projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
btrfs: hide test-only member under ifdef
[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 /*
64 * Control how reservations are dealt with.
65 *
66 * RESERVE_FREE - freeing a reservation.
67 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
68 * ENOSPC accounting
69 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
70 * bytes_may_use as the ENOSPC accounting is done elsewhere
71 */
72 enum {
73 RESERVE_FREE = 0,
74 RESERVE_ALLOC = 1,
75 RESERVE_ALLOC_NO_ACCOUNT = 2,
76 };
77
78 static int update_block_group(struct btrfs_trans_handle *trans,
79 struct btrfs_root *root, u64 bytenr,
80 u64 num_bytes, int alloc);
81 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
82 struct btrfs_root *root,
83 struct btrfs_delayed_ref_node *node, u64 parent,
84 u64 root_objectid, u64 owner_objectid,
85 u64 owner_offset, int refs_to_drop,
86 struct btrfs_delayed_extent_op *extra_op);
87 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
88 struct extent_buffer *leaf,
89 struct btrfs_extent_item *ei);
90 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
91 struct btrfs_root *root,
92 u64 parent, u64 root_objectid,
93 u64 flags, u64 owner, u64 offset,
94 struct btrfs_key *ins, int ref_mod);
95 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
96 struct btrfs_root *root,
97 u64 parent, u64 root_objectid,
98 u64 flags, struct btrfs_disk_key *key,
99 int level, struct btrfs_key *ins);
100 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
101 struct btrfs_root *extent_root, u64 flags,
102 int force);
103 static int find_next_key(struct btrfs_path *path, int level,
104 struct btrfs_key *key);
105 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
106 int dump_block_groups);
107 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
108 u64 num_bytes, int reserve,
109 int delalloc);
110 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
111 u64 num_bytes);
112 int btrfs_pin_extent(struct btrfs_root *root,
113 u64 bytenr, u64 num_bytes, int reserved);
114 static int __reserve_metadata_bytes(struct btrfs_root *root,
115 struct btrfs_space_info *space_info,
116 u64 orig_bytes,
117 enum btrfs_reserve_flush_enum flush);
118 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
119 struct btrfs_space_info *space_info,
120 u64 num_bytes);
121 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
122 struct btrfs_space_info *space_info,
123 u64 num_bytes);
124
125 static noinline int
126 block_group_cache_done(struct btrfs_block_group_cache *cache)
127 {
128 smp_mb();
129 return cache->cached == BTRFS_CACHE_FINISHED ||
130 cache->cached == BTRFS_CACHE_ERROR;
131 }
132
133 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
134 {
135 return (cache->flags & bits) == bits;
136 }
137
138 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
139 {
140 atomic_inc(&cache->count);
141 }
142
143 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
144 {
145 if (atomic_dec_and_test(&cache->count)) {
146 WARN_ON(cache->pinned > 0);
147 WARN_ON(cache->reserved > 0);
148 kfree(cache->free_space_ctl);
149 kfree(cache);
150 }
151 }
152
153 /*
154 * this adds the block group to the fs_info rb tree for the block group
155 * cache
156 */
157 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
158 struct btrfs_block_group_cache *block_group)
159 {
160 struct rb_node **p;
161 struct rb_node *parent = NULL;
162 struct btrfs_block_group_cache *cache;
163
164 spin_lock(&info->block_group_cache_lock);
165 p = &info->block_group_cache_tree.rb_node;
166
167 while (*p) {
168 parent = *p;
169 cache = rb_entry(parent, struct btrfs_block_group_cache,
170 cache_node);
171 if (block_group->key.objectid < cache->key.objectid) {
172 p = &(*p)->rb_left;
173 } else if (block_group->key.objectid > cache->key.objectid) {
174 p = &(*p)->rb_right;
175 } else {
176 spin_unlock(&info->block_group_cache_lock);
177 return -EEXIST;
178 }
179 }
180
181 rb_link_node(&block_group->cache_node, parent, p);
182 rb_insert_color(&block_group->cache_node,
183 &info->block_group_cache_tree);
184
185 if (info->first_logical_byte > block_group->key.objectid)
186 info->first_logical_byte = block_group->key.objectid;
187
188 spin_unlock(&info->block_group_cache_lock);
189
190 return 0;
191 }
192
193 /*
194 * This will return the block group at or after bytenr if contains is 0, else
195 * it will return the block group that contains the bytenr
196 */
197 static struct btrfs_block_group_cache *
198 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
199 int contains)
200 {
201 struct btrfs_block_group_cache *cache, *ret = NULL;
202 struct rb_node *n;
203 u64 end, start;
204
205 spin_lock(&info->block_group_cache_lock);
206 n = info->block_group_cache_tree.rb_node;
207
208 while (n) {
209 cache = rb_entry(n, struct btrfs_block_group_cache,
210 cache_node);
211 end = cache->key.objectid + cache->key.offset - 1;
212 start = cache->key.objectid;
213
214 if (bytenr < start) {
215 if (!contains && (!ret || start < ret->key.objectid))
216 ret = cache;
217 n = n->rb_left;
218 } else if (bytenr > start) {
219 if (contains && bytenr <= end) {
220 ret = cache;
221 break;
222 }
223 n = n->rb_right;
224 } else {
225 ret = cache;
226 break;
227 }
228 }
229 if (ret) {
230 btrfs_get_block_group(ret);
231 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
232 info->first_logical_byte = ret->key.objectid;
233 }
234 spin_unlock(&info->block_group_cache_lock);
235
236 return ret;
237 }
238
239 static int add_excluded_extent(struct btrfs_root *root,
240 u64 start, u64 num_bytes)
241 {
242 u64 end = start + num_bytes - 1;
243 set_extent_bits(&root->fs_info->freed_extents[0],
244 start, end, EXTENT_UPTODATE);
245 set_extent_bits(&root->fs_info->freed_extents[1],
246 start, end, EXTENT_UPTODATE);
247 return 0;
248 }
249
250 static void free_excluded_extents(struct btrfs_root *root,
251 struct btrfs_block_group_cache *cache)
252 {
253 u64 start, end;
254
255 start = cache->key.objectid;
256 end = start + cache->key.offset - 1;
257
258 clear_extent_bits(&root->fs_info->freed_extents[0],
259 start, end, EXTENT_UPTODATE);
260 clear_extent_bits(&root->fs_info->freed_extents[1],
261 start, end, EXTENT_UPTODATE);
262 }
263
264 static int exclude_super_stripes(struct btrfs_root *root,
265 struct btrfs_block_group_cache *cache)
266 {
267 u64 bytenr;
268 u64 *logical;
269 int stripe_len;
270 int i, nr, ret;
271
272 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
273 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
274 cache->bytes_super += stripe_len;
275 ret = add_excluded_extent(root, cache->key.objectid,
276 stripe_len);
277 if (ret)
278 return ret;
279 }
280
281 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
282 bytenr = btrfs_sb_offset(i);
283 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
284 cache->key.objectid, bytenr,
285 0, &logical, &nr, &stripe_len);
286 if (ret)
287 return ret;
288
289 while (nr--) {
290 u64 start, len;
291
292 if (logical[nr] > cache->key.objectid +
293 cache->key.offset)
294 continue;
295
296 if (logical[nr] + stripe_len <= cache->key.objectid)
297 continue;
298
299 start = logical[nr];
300 if (start < cache->key.objectid) {
301 start = cache->key.objectid;
302 len = (logical[nr] + stripe_len) - start;
303 } else {
304 len = min_t(u64, stripe_len,
305 cache->key.objectid +
306 cache->key.offset - start);
307 }
308
309 cache->bytes_super += len;
310 ret = add_excluded_extent(root, start, len);
311 if (ret) {
312 kfree(logical);
313 return ret;
314 }
315 }
316
317 kfree(logical);
318 }
319 return 0;
320 }
321
322 static struct btrfs_caching_control *
323 get_caching_control(struct btrfs_block_group_cache *cache)
324 {
325 struct btrfs_caching_control *ctl;
326
327 spin_lock(&cache->lock);
328 if (!cache->caching_ctl) {
329 spin_unlock(&cache->lock);
330 return NULL;
331 }
332
333 ctl = cache->caching_ctl;
334 atomic_inc(&ctl->count);
335 spin_unlock(&cache->lock);
336 return ctl;
337 }
338
339 static void put_caching_control(struct btrfs_caching_control *ctl)
340 {
341 if (atomic_dec_and_test(&ctl->count))
342 kfree(ctl);
343 }
344
345 #ifdef CONFIG_BTRFS_DEBUG
346 static void fragment_free_space(struct btrfs_root *root,
347 struct btrfs_block_group_cache *block_group)
348 {
349 u64 start = block_group->key.objectid;
350 u64 len = block_group->key.offset;
351 u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ?
352 root->nodesize : root->sectorsize;
353 u64 step = chunk << 1;
354
355 while (len > chunk) {
356 btrfs_remove_free_space(block_group, start, chunk);
357 start += step;
358 if (len < step)
359 len = 0;
360 else
361 len -= step;
362 }
363 }
364 #endif
365
366 /*
367 * this is only called by cache_block_group, since we could have freed extents
368 * we need to check the pinned_extents for any extents that can't be used yet
369 * since their free space will be released as soon as the transaction commits.
370 */
371 u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
372 struct btrfs_fs_info *info, u64 start, u64 end)
373 {
374 u64 extent_start, extent_end, size, total_added = 0;
375 int ret;
376
377 while (start < end) {
378 ret = find_first_extent_bit(info->pinned_extents, start,
379 &extent_start, &extent_end,
380 EXTENT_DIRTY | EXTENT_UPTODATE,
381 NULL);
382 if (ret)
383 break;
384
385 if (extent_start <= start) {
386 start = extent_end + 1;
387 } else if (extent_start > start && extent_start < end) {
388 size = extent_start - start;
389 total_added += size;
390 ret = btrfs_add_free_space(block_group, start,
391 size);
392 BUG_ON(ret); /* -ENOMEM or logic error */
393 start = extent_end + 1;
394 } else {
395 break;
396 }
397 }
398
399 if (start < end) {
400 size = end - start;
401 total_added += size;
402 ret = btrfs_add_free_space(block_group, start, size);
403 BUG_ON(ret); /* -ENOMEM or logic error */
404 }
405
406 return total_added;
407 }
408
409 static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl)
410 {
411 struct btrfs_block_group_cache *block_group;
412 struct btrfs_fs_info *fs_info;
413 struct btrfs_root *extent_root;
414 struct btrfs_path *path;
415 struct extent_buffer *leaf;
416 struct btrfs_key key;
417 u64 total_found = 0;
418 u64 last = 0;
419 u32 nritems;
420 int ret;
421 bool wakeup = true;
422
423 block_group = caching_ctl->block_group;
424 fs_info = block_group->fs_info;
425 extent_root = fs_info->extent_root;
426
427 path = btrfs_alloc_path();
428 if (!path)
429 return -ENOMEM;
430
431 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
432
433 #ifdef CONFIG_BTRFS_DEBUG
434 /*
435 * If we're fragmenting we don't want to make anybody think we can
436 * allocate from this block group until we've had a chance to fragment
437 * the free space.
438 */
439 if (btrfs_should_fragment_free_space(extent_root, block_group))
440 wakeup = false;
441 #endif
442 /*
443 * We don't want to deadlock with somebody trying to allocate a new
444 * extent for the extent root while also trying to search the extent
445 * root to add free space. So we skip locking and search the commit
446 * root, since its read-only
447 */
448 path->skip_locking = 1;
449 path->search_commit_root = 1;
450 path->reada = READA_FORWARD;
451
452 key.objectid = last;
453 key.offset = 0;
454 key.type = BTRFS_EXTENT_ITEM_KEY;
455
456 next:
457 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
458 if (ret < 0)
459 goto out;
460
461 leaf = path->nodes[0];
462 nritems = btrfs_header_nritems(leaf);
463
464 while (1) {
465 if (btrfs_fs_closing(fs_info) > 1) {
466 last = (u64)-1;
467 break;
468 }
469
470 if (path->slots[0] < nritems) {
471 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
472 } else {
473 ret = find_next_key(path, 0, &key);
474 if (ret)
475 break;
476
477 if (need_resched() ||
478 rwsem_is_contended(&fs_info->commit_root_sem)) {
479 if (wakeup)
480 caching_ctl->progress = last;
481 btrfs_release_path(path);
482 up_read(&fs_info->commit_root_sem);
483 mutex_unlock(&caching_ctl->mutex);
484 cond_resched();
485 mutex_lock(&caching_ctl->mutex);
486 down_read(&fs_info->commit_root_sem);
487 goto next;
488 }
489
490 ret = btrfs_next_leaf(extent_root, path);
491 if (ret < 0)
492 goto out;
493 if (ret)
494 break;
495 leaf = path->nodes[0];
496 nritems = btrfs_header_nritems(leaf);
497 continue;
498 }
499
500 if (key.objectid < last) {
501 key.objectid = last;
502 key.offset = 0;
503 key.type = BTRFS_EXTENT_ITEM_KEY;
504
505 if (wakeup)
506 caching_ctl->progress = last;
507 btrfs_release_path(path);
508 goto next;
509 }
510
511 if (key.objectid < block_group->key.objectid) {
512 path->slots[0]++;
513 continue;
514 }
515
516 if (key.objectid >= block_group->key.objectid +
517 block_group->key.offset)
518 break;
519
520 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
521 key.type == BTRFS_METADATA_ITEM_KEY) {
522 total_found += add_new_free_space(block_group,
523 fs_info, last,
524 key.objectid);
525 if (key.type == BTRFS_METADATA_ITEM_KEY)
526 last = key.objectid +
527 fs_info->tree_root->nodesize;
528 else
529 last = key.objectid + key.offset;
530
531 if (total_found > CACHING_CTL_WAKE_UP) {
532 total_found = 0;
533 if (wakeup)
534 wake_up(&caching_ctl->wait);
535 }
536 }
537 path->slots[0]++;
538 }
539 ret = 0;
540
541 total_found += add_new_free_space(block_group, fs_info, last,
542 block_group->key.objectid +
543 block_group->key.offset);
544 caching_ctl->progress = (u64)-1;
545
546 out:
547 btrfs_free_path(path);
548 return ret;
549 }
550
551 static noinline void caching_thread(struct btrfs_work *work)
552 {
553 struct btrfs_block_group_cache *block_group;
554 struct btrfs_fs_info *fs_info;
555 struct btrfs_caching_control *caching_ctl;
556 struct btrfs_root *extent_root;
557 int ret;
558
559 caching_ctl = container_of(work, struct btrfs_caching_control, work);
560 block_group = caching_ctl->block_group;
561 fs_info = block_group->fs_info;
562 extent_root = fs_info->extent_root;
563
564 mutex_lock(&caching_ctl->mutex);
565 down_read(&fs_info->commit_root_sem);
566
567 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
568 ret = load_free_space_tree(caching_ctl);
569 else
570 ret = load_extent_tree_free(caching_ctl);
571
572 spin_lock(&block_group->lock);
573 block_group->caching_ctl = NULL;
574 block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED;
575 spin_unlock(&block_group->lock);
576
577 #ifdef CONFIG_BTRFS_DEBUG
578 if (btrfs_should_fragment_free_space(extent_root, block_group)) {
579 u64 bytes_used;
580
581 spin_lock(&block_group->space_info->lock);
582 spin_lock(&block_group->lock);
583 bytes_used = block_group->key.offset -
584 btrfs_block_group_used(&block_group->item);
585 block_group->space_info->bytes_used += bytes_used >> 1;
586 spin_unlock(&block_group->lock);
587 spin_unlock(&block_group->space_info->lock);
588 fragment_free_space(extent_root, block_group);
589 }
590 #endif
591
592 caching_ctl->progress = (u64)-1;
593
594 up_read(&fs_info->commit_root_sem);
595 free_excluded_extents(fs_info->extent_root, block_group);
596 mutex_unlock(&caching_ctl->mutex);
597
598 wake_up(&caching_ctl->wait);
599
600 put_caching_control(caching_ctl);
601 btrfs_put_block_group(block_group);
602 }
603
604 static int cache_block_group(struct btrfs_block_group_cache *cache,
605 int load_cache_only)
606 {
607 DEFINE_WAIT(wait);
608 struct btrfs_fs_info *fs_info = cache->fs_info;
609 struct btrfs_caching_control *caching_ctl;
610 int ret = 0;
611
612 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
613 if (!caching_ctl)
614 return -ENOMEM;
615
616 INIT_LIST_HEAD(&caching_ctl->list);
617 mutex_init(&caching_ctl->mutex);
618 init_waitqueue_head(&caching_ctl->wait);
619 caching_ctl->block_group = cache;
620 caching_ctl->progress = cache->key.objectid;
621 atomic_set(&caching_ctl->count, 1);
622 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
623 caching_thread, NULL, NULL);
624
625 spin_lock(&cache->lock);
626 /*
627 * This should be a rare occasion, but this could happen I think in the
628 * case where one thread starts to load the space cache info, and then
629 * some other thread starts a transaction commit which tries to do an
630 * allocation while the other thread is still loading the space cache
631 * info. The previous loop should have kept us from choosing this block
632 * group, but if we've moved to the state where we will wait on caching
633 * block groups we need to first check if we're doing a fast load here,
634 * so we can wait for it to finish, otherwise we could end up allocating
635 * from a block group who's cache gets evicted for one reason or
636 * another.
637 */
638 while (cache->cached == BTRFS_CACHE_FAST) {
639 struct btrfs_caching_control *ctl;
640
641 ctl = cache->caching_ctl;
642 atomic_inc(&ctl->count);
643 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
644 spin_unlock(&cache->lock);
645
646 schedule();
647
648 finish_wait(&ctl->wait, &wait);
649 put_caching_control(ctl);
650 spin_lock(&cache->lock);
651 }
652
653 if (cache->cached != BTRFS_CACHE_NO) {
654 spin_unlock(&cache->lock);
655 kfree(caching_ctl);
656 return 0;
657 }
658 WARN_ON(cache->caching_ctl);
659 cache->caching_ctl = caching_ctl;
660 cache->cached = BTRFS_CACHE_FAST;
661 spin_unlock(&cache->lock);
662
663 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
664 mutex_lock(&caching_ctl->mutex);
665 ret = load_free_space_cache(fs_info, cache);
666
667 spin_lock(&cache->lock);
668 if (ret == 1) {
669 cache->caching_ctl = NULL;
670 cache->cached = BTRFS_CACHE_FINISHED;
671 cache->last_byte_to_unpin = (u64)-1;
672 caching_ctl->progress = (u64)-1;
673 } else {
674 if (load_cache_only) {
675 cache->caching_ctl = NULL;
676 cache->cached = BTRFS_CACHE_NO;
677 } else {
678 cache->cached = BTRFS_CACHE_STARTED;
679 cache->has_caching_ctl = 1;
680 }
681 }
682 spin_unlock(&cache->lock);
683 #ifdef CONFIG_BTRFS_DEBUG
684 if (ret == 1 &&
685 btrfs_should_fragment_free_space(fs_info->extent_root,
686 cache)) {
687 u64 bytes_used;
688
689 spin_lock(&cache->space_info->lock);
690 spin_lock(&cache->lock);
691 bytes_used = cache->key.offset -
692 btrfs_block_group_used(&cache->item);
693 cache->space_info->bytes_used += bytes_used >> 1;
694 spin_unlock(&cache->lock);
695 spin_unlock(&cache->space_info->lock);
696 fragment_free_space(fs_info->extent_root, cache);
697 }
698 #endif
699 mutex_unlock(&caching_ctl->mutex);
700
701 wake_up(&caching_ctl->wait);
702 if (ret == 1) {
703 put_caching_control(caching_ctl);
704 free_excluded_extents(fs_info->extent_root, cache);
705 return 0;
706 }
707 } else {
708 /*
709 * We're either using the free space tree or no caching at all.
710 * Set cached to the appropriate value and wakeup any waiters.
711 */
712 spin_lock(&cache->lock);
713 if (load_cache_only) {
714 cache->caching_ctl = NULL;
715 cache->cached = BTRFS_CACHE_NO;
716 } else {
717 cache->cached = BTRFS_CACHE_STARTED;
718 cache->has_caching_ctl = 1;
719 }
720 spin_unlock(&cache->lock);
721 wake_up(&caching_ctl->wait);
722 }
723
724 if (load_cache_only) {
725 put_caching_control(caching_ctl);
726 return 0;
727 }
728
729 down_write(&fs_info->commit_root_sem);
730 atomic_inc(&caching_ctl->count);
731 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
732 up_write(&fs_info->commit_root_sem);
733
734 btrfs_get_block_group(cache);
735
736 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
737
738 return ret;
739 }
740
741 /*
742 * return the block group that starts at or after bytenr
743 */
744 static struct btrfs_block_group_cache *
745 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
746 {
747 struct btrfs_block_group_cache *cache;
748
749 cache = block_group_cache_tree_search(info, bytenr, 0);
750
751 return cache;
752 }
753
754 /*
755 * return the block group that contains the given bytenr
756 */
757 struct btrfs_block_group_cache *btrfs_lookup_block_group(
758 struct btrfs_fs_info *info,
759 u64 bytenr)
760 {
761 struct btrfs_block_group_cache *cache;
762
763 cache = block_group_cache_tree_search(info, bytenr, 1);
764
765 return cache;
766 }
767
768 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
769 u64 flags)
770 {
771 struct list_head *head = &info->space_info;
772 struct btrfs_space_info *found;
773
774 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
775
776 rcu_read_lock();
777 list_for_each_entry_rcu(found, head, list) {
778 if (found->flags & flags) {
779 rcu_read_unlock();
780 return found;
781 }
782 }
783 rcu_read_unlock();
784 return NULL;
785 }
786
787 /*
788 * after adding space to the filesystem, we need to clear the full flags
789 * on all the space infos.
790 */
791 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
792 {
793 struct list_head *head = &info->space_info;
794 struct btrfs_space_info *found;
795
796 rcu_read_lock();
797 list_for_each_entry_rcu(found, head, list)
798 found->full = 0;
799 rcu_read_unlock();
800 }
801
802 /* simple helper to search for an existing data extent at a given offset */
803 int btrfs_lookup_data_extent(struct btrfs_root *root, u64 start, u64 len)
804 {
805 int ret;
806 struct btrfs_key key;
807 struct btrfs_path *path;
808
809 path = btrfs_alloc_path();
810 if (!path)
811 return -ENOMEM;
812
813 key.objectid = start;
814 key.offset = len;
815 key.type = BTRFS_EXTENT_ITEM_KEY;
816 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
817 0, 0);
818 btrfs_free_path(path);
819 return ret;
820 }
821
822 /*
823 * helper function to lookup reference count and flags of a tree block.
824 *
825 * the head node for delayed ref is used to store the sum of all the
826 * reference count modifications queued up in the rbtree. the head
827 * node may also store the extent flags to set. This way you can check
828 * to see what the reference count and extent flags would be if all of
829 * the delayed refs are not processed.
830 */
831 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
832 struct btrfs_root *root, u64 bytenr,
833 u64 offset, int metadata, u64 *refs, u64 *flags)
834 {
835 struct btrfs_delayed_ref_head *head;
836 struct btrfs_delayed_ref_root *delayed_refs;
837 struct btrfs_path *path;
838 struct btrfs_extent_item *ei;
839 struct extent_buffer *leaf;
840 struct btrfs_key key;
841 u32 item_size;
842 u64 num_refs;
843 u64 extent_flags;
844 int ret;
845
846 /*
847 * If we don't have skinny metadata, don't bother doing anything
848 * different
849 */
850 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
851 offset = root->nodesize;
852 metadata = 0;
853 }
854
855 path = btrfs_alloc_path();
856 if (!path)
857 return -ENOMEM;
858
859 if (!trans) {
860 path->skip_locking = 1;
861 path->search_commit_root = 1;
862 }
863
864 search_again:
865 key.objectid = bytenr;
866 key.offset = offset;
867 if (metadata)
868 key.type = BTRFS_METADATA_ITEM_KEY;
869 else
870 key.type = BTRFS_EXTENT_ITEM_KEY;
871
872 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
873 &key, path, 0, 0);
874 if (ret < 0)
875 goto out_free;
876
877 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
878 if (path->slots[0]) {
879 path->slots[0]--;
880 btrfs_item_key_to_cpu(path->nodes[0], &key,
881 path->slots[0]);
882 if (key.objectid == bytenr &&
883 key.type == BTRFS_EXTENT_ITEM_KEY &&
884 key.offset == root->nodesize)
885 ret = 0;
886 }
887 }
888
889 if (ret == 0) {
890 leaf = path->nodes[0];
891 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
892 if (item_size >= sizeof(*ei)) {
893 ei = btrfs_item_ptr(leaf, path->slots[0],
894 struct btrfs_extent_item);
895 num_refs = btrfs_extent_refs(leaf, ei);
896 extent_flags = btrfs_extent_flags(leaf, ei);
897 } else {
898 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
899 struct btrfs_extent_item_v0 *ei0;
900 BUG_ON(item_size != sizeof(*ei0));
901 ei0 = btrfs_item_ptr(leaf, path->slots[0],
902 struct btrfs_extent_item_v0);
903 num_refs = btrfs_extent_refs_v0(leaf, ei0);
904 /* FIXME: this isn't correct for data */
905 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
906 #else
907 BUG();
908 #endif
909 }
910 BUG_ON(num_refs == 0);
911 } else {
912 num_refs = 0;
913 extent_flags = 0;
914 ret = 0;
915 }
916
917 if (!trans)
918 goto out;
919
920 delayed_refs = &trans->transaction->delayed_refs;
921 spin_lock(&delayed_refs->lock);
922 head = btrfs_find_delayed_ref_head(trans, bytenr);
923 if (head) {
924 if (!mutex_trylock(&head->mutex)) {
925 atomic_inc(&head->node.refs);
926 spin_unlock(&delayed_refs->lock);
927
928 btrfs_release_path(path);
929
930 /*
931 * Mutex was contended, block until it's released and try
932 * again
933 */
934 mutex_lock(&head->mutex);
935 mutex_unlock(&head->mutex);
936 btrfs_put_delayed_ref(&head->node);
937 goto search_again;
938 }
939 spin_lock(&head->lock);
940 if (head->extent_op && head->extent_op->update_flags)
941 extent_flags |= head->extent_op->flags_to_set;
942 else
943 BUG_ON(num_refs == 0);
944
945 num_refs += head->node.ref_mod;
946 spin_unlock(&head->lock);
947 mutex_unlock(&head->mutex);
948 }
949 spin_unlock(&delayed_refs->lock);
950 out:
951 WARN_ON(num_refs == 0);
952 if (refs)
953 *refs = num_refs;
954 if (flags)
955 *flags = extent_flags;
956 out_free:
957 btrfs_free_path(path);
958 return ret;
959 }
960
961 /*
962 * Back reference rules. Back refs have three main goals:
963 *
964 * 1) differentiate between all holders of references to an extent so that
965 * when a reference is dropped we can make sure it was a valid reference
966 * before freeing the extent.
967 *
968 * 2) Provide enough information to quickly find the holders of an extent
969 * if we notice a given block is corrupted or bad.
970 *
971 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
972 * maintenance. This is actually the same as #2, but with a slightly
973 * different use case.
974 *
975 * There are two kinds of back refs. The implicit back refs is optimized
976 * for pointers in non-shared tree blocks. For a given pointer in a block,
977 * back refs of this kind provide information about the block's owner tree
978 * and the pointer's key. These information allow us to find the block by
979 * b-tree searching. The full back refs is for pointers in tree blocks not
980 * referenced by their owner trees. The location of tree block is recorded
981 * in the back refs. Actually the full back refs is generic, and can be
982 * used in all cases the implicit back refs is used. The major shortcoming
983 * of the full back refs is its overhead. Every time a tree block gets
984 * COWed, we have to update back refs entry for all pointers in it.
985 *
986 * For a newly allocated tree block, we use implicit back refs for
987 * pointers in it. This means most tree related operations only involve
988 * implicit back refs. For a tree block created in old transaction, the
989 * only way to drop a reference to it is COW it. So we can detect the
990 * event that tree block loses its owner tree's reference and do the
991 * back refs conversion.
992 *
993 * When a tree block is COWed through a tree, there are four cases:
994 *
995 * The reference count of the block is one and the tree is the block's
996 * owner tree. Nothing to do in this case.
997 *
998 * The reference count of the block is one and the tree is not the
999 * block's owner tree. In this case, full back refs is used for pointers
1000 * in the block. Remove these full back refs, add implicit back refs for
1001 * every pointers in the new block.
1002 *
1003 * The reference count of the block is greater than one and the tree is
1004 * the block's owner tree. In this case, implicit back refs is used for
1005 * pointers in the block. Add full back refs for every pointers in the
1006 * block, increase lower level extents' reference counts. The original
1007 * implicit back refs are entailed to the new block.
1008 *
1009 * The reference count of the block is greater than one and the tree is
1010 * not the block's owner tree. Add implicit back refs for every pointer in
1011 * the new block, increase lower level extents' reference count.
1012 *
1013 * Back Reference Key composing:
1014 *
1015 * The key objectid corresponds to the first byte in the extent,
1016 * The key type is used to differentiate between types of back refs.
1017 * There are different meanings of the key offset for different types
1018 * of back refs.
1019 *
1020 * File extents can be referenced by:
1021 *
1022 * - multiple snapshots, subvolumes, or different generations in one subvol
1023 * - different files inside a single subvolume
1024 * - different offsets inside a file (bookend extents in file.c)
1025 *
1026 * The extent ref structure for the implicit back refs has fields for:
1027 *
1028 * - Objectid of the subvolume root
1029 * - objectid of the file holding the reference
1030 * - original offset in the file
1031 * - how many bookend extents
1032 *
1033 * The key offset for the implicit back refs is hash of the first
1034 * three fields.
1035 *
1036 * The extent ref structure for the full back refs has field for:
1037 *
1038 * - number of pointers in the tree leaf
1039 *
1040 * The key offset for the implicit back refs is the first byte of
1041 * the tree leaf
1042 *
1043 * When a file extent is allocated, The implicit back refs is used.
1044 * the fields are filled in:
1045 *
1046 * (root_key.objectid, inode objectid, offset in file, 1)
1047 *
1048 * When a file extent is removed file truncation, we find the
1049 * corresponding implicit back refs and check the following fields:
1050 *
1051 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1052 *
1053 * Btree extents can be referenced by:
1054 *
1055 * - Different subvolumes
1056 *
1057 * Both the implicit back refs and the full back refs for tree blocks
1058 * only consist of key. The key offset for the implicit back refs is
1059 * objectid of block's owner tree. The key offset for the full back refs
1060 * is the first byte of parent block.
1061 *
1062 * When implicit back refs is used, information about the lowest key and
1063 * level of the tree block are required. These information are stored in
1064 * tree block info structure.
1065 */
1066
1067 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1068 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
1069 struct btrfs_root *root,
1070 struct btrfs_path *path,
1071 u64 owner, u32 extra_size)
1072 {
1073 struct btrfs_extent_item *item;
1074 struct btrfs_extent_item_v0 *ei0;
1075 struct btrfs_extent_ref_v0 *ref0;
1076 struct btrfs_tree_block_info *bi;
1077 struct extent_buffer *leaf;
1078 struct btrfs_key key;
1079 struct btrfs_key found_key;
1080 u32 new_size = sizeof(*item);
1081 u64 refs;
1082 int ret;
1083
1084 leaf = path->nodes[0];
1085 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1086
1087 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1088 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1089 struct btrfs_extent_item_v0);
1090 refs = btrfs_extent_refs_v0(leaf, ei0);
1091
1092 if (owner == (u64)-1) {
1093 while (1) {
1094 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1095 ret = btrfs_next_leaf(root, path);
1096 if (ret < 0)
1097 return ret;
1098 BUG_ON(ret > 0); /* Corruption */
1099 leaf = path->nodes[0];
1100 }
1101 btrfs_item_key_to_cpu(leaf, &found_key,
1102 path->slots[0]);
1103 BUG_ON(key.objectid != found_key.objectid);
1104 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1105 path->slots[0]++;
1106 continue;
1107 }
1108 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1109 struct btrfs_extent_ref_v0);
1110 owner = btrfs_ref_objectid_v0(leaf, ref0);
1111 break;
1112 }
1113 }
1114 btrfs_release_path(path);
1115
1116 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1117 new_size += sizeof(*bi);
1118
1119 new_size -= sizeof(*ei0);
1120 ret = btrfs_search_slot(trans, root, &key, path,
1121 new_size + extra_size, 1);
1122 if (ret < 0)
1123 return ret;
1124 BUG_ON(ret); /* Corruption */
1125
1126 btrfs_extend_item(root, path, new_size);
1127
1128 leaf = path->nodes[0];
1129 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1130 btrfs_set_extent_refs(leaf, item, refs);
1131 /* FIXME: get real generation */
1132 btrfs_set_extent_generation(leaf, item, 0);
1133 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1134 btrfs_set_extent_flags(leaf, item,
1135 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1136 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1137 bi = (struct btrfs_tree_block_info *)(item + 1);
1138 /* FIXME: get first key of the block */
1139 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1140 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1141 } else {
1142 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1143 }
1144 btrfs_mark_buffer_dirty(leaf);
1145 return 0;
1146 }
1147 #endif
1148
1149 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1150 {
1151 u32 high_crc = ~(u32)0;
1152 u32 low_crc = ~(u32)0;
1153 __le64 lenum;
1154
1155 lenum = cpu_to_le64(root_objectid);
1156 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1157 lenum = cpu_to_le64(owner);
1158 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1159 lenum = cpu_to_le64(offset);
1160 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1161
1162 return ((u64)high_crc << 31) ^ (u64)low_crc;
1163 }
1164
1165 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1166 struct btrfs_extent_data_ref *ref)
1167 {
1168 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1169 btrfs_extent_data_ref_objectid(leaf, ref),
1170 btrfs_extent_data_ref_offset(leaf, ref));
1171 }
1172
1173 static int match_extent_data_ref(struct extent_buffer *leaf,
1174 struct btrfs_extent_data_ref *ref,
1175 u64 root_objectid, u64 owner, u64 offset)
1176 {
1177 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1178 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1179 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1180 return 0;
1181 return 1;
1182 }
1183
1184 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1185 struct btrfs_root *root,
1186 struct btrfs_path *path,
1187 u64 bytenr, u64 parent,
1188 u64 root_objectid,
1189 u64 owner, u64 offset)
1190 {
1191 struct btrfs_key key;
1192 struct btrfs_extent_data_ref *ref;
1193 struct extent_buffer *leaf;
1194 u32 nritems;
1195 int ret;
1196 int recow;
1197 int err = -ENOENT;
1198
1199 key.objectid = bytenr;
1200 if (parent) {
1201 key.type = BTRFS_SHARED_DATA_REF_KEY;
1202 key.offset = parent;
1203 } else {
1204 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1205 key.offset = hash_extent_data_ref(root_objectid,
1206 owner, offset);
1207 }
1208 again:
1209 recow = 0;
1210 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1211 if (ret < 0) {
1212 err = ret;
1213 goto fail;
1214 }
1215
1216 if (parent) {
1217 if (!ret)
1218 return 0;
1219 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1220 key.type = BTRFS_EXTENT_REF_V0_KEY;
1221 btrfs_release_path(path);
1222 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1223 if (ret < 0) {
1224 err = ret;
1225 goto fail;
1226 }
1227 if (!ret)
1228 return 0;
1229 #endif
1230 goto fail;
1231 }
1232
1233 leaf = path->nodes[0];
1234 nritems = btrfs_header_nritems(leaf);
1235 while (1) {
1236 if (path->slots[0] >= nritems) {
1237 ret = btrfs_next_leaf(root, path);
1238 if (ret < 0)
1239 err = ret;
1240 if (ret)
1241 goto fail;
1242
1243 leaf = path->nodes[0];
1244 nritems = btrfs_header_nritems(leaf);
1245 recow = 1;
1246 }
1247
1248 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1249 if (key.objectid != bytenr ||
1250 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1251 goto fail;
1252
1253 ref = btrfs_item_ptr(leaf, path->slots[0],
1254 struct btrfs_extent_data_ref);
1255
1256 if (match_extent_data_ref(leaf, ref, root_objectid,
1257 owner, offset)) {
1258 if (recow) {
1259 btrfs_release_path(path);
1260 goto again;
1261 }
1262 err = 0;
1263 break;
1264 }
1265 path->slots[0]++;
1266 }
1267 fail:
1268 return err;
1269 }
1270
1271 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1272 struct btrfs_root *root,
1273 struct btrfs_path *path,
1274 u64 bytenr, u64 parent,
1275 u64 root_objectid, u64 owner,
1276 u64 offset, int refs_to_add)
1277 {
1278 struct btrfs_key key;
1279 struct extent_buffer *leaf;
1280 u32 size;
1281 u32 num_refs;
1282 int ret;
1283
1284 key.objectid = bytenr;
1285 if (parent) {
1286 key.type = BTRFS_SHARED_DATA_REF_KEY;
1287 key.offset = parent;
1288 size = sizeof(struct btrfs_shared_data_ref);
1289 } else {
1290 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1291 key.offset = hash_extent_data_ref(root_objectid,
1292 owner, offset);
1293 size = sizeof(struct btrfs_extent_data_ref);
1294 }
1295
1296 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1297 if (ret && ret != -EEXIST)
1298 goto fail;
1299
1300 leaf = path->nodes[0];
1301 if (parent) {
1302 struct btrfs_shared_data_ref *ref;
1303 ref = btrfs_item_ptr(leaf, path->slots[0],
1304 struct btrfs_shared_data_ref);
1305 if (ret == 0) {
1306 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1307 } else {
1308 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1309 num_refs += refs_to_add;
1310 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1311 }
1312 } else {
1313 struct btrfs_extent_data_ref *ref;
1314 while (ret == -EEXIST) {
1315 ref = btrfs_item_ptr(leaf, path->slots[0],
1316 struct btrfs_extent_data_ref);
1317 if (match_extent_data_ref(leaf, ref, root_objectid,
1318 owner, offset))
1319 break;
1320 btrfs_release_path(path);
1321 key.offset++;
1322 ret = btrfs_insert_empty_item(trans, root, path, &key,
1323 size);
1324 if (ret && ret != -EEXIST)
1325 goto fail;
1326
1327 leaf = path->nodes[0];
1328 }
1329 ref = btrfs_item_ptr(leaf, path->slots[0],
1330 struct btrfs_extent_data_ref);
1331 if (ret == 0) {
1332 btrfs_set_extent_data_ref_root(leaf, ref,
1333 root_objectid);
1334 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1335 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1336 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1337 } else {
1338 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1339 num_refs += refs_to_add;
1340 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1341 }
1342 }
1343 btrfs_mark_buffer_dirty(leaf);
1344 ret = 0;
1345 fail:
1346 btrfs_release_path(path);
1347 return ret;
1348 }
1349
1350 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1351 struct btrfs_root *root,
1352 struct btrfs_path *path,
1353 int refs_to_drop, int *last_ref)
1354 {
1355 struct btrfs_key key;
1356 struct btrfs_extent_data_ref *ref1 = NULL;
1357 struct btrfs_shared_data_ref *ref2 = NULL;
1358 struct extent_buffer *leaf;
1359 u32 num_refs = 0;
1360 int ret = 0;
1361
1362 leaf = path->nodes[0];
1363 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1364
1365 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1366 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1367 struct btrfs_extent_data_ref);
1368 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1369 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1370 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1371 struct btrfs_shared_data_ref);
1372 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1373 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1374 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1375 struct btrfs_extent_ref_v0 *ref0;
1376 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1377 struct btrfs_extent_ref_v0);
1378 num_refs = btrfs_ref_count_v0(leaf, ref0);
1379 #endif
1380 } else {
1381 BUG();
1382 }
1383
1384 BUG_ON(num_refs < refs_to_drop);
1385 num_refs -= refs_to_drop;
1386
1387 if (num_refs == 0) {
1388 ret = btrfs_del_item(trans, root, path);
1389 *last_ref = 1;
1390 } else {
1391 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1392 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1393 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1394 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1395 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1396 else {
1397 struct btrfs_extent_ref_v0 *ref0;
1398 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1399 struct btrfs_extent_ref_v0);
1400 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1401 }
1402 #endif
1403 btrfs_mark_buffer_dirty(leaf);
1404 }
1405 return ret;
1406 }
1407
1408 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
1409 struct btrfs_extent_inline_ref *iref)
1410 {
1411 struct btrfs_key key;
1412 struct extent_buffer *leaf;
1413 struct btrfs_extent_data_ref *ref1;
1414 struct btrfs_shared_data_ref *ref2;
1415 u32 num_refs = 0;
1416
1417 leaf = path->nodes[0];
1418 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1419 if (iref) {
1420 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1421 BTRFS_EXTENT_DATA_REF_KEY) {
1422 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1423 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1424 } else {
1425 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1426 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1427 }
1428 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1429 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1430 struct btrfs_extent_data_ref);
1431 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1432 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1433 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1434 struct btrfs_shared_data_ref);
1435 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1436 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1437 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1438 struct btrfs_extent_ref_v0 *ref0;
1439 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1440 struct btrfs_extent_ref_v0);
1441 num_refs = btrfs_ref_count_v0(leaf, ref0);
1442 #endif
1443 } else {
1444 WARN_ON(1);
1445 }
1446 return num_refs;
1447 }
1448
1449 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1450 struct btrfs_root *root,
1451 struct btrfs_path *path,
1452 u64 bytenr, u64 parent,
1453 u64 root_objectid)
1454 {
1455 struct btrfs_key key;
1456 int ret;
1457
1458 key.objectid = bytenr;
1459 if (parent) {
1460 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1461 key.offset = parent;
1462 } else {
1463 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1464 key.offset = root_objectid;
1465 }
1466
1467 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1468 if (ret > 0)
1469 ret = -ENOENT;
1470 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1471 if (ret == -ENOENT && parent) {
1472 btrfs_release_path(path);
1473 key.type = BTRFS_EXTENT_REF_V0_KEY;
1474 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1475 if (ret > 0)
1476 ret = -ENOENT;
1477 }
1478 #endif
1479 return ret;
1480 }
1481
1482 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1483 struct btrfs_root *root,
1484 struct btrfs_path *path,
1485 u64 bytenr, u64 parent,
1486 u64 root_objectid)
1487 {
1488 struct btrfs_key key;
1489 int ret;
1490
1491 key.objectid = bytenr;
1492 if (parent) {
1493 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1494 key.offset = parent;
1495 } else {
1496 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1497 key.offset = root_objectid;
1498 }
1499
1500 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1501 btrfs_release_path(path);
1502 return ret;
1503 }
1504
1505 static inline int extent_ref_type(u64 parent, u64 owner)
1506 {
1507 int type;
1508 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1509 if (parent > 0)
1510 type = BTRFS_SHARED_BLOCK_REF_KEY;
1511 else
1512 type = BTRFS_TREE_BLOCK_REF_KEY;
1513 } else {
1514 if (parent > 0)
1515 type = BTRFS_SHARED_DATA_REF_KEY;
1516 else
1517 type = BTRFS_EXTENT_DATA_REF_KEY;
1518 }
1519 return type;
1520 }
1521
1522 static int find_next_key(struct btrfs_path *path, int level,
1523 struct btrfs_key *key)
1524
1525 {
1526 for (; level < BTRFS_MAX_LEVEL; level++) {
1527 if (!path->nodes[level])
1528 break;
1529 if (path->slots[level] + 1 >=
1530 btrfs_header_nritems(path->nodes[level]))
1531 continue;
1532 if (level == 0)
1533 btrfs_item_key_to_cpu(path->nodes[level], key,
1534 path->slots[level] + 1);
1535 else
1536 btrfs_node_key_to_cpu(path->nodes[level], key,
1537 path->slots[level] + 1);
1538 return 0;
1539 }
1540 return 1;
1541 }
1542
1543 /*
1544 * look for inline back ref. if back ref is found, *ref_ret is set
1545 * to the address of inline back ref, and 0 is returned.
1546 *
1547 * if back ref isn't found, *ref_ret is set to the address where it
1548 * should be inserted, and -ENOENT is returned.
