projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Btrfs: fix race between block group relocation and nocow writes
[deliverable/linux.git] / fs / btrfs / extent-tree.c
1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
27 #include <linux/percpu_counter.h>
28 #include "hash.h"
29 #include "tree-log.h"
30 #include "disk-io.h"
31 #include "print-tree.h"
32 #include "volumes.h"
33 #include "raid56.h"
34 #include "locking.h"
35 #include "free-space-cache.h"
36 #include "free-space-tree.h"
37 #include "math.h"
38 #include "sysfs.h"
39 #include "qgroup.h"
40
41 #undef SCRAMBLE_DELAYED_REFS
42
43 /*
44 * control flags for do_chunk_alloc's force field
45 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
46 * if we really need one.
47 *
48 * CHUNK_ALLOC_LIMITED means to only try and allocate one
49 * if we have very few chunks already allocated. This is
50 * used as part of the clustering code to help make sure
51 * we have a good pool of storage to cluster in, without
52 * filling the FS with empty chunks
53 *
54 * CHUNK_ALLOC_FORCE means it must try to allocate one
55 *
56 */
57 enum {
58 CHUNK_ALLOC_NO_FORCE = 0,
59 CHUNK_ALLOC_LIMITED = 1,
60 CHUNK_ALLOC_FORCE = 2,
61 };
62
63 /*
64 * Control how reservations are dealt with.
65 *
66 * RESERVE_FREE - freeing a reservation.
67 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
68 * ENOSPC accounting
69 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
70 * bytes_may_use as the ENOSPC accounting is done elsewhere
71 */
72 enum {
73 RESERVE_FREE = 0,
74 RESERVE_ALLOC = 1,
75 RESERVE_ALLOC_NO_ACCOUNT = 2,
76 };
77
78 static int update_block_group(struct btrfs_trans_handle *trans,
79 struct btrfs_root *root, u64 bytenr,
80 u64 num_bytes, int alloc);
81 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
82 struct btrfs_root *root,
83 struct btrfs_delayed_ref_node *node, u64 parent,
84 u64 root_objectid, u64 owner_objectid,
85 u64 owner_offset, int refs_to_drop,
86 struct btrfs_delayed_extent_op *extra_op);
87 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
88 struct extent_buffer *leaf,
89 struct btrfs_extent_item *ei);
90 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
91 struct btrfs_root *root,
92 u64 parent, u64 root_objectid,
93 u64 flags, u64 owner, u64 offset,
94 struct btrfs_key *ins, int ref_mod);
95 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
96 struct btrfs_root *root,
97 u64 parent, u64 root_objectid,
98 u64 flags, struct btrfs_disk_key *key,
99 int level, struct btrfs_key *ins);
100 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
101 struct btrfs_root *extent_root, u64 flags,
102 int force);
103 static int find_next_key(struct btrfs_path *path, int level,
104 struct btrfs_key *key);
105 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
106 int dump_block_groups);
107 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
108 u64 num_bytes, int reserve,
109 int delalloc);
110 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
111 u64 num_bytes);
112 int btrfs_pin_extent(struct btrfs_root *root,
113 u64 bytenr, u64 num_bytes, int reserved);
114
115 static noinline int
116 block_group_cache_done(struct btrfs_block_group_cache *cache)
117 {
118 smp_mb();
119 return cache->cached == BTRFS_CACHE_FINISHED ||
120 cache->cached == BTRFS_CACHE_ERROR;
121 }
122
123 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
124 {
125 return (cache->flags & bits) == bits;
126 }
127
128 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
129 {
130 atomic_inc(&cache->count);
131 }
132
133 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
134 {
135 if (atomic_dec_and_test(&cache->count)) {
136 WARN_ON(cache->pinned > 0);
137 WARN_ON(cache->reserved > 0);
138 kfree(cache->free_space_ctl);
139 kfree(cache);
140 }
141 }
142
143 /*
144 * this adds the block group to the fs_info rb tree for the block group
145 * cache
146 */
147 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
148 struct btrfs_block_group_cache *block_group)
149 {
150 struct rb_node **p;
151 struct rb_node *parent = NULL;
152 struct btrfs_block_group_cache *cache;
153
154 spin_lock(&info->block_group_cache_lock);
155 p = &info->block_group_cache_tree.rb_node;
156
157 while (*p) {
158 parent = *p;
159 cache = rb_entry(parent, struct btrfs_block_group_cache,
160 cache_node);
161 if (block_group->key.objectid < cache->key.objectid) {
162 p = &(*p)->rb_left;
163 } else if (block_group->key.objectid > cache->key.objectid) {
164 p = &(*p)->rb_right;
165 } else {
166 spin_unlock(&info->block_group_cache_lock);
