?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Btrfs: add tracepoints for flush events
[linux/fpc-iii.git] / fs / btrfs / extent-tree.c
blob31ded6aae1f098d7ca8963d182d1f42547555a22
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_fs_info *fs_info, 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 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
4948 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4949 }
4950
4951 static void wake_all_tickets(struct list_head *head)
4952 {
4953 struct reserve_ticket *ticket;
4954
4955 while (!list_empty(head)) {
4956 ticket = list_first_entry(head, struct reserve_ticket, list);
4957 list_del_init(&ticket->list);
4958 ticket->error = -ENOSPC;
4959 wake_up(&ticket->wait);
4960 }
4961 }
4962
4963 /*
4964 * This is for normal flushers, we can wait all goddamned day if we want to. We
4965 * will loop and continuously try to flush as long as we are making progress.
4966 * We count progress as clearing off tickets each time we have to loop.
4967 */
4968 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4969 {
4970 struct reserve_ticket *last_ticket = NULL;
4971 struct btrfs_fs_info *fs_info;
4972 struct btrfs_space_info *space_info;
4973 u64 to_reclaim;
4974 int flush_state;
4975 int commit_cycles = 0;
4976
4977 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4978 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4979
4980 spin_lock(&space_info->lock);
4981 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4982 space_info);
4983 if (!to_reclaim) {
4984 space_info->flush = 0;
4985 spin_unlock(&space_info->lock);
4986 return;
4987 }
4988 last_ticket = list_first_entry(&space_info->tickets,
4989 struct reserve_ticket, list);
4990 spin_unlock(&space_info->lock);
4991
4992 flush_state = FLUSH_DELAYED_ITEMS_NR;
4993 do {
4994 struct reserve_ticket *ticket;
4995 int ret;
4996
4997 ret = flush_space(fs_info->fs_root, space_info, to_reclaim,
4998 to_reclaim, flush_state);
4999 spin_lock(&space_info->lock);
5000 if (list_empty(&space_info->tickets)) {
5001 space_info->flush = 0;
5002 spin_unlock(&space_info->lock);
5003 return;
5004 }
5005 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
5006 space_info);
5007 ticket = list_first_entry(&space_info->tickets,
5008 struct reserve_ticket, list);
5009 if (last_ticket == ticket) {
5010 flush_state++;
5011 } else {
5012 last_ticket = ticket;
5013 flush_state = FLUSH_DELAYED_ITEMS_NR;
5014 if (commit_cycles)
5015 commit_cycles--;
5016 }
5017
5018 if (flush_state > COMMIT_TRANS) {
5019 commit_cycles++;
5020 if (commit_cycles > 2) {
5021 wake_all_tickets(&space_info->tickets);
5022 space_info->flush = 0;
5023 } else {
5024 flush_state = FLUSH_DELAYED_ITEMS_NR;
5025 }
5026 }
5027 spin_unlock(&space_info->lock);
5028 } while (flush_state <= COMMIT_TRANS);
5029 }
5030
5031 void btrfs_init_async_reclaim_work(struct work_struct *work)
5032 {
5033 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
5034 }
5035
5036 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
5037 struct btrfs_space_info *space_info,
5038 struct reserve_ticket *ticket)
5039 {
5040 u64 to_reclaim;
5041 int flush_state = FLUSH_DELAYED_ITEMS_NR;
5042
5043 spin_lock(&space_info->lock);
5044 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
5045 space_info);
5046 if (!to_reclaim) {
5047 spin_unlock(&space_info->lock);
5048 return;
5049 }
5050 spin_unlock(&space_info->lock);
5051
5052 do {
5053 flush_space(fs_info->fs_root, space_info, to_reclaim,
5054 to_reclaim, flush_state);
5055 flush_state++;
5056 spin_lock(&space_info->lock);
5057 if (ticket->bytes == 0) {
5058 spin_unlock(&space_info->lock);
5059 return;
5060 }
5061 spin_unlock(&space_info->lock);
5062
5063 /*
5064 * Priority flushers can't wait on delalloc without
5065 * deadlocking.
5066 */
5067 if (flush_state == FLUSH_DELALLOC ||
5068 flush_state == FLUSH_DELALLOC_WAIT)
5069 flush_state = ALLOC_CHUNK;
5070 } while (flush_state < COMMIT_TRANS);
5071 }
5072
5073 static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
5074 struct btrfs_space_info *space_info,
5075 struct reserve_ticket *ticket, u64 orig_bytes)
5076
5077 {
5078 DEFINE_WAIT(wait);
5079 int ret = 0;
5080
5081 spin_lock(&space_info->lock);
5082 while (ticket->bytes > 0 && ticket->error == 0) {
5083 ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE);
5084 if (ret) {
5085 ret = -EINTR;
5086 break;
5087 }
5088 spin_unlock(&space_info->lock);
5089
5090 schedule();
5091
5092 finish_wait(&ticket->wait, &wait);
5093 spin_lock(&space_info->lock);
5094 }
5095 if (!ret)
5096 ret = ticket->error;
5097 if (!list_empty(&ticket->list))
5098 list_del_init(&ticket->list);
5099 if (ticket->bytes && ticket->bytes < orig_bytes) {
5100 u64 num_bytes = orig_bytes - ticket->bytes;
5101 space_info->bytes_may_use -= num_bytes;
5102 trace_btrfs_space_reservation(fs_info, "space_info",
5103 space_info->flags, num_bytes, 0);
5104 }
5105 spin_unlock(&space_info->lock);
5106
5107 return ret;
5108 }
5109
5110 /**
5111 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5112 * @root - the root we're allocating for
5113 * @space_info - the space info we want to allocate from
5114 * @orig_bytes - the number of bytes we want
5115 * @flush - whether or not we can flush to make our reservation
5116 *
5117 * This will reserve orig_bytes number of bytes from the space info associated
5118 * with the block_rsv. If there is not enough space it will make an attempt to
5119 * flush out space to make room. It will do this by flushing delalloc if
5120 * possible or committing the transaction. If flush is 0 then no attempts to
5121 * regain reservations will be made and this will fail if there is not enough
5122 * space already.
5123 */
5124 static int __reserve_metadata_bytes(struct btrfs_root *root,
5125 struct btrfs_space_info *space_info,
5126 u64 orig_bytes,
5127 enum btrfs_reserve_flush_enum flush)
5128 {
5129 struct reserve_ticket ticket;
5130 u64 used;
5131 int ret = 0;
5132
5133 ASSERT(orig_bytes);
5134 spin_lock(&space_info->lock);
5135 ret = -ENOSPC;
5136 used = space_info->bytes_used + space_info->bytes_reserved +
5137 space_info->bytes_pinned + space_info->bytes_readonly +
5138 space_info->bytes_may_use;
5139
5140 /*
5141 * If we have enough space then hooray, make our reservation and carry
5142 * on. If not see if we can overcommit, and if we can, hooray carry on.
5143 * If not things get more complicated.
5144 */
5145 if (used + orig_bytes <= space_info->total_bytes) {
5146 space_info->bytes_may_use += orig_bytes;
5147 trace_btrfs_space_reservation(root->fs_info, "space_info",
5148 space_info->flags, orig_bytes,
5149 1);
5150 ret = 0;
5151 } else if (can_overcommit(root, space_info, orig_bytes, flush)) {
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 }
5158
5159 /*
5160 * If we couldn't make a reservation then setup our reservation ticket
5161 * and kick the async worker if it's not already running.
5162 *
5163 * If we are a priority flusher then we just need to add our ticket to
5164 * the list and we will do our own flushing further down.
5165 */
5166 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5167 ticket.bytes = orig_bytes;
5168 ticket.error = 0;
5169 init_waitqueue_head(&ticket.wait);
5170 if (flush == BTRFS_RESERVE_FLUSH_ALL) {
5171 list_add_tail(&ticket.list, &space_info->tickets);
5172 if (!space_info->flush) {
5173 space_info->flush = 1;
5174 trace_btrfs_trigger_flush(root->fs_info,
5175 space_info->flags,
5176 orig_bytes, flush,
5177 "enospc");
5178 queue_work(system_unbound_wq,
5179 &root->fs_info->async_reclaim_work);
5180 }
5181 } else {
5182 list_add_tail(&ticket.list,
5183 &space_info->priority_tickets);
5184 }
5185 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5186 used += orig_bytes;
5187 /*
5188 * We will do the space reservation dance during log replay,
5189 * which means we won't have fs_info->fs_root set, so don't do
5190 * the async reclaim as we will panic.
5191 */
5192 if (!root->fs_info->log_root_recovering &&
5193 need_do_async_reclaim(space_info, root->fs_info, used) &&
5194 !work_busy(&root->fs_info->async_reclaim_work)) {
5195 trace_btrfs_trigger_flush(root->fs_info,
5196 space_info->flags,
5197 orig_bytes, flush,
5198 "preempt");
5199 queue_work(system_unbound_wq,
5200 &root->fs_info->async_reclaim_work);
5201 }
5202 }
5203 spin_unlock(&space_info->lock);
5204 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5205 return ret;
5206
5207 if (flush == BTRFS_RESERVE_FLUSH_ALL)
5208 return wait_reserve_ticket(root->fs_info, space_info, &ticket,
5209 orig_bytes);
5210
5211 ret = 0;
5212 priority_reclaim_metadata_space(root->fs_info, space_info, &ticket);
5213 spin_lock(&space_info->lock);
5214 if (ticket.bytes) {
5215 if (ticket.bytes < orig_bytes) {
5216 u64 num_bytes = orig_bytes - ticket.bytes;
5217 space_info->bytes_may_use -= num_bytes;
5218 trace_btrfs_space_reservation(root->fs_info,
5219 "space_info", space_info->flags,
5220 num_bytes, 0);
5221
5222 }
5223 list_del_init(&ticket.list);
5224 ret = -ENOSPC;
5225 }
5226 spin_unlock(&space_info->lock);
5227 ASSERT(list_empty(&ticket.list));
5228 return ret;
5229 }
5230
5231 /**
5232 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5233 * @root - the root we're allocating for
5234 * @block_rsv - the block_rsv we're allocating for
5235 * @orig_bytes - the number of bytes we want
5236 * @flush - whether or not we can flush to make our reservation
5237 *
5238 * This will reserve orgi_bytes number of bytes from the space info associated
5239 * with the block_rsv. If there is not enough space it will make an attempt to
5240 * flush out space to make room. It will do this by flushing delalloc if
5241 * possible or committing the transaction. If flush is 0 then no attempts to
5242 * regain reservations will be made and this will fail if there is not enough
5243 * space already.
5244 */
5245 static int reserve_metadata_bytes(struct btrfs_root *root,
5246 struct btrfs_block_rsv *block_rsv,
5247 u64 orig_bytes,
5248 enum btrfs_reserve_flush_enum flush)
5249 {
5250 int ret;
5251
5252 ret = __reserve_metadata_bytes(root, block_rsv->space_info, orig_bytes,
5253 flush);
5254 if (ret == -ENOSPC &&
5255 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5256 struct btrfs_block_rsv *global_rsv =
5257 &root->fs_info->global_block_rsv;
5258
5259 if (block_rsv != global_rsv &&
5260 !block_rsv_use_bytes(global_rsv, orig_bytes))
5261 ret = 0;
5262 }
5263 if (ret == -ENOSPC)
5264 trace_btrfs_space_reservation(root->fs_info,
5265 "space_info:enospc",
5266 block_rsv->space_info->flags,
5267 orig_bytes, 1);
5268 return ret;
5269 }
5270
5271 static struct btrfs_block_rsv *get_block_rsv(
5272 const struct btrfs_trans_handle *trans,
5273 const struct btrfs_root *root)
5274 {
5275 struct btrfs_block_rsv *block_rsv = NULL;
5276
5277 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5278 (root == root->fs_info->csum_root && trans->adding_csums) ||
5279 (root == root->fs_info->uuid_root))
5280 block_rsv = trans->block_rsv;
5281
5282 if (!block_rsv)
5283 block_rsv = root->block_rsv;
5284
5285 if (!block_rsv)
5286 block_rsv = &root->fs_info->empty_block_rsv;
5287
5288 return block_rsv;
5289 }
5290
5291 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5292 u64 num_bytes)
5293 {
5294 int ret = -ENOSPC;
5295 spin_lock(&block_rsv->lock);
5296 if (block_rsv->reserved >= num_bytes) {
5297 block_rsv->reserved -= num_bytes;
5298 if (block_rsv->reserved < block_rsv->size)
5299 block_rsv->full = 0;
5300 ret = 0;
5301 }
5302 spin_unlock(&block_rsv->lock);
5303 return ret;
5304 }
5305
5306 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5307 u64 num_bytes, int update_size)
5308 {
5309 spin_lock(&block_rsv->lock);
5310 block_rsv->reserved += num_bytes;
5311 if (update_size)
5312 block_rsv->size += num_bytes;
5313 else if (block_rsv->reserved >= block_rsv->size)
5314 block_rsv->full = 1;
5315 spin_unlock(&block_rsv->lock);
5316 }
5317
5318 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5319 struct btrfs_block_rsv *dest, u64 num_bytes,
5320 int min_factor)
5321 {
5322 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5323 u64 min_bytes;
5324
5325 if (global_rsv->space_info != dest->space_info)
5326 return -ENOSPC;
5327
5328 spin_lock(&global_rsv->lock);
5329 min_bytes = div_factor(global_rsv->size, min_factor);
5330 if (global_rsv->reserved < min_bytes + num_bytes) {
5331 spin_unlock(&global_rsv->lock);
5332 return -ENOSPC;
5333 }
5334 global_rsv->reserved -= num_bytes;
5335 if (global_rsv->reserved < global_rsv->size)
5336 global_rsv->full = 0;
5337 spin_unlock(&global_rsv->lock);
5338
5339 block_rsv_add_bytes(dest, num_bytes, 1);
5340 return 0;
5341 }
5342
5343 /*
5344 * This is for space we already have accounted in space_info->bytes_may_use, so
5345 * basically when we're returning space from block_rsv's.
5346 */
5347 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
5348 struct btrfs_space_info *space_info,
5349 u64 num_bytes)
5350 {
5351 struct reserve_ticket *ticket;
5352 struct list_head *head;
5353 u64 used;
5354 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH;
5355 bool check_overcommit = false;
5356
5357 spin_lock(&space_info->lock);
5358 head = &space_info->priority_tickets;
5359
5360 /*
5361 * If we are over our limit then we need to check and see if we can
5362 * overcommit, and if we can't then we just need to free up our space
5363 * and not satisfy any requests.
5364 */
5365 used = space_info->bytes_used + space_info->bytes_reserved +
5366 space_info->bytes_pinned + space_info->bytes_readonly +
5367 space_info->bytes_may_use;
5368 if (used - num_bytes >= space_info->total_bytes)
5369 check_overcommit = true;
5370 again:
5371 while (!list_empty(head) && num_bytes) {
5372 ticket = list_first_entry(head, struct reserve_ticket,
5373 list);
5374 /*
5375 * We use 0 bytes because this space is already reserved, so
5376 * adding the ticket space would be a double count.
5377 */
5378 if (check_overcommit &&
5379 !can_overcommit(fs_info->extent_root, space_info, 0,
5380 flush))
5381 break;
5382 if (num_bytes >= ticket->bytes) {
5383 list_del_init(&ticket->list);
5384 num_bytes -= ticket->bytes;
5385 ticket->bytes = 0;
5386 wake_up(&ticket->wait);
5387 } else {
5388 ticket->bytes -= num_bytes;
5389 num_bytes = 0;
5390 }
5391 }
5392
5393 if (num_bytes && head == &space_info->priority_tickets) {
5394 head = &space_info->tickets;
5395 flush = BTRFS_RESERVE_FLUSH_ALL;
5396 goto again;
5397 }
5398 space_info->bytes_may_use -= num_bytes;
5399 trace_btrfs_space_reservation(fs_info, "space_info",
5400 space_info->flags, num_bytes, 0);
5401 spin_unlock(&space_info->lock);
5402 }
5403
5404 /*
5405 * This is for newly allocated space that isn't accounted in
5406 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5407 * we use this helper.
5408 */
5409 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
5410 struct btrfs_space_info *space_info,
5411 u64 num_bytes)
5412 {
5413 struct reserve_ticket *ticket;
5414 struct list_head *head = &space_info->priority_tickets;
5415
5416 again:
5417 while (!list_empty(head) && num_bytes) {
5418 ticket = list_first_entry(head, struct reserve_ticket,
5419 list);
5420 if (num_bytes >= ticket->bytes) {
5421 trace_btrfs_space_reservation(fs_info, "space_info",
5422 space_info->flags,
5423 ticket->bytes, 1);
5424 list_del_init(&ticket->list);
5425 num_bytes -= ticket->bytes;
5426 space_info->bytes_may_use += ticket->bytes;
5427 ticket->bytes = 0;
5428 wake_up(&ticket->wait);
5429 } else {
5430 trace_btrfs_space_reservation(fs_info, "space_info",
5431 space_info->flags,
5432 num_bytes, 1);
5433 space_info->bytes_may_use += num_bytes;
5434 ticket->bytes -= num_bytes;
5435 num_bytes = 0;
5436 }
5437 }
5438
5439 if (num_bytes && head == &space_info->priority_tickets) {
5440 head = &space_info->tickets;
5441 goto again;
5442 }
5443 }
5444
5445 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5446 struct btrfs_block_rsv *block_rsv,
5447 struct btrfs_block_rsv *dest, u64 num_bytes)
5448 {
5449 struct btrfs_space_info *space_info = block_rsv->space_info;
5450
5451 spin_lock(&block_rsv->lock);
5452 if (num_bytes == (u64)-1)
5453 num_bytes = block_rsv->size;
5454 block_rsv->size -= num_bytes;
5455 if (block_rsv->reserved >= block_rsv->size) {
5456 num_bytes = block_rsv->reserved - block_rsv->size;
5457 block_rsv->reserved = block_rsv->size;
5458 block_rsv->full = 1;
5459 } else {
5460 num_bytes = 0;
5461 }
5462 spin_unlock(&block_rsv->lock);
5463
5464 if (num_bytes > 0) {
5465 if (dest) {
5466 spin_lock(&dest->lock);
5467 if (!dest->full) {
5468 u64 bytes_to_add;
5469
5470 bytes_to_add = dest->size - dest->reserved;
5471 bytes_to_add = min(num_bytes, bytes_to_add);
5472 dest->reserved += bytes_to_add;
5473 if (dest->reserved >= dest->size)
5474 dest->full = 1;
5475 num_bytes -= bytes_to_add;
5476 }
5477 spin_unlock(&dest->lock);
5478 }
5479 if (num_bytes)
5480 space_info_add_old_bytes(fs_info, space_info,
5481 num_bytes);
5482 }
5483 }
5484
5485 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
5486 struct btrfs_block_rsv *dst, u64 num_bytes,
5487 int update_size)
5488 {
5489 int ret;
5490
5491 ret = block_rsv_use_bytes(src, num_bytes);
5492 if (ret)
5493 return ret;
5494
5495 block_rsv_add_bytes(dst, num_bytes, update_size);
5496 return 0;
5497 }
5498
5499 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5500 {
5501 memset(rsv, 0, sizeof(*rsv));
5502 spin_lock_init(&rsv->lock);
5503 rsv->type = type;
5504 }
5505
5506 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
5507 unsigned short type)
5508 {
5509 struct btrfs_block_rsv *block_rsv;
5510 struct btrfs_fs_info *fs_info = root->fs_info;
5511
5512 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5513 if (!block_rsv)
5514 return NULL;
5515
5516 btrfs_init_block_rsv(block_rsv, type);
5517 block_rsv->space_info = __find_space_info(fs_info,
5518 BTRFS_BLOCK_GROUP_METADATA);
5519 return block_rsv;
5520 }
5521
5522 void btrfs_free_block_rsv(struct btrfs_root *root,
5523 struct btrfs_block_rsv *rsv)
5524 {
5525 if (!rsv)
5526 return;
5527 btrfs_block_rsv_release(root, rsv, (u64)-1);
5528 kfree(rsv);
5529 }
5530
5531 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5532 {
5533 kfree(rsv);
5534 }
5535
5536 int btrfs_block_rsv_add(struct btrfs_root *root,
5537 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5538 enum btrfs_reserve_flush_enum flush)
5539 {
5540 int ret;
5541
5542 if (num_bytes == 0)
5543 return 0;
5544
5545 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5546 if (!ret) {
5547 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5548 return 0;
5549 }
5550
5551 return ret;
5552 }
5553
5554 int btrfs_block_rsv_check(struct btrfs_root *root,
5555 struct btrfs_block_rsv *block_rsv, int min_factor)
5556 {
5557 u64 num_bytes = 0;
5558 int ret = -ENOSPC;
5559
5560 if (!block_rsv)
5561 return 0;
5562
5563 spin_lock(&block_rsv->lock);
5564 num_bytes = div_factor(block_rsv->size, min_factor);
5565 if (block_rsv->reserved >= num_bytes)
5566 ret = 0;
5567 spin_unlock(&block_rsv->lock);
5568
5569 return ret;
5570 }
5571
5572 int btrfs_block_rsv_refill(struct btrfs_root *root,
5573 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5574 enum btrfs_reserve_flush_enum flush)
5575 {
5576 u64 num_bytes = 0;
5577 int ret = -ENOSPC;
5578
5579 if (!block_rsv)
5580 return 0;
5581
5582 spin_lock(&block_rsv->lock);
5583 num_bytes = min_reserved;
5584 if (block_rsv->reserved >= num_bytes)
5585 ret = 0;
5586 else
5587 num_bytes -= block_rsv->reserved;
5588 spin_unlock(&block_rsv->lock);
5589
5590 if (!ret)
5591 return 0;
5592
5593 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5594 if (!ret) {
5595 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5596 return 0;
5597 }
5598
5599 return ret;
5600 }
5601
5602 void btrfs_block_rsv_release(struct btrfs_root *root,
5603 struct btrfs_block_rsv *block_rsv,
5604 u64 num_bytes)
5605 {
5606 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5607 if (global_rsv == block_rsv ||
5608 block_rsv->space_info != global_rsv->space_info)
5609 global_rsv = NULL;
5610 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
5611 num_bytes);
5612 }
5613
5614 /*
5615 * helper to calculate size of global block reservation.
5616 * the desired value is sum of space used by extent tree,
5617 * checksum tree and root tree
5618 */
5619 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
5620 {
5621 struct btrfs_space_info *sinfo;
5622 u64 num_bytes;
5623 u64 meta_used;
5624 u64 data_used;
5625 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
5626
5627 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
5628 spin_lock(&sinfo->lock);
5629 data_used = sinfo->bytes_used;
5630 spin_unlock(&sinfo->lock);
5631
5632 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5633 spin_lock(&sinfo->lock);
5634 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
5635 data_used = 0;
5636 meta_used = sinfo->bytes_used;
5637 spin_unlock(&sinfo->lock);
5638
5639 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
5640 csum_size * 2;
5641 num_bytes += div_u64(data_used + meta_used, 50);
5642
5643 if (num_bytes * 3 > meta_used)
5644 num_bytes = div_u64(meta_used, 3);
5645
5646 return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
5647 }
5648
5649 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5650 {
5651 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5652 struct btrfs_space_info *sinfo = block_rsv->space_info;
5653 u64 num_bytes;
5654
5655 num_bytes = calc_global_metadata_size(fs_info);
5656
5657 spin_lock(&sinfo->lock);
5658 spin_lock(&block_rsv->lock);
5659
5660 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5661
5662 if (block_rsv->reserved < block_rsv->size) {
5663 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
5664 sinfo->bytes_reserved + sinfo->bytes_readonly +
5665 sinfo->bytes_may_use;
5666 if (sinfo->total_bytes > num_bytes) {
5667 num_bytes = sinfo->total_bytes - num_bytes;
5668 num_bytes = min(num_bytes,
5669 block_rsv->size - block_rsv->reserved);
5670 block_rsv->reserved += num_bytes;
5671 sinfo->bytes_may_use += num_bytes;
5672 trace_btrfs_space_reservation(fs_info, "space_info",
5673 sinfo->flags, num_bytes,
5674 1);
5675 }
5676 } else if (block_rsv->reserved > block_rsv->size) {
5677 num_bytes = block_rsv->reserved - block_rsv->size;
5678 sinfo->bytes_may_use -= num_bytes;
5679 trace_btrfs_space_reservation(fs_info, "space_info",
5680 sinfo->flags, num_bytes, 0);
5681 block_rsv->reserved = block_rsv->size;
5682 }
5683
5684 if (block_rsv->reserved == block_rsv->size)
5685 block_rsv->full = 1;
5686 else
5687 block_rsv->full = 0;
5688
5689 spin_unlock(&block_rsv->lock);
5690 spin_unlock(&sinfo->lock);
5691 }
5692
5693 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5694 {
5695 struct btrfs_space_info *space_info;
5696
5697 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5698 fs_info->chunk_block_rsv.space_info = space_info;
5699
5700 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5701 fs_info->global_block_rsv.space_info = space_info;
5702 fs_info->delalloc_block_rsv.space_info = space_info;
5703 fs_info->trans_block_rsv.space_info = space_info;
5704 fs_info->empty_block_rsv.space_info = space_info;
5705 fs_info->delayed_block_rsv.space_info = space_info;
5706
5707 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5708 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5709 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5710 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5711 if (fs_info->quota_root)
5712 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5713 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5714
5715 update_global_block_rsv(fs_info);
5716 }
5717
5718 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5719 {
5720 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5721 (u64)-1);
5722 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5723 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5724 WARN_ON(fs_info->trans_block_rsv.size > 0);
5725 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5726 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5727 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5728 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5729 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5730 }
5731
5732 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5733 struct btrfs_root *root)
5734 {
5735 if (!trans->block_rsv)
5736 return;
5737
5738 if (!trans->bytes_reserved)
5739 return;
5740
5741 trace_btrfs_space_reservation(root->fs_info, "transaction",
5742 trans->transid, trans->bytes_reserved, 0);
5743 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
5744 trans->bytes_reserved = 0;
5745 }
5746
5747 /*
5748 * To be called after all the new block groups attached to the transaction
5749 * handle have been created (btrfs_create_pending_block_groups()).
