group;
9952 u64 last = 0;
9953
9954 while (1) {
9955 struct inode *inode;
9956
9957 block_group = btrfs_lookup_first_block_group(info, last);
9958 while (block_group) {
9959 spin_lock(&block_group->lock);
9960 if (block_group->iref)
9961 break;
9962 spin_unlock(&block_group->lock);
9963 block_group = next_block_group(info->tree_root,
9964 block_group);
9965 }
9966 if (!block_group) {
9967 if (last == 0)
9968 break;
9969 last = 0;
9970 continue;
9971 }
9972
9973 inode = block_group->inode;
9974 block_group->iref = 0;
9975 block_group->inode = NULL;
9976 spin_unlock(&block_group->lock);
9977 ASSERT(block_group->io_ctl.inode == NULL);
9978 iput(inode);
9979 last = block_group->key.objectid + block_group->key.offset;
9980 btrfs_put_block_group(block_group);
9981 }
9982 }
9983
9984 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9985 {
9986 struct btrfs_block_group_cache *block_group;
9987 struct btrfs_space_info *space_info;
9988 struct btrfs_caching_control *caching_ctl;
9989 struct rb_node *n;
9990
9991 down_write(&info->commit_root_sem);
9992 while (!list_empty(&info->caching_block_groups)) {
9993 caching_ctl = list_entry(info->caching_block_groups.next,
9994 struct btrfs_caching_control, list);
9995 list_del(&caching_ctl->list);
9996 put_caching_control(caching_ctl);
9997 }
9998 up_write(&info->commit_root_sem);
9999
10000 spin_lock(&info->unused_bgs_lock);
10001 while (!list_empty(&info->unused_bgs)) {
10002 block_group = list_first_entry(&info->unused_bgs,
10003 struct btrfs_block_group_cache,
10004 bg_list);
10005 list_del_init(&block_group->bg_list);
10006 btrfs_put_block_group(block_group);
10007 }
10008 spin_unlock(&info->unused_bgs_lock);
10009
10010 spin_lock(&info->block_group_cache_lock);
10011 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
10012 block_group = rb_entry(n, struct btrfs_block_group_cache,
10013 cache_node);
10014 rb_erase(&block_group->cache_node,
10015 &info->block_group_cache_tree);
10016 RB_CLEAR_NODE(&block_group->cache_node);
10017 spin_unlock(&info->block_group_cache_lock);
10018
10019 down_write(&block_group->space_info->groups_sem);
10020 list_del(&block_group->list);
10021 up_write(&block_group->space_info->groups_sem);
10022
10023 if (block_group->cached == BTRFS_CACHE_STARTED)
10024 wait_block_group_cache_done(block_group);
10025
10026 /*
10027 * We haven't cached this block group, which means we could
10028 * possibly have excluded extents on this block group.
10029 */
10030 if (block_group->cached == BTRFS_CACHE_NO ||
10031 block_group->cached == BTRFS_CACHE_ERROR)
10032 free_excluded_extents(info->extent_root, block_group);
10033
10034 btrfs_remove_free_space_cache(block_group);
10035 ASSERT(list_empty(&block_group->dirty_list));
10036 ASSERT(list_empty(&block_group->io_list));
10037 ASSERT(list_empty(&block_group->bg_list));
10038 ASSERT(atomic_read(&block_group->count) == 1);
10039 btrfs_put_block_group(block_group);
10040
10041 spin_lock(&info->block_group_cache_lock);
10042 }
10043 spin_unlock(&info->block_group_cache_lock);
10044
10045 /* now that all the block groups are freed, go through and
10046 * free all the space_info structs. This is only called during
10047 * the final stages of unmount, and so we know nobody is
10048 * using them. We call synchronize_rcu() once before we start,
10049 * just to be on the safe side.
10050 */
10051 synchronize_rcu();
10052
10053 release_global_block_rsv(info);
10054
10055 while (!list_empty(&info->space_info)) {
10056 int i;
10057
10058 space_info = list_entry(info->space_info.next,
10059 struct btrfs_space_info,
10060 list);
10061
10062 /*
10063 * Do not hide this behind enospc_debug, this is actually
10064 * important and indicates a real bug if this happens.