167 return -EEXIST;
168 }
169 }
170
171 rb_link_node(&block_group->cache_node, parent, p);
172 rb_insert_color(&block_group->cache_node,
173 &info->block_group_cache_tree);
174
175 if (info->first_logical_byte > block_group->key.objectid)
176 info->first_logical_byte = block_group->key.objectid;
177
178 spin_unlock(&info->block_group_cache_lock);
179
180 return 0;
181 }
182
183 /*
184 * This will return the block group at or after bytenr if contains is 0, else
185 * it will return the block group that contains the bytenr
186 */
187 static struct btrfs_block_group_cache *
188 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
189 int contains)
190 {
191 struct btrfs_block_group_cache *cache, *ret = NULL;
192 struct rb_node *n;
193 u64 end, start;
194
195 spin_lock(&info->block_group_cache_lock);
196 n = info->block_group_cache_tree.rb_node;
197
198 while (n) {
199 cache = rb_entry(n, struct btrfs_block_group_cache,
200 cache_node);
201 end = cache->key.objectid + cache->key.offset - 1;
202 start = cache->key.objectid;
203
204 if (bytenr < start) {
205 if (!contains && (!ret || start < ret->key.objectid))
206 ret = cache;
207 n = n->rb_left;
208 } else if (bytenr > start) {
209 if (contains && bytenr <= end) {
210 ret = cache;
211 break;
212 }
213 n = n->rb_right;
214 } else {
215 ret = cache;
216 break;
217 }
218 }
219 if (ret) {
220 btrfs_get_block_group(ret);
221 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
222 info->first_logical_byte = ret->key.objectid;
223 }
224 spin_unlock(&info->block_group_cache_lock);
225
226 return ret;
227 }
228
229 static int add_excluded_extent(struct btrfs_root *root,
230 u64 start, u64 num_bytes)
231 {
232 u64 end = start + num_bytes - 1;
233 set_extent_bits(&root->fs_info->freed_extents[0],
234 start, end, EXTENT_UPTODATE, GFP_NOFS);
235 set_extent_bits(&root->fs_info->freed_extents[1],
236 start, end, EXTENT_UPTODATE, GFP_NOFS);
237 return 0;
238 }
239
240 static void free_excluded_extents(struct btrfs_root *root,
241 struct btrfs_block_group_cache *cache)
242 {
243 u64 start, end;
244
245 start = cache->key.objectid;
246 end = start + cache->key.offset - 1;
247
248 clear_extent_bits(&root->fs_info->freed_extents[0],
249 start, end, EXTENT_UPTODATE, GFP_NOFS);
250 clear_extent_bits(&root->fs_info->freed_extents[1],
251 start, end, EXTENT_UPTODATE, GFP_NOFS);
252 }
253
254 static int exclude_super_stripes(struct btrfs_root *root,
255 struct btrfs_block_group_cache *cache)
256 {
257 u64 bytenr;
258 u64 *logical;
259 int stripe_len;
260 int i, nr, ret;
261
262 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
263 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
264 cache->bytes_super += stripe_len;
265 ret = add_excluded_extent(root, cache->key.objectid,
266 stripe_len);
267 if (ret)
268 return ret;
269 }
270
271 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
272 bytenr = btrfs_sb_offset(i);
273 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
274 cache->key.objectid, bytenr,
275 0, &logical, &nr, &stripe_len);
276 if (ret)
277 return ret;
278
279 while (nr--) {
280 u64 start, len;
281
282 if (logical[nr] > cache->key.objectid +
283 cache->key.offset)
284 continue;
285
286 if (logical[nr] + stripe_len <= cache->key.objectid)
287 continue;
288
289 start = logical[nr];
290 if (start < cache->key.objectid) {
291 start = cache->key.objectid;
292 len = (logical[nr] + stripe_len) - start;
293 } else {
294 len = min_t(u64, stripe_len,
295 cache->key.objectid +
296 cache->key.offset - start);
297 }
298
299 cache->bytes_super += len;
300 ret = add_excluded_extent(root, start, len);
301 if (ret) {
302 kfree(logical);
303 return ret;
304 }
305 }
306
307 kfree(logical);
308 }
309 return 0;
310 }
311
312 static struct btrfs_caching_control *
313 get_caching_control(struct btrfs_block_group_cache *cache)
314 {
315 struct btrfs_caching_control *ctl;
316
317 spin_lock(&cache->lock);
318 if (!cache->caching_ctl) {
319 spin_unlock(&cache->lock);
320 return NULL;
321 }
322
323 ctl = cache->caching_ctl;
324 atomic_inc(&ctl->count);
325 spin_unlock(&cache->lock);
326 return ctl;
327 }
328
329 static void put_caching_control(struct btrfs_caching_control *ctl)
330 {
331 if (atomic_dec_and_test(&ctl->count))
332 kfree(ctl);
333 }
334
335 #ifdef CONFIG_BTRFS_DEBUG
336 static void fragment_free_space(struct btrfs_root *root,
337 struct btrfs_block_group_cache *block_group)
338 {
339 u64 start = block_group->key.objectid;
340 u64 len = block_group->key.offset;
341 u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ?