5750 */
5751 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5752 {
5753 struct btrfs_fs_info *fs_info = trans->root->fs_info;
5754
5755 if (!trans->chunk_bytes_reserved)
5756 return;
5757
5758 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5759
5760 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5761 trans->chunk_bytes_reserved);
5762 trans->chunk_bytes_reserved = 0;
5763 }
5764
5765 /* Can only return 0 or -ENOSPC */
5766 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5767 struct inode *inode)
5768 {
5769 struct btrfs_root *root = BTRFS_I(inode)->root;
5770 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
5771 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5772
5773 /*
5774 * We need to hold space in order to delete our orphan item once we've
5775 * added it, so this takes the reservation so we can release it later
5776 * when we are truly done with the orphan item.
5777 */
5778 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5779 trace_btrfs_space_reservation(root->fs_info, "orphan",
5780 btrfs_ino(inode), num_bytes, 1);
5781 return btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
5782 }
5783
5784 void btrfs_orphan_release_metadata(struct inode *inode)
5785 {
5786 struct btrfs_root *root = BTRFS_I(inode)->root;
5787 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5788 trace_btrfs_space_reservation(root->fs_info, "orphan",
5789 btrfs_ino(inode), num_bytes, 0);
5790 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
5791 }
5792
5793 /*
5794 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5795 * root: the root of the parent directory
5796 * rsv: block reservation
5797 * items: the number of items that we need do reservation
5798 * qgroup_reserved: used to return the reserved size in qgroup
5799 *
5800 * This function is used to reserve the space for snapshot/subvolume
5801 * creation and deletion. Those operations are different with the
5802 * common file/directory operations, they change two fs/file trees
5803 * and root tree, the number of items that the qgroup reserves is
5804 * different with the free space reservation. So we can not use
5805 * the space reservation mechanism in start_transaction().
5806 */
5807 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5808 struct btrfs_block_rsv *rsv,
5809 int items,
5810 u64 *qgroup_reserved,
5811 bool use_global_rsv)
5812 {
5813 u64 num_bytes;
5814 int ret;
5815 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5816
5817 if (root->fs_info->quota_enabled) {
5818 /* One for parent inode, two for dir entries */
5819 num_bytes = 3 * root->nodesize;
5820 ret = btrfs_qgroup_reserve_meta(root, num_bytes);
5821 if (ret)
5822 return ret;
5823 } else {
5824 num_bytes = 0;
5825 }
5826
5827 *qgroup_reserved = num_bytes;
5828
5829 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5830 rsv->space_info = __find_space_info(root->fs_info,
5831 BTRFS_BLOCK_GROUP_METADATA);
5832 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5833 BTRFS_RESERVE_FLUSH_ALL);
5834
5835 if (ret == -ENOSPC && use_global_rsv)
5836 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1);
5837
5838 if (ret && *qgroup_reserved)
5839 btrfs_qgroup_free_meta(root, *qgroup_reserved);
5840
5841 return ret;
5842 }
5843
5844 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5845 struct btrfs_block_rsv *rsv,
5846 u64 qgroup_reserved)
5847 {
5848 btrfs_block_rsv_release(root, rsv, (u64)-1);
5849 }
5850
5851 /**
5852 * drop_outstanding_extent - drop an outstanding extent
5853 * @inode: the inode we're dropping the extent for
5854 * @num_bytes: the number of bytes we're releasing.
5855 *
5856 * This is called when we are freeing up an outstanding extent, either called
5857 * after an error or after an extent is written. This will return the number of
5858 * reserved extents that need to be freed. This must be called with
5859 * BTRFS_I(inode)->lock held.
5860 */
5861 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5862 {
5863 unsigned drop_inode_space = 0;
5864 unsigned dropped_extents = 0;
5865 unsigned num_extents = 0;
5866
5867 num_extents = (unsigned)div64_u64(num_bytes +
5868 BTRFS_MAX_EXTENT_SIZE - 1,
5869 BTRFS_MAX_EXTENT_SIZE);
5870 ASSERT(num_extents);
5871 ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5872 BTRFS_I(inode)->outstanding_extents -= num_extents;
5873
5874 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5875 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5876 &BTRFS_I(inode)->runtime_flags))
5877 drop_inode_space = 1;
5878
5879 /*
5880 * If we have more or the same amount of outstanding extents than we have
5881 * reserved then we need to leave the reserved extents count alone.
5882 */
5883 if (BTRFS_I(inode)->outstanding_extents >=
5884 BTRFS_I(inode)->reserved_extents)
5885 return drop_inode_space;
5886
5887 dropped_extents = BTRFS_I(inode)->reserved_extents -
5888 BTRFS_I(inode)->outstanding_extents;
5889 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5890 return dropped_extents + drop_inode_space;
5891 }
5892
5893 /**
5894 * calc_csum_metadata_size - return the amount of metadata space that must be
5895 * reserved/freed for the given bytes.
5896 * @inode: the inode we're manipulating
5897 * @num_bytes: the number of bytes in question
5898 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5899 *
5900 * This adjusts the number of csum_bytes in the inode and then returns the
5901 * correct amount of metadata that must either be reserved or freed. We
5902 * calculate how many checksums we can fit into one leaf and then divide the
5903 * number of bytes that will need to be checksumed by this value to figure out
5904 * how many checksums will be required. If we are adding bytes then the number
5905 * may go up and we will return the number of additional bytes that must be
5906 * reserved. If it is going down we will return the number of bytes that must
5907 * be freed.
5908 *
5909 * This must be called with BTRFS_I(inode)->lock held.
5910 */
5911 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5912 int reserve)
5913 {
5914 struct btrfs_root *root = BTRFS_I(inode)->root;
5915 u64 old_csums, num_csums;
5916
5917 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5918 BTRFS_I(inode)->csum_bytes == 0)
5919 return 0;
5920
5921 old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5922 if (reserve)
5923 BTRFS_I(inode)->csum_bytes += num_bytes;
5924 else
5925 BTRFS_I(inode)->csum_bytes -= num_bytes;
5926 num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5927
5928 /* No change, no need to reserve more */
5929 if (old_csums == num_csums)
5930 return 0;
5931
5932 if (reserve)
5933 return btrfs_calc_trans_metadata_size(root,
5934 num_csums - old_csums);
5935
5936 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5937 }
5938
5939 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5940 {
5941 struct btrfs_root *root = BTRFS_I(inode)->root;
5942 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5943 u64 to_reserve = 0;
5944 u64 csum_bytes;
5945 unsigned nr_extents = 0;
5946 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5947 int ret = 0;
5948 bool delalloc_lock = true;
5949 u64 to_free = 0;
5950 unsigned dropped;
5951 bool release_extra = false;
5952
5953 /* If we are a free space inode we need to not flush since we will be in
5954 * the middle of a transaction commit. We also don't need the delalloc
5955 * mutex since we won't race with anybody. We need this mostly to make
5956 * lockdep shut its filthy mouth.
5957 */
5958 if (btrfs_is_free_space_inode(inode)) {
5959 flush = BTRFS_RESERVE_NO_FLUSH;
5960 delalloc_lock = false;
5961 }
5962
5963 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5964 btrfs_transaction_in_commit(root->fs_info))
5965 schedule_timeout(1);
5966
5967 if (delalloc_lock)
5968 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5969
5970 num_bytes = ALIGN(num_bytes, root->sectorsize);
5971
5972 spin_lock(&BTRFS_I(inode)->lock);
5973 nr_extents = (unsigned)div64_u64(num_bytes +
5974 BTRFS_MAX_EXTENT_SIZE - 1,
5975 BTRFS_MAX_EXTENT_SIZE);
5976 BTRFS_I(inode)->outstanding_extents += nr_extents;
5977
5978 nr_extents = 0;
5979 if (BTRFS_I(inode)->outstanding_extents >
5980 BTRFS_I(inode)->reserved_extents)
5981 nr_extents += BTRFS_I(inode)->outstanding_extents -
5982 BTRFS_I(inode)->reserved_extents;
5983
5984 /* We always want to reserve a slot for updating the inode. */
5985 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents + 1);
5986 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5987 csum_bytes = BTRFS_I(inode)->csum_bytes;
5988 spin_unlock(&BTRFS_I(inode)->lock);
5989
5990 if (root->fs_info->quota_enabled) {
5991 ret = btrfs_qgroup_reserve_meta(root,
5992 nr_extents * root->nodesize);
5993 if (ret)
5994 goto out_fail;
5995 }
5996
5997 ret = btrfs_block_rsv_add(root, block_rsv, to_reserve, flush);
5998 if (unlikely(ret)) {
5999 btrfs_qgroup_free_meta(root, nr_extents * root->nodesize);
6000 goto out_fail;
6001 }
6002
6003 spin_lock(&BTRFS_I(inode)->lock);
6004 if (test_and_set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
6005 &BTRFS_I(inode)->runtime_flags)) {
6006 to_reserve -= btrfs_calc_trans_metadata_size(root, 1);
6007 release_extra = true;
6008 }
6009 BTRFS_I(inode)->reserved_extents += nr_extents;
6010 spin_unlock(&BTRFS_I(inode)->lock);
6011
6012 if (delalloc_lock)
6013 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
6014
6015 if (to_reserve)
6016 trace_btrfs_space_reservation(root->fs_info, "delalloc",
6017 btrfs_ino(inode), to_reserve, 1);
6018 if (release_extra)
6019 btrfs_block_rsv_release(root, block_rsv,
6020 btrfs_calc_trans_metadata_size(root,
6021 1));
6022 return 0;
6023
6024 out_fail:
6025 spin_lock(&BTRFS_I(inode)->lock);
6026 dropped = drop_outstanding_extent(inode, num_bytes);
6027 /*
6028 * If the inodes csum_bytes is the same as the original
6029 * csum_bytes then we know we haven't raced with any free()ers
6030 * so we can just reduce our inodes csum bytes and carry on.
6031 */
6032 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
6033 calc_csum_metadata_size(inode, num_bytes, 0);
6034 } else {
6035 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
6036 u64 bytes;
6037
6038 /*
6039 * This is tricky, but first we need to figure out how much we
6040 * freed from any free-ers that occurred during this
6041 * reservation, so we reset ->csum_bytes to the csum_bytes
6042 * before we dropped our lock, and then call the free for the
6043 * number of bytes that were freed while we were trying our
6044 * reservation.
6045 */
6046 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
6047 BTRFS_I(inode)->csum_bytes = csum_bytes;
6048 to_free = calc_csum_metadata_size(inode, bytes, 0);
6049
6050
6051 /*
6052 * Now we need to see how much we would have freed had we not
6053 * been making this reservation and our ->csum_bytes were not
6054 * artificially inflated.
6055 */
6056 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
6057 bytes = csum_bytes - orig_csum_bytes;
6058 bytes = calc_csum_metadata_size(inode, bytes, 0);
6059
6060 /*
6061 * Now reset ->csum_bytes to what it should be. If bytes is
6062 * more than to_free then we would have freed more space had we
6063 * not had an artificially high ->csum_bytes, so we need to free
6064 * the remainder. If bytes is the same or less then we don't
6065 * need to do anything, the other free-ers did the correct
6066 * thing.
6067 */
6068 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
6069 if (bytes > to_free)
6070 to_free = bytes - to_free;
6071 else
6072 to_free = 0;
6073 }
6074 spin_unlock(&BTRFS_I(inode)->lock);
6075 if (dropped)
6076 to_free += btrfs_calc_trans_metadata_size(root, dropped);
6077
6078 if (to_free) {
6079 btrfs_block_rsv_release(root, block_rsv, to_free);
6080 trace_btrfs_space_reservation(root->fs_info, "delalloc",
6081 btrfs_ino(inode), to_free, 0);
6082 }
6083 if (delalloc_lock)
6084 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
6085 return ret;
6086 }
6087
6088 /**
6089 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6090 * @inode: the inode to release the reservation for
6091 * @num_bytes: the number of bytes we're releasing
6092 *
6093 * This will release the metadata reservation for an inode. This can be called
6094 * once we complete IO for a given set of bytes to release their metadata
6095 * reservations.
6096 */
6097 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
6098 {
6099 struct btrfs_root *root = BTRFS_I(inode)->root;
6100 u64 to_free = 0;
6101 unsigned dropped;
6102
6103 num_bytes = ALIGN(num_bytes, root->sectorsize);
6104 spin_lock(&BTRFS_I(inode)->lock);
6105 dropped = drop_outstanding_extent(inode, num_bytes);
6106
6107 if (num_bytes)
6108 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
6109 spin_unlock(&BTRFS_I(inode)->lock);
6110 if (dropped > 0)
6111 to_free += btrfs_calc_trans_metadata_size(root, dropped);
6112
6113 if (btrfs_test_is_dummy_root(root))
6114 return;
6115
6116 trace_btrfs_space_reservation(root->fs_info, "delalloc",
6117 btrfs_ino(inode), to_free, 0);
6118
6119 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
6120 to_free);
6121 }
6122
6123 /**
6124 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6125 * delalloc
6126 * @inode: inode we're writing to
6127 * @start: start range we are writing to
6128 * @len: how long the range we are writing to
6129 *
6130 * TODO: This function will finally replace old btrfs_delalloc_reserve_space()
6131 *
6132 * This will do the following things
6133 *
6134 * o reserve space in data space info for num bytes
6135 * and reserve precious corresponding qgroup space
6136 * (Done in check_data_free_space)
6137 *
6138 * o reserve space for metadata space, based on the number of outstanding
6139 * extents and how much csums will be needed
6140 * also reserve metadata space in a per root over-reserve method.
6141 * o add to the inodes->delalloc_bytes
6142 * o add it to the fs_info's delalloc inodes list.
6143 * (Above 3 all done in delalloc_reserve_metadata)
6144 *
6145 * Return 0 for success
6146 * Return <0 for error(-ENOSPC or -EQUOT)
6147 */
6148 int btrfs_delalloc_reserve_space(struct inode *inode, u64 start, u64 len)
6149 {
6150 int ret;
6151
6152 ret = btrfs_check_data_free_space(inode, start, len);
6153 if (ret < 0)
6154 return ret;
6155 ret = btrfs_delalloc_reserve_metadata(inode, len);
6156 if (ret < 0)
6157 btrfs_free_reserved_data_space(inode, start, len);
6158 return ret;
6159 }
6160
6161 /**
6162 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6163 * @inode: inode we're releasing space for
6164 * @start: start position of the space already reserved
6165 * @len: the len of the space already reserved
6166 *
6167 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
6168 * called in the case that we don't need the metadata AND data reservations
6169 * anymore. So if there is an error or we insert an inline extent.
6170 *
6171 * This function will release the metadata space that was not used and will
6172 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6173 * list if there are no delalloc bytes left.
6174 * Also it will handle the qgroup reserved space.
6175 */
6176 void btrfs_delalloc_release_space(struct inode *inode, u64 start, u64 len)
6177 {
6178 btrfs_delalloc_release_metadata(inode, len);
6179 btrfs_free_reserved_data_space(inode, start, len);
6180 }
6181
6182 static int update_block_group(struct btrfs_trans_handle *trans,
6183 struct btrfs_root *root, u64 bytenr,
6184 u64 num_bytes, int alloc)
6185 {
6186 struct btrfs_block_group_cache *cache = NULL;
6187 struct btrfs_fs_info *info = root->fs_info;
6188 u64 total = num_bytes;
6189 u64 old_val;
6190 u64 byte_in_group;
6191 int factor;
6192
6193 /* block accounting for super block */
6194 spin_lock(&info->delalloc_root_lock);
6195 old_val = btrfs_super_bytes_used(info->super_copy);
6196 if (alloc)
6197 old_val += num_bytes;
6198 else
6199 old_val -= num_bytes;
6200 btrfs_set_super_bytes_used(info->super_copy, old_val);
6201 spin_unlock(&info->delalloc_root_lock);
6202
6203 while (total) {
6204 cache = btrfs_lookup_block_group(info, bytenr);
6205 if (!cache)
6206 return -ENOENT;
6207 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
6208 BTRFS_BLOCK_GROUP_RAID1 |
6209 BTRFS_BLOCK_GROUP_RAID10))
6210 factor = 2;
6211 else
6212 factor = 1;
6213 /*
6214 * If this block group has free space cache written out, we
6215 * need to make sure to load it if we are removing space. This
6216 * is because we need the unpinning stage to actually add the
6217 * space back to the block group, otherwise we will leak space.
6218 */
6219 if (!alloc && cache->cached == BTRFS_CACHE_NO)
6220 cache_block_group(cache, 1);
6221
6222 byte_in_group = bytenr - cache->key.objectid;
6223 WARN_ON(byte_in_group > cache->key.offset);
6224
6225 spin_lock(&cache->space_info->lock);
6226 spin_lock(&cache->lock);
6227
6228 if (btrfs_test_opt(root, SPACE_CACHE) &&
6229 cache->disk_cache_state < BTRFS_DC_CLEAR)
6230 cache->disk_cache_state = BTRFS_DC_CLEAR;
6231
6232 old_val = btrfs_block_group_used(&cache->item);
6233 num_bytes = min(total, cache->key.offset - byte_in_group);
6234 if (alloc) {
6235 old_val += num_bytes;
6236 btrfs_set_block_group_used(&cache->item, old_val);
6237 cache->reserved -= num_bytes;
6238 cache->space_info->bytes_reserved -= num_bytes;
6239 cache->space_info->bytes_used += num_bytes;
6240 cache->space_info->disk_used += num_bytes * factor;
6241 spin_unlock(&cache->lock);
6242 spin_unlock(&cache->space_info->lock);
6243 } else {
6244 old_val -= num_bytes;
6245 btrfs_set_block_group_used(&cache->item, old_val);
6246 cache->pinned += num_bytes;
6247 cache->space_info->bytes_pinned += num_bytes;
6248 cache->space_info->bytes_used -= num_bytes;
6249 cache->space_info->disk_used -= num_bytes * factor;
6250 spin_unlock(&cache->lock);
6251 spin_unlock(&cache->space_info->lock);
6252
6253 trace_btrfs_space_reservation(root->fs_info, "pinned",
6254 cache->space_info->flags,
6255 num_bytes, 1);
6256 set_extent_dirty(info->pinned_extents,
6257 bytenr, bytenr + num_bytes - 1,
6258 GFP_NOFS | __GFP_NOFAIL);
6259 }
6260
6261 spin_lock(&trans->transaction->dirty_bgs_lock);
6262 if (list_empty(&cache->dirty_list)) {
6263 list_add_tail(&cache->dirty_list,
6264 &trans->transaction->dirty_bgs);
6265 trans->transaction->num_dirty_bgs++;
6266 btrfs_get_block_group(cache);
6267 }
6268 spin_unlock(&trans->transaction->dirty_bgs_lock);
6269
6270 /*
6271 * No longer have used bytes in this block group, queue it for
6272 * deletion. We do this after adding the block group to the
6273 * dirty list to avoid races between cleaner kthread and space
6274 * cache writeout.
6275 */
6276 if (!alloc && old_val == 0) {
6277 spin_lock(&info->unused_bgs_lock);
6278 if (list_empty(&cache->bg_list)) {
6279 btrfs_get_block_group(cache);
6280 list_add_tail(&cache->bg_list,
6281 &info->unused_bgs);
6282 }
6283 spin_unlock(&info->unused_bgs_lock);
6284 }
6285
6286 btrfs_put_block_group(cache);
6287 total -= num_bytes;
6288 bytenr += num_bytes;
6289 }
6290 return 0;
6291 }
6292
6293 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
6294 {
6295 struct btrfs_block_group_cache *cache;
6296 u64 bytenr;
6297
6298 spin_lock(&root->fs_info->block_group_cache_lock);
6299 bytenr = root->fs_info->first_logical_byte;
6300 spin_unlock(&root->fs_info->block_group_cache_lock);
6301
6302 if (bytenr < (u64)-1)
6303 return bytenr;
6304
6305 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
6306 if (!cache)
6307 return 0;
6308
6309 bytenr = cache->key.objectid;
6310 btrfs_put_block_group(cache);
6311
6312 return bytenr;
6313 }
6314
6315 static int pin_down_extent(struct btrfs_root *root,
6316 struct btrfs_block_group_cache *cache,
6317 u64 bytenr, u64 num_bytes, int reserved)
6318 {
6319 spin_lock(&cache->space_info->lock);
6320 spin_lock(&cache->lock);
6321 cache->pinned += num_bytes;
6322 cache->space_info->bytes_pinned += num_bytes;
6323 if (reserved) {
6324 cache->reserved -= num_bytes;
6325 cache->space_info->bytes_reserved -= num_bytes;
6326 }
6327 spin_unlock(&cache->lock);
6328 spin_unlock(&cache->space_info->lock);
6329
6330 trace_btrfs_space_reservation(root->fs_info, "pinned",
6331 cache->space_info->flags, num_bytes, 1);
6332 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
6333 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6334 if (reserved)
6335 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
6336 return 0;
6337 }
6338
6339 /*
6340 * this function must be called within transaction
6341 */
6342 int btrfs_pin_extent(struct btrfs_root *root,
6343 u64 bytenr, u64 num_bytes, int reserved)
6344 {
6345 struct btrfs_block_group_cache *cache;
6346
6347 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
6348 BUG_ON(!cache); /* Logic error */
6349
6350 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
6351
6352 btrfs_put_block_group(cache);
6353 return 0;
6354 }
6355
6356 /*
6357 * this function must be called within transaction
6358 */
6359 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
6360 u64 bytenr, u64 num_bytes)
6361 {
6362 struct btrfs_block_group_cache *cache;
6363 int ret;
6364
6365 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
6366 if (!cache)
6367 return -EINVAL;
6368
6369 /*
6370 * pull in the free space cache (if any) so that our pin
6371 * removes the free space from the cache. We have load_only set
6372 * to one because the slow code to read in the free extents does check
6373 * the pinned extents.
6374 */
6375 cache_block_group(cache, 1);
6376
6377 pin_down_extent(root, cache, bytenr, num_bytes, 0);
6378
6379 /* remove us from the free space cache (if we're there at all) */
6380 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6381 btrfs_put_block_group(cache);
6382 return ret;
6383 }
6384
6385 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
6386 {
6387 int ret;
6388 struct btrfs_block_group_cache *block_group;
6389 struct btrfs_caching_control *caching_ctl;
6390
6391 block_group = btrfs_lookup_block_group(root->fs_info, start);
6392 if (!block_group)
6393 return -EINVAL;
6394
6395 cache_block_group(block_group, 0);
6396 caching_ctl = get_caching_control(block_group);
6397
6398 if (!caching_ctl) {
6399 /* Logic error */
6400 BUG_ON(!block_group_cache_done(block_group));
6401 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6402 } else {
6403 mutex_lock(&caching_ctl->mutex);
6404
6405 if (start >= caching_ctl->progress) {
6406 ret = add_excluded_extent(root, start, num_bytes);
6407 } else if (start + num_bytes <= caching_ctl->progress) {
6408 ret = btrfs_remove_free_space(block_group,
6409 start, num_bytes);
6410 } else {
6411 num_bytes = caching_ctl->progress - start;
6412 ret = btrfs_remove_free_space(block_group,
6413 start, num_bytes);
6414 if (ret)
6415 goto out_lock;
6416
6417 num_bytes = (start + num_bytes) -
6418 caching_ctl->progress;
6419 start = caching_ctl->progress;
6420 ret = add_excluded_extent(root, start, num_bytes);
6421 }
6422 out_lock:
6423 mutex_unlock(&caching_ctl->mutex);
6424 put_caching_control(caching_ctl);
6425 }
6426 btrfs_put_block_group(block_group);
6427 return ret;
6428 }
6429
6430 int btrfs_exclude_logged_extents(struct btrfs_root *log,
6431 struct extent_buffer *eb)
6432 {
6433 struct btrfs_file_extent_item *item;
6434 struct btrfs_key key;
6435 int found_type;
6436 int i;
6437
6438 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
6439 return 0;
6440
6441 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6442 btrfs_item_key_to_cpu(eb, &key, i);
6443 if (key.type != BTRFS_EXTENT_DATA_KEY)
6444 continue;
6445 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6446 found_type = btrfs_file_extent_type(eb, item);
6447 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6448 continue;
6449 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6450 continue;
6451 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6452 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6453 __exclude_logged_extent(log, key.objectid, key.offset);
6454 }
6455
6456 return 0;
6457 }
6458
6459 static void
6460 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
6461 {
6462 atomic_inc(&bg->reservations);
6463 }
6464
6465 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
6466 const u64 start)
6467 {
6468 struct btrfs_block_group_cache *bg;
6469
6470 bg = btrfs_lookup_block_group(fs_info, start);
6471 ASSERT(bg);
6472 if (atomic_dec_and_test(&bg->reservations))
6473 wake_up_atomic_t(&bg->reservations);
6474 btrfs_put_block_group(bg);
6475 }
6476
6477 static int btrfs_wait_bg_reservations_atomic_t(atomic_t *a)
6478 {
6479 schedule();
6480 return 0;
6481 }
6482
6483 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg)
6484 {
6485 struct btrfs_space_info *space_info = bg->space_info;
6486
6487 ASSERT(bg->ro);
6488
6489 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
6490 return;
6491
6492 /*
6493 * Our block group is read only but before we set it to read only,
6494 * some task might have had allocated an extent from it already, but it
6495 * has not yet created a respective ordered extent (and added it to a
6496 * root's list of ordered extents).