1549 *
1550 * if insert is true and there are too many inline back refs, the path
1551 * points to the extent item, and -EAGAIN is returned.
1552 *
1553 * NOTE: inline back refs are ordered in the same way that back ref
1554 * items in the tree are ordered.
1555 */
1556 static noinline_for_stack
1557 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1558 struct btrfs_root *root,
1559 struct btrfs_path *path,
1560 struct btrfs_extent_inline_ref **ref_ret,
1561 u64 bytenr, u64 num_bytes,
1562 u64 parent, u64 root_objectid,
1563 u64 owner, u64 offset, int insert)
1564 {
1565 struct btrfs_key key;
1566 struct extent_buffer *leaf;
1567 struct btrfs_extent_item *ei;
1568 struct btrfs_extent_inline_ref *iref;
1569 u64 flags;
1570 u64 item_size;
1571 unsigned long ptr;
1572 unsigned long end;
1573 int extra_size;
1574 int type;
1575 int want;
1576 int ret;
1577 int err = 0;
1578 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1579 SKINNY_METADATA);
1580
1581 key.objectid = bytenr;
1582 key.type = BTRFS_EXTENT_ITEM_KEY;
1583 key.offset = num_bytes;
1584
1585 want = extent_ref_type(parent, owner);
1586 if (insert) {
1587 extra_size = btrfs_extent_inline_ref_size(want);
1588 path->keep_locks = 1;
1589 } else
1590 extra_size = -1;
1591
1592 /*
1593 * Owner is our parent level, so we can just add one to get the level
1594 * for the block we are interested in.
1595 */
1596 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1597 key.type = BTRFS_METADATA_ITEM_KEY;
1598 key.offset = owner;
1599 }
1600
1601 again:
1602 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1603 if (ret < 0) {
1604 err = ret;
1605 goto out;
1606 }
1607
1608 /*
1609 * We may be a newly converted file system which still has the old fat
1610 * extent entries for metadata, so try and see if we have one of those.
1611 */
1612 if (ret > 0 && skinny_metadata) {
1613 skinny_metadata = false;
1614 if (path->slots[0]) {
1615 path->slots[0]--;
1616 btrfs_item_key_to_cpu(path->nodes[0], &key,
1617 path->slots[0]);
1618 if (key.objectid == bytenr &&
1619 key.type == BTRFS_EXTENT_ITEM_KEY &&
1620 key.offset == num_bytes)
1621 ret = 0;
1622 }
1623 if (ret) {
1624 key.objectid = bytenr;
1625 key.type = BTRFS_EXTENT_ITEM_KEY;
1626 key.offset = num_bytes;
1627 btrfs_release_path(path);
1628 goto again;
1629 }
1630 }
1631
1632 if (ret && !insert) {
1633 err = -ENOENT;
1634 goto out;
1635 } else if (WARN_ON(ret)) {
1636 err = -EIO;
1637 goto out;
1638 }
1639
1640 leaf = path->nodes[0];
1641 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1642 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1643 if (item_size < sizeof(*ei)) {
1644 if (!insert) {
1645 err = -ENOENT;
1646 goto out;
1647 }
1648 ret = convert_extent_item_v0(trans, root, path, owner,
1649 extra_size);
1650 if (ret < 0) {
1651 err = ret;
1652 goto out;
1653 }
1654 leaf = path->nodes[0];
1655 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1656 }
1657 #endif
1658 BUG_ON(item_size < sizeof(*ei));
1659
1660 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1661 flags = btrfs_extent_flags(leaf, ei);
1662
1663 ptr = (unsigned long)(ei + 1);
1664 end = (unsigned long)ei + item_size;
1665
1666 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1667 ptr += sizeof(struct btrfs_tree_block_info);
1668 BUG_ON(ptr > end);
1669 }
1670
1671 err = -ENOENT;
1672 while (1) {
1673 if (ptr >= end) {
1674 WARN_ON(ptr > end);
1675 break;
1676 }
1677 iref = (struct btrfs_extent_inline_ref *)ptr;
1678 type = btrfs_extent_inline_ref_type(leaf, iref);
1679 if (want < type)
1680 break;
1681 if (want > type) {
1682 ptr += btrfs_extent_inline_ref_size(type);
1683 continue;
1684 }
1685
1686 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1687 struct btrfs_extent_data_ref *dref;
1688 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1689 if (match_extent_data_ref(leaf, dref, root_objectid,
1690 owner, offset)) {
1691 err = 0;
1692 break;
1693 }
1694 if (hash_extent_data_ref_item(leaf, dref) <
1695 hash_extent_data_ref(root_objectid, owner, offset))
1696 break;
1697 } else {
1698 u64 ref_offset;
1699 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1700 if (parent > 0) {
1701 if (parent == ref_offset) {
1702 err = 0;
1703 break;
1704 }
1705 if (ref_offset < parent)
1706 break;
1707 } else {
1708 if (root_objectid == ref_offset) {
1709 err = 0;
1710 break;
1711 }
1712 if (ref_offset < root_objectid)
1713 break;
1714 }
1715 }
1716 ptr += btrfs_extent_inline_ref_size(type);
1717 }
1718 if (err == -ENOENT && insert) {
1719 if (item_size + extra_size >=
1720 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1721 err = -EAGAIN;
1722 goto out;
1723 }
1724 /*
1725 * To add new inline back ref, we have to make sure
1726 * there is no corresponding back ref item.
1727 * For simplicity, we just do not add new inline back
1728 * ref if there is any kind of item for this block
1729 */
1730 if (find_next_key(path, 0, &key) == 0 &&
1731 key.objectid == bytenr &&
1732 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1733 err = -EAGAIN;
1734 goto out;
1735 }
1736 }
1737 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1738 out:
1739 if (insert) {
1740 path->keep_locks = 0;
1741 btrfs_unlock_up_safe(path, 1);
1742 }
1743 return err;
1744 }
1745
1746 /*
1747 * helper to add new inline back ref
1748 */
1749 static noinline_for_stack
1750 void setup_inline_extent_backref(struct btrfs_root *root,
1751 struct btrfs_path *path,
1752 struct btrfs_extent_inline_ref *iref,
1753 u64 parent, u64 root_objectid,
1754 u64 owner, u64 offset, int refs_to_add,
1755 struct btrfs_delayed_extent_op *extent_op)
1756 {
1757 struct extent_buffer *leaf;
1758 struct btrfs_extent_item *ei;
1759 unsigned long ptr;
1760 unsigned long end;
1761 unsigned long item_offset;
1762 u64 refs;
1763 int size;
1764 int type;
1765
1766 leaf = path->nodes[0];
1767 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1768 item_offset = (unsigned long)iref - (unsigned long)ei;
1769
1770 type = extent_ref_type(parent, owner);
1771 size = btrfs_extent_inline_ref_size(type);
1772
1773 btrfs_extend_item(root, path, size);
1774
1775 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1776 refs = btrfs_extent_refs(leaf, ei);
1777 refs += refs_to_add;
1778 btrfs_set_extent_refs(leaf, ei, refs);
1779 if (extent_op)
1780 __run_delayed_extent_op(extent_op, leaf, ei);
1781
1782 ptr = (unsigned long)ei + item_offset;
1783 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1784 if (ptr < end - size)
1785 memmove_extent_buffer(leaf, ptr + size, ptr,
1786 end - size - ptr);
1787
1788 iref = (struct btrfs_extent_inline_ref *)ptr;
1789 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1790 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1791 struct btrfs_extent_data_ref *dref;
1792 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1793 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1794 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1795 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1796 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1797 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1798 struct btrfs_shared_data_ref *sref;
1799 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1800 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1801 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1802 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1803 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1804 } else {
1805 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1806 }
1807 btrfs_mark_buffer_dirty(leaf);
1808 }
1809
1810 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1811 struct btrfs_root *root,
1812 struct btrfs_path *path,
1813 struct btrfs_extent_inline_ref **ref_ret,
1814 u64 bytenr, u64 num_bytes, u64 parent,
1815 u64 root_objectid, u64 owner, u64 offset)
1816 {
1817 int ret;
1818
1819 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1820 bytenr, num_bytes, parent,
1821 root_objectid, owner, offset, 0);
1822 if (ret != -ENOENT)
1823 return ret;
1824
1825 btrfs_release_path(path);
1826 *ref_ret = NULL;
1827
1828 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1829 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1830 root_objectid);
1831 } else {
1832 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1833 root_objectid, owner, offset);
1834 }
1835 return ret;
1836 }
1837
1838 /*
1839 * helper to update/remove inline back ref
1840 */
1841 static noinline_for_stack
1842 void update_inline_extent_backref(struct btrfs_root *root,
1843 struct btrfs_path *path,
1844 struct btrfs_extent_inline_ref *iref,
1845 int refs_to_mod,
1846 struct btrfs_delayed_extent_op *extent_op,
1847 int *last_ref)
1848 {
1849 struct extent_buffer *leaf;
1850 struct btrfs_extent_item *ei;
1851 struct btrfs_extent_data_ref *dref = NULL;
1852 struct btrfs_shared_data_ref *sref = NULL;
1853 unsigned long ptr;
1854 unsigned long end;
1855 u32 item_size;
1856 int size;
1857 int type;
1858 u64 refs;
1859
1860 leaf = path->nodes[0];
1861 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1862 refs = btrfs_extent_refs(leaf, ei);
1863 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1864 refs += refs_to_mod;
1865 btrfs_set_extent_refs(leaf, ei, refs);
1866 if (extent_op)
1867 __run_delayed_extent_op(extent_op, leaf, ei);
1868
1869 type = btrfs_extent_inline_ref_type(leaf, iref);
1870
1871 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1872 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1873 refs = btrfs_extent_data_ref_count(leaf, dref);
1874 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1875 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1876 refs = btrfs_shared_data_ref_count(leaf, sref);
1877 } else {
1878 refs = 1;
1879 BUG_ON(refs_to_mod != -1);
1880 }
1881
1882 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1883 refs += refs_to_mod;
1884
1885 if (refs > 0) {
1886 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1887 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1888 else
1889 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1890 } else {
1891 *last_ref = 1;
1892 size = btrfs_extent_inline_ref_size(type);
1893 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1894 ptr = (unsigned long)iref;
1895 end = (unsigned long)ei + item_size;
1896 if (ptr + size < end)
1897 memmove_extent_buffer(leaf, ptr, ptr + size,
1898 end - ptr - size);
1899 item_size -= size;
1900 btrfs_truncate_item(root, path, item_size, 1);
1901 }
1902 btrfs_mark_buffer_dirty(leaf);
1903 }
1904
1905 static noinline_for_stack
1906 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1907 struct btrfs_root *root,
1908 struct btrfs_path *path,
1909 u64 bytenr, u64 num_bytes, u64 parent,
1910 u64 root_objectid, u64 owner,
1911 u64 offset, int refs_to_add,
1912 struct btrfs_delayed_extent_op *extent_op)
1913 {
1914 struct btrfs_extent_inline_ref *iref;
1915 int ret;
1916
1917 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1918 bytenr, num_bytes, parent,
1919 root_objectid, owner, offset, 1);
1920 if (ret == 0) {
1921 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1922 update_inline_extent_backref(root, path, iref,
1923 refs_to_add, extent_op, NULL);
1924 } else if (ret == -ENOENT) {
1925 setup_inline_extent_backref(root, path, iref, parent,
1926 root_objectid, owner, offset,
1927 refs_to_add, extent_op);
1928 ret = 0;
1929 }
1930 return ret;
1931 }
1932
1933 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1934 struct btrfs_root *root,
1935 struct btrfs_path *path,
1936 u64 bytenr, u64 parent, u64 root_objectid,
1937 u64 owner, u64 offset, int refs_to_add)
1938 {
1939 int ret;
1940 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1941 BUG_ON(refs_to_add != 1);
1942 ret = insert_tree_block_ref(trans, root, path, bytenr,
1943 parent, root_objectid);
1944 } else {
1945 ret = insert_extent_data_ref(trans, root, path, bytenr,
1946 parent, root_objectid,
1947 owner, offset, refs_to_add);
1948 }
1949 return ret;
1950 }
1951
1952 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1953 struct btrfs_root *root,
1954 struct btrfs_path *path,
1955 struct btrfs_extent_inline_ref *iref,
1956 int refs_to_drop, int is_data, int *last_ref)
1957 {
1958 int ret = 0;
1959
1960 BUG_ON(!is_data && refs_to_drop != 1);
1961 if (iref) {
1962 update_inline_extent_backref(root, path, iref,
1963 -refs_to_drop, NULL, last_ref);
1964 } else if (is_data) {
1965 ret = remove_extent_data_ref(trans, root, path, refs_to_drop,
1966 last_ref);
1967 } else {
1968 *last_ref = 1;
1969 ret = btrfs_del_item(trans, root, path);
1970 }
1971 return ret;
1972 }
1973
1974 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
1975 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1976 u64 *discarded_bytes)
1977 {
1978 int j, ret = 0;
1979 u64 bytes_left, end;
1980 u64 aligned_start = ALIGN(start, 1 << 9);
1981
1982 if (WARN_ON(start != aligned_start)) {
1983 len -= aligned_start - start;
1984 len = round_down(len, 1 << 9);
1985 start = aligned_start;
1986 }
1987
1988 *discarded_bytes = 0;
1989
1990 if (!len)
1991 return 0;
1992
1993 end = start + len;
1994 bytes_left = len;
1995
1996 /* Skip any superblocks on this device. */
1997 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1998 u64 sb_start = btrfs_sb_offset(j);
1999 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
2000 u64 size = sb_start - start;
2001
2002 if (!in_range(sb_start, start, bytes_left) &&
2003 !in_range(sb_end, start, bytes_left) &&
2004 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
2005 continue;
2006
2007 /*
2008 * Superblock spans beginning of range. Adjust start and
2009 * try again.
2010 */
2011 if (sb_start <= start) {
2012 start += sb_end - start;
2013 if (start > end) {
2014 bytes_left = 0;
2015 break;
2016 }
2017 bytes_left = end - start;
2018 continue;
2019 }
2020
2021 if (size) {
2022 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
2023 GFP_NOFS, 0);
2024 if (!ret)
2025 *discarded_bytes += size;
2026 else if (ret != -EOPNOTSUPP)
2027 return ret;
2028 }
2029
2030 start = sb_end;
2031 if (start > end) {
2032 bytes_left = 0;
2033 break;
2034 }
2035 bytes_left = end - start;
2036 }
2037
2038 if (bytes_left) {
2039 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
2040 GFP_NOFS, 0);
2041 if (!ret)
2042 *discarded_bytes += bytes_left;
2043 }
2044 return ret;
2045 }
2046
2047 int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
2048 u64 num_bytes, u64 *actual_bytes)
2049 {
2050 int ret;
2051 u64 discarded_bytes = 0;
2052 struct btrfs_bio *bbio = NULL;
2053
2054
2055 /*
2056 * Avoid races with device replace and make sure our bbio has devices
2057 * associated to its stripes that don't go away while we are discarding.
2058 */
2059 btrfs_bio_counter_inc_blocked(root->fs_info);
2060 /* Tell the block device(s) that the sectors can be discarded */
2061 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
2062 bytenr, &num_bytes, &bbio, 0);
2063 /* Error condition is -ENOMEM */
2064 if (!ret) {
2065 struct btrfs_bio_stripe *stripe = bbio->stripes;
2066 int i;
2067
2068
2069 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
2070 u64 bytes;
2071 if (!stripe->dev->can_discard)
2072 continue;
2073
2074 ret = btrfs_issue_discard(stripe->dev->bdev,
2075 stripe->physical,
2076 stripe->length,
2077 &bytes);
2078 if (!ret)
2079 discarded_bytes += bytes;
2080 else if (ret != -EOPNOTSUPP)
2081 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2082
2083 /*
2084 * Just in case we get back EOPNOTSUPP for some reason,
2085 * just ignore the return value so we don't screw up
2086 * people calling discard_extent.
2087 */
2088 ret = 0;
2089 }
2090 btrfs_put_bbio(bbio);
2091 }
2092 btrfs_bio_counter_dec(root->fs_info);
2093
2094 if (actual_bytes)
2095 *actual_bytes = discarded_bytes;
2096
2097
2098 if (ret == -EOPNOTSUPP)
2099 ret = 0;
2100 return ret;
2101 }
2102
2103 /* Can return -ENOMEM */
2104 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2105 struct btrfs_root *root,
2106 u64 bytenr, u64 num_bytes, u64 parent,
2107 u64 root_objectid, u64 owner, u64 offset)
2108 {
2109 int ret;
2110 struct btrfs_fs_info *fs_info = root->fs_info;
2111
2112 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2113 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2114
2115 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2116 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
2117 num_bytes,
2118 parent, root_objectid, (int)owner,
2119 BTRFS_ADD_DELAYED_REF, NULL);
2120 } else {
2121 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
2122 num_bytes, parent, root_objectid,
2123 owner, offset, 0,
2124 BTRFS_ADD_DELAYED_REF, NULL);
2125 }
2126 return ret;
2127 }
2128
2129 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2130 struct btrfs_root *root,
2131 struct btrfs_delayed_ref_node *node,
2132 u64 parent, u64 root_objectid,
2133 u64 owner, u64 offset, int refs_to_add,
2134 struct btrfs_delayed_extent_op *extent_op)
2135 {
2136 struct btrfs_fs_info *fs_info = root->fs_info;
2137 struct btrfs_path *path;
2138 struct extent_buffer *leaf;
2139 struct btrfs_extent_item *item;
2140 struct btrfs_key key;
2141 u64 bytenr = node->bytenr;
2142 u64 num_bytes = node->num_bytes;
2143 u64 refs;
2144 int ret;
2145
2146 path = btrfs_alloc_path();
2147 if (!path)
2148 return -ENOMEM;
2149
2150 path->reada = READA_FORWARD;
2151 path->leave_spinning = 1;
2152 /* this will setup the path even if it fails to insert the back ref */
2153 ret = insert_inline_extent_backref(trans, fs_info->extent_root, path,
2154 bytenr, num_bytes, parent,
2155 root_objectid, owner, offset,
2156 refs_to_add, extent_op);
2157 if ((ret < 0 && ret != -EAGAIN) || !ret)
2158 goto out;
2159
2160 /*
2161 * Ok we had -EAGAIN which means we didn't have space to insert and
2162 * inline extent ref, so just update the reference count and add a
2163 * normal backref.
2164 */
2165 leaf = path->nodes[0];
2166 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2167 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2168 refs = btrfs_extent_refs(leaf, item);
2169 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2170 if (extent_op)
2171 __run_delayed_extent_op(extent_op, leaf, item);
2172
2173 btrfs_mark_buffer_dirty(leaf);
2174 btrfs_release_path(path);
2175
2176 path->reada = READA_FORWARD;
2177 path->leave_spinning = 1;
2178 /* now insert the actual backref */
2179 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2180 path, bytenr, parent, root_objectid,
2181 owner, offset, refs_to_add);
2182 if (ret)
2183 btrfs_abort_transaction(trans, root, ret);
2184 out:
2185 btrfs_free_path(path);
2186 return ret;
2187 }
2188
2189 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2190 struct btrfs_root *root,
2191 struct btrfs_delayed_ref_node *node,
2192 struct btrfs_delayed_extent_op *extent_op,
2193 int insert_reserved)
2194 {
2195 int ret = 0;
2196 struct btrfs_delayed_data_ref *ref;
2197 struct btrfs_key ins;
2198 u64 parent = 0;
2199 u64 ref_root = 0;
2200 u64 flags = 0;
2201
2202 ins.objectid = node->bytenr;
2203 ins.offset = node->num_bytes;
2204 ins.type = BTRFS_EXTENT_ITEM_KEY;
2205
2206 ref = btrfs_delayed_node_to_data_ref(node);
2207 trace_run_delayed_data_ref(node, ref, node->action);
2208
2209 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2210 parent = ref->parent;
2211 ref_root = ref->root;
2212
2213 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2214 if (extent_op)
2215 flags |= extent_op->flags_to_set;
2216 ret = alloc_reserved_file_extent(trans, root,
2217 parent, ref_root, flags,
2218 ref->objectid, ref->offset,
2219 &ins, node->ref_mod);
2220 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2221 ret = __btrfs_inc_extent_ref(trans, root, node, parent,
2222 ref_root, ref->objectid,
2223 ref->offset, node->ref_mod,
2224 extent_op);
2225 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2226 ret = __btrfs_free_extent(trans, root, node, parent,
2227 ref_root, ref->objectid,
2228 ref->offset, node->ref_mod,
2229 extent_op);
2230 } else {
2231 BUG();
2232 }
2233 return ret;
2234 }
2235
2236 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2237 struct extent_buffer *leaf,
2238 struct btrfs_extent_item *ei)
2239 {
2240 u64 flags = btrfs_extent_flags(leaf, ei);
2241 if (extent_op->update_flags) {
2242 flags |= extent_op->flags_to_set;
2243 btrfs_set_extent_flags(leaf, ei, flags);
2244 }
2245
2246 if (extent_op->update_key) {
2247 struct btrfs_tree_block_info *bi;
2248 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2249 bi = (struct btrfs_tree_block_info *)(ei + 1);
2250 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2251 }
2252 }
2253
2254 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2255 struct btrfs_root *root,
2256 struct btrfs_delayed_ref_node *node,
2257 struct btrfs_delayed_extent_op *extent_op)
2258 {
2259 struct btrfs_key key;
2260 struct btrfs_path *path;
2261 struct btrfs_extent_item *ei;
2262 struct extent_buffer *leaf;
2263 u32 item_size;
2264 int ret;
2265 int err = 0;
2266 int metadata = !extent_op->is_data;
2267
2268 if (trans->aborted)
2269 return 0;
2270
2271 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2272 metadata = 0;
2273
2274 path = btrfs_alloc_path();
2275 if (!path)
2276 return -ENOMEM;
2277
2278 key.objectid = node->bytenr;
2279
2280 if (metadata) {
2281 key.type = BTRFS_METADATA_ITEM_KEY;
2282 key.offset = extent_op->level;
2283 } else {
2284 key.type = BTRFS_EXTENT_ITEM_KEY;
2285 key.offset = node->num_bytes;
2286 }
2287
2288 again:
2289 path->reada = READA_FORWARD;
2290 path->leave_spinning = 1;
2291 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2292 path, 0, 1);
2293 if (ret < 0) {
2294 err = ret;
2295 goto out;
2296 }
2297 if (ret > 0) {
2298 if (metadata) {
2299 if (path->slots[0] > 0) {
2300 path->slots[0]--;
2301 btrfs_item_key_to_cpu(path->nodes[0], &key,
2302 path->slots[0]);
2303 if (key.objectid == node->bytenr &&
2304 key.type == BTRFS_EXTENT_ITEM_KEY &&
2305 key.offset == node->num_bytes)
2306 ret = 0;
2307 }
2308 if (ret > 0) {
2309 btrfs_release_path(path);
2310 metadata = 0;
2311
2312 key.objectid = node->bytenr;
2313 key.offset = node->num_bytes;
2314 key.type = BTRFS_EXTENT_ITEM_KEY;
2315 goto again;
2316 }
2317 } else {
2318 err = -EIO;
2319 goto out;
2320 }
2321 }
2322
2323 leaf = path->nodes[0];
2324 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2325 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2326 if (item_size < sizeof(*ei)) {
2327 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2328 path, (u64)-1, 0);
2329 if (ret < 0) {
2330 err = ret;
2331 goto out;
2332 }
2333 leaf = path->nodes[0];
2334 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2335 }
2336 #endif
2337 BUG_ON(item_size < sizeof(*ei));
2338 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2339 __run_delayed_extent_op(extent_op, leaf, ei);
2340
2341 btrfs_mark_buffer_dirty(leaf);
2342 out:
2343 btrfs_free_path(path);
2344 return err;
2345 }
2346
2347 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2348 struct btrfs_root *root,
2349 struct btrfs_delayed_ref_node *node,
2350 struct btrfs_delayed_extent_op *extent_op,
2351 int insert_reserved)
2352 {
2353 int ret = 0;
2354 struct btrfs_delayed_tree_ref *ref;
2355 struct btrfs_key ins;
2356 u64 parent = 0;
2357 u64 ref_root = 0;
2358 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2359 SKINNY_METADATA);
2360
2361 ref = btrfs_delayed_node_to_tree_ref(node);
2362 trace_run_delayed_tree_ref(node, ref, node->action);
2363
2364 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2365 parent = ref->parent;
2366 ref_root = ref->root;
2367
2368 ins.objectid = node->bytenr;
2369 if (skinny_metadata) {
2370 ins.offset = ref->level;
2371 ins.type = BTRFS_METADATA_ITEM_KEY;
2372 } else {
2373 ins.offset = node->num_bytes;
2374 ins.type = BTRFS_EXTENT_ITEM_KEY;
2375 }
2376
2377 BUG_ON(node->ref_mod != 1);
2378 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2379 BUG_ON(!extent_op || !extent_op->update_flags);
2380 ret = alloc_reserved_tree_block(trans, root,
2381 parent, ref_root,
2382 extent_op->flags_to_set,
2383 &extent_op->key,
2384 ref->level, &ins);
2385 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2386 ret = __btrfs_inc_extent_ref(trans, root, node,
2387 parent, ref_root,
2388 ref->level, 0, 1,
2389 extent_op);
2390 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2391 ret = __btrfs_free_extent(trans, root, node,
2392 parent, ref_root,
2393 ref->level, 0, 1, extent_op);
2394 } else {
2395 BUG();
2396 }
2397 return ret;
2398 }
2399
2400 /* helper function to actually process a single delayed ref entry */
2401 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2402 struct btrfs_root *root,
2403 struct btrfs_delayed_ref_node *node,
2404 struct btrfs_delayed_extent_op *extent_op,
2405 int insert_reserved)
2406 {
2407 int ret = 0;
2408
2409 if (trans->aborted) {
2410 if (insert_reserved)
2411 btrfs_pin_extent(root, node->bytenr,
2412 node->num_bytes, 1);
2413 return 0;
2414 }
2415
2416 if (btrfs_delayed_ref_is_head(node)) {
2417 struct btrfs_delayed_ref_head *head;
2418 /*
2419 * we've hit the end of the chain and we were supposed
2420 * to insert this extent into the tree. But, it got
2421 * deleted before we ever needed to insert it, so all
2422 * we have to do is clean up the accounting
2423 */
2424 BUG_ON(extent_op);
2425 head = btrfs_delayed_node_to_head(node);
2426 trace_run_delayed_ref_head(node, head, node->action);
2427
2428 if (insert_reserved) {
2429 btrfs_pin_extent(root, node->bytenr,
2430 node->num_bytes, 1);
2431 if (head->is_data) {
2432 ret = btrfs_del_csums(trans, root,
2433 node->bytenr,
2434 node->num_bytes);
2435 }
2436 }
2437
2438 /* Also free its reserved qgroup space */
2439 btrfs_qgroup_free_delayed_ref(root->fs_info,
2440 head->qgroup_ref_root,
2441 head->qgroup_reserved);
2442 return ret;
2443 }
2444
2445 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2446 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2447 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2448 insert_reserved);
2449 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2450 node->type == BTRFS_SHARED_DATA_REF_KEY)
2451 ret = run_delayed_data_ref(trans, root, node, extent_op,
2452 insert_reserved);
2453 else
2454 BUG();
2455 return ret;
2456 }
2457
2458 static inline struct btrfs_delayed_ref_node *
2459 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2460 {
2461 struct btrfs_delayed_ref_node *ref;
2462
2463 if (list_empty(&head->ref_list))
2464 return NULL;
2465
2466 /*
2467 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2468 * This is to prevent a ref count from going down to zero, which deletes
2469 * the extent item from the extent tree, when there still are references
2470 * to add, which would fail because they would not find the extent item.
2471 */
2472 list_for_each_entry(ref, &head->ref_list, list) {
2473 if (ref->action == BTRFS_ADD_DELAYED_REF)
2474 return ref;
2475 }
2476
2477 return list_entry(head->ref_list.next, struct btrfs_delayed_ref_node,
2478 list);
2479 }
2480
2481 /*
2482 * Returns 0 on success or if called with an already aborted transaction.
2483 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2484 */
2485 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2486 struct btrfs_root *root,
2487 unsigned long nr)
2488 {
2489 struct btrfs_delayed_ref_root *delayed_refs;
2490 struct btrfs_delayed_ref_node *ref;
2491 struct btrfs_delayed_ref_head *locked_ref = NULL;
2492 struct btrfs_delayed_extent_op *extent_op;
2493 struct btrfs_fs_info *fs_info = root->fs_info;
2494 ktime_t start = ktime_get();
2495 int ret;
2496 unsigned long count = 0;
2497 unsigned long actual_count = 0;
2498 int must_insert_reserved = 0;
2499
2500 delayed_refs = &trans->transaction->delayed_refs;
2501 while (1) {
2502 if (!locked_ref) {
2503 if (count >= nr)
2504 break;
2505
2506 spin_lock(&delayed_refs->lock);
2507 locked_ref = btrfs_select_ref_head(trans);
2508 if (!locked_ref) {
2509 spin_unlock(&delayed_refs->lock);
2510 break;
2511 }
2512
2513 /* grab the lock that says we are going to process
2514 * all the refs for this head */
2515 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2516 spin_unlock(&delayed_refs->lock);
2517 /*
2518 * we may have dropped the spin lock to get the head
2519 * mutex lock, and that might have given someone else
2520 * time to free the head. If that's true, it has been
2521 * removed from our list and we can move on.
2522 */
2523 if (ret == -EAGAIN) {
2524 locked_ref = NULL;
2525 count++;
2526 continue;
2527 }
2528 }
2529
2530 /*
2531 * We need to try and merge add/drops of the same ref since we
2532 * can run into issues with relocate dropping the implicit ref
2533 * and then it being added back again before the drop can
2534 * finish. If we merged anything we need to re-loop so we can
2535 * get a good ref.
2536 * Or we can get node references of the same type that weren't
2537 * merged when created due to bumps in the tree mod seq, and
2538 * we need to merge them to prevent adding an inline extent
2539 * backref before dropping it (triggering a BUG_ON at
2540 * insert_inline_extent_backref()).
2541 */
2542 spin_lock(&locked_ref->lock);
2543 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2544 locked_ref);
2545
2546 /*
2547 * locked_ref is the head node, so we have to go one
2548 * node back for any delayed ref updates
2549 */
2550 ref = select_delayed_ref(locked_ref);
2551
2552 if (ref && ref->seq &&
2553 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2554 spin_unlock(&locked_ref->lock);
2555 btrfs_delayed_ref_unlock(locked_ref);
2556 spin_lock(&delayed_refs->lock);
2557 locked_ref->processing = 0;
2558 delayed_refs->num_heads_ready++;
2559 spin_unlock(&delayed_refs->lock);
2560 locked_ref = NULL;
2561 cond_resched();
2562 count++;
2563 continue;
2564 }
2565
2566 /*
2567 * record the must insert reserved flag before we
2568 * drop the spin lock.
2569 */
2570 must_insert_reserved = locked_ref->must_insert_reserved;
2571 locked_ref->must_insert_reserved = 0;
2572
2573 extent_op = locked_ref->extent_op;
2574 locked_ref->extent_op = NULL;
2575
2576 if (!ref) {
2577
2578
2579 /* All delayed refs have been processed, Go ahead
2580 * and send the head node to run_one_delayed_ref,
2581 * so that any accounting fixes can happen
2582 */
2583 ref = &locked_ref->node;
2584
2585 if (extent_op && must_insert_reserved) {
2586 btrfs_free_delayed_extent_op(extent_op);
2587 extent_op = NULL;
2588 }
2589
2590 if (extent_op) {
2591 spin_unlock(&locked_ref->lock);
2592 ret = run_delayed_extent_op(trans, root,
2593 ref, extent_op);
2594 btrfs_free_delayed_extent_op(extent_op);
2595
2596 if (ret) {
2597 /*
2598 * Need to reset must_insert_reserved if
2599 * there was an error so the abort stuff
2600 * can cleanup the reserved space
2601 * properly.
2602 */
2603 if (must_insert_reserved)
2604 locked_ref->must_insert_reserved = 1;
2605 locked_ref->processing = 0;
2606 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2607 btrfs_delayed_ref_unlock(locked_ref);
2608 return ret;
2609 }
2610 continue;
2611 }
2612
2613 /*
2614 * Need to drop our head ref lock and re-acquire the
2615 * delayed ref lock and then re-check to make sure
2616 * nobody got added.
2617 */
2618 spin_unlock(&locked_ref->lock);
2619 spin_lock(&delayed_refs->lock);
2620 spin_lock(&locked_ref->lock);
2621 if (!list_empty(&locked_ref->ref_list) ||
2622 locked_ref->extent_op) {
2623 spin_unlock(&locked_ref->lock);
2624 spin_unlock(&delayed_refs->lock);
2625 continue;
2626 }
2627 ref->in_tree = 0;
2628 delayed_refs->num_heads--;
2629 rb_erase(&locked_ref->href_node,
2630 &delayed_refs->href_root);
2631 spin_unlock(&delayed_refs->lock);
2632 } else {
2633 actual_count++;
2634 ref->in_tree = 0;
2635 list_del(&ref->list);
2636 }
2637 atomic_dec(&delayed_refs->num_entries);
2638
2639 if (!btrfs_delayed_ref_is_head(ref)) {
2640 /*
2641 * when we play the delayed ref, also correct the
2642 * ref_mod on head
2643 */
2644 switch (ref->action) {
2645 case BTRFS_ADD_DELAYED_REF:
2646 case BTRFS_ADD_DELAYED_EXTENT:
2647 locked_ref->node.ref_mod -= ref->ref_mod;
2648 break;
2649 case BTRFS_DROP_DELAYED_REF:
2650 locked_ref->node.ref_mod += ref->ref_mod;
2651 break;
2652 default:
2653 WARN_ON(1);
2654 }
2655 }
2656 spin_unlock(&locked_ref->lock);
2657
2658 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2659 must_insert_reserved);
2660
2661 btrfs_free_delayed_extent_op(extent_op);
2662 if (ret) {
2663 locked_ref->processing = 0;
2664 btrfs_delayed_ref_unlock(locked_ref);
2665 btrfs_put_delayed_ref(ref);
2666 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2667 return ret;
2668 }
2669
2670 /*
2671 * If this node is a head, that means all the refs in this head
2672 * have been dealt with, and we will pick the next head to deal
2673 * with, so we must unlock the head and drop it from the cluster
2674 * list before we release it.
2675 */
2676 if (btrfs_delayed_ref_is_head(ref)) {
2677 if (locked_ref->is_data &&
2678 locked_ref->total_ref_mod < 0) {
2679 spin_lock(&delayed_refs->lock);
2680 delayed_refs->pending_csums -= ref->num_bytes;
2681 spin_unlock(&delayed_refs->lock);
2682 }
2683 btrfs_delayed_ref_unlock(locked_ref);
2684 locked_ref = NULL;
2685 }
2686 btrfs_put_delayed_ref(ref);
2687 count++;
2688 cond_resched();
2689 }
2690
2691 /*
2692 * We don't want to include ref heads since we can have empty ref heads
2693 * and those will drastically skew our runtime down since we just do
2694 * accounting, no actual extent tree updates.
2695 */
2696 if (actual_count > 0) {
2697 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2698 u64 avg;
2699
2700 /*
2701 * We weigh the current average higher than our current runtime
2702 * to avoid large swings in the average.
2703 */
2704 spin_lock(&delayed_refs->lock);
2705 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2706 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2707 spin_unlock(&delayed_refs->lock);
2708 }
2709 return 0;
2710 }
2711
2712 #ifdef SCRAMBLE_DELAYED_REFS
2713 /*
2714 * Normally delayed refs get processed in ascending bytenr order. This
2715 * correlates in most cases to the order added. To expose dependencies on this
2716 * order, we start to process the tree in the middle instead of the beginning
2717 */
2718 static u64 find_middle(struct rb_root *root)
2719 {
2720 struct rb_node *n = root->rb_node;
2721 struct btrfs_delayed_ref_node *entry;
2722 int alt = 1;
2723 u64 middle;
2724 u64 first = 0, last = 0;
2725
2726 n = rb_first(root);
2727 if (n) {
2728 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2729 first = entry->bytenr;
2730 }
2731 n = rb_last(root);
2732 if (n) {
2733 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2734 last = entry->bytenr;
2735 }
2736 n = root->rb_node;
2737
2738 while (n) {
2739 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2740 WARN_ON(!entry->in_tree);
2741
2742 middle = entry->bytenr;
2743
2744 if (alt)
2745 n = n->rb_left;
2746 else
2747 n = n->rb_right;
2748
2749 alt = 1 - alt;
2750 }
2751 return middle;
2752 }
2753 #endif
2754
2755 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2756 {
2757 u64 num_bytes;
2758
2759 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2760 sizeof(struct btrfs_extent_inline_ref));
2761 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2762 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2763
2764 /*
2765 * We don't ever fill up leaves all the way so multiply by 2 just to be
2766 * closer to what we're really going to want to use.
2767 */
2768 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2769 }
2770
2771 /*
2772 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2773 * would require to store the csums for that many bytes.
2774 */
2775 u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes)
2776 {
2777 u64 csum_size;
2778 u64 num_csums_per_leaf;
2779 u64 num_csums;
2780
2781 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
2782 num_csums_per_leaf = div64_u64(csum_size,
2783 (u64)btrfs_super_csum_size(root->fs_info->super_copy));
2784 num_csums = div64_u64(csum_bytes, root->sectorsize);
2785 num_csums += num_csums_per_leaf - 1;
2786 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2787 return num_csums;
2788 }
2789
2790 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2791 struct btrfs_root *root)
2792 {
2793 struct btrfs_block_rsv *global_rsv;
2794 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2795 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2796 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2797 u64 num_bytes, num_dirty_bgs_bytes;
2798 int ret = 0;
2799
2800 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2801 num_heads = heads_to_leaves(root, num_heads);
2802 if (num_heads > 1)
2803 num_bytes += (num_heads - 1) * root->nodesize;
2804 num_bytes <<= 1;
2805 num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize;
2806 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root,
2807 num_dirty_bgs);
2808 global_rsv = &root->fs_info->global_block_rsv;
2809
2810 /*
2811 * If we can't allocate any more chunks lets make sure we have _lots_ of
2812 * wiggle room since running delayed refs can create more delayed refs.
2813 */
2814 if (global_rsv->space_info->full) {
2815 num_dirty_bgs_bytes <<= 1;
2816 num_bytes <<= 1;
2817 }
2818
2819 spin_lock(&global_rsv->lock);
2820 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2821 ret = 1;
2822 spin_unlock(&global_rsv->lock);
2823 return ret;
2824 }
2825
2826 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2827 struct btrfs_root *root)
2828 {
2829 struct btrfs_fs_info *fs_info = root->fs_info;
2830 u64 num_entries =
2831 atomic_read(&trans->transaction->delayed_refs.num_entries);
2832 u64 avg_runtime;
2833 u64 val;
2834
2835 smp_mb();
2836 avg_runtime = fs_info->avg_delayed_ref_runtime;
2837 val = num_entries * avg_runtime;
2838 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2839 return 1;
2840 if (val >= NSEC_PER_SEC / 2)
2841 return 2;
2842
2843 return btrfs_check_space_for_delayed_refs(trans, root);
2844 }
2845
2846 struct async_delayed_refs {
2847 struct btrfs_root *root;
2848 u64 transid;
2849 int count;
2850 int error;
2851 int sync;
2852 struct completion wait;
2853 struct btrfs_work work;
2854 };
2855
2856 static void delayed_ref_async_start(struct btrfs_work *work)
2857 {
2858 struct async_delayed_refs *async;
2859 struct btrfs_trans_handle *trans;
2860 int ret;
2861
2862 async = container_of(work, struct async_delayed_refs, work);
2863
2864 /* if the commit is already started, we don't need to wait here */
2865 if (btrfs_transaction_blocked(async->root->fs_info))
2866 goto done;
2867
2868 trans = btrfs_join_transaction(async->root);
2869 if (IS_ERR(trans)) {
2870 async->error = PTR_ERR(trans);
2871 goto done;
2872 }
2873
2874 /*
2875 * trans->sync means that when we call end_transaction, we won't
2876 * wait on delayed refs
2877 */
2878 trans->sync = true;
2879
2880 /* Don't bother flushing if we got into a different transaction */
2881 if (trans->transid > async->transid)
2882 goto end;
2883
2884 ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2885 if (ret)
2886 async->error = ret;
2887 end:
2888 ret = btrfs_end_transaction(trans, async->root);
2889 if (ret && !async->error)
2890 async->error = ret;
2891 done:
2892 if (async->sync)
2893 complete(&async->wait);
2894 else
2895 kfree(async);
2896 }
2897
2898 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2899 unsigned long count, u64 transid, int wait)
2900 {
2901 struct async_delayed_refs *async;
2902 int ret;
2903
2904 async = kmalloc(sizeof(*async), GFP_NOFS);
2905 if (!async)
2906 return -ENOMEM;
2907
2908 async->root = root->fs_info->tree_root;
2909 async->count = count;
2910 async->error = 0;
2911 async->transid = transid;
2912 if (wait)
2913 async->sync = 1;
2914 else
2915 async->sync = 0;
2916 init_completion(&async->wait);
2917
2918 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2919 delayed_ref_async_start, NULL, NULL);
2920
2921 btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2922
2923 if (wait) {
2924 wait_for_completion(&async->wait);
2925 ret = async->error;
2926 kfree(async);
2927 return ret;
2928 }
2929 return 0;
2930 }
2931
2932 /*
2933 * this starts processing the delayed reference count updates and
2934 * extent insertions we have queued up so far. count can be
2935 * 0, which means to process everything in the tree at the start
2936 * of the run (but not newly added entries), or it can be some target
2937 * number you'd like to process.