342 root->nodesize : root->sectorsize;
343 u64 step = chunk << 1;
344
345 while (len > chunk) {
346 btrfs_remove_free_space(block_group, start, chunk);
347 start += step;
348 if (len < step)
349 len = 0;
350 else
351 len -= step;
352 }
353 }
354 #endif
355
356 /*
357 * this is only called by cache_block_group, since we could have freed extents
358 * we need to check the pinned_extents for any extents that can't be used yet
359 * since their free space will be released as soon as the transaction commits.
360 */
361 u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
362 struct btrfs_fs_info *info, u64 start, u64 end)
363 {
364 u64 extent_start, extent_end, size, total_added = 0;
365 int ret;
366
367 while (start < end) {
368 ret = find_first_extent_bit(info->pinned_extents, start,
369 &extent_start, &extent_end,
370 EXTENT_DIRTY | EXTENT_UPTODATE,
371 NULL);
372 if (ret)
373 break;
374
375 if (extent_start <= start) {
376 start = extent_end + 1;
377 } else if (extent_start > start && extent_start < end) {
378 size = extent_start - start;
379 total_added += size;
380 ret = btrfs_add_free_space(block_group, start,
381 size);
382 BUG_ON(ret); /* -ENOMEM or logic error */
383 start = extent_end + 1;
384 } else {
385 break;
386 }
387 }
388
389 if (start < end) {
390 size = end - start;
391 total_added += size;
392 ret = btrfs_add_free_space(block_group, start, size);
393 BUG_ON(ret); /* -ENOMEM or logic error */
394 }
395
396 return total_added;
397 }
398
399 static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl)
400 {
401 struct btrfs_block_group_cache *block_group;
402 struct btrfs_fs_info *fs_info;
403 struct btrfs_root *extent_root;
404 struct btrfs_path *path;
405 struct extent_buffer *leaf;
406 struct btrfs_key key;
407 u64 total_found = 0;
408 u64 last = 0;
409 u32 nritems;
410 int ret;
411 bool wakeup = true;
412
413 block_group = caching_ctl->block_group;
414 fs_info = block_group->fs_info;
415 extent_root = fs_info->extent_root;
416
417 path = btrfs_alloc_path();
418 if (!path)
419 return -ENOMEM;
420
421 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
422
423 #ifdef CONFIG_BTRFS_DEBUG
424 /*
425 * If we're fragmenting we don't want to make anybody think we can
426 * allocate from this block group until we've had a chance to fragment
427 * the free space.
428 */
429 if (btrfs_should_fragment_free_space(extent_root, block_group))
430 wakeup = false;
431 #endif
432 /*
433 * We don't want to deadlock with somebody trying to allocate a new
434 * extent for the extent root while also trying to search the extent
435 * root to add free space. So we skip locking and search the commit
436 * root, since its read-only
437 */
438 path->skip_locking = 1;
439 path->search_commit_root = 1;
440 path->reada = READA_FORWARD;
441
442 key.objectid = last;
443 key.offset = 0;
444 key.type = BTRFS_EXTENT_ITEM_KEY;
445
446 next:
447 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
448 if (ret < 0)
449 goto out;
450
451 leaf = path->nodes[0];
452 nritems = btrfs_header_nritems(leaf);
453
454 while (1) {
455 if (btrfs_fs_closing(fs_info) > 1) {
456 last = (u64)-1;
457 break;
458 }
459
460 if (path->slots[0] < nritems) {
461 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
462 } else {
463 ret = find_next_key(path, 0, &key);
464 if (ret)
465 break;
466
467 if (need_resched() ||
468 rwsem_is_contended(&fs_info->commit_root_sem)) {
469 if (wakeup)
470 caching_ctl->progress = last;
471 btrfs_release_path(path);
472 up_read(&fs_info->commit_root_sem);
473 mutex_unlock(&caching_ctl->mutex);
474 cond_resched();
475 mutex_lock(&caching_ctl->mutex);
476 down_read(&fs_info->commit_root_sem);
477 goto next;
478 }
479
480 ret = btrfs_next_leaf(extent_root, path);
481 if (ret < 0)
482 goto out;
483 if (ret)
484 break;
485 leaf = path->nodes[0];
486 nritems = btrfs_header_nritems(leaf);
487 continue;
488 }
489
490 if (key.objectid < last) {
491 key.objectid = last;
492 key.offset = 0;
493 key.type = BTRFS_EXTENT_ITEM_KEY;
494
495 if (wakeup)
496 caching_ctl->progress = last;
497 btrfs_release_path(path);