6497 * Therefore wait for any task currently allocating extents, since the
6498 * block group's reservations counter is incremented while a read lock
6499 * on the groups' semaphore is held and decremented after releasing
6500 * the read access on that semaphore and creating the ordered extent.
6501 */
6502 down_write(&space_info->groups_sem);
6503 up_write(&space_info->groups_sem);
6504
6505 wait_on_atomic_t(&bg->reservations,
6506 btrfs_wait_bg_reservations_atomic_t,
6507 TASK_UNINTERRUPTIBLE);
6508 }
6509
6510 /**
6511 * btrfs_update_reserved_bytes - update the block_group and space info counters
6512 * @cache: The cache we are manipulating
6513 * @num_bytes: The number of bytes in question
6514 * @reserve: One of the reservation enums
6515 * @delalloc: The blocks are allocated for the delalloc write
6516 *
6517 * This is called by the allocator when it reserves space, or by somebody who is
6518 * freeing space that was never actually used on disk. For example if you
6519 * reserve some space for a new leaf in transaction A and before transaction A
6520 * commits you free that leaf, you call this with reserve set to 0 in order to
6521 * clear the reservation.
6522 *
6523 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
6524 * ENOSPC accounting. For data we handle the reservation through clearing the
6525 * delalloc bits in the io_tree. We have to do this since we could end up
6526 * allocating less disk space for the amount of data we have reserved in the
6527 * case of compression.
6528 *
6529 * If this is a reservation and the block group has become read only we cannot
6530 * make the reservation and return -EAGAIN, otherwise this function always
6531 * succeeds.
6532 */
6533 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
6534 u64 num_bytes, int reserve, int delalloc)
6535 {
6536 struct btrfs_space_info *space_info = cache->space_info;
6537 int ret = 0;
6538
6539 spin_lock(&space_info->lock);
6540 spin_lock(&cache->lock);
6541 if (reserve != RESERVE_FREE) {
6542 if (cache->ro) {
6543 ret = -EAGAIN;
6544 } else {
6545 cache->reserved += num_bytes;
6546 space_info->bytes_reserved += num_bytes;
6547 if (reserve == RESERVE_ALLOC) {
6548 trace_btrfs_space_reservation(cache->fs_info,
6549 "space_info", space_info->flags,
6550 num_bytes, 0);
6551 space_info->bytes_may_use -= num_bytes;
6552 }
6553
6554 if (delalloc)
6555 cache->delalloc_bytes += num_bytes;
6556 }
6557 } else {
6558 if (cache->ro)
6559 space_info->bytes_readonly += num_bytes;
6560 cache->reserved -= num_bytes;
6561 space_info->bytes_reserved -= num_bytes;
6562
6563 if (delalloc)
6564 cache->delalloc_bytes -= num_bytes;
6565 }
6566 spin_unlock(&cache->lock);
6567 spin_unlock(&space_info->lock);
6568 return ret;
6569 }
6570
6571 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
6572 struct btrfs_root *root)
6573 {
6574 struct btrfs_fs_info *fs_info = root->fs_info;
6575 struct btrfs_caching_control *next;
6576 struct btrfs_caching_control *caching_ctl;
6577 struct btrfs_block_group_cache *cache;
6578
6579 down_write(&fs_info->commit_root_sem);
6580
6581 list_for_each_entry_safe(caching_ctl, next,
6582 &fs_info->caching_block_groups, list) {
6583 cache = caching_ctl->block_group;
6584 if (block_group_cache_done(cache)) {
6585 cache->last_byte_to_unpin = (u64)-1;
6586 list_del_init(&caching_ctl->list);
6587 put_caching_control(caching_ctl);
6588 } else {
6589 cache->last_byte_to_unpin = caching_ctl->progress;
6590 }
6591 }
6592
6593 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6594 fs_info->pinned_extents = &fs_info->freed_extents[1];
6595 else
6596 fs_info->pinned_extents = &fs_info->freed_extents[0];
6597
6598 up_write(&fs_info->commit_root_sem);
6599
6600 update_global_block_rsv(fs_info);
6601 }
6602
6603 /*
6604 * Returns the free cluster for the given space info and sets empty_cluster to
6605 * what it should be based on the mount options.
6606 */
6607 static struct btrfs_free_cluster *
6608 fetch_cluster_info(struct btrfs_root *root, struct btrfs_space_info *space_info,
6609 u64 *empty_cluster)
6610 {
6611 struct btrfs_free_cluster *ret = NULL;
6612 bool ssd = btrfs_test_opt(root, SSD);
6613
6614 *empty_cluster = 0;
6615 if (btrfs_mixed_space_info(space_info))
6616 return ret;
6617
6618 if (ssd)
6619 *empty_cluster = SZ_2M;
6620 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6621 ret = &root->fs_info->meta_alloc_cluster;
6622 if (!ssd)
6623 *empty_cluster = SZ_64K;
6624 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && ssd) {
6625 ret = &root->fs_info->data_alloc_cluster;
6626 }
6627
6628 return ret;
6629 }
6630
6631 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
6632 const bool return_free_space)
6633 {
6634 struct btrfs_fs_info *fs_info = root->fs_info;
6635 struct btrfs_block_group_cache *cache = NULL;
6636 struct btrfs_space_info *space_info;
6637 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6638 struct btrfs_free_cluster *cluster = NULL;
6639 u64 len;
6640 u64 total_unpinned = 0;
6641 u64 empty_cluster = 0;
6642 bool readonly;
6643
6644 while (start <= end) {
6645 readonly = false;
6646 if (!cache ||
6647 start >= cache->key.objectid + cache->key.offset) {
6648 if (cache)
6649 btrfs_put_block_group(cache);
6650 total_unpinned = 0;
6651 cache = btrfs_lookup_block_group(fs_info, start);
6652 BUG_ON(!cache); /* Logic error */
6653
6654 cluster = fetch_cluster_info(root,
6655 cache->space_info,
6656 &empty_cluster);
6657 empty_cluster <<= 1;
6658 }
6659
6660 len = cache->key.objectid + cache->key.offset - start;
6661 len = min(len, end + 1 - start);
6662
6663 if (start < cache->last_byte_to_unpin) {
6664 len = min(len, cache->last_byte_to_unpin - start);
6665 if (return_free_space)
6666 btrfs_add_free_space(cache, start, len);
6667 }
6668
6669 start += len;
6670 total_unpinned += len;
6671 space_info = cache->space_info;
6672
6673 /*
6674 * If this space cluster has been marked as fragmented and we've
6675 * unpinned enough in this block group to potentially allow a
6676 * cluster to be created inside of it go ahead and clear the
6677 * fragmented check.
6678 */
6679 if (cluster && cluster->fragmented &&
6680 total_unpinned > empty_cluster) {
6681 spin_lock(&cluster->lock);
6682 cluster->fragmented = 0;
6683 spin_unlock(&cluster->lock);
6684 }
6685
6686 spin_lock(&space_info->lock);
6687 spin_lock(&cache->lock);
6688 cache->pinned -= len;
6689 space_info->bytes_pinned -= len;
6690
6691 trace_btrfs_space_reservation(fs_info, "pinned",
6692 space_info->flags, len, 0);
6693 space_info->max_extent_size = 0;
6694 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6695 if (cache->ro) {
6696 space_info->bytes_readonly += len;
6697 readonly = true;
6698 }
6699 spin_unlock(&cache->lock);
6700 if (!readonly && return_free_space &&
6701 global_rsv->space_info == space_info) {
6702 u64 to_add = len;
6703 WARN_ON(!return_free_space);
6704 spin_lock(&global_rsv->lock);
6705 if (!global_rsv->full) {
6706 to_add = min(len, global_rsv->size -
6707 global_rsv->reserved);
6708 global_rsv->reserved += to_add;
6709 space_info->bytes_may_use += to_add;
6710 if (global_rsv->reserved >= global_rsv->size)
6711 global_rsv->full = 1;
6712 trace_btrfs_space_reservation(fs_info,
6713 "space_info",
6714 space_info->flags,
6715 to_add, 1);
6716 len -= to_add;
6717 }
6718 spin_unlock(&global_rsv->lock);
6719 /* Add to any tickets we may have */
6720 if (len)
6721 space_info_add_new_bytes(fs_info, space_info,
6722 len);
6723 }
6724 spin_unlock(&space_info->lock);
6725 }
6726
6727 if (cache)
6728 btrfs_put_block_group(cache);
6729 return 0;
6730 }
6731
6732 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6733 struct btrfs_root *root)
6734 {
6735 struct btrfs_fs_info *fs_info = root->fs_info;
6736 struct btrfs_block_group_cache *block_group, *tmp;
6737 struct list_head *deleted_bgs;
6738 struct extent_io_tree *unpin;
6739 u64 start;
6740 u64 end;
6741 int ret;
6742
6743 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6744 unpin = &fs_info->freed_extents[1];
6745 else
6746 unpin = &fs_info->freed_extents[0];
6747
6748 while (!trans->aborted) {
6749 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6750 ret = find_first_extent_bit(unpin, 0, &start, &end,
6751 EXTENT_DIRTY, NULL);
6752 if (ret) {
6753 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6754 break;
6755 }
6756
6757 if (btrfs_test_opt(root, DISCARD))
6758 ret = btrfs_discard_extent(root, start,
6759 end + 1 - start, NULL);
6760
6761 clear_extent_dirty(unpin, start, end);
6762 unpin_extent_range(root, start, end, true);
6763 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6764 cond_resched();
6765 }
6766
6767 /*
6768 * Transaction is finished. We don't need the lock anymore. We
6769 * do need to clean up the block groups in case of a transaction
6770 * abort.
6771 */
6772 deleted_bgs = &trans->transaction->deleted_bgs;
6773 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6774 u64 trimmed = 0;
6775
6776 ret = -EROFS;
6777 if (!trans->aborted)
6778 ret = btrfs_discard_extent(root,
6779 block_group->key.objectid,
6780 block_group->key.offset,
6781 &trimmed);
6782
6783 list_del_init(&block_group->bg_list);
6784 btrfs_put_block_group_trimming(block_group);
6785 btrfs_put_block_group(block_group);
6786
6787 if (ret) {
6788 const char *errstr = btrfs_decode_error(ret);
6789 btrfs_warn(fs_info,
6790 "Discard failed while removing blockgroup: errno=%d %s\n",
6791 ret, errstr);
6792 }
6793 }
6794
6795 return 0;
6796 }
6797
6798 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
6799 u64 owner, u64 root_objectid)
6800 {
6801 struct btrfs_space_info *space_info;
6802 u64 flags;
6803
6804 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6805 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
6806 flags = BTRFS_BLOCK_GROUP_SYSTEM;
6807 else
6808 flags = BTRFS_BLOCK_GROUP_METADATA;
6809 } else {
6810 flags = BTRFS_BLOCK_GROUP_DATA;
6811 }
6812
6813 space_info = __find_space_info(fs_info, flags);
6814 BUG_ON(!space_info); /* Logic bug */
6815 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
6816 }
6817
6818
6819 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6820 struct btrfs_root *root,
6821 struct btrfs_delayed_ref_node *node, u64 parent,
6822 u64 root_objectid, u64 owner_objectid,
6823 u64 owner_offset, int refs_to_drop,
6824 struct btrfs_delayed_extent_op *extent_op)
6825 {
6826 struct btrfs_key key;
6827 struct btrfs_path *path;
6828 struct btrfs_fs_info *info = root->fs_info;
6829 struct btrfs_root *extent_root = info->extent_root;
6830 struct extent_buffer *leaf;
6831 struct btrfs_extent_item *ei;
6832 struct btrfs_extent_inline_ref *iref;
6833 int ret;
6834 int is_data;
6835 int extent_slot = 0;
6836 int found_extent = 0;
6837 int num_to_del = 1;
6838 u32 item_size;
6839 u64 refs;
6840 u64 bytenr = node->bytenr;
6841 u64 num_bytes = node->num_bytes;
6842 int last_ref = 0;
6843 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6844 SKINNY_METADATA);
6845
6846 path = btrfs_alloc_path();
6847 if (!path)
6848 return -ENOMEM;
6849
6850 path->reada = READA_FORWARD;
6851 path->leave_spinning = 1;
6852
6853 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6854 BUG_ON(!is_data && refs_to_drop != 1);
6855
6856 if (is_data)
6857 skinny_metadata = 0;
6858
6859 ret = lookup_extent_backref(trans, extent_root, path, &iref,
6860 bytenr, num_bytes, parent,
6861 root_objectid, owner_objectid,
6862 owner_offset);
6863 if (ret == 0) {
6864 extent_slot = path->slots[0];
6865 while (extent_slot >= 0) {
6866 btrfs_item_key_to_cpu(path->nodes[0], &key,
6867 extent_slot);
6868 if (key.objectid != bytenr)
6869 break;
6870 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6871 key.offset == num_bytes) {
6872 found_extent = 1;
6873 break;
6874 }
6875 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6876 key.offset == owner_objectid) {
6877 found_extent = 1;
6878 break;
6879 }
6880 if (path->slots[0] - extent_slot > 5)
6881 break;
6882 extent_slot--;
6883 }
6884 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6885 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6886 if (found_extent && item_size < sizeof(*ei))
6887 found_extent = 0;
6888 #endif
6889 if (!found_extent) {
6890 BUG_ON(iref);
6891 ret = remove_extent_backref(trans, extent_root, path,
6892 NULL, refs_to_drop,
6893 is_data, &last_ref);
6894 if (ret) {
6895 btrfs_abort_transaction(trans, extent_root, ret);
6896 goto out;
6897 }
6898 btrfs_release_path(path);
6899 path->leave_spinning = 1;
6900
6901 key.objectid = bytenr;
6902 key.type = BTRFS_EXTENT_ITEM_KEY;
6903 key.offset = num_bytes;
6904
6905 if (!is_data && skinny_metadata) {
6906 key.type = BTRFS_METADATA_ITEM_KEY;
6907 key.offset = owner_objectid;
6908 }
6909
6910 ret = btrfs_search_slot(trans, extent_root,
6911 &key, path, -1, 1);
6912 if (ret > 0 && skinny_metadata && path->slots[0]) {
6913 /*
6914 * Couldn't find our skinny metadata item,
6915 * see if we have ye olde extent item.
6916 */
6917 path->slots[0]--;
6918 btrfs_item_key_to_cpu(path->nodes[0], &key,
6919 path->slots[0]);
6920 if (key.objectid == bytenr &&
6921 key.type == BTRFS_EXTENT_ITEM_KEY &&
6922 key.offset == num_bytes)
6923 ret = 0;
6924 }
6925
6926 if (ret > 0 && skinny_metadata) {
6927 skinny_metadata = false;
6928 key.objectid = bytenr;
6929 key.type = BTRFS_EXTENT_ITEM_KEY;
6930 key.offset = num_bytes;
6931 btrfs_release_path(path);
6932 ret = btrfs_search_slot(trans, extent_root,
6933 &key, path, -1, 1);
6934 }
6935
6936 if (ret) {
6937 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6938 ret, bytenr);
6939 if (ret > 0)
6940 btrfs_print_leaf(extent_root,
6941 path->nodes[0]);
6942 }
6943 if (ret < 0) {
6944 btrfs_abort_transaction(trans, extent_root, ret);
6945 goto out;
6946 }
6947 extent_slot = path->slots[0];
6948 }
6949 } else if (WARN_ON(ret == -ENOENT)) {
6950 btrfs_print_leaf(extent_root, path->nodes[0]);
6951 btrfs_err(info,
6952 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6953 bytenr, parent, root_objectid, owner_objectid,
6954 owner_offset);
6955 btrfs_abort_transaction(trans, extent_root, ret);
6956 goto out;
6957 } else {
6958 btrfs_abort_transaction(trans, extent_root, ret);
6959 goto out;
6960 }
6961
6962 leaf = path->nodes[0];
6963 item_size = btrfs_item_size_nr(leaf, extent_slot);
6964 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6965 if (item_size < sizeof(*ei)) {
6966 BUG_ON(found_extent || extent_slot != path->slots[0]);
6967 ret = convert_extent_item_v0(trans, extent_root, path,
6968 owner_objectid, 0);
6969 if (ret < 0) {
6970 btrfs_abort_transaction(trans, extent_root, ret);
6971 goto out;
6972 }
6973
6974 btrfs_release_path(path);
6975 path->leave_spinning = 1;
6976
6977 key.objectid = bytenr;
6978 key.type = BTRFS_EXTENT_ITEM_KEY;
6979 key.offset = num_bytes;
6980
6981 ret = btrfs_search_slot(trans, extent_root, &key, path,
6982 -1, 1);
6983 if (ret) {
6984 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6985 ret, bytenr);
6986 btrfs_print_leaf(extent_root, path->nodes[0]);
6987 }
6988 if (ret < 0) {
6989 btrfs_abort_transaction(trans, extent_root, ret);
6990 goto out;
6991 }
6992
6993 extent_slot = path->slots[0];
6994 leaf = path->nodes[0];
6995 item_size = btrfs_item_size_nr(leaf, extent_slot);
6996 }
6997 #endif
6998 BUG_ON(item_size < sizeof(*ei));
6999 ei = btrfs_item_ptr(leaf, extent_slot,
7000 struct btrfs_extent_item);
7001 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
7002 key.type == BTRFS_EXTENT_ITEM_KEY) {
7003 struct btrfs_tree_block_info *bi;
7004 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
7005 bi = (struct btrfs_tree_block_info *)(ei + 1);
7006 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
7007 }
7008
7009 refs = btrfs_extent_refs(leaf, ei);
7010 if (refs < refs_to_drop) {
7011 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
7012 "for bytenr %Lu", refs_to_drop, refs, bytenr);
7013 ret = -EINVAL;
7014 btrfs_abort_transaction(trans, extent_root, ret);
7015 goto out;
7016 }
7017 refs -= refs_to_drop;
7018
7019 if (refs > 0) {
7020 if (extent_op)
7021 __run_delayed_extent_op(extent_op, leaf, ei);
7022 /*
7023 * In the case of inline back ref, reference count will
7024 * be updated by remove_extent_backref
7025 */
7026 if (iref) {
7027 BUG_ON(!found_extent);
7028 } else {
7029 btrfs_set_extent_refs(leaf, ei, refs);
7030 btrfs_mark_buffer_dirty(leaf);
7031 }
7032 if (found_extent) {
7033 ret = remove_extent_backref(trans, extent_root, path,
7034 iref, refs_to_drop,
7035 is_data, &last_ref);
7036 if (ret) {
7037 btrfs_abort_transaction(trans, extent_root, ret);
7038 goto out;
7039 }
7040 }
7041 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
7042 root_objectid);
7043 } else {
7044 if (found_extent) {
7045 BUG_ON(is_data && refs_to_drop !=
7046 extent_data_ref_count(path, iref));
7047 if (iref) {
7048 BUG_ON(path->slots[0] != extent_slot);
7049 } else {
7050 BUG_ON(path->slots[0] != extent_slot + 1);
7051 path->slots[0] = extent_slot;
7052 num_to_del = 2;
7053 }
7054 }
7055
7056 last_ref = 1;
7057 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
7058 num_to_del);
7059 if (ret) {
7060 btrfs_abort_transaction(trans, extent_root, ret);
7061 goto out;
7062 }
7063 btrfs_release_path(path);
7064
7065 if (is_data) {
7066 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
7067 if (ret) {
7068 btrfs_abort_transaction(trans, extent_root, ret);
7069 goto out;
7070 }
7071 }
7072
7073 ret = add_to_free_space_tree(trans, root->fs_info, bytenr,
7074 num_bytes);
7075 if (ret) {
7076 btrfs_abort_transaction(trans, extent_root, ret);
7077 goto out;
7078 }
7079
7080 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
7081 if (ret) {
7082 btrfs_abort_transaction(trans, extent_root, ret);
7083 goto out;
7084 }
7085 }
7086 btrfs_release_path(path);
7087
7088 out:
7089 btrfs_free_path(path);
7090 return ret;
7091 }
7092
7093 /*
7094 * when we free an block, it is possible (and likely) that we free the last
7095 * delayed ref for that extent as well. This searches the delayed ref tree for
7096 * a given extent, and if there are no other delayed refs to be processed, it
7097 * removes it from the tree.
7098 */
7099 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
7100 struct btrfs_root *root, u64 bytenr)
7101 {
7102 struct btrfs_delayed_ref_head *head;
7103 struct btrfs_delayed_ref_root *delayed_refs;
7104 int ret = 0;
7105
7106 delayed_refs = &trans->transaction->delayed_refs;
7107 spin_lock(&delayed_refs->lock);
7108 head = btrfs_find_delayed_ref_head(trans, bytenr);
7109 if (!head)
7110 goto out_delayed_unlock;
7111
7112 spin_lock(&head->lock);
7113 if (!list_empty(&head->ref_list))
7114 goto out;
7115
7116 if (head->extent_op) {
7117 if (!head->must_insert_reserved)
7118 goto out;
7119 btrfs_free_delayed_extent_op(head->extent_op);
7120 head->extent_op = NULL;
7121 }
7122
7123 /*
7124 * waiting for the lock here would deadlock. If someone else has it
7125 * locked they are already in the process of dropping it anyway
7126 */
7127 if (!mutex_trylock(&head->mutex))
7128 goto out;
7129
7130 /*
7131 * at this point we have a head with no other entries. Go
7132 * ahead and process it.
7133 */
7134 head->node.in_tree = 0;
7135 rb_erase(&head->href_node, &delayed_refs->href_root);
7136
7137 atomic_dec(&delayed_refs->num_entries);
7138
7139 /*
7140 * we don't take a ref on the node because we're removing it from the
7141 * tree, so we just steal the ref the tree was holding.
7142 */
7143 delayed_refs->num_heads--;
7144 if (head->processing == 0)
7145 delayed_refs->num_heads_ready--;
7146 head->processing = 0;
7147 spin_unlock(&head->lock);
7148 spin_unlock(&delayed_refs->lock);
7149
7150 BUG_ON(head->extent_op);
7151 if (head->must_insert_reserved)
7152 ret = 1;
7153
7154 mutex_unlock(&head->mutex);
7155 btrfs_put_delayed_ref(&head->node);
7156 return ret;
7157 out:
7158 spin_unlock(&head->lock);
7159
7160 out_delayed_unlock:
7161 spin_unlock(&delayed_refs->lock);
7162 return 0;
7163 }
7164
7165 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
7166 struct btrfs_root *root,
7167 struct extent_buffer *buf,
7168 u64 parent, int last_ref)
7169 {
7170 int pin = 1;
7171 int ret;
7172
7173 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7174 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7175 buf->start, buf->len,
7176 parent, root->root_key.objectid,
7177 btrfs_header_level(buf),
7178 BTRFS_DROP_DELAYED_REF, NULL);
7179 BUG_ON(ret); /* -ENOMEM */
7180 }
7181
7182 if (!last_ref)
7183 return;
7184
7185 if (btrfs_header_generation(buf) == trans->transid) {
7186 struct btrfs_block_group_cache *cache;
7187
7188 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7189 ret = check_ref_cleanup(trans, root, buf->start);
7190 if (!ret)
7191 goto out;
7192 }
7193
7194 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
7195
7196 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
7197 pin_down_extent(root, cache, buf->start, buf->len, 1);
7198 btrfs_put_block_group(cache);
7199 goto out;
7200 }
7201
7202 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
7203
7204 btrfs_add_free_space(cache, buf->start, buf->len);
7205 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
7206 btrfs_put_block_group(cache);
7207 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
7208 pin = 0;
7209 }
7210 out:
7211 if (pin)
7212 add_pinned_bytes(root->fs_info, buf->len,
7213 btrfs_header_level(buf),
7214 root->root_key.objectid);
7215
7216 /*
7217 * Deleting the buffer, clear the corrupt flag since it doesn't matter
7218 * anymore.
7219 */
7220 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
7221 }
7222
7223 /* Can return -ENOMEM */
7224 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7225 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
7226 u64 owner, u64 offset)
7227 {
7228 int ret;
7229 struct btrfs_fs_info *fs_info = root->fs_info;
7230
7231 if (btrfs_test_is_dummy_root(root))
7232 return 0;
7233
7234 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
7235
7236 /*
7237 * tree log blocks never actually go into the extent allocation
7238 * tree, just update pinning info and exit early.