10065 */
10066 if (WARN_ON(space_info->bytes_pinned > 0 ||
10067 space_info->bytes_reserved > 0 ||
10068 space_info->bytes_may_use > 0))
10069 dump_space_info(space_info, 0, 0);
10070 list_del(&space_info->list);
10071 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
10072 struct kobject *kobj;
10073 kobj = space_info->block_group_kobjs[i];
10074 space_info->block_group_kobjs[i] = NULL;
10075 if (kobj) {
10076 kobject_del(kobj);
10077 kobject_put(kobj);
10078 }
10079 }
10080 kobject_del(&space_info->kobj);
10081 kobject_put(&space_info->kobj);
10082 }
10083 return 0;
10084 }
10085
10086 static void __link_block_group(struct btrfs_space_info *space_info,
10087 struct btrfs_block_group_cache *cache)
10088 {
10089 int index = get_block_group_index(cache);
10090 bool first = false;
10091
10092 down_write(&space_info->groups_sem);
10093 if (list_empty(&space_info->block_groups[index]))
10094 first = true;
10095 list_add_tail(&cache->list, &space_info->block_groups[index]);
10096 up_write(&space_info->groups_sem);
10097
10098 if (first) {
10099 struct raid_kobject *rkobj;
10100 int ret;
10101
10102 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
10103 if (!rkobj)
10104 goto out_err;
10105 rkobj->raid_type = index;
10106 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
10107 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
10108 "%s", get_raid_name(index));
10109 if (ret) {
10110 kobject_put(&rkobj->kobj);
10111 goto out_err;
10112 }
10113 space_info->block_group_kobjs[index] = &rkobj->kobj;
10114 }
10115
10116 return;
10117 out_err:
10118 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
10119 }
10120
10121 static struct btrfs_block_group_cache *
10122 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
10123 {
10124 struct btrfs_block_group_cache *cache;
10125
10126 cache = kzalloc(sizeof(*cache), GFP_NOFS);
10127 if (!cache)
10128 return NULL;
10129
10130 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
10131 GFP_NOFS);
10132 if (!cache->free_space_ctl) {
10133 kfree(cache);
10134 return NULL;
10135 }
10136
10137 cache->key.objectid = start;
10138 cache->key.offset = size;
10139 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10140
10141 cache->sectorsize = root->sectorsize;
10142 cache->fs_info = root->fs_info;
10143 cache->full_stripe_len = btrfs_full_stripe_len(root,
10144 &root->fs_info->mapping_tree,
10145 start);
10146 set_free_space_tree_thresholds(cache);
10147
10148 atomic_set(&cache->count, 1);
10149 spin_lock_init(&cache->lock);
10150 init_rwsem(&cache->data_rwsem);
10151 INIT_LIST_HEAD(&cache->list);
10152 INIT_LIST_HEAD(&cache->cluster_list);
10153 INIT_LIST_HEAD(&cache->bg_list);
10154 INIT_LIST_HEAD(&cache->ro_list);
10155 INIT_LIST_HEAD(&cache->dirty_list);
10156 INIT_LIST_HEAD(&cache->io_list);
10157 btrfs_init_free_space_ctl(cache);
10158 atomic_set(&cache->trimming, 0);
10159 mutex_init(&cache->free_space_lock);
10160
10161 return cache;
10162 }
10163
10164 int btrfs_read_block_groups(struct btrfs_root *root)
10165 {
10166 struct btrfs_path *path;
10167 int ret;
10168 struct btrfs_block_group_cache *cache;
10169 struct btrfs_fs_info *info = root->fs_info;
10170 struct btrfs_space_info *space_info;
10171 struct btrfs_key key;
10172 struct btrfs_key found_key;
10173 struct extent_buffer *leaf;
10174 int need_clear = 0;
10175 u64 cache_gen;
10176 u64 feature;
10177 int mixed;
10178
10179 feature = btrfs_super_incompat_flags(info->super_copy);
10180 mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS);
10181
10182 root = info->extent_root;
10183 key.objectid = 0;
10184 key.offset = 0;
10185 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10186 path = btrfs_alloc_path();
10187 if (!path)
10188 return -ENOMEM;
10189 path->reada = READA_FORWARD;
10190
10191 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
10192 if (btrfs_test_opt(root->fs_info, SPACE_CACHE) &&
10193 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
10194 need_clear = 1;
10195 if (btrfs_test_opt(root->fs_info, CLEAR_CACHE))
10196 need_clear = 1;
10197
10198 while (1) {
10199 ret = find_first_block_group(root, path, &key);
10200 if (ret > 0)
10201 break;
10202 if (ret != 0)
10203 goto error;
10204
10205 leaf = path->nodes[0];
10206 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
10207
10208 cache = btrfs_create_block_group_cache(root, found_key.objectid,
10209 found_key.offset);
10210 if (!cache) {
10211 ret = -ENOMEM;
10212 goto error;
10213 }
10214
10215 if (need_clear) {
10216 /*
10217 * When we mount with old space cache, we need to
10218 * set BTRFS_DC_CLEAR and set dirty flag.
10219 *
10220 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10221 * truncate the old free space cache inode and
10222 * setup a new one.
10223 * b) Setting 'dirty flag' makes sure that we flush
10224 * the new space cache info onto disk.
10225 */
10226 if (btrfs_test_opt(root->fs_info, SPACE_CACHE))
10227 cache->disk_cache_state = BTRFS_DC_CLEAR;
10228 }
10229
10230 read_extent_buffer(leaf, &cache->item,
10231 btrfs_item_ptr_offset(leaf, path->slots[0]),
10232 sizeof(cache->item));
10233 cache->flags = btrfs_block_group_flags(&cache->item);
10234 if (!mixed &&
10235 ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) &&
10236 (cache->flags & BTRFS_BLOCK_GROUP_DATA))) {
10237 btrfs_err(info,
10238 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
10239 cache->key.objectid);
10240 ret = -EINVAL;
10241 goto error;
10242 }
10243
10244 key.objectid = found_key.objectid + found_key.offset;
10245 btrfs_release_path(path);
10246
10247 /*
10248 * We need to exclude the super stripes now so that the space
10249 * info has super bytes accounted for, otherwise we'll think
10250 * we have more space than we actually do.
10251 */
10252 ret = exclude_super_stripes(root, cache);
10253 if (ret) {
10254 /*
10255 * We may have excluded something, so call this just in
10256 * case.
10257 */
10258 free_excluded_extents(root, cache);
10259 btrfs_put_block_group(cache);
10260 goto error;
10261 }
10262
10263 /*
10264 * check for two cases, either we are full, and therefore
10265 * don't need to bother with the caching work since we won't
10266 * find any space, or we are empty, and we can just add all
10267 * the space in and be done with it. This saves us _alot_ of
10268 * time, particularly in the full case.