498 goto next;
499 }
500
501 if (key.objectid < block_group->key.objectid) {
502 path->slots[0]++;
503 continue;
504 }
505
506 if (key.objectid >= block_group->key.objectid +
507 block_group->key.offset)
508 break;
509
510 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
511 key.type == BTRFS_METADATA_ITEM_KEY) {
512 total_found += add_new_free_space(block_group,
513 fs_info, last,
514 key.objectid);
515 if (key.type == BTRFS_METADATA_ITEM_KEY)
516 last = key.objectid +
517 fs_info->tree_root->nodesize;
518 else
519 last = key.objectid + key.offset;
520
521 if (total_found > CACHING_CTL_WAKE_UP) {
522 total_found = 0;
523 if (wakeup)
524 wake_up(&caching_ctl->wait);
525 }
526 }
527 path->slots[0]++;
528 }
529 ret = 0;
530
531 total_found += add_new_free_space(block_group, fs_info, last,
532 block_group->key.objectid +
533 block_group->key.offset);
534 caching_ctl->progress = (u64)-1;
535
536 out:
537 btrfs_free_path(path);
538 return ret;
539 }
540
541 static noinline void caching_thread(struct btrfs_work *work)
542 {
543 struct btrfs_block_group_cache *block_group;
544 struct btrfs_fs_info *fs_info;
545 struct btrfs_caching_control *caching_ctl;
546 struct btrfs_root *extent_root;
547 int ret;
548
549 caching_ctl = container_of(work, struct btrfs_caching_control, work);
550 block_group = caching_ctl->block_group;
551 fs_info = block_group->fs_info;
552 extent_root = fs_info->extent_root;
553
554 mutex_lock(&caching_ctl->mutex);
555 down_read(&fs_info->commit_root_sem);
556
557 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
558 ret = load_free_space_tree(caching_ctl);
559 else
560 ret = load_extent_tree_free(caching_ctl);
561
562 spin_lock(&block_group->lock);
563 block_group->caching_ctl = NULL;
564 block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED;
565 spin_unlock(&block_group->lock);
566
567 #ifdef CONFIG_BTRFS_DEBUG
568 if (btrfs_should_fragment_free_space(extent_root, block_group)) {
569 u64 bytes_used;
570
571 spin_lock(&block_group->space_info->lock);
572 spin_lock(&block_group->lock);
573 bytes_used = block_group->key.offset -
574 btrfs_block_group_used(&block_group->item);
575 block_group->space_info->bytes_used += bytes_used >> 1;
576 spin_unlock(&block_group->lock);
577 spin_unlock(&block_group->space_info->lock);
578 fragment_free_space(extent_root, block_group);
579 }
580 #endif
581
582 caching_ctl->progress = (u64)-1;
583
584 up_read(&fs_info->commit_root_sem);
585 free_excluded_extents(fs_info->extent_root, block_group);
586 mutex_unlock(&caching_ctl->mutex);
587
588 wake_up(&caching_ctl->wait);
589
590 put_caching_control(caching_ctl);
591 btrfs_put_block_group(block_group);
592 }
593
594 static int cache_block_group(struct btrfs_block_group_cache *cache,
595 int load_cache_only)
596 {
597 DEFINE_WAIT(wait);
598 struct btrfs_fs_info *fs_info = cache->fs_info;
599 struct btrfs_caching_control *caching_ctl;
600 int ret = 0;
601
602 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
603 if (!caching_ctl)
604 return -ENOMEM;
605
606 INIT_LIST_HEAD(&caching_ctl->list);
607 mutex_init(&caching_ctl->mutex);
608 init_waitqueue_head(&caching_ctl->wait);
609 caching_ctl->block_group = cache;
610 caching_ctl->progress = cache->key.objectid;
611 atomic_set(&caching_ctl->count, 1);
612 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
613 caching_thread, NULL, NULL);
614
615 spin_lock(&cache->lock);
616 /*
617 * This should be a rare occasion, but this could happen I think in the
618 * case where one thread starts to load the space cache info, and then
619 * some other thread starts a transaction commit which tries to do an
620 * allocation while the other thread is still loading the space cache
621 * info. The previous loop should have kept us from choosing this block
622 * group, but if we've moved to the state where we will wait on caching
623 * block groups we need to first check if we're doing a fast load here,
624 * so we can wait for it to finish, otherwise we could end up allocating
625 * from a block group who's cache gets evicted for one reason or
626 * another.