7239 */
7240 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
7241 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
7242 /* unlocks the pinned mutex */
7243 btrfs_pin_extent(root, bytenr, num_bytes, 1);
7244 ret = 0;
7245 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
7246 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
7247 num_bytes,
7248 parent, root_objectid, (int)owner,
7249 BTRFS_DROP_DELAYED_REF, NULL);
7250 } else {
7251 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
7252 num_bytes,
7253 parent, root_objectid, owner,
7254 offset, 0,
7255 BTRFS_DROP_DELAYED_REF, NULL);
7256 }
7257 return ret;
7258 }
7259
7260 /*
7261 * when we wait for progress in the block group caching, its because
7262 * our allocation attempt failed at least once. So, we must sleep
7263 * and let some progress happen before we try again.
7264 *
7265 * This function will sleep at least once waiting for new free space to
7266 * show up, and then it will check the block group free space numbers
7267 * for our min num_bytes. Another option is to have it go ahead
7268 * and look in the rbtree for a free extent of a given size, but this
7269 * is a good start.
7270 *
7271 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7272 * any of the information in this block group.
7273 */
7274 static noinline void
7275 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
7276 u64 num_bytes)
7277 {
7278 struct btrfs_caching_control *caching_ctl;
7279
7280 caching_ctl = get_caching_control(cache);
7281 if (!caching_ctl)
7282 return;
7283
7284 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
7285 (cache->free_space_ctl->free_space >= num_bytes));
7286
7287 put_caching_control(caching_ctl);
7288 }
7289
7290 static noinline int
7291 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
7292 {
7293 struct btrfs_caching_control *caching_ctl;
7294 int ret = 0;
7295
7296 caching_ctl = get_caching_control(cache);
7297 if (!caching_ctl)
7298 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
7299
7300 wait_event(caching_ctl->wait, block_group_cache_done(cache));
7301 if (cache->cached == BTRFS_CACHE_ERROR)
7302 ret = -EIO;
7303 put_caching_control(caching_ctl);
7304 return ret;
7305 }
7306
7307 int __get_raid_index(u64 flags)
7308 {
7309 if (flags & BTRFS_BLOCK_GROUP_RAID10)
7310 return BTRFS_RAID_RAID10;
7311 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
7312 return BTRFS_RAID_RAID1;
7313 else if (flags & BTRFS_BLOCK_GROUP_DUP)
7314 return BTRFS_RAID_DUP;
7315 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
7316 return BTRFS_RAID_RAID0;
7317 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
7318 return BTRFS_RAID_RAID5;
7319 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
7320 return BTRFS_RAID_RAID6;
7321
7322 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
7323 }
7324
7325 int get_block_group_index(struct btrfs_block_group_cache *cache)
7326 {
7327 return __get_raid_index(cache->flags);
7328 }
7329
7330 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
7331 [BTRFS_RAID_RAID10] = "raid10",
7332 [BTRFS_RAID_RAID1] = "raid1",
7333 [BTRFS_RAID_DUP] = "dup",
7334 [BTRFS_RAID_RAID0] = "raid0",
7335 [BTRFS_RAID_SINGLE] = "single",
7336 [BTRFS_RAID_RAID5] = "raid5",
7337 [BTRFS_RAID_RAID6] = "raid6",
7338 };
7339
7340 static const char *get_raid_name(enum btrfs_raid_types type)
7341 {
7342 if (type >= BTRFS_NR_RAID_TYPES)
7343 return NULL;
7344
7345 return btrfs_raid_type_names[type];
7346 }
7347
7348 enum btrfs_loop_type {
7349 LOOP_CACHING_NOWAIT = 0,
7350 LOOP_CACHING_WAIT = 1,
7351 LOOP_ALLOC_CHUNK = 2,
7352 LOOP_NO_EMPTY_SIZE = 3,
7353 };
7354
7355 static inline void
7356 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7357 int delalloc)
7358 {
7359 if (delalloc)
7360 down_read(&cache->data_rwsem);
7361 }
7362
7363 static inline void
7364 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7365 int delalloc)
7366 {
7367 btrfs_get_block_group(cache);
7368 if (delalloc)
7369 down_read(&cache->data_rwsem);
7370 }
7371
7372 static struct btrfs_block_group_cache *
7373 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7374 struct btrfs_free_cluster *cluster,
7375 int delalloc)
7376 {
7377 struct btrfs_block_group_cache *used_bg = NULL;
7378
7379 spin_lock(&cluster->refill_lock);
7380 while (1) {
7381 used_bg = cluster->block_group;
7382 if (!used_bg)
7383 return NULL;
7384
7385 if (used_bg == block_group)
7386 return used_bg;
7387
7388 btrfs_get_block_group(used_bg);
7389
7390 if (!delalloc)
7391 return used_bg;
7392
7393 if (down_read_trylock(&used_bg->data_rwsem))
7394 return used_bg;
7395
7396 spin_unlock(&cluster->refill_lock);
7397
7398 down_read(&used_bg->data_rwsem);
7399
7400 spin_lock(&cluster->refill_lock);
7401 if (used_bg == cluster->block_group)
7402 return used_bg;
7403
7404 up_read(&used_bg->data_rwsem);
7405 btrfs_put_block_group(used_bg);
7406 }
7407 }
7408
7409 static inline void
7410 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7411 int delalloc)
7412 {
7413 if (delalloc)
7414 up_read(&cache->data_rwsem);
7415 btrfs_put_block_group(cache);
7416 }
7417
7418 /*
7419 * walks the btree of allocated extents and find a hole of a given size.
7420 * The key ins is changed to record the hole:
7421 * ins->objectid == start position
7422 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7423 * ins->offset == the size of the hole.
7424 * Any available blocks before search_start are skipped.
7425 *
7426 * If there is no suitable free space, we will record the max size of
7427 * the free space extent currently.
7428 */
7429 static noinline int find_free_extent(struct btrfs_root *orig_root,
7430 u64 num_bytes, u64 empty_size,
7431 u64 hint_byte, struct btrfs_key *ins,
7432 u64 flags, int delalloc)
7433 {
7434 int ret = 0;
7435 struct btrfs_root *root = orig_root->fs_info->extent_root;
7436 struct btrfs_free_cluster *last_ptr = NULL;
7437 struct btrfs_block_group_cache *block_group = NULL;
7438 u64 search_start = 0;
7439 u64 max_extent_size = 0;
7440 u64 empty_cluster = 0;
7441 struct btrfs_space_info *space_info;
7442 int loop = 0;
7443 int index = __get_raid_index(flags);
7444 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
7445 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
7446 bool failed_cluster_refill = false;
7447 bool failed_alloc = false;
7448 bool use_cluster = true;
7449 bool have_caching_bg = false;
7450 bool orig_have_caching_bg = false;
7451 bool full_search = false;
7452
7453 WARN_ON(num_bytes < root->sectorsize);
7454 ins->type = BTRFS_EXTENT_ITEM_KEY;
7455 ins->objectid = 0;
7456 ins->offset = 0;
7457
7458 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
7459
7460 space_info = __find_space_info(root->fs_info, flags);
7461 if (!space_info) {
7462 btrfs_err(root->fs_info, "No space info for %llu", flags);
7463 return -ENOSPC;
7464 }
7465
7466 /*
7467 * If our free space is heavily fragmented we may not be able to make
7468 * big contiguous allocations, so instead of doing the expensive search
7469 * for free space, simply return ENOSPC with our max_extent_size so we
7470 * can go ahead and search for a more manageable chunk.
7471 *
7472 * If our max_extent_size is large enough for our allocation simply
7473 * disable clustering since we will likely not be able to find enough
7474 * space to create a cluster and induce latency trying.
7475 */
7476 if (unlikely(space_info->max_extent_size)) {
7477 spin_lock(&space_info->lock);
7478 if (space_info->max_extent_size &&
7479 num_bytes > space_info->max_extent_size) {
7480 ins->offset = space_info->max_extent_size;
7481 spin_unlock(&space_info->lock);
7482 return -ENOSPC;
7483 } else if (space_info->max_extent_size) {
7484 use_cluster = false;
7485 }
7486 spin_unlock(&space_info->lock);
7487 }
7488
7489 last_ptr = fetch_cluster_info(orig_root, space_info, &empty_cluster);
7490 if (last_ptr) {
7491 spin_lock(&last_ptr->lock);
7492 if (last_ptr->block_group)
7493 hint_byte = last_ptr->window_start;
7494 if (last_ptr->fragmented) {
7495 /*
7496 * We still set window_start so we can keep track of the
7497 * last place we found an allocation to try and save
7498 * some time.
7499 */
7500 hint_byte = last_ptr->window_start;
7501 use_cluster = false;
7502 }
7503 spin_unlock(&last_ptr->lock);
7504 }
7505
7506 search_start = max(search_start, first_logical_byte(root, 0));
7507 search_start = max(search_start, hint_byte);
7508 if (search_start == hint_byte) {
7509 block_group = btrfs_lookup_block_group(root->fs_info,
7510 search_start);
7511 /*
7512 * we don't want to use the block group if it doesn't match our
7513 * allocation bits, or if its not cached.
7514 *
7515 * However if we are re-searching with an ideal block group
7516 * picked out then we don't care that the block group is cached.
7517 */
7518 if (block_group && block_group_bits(block_group, flags) &&
7519 block_group->cached != BTRFS_CACHE_NO) {
7520 down_read(&space_info->groups_sem);
7521 if (list_empty(&block_group->list) ||
7522 block_group->ro) {
7523 /*
7524 * someone is removing this block group,
7525 * we can't jump into the have_block_group
7526 * target because our list pointers are not
7527 * valid
7528 */
7529 btrfs_put_block_group(block_group);
7530 up_read(&space_info->groups_sem);
7531 } else {
7532 index = get_block_group_index(block_group);
7533 btrfs_lock_block_group(block_group, delalloc);
7534 goto have_block_group;
7535 }
7536 } else if (block_group) {
7537 btrfs_put_block_group(block_group);
7538 }
7539 }
7540 search:
7541 have_caching_bg = false;
7542 if (index == 0 || index == __get_raid_index(flags))
7543 full_search = true;
7544 down_read(&space_info->groups_sem);
7545 list_for_each_entry(block_group, &space_info->block_groups[index],
7546 list) {
7547 u64 offset;
7548 int cached;
7549
7550 btrfs_grab_block_group(block_group, delalloc);
7551 search_start = block_group->key.objectid;
7552
7553 /*
7554 * this can happen if we end up cycling through all the
7555 * raid types, but we want to make sure we only allocate
7556 * for the proper type.
7557 */
7558 if (!block_group_bits(block_group, flags)) {
7559 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7560 BTRFS_BLOCK_GROUP_RAID1 |
7561 BTRFS_BLOCK_GROUP_RAID5 |
7562 BTRFS_BLOCK_GROUP_RAID6 |
7563 BTRFS_BLOCK_GROUP_RAID10;
7564
7565 /*
7566 * if they asked for extra copies and this block group
7567 * doesn't provide them, bail. This does allow us to
7568 * fill raid0 from raid1.
7569 */
7570 if ((flags & extra) && !(block_group->flags & extra))
7571 goto loop;
7572 }
7573
7574 have_block_group:
7575 cached = block_group_cache_done(block_group);
7576 if (unlikely(!cached)) {
7577 have_caching_bg = true;
7578 ret = cache_block_group(block_group, 0);
7579 BUG_ON(ret < 0);
7580 ret = 0;
7581 }
7582
7583 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7584 goto loop;
7585 if (unlikely(block_group->ro))
7586 goto loop;
7587
7588 /*
7589 * Ok we want to try and use the cluster allocator, so
7590 * lets look there
7591 */
7592 if (last_ptr && use_cluster) {
7593 struct btrfs_block_group_cache *used_block_group;
7594 unsigned long aligned_cluster;
7595 /*
7596 * the refill lock keeps out other
7597 * people trying to start a new cluster
7598 */
7599 used_block_group = btrfs_lock_cluster(block_group,
7600 last_ptr,
7601 delalloc);
7602 if (!used_block_group)
7603 goto refill_cluster;
7604
7605 if (used_block_group != block_group &&
7606 (used_block_group->ro ||
7607 !block_group_bits(used_block_group, flags)))
7608 goto release_cluster;
7609
7610 offset = btrfs_alloc_from_cluster(used_block_group,
7611 last_ptr,
7612 num_bytes,
7613 used_block_group->key.objectid,
7614 &max_extent_size);
7615 if (offset) {
7616 /* we have a block, we're done */
7617 spin_unlock(&last_ptr->refill_lock);
7618 trace_btrfs_reserve_extent_cluster(root,
7619 used_block_group,
7620 search_start, num_bytes);
7621 if (used_block_group != block_group) {
7622 btrfs_release_block_group(block_group,
7623 delalloc);
7624 block_group = used_block_group;
7625 }
7626 goto checks;
7627 }
7628
7629 WARN_ON(last_ptr->block_group != used_block_group);
7630 release_cluster:
7631 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7632 * set up a new clusters, so lets just skip it
7633 * and let the allocator find whatever block
7634 * it can find. If we reach this point, we
7635 * will have tried the cluster allocator
7636 * plenty of times and not have found
7637 * anything, so we are likely way too
7638 * fragmented for the clustering stuff to find
7639 * anything.
7640 *
7641 * However, if the cluster is taken from the
7642 * current block group, release the cluster
7643 * first, so that we stand a better chance of
7644 * succeeding in the unclustered
7645 * allocation. */
7646 if (loop >= LOOP_NO_EMPTY_SIZE &&
7647 used_block_group != block_group) {
7648 spin_unlock(&last_ptr->refill_lock);
7649 btrfs_release_block_group(used_block_group,
7650 delalloc);
7651 goto unclustered_alloc;
7652 }
7653
7654 /*
7655 * this cluster didn't work out, free it and
7656 * start over
7657 */
7658 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7659
7660 if (used_block_group != block_group)
7661 btrfs_release_block_group(used_block_group,
7662 delalloc);
7663 refill_cluster:
7664 if (loop >= LOOP_NO_EMPTY_SIZE) {
7665 spin_unlock(&last_ptr->refill_lock);
7666 goto unclustered_alloc;
7667 }
7668
7669 aligned_cluster = max_t(unsigned long,
7670 empty_cluster + empty_size,
7671 block_group->full_stripe_len);
7672
7673 /* allocate a cluster in this block group */
7674 ret = btrfs_find_space_cluster(root, block_group,
7675 last_ptr, search_start,
7676 num_bytes,
7677 aligned_cluster);
7678 if (ret == 0) {
7679 /*
7680 * now pull our allocation out of this
7681 * cluster
7682 */
7683 offset = btrfs_alloc_from_cluster(block_group,
7684 last_ptr,
7685 num_bytes,
7686 search_start,
7687 &max_extent_size);
7688 if (offset) {
7689 /* we found one, proceed */
7690 spin_unlock(&last_ptr->refill_lock);
7691 trace_btrfs_reserve_extent_cluster(root,
7692 block_group, search_start,
7693 num_bytes);
7694 goto checks;
7695 }
7696 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7697 && !failed_cluster_refill) {
7698 spin_unlock(&last_ptr->refill_lock);
7699
7700 failed_cluster_refill = true;
7701 wait_block_group_cache_progress(block_group,
7702 num_bytes + empty_cluster + empty_size);
7703 goto have_block_group;
7704 }
7705
7706 /*
7707 * at this point we either didn't find a cluster
7708 * or we weren't able to allocate a block from our
7709 * cluster. Free the cluster we've been trying
7710 * to use, and go to the next block group
7711 */
7712 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7713 spin_unlock(&last_ptr->refill_lock);
7714 goto loop;
7715 }
7716
7717 unclustered_alloc:
7718 /*
7719 * We are doing an unclustered alloc, set the fragmented flag so
7720 * we don't bother trying to setup a cluster again until we get
7721 * more space.
7722 */
7723 if (unlikely(last_ptr)) {
7724 spin_lock(&last_ptr->lock);
7725 last_ptr->fragmented = 1;
7726 spin_unlock(&last_ptr->lock);
7727 }
7728 spin_lock(&block_group->free_space_ctl->tree_lock);
7729 if (cached &&
7730 block_group->free_space_ctl->free_space <
7731 num_bytes + empty_cluster + empty_size) {
7732 if (block_group->free_space_ctl->free_space >
7733 max_extent_size)
7734 max_extent_size =
7735 block_group->free_space_ctl->free_space;
7736 spin_unlock(&block_group->free_space_ctl->tree_lock);
7737 goto loop;
7738 }
7739 spin_unlock(&block_group->free_space_ctl->tree_lock);
7740
7741 offset = btrfs_find_space_for_alloc(block_group, search_start,
7742 num_bytes, empty_size,
7743 &max_extent_size);
7744 /*
7745 * If we didn't find a chunk, and we haven't failed on this
7746 * block group before, and this block group is in the middle of
7747 * caching and we are ok with waiting, then go ahead and wait
7748 * for progress to be made, and set failed_alloc to true.
7749 *
7750 * If failed_alloc is true then we've already waited on this
7751 * block group once and should move on to the next block group.
7752 */
7753 if (!offset && !failed_alloc && !cached &&
7754 loop > LOOP_CACHING_NOWAIT) {
7755 wait_block_group_cache_progress(block_group,
7756 num_bytes + empty_size);
7757 failed_alloc = true;
7758 goto have_block_group;
7759 } else if (!offset) {
7760 goto loop;
7761 }
7762 checks:
7763 search_start = ALIGN(offset, root->stripesize);
7764
7765 /* move on to the next group */
7766 if (search_start + num_bytes >
7767 block_group->key.objectid + block_group->key.offset) {
7768 btrfs_add_free_space(block_group, offset, num_bytes);
7769 goto loop;
7770 }
7771
7772 if (offset < search_start)
7773 btrfs_add_free_space(block_group, offset,
7774 search_start - offset);
7775 BUG_ON(offset > search_start);
7776
7777 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
7778 alloc_type, delalloc);
7779 if (ret == -EAGAIN) {
7780 btrfs_add_free_space(block_group, offset, num_bytes);
7781 goto loop;
7782 }
7783 btrfs_inc_block_group_reservations(block_group);
7784
7785 /* we are all good, lets return */
7786 ins->objectid = search_start;
7787 ins->offset = num_bytes;
7788
7789 trace_btrfs_reserve_extent(orig_root, block_group,
7790 search_start, num_bytes);
7791 btrfs_release_block_group(block_group, delalloc);
7792 break;
7793 loop:
7794 failed_cluster_refill = false;
7795 failed_alloc = false;
7796 BUG_ON(index != get_block_group_index(block_group));
7797 btrfs_release_block_group(block_group, delalloc);
7798 }
7799 up_read(&space_info->groups_sem);
7800
7801 if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg
7802 && !orig_have_caching_bg)
7803 orig_have_caching_bg = true;
7804
7805 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7806 goto search;
7807
7808 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7809 goto search;
7810
7811 /*
7812 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7813 * caching kthreads as we move along
7814 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7815 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7816 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7817 * again
7818 */
7819 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7820 index = 0;
7821 if (loop == LOOP_CACHING_NOWAIT) {
7822 /*
7823 * We want to skip the LOOP_CACHING_WAIT step if we
7824 * don't have any uncached bgs and we've already done a
7825 * full search through.
7826 */
7827 if (orig_have_caching_bg || !full_search)
7828 loop = LOOP_CACHING_WAIT;
7829 else
7830 loop = LOOP_ALLOC_CHUNK;
7831 } else {
7832 loop++;
7833 }
7834
7835 if (loop == LOOP_ALLOC_CHUNK) {
7836 struct btrfs_trans_handle *trans;
7837 int exist = 0;
7838
7839 trans = current->journal_info;
7840 if (trans)
7841 exist = 1;
7842 else
7843 trans = btrfs_join_transaction(root);
7844
7845 if (IS_ERR(trans)) {
7846 ret = PTR_ERR(trans);
7847 goto out;
7848 }
7849
7850 ret = do_chunk_alloc(trans, root, flags,
7851 CHUNK_ALLOC_FORCE);
7852
7853 /*
7854 * If we can't allocate a new chunk we've already looped
7855 * through at least once, move on to the NO_EMPTY_SIZE
7856 * case.
7857 */
7858 if (ret == -ENOSPC)
7859 loop = LOOP_NO_EMPTY_SIZE;
7860
7861 /*
7862 * Do not bail out on ENOSPC since we
7863 * can do more things.
7864 */
7865 if (ret < 0 && ret != -ENOSPC)
7866 btrfs_abort_transaction(trans,
7867 root, ret);
7868 else
7869 ret = 0;
7870 if (!exist)
7871 btrfs_end_transaction(trans, root);
7872 if (ret)
7873 goto out;
7874 }
7875
7876 if (loop == LOOP_NO_EMPTY_SIZE) {
7877 /*
7878 * Don't loop again if we already have no empty_size and
7879 * no empty_cluster.