10269 */
10270 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
10271 cache->last_byte_to_unpin = (u64)-1;
10272 cache->cached = BTRFS_CACHE_FINISHED;
10273 free_excluded_extents(root, cache);
10274 } else if (btrfs_block_group_used(&cache->item) == 0) {
10275 cache->last_byte_to_unpin = (u64)-1;
10276 cache->cached = BTRFS_CACHE_FINISHED;
10277 add_new_free_space(cache, root->fs_info,
10278 found_key.objectid,
10279 found_key.objectid +
10280 found_key.offset);
10281 free_excluded_extents(root, cache);
10282 }
10283
10284 ret = btrfs_add_block_group_cache(root->fs_info, cache);
10285 if (ret) {
10286 btrfs_remove_free_space_cache(cache);
10287 btrfs_put_block_group(cache);
10288 goto error;
10289 }
10290
10291 trace_btrfs_add_block_group(root->fs_info, cache, 0);
10292 ret = update_space_info(info, cache->flags, found_key.offset,
10293 btrfs_block_group_used(&cache->item),
10294 cache->bytes_super, &space_info);
10295 if (ret) {
10296 btrfs_remove_free_space_cache(cache);
10297 spin_lock(&info->block_group_cache_lock);
10298 rb_erase(&cache->cache_node,
10299 &info->block_group_cache_tree);
10300 RB_CLEAR_NODE(&cache->cache_node);
10301 spin_unlock(&info->block_group_cache_lock);
10302 btrfs_put_block_group(cache);
10303 goto error;
10304 }
10305
10306 cache->space_info = space_info;
10307
10308 __link_block_group(space_info, cache);
10309
10310 set_avail_alloc_bits(root->fs_info, cache->flags);
10311 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
10312 inc_block_group_ro(cache, 1);
10313 } else if (btrfs_block_group_used(&cache->item) == 0) {
10314 spin_lock(&info->unused_bgs_lock);
10315 /* Should always be true but just in case. */
10316 if (list_empty(&cache->bg_list)) {
10317 btrfs_get_block_group(cache);
10318 list_add_tail(&cache->bg_list,
10319 &info->unused_bgs);
10320 }
10321 spin_unlock(&info->unused_bgs_lock);
10322 }
10323 }
10324
10325 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
10326 if (!(get_alloc_profile(root, space_info->flags) &
10327 (BTRFS_BLOCK_GROUP_RAID10 |
10328 BTRFS_BLOCK_GROUP_RAID1 |
10329 BTRFS_BLOCK_GROUP_RAID5 |
10330 BTRFS_BLOCK_GROUP_RAID6 |
10331 BTRFS_BLOCK_GROUP_DUP)))
10332 continue;
10333 /*
10334 * avoid allocating from un-mirrored block group if there are
10335 * mirrored block groups.
10336 */
10337 list_for_each_entry(cache,
10338 &space_info->block_groups[BTRFS_RAID_RAID0],
10339 list)
10340 inc_block_group_ro(cache, 1);
10341 list_for_each_entry(cache,
10342 &space_info->block_groups[BTRFS_RAID_SINGLE],
10343 list)
10344 inc_block_group_ro(cache, 1);
10345 }
10346
10347 init_global_block_rsv(info);
10348 ret = 0;
10349 error:
10350 btrfs_free_path(path);
10351 return ret;
10352 }
10353
10354 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
10355 struct btrfs_root *root)
10356 {
10357 struct btrfs_block_group_cache *block_group, *tmp;
10358 struct btrfs_root *extent_root = root->fs_info->extent_root;
10359 struct btrfs_block_group_item item;
10360 struct btrfs_key key;
10361 int ret = 0;
10362 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
10363
10364 trans->can_flush_pending_bgs = false;
10365 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
10366 if (ret)
10367 goto next;
10368
10369 spin_lock(&block_group->lock);
10370 memcpy(&item, &block_group->item, sizeof(item));
10371 memcpy(&key, &block_group->key, sizeof(key));
10372 spin_unlock(&block_group->lock);
10373
10374 ret = btrfs_insert_item(trans, extent_root, &key, &item,
10375 sizeof(item));
10376 if (ret)
10377 btrfs_abort_transaction(trans, ret);
10378 ret = btrfs_finish_chunk_alloc(trans, extent_root,
10379 key.objectid, key.offset);
10380 if (ret)
10381 btrfs_abort_transaction(trans, ret);
10382 add_block_group_free_space(trans, root->fs_info, block_group);
10383 /* already aborted the transaction if it failed. */
10384 next:
10385 list_del_init(&block_group->bg_list);
10386 }
10387 trans->can_flush_pending_bgs = can_flush_pending_bgs;
10388 }
10389
10390 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
10391 struct btrfs_root *root, u64 bytes_used,
10392 u64 type, u64 chunk_objectid, u64 chunk_offset,
10393 u64 size)
10394 {
10395 int ret;
10396 struct btrfs_root *extent_root;
10397 struct btrfs_block_group_cache *cache;
10398 extent_root = root->fs_info->extent_root;
10399
10400 btrfs_set_log_full_commit(root->fs_info, trans);
10401
10402 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
10403 if (!cache)
10404 return -ENOMEM;
10405
10406 btrfs_set_block_group_used(&cache->item, bytes_used);
10407 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
10408 btrfs_set_block_group_flags(&cache->item, type);
10409
10410 cache->flags = type;
10411 cache->last_byte_to_unpin = (u64)-1;
10412 cache->cached = BTRFS_CACHE_FINISHED;
10413 cache->needs_free_space = 1;
10414 ret = exclude_super_stripes(root, cache);
10415 if (ret) {
10416 /*
10417 * We may have excluded something, so call this just in
10418 * case.
10419 */
10420 free_excluded_extents(root, cache);
10421 btrfs_put_block_group(cache);
10422 return ret;
10423 }
10424
10425 add_new_free_space(cache, root->fs_info, chunk_offset,
10426 chunk_offset + size);
10427
10428 free_excluded_extents(root, cache);
10429
10430 #ifdef CONFIG_BTRFS_DEBUG
10431 if (btrfs_should_fragment_free_space(root, cache)) {
10432 u64 new_bytes_used = size - bytes_used;
10433
10434 bytes_used += new_bytes_used >> 1;
10435 fragment_free_space(root, cache);
10436 }
10437 #endif
10438 /*
10439 * Call to ensure the corresponding space_info object is created and
10440 * assigned to our block group, but don't update its counters just yet.