627 */
628 while (cache->cached == BTRFS_CACHE_FAST) {
629 struct btrfs_caching_control *ctl;
630
631 ctl = cache->caching_ctl;
632 atomic_inc(&ctl->count);
633 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
634 spin_unlock(&cache->lock);
635
636 schedule();
637
638 finish_wait(&ctl->wait, &wait);
639 put_caching_control(ctl);
640 spin_lock(&cache->lock);
641 }
642
643 if (cache->cached != BTRFS_CACHE_NO) {
644 spin_unlock(&cache->lock);
645 kfree(caching_ctl);
646 return 0;
647 }
648 WARN_ON(cache->caching_ctl);
649 cache->caching_ctl = caching_ctl;
650 cache->cached = BTRFS_CACHE_FAST;
651 spin_unlock(&cache->lock);
652
653 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
654 mutex_lock(&caching_ctl->mutex);
655 ret = load_free_space_cache(fs_info, cache);
656
657 spin_lock(&cache->lock);
658 if (ret == 1) {
659 cache->caching_ctl = NULL;
660 cache->cached = BTRFS_CACHE_FINISHED;
661 cache->last_byte_to_unpin = (u64)-1;
662 caching_ctl->progress = (u64)-1;
663 } else {
664 if (load_cache_only) {
665 cache->caching_ctl = NULL;
666 cache->cached = BTRFS_CACHE_NO;
667 } else {
668 cache->cached = BTRFS_CACHE_STARTED;
669 cache->has_caching_ctl = 1;
670 }
671 }
672 spin_unlock(&cache->lock);
673 #ifdef CONFIG_BTRFS_DEBUG
674 if (ret == 1 &&
675 btrfs_should_fragment_free_space(fs_info->extent_root,
676 cache)) {
677 u64 bytes_used;
678
679 spin_lock(&cache->space_info->lock);
680 spin_lock(&cache->lock);
681 bytes_used = cache->key.offset -
682 btrfs_block_group_used(&cache->item);
683 cache->space_info->bytes_used += bytes_used >> 1;
684 spin_unlock(&cache->lock);
685 spin_unlock(&cache->space_info->lock);
686 fragment_free_space(fs_info->extent_root, cache);
687 }
688 #endif
689 mutex_unlock(&caching_ctl->mutex);
690
691 wake_up(&caching_ctl->wait);
692 if (ret == 1) {
693 put_caching_control(caching_ctl);
694 free_excluded_extents(fs_info->extent_root, cache);
695 return 0;
696 }
697 } else {
698 /*
699 * We're either using the free space tree or no caching at all.
700 * Set cached to the appropriate value and wakeup any waiters.
701 */
702 spin_lock(&cache->lock);
703 if (load_cache_only) {
704 cache->caching_ctl = NULL;
705 cache->cached = BTRFS_CACHE_NO;
706 } else {
707 cache->cached = BTRFS_CACHE_STARTED;
708 cache->has_caching_ctl = 1;
709 }
710 spin_unlock(&cache->lock);
711 wake_up(&caching_ctl->wait);
712 }
713
714 if (load_cache_only) {
715 put_caching_control(caching_ctl);
716 return 0;
717 }
718
719 down_write(&fs_info->commit_root_sem);
720 atomic_inc(&caching_ctl->count);
721 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
722 up_write(&fs_info->commit_root_sem);
723
724 btrfs_get_block_group(cache);
725
726 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
727
728 return ret;
729 }
730
731 /*
732 * return the block group that starts at or after bytenr
733 */
734 static struct btrfs_block_group_cache *
735 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
736 {
737 struct btrfs_block_group_cache *cache;
738
739 cache = block_group_cache_tree_search(info, bytenr, 0);
740
741 return cache;
742 }
743
744 /*
745 * return the block group that contains the given bytenr
746 */
747 struct btrfs_block_group_cache *btrfs_lookup_block_group(
748 struct btrfs_fs_info *info,
749 u64 bytenr)
750 {
751 struct btrfs_block_group_cache *cache;
752
753 cache = block_group_cache_tree_search(info, bytenr, 1);
754
755 return cache;
756 }
757
758 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
759 u64 flags)
760 {
761 struct list_head *head = &info->space_info;
762 struct btrfs_space_info *found;
763
764 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
765
766 rcu_read_lock();
767 list_for_each_entry_rcu(found, head, list) {
768 if (found->flags & flags) {
769 rcu_read_unlock();
770 return found;
771 }
772 }
773 rcu_read_unlock();
774 return NULL;
775 }
776
777 /*
778 * after adding space to the filesystem, we need to clear the full flags
779 * on all the space infos.
780 */
781 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
782 {
783 struct list_head *head = &info->space_info;
784 struct btrfs_space_info *found;
785
786 rcu_read_lock();
787 list_for_each_entry_rcu(found, head, list)
788 found->full = 0;
789 rcu_read_unlock();
790 }
791
792 /* simple helper to search for an existing data extent at a given offset */
793 int btrfs_lookup_data_extent(struct btrfs_root *root, u64 start, u64 len)
794 {
795 int ret;
796 struct btrfs_key key;
797 struct btrfs_path *path;
798
799 path = btrfs_alloc_path();
800 if (!path)
801 return -ENOMEM;
802
803 key.objectid = start;
804 key.offset = len;
805 key.type = BTRFS_EXTENT_ITEM_KEY;
806 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
807 0, 0);
808 btrfs_free_path(path);
809 return ret;
810 }
811
812 /*
813 * helper function to lookup reference count and flags of a tree block.