7880 */
7881 if (empty_size == 0 &&
7882 empty_cluster == 0) {
7883 ret = -ENOSPC;
7884 goto out;
7885 }
7886 empty_size = 0;
7887 empty_cluster = 0;
7888 }
7889
7890 goto search;
7891 } else if (!ins->objectid) {
7892 ret = -ENOSPC;
7893 } else if (ins->objectid) {
7894 if (!use_cluster && last_ptr) {
7895 spin_lock(&last_ptr->lock);
7896 last_ptr->window_start = ins->objectid;
7897 spin_unlock(&last_ptr->lock);
7898 }
7899 ret = 0;
7900 }
7901 out:
7902 if (ret == -ENOSPC) {
7903 spin_lock(&space_info->lock);
7904 space_info->max_extent_size = max_extent_size;
7905 spin_unlock(&space_info->lock);
7906 ins->offset = max_extent_size;
7907 }
7908 return ret;
7909 }
7910
7911 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
7912 int dump_block_groups)
7913 {
7914 struct btrfs_block_group_cache *cache;
7915 int index = 0;
7916
7917 spin_lock(&info->lock);
7918 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
7919 info->flags,
7920 info->total_bytes - info->bytes_used - info->bytes_pinned -
7921 info->bytes_reserved - info->bytes_readonly,
7922 (info->full) ? "" : "not ");
7923 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7924 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7925 info->total_bytes, info->bytes_used, info->bytes_pinned,
7926 info->bytes_reserved, info->bytes_may_use,
7927 info->bytes_readonly);
7928 spin_unlock(&info->lock);
7929
7930 if (!dump_block_groups)
7931 return;
7932
7933 down_read(&info->groups_sem);
7934 again:
7935 list_for_each_entry(cache, &info->block_groups[index], list) {
7936 spin_lock(&cache->lock);
7937 printk(KERN_INFO "BTRFS: "
7938 "block group %llu has %llu bytes, "
7939 "%llu used %llu pinned %llu reserved %s\n",
7940 cache->key.objectid, cache->key.offset,
7941 btrfs_block_group_used(&cache->item), cache->pinned,
7942 cache->reserved, cache->ro ? "[readonly]" : "");
7943 btrfs_dump_free_space(cache, bytes);
7944 spin_unlock(&cache->lock);
7945 }
7946 if (++index < BTRFS_NR_RAID_TYPES)
7947 goto again;
7948 up_read(&info->groups_sem);
7949 }
7950
7951 int btrfs_reserve_extent(struct btrfs_root *root,
7952 u64 num_bytes, u64 min_alloc_size,
7953 u64 empty_size, u64 hint_byte,
7954 struct btrfs_key *ins, int is_data, int delalloc)
7955 {
7956 bool final_tried = num_bytes == min_alloc_size;
7957 u64 flags;
7958 int ret;
7959
7960 flags = btrfs_get_alloc_profile(root, is_data);
7961 again:
7962 WARN_ON(num_bytes < root->sectorsize);
7963 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
7964 flags, delalloc);
7965 if (!ret && !is_data) {
7966 btrfs_dec_block_group_reservations(root->fs_info,
7967 ins->objectid);
7968 } else if (ret == -ENOSPC) {
7969 if (!final_tried && ins->offset) {
7970 num_bytes = min(num_bytes >> 1, ins->offset);
7971 num_bytes = round_down(num_bytes, root->sectorsize);
7972 num_bytes = max(num_bytes, min_alloc_size);
7973 if (num_bytes == min_alloc_size)
7974 final_tried = true;
7975 goto again;
7976 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7977 struct btrfs_space_info *sinfo;
7978
7979 sinfo = __find_space_info(root->fs_info, flags);
7980 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
7981 flags, num_bytes);
7982 if (sinfo)
7983 dump_space_info(sinfo, num_bytes, 1);
7984 }
7985 }
7986
7987 return ret;
7988 }
7989
7990 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
7991 u64 start, u64 len,
7992 int pin, int delalloc)
7993 {
7994 struct btrfs_block_group_cache *cache;
7995 int ret = 0;
7996
7997 cache = btrfs_lookup_block_group(root->fs_info, start);
7998 if (!cache) {
7999 btrfs_err(root->fs_info, "Unable to find block group for %llu",
8000 start);
8001 return -ENOSPC;
8002 }
8003
8004 if (pin)
8005 pin_down_extent(root, cache, start, len, 1);
8006 else {
8007 if (btrfs_test_opt(root, DISCARD))
8008 ret = btrfs_discard_extent(root, start, len, NULL);
8009 btrfs_add_free_space(cache, start, len);
8010 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
8011 }
8012
8013 btrfs_put_block_group(cache);
8014
8015 trace_btrfs_reserved_extent_free(root, start, len);
8016
8017 return ret;
8018 }
8019
8020 int btrfs_free_reserved_extent(struct btrfs_root *root,
8021 u64 start, u64 len, int delalloc)
8022 {
8023 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
8024 }
8025
8026 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
8027 u64 start, u64 len)
8028 {
8029 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
8030 }
8031
8032 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8033 struct btrfs_root *root,
8034 u64 parent, u64 root_objectid,
8035 u64 flags, u64 owner, u64 offset,
8036 struct btrfs_key *ins, int ref_mod)
8037 {
8038 int ret;
8039 struct btrfs_fs_info *fs_info = root->fs_info;
8040 struct btrfs_extent_item *extent_item;
8041 struct btrfs_extent_inline_ref *iref;
8042 struct btrfs_path *path;
8043 struct extent_buffer *leaf;
8044 int type;
8045 u32 size;
8046
8047 if (parent > 0)
8048 type = BTRFS_SHARED_DATA_REF_KEY;
8049 else
8050 type = BTRFS_EXTENT_DATA_REF_KEY;
8051
8052 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
8053
8054 path = btrfs_alloc_path();
8055 if (!path)
8056 return -ENOMEM;
8057
8058 path->leave_spinning = 1;
8059 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8060 ins, size);
8061 if (ret) {
8062 btrfs_free_path(path);
8063 return ret;
8064 }
8065
8066 leaf = path->nodes[0];
8067 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8068 struct btrfs_extent_item);
8069 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
8070 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8071 btrfs_set_extent_flags(leaf, extent_item,
8072 flags | BTRFS_EXTENT_FLAG_DATA);
8073
8074 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8075 btrfs_set_extent_inline_ref_type(leaf, iref, type);
8076 if (parent > 0) {
8077 struct btrfs_shared_data_ref *ref;
8078 ref = (struct btrfs_shared_data_ref *)(iref + 1);
8079 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8080 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
8081 } else {
8082 struct btrfs_extent_data_ref *ref;
8083 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
8084 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
8085 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
8086 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
8087 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
8088 }
8089
8090 btrfs_mark_buffer_dirty(path->nodes[0]);
8091 btrfs_free_path(path);
8092
8093 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8094 ins->offset);
8095 if (ret)
8096 return ret;
8097
8098 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
8099 if (ret) { /* -ENOENT, logic error */
8100 btrfs_err(fs_info, "update block group failed for %llu %llu",
8101 ins->objectid, ins->offset);
8102 BUG();
8103 }
8104 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
8105 return ret;
8106 }
8107
8108 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
8109 struct btrfs_root *root,
8110 u64 parent, u64 root_objectid,
8111 u64 flags, struct btrfs_disk_key *key,
8112 int level, struct btrfs_key *ins)
8113 {
8114 int ret;
8115 struct btrfs_fs_info *fs_info = root->fs_info;
8116 struct btrfs_extent_item *extent_item;
8117 struct btrfs_tree_block_info *block_info;
8118 struct btrfs_extent_inline_ref *iref;
8119 struct btrfs_path *path;
8120 struct extent_buffer *leaf;
8121 u32 size = sizeof(*extent_item) + sizeof(*iref);
8122 u64 num_bytes = ins->offset;
8123 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
8124 SKINNY_METADATA);
8125
8126 if (!skinny_metadata)
8127 size += sizeof(*block_info);
8128
8129 path = btrfs_alloc_path();
8130 if (!path) {
8131 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
8132 root->nodesize);
8133 return -ENOMEM;
8134 }
8135
8136 path->leave_spinning = 1;
8137 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8138 ins, size);
8139 if (ret) {
8140 btrfs_free_path(path);
8141 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
8142 root->nodesize);
8143 return ret;
8144 }
8145
8146 leaf = path->nodes[0];
8147 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8148 struct btrfs_extent_item);
8149 btrfs_set_extent_refs(leaf, extent_item, 1);
8150 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8151 btrfs_set_extent_flags(leaf, extent_item,
8152 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
8153
8154 if (skinny_metadata) {
8155 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8156 num_bytes = root->nodesize;
8157 } else {
8158 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
8159 btrfs_set_tree_block_key(leaf, block_info, key);
8160 btrfs_set_tree_block_level(leaf, block_info, level);
8161 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
8162 }
8163
8164 if (parent > 0) {
8165 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
8166 btrfs_set_extent_inline_ref_type(leaf, iref,
8167 BTRFS_SHARED_BLOCK_REF_KEY);
8168 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8169 } else {
8170 btrfs_set_extent_inline_ref_type(leaf, iref,
8171 BTRFS_TREE_BLOCK_REF_KEY);
8172 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
8173 }
8174
8175 btrfs_mark_buffer_dirty(leaf);
8176 btrfs_free_path(path);
8177
8178 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8179 num_bytes);
8180 if (ret)
8181 return ret;
8182
8183 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
8184 1);
8185 if (ret) { /* -ENOENT, logic error */
8186 btrfs_err(fs_info, "update block group failed for %llu %llu",
8187 ins->objectid, ins->offset);
8188 BUG();
8189 }
8190
8191 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
8192 return ret;
8193 }
8194
8195 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8196 struct btrfs_root *root,
8197 u64 root_objectid, u64 owner,
8198 u64 offset, u64 ram_bytes,
8199 struct btrfs_key *ins)
8200 {
8201 int ret;
8202
8203 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
8204
8205 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
8206 ins->offset, 0,
8207 root_objectid, owner, offset,
8208 ram_bytes, BTRFS_ADD_DELAYED_EXTENT,
8209 NULL);
8210 return ret;
8211 }
8212
8213 /*
8214 * this is used by the tree logging recovery code. It records that
8215 * an extent has been allocated and makes sure to clear the free
8216 * space cache bits as well
8217 */
8218 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
8219 struct btrfs_root *root,
8220 u64 root_objectid, u64 owner, u64 offset,
8221 struct btrfs_key *ins)
8222 {
8223 int ret;
8224 struct btrfs_block_group_cache *block_group;
8225
8226 /*
8227 * Mixed block groups will exclude before processing the log so we only
8228 * need to do the exclude dance if this fs isn't mixed.
8229 */
8230 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
8231 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
8232 if (ret)
8233 return ret;
8234 }
8235
8236 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
8237 if (!block_group)
8238 return -EINVAL;
8239
8240 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
8241 RESERVE_ALLOC_NO_ACCOUNT, 0);
8242 BUG_ON(ret); /* logic error */
8243 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
8244 0, owner, offset, ins, 1);
8245 btrfs_put_block_group(block_group);
8246 return ret;
8247 }
8248
8249 static struct extent_buffer *
8250 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
8251 u64 bytenr, int level)
8252 {
8253 struct extent_buffer *buf;
8254
8255 buf = btrfs_find_create_tree_block(root, bytenr);
8256 if (IS_ERR(buf))
8257 return buf;
8258
8259 btrfs_set_header_generation(buf, trans->transid);
8260 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
8261 btrfs_tree_lock(buf);
8262 clean_tree_block(trans, root->fs_info, buf);
8263 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
8264
8265 btrfs_set_lock_blocking(buf);
8266 set_extent_buffer_uptodate(buf);
8267
8268 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8269 buf->log_index = root->log_transid % 2;
8270 /*
8271 * we allow two log transactions at a time, use different
8272 * EXENT bit to differentiate dirty pages.
8273 */
8274 if (buf->log_index == 0)
8275 set_extent_dirty(&root->dirty_log_pages, buf->start,
8276 buf->start + buf->len - 1, GFP_NOFS);
8277 else
8278 set_extent_new(&root->dirty_log_pages, buf->start,
8279 buf->start + buf->len - 1);
8280 } else {
8281 buf->log_index = -1;
8282 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
8283 buf->start + buf->len - 1, GFP_NOFS);
8284 }
8285 trans->dirty = true;
8286 /* this returns a buffer locked for blocking */
8287 return buf;
8288 }
8289
8290 static struct btrfs_block_rsv *
8291 use_block_rsv(struct btrfs_trans_handle *trans,
8292 struct btrfs_root *root, u32 blocksize)
8293 {
8294 struct btrfs_block_rsv *block_rsv;
8295 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
8296 int ret;
8297 bool global_updated = false;
8298
8299 block_rsv = get_block_rsv(trans, root);
8300
8301 if (unlikely(block_rsv->size == 0))
8302 goto try_reserve;
8303 again:
8304 ret = block_rsv_use_bytes(block_rsv, blocksize);
8305 if (!ret)
8306 return block_rsv;
8307
8308 if (block_rsv->failfast)
8309 return ERR_PTR(ret);
8310
8311 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
8312 global_updated = true;
8313 update_global_block_rsv(root->fs_info);
8314 goto again;
8315 }
8316
8317 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
8318 static DEFINE_RATELIMIT_STATE(_rs,
8319 DEFAULT_RATELIMIT_INTERVAL * 10,
8320 /*DEFAULT_RATELIMIT_BURST*/ 1);
8321 if (__ratelimit(&_rs))
8322 WARN(1, KERN_DEBUG
8323 "BTRFS: block rsv returned %d\n", ret);
8324 }
8325 try_reserve:
8326 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
8327 BTRFS_RESERVE_NO_FLUSH);
8328 if (!ret)
8329 return block_rsv;
8330 /*
8331 * If we couldn't reserve metadata bytes try and use some from
8332 * the global reserve if its space type is the same as the global
8333 * reservation.
8334 */
8335 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
8336 block_rsv->space_info == global_rsv->space_info) {
8337 ret = block_rsv_use_bytes(global_rsv, blocksize);
8338 if (!ret)
8339 return global_rsv;
8340 }
8341 return ERR_PTR(ret);
8342 }
8343
8344 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
8345 struct btrfs_block_rsv *block_rsv, u32 blocksize)
8346 {
8347 block_rsv_add_bytes(block_rsv, blocksize, 0);
8348 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
8349 }
8350
8351 /*
8352 * finds a free extent and does all the dirty work required for allocation
8353 * returns the tree buffer or an ERR_PTR on error.
8354 */
8355 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8356 struct btrfs_root *root,
8357 u64 parent, u64 root_objectid,
8358 struct btrfs_disk_key *key, int level,
8359 u64 hint, u64 empty_size)
8360 {
8361 struct btrfs_key ins;
8362 struct btrfs_block_rsv *block_rsv;
8363 struct extent_buffer *buf;
8364 struct btrfs_delayed_extent_op *extent_op;
8365 u64 flags = 0;
8366 int ret;
8367 u32 blocksize = root->nodesize;
8368 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
8369 SKINNY_METADATA);
8370
8371 if (btrfs_test_is_dummy_root(root)) {
8372 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8373 level);
8374 if (!IS_ERR(buf))
8375 root->alloc_bytenr += blocksize;
8376 return buf;
8377 }
8378
8379 block_rsv = use_block_rsv(trans, root, blocksize);
8380 if (IS_ERR(block_rsv))
8381 return ERR_CAST(block_rsv);
8382
8383 ret = btrfs_reserve_extent(root, blocksize, blocksize,
8384 empty_size, hint, &ins, 0, 0);
8385 if (ret)
8386 goto out_unuse;
8387
8388 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
8389 if (IS_ERR(buf)) {
8390 ret = PTR_ERR(buf);
8391 goto out_free_reserved;
8392 }
8393
8394 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8395 if (parent == 0)
8396 parent = ins.objectid;
8397 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8398 } else
8399 BUG_ON(parent > 0);
8400
8401 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8402 extent_op = btrfs_alloc_delayed_extent_op();
8403 if (!extent_op) {
8404 ret = -ENOMEM;
8405 goto out_free_buf;
8406 }
8407 if (key)
8408 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8409 else
8410 memset(&extent_op->key, 0, sizeof(extent_op->key));
8411 extent_op->flags_to_set = flags;
8412 extent_op->update_key = skinny_metadata ? false : true;
8413 extent_op->update_flags = true;
8414 extent_op->is_data = false;
8415 extent_op->level = level;
8416
8417 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
8418 ins.objectid, ins.offset,
8419 parent, root_objectid, level,
8420 BTRFS_ADD_DELAYED_EXTENT,
8421 extent_op);
8422 if (ret)
8423 goto out_free_delayed;
8424 }
8425 return buf;
8426
8427 out_free_delayed:
8428 btrfs_free_delayed_extent_op(extent_op);
8429 out_free_buf:
8430 free_extent_buffer(buf);
8431 out_free_reserved:
8432 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0);
8433 out_unuse:
8434 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
8435 return ERR_PTR(ret);
8436 }
8437
8438 struct walk_control {
8439 u64 refs[BTRFS_MAX_LEVEL];
8440 u64 flags[BTRFS_MAX_LEVEL];
8441 struct btrfs_key update_progress;
8442 int stage;
8443 int level;
8444 int shared_level;
8445 int update_ref;
8446 int keep_locks;
8447 int reada_slot;
8448 int reada_count;
8449 int for_reloc;
8450 };
8451
8452 #define DROP_REFERENCE 1
8453 #define UPDATE_BACKREF 2
8454
8455 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8456 struct btrfs_root *root,
8457 struct walk_control *wc,
8458 struct btrfs_path *path)
8459 {
8460 u64 bytenr;
8461 u64 generation;
8462 u64 refs;
8463 u64 flags;
8464 u32 nritems;
8465 u32 blocksize;
8466 struct btrfs_key key;
8467 struct extent_buffer *eb;
8468 int ret;
8469 int slot;
8470 int nread = 0;
8471
8472 if (path->slots[wc->level] < wc->reada_slot) {
8473 wc->reada_count = wc->reada_count * 2 / 3;
8474 wc->reada_count = max(wc->reada_count, 2);
8475 } else {
8476 wc->reada_count = wc->reada_count * 3 / 2;
8477 wc->reada_count = min_t(int, wc->reada_count,
8478 BTRFS_NODEPTRS_PER_BLOCK(root));
8479 }
8480
8481 eb = path->nodes[wc->level];
8482 nritems = btrfs_header_nritems(eb);
8483 blocksize = root->nodesize;
8484
8485 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8486 if (nread >= wc->reada_count)
8487 break;
8488
8489 cond_resched();
8490 bytenr = btrfs_node_blockptr(eb, slot);
8491 generation = btrfs_node_ptr_generation(eb, slot);
8492
8493 if (slot == path->slots[wc->level])
8494 goto reada;
8495
8496 if (wc->stage == UPDATE_BACKREF &&
8497 generation <= root->root_key.offset)
8498 continue;
8499
8500 /* We don't lock the tree block, it's OK to be racy here */
8501 ret = btrfs_lookup_extent_info(trans, root, bytenr,
8502 wc->level - 1, 1, &refs,
8503 &flags);
8504 /* We don't care about errors in readahead. */
8505 if (ret < 0)
8506 continue;
8507 BUG_ON(refs == 0);
8508
8509 if (wc->stage == DROP_REFERENCE) {
8510 if (refs == 1)
8511 goto reada;
8512
8513 if (wc->level == 1 &&
8514 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8515 continue;
8516 if (!wc->update_ref ||
8517 generation <= root->root_key.offset)
8518 continue;
8519 btrfs_node_key_to_cpu(eb, &key, slot);
8520 ret = btrfs_comp_cpu_keys(&key,
8521 &wc->update_progress);
8522 if (ret < 0)
8523 continue;
8524 } else {
8525 if (wc->level == 1 &&
8526 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8527 continue;
8528 }
8529 reada:
8530 readahead_tree_block(root, bytenr);
8531 nread++;
8532 }
8533 wc->reada_slot = slot;
8534 }
8535
8536 /*
8537 * These may not be seen by the usual inc/dec ref code so we have to
8538 * add them here.
8539 */
8540 static int record_one_subtree_extent(struct btrfs_trans_handle *trans,
8541 struct btrfs_root *root, u64 bytenr,
8542 u64 num_bytes)
8543 {
8544 struct btrfs_qgroup_extent_record *qrecord;
8545 struct btrfs_delayed_ref_root *delayed_refs;
8546
8547 qrecord = kmalloc(sizeof(*qrecord), GFP_NOFS);
8548 if (!qrecord)
8549 return -ENOMEM;
8550
8551 qrecord->bytenr = bytenr;
8552 qrecord->num_bytes = num_bytes;
8553 qrecord->old_roots = NULL;
8554
8555 delayed_refs = &trans->transaction->delayed_refs;
8556 spin_lock(&delayed_refs->lock);
8557 if (btrfs_qgroup_insert_dirty_extent(delayed_refs, qrecord))
8558 kfree(qrecord);
8559 spin_unlock(&delayed_refs->lock);
8560
8561 return 0;
8562 }
8563
8564 static int account_leaf_items(struct btrfs_trans_handle *trans,
8565 struct btrfs_root *root,
8566 struct extent_buffer *eb)
8567 {
8568 int nr = btrfs_header_nritems(eb);
8569 int i, extent_type, ret;
8570 struct btrfs_key key;
8571 struct btrfs_file_extent_item *fi;
8572 u64 bytenr, num_bytes;
8573
8574 /* We can be called directly from walk_up_proc() */
8575 if (!root->fs_info->quota_enabled)
8576 return 0;
8577
8578 for (i = 0; i < nr; i++) {
8579 btrfs_item_key_to_cpu(eb, &key, i);
8580
8581 if (key.type != BTRFS_EXTENT_DATA_KEY)
8582 continue;
8583
8584 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
8585 /* filter out non qgroup-accountable extents */
8586 extent_type = btrfs_file_extent_type(eb, fi);
8587
8588 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
8589 continue;
8590
8591 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
8592 if (!bytenr)
8593 continue;
8594
8595 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
8596
8597 ret = record_one_subtree_extent(trans, root, bytenr, num_bytes);
8598 if (ret)
8599 return ret;
8600 }
8601 return 0;
8602 }
8603
8604 /*
8605 * Walk up the tree from the bottom, freeing leaves and any interior
8606 * nodes which have had all slots visited. If a node (leaf or
8607 * interior) is freed, the node above it will have it's slot
8608 * incremented. The root node will never be freed.
8609 *
8610 * At the end of this function, we should have a path which has all
8611 * slots incremented to the next position for a search. If we need to
8612 * read a new node it will be NULL and the node above it will have the
8613 * correct slot selected for a later read.
8614 *
8615 * If we increment the root nodes slot counter past the number of
8616 * elements, 1 is returned to signal completion of the search.
8617 */
8618 static int adjust_slots_upwards(struct btrfs_root *root,
8619 struct btrfs_path *path, int root_level)
8620 {
8621 int level = 0;
8622 int nr, slot;
8623 struct extent_buffer *eb;
8624
8625 if (root_level == 0)
8626 return 1;
8627
8628 while (level <= root_level) {
8629 eb = path->nodes[level];
8630 nr = btrfs_header_nritems(eb);
8631 path->slots[level]++;
8632 slot = path->slots[level];
8633 if (slot >= nr || level == 0) {
8634 /*
8635 * Don't free the root - we will detect this
8636 * condition after our loop and return a
8637 * positive value for caller to stop walking the tree.
8638 */
8639 if (level != root_level) {
8640 btrfs_tree_unlock_rw(eb, path->locks[level]);
8641 path->locks[level] = 0;
8642
8643 free_extent_buffer(eb);
8644 path->nodes[level] = NULL;
8645 path->slots[level] = 0;
8646 }
8647 } else {
8648 /*
8649 * We have a valid slot to walk back down
8650 * from. Stop here so caller can process these
8651 * new nodes.
8652 */
8653 break;
8654 }
8655
8656 level++;
8657 }
8658
8659 eb = path->nodes[root_level];
8660 if (path->slots[root_level] >= btrfs_header_nritems(eb))
8661 return 1;
8662
8663 return 0;
8664 }
8665
8666 /*
8667 * root_eb is the subtree root and is locked before this function is called.
8668 */
8669 static int account_shared_subtree(struct btrfs_trans_handle *trans,
8670 struct btrfs_root *root,
8671 struct extent_buffer *root_eb,
8672 u64 root_gen,
8673 int root_level)
8674 {
8675 int ret = 0;
8676 int level;
8677 struct extent_buffer *eb = root_eb;
8678 struct btrfs_path *path = NULL;
8679
8680 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
8681 BUG_ON(root_eb == NULL);
8682
8683 if (!root->fs_info->quota_enabled)
8684 return 0;
8685
8686 if (!extent_buffer_uptodate(root_eb)) {
8687 ret = btrfs_read_buffer(root_eb, root_gen);
8688 if (ret)
8689 goto out;
8690 }
8691
8692 if (root_level == 0) {
8693 ret = account_leaf_items(trans, root, root_eb);
8694 goto out;
8695 }
8696
8697 path = btrfs_alloc_path();
8698 if (!path)
8699 return -ENOMEM;
8700
8701 /*
8702 * Walk down the tree. Missing extent blocks are filled in as
8703 * we go. Metadata is accounted every time we read a new
8704 * extent block.
8705 *
8706 * When we reach a leaf, we account for file extent items in it,
8707 * walk back up the tree (adjusting slot pointers as we go)
8708 * and restart the search process.
8709 */
8710 extent_buffer_get(root_eb); /* For path */
8711 path->nodes[root_level] = root_eb;
8712 path->slots[root_level] = 0;
8713 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
8714 walk_down:
8715 level = root_level;
8716 while (level >= 0) {
8717 if (path->nodes[level] == NULL) {
8718 int parent_slot;
8719 u64 child_gen;
8720 u64 child_bytenr;
8721
8722 /* We need to get child blockptr/gen from
8723 * parent before we can read it. */
8724 eb = path->nodes[level + 1];
8725 parent_slot = path->slots[level + 1];
8726 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
8727 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
8728
8729 eb = read_tree_block(root, child_bytenr, child_gen);
8730 if (IS_ERR(eb)) {
8731 ret = PTR_ERR(eb);
8732 goto out;
8733 } else if (!extent_buffer_uptodate(eb)) {
8734 free_extent_buffer(eb);
8735 ret = -EIO;
8736 goto out;
8737 }
8738
8739 path->nodes[level] = eb;
8740 path->slots[level] = 0;
8741
8742 btrfs_tree_read_lock(eb);
8743 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
8744 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
8745
8746 ret = record_one_subtree_extent(trans, root, child_bytenr,
8747 root->nodesize);
8748 if (ret)
8749 goto out;
8750 }
8751
8752 if (level == 0) {
8753 ret = account_leaf_items(trans, root, path->nodes[level]);
8754 if (ret)
8755 goto out;
8756
8757 /* Nonzero return here means we completed our search */
8758 ret = adjust_slots_upwards(root, path, root_level);
8759 if (ret)
8760 break;
8761
8762 /* Restart search with new slots */
8763 goto walk_down;
8764 }
8765
8766 level--;
8767 }
8768
8769 ret = 0;
8770 out:
8771 btrfs_free_path(path);
8772
8773 return ret;
8774 }
8775
8776 /*
8777 * helper to process tree block while walking down the tree.
8778 *
8779 * when wc->stage == UPDATE_BACKREF, this function updates
8780 * back refs for pointers in the block.
8781 *
8782 * NOTE: return value 1 means we should stop walking down.
8783 */
8784 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8785 struct btrfs_root *root,
8786 struct btrfs_path *path,
8787 struct walk_control *wc, int lookup_info)
8788 {
8789 int level = wc->level;
8790 struct extent_buffer *eb = path->nodes[level];
8791 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8792 int ret;
8793
8794 if (wc->stage == UPDATE_BACKREF &&
8795 btrfs_header_owner(eb) != root->root_key.objectid)
8796 return 1;
8797
8798 /*
8799 * when reference count of tree block is 1, it won't increase
8800 * again. once full backref flag is set, we never clear it.
8801 */
8802 if (lookup_info &&
8803 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8804 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8805 BUG_ON(!path->locks[level]);
8806 ret = btrfs_lookup_extent_info(trans, root,
8807 eb->start, level, 1,
8808 &wc->refs[level],
8809 &wc->flags[level]);
8810 BUG_ON(ret == -ENOMEM);
8811 if (ret)
8812 return ret;
8813 BUG_ON(wc->refs[level] == 0);
8814 }
8815
8816 if (wc->stage == DROP_REFERENCE) {
8817 if (wc->refs[level] > 1)
8818 return 1;
8819
8820 if (path->locks[level] && !wc->keep_locks) {
8821 btrfs_tree_unlock_rw(eb, path->locks[level]);
8822 path->locks[level] = 0;
8823 }
8824 return 0;
8825 }
8826
8827 /* wc->stage == UPDATE_BACKREF */
8828 if (!(wc->flags[level] & flag)) {
8829 BUG_ON(!path->locks[level]);
8830 ret = btrfs_inc_ref(trans, root, eb, 1);
8831 BUG_ON(ret); /* -ENOMEM */
8832 ret = btrfs_dec_ref(trans, root, eb, 0);
8833 BUG_ON(ret); /* -ENOMEM */
8834 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
8835 eb->len, flag,
8836 btrfs_header_level(eb), 0);
8837 BUG_ON(ret); /* -ENOMEM */
8838 wc->flags[level] |= flag;
8839 }
8840
8841 /*
8842 * the block is shared by multiple trees, so it's not good to
8843 * keep the tree lock
8844 */
8845 if (path->locks[level] && level > 0) {
8846 btrfs_tree_unlock_rw(eb, path->locks[level]);
8847 path->locks[level] = 0;
8848 }
8849 return 0;
8850 }
8851
8852 /*
8853 * helper to process tree block pointer.