10441 * We want our bg to be added to the rbtree with its ->space_info set.
10442 */
10443 ret = update_space_info(root->fs_info, cache->flags, 0, 0, 0,
10444 &cache->space_info);
10445 if (ret) {
10446 btrfs_remove_free_space_cache(cache);
10447 btrfs_put_block_group(cache);
10448 return ret;
10449 }
10450
10451 ret = btrfs_add_block_group_cache(root->fs_info, cache);
10452 if (ret) {
10453 btrfs_remove_free_space_cache(cache);
10454 btrfs_put_block_group(cache);
10455 return ret;
10456 }
10457
10458 /*
10459 * Now that our block group has its ->space_info set and is inserted in
10460 * the rbtree, update the space info's counters.
10461 */
10462 trace_btrfs_add_block_group(root->fs_info, cache, 1);
10463 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
10464 cache->bytes_super, &cache->space_info);
10465 if (ret) {
10466 btrfs_remove_free_space_cache(cache);
10467 spin_lock(&root->fs_info->block_group_cache_lock);
10468 rb_erase(&cache->cache_node,
10469 &root->fs_info->block_group_cache_tree);
10470 RB_CLEAR_NODE(&cache->cache_node);
10471 spin_unlock(&root->fs_info->block_group_cache_lock);
10472 btrfs_put_block_group(cache);
10473 return ret;
10474 }
10475 update_global_block_rsv(root->fs_info);
10476
10477 __link_block_group(cache->space_info, cache);
10478
10479 list_add_tail(&cache->bg_list, &trans->new_bgs);
10480
10481 set_avail_alloc_bits(extent_root->fs_info, type);
10482 return 0;
10483 }
10484
10485 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10486 {
10487 u64 extra_flags = chunk_to_extended(flags) &
10488 BTRFS_EXTENDED_PROFILE_MASK;
10489
10490 write_seqlock(&fs_info->profiles_lock);
10491 if (flags & BTRFS_BLOCK_GROUP_DATA)
10492 fs_info->avail_data_alloc_bits &= ~extra_flags;
10493 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10494 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10495 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10496 fs_info->avail_system_alloc_bits &= ~extra_flags;
10497 write_sequnlock(&fs_info->profiles_lock);
10498 }
10499
10500 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10501 struct btrfs_root *root, u64 group_start,
10502 struct extent_map *em)
10503 {
10504 struct btrfs_path *path;
10505 struct btrfs_block_group_cache *block_group;
10506 struct btrfs_free_cluster *cluster;
10507 struct btrfs_root *tree_root = root->fs_info->tree_root;
10508 struct btrfs_key key;
10509 struct inode *inode;
10510 struct kobject *kobj = NULL;
10511 int ret;
10512 int index;
10513 int factor;
10514 struct btrfs_caching_control *caching_ctl = NULL;
10515 bool remove_em;
10516
10517 root = root->fs_info->extent_root;
10518
10519 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
10520 BUG_ON(!block_group);
10521 BUG_ON(!block_group->ro);
10522
10523 /*
10524 * Free the reserved super bytes from this block group before
10525 * remove it.
10526 */
10527 free_excluded_extents(root, block_group);
10528
10529 memcpy(&key, &block_group->key, sizeof(key));
10530 index = get_block_group_index(block_group);
10531 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
10532 BTRFS_BLOCK_GROUP_RAID1 |
10533 BTRFS_BLOCK_GROUP_RAID10))
10534 factor = 2;
10535 else
10536 factor = 1;
10537
10538 /* make sure this block group isn't part of an allocation cluster */
10539 cluster = &root->fs_info->data_alloc_cluster;
10540 spin_lock(&cluster->refill_lock);
10541 btrfs_return_cluster_to_free_space(block_group, cluster);
10542 spin_unlock(&cluster->refill_lock);
10543
10544 /*
10545 * make sure this block group isn't part of a metadata
10546 * allocation cluster
10547 */
10548 cluster = &root->fs_info->meta_alloc_cluster;
10549 spin_lock(&cluster->refill_lock);
10550 btrfs_return_cluster_to_free_space(block_group, cluster);
10551 spin_unlock(&cluster->refill_lock);
10552
10553 path = btrfs_alloc_path();
10554 if (!path) {
10555 ret = -ENOMEM;
10556 goto out;
10557 }
10558
10559 /*
10560 * get the inode first so any iput calls done for the io_list
10561 * aren't the final iput (no unlinks allowed now)
10562 */
10563 inode = lookup_free_space_inode(tree_root, block_group, path);
10564
10565 mutex_lock(&trans->transaction->cache_write_mutex);
10566 /*
10567 * make sure our free spache cache IO is done before remove the
10568 * free space inode
10569 */
10570 spin_lock(&trans->transaction->dirty_bgs_lock);
10571 if (!list_empty(&block_group->io_list)) {
10572 list_del_init(&block_group->io_list);
10573
10574 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10575
10576 spin_unlock(&trans->transaction->dirty_bgs_lock);
10577 btrfs_wait_cache_io(root, trans, block_group,
10578 &block_group->io_ctl, path,
10579 block_group->key.objectid);
10580 btrfs_put_block_group(block_group);
10581 spin_lock(&trans->transaction->dirty_bgs_lock);
10582 }
10583
10584 if (!list_empty(&block_group->dirty_list)) {
10585 list_del_init(&block_group->dirty_list);
10586 btrfs_put_block_group(block_group);
10587 }
10588 spin_unlock(&trans->transaction->dirty_bgs_lock);
10589 mutex_unlock(&trans->transaction->cache_write_mutex);
10590
10591 if (!IS_ERR(inode)) {
10592 ret = btrfs_orphan_add(trans, inode);
10593 if (ret) {
10594 btrfs_add_delayed_iput(inode);