2938 *
2939 * Returns 0 on success or if called with an aborted transaction
2940 * Returns <0 on error and aborts the transaction
2941 */
2942 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2943 struct btrfs_root *root, unsigned long count)
2944 {
2945 struct rb_node *node;
2946 struct btrfs_delayed_ref_root *delayed_refs;
2947 struct btrfs_delayed_ref_head *head;
2948 int ret;
2949 int run_all = count == (unsigned long)-1;
2950 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
2951
2952 /* We'll clean this up in btrfs_cleanup_transaction */
2953 if (trans->aborted)
2954 return 0;
2955
2956 if (root->fs_info->creating_free_space_tree)
2957 return 0;
2958
2959 if (root == root->fs_info->extent_root)
2960 root = root->fs_info->tree_root;
2961
2962 delayed_refs = &trans->transaction->delayed_refs;
2963 if (count == 0)
2964 count = atomic_read(&delayed_refs->num_entries) * 2;
2965
2966 again:
2967 #ifdef SCRAMBLE_DELAYED_REFS
2968 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2969 #endif
2970 trans->can_flush_pending_bgs = false;
2971 ret = __btrfs_run_delayed_refs(trans, root, count);
2972 if (ret < 0) {
2973 btrfs_abort_transaction(trans, root, ret);
2974 return ret;
2975 }
2976
2977 if (run_all) {
2978 if (!list_empty(&trans->new_bgs))
2979 btrfs_create_pending_block_groups(trans, root);
2980
2981 spin_lock(&delayed_refs->lock);
2982 node = rb_first(&delayed_refs->href_root);
2983 if (!node) {
2984 spin_unlock(&delayed_refs->lock);
2985 goto out;
2986 }
2987 count = (unsigned long)-1;
2988
2989 while (node) {
2990 head = rb_entry(node, struct btrfs_delayed_ref_head,
2991 href_node);
2992 if (btrfs_delayed_ref_is_head(&head->node)) {
2993 struct btrfs_delayed_ref_node *ref;
2994
2995 ref = &head->node;
2996 atomic_inc(&ref->refs);
2997
2998 spin_unlock(&delayed_refs->lock);
2999 /*
3000 * Mutex was contended, block until it's
3001 * released and try again
3002 */
3003 mutex_lock(&head->mutex);
3004 mutex_unlock(&head->mutex);
3005
3006 btrfs_put_delayed_ref(ref);
3007 cond_resched();
3008 goto again;
3009 } else {
3010 WARN_ON(1);
3011 }
3012 node = rb_next(node);
3013 }
3014 spin_unlock(&delayed_refs->lock);
3015 cond_resched();
3016 goto again;
3017 }
3018 out:
3019 assert_qgroups_uptodate(trans);
3020 trans->can_flush_pending_bgs = can_flush_pending_bgs;
3021 return 0;
3022 }
3023
3024 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
3025 struct btrfs_root *root,
3026 u64 bytenr, u64 num_bytes, u64 flags,
3027 int level, int is_data)
3028 {
3029 struct btrfs_delayed_extent_op *extent_op;
3030 int ret;
3031
3032 extent_op = btrfs_alloc_delayed_extent_op();
3033 if (!extent_op)
3034 return -ENOMEM;
3035
3036 extent_op->flags_to_set = flags;
3037 extent_op->update_flags = true;
3038 extent_op->update_key = false;
3039 extent_op->is_data = is_data ? true : false;
3040 extent_op->level = level;
3041
3042 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
3043 num_bytes, extent_op);
3044 if (ret)
3045 btrfs_free_delayed_extent_op(extent_op);
3046 return ret;
3047 }
3048
3049 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
3050 struct btrfs_root *root,
3051 struct btrfs_path *path,
3052 u64 objectid, u64 offset, u64 bytenr)
3053 {
3054 struct btrfs_delayed_ref_head *head;
3055 struct btrfs_delayed_ref_node *ref;
3056 struct btrfs_delayed_data_ref *data_ref;
3057 struct btrfs_delayed_ref_root *delayed_refs;
3058 int ret = 0;
3059
3060 delayed_refs = &trans->transaction->delayed_refs;
3061 spin_lock(&delayed_refs->lock);
3062 head = btrfs_find_delayed_ref_head(trans, bytenr);
3063 if (!head) {
3064 spin_unlock(&delayed_refs->lock);
3065 return 0;
3066 }
3067
3068 if (!mutex_trylock(&head->mutex)) {
3069 atomic_inc(&head->node.refs);
3070 spin_unlock(&delayed_refs->lock);
3071
3072 btrfs_release_path(path);
3073
3074 /*
3075 * Mutex was contended, block until it's released and let
3076 * caller try again
3077 */
3078 mutex_lock(&head->mutex);
3079 mutex_unlock(&head->mutex);
3080 btrfs_put_delayed_ref(&head->node);
3081 return -EAGAIN;
3082 }
3083 spin_unlock(&delayed_refs->lock);
3084
3085 spin_lock(&head->lock);
3086 list_for_each_entry(ref, &head->ref_list, list) {
3087 /* If it's a shared ref we know a cross reference exists */
3088 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3089 ret = 1;
3090 break;
3091 }
3092
3093 data_ref = btrfs_delayed_node_to_data_ref(ref);
3094
3095 /*
3096 * If our ref doesn't match the one we're currently looking at
3097 * then we have a cross reference.
3098 */
3099 if (data_ref->root != root->root_key.objectid ||
3100 data_ref->objectid != objectid ||
3101 data_ref->offset != offset) {
3102 ret = 1;
3103 break;
3104 }
3105 }
3106 spin_unlock(&head->lock);
3107 mutex_unlock(&head->mutex);
3108 return ret;
3109 }
3110
3111 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
3112 struct btrfs_root *root,
3113 struct btrfs_path *path,
3114 u64 objectid, u64 offset, u64 bytenr)
3115 {
3116 struct btrfs_root *extent_root = root->fs_info->extent_root;
3117 struct extent_buffer *leaf;
3118 struct btrfs_extent_data_ref *ref;
3119 struct btrfs_extent_inline_ref *iref;
3120 struct btrfs_extent_item *ei;
3121 struct btrfs_key key;
3122 u32 item_size;
3123 int ret;
3124
3125 key.objectid = bytenr;
3126 key.offset = (u64)-1;
3127 key.type = BTRFS_EXTENT_ITEM_KEY;
3128
3129 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3130 if (ret < 0)
3131 goto out;
3132 BUG_ON(ret == 0); /* Corruption */
3133
3134 ret = -ENOENT;
3135 if (path->slots[0] == 0)
3136 goto out;
3137
3138 path->slots[0]--;
3139 leaf = path->nodes[0];
3140 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3141
3142 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3143 goto out;
3144
3145 ret = 1;
3146 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3147 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3148 if (item_size < sizeof(*ei)) {
3149 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3150 goto out;
3151 }
3152 #endif
3153 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3154
3155 if (item_size != sizeof(*ei) +
3156 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3157 goto out;
3158
3159 if (btrfs_extent_generation(leaf, ei) <=
3160 btrfs_root_last_snapshot(&root->root_item))
3161 goto out;
3162
3163 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3164 if (btrfs_extent_inline_ref_type(leaf, iref) !=
3165 BTRFS_EXTENT_DATA_REF_KEY)
3166 goto out;
3167
3168 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3169 if (btrfs_extent_refs(leaf, ei) !=
3170 btrfs_extent_data_ref_count(leaf, ref) ||
3171 btrfs_extent_data_ref_root(leaf, ref) !=
3172 root->root_key.objectid ||
3173 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3174 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3175 goto out;
3176
3177 ret = 0;
3178 out:
3179 return ret;
3180 }
3181
3182 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
3183 struct btrfs_root *root,
3184 u64 objectid, u64 offset, u64 bytenr)
3185 {
3186 struct btrfs_path *path;
3187 int ret;
3188 int ret2;
3189
3190 path = btrfs_alloc_path();
3191 if (!path)
3192 return -ENOENT;
3193
3194 do {
3195 ret = check_committed_ref(trans, root, path, objectid,
3196 offset, bytenr);
3197 if (ret && ret != -ENOENT)
3198 goto out;
3199
3200 ret2 = check_delayed_ref(trans, root, path, objectid,
3201 offset, bytenr);
3202 } while (ret2 == -EAGAIN);
3203
3204 if (ret2 && ret2 != -ENOENT) {
3205 ret = ret2;
3206 goto out;
3207 }
3208
3209 if (ret != -ENOENT || ret2 != -ENOENT)
3210 ret = 0;
3211 out:
3212 btrfs_free_path(path);
3213 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3214 WARN_ON(ret > 0);
3215 return ret;
3216 }
3217
3218 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3219 struct btrfs_root *root,
3220 struct extent_buffer *buf,
3221 int full_backref, int inc)
3222 {
3223 u64 bytenr;
3224 u64 num_bytes;
3225 u64 parent;
3226 u64 ref_root;
3227 u32 nritems;
3228 struct btrfs_key key;
3229 struct btrfs_file_extent_item *fi;
3230 int i;
3231 int level;
3232 int ret = 0;
3233 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3234 u64, u64, u64, u64, u64, u64);
3235
3236
3237 if (btrfs_test_is_dummy_root(root))
3238 return 0;
3239
3240 ref_root = btrfs_header_owner(buf);
3241 nritems = btrfs_header_nritems(buf);
3242 level = btrfs_header_level(buf);
3243
3244 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3245 return 0;
3246
3247 if (inc)
3248 process_func = btrfs_inc_extent_ref;
3249 else
3250 process_func = btrfs_free_extent;
3251
3252 if (full_backref)
3253 parent = buf->start;
3254 else
3255 parent = 0;
3256
3257 for (i = 0; i < nritems; i++) {
3258 if (level == 0) {
3259 btrfs_item_key_to_cpu(buf, &key, i);
3260 if (key.type != BTRFS_EXTENT_DATA_KEY)
3261 continue;
3262 fi = btrfs_item_ptr(buf, i,
3263 struct btrfs_file_extent_item);
3264 if (btrfs_file_extent_type(buf, fi) ==
3265 BTRFS_FILE_EXTENT_INLINE)
3266 continue;
3267 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3268 if (bytenr == 0)
3269 continue;
3270
3271 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3272 key.offset -= btrfs_file_extent_offset(buf, fi);
3273 ret = process_func(trans, root, bytenr, num_bytes,
3274 parent, ref_root, key.objectid,
3275 key.offset);
3276 if (ret)
3277 goto fail;
3278 } else {
3279 bytenr = btrfs_node_blockptr(buf, i);
3280 num_bytes = root->nodesize;
3281 ret = process_func(trans, root, bytenr, num_bytes,
3282 parent, ref_root, level - 1, 0);
3283 if (ret)
3284 goto fail;
3285 }
3286 }
3287 return 0;
3288 fail:
3289 return ret;
3290 }
3291
3292 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3293 struct extent_buffer *buf, int full_backref)
3294 {
3295 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3296 }
3297
3298 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3299 struct extent_buffer *buf, int full_backref)
3300 {
3301 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3302 }
3303
3304 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3305 struct btrfs_root *root,
3306 struct btrfs_path *path,
3307 struct btrfs_block_group_cache *cache)
3308 {
3309 int ret;
3310 struct btrfs_root *extent_root = root->fs_info->extent_root;
3311 unsigned long bi;
3312 struct extent_buffer *leaf;
3313
3314 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3315 if (ret) {
3316 if (ret > 0)
3317 ret = -ENOENT;
3318 goto fail;
3319 }
3320
3321 leaf = path->nodes[0];
3322 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3323 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3324 btrfs_mark_buffer_dirty(leaf);
3325 fail:
3326 btrfs_release_path(path);
3327 return ret;
3328
3329 }
3330
3331 static struct btrfs_block_group_cache *
3332 next_block_group(struct btrfs_root *root,
3333 struct btrfs_block_group_cache *cache)
3334 {
3335 struct rb_node *node;
3336
3337 spin_lock(&root->fs_info->block_group_cache_lock);
3338
3339 /* If our block group was removed, we need a full search. */
3340 if (RB_EMPTY_NODE(&cache->cache_node)) {
3341 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3342
3343 spin_unlock(&root->fs_info->block_group_cache_lock);
3344 btrfs_put_block_group(cache);
3345 cache = btrfs_lookup_first_block_group(root->fs_info,
3346 next_bytenr);
3347 return cache;
3348 }
3349 node = rb_next(&cache->cache_node);
3350 btrfs_put_block_group(cache);
3351 if (node) {
3352 cache = rb_entry(node, struct btrfs_block_group_cache,
3353 cache_node);
3354 btrfs_get_block_group(cache);
3355 } else
3356 cache = NULL;
3357 spin_unlock(&root->fs_info->block_group_cache_lock);
3358 return cache;
3359 }
3360
3361 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3362 struct btrfs_trans_handle *trans,
3363 struct btrfs_path *path)
3364 {
3365 struct btrfs_root *root = block_group->fs_info->tree_root;
3366 struct inode *inode = NULL;
3367 u64 alloc_hint = 0;
3368 int dcs = BTRFS_DC_ERROR;
3369 u64 num_pages = 0;
3370 int retries = 0;
3371 int ret = 0;
3372
3373 /*
3374 * If this block group is smaller than 100 megs don't bother caching the
3375 * block group.
3376 */
3377 if (block_group->key.offset < (100 * SZ_1M)) {
3378 spin_lock(&block_group->lock);
3379 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3380 spin_unlock(&block_group->lock);
3381 return 0;
3382 }
3383
3384 if (trans->aborted)
3385 return 0;
3386 again:
3387 inode = lookup_free_space_inode(root, block_group, path);
3388 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3389 ret = PTR_ERR(inode);
3390 btrfs_release_path(path);
3391 goto out;
3392 }
3393
3394 if (IS_ERR(inode)) {
3395 BUG_ON(retries);
3396 retries++;
3397
3398 if (block_group->ro)
3399 goto out_free;
3400
3401 ret = create_free_space_inode(root, trans, block_group, path);
3402 if (ret)
3403 goto out_free;
3404 goto again;
3405 }
3406
3407 /* We've already setup this transaction, go ahead and exit */
3408 if (block_group->cache_generation == trans->transid &&
3409 i_size_read(inode)) {
3410 dcs = BTRFS_DC_SETUP;
3411 goto out_put;
3412 }
3413
3414 /*
3415 * We want to set the generation to 0, that way if anything goes wrong
3416 * from here on out we know not to trust this cache when we load up next
3417 * time.
3418 */
3419 BTRFS_I(inode)->generation = 0;
3420 ret = btrfs_update_inode(trans, root, inode);
3421 if (ret) {
3422 /*
3423 * So theoretically we could recover from this, simply set the
3424 * super cache generation to 0 so we know to invalidate the
3425 * cache, but then we'd have to keep track of the block groups
3426 * that fail this way so we know we _have_ to reset this cache
3427 * before the next commit or risk reading stale cache. So to
3428 * limit our exposure to horrible edge cases lets just abort the
3429 * transaction, this only happens in really bad situations
3430 * anyway.
3431 */
3432 btrfs_abort_transaction(trans, root, ret);
3433 goto out_put;
3434 }
3435 WARN_ON(ret);
3436
3437 if (i_size_read(inode) > 0) {
3438 ret = btrfs_check_trunc_cache_free_space(root,
3439 &root->fs_info->global_block_rsv);
3440 if (ret)
3441 goto out_put;
3442
3443 ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode);
3444 if (ret)
3445 goto out_put;
3446 }
3447
3448 spin_lock(&block_group->lock);
3449 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3450 !btrfs_test_opt(root, SPACE_CACHE)) {
3451 /*
3452 * don't bother trying to write stuff out _if_
3453 * a) we're not cached,
3454 * b) we're with nospace_cache mount option.
3455 */
3456 dcs = BTRFS_DC_WRITTEN;
3457 spin_unlock(&block_group->lock);
3458 goto out_put;
3459 }
3460 spin_unlock(&block_group->lock);
3461
3462 /*
3463 * We hit an ENOSPC when setting up the cache in this transaction, just
3464 * skip doing the setup, we've already cleared the cache so we're safe.
3465 */
3466 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3467 ret = -ENOSPC;
3468 goto out_put;
3469 }
3470
3471 /*
3472 * Try to preallocate enough space based on how big the block group is.
3473 * Keep in mind this has to include any pinned space which could end up
3474 * taking up quite a bit since it's not folded into the other space
3475 * cache.
3476 */
3477 num_pages = div_u64(block_group->key.offset, SZ_256M);
3478 if (!num_pages)
3479 num_pages = 1;
3480
3481 num_pages *= 16;
3482 num_pages *= PAGE_SIZE;
3483
3484 ret = btrfs_check_data_free_space(inode, 0, num_pages);
3485 if (ret)
3486 goto out_put;
3487
3488 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3489 num_pages, num_pages,
3490 &alloc_hint);
3491 /*
3492 * Our cache requires contiguous chunks so that we don't modify a bunch
3493 * of metadata or split extents when writing the cache out, which means
3494 * we can enospc if we are heavily fragmented in addition to just normal
3495 * out of space conditions. So if we hit this just skip setting up any
3496 * other block groups for this transaction, maybe we'll unpin enough
3497 * space the next time around.
3498 */
3499 if (!ret)
3500 dcs = BTRFS_DC_SETUP;
3501 else if (ret == -ENOSPC)
3502 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3503 btrfs_free_reserved_data_space(inode, 0, num_pages);
3504
3505 out_put:
3506 iput(inode);
3507 out_free:
3508 btrfs_release_path(path);
3509 out:
3510 spin_lock(&block_group->lock);
3511 if (!ret && dcs == BTRFS_DC_SETUP)
3512 block_group->cache_generation = trans->transid;
3513 block_group->disk_cache_state = dcs;
3514 spin_unlock(&block_group->lock);
3515
3516 return ret;
3517 }
3518
3519 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3520 struct btrfs_root *root)
3521 {
3522 struct btrfs_block_group_cache *cache, *tmp;
3523 struct btrfs_transaction *cur_trans = trans->transaction;
3524 struct btrfs_path *path;
3525
3526 if (list_empty(&cur_trans->dirty_bgs) ||
3527 !btrfs_test_opt(root, SPACE_CACHE))
3528 return 0;
3529
3530 path = btrfs_alloc_path();
3531 if (!path)
3532 return -ENOMEM;
3533
3534 /* Could add new block groups, use _safe just in case */
3535 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3536 dirty_list) {
3537 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3538 cache_save_setup(cache, trans, path);
3539 }
3540
3541 btrfs_free_path(path);
3542 return 0;
3543 }
3544
3545 /*
3546 * transaction commit does final block group cache writeback during a
3547 * critical section where nothing is allowed to change the FS. This is
3548 * required in order for the cache to actually match the block group,
3549 * but can introduce a lot of latency into the commit.
3550 *
3551 * So, btrfs_start_dirty_block_groups is here to kick off block group
3552 * cache IO. There's a chance we'll have to redo some of it if the
3553 * block group changes again during the commit, but it greatly reduces
3554 * the commit latency by getting rid of the easy block groups while
3555 * we're still allowing others to join the commit.
3556 */
3557 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3558 struct btrfs_root *root)
3559 {
3560 struct btrfs_block_group_cache *cache;
3561 struct btrfs_transaction *cur_trans = trans->transaction;
3562 int ret = 0;
3563 int should_put;
3564 struct btrfs_path *path = NULL;
3565 LIST_HEAD(dirty);
3566 struct list_head *io = &cur_trans->io_bgs;
3567 int num_started = 0;
3568 int loops = 0;
3569
3570 spin_lock(&cur_trans->dirty_bgs_lock);
3571 if (list_empty(&cur_trans->dirty_bgs)) {
3572 spin_unlock(&cur_trans->dirty_bgs_lock);
3573 return 0;
3574 }
3575 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3576 spin_unlock(&cur_trans->dirty_bgs_lock);
3577
3578 again:
3579 /*
3580 * make sure all the block groups on our dirty list actually
3581 * exist
3582 */
3583 btrfs_create_pending_block_groups(trans, root);
3584
3585 if (!path) {
3586 path = btrfs_alloc_path();
3587 if (!path)
3588 return -ENOMEM;
3589 }
3590
3591 /*
3592 * cache_write_mutex is here only to save us from balance or automatic
3593 * removal of empty block groups deleting this block group while we are
3594 * writing out the cache
3595 */
3596 mutex_lock(&trans->transaction->cache_write_mutex);
3597 while (!list_empty(&dirty)) {
3598 cache = list_first_entry(&dirty,
3599 struct btrfs_block_group_cache,
3600 dirty_list);
3601 /*
3602 * this can happen if something re-dirties a block
3603 * group that is already under IO. Just wait for it to
3604 * finish and then do it all again
3605 */
3606 if (!list_empty(&cache->io_list)) {
3607 list_del_init(&cache->io_list);
3608 btrfs_wait_cache_io(root, trans, cache,
3609 &cache->io_ctl, path,
3610 cache->key.objectid);
3611 btrfs_put_block_group(cache);
3612 }
3613
3614
3615 /*
3616 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3617 * if it should update the cache_state. Don't delete
3618 * until after we wait.
3619 *
3620 * Since we're not running in the commit critical section
3621 * we need the dirty_bgs_lock to protect from update_block_group
3622 */
3623 spin_lock(&cur_trans->dirty_bgs_lock);
3624 list_del_init(&cache->dirty_list);
3625 spin_unlock(&cur_trans->dirty_bgs_lock);
3626
3627 should_put = 1;
3628
3629 cache_save_setup(cache, trans, path);
3630
3631 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3632 cache->io_ctl.inode = NULL;
3633 ret = btrfs_write_out_cache(root, trans, cache, path);
3634 if (ret == 0 && cache->io_ctl.inode) {
3635 num_started++;
3636 should_put = 0;
3637
3638 /*
3639 * the cache_write_mutex is protecting
3640 * the io_list
3641 */
3642 list_add_tail(&cache->io_list, io);
3643 } else {
3644 /*
3645 * if we failed to write the cache, the
3646 * generation will be bad and life goes on
3647 */
3648 ret = 0;
3649 }
3650 }
3651 if (!ret) {
3652 ret = write_one_cache_group(trans, root, path, cache);
3653 /*
3654 * Our block group might still be attached to the list
3655 * of new block groups in the transaction handle of some
3656 * other task (struct btrfs_trans_handle->new_bgs). This
3657 * means its block group item isn't yet in the extent
3658 * tree. If this happens ignore the error, as we will
3659 * try again later in the critical section of the
3660 * transaction commit.
3661 */
3662 if (ret == -ENOENT) {
3663 ret = 0;
3664 spin_lock(&cur_trans->dirty_bgs_lock);
3665 if (list_empty(&cache->dirty_list)) {
3666 list_add_tail(&cache->dirty_list,
3667 &cur_trans->dirty_bgs);
3668 btrfs_get_block_group(cache);
3669 }
3670 spin_unlock(&cur_trans->dirty_bgs_lock);
3671 } else if (ret) {
3672 btrfs_abort_transaction(trans, root, ret);
3673 }
3674 }
3675
3676 /* if its not on the io list, we need to put the block group */
3677 if (should_put)
3678 btrfs_put_block_group(cache);
3679
3680 if (ret)
3681 break;
3682
3683 /*
3684 * Avoid blocking other tasks for too long. It might even save
3685 * us from writing caches for block groups that are going to be
3686 * removed.
3687 */
3688 mutex_unlock(&trans->transaction->cache_write_mutex);
3689 mutex_lock(&trans->transaction->cache_write_mutex);
3690 }
3691 mutex_unlock(&trans->transaction->cache_write_mutex);
3692
3693 /*
3694 * go through delayed refs for all the stuff we've just kicked off
3695 * and then loop back (just once)
3696 */
3697 ret = btrfs_run_delayed_refs(trans, root, 0);
3698 if (!ret && loops == 0) {
3699 loops++;
3700 spin_lock(&cur_trans->dirty_bgs_lock);
3701 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3702 /*
3703 * dirty_bgs_lock protects us from concurrent block group
3704 * deletes too (not just cache_write_mutex).
3705 */
3706 if (!list_empty(&dirty)) {
3707 spin_unlock(&cur_trans->dirty_bgs_lock);
3708 goto again;
3709 }
3710 spin_unlock(&cur_trans->dirty_bgs_lock);
3711 }
3712
3713 btrfs_free_path(path);
3714 return ret;
3715 }
3716
3717 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3718 struct btrfs_root *root)
3719 {
3720 struct btrfs_block_group_cache *cache;
3721 struct btrfs_transaction *cur_trans = trans->transaction;
3722 int ret = 0;
3723 int should_put;
3724 struct btrfs_path *path;
3725 struct list_head *io = &cur_trans->io_bgs;
3726 int num_started = 0;
3727
3728 path = btrfs_alloc_path();
3729 if (!path)
3730 return -ENOMEM;
3731
3732 /*
3733 * Even though we are in the critical section of the transaction commit,
3734 * we can still have concurrent tasks adding elements to this
3735 * transaction's list of dirty block groups. These tasks correspond to
3736 * endio free space workers started when writeback finishes for a
3737 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3738 * allocate new block groups as a result of COWing nodes of the root
3739 * tree when updating the free space inode. The writeback for the space
3740 * caches is triggered by an earlier call to
3741 * btrfs_start_dirty_block_groups() and iterations of the following
3742 * loop.
3743 * Also we want to do the cache_save_setup first and then run the
3744 * delayed refs to make sure we have the best chance at doing this all
3745 * in one shot.
3746 */
3747 spin_lock(&cur_trans->dirty_bgs_lock);
3748 while (!list_empty(&cur_trans->dirty_bgs)) {
3749 cache = list_first_entry(&cur_trans->dirty_bgs,
3750 struct btrfs_block_group_cache,
3751 dirty_list);
3752
3753 /*
3754 * this can happen if cache_save_setup re-dirties a block
3755 * group that is already under IO. Just wait for it to
3756 * finish and then do it all again
3757 */
3758 if (!list_empty(&cache->io_list)) {
3759 spin_unlock(&cur_trans->dirty_bgs_lock);
3760 list_del_init(&cache->io_list);
3761 btrfs_wait_cache_io(root, trans, cache,
3762 &cache->io_ctl, path,
3763 cache->key.objectid);
3764 btrfs_put_block_group(cache);
3765 spin_lock(&cur_trans->dirty_bgs_lock);
3766 }
3767
3768 /*
3769 * don't remove from the dirty list until after we've waited
3770 * on any pending IO
3771 */
3772 list_del_init(&cache->dirty_list);
3773 spin_unlock(&cur_trans->dirty_bgs_lock);
3774 should_put = 1;
3775
3776 cache_save_setup(cache, trans, path);
3777
3778 if (!ret)
3779 ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1);
3780
3781 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3782 cache->io_ctl.inode = NULL;
3783 ret = btrfs_write_out_cache(root, trans, cache, path);
3784 if (ret == 0 && cache->io_ctl.inode) {
3785 num_started++;
3786 should_put = 0;
3787 list_add_tail(&cache->io_list, io);
3788 } else {
3789 /*
3790 * if we failed to write the cache, the
3791 * generation will be bad and life goes on
3792 */
3793 ret = 0;
3794 }
3795 }
3796 if (!ret) {
3797 ret = write_one_cache_group(trans, root, path, cache);
3798 /*
3799 * One of the free space endio workers might have
3800 * created a new block group while updating a free space
3801 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3802 * and hasn't released its transaction handle yet, in
3803 * which case the new block group is still attached to
3804 * its transaction handle and its creation has not
3805 * finished yet (no block group item in the extent tree
3806 * yet, etc). If this is the case, wait for all free
3807 * space endio workers to finish and retry. This is a
3808 * a very rare case so no need for a more efficient and
3809 * complex approach.
3810 */
3811 if (ret == -ENOENT) {
3812 wait_event(cur_trans->writer_wait,
3813 atomic_read(&cur_trans->num_writers) == 1);
3814 ret = write_one_cache_group(trans, root, path,
3815 cache);
3816 }
3817 if (ret)
3818 btrfs_abort_transaction(trans, root, ret);
3819 }
3820
3821 /* if its not on the io list, we need to put the block group */
3822 if (should_put)
3823 btrfs_put_block_group(cache);
3824 spin_lock(&cur_trans->dirty_bgs_lock);
3825 }
3826 spin_unlock(&cur_trans->dirty_bgs_lock);
3827
3828 while (!list_empty(io)) {
3829 cache = list_first_entry(io, struct btrfs_block_group_cache,
3830 io_list);
3831 list_del_init(&cache->io_list);
3832 btrfs_wait_cache_io(root, trans, cache,
3833 &cache->io_ctl, path, cache->key.objectid);
3834 btrfs_put_block_group(cache);
3835 }
3836
3837 btrfs_free_path(path);
3838 return ret;
3839 }
3840
3841 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3842 {
3843 struct btrfs_block_group_cache *block_group;
3844 int readonly = 0;
3845
3846 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3847 if (!block_group || block_group->ro)
3848 readonly = 1;
3849 if (block_group)
3850 btrfs_put_block_group(block_group);
3851 return readonly;
3852 }
3853
3854 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3855 {
3856 struct btrfs_block_group_cache *bg;
3857 bool ret = true;
3858
3859 bg = btrfs_lookup_block_group(fs_info, bytenr);
3860 if (!bg)
3861 return false;
3862
3863 spin_lock(&bg->lock);
3864 if (bg->ro)
3865 ret = false;
3866 else
3867 atomic_inc(&bg->nocow_writers);
3868 spin_unlock(&bg->lock);
3869
3870 /* no put on block group, done by btrfs_dec_nocow_writers */
3871 if (!ret)
3872 btrfs_put_block_group(bg);
3873
3874 return ret;
3875
3876 }
3877
3878 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3879 {
3880 struct btrfs_block_group_cache *bg;
3881
3882 bg = btrfs_lookup_block_group(fs_info, bytenr);
3883 ASSERT(bg);
3884 if (atomic_dec_and_test(&bg->nocow_writers))
3885 wake_up_atomic_t(&bg->nocow_writers);
3886 /*
3887 * Once for our lookup and once for the lookup done by a previous call
3888 * to btrfs_inc_nocow_writers()
3889 */
3890 btrfs_put_block_group(bg);
3891 btrfs_put_block_group(bg);
3892 }
3893
3894 static int btrfs_wait_nocow_writers_atomic_t(atomic_t *a)
3895 {
3896 schedule();
3897 return 0;
3898 }
3899
3900 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
3901 {
3902 wait_on_atomic_t(&bg->nocow_writers,
3903 btrfs_wait_nocow_writers_atomic_t,
3904 TASK_UNINTERRUPTIBLE);
3905 }
3906
3907 static const char *alloc_name(u64 flags)
3908 {
3909 switch (flags) {
3910 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3911 return "mixed";
3912 case BTRFS_BLOCK_GROUP_METADATA:
3913 return "metadata";
3914 case BTRFS_BLOCK_GROUP_DATA:
3915 return "data";
3916 case BTRFS_BLOCK_GROUP_SYSTEM:
3917 return "system";
3918 default:
3919 WARN_ON(1);
3920 return "invalid-combination";
3921 };
3922 }
3923
3924 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3925 u64 total_bytes, u64 bytes_used,
3926 u64 bytes_readonly,
3927 struct btrfs_space_info **space_info)
3928 {
3929 struct btrfs_space_info *found;
3930 int i;
3931 int factor;
3932 int ret;
3933
3934 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3935 BTRFS_BLOCK_GROUP_RAID10))
3936 factor = 2;
3937 else
3938 factor = 1;
3939
3940 found = __find_space_info(info, flags);
3941 if (found) {
3942 spin_lock(&found->lock);
3943 found->total_bytes += total_bytes;
3944 found->disk_total += total_bytes * factor;
3945 found->bytes_used += bytes_used;
3946 found->disk_used += bytes_used * factor;
3947 found->bytes_readonly += bytes_readonly;
3948 if (total_bytes > 0)
3949 found->full = 0;
3950 space_info_add_new_bytes(info, found, total_bytes -
3951 bytes_used - bytes_readonly);
3952 spin_unlock(&found->lock);
3953 *space_info = found;
3954 return 0;
3955 }
3956 found = kzalloc(sizeof(*found), GFP_NOFS);
3957 if (!found)
3958 return -ENOMEM;
3959
3960 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3961 if (ret) {
3962 kfree(found);
3963 return ret;
3964 }
3965
3966 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3967 INIT_LIST_HEAD(&found->block_groups[i]);
3968 init_rwsem(&found->groups_sem);
3969 spin_lock_init(&found->lock);
3970 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3971 found->total_bytes = total_bytes;
3972 found->disk_total = total_bytes * factor;
3973 found->bytes_used = bytes_used;
3974 found->disk_used = bytes_used * factor;
3975 found->bytes_pinned = 0;
3976 found->bytes_reserved = 0;
3977 found->bytes_readonly = bytes_readonly;
3978 found->bytes_may_use = 0;
3979 found->full = 0;
3980 found->max_extent_size = 0;
3981 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3982 found->chunk_alloc = 0;
3983 found->flush = 0;
3984 init_waitqueue_head(&found->wait);
3985 INIT_LIST_HEAD(&found->ro_bgs);
3986 INIT_LIST_HEAD(&found->tickets);
3987 INIT_LIST_HEAD(&found->priority_tickets);
3988
3989 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3990 info->space_info_kobj, "%s",
3991 alloc_name(found->flags));
3992 if (ret) {
3993 kfree(found);
3994 return ret;
3995 }
3996
3997 *space_info = found;
3998 list_add_rcu(&found->list, &info->space_info);
3999 if (flags & BTRFS_BLOCK_GROUP_DATA)
4000 info->data_sinfo = found;
4001
4002 return ret;
4003 }
4004
4005 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
4006 {
4007 u64 extra_flags = chunk_to_extended(flags) &
4008 BTRFS_EXTENDED_PROFILE_MASK;
4009
4010 write_seqlock(&fs_info->profiles_lock);
4011 if (flags & BTRFS_BLOCK_GROUP_DATA)
4012 fs_info->avail_data_alloc_bits |= extra_flags;
4013 if (flags & BTRFS_BLOCK_GROUP_METADATA)
4014 fs_info->avail_metadata_alloc_bits |= extra_flags;
4015 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4016 fs_info->avail_system_alloc_bits |= extra_flags;
4017 write_sequnlock(&fs_info->profiles_lock);
4018 }
4019
4020 /*
4021 * returns target flags in extended format or 0 if restripe for this
4022 * chunk_type is not in progress
4023 *
4024 * should be called with either volume_mutex or balance_lock held
4025 */
4026 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
4027 {
4028 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4029 u64 target = 0;
4030
4031 if (!bctl)
4032 return 0;
4033
4034 if (flags & BTRFS_BLOCK_GROUP_DATA &&
4035 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4036 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
4037 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
4038 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4039 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
4040 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
4041 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4042 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
4043 }
4044
4045 return target;
4046 }
4047
4048 /*
4049 * @flags: available profiles in extended format (see ctree.h)
4050 *
4051 * Returns reduced profile in chunk format. If profile changing is in
4052 * progress (either running or paused) picks the target profile (if it's
4053 * already available), otherwise falls back to plain reducing.
4054 */
4055 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
4056 {
4057 u64 num_devices = root->fs_info->fs_devices->rw_devices;
4058 u64 target;
4059 u64 raid_type;
4060 u64 allowed = 0;
4061
4062 /*
4063 * see if restripe for this chunk_type is in progress, if so
4064 * try to reduce to the target profile
4065 */
4066 spin_lock(&root->fs_info->balance_lock);
4067 target = get_restripe_target(root->fs_info, flags);
4068 if (target) {
4069 /* pick target profile only if it's already available */
4070 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
4071 spin_unlock(&root->fs_info->balance_lock);
4072 return extended_to_chunk(target);
4073 }
4074 }
4075 spin_unlock(&root->fs_info->balance_lock);
4076
4077 /* First, mask out the RAID levels which aren't possible */
4078 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4079 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
4080 allowed |= btrfs_raid_group[raid_type];
4081 }
4082 allowed &= flags;
4083
4084 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
4085 allowed = BTRFS_BLOCK_GROUP_RAID6;
4086 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4087 allowed = BTRFS_BLOCK_GROUP_RAID5;
4088 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4089 allowed = BTRFS_BLOCK_GROUP_RAID10;
4090 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4091 allowed = BTRFS_BLOCK_GROUP_RAID1;
4092 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4093 allowed = BTRFS_BLOCK_GROUP_RAID0;
4094
4095 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4096
4097 return extended_to_chunk(flags | allowed);
4098 }
4099
4100 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
4101 {
4102 unsigned seq;
4103 u64 flags;
4104
4105 do {
4106 flags = orig_flags;
4107 seq = read_seqbegin(&root->fs_info->profiles_lock);
4108
4109 if (flags & BTRFS_BLOCK_GROUP_DATA)
4110 flags |= root->fs_info->avail_data_alloc_bits;
4111 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4112 flags |= root->fs_info->avail_system_alloc_bits;
4113 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4114 flags |= root->fs_info->avail_metadata_alloc_bits;
4115 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
4116
4117 return btrfs_reduce_alloc_profile(root, flags);
4118 }
4119
4120 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
4121 {
4122 u64 flags;
4123 u64 ret;
4124
4125 if (data)
4126 flags = BTRFS_BLOCK_GROUP_DATA;
4127 else if (root == root->fs_info->chunk_root)
4128 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4129 else
4130 flags = BTRFS_BLOCK_GROUP_METADATA;
4131
4132 ret = get_alloc_profile(root, flags);
4133 return ret;
4134 }
4135
4136 int btrfs_alloc_data_chunk_ondemand(struct inode *inode, u64 bytes)
4137 {
4138 struct btrfs_space_info *data_sinfo;
4139 struct btrfs_root *root = BTRFS_I(inode)->root;
4140 struct btrfs_fs_info *fs_info = root->fs_info;
4141 u64 used;
4142 int ret = 0;
4143 int need_commit = 2;
4144 int have_pinned_space;
4145
4146 /* make sure bytes are sectorsize aligned */
4147 bytes = ALIGN(bytes, root->sectorsize);
4148
4149 if (btrfs_is_free_space_inode(inode)) {
4150 need_commit = 0;
4151 ASSERT(current->journal_info);
4152 }
4153
4154 data_sinfo = fs_info->data_sinfo;
4155 if (!data_sinfo)
4156 goto alloc;
4157
4158 again:
4159 /* make sure we have enough space to handle the data first */
4160 spin_lock(&data_sinfo->lock);
4161 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
4162 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
4163 data_sinfo->bytes_may_use;
4164
4165 if (used + bytes > data_sinfo->total_bytes) {
4166 struct btrfs_trans_handle *trans;
4167
4168 /*
4169 * if we don't have enough free bytes in this space then we need
4170 * to alloc a new chunk.
4171 */
4172 if (!data_sinfo->full) {
4173 u64 alloc_target;
4174
4175 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4176 spin_unlock(&data_sinfo->lock);
4177 alloc:
4178 alloc_target = btrfs_get_alloc_profile(root, 1);
4179 /*
4180 * It is ugly that we don't call nolock join
4181 * transaction for the free space inode case here.
4182 * But it is safe because we only do the data space
4183 * reservation for the free space cache in the
4184 * transaction context, the common join transaction
4185 * just increase the counter of the current transaction
4186 * handler, doesn't try to acquire the trans_lock of
4187 * the fs.
4188 */
4189 trans = btrfs_join_transaction(root);
4190 if (IS_ERR(trans))
4191 return PTR_ERR(trans);
4192
4193 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4194 alloc_target,
4195 CHUNK_ALLOC_NO_FORCE);
4196 btrfs_end_transaction(trans, root);
4197 if (ret < 0) {
4198 if (ret != -ENOSPC)
4199 return ret;
4200 else {
4201 have_pinned_space = 1;
4202 goto commit_trans;
4203 }
4204 }
4205
4206 if (!data_sinfo)
4207 data_sinfo = fs_info->data_sinfo;
4208
4209 goto again;
4210 }
4211
4212 /*
4213 * If we don't have enough pinned space to deal with this
4214 * allocation, and no removed chunk in current transaction,
4215 * don't bother committing the transaction.
4216 */
4217 have_pinned_space = percpu_counter_compare(
4218 &data_sinfo->total_bytes_pinned,
4219 used + bytes - data_sinfo->total_bytes);
4220 spin_unlock(&data_sinfo->lock);
4221
4222 /* commit the current transaction and try again */
4223 commit_trans:
4224 if (need_commit &&
4225 !atomic_read(&root->fs_info->open_ioctl_trans)) {
4226 need_commit--;
4227
4228 if (need_commit > 0) {
4229 btrfs_start_delalloc_roots(fs_info, 0, -1);
4230 btrfs_wait_ordered_roots(fs_info, -1, 0, (u64)-1);
4231 }
4232
4233 trans = btrfs_join_transaction(root);
4234 if (IS_ERR(trans))
4235 return PTR_ERR(trans);
4236 if (have_pinned_space >= 0 ||
4237 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4238 &trans->transaction->flags) ||
4239 need_commit > 0) {
4240 ret = btrfs_commit_transaction(trans, root);
4241 if (ret)
4242 return ret;
4243 /*
4244 * The cleaner kthread might still be doing iput
4245 * operations. Wait for it to finish so that
4246 * more space is released.
4247 */
4248 mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex);
4249 mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex);
4250 goto again;
4251 } else {
4252 btrfs_end_transaction(trans, root);
4253 }
4254 }
4255
4256 trace_btrfs_space_reservation(root->fs_info,
4257 "space_info:enospc",
4258 data_sinfo->flags, bytes, 1);
4259 return -ENOSPC;
4260 }
4261 data_sinfo->bytes_may_use += bytes;
4262 trace_btrfs_space_reservation(root->fs_info, "space_info",
4263 data_sinfo->flags, bytes, 1);
4264 spin_unlock(&data_sinfo->lock);
4265
4266 return ret;
4267 }
4268
4269 /*
4270 * New check_data_free_space() with ability for precious data reservation
4271 * Will replace old btrfs_check_data_free_space(), but for patch split,
4272 * add a new function first and then replace it.
4273 */
4274 int btrfs_check_data_free_space(struct inode *inode, u64 start, u64 len)
4275 {
4276 struct btrfs_root *root = BTRFS_I(inode)->root;
4277 int ret;
4278
4279 /* align the range */
4280 len = round_up(start + len, root->sectorsize) -
4281 round_down(start, root->sectorsize);
4282 start = round_down(start, root->sectorsize);
4283
4284 ret = btrfs_alloc_data_chunk_ondemand(inode, len);
4285 if (ret < 0)
4286 return ret;
4287
4288 /*
4289 * Use new btrfs_qgroup_reserve_data to reserve precious data space
4290 *
4291 * TODO: Find a good method to avoid reserve data space for NOCOW
4292 * range, but don't impact performance on quota disable case.
4293 */
4294 ret = btrfs_qgroup_reserve_data(inode, start, len);
4295 return ret;
4296 }
4297
4298 /*
4299 * Called if we need to clear a data reservation for this inode
4300 * Normally in a error case.
4301 *
4302 * This one will *NOT* use accurate qgroup reserved space API, just for case
4303 * which we can't sleep and is sure it won't affect qgroup reserved space.
4304 * Like clear_bit_hook().
4305 */
4306 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4307 u64 len)
4308 {
4309 struct btrfs_root *root = BTRFS_I(inode)->root;
4310 struct btrfs_space_info *data_sinfo;
4311
4312 /* Make sure the range is aligned to sectorsize */
4313 len = round_up(start + len, root->sectorsize) -
4314 round_down(start, root->sectorsize);
4315 start = round_down(start, root->sectorsize);
4316
4317 data_sinfo = root->fs_info->data_sinfo;
4318 spin_lock(&data_sinfo->lock);
4319 if (WARN_ON(data_sinfo->bytes_may_use < len))
4320 data_sinfo->bytes_may_use = 0;
4321 else
4322 data_sinfo->bytes_may_use -= len;
4323 trace_btrfs_space_reservation(root->fs_info, "space_info",
4324 data_sinfo->flags, len, 0);
4325 spin_unlock(&data_sinfo->lock);
4326 }
4327
4328 /*
4329 * Called if we need to clear a data reservation for this inode
4330 * Normally in a error case.
4331 *
4332 * This one will handle the per-inode data rsv map for accurate reserved
4333 * space framework.
4334 */
4335 void btrfs_free_reserved_data_space(struct inode *inode, u64 start, u64 len)
4336 {
4337 btrfs_free_reserved_data_space_noquota(inode, start, len);
4338 btrfs_qgroup_free_data(inode, start, len);
4339 }
4340
4341 static void force_metadata_allocation(struct btrfs_fs_info *info)
4342 {
4343 struct list_head *head = &info->space_info;
4344 struct btrfs_space_info *found;
4345
4346 rcu_read_lock();
4347 list_for_each_entry_rcu(found, head, list) {
4348 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4349 found->force_alloc = CHUNK_ALLOC_FORCE;
4350 }
4351 rcu_read_unlock();
4352 }
4353
4354 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4355 {
4356 return (global->size << 1);
4357 }
4358
4359 static int should_alloc_chunk(struct btrfs_root *root,
4360 struct btrfs_space_info *sinfo, int force)
4361 {
4362 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4363 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
4364 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
4365 u64 thresh;
4366
4367 if (force == CHUNK_ALLOC_FORCE)
4368 return 1;
4369
4370 /*
4371 * We need to take into account the global rsv because for all intents
4372 * and purposes it's used space. Don't worry about locking the
4373 * global_rsv, it doesn't change except when the transaction commits.
4374 */
4375 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4376 num_allocated += calc_global_rsv_need_space(global_rsv);
4377
4378 /*
4379 * in limited mode, we want to have some free space up to
4380 * about 1% of the FS size.
4381 */
4382 if (force == CHUNK_ALLOC_LIMITED) {
4383 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
4384 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4385
4386 if (num_bytes - num_allocated < thresh)
4387 return 1;
4388 }
4389
4390 if (num_allocated + SZ_2M < div_factor(num_bytes, 8))
4391 return 0;
4392 return 1;
4393 }
4394
4395 static u64 get_profile_num_devs(struct btrfs_root *root, u64 type)
4396 {
4397 u64 num_dev;
4398
4399 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4400 BTRFS_BLOCK_GROUP_RAID0 |
4401 BTRFS_BLOCK_GROUP_RAID5 |
4402 BTRFS_BLOCK_GROUP_RAID6))
4403 num_dev = root->fs_info->fs_devices->rw_devices;
4404 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4405 num_dev = 2;
4406 else
4407 num_dev = 1; /* DUP or single */
4408
4409 return num_dev;
4410 }
4411
4412 /*
4413 * If @is_allocation is true, reserve space in the system space info necessary
4414 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4415 * removing a chunk.