8854 *
8855 * when wc->stage == DROP_REFERENCE, this function checks
8856 * reference count of the block pointed to. if the block
8857 * is shared and we need update back refs for the subtree
8858 * rooted at the block, this function changes wc->stage to
8859 * UPDATE_BACKREF. if the block is shared and there is no
8860 * need to update back, this function drops the reference
8861 * to the block.
8862 *
8863 * NOTE: return value 1 means we should stop walking down.
8864 */
8865 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8866 struct btrfs_root *root,
8867 struct btrfs_path *path,
8868 struct walk_control *wc, int *lookup_info)
8869 {
8870 u64 bytenr;
8871 u64 generation;
8872 u64 parent;
8873 u32 blocksize;
8874 struct btrfs_key key;
8875 struct extent_buffer *next;
8876 int level = wc->level;
8877 int reada = 0;
8878 int ret = 0;
8879 bool need_account = false;
8880
8881 generation = btrfs_node_ptr_generation(path->nodes[level],
8882 path->slots[level]);
8883 /*
8884 * if the lower level block was created before the snapshot
8885 * was created, we know there is no need to update back refs
8886 * for the subtree
8887 */
8888 if (wc->stage == UPDATE_BACKREF &&
8889 generation <= root->root_key.offset) {
8890 *lookup_info = 1;
8891 return 1;
8892 }
8893
8894 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8895 blocksize = root->nodesize;
8896
8897 next = btrfs_find_tree_block(root->fs_info, bytenr);
8898 if (!next) {
8899 next = btrfs_find_create_tree_block(root, bytenr);
8900 if (IS_ERR(next))
8901 return PTR_ERR(next);
8902
8903 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8904 level - 1);
8905 reada = 1;
8906 }
8907 btrfs_tree_lock(next);
8908 btrfs_set_lock_blocking(next);
8909
8910 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
8911 &wc->refs[level - 1],
8912 &wc->flags[level - 1]);
8913 if (ret < 0) {
8914 btrfs_tree_unlock(next);
8915 return ret;
8916 }
8917
8918 if (unlikely(wc->refs[level - 1] == 0)) {
8919 btrfs_err(root->fs_info, "Missing references.");
8920 BUG();
8921 }
8922 *lookup_info = 0;
8923
8924 if (wc->stage == DROP_REFERENCE) {
8925 if (wc->refs[level - 1] > 1) {
8926 need_account = true;
8927 if (level == 1 &&
8928 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8929 goto skip;
8930
8931 if (!wc->update_ref ||
8932 generation <= root->root_key.offset)
8933 goto skip;
8934
8935 btrfs_node_key_to_cpu(path->nodes[level], &key,
8936 path->slots[level]);
8937 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8938 if (ret < 0)
8939 goto skip;
8940
8941 wc->stage = UPDATE_BACKREF;
8942 wc->shared_level = level - 1;
8943 }
8944 } else {
8945 if (level == 1 &&
8946 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8947 goto skip;
8948 }
8949
8950 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8951 btrfs_tree_unlock(next);
8952 free_extent_buffer(next);
8953 next = NULL;
8954 *lookup_info = 1;
8955 }
8956
8957 if (!next) {
8958 if (reada && level == 1)
8959 reada_walk_down(trans, root, wc, path);
8960 next = read_tree_block(root, bytenr, generation);
8961 if (IS_ERR(next)) {
8962 return PTR_ERR(next);
8963 } else if (!extent_buffer_uptodate(next)) {
8964 free_extent_buffer(next);
8965 return -EIO;
8966 }
8967 btrfs_tree_lock(next);
8968 btrfs_set_lock_blocking(next);
8969 }
8970
8971 level--;
8972 BUG_ON(level != btrfs_header_level(next));
8973 path->nodes[level] = next;
8974 path->slots[level] = 0;
8975 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8976 wc->level = level;
8977 if (wc->level == 1)
8978 wc->reada_slot = 0;
8979 return 0;
8980 skip:
8981 wc->refs[level - 1] = 0;
8982 wc->flags[level - 1] = 0;
8983 if (wc->stage == DROP_REFERENCE) {
8984 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8985 parent = path->nodes[level]->start;
8986 } else {
8987 BUG_ON(root->root_key.objectid !=
8988 btrfs_header_owner(path->nodes[level]));
8989 parent = 0;
8990 }
8991
8992 if (need_account) {
8993 ret = account_shared_subtree(trans, root, next,
8994 generation, level - 1);
8995 if (ret) {
8996 btrfs_err_rl(root->fs_info,
8997 "Error "
8998 "%d accounting shared subtree. Quota "
8999 "is out of sync, rescan required.",
9000 ret);
9001 }
9002 }
9003 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
9004 root->root_key.objectid, level - 1, 0);
9005 BUG_ON(ret); /* -ENOMEM */
9006 }
9007 btrfs_tree_unlock(next);
9008 free_extent_buffer(next);
9009 *lookup_info = 1;
9010 return 1;
9011 }
9012
9013 /*
9014 * helper to process tree block while walking up the tree.
9015 *
9016 * when wc->stage == DROP_REFERENCE, this function drops
9017 * reference count on the block.
9018 *
9019 * when wc->stage == UPDATE_BACKREF, this function changes
9020 * wc->stage back to DROP_REFERENCE if we changed wc->stage
9021 * to UPDATE_BACKREF previously while processing the block.
9022 *
9023 * NOTE: return value 1 means we should stop walking up.
9024 */
9025 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
9026 struct btrfs_root *root,
9027 struct btrfs_path *path,
9028 struct walk_control *wc)
9029 {
9030 int ret;
9031 int level = wc->level;
9032 struct extent_buffer *eb = path->nodes[level];
9033 u64 parent = 0;
9034
9035 if (wc->stage == UPDATE_BACKREF) {
9036 BUG_ON(wc->shared_level < level);
9037 if (level < wc->shared_level)
9038 goto out;
9039
9040 ret = find_next_key(path, level + 1, &wc->update_progress);
9041 if (ret > 0)
9042 wc->update_ref = 0;
9043
9044 wc->stage = DROP_REFERENCE;
9045 wc->shared_level = -1;
9046 path->slots[level] = 0;
9047
9048 /*
9049 * check reference count again if the block isn't locked.
9050 * we should start walking down the tree again if reference
9051 * count is one.
9052 */
9053 if (!path->locks[level]) {
9054 BUG_ON(level == 0);
9055 btrfs_tree_lock(eb);
9056 btrfs_set_lock_blocking(eb);
9057 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9058
9059 ret = btrfs_lookup_extent_info(trans, root,
9060 eb->start, level, 1,
9061 &wc->refs[level],
9062 &wc->flags[level]);
9063 if (ret < 0) {
9064 btrfs_tree_unlock_rw(eb, path->locks[level]);
9065 path->locks[level] = 0;
9066 return ret;
9067 }
9068 BUG_ON(wc->refs[level] == 0);
9069 if (wc->refs[level] == 1) {
9070 btrfs_tree_unlock_rw(eb, path->locks[level]);
9071 path->locks[level] = 0;
9072 return 1;
9073 }
9074 }
9075 }
9076
9077 /* wc->stage == DROP_REFERENCE */
9078 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
9079
9080 if (wc->refs[level] == 1) {
9081 if (level == 0) {
9082 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9083 ret = btrfs_dec_ref(trans, root, eb, 1);
9084 else
9085 ret = btrfs_dec_ref(trans, root, eb, 0);
9086 BUG_ON(ret); /* -ENOMEM */
9087 ret = account_leaf_items(trans, root, eb);
9088 if (ret) {
9089 btrfs_err_rl(root->fs_info,
9090 "error "
9091 "%d accounting leaf items. Quota "
9092 "is out of sync, rescan required.",
9093 ret);
9094 }
9095 }
9096 /* make block locked assertion in clean_tree_block happy */
9097 if (!path->locks[level] &&
9098 btrfs_header_generation(eb) == trans->transid) {
9099 btrfs_tree_lock(eb);
9100 btrfs_set_lock_blocking(eb);
9101 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9102 }
9103 clean_tree_block(trans, root->fs_info, eb);
9104 }
9105
9106 if (eb == root->node) {
9107 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9108 parent = eb->start;
9109 else
9110 BUG_ON(root->root_key.objectid !=
9111 btrfs_header_owner(eb));
9112 } else {
9113 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9114 parent = path->nodes[level + 1]->start;
9115 else
9116 BUG_ON(root->root_key.objectid !=
9117 btrfs_header_owner(path->nodes[level + 1]));
9118 }
9119
9120 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
9121 out:
9122 wc->refs[level] = 0;
9123 wc->flags[level] = 0;
9124 return 0;
9125 }
9126
9127 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
9128 struct btrfs_root *root,
9129 struct btrfs_path *path,
9130 struct walk_control *wc)
9131 {
9132 int level = wc->level;
9133 int lookup_info = 1;
9134 int ret;
9135
9136 while (level >= 0) {
9137 ret = walk_down_proc(trans, root, path, wc, lookup_info);
9138 if (ret > 0)
9139 break;
9140
9141 if (level == 0)
9142 break;
9143
9144 if (path->slots[level] >=
9145 btrfs_header_nritems(path->nodes[level]))
9146 break;
9147
9148 ret = do_walk_down(trans, root, path, wc, &lookup_info);
9149 if (ret > 0) {
9150 path->slots[level]++;
9151 continue;
9152 } else if (ret < 0)
9153 return ret;
9154 level = wc->level;
9155 }
9156 return 0;
9157 }
9158
9159 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
9160 struct btrfs_root *root,
9161 struct btrfs_path *path,
9162 struct walk_control *wc, int max_level)
9163 {
9164 int level = wc->level;
9165 int ret;
9166
9167 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
9168 while (level < max_level && path->nodes[level]) {
9169 wc->level = level;
9170 if (path->slots[level] + 1 <
9171 btrfs_header_nritems(path->nodes[level])) {
9172 path->slots[level]++;
9173 return 0;
9174 } else {
9175 ret = walk_up_proc(trans, root, path, wc);
9176 if (ret > 0)
9177 return 0;
9178
9179 if (path->locks[level]) {
9180 btrfs_tree_unlock_rw(path->nodes[level],
9181 path->locks[level]);
9182 path->locks[level] = 0;
9183 }
9184 free_extent_buffer(path->nodes[level]);
9185 path->nodes[level] = NULL;
9186 level++;
9187 }
9188 }
9189 return 1;
9190 }
9191
9192 /*
9193 * drop a subvolume tree.
9194 *
9195 * this function traverses the tree freeing any blocks that only
9196 * referenced by the tree.
9197 *
9198 * when a shared tree block is found. this function decreases its
9199 * reference count by one. if update_ref is true, this function
9200 * also make sure backrefs for the shared block and all lower level
9201 * blocks are properly updated.
9202 *
9203 * If called with for_reloc == 0, may exit early with -EAGAIN
9204 */
9205 int btrfs_drop_snapshot(struct btrfs_root *root,
9206 struct btrfs_block_rsv *block_rsv, int update_ref,
9207 int for_reloc)
9208 {
9209 struct btrfs_path *path;
9210 struct btrfs_trans_handle *trans;
9211 struct btrfs_root *tree_root = root->fs_info->tree_root;
9212 struct btrfs_root_item *root_item = &root->root_item;
9213 struct walk_control *wc;
9214 struct btrfs_key key;
9215 int err = 0;
9216 int ret;
9217 int level;
9218 bool root_dropped = false;
9219
9220 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
9221
9222 path = btrfs_alloc_path();
9223 if (!path) {
9224 err = -ENOMEM;
9225 goto out;
9226 }
9227
9228 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9229 if (!wc) {
9230 btrfs_free_path(path);
9231 err = -ENOMEM;
9232 goto out;
9233 }
9234
9235 trans = btrfs_start_transaction(tree_root, 0);
9236 if (IS_ERR(trans)) {
9237 err = PTR_ERR(trans);
9238 goto out_free;
9239 }
9240
9241 if (block_rsv)
9242 trans->block_rsv = block_rsv;
9243
9244 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
9245 level = btrfs_header_level(root->node);
9246 path->nodes[level] = btrfs_lock_root_node(root);
9247 btrfs_set_lock_blocking(path->nodes[level]);
9248 path->slots[level] = 0;
9249 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9250 memset(&wc->update_progress, 0,
9251 sizeof(wc->update_progress));
9252 } else {
9253 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
9254 memcpy(&wc->update_progress, &key,
9255 sizeof(wc->update_progress));
9256
9257 level = root_item->drop_level;
9258 BUG_ON(level == 0);
9259 path->lowest_level = level;
9260 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
9261 path->lowest_level = 0;
9262 if (ret < 0) {
9263 err = ret;
9264 goto out_end_trans;
9265 }
9266 WARN_ON(ret > 0);
9267
9268 /*
9269 * unlock our path, this is safe because only this
9270 * function is allowed to delete this snapshot
9271 */
9272 btrfs_unlock_up_safe(path, 0);
9273
9274 level = btrfs_header_level(root->node);
9275 while (1) {
9276 btrfs_tree_lock(path->nodes[level]);
9277 btrfs_set_lock_blocking(path->nodes[level]);
9278 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9279
9280 ret = btrfs_lookup_extent_info(trans, root,
9281 path->nodes[level]->start,
9282 level, 1, &wc->refs[level],
9283 &wc->flags[level]);
9284 if (ret < 0) {
9285 err = ret;
9286 goto out_end_trans;
9287 }
9288 BUG_ON(wc->refs[level] == 0);
9289
9290 if (level == root_item->drop_level)
9291 break;
9292
9293 btrfs_tree_unlock(path->nodes[level]);
9294 path->locks[level] = 0;
9295 WARN_ON(wc->refs[level] != 1);
9296 level--;
9297 }
9298 }
9299
9300 wc->level = level;
9301 wc->shared_level = -1;
9302 wc->stage = DROP_REFERENCE;
9303 wc->update_ref = update_ref;
9304 wc->keep_locks = 0;
9305 wc->for_reloc = for_reloc;
9306 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
9307
9308 while (1) {
9309
9310 ret = walk_down_tree(trans, root, path, wc);
9311 if (ret < 0) {
9312 err = ret;
9313 break;
9314 }
9315
9316 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
9317 if (ret < 0) {
9318 err = ret;
9319 break;
9320 }
9321
9322 if (ret > 0) {
9323 BUG_ON(wc->stage != DROP_REFERENCE);
9324 break;
9325 }
9326
9327 if (wc->stage == DROP_REFERENCE) {
9328 level = wc->level;
9329 btrfs_node_key(path->nodes[level],
9330 &root_item->drop_progress,
9331 path->slots[level]);
9332 root_item->drop_level = level;
9333 }
9334
9335 BUG_ON(wc->level == 0);
9336 if (btrfs_should_end_transaction(trans, tree_root) ||
9337 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
9338 ret = btrfs_update_root(trans, tree_root,
9339 &root->root_key,
9340 root_item);
9341 if (ret) {
9342 btrfs_abort_transaction(trans, tree_root, ret);
9343 err = ret;
9344 goto out_end_trans;
9345 }
9346
9347 btrfs_end_transaction_throttle(trans, tree_root);
9348 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
9349 pr_debug("BTRFS: drop snapshot early exit\n");
9350 err = -EAGAIN;
9351 goto out_free;
9352 }
9353
9354 trans = btrfs_start_transaction(tree_root, 0);
9355 if (IS_ERR(trans)) {
9356 err = PTR_ERR(trans);
9357 goto out_free;
9358 }
9359 if (block_rsv)
9360 trans->block_rsv = block_rsv;
9361 }
9362 }
9363 btrfs_release_path(path);
9364 if (err)
9365 goto out_end_trans;
9366
9367 ret = btrfs_del_root(trans, tree_root, &root->root_key);
9368 if (ret) {
9369 btrfs_abort_transaction(trans, tree_root, ret);
9370 goto out_end_trans;
9371 }
9372
9373 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9374 ret = btrfs_find_root(tree_root, &root->root_key, path,
9375 NULL, NULL);
9376 if (ret < 0) {
9377 btrfs_abort_transaction(trans, tree_root, ret);
9378 err = ret;
9379 goto out_end_trans;
9380 } else if (ret > 0) {
9381 /* if we fail to delete the orphan item this time
9382 * around, it'll get picked up the next time.
9383 *
9384 * The most common failure here is just -ENOENT.
9385 */
9386 btrfs_del_orphan_item(trans, tree_root,
9387 root->root_key.objectid);
9388 }
9389 }
9390
9391 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9392 btrfs_add_dropped_root(trans, root);
9393 } else {
9394 free_extent_buffer(root->node);
9395 free_extent_buffer(root->commit_root);
9396 btrfs_put_fs_root(root);
9397 }
9398 root_dropped = true;
9399 out_end_trans:
9400 btrfs_end_transaction_throttle(trans, tree_root);
9401 out_free:
9402 kfree(wc);
9403 btrfs_free_path(path);
9404 out:
9405 /*
9406 * So if we need to stop dropping the snapshot for whatever reason we
9407 * need to make sure to add it back to the dead root list so that we
9408 * keep trying to do the work later. This also cleans up roots if we
9409 * don't have it in the radix (like when we recover after a power fail
9410 * or unmount) so we don't leak memory.
9411 */
9412 if (!for_reloc && root_dropped == false)
9413 btrfs_add_dead_root(root);
9414 if (err && err != -EAGAIN)
9415 btrfs_handle_fs_error(root->fs_info, err, NULL);
9416 return err;
9417 }
9418
9419 /*
9420 * drop subtree rooted at tree block 'node'.
9421 *
9422 * NOTE: this function will unlock and release tree block 'node'
9423 * only used by relocation code
9424 */
9425 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9426 struct btrfs_root *root,
9427 struct extent_buffer *node,
9428 struct extent_buffer *parent)
9429 {
9430 struct btrfs_path *path;
9431 struct walk_control *wc;
9432 int level;
9433 int parent_level;
9434 int ret = 0;
9435 int wret;
9436
9437 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9438
9439 path = btrfs_alloc_path();
9440 if (!path)
9441 return -ENOMEM;
9442
9443 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9444 if (!wc) {
9445 btrfs_free_path(path);
9446 return -ENOMEM;
9447 }
9448
9449 btrfs_assert_tree_locked(parent);
9450 parent_level = btrfs_header_level(parent);
9451 extent_buffer_get(parent);
9452 path->nodes[parent_level] = parent;
9453 path->slots[parent_level] = btrfs_header_nritems(parent);
9454
9455 btrfs_assert_tree_locked(node);
9456 level = btrfs_header_level(node);
9457 path->nodes[level] = node;
9458 path->slots[level] = 0;
9459 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9460
9461 wc->refs[parent_level] = 1;
9462 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9463 wc->level = level;
9464 wc->shared_level = -1;
9465 wc->stage = DROP_REFERENCE;
9466 wc->update_ref = 0;
9467 wc->keep_locks = 1;
9468 wc->for_reloc = 1;
9469 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
9470
9471 while (1) {
9472 wret = walk_down_tree(trans, root, path, wc);
9473 if (wret < 0) {
9474 ret = wret;
9475 break;
9476 }
9477
9478 wret = walk_up_tree(trans, root, path, wc, parent_level);
9479 if (wret < 0)
9480 ret = wret;
9481 if (wret != 0)
9482 break;
9483 }
9484
9485 kfree(wc);
9486 btrfs_free_path(path);
9487 return ret;
9488 }
9489
9490 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
9491 {
9492 u64 num_devices;
9493 u64 stripped;
9494
9495 /*
9496 * if restripe for this chunk_type is on pick target profile and
9497 * return, otherwise do the usual balance
9498 */
9499 stripped = get_restripe_target(root->fs_info, flags);
9500 if (stripped)
9501 return extended_to_chunk(stripped);
9502
9503 num_devices = root->fs_info->fs_devices->rw_devices;
9504
9505 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9506 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9507 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9508
9509 if (num_devices == 1) {
9510 stripped |= BTRFS_BLOCK_GROUP_DUP;
9511 stripped = flags & ~stripped;
9512
9513 /* turn raid0 into single device chunks */
9514 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9515 return stripped;
9516
9517 /* turn mirroring into duplication */
9518 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9519 BTRFS_BLOCK_GROUP_RAID10))
9520 return stripped | BTRFS_BLOCK_GROUP_DUP;
9521 } else {
9522 /* they already had raid on here, just return */
9523 if (flags & stripped)
9524 return flags;
9525
9526 stripped |= BTRFS_BLOCK_GROUP_DUP;
9527 stripped = flags & ~stripped;
9528
9529 /* switch duplicated blocks with raid1 */
9530 if (flags & BTRFS_BLOCK_GROUP_DUP)
9531 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9532
9533 /* this is drive concat, leave it alone */
9534 }
9535
9536 return flags;
9537 }
9538
9539 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9540 {
9541 struct btrfs_space_info *sinfo = cache->space_info;
9542 u64 num_bytes;
9543 u64 min_allocable_bytes;
9544 int ret = -ENOSPC;
9545
9546 /*
9547 * We need some metadata space and system metadata space for
9548 * allocating chunks in some corner cases until we force to set
9549 * it to be readonly.
9550 */
9551 if ((sinfo->flags &
9552 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9553 !force)
9554 min_allocable_bytes = SZ_1M;
9555 else
9556 min_allocable_bytes = 0;
9557
9558 spin_lock(&sinfo->lock);
9559 spin_lock(&cache->lock);
9560
9561 if (cache->ro) {
9562 cache->ro++;
9563 ret = 0;
9564 goto out;
9565 }
9566
9567 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9568 cache->bytes_super - btrfs_block_group_used(&cache->item);
9569
9570 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
9571 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
9572 min_allocable_bytes <= sinfo->total_bytes) {
9573 sinfo->bytes_readonly += num_bytes;
9574 cache->ro++;
9575 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9576 ret = 0;
9577 }
9578 out:
9579 spin_unlock(&cache->lock);
9580 spin_unlock(&sinfo->lock);
9581 return ret;
9582 }
9583
9584 int btrfs_inc_block_group_ro(struct btrfs_root *root,
9585 struct btrfs_block_group_cache *cache)
9586
9587 {
9588 struct btrfs_trans_handle *trans;
9589 u64 alloc_flags;
9590 int ret;
9591
9592 again:
9593 trans = btrfs_join_transaction(root);
9594 if (IS_ERR(trans))
9595 return PTR_ERR(trans);
9596
9597 /*
9598 * we're not allowed to set block groups readonly after the dirty
9599 * block groups cache has started writing. If it already started,
9600 * back off and let this transaction commit
9601 */
9602 mutex_lock(&root->fs_info->ro_block_group_mutex);
9603 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9604 u64 transid = trans->transid;
9605
9606 mutex_unlock(&root->fs_info->ro_block_group_mutex);
9607 btrfs_end_transaction(trans, root);
9608
9609 ret = btrfs_wait_for_commit(root, transid);
9610 if (ret)
9611 return ret;
9612 goto again;
9613 }
9614
9615 /*
9616 * if we are changing raid levels, try to allocate a corresponding
9617 * block group with the new raid level.
9618 */
9619 alloc_flags = update_block_group_flags(root, cache->flags);
9620 if (alloc_flags != cache->flags) {
9621 ret = do_chunk_alloc(trans, root, alloc_flags,
9622 CHUNK_ALLOC_FORCE);
9623 /*
9624 * ENOSPC is allowed here, we may have enough space
9625 * already allocated at the new raid level to
9626 * carry on
9627 */
9628 if (ret == -ENOSPC)
9629 ret = 0;
9630 if (ret < 0)
9631 goto out;
9632 }
9633
9634 ret = inc_block_group_ro(cache, 0);
9635 if (!ret)
9636 goto out;
9637 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
9638 ret = do_chunk_alloc(trans, root, alloc_flags,
9639 CHUNK_ALLOC_FORCE);
9640 if (ret < 0)
9641 goto out;
9642 ret = inc_block_group_ro(cache, 0);
9643 out:
9644 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9645 alloc_flags = update_block_group_flags(root, cache->flags);
9646 lock_chunks(root->fs_info->chunk_root);
9647 check_system_chunk(trans, root, alloc_flags);
9648 unlock_chunks(root->fs_info->chunk_root);
9649 }
9650 mutex_unlock(&root->fs_info->ro_block_group_mutex);
9651
9652 btrfs_end_transaction(trans, root);
9653 return ret;
9654 }
9655
9656 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
9657 struct btrfs_root *root, u64 type)
9658 {
9659 u64 alloc_flags = get_alloc_profile(root, type);
9660 return do_chunk_alloc(trans, root, alloc_flags,
9661 CHUNK_ALLOC_FORCE);
9662 }
9663
9664 /*
9665 * helper to account the unused space of all the readonly block group in the
9666 * space_info. takes mirrors into account.