10595 goto out;
10596 }
10597 clear_nlink(inode);
10598 /* One for the block groups ref */
10599 spin_lock(&block_group->lock);
10600 if (block_group->iref) {
10601 block_group->iref = 0;
10602 block_group->inode = NULL;
10603 spin_unlock(&block_group->lock);
10604 iput(inode);
10605 } else {
10606 spin_unlock(&block_group->lock);
10607 }
10608 /* One for our lookup ref */
10609 btrfs_add_delayed_iput(inode);
10610 }
10611
10612 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10613 key.offset = block_group->key.objectid;
10614 key.type = 0;
10615
10616 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10617 if (ret < 0)
10618 goto out;
10619 if (ret > 0)
10620 btrfs_release_path(path);
10621 if (ret == 0) {
10622 ret = btrfs_del_item(trans, tree_root, path);
10623 if (ret)
10624 goto out;
10625 btrfs_release_path(path);
10626 }
10627
10628 spin_lock(&root->fs_info->block_group_cache_lock);
10629 rb_erase(&block_group->cache_node,
10630 &root->fs_info->block_group_cache_tree);
10631 RB_CLEAR_NODE(&block_group->cache_node);
10632
10633 if (root->fs_info->first_logical_byte == block_group->key.objectid)
10634 root->fs_info->first_logical_byte = (u64)-1;
10635 spin_unlock(&root->fs_info->block_group_cache_lock);
10636
10637 down_write(&block_group->space_info->groups_sem);
10638 /*
10639 * we must use list_del_init so people can check to see if they
10640 * are still on the list after taking the semaphore
10641 */
10642 list_del_init(&block_group->list);
10643 if (list_empty(&block_group->space_info->block_groups[index])) {
10644 kobj = block_group->space_info->block_group_kobjs[index];
10645 block_group->space_info->block_group_kobjs[index] = NULL;
10646 clear_avail_alloc_bits(root->fs_info, block_group->flags);
10647 }
10648 up_write(&block_group->space_info->groups_sem);
10649 if (kobj) {
10650 kobject_del(kobj);
10651 kobject_put(kobj);
10652 }
10653
10654 if (block_group->has_caching_ctl)
10655 caching_ctl = get_caching_control(block_group);
10656 if (block_group->cached == BTRFS_CACHE_STARTED)
10657 wait_block_group_cache_done(block_group);
10658 if (block_group->has_caching_ctl) {
10659 down_write(&root->fs_info->commit_root_sem);
10660 if (!caching_ctl) {
10661 struct btrfs_caching_control *ctl;
10662
10663 list_for_each_entry(ctl,
10664 &root->fs_info->caching_block_groups, list)
10665 if (ctl->block_group == block_group) {
10666 caching_ctl = ctl;
10667 atomic_inc(&caching_ctl->count);
10668 break;
10669 }
10670 }
10671 if (caching_ctl)
10672 list_del_init(&caching_ctl->list);
10673 up_write(&root->fs_info->commit_root_sem);
10674 if (caching_ctl) {
10675 /* Once for the caching bgs list and once for us. */
10676 put_caching_control(caching_ctl);
10677 put_caching_control(caching_ctl);
10678 }
10679 }
10680
10681 spin_lock(&trans->transaction->dirty_bgs_lock);
10682 if (!list_empty(&block_group->dirty_list)) {
10683 WARN_ON(1);
10684 }
10685 if (!list_empty(&block_group->io_list)) {
10686 WARN_ON(1);
10687 }
10688 spin_unlock(&trans->transaction->dirty_bgs_lock);
10689 btrfs_remove_free_space_cache(block_group);
10690
10691 spin_lock(&block_group->space_info->lock);
10692 list_del_init(&block_group->ro_list);
10693
10694 if (btrfs_test_opt(root->fs_info, ENOSPC_DEBUG)) {
10695 WARN_ON(block_group->space_info->total_bytes
10696 < block_group->key.offset);
10697 WARN_ON(block_group->space_info->bytes_readonly
10698 < block_group->key.offset);
10699 WARN_ON(block_group->space_info->disk_total
10700 < block_group->key.offset * factor);
10701 }
10702 block_group->space_info->total_bytes -= block_group->key.offset;
10703 block_group->space_info->bytes_readonly -= block_group->key.offset;
10704 block_group->space_info->disk_total -= block_group->key.offset * factor;
10705
10706 spin_unlock(&block_group->space_info->lock);
10707
10708 memcpy(&key, &block_group->key, sizeof(key));
10709
10710 lock_chunks(root);
10711 if (!list_empty(&em->list)) {
10712 /* We're in the transaction->pending_chunks list. */
10713 free_extent_map(em);
10714 }
10715 spin_lock(&block_group->lock);
10716 block_group->removed = 1;
10717 /*
10718 * At this point trimming can't start on this block group, because we
10719 * removed the block group from the tree fs_info->block_group_cache_tree
10720 * so no one can't find it anymore and even if someone already got this
10721 * block group before we removed it from the rbtree, they have already
10722 * incremented block_group->trimming - if they didn't, they won't find
10723 * any free space entries because we already removed them all when we
10724 * called btrfs_remove_free_space_cache().
10725 *
10726 * And we must not remove the extent map from the fs_info->mapping_tree
10727 * to prevent the same logical address range and physical device space
10728 * ranges from being reused for a new block group. This is because our
10729 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10730 * completely transactionless, so while it is trimming a range the
10731 * currently running transaction might finish and a new one start,
10732 * allowing for new block groups to be created that can reuse the same
10733 * physical device locations unless we take this special care.