814 *
815 * the head node for delayed ref is used to store the sum of all the
816 * reference count modifications queued up in the rbtree. the head
817 * node may also store the extent flags to set. This way you can check
818 * to see what the reference count and extent flags would be if all of
819 * the delayed refs are not processed.
820 */
821 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
822 struct btrfs_root *root, u64 bytenr,
823 u64 offset, int metadata, u64 *refs, u64 *flags)
824 {
825 struct btrfs_delayed_ref_head *head;
826 struct btrfs_delayed_ref_root *delayed_refs;
827 struct btrfs_path *path;
828 struct btrfs_extent_item *ei;
829 struct extent_buffer *leaf;
830 struct btrfs_key key;
831 u32 item_size;
832 u64 num_refs;
833 u64 extent_flags;
834 int ret;
835
836 /*
837 * If we don't have skinny metadata, don't bother doing anything
838 * different
839 */
840 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
841 offset = root->nodesize;
842 metadata = 0;
843 }
844
845 path = btrfs_alloc_path();
846 if (!path)
847 return -ENOMEM;
848
849 if (!trans) {
850 path->skip_locking = 1;
851 path->search_commit_root = 1;
852 }
853
854 search_again:
855 key.objectid = bytenr;
856 key.offset = offset;
857 if (metadata)
858 key.type = BTRFS_METADATA_ITEM_KEY;
859 else
860 key.type = BTRFS_EXTENT_ITEM_KEY;
861
862 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
863 &key, path, 0, 0);
864 if (ret < 0)
865 goto out_free;
866
867 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
868 if (path->slots[0]) {
869 path->slots[0]--;
870 btrfs_item_key_to_cpu(path->nodes[0], &key,
871 path->slots[0]);
872 if (key.objectid == bytenr &&
873 key.type == BTRFS_EXTENT_ITEM_KEY &&
874 key.offset == root->nodesize)
875 ret = 0;
876 }
877 }
878
879 if (ret == 0) {
880 leaf = path->nodes[0];
881 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
882 if (item_size >= sizeof(*ei)) {
883 ei = btrfs_item_ptr(leaf, path->slots[0],
884 struct btrfs_extent_item);
885 num_refs = btrfs_extent_refs(leaf, ei);
886 extent_flags = btrfs_extent_flags(leaf, ei);
887 } else {
888 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
889 struct btrfs_extent_item_v0 *ei0;
890 BUG_ON(item_size != sizeof(*ei0));
891 ei0 = btrfs_item_ptr(leaf, path->slots[0],
892 struct btrfs_extent_item_v0);
893 num_refs = btrfs_extent_refs_v0(leaf, ei0);
894 /* FIXME: this isn't correct for data */
895 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
896 #else
897 BUG();
898 #endif
899 }
900 BUG_ON(num_refs == 0);
901 } else {
902 num_refs = 0;
903 extent_flags = 0;
904 ret = 0;
905 }
906
907 if (!trans)
908 goto out;
909
910 delayed_refs = &trans->transaction->delayed_refs;
911 spin_lock(&delayed_refs->lock);
912 head = btrfs_find_delayed_ref_head(trans, bytenr);
913 if (head) {
914 if (!mutex_trylock(&head->mutex)) {
915 atomic_inc(&head->node.refs);
916 spin_unlock(&delayed_refs->lock);
917
918 btrfs_release_path(path);
919
920 /*
921 * Mutex was contended, block until it's released and try
922 * again
923 */
924 mutex_lock(&head->mutex);
925 mutex_unlock(&head->mutex);
926 btrfs_put_delayed_ref(&head->node);
927 goto search_again;
928 }
929 spin_lock(&head->lock);
930 if (head->extent_op && head->extent_op->update_flags)
931 extent_flags |= head->extent_op->flags_to_set;
932 else
933 BUG_ON(num_refs == 0);
934
935 num_refs += head->node.ref_mod;
936 spin_unlock(&head->lock);
937 mutex_unlock(&head->mutex);
938 }
939 spin_unlock(&delayed_refs->lock);
940 out:
941 WARN_ON(num_refs == 0);
942 if (refs)
943 *refs = num_refs;
944 if (flags)
945 *flags = extent_flags;
946 out_free:
947 btrfs_free_path(path);
948 return ret;
949 }
950
951 /*
952 * Back reference rules. Back refs have three main goals:
953 *
954 * 1) differentiate between all holders of references to an extent so that
955 * when a reference is dropped we can make sure it was a valid reference
956 * before freeing the extent.
957 *
958 * 2) Provide enough information to quickly find the holders of an extent
959 * if we notice a given block is corrupted or bad.
960 *
961 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
962 * maintenance. This is actually the same as #2, but with a slightly
963 * different use case.