9667 */
9668 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9669 {
9670 struct btrfs_block_group_cache *block_group;
9671 u64 free_bytes = 0;
9672 int factor;
9673
9674 /* It's df, we don't care if it's racy */
9675 if (list_empty(&sinfo->ro_bgs))
9676 return 0;
9677
9678 spin_lock(&sinfo->lock);
9679 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9680 spin_lock(&block_group->lock);
9681
9682 if (!block_group->ro) {
9683 spin_unlock(&block_group->lock);
9684 continue;
9685 }
9686
9687 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
9688 BTRFS_BLOCK_GROUP_RAID10 |
9689 BTRFS_BLOCK_GROUP_DUP))
9690 factor = 2;
9691 else
9692 factor = 1;
9693
9694 free_bytes += (block_group->key.offset -
9695 btrfs_block_group_used(&block_group->item)) *
9696 factor;
9697
9698 spin_unlock(&block_group->lock);
9699 }
9700 spin_unlock(&sinfo->lock);
9701
9702 return free_bytes;
9703 }
9704
9705 void btrfs_dec_block_group_ro(struct btrfs_root *root,
9706 struct btrfs_block_group_cache *cache)
9707 {
9708 struct btrfs_space_info *sinfo = cache->space_info;
9709 u64 num_bytes;
9710
9711 BUG_ON(!cache->ro);
9712
9713 spin_lock(&sinfo->lock);
9714 spin_lock(&cache->lock);
9715 if (!--cache->ro) {
9716 num_bytes = cache->key.offset - cache->reserved -
9717 cache->pinned - cache->bytes_super -
9718 btrfs_block_group_used(&cache->item);
9719 sinfo->bytes_readonly -= num_bytes;
9720 list_del_init(&cache->ro_list);
9721 }
9722 spin_unlock(&cache->lock);
9723 spin_unlock(&sinfo->lock);
9724 }
9725
9726 /*
9727 * checks to see if its even possible to relocate this block group.
9728 *
9729 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9730 * ok to go ahead and try.
9731 */
9732 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
9733 {
9734 struct btrfs_block_group_cache *block_group;
9735 struct btrfs_space_info *space_info;
9736 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
9737 struct btrfs_device *device;
9738 struct btrfs_trans_handle *trans;
9739 u64 min_free;
9740 u64 dev_min = 1;
9741 u64 dev_nr = 0;
9742 u64 target;
9743 int debug;
9744 int index;
9745 int full = 0;
9746 int ret = 0;
9747
9748 debug = btrfs_test_opt(root, ENOSPC_DEBUG);
9749
9750 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
9751
9752 /* odd, couldn't find the block group, leave it alone */
9753 if (!block_group) {
9754 if (debug)
9755 btrfs_warn(root->fs_info,
9756 "can't find block group for bytenr %llu",
9757 bytenr);
9758 return -1;
9759 }
9760
9761 min_free = btrfs_block_group_used(&block_group->item);
9762
9763 /* no bytes used, we're good */
9764 if (!min_free)
9765 goto out;
9766
9767 space_info = block_group->space_info;
9768 spin_lock(&space_info->lock);
9769
9770 full = space_info->full;
9771
9772 /*
9773 * if this is the last block group we have in this space, we can't
9774 * relocate it unless we're able to allocate a new chunk below.
9775 *
9776 * Otherwise, we need to make sure we have room in the space to handle
9777 * all of the extents from this block group. If we can, we're good
9778 */
9779 if ((space_info->total_bytes != block_group->key.offset) &&
9780 (space_info->bytes_used + space_info->bytes_reserved +
9781 space_info->bytes_pinned + space_info->bytes_readonly +
9782 min_free < space_info->total_bytes)) {
9783 spin_unlock(&space_info->lock);
9784 goto out;
9785 }
9786 spin_unlock(&space_info->lock);
9787
9788 /*
9789 * ok we don't have enough space, but maybe we have free space on our
9790 * devices to allocate new chunks for relocation, so loop through our
9791 * alloc devices and guess if we have enough space. if this block
9792 * group is going to be restriped, run checks against the target
9793 * profile instead of the current one.
9794 */
9795 ret = -1;
9796
9797 /*
9798 * index:
9799 * 0: raid10
9800 * 1: raid1
9801 * 2: dup
9802 * 3: raid0
9803 * 4: single
9804 */
9805 target = get_restripe_target(root->fs_info, block_group->flags);
9806 if (target) {
9807 index = __get_raid_index(extended_to_chunk(target));
9808 } else {
9809 /*
9810 * this is just a balance, so if we were marked as full
9811 * we know there is no space for a new chunk
9812 */
9813 if (full) {
9814 if (debug)
9815 btrfs_warn(root->fs_info,
9816 "no space to alloc new chunk for block group %llu",
9817 block_group->key.objectid);
9818 goto out;
9819 }
9820
9821 index = get_block_group_index(block_group);
9822 }
9823
9824 if (index == BTRFS_RAID_RAID10) {
9825 dev_min = 4;
9826 /* Divide by 2 */
9827 min_free >>= 1;
9828 } else if (index == BTRFS_RAID_RAID1) {
9829 dev_min = 2;
9830 } else if (index == BTRFS_RAID_DUP) {
9831 /* Multiply by 2 */
9832 min_free <<= 1;
9833 } else if (index == BTRFS_RAID_RAID0) {
9834 dev_min = fs_devices->rw_devices;
9835 min_free = div64_u64(min_free, dev_min);
9836 }
9837
9838 /* We need to do this so that we can look at pending chunks */
9839 trans = btrfs_join_transaction(root);
9840 if (IS_ERR(trans)) {
9841 ret = PTR_ERR(trans);
9842 goto out;
9843 }
9844
9845 mutex_lock(&root->fs_info->chunk_mutex);
9846 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9847 u64 dev_offset;
9848
9849 /*
9850 * check to make sure we can actually find a chunk with enough
9851 * space to fit our block group in.
9852 */
9853 if (device->total_bytes > device->bytes_used + min_free &&
9854 !device->is_tgtdev_for_dev_replace) {
9855 ret = find_free_dev_extent(trans, device, min_free,
9856 &dev_offset, NULL);
9857 if (!ret)
9858 dev_nr++;
9859
9860 if (dev_nr >= dev_min)
9861 break;
9862
9863 ret = -1;
9864 }
9865 }
9866 if (debug && ret == -1)
9867 btrfs_warn(root->fs_info,
9868 "no space to allocate a new chunk for block group %llu",
9869 block_group->key.objectid);
9870 mutex_unlock(&root->fs_info->chunk_mutex);
9871 btrfs_end_transaction(trans, root);
9872 out:
9873 btrfs_put_block_group(block_group);
9874 return ret;
9875 }
9876
9877 static int find_first_block_group(struct btrfs_root *root,
9878 struct btrfs_path *path, struct btrfs_key *key)
9879 {
9880 int ret = 0;
9881 struct btrfs_key found_key;
9882 struct extent_buffer *leaf;
9883 int slot;
9884
9885 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9886 if (ret < 0)
9887 goto out;
9888
9889 while (1) {
9890 slot = path->slots[0];
9891 leaf = path->nodes[0];
9892 if (slot >= btrfs_header_nritems(leaf)) {
9893 ret = btrfs_next_leaf(root, path);
9894 if (ret == 0)
9895 continue;
9896 if (ret < 0)
9897 goto out;
9898 break;
9899 }
9900 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9901
9902 if (found_key.objectid >= key->objectid &&
9903 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9904 ret = 0;
9905 goto out;
9906 }
9907 path->slots[0]++;
9908 }
9909 out:
9910 return ret;
9911 }
9912
9913 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9914 {
9915 struct btrfs_block_group_cache *block_group;
9916 u64 last = 0;
9917
9918 while (1) {
9919 struct inode *inode;
9920
9921 block_group = btrfs_lookup_first_block_group(info, last);
9922 while (block_group) {
9923 spin_lock(&block_group->lock);
9924 if (block_group->iref)
9925 break;
9926 spin_unlock(&block_group->lock);
9927 block_group = next_block_group(info->tree_root,
9928 block_group);
9929 }
9930 if (!block_group) {
9931 if (last == 0)
9932 break;
9933 last = 0;
9934 continue;
9935 }
9936
9937 inode = block_group->inode;
9938 block_group->iref = 0;
9939 block_group->inode = NULL;
9940 spin_unlock(&block_group->lock);
9941 iput(inode);
9942 last = block_group->key.objectid + block_group->key.offset;
9943 btrfs_put_block_group(block_group);
9944 }
9945 }
9946
9947 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9948 {
9949 struct btrfs_block_group_cache *block_group;
9950 struct btrfs_space_info *space_info;
9951 struct btrfs_caching_control *caching_ctl;
9952 struct rb_node *n;
9953
9954 down_write(&info->commit_root_sem);
9955 while (!list_empty(&info->caching_block_groups)) {
9956 caching_ctl = list_entry(info->caching_block_groups.next,
9957 struct btrfs_caching_control, list);
9958 list_del(&caching_ctl->list);
9959 put_caching_control(caching_ctl);
9960 }
9961 up_write(&info->commit_root_sem);
9962
9963 spin_lock(&info->unused_bgs_lock);
9964 while (!list_empty(&info->unused_bgs)) {
9965 block_group = list_first_entry(&info->unused_bgs,
9966 struct btrfs_block_group_cache,
9967 bg_list);
9968 list_del_init(&block_group->bg_list);
9969 btrfs_put_block_group(block_group);
9970 }
9971 spin_unlock(&info->unused_bgs_lock);
9972
9973 spin_lock(&info->block_group_cache_lock);
9974 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9975 block_group = rb_entry(n, struct btrfs_block_group_cache,
9976 cache_node);
9977 rb_erase(&block_group->cache_node,
9978 &info->block_group_cache_tree);
9979 RB_CLEAR_NODE(&block_group->cache_node);
9980 spin_unlock(&info->block_group_cache_lock);
9981
9982 down_write(&block_group->space_info->groups_sem);
9983 list_del(&block_group->list);
9984 up_write(&block_group->space_info->groups_sem);
9985
9986 if (block_group->cached == BTRFS_CACHE_STARTED)
9987 wait_block_group_cache_done(block_group);
9988
9989 /*
9990 * We haven't cached this block group, which means we could
9991 * possibly have excluded extents on this block group.
9992 */
9993 if (block_group->cached == BTRFS_CACHE_NO ||
9994 block_group->cached == BTRFS_CACHE_ERROR)
9995 free_excluded_extents(info->extent_root, block_group);
9996
9997 btrfs_remove_free_space_cache(block_group);
9998 btrfs_put_block_group(block_group);
9999
10000 spin_lock(&info->block_group_cache_lock);
10001 }
10002 spin_unlock(&info->block_group_cache_lock);
10003
10004 /* now that all the block groups are freed, go through and
10005 * free all the space_info structs. This is only called during
10006 * the final stages of unmount, and so we know nobody is
10007 * using them. We call synchronize_rcu() once before we start,
10008 * just to be on the safe side.
10009 */
10010 synchronize_rcu();
10011
10012 release_global_block_rsv(info);
10013
10014 while (!list_empty(&info->space_info)) {
10015 int i;
10016
10017 space_info = list_entry(info->space_info.next,
10018 struct btrfs_space_info,
10019 list);
10020
10021 /*
10022 * Do not hide this behind enospc_debug, this is actually
10023 * important and indicates a real bug if this happens.
10024 */
10025 if (WARN_ON(space_info->bytes_pinned > 0 ||
10026 space_info->bytes_reserved > 0 ||
10027 space_info->bytes_may_use > 0))
10028 dump_space_info(space_info, 0, 0);
10029 list_del(&space_info->list);
10030 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
10031 struct kobject *kobj;
10032 kobj = space_info->block_group_kobjs[i];
10033 space_info->block_group_kobjs[i] = NULL;
10034 if (kobj) {
10035 kobject_del(kobj);
10036 kobject_put(kobj);
10037 }
10038 }
10039 kobject_del(&space_info->kobj);
10040 kobject_put(&space_info->kobj);
10041 }
10042 return 0;
10043 }
10044
10045 static void __link_block_group(struct btrfs_space_info *space_info,
10046 struct btrfs_block_group_cache *cache)
10047 {
10048 int index = get_block_group_index(cache);
10049 bool first = false;
10050
10051 down_write(&space_info->groups_sem);
10052 if (list_empty(&space_info->block_groups[index]))
10053 first = true;
10054 list_add_tail(&cache->list, &space_info->block_groups[index]);
10055 up_write(&space_info->groups_sem);
10056
10057 if (first) {
10058 struct raid_kobject *rkobj;
10059 int ret;
10060
10061 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
10062 if (!rkobj)
10063 goto out_err;
10064 rkobj->raid_type = index;
10065 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
10066 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
10067 "%s", get_raid_name(index));
10068 if (ret) {
10069 kobject_put(&rkobj->kobj);
10070 goto out_err;
10071 }
10072 space_info->block_group_kobjs[index] = &rkobj->kobj;
10073 }
10074
10075 return;
10076 out_err:
10077 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
10078 }
10079
10080 static struct btrfs_block_group_cache *
10081 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
10082 {
10083 struct btrfs_block_group_cache *cache;
10084
10085 cache = kzalloc(sizeof(*cache), GFP_NOFS);
10086 if (!cache)
10087 return NULL;
10088
10089 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
10090 GFP_NOFS);
10091 if (!cache->free_space_ctl) {
10092 kfree(cache);
10093 return NULL;
10094 }
10095
10096 cache->key.objectid = start;
10097 cache->key.offset = size;
10098 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10099
10100 cache->sectorsize = root->sectorsize;
10101 cache->fs_info = root->fs_info;
10102 cache->full_stripe_len = btrfs_full_stripe_len(root,
10103 &root->fs_info->mapping_tree,
10104 start);
10105 set_free_space_tree_thresholds(cache);
10106
10107 atomic_set(&cache->count, 1);
10108 spin_lock_init(&cache->lock);
10109 init_rwsem(&cache->data_rwsem);
10110 INIT_LIST_HEAD(&cache->list);
10111 INIT_LIST_HEAD(&cache->cluster_list);
10112 INIT_LIST_HEAD(&cache->bg_list);
10113 INIT_LIST_HEAD(&cache->ro_list);
10114 INIT_LIST_HEAD(&cache->dirty_list);
10115 INIT_LIST_HEAD(&cache->io_list);
10116 btrfs_init_free_space_ctl(cache);
10117 atomic_set(&cache->trimming, 0);
10118 mutex_init(&cache->free_space_lock);
10119
10120 return cache;
10121 }
10122
10123 int btrfs_read_block_groups(struct btrfs_root *root)
10124 {
10125 struct btrfs_path *path;
10126 int ret;
10127 struct btrfs_block_group_cache *cache;
10128 struct btrfs_fs_info *info = root->fs_info;
10129 struct btrfs_space_info *space_info;
10130 struct btrfs_key key;
10131 struct btrfs_key found_key;
10132 struct extent_buffer *leaf;
10133 int need_clear = 0;
10134 u64 cache_gen;
10135
10136 root = info->extent_root;
10137 key.objectid = 0;
10138 key.offset = 0;
10139 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10140 path = btrfs_alloc_path();
10141 if (!path)
10142 return -ENOMEM;
10143 path->reada = READA_FORWARD;
10144
10145 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
10146 if (btrfs_test_opt(root, SPACE_CACHE) &&
10147 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
10148 need_clear = 1;
10149 if (btrfs_test_opt(root, CLEAR_CACHE))
10150 need_clear = 1;
10151
10152 while (1) {
10153 ret = find_first_block_group(root, path, &key);
10154 if (ret > 0)
10155 break;
10156 if (ret != 0)
10157 goto error;
10158
10159 leaf = path->nodes[0];
10160 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
10161
10162 cache = btrfs_create_block_group_cache(root, found_key.objectid,
10163 found_key.offset);
10164 if (!cache) {
10165 ret = -ENOMEM;
10166 goto error;
10167 }
10168
10169 if (need_clear) {
10170 /*
10171 * When we mount with old space cache, we need to
10172 * set BTRFS_DC_CLEAR and set dirty flag.
10173 *
10174 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10175 * truncate the old free space cache inode and
10176 * setup a new one.
10177 * b) Setting 'dirty flag' makes sure that we flush
10178 * the new space cache info onto disk.
10179 */
10180 if (btrfs_test_opt(root, SPACE_CACHE))
10181 cache->disk_cache_state = BTRFS_DC_CLEAR;
10182 }
10183
10184 read_extent_buffer(leaf, &cache->item,
10185 btrfs_item_ptr_offset(leaf, path->slots[0]),
10186 sizeof(cache->item));
10187 cache->flags = btrfs_block_group_flags(&cache->item);
10188
10189 key.objectid = found_key.objectid + found_key.offset;
10190 btrfs_release_path(path);
10191
10192 /*
10193 * We need to exclude the super stripes now so that the space
10194 * info has super bytes accounted for, otherwise we'll think
10195 * we have more space than we actually do.
10196 */
10197 ret = exclude_super_stripes(root, cache);
10198 if (ret) {
10199 /*
10200 * We may have excluded something, so call this just in
10201 * case.
10202 */
10203 free_excluded_extents(root, cache);
10204 btrfs_put_block_group(cache);
10205 goto error;
10206 }
10207
10208 /*
10209 * check for two cases, either we are full, and therefore
10210 * don't need to bother with the caching work since we won't
10211 * find any space, or we are empty, and we can just add all
10212 * the space in and be done with it. This saves us _alot_ of
10213 * time, particularly in the full case.
10214 */
10215 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
10216 cache->last_byte_to_unpin = (u64)-1;
10217 cache->cached = BTRFS_CACHE_FINISHED;
10218 free_excluded_extents(root, cache);
10219 } else if (btrfs_block_group_used(&cache->item) == 0) {
10220 cache->last_byte_to_unpin = (u64)-1;
10221 cache->cached = BTRFS_CACHE_FINISHED;
10222 add_new_free_space(cache, root->fs_info,
10223 found_key.objectid,
10224 found_key.objectid +
10225 found_key.offset);
10226 free_excluded_extents(root, cache);
10227 }
10228
10229 ret = btrfs_add_block_group_cache(root->fs_info, cache);
10230 if (ret) {
10231 btrfs_remove_free_space_cache(cache);
10232 btrfs_put_block_group(cache);
10233 goto error;
10234 }
10235
10236 trace_btrfs_add_block_group(root->fs_info, cache, 0);
10237 ret = update_space_info(info, cache->flags, found_key.offset,
10238 btrfs_block_group_used(&cache->item),
10239 cache->bytes_super, &space_info);
10240 if (ret) {
10241 btrfs_remove_free_space_cache(cache);
10242 spin_lock(&info->block_group_cache_lock);
10243 rb_erase(&cache->cache_node,
10244 &info->block_group_cache_tree);
10245 RB_CLEAR_NODE(&cache->cache_node);
10246 spin_unlock(&info->block_group_cache_lock);
10247 btrfs_put_block_group(cache);
10248 goto error;
10249 }
10250
10251 cache->space_info = space_info;
10252
10253 __link_block_group(space_info, cache);
10254
10255 set_avail_alloc_bits(root->fs_info, cache->flags);
10256 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
10257 inc_block_group_ro(cache, 1);
10258 } else if (btrfs_block_group_used(&cache->item) == 0) {
10259 spin_lock(&info->unused_bgs_lock);
10260 /* Should always be true but just in case. */
10261 if (list_empty(&cache->bg_list)) {
10262 btrfs_get_block_group(cache);
10263 list_add_tail(&cache->bg_list,
10264 &info->unused_bgs);
10265 }
10266 spin_unlock(&info->unused_bgs_lock);
10267 }
10268 }
10269
10270 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
10271 if (!(get_alloc_profile(root, space_info->flags) &
10272 (BTRFS_BLOCK_GROUP_RAID10 |
10273 BTRFS_BLOCK_GROUP_RAID1 |
10274 BTRFS_BLOCK_GROUP_RAID5 |
10275 BTRFS_BLOCK_GROUP_RAID6 |
10276 BTRFS_BLOCK_GROUP_DUP)))
10277 continue;
10278 /*
10279 * avoid allocating from un-mirrored block group if there are
10280 * mirrored block groups.
10281 */
10282 list_for_each_entry(cache,
10283 &space_info->block_groups[BTRFS_RAID_RAID0],
10284 list)
10285 inc_block_group_ro(cache, 1);
10286 list_for_each_entry(cache,
10287 &space_info->block_groups[BTRFS_RAID_SINGLE],
10288 list)
10289 inc_block_group_ro(cache, 1);
10290 }
10291
10292 init_global_block_rsv(info);
10293 ret = 0;
10294 error:
10295 btrfs_free_path(path);
10296 return ret;
10297 }
10298
10299 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
10300 struct btrfs_root *root)
10301 {
10302 struct btrfs_block_group_cache *block_group, *tmp;
10303 struct btrfs_root *extent_root = root->fs_info->extent_root;
10304 struct btrfs_block_group_item item;
10305 struct btrfs_key key;
10306 int ret = 0;
10307 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
10308
10309 trans->can_flush_pending_bgs = false;
10310 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
10311 if (ret)
10312 goto next;
10313
10314 spin_lock(&block_group->lock);
10315 memcpy(&item, &block_group->item, sizeof(item));
10316 memcpy(&key, &block_group->key, sizeof(key));
10317 spin_unlock(&block_group->lock);
10318
10319 ret = btrfs_insert_item(trans, extent_root, &key, &item,
10320 sizeof(item));
10321 if (ret)
10322 btrfs_abort_transaction(trans, extent_root, ret);
10323 ret = btrfs_finish_chunk_alloc(trans, extent_root,
10324 key.objectid, key.offset);
10325 if (ret)
10326 btrfs_abort_transaction(trans, extent_root, ret);
10327 add_block_group_free_space(trans, root->fs_info, block_group);
10328 /* already aborted the transaction if it failed. */
10329 next:
10330 list_del_init(&block_group->bg_list);
10331 }
10332 trans->can_flush_pending_bgs = can_flush_pending_bgs;
10333 }
10334
10335 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
10336 struct btrfs_root *root, u64 bytes_used,
10337 u64 type, u64 chunk_objectid, u64 chunk_offset,
10338 u64 size)
10339 {
10340 int ret;
10341 struct btrfs_root *extent_root;
10342 struct btrfs_block_group_cache *cache;
10343 extent_root = root->fs_info->extent_root;
10344
10345 btrfs_set_log_full_commit(root->fs_info, trans);
10346
10347 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
10348 if (!cache)
10349 return -ENOMEM;
10350
10351 btrfs_set_block_group_used(&cache->item, bytes_used);
10352 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
10353 btrfs_set_block_group_flags(&cache->item, type);
10354
10355 cache->flags = type;
10356 cache->last_byte_to_unpin = (u64)-1;
10357 cache->cached = BTRFS_CACHE_FINISHED;
10358 cache->needs_free_space = 1;
10359 ret = exclude_super_stripes(root, cache);
10360 if (ret) {
10361 /*
10362 * We may have excluded something, so call this just in
10363 * case.
10364 */
10365 free_excluded_extents(root, cache);
10366 btrfs_put_block_group(cache);
10367 return ret;
10368 }
10369
10370 add_new_free_space(cache, root->fs_info, chunk_offset,
10371 chunk_offset + size);
10372
10373 free_excluded_extents(root, cache);
10374
10375 #ifdef CONFIG_BTRFS_DEBUG
10376 if (btrfs_should_fragment_free_space(root, cache)) {
10377 u64 new_bytes_used = size - bytes_used;
10378
10379 bytes_used += new_bytes_used >> 1;
10380 fragment_free_space(root, cache);
10381 }
10382 #endif
10383 /*
10384 * Call to ensure the corresponding space_info object is created and
10385 * assigned to our block group, but don't update its counters just yet.
10386 * We want our bg to be added to the rbtree with its ->space_info set.
10387 */
10388 ret = update_space_info(root->fs_info, cache->flags, 0, 0, 0,
10389 &cache->space_info);
10390 if (ret) {
10391 btrfs_remove_free_space_cache(cache);
10392 btrfs_put_block_group(cache);
10393 return ret;
10394 }
10395
10396 ret = btrfs_add_block_group_cache(root->fs_info, cache);
10397 if (ret) {
10398 btrfs_remove_free_space_cache(cache);
10399 btrfs_put_block_group(cache);
10400 return ret;
10401 }
10402
10403 /*
10404 * Now that our block group has its ->space_info set and is inserted in
10405 * the rbtree, update the space info's counters.
10406 */
10407 trace_btrfs_add_block_group(root->fs_info, cache, 1);
10408 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
10409 cache->bytes_super, &cache->space_info);
10410 if (ret) {
10411 btrfs_remove_free_space_cache(cache);
10412 spin_lock(&root->fs_info->block_group_cache_lock);
10413 rb_erase(&cache->cache_node,
10414 &root->fs_info->block_group_cache_tree);
10415 RB_CLEAR_NODE(&cache->cache_node);
10416 spin_unlock(&root->fs_info->block_group_cache_lock);
10417 btrfs_put_block_group(cache);
10418 return ret;
10419 }
10420 update_global_block_rsv(root->fs_info);
10421
10422 __link_block_group(cache->space_info, cache);
10423
10424 list_add_tail(&cache->bg_list, &trans->new_bgs);
10425
10426 set_avail_alloc_bits(extent_root->fs_info, type);
10427 return 0;
10428 }
10429
10430 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10431 {
10432 u64 extra_flags = chunk_to_extended(flags) &
10433 BTRFS_EXTENDED_PROFILE_MASK;
10434
10435 write_seqlock(&fs_info->profiles_lock);
10436 if (flags & BTRFS_BLOCK_GROUP_DATA)
10437 fs_info->avail_data_alloc_bits &= ~extra_flags;
10438 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10439 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10440 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10441 fs_info->avail_system_alloc_bits &= ~extra_flags;
10442 write_sequnlock(&fs_info->profiles_lock);
10443 }
10444
10445 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10446 struct btrfs_root *root, u64 group_start,
10447 struct extent_map *em)
10448 {
10449 struct btrfs_path *path;
10450 struct btrfs_block_group_cache *block_group;
10451 struct btrfs_free_cluster *cluster;
10452 struct btrfs_root *tree_root = root->fs_info->tree_root;
10453 struct btrfs_key key;
10454 struct inode *inode;
10455 struct kobject *kobj = NULL;
10456 int ret;
10457 int index;
10458 int factor;
10459 struct btrfs_caching_control *caching_ctl = NULL;
10460 bool remove_em;
10461
10462 root = root->fs_info->extent_root;
10463
10464 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
10465 BUG_ON(!block_group);
10466 BUG_ON(!block_group->ro);
10467
10468 /*
10469 * Free the reserved super bytes from this block group before
10470 * remove it.