4416 */
4417 void check_system_chunk(struct btrfs_trans_handle *trans,
4418 struct btrfs_root *root,
4419 u64 type)
4420 {
4421 struct btrfs_space_info *info;
4422 u64 left;
4423 u64 thresh;
4424 int ret = 0;
4425 u64 num_devs;
4426
4427 /*
4428 * Needed because we can end up allocating a system chunk and for an
4429 * atomic and race free space reservation in the chunk block reserve.
4430 */
4431 ASSERT(mutex_is_locked(&root->fs_info->chunk_mutex));
4432
4433 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4434 spin_lock(&info->lock);
4435 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
4436 info->bytes_reserved - info->bytes_readonly -
4437 info->bytes_may_use;
4438 spin_unlock(&info->lock);
4439
4440 num_devs = get_profile_num_devs(root, type);
4441
4442 /* num_devs device items to update and 1 chunk item to add or remove */
4443 thresh = btrfs_calc_trunc_metadata_size(root, num_devs) +
4444 btrfs_calc_trans_metadata_size(root, 1);
4445
4446 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
4447 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
4448 left, thresh, type);
4449 dump_space_info(info, 0, 0);
4450 }
4451
4452 if (left < thresh) {
4453 u64 flags;
4454
4455 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
4456 /*
4457 * Ignore failure to create system chunk. We might end up not
4458 * needing it, as we might not need to COW all nodes/leafs from
4459 * the paths we visit in the chunk tree (they were already COWed
4460 * or created in the current transaction for example).
4461 */
4462 ret = btrfs_alloc_chunk(trans, root, flags);
4463 }
4464
4465 if (!ret) {
4466 ret = btrfs_block_rsv_add(root->fs_info->chunk_root,
4467 &root->fs_info->chunk_block_rsv,
4468 thresh, BTRFS_RESERVE_NO_FLUSH);
4469 if (!ret)
4470 trans->chunk_bytes_reserved += thresh;
4471 }
4472 }
4473
4474 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4475 struct btrfs_root *extent_root, u64 flags, int force)
4476 {
4477 struct btrfs_space_info *space_info;
4478 struct btrfs_fs_info *fs_info = extent_root->fs_info;
4479 int wait_for_alloc = 0;
4480 int ret = 0;
4481
4482 /* Don't re-enter if we're already allocating a chunk */
4483 if (trans->allocating_chunk)
4484 return -ENOSPC;
4485
4486 space_info = __find_space_info(extent_root->fs_info, flags);
4487 if (!space_info) {
4488 ret = update_space_info(extent_root->fs_info, flags,
4489 0, 0, 0, &space_info);
4490 BUG_ON(ret); /* -ENOMEM */
4491 }
4492 BUG_ON(!space_info); /* Logic error */
4493
4494 again:
4495 spin_lock(&space_info->lock);
4496 if (force < space_info->force_alloc)
4497 force = space_info->force_alloc;
4498 if (space_info->full) {
4499 if (should_alloc_chunk(extent_root, space_info, force))
4500 ret = -ENOSPC;
4501 else
4502 ret = 0;
4503 spin_unlock(&space_info->lock);
4504 return ret;
4505 }
4506
4507 if (!should_alloc_chunk(extent_root, space_info, force)) {
4508 spin_unlock(&space_info->lock);
4509 return 0;
4510 } else if (space_info->chunk_alloc) {
4511 wait_for_alloc = 1;
4512 } else {
4513 space_info->chunk_alloc = 1;
4514 }
4515
4516 spin_unlock(&space_info->lock);
4517
4518 mutex_lock(&fs_info->chunk_mutex);
4519
4520 /*
4521 * The chunk_mutex is held throughout the entirety of a chunk
4522 * allocation, so once we've acquired the chunk_mutex we know that the
4523 * other guy is done and we need to recheck and see if we should
4524 * allocate.
4525 */
4526 if (wait_for_alloc) {
4527 mutex_unlock(&fs_info->chunk_mutex);
4528 wait_for_alloc = 0;
4529 goto again;
4530 }
4531
4532 trans->allocating_chunk = true;
4533
4534 /*
4535 * If we have mixed data/metadata chunks we want to make sure we keep
4536 * allocating mixed chunks instead of individual chunks.
4537 */
4538 if (btrfs_mixed_space_info(space_info))
4539 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4540
4541 /*
4542 * if we're doing a data chunk, go ahead and make sure that
4543 * we keep a reasonable number of metadata chunks allocated in the
4544 * FS as well.
4545 */
4546 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4547 fs_info->data_chunk_allocations++;
4548 if (!(fs_info->data_chunk_allocations %
4549 fs_info->metadata_ratio))
4550 force_metadata_allocation(fs_info);
4551 }
4552
4553 /*
4554 * Check if we have enough space in SYSTEM chunk because we may need
4555 * to update devices.
4556 */
4557 check_system_chunk(trans, extent_root, flags);
4558
4559 ret = btrfs_alloc_chunk(trans, extent_root, flags);
4560 trans->allocating_chunk = false;
4561
4562 spin_lock(&space_info->lock);
4563 if (ret < 0 && ret != -ENOSPC)
4564 goto out;
4565 if (ret)
4566 space_info->full = 1;
4567 else
4568 ret = 1;
4569
4570 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4571 out:
4572 space_info->chunk_alloc = 0;
4573 spin_unlock(&space_info->lock);
4574 mutex_unlock(&fs_info->chunk_mutex);
4575 /*
4576 * When we allocate a new chunk we reserve space in the chunk block
4577 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4578 * add new nodes/leafs to it if we end up needing to do it when
4579 * inserting the chunk item and updating device items as part of the
4580 * second phase of chunk allocation, performed by
4581 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4582 * large number of new block groups to create in our transaction
4583 * handle's new_bgs list to avoid exhausting the chunk block reserve
4584 * in extreme cases - like having a single transaction create many new
4585 * block groups when starting to write out the free space caches of all
4586 * the block groups that were made dirty during the lifetime of the
4587 * transaction.
4588 */
4589 if (trans->can_flush_pending_bgs &&
4590 trans->chunk_bytes_reserved >= (u64)SZ_2M) {
4591 btrfs_create_pending_block_groups(trans, trans->root);
4592 btrfs_trans_release_chunk_metadata(trans);
4593 }
4594 return ret;
4595 }
4596
4597 static int can_overcommit(struct btrfs_root *root,
4598 struct btrfs_space_info *space_info, u64 bytes,
4599 enum btrfs_reserve_flush_enum flush)
4600 {
4601 struct btrfs_block_rsv *global_rsv;
4602 u64 profile;
4603 u64 space_size;
4604 u64 avail;
4605 u64 used;
4606
4607 /* Don't overcommit when in mixed mode. */
4608 if (space_info->flags & BTRFS_BLOCK_GROUP_DATA)
4609 return 0;
4610
4611 BUG_ON(root->fs_info == NULL);
4612 global_rsv = &root->fs_info->global_block_rsv;
4613 profile = btrfs_get_alloc_profile(root, 0);
4614 used = space_info->bytes_used + space_info->bytes_reserved +
4615 space_info->bytes_pinned + space_info->bytes_readonly;
4616
4617 /*
4618 * We only want to allow over committing if we have lots of actual space
4619 * free, but if we don't have enough space to handle the global reserve
4620 * space then we could end up having a real enospc problem when trying
4621 * to allocate a chunk or some other such important allocation.
4622 */
4623 spin_lock(&global_rsv->lock);
4624 space_size = calc_global_rsv_need_space(global_rsv);
4625 spin_unlock(&global_rsv->lock);
4626 if (used + space_size >= space_info->total_bytes)
4627 return 0;
4628
4629 used += space_info->bytes_may_use;
4630
4631 spin_lock(&root->fs_info->free_chunk_lock);
4632 avail = root->fs_info->free_chunk_space;
4633 spin_unlock(&root->fs_info->free_chunk_lock);
4634
4635 /*
4636 * If we have dup, raid1 or raid10 then only half of the free
4637 * space is actually useable. For raid56, the space info used
4638 * doesn't include the parity drive, so we don't have to
4639 * change the math
4640 */
4641 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4642 BTRFS_BLOCK_GROUP_RAID1 |
4643 BTRFS_BLOCK_GROUP_RAID10))
4644 avail >>= 1;
4645
4646 /*
4647 * If we aren't flushing all things, let us overcommit up to
4648 * 1/2th of the space. If we can flush, don't let us overcommit
4649 * too much, let it overcommit up to 1/8 of the space.
4650 */
4651 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4652 avail >>= 3;
4653 else
4654 avail >>= 1;
4655
4656 if (used + bytes < space_info->total_bytes + avail)
4657 return 1;
4658 return 0;
4659 }
4660
4661 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4662 unsigned long nr_pages, int nr_items)
4663 {
4664 struct super_block *sb = root->fs_info->sb;
4665
4666 if (down_read_trylock(&sb->s_umount)) {
4667 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4668 up_read(&sb->s_umount);
4669 } else {
4670 /*
4671 * We needn't worry the filesystem going from r/w to r/o though
4672 * we don't acquire ->s_umount mutex, because the filesystem
4673 * should guarantee the delalloc inodes list be empty after
4674 * the filesystem is readonly(all dirty pages are written to
4675 * the disk).
4676 */
4677 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4678 if (!current->journal_info)
4679 btrfs_wait_ordered_roots(root->fs_info, nr_items,
4680 0, (u64)-1);
4681 }
4682 }
4683
4684 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4685 {
4686 u64 bytes;
4687 int nr;
4688
4689 bytes = btrfs_calc_trans_metadata_size(root, 1);
4690 nr = (int)div64_u64(to_reclaim, bytes);
4691 if (!nr)
4692 nr = 1;
4693 return nr;
4694 }
4695
4696 #define EXTENT_SIZE_PER_ITEM SZ_256K
4697
4698 /*
4699 * shrink metadata reservation for delalloc
4700 */
4701 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4702 bool wait_ordered)
4703 {
4704 struct btrfs_block_rsv *block_rsv;
4705 struct btrfs_space_info *space_info;
4706 struct btrfs_trans_handle *trans;
4707 u64 delalloc_bytes;
4708 u64 max_reclaim;
4709 long time_left;
4710 unsigned long nr_pages;
4711 int loops;
4712 int items;
4713 enum btrfs_reserve_flush_enum flush;
4714
4715 /* Calc the number of the pages we need flush for space reservation */
4716 items = calc_reclaim_items_nr(root, to_reclaim);
4717 to_reclaim = (u64)items * EXTENT_SIZE_PER_ITEM;
4718
4719 trans = (struct btrfs_trans_handle *)current->journal_info;
4720 block_rsv = &root->fs_info->delalloc_block_rsv;
4721 space_info = block_rsv->space_info;
4722
4723 delalloc_bytes = percpu_counter_sum_positive(
4724 &root->fs_info->delalloc_bytes);
4725 if (delalloc_bytes == 0) {
4726 if (trans)
4727 return;
4728 if (wait_ordered)
4729 btrfs_wait_ordered_roots(root->fs_info, items,
4730 0, (u64)-1);
4731 return;
4732 }
4733
4734 loops = 0;
4735 while (delalloc_bytes && loops < 3) {
4736 max_reclaim = min(delalloc_bytes, to_reclaim);
4737 nr_pages = max_reclaim >> PAGE_SHIFT;
4738 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4739 /*
4740 * We need to wait for the async pages to actually start before
4741 * we do anything.
4742 */
4743 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4744 if (!max_reclaim)
4745 goto skip_async;
4746
4747 if (max_reclaim <= nr_pages)
4748 max_reclaim = 0;
4749 else
4750 max_reclaim -= nr_pages;
4751
4752 wait_event(root->fs_info->async_submit_wait,
4753 atomic_read(&root->fs_info->async_delalloc_pages) <=
4754 (int)max_reclaim);
4755 skip_async:
4756 if (!trans)
4757 flush = BTRFS_RESERVE_FLUSH_ALL;
4758 else
4759 flush = BTRFS_RESERVE_NO_FLUSH;
4760 spin_lock(&space_info->lock);
4761 if (can_overcommit(root, space_info, orig, flush)) {
4762 spin_unlock(&space_info->lock);
4763 break;
4764 }
4765 if (list_empty(&space_info->tickets) &&
4766 list_empty(&space_info->priority_tickets)) {
4767 spin_unlock(&space_info->lock);
4768 break;
4769 }
4770 spin_unlock(&space_info->lock);
4771
4772 loops++;
4773 if (wait_ordered && !trans) {
4774 btrfs_wait_ordered_roots(root->fs_info, items,
4775 0, (u64)-1);
4776 } else {
4777 time_left = schedule_timeout_killable(1);
4778 if (time_left)
4779 break;
4780 }
4781 delalloc_bytes = percpu_counter_sum_positive(
4782 &root->fs_info->delalloc_bytes);
4783 }
4784 }
4785
4786 /**
4787 * maybe_commit_transaction - possibly commit the transaction if its ok to
4788 * @root - the root we're allocating for
4789 * @bytes - the number of bytes we want to reserve
4790 * @force - force the commit
4791 *
4792 * This will check to make sure that committing the transaction will actually
4793 * get us somewhere and then commit the transaction if it does. Otherwise it
4794 * will return -ENOSPC.
4795 */
4796 static int may_commit_transaction(struct btrfs_root *root,
4797 struct btrfs_space_info *space_info,
4798 u64 bytes, int force)
4799 {
4800 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4801 struct btrfs_trans_handle *trans;
4802
4803 trans = (struct btrfs_trans_handle *)current->journal_info;
4804 if (trans)
4805 return -EAGAIN;
4806
4807 if (force)
4808 goto commit;
4809
4810 /* See if there is enough pinned space to make this reservation */
4811 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4812 bytes) >= 0)
4813 goto commit;
4814
4815 /*
4816 * See if there is some space in the delayed insertion reservation for
4817 * this reservation.
4818 */
4819 if (space_info != delayed_rsv->space_info)
4820 return -ENOSPC;
4821
4822 spin_lock(&delayed_rsv->lock);
4823 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4824 bytes - delayed_rsv->size) >= 0) {
4825 spin_unlock(&delayed_rsv->lock);
4826 return -ENOSPC;
4827 }
4828 spin_unlock(&delayed_rsv->lock);
4829
4830 commit:
4831 trans = btrfs_join_transaction(root);
4832 if (IS_ERR(trans))
4833 return -ENOSPC;
4834
4835 return btrfs_commit_transaction(trans, root);
4836 }
4837
4838 struct reserve_ticket {
4839 u64 bytes;
4840 int error;
4841 struct list_head list;
4842 wait_queue_head_t wait;
4843 };
4844
4845 static int flush_space(struct btrfs_root *root,
4846 struct btrfs_space_info *space_info, u64 num_bytes,
4847 u64 orig_bytes, int state)
4848 {
4849 struct btrfs_trans_handle *trans;
4850 int nr;
4851 int ret = 0;
4852
4853 switch (state) {
4854 case FLUSH_DELAYED_ITEMS_NR:
4855 case FLUSH_DELAYED_ITEMS:
4856 if (state == FLUSH_DELAYED_ITEMS_NR)
4857 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4858 else
4859 nr = -1;
4860
4861 trans = btrfs_join_transaction(root);
4862 if (IS_ERR(trans)) {
4863 ret = PTR_ERR(trans);
4864 break;
4865 }
4866 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4867 btrfs_end_transaction(trans, root);
4868 break;
4869 case FLUSH_DELALLOC:
4870 case FLUSH_DELALLOC_WAIT:
4871 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4872 state == FLUSH_DELALLOC_WAIT);
4873 break;
4874 case ALLOC_CHUNK:
4875 trans = btrfs_join_transaction(root);
4876 if (IS_ERR(trans)) {
4877 ret = PTR_ERR(trans);
4878 break;
4879 }
4880 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4881 btrfs_get_alloc_profile(root, 0),
4882 CHUNK_ALLOC_NO_FORCE);
4883 btrfs_end_transaction(trans, root);
4884 if (ret == -ENOSPC)
4885 ret = 0;
4886 break;
4887 case COMMIT_TRANS:
4888 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4889 break;
4890 default:
4891 ret = -ENOSPC;
4892 break;
4893 }
4894
4895 trace_btrfs_flush_space(root->fs_info, space_info->flags, num_bytes,
4896 orig_bytes, state, ret);
4897 return ret;
4898 }
4899
4900 static inline u64
4901 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4902 struct btrfs_space_info *space_info)
4903 {
4904 struct reserve_ticket *ticket;
4905 u64 used;
4906 u64 expected;
4907 u64 to_reclaim = 0;
4908
4909 to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
4910 if (can_overcommit(root, space_info, to_reclaim,
4911 BTRFS_RESERVE_FLUSH_ALL))
4912 return 0;
4913
4914 list_for_each_entry(ticket, &space_info->tickets, list)
4915 to_reclaim += ticket->bytes;
4916 list_for_each_entry(ticket, &space_info->priority_tickets, list)
4917 to_reclaim += ticket->bytes;
4918 if (to_reclaim)
4919 return to_reclaim;
4920
4921 used = space_info->bytes_used + space_info->bytes_reserved +
4922 space_info->bytes_pinned + space_info->bytes_readonly +
4923 space_info->bytes_may_use;
4924 if (can_overcommit(root, space_info, SZ_1M, BTRFS_RESERVE_FLUSH_ALL))
4925 expected = div_factor_fine(space_info->total_bytes, 95);
4926 else
4927 expected = div_factor_fine(space_info->total_bytes, 90);
4928
4929 if (used > expected)
4930 to_reclaim = used - expected;
4931 else
4932 to_reclaim = 0;
4933 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4934 space_info->bytes_reserved);
4935 return to_reclaim;
4936 }
4937
4938 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4939 struct btrfs_root *root, u64 used)
4940 {
4941 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4942
4943 /* If we're just plain full then async reclaim just slows us down. */
4944 if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
4945 return 0;
4946
4947 if (!btrfs_calc_reclaim_metadata_size(root, space_info))
4948 return 0;
4949
4950 return (used >= thresh && !btrfs_fs_closing(root->fs_info) &&
4951 !test_bit(BTRFS_FS_STATE_REMOUNTING,
4952 &root->fs_info->fs_state));
4953 }
4954
4955 static void wake_all_tickets(struct list_head *head)
4956 {
4957 struct reserve_ticket *ticket;
4958
4959 while (!list_empty(head)) {
4960 ticket = list_first_entry(head, struct reserve_ticket, list);
4961 list_del_init(&ticket->list);
4962 ticket->error = -ENOSPC;
4963 wake_up(&ticket->wait);
4964 }
4965 }
4966
4967 /*
4968 * This is for normal flushers, we can wait all goddamned day if we want to. We
4969 * will loop and continuously try to flush as long as we are making progress.
4970 * We count progress as clearing off tickets each time we have to loop.
4971 */
4972 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4973 {
4974 struct reserve_ticket *last_ticket = NULL;
4975 struct btrfs_fs_info *fs_info;
4976 struct btrfs_space_info *space_info;
4977 u64 to_reclaim;
4978 int flush_state;
4979 int commit_cycles = 0;
4980
4981 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4982 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4983
4984 spin_lock(&space_info->lock);
4985 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4986 space_info);
4987 if (!to_reclaim) {
4988 space_info->flush = 0;
4989 spin_unlock(&space_info->lock);
4990 return;
4991 }
4992 last_ticket = list_first_entry(&space_info->tickets,
4993 struct reserve_ticket, list);
4994 spin_unlock(&space_info->lock);
4995
4996 flush_state = FLUSH_DELAYED_ITEMS_NR;
4997 do {
4998 struct reserve_ticket *ticket;
4999 int ret;
5000
5001 ret = flush_space(fs_info->fs_root, space_info, to_reclaim,
5002 to_reclaim, flush_state);
5003 spin_lock(&space_info->lock);
5004 if (list_empty(&space_info->tickets)) {
5005 space_info->flush = 0;
5006 spin_unlock(&space_info->lock);
5007 return;
5008 }
5009 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
5010 space_info);
5011 ticket = list_first_entry(&space_info->tickets,
5012 struct reserve_ticket, list);
5013 if (last_ticket == ticket) {
5014 flush_state++;
5015 } else {
5016 last_ticket = ticket;
5017 flush_state = FLUSH_DELAYED_ITEMS_NR;
5018 if (commit_cycles)
5019 commit_cycles--;
5020 }
5021
5022 if (flush_state > COMMIT_TRANS) {
5023 commit_cycles++;
5024 if (commit_cycles > 2) {
5025 wake_all_tickets(&space_info->tickets);
5026 space_info->flush = 0;
5027 } else {
5028 flush_state = FLUSH_DELAYED_ITEMS_NR;
5029 }
5030 }
5031 spin_unlock(&space_info->lock);
5032 } while (flush_state <= COMMIT_TRANS);
5033 }
5034
5035 void btrfs_init_async_reclaim_work(struct work_struct *work)
5036 {
5037 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
5038 }
5039
5040 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
5041 struct btrfs_space_info *space_info,
5042 struct reserve_ticket *ticket)
5043 {
5044 u64 to_reclaim;
5045 int flush_state = FLUSH_DELAYED_ITEMS_NR;
5046
5047 spin_lock(&space_info->lock);
5048 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
5049 space_info);
5050 if (!to_reclaim) {
5051 spin_unlock(&space_info->lock);
5052 return;
5053 }
5054 spin_unlock(&space_info->lock);
5055
5056 do {
5057 flush_space(fs_info->fs_root, space_info, to_reclaim,
5058 to_reclaim, flush_state);
5059 flush_state++;
5060 spin_lock(&space_info->lock);
5061 if (ticket->bytes == 0) {
5062 spin_unlock(&space_info->lock);
5063 return;
5064 }
5065 spin_unlock(&space_info->lock);
5066
5067 /*
5068 * Priority flushers can't wait on delalloc without
5069 * deadlocking.
5070 */
5071 if (flush_state == FLUSH_DELALLOC ||
5072 flush_state == FLUSH_DELALLOC_WAIT)
5073 flush_state = ALLOC_CHUNK;
5074 } while (flush_state < COMMIT_TRANS);
5075 }
5076
5077 static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
5078 struct btrfs_space_info *space_info,
5079 struct reserve_ticket *ticket, u64 orig_bytes)
5080
5081 {
5082 DEFINE_WAIT(wait);
5083 int ret = 0;
5084
5085 spin_lock(&space_info->lock);
5086 while (ticket->bytes > 0 && ticket->error == 0) {
5087 ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE);
5088 if (ret) {
5089 ret = -EINTR;
5090 break;
5091 }
5092 spin_unlock(&space_info->lock);
5093
5094 schedule();
5095
5096 finish_wait(&ticket->wait, &wait);
5097 spin_lock(&space_info->lock);
5098 }
5099 if (!ret)
5100 ret = ticket->error;
5101 if (!list_empty(&ticket->list))
5102 list_del_init(&ticket->list);
5103 if (ticket->bytes && ticket->bytes < orig_bytes) {
5104 u64 num_bytes = orig_bytes - ticket->bytes;
5105 space_info->bytes_may_use -= num_bytes;
5106 trace_btrfs_space_reservation(fs_info, "space_info",
5107 space_info->flags, num_bytes, 0);
5108 }
5109 spin_unlock(&space_info->lock);
5110
5111 return ret;
5112 }
5113
5114 /**
5115 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5116 * @root - the root we're allocating for
5117 * @space_info - the space info we want to allocate from
5118 * @orig_bytes - the number of bytes we want
5119 * @flush - whether or not we can flush to make our reservation
5120 *
5121 * This will reserve orig_bytes number of bytes from the space info associated
5122 * with the block_rsv. If there is not enough space it will make an attempt to
5123 * flush out space to make room. It will do this by flushing delalloc if
5124 * possible or committing the transaction. If flush is 0 then no attempts to
5125 * regain reservations will be made and this will fail if there is not enough
5126 * space already.
5127 */
5128 static int __reserve_metadata_bytes(struct btrfs_root *root,
5129 struct btrfs_space_info *space_info,
5130 u64 orig_bytes,
5131 enum btrfs_reserve_flush_enum flush)
5132 {
5133 struct reserve_ticket ticket;
5134 u64 used;
5135 int ret = 0;
5136
5137 ASSERT(orig_bytes);
5138 ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL);
5139
5140 spin_lock(&space_info->lock);
5141 ret = -ENOSPC;
5142 used = space_info->bytes_used + space_info->bytes_reserved +
5143 space_info->bytes_pinned + space_info->bytes_readonly +
5144 space_info->bytes_may_use;
5145
5146 /*
5147 * If we have enough space then hooray, make our reservation and carry
5148 * on. If not see if we can overcommit, and if we can, hooray carry on.
5149 * If not things get more complicated.
5150 */
5151 if (used + orig_bytes <= space_info->total_bytes) {
5152 space_info->bytes_may_use += orig_bytes;
5153 trace_btrfs_space_reservation(root->fs_info, "space_info",
5154 space_info->flags, orig_bytes,
5155 1);
5156 ret = 0;
5157 } else if (can_overcommit(root, space_info, orig_bytes, flush)) {
5158 space_info->bytes_may_use += orig_bytes;
5159 trace_btrfs_space_reservation(root->fs_info, "space_info",
5160 space_info->flags, orig_bytes,
5161 1);
5162 ret = 0;
5163 }
5164
5165 /*
5166 * If we couldn't make a reservation then setup our reservation ticket
5167 * and kick the async worker if it's not already running.
5168 *
5169 * If we are a priority flusher then we just need to add our ticket to
5170 * the list and we will do our own flushing further down.
5171 */
5172 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5173 ticket.bytes = orig_bytes;
5174 ticket.error = 0;
5175 init_waitqueue_head(&ticket.wait);
5176 if (flush == BTRFS_RESERVE_FLUSH_ALL) {
5177 list_add_tail(&ticket.list, &space_info->tickets);
5178 if (!space_info->flush) {
5179 space_info->flush = 1;
5180 trace_btrfs_trigger_flush(root->fs_info,
5181 space_info->flags,
5182 orig_bytes, flush,
5183 "enospc");
5184 queue_work(system_unbound_wq,
5185 &root->fs_info->async_reclaim_work);
5186 }
5187 } else {
5188 list_add_tail(&ticket.list,
5189 &space_info->priority_tickets);
5190 }
5191 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5192 used += orig_bytes;
5193 /*
5194 * We will do the space reservation dance during log replay,
5195 * which means we won't have fs_info->fs_root set, so don't do
5196 * the async reclaim as we will panic.
5197 */
5198 if (!root->fs_info->log_root_recovering &&
5199 need_do_async_reclaim(space_info, root, used) &&
5200 !work_busy(&root->fs_info->async_reclaim_work)) {
5201 trace_btrfs_trigger_flush(root->fs_info,
5202 space_info->flags,
5203 orig_bytes, flush,
5204 "preempt");
5205 queue_work(system_unbound_wq,
5206 &root->fs_info->async_reclaim_work);
5207 }
5208 }
5209 spin_unlock(&space_info->lock);
5210 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5211 return ret;
5212
5213 if (flush == BTRFS_RESERVE_FLUSH_ALL)
5214 return wait_reserve_ticket(root->fs_info, space_info, &ticket,
5215 orig_bytes);
5216
5217 ret = 0;
5218 priority_reclaim_metadata_space(root->fs_info, space_info, &ticket);
5219 spin_lock(&space_info->lock);
5220 if (ticket.bytes) {
5221 if (ticket.bytes < orig_bytes) {
5222 u64 num_bytes = orig_bytes - ticket.bytes;
5223 space_info->bytes_may_use -= num_bytes;
5224 trace_btrfs_space_reservation(root->fs_info,
5225 "space_info", space_info->flags,
5226 num_bytes, 0);
5227
5228 }
5229 list_del_init(&ticket.list);
5230 ret = -ENOSPC;
5231 }
5232 spin_unlock(&space_info->lock);
5233 ASSERT(list_empty(&ticket.list));
5234 return ret;
5235 }
5236
5237 /**
5238 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5239 * @root - the root we're allocating for
5240 * @block_rsv - the block_rsv we're allocating for
5241 * @orig_bytes - the number of bytes we want
5242 * @flush - whether or not we can flush to make our reservation
5243 *
5244 * This will reserve orgi_bytes number of bytes from the space info associated
5245 * with the block_rsv. If there is not enough space it will make an attempt to
5246 * flush out space to make room. It will do this by flushing delalloc if
5247 * possible or committing the transaction. If flush is 0 then no attempts to
5248 * regain reservations will be made and this will fail if there is not enough
5249 * space already.
5250 */
5251 static int reserve_metadata_bytes(struct btrfs_root *root,
5252 struct btrfs_block_rsv *block_rsv,
5253 u64 orig_bytes,
5254 enum btrfs_reserve_flush_enum flush)
5255 {
5256 int ret;
5257
5258 ret = __reserve_metadata_bytes(root, block_rsv->space_info, orig_bytes,
5259 flush);
5260 if (ret == -ENOSPC &&
5261 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5262 struct btrfs_block_rsv *global_rsv =
5263 &root->fs_info->global_block_rsv;
5264
5265 if (block_rsv != global_rsv &&
5266 !block_rsv_use_bytes(global_rsv, orig_bytes))
5267 ret = 0;
5268 }
5269 if (ret == -ENOSPC)
5270 trace_btrfs_space_reservation(root->fs_info,
5271 "space_info:enospc",
5272 block_rsv->space_info->flags,
5273 orig_bytes, 1);
5274 return ret;
5275 }
5276
5277 static struct btrfs_block_rsv *get_block_rsv(
5278 const struct btrfs_trans_handle *trans,
5279 const struct btrfs_root *root)
5280 {
5281 struct btrfs_block_rsv *block_rsv = NULL;
5282
5283 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5284 (root == root->fs_info->csum_root && trans->adding_csums) ||
5285 (root == root->fs_info->uuid_root))
5286 block_rsv = trans->block_rsv;
5287
5288 if (!block_rsv)
5289 block_rsv = root->block_rsv;
5290
5291 if (!block_rsv)
5292 block_rsv = &root->fs_info->empty_block_rsv;
5293
5294 return block_rsv;
5295 }
5296
5297 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5298 u64 num_bytes)
5299 {
5300 int ret = -ENOSPC;
5301 spin_lock(&block_rsv->lock);
5302 if (block_rsv->reserved >= num_bytes) {
5303 block_rsv->reserved -= num_bytes;
5304 if (block_rsv->reserved < block_rsv->size)
5305 block_rsv->full = 0;
5306 ret = 0;
5307 }
5308 spin_unlock(&block_rsv->lock);
5309 return ret;
5310 }
5311
5312 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5313 u64 num_bytes, int update_size)
5314 {
5315 spin_lock(&block_rsv->lock);
5316 block_rsv->reserved += num_bytes;
5317 if (update_size)
5318 block_rsv->size += num_bytes;
5319 else if (block_rsv->reserved >= block_rsv->size)
5320 block_rsv->full = 1;
5321 spin_unlock(&block_rsv->lock);
5322 }
5323
5324 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5325 struct btrfs_block_rsv *dest, u64 num_bytes,
5326 int min_factor)
5327 {
5328 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5329 u64 min_bytes;
5330
5331 if (global_rsv->space_info != dest->space_info)
5332 return -ENOSPC;
5333
5334 spin_lock(&global_rsv->lock);
5335 min_bytes = div_factor(global_rsv->size, min_factor);
5336 if (global_rsv->reserved < min_bytes + num_bytes) {
5337 spin_unlock(&global_rsv->lock);
5338 return -ENOSPC;
5339 }
5340 global_rsv->reserved -= num_bytes;
5341 if (global_rsv->reserved < global_rsv->size)
5342 global_rsv->full = 0;
5343 spin_unlock(&global_rsv->lock);
5344
5345 block_rsv_add_bytes(dest, num_bytes, 1);
5346 return 0;
5347 }
5348
5349 /*
5350 * This is for space we already have accounted in space_info->bytes_may_use, so
5351 * basically when we're returning space from block_rsv's.
5352 */
5353 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
5354 struct btrfs_space_info *space_info,
5355 u64 num_bytes)
5356 {
5357 struct reserve_ticket *ticket;
5358 struct list_head *head;
5359 u64 used;
5360 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH;
5361 bool check_overcommit = false;
5362
5363 spin_lock(&space_info->lock);
5364 head = &space_info->priority_tickets;
5365
5366 /*
5367 * If we are over our limit then we need to check and see if we can
5368 * overcommit, and if we can't then we just need to free up our space
5369 * and not satisfy any requests.
5370 */
5371 used = space_info->bytes_used + space_info->bytes_reserved +
5372 space_info->bytes_pinned + space_info->bytes_readonly +
5373 space_info->bytes_may_use;
5374 if (used - num_bytes >= space_info->total_bytes)
5375 check_overcommit = true;
5376 again:
5377 while (!list_empty(head) && num_bytes) {
5378 ticket = list_first_entry(head, struct reserve_ticket,
5379 list);
5380 /*
5381 * We use 0 bytes because this space is already reserved, so
5382 * adding the ticket space would be a double count.
5383 */
5384 if (check_overcommit &&
5385 !can_overcommit(fs_info->extent_root, space_info, 0,
5386 flush))
5387 break;
5388 if (num_bytes >= ticket->bytes) {
5389 list_del_init(&ticket->list);
5390 num_bytes -= ticket->bytes;
5391 ticket->bytes = 0;
5392 wake_up(&ticket->wait);
5393 } else {
5394 ticket->bytes -= num_bytes;
5395 num_bytes = 0;
5396 }
5397 }
5398
5399 if (num_bytes && head == &space_info->priority_tickets) {
5400 head = &space_info->tickets;
5401 flush = BTRFS_RESERVE_FLUSH_ALL;
5402 goto again;
5403 }
5404 space_info->bytes_may_use -= num_bytes;
5405 trace_btrfs_space_reservation(fs_info, "space_info",
5406 space_info->flags, num_bytes, 0);
5407 spin_unlock(&space_info->lock);
5408 }
5409
5410 /*
5411 * This is for newly allocated space that isn't accounted in
5412 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5413 * we use this helper.
5414 */
5415 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
5416 struct btrfs_space_info *space_info,
5417 u64 num_bytes)
5418 {
5419 struct reserve_ticket *ticket;
5420 struct list_head *head = &space_info->priority_tickets;
5421
5422 again:
5423 while (!list_empty(head) && num_bytes) {
5424 ticket = list_first_entry(head, struct reserve_ticket,
5425 list);
5426 if (num_bytes >= ticket->bytes) {
5427 trace_btrfs_space_reservation(fs_info, "space_info",
5428 space_info->flags,
5429 ticket->bytes, 1);
5430 list_del_init(&ticket->list);
5431 num_bytes -= ticket->bytes;
5432 space_info->bytes_may_use += ticket->bytes;
5433 ticket->bytes = 0;
5434 wake_up(&ticket->wait);
5435 } else {
5436 trace_btrfs_space_reservation(fs_info, "space_info",
5437 space_info->flags,
5438 num_bytes, 1);
5439 space_info->bytes_may_use += num_bytes;
5440 ticket->bytes -= num_bytes;
5441 num_bytes = 0;
5442 }
5443 }
5444
5445 if (num_bytes && head == &space_info->priority_tickets) {
5446 head = &space_info->tickets;
5447 goto again;
5448 }
5449 }
5450
5451 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5452 struct btrfs_block_rsv *block_rsv,
5453 struct btrfs_block_rsv *dest, u64 num_bytes)
5454 {
5455 struct btrfs_space_info *space_info = block_rsv->space_info;
5456
5457 spin_lock(&block_rsv->lock);
5458 if (num_bytes == (u64)-1)
5459 num_bytes = block_rsv->size;
5460 block_rsv->size -= num_bytes;
5461 if (block_rsv->reserved >= block_rsv->size) {
5462 num_bytes = block_rsv->reserved - block_rsv->size;
5463 block_rsv->reserved = block_rsv->size;
5464 block_rsv->full = 1;
5465 } else {
5466 num_bytes = 0;
5467 }
5468 spin_unlock(&block_rsv->lock);
5469
5470 if (num_bytes > 0) {
5471 if (dest) {
5472 spin_lock(&dest->lock);
5473 if (!dest->full) {
5474 u64 bytes_to_add;
5475
5476 bytes_to_add = dest->size - dest->reserved;
5477 bytes_to_add = min(num_bytes, bytes_to_add);
5478 dest->reserved += bytes_to_add;
5479 if (dest->reserved >= dest->size)
5480 dest->full = 1;
5481 num_bytes -= bytes_to_add;
5482 }
5483 spin_unlock(&dest->lock);
5484 }
5485 if (num_bytes)
5486 space_info_add_old_bytes(fs_info, space_info,
5487 num_bytes);
5488 }
5489 }
5490
5491 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
5492 struct btrfs_block_rsv *dst, u64 num_bytes,
5493 int update_size)
5494 {
5495 int ret;
5496
5497 ret = block_rsv_use_bytes(src, num_bytes);
5498 if (ret)
5499 return ret;
5500
5501 block_rsv_add_bytes(dst, num_bytes, update_size);
5502 return 0;
5503 }
5504
5505 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5506 {
5507 memset(rsv, 0, sizeof(*rsv));
5508 spin_lock_init(&rsv->lock);
5509 rsv->type = type;
5510 }
5511
5512 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
5513 unsigned short type)
5514 {
5515 struct btrfs_block_rsv *block_rsv;
5516 struct btrfs_fs_info *fs_info = root->fs_info;
5517
5518 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5519 if (!block_rsv)
5520 return NULL;
5521
5522 btrfs_init_block_rsv(block_rsv, type);
5523 block_rsv->space_info = __find_space_info(fs_info,
5524 BTRFS_BLOCK_GROUP_METADATA);
5525 return block_rsv;
5526 }
5527
5528 void btrfs_free_block_rsv(struct btrfs_root *root,
5529 struct btrfs_block_rsv *rsv)
5530 {
5531 if (!rsv)
5532 return;
5533 btrfs_block_rsv_release(root, rsv, (u64)-1);
5534 kfree(rsv);
5535 }
5536
5537 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5538 {
5539 kfree(rsv);
5540 }
5541
5542 int btrfs_block_rsv_add(struct btrfs_root *root,
5543 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5544 enum btrfs_reserve_flush_enum flush)
5545 {
5546 int ret;
5547
5548 if (num_bytes == 0)
5549 return 0;
5550
5551 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5552 if (!ret) {
5553 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5554 return 0;
5555 }
5556
5557 return ret;
5558 }
5559
5560 int btrfs_block_rsv_check(struct btrfs_root *root,
5561 struct btrfs_block_rsv *block_rsv, int min_factor)
5562 {
5563 u64 num_bytes = 0;
5564 int ret = -ENOSPC;
5565
5566 if (!block_rsv)
5567 return 0;
5568
5569 spin_lock(&block_rsv->lock);
5570 num_bytes = div_factor(block_rsv->size, min_factor);
5571 if (block_rsv->reserved >= num_bytes)
5572 ret = 0;
5573 spin_unlock(&block_rsv->lock);
5574
5575 return ret;
5576 }
5577
5578 int btrfs_block_rsv_refill(struct btrfs_root *root,
5579 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5580 enum btrfs_reserve_flush_enum flush)
5581 {
5582 u64 num_bytes = 0;
5583 int ret = -ENOSPC;
5584
5585 if (!block_rsv)
5586 return 0;
5587
5588 spin_lock(&block_rsv->lock);
5589 num_bytes = min_reserved;
5590 if (block_rsv->reserved >= num_bytes)
5591 ret = 0;
5592 else
5593 num_bytes -= block_rsv->reserved;
5594 spin_unlock(&block_rsv->lock);
5595
5596 if (!ret)
5597 return 0;
5598
5599 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5600 if (!ret) {
5601 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5602 return 0;
5603 }
5604
5605 return ret;
5606 }
5607
5608 void btrfs_block_rsv_release(struct btrfs_root *root,
5609 struct btrfs_block_rsv *block_rsv,
5610 u64 num_bytes)
5611 {
5612 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5613 if (global_rsv == block_rsv ||
5614 block_rsv->space_info != global_rsv->space_info)
5615 global_rsv = NULL;
5616 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
5617 num_bytes);
5618 }
5619
5620 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5621 {
5622 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5623 struct btrfs_space_info *sinfo = block_rsv->space_info;
5624 u64 num_bytes;
5625
5626 /*
5627 * The global block rsv is based on the size of the extent tree, the
5628 * checksum tree and the root tree. If the fs is empty we want to set
5629 * it to a minimal amount for safety.
5630 */
5631 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
5632 btrfs_root_used(&fs_info->csum_root->root_item) +
5633 btrfs_root_used(&fs_info->tree_root->root_item);
5634 num_bytes = max_t(u64, num_bytes, SZ_16M);
5635
5636 spin_lock(&sinfo->lock);
5637 spin_lock(&block_rsv->lock);
5638
5639 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5640
5641 if (block_rsv->reserved < block_rsv->size) {
5642 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
5643 sinfo->bytes_reserved + sinfo->bytes_readonly +
5644 sinfo->bytes_may_use;
5645 if (sinfo->total_bytes > num_bytes) {
5646 num_bytes = sinfo->total_bytes - num_bytes;
5647 num_bytes = min(num_bytes,
5648 block_rsv->size - block_rsv->reserved);
5649 block_rsv->reserved += num_bytes;
5650 sinfo->bytes_may_use += num_bytes;
5651 trace_btrfs_space_reservation(fs_info, "space_info",
5652 sinfo->flags, num_bytes,
5653 1);
5654 }
5655 } else if (block_rsv->reserved > block_rsv->size) {
5656 num_bytes = block_rsv->reserved - block_rsv->size;
5657 sinfo->bytes_may_use -= num_bytes;
5658 trace_btrfs_space_reservation(fs_info, "space_info",
5659 sinfo->flags, num_bytes, 0);
5660 block_rsv->reserved = block_rsv->size;
5661 }
5662
5663 if (block_rsv->reserved == block_rsv->size)
5664 block_rsv->full = 1;
5665 else
5666 block_rsv->full = 0;
5667
5668 spin_unlock(&block_rsv->lock);
5669 spin_unlock(&sinfo->lock);
5670 }
5671
5672 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5673 {
5674 struct btrfs_space_info *space_info;
5675
5676 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5677 fs_info->chunk_block_rsv.space_info = space_info;
5678
5679 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5680 fs_info->global_block_rsv.space_info = space_info;
5681 fs_info->delalloc_block_rsv.space_info = space_info;
5682 fs_info->trans_block_rsv.space_info = space_info;
5683 fs_info->empty_block_rsv.space_info = space_info;
5684 fs_info->delayed_block_rsv.space_info = space_info;
5685
5686 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5687 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5688 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5689 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5690 if (fs_info->quota_root)
5691 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5692 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5693
5694 update_global_block_rsv(fs_info);
5695 }
5696
5697 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5698 {
5699 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5700 (u64)-1);
5701 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5702 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5703 WARN_ON(fs_info->trans_block_rsv.size > 0);
5704 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5705 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5706 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5707 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5708 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5709 }
5710
5711 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5712 struct btrfs_root *root)
5713 {
5714 if (!trans->block_rsv)
5715 return;
5716
5717 if (!trans->bytes_reserved)
5718 return;
5719
5720 trace_btrfs_space_reservation(root->fs_info, "transaction",
5721 trans->transid, trans->bytes_reserved, 0);
5722 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
5723 trans->bytes_reserved = 0;
5724 }
5725
5726 /*
5727 * To be called after all the new block groups attached to the transaction
5728 * handle have been created (btrfs_create_pending_block_groups()).