10734 *
10735 * There may also be an implicit trim operation if the file system
10736 * is mounted with -odiscard. The same protections must remain
10737 * in place until the extents have been discarded completely when
10738 * the transaction commit has completed.
10739 */
10740 remove_em = (atomic_read(&block_group->trimming) == 0);
10741 /*
10742 * Make sure a trimmer task always sees the em in the pinned_chunks list
10743 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10744 * before checking block_group->removed).
10745 */
10746 if (!remove_em) {
10747 /*
10748 * Our em might be in trans->transaction->pending_chunks which
10749 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10750 * and so is the fs_info->pinned_chunks list.
10751 *
10752 * So at this point we must be holding the chunk_mutex to avoid
10753 * any races with chunk allocation (more specifically at
10754 * volumes.c:contains_pending_extent()), to ensure it always
10755 * sees the em, either in the pending_chunks list or in the
10756 * pinned_chunks list.
10757 */
10758 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
10759 }
10760 spin_unlock(&block_group->lock);
10761
10762 if (remove_em) {
10763 struct extent_map_tree *em_tree;
10764
10765 em_tree = &root->fs_info->mapping_tree.map_tree;
10766 write_lock(&em_tree->lock);
10767 /*
10768 * The em might be in the pending_chunks list, so make sure the
10769 * chunk mutex is locked, since remove_extent_mapping() will
10770 * delete us from that list.
10771 */
10772 remove_extent_mapping(em_tree, em);
10773 write_unlock(&em_tree->lock);
10774 /* once for the tree */
10775 free_extent_map(em);
10776 }
10777
10778 unlock_chunks(root);
10779
10780 ret = remove_block_group_free_space(trans, root->fs_info, block_group);
10781 if (ret)
10782 goto out;
10783
10784 btrfs_put_block_group(block_group);
10785 btrfs_put_block_group(block_group);
10786
10787 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10788 if (ret > 0)
10789 ret = -EIO;
10790 if (ret < 0)
10791 goto out;
10792
10793 ret = btrfs_del_item(trans, root, path);
10794 out:
10795 btrfs_free_path(path);
10796 return ret;
10797 }
10798
10799 struct btrfs_trans_handle *
10800 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10801 const u64 chunk_offset)
10802 {
10803 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10804 struct extent_map *em;
10805 struct map_lookup *map;
10806 unsigned int num_items;
10807
10808 read_lock(&em_tree->lock);
10809 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10810 read_unlock(&em_tree->lock);
10811 ASSERT(em && em->start == chunk_offset);
10812
10813 /*
10814 * We need to reserve 3 + N units from the metadata space info in order
10815 * to remove a block group (done at btrfs_remove_chunk() and at
10816 * btrfs_remove_block_group()), which are used for:
10817 *
10818 * 1 unit for adding the free space inode's orphan (located in the tree
10819 * of tree roots).
10820 * 1 unit for deleting the block group item (located in the extent
10821 * tree).
10822 * 1 unit for deleting the free space item (located in tree of tree
10823 * roots).
10824 * N units for deleting N device extent items corresponding to each
10825 * stripe (located in the device tree).
10826 *
10827 * In order to remove a block group we also need to reserve units in the
10828 * system space info in order to update the chunk tree (update one or
10829 * more device items and remove one chunk item), but this is done at
10830 * btrfs_remove_chunk() through a call to check_system_chunk().
10831 */
10832 map = em->map_lookup;
10833 num_items = 3 + map->num_stripes;
10834 free_extent_map(em);
10835
10836 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10837 num_items, 1);
10838 }
10839
10840 /*
10841 * Process the unused_bgs list and remove any that don't have any allocated
10842 * space inside of them.
10843 */
10844 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10845 {
10846 struct btrfs_block_group_cache *block_group;
10847 struct btrfs_space_info *space_info;
10848 struct btrfs_root *root = fs_info->extent_root;
10849 struct btrfs_trans_handle *trans;
10850 int ret = 0;
10851
10852 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
10853 return;
10854
10855 spin_lock(&fs_info->unused_bgs_lock);
10856 while (!list_empty(&fs_info->unused_bgs)) {
10857 u64 start, end;
10858 int trimming;
10859
10860 block_group = list_first_entry(&fs_info->unused_bgs,
10861 struct btrfs_block_group_cache,
10862 bg_list);
10863 list_del_init(&block_group->bg_list);
10864
10865 space_info = block_group->space_info;
10866
10867 if (ret || btrfs_mixed_space_info(space_info)) {
10868 btrfs_put_block_group(block_group);
10869 continue;
10870 }
10871 spin_unlock(&fs_info->unused_bgs_lock);
10872
10873 down_write(&root->fs_info->bg_delete_sem);
10874
10875 /* Don't want to race with allocators so take the groups_sem */
10876 down_write(&space_info->groups_sem);
10877 spin_lock(&block_group->lock);
10878 if (block_group->reserved ||
10879 btrfs_block_group_used(&block_group->item) ||
10880 (block_group->ro && !block_group->removed) ||
10881 list_is_singular(&block_group->list)) {
10882 /*
10883 * We want to bail if we made new allocations or have
10884 * outstanding allocations in this block group. We do
10885 * the ro check in case balance is currently acting on
10886 * this block group.
10887 */
10888 spin_unlock(&block_group->lock);
10889 up_write(&space_info->groups_sem);
10890 goto next;
10891 }
10892 spin_unlock(&block_group->lock);
10893
10894 /* We don't want to force the issue, only flip if it's ok. */
10895 ret = inc_block_group_ro(block_group, 0);
10896 up_write(&space_info->groups_sem);
10897 if (ret < 0) {
10898 ret = 0;
10899 goto next;
10900 }
10901
10902 /*
10903 * Want to do this before we do anything else so we can recover
10904 * properly if we fail to join the transaction.