10471 */
10472 free_excluded_extents(root, block_group);
10473
10474 memcpy(&key, &block_group->key, sizeof(key));
10475 index = get_block_group_index(block_group);
10476 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
10477 BTRFS_BLOCK_GROUP_RAID1 |
10478 BTRFS_BLOCK_GROUP_RAID10))
10479 factor = 2;
10480 else
10481 factor = 1;
10482
10483 /* make sure this block group isn't part of an allocation cluster */
10484 cluster = &root->fs_info->data_alloc_cluster;
10485 spin_lock(&cluster->refill_lock);
10486 btrfs_return_cluster_to_free_space(block_group, cluster);
10487 spin_unlock(&cluster->refill_lock);
10488
10489 /*
10490 * make sure this block group isn't part of a metadata
10491 * allocation cluster
10492 */
10493 cluster = &root->fs_info->meta_alloc_cluster;
10494 spin_lock(&cluster->refill_lock);
10495 btrfs_return_cluster_to_free_space(block_group, cluster);
10496 spin_unlock(&cluster->refill_lock);
10497
10498 path = btrfs_alloc_path();
10499 if (!path) {
10500 ret = -ENOMEM;
10501 goto out;
10502 }
10503
10504 /*
10505 * get the inode first so any iput calls done for the io_list
10506 * aren't the final iput (no unlinks allowed now)
10507 */
10508 inode = lookup_free_space_inode(tree_root, block_group, path);
10509
10510 mutex_lock(&trans->transaction->cache_write_mutex);
10511 /*
10512 * make sure our free spache cache IO is done before remove the
10513 * free space inode
10514 */
10515 spin_lock(&trans->transaction->dirty_bgs_lock);
10516 if (!list_empty(&block_group->io_list)) {
10517 list_del_init(&block_group->io_list);
10518
10519 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10520
10521 spin_unlock(&trans->transaction->dirty_bgs_lock);
10522 btrfs_wait_cache_io(root, trans, block_group,
10523 &block_group->io_ctl, path,
10524 block_group->key.objectid);
10525 btrfs_put_block_group(block_group);
10526 spin_lock(&trans->transaction->dirty_bgs_lock);
10527 }
10528
10529 if (!list_empty(&block_group->dirty_list)) {
10530 list_del_init(&block_group->dirty_list);
10531 btrfs_put_block_group(block_group);
10532 }
10533 spin_unlock(&trans->transaction->dirty_bgs_lock);
10534 mutex_unlock(&trans->transaction->cache_write_mutex);
10535
10536 if (!IS_ERR(inode)) {
10537 ret = btrfs_orphan_add(trans, inode);
10538 if (ret) {
10539 btrfs_add_delayed_iput(inode);
10540 goto out;
10541 }
10542 clear_nlink(inode);
10543 /* One for the block groups ref */
10544 spin_lock(&block_group->lock);
10545 if (block_group->iref) {
10546 block_group->iref = 0;
10547 block_group->inode = NULL;
10548 spin_unlock(&block_group->lock);
10549 iput(inode);
10550 } else {
10551 spin_unlock(&block_group->lock);
10552 }
10553 /* One for our lookup ref */
10554 btrfs_add_delayed_iput(inode);
10555 }
10556
10557 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10558 key.offset = block_group->key.objectid;
10559 key.type = 0;
10560
10561 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10562 if (ret < 0)
10563 goto out;
10564 if (ret > 0)
10565 btrfs_release_path(path);
10566 if (ret == 0) {
10567 ret = btrfs_del_item(trans, tree_root, path);
10568 if (ret)
10569 goto out;
10570 btrfs_release_path(path);
10571 }
10572
10573 spin_lock(&root->fs_info->block_group_cache_lock);
10574 rb_erase(&block_group->cache_node,
10575 &root->fs_info->block_group_cache_tree);
10576 RB_CLEAR_NODE(&block_group->cache_node);
10577
10578 if (root->fs_info->first_logical_byte == block_group->key.objectid)
10579 root->fs_info->first_logical_byte = (u64)-1;
10580 spin_unlock(&root->fs_info->block_group_cache_lock);
10581
10582 down_write(&block_group->space_info->groups_sem);
10583 /*
10584 * we must use list_del_init so people can check to see if they
10585 * are still on the list after taking the semaphore
10586 */
10587 list_del_init(&block_group->list);
10588 if (list_empty(&block_group->space_info->block_groups[index])) {
10589 kobj = block_group->space_info->block_group_kobjs[index];
10590 block_group->space_info->block_group_kobjs[index] = NULL;
10591 clear_avail_alloc_bits(root->fs_info, block_group->flags);
10592 }
10593 up_write(&block_group->space_info->groups_sem);
10594 if (kobj) {
10595 kobject_del(kobj);
10596 kobject_put(kobj);
10597 }
10598
10599 if (block_group->has_caching_ctl)
10600 caching_ctl = get_caching_control(block_group);
10601 if (block_group->cached == BTRFS_CACHE_STARTED)
10602 wait_block_group_cache_done(block_group);
10603 if (block_group->has_caching_ctl) {
10604 down_write(&root->fs_info->commit_root_sem);
10605 if (!caching_ctl) {
10606 struct btrfs_caching_control *ctl;
10607
10608 list_for_each_entry(ctl,
10609 &root->fs_info->caching_block_groups, list)
10610 if (ctl->block_group == block_group) {
10611 caching_ctl = ctl;
10612 atomic_inc(&caching_ctl->count);
10613 break;
10614 }
10615 }
10616 if (caching_ctl)
10617 list_del_init(&caching_ctl->list);
10618 up_write(&root->fs_info->commit_root_sem);
10619 if (caching_ctl) {
10620 /* Once for the caching bgs list and once for us. */
10621 put_caching_control(caching_ctl);
10622 put_caching_control(caching_ctl);
10623 }
10624 }
10625
10626 spin_lock(&trans->transaction->dirty_bgs_lock);
10627 if (!list_empty(&block_group->dirty_list)) {
10628 WARN_ON(1);
10629 }
10630 if (!list_empty(&block_group->io_list)) {
10631 WARN_ON(1);
10632 }
10633 spin_unlock(&trans->transaction->dirty_bgs_lock);
10634 btrfs_remove_free_space_cache(block_group);
10635
10636 spin_lock(&block_group->space_info->lock);
10637 list_del_init(&block_group->ro_list);
10638
10639 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
10640 WARN_ON(block_group->space_info->total_bytes
10641 < block_group->key.offset);
10642 WARN_ON(block_group->space_info->bytes_readonly
10643 < block_group->key.offset);
10644 WARN_ON(block_group->space_info->disk_total
10645 < block_group->key.offset * factor);
10646 }
10647 block_group->space_info->total_bytes -= block_group->key.offset;
10648 block_group->space_info->bytes_readonly -= block_group->key.offset;
10649 block_group->space_info->disk_total -= block_group->key.offset * factor;
10650
10651 spin_unlock(&block_group->space_info->lock);
10652
10653 memcpy(&key, &block_group->key, sizeof(key));
10654
10655 lock_chunks(root);
10656 if (!list_empty(&em->list)) {
10657 /* We're in the transaction->pending_chunks list. */
10658 free_extent_map(em);
10659 }
10660 spin_lock(&block_group->lock);
10661 block_group->removed = 1;
10662 /*
10663 * At this point trimming can't start on this block group, because we
10664 * removed the block group from the tree fs_info->block_group_cache_tree
10665 * so no one can't find it anymore and even if someone already got this
10666 * block group before we removed it from the rbtree, they have already
10667 * incremented block_group->trimming - if they didn't, they won't find
10668 * any free space entries because we already removed them all when we
10669 * called btrfs_remove_free_space_cache().
10670 *
10671 * And we must not remove the extent map from the fs_info->mapping_tree
10672 * to prevent the same logical address range and physical device space
10673 * ranges from being reused for a new block group. This is because our
10674 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10675 * completely transactionless, so while it is trimming a range the
10676 * currently running transaction might finish and a new one start,
10677 * allowing for new block groups to be created that can reuse the same
10678 * physical device locations unless we take this special care.
10679 *
10680 * There may also be an implicit trim operation if the file system
10681 * is mounted with -odiscard. The same protections must remain
10682 * in place until the extents have been discarded completely when
10683 * the transaction commit has completed.
10684 */
10685 remove_em = (atomic_read(&block_group->trimming) == 0);
10686 /*
10687 * Make sure a trimmer task always sees the em in the pinned_chunks list
10688 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10689 * before checking block_group->removed).
10690 */
10691 if (!remove_em) {
10692 /*
10693 * Our em might be in trans->transaction->pending_chunks which
10694 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10695 * and so is the fs_info->pinned_chunks list.
10696 *
10697 * So at this point we must be holding the chunk_mutex to avoid
10698 * any races with chunk allocation (more specifically at
10699 * volumes.c:contains_pending_extent()), to ensure it always
10700 * sees the em, either in the pending_chunks list or in the
10701 * pinned_chunks list.
10702 */
10703 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
10704 }
10705 spin_unlock(&block_group->lock);
10706
10707 if (remove_em) {
10708 struct extent_map_tree *em_tree;
10709
10710 em_tree = &root->fs_info->mapping_tree.map_tree;
10711 write_lock(&em_tree->lock);
10712 /*
10713 * The em might be in the pending_chunks list, so make sure the
10714 * chunk mutex is locked, since remove_extent_mapping() will
10715 * delete us from that list.
10716 */
10717 remove_extent_mapping(em_tree, em);
10718 write_unlock(&em_tree->lock);
10719 /* once for the tree */
10720 free_extent_map(em);
10721 }
10722
10723 unlock_chunks(root);
10724
10725 ret = remove_block_group_free_space(trans, root->fs_info, block_group);
10726 if (ret)
10727 goto out;
10728
10729 btrfs_put_block_group(block_group);
10730 btrfs_put_block_group(block_group);
10731
10732 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10733 if (ret > 0)
10734 ret = -EIO;
10735 if (ret < 0)
10736 goto out;
10737
10738 ret = btrfs_del_item(trans, root, path);
10739 out:
10740 btrfs_free_path(path);
10741 return ret;
10742 }
10743
10744 struct btrfs_trans_handle *
10745 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10746 const u64 chunk_offset)
10747 {
10748 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10749 struct extent_map *em;
10750 struct map_lookup *map;
10751 unsigned int num_items;
10752
10753 read_lock(&em_tree->lock);
10754 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10755 read_unlock(&em_tree->lock);
10756 ASSERT(em && em->start == chunk_offset);
10757
10758 /*
10759 * We need to reserve 3 + N units from the metadata space info in order
10760 * to remove a block group (done at btrfs_remove_chunk() and at
10761 * btrfs_remove_block_group()), which are used for:
10762 *
10763 * 1 unit for adding the free space inode's orphan (located in the tree
10764 * of tree roots).
10765 * 1 unit for deleting the block group item (located in the extent
10766 * tree).
10767 * 1 unit for deleting the free space item (located in tree of tree
10768 * roots).
10769 * N units for deleting N device extent items corresponding to each
10770 * stripe (located in the device tree).
10771 *
10772 * In order to remove a block group we also need to reserve units in the
10773 * system space info in order to update the chunk tree (update one or
10774 * more device items and remove one chunk item), but this is done at
10775 * btrfs_remove_chunk() through a call to check_system_chunk().
10776 */
10777 map = em->map_lookup;
10778 num_items = 3 + map->num_stripes;
10779 free_extent_map(em);
10780
10781 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10782 num_items, 1);
10783 }
10784
10785 /*
10786 * Process the unused_bgs list and remove any that don't have any allocated
10787 * space inside of them.
10788 */
10789 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10790 {
10791 struct btrfs_block_group_cache *block_group;
10792 struct btrfs_space_info *space_info;
10793 struct btrfs_root *root = fs_info->extent_root;
10794 struct btrfs_trans_handle *trans;
10795 int ret = 0;
10796
10797 if (!fs_info->open)
10798 return;
10799
10800 spin_lock(&fs_info->unused_bgs_lock);
10801 while (!list_empty(&fs_info->unused_bgs)) {
10802 u64 start, end;
10803 int trimming;
10804
10805 block_group = list_first_entry(&fs_info->unused_bgs,
10806 struct btrfs_block_group_cache,
10807 bg_list);
10808 list_del_init(&block_group->bg_list);
10809
10810 space_info = block_group->space_info;
10811
10812 if (ret || btrfs_mixed_space_info(space_info)) {
10813 btrfs_put_block_group(block_group);
10814 continue;
10815 }
10816 spin_unlock(&fs_info->unused_bgs_lock);
10817
10818 mutex_lock(&fs_info->delete_unused_bgs_mutex);
10819
10820 /* Don't want to race with allocators so take the groups_sem */
10821 down_write(&space_info->groups_sem);
10822 spin_lock(&block_group->lock);
10823 if (block_group->reserved ||
10824 btrfs_block_group_used(&block_group->item) ||
10825 block_group->ro ||
10826 list_is_singular(&block_group->list)) {
10827 /*
10828 * We want to bail if we made new allocations or have
10829 * outstanding allocations in this block group. We do
10830 * the ro check in case balance is currently acting on
10831 * this block group.
10832 */
10833 spin_unlock(&block_group->lock);
10834 up_write(&space_info->groups_sem);
10835 goto next;
10836 }
10837 spin_unlock(&block_group->lock);
10838
10839 /* We don't want to force the issue, only flip if it's ok. */
10840 ret = inc_block_group_ro(block_group, 0);
10841 up_write(&space_info->groups_sem);
10842 if (ret < 0) {
10843 ret = 0;
10844 goto next;
10845 }
10846
10847 /*
10848 * Want to do this before we do anything else so we can recover
10849 * properly if we fail to join the transaction.
10850 */
10851 trans = btrfs_start_trans_remove_block_group(fs_info,
10852 block_group->key.objectid);
10853 if (IS_ERR(trans)) {
10854 btrfs_dec_block_group_ro(root, block_group);
10855 ret = PTR_ERR(trans);
10856 goto next;
10857 }
10858
10859 /*
10860 * We could have pending pinned extents for this block group,
10861 * just delete them, we don't care about them anymore.
10862 */
10863 start = block_group->key.objectid;
10864 end = start + block_group->key.offset - 1;
10865 /*
10866 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10867 * btrfs_finish_extent_commit(). If we are at transaction N,
10868 * another task might be running finish_extent_commit() for the
10869 * previous transaction N - 1, and have seen a range belonging
10870 * to the block group in freed_extents[] before we were able to
10871 * clear the whole block group range from freed_extents[]. This
10872 * means that task can lookup for the block group after we
10873 * unpinned it from freed_extents[] and removed it, leading to
10874 * a BUG_ON() at btrfs_unpin_extent_range().
10875 */
10876 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10877 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10878 EXTENT_DIRTY);
10879 if (ret) {
10880 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10881 btrfs_dec_block_group_ro(root, block_group);
10882 goto end_trans;
10883 }
10884 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10885 EXTENT_DIRTY);
10886 if (ret) {
10887 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10888 btrfs_dec_block_group_ro(root, block_group);
10889 goto end_trans;
10890 }
10891 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10892
10893 /* Reset pinned so btrfs_put_block_group doesn't complain */
10894 spin_lock(&space_info->lock);
10895 spin_lock(&block_group->lock);
10896
10897 space_info->bytes_pinned -= block_group->pinned;
10898 space_info->bytes_readonly += block_group->pinned;
10899 percpu_counter_add(&space_info->total_bytes_pinned,
10900 -block_group->pinned);
10901 block_group->pinned = 0;
10902
10903 spin_unlock(&block_group->lock);
10904 spin_unlock(&space_info->lock);
10905
10906 /* DISCARD can flip during remount */
10907 trimming = btrfs_test_opt(root, DISCARD);
10908
10909 /* Implicit trim during transaction commit. */
10910 if (trimming)
10911 btrfs_get_block_group_trimming(block_group);
10912
10913 /*
10914 * Btrfs_remove_chunk will abort the transaction if things go
10915 * horribly wrong.
10916 */
10917 ret = btrfs_remove_chunk(trans, root,
10918 block_group->key.objectid);
10919
10920 if (ret) {
10921 if (trimming)
10922 btrfs_put_block_group_trimming(block_group);
10923 goto end_trans;
10924 }
10925
10926 /*
10927 * If we're not mounted with -odiscard, we can just forget
10928 * about this block group. Otherwise we'll need to wait
10929 * until transaction commit to do the actual discard.
10930 */
10931 if (trimming) {
10932 spin_lock(&fs_info->unused_bgs_lock);
10933 /*
10934 * A concurrent scrub might have added us to the list
10935 * fs_info->unused_bgs, so use a list_move operation
10936 * to add the block group to the deleted_bgs list.
10937 */
10938 list_move(&block_group->bg_list,
10939 &trans->transaction->deleted_bgs);
10940 spin_unlock(&fs_info->unused_bgs_lock);
10941 btrfs_get_block_group(block_group);
10942 }
10943 end_trans:
10944 btrfs_end_transaction(trans, root);
10945 next:
10946 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
10947 btrfs_put_block_group(block_group);
10948 spin_lock(&fs_info->unused_bgs_lock);
10949 }
10950 spin_unlock(&fs_info->unused_bgs_lock);
10951 }
10952
10953 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10954 {
10955 struct btrfs_space_info *space_info;
10956 struct btrfs_super_block *disk_super;
10957 u64 features;
10958 u64 flags;
10959 int mixed = 0;
10960 int ret;
10961
10962 disk_super = fs_info->super_copy;
10963 if (!btrfs_super_root(disk_super))
10964 return -EINVAL;
10965
10966 features = btrfs_super_incompat_flags(disk_super);
10967 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10968 mixed = 1;
10969
10970 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10971 ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info);
10972 if (ret)
10973 goto out;
10974
10975 if (mixed) {
10976 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10977 ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info);
10978 } else {
10979 flags = BTRFS_BLOCK_GROUP_METADATA;
10980 ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info);
10981 if (ret)
10982 goto out;
10983
10984 flags = BTRFS_BLOCK_GROUP_DATA;
10985 ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info);
10986 }
10987 out:
10988 return ret;
10989 }
10990
10991 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
10992 {
10993 return unpin_extent_range(root, start, end, false);
10994 }
10995
10996 /*
10997 * It used to be that old block groups would be left around forever.
10998 * Iterating over them would be enough to trim unused space. Since we
10999 * now automatically remove them, we also need to iterate over unallocated
11000 * space.
11001 *
11002 * We don't want a transaction for this since the discard may take a
11003 * substantial amount of time. We don't require that a transaction be
11004 * running, but we do need to take a running transaction into account
11005 * to ensure that we're not discarding chunks that were released in
11006 * the current transaction.
11007 *
11008 * Holding the chunks lock will prevent other threads from allocating
11009 * or releasing chunks, but it won't prevent a running transaction
11010 * from committing and releasing the memory that the pending chunks
11011 * list head uses. For that, we need to take a reference to the
11012 * transaction.
11013 */
11014 static int btrfs_trim_free_extents(struct btrfs_device *device,
11015 u64 minlen, u64 *trimmed)
11016 {
11017 u64 start = 0, len = 0;
11018 int ret;
11019
11020 *trimmed = 0;
11021
11022 /* Not writeable = nothing to do. */
11023 if (!device->writeable)
11024 return 0;
11025
11026 /* No free space = nothing to do. */
11027 if (device->total_bytes <= device->bytes_used)
11028 return 0;
11029
11030 ret = 0;
11031
11032 while (1) {
11033 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
11034 struct btrfs_transaction *trans;
11035 u64 bytes;
11036
11037 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
11038 if (ret)
11039 return ret;
11040
11041 down_read(&fs_info->commit_root_sem);
11042
11043 spin_lock(&fs_info->trans_lock);
11044 trans = fs_info->running_transaction;
11045 if (trans)
11046 atomic_inc(&trans->use_count);
11047 spin_unlock(&fs_info->trans_lock);
11048
11049 ret = find_free_dev_extent_start(trans, device, minlen, start,
11050 &start, &len);
11051 if (trans)
11052 btrfs_put_transaction(trans);
11053
11054 if (ret) {
11055 up_read(&fs_info->commit_root_sem);
11056 mutex_unlock(&fs_info->chunk_mutex);
11057 if (ret == -ENOSPC)
11058 ret = 0;
11059 break;
11060 }
11061
11062 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
11063 up_read(&fs_info->commit_root_sem);
11064 mutex_unlock(&fs_info->chunk_mutex);
11065
11066 if (ret)
11067 break;
11068
11069 start += len;
11070 *trimmed += bytes;
11071
11072 if (fatal_signal_pending(current)) {
11073 ret = -ERESTARTSYS;
11074 break;
11075 }
11076
11077 cond_resched();
11078 }
11079
11080 return ret;
11081 }
11082
11083 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
11084 {
11085 struct btrfs_fs_info *fs_info = root->fs_info;
11086 struct btrfs_block_group_cache *cache = NULL;
11087 struct btrfs_device *device;
11088 struct list_head *devices;
11089 u64 group_trimmed;
11090 u64 start;
11091 u64 end;
11092 u64 trimmed = 0;
11093 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
11094 int ret = 0;
11095
11096 /*
11097 * try to trim all FS space, our block group may start from non-zero.
11098 */
11099 if (range->len == total_bytes)
11100 cache = btrfs_lookup_first_block_group(fs_info, range->start);
11101 else
11102 cache = btrfs_lookup_block_group(fs_info, range->start);
11103
11104 while (cache) {
11105 if (cache->key.objectid >= (range->start + range->len)) {
11106 btrfs_put_block_group(cache);
11107 break;
11108 }
11109
11110 start = max(range->start, cache->key.objectid);
11111 end = min(range->start + range->len,
11112 cache->key.objectid + cache->key.offset);
11113
11114 if (end - start >= range->minlen) {
11115 if (!block_group_cache_done(cache)) {
11116 ret = cache_block_group(cache, 0);
11117 if (ret) {
11118 btrfs_put_block_group(cache);
11119 break;
11120 }
11121 ret = wait_block_group_cache_done(cache);
11122 if (ret) {
11123 btrfs_put_block_group(cache);
11124 break;
11125 }
11126 }
11127 ret = btrfs_trim_block_group(cache,
11128 &group_trimmed,
11129 start,
11130 end,
11131 range->minlen);
11132
11133 trimmed += group_trimmed;
11134 if (ret) {
11135 btrfs_put_block_group(cache);
11136 break;
11137 }
11138 }
11139
11140 cache = next_block_group(fs_info->tree_root, cache);
11141 }
11142
11143 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
11144 devices = &root->fs_info->fs_devices->alloc_list;
11145 list_for_each_entry(device, devices, dev_alloc_list) {
11146 ret = btrfs_trim_free_extents(device, range->minlen,
11147 &group_trimmed);
11148 if (ret)
11149 break;
11150
11151 trimmed += group_trimmed;
11152 }
11153 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
11154
11155 range->len = trimmed;
11156 return ret;
11157 }
11158
11159 /*
11160 * btrfs_{start,end}_write_no_snapshoting() are similar to
11161 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
11162 * data into the page cache through nocow before the subvolume is snapshoted,
11163 * but flush the data into disk after the snapshot creation, or to prevent
11164 * operations while snapshoting is ongoing and that cause the snapshot to be
11165 * inconsistent (writes followed by expanding truncates for example).
11166 */
11167 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
11168 {
11169 percpu_counter_dec(&root->subv_writers->counter);
11170 /*
11171 * Make sure counter is updated before we wake up waiters.
11172 */
11173 smp_mb();
11174 if (waitqueue_active(&root->subv_writers->wait))
11175 wake_up(&root->subv_writers->wait);
11176 }
11177
11178 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
11179 {
11180 if (atomic_read(&root->will_be_snapshoted))
11181 return 0;
11182
11183 percpu_counter_inc(&root->subv_writers->counter);
11184 /*
11185 * Make sure counter is updated before we check for snapshot creation.
11186 */
11187 smp_mb();
11188 if (atomic_read(&root->will_be_snapshoted)) {
11189 btrfs_end_write_no_snapshoting(root);
11190 return 0;
11191 }
11192 return 1;
11193 }
11194
11195 static int wait_snapshoting_atomic_t(atomic_t *a)
11196 {
11197 schedule();
11198 return 0;
11199 }
11200
11201 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
11202 {
11203 while (true) {
11204 int ret;
11205
11206 ret = btrfs_start_write_no_snapshoting(root);
11207 if (ret)
11208 break;
11209 wait_on_atomic_t(&root->will_be_snapshoted,
11210 wait_snapshoting_atomic_t,
11211 TASK_UNINTERRUPTIBLE);
11212 }
11213 }
Linux with added FPC-III support