5729 */
5730 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5731 {
5732 struct btrfs_fs_info *fs_info = trans->root->fs_info;
5733
5734 if (!trans->chunk_bytes_reserved)
5735 return;
5736
5737 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5738
5739 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5740 trans->chunk_bytes_reserved);
5741 trans->chunk_bytes_reserved = 0;
5742 }
5743
5744 /* Can only return 0 or -ENOSPC */
5745 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5746 struct inode *inode)
5747 {
5748 struct btrfs_root *root = BTRFS_I(inode)->root;
5749 /*
5750 * We always use trans->block_rsv here as we will have reserved space
5751 * for our orphan when starting the transaction, using get_block_rsv()
5752 * here will sometimes make us choose the wrong block rsv as we could be
5753 * doing a reloc inode for a non refcounted root.
5754 */
5755 struct btrfs_block_rsv *src_rsv = trans->block_rsv;
5756 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5757
5758 /*
5759 * We need to hold space in order to delete our orphan item once we've
5760 * added it, so this takes the reservation so we can release it later
5761 * when we are truly done with the orphan item.
5762 */
5763 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5764 trace_btrfs_space_reservation(root->fs_info, "orphan",
5765 btrfs_ino(inode), num_bytes, 1);
5766 return btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
5767 }
5768
5769 void btrfs_orphan_release_metadata(struct inode *inode)
5770 {
5771 struct btrfs_root *root = BTRFS_I(inode)->root;
5772 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5773 trace_btrfs_space_reservation(root->fs_info, "orphan",
5774 btrfs_ino(inode), num_bytes, 0);
5775 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
5776 }
5777
5778 /*
5779 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5780 * root: the root of the parent directory
5781 * rsv: block reservation
5782 * items: the number of items that we need do reservation
5783 * qgroup_reserved: used to return the reserved size in qgroup
5784 *
5785 * This function is used to reserve the space for snapshot/subvolume
5786 * creation and deletion. Those operations are different with the
5787 * common file/directory operations, they change two fs/file trees
5788 * and root tree, the number of items that the qgroup reserves is
5789 * different with the free space reservation. So we can not use
5790 * the space reservation mechanism in start_transaction().
5791 */
5792 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5793 struct btrfs_block_rsv *rsv,
5794 int items,
5795 u64 *qgroup_reserved,
5796 bool use_global_rsv)
5797 {
5798 u64 num_bytes;
5799 int ret;
5800 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5801
5802 if (root->fs_info->quota_enabled) {
5803 /* One for parent inode, two for dir entries */
5804 num_bytes = 3 * root->nodesize;
5805 ret = btrfs_qgroup_reserve_meta(root, num_bytes);
5806 if (ret)
5807 return ret;
5808 } else {
5809 num_bytes = 0;
5810 }
5811
5812 *qgroup_reserved = num_bytes;
5813
5814 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5815 rsv->space_info = __find_space_info(root->fs_info,
5816 BTRFS_BLOCK_GROUP_METADATA);
5817 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5818 BTRFS_RESERVE_FLUSH_ALL);
5819
5820 if (ret == -ENOSPC && use_global_rsv)
5821 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1);
5822
5823 if (ret && *qgroup_reserved)
5824 btrfs_qgroup_free_meta(root, *qgroup_reserved);
5825
5826 return ret;
5827 }
5828
5829 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5830 struct btrfs_block_rsv *rsv,
5831 u64 qgroup_reserved)
5832 {
5833 btrfs_block_rsv_release(root, rsv, (u64)-1);
5834 }
5835
5836 /**
5837 * drop_outstanding_extent - drop an outstanding extent
5838 * @inode: the inode we're dropping the extent for
5839 * @num_bytes: the number of bytes we're releasing.
5840 *
5841 * This is called when we are freeing up an outstanding extent, either called
5842 * after an error or after an extent is written. This will return the number of
5843 * reserved extents that need to be freed. This must be called with
5844 * BTRFS_I(inode)->lock held.
5845 */
5846 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5847 {
5848 unsigned drop_inode_space = 0;
5849 unsigned dropped_extents = 0;
5850 unsigned num_extents = 0;
5851
5852 num_extents = (unsigned)div64_u64(num_bytes +
5853 BTRFS_MAX_EXTENT_SIZE - 1,
5854 BTRFS_MAX_EXTENT_SIZE);
5855 ASSERT(num_extents);
5856 ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5857 BTRFS_I(inode)->outstanding_extents -= num_extents;
5858
5859 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5860 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5861 &BTRFS_I(inode)->runtime_flags))
5862 drop_inode_space = 1;
5863
5864 /*
5865 * If we have more or the same amount of outstanding extents than we have
5866 * reserved then we need to leave the reserved extents count alone.
5867 */
5868 if (BTRFS_I(inode)->outstanding_extents >=
5869 BTRFS_I(inode)->reserved_extents)
5870 return drop_inode_space;
5871
5872 dropped_extents = BTRFS_I(inode)->reserved_extents -
5873 BTRFS_I(inode)->outstanding_extents;
5874 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5875 return dropped_extents + drop_inode_space;
5876 }
5877
5878 /**
5879 * calc_csum_metadata_size - return the amount of metadata space that must be
5880 * reserved/freed for the given bytes.
5881 * @inode: the inode we're manipulating
5882 * @num_bytes: the number of bytes in question
5883 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5884 *
5885 * This adjusts the number of csum_bytes in the inode and then returns the
5886 * correct amount of metadata that must either be reserved or freed. We
5887 * calculate how many checksums we can fit into one leaf and then divide the
5888 * number of bytes that will need to be checksumed by this value to figure out
5889 * how many checksums will be required. If we are adding bytes then the number
5890 * may go up and we will return the number of additional bytes that must be
5891 * reserved. If it is going down we will return the number of bytes that must
5892 * be freed.
5893 *
5894 * This must be called with BTRFS_I(inode)->lock held.
5895 */
5896 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5897 int reserve)
5898 {
5899 struct btrfs_root *root = BTRFS_I(inode)->root;
5900 u64 old_csums, num_csums;
5901
5902 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5903 BTRFS_I(inode)->csum_bytes == 0)
5904 return 0;
5905
5906 old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5907 if (reserve)
5908 BTRFS_I(inode)->csum_bytes += num_bytes;
5909 else
5910 BTRFS_I(inode)->csum_bytes -= num_bytes;
5911 num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5912
5913 /* No change, no need to reserve more */
5914 if (old_csums == num_csums)
5915 return 0;
5916
5917 if (reserve)
5918 return btrfs_calc_trans_metadata_size(root,
5919 num_csums - old_csums);
5920
5921 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5922 }
5923
5924 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5925 {
5926 struct btrfs_root *root = BTRFS_I(inode)->root;
5927 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5928 u64 to_reserve = 0;
5929 u64 csum_bytes;
5930 unsigned nr_extents = 0;
5931 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5932 int ret = 0;
5933 bool delalloc_lock = true;
5934 u64 to_free = 0;
5935 unsigned dropped;
5936 bool release_extra = false;
5937
5938 /* If we are a free space inode we need to not flush since we will be in
5939 * the middle of a transaction commit. We also don't need the delalloc
5940 * mutex since we won't race with anybody. We need this mostly to make
5941 * lockdep shut its filthy mouth.
5942 *
5943 * If we have a transaction open (can happen if we call truncate_block
5944 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
5945 */
5946 if (btrfs_is_free_space_inode(inode)) {
5947 flush = BTRFS_RESERVE_NO_FLUSH;
5948 delalloc_lock = false;
5949 } else if (current->journal_info) {
5950 flush = BTRFS_RESERVE_FLUSH_LIMIT;
5951 }
5952
5953 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5954 btrfs_transaction_in_commit(root->fs_info))
5955 schedule_timeout(1);
5956
5957 if (delalloc_lock)
5958 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5959
5960 num_bytes = ALIGN(num_bytes, root->sectorsize);
5961
5962 spin_lock(&BTRFS_I(inode)->lock);
5963 nr_extents = (unsigned)div64_u64(num_bytes +
5964 BTRFS_MAX_EXTENT_SIZE - 1,
5965 BTRFS_MAX_EXTENT_SIZE);
5966 BTRFS_I(inode)->outstanding_extents += nr_extents;
5967
5968 nr_extents = 0;
5969 if (BTRFS_I(inode)->outstanding_extents >
5970 BTRFS_I(inode)->reserved_extents)
5971 nr_extents += BTRFS_I(inode)->outstanding_extents -
5972 BTRFS_I(inode)->reserved_extents;
5973
5974 /* We always want to reserve a slot for updating the inode. */
5975 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents + 1);
5976 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5977 csum_bytes = BTRFS_I(inode)->csum_bytes;
5978 spin_unlock(&BTRFS_I(inode)->lock);
5979
5980 if (root->fs_info->quota_enabled) {
5981 ret = btrfs_qgroup_reserve_meta(root,
5982 nr_extents * root->nodesize);
5983 if (ret)
5984 goto out_fail;
5985 }
5986
5987 ret = btrfs_block_rsv_add(root, block_rsv, to_reserve, flush);
5988 if (unlikely(ret)) {
5989 btrfs_qgroup_free_meta(root, nr_extents * root->nodesize);
5990 goto out_fail;
5991 }
5992
5993 spin_lock(&BTRFS_I(inode)->lock);
5994 if (test_and_set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5995 &BTRFS_I(inode)->runtime_flags)) {
5996 to_reserve -= btrfs_calc_trans_metadata_size(root, 1);
5997 release_extra = true;
5998 }
5999 BTRFS_I(inode)->reserved_extents += nr_extents;
6000 spin_unlock(&BTRFS_I(inode)->lock);
6001
6002 if (delalloc_lock)
6003 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
6004
6005 if (to_reserve)
6006 trace_btrfs_space_reservation(root->fs_info, "delalloc",
6007 btrfs_ino(inode), to_reserve, 1);
6008 if (release_extra)
6009 btrfs_block_rsv_release(root, block_rsv,
6010 btrfs_calc_trans_metadata_size(root,
6011 1));
6012 return 0;
6013
6014 out_fail:
6015 spin_lock(&BTRFS_I(inode)->lock);
6016 dropped = drop_outstanding_extent(inode, num_bytes);
6017 /*
6018 * If the inodes csum_bytes is the same as the original
6019 * csum_bytes then we know we haven't raced with any free()ers
6020 * so we can just reduce our inodes csum bytes and carry on.
6021 */
6022 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
6023 calc_csum_metadata_size(inode, num_bytes, 0);
6024 } else {
6025 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
6026 u64 bytes;
6027
6028 /*
6029 * This is tricky, but first we need to figure out how much we
6030 * freed from any free-ers that occurred during this
6031 * reservation, so we reset ->csum_bytes to the csum_bytes
6032 * before we dropped our lock, and then call the free for the
6033 * number of bytes that were freed while we were trying our
6034 * reservation.
6035 */
6036 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
6037 BTRFS_I(inode)->csum_bytes = csum_bytes;
6038 to_free = calc_csum_metadata_size(inode, bytes, 0);
6039
6040
6041 /*
6042 * Now we need to see how much we would have freed had we not
6043 * been making this reservation and our ->csum_bytes were not
6044 * artificially inflated.
6045 */
6046 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
6047 bytes = csum_bytes - orig_csum_bytes;
6048 bytes = calc_csum_metadata_size(inode, bytes, 0);
6049
6050 /*
6051 * Now reset ->csum_bytes to what it should be. If bytes is
6052 * more than to_free then we would have freed more space had we
6053 * not had an artificially high ->csum_bytes, so we need to free
6054 * the remainder. If bytes is the same or less then we don't
6055 * need to do anything, the other free-ers did the correct
6056 * thing.
6057 */
6058 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
6059 if (bytes > to_free)
6060 to_free = bytes - to_free;
6061 else
6062 to_free = 0;
6063 }
6064 spin_unlock(&BTRFS_I(inode)->lock);
6065 if (dropped)
6066 to_free += btrfs_calc_trans_metadata_size(root, dropped);
6067
6068 if (to_free) {
6069 btrfs_block_rsv_release(root, block_rsv, to_free);
6070 trace_btrfs_space_reservation(root->fs_info, "delalloc",
6071 btrfs_ino(inode), to_free, 0);
6072 }
6073 if (delalloc_lock)
6074 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
6075 return ret;
6076 }
6077
6078 /**
6079 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6080 * @inode: the inode to release the reservation for
6081 * @num_bytes: the number of bytes we're releasing
6082 *
6083 * This will release the metadata reservation for an inode. This can be called
6084 * once we complete IO for a given set of bytes to release their metadata
6085 * reservations.
6086 */
6087 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
6088 {
6089 struct btrfs_root *root = BTRFS_I(inode)->root;
6090 u64 to_free = 0;
6091 unsigned dropped;
6092
6093 num_bytes = ALIGN(num_bytes, root->sectorsize);
6094 spin_lock(&BTRFS_I(inode)->lock);
6095 dropped = drop_outstanding_extent(inode, num_bytes);
6096
6097 if (num_bytes)
6098 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
6099 spin_unlock(&BTRFS_I(inode)->lock);
6100 if (dropped > 0)
6101 to_free += btrfs_calc_trans_metadata_size(root, dropped);
6102
6103 if (btrfs_test_is_dummy_root(root))
6104 return;
6105
6106 trace_btrfs_space_reservation(root->fs_info, "delalloc",
6107 btrfs_ino(inode), to_free, 0);
6108
6109 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
6110 to_free);
6111 }
6112
6113 /**
6114 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6115 * delalloc
6116 * @inode: inode we're writing to
6117 * @start: start range we are writing to
6118 * @len: how long the range we are writing to
6119 *
6120 * TODO: This function will finally replace old btrfs_delalloc_reserve_space()
6121 *
6122 * This will do the following things
6123 *
6124 * o reserve space in data space info for num bytes
6125 * and reserve precious corresponding qgroup space
6126 * (Done in check_data_free_space)
6127 *
6128 * o reserve space for metadata space, based on the number of outstanding
6129 * extents and how much csums will be needed
6130 * also reserve metadata space in a per root over-reserve method.
6131 * o add to the inodes->delalloc_bytes
6132 * o add it to the fs_info's delalloc inodes list.
6133 * (Above 3 all done in delalloc_reserve_metadata)
6134 *
6135 * Return 0 for success
6136 * Return <0 for error(-ENOSPC or -EQUOT)
6137 */
6138 int btrfs_delalloc_reserve_space(struct inode *inode, u64 start, u64 len)
6139 {
6140 int ret;
6141
6142 ret = btrfs_check_data_free_space(inode, start, len);
6143 if (ret < 0)
6144 return ret;
6145 ret = btrfs_delalloc_reserve_metadata(inode, len);
6146 if (ret < 0)
6147 btrfs_free_reserved_data_space(inode, start, len);
6148 return ret;
6149 }
6150
6151 /**
6152 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6153 * @inode: inode we're releasing space for
6154 * @start: start position of the space already reserved
6155 * @len: the len of the space already reserved
6156 *
6157 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
6158 * called in the case that we don't need the metadata AND data reservations
6159 * anymore. So if there is an error or we insert an inline extent.
6160 *
6161 * This function will release the metadata space that was not used and will
6162 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6163 * list if there are no delalloc bytes left.
6164 * Also it will handle the qgroup reserved space.
6165 */
6166 void btrfs_delalloc_release_space(struct inode *inode, u64 start, u64 len)
6167 {
6168 btrfs_delalloc_release_metadata(inode, len);
6169 btrfs_free_reserved_data_space(inode, start, len);
6170 }
6171
6172 static int update_block_group(struct btrfs_trans_handle *trans,
6173 struct btrfs_root *root, u64 bytenr,
6174 u64 num_bytes, int alloc)
6175 {
6176 struct btrfs_block_group_cache *cache = NULL;
6177 struct btrfs_fs_info *info = root->fs_info;
6178 u64 total = num_bytes;
6179 u64 old_val;
6180 u64 byte_in_group;
6181 int factor;
6182
6183 /* block accounting for super block */
6184 spin_lock(&info->delalloc_root_lock);
6185 old_val = btrfs_super_bytes_used(info->super_copy);
6186 if (alloc)
6187 old_val += num_bytes;
6188 else
6189 old_val -= num_bytes;
6190 btrfs_set_super_bytes_used(info->super_copy, old_val);
6191 spin_unlock(&info->delalloc_root_lock);
6192
6193 while (total) {
6194 cache = btrfs_lookup_block_group(info, bytenr);
6195 if (!cache)
6196 return -ENOENT;
6197 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
6198 BTRFS_BLOCK_GROUP_RAID1 |
6199 BTRFS_BLOCK_GROUP_RAID10))
6200 factor = 2;
6201 else
6202 factor = 1;
6203 /*
6204 * If this block group has free space cache written out, we
6205 * need to make sure to load it if we are removing space. This
6206 * is because we need the unpinning stage to actually add the
6207 * space back to the block group, otherwise we will leak space.
6208 */
6209 if (!alloc && cache->cached == BTRFS_CACHE_NO)
6210 cache_block_group(cache, 1);
6211
6212 byte_in_group = bytenr - cache->key.objectid;
6213 WARN_ON(byte_in_group > cache->key.offset);
6214
6215 spin_lock(&cache->space_info->lock);
6216 spin_lock(&cache->lock);
6217
6218 if (btrfs_test_opt(root, SPACE_CACHE) &&
6219 cache->disk_cache_state < BTRFS_DC_CLEAR)
6220 cache->disk_cache_state = BTRFS_DC_CLEAR;
6221
6222 old_val = btrfs_block_group_used(&cache->item);
6223 num_bytes = min(total, cache->key.offset - byte_in_group);
6224 if (alloc) {
6225 old_val += num_bytes;
6226 btrfs_set_block_group_used(&cache->item, old_val);
6227 cache->reserved -= num_bytes;
6228 cache->space_info->bytes_reserved -= num_bytes;
6229 cache->space_info->bytes_used += num_bytes;
6230 cache->space_info->disk_used += num_bytes * factor;
6231 spin_unlock(&cache->lock);
6232 spin_unlock(&cache->space_info->lock);
6233 } else {
6234 old_val -= num_bytes;
6235 btrfs_set_block_group_used(&cache->item, old_val);
6236 cache->pinned += num_bytes;
6237 cache->space_info->bytes_pinned += num_bytes;
6238 cache->space_info->bytes_used -= num_bytes;
6239 cache->space_info->disk_used -= num_bytes * factor;
6240 spin_unlock(&cache->lock);
6241 spin_unlock(&cache->space_info->lock);
6242
6243 trace_btrfs_space_reservation(root->fs_info, "pinned",
6244 cache->space_info->flags,
6245 num_bytes, 1);
6246 set_extent_dirty(info->pinned_extents,
6247 bytenr, bytenr + num_bytes - 1,
6248 GFP_NOFS | __GFP_NOFAIL);
6249 }
6250
6251 spin_lock(&trans->transaction->dirty_bgs_lock);
6252 if (list_empty(&cache->dirty_list)) {
6253 list_add_tail(&cache->dirty_list,
6254 &trans->transaction->dirty_bgs);
6255 trans->transaction->num_dirty_bgs++;
6256 btrfs_get_block_group(cache);
6257 }
6258 spin_unlock(&trans->transaction->dirty_bgs_lock);
6259
6260 /*
6261 * No longer have used bytes in this block group, queue it for
6262 * deletion. We do this after adding the block group to the
6263 * dirty list to avoid races between cleaner kthread and space
6264 * cache writeout.
6265 */
6266 if (!alloc && old_val == 0) {
6267 spin_lock(&info->unused_bgs_lock);
6268 if (list_empty(&cache->bg_list)) {
6269 btrfs_get_block_group(cache);
6270 list_add_tail(&cache->bg_list,
6271 &info->unused_bgs);
6272 }
6273 spin_unlock(&info->unused_bgs_lock);
6274 }
6275
6276 btrfs_put_block_group(cache);
6277 total -= num_bytes;
6278 bytenr += num_bytes;
6279 }
6280 return 0;
6281 }
6282
6283 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
6284 {
6285 struct btrfs_block_group_cache *cache;
6286 u64 bytenr;
6287
6288 spin_lock(&root->fs_info->block_group_cache_lock);
6289 bytenr = root->fs_info->first_logical_byte;
6290 spin_unlock(&root->fs_info->block_group_cache_lock);
6291
6292 if (bytenr < (u64)-1)
6293 return bytenr;
6294
6295 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
6296 if (!cache)
6297 return 0;
6298
6299 bytenr = cache->key.objectid;
6300 btrfs_put_block_group(cache);
6301
6302 return bytenr;
6303 }
6304
6305 static int pin_down_extent(struct btrfs_root *root,
6306 struct btrfs_block_group_cache *cache,
6307 u64 bytenr, u64 num_bytes, int reserved)
6308 {
6309 spin_lock(&cache->space_info->lock);
6310 spin_lock(&cache->lock);
6311 cache->pinned += num_bytes;
6312 cache->space_info->bytes_pinned += num_bytes;
6313 if (reserved) {
6314 cache->reserved -= num_bytes;
6315 cache->space_info->bytes_reserved -= num_bytes;
6316 }
6317 spin_unlock(&cache->lock);
6318 spin_unlock(&cache->space_info->lock);
6319
6320 trace_btrfs_space_reservation(root->fs_info, "pinned",
6321 cache->space_info->flags, num_bytes, 1);
6322 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
6323 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6324 return 0;
6325 }
6326
6327 /*
6328 * this function must be called within transaction
6329 */
6330 int btrfs_pin_extent(struct btrfs_root *root,
6331 u64 bytenr, u64 num_bytes, int reserved)
6332 {
6333 struct btrfs_block_group_cache *cache;
6334
6335 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
6336 BUG_ON(!cache); /* Logic error */
6337
6338 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
6339
6340 btrfs_put_block_group(cache);
6341 return 0;
6342 }
6343
6344 /*
6345 * this function must be called within transaction
6346 */
6347 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
6348 u64 bytenr, u64 num_bytes)
6349 {
6350 struct btrfs_block_group_cache *cache;
6351 int ret;
6352
6353 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
6354 if (!cache)
6355 return -EINVAL;
6356
6357 /*
6358 * pull in the free space cache (if any) so that our pin
6359 * removes the free space from the cache. We have load_only set
6360 * to one because the slow code to read in the free extents does check
6361 * the pinned extents.
6362 */
6363 cache_block_group(cache, 1);
6364
6365 pin_down_extent(root, cache, bytenr, num_bytes, 0);
6366
6367 /* remove us from the free space cache (if we're there at all) */
6368 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6369 btrfs_put_block_group(cache);
6370 return ret;
6371 }
6372
6373 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
6374 {
6375 int ret;
6376 struct btrfs_block_group_cache *block_group;
6377 struct btrfs_caching_control *caching_ctl;
6378
6379 block_group = btrfs_lookup_block_group(root->fs_info, start);
6380 if (!block_group)
6381 return -EINVAL;
6382
6383 cache_block_group(block_group, 0);
6384 caching_ctl = get_caching_control(block_group);
6385
6386 if (!caching_ctl) {
6387 /* Logic error */
6388 BUG_ON(!block_group_cache_done(block_group));
6389 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6390 } else {
6391 mutex_lock(&caching_ctl->mutex);
6392
6393 if (start >= caching_ctl->progress) {
6394 ret = add_excluded_extent(root, start, num_bytes);
6395 } else if (start + num_bytes <= caching_ctl->progress) {
6396 ret = btrfs_remove_free_space(block_group,
6397 start, num_bytes);
6398 } else {
6399 num_bytes = caching_ctl->progress - start;
6400 ret = btrfs_remove_free_space(block_group,
6401 start, num_bytes);
6402 if (ret)
6403 goto out_lock;
6404
6405 num_bytes = (start + num_bytes) -
6406 caching_ctl->progress;
6407 start = caching_ctl->progress;
6408 ret = add_excluded_extent(root, start, num_bytes);
6409 }
6410 out_lock:
6411 mutex_unlock(&caching_ctl->mutex);
6412 put_caching_control(caching_ctl);
6413 }
6414 btrfs_put_block_group(block_group);
6415 return ret;
6416 }
6417
6418 int btrfs_exclude_logged_extents(struct btrfs_root *log,
6419 struct extent_buffer *eb)
6420 {
6421 struct btrfs_file_extent_item *item;
6422 struct btrfs_key key;
6423 int found_type;
6424 int i;
6425
6426 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
6427 return 0;
6428
6429 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6430 btrfs_item_key_to_cpu(eb, &key, i);
6431 if (key.type != BTRFS_EXTENT_DATA_KEY)
6432 continue;
6433 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6434 found_type = btrfs_file_extent_type(eb, item);
6435 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6436 continue;
6437 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6438 continue;
6439 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6440 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6441 __exclude_logged_extent(log, key.objectid, key.offset);
6442 }
6443
6444 return 0;
6445 }
6446
6447 static void
6448 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
6449 {
6450 atomic_inc(&bg->reservations);
6451 }
6452
6453 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
6454 const u64 start)
6455 {
6456 struct btrfs_block_group_cache *bg;
6457
6458 bg = btrfs_lookup_block_group(fs_info, start);
6459 ASSERT(bg);
6460 if (atomic_dec_and_test(&bg->reservations))
6461 wake_up_atomic_t(&bg->reservations);
6462 btrfs_put_block_group(bg);
6463 }
6464
6465 static int btrfs_wait_bg_reservations_atomic_t(atomic_t *a)
6466 {
6467 schedule();
6468 return 0;
6469 }
6470
6471 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg)
6472 {
6473 struct btrfs_space_info *space_info = bg->space_info;
6474
6475 ASSERT(bg->ro);
6476
6477 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
6478 return;
6479
6480 /*
6481 * Our block group is read only but before we set it to read only,
6482 * some task might have had allocated an extent from it already, but it
6483 * has not yet created a respective ordered extent (and added it to a
6484 * root's list of ordered extents).
6485 * Therefore wait for any task currently allocating extents, since the
6486 * block group's reservations counter is incremented while a read lock
6487 * on the groups' semaphore is held and decremented after releasing
6488 * the read access on that semaphore and creating the ordered extent.
6489 */
6490 down_write(&space_info->groups_sem);
6491 up_write(&space_info->groups_sem);
6492
6493 wait_on_atomic_t(&bg->reservations,
6494 btrfs_wait_bg_reservations_atomic_t,
6495 TASK_UNINTERRUPTIBLE);
6496 }
6497
6498 /**
6499 * btrfs_update_reserved_bytes - update the block_group and space info counters
6500 * @cache: The cache we are manipulating
6501 * @num_bytes: The number of bytes in question
6502 * @reserve: One of the reservation enums
6503 * @delalloc: The blocks are allocated for the delalloc write
6504 *
6505 * This is called by the allocator when it reserves space, or by somebody who is
6506 * freeing space that was never actually used on disk. For example if you
6507 * reserve some space for a new leaf in transaction A and before transaction A
6508 * commits you free that leaf, you call this with reserve set to 0 in order to
6509 * clear the reservation.
6510 *
6511 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
6512 * ENOSPC accounting. For data we handle the reservation through clearing the
6513 * delalloc bits in the io_tree. We have to do this since we could end up
6514 * allocating less disk space for the amount of data we have reserved in the
6515 * case of compression.
6516 *
6517 * If this is a reservation and the block group has become read only we cannot
6518 * make the reservation and return -EAGAIN, otherwise this function always
6519 * succeeds.
6520 */
6521 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
6522 u64 num_bytes, int reserve, int delalloc)
6523 {
6524 struct btrfs_space_info *space_info = cache->space_info;
6525 int ret = 0;
6526
6527 spin_lock(&space_info->lock);
6528 spin_lock(&cache->lock);
6529 if (reserve != RESERVE_FREE) {
6530 if (cache->ro) {
6531 ret = -EAGAIN;
6532 } else {
6533 cache->reserved += num_bytes;
6534 space_info->bytes_reserved += num_bytes;
6535 if (reserve == RESERVE_ALLOC) {
6536 trace_btrfs_space_reservation(cache->fs_info,
6537 "space_info", space_info->flags,
6538 num_bytes, 0);
6539 space_info->bytes_may_use -= num_bytes;
6540 }
6541
6542 if (delalloc)
6543 cache->delalloc_bytes += num_bytes;
6544 }
6545 } else {
6546 if (cache->ro)
6547 space_info->bytes_readonly += num_bytes;
6548 cache->reserved -= num_bytes;
6549 space_info->bytes_reserved -= num_bytes;
6550
6551 if (delalloc)
6552 cache->delalloc_bytes -= num_bytes;
6553 }
6554 spin_unlock(&cache->lock);
6555 spin_unlock(&space_info->lock);
6556 return ret;
6557 }
6558
6559 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
6560 struct btrfs_root *root)
6561 {
6562 struct btrfs_fs_info *fs_info = root->fs_info;
6563 struct btrfs_caching_control *next;
6564 struct btrfs_caching_control *caching_ctl;
6565 struct btrfs_block_group_cache *cache;
6566
6567 down_write(&fs_info->commit_root_sem);
6568
6569 list_for_each_entry_safe(caching_ctl, next,
6570 &fs_info->caching_block_groups, list) {
6571 cache = caching_ctl->block_group;
6572 if (block_group_cache_done(cache)) {
6573 cache->last_byte_to_unpin = (u64)-1;
6574 list_del_init(&caching_ctl->list);
6575 put_caching_control(caching_ctl);
6576 } else {
6577 cache->last_byte_to_unpin = caching_ctl->progress;
6578 }
6579 }
6580
6581 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6582 fs_info->pinned_extents = &fs_info->freed_extents[1];
6583 else
6584 fs_info->pinned_extents = &fs_info->freed_extents[0];
6585
6586 up_write(&fs_info->commit_root_sem);
6587
6588 update_global_block_rsv(fs_info);
6589 }
6590
6591 /*
6592 * Returns the free cluster for the given space info and sets empty_cluster to
6593 * what it should be based on the mount options.
6594 */
6595 static struct btrfs_free_cluster *
6596 fetch_cluster_info(struct btrfs_root *root, struct btrfs_space_info *space_info,
6597 u64 *empty_cluster)
6598 {
6599 struct btrfs_free_cluster *ret = NULL;
6600 bool ssd = btrfs_test_opt(root, SSD);
6601
6602 *empty_cluster = 0;
6603 if (btrfs_mixed_space_info(space_info))
6604 return ret;
6605
6606 if (ssd)
6607 *empty_cluster = SZ_2M;
6608 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6609 ret = &root->fs_info->meta_alloc_cluster;
6610 if (!ssd)
6611 *empty_cluster = SZ_64K;
6612 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && ssd) {
6613 ret = &root->fs_info->data_alloc_cluster;
6614 }
6615
6616 return ret;
6617 }
6618
6619 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
6620 const bool return_free_space)
6621 {
6622 struct btrfs_fs_info *fs_info = root->fs_info;
6623 struct btrfs_block_group_cache *cache = NULL;
6624 struct btrfs_space_info *space_info;
6625 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6626 struct btrfs_free_cluster *cluster = NULL;
6627 u64 len;
6628 u64 total_unpinned = 0;
6629 u64 empty_cluster = 0;
6630 bool readonly;
6631
6632 while (start <= end) {
6633 readonly = false;
6634 if (!cache ||
6635 start >= cache->key.objectid + cache->key.offset) {
6636 if (cache)
6637 btrfs_put_block_group(cache);
6638 total_unpinned = 0;
6639 cache = btrfs_lookup_block_group(fs_info, start);
6640 BUG_ON(!cache); /* Logic error */
6641
6642 cluster = fetch_cluster_info(root,
6643 cache->space_info,
6644 &empty_cluster);
6645 empty_cluster <<= 1;
6646 }
6647
6648 len = cache->key.objectid + cache->key.offset - start;
6649 len = min(len, end + 1 - start);
6650
6651 if (start < cache->last_byte_to_unpin) {
6652 len = min(len, cache->last_byte_to_unpin - start);
6653 if (return_free_space)
6654 btrfs_add_free_space(cache, start, len);
6655 }
6656
6657 start += len;
6658 total_unpinned += len;
6659 space_info = cache->space_info;
6660
6661 /*
6662 * If this space cluster has been marked as fragmented and we've
6663 * unpinned enough in this block group to potentially allow a
6664 * cluster to be created inside of it go ahead and clear the
6665 * fragmented check.
6666 */
6667 if (cluster && cluster->fragmented &&
6668 total_unpinned > empty_cluster) {
6669 spin_lock(&cluster->lock);
6670 cluster->fragmented = 0;
6671 spin_unlock(&cluster->lock);
6672 }
6673
6674 spin_lock(&space_info->lock);
6675 spin_lock(&cache->lock);
6676 cache->pinned -= len;
6677 space_info->bytes_pinned -= len;
6678
6679 trace_btrfs_space_reservation(fs_info, "pinned",
6680 space_info->flags, len, 0);
6681 space_info->max_extent_size = 0;
6682 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6683 if (cache->ro) {
6684 space_info->bytes_readonly += len;
6685 readonly = true;
6686 }
6687 spin_unlock(&cache->lock);
6688 if (!readonly && return_free_space &&
6689 global_rsv->space_info == space_info) {
6690 u64 to_add = len;
6691 WARN_ON(!return_free_space);
6692 spin_lock(&global_rsv->lock);
6693 if (!global_rsv->full) {
6694 to_add = min(len, global_rsv->size -
6695 global_rsv->reserved);
6696 global_rsv->reserved += to_add;
6697 space_info->bytes_may_use += to_add;
6698 if (global_rsv->reserved >= global_rsv->size)
6699 global_rsv->full = 1;
6700 trace_btrfs_space_reservation(fs_info,
6701 "space_info",
6702 space_info->flags,
6703 to_add, 1);
6704 len -= to_add;
6705 }
6706 spin_unlock(&global_rsv->lock);
6707 /* Add to any tickets we may have */
6708 if (len)
6709 space_info_add_new_bytes(fs_info, space_info,
6710 len);
6711 }
6712 spin_unlock(&space_info->lock);
6713 }
6714
6715 if (cache)
6716 btrfs_put_block_group(cache);
6717 return 0;
6718 }
6719
6720 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6721 struct btrfs_root *root)
6722 {
6723 struct btrfs_fs_info *fs_info = root->fs_info;
6724 struct btrfs_block_group_cache *block_group, *tmp;
6725 struct list_head *deleted_bgs;
6726 struct extent_io_tree *unpin;
6727 u64 start;
6728 u64 end;
6729 int ret;
6730
6731 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6732 unpin = &fs_info->freed_extents[1];
6733 else
6734 unpin = &fs_info->freed_extents[0];
6735
6736 while (!trans->aborted) {
6737 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6738 ret = find_first_extent_bit(unpin, 0, &start, &end,
6739 EXTENT_DIRTY, NULL);
6740 if (ret) {
6741 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6742 break;
6743 }
6744
6745 if (btrfs_test_opt(root, DISCARD))
6746 ret = btrfs_discard_extent(root, start,
6747 end + 1 - start, NULL);
6748
6749 clear_extent_dirty(unpin, start, end);
6750 unpin_extent_range(root, start, end, true);
6751 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6752 cond_resched();
6753 }
6754
6755 /*
6756 * Transaction is finished. We don't need the lock anymore. We
6757 * do need to clean up the block groups in case of a transaction
6758 * abort.
6759 */
6760 deleted_bgs = &trans->transaction->deleted_bgs;
6761 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6762 u64 trimmed = 0;
6763
6764 ret = -EROFS;
6765 if (!trans->aborted)
6766 ret = btrfs_discard_extent(root,
6767 block_group->key.objectid,
6768 block_group->key.offset,
6769 &trimmed);
6770
6771 list_del_init(&block_group->bg_list);
6772 btrfs_put_block_group_trimming(block_group);
6773 btrfs_put_block_group(block_group);
6774
6775 if (ret) {
6776 const char *errstr = btrfs_decode_error(ret);
6777 btrfs_warn(fs_info,
6778 "Discard failed while removing blockgroup: errno=%d %s\n",
6779 ret, errstr);
6780 }
6781 }
6782
6783 return 0;
6784 }
6785
6786 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
6787 u64 owner, u64 root_objectid)
6788 {
6789 struct btrfs_space_info *space_info;
6790 u64 flags;
6791
6792 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6793 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
6794 flags = BTRFS_BLOCK_GROUP_SYSTEM;
6795 else
6796 flags = BTRFS_BLOCK_GROUP_METADATA;
6797 } else {
6798 flags = BTRFS_BLOCK_GROUP_DATA;
6799 }
6800
6801 space_info = __find_space_info(fs_info, flags);
6802 BUG_ON(!space_info); /* Logic bug */
6803 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
6804 }
6805
6806
6807 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6808 struct btrfs_root *root,
6809 struct btrfs_delayed_ref_node *node, u64 parent,
6810 u64 root_objectid, u64 owner_objectid,
6811 u64 owner_offset, int refs_to_drop,
6812 struct btrfs_delayed_extent_op *extent_op)
6813 {
6814 struct btrfs_key key;
6815 struct btrfs_path *path;
6816 struct btrfs_fs_info *info = root->fs_info;
6817 struct btrfs_root *extent_root = info->extent_root;
6818 struct extent_buffer *leaf;
6819 struct btrfs_extent_item *ei;
6820 struct btrfs_extent_inline_ref *iref;
6821 int ret;
6822 int is_data;
6823 int extent_slot = 0;
6824 int found_extent = 0;
6825 int num_to_del = 1;
6826 u32 item_size;
6827 u64 refs;
6828 u64 bytenr = node->bytenr;
6829 u64 num_bytes = node->num_bytes;
6830 int last_ref = 0;
6831 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6832 SKINNY_METADATA);
6833
6834 path = btrfs_alloc_path();
6835 if (!path)
6836 return -ENOMEM;
6837
6838 path->reada = READA_FORWARD;
6839 path->leave_spinning = 1;
6840
6841 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6842 BUG_ON(!is_data && refs_to_drop != 1);
6843
6844 if (is_data)
6845 skinny_metadata = 0;
6846
6847 ret = lookup_extent_backref(trans, extent_root, path, &iref,
6848 bytenr, num_bytes, parent,
6849 root_objectid, owner_objectid,
6850 owner_offset);
6851 if (ret == 0) {
6852 extent_slot = path->slots[0];
6853 while (extent_slot >= 0) {
6854 btrfs_item_key_to_cpu(path->nodes[0], &key,
6855 extent_slot);
6856 if (key.objectid != bytenr)
6857 break;
6858 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6859 key.offset == num_bytes) {
6860 found_extent = 1;
6861 break;
6862 }
6863 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6864 key.offset == owner_objectid) {
6865 found_extent = 1;
6866 break;
6867 }
6868 if (path->slots[0] - extent_slot > 5)
6869 break;
6870 extent_slot--;
6871 }
6872 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6873 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6874 if (found_extent && item_size < sizeof(*ei))
6875 found_extent = 0;
6876 #endif
6877 if (!found_extent) {
6878 BUG_ON(iref);
6879 ret = remove_extent_backref(trans, extent_root, path,
6880 NULL, refs_to_drop,
6881 is_data, &last_ref);
6882 if (ret) {
6883 btrfs_abort_transaction(trans, extent_root, ret);
6884 goto out;
6885 }
6886 btrfs_release_path(path);
6887 path->leave_spinning = 1;
6888
6889 key.objectid = bytenr;
6890 key.type = BTRFS_EXTENT_ITEM_KEY;
6891 key.offset = num_bytes;
6892
6893 if (!is_data && skinny_metadata) {
6894 key.type = BTRFS_METADATA_ITEM_KEY;
6895 key.offset = owner_objectid;
6896 }
6897
6898 ret = btrfs_search_slot(trans, extent_root,
6899 &key, path, -1, 1);
6900 if (ret > 0 && skinny_metadata && path->slots[0]) {
6901 /*
6902 * Couldn't find our skinny metadata item,
6903 * see if we have ye olde extent item.
6904 */
6905 path->slots[0]--;
6906 btrfs_item_key_to_cpu(path->nodes[0], &key,
6907 path->slots[0]);
6908 if (key.objectid == bytenr &&
6909 key.type == BTRFS_EXTENT_ITEM_KEY &&
6910 key.offset == num_bytes)
6911 ret = 0;
6912 }
6913
6914 if (ret > 0 && skinny_metadata) {
6915 skinny_metadata = false;
6916 key.objectid = bytenr;
6917 key.type = BTRFS_EXTENT_ITEM_KEY;
6918 key.offset = num_bytes;
6919 btrfs_release_path(path);
6920 ret = btrfs_search_slot(trans, extent_root,
6921 &key, path, -1, 1);
6922 }
6923
6924 if (ret) {
6925 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6926 ret, bytenr);
6927 if (ret > 0)
6928 btrfs_print_leaf(extent_root,
6929 path->nodes[0]);
6930 }
6931 if (ret < 0) {
6932 btrfs_abort_transaction(trans, extent_root, ret);
6933 goto out;
6934 }
6935 extent_slot = path->slots[0];
6936 }
6937 } else if (WARN_ON(ret == -ENOENT)) {
6938 btrfs_print_leaf(extent_root, path->nodes[0]);
6939 btrfs_err(info,
6940 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6941 bytenr, parent, root_objectid, owner_objectid,
6942 owner_offset);
6943 btrfs_abort_transaction(trans, extent_root, ret);
6944 goto out;
6945 } else {
6946 btrfs_abort_transaction(trans, extent_root, ret);
6947 goto out;
6948 }
6949
6950 leaf = path->nodes[0];
6951 item_size = btrfs_item_size_nr(leaf, extent_slot);
6952 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6953 if (item_size < sizeof(*ei)) {
6954 BUG_ON(found_extent || extent_slot != path->slots[0]);
6955 ret = convert_extent_item_v0(trans, extent_root, path,
6956 owner_objectid, 0);
6957 if (ret < 0) {
6958 btrfs_abort_transaction(trans, extent_root, ret);
6959 goto out;
6960 }
6961
6962 btrfs_release_path(path);
6963 path->leave_spinning = 1;
6964
6965 key.objectid = bytenr;
6966 key.type = BTRFS_EXTENT_ITEM_KEY;
6967 key.offset = num_bytes;
6968
6969 ret = btrfs_search_slot(trans, extent_root, &key, path,
6970 -1, 1);
6971 if (ret) {
6972 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6973 ret, bytenr);
6974 btrfs_print_leaf(extent_root, path->nodes[0]);
6975 }
6976 if (ret < 0) {
6977 btrfs_abort_transaction(trans, extent_root, ret);
6978 goto out;
6979 }
6980
6981 extent_slot = path->slots[0];
6982 leaf = path->nodes[0];
6983 item_size = btrfs_item_size_nr(leaf, extent_slot);
6984 }
6985 #endif
6986 BUG_ON(item_size < sizeof(*ei));
6987 ei = btrfs_item_ptr(leaf, extent_slot,
6988 struct btrfs_extent_item);
6989 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6990 key.type == BTRFS_EXTENT_ITEM_KEY) {
6991 struct btrfs_tree_block_info *bi;
6992 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6993 bi = (struct btrfs_tree_block_info *)(ei + 1);
6994 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6995 }
6996
6997 refs = btrfs_extent_refs(leaf, ei);
6998 if (refs < refs_to_drop) {
6999 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
7000 "for bytenr %Lu", refs_to_drop, refs, bytenr);
7001 ret = -EINVAL;
7002 btrfs_abort_transaction(trans, extent_root, ret);
7003 goto out;
7004 }
7005 refs -= refs_to_drop;
7006
7007 if (refs > 0) {
7008 if (extent_op)
7009 __run_delayed_extent_op(extent_op, leaf, ei);
7010 /*
7011 * In the case of inline back ref, reference count will
7012 * be updated by remove_extent_backref
7013 */
7014 if (iref) {
7015 BUG_ON(!found_extent);
7016 } else {
7017 btrfs_set_extent_refs(leaf, ei, refs);
7018 btrfs_mark_buffer_dirty(leaf);
7019 }
7020 if (found_extent) {
7021 ret = remove_extent_backref(trans, extent_root, path,
7022 iref, refs_to_drop,
7023 is_data, &last_ref);
7024 if (ret) {
7025 btrfs_abort_transaction(trans, extent_root, ret);
7026 goto out;
7027 }
7028 }
7029 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
7030 root_objectid);
7031 } else {
7032 if (found_extent) {
7033 BUG_ON(is_data && refs_to_drop !=
7034 extent_data_ref_count(path, iref));
7035 if (iref) {
7036 BUG_ON(path->slots[0] != extent_slot);
7037 } else {
7038 BUG_ON(path->slots[0] != extent_slot + 1);
7039 path->slots[0] = extent_slot;
7040 num_to_del = 2;
7041 }
7042 }
7043
7044 last_ref = 1;
7045 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
7046 num_to_del);
7047 if (ret) {
7048 btrfs_abort_transaction(trans, extent_root, ret);
7049 goto out;
7050 }
7051 btrfs_release_path(path);
7052
7053 if (is_data) {
7054 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
7055 if (ret) {
7056 btrfs_abort_transaction(trans, extent_root, ret);
7057 goto out;
7058 }
7059 }
7060
7061 ret = add_to_free_space_tree(trans, root->fs_info, bytenr,
7062 num_bytes);
7063 if (ret) {
7064 btrfs_abort_transaction(trans, extent_root, ret);
7065 goto out;
7066 }
7067
7068 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
7069 if (ret) {
7070 btrfs_abort_transaction(trans, extent_root, ret);
7071 goto out;
7072 }
7073 }
7074 btrfs_release_path(path);
7075
7076 out:
7077 btrfs_free_path(path);
7078 return ret;
7079 }
7080
7081 /*
7082 * when we free an block, it is possible (and likely) that we free the last
7083 * delayed ref for that extent as well. This searches the delayed ref tree for
7084 * a given extent, and if there are no other delayed refs to be processed, it
7085 * removes it from the tree.