10905 */
10906 trans = btrfs_start_trans_remove_block_group(fs_info,
10907 block_group->key.objectid);
10908 if (IS_ERR(trans)) {
10909 btrfs_dec_block_group_ro(root, block_group);
10910 ret = PTR_ERR(trans);
10911 goto next;
10912 }
10913
10914 /*
10915 * We could have pending pinned extents for this block group,
10916 * just delete them, we don't care about them anymore.
10917 */
10918 start = block_group->key.objectid;
10919 end = start + block_group->key.offset - 1;
10920 /*
10921 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10922 * btrfs_finish_extent_commit(). If we are at transaction N,
10923 * another task might be running finish_extent_commit() for the
10924 * previous transaction N - 1, and have seen a range belonging
10925 * to the block group in freed_extents[] before we were able to
10926 * clear the whole block group range from freed_extents[]. This
10927 * means that task can lookup for the block group after we
10928 * unpinned it from freed_extents[] and removed it, leading to
10929 * a BUG_ON() at btrfs_unpin_extent_range().
10930 */
10931 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10932 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10933 EXTENT_DIRTY);
10934 if (ret) {
10935 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10936 btrfs_dec_block_group_ro(root, block_group);
10937 goto end_trans;
10938 }
10939 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10940 EXTENT_DIRTY);
10941 if (ret) {
10942 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10943 btrfs_dec_block_group_ro(root, block_group);
10944 goto end_trans;
10945 }
10946 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10947
10948 /* Reset pinned so btrfs_put_block_group doesn't complain */
10949 spin_lock(&space_info->lock);
10950 spin_lock(&block_group->lock);
10951
10952 space_info->bytes_pinned -= block_group->pinned;
10953 space_info->bytes_readonly += block_group->pinned;
10954 percpu_counter_add(&space_info->total_bytes_pinned,
10955 -block_group->pinned);
10956 block_group->pinned = 0;
10957
10958 spin_unlock(&block_group->lock);
10959 spin_unlock(&space_info->lock);
10960
10961 /* DISCARD can flip during remount */
10962 trimming = btrfs_test_opt(root->fs_info, DISCARD);
10963
10964 /* Implicit trim during transaction commit. */
10965 if (trimming)
10966 btrfs_get_block_group_trimming(block_group);
10967
10968 /*
10969 * Btrfs_remove_chunk will abort the transaction if things go
10970 * horribly wrong.
10971 */
10972 ret = btrfs_remove_chunk(trans, root,
10973 block_group->key.objectid);
10974
10975 if (ret) {
10976 if (trimming)
10977 btrfs_put_block_group_trimming(block_group);
10978 goto end_trans;
10979 }
10980
10981 /*
10982 * If we're not mounted with -odiscard, we can just forget
10983 * about this block group. Otherwise we'll need to wait
10984 * until transaction commit to do the actual discard.
10985 */
10986 if (trimming) {
10987 spin_lock(&fs_info->unused_bgs_lock);
10988 /*
10989 * A concurrent scrub might have added us to the list
10990 * fs_info->unused_bgs, so use a list_move operation
10991 * to add the block group to the deleted_bgs list.
10992 */
10993 list_move(&block_group->bg_list,
10994 &trans->transaction->deleted_bgs);
10995 spin_unlock(&fs_info->unused_bgs_lock);
10996 btrfs_get_block_group(block_group);
10997 }
10998 end_trans:
10999 btrfs_end_transaction(trans, root);
11000 next:
11001 up_write(&root->fs_info->bg_delete_sem);
11002 btrfs_put_block_group(block_group);
11003 spin_lock(&fs_info->unused_bgs_lock);
11004 }
11005 spin_unlock(&fs_info->unused_bgs_lock);
11006 }
11007
11008 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
11009 {
11010 struct btrfs_space_info *space_info;
11011 struct btrfs_super_block *disk_super;
11012 u64 features;
11013 u64 flags;
11014 int mixed = 0;
11015 int ret;
11016
11017 disk_super = fs_info->super_copy;
11018 if (!btrfs_super_root(disk_super))
11019 return -EINVAL;
11020
11021 features = btrfs_super_incompat_flags(disk_super);
11022 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
11023 mixed = 1;
11024
11025 flags = BTRFS_BLOCK_GROUP_SYSTEM;
11026 ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info);
11027 if (ret)
11028 goto out;
11029
11030 if (mixed) {
11031 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
11032 ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info);
11033 } else {
11034 flags = BTRFS_BLOCK_GROUP_METADATA;
11035 ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info);
11036 if (ret)
11037 goto out;
11038
11039 flags = BTRFS_BLOCK_GROUP_DATA;
11040 ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info);
11041 }
11042 out:
11043 return ret;
11044 }
11045
11046 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
11047 {
11048 return unpin_extent_range(root, start, end, false);
11049 }
11050
11051 /*
11052 * It used to be that old block groups would be left around forever.
11053 * Iterating over them would be enough to trim unused space. Since we
11054 * now automatically remove them, we also need to iterate over unallocated
11055 * space.
11056 *
11057 * We don't want a transaction for this since the discard may take a
11058 * substantial amount of time. We don't require that a transaction be
11059 * running, but we do need to take a running transaction into account
11060 * to ensure that we're not discarding chunks that were released in
11061 * the current transaction.
11062 *
11063 * Holding the chunks lock will prevent other threads from allocating
11064 * or releasing chunks, but it won't prevent a running transaction
11065 * from committing and releasing the memory that the pending chunks
11066 * list head uses. For that, we need to take a reference to the
11067 * transaction.