964 *
965 * There are two kinds of back refs. The implicit back refs is optimized
966 * for pointers in non-shared tree blocks. For a given pointer in a block,
967 * back refs of this kind provide information about the block's owner tree
968 * and the pointer's key. These information allow us to find the block by
969 * b-tree searching. The full back refs is for pointers in tree blocks not
970 * referenced by their owner trees. The location of tree block is recorded
971 * in the back refs. Actually the full back refs is generic, and can be
972 * used in all cases the implicit back refs is used. The major shortcoming
973 * of the full back refs is its overhead. Every time a tree block gets
974 * COWed, we have to update back refs entry for all pointers in it.
975 *
976 * For a newly allocated tree block, we use implicit back refs for
977 * pointers in it. This means most tree related operations only involve
978 * implicit back refs. For a tree block created in old transaction, the
979 * only way to drop a reference to it is COW it. So we can detect the
980 * event that tree block loses its owner tree's reference and do the
981 * back refs conversion.
982 *
983 * When a tree block is COW'd through a tree, there are four cases:
984 *
985 * The reference count of the block is one and the tree is the block's
986 * owner tree. Nothing to do in this case.
987 *
988 * The reference count of the block is one and the tree is not the
989 * block's owner tree. In this case, full back refs is used for pointers
990 * in the block. Remove these full back refs, add implicit back refs for
991 * every pointers in the new block.
992 *
993 * The reference count of the block is greater than one and the tree is
994 * the block's owner tree. In this case, implicit back refs is used for
995 * pointers in the block. Add full back refs for every pointers in the
996 * block, increase lower level extents' reference counts. The original
997 * implicit back refs are entailed to the new block.
998 *
999 * The reference count of the block is greater than one and the tree is
1000 * not the block's owner tree. Add implicit back refs for every pointer in
1001 * the new block, increase lower level extents' reference count.
1002 *
1003 * Back Reference Key composing:
1004 *
1005 * The key objectid corresponds to the first byte in the extent,
1006 * The key type is used to differentiate between types of back refs.
1007 * There are different meanings of the key offset for different types
1008 * of back refs.
1009 *
1010 * File extents can be referenced by:
1011 *
1012 * - multiple snapshots, subvolumes, or different generations in one subvol
1013 * - different files inside a single subvolume
1014 * - different offsets inside a file (bookend extents in file.c)
1015 *
1016 * The extent ref structure for the implicit back refs has fields for:
1017 *
1018 * - Objectid of the subvolume root
1019 * - objectid of the file holding the reference
1020 * - original offset in the file
1021 * - how many bookend extents
1022 *
1023 * The key offset for the implicit back refs is hash of the first
1024 * three fields.
1025 *
1026 * The extent ref structure for the full back refs has field for:
1027 *
1028 * - number of pointers in the tree leaf
1029 *
1030 * The key offset for the implicit back refs is the first byte of
1031 * the tree leaf
1032 *
1033 * When a file extent is allocated, The implicit back refs is used.
1034 * the fields are filled in:
1035 *
1036 * (root_key.objectid, inode objectid, offset in file, 1)
1037 *
1038 * When a file extent is removed file truncation, we find the
1039 * corresponding implicit back refs and check the following fields:
1040 *
1041 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1042 *
1043 * Btree extents can be referenced by:
1044 *
1045 * - Different subvolumes
1046 *
1047 * Both the implicit back refs and the full back refs for tree blocks
1048 * only consist of key. The key offset for the implicit back refs is
1049 * objectid of block's owner tree. The key offset for the full back refs
1050 * is the first byte of parent block.
1051 *
1052 * When implicit back refs is used, information about the lowest key and
1053 * level of the tree block are required. These information are stored in
1054 * tree block info structure.