7086 */
7087 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
7088 struct btrfs_root *root, u64 bytenr)
7089 {
7090 struct btrfs_delayed_ref_head *head;
7091 struct btrfs_delayed_ref_root *delayed_refs;
7092 int ret = 0;
7093
7094 delayed_refs = &trans->transaction->delayed_refs;
7095 spin_lock(&delayed_refs->lock);
7096 head = btrfs_find_delayed_ref_head(trans, bytenr);
7097 if (!head)
7098 goto out_delayed_unlock;
7099
7100 spin_lock(&head->lock);
7101 if (!list_empty(&head->ref_list))
7102 goto out;
7103
7104 if (head->extent_op) {
7105 if (!head->must_insert_reserved)
7106 goto out;
7107 btrfs_free_delayed_extent_op(head->extent_op);
7108 head->extent_op = NULL;
7109 }
7110
7111 /*
7112 * waiting for the lock here would deadlock. If someone else has it
7113 * locked they are already in the process of dropping it anyway
7114 */
7115 if (!mutex_trylock(&head->mutex))
7116 goto out;
7117
7118 /*
7119 * at this point we have a head with no other entries. Go
7120 * ahead and process it.
7121 */
7122 head->node.in_tree = 0;
7123 rb_erase(&head->href_node, &delayed_refs->href_root);
7124
7125 atomic_dec(&delayed_refs->num_entries);
7126
7127 /*
7128 * we don't take a ref on the node because we're removing it from the
7129 * tree, so we just steal the ref the tree was holding.
7130 */
7131 delayed_refs->num_heads--;
7132 if (head->processing == 0)
7133 delayed_refs->num_heads_ready--;
7134 head->processing = 0;
7135 spin_unlock(&head->lock);
7136 spin_unlock(&delayed_refs->lock);
7137
7138 BUG_ON(head->extent_op);
7139 if (head->must_insert_reserved)
7140 ret = 1;
7141
7142 mutex_unlock(&head->mutex);
7143 btrfs_put_delayed_ref(&head->node);
7144 return ret;
7145 out:
7146 spin_unlock(&head->lock);
7147
7148 out_delayed_unlock:
7149 spin_unlock(&delayed_refs->lock);
7150 return 0;
7151 }
7152
7153 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
7154 struct btrfs_root *root,
7155 struct extent_buffer *buf,
7156 u64 parent, int last_ref)
7157 {
7158 int pin = 1;
7159 int ret;
7160
7161 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7162 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7163 buf->start, buf->len,
7164 parent, root->root_key.objectid,
7165 btrfs_header_level(buf),
7166 BTRFS_DROP_DELAYED_REF, NULL);
7167 BUG_ON(ret); /* -ENOMEM */
7168 }
7169
7170 if (!last_ref)
7171 return;
7172
7173 if (btrfs_header_generation(buf) == trans->transid) {
7174 struct btrfs_block_group_cache *cache;
7175
7176 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7177 ret = check_ref_cleanup(trans, root, buf->start);
7178 if (!ret)
7179 goto out;
7180 }
7181
7182 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
7183
7184 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
7185 pin_down_extent(root, cache, buf->start, buf->len, 1);
7186 btrfs_put_block_group(cache);
7187 goto out;
7188 }
7189
7190 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
7191
7192 btrfs_add_free_space(cache, buf->start, buf->len);
7193 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
7194 btrfs_put_block_group(cache);
7195 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
7196 pin = 0;
7197 }
7198 out:
7199 if (pin)
7200 add_pinned_bytes(root->fs_info, buf->len,
7201 btrfs_header_level(buf),
7202 root->root_key.objectid);
7203
7204 /*
7205 * Deleting the buffer, clear the corrupt flag since it doesn't matter
7206 * anymore.
7207 */
7208 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
7209 }
7210
7211 /* Can return -ENOMEM */
7212 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7213 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
7214 u64 owner, u64 offset)
7215 {
7216 int ret;
7217 struct btrfs_fs_info *fs_info = root->fs_info;
7218
7219 if (btrfs_test_is_dummy_root(root))
7220 return 0;
7221
7222 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
7223
7224 /*
7225 * tree log blocks never actually go into the extent allocation
7226 * tree, just update pinning info and exit early.
7227 */
7228 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
7229 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
7230 /* unlocks the pinned mutex */
7231 btrfs_pin_extent(root, bytenr, num_bytes, 1);
7232 ret = 0;
7233 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
7234 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
7235 num_bytes,
7236 parent, root_objectid, (int)owner,
7237 BTRFS_DROP_DELAYED_REF, NULL);
7238 } else {
7239 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
7240 num_bytes,
7241 parent, root_objectid, owner,
7242 offset, 0,
7243 BTRFS_DROP_DELAYED_REF, NULL);
7244 }
7245 return ret;
7246 }
7247
7248 /*
7249 * when we wait for progress in the block group caching, its because
7250 * our allocation attempt failed at least once. So, we must sleep
7251 * and let some progress happen before we try again.
7252 *
7253 * This function will sleep at least once waiting for new free space to
7254 * show up, and then it will check the block group free space numbers
7255 * for our min num_bytes. Another option is to have it go ahead
7256 * and look in the rbtree for a free extent of a given size, but this
7257 * is a good start.
7258 *
7259 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7260 * any of the information in this block group.
7261 */
7262 static noinline void
7263 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
7264 u64 num_bytes)
7265 {
7266 struct btrfs_caching_control *caching_ctl;
7267
7268 caching_ctl = get_caching_control(cache);
7269 if (!caching_ctl)
7270 return;
7271
7272 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
7273 (cache->free_space_ctl->free_space >= num_bytes));
7274
7275 put_caching_control(caching_ctl);
7276 }
7277
7278 static noinline int
7279 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
7280 {
7281 struct btrfs_caching_control *caching_ctl;
7282 int ret = 0;
7283
7284 caching_ctl = get_caching_control(cache);
7285 if (!caching_ctl)
7286 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
7287
7288 wait_event(caching_ctl->wait, block_group_cache_done(cache));
7289 if (cache->cached == BTRFS_CACHE_ERROR)
7290 ret = -EIO;
7291 put_caching_control(caching_ctl);
7292 return ret;
7293 }
7294
7295 int __get_raid_index(u64 flags)
7296 {
7297 if (flags & BTRFS_BLOCK_GROUP_RAID10)
7298 return BTRFS_RAID_RAID10;
7299 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
7300 return BTRFS_RAID_RAID1;
7301 else if (flags & BTRFS_BLOCK_GROUP_DUP)
7302 return BTRFS_RAID_DUP;
7303 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
7304 return BTRFS_RAID_RAID0;
7305 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
7306 return BTRFS_RAID_RAID5;
7307 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
7308 return BTRFS_RAID_RAID6;
7309
7310 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
7311 }
7312
7313 int get_block_group_index(struct btrfs_block_group_cache *cache)
7314 {
7315 return __get_raid_index(cache->flags);
7316 }
7317
7318 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
7319 [BTRFS_RAID_RAID10] = "raid10",
7320 [BTRFS_RAID_RAID1] = "raid1",
7321 [BTRFS_RAID_DUP] = "dup",
7322 [BTRFS_RAID_RAID0] = "raid0",
7323 [BTRFS_RAID_SINGLE] = "single",
7324 [BTRFS_RAID_RAID5] = "raid5",
7325 [BTRFS_RAID_RAID6] = "raid6",
7326 };
7327
7328 static const char *get_raid_name(enum btrfs_raid_types type)
7329 {
7330 if (type >= BTRFS_NR_RAID_TYPES)
7331 return NULL;
7332
7333 return btrfs_raid_type_names[type];
7334 }
7335
7336 enum btrfs_loop_type {
7337 LOOP_CACHING_NOWAIT = 0,
7338 LOOP_CACHING_WAIT = 1,
7339 LOOP_ALLOC_CHUNK = 2,
7340 LOOP_NO_EMPTY_SIZE = 3,
7341 };
7342
7343 static inline void
7344 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7345 int delalloc)
7346 {
7347 if (delalloc)
7348 down_read(&cache->data_rwsem);
7349 }
7350
7351 static inline void
7352 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7353 int delalloc)
7354 {
7355 btrfs_get_block_group(cache);
7356 if (delalloc)
7357 down_read(&cache->data_rwsem);
7358 }
7359
7360 static struct btrfs_block_group_cache *
7361 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7362 struct btrfs_free_cluster *cluster,
7363 int delalloc)
7364 {
7365 struct btrfs_block_group_cache *used_bg = NULL;
7366
7367 spin_lock(&cluster->refill_lock);
7368 while (1) {
7369 used_bg = cluster->block_group;
7370 if (!used_bg)
7371 return NULL;
7372
7373 if (used_bg == block_group)
7374 return used_bg;
7375
7376 btrfs_get_block_group(used_bg);
7377
7378 if (!delalloc)
7379 return used_bg;
7380
7381 if (down_read_trylock(&used_bg->data_rwsem))
7382 return used_bg;
7383
7384 spin_unlock(&cluster->refill_lock);
7385
7386 down_read(&used_bg->data_rwsem);
7387
7388 spin_lock(&cluster->refill_lock);
7389 if (used_bg == cluster->block_group)
7390 return used_bg;
7391
7392 up_read(&used_bg->data_rwsem);
7393 btrfs_put_block_group(used_bg);
7394 }
7395 }
7396
7397 static inline void
7398 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7399 int delalloc)
7400 {
7401 if (delalloc)
7402 up_read(&cache->data_rwsem);
7403 btrfs_put_block_group(cache);
7404 }
7405
7406 /*
7407 * walks the btree of allocated extents and find a hole of a given size.
7408 * The key ins is changed to record the hole:
7409 * ins->objectid == start position
7410 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7411 * ins->offset == the size of the hole.
7412 * Any available blocks before search_start are skipped.
7413 *
7414 * If there is no suitable free space, we will record the max size of
7415 * the free space extent currently.
7416 */
7417 static noinline int find_free_extent(struct btrfs_root *orig_root,
7418 u64 num_bytes, u64 empty_size,
7419 u64 hint_byte, struct btrfs_key *ins,
7420 u64 flags, int delalloc)
7421 {
7422 int ret = 0;
7423 struct btrfs_root *root = orig_root->fs_info->extent_root;
7424 struct btrfs_free_cluster *last_ptr = NULL;
7425 struct btrfs_block_group_cache *block_group = NULL;
7426 u64 search_start = 0;
7427 u64 max_extent_size = 0;
7428 u64 empty_cluster = 0;
7429 struct btrfs_space_info *space_info;
7430 int loop = 0;
7431 int index = __get_raid_index(flags);
7432 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
7433 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
7434 bool failed_cluster_refill = false;
7435 bool failed_alloc = false;
7436 bool use_cluster = true;
7437 bool have_caching_bg = false;
7438 bool orig_have_caching_bg = false;
7439 bool full_search = false;
7440
7441 WARN_ON(num_bytes < root->sectorsize);
7442 ins->type = BTRFS_EXTENT_ITEM_KEY;
7443 ins->objectid = 0;
7444 ins->offset = 0;
7445
7446 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
7447
7448 space_info = __find_space_info(root->fs_info, flags);
7449 if (!space_info) {
7450 btrfs_err(root->fs_info, "No space info for %llu", flags);
7451 return -ENOSPC;
7452 }
7453
7454 /*
7455 * If our free space is heavily fragmented we may not be able to make
7456 * big contiguous allocations, so instead of doing the expensive search
7457 * for free space, simply return ENOSPC with our max_extent_size so we
7458 * can go ahead and search for a more manageable chunk.
7459 *
7460 * If our max_extent_size is large enough for our allocation simply
7461 * disable clustering since we will likely not be able to find enough
7462 * space to create a cluster and induce latency trying.
7463 */
7464 if (unlikely(space_info->max_extent_size)) {
7465 spin_lock(&space_info->lock);
7466 if (space_info->max_extent_size &&
7467 num_bytes > space_info->max_extent_size) {
7468 ins->offset = space_info->max_extent_size;
7469 spin_unlock(&space_info->lock);
7470 return -ENOSPC;
7471 } else if (space_info->max_extent_size) {
7472 use_cluster = false;
7473 }
7474 spin_unlock(&space_info->lock);
7475 }
7476
7477 last_ptr = fetch_cluster_info(orig_root, space_info, &empty_cluster);
7478 if (last_ptr) {
7479 spin_lock(&last_ptr->lock);
7480 if (last_ptr->block_group)
7481 hint_byte = last_ptr->window_start;
7482 if (last_ptr->fragmented) {
7483 /*
7484 * We still set window_start so we can keep track of the
7485 * last place we found an allocation to try and save
7486 * some time.
7487 */
7488 hint_byte = last_ptr->window_start;
7489 use_cluster = false;
7490 }
7491 spin_unlock(&last_ptr->lock);
7492 }
7493
7494 search_start = max(search_start, first_logical_byte(root, 0));
7495 search_start = max(search_start, hint_byte);
7496 if (search_start == hint_byte) {
7497 block_group = btrfs_lookup_block_group(root->fs_info,
7498 search_start);
7499 /*
7500 * we don't want to use the block group if it doesn't match our
7501 * allocation bits, or if its not cached.
7502 *
7503 * However if we are re-searching with an ideal block group
7504 * picked out then we don't care that the block group is cached.
7505 */
7506 if (block_group && block_group_bits(block_group, flags) &&
7507 block_group->cached != BTRFS_CACHE_NO) {
7508 down_read(&space_info->groups_sem);
7509 if (list_empty(&block_group->list) ||
7510 block_group->ro) {
7511 /*
7512 * someone is removing this block group,
7513 * we can't jump into the have_block_group
7514 * target because our list pointers are not
7515 * valid
7516 */
7517 btrfs_put_block_group(block_group);
7518 up_read(&space_info->groups_sem);
7519 } else {
7520 index = get_block_group_index(block_group);
7521 btrfs_lock_block_group(block_group, delalloc);
7522 goto have_block_group;
7523 }
7524 } else if (block_group) {
7525 btrfs_put_block_group(block_group);
7526 }
7527 }
7528 search:
7529 have_caching_bg = false;
7530 if (index == 0 || index == __get_raid_index(flags))
7531 full_search = true;
7532 down_read(&space_info->groups_sem);
7533 list_for_each_entry(block_group, &space_info->block_groups[index],
7534 list) {
7535 u64 offset;
7536 int cached;
7537
7538 btrfs_grab_block_group(block_group, delalloc);
7539 search_start = block_group->key.objectid;
7540
7541 /*
7542 * this can happen if we end up cycling through all the
7543 * raid types, but we want to make sure we only allocate
7544 * for the proper type.
7545 */
7546 if (!block_group_bits(block_group, flags)) {
7547 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7548 BTRFS_BLOCK_GROUP_RAID1 |
7549 BTRFS_BLOCK_GROUP_RAID5 |
7550 BTRFS_BLOCK_GROUP_RAID6 |
7551 BTRFS_BLOCK_GROUP_RAID10;
7552
7553 /*
7554 * if they asked for extra copies and this block group
7555 * doesn't provide them, bail. This does allow us to
7556 * fill raid0 from raid1.
7557 */
7558 if ((flags & extra) && !(block_group->flags & extra))
7559 goto loop;
7560 }
7561
7562 have_block_group:
7563 cached = block_group_cache_done(block_group);
7564 if (unlikely(!cached)) {
7565 have_caching_bg = true;
7566 ret = cache_block_group(block_group, 0);
7567 BUG_ON(ret < 0);
7568 ret = 0;
7569 }
7570
7571 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7572 goto loop;
7573 if (unlikely(block_group->ro))
7574 goto loop;
7575
7576 /*
7577 * Ok we want to try and use the cluster allocator, so
7578 * lets look there
7579 */
7580 if (last_ptr && use_cluster) {
7581 struct btrfs_block_group_cache *used_block_group;
7582 unsigned long aligned_cluster;
7583 /*
7584 * the refill lock keeps out other
7585 * people trying to start a new cluster
7586 */
7587 used_block_group = btrfs_lock_cluster(block_group,
7588 last_ptr,
7589 delalloc);
7590 if (!used_block_group)
7591 goto refill_cluster;
7592
7593 if (used_block_group != block_group &&
7594 (used_block_group->ro ||
7595 !block_group_bits(used_block_group, flags)))
7596 goto release_cluster;
7597
7598 offset = btrfs_alloc_from_cluster(used_block_group,
7599 last_ptr,
7600 num_bytes,
7601 used_block_group->key.objectid,
7602 &max_extent_size);
7603 if (offset) {
7604 /* we have a block, we're done */
7605 spin_unlock(&last_ptr->refill_lock);
7606 trace_btrfs_reserve_extent_cluster(root,
7607 used_block_group,
7608 search_start, num_bytes);
7609 if (used_block_group != block_group) {
7610 btrfs_release_block_group(block_group,
7611 delalloc);
7612 block_group = used_block_group;
7613 }
7614 goto checks;
7615 }
7616
7617 WARN_ON(last_ptr->block_group != used_block_group);
7618 release_cluster:
7619 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7620 * set up a new clusters, so lets just skip it
7621 * and let the allocator find whatever block
7622 * it can find. If we reach this point, we
7623 * will have tried the cluster allocator
7624 * plenty of times and not have found
7625 * anything, so we are likely way too
7626 * fragmented for the clustering stuff to find
7627 * anything.
7628 *
7629 * However, if the cluster is taken from the
7630 * current block group, release the cluster
7631 * first, so that we stand a better chance of
7632 * succeeding in the unclustered
7633 * allocation. */
7634 if (loop >= LOOP_NO_EMPTY_SIZE &&
7635 used_block_group != block_group) {
7636 spin_unlock(&last_ptr->refill_lock);
7637 btrfs_release_block_group(used_block_group,
7638 delalloc);
7639 goto unclustered_alloc;
7640 }
7641
7642 /*
7643 * this cluster didn't work out, free it and
7644 * start over
7645 */
7646 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7647
7648 if (used_block_group != block_group)
7649 btrfs_release_block_group(used_block_group,
7650 delalloc);
7651 refill_cluster:
7652 if (loop >= LOOP_NO_EMPTY_SIZE) {
7653 spin_unlock(&last_ptr->refill_lock);
7654 goto unclustered_alloc;
7655 }
7656
7657 aligned_cluster = max_t(unsigned long,
7658 empty_cluster + empty_size,
7659 block_group->full_stripe_len);
7660
7661 /* allocate a cluster in this block group */
7662 ret = btrfs_find_space_cluster(root, block_group,
7663 last_ptr, search_start,
7664 num_bytes,
7665 aligned_cluster);
7666 if (ret == 0) {
7667 /*
7668 * now pull our allocation out of this
7669 * cluster
7670 */
7671 offset = btrfs_alloc_from_cluster(block_group,
7672 last_ptr,
7673 num_bytes,
7674 search_start,
7675 &max_extent_size);
7676 if (offset) {
7677 /* we found one, proceed */
7678 spin_unlock(&last_ptr->refill_lock);
7679 trace_btrfs_reserve_extent_cluster(root,
7680 block_group, search_start,
7681 num_bytes);
7682 goto checks;
7683 }
7684 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7685 && !failed_cluster_refill) {
7686 spin_unlock(&last_ptr->refill_lock);
7687
7688 failed_cluster_refill = true;
7689 wait_block_group_cache_progress(block_group,
7690 num_bytes + empty_cluster + empty_size);
7691 goto have_block_group;
7692 }
7693
7694 /*
7695 * at this point we either didn't find a cluster
7696 * or we weren't able to allocate a block from our
7697 * cluster. Free the cluster we've been trying
7698 * to use, and go to the next block group
7699 */
7700 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7701 spin_unlock(&last_ptr->refill_lock);
7702 goto loop;
7703 }
7704
7705 unclustered_alloc:
7706 /*
7707 * We are doing an unclustered alloc, set the fragmented flag so
7708 * we don't bother trying to setup a cluster again until we get
7709 * more space.
7710 */
7711 if (unlikely(last_ptr)) {
7712 spin_lock(&last_ptr->lock);
7713 last_ptr->fragmented = 1;
7714 spin_unlock(&last_ptr->lock);
7715 }
7716 spin_lock(&block_group->free_space_ctl->tree_lock);
7717 if (cached &&
7718 block_group->free_space_ctl->free_space <
7719 num_bytes + empty_cluster + empty_size) {
7720 if (block_group->free_space_ctl->free_space >
7721 max_extent_size)
7722 max_extent_size =
7723 block_group->free_space_ctl->free_space;
7724 spin_unlock(&block_group->free_space_ctl->tree_lock);
7725 goto loop;
7726 }
7727 spin_unlock(&block_group->free_space_ctl->tree_lock);
7728
7729 offset = btrfs_find_space_for_alloc(block_group, search_start,
7730 num_bytes, empty_size,
7731 &max_extent_size);
7732 /*
7733 * If we didn't find a chunk, and we haven't failed on this
7734 * block group before, and this block group is in the middle of
7735 * caching and we are ok with waiting, then go ahead and wait
7736 * for progress to be made, and set failed_alloc to true.
7737 *
7738 * If failed_alloc is true then we've already waited on this
7739 * block group once and should move on to the next block group.
7740 */
7741 if (!offset && !failed_alloc && !cached &&
7742 loop > LOOP_CACHING_NOWAIT) {
7743 wait_block_group_cache_progress(block_group,
7744 num_bytes + empty_size);
7745 failed_alloc = true;
7746 goto have_block_group;
7747 } else if (!offset) {
7748 goto loop;
7749 }
7750 checks:
7751 search_start = ALIGN(offset, root->stripesize);
7752
7753 /* move on to the next group */
7754 if (search_start + num_bytes >
7755 block_group->key.objectid + block_group->key.offset) {
7756 btrfs_add_free_space(block_group, offset, num_bytes);
7757 goto loop;
7758 }
7759
7760 if (offset < search_start)
7761 btrfs_add_free_space(block_group, offset,
7762 search_start - offset);
7763 BUG_ON(offset > search_start);
7764
7765 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
7766 alloc_type, delalloc);
7767 if (ret == -EAGAIN) {
7768 btrfs_add_free_space(block_group, offset, num_bytes);
7769 goto loop;
7770 }
7771 btrfs_inc_block_group_reservations(block_group);
7772
7773 /* we are all good, lets return */
7774 ins->objectid = search_start;
7775 ins->offset = num_bytes;
7776
7777 trace_btrfs_reserve_extent(orig_root, block_group,
7778 search_start, num_bytes);
7779 btrfs_release_block_group(block_group, delalloc);
7780 break;
7781 loop:
7782 failed_cluster_refill = false;
7783 failed_alloc = false;
7784 BUG_ON(index != get_block_group_index(block_group));
7785 btrfs_release_block_group(block_group, delalloc);
7786 }
7787 up_read(&space_info->groups_sem);
7788
7789 if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg
7790 && !orig_have_caching_bg)
7791 orig_have_caching_bg = true;
7792
7793 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7794 goto search;
7795
7796 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7797 goto search;
7798
7799 /*
7800 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7801 * caching kthreads as we move along
7802 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7803 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7804 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7805 * again
7806 */
7807 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7808 index = 0;
7809 if (loop == LOOP_CACHING_NOWAIT) {
7810 /*
7811 * We want to skip the LOOP_CACHING_WAIT step if we
7812 * don't have any uncached bgs and we've already done a
7813 * full search through.
7814 */
7815 if (orig_have_caching_bg || !full_search)
7816 loop = LOOP_CACHING_WAIT;
7817 else
7818 loop = LOOP_ALLOC_CHUNK;
7819 } else {
7820 loop++;
7821 }
7822
7823 if (loop == LOOP_ALLOC_CHUNK) {
7824 struct btrfs_trans_handle *trans;
7825 int exist = 0;
7826
7827 trans = current->journal_info;
7828 if (trans)
7829 exist = 1;
7830 else
7831 trans = btrfs_join_transaction(root);
7832
7833 if (IS_ERR(trans)) {
7834 ret = PTR_ERR(trans);
7835 goto out;
7836 }
7837
7838 ret = do_chunk_alloc(trans, root, flags,
7839 CHUNK_ALLOC_FORCE);
7840
7841 /*
7842 * If we can't allocate a new chunk we've already looped
7843 * through at least once, move on to the NO_EMPTY_SIZE
7844 * case.
7845 */
7846 if (ret == -ENOSPC)
7847 loop = LOOP_NO_EMPTY_SIZE;
7848
7849 /*
7850 * Do not bail out on ENOSPC since we
7851 * can do more things.
7852 */
7853 if (ret < 0 && ret != -ENOSPC)
7854 btrfs_abort_transaction(trans,
7855 root, ret);
7856 else
7857 ret = 0;
7858 if (!exist)
7859 btrfs_end_transaction(trans, root);
7860 if (ret)
7861 goto out;
7862 }
7863
7864 if (loop == LOOP_NO_EMPTY_SIZE) {
7865 /*
7866 * Don't loop again if we already have no empty_size and
7867 * no empty_cluster.
7868 */
7869 if (empty_size == 0 &&
7870 empty_cluster == 0) {
7871 ret = -ENOSPC;
7872 goto out;
7873 }
7874 empty_size = 0;
7875 empty_cluster = 0;
7876 }
7877
7878 goto search;
7879 } else if (!ins->objectid) {
7880 ret = -ENOSPC;
7881 } else if (ins->objectid) {
7882 if (!use_cluster && last_ptr) {
7883 spin_lock(&last_ptr->lock);
7884 last_ptr->window_start = ins->objectid;
7885 spin_unlock(&last_ptr->lock);
7886 }
7887 ret = 0;
7888 }
7889 out:
7890 if (ret == -ENOSPC) {
7891 spin_lock(&space_info->lock);
7892 space_info->max_extent_size = max_extent_size;
7893 spin_unlock(&space_info->lock);
7894 ins->offset = max_extent_size;
7895 }
7896 return ret;
7897 }
7898
7899 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
7900 int dump_block_groups)
7901 {
7902 struct btrfs_block_group_cache *cache;
7903 int index = 0;
7904
7905 spin_lock(&info->lock);
7906 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
7907 info->flags,
7908 info->total_bytes - info->bytes_used - info->bytes_pinned -
7909 info->bytes_reserved - info->bytes_readonly -
7910 info->bytes_may_use, (info->full) ? "" : "not ");
7911 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7912 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7913 info->total_bytes, info->bytes_used, info->bytes_pinned,
7914 info->bytes_reserved, info->bytes_may_use,
7915 info->bytes_readonly);
7916 spin_unlock(&info->lock);
7917
7918 if (!dump_block_groups)
7919 return;
7920
7921 down_read(&info->groups_sem);
7922 again:
7923 list_for_each_entry(cache, &info->block_groups[index], list) {
7924 spin_lock(&cache->lock);
7925 printk(KERN_INFO "BTRFS: "
7926 "block group %llu has %llu bytes, "
7927 "%llu used %llu pinned %llu reserved %s\n",
7928 cache->key.objectid, cache->key.offset,
7929 btrfs_block_group_used(&cache->item), cache->pinned,
7930 cache->reserved, cache->ro ? "[readonly]" : "");
7931 btrfs_dump_free_space(cache, bytes);
7932 spin_unlock(&cache->lock);
7933 }
7934 if (++index < BTRFS_NR_RAID_TYPES)
7935 goto again;
7936 up_read(&info->groups_sem);
7937 }
7938
7939 int btrfs_reserve_extent(struct btrfs_root *root,
7940 u64 num_bytes, u64 min_alloc_size,
7941 u64 empty_size, u64 hint_byte,
7942 struct btrfs_key *ins, int is_data, int delalloc)
7943 {
7944 bool final_tried = num_bytes == min_alloc_size;
7945 u64 flags;
7946 int ret;
7947
7948 flags = btrfs_get_alloc_profile(root, is_data);
7949 again:
7950 WARN_ON(num_bytes < root->sectorsize);
7951 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
7952 flags, delalloc);
7953 if (!ret && !is_data) {
7954 btrfs_dec_block_group_reservations(root->fs_info,
7955 ins->objectid);
7956 } else if (ret == -ENOSPC) {
7957 if (!final_tried && ins->offset) {
7958 num_bytes = min(num_bytes >> 1, ins->offset);
7959 num_bytes = round_down(num_bytes, root->sectorsize);
7960 num_bytes = max(num_bytes, min_alloc_size);
7961 if (num_bytes == min_alloc_size)
7962 final_tried = true;
7963 goto again;
7964 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7965 struct btrfs_space_info *sinfo;
7966
7967 sinfo = __find_space_info(root->fs_info, flags);
7968 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
7969 flags, num_bytes);
7970 if (sinfo)
7971 dump_space_info(sinfo, num_bytes, 1);
7972 }
7973 }
7974
7975 return ret;
7976 }
7977
7978 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
7979 u64 start, u64 len,
7980 int pin, int delalloc)
7981 {
7982 struct btrfs_block_group_cache *cache;
7983 int ret = 0;
7984
7985 cache = btrfs_lookup_block_group(root->fs_info, start);
7986 if (!cache) {
7987 btrfs_err(root->fs_info, "Unable to find block group for %llu",
7988 start);
7989 return -ENOSPC;
7990 }
7991
7992 if (pin)
7993 pin_down_extent(root, cache, start, len, 1);
7994 else {
7995 if (btrfs_test_opt(root, DISCARD))
7996 ret = btrfs_discard_extent(root, start, len, NULL);
7997 btrfs_add_free_space(cache, start, len);
7998 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
7999 trace_btrfs_reserved_extent_free(root, start, len);
8000 }
8001
8002 btrfs_put_block_group(cache);
8003 return ret;
8004 }
8005
8006 int btrfs_free_reserved_extent(struct btrfs_root *root,
8007 u64 start, u64 len, int delalloc)
8008 {
8009 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
8010 }
8011
8012 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
8013 u64 start, u64 len)
8014 {
8015 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
8016 }
8017
8018 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8019 struct btrfs_root *root,
8020 u64 parent, u64 root_objectid,
8021 u64 flags, u64 owner, u64 offset,
8022 struct btrfs_key *ins, int ref_mod)
8023 {
8024 int ret;
8025 struct btrfs_fs_info *fs_info = root->fs_info;
8026 struct btrfs_extent_item *extent_item;
8027 struct btrfs_extent_inline_ref *iref;
8028 struct btrfs_path *path;
8029 struct extent_buffer *leaf;
8030 int type;
8031 u32 size;
8032
8033 if (parent > 0)
8034 type = BTRFS_SHARED_DATA_REF_KEY;
8035 else
8036 type = BTRFS_EXTENT_DATA_REF_KEY;
8037
8038 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
8039
8040 path = btrfs_alloc_path();
8041 if (!path)
8042 return -ENOMEM;
8043
8044 path->leave_spinning = 1;
8045 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8046 ins, size);
8047 if (ret) {
8048 btrfs_free_path(path);
8049 return ret;
8050 }
8051
8052 leaf = path->nodes[0];
8053 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8054 struct btrfs_extent_item);
8055 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
8056 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8057 btrfs_set_extent_flags(leaf, extent_item,
8058 flags | BTRFS_EXTENT_FLAG_DATA);
8059
8060 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8061 btrfs_set_extent_inline_ref_type(leaf, iref, type);
8062 if (parent > 0) {
8063 struct btrfs_shared_data_ref *ref;
8064 ref = (struct btrfs_shared_data_ref *)(iref + 1);
8065 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8066 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
8067 } else {
8068 struct btrfs_extent_data_ref *ref;
8069 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
8070 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
8071 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
8072 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
8073 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
8074 }
8075
8076 btrfs_mark_buffer_dirty(path->nodes[0]);
8077 btrfs_free_path(path);
8078
8079 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8080 ins->offset);
8081 if (ret)
8082 return ret;
8083
8084 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
8085 if (ret) { /* -ENOENT, logic error */
8086 btrfs_err(fs_info, "update block group failed for %llu %llu",
8087 ins->objectid, ins->offset);
8088 BUG();
8089 }
8090 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
8091 return ret;
8092 }
8093
8094 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
8095 struct btrfs_root *root,
8096 u64 parent, u64 root_objectid,
8097 u64 flags, struct btrfs_disk_key *key,
8098 int level, struct btrfs_key *ins)
8099 {
8100 int ret;
8101 struct btrfs_fs_info *fs_info = root->fs_info;
8102 struct btrfs_extent_item *extent_item;
8103 struct btrfs_tree_block_info *block_info;
8104 struct btrfs_extent_inline_ref *iref;
8105 struct btrfs_path *path;
8106 struct extent_buffer *leaf;
8107 u32 size = sizeof(*extent_item) + sizeof(*iref);
8108 u64 num_bytes = ins->offset;
8109 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
8110 SKINNY_METADATA);
8111
8112 if (!skinny_metadata)
8113 size += sizeof(*block_info);
8114
8115 path = btrfs_alloc_path();
8116 if (!path) {
8117 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
8118 root->nodesize);
8119 return -ENOMEM;
8120 }
8121
8122 path->leave_spinning = 1;
8123 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8124 ins, size);
8125 if (ret) {
8126 btrfs_free_path(path);
8127 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
8128 root->nodesize);
8129 return ret;
8130 }
8131
8132 leaf = path->nodes[0];
8133 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8134 struct btrfs_extent_item);
8135 btrfs_set_extent_refs(leaf, extent_item, 1);
8136 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8137 btrfs_set_extent_flags(leaf, extent_item,
8138 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
8139
8140 if (skinny_metadata) {
8141 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8142 num_bytes = root->nodesize;
8143 } else {
8144 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
8145 btrfs_set_tree_block_key(leaf, block_info, key);
8146 btrfs_set_tree_block_level(leaf, block_info, level);
8147 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
8148 }
8149
8150 if (parent > 0) {
8151 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
8152 btrfs_set_extent_inline_ref_type(leaf, iref,
8153 BTRFS_SHARED_BLOCK_REF_KEY);
8154 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8155 } else {
8156 btrfs_set_extent_inline_ref_type(leaf, iref,
8157 BTRFS_TREE_BLOCK_REF_KEY);
8158 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
8159 }
8160
8161 btrfs_mark_buffer_dirty(leaf);
8162 btrfs_free_path(path);
8163
8164 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8165 num_bytes);
8166 if (ret)
8167 return ret;
8168
8169 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
8170 1);
8171 if (ret) { /* -ENOENT, logic error */
8172 btrfs_err(fs_info, "update block group failed for %llu %llu",
8173 ins->objectid, ins->offset);
8174 BUG();
8175 }
8176
8177 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
8178 return ret;
8179 }
8180
8181 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8182 struct btrfs_root *root,
8183 u64 root_objectid, u64 owner,
8184 u64 offset, u64 ram_bytes,
8185 struct btrfs_key *ins)
8186 {
8187 int ret;
8188
8189 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
8190
8191 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
8192 ins->offset, 0,
8193 root_objectid, owner, offset,
8194 ram_bytes, BTRFS_ADD_DELAYED_EXTENT,
8195 NULL);
8196 return ret;
8197 }
8198
8199 /*
8200 * this is used by the tree logging recovery code. It records that
8201 * an extent has been allocated and makes sure to clear the free
8202 * space cache bits as well
8203 */
8204 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
8205 struct btrfs_root *root,
8206 u64 root_objectid, u64 owner, u64 offset,
8207 struct btrfs_key *ins)
8208 {
8209 int ret;
8210 struct btrfs_block_group_cache *block_group;
8211
8212 /*
8213 * Mixed block groups will exclude before processing the log so we only
8214 * need to do the exclude dance if this fs isn't mixed.
8215 */
8216 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
8217 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
8218 if (ret)
8219 return ret;
8220 }
8221
8222 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
8223 if (!block_group)
8224 return -EINVAL;
8225
8226 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
8227 RESERVE_ALLOC_NO_ACCOUNT, 0);
8228 BUG_ON(ret); /* logic error */
8229 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
8230 0, owner, offset, ins, 1);
8231 btrfs_put_block_group(block_group);
8232 return ret;
8233 }
8234
8235 static struct extent_buffer *
8236 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
8237 u64 bytenr, int level)
8238 {
8239 struct extent_buffer *buf;
8240
8241 buf = btrfs_find_create_tree_block(root, bytenr);
8242 if (IS_ERR(buf))
8243 return buf;
8244
8245 btrfs_set_header_generation(buf, trans->transid);
8246 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
8247 btrfs_tree_lock(buf);
8248 clean_tree_block(trans, root->fs_info, buf);
8249 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
8250
8251 btrfs_set_lock_blocking(buf);
8252 set_extent_buffer_uptodate(buf);
8253
8254 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8255 buf->log_index = root->log_transid % 2;
8256 /*
8257 * we allow two log transactions at a time, use different
8258 * EXENT bit to differentiate dirty pages.
8259 */
8260 if (buf->log_index == 0)
8261 set_extent_dirty(&root->dirty_log_pages, buf->start,
8262 buf->start + buf->len - 1, GFP_NOFS);
8263 else
8264 set_extent_new(&root->dirty_log_pages, buf->start,
8265 buf->start + buf->len - 1);
8266 } else {
8267 buf->log_index = -1;
8268 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
8269 buf->start + buf->len - 1, GFP_NOFS);
8270 }
8271 trans->dirty = true;
8272 /* this returns a buffer locked for blocking */
8273 return buf;
8274 }
8275
8276 static struct btrfs_block_rsv *
8277 use_block_rsv(struct btrfs_trans_handle *trans,
8278 struct btrfs_root *root, u32 blocksize)
8279 {
8280 struct btrfs_block_rsv *block_rsv;
8281 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
8282 int ret;
8283 bool global_updated = false;
8284
8285 block_rsv = get_block_rsv(trans, root);
8286
8287 if (unlikely(block_rsv->size == 0))
8288 goto try_reserve;
8289 again:
8290 ret = block_rsv_use_bytes(block_rsv, blocksize);
8291 if (!ret)
8292 return block_rsv;
8293
8294 if (block_rsv->failfast)
8295 return ERR_PTR(ret);
8296
8297 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
8298 global_updated = true;
8299 update_global_block_rsv(root->fs_info);
8300 goto again;
8301 }
8302
8303 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
8304 static DEFINE_RATELIMIT_STATE(_rs,
8305 DEFAULT_RATELIMIT_INTERVAL * 10,
8306 /*DEFAULT_RATELIMIT_BURST*/ 1);
8307 if (__ratelimit(&_rs))
8308 WARN(1, KERN_DEBUG
8309 "BTRFS: block rsv returned %d\n", ret);
8310 }
8311 try_reserve:
8312 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
8313 BTRFS_RESERVE_NO_FLUSH);
8314 if (!ret)
8315 return block_rsv;
8316 /*
8317 * If we couldn't reserve metadata bytes try and use some from
8318 * the global reserve if its space type is the same as the global
8319 * reservation.
8320 */
8321 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
8322 block_rsv->space_info == global_rsv->space_info) {
8323 ret = block_rsv_use_bytes(global_rsv, blocksize);
8324 if (!ret)
8325 return global_rsv;
8326 }
8327 return ERR_PTR(ret);
8328 }
8329
8330 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
8331 struct btrfs_block_rsv *block_rsv, u32 blocksize)
8332 {
8333 block_rsv_add_bytes(block_rsv, blocksize, 0);
8334 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
8335 }
8336
8337 /*
8338 * finds a free extent and does all the dirty work required for allocation
8339 * returns the tree buffer or an ERR_PTR on error.
8340 */
8341 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8342 struct btrfs_root *root,
8343 u64 parent, u64 root_objectid,
8344 struct btrfs_disk_key *key, int level,
8345 u64 hint, u64 empty_size)
8346 {
8347 struct btrfs_key ins;
8348 struct btrfs_block_rsv *block_rsv;
8349 struct extent_buffer *buf;
8350 struct btrfs_delayed_extent_op *extent_op;
8351 u64 flags = 0;
8352 int ret;
8353 u32 blocksize = root->nodesize;
8354 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
8355 SKINNY_METADATA);
8356
8357 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8358 if (btrfs_test_is_dummy_root(root)) {
8359 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8360 level);
8361 if (!IS_ERR(buf))
8362 root->alloc_bytenr += blocksize;
8363 return buf;
8364 }
8365 #endif
8366
8367 block_rsv = use_block_rsv(trans, root, blocksize);
8368 if (IS_ERR(block_rsv))
8369 return ERR_CAST(block_rsv);
8370
8371 ret = btrfs_reserve_extent(root, blocksize, blocksize,
8372 empty_size, hint, &ins, 0, 0);
8373 if (ret)
8374 goto out_unuse;
8375
8376 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
8377 if (IS_ERR(buf)) {
8378 ret = PTR_ERR(buf);
8379 goto out_free_reserved;
8380 }
8381
8382 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8383 if (parent == 0)
8384 parent = ins.objectid;
8385 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8386 } else
8387 BUG_ON(parent > 0);
8388
8389 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8390 extent_op = btrfs_alloc_delayed_extent_op();
8391 if (!extent_op) {
8392 ret = -ENOMEM;
8393 goto out_free_buf;
8394 }
8395 if (key)
8396 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8397 else
8398 memset(&extent_op->key, 0, sizeof(extent_op->key));
8399 extent_op->flags_to_set = flags;
8400 extent_op->update_key = skinny_metadata ? false : true;
8401 extent_op->update_flags = true;
8402 extent_op->is_data = false;
8403 extent_op->level = level;
8404
8405 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
8406 ins.objectid, ins.offset,
8407 parent, root_objectid, level,
8408 BTRFS_ADD_DELAYED_EXTENT,
8409 extent_op);
8410 if (ret)
8411 goto out_free_delayed;
8412 }
8413 return buf;
8414
8415 out_free_delayed:
8416 btrfs_free_delayed_extent_op(extent_op);
8417 out_free_buf:
8418 free_extent_buffer(buf);
8419 out_free_reserved:
8420 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0);
8421 out_unuse:
8422 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
8423 return ERR_PTR(ret);
8424 }
8425
8426 struct walk_control {
8427 u64 refs[BTRFS_MAX_LEVEL];
8428 u64 flags[BTRFS_MAX_LEVEL];
8429 struct btrfs_key update_progress;
8430 int stage;
8431 int level;
8432 int shared_level;
8433 int update_ref;
8434 int keep_locks;
8435 int reada_slot;
8436 int reada_count;
8437 int for_reloc;
8438 };
8439
8440 #define DROP_REFERENCE 1
8441 #define UPDATE_BACKREF 2
8442
8443 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8444 struct btrfs_root *root,
8445 struct walk_control *wc,
8446 struct btrfs_path *path)
8447 {
8448 u64 bytenr;
8449 u64 generation;
8450 u64 refs;
8451 u64 flags;
8452 u32 nritems;
8453 u32 blocksize;
8454 struct btrfs_key key;
8455 struct extent_buffer *eb;
8456 int ret;
8457 int slot;
8458 int nread = 0;
8459
8460 if (path->slots[wc->level] < wc->reada_slot) {
8461 wc->reada_count = wc->reada_count * 2 / 3;
8462 wc->reada_count = max(wc->reada_count, 2);
8463 } else {
8464 wc->reada_count = wc->reada_count * 3 / 2;
8465 wc->reada_count = min_t(int, wc->reada_count,
8466 BTRFS_NODEPTRS_PER_BLOCK(root));
8467 }
8468
8469 eb = path->nodes[wc->level];
8470 nritems = btrfs_header_nritems(eb);
8471 blocksize = root->nodesize;
8472
8473 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8474 if (nread >= wc->reada_count)
8475 break;
8476
8477 cond_resched();
8478 bytenr = btrfs_node_blockptr(eb, slot);
8479 generation = btrfs_node_ptr_generation(eb, slot);
8480
8481 if (slot == path->slots[wc->level])
8482 goto reada;
8483
8484 if (wc->stage == UPDATE_BACKREF &&
8485 generation <= root->root_key.offset)
8486 continue;
8487
8488 /* We don't lock the tree block, it's OK to be racy here */
8489 ret = btrfs_lookup_extent_info(trans, root, bytenr,
8490 wc->level - 1, 1, &refs,
8491 &flags);
8492 /* We don't care about errors in readahead. */
8493 if (ret < 0)
8494 continue;
8495 BUG_ON(refs == 0);
8496
8497 if (wc->stage == DROP_REFERENCE) {
8498 if (refs == 1)
8499 goto reada;
8500
8501 if (wc->level == 1 &&
8502 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8503 continue;
8504 if (!wc->update_ref ||
8505 generation <= root->root_key.offset)
8506 continue;
8507 btrfs_node_key_to_cpu(eb, &key, slot);
8508 ret = btrfs_comp_cpu_keys(&key,
8509 &wc->update_progress);
8510 if (ret < 0)
8511 continue;
8512 } else {
8513 if (wc->level == 1 &&
8514 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8515 continue;
8516 }
8517 reada:
8518 readahead_tree_block(root, bytenr);
8519 nread++;
8520 }
8521 wc->reada_slot = slot;
8522 }
8523
8524 /*
8525 * These may not be seen by the usual inc/dec ref code so we have to
8526 * add them here.