11068 */
11069 static int btrfs_trim_free_extents(struct btrfs_device *device,
11070 u64 minlen, u64 *trimmed)
11071 {
11072 u64 start = 0, len = 0;
11073 int ret;
11074
11075 *trimmed = 0;
11076
11077 /* Not writeable = nothing to do. */
11078 if (!device->writeable)
11079 return 0;
11080
11081 /* No free space = nothing to do. */
11082 if (device->total_bytes <= device->bytes_used)
11083 return 0;
11084
11085 ret = 0;
11086
11087 while (1) {
11088 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
11089 struct btrfs_transaction *trans;
11090 u64 bytes;
11091
11092 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
11093 if (ret)
11094 return ret;
11095
11096 down_read(&fs_info->commit_root_sem);
11097
11098 spin_lock(&fs_info->trans_lock);
11099 trans = fs_info->running_transaction;
11100 if (trans)
11101 atomic_inc(&trans->use_count);
11102 spin_unlock(&fs_info->trans_lock);
11103
11104 ret = find_free_dev_extent_start(trans, device, minlen, start,
11105 &start, &len);
11106 if (trans)
11107 btrfs_put_transaction(trans);
11108
11109 if (ret) {
11110 up_read(&fs_info->commit_root_sem);
11111 mutex_unlock(&fs_info->chunk_mutex);
11112 if (ret == -ENOSPC)
11113 ret = 0;
11114 break;
11115 }
11116
11117 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
11118 up_read(&fs_info->commit_root_sem);
11119 mutex_unlock(&fs_info->chunk_mutex);
11120
11121 if (ret)
11122 break;
11123
11124 start += len;
11125 *trimmed += bytes;
11126
11127 if (fatal_signal_pending(current)) {
11128 ret = -ERESTARTSYS;
11129 break;
11130 }
11131
11132 cond_resched();
11133 }
11134
11135 return ret;
11136 }
11137
11138 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
11139 {
11140 struct btrfs_fs_info *fs_info = root->fs_info;
11141 struct btrfs_block_group_cache *cache = NULL;
11142 struct btrfs_device *device;
11143 struct list_head *devices;
11144 u64 group_trimmed;
11145 u64 start;
11146 u64 end;
11147 u64 trimmed = 0;
11148 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
11149 int ret = 0;
11150
11151 /*
11152 * try to trim all FS space, our block group may start from non-zero.
11153 */
11154 if (range->len == total_bytes)
11155 cache = btrfs_lookup_first_block_group(fs_info, range->start);
11156 else
11157 cache = btrfs_lookup_block_group(fs_info, range->start);
11158
11159 while (cache) {
11160 if (cache->key.objectid >= (range->start + range->len)) {
11161 btrfs_put_block_group(cache);
11162 break;
11163 }
11164
11165 start = max(range->start, cache->key.objectid);
11166 end = min(range->start + range->len,
11167 cache->key.objectid + cache->key.offset);
11168
11169 if (end - start >= range->minlen) {
11170 if (!block_group_cache_done(cache)) {
11171 ret = cache_block_group(cache, 0);
11172 if (ret) {
11173 btrfs_put_block_group(cache);
11174 break;
11175 }
11176 ret = wait_block_group_cache_done(cache);
11177 if (ret) {
11178 btrfs_put_block_group(cache);
11179 break;
11180 }
11181 }
11182 ret = btrfs_trim_block_group(cache,
11183 &group_trimmed,
11184 start,
11185 end,
11186 range->minlen);
11187
11188 trimmed += group_trimmed;
11189 if (ret) {
11190 btrfs_put_block_group(cache);
11191 break;
11192 }
11193 }
11194
11195 cache = next_block_group(fs_info->tree_root, cache);
11196 }
11197
11198 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
11199 devices = &root->fs_info->fs_devices->alloc_list;
11200 list_for_each_entry(device, devices, dev_alloc_list) {
11201 ret = btrfs_trim_free_extents(device, range->minlen,
11202 &group_trimmed);
11203 if (ret)
11204 break;
11205
11206 trimmed += group_trimmed;
11207 }
11208 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
11209
11210 range->len = trimmed;
11211 return ret;
11212 }
11213
11214 /*
11215 * btrfs_{start,end}_write_no_snapshoting() are similar to
11216 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
11217 * data into the page cache through nocow before the subvolume is snapshoted,
11218 * but flush the data into disk after the snapshot creation, or to prevent
11219 * operations while snapshoting is ongoing and that cause the snapshot to be
11220 * inconsistent (writes followed by expanding truncates for example).
11221 */
11222 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
11223 {
11224 percpu_counter_dec(&root->subv_writers->counter);
11225 /*
11226 * Make sure counter is updated before we wake up waiters.
11227 */
11228 smp_mb();
11229 if (waitqueue_active(&root->subv_writers->wait))
11230 wake_up(&root->subv_writers->wait);
11231 }
11232
11233 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
11234 {
11235 if (atomic_read(&root->will_be_snapshoted))
11236 return 0;
11237
11238 percpu_counter_inc(&root->subv_writers->counter);
11239 /*
11240 * Make sure counter is updated before we check for snapshot creation.
11241 */
11242 smp_mb();
11243 if (atomic_read(&root->will_be_snapshoted)) {
11244 btrfs_end_write_no_snapshoting(root);
11245 return 0;
11246 }
11247 return 1;
11248 }
11249
11250 static int wait_snapshoting_atomic_t(atomic_t *a)
11251 {
11252 schedule();
11253 return 0;
11254 }
11255
11256 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
11257 {
11258 while (true) {
11259 int ret;
11260
11261 ret = btrfs_start_write_no_snapshoting(root);
11262 if (ret)
11263 break;
11264 wait_on_atomic_t(&root->will_be_snapshoted,
11265 wait_snapshoting_atomic_t,
11266 TASK_UNINTERRUPTIBLE);
11267 }
11268 }
Linux kernel - Deliverables
This page took 0.284743 seconds and 5 git commands to generate.