8527 */
8528 static int record_one_subtree_extent(struct btrfs_trans_handle *trans,
8529 struct btrfs_root *root, u64 bytenr,
8530 u64 num_bytes)
8531 {
8532 struct btrfs_qgroup_extent_record *qrecord;
8533 struct btrfs_delayed_ref_root *delayed_refs;
8534
8535 qrecord = kmalloc(sizeof(*qrecord), GFP_NOFS);
8536 if (!qrecord)
8537 return -ENOMEM;
8538
8539 qrecord->bytenr = bytenr;
8540 qrecord->num_bytes = num_bytes;
8541 qrecord->old_roots = NULL;
8542
8543 delayed_refs = &trans->transaction->delayed_refs;
8544 spin_lock(&delayed_refs->lock);
8545 if (btrfs_qgroup_insert_dirty_extent(delayed_refs, qrecord))
8546 kfree(qrecord);
8547 spin_unlock(&delayed_refs->lock);
8548
8549 return 0;
8550 }
8551
8552 static int account_leaf_items(struct btrfs_trans_handle *trans,
8553 struct btrfs_root *root,
8554 struct extent_buffer *eb)
8555 {
8556 int nr = btrfs_header_nritems(eb);
8557 int i, extent_type, ret;
8558 struct btrfs_key key;
8559 struct btrfs_file_extent_item *fi;
8560 u64 bytenr, num_bytes;
8561
8562 /* We can be called directly from walk_up_proc() */
8563 if (!root->fs_info->quota_enabled)
8564 return 0;
8565
8566 for (i = 0; i < nr; i++) {
8567 btrfs_item_key_to_cpu(eb, &key, i);
8568
8569 if (key.type != BTRFS_EXTENT_DATA_KEY)
8570 continue;
8571
8572 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
8573 /* filter out non qgroup-accountable extents */
8574 extent_type = btrfs_file_extent_type(eb, fi);
8575
8576 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
8577 continue;
8578
8579 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
8580 if (!bytenr)
8581 continue;
8582
8583 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
8584
8585 ret = record_one_subtree_extent(trans, root, bytenr, num_bytes);
8586 if (ret)
8587 return ret;
8588 }
8589 return 0;
8590 }
8591
8592 /*
8593 * Walk up the tree from the bottom, freeing leaves and any interior
8594 * nodes which have had all slots visited. If a node (leaf or
8595 * interior) is freed, the node above it will have it's slot
8596 * incremented. The root node will never be freed.
8597 *
8598 * At the end of this function, we should have a path which has all
8599 * slots incremented to the next position for a search. If we need to
8600 * read a new node it will be NULL and the node above it will have the
8601 * correct slot selected for a later read.
8602 *
8603 * If we increment the root nodes slot counter past the number of
8604 * elements, 1 is returned to signal completion of the search.
8605 */
8606 static int adjust_slots_upwards(struct btrfs_root *root,
8607 struct btrfs_path *path, int root_level)
8608 {
8609 int level = 0;
8610 int nr, slot;
8611 struct extent_buffer *eb;
8612
8613 if (root_level == 0)
8614 return 1;
8615
8616 while (level <= root_level) {
8617 eb = path->nodes[level];
8618 nr = btrfs_header_nritems(eb);
8619 path->slots[level]++;
8620 slot = path->slots[level];
8621 if (slot >= nr || level == 0) {
8622 /*
8623 * Don't free the root - we will detect this
8624 * condition after our loop and return a
8625 * positive value for caller to stop walking the tree.
8626 */
8627 if (level != root_level) {
8628 btrfs_tree_unlock_rw(eb, path->locks[level]);
8629 path->locks[level] = 0;
8630
8631 free_extent_buffer(eb);
8632 path->nodes[level] = NULL;
8633 path->slots[level] = 0;
8634 }
8635 } else {
8636 /*
8637 * We have a valid slot to walk back down
8638 * from. Stop here so caller can process these
8639 * new nodes.
8640 */
8641 break;
8642 }
8643
8644 level++;
8645 }
8646
8647 eb = path->nodes[root_level];
8648 if (path->slots[root_level] >= btrfs_header_nritems(eb))
8649 return 1;
8650
8651 return 0;
8652 }
8653
8654 /*
8655 * root_eb is the subtree root and is locked before this function is called.
8656 */
8657 static int account_shared_subtree(struct btrfs_trans_handle *trans,
8658 struct btrfs_root *root,
8659 struct extent_buffer *root_eb,
8660 u64 root_gen,
8661 int root_level)
8662 {
8663 int ret = 0;
8664 int level;
8665 struct extent_buffer *eb = root_eb;
8666 struct btrfs_path *path = NULL;
8667
8668 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
8669 BUG_ON(root_eb == NULL);
8670
8671 if (!root->fs_info->quota_enabled)
8672 return 0;
8673
8674 if (!extent_buffer_uptodate(root_eb)) {
8675 ret = btrfs_read_buffer(root_eb, root_gen);
8676 if (ret)
8677 goto out;
8678 }
8679
8680 if (root_level == 0) {
8681 ret = account_leaf_items(trans, root, root_eb);
8682 goto out;
8683 }
8684
8685 path = btrfs_alloc_path();
8686 if (!path)
8687 return -ENOMEM;
8688
8689 /*
8690 * Walk down the tree. Missing extent blocks are filled in as
8691 * we go. Metadata is accounted every time we read a new
8692 * extent block.
8693 *
8694 * When we reach a leaf, we account for file extent items in it,
8695 * walk back up the tree (adjusting slot pointers as we go)
8696 * and restart the search process.
8697 */
8698 extent_buffer_get(root_eb); /* For path */
8699 path->nodes[root_level] = root_eb;
8700 path->slots[root_level] = 0;
8701 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
8702 walk_down:
8703 level = root_level;
8704 while (level >= 0) {
8705 if (path->nodes[level] == NULL) {
8706 int parent_slot;
8707 u64 child_gen;
8708 u64 child_bytenr;
8709
8710 /* We need to get child blockptr/gen from
8711 * parent before we can read it. */
8712 eb = path->nodes[level + 1];
8713 parent_slot = path->slots[level + 1];
8714 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
8715 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
8716
8717 eb = read_tree_block(root, child_bytenr, child_gen);
8718 if (IS_ERR(eb)) {
8719 ret = PTR_ERR(eb);
8720 goto out;
8721 } else if (!extent_buffer_uptodate(eb)) {
8722 free_extent_buffer(eb);
8723 ret = -EIO;
8724 goto out;
8725 }
8726
8727 path->nodes[level] = eb;
8728 path->slots[level] = 0;
8729
8730 btrfs_tree_read_lock(eb);
8731 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
8732 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
8733
8734 ret = record_one_subtree_extent(trans, root, child_bytenr,
8735 root->nodesize);
8736 if (ret)
8737 goto out;
8738 }
8739
8740 if (level == 0) {
8741 ret = account_leaf_items(trans, root, path->nodes[level]);
8742 if (ret)
8743 goto out;
8744
8745 /* Nonzero return here means we completed our search */
8746 ret = adjust_slots_upwards(root, path, root_level);
8747 if (ret)
8748 break;
8749
8750 /* Restart search with new slots */
8751 goto walk_down;
8752 }
8753
8754 level--;
8755 }
8756
8757 ret = 0;
8758 out:
8759 btrfs_free_path(path);
8760
8761 return ret;
8762 }
8763
8764 /*
8765 * helper to process tree block while walking down the tree.
8766 *
8767 * when wc->stage == UPDATE_BACKREF, this function updates
8768 * back refs for pointers in the block.
8769 *
8770 * NOTE: return value 1 means we should stop walking down.
8771 */
8772 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8773 struct btrfs_root *root,
8774 struct btrfs_path *path,
8775 struct walk_control *wc, int lookup_info)
8776 {
8777 int level = wc->level;
8778 struct extent_buffer *eb = path->nodes[level];
8779 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8780 int ret;
8781
8782 if (wc->stage == UPDATE_BACKREF &&
8783 btrfs_header_owner(eb) != root->root_key.objectid)
8784 return 1;
8785
8786 /*
8787 * when reference count of tree block is 1, it won't increase
8788 * again. once full backref flag is set, we never clear it.
8789 */
8790 if (lookup_info &&
8791 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8792 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8793 BUG_ON(!path->locks[level]);
8794 ret = btrfs_lookup_extent_info(trans, root,
8795 eb->start, level, 1,
8796 &wc->refs[level],
8797 &wc->flags[level]);
8798 BUG_ON(ret == -ENOMEM);
8799 if (ret)
8800 return ret;
8801 BUG_ON(wc->refs[level] == 0);
8802 }
8803
8804 if (wc->stage == DROP_REFERENCE) {
8805 if (wc->refs[level] > 1)
8806 return 1;
8807
8808 if (path->locks[level] && !wc->keep_locks) {
8809 btrfs_tree_unlock_rw(eb, path->locks[level]);
8810 path->locks[level] = 0;
8811 }
8812 return 0;
8813 }
8814
8815 /* wc->stage == UPDATE_BACKREF */
8816 if (!(wc->flags[level] & flag)) {
8817 BUG_ON(!path->locks[level]);
8818 ret = btrfs_inc_ref(trans, root, eb, 1);
8819 BUG_ON(ret); /* -ENOMEM */
8820 ret = btrfs_dec_ref(trans, root, eb, 0);
8821 BUG_ON(ret); /* -ENOMEM */
8822 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
8823 eb->len, flag,
8824 btrfs_header_level(eb), 0);
8825 BUG_ON(ret); /* -ENOMEM */
8826 wc->flags[level] |= flag;
8827 }
8828
8829 /*
8830 * the block is shared by multiple trees, so it's not good to
8831 * keep the tree lock
8832 */
8833 if (path->locks[level] && level > 0) {
8834 btrfs_tree_unlock_rw(eb, path->locks[level]);
8835 path->locks[level] = 0;
8836 }
8837 return 0;
8838 }
8839
8840 /*
8841 * helper to process tree block pointer.
8842 *
8843 * when wc->stage == DROP_REFERENCE, this function checks
8844 * reference count of the block pointed to. if the block
8845 * is shared and we need update back refs for the subtree
8846 * rooted at the block, this function changes wc->stage to
8847 * UPDATE_BACKREF. if the block is shared and there is no
8848 * need to update back, this function drops the reference
8849 * to the block.
8850 *
8851 * NOTE: return value 1 means we should stop walking down.
8852 */
8853 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8854 struct btrfs_root *root,
8855 struct btrfs_path *path,
8856 struct walk_control *wc, int *lookup_info)
8857 {
8858 u64 bytenr;
8859 u64 generation;
8860 u64 parent;
8861 u32 blocksize;
8862 struct btrfs_key key;
8863 struct extent_buffer *next;
8864 int level = wc->level;
8865 int reada = 0;
8866 int ret = 0;
8867 bool need_account = false;
8868
8869 generation = btrfs_node_ptr_generation(path->nodes[level],
8870 path->slots[level]);
8871 /*
8872 * if the lower level block was created before the snapshot
8873 * was created, we know there is no need to update back refs
8874 * for the subtree
8875 */
8876 if (wc->stage == UPDATE_BACKREF &&
8877 generation <= root->root_key.offset) {
8878 *lookup_info = 1;
8879 return 1;
8880 }
8881
8882 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8883 blocksize = root->nodesize;
8884
8885 next = btrfs_find_tree_block(root->fs_info, bytenr);
8886 if (!next) {
8887 next = btrfs_find_create_tree_block(root, bytenr);
8888 if (IS_ERR(next))
8889 return PTR_ERR(next);
8890
8891 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8892 level - 1);
8893 reada = 1;
8894 }
8895 btrfs_tree_lock(next);
8896 btrfs_set_lock_blocking(next);
8897
8898 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
8899 &wc->refs[level - 1],
8900 &wc->flags[level - 1]);
8901 if (ret < 0) {
8902 btrfs_tree_unlock(next);
8903 return ret;
8904 }
8905
8906 if (unlikely(wc->refs[level - 1] == 0)) {
8907 btrfs_err(root->fs_info, "Missing references.");
8908 BUG();
8909 }
8910 *lookup_info = 0;
8911
8912 if (wc->stage == DROP_REFERENCE) {
8913 if (wc->refs[level - 1] > 1) {
8914 need_account = true;
8915 if (level == 1 &&
8916 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8917 goto skip;
8918
8919 if (!wc->update_ref ||
8920 generation <= root->root_key.offset)
8921 goto skip;
8922
8923 btrfs_node_key_to_cpu(path->nodes[level], &key,
8924 path->slots[level]);
8925 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8926 if (ret < 0)
8927 goto skip;
8928
8929 wc->stage = UPDATE_BACKREF;
8930 wc->shared_level = level - 1;
8931 }
8932 } else {
8933 if (level == 1 &&
8934 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8935 goto skip;
8936 }
8937
8938 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8939 btrfs_tree_unlock(next);
8940 free_extent_buffer(next);
8941 next = NULL;
8942 *lookup_info = 1;
8943 }
8944
8945 if (!next) {
8946 if (reada && level == 1)
8947 reada_walk_down(trans, root, wc, path);
8948 next = read_tree_block(root, bytenr, generation);
8949 if (IS_ERR(next)) {
8950 return PTR_ERR(next);
8951 } else if (!extent_buffer_uptodate(next)) {
8952 free_extent_buffer(next);
8953 return -EIO;
8954 }
8955 btrfs_tree_lock(next);
8956 btrfs_set_lock_blocking(next);
8957 }
8958
8959 level--;
8960 BUG_ON(level != btrfs_header_level(next));
8961 path->nodes[level] = next;
8962 path->slots[level] = 0;
8963 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8964 wc->level = level;
8965 if (wc->level == 1)
8966 wc->reada_slot = 0;
8967 return 0;
8968 skip:
8969 wc->refs[level - 1] = 0;
8970 wc->flags[level - 1] = 0;
8971 if (wc->stage == DROP_REFERENCE) {
8972 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8973 parent = path->nodes[level]->start;
8974 } else {
8975 BUG_ON(root->root_key.objectid !=
8976 btrfs_header_owner(path->nodes[level]));
8977 parent = 0;
8978 }
8979
8980 if (need_account) {
8981 ret = account_shared_subtree(trans, root, next,
8982 generation, level - 1);
8983 if (ret) {
8984 btrfs_err_rl(root->fs_info,
8985 "Error "
8986 "%d accounting shared subtree. Quota "
8987 "is out of sync, rescan required.",
8988 ret);
8989 }
8990 }
8991 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
8992 root->root_key.objectid, level - 1, 0);
8993 BUG_ON(ret); /* -ENOMEM */
8994 }
8995 btrfs_tree_unlock(next);
8996 free_extent_buffer(next);
8997 *lookup_info = 1;
8998 return 1;
8999 }
9000
9001 /*
9002 * helper to process tree block while walking up the tree.
9003 *
9004 * when wc->stage == DROP_REFERENCE, this function drops
9005 * reference count on the block.
9006 *
9007 * when wc->stage == UPDATE_BACKREF, this function changes
9008 * wc->stage back to DROP_REFERENCE if we changed wc->stage
9009 * to UPDATE_BACKREF previously while processing the block.
9010 *
9011 * NOTE: return value 1 means we should stop walking up.
9012 */
9013 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
9014 struct btrfs_root *root,
9015 struct btrfs_path *path,
9016 struct walk_control *wc)
9017 {
9018 int ret;
9019 int level = wc->level;
9020 struct extent_buffer *eb = path->nodes[level];
9021 u64 parent = 0;
9022
9023 if (wc->stage == UPDATE_BACKREF) {
9024 BUG_ON(wc->shared_level < level);
9025 if (level < wc->shared_level)
9026 goto out;
9027
9028 ret = find_next_key(path, level + 1, &wc->update_progress);
9029 if (ret > 0)
9030 wc->update_ref = 0;
9031
9032 wc->stage = DROP_REFERENCE;
9033 wc->shared_level = -1;
9034 path->slots[level] = 0;
9035
9036 /*
9037 * check reference count again if the block isn't locked.
9038 * we should start walking down the tree again if reference
9039 * count is one.
9040 */
9041 if (!path->locks[level]) {
9042 BUG_ON(level == 0);
9043 btrfs_tree_lock(eb);
9044 btrfs_set_lock_blocking(eb);
9045 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9046
9047 ret = btrfs_lookup_extent_info(trans, root,
9048 eb->start, level, 1,
9049 &wc->refs[level],
9050 &wc->flags[level]);
9051 if (ret < 0) {
9052 btrfs_tree_unlock_rw(eb, path->locks[level]);
9053 path->locks[level] = 0;
9054 return ret;
9055 }
9056 BUG_ON(wc->refs[level] == 0);
9057 if (wc->refs[level] == 1) {
9058 btrfs_tree_unlock_rw(eb, path->locks[level]);
9059 path->locks[level] = 0;
9060 return 1;
9061 }
9062 }
9063 }
9064
9065 /* wc->stage == DROP_REFERENCE */
9066 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
9067
9068 if (wc->refs[level] == 1) {
9069 if (level == 0) {
9070 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9071 ret = btrfs_dec_ref(trans, root, eb, 1);
9072 else
9073 ret = btrfs_dec_ref(trans, root, eb, 0);
9074 BUG_ON(ret); /* -ENOMEM */
9075 ret = account_leaf_items(trans, root, eb);
9076 if (ret) {
9077 btrfs_err_rl(root->fs_info,
9078 "error "
9079 "%d accounting leaf items. Quota "
9080 "is out of sync, rescan required.",
9081 ret);
9082 }
9083 }
9084 /* make block locked assertion in clean_tree_block happy */
9085 if (!path->locks[level] &&
9086 btrfs_header_generation(eb) == trans->transid) {
9087 btrfs_tree_lock(eb);
9088 btrfs_set_lock_blocking(eb);
9089 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9090 }
9091 clean_tree_block(trans, root->fs_info, eb);
9092 }
9093
9094 if (eb == root->node) {
9095 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9096 parent = eb->start;
9097 else
9098 BUG_ON(root->root_key.objectid !=
9099 btrfs_header_owner(eb));
9100 } else {
9101 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9102 parent = path->nodes[level + 1]->start;
9103 else
9104 BUG_ON(root->root_key.objectid !=
9105 btrfs_header_owner(path->nodes[level + 1]));
9106 }
9107
9108 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
9109 out:
9110 wc->refs[level] = 0;
9111 wc->flags[level] = 0;
9112 return 0;
9113 }
9114
9115 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
9116 struct btrfs_root *root,
9117 struct btrfs_path *path,
9118 struct walk_control *wc)
9119 {
9120 int level = wc->level;
9121 int lookup_info = 1;
9122 int ret;
9123
9124 while (level >= 0) {
9125 ret = walk_down_proc(trans, root, path, wc, lookup_info);
9126 if (ret > 0)
9127 break;
9128
9129 if (level == 0)
9130 break;
9131
9132 if (path->slots[level] >=
9133 btrfs_header_nritems(path->nodes[level]))
9134 break;
9135
9136 ret = do_walk_down(trans, root, path, wc, &lookup_info);
9137 if (ret > 0) {
9138 path->slots[level]++;
9139 continue;
9140 } else if (ret < 0)
9141 return ret;
9142 level = wc->level;
9143 }
9144 return 0;
9145 }
9146
9147 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
9148 struct btrfs_root *root,
9149 struct btrfs_path *path,
9150 struct walk_control *wc, int max_level)
9151 {
9152 int level = wc->level;
9153 int ret;
9154
9155 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
9156 while (level < max_level && path->nodes[level]) {
9157 wc->level = level;
9158 if (path->slots[level] + 1 <
9159 btrfs_header_nritems(path->nodes[level])) {
9160 path->slots[level]++;
9161 return 0;
9162 } else {
9163 ret = walk_up_proc(trans, root, path, wc);
9164 if (ret > 0)
9165 return 0;
9166
9167 if (path->locks[level]) {
9168 btrfs_tree_unlock_rw(path->nodes[level],
9169 path->locks[level]);
9170 path->locks[level] = 0;
9171 }
9172 free_extent_buffer(path->nodes[level]);
9173 path->nodes[level] = NULL;
9174 level++;
9175 }
9176 }
9177 return 1;
9178 }
9179
9180 /*
9181 * drop a subvolume tree.
9182 *
9183 * this function traverses the tree freeing any blocks that only
9184 * referenced by the tree.
9185 *
9186 * when a shared tree block is found. this function decreases its
9187 * reference count by one. if update_ref is true, this function
9188 * also make sure backrefs for the shared block and all lower level
9189 * blocks are properly updated.
9190 *
9191 * If called with for_reloc == 0, may exit early with -EAGAIN
9192 */
9193 int btrfs_drop_snapshot(struct btrfs_root *root,
9194 struct btrfs_block_rsv *block_rsv, int update_ref,
9195 int for_reloc)
9196 {
9197 struct btrfs_path *path;
9198 struct btrfs_trans_handle *trans;
9199 struct btrfs_root *tree_root = root->fs_info->tree_root;
9200 struct btrfs_root_item *root_item = &root->root_item;
9201 struct walk_control *wc;
9202 struct btrfs_key key;
9203 int err = 0;
9204 int ret;
9205 int level;
9206 bool root_dropped = false;
9207
9208 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
9209
9210 path = btrfs_alloc_path();
9211 if (!path) {
9212 err = -ENOMEM;
9213 goto out;
9214 }
9215
9216 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9217 if (!wc) {
9218 btrfs_free_path(path);
9219 err = -ENOMEM;
9220 goto out;
9221 }
9222
9223 trans = btrfs_start_transaction(tree_root, 0);
9224 if (IS_ERR(trans)) {
9225 err = PTR_ERR(trans);
9226 goto out_free;
9227 }
9228
9229 if (block_rsv)
9230 trans->block_rsv = block_rsv;
9231
9232 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
9233 level = btrfs_header_level(root->node);
9234 path->nodes[level] = btrfs_lock_root_node(root);
9235 btrfs_set_lock_blocking(path->nodes[level]);
9236 path->slots[level] = 0;
9237 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9238 memset(&wc->update_progress, 0,
9239 sizeof(wc->update_progress));
9240 } else {
9241 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
9242 memcpy(&wc->update_progress, &key,
9243 sizeof(wc->update_progress));
9244
9245 level = root_item->drop_level;
9246 BUG_ON(level == 0);
9247 path->lowest_level = level;
9248 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
9249 path->lowest_level = 0;
9250 if (ret < 0) {
9251 err = ret;
9252 goto out_end_trans;
9253 }
9254 WARN_ON(ret > 0);
9255
9256 /*
9257 * unlock our path, this is safe because only this
9258 * function is allowed to delete this snapshot
9259 */
9260 btrfs_unlock_up_safe(path, 0);
9261
9262 level = btrfs_header_level(root->node);
9263 while (1) {
9264 btrfs_tree_lock(path->nodes[level]);
9265 btrfs_set_lock_blocking(path->nodes[level]);
9266 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9267
9268 ret = btrfs_lookup_extent_info(trans, root,
9269 path->nodes[level]->start,
9270 level, 1, &wc->refs[level],
9271 &wc->flags[level]);
9272 if (ret < 0) {
9273 err = ret;
9274 goto out_end_trans;
9275 }
9276 BUG_ON(wc->refs[level] == 0);
9277
9278 if (level == root_item->drop_level)
9279 break;
9280
9281 btrfs_tree_unlock(path->nodes[level]);
9282 path->locks[level] = 0;
9283 WARN_ON(wc->refs[level] != 1);
9284 level--;
9285 }
9286 }
9287
9288 wc->level = level;
9289 wc->shared_level = -1;
9290 wc->stage = DROP_REFERENCE;
9291 wc->update_ref = update_ref;
9292 wc->keep_locks = 0;
9293 wc->for_reloc = for_reloc;
9294 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
9295
9296 while (1) {
9297
9298 ret = walk_down_tree(trans, root, path, wc);
9299 if (ret < 0) {
9300 err = ret;
9301 break;
9302 }
9303
9304 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
9305 if (ret < 0) {
9306 err = ret;
9307 break;
9308 }
9309
9310 if (ret > 0) {
9311 BUG_ON(wc->stage != DROP_REFERENCE);
9312 break;
9313 }
9314
9315 if (wc->stage == DROP_REFERENCE) {
9316 level = wc->level;
9317 btrfs_node_key(path->nodes[level],
9318 &root_item->drop_progress,
9319 path->slots[level]);
9320 root_item->drop_level = level;
9321 }
9322
9323 BUG_ON(wc->level == 0);
9324 if (btrfs_should_end_transaction(trans, tree_root) ||
9325 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
9326 ret = btrfs_update_root(trans, tree_root,
9327 &root->root_key,
9328 root_item);
9329 if (ret) {
9330 btrfs_abort_transaction(trans, tree_root, ret);
9331 err = ret;
9332 goto out_end_trans;
9333 }
9334
9335 btrfs_end_transaction_throttle(trans, tree_root);
9336 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
9337 pr_debug("BTRFS: drop snapshot early exit\n");
9338 err = -EAGAIN;
9339 goto out_free;
9340 }
9341
9342 trans = btrfs_start_transaction(tree_root, 0);
9343 if (IS_ERR(trans)) {
9344 err = PTR_ERR(trans);
9345 goto out_free;
9346 }
9347 if (block_rsv)
9348 trans->block_rsv = block_rsv;
9349 }
9350 }
9351 btrfs_release_path(path);
9352 if (err)
9353 goto out_end_trans;
9354
9355 ret = btrfs_del_root(trans, tree_root, &root->root_key);
9356 if (ret) {
9357 btrfs_abort_transaction(trans, tree_root, ret);
9358 goto out_end_trans;
9359 }
9360
9361 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9362 ret = btrfs_find_root(tree_root, &root->root_key, path,
9363 NULL, NULL);
9364 if (ret < 0) {
9365 btrfs_abort_transaction(trans, tree_root, ret);
9366 err = ret;
9367 goto out_end_trans;
9368 } else if (ret > 0) {
9369 /* if we fail to delete the orphan item this time
9370 * around, it'll get picked up the next time.
9371 *
9372 * The most common failure here is just -ENOENT.
9373 */
9374 btrfs_del_orphan_item(trans, tree_root,
9375 root->root_key.objectid);
9376 }
9377 }
9378
9379 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9380 btrfs_add_dropped_root(trans, root);
9381 } else {
9382 free_extent_buffer(root->node);
9383 free_extent_buffer(root->commit_root);
9384 btrfs_put_fs_root(root);
9385 }
9386 root_dropped = true;
9387 out_end_trans:
9388 btrfs_end_transaction_throttle(trans, tree_root);
9389 out_free:
9390 kfree(wc);
9391 btrfs_free_path(path);
9392 out:
9393 /*
9394 * So if we need to stop dropping the snapshot for whatever reason we
9395 * need to make sure to add it back to the dead root list so that we
9396 * keep trying to do the work later. This also cleans up roots if we
9397 * don't have it in the radix (like when we recover after a power fail
9398 * or unmount) so we don't leak memory.
9399 */
9400 if (!for_reloc && root_dropped == false)
9401 btrfs_add_dead_root(root);
9402 if (err && err != -EAGAIN)
9403 btrfs_handle_fs_error(root->fs_info, err, NULL);
9404 return err;
9405 }
9406
9407 /*
9408 * drop subtree rooted at tree block 'node'.
9409 *
9410 * NOTE: this function will unlock and release tree block 'node'
9411 * only used by relocation code
9412 */
9413 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9414 struct btrfs_root *root,
9415 struct extent_buffer *node,
9416 struct extent_buffer *parent)
9417 {
9418 struct btrfs_path *path;
9419 struct walk_control *wc;
9420 int level;
9421 int parent_level;
9422 int ret = 0;
9423 int wret;
9424
9425 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9426
9427 path = btrfs_alloc_path();
9428 if (!path)
9429 return -ENOMEM;
9430
9431 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9432 if (!wc) {
9433 btrfs_free_path(path);
9434 return -ENOMEM;
9435 }
9436
9437 btrfs_assert_tree_locked(parent);
9438 parent_level = btrfs_header_level(parent);
9439 extent_buffer_get(parent);
9440 path->nodes[parent_level] = parent;
9441 path->slots[parent_level] = btrfs_header_nritems(parent);
9442
9443 btrfs_assert_tree_locked(node);
9444 level = btrfs_header_level(node);
9445 path->nodes[level] = node;
9446 path->slots[level] = 0;
9447 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9448
9449 wc->refs[parent_level] = 1;
9450 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9451 wc->level = level;
9452 wc->shared_level = -1;
9453 wc->stage = DROP_REFERENCE;
9454 wc->update_ref = 0;
9455 wc->keep_locks = 1;
9456 wc->for_reloc = 1;
9457 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
9458
9459 while (1) {
9460 wret = walk_down_tree(trans, root, path, wc);
9461 if (wret < 0) {
9462 ret = wret;
9463 break;
9464 }
9465
9466 wret = walk_up_tree(trans, root, path, wc, parent_level);
9467 if (wret < 0)
9468 ret = wret;
9469 if (wret != 0)
9470 break;
9471 }
9472
9473 kfree(wc);
9474 btrfs_free_path(path);
9475 return ret;
9476 }
9477
9478 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
9479 {
9480 u64 num_devices;
9481 u64 stripped;
9482
9483 /*
9484 * if restripe for this chunk_type is on pick target profile and
9485 * return, otherwise do the usual balance
9486 */
9487 stripped = get_restripe_target(root->fs_info, flags);
9488 if (stripped)
9489 return extended_to_chunk(stripped);
9490
9491 num_devices = root->fs_info->fs_devices->rw_devices;
9492
9493 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9494 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9495 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9496
9497 if (num_devices == 1) {
9498 stripped |= BTRFS_BLOCK_GROUP_DUP;
9499 stripped = flags & ~stripped;
9500
9501 /* turn raid0 into single device chunks */
9502 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9503 return stripped;
9504
9505 /* turn mirroring into duplication */
9506 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9507 BTRFS_BLOCK_GROUP_RAID10))
9508 return stripped | BTRFS_BLOCK_GROUP_DUP;
9509 } else {
9510 /* they already had raid on here, just return */
9511 if (flags & stripped)
9512 return flags;
9513
9514 stripped |= BTRFS_BLOCK_GROUP_DUP;
9515 stripped = flags & ~stripped;
9516
9517 /* switch duplicated blocks with raid1 */
9518 if (flags & BTRFS_BLOCK_GROUP_DUP)
9519 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9520
9521 /* this is drive concat, leave it alone */
9522 }
9523
9524 return flags;
9525 }
9526
9527 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9528 {
9529 struct btrfs_space_info *sinfo = cache->space_info;
9530 u64 num_bytes;
9531 u64 min_allocable_bytes;
9532 int ret = -ENOSPC;
9533
9534 /*
9535 * We need some metadata space and system metadata space for
9536 * allocating chunks in some corner cases until we force to set
9537 * it to be readonly.
9538 */
9539 if ((sinfo->flags &
9540 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9541 !force)
9542 min_allocable_bytes = SZ_1M;
9543 else
9544 min_allocable_bytes = 0;
9545
9546 spin_lock(&sinfo->lock);
9547 spin_lock(&cache->lock);
9548
9549 if (cache->ro) {
9550 cache->ro++;
9551 ret = 0;
9552 goto out;
9553 }
9554
9555 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9556 cache->bytes_super - btrfs_block_group_used(&cache->item);
9557
9558 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
9559 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
9560 min_allocable_bytes <= sinfo->total_bytes) {
9561 sinfo->bytes_readonly += num_bytes;
9562 cache->ro++;
9563 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9564 ret = 0;
9565 }
9566 out:
9567 spin_unlock(&cache->lock);
9568 spin_unlock(&sinfo->lock);
9569 return ret;
9570 }
9571
9572 int btrfs_inc_block_group_ro(struct btrfs_root *root,
9573 struct btrfs_block_group_cache *cache)
9574
9575 {
9576 struct btrfs_trans_handle *trans;
9577 u64 alloc_flags;
9578 int ret;
9579
9580 again:
9581 trans = btrfs_join_transaction(root);
9582 if (IS_ERR(trans))
9583 return PTR_ERR(trans);
9584
9585 /*
9586 * we're not allowed to set block groups readonly after the dirty
9587 * block groups cache has started writing. If it already started,
9588 * back off and let this transaction commit
9589 */
9590 mutex_lock(&root->fs_info->ro_block_group_mutex);
9591 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9592 u64 transid = trans->transid;
9593
9594 mutex_unlock(&root->fs_info->ro_block_group_mutex);
9595 btrfs_end_transaction(trans, root);
9596
9597 ret = btrfs_wait_for_commit(root, transid);
9598 if (ret)
9599 return ret;
9600 goto again;
9601 }
9602
9603 /*
9604 * if we are changing raid levels, try to allocate a corresponding
9605 * block group with the new raid level.
9606 */
9607 alloc_flags = update_block_group_flags(root, cache->flags);
9608 if (alloc_flags != cache->flags) {
9609 ret = do_chunk_alloc(trans, root, alloc_flags,
9610 CHUNK_ALLOC_FORCE);
9611 /*
9612 * ENOSPC is allowed here, we may have enough space
9613 * already allocated at the new raid level to
9614 * carry on
9615 */
9616 if (ret == -ENOSPC)
9617 ret = 0;
9618 if (ret < 0)
9619 goto out;
9620 }
9621
9622 ret = inc_block_group_ro(cache, 0);
9623 if (!ret)
9624 goto out;
9625 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
9626 ret = do_chunk_alloc(trans, root, alloc_flags,
9627 CHUNK_ALLOC_FORCE);
9628 if (ret < 0)
9629 goto out;
9630 ret = inc_block_group_ro(cache, 0);
9631 out:
9632 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9633 alloc_flags = update_block_group_flags(root, cache->flags);
9634 lock_chunks(root->fs_info->chunk_root);
9635 check_system_chunk(trans, root, alloc_flags);
9636 unlock_chunks(root->fs_info->chunk_root);
9637 }
9638 mutex_unlock(&root->fs_info->ro_block_group_mutex);
9639
9640 btrfs_end_transaction(trans, root);
9641 return ret;
9642 }
9643
9644 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
9645 struct btrfs_root *root, u64 type)
9646 {
9647 u64 alloc_flags = get_alloc_profile(root, type);
9648 return do_chunk_alloc(trans, root, alloc_flags,
9649 CHUNK_ALLOC_FORCE);
9650 }
9651
9652 /*
9653 * helper to account the unused space of all the readonly block group in the
9654 * space_info. takes mirrors into account.
9655 */
9656 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9657 {
9658 struct btrfs_block_group_cache *block_group;
9659 u64 free_bytes = 0;
9660 int factor;
9661
9662 /* It's df, we don't care if it's racy */
9663 if (list_empty(&sinfo->ro_bgs))
9664 return 0;
9665
9666 spin_lock(&sinfo->lock);
9667 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9668 spin_lock(&block_group->lock);
9669
9670 if (!block_group->ro) {
9671 spin_unlock(&block_group->lock);
9672 continue;
9673 }
9674
9675 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
9676 BTRFS_BLOCK_GROUP_RAID10 |
9677 BTRFS_BLOCK_GROUP_DUP))
9678 factor = 2;
9679 else
9680 factor = 1;
9681
9682 free_bytes += (block_group->key.offset -
9683 btrfs_block_group_used(&block_group->item)) *
9684 factor;
9685
9686 spin_unlock(&block_group->lock);
9687 }
9688 spin_unlock(&sinfo->lock);
9689
9690 return free_bytes;
9691 }
9692
9693 void btrfs_dec_block_group_ro(struct btrfs_root *root,
9694 struct btrfs_block_group_cache *cache)
9695 {
9696 struct btrfs_space_info *sinfo = cache->space_info;
9697 u64 num_bytes;
9698
9699 BUG_ON(!cache->ro);
9700
9701 spin_lock(&sinfo->lock);
9702 spin_lock(&cache->lock);
9703 if (!--cache->ro) {
9704 num_bytes = cache->key.offset - cache->reserved -
9705 cache->pinned - cache->bytes_super -
9706 btrfs_block_group_used(&cache->item);
9707 sinfo->bytes_readonly -= num_bytes;
9708 list_del_init(&cache->ro_list);
9709 }
9710 spin_unlock(&cache->lock);
9711 spin_unlock(&sinfo->lock);
9712 }
9713
9714 /*
9715 * checks to see if its even possible to relocate this block group.
9716 *
9717 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9718 * ok to go ahead and try.
9719 */
9720 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
9721 {
9722 struct btrfs_block_group_cache *block_group;
9723 struct btrfs_space_info *space_info;
9724 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
9725 struct btrfs_device *device;
9726 struct btrfs_trans_handle *trans;
9727 u64 min_free;
9728 u64 dev_min = 1;
9729 u64 dev_nr = 0;
9730 u64 target;
9731 int debug;
9732 int index;
9733 int full = 0;
9734 int ret = 0;
9735
9736 debug = btrfs_test_opt(root, ENOSPC_DEBUG);
9737
9738 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
9739
9740 /* odd, couldn't find the block group, leave it alone */
9741 if (!block_group) {
9742 if (debug)
9743 btrfs_warn(root->fs_info,
9744 "can't find block group for bytenr %llu",
9745 bytenr);
9746 return -1;
9747 }
9748
9749 min_free = btrfs_block_group_used(&block_group->item);
9750
9751 /* no bytes used, we're good */
9752 if (!min_free)
9753 goto out;
9754
9755 space_info = block_group->space_info;
9756 spin_lock(&space_info->lock);
9757
9758 full = space_info->full;
9759
9760 /*
9761 * if this is the last block group we have in this space, we can't
9762 * relocate it unless we're able to allocate a new chunk below.
9763 *
9764 * Otherwise, we need to make sure we have room in the space to handle
9765 * all of the extents from this block group. If we can, we're good
9766 */
9767 if ((space_info->total_bytes != block_group->key.offset) &&
9768 (space_info->bytes_used + space_info->bytes_reserved +
9769 space_info->bytes_pinned + space_info->bytes_readonly +
9770 min_free < space_info->total_bytes)) {
9771 spin_unlock(&space_info->lock);
9772 goto out;
9773 }
9774 spin_unlock(&space_info->lock);
9775
9776 /*
9777 * ok we don't have enough space, but maybe we have free space on our
9778 * devices to allocate new chunks for relocation, so loop through our
9779 * alloc devices and guess if we have enough space. if this block
9780 * group is going to be restriped, run checks against the target
9781 * profile instead of the current one.
9782 */
9783 ret = -1;
9784
9785 /*
9786 * index:
9787 * 0: raid10
9788 * 1: raid1
9789 * 2: dup
9790 * 3: raid0
9791 * 4: single
9792 */
9793 target = get_restripe_target(root->fs_info, block_group->flags);
9794 if (target) {
9795 index = __get_raid_index(extended_to_chunk(target));
9796 } else {
9797 /*
9798 * this is just a balance, so if we were marked as full
9799 * we know there is no space for a new chunk
9800 */
9801 if (full) {
9802 if (debug)
9803 btrfs_warn(root->fs_info,
9804 "no space to alloc new chunk for block group %llu",
9805 block_group->key.objectid);
9806 goto out;
9807 }
9808
9809 index = get_block_group_index(block_group);
9810 }
9811
9812 if (index == BTRFS_RAID_RAID10) {
9813 dev_min = 4;
9814 /* Divide by 2 */
9815 min_free >>= 1;
9816 } else if (index == BTRFS_RAID_RAID1) {
9817 dev_min = 2;
9818 } else if (index == BTRFS_RAID_DUP) {
9819 /* Multiply by 2 */
9820 min_free <<= 1;
9821 } else if (index == BTRFS_RAID_RAID0) {
9822 dev_min = fs_devices->rw_devices;
9823 min_free = div64_u64(min_free, dev_min);
9824 }
9825
9826 /* We need to do this so that we can look at pending chunks */
9827 trans = btrfs_join_transaction(root);
9828 if (IS_ERR(trans)) {
9829 ret = PTR_ERR(trans);
9830 goto out;
9831 }
9832
9833 mutex_lock(&root->fs_info->chunk_mutex);
9834 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9835 u64 dev_offset;
9836
9837 /*
9838 * check to make sure we can actually find a chunk with enough
9839 * space to fit our block group in.
9840 */
9841 if (device->total_bytes > device->bytes_used + min_free &&
9842 !device->is_tgtdev_for_dev_replace) {
9843 ret = find_free_dev_extent(trans, device, min_free,
9844 &dev_offset, NULL);
9845 if (!ret)
9846 dev_nr++;
9847
9848 if (dev_nr >= dev_min)
9849 break;
9850
9851 ret = -1;
9852 }
9853 }
9854 if (debug && ret == -1)
9855 btrfs_warn(root->fs_info,
9856 "no space to allocate a new chunk for block group %llu",
9857 block_group->key.objectid);
9858 mutex_unlock(&root->fs_info->chunk_mutex);
9859 btrfs_end_transaction(trans, root);
9860 out:
9861 btrfs_put_block_group(block_group);
9862 return ret;
9863 }
9864
9865 static int find_first_block_group(struct btrfs_root *root,
9866 struct btrfs_path *path, struct btrfs_key *key)
9867 {
9868 int ret = 0;
9869 struct btrfs_key found_key;
9870 struct extent_buffer *leaf;
9871 int slot;
9872
9873 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9874 if (ret < 0)
9875 goto out;
9876
9877 while (1) {
9878 slot = path->slots[0];
9879 leaf = path->nodes[0];
9880 if (slot >= btrfs_header_nritems(leaf)) {
9881 ret = btrfs_next_leaf(root, path);
9882 if (ret == 0)
9883 continue;
9884 if (ret < 0)
9885 goto out;
9886 break;
9887 }
9888 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9889
9890 if (found_key.objectid >= key->objectid &&
9891 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9892 struct extent_map_tree *em_tree;
9893 struct extent_map *em;
9894
9895 em_tree = &root->fs_info->mapping_tree.map_tree;
9896 read_lock(&em_tree->lock);
9897 em = lookup_extent_mapping(em_tree, found_key.objectid,
9898 found_key.offset);
9899 read_unlock(&em_tree->lock);
9900 if (!em) {
9901 btrfs_err(root->fs_info,
9902 "logical %llu len %llu found bg but no related chunk",
9903 found_key.objectid, found_key.offset);
9904 ret = -ENOENT;
9905 } else {
9906 ret = 0;
9907 }
9908 goto out;
9909 }
9910 path->slots[0]++;
9911 }
9912 out:
9913 return ret;
9914 }
9915
9916 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9917 {
9918 struct btrfs_block_group_cache *block_group;
9919 u64 last = 0;
9920
9921 while (1) {
9922 struct inode *inode;
9923
9924 block_group = btrfs_lookup_first_block_group(info, last);
9925 while (block_group) {
9926 spin_lock(&block_group->lock);
9927 if (block_group->iref)
9928 break;
9929 spin_unlock(&block_group->lock);
9930 block_group = next_block_group(info->tree_root,
9931 block_group);
9932 }
9933 if (!block_group) {
9934 if (last == 0)
9935 break;
9936 last = 0;
9937 continue;
9938 }
9939
9940 inode = block_group->inode;
9941 block_group->iref = 0;
9942 block_group->inode = NULL;
9943 spin_unlock(&block_group->lock);
9944 iput(inode);
9945 last = block_group->key.objectid + block_group->key.offset;
9946 btrfs_put_block_group(block_group);
9947 }
9948 }
9949
9950 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9951 {
9952 struct btrfs_block_group_cache *block_group;
9953 struct btrfs_space_info *space_info;
9954 struct btrfs_caching_control *caching_ctl;
9955 struct rb_node *n;
9956
9957 down_write(&info->commit_root_sem);
9958 while (!list_empty(&info->caching_block_groups)) {
9959 caching_ctl = list_entry(info->caching_block_groups.next,
9960 struct btrfs_caching_control, list);
9961 list_del(&caching_ctl->list);
9962 put_caching_control(caching_ctl);
9963 }
9964 up_write(&info->commit_root_sem);
9965
9966 spin_lock(&info->unused_bgs_lock);
9967 while (!list_empty(&info->unused_bgs)) {
9968 block_group = list_first_entry(&info->unused_bgs,
9969 struct btrfs_block_group_cache,
9970 bg_list);
9971 list_del_init(&block_group->bg_list);
9972 btrfs_put_block_group(block_group);
9973 }
9974 spin_unlock(&info->unused_bgs_lock);
9975
9976 spin_lock(&info->block_group_cache_lock);
9977 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9978 block_group = rb_entry(n, struct btrfs_block_group_cache,
9979 cache_node);
9980 rb_erase(&block_group->cache_node,
9981 &info->block_group_cache_tree);
9982 RB_CLEAR_NODE(&block_group->cache_node);
9983 spin_unlock(&info->block_group_cache_lock);
9984
9985 down_write(&block_group->space_info->groups_sem);
9986 list_del(&block_group->list);
9987 up_write(&block_group->space_info->groups_sem);
9988
9989 if (block_group->cached == BTRFS_CACHE_STARTED)
9990 wait_block_group_cache_done(block_group);
9991
9992 /*
9993 * We haven't cached this block group, which means we could
9994 * possibly have excluded extents on this block group.
9995 */
9996 if (block_group->cached == BTRFS_CACHE_NO ||
9997 block_group->cached == BTRFS_CACHE_ERROR)
9998 free_excluded_extents(info->extent_root, block_group);
9999
10000 btrfs_remove_free_space_cache(block_group);
10001 btrfs_put_block_group(block_group);
10002
10003 spin_lock(&info->block_group_cache_lock);
10004 }
10005 spin_unlock(&info->block_group_cache_lock);
10006
10007 /* now that all the block groups are freed, go through and
10008 * free all the space_info structs. This is only called during
10009 * the final stages of unmount, and so we know nobody is
10010 * using them. We call synchronize_rcu() once before we start,
10011 * just to be on the safe side.
10012 */
10013 synchronize_rcu();
10014
10015 release_global_block_rsv(info);
10016
10017 while (!list_empty(&info->space_info)) {
10018 int i;
10019
10020 space_info = list_entry(info->space_info.next,
10021 struct btrfs_space_info,
10022 list);
10023
10024 /*
10025 * Do not hide this behind enospc_debug, this is actually
10026 * important and indicates a real bug if this happens.
10027 */
10028 if (WARN_ON(space_info->bytes_pinned > 0 ||
10029 space_info->bytes_reserved > 0 ||
10030 space_info->bytes_may_use > 0))
10031 dump_space_info(space_info, 0, 0);
10032 list_del(&space_info->list);
10033 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
10034 struct kobject *kobj;
10035 kobj = space_info->block_group_kobjs[i];
10036 space_info->block_group_kobjs[i] = NULL;
10037 if (kobj) {
10038 kobject_del(kobj);
10039 kobject_put(kobj);
10040 }
10041 }
10042 kobject_del(&space_info->kobj);
10043 kobject_put(&space_info->kobj);
10044 }
10045 return 0;
10046 }
10047
10048 static void __link_block_group(struct btrfs_space_info *space_info,
10049 struct btrfs_block_group_cache *cache)
10050 {
10051 int index = get_block_group_index(cache);
10052 bool first = false;
10053
10054 down_write(&space_info->groups_sem);
10055 if (list_empty(&space_info->block_groups[index]))
10056 first = true;
10057 list_add_tail(&cache->list, &space_info->block_groups[index]);
10058 up_write(&space_info->groups_sem);
10059
10060 if (first) {
10061 struct raid_kobject *rkobj;
10062 int ret;
10063
10064 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
10065 if (!rkobj)
10066 goto out_err;
10067 rkobj->raid_type = index;
10068 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
10069 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
10070 "%s", get_raid_name(index));
10071 if (ret) {
10072 kobject_put(&rkobj->kobj);
10073 goto out_err;
10074 }
10075 space_info->block_group_kobjs[index] = &rkobj->kobj;
10076 }
10077
10078 return;
10079 out_err:
10080 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
10081 }
10082
10083 static struct btrfs_block_group_cache *
10084 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
10085 {
10086 struct btrfs_block_group_cache *cache;
10087
10088 cache = kzalloc(sizeof(*cache), GFP_NOFS);
10089 if (!cache)
10090 return NULL;
10091
10092 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
10093 GFP_NOFS);
10094 if (!cache->free_space_ctl) {
10095 kfree(cache);
10096 return NULL;
10097 }
10098
10099 cache->key.objectid = start;
10100 cache->key.offset = size;
10101 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10102
10103 cache->sectorsize = root->sectorsize;
10104 cache->fs_info = root->fs_info;
10105 cache->full_stripe_len = btrfs_full_stripe_len(root,
10106 &root->fs_info->mapping_tree,
10107 start);
10108 set_free_space_tree_thresholds(cache);
10109
10110 atomic_set(&cache->count, 1);
10111 spin_lock_init(&cache->lock);
10112 init_rwsem(&cache->data_rwsem);
10113 INIT_LIST_HEAD(&cache->list);
10114 INIT_LIST_HEAD(&cache->cluster_list);
10115 INIT_LIST_HEAD(&cache->bg_list);
10116 INIT_LIST_HEAD(&cache->ro_list);
10117 INIT_LIST_HEAD(&cache->dirty_list);
10118 INIT_LIST_HEAD(&cache->io_list);
10119 btrfs_init_free_space_ctl(cache);
10120 atomic_set(&cache->trimming, 0);
10121 mutex_init(&cache->free_space_lock);
10122
10123 return cache;
10124 }
10125
10126 int btrfs_read_block_groups(struct btrfs_root *root)
10127 {
10128 struct btrfs_path *path;
10129 int ret;
10130 struct btrfs_block_group_cache *cache;
10131 struct btrfs_fs_info *info = root->fs_info;
10132 struct btrfs_space_info *space_info;
10133 struct btrfs_key key;
10134 struct btrfs_key found_key;
10135 struct extent_buffer *leaf;
10136 int need_clear = 0;
10137 u64 cache_gen;
10138
10139 root = info->extent_root;
10140 key.objectid = 0;
10141 key.offset = 0;
10142 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10143 path = btrfs_alloc_path();
10144 if (!path)
10145 return -ENOMEM;
10146 path->reada = READA_FORWARD;
10147
10148 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
10149 if (btrfs_test_opt(root, SPACE_CACHE) &&
10150 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
10151 need_clear = 1;
10152 if (btrfs_test_opt(root, CLEAR_CACHE))
10153 need_clear = 1;
10154
10155 while (1) {
10156 ret = find_first_block_group(root, path, &key);
10157 if (ret > 0)
10158 break;
10159 if (ret != 0)
10160 goto error;
10161
10162 leaf = path->nodes[0];
10163 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
10164
10165 cache = btrfs_create_block_group_cache(root, found_key.objectid,
10166 found_key.offset);
10167 if (!cache) {
10168 ret = -ENOMEM;
10169 goto error;
10170 }
10171
10172 if (need_clear) {
10173 /*
10174 * When we mount with old space cache, we need to
10175 * set BTRFS_DC_CLEAR and set dirty flag.
10176 *
10177 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10178 * truncate the old free space cache inode and
10179 * setup a new one.
10180 * b) Setting 'dirty flag' makes sure that we flush
10181 * the new space cache info onto disk.
10182 */
10183 if (btrfs_test_opt(root, SPACE_CACHE))
10184 cache->disk_cache_state = BTRFS_DC_CLEAR;
10185 }
10186
10187 read_extent_buffer(leaf, &cache->item,
10188 btrfs_item_ptr_offset(leaf, path->slots[0]),
10189 sizeof(cache->item));
10190 cache->flags = btrfs_block_group_flags(&cache->item);
10191
10192 key.objectid = found_key.objectid + found_key.offset;
10193 btrfs_release_path(path);
10194
10195 /*
10196 * We need to exclude the super stripes now so that the space
10197 * info has super bytes accounted for, otherwise we'll think
10198 * we have more space than we actually do.
10199 */
10200 ret = exclude_super_stripes(root, cache);
10201 if (ret) {
10202 /*
10203 * We may have excluded something, so call this just in
10204 * case.
10205 */
10206 free_excluded_extents(root, cache);
10207 btrfs_put_block_group(cache);
10208 goto error;
10209 }
10210
10211 /*
10212 * check for two cases, either we are full, and therefore
10213 * don't need to bother with the caching work since we won't
10214 * find any space, or we are empty, and we can just add all
10215 * the space in and be done with it. This saves us _alot_ of
10216 * time, particularly in the full case.
10217 */
10218 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
10219 cache->last_byte_to_unpin = (u64)-1;
10220 cache->cached = BTRFS_CACHE_FINISHED;
10221 free_excluded_extents(root, cache);
10222 } else if (btrfs_block_group_used(&cache->item) == 0) {
10223 cache->last_byte_to_unpin = (u64)-1;
10224 cache->cached = BTRFS_CACHE_FINISHED;
10225 add_new_free_space(cache, root->fs_info,
10226 found_key.objectid,
10227 found_key.objectid +
10228 found_key.offset);
10229 free_excluded_extents(root, cache);
10230 }
10231
10232 ret = btrfs_add_block_group_cache(root->fs_info, cache);
10233 if (ret) {
10234 btrfs_remove_free_space_cache(cache);
10235 btrfs_put_block_group(cache);
10236 goto error;
10237 }
10238
10239 trace_btrfs_add_block_group(root->fs_info, cache, 0);
10240 ret = update_space_info(info, cache->flags, found_key.offset,
10241 btrfs_block_group_used(&cache->item),
10242 cache->bytes_super, &space_info);
10243 if (ret) {
10244 btrfs_remove_free_space_cache(cache);
10245 spin_lock(&info->block_group_cache_lock);
10246 rb_erase(&cache->cache_node,
10247 &info->block_group_cache_tree);
10248 RB_CLEAR_NODE(&cache->cache_node);
10249 spin_unlock(&info->block_group_cache_lock);
10250 btrfs_put_block_group(cache);
10251 goto error;
10252 }
10253
10254 cache->space_info = space_info;
10255
10256 __link_block_group(space_info, cache);
10257
10258 set_avail_alloc_bits(root->fs_info, cache->flags);
10259 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
10260 inc_block_group_ro(cache, 1);
10261 } else if (btrfs_block_group_used(&cache->item) == 0) {
10262 spin_lock(&info->unused_bgs_lock);
10263 /* Should always be true but just in case. */
10264 if (list_empty(&cache->bg_list)) {
10265 btrfs_get_block_group(cache);
10266 list_add_tail(&cache->bg_list,
10267 &info->unused_bgs);
10268 }
10269 spin_unlock(&info->unused_bgs_lock);
10270 }
10271 }
10272
10273 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
10274 if (!(get_alloc_profile(root, space_info->flags) &
10275 (BTRFS_BLOCK_GROUP_RAID10 |
10276 BTRFS_BLOCK_GROUP_RAID1 |
10277 BTRFS_BLOCK_GROUP_RAID5 |
10278 BTRFS_BLOCK_GROUP_RAID6 |
10279 BTRFS_BLOCK_GROUP_DUP)))
10280 continue;
10281 /*
10282 * avoid allocating from un-mirrored block group if there are
10283 * mirrored block groups.
10284 */
10285 list_for_each_entry(cache,
10286 &space_info->block_groups[BTRFS_RAID_RAID0],
10287 list)
10288 inc_block_group_ro(cache, 1);
10289 list_for_each_entry(cache,
10290 &space_info->block_groups[BTRFS_RAID_SINGLE],
10291 list)
10292 inc_block_group_ro(cache, 1);
10293 }
10294
10295 init_global_block_rsv(info);
10296 ret = 0;
10297 error:
10298 btrfs_free_path(path);
10299 return ret;
10300 }
10301
10302 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
10303 struct btrfs_root *root)
10304 {
10305 struct btrfs_block_group_cache *block_group, *tmp;
10306 struct btrfs_root *extent_root = root->fs_info->extent_root;
10307 struct btrfs_block_group_item item;
10308 struct btrfs_key key;
10309 int ret = 0;
10310 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
10311
10312 trans->can_flush_pending_bgs = false;
10313 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
10314 if (ret)
10315 goto next;
10316
10317 spin_lock(&block_group->lock);
10318 memcpy(&item, &block_group->item, sizeof(item));
10319 memcpy(&key, &block_group->key, sizeof(key));
10320 spin_unlock(&block_group->lock);
10321
10322 ret = btrfs_insert_item(trans, extent_root, &key, &item,
10323 sizeof(item));
10324 if (ret)
10325 btrfs_abort_transaction(trans, extent_root, ret);
10326 ret = btrfs_finish_chunk_alloc(trans, extent_root,
10327 key.objectid, key.offset);
10328 if (ret)
10329 btrfs_abort_transaction(trans, extent_root, ret);
10330 add_block_group_free_space(trans, root->fs_info, block_group);
10331 /* already aborted the transaction if it failed. */
10332 next:
10333 list_del_init(&block_group->bg_list);
10334 }
10335 trans->can_flush_pending_bgs = can_flush_pending_bgs;
10336 }
10337
10338 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
10339 struct btrfs_root *root, u64 bytes_used,
10340 u64 type, u64 chunk_objectid, u64 chunk_offset,
10341 u64 size)
10342 {
10343 int ret;
10344 struct btrfs_root *extent_root;
10345 struct btrfs_block_group_cache *cache;
10346 extent_root = root->fs_info->extent_root;
10347
10348 btrfs_set_log_full_commit(root->fs_info, trans);
10349
10350 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
10351 if (!cache)
10352 return -ENOMEM;
10353
10354 btrfs_set_block_group_used(&cache->item, bytes_used);
10355 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
10356 btrfs_set_block_group_flags(&cache->item, type);
10357
10358 cache->flags = type;
10359 cache->last_byte_to_unpin = (u64)-1;
10360 cache->cached = BTRFS_CACHE_FINISHED;
10361 cache->needs_free_space = 1;
10362 ret = exclude_super_stripes(root, cache);
10363 if (ret) {
10364 /*
10365 * We may have excluded something, so call this just in
10366 * case.
10367 */
10368 free_excluded_extents(root, cache);
10369 btrfs_put_block_group(cache);
10370 return ret;
10371 }
10372
10373 add_new_free_space(cache, root->fs_info, chunk_offset,
10374 chunk_offset + size);
10375
10376 free_excluded_extents(root, cache);
10377
10378 #ifdef CONFIG_BTRFS_DEBUG
10379 if (btrfs_should_fragment_free_space(root, cache)) {
10380 u64 new_bytes_used = size - bytes_used;
10381
10382 bytes_used += new_bytes_used >> 1;
10383 fragment_free_space(root, cache);
10384 }
10385 #endif
10386 /*
10387 * Call to ensure the corresponding space_info object is created and
10388 * assigned to our block group, but don't update its counters just yet.
10389 * We want our bg to be added to the rbtree with its ->space_info set.
10390 */
10391 ret = update_space_info(root->fs_info, cache->flags, 0, 0, 0,
10392 &cache->space_info);
10393 if (ret) {
10394 btrfs_remove_free_space_cache(cache);
10395 btrfs_put_block_group(cache);
10396 return ret;
10397 }
10398
10399 ret = btrfs_add_block_group_cache(root->fs_info, cache);
10400 if (ret) {
10401 btrfs_remove_free_space_cache(cache);
10402 btrfs_put_block_group(cache);
10403 return ret;
10404 }
10405
10406 /*
10407 * Now that our block group has its ->space_info set and is inserted in
10408 * the rbtree, update the space info's counters.
10409 */
10410 trace_btrfs_add_block_group(root->fs_info, cache, 1);
10411 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
10412 cache->bytes_super, &cache->space_info);
10413 if (ret) {
10414 btrfs_remove_free_space_cache(cache);
10415 spin_lock(&root->fs_info->block_group_cache_lock);
10416 rb_erase(&cache->cache_node,
10417 &root->fs_info->block_group_cache_tree);
10418 RB_CLEAR_NODE(&cache->cache_node);
10419 spin_unlock(&root->fs_info->block_group_cache_lock);
10420 btrfs_put_block_group(cache);
10421 return ret;
10422 }
10423 update_global_block_rsv(root->fs_info);
10424
10425 __link_block_group(cache->space_info, cache);
10426
10427 list_add_tail(&cache->bg_list, &trans->new_bgs);
10428
10429 set_avail_alloc_bits(extent_root->fs_info, type);
10430 return 0;
10431 }
10432
10433 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10434 {
10435 u64 extra_flags = chunk_to_extended(flags) &
10436 BTRFS_EXTENDED_PROFILE_MASK;
10437
10438 write_seqlock(&fs_info->profiles_lock);
10439 if (flags & BTRFS_BLOCK_GROUP_DATA)
10440 fs_info->avail_data_alloc_bits &= ~extra_flags;
10441 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10442 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10443 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10444 fs_info->avail_system_alloc_bits &= ~extra_flags;
10445 write_sequnlock(&fs_info->profiles_lock);
10446 }
10447
10448 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10449 struct btrfs_root *root, u64 group_start,
10450 struct extent_map *em)
10451 {
10452 struct btrfs_path *path;
10453 struct btrfs_block_group_cache *block_group;
10454 struct btrfs_free_cluster *cluster;
10455 struct btrfs_root *tree_root = root->fs_info->tree_root;
10456 struct btrfs_key key;
10457 struct inode *inode;
10458 struct kobject *kobj = NULL;
10459 int ret;
10460 int index;
10461 int factor;
10462 struct btrfs_caching_control *caching_ctl = NULL;
10463 bool remove_em;
10464
10465 root = root->fs_info->extent_root;
10466
10467 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
10468 BUG_ON(!block_group);
10469 BUG_ON(!block_group->ro);
10470
10471 /*
10472 * Free the reserved super bytes from this block group before
10473 * remove it.
10474 */
10475 free_excluded_extents(root, block_group);
10476
10477 memcpy(&key, &block_group->key, sizeof(key));
10478 index = get_block_group_index(block_group);
10479 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
10480 BTRFS_BLOCK_GROUP_RAID1 |
10481 BTRFS_BLOCK_GROUP_RAID10))
10482 factor = 2;
10483 else
10484 factor = 1;
10485
10486 /* make sure this block group isn't part of an allocation cluster */
10487 cluster = &root->fs_info->data_alloc_cluster;
10488 spin_lock(&cluster->refill_lock);
10489 btrfs_return_cluster_to_free_space(block_group, cluster);
10490 spin_unlock(&cluster->refill_lock);
10491
10492 /*
10493 * make sure this block group isn't part of a metadata
10494 * allocation cluster
10495 */
10496 cluster = &root->fs_info->meta_alloc_cluster;
10497 spin_lock(&cluster->refill_lock);
10498 btrfs_return_cluster_to_free_space(block_group, cluster);
10499 spin_unlock(&cluster->refill_lock);
10500
10501 path = btrfs_alloc_path();
10502 if (!path) {
10503 ret = -ENOMEM;
10504 goto out;
10505 }
10506
10507 /*
10508 * get the inode first so any iput calls done for the io_list
10509 * aren't the final iput (no unlinks allowed now)
10510 */
10511 inode = lookup_free_space_inode(tree_root, block_group, path);
10512
10513 mutex_lock(&trans->transaction->cache_write_mutex);
10514 /*
10515 * make sure our free spache cache IO is done before remove the
10516 * free space inode
10517 */
10518 spin_lock(&trans->transaction->dirty_bgs_lock);
10519 if (!list_empty(&block_group->io_list)) {
10520 list_del_init(&block_group->io_list);
10521
10522 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10523
10524 spin_unlock(&trans->transaction->dirty_bgs_lock);
10525 btrfs_wait_cache_io(root, trans, block_group,
10526 &block_group->io_ctl, path,
10527 block_group->key.objectid);
10528 btrfs_put_block_group(block_group);
10529 spin_lock(&trans->transaction->dirty_bgs_lock);
10530 }
10531
10532 if (!list_empty(&block_group->dirty_list)) {
10533 list_del_init(&block_group->dirty_list);
10534 btrfs_put_block_group(block_group);
10535 }
10536 spin_unlock(&trans->transaction->dirty_bgs_lock);
10537 mutex_unlock(&trans->transaction->cache_write_mutex);
10538
10539 if (!IS_ERR(inode)) {
10540 ret = btrfs_orphan_add(trans, inode);
10541 if (ret) {
10542 btrfs_add_delayed_iput(inode);
10543 goto out;
10544 }
10545 clear_nlink(inode);
10546 /* One for the block groups ref */
10547 spin_lock(&block_group->lock);
10548 if (block_group->iref) {
10549 block_group->iref = 0;
10550 block_group->inode = NULL;
10551 spin_unlock(&block_group->lock);
10552 iput(inode);
10553 } else {
10554 spin_unlock(&block_group->lock);
10555 }
10556 /* One for our lookup ref */
10557 btrfs_add_delayed_iput(inode);
10558 }
10559
10560 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10561 key.offset = block_group->key.objectid;
10562 key.type = 0;
10563
10564 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10565 if (ret < 0)
10566 goto out;
10567 if (ret > 0)
10568 btrfs_release_path(path);
10569 if (ret == 0) {
10570 ret = btrfs_del_item(trans, tree_root, path);
10571 if (ret)
10572 goto out;
10573 btrfs_release_path(path);
10574 }
10575
10576 spin_lock(&root->fs_info->block_group_cache_lock);
10577 rb_erase(&block_group->cache_node,
10578 &root->fs_info->block_group_cache_tree);
10579 RB_CLEAR_NODE(&block_group->cache_node);
10580
10581 if (root->fs_info->first_logical_byte == block_group->key.objectid)
10582 root->fs_info->first_logical_byte = (u64)-1;
10583 spin_unlock(&root->fs_info->block_group_cache_lock);
10584
10585 down_write(&block_group->space_info->groups_sem);
10586 /*
10587 * we must use list_del_init so people can check to see if they
10588 * are still on the list after taking the semaphore
10589 */
10590 list_del_init(&block_group->list);
10591 if (list_empty(&block_group->space_info->block_groups[index])) {
10592 kobj = block_group->space_info->block_group_kobjs[index];
10593 block_group->space_info->block_group_kobjs[index] = NULL;
10594 clear_avail_alloc_bits(root->fs_info, block_group->flags);
10595 }
10596 up_write(&block_group->space_info->groups_sem);
10597 if (kobj) {
10598 kobject_del(kobj);
10599 kobject_put(kobj);
10600 }
10601
10602 if (block_group->has_caching_ctl)
10603 caching_ctl = get_caching_control(block_group);
10604 if (block_group->cached == BTRFS_CACHE_STARTED)
10605 wait_block_group_cache_done(block_group);
10606 if (block_group->has_caching_ctl) {
10607 down_write(&root->fs_info->commit_root_sem);
10608 if (!caching_ctl) {
10609 struct btrfs_caching_control *ctl;
10610
10611 list_for_each_entry(ctl,
10612 &root->fs_info->caching_block_groups, list)
10613 if (ctl->block_group == block_group) {
10614 caching_ctl = ctl;
10615 atomic_inc(&caching_ctl->count);
10616 break;
10617 }
10618 }
10619 if (caching_ctl)
10620 list_del_init(&caching_ctl->list);
10621 up_write(&root->fs_info->commit_root_sem);
10622 if (caching_ctl) {
10623 /* Once for the caching bgs list and once for us. */
10624 put_caching_control(caching_ctl);
10625 put_caching_control(caching_ctl);
10626 }
10627 }
10628
10629 spin_lock(&trans->transaction->dirty_bgs_lock);
10630 if (!list_empty(&block_group->dirty_list)) {
10631 WARN_ON(1);
10632 }
10633 if (!list_empty(&block_group->io_list)) {
10634 WARN_ON(1);
10635 }
10636 spin_unlock(&trans->transaction->dirty_bgs_lock);
10637 btrfs_remove_free_space_cache(block_group);
10638
10639 spin_lock(&block_group->space_info->lock);
10640 list_del_init(&block_group->ro_list);
10641
10642 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
10643 WARN_ON(block_group->space_info->total_bytes
10644 < block_group->key.offset);
10645 WARN_ON(block_group->space_info->bytes_readonly
10646 < block_group->key.offset);
10647 WARN_ON(block_group->space_info->disk_total
10648 < block_group->key.offset * factor);
10649 }
10650 block_group->space_info->total_bytes -= block_group->key.offset;
10651 block_group->space_info->bytes_readonly -= block_group->key.offset;
10652 block_group->space_info->disk_total -= block_group->key.offset * factor;
10653
10654 spin_unlock(&block_group->space_info->lock);
10655
10656 memcpy(&key, &block_group->key, sizeof(key));
10657
10658 lock_chunks(root);
10659 if (!list_empty(&em->list)) {
10660 /* We're in the transaction->pending_chunks list. */
10661 free_extent_map(em);
10662 }
10663 spin_lock(&block_group->lock);
10664 block_group->removed = 1;
10665 /*
10666 * At this point trimming can't start on this block group, because we
10667 * removed the block group from the tree fs_info->block_group_cache_tree
10668 * so no one can't find it anymore and even if someone already got this
10669 * block group before we removed it from the rbtree, they have already
10670 * incremented block_group->trimming - if they didn't, they won't find
10671 * any free space entries because we already removed them all when we
10672 * called btrfs_remove_free_space_cache().
10673 *
10674 * And we must not remove the extent map from the fs_info->mapping_tree
10675 * to prevent the same logical address range and physical device space
10676 * ranges from being reused for a new block group. This is because our
10677 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10678 * completely transactionless, so while it is trimming a range the
10679 * currently running transaction might finish and a new one start,
10680 * allowing for new block groups to be created that can reuse the same
10681 * physical device locations unless we take this special care.
10682 *
10683 * There may also be an implicit trim operation if the file system
10684 * is mounted with -odiscard. The same protections must remain
10685 * in place until the extents have been discarded completely when
10686 * the transaction commit has completed.
10687 */
10688 remove_em = (atomic_read(&block_group->trimming) == 0);
10689 /*
10690 * Make sure a trimmer task always sees the em in the pinned_chunks list
10691 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10692 * before checking block_group->removed).
10693 */
10694 if (!remove_em) {
10695 /*
10696 * Our em might be in trans->transaction->pending_chunks which
10697 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10698 * and so is the fs_info->pinned_chunks list.
10699 *
10700 * So at this point we must be holding the chunk_mutex to avoid
10701 * any races with chunk allocation (more specifically at
10702 * volumes.c:contains_pending_extent()), to ensure it always
10703 * sees the em, either in the pending_chunks list or in the
10704 * pinned_chunks list.
10705 */
10706 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
10707 }
10708 spin_unlock(&block_group->lock);
10709
10710 if (remove_em) {
10711 struct extent_map_tree *em_tree;
10712
10713 em_tree = &root->fs_info->mapping_tree.map_tree;
10714 write_lock(&em_tree->lock);
10715 /*
10716 * The em might be in the pending_chunks list, so make sure the
10717 * chunk mutex is locked, since remove_extent_mapping() will
10718 * delete us from that list.
10719 */
10720 remove_extent_mapping(em_tree, em);
10721 write_unlock(&em_tree->lock);
10722 /* once for the tree */
10723 free_extent_map(em);
10724 }
10725
10726 unlock_chunks(root);
10727
10728 ret = remove_block_group_free_space(trans, root->fs_info, block_group);
10729 if (ret)
10730 goto out;
10731
10732 btrfs_put_block_group(block_group);
10733 btrfs_put_block_group(block_group);
10734
10735 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10736 if (ret > 0)
10737 ret = -EIO;
10738 if (ret < 0)
10739 goto out;
10740
10741 ret = btrfs_del_item(trans, root, path);
10742 out:
10743 btrfs_free_path(path);
10744 return ret;
10745 }
10746
10747 struct btrfs_trans_handle *
10748 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10749 const u64 chunk_offset)
10750 {
10751 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10752 struct extent_map *em;
10753 struct map_lookup *map;
10754 unsigned int num_items;
10755
10756 read_lock(&em_tree->lock);
10757 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10758 read_unlock(&em_tree->lock);
10759 ASSERT(em && em->start == chunk_offset);
10760
10761 /*
10762 * We need to reserve 3 + N units from the metadata space info in order
10763 * to remove a block group (done at btrfs_remove_chunk() and at
10764 * btrfs_remove_block_group()), which are used for:
10765 *
10766 * 1 unit for adding the free space inode's orphan (located in the tree
10767 * of tree roots).
10768 * 1 unit for deleting the block group item (located in the extent
10769 * tree).
10770 * 1 unit for deleting the free space item (located in tree of tree
10771 * roots).
10772 * N units for deleting N device extent items corresponding to each
10773 * stripe (located in the device tree).
10774 *
10775 * In order to remove a block group we also need to reserve units in the
10776 * system space info in order to update the chunk tree (update one or
10777 * more device items and remove one chunk item), but this is done at
10778 * btrfs_remove_chunk() through a call to check_system_chunk().
10779 */
10780 map = em->map_lookup;
10781 num_items = 3 + map->num_stripes;
10782 free_extent_map(em);
10783
10784 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10785 num_items, 1);
10786 }
10787
10788 /*
10789 * Process the unused_bgs list and remove any that don't have any allocated
10790 * space inside of them.
10791 */
10792 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10793 {
10794 struct btrfs_block_group_cache *block_group;
10795 struct btrfs_space_info *space_info;
10796 struct btrfs_root *root = fs_info->extent_root;
10797 struct btrfs_trans_handle *trans;
10798 int ret = 0;
10799
10800 if (!fs_info->open)
10801 return;
10802
10803 spin_lock(&fs_info->unused_bgs_lock);
10804 while (!list_empty(&fs_info->unused_bgs)) {
10805 u64 start, end;
10806 int trimming;
10807
10808 block_group = list_first_entry(&fs_info->unused_bgs,
10809 struct btrfs_block_group_cache,
10810 bg_list);
10811 list_del_init(&block_group->bg_list);
10812
10813 space_info = block_group->space_info;
10814
10815 if (ret || btrfs_mixed_space_info(space_info)) {
10816 btrfs_put_block_group(block_group);
10817 continue;
10818 }
10819 spin_unlock(&fs_info->unused_bgs_lock);
10820
10821 mutex_lock(&fs_info->delete_unused_bgs_mutex);
10822
10823 /* Don't want to race with allocators so take the groups_sem */
10824 down_write(&space_info->groups_sem);
10825 spin_lock(&block_group->lock);
10826 if (block_group->reserved ||
10827 btrfs_block_group_used(&block_group->item) ||
10828 block_group->ro ||
10829 list_is_singular(&block_group->list)) {
10830 /*
10831 * We want to bail if we made new allocations or have
10832 * outstanding allocations in this block group. We do
10833 * the ro check in case balance is currently acting on
10834 * this block group.
10835 */
10836 spin_unlock(&block_group->lock);
10837 up_write(&space_info->groups_sem);
10838 goto next;
10839 }
10840 spin_unlock(&block_group->lock);
10841
10842 /* We don't want to force the issue, only flip if it's ok. */
10843 ret = inc_block_group_ro(block_group, 0);
10844 up_write(&space_info->groups_sem);
10845 if (ret < 0) {
10846 ret = 0;
10847 goto next;
10848 }
10849
10850 /*
10851 * Want to do this before we do anything else so we can recover
10852 * properly if we fail to join the transaction.
10853 */
10854 trans = btrfs_start_trans_remove_block_group(fs_info,
10855 block_group->key.objectid);
10856 if (IS_ERR(trans)) {
10857 btrfs_dec_block_group_ro(root, block_group);
10858 ret = PTR_ERR(trans);
10859 goto next;
10860 }
10861
10862 /*
10863 * We could have pending pinned extents for this block group,
10864 * just delete them, we don't care about them anymore.
10865 */
10866 start = block_group->key.objectid;
10867 end = start + block_group->key.offset - 1;
10868 /*
10869 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10870 * btrfs_finish_extent_commit(). If we are at transaction N,
10871 * another task might be running finish_extent_commit() for the
10872 * previous transaction N - 1, and have seen a range belonging
10873 * to the block group in freed_extents[] before we were able to
10874 * clear the whole block group range from freed_extents[]. This
10875 * means that task can lookup for the block group after we
10876 * unpinned it from freed_extents[] and removed it, leading to
10877 * a BUG_ON() at btrfs_unpin_extent_range().
10878 */
10879 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10880 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10881 EXTENT_DIRTY);
10882 if (ret) {
10883 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10884 btrfs_dec_block_group_ro(root, block_group);
10885 goto end_trans;
10886 }
10887 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10888 EXTENT_DIRTY);
10889 if (ret) {
10890 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10891 btrfs_dec_block_group_ro(root, block_group);
10892 goto end_trans;
10893 }
10894 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10895
10896 /* Reset pinned so btrfs_put_block_group doesn't complain */
10897 spin_lock(&space_info->lock);
10898 spin_lock(&block_group->lock);
10899
10900 space_info->bytes_pinned -= block_group->pinned;
10901 space_info->bytes_readonly += block_group->pinned;
10902 percpu_counter_add(&space_info->total_bytes_pinned,
10903 -block_group->pinned);
10904 block_group->pinned = 0;
10905
10906 spin_unlock(&block_group->lock);
10907 spin_unlock(&space_info->lock);
10908
10909 /* DISCARD can flip during remount */
10910 trimming = btrfs_test_opt(root, DISCARD);
10911
10912 /* Implicit trim during transaction commit. */
10913 if (trimming)
10914 btrfs_get_block_group_trimming(block_group);
10915
10916 /*
10917 * Btrfs_remove_chunk will abort the transaction if things go
10918 * horribly wrong.
10919 */
10920 ret = btrfs_remove_chunk(trans, root,
10921 block_group->key.objectid);
10922
10923 if (ret) {
10924 if (trimming)
10925 btrfs_put_block_group_trimming(block_group);
10926 goto end_trans;
10927 }
10928
10929 /*
10930 * If we're not mounted with -odiscard, we can just forget
10931 * about this block group. Otherwise we'll need to wait
10932 * until transaction commit to do the actual discard.
10933 */
10934 if (trimming) {
10935 spin_lock(&fs_info->unused_bgs_lock);
10936 /*
10937 * A concurrent scrub might have added us to the list
10938 * fs_info->unused_bgs, so use a list_move operation
10939 * to add the block group to the deleted_bgs list.
10940 */
10941 list_move(&block_group->bg_list,
10942 &trans->transaction->deleted_bgs);
10943 spin_unlock(&fs_info->unused_bgs_lock);
10944 btrfs_get_block_group(block_group);
10945 }
10946 end_trans:
10947 btrfs_end_transaction(trans, root);
10948 next:
10949 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
10950 btrfs_put_block_group(block_group);
10951 spin_lock(&fs_info->unused_bgs_lock);
10952 }
10953 spin_unlock(&fs_info->unused_bgs_lock);
10954 }
10955
10956 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10957 {
10958 struct btrfs_space_info *space_info;
10959 struct btrfs_super_block *disk_super;
10960 u64 features;
10961 u64 flags;
10962 int mixed = 0;
10963 int ret;
10964
10965 disk_super = fs_info->super_copy;
10966 if (!btrfs_super_root(disk_super))
10967 return -EINVAL;
10968
10969 features = btrfs_super_incompat_flags(disk_super);
10970 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10971 mixed = 1;
10972
10973 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10974 ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info);
10975 if (ret)
10976 goto out;
10977
10978 if (mixed) {
10979 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10980 ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info);
10981 } else {
10982 flags = BTRFS_BLOCK_GROUP_METADATA;
10983 ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info);
10984 if (ret)
10985 goto out;
10986
10987 flags = BTRFS_BLOCK_GROUP_DATA;
10988 ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info);
10989 }
10990 out:
10991 return ret;
10992 }
10993
10994 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
10995 {
10996 return unpin_extent_range(root, start, end, false);
10997 }
10998
10999 /*
11000 * It used to be that old block groups would be left around forever.
11001 * Iterating over them would be enough to trim unused space. Since we
11002 * now automatically remove them, we also need to iterate over unallocated
11003 * space.
11004 *
11005 * We don't want a transaction for this since the discard may take a
11006 * substantial amount of time. We don't require that a transaction be
11007 * running, but we do need to take a running transaction into account
11008 * to ensure that we're not discarding chunks that were released in
11009 * the current transaction.
11010 *
11011 * Holding the chunks lock will prevent other threads from allocating
11012 * or releasing chunks, but it won't prevent a running transaction
11013 * from committing and releasing the memory that the pending chunks
11014 * list head uses. For that, we need to take a reference to the
11015 * transaction.
11016 */
11017 static int btrfs_trim_free_extents(struct btrfs_device *device,
11018 u64 minlen, u64 *trimmed)
11019 {
11020 u64 start = 0, len = 0;
11021 int ret;
11022
11023 *trimmed = 0;
11024
11025 /* Not writeable = nothing to do. */
11026 if (!device->writeable)
11027 return 0;
11028
11029 /* No free space = nothing to do. */
11030 if (device->total_bytes <= device->bytes_used)
11031 return 0;
11032
11033 ret = 0;
11034
11035 while (1) {
11036 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
11037 struct btrfs_transaction *trans;
11038 u64 bytes;
11039
11040 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
11041 if (ret)
11042 return ret;
11043
11044 down_read(&fs_info->commit_root_sem);
11045
11046 spin_lock(&fs_info->trans_lock);
11047 trans = fs_info->running_transaction;
11048 if (trans)
11049 atomic_inc(&trans->use_count);
11050 spin_unlock(&fs_info->trans_lock);
11051
11052 ret = find_free_dev_extent_start(trans, device, minlen, start,
11053 &start, &len);
11054 if (trans)
11055 btrfs_put_transaction(trans);
11056
11057 if (ret) {
11058 up_read(&fs_info->commit_root_sem);
11059 mutex_unlock(&fs_info->chunk_mutex);
11060 if (ret == -ENOSPC)
11061 ret = 0;
11062 break;
11063 }
11064
11065 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
11066 up_read(&fs_info->commit_root_sem);
11067 mutex_unlock(&fs_info->chunk_mutex);
11068
11069 if (ret)
11070 break;
11071
11072 start += len;
11073 *trimmed += bytes;
11074
11075 if (fatal_signal_pending(current)) {
11076 ret = -ERESTARTSYS;
11077 break;
11078 }
11079
11080 cond_resched();
11081 }
11082
11083 return ret;
11084 }
11085
11086 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
11087 {
11088 struct btrfs_fs_info *fs_info = root->fs_info;
11089 struct btrfs_block_group_cache *cache = NULL;
11090 struct btrfs_device *device;
11091 struct list_head *devices;
11092 u64 group_trimmed;
11093 u64 start;
11094 u64 end;
11095 u64 trimmed = 0;
11096 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
11097 int ret = 0;
11098
11099 /*
11100 * try to trim all FS space, our block group may start from non-zero.
11101 */
11102 if (range->len == total_bytes)
11103 cache = btrfs_lookup_first_block_group(fs_info, range->start);
11104 else
11105 cache = btrfs_lookup_block_group(fs_info, range->start);
11106
11107 while (cache) {
11108 if (cache->key.objectid >= (range->start + range->len)) {
11109 btrfs_put_block_group(cache);
11110 break;
11111 }
11112
11113 start = max(range->start, cache->key.objectid);
11114 end = min(range->start + range->len,
11115 cache->key.objectid + cache->key.offset);
11116
11117 if (end - start >= range->minlen) {
11118 if (!block_group_cache_done(cache)) {
11119 ret = cache_block_group(cache, 0);
11120 if (ret) {
11121 btrfs_put_block_group(cache);
11122 break;
11123 }
11124 ret = wait_block_group_cache_done(cache);
11125 if (ret) {
11126 btrfs_put_block_group(cache);
11127 break;
11128 }
11129 }
11130 ret = btrfs_trim_block_group(cache,
11131 &group_trimmed,
11132 start,
11133 end,
11134 range->minlen);
11135
11136 trimmed += group_trimmed;
11137 if (ret) {
11138 btrfs_put_block_group(cache);
11139 break;
11140 }
11141 }
11142
11143 cache = next_block_group(fs_info->tree_root, cache);
11144 }
11145
11146 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
11147 devices = &root->fs_info->fs_devices->alloc_list;
11148 list_for_each_entry(device, devices, dev_alloc_list) {
11149 ret = btrfs_trim_free_extents(device, range->minlen,
11150 &group_trimmed);
11151 if (ret)
11152 break;
11153
11154 trimmed += group_trimmed;
11155 }
11156 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
11157
11158 range->len = trimmed;
11159 return ret;
11160 }
11161
11162 /*
11163 * btrfs_{start,end}_write_no_snapshoting() are similar to
11164 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
11165 * data into the page cache through nocow before the subvolume is snapshoted,
11166 * but flush the data into disk after the snapshot creation, or to prevent
11167 * operations while snapshoting is ongoing and that cause the snapshot to be
11168 * inconsistent (writes followed by expanding truncates for example).
11169 */
11170 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
11171 {
11172 percpu_counter_dec(&root->subv_writers->counter);
11173 /*
11174 * Make sure counter is updated before we wake up waiters.
11175 */
11176 smp_mb();
11177 if (waitqueue_active(&root->subv_writers->wait))
11178 wake_up(&root->subv_writers->wait);
11179 }
11180
11181 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
11182 {
11183 if (atomic_read(&root->will_be_snapshoted))
11184 return 0;
11185
11186 percpu_counter_inc(&root->subv_writers->counter);
11187 /*
11188 * Make sure counter is updated before we check for snapshot creation.
11189 */
11190 smp_mb();
11191 if (atomic_read(&root->will_be_snapshoted)) {
11192 btrfs_end_write_no_snapshoting(root);
11193 return 0;
11194 }
11195 return 1;
11196 }
11197
11198 static int wait_snapshoting_atomic_t(atomic_t *a)
11199 {
11200 schedule();
11201 return 0;
11202 }
11203
11204 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
11205 {
11206 while (true) {
11207 int ret;
11208
11209 ret = btrfs_start_write_no_snapshoting(root);
11210 if (ret)
11211 break;
11212 wait_on_atomic_t(&root->will_be_snapshoted,
11213 wait_snapshoting_atomic_t,
11214 TASK_UNINTERRUPTIBLE);
11215 }
11216 }
Linux kernel - Deliverables
This page took 0.459889 seconds and 5 git commands to generate.