1055 */
1056
1057 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1058 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
1059 struct btrfs_root *root,
1060 struct btrfs_path *path,
1061 u64 owner, u32 extra_size)
1062 {
1063 struct btrfs_extent_item *item;
1064 struct btrfs_extent_item_v0 *ei0;
1065 struct btrfs_extent_ref_v0 *ref0;
1066 struct btrfs_tree_block_info *bi;
1067 struct extent_buffer *leaf;
1068 struct btrfs_key key;
1069 struct btrfs_key found_key;
1070 u32 new_size = sizeof(*item);
1071 u64 refs;
1072 int ret;
1073
1074 leaf = path->nodes[0];
1075 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1076
1077 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1078 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1079 struct btrfs_extent_item_v0);
1080 refs = btrfs_extent_refs_v0(leaf, ei0);
1081
1082 if (owner == (u64)-1) {
1083 while (1) {
1084 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1085 ret = btrfs_next_leaf(root, path);
1086 if (ret < 0)
1087 return ret;
1088 BUG_ON(ret > 0); /* Corruption */
1089 leaf = path->nodes[0];
1090 }
1091 btrfs_item_key_to_cpu(leaf, &found_key,
1092 path->slots[0]);
1093 BUG_ON(key.objectid != found_key.objectid);
1094 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1095 path->slots[0]++;
1096 continue;
1097 }
1098 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1099 struct btrfs_extent_ref_v0);
1100 owner = btrfs_ref_objectid_v0(leaf, ref0);
1101 break;
1102 }
1103 }
1104 btrfs_release_path(path);
1105
1106 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1107 new_size += sizeof(*bi);
1108
1109 new_size -= sizeof(*ei0);
1110 ret = btrfs_search_slot(trans, root, &key, path,
1111 new_size + extra_size, 1);
1112 if (ret < 0)
1113 return ret;
1114 BUG_ON(ret); /* Corruption */
1115
1116 btrfs_extend_item(root, path, new_size);
1117
1118 leaf = path->nodes[0];
1119 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1120 btrfs_set_extent_refs(leaf, item, refs);
1121 /* FIXME: get real generation */
1122 btrfs_set_extent_generation(leaf, item, 0);
1123 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1124 btrfs_set_extent_flags(leaf, item,
1125 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1126 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1127 bi = (struct btrfs_tree_block_info *)(item + 1);
1128 /* FIXME: get first key of the block */
1129 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1130 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1131 } else {
1132 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1133 }
1134 btrfs_mark_buffer_dirty(leaf);
1135 return 0;
1136 }
1137 #endif
1138
1139 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1140 {
1141 u32 high_crc = ~(u32)0;
1142 u32 low_crc = ~(u32)0;
1143 __le64 lenum;
1144
1145 lenum = cpu_to_le64(root_objectid);
1146 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1147 lenum = cpu_to_le64(owner);
1148 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1149 lenum = cpu_to_le64(offset);
1150 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1151
1152 return ((u64)high_crc << 31) ^ (u64)low_crc;
1153 }
1154
1155 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1156 struct btrfs_extent_data_ref *ref)
1157 {
1158 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1159 btrfs_extent_data_ref_objectid(leaf, ref),
1160 btrfs_extent_data_ref_offset(leaf, ref));
1161 }
1162
1163 static int match_extent_data_ref(struct extent_buffer *leaf,
1164 struct btrfs_extent_data_ref *ref,
1165 u64 root_objectid, u64 owner, u64 offset)
1166 {
1167 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1168 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1169 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1170 return 0;
1171 return 1;
1172 }
1173
1174 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1175 struct btrfs_root *root,
1176 struct btrfs_path *path,
1177 u64 bytenr, u64 parent,
1178 u64 root_objectid,
1179 u64 owner, u64 offset)
1180 {
1181 struct btrfs_key key;
1182 struct btrfs_extent_data_ref *ref;
1183 struct extent_buffer *leaf;
1184 u32 nritems;
1185 int ret;
1186 int recow;
1187 int err = -ENOENT;
1188
1189 key.objectid = bytenr;
1190 if (parent) {
1191 key.type = BTRFS_SHARED_DATA_REF_KEY;
1192 key.offset = parent;
1193 } else {
1194 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1195 key.offset = hash_extent_data_ref(root_objectid,
1196 owner, offset);
1197 }
1198 again:
1199 recow = 0;
1200 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1201 if (ret < 0) {
1202 err = ret;
1203 goto fail;
1204 }
1205
1206 if (parent) {
1207 if (!ret)
1208 return 0;
1209 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1210 key.type = BTRFS_EXTENT_REF_V0_KEY;
1211 btrfs_release_path(path);
1212 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1213 if (ret < 0) {
1214 err = ret;
1215 goto fail;
1216 }
1217 if (!ret)
1218 return 0;
1219 #endif
1220 goto fail;
1221 }
1222
1223 leaf = path->nodes[0];
1224 nritems = btrfs_header_nritems(leaf);
1225 while (1) {
1226 if (path->slots[0] >= nritems) {
1227 ret = btrfs_next_leaf(root, path);
1228 if (ret < 0)
1229 err = ret;
1230 if (ret)
1231 goto fail;
1232
1233 leaf = path->nodes[0];
1234 nritems = btrfs_header_nritems(leaf);
1235 recow = 1;
1236 }
1237
1238 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1239 if (key.objectid != bytenr ||
1240 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1241 goto fail;
1242
1243 ref = btrfs_item_ptr(leaf, path->slots[0],
1244 struct btrfs_extent_data_ref);
1245
1246 if (match_extent_data_ref(leaf, ref, root_objectid,
1247 owner, offset)) {
1248 if (recow) {
1249 btrfs_release_path(path);
1250 goto again;
1251 }
1252 err = 0;
1253 break;
1254 }
1255 path->slots[0]++;
1256 }
1257 fail:
1258 return err;
1259 }
1260
1261 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1262 struct btrfs_root *root,
1263 struct btrfs_path *path,
1264 u64 bytenr, u64 parent,
1265 u64 root_objectid, u64 owner,
1266 u64 offset, int refs_to_add)
1267 {
1268 struct btrfs_key key;
1269 struct extent_buffer *leaf;
1270 u32 size;
1271 u32 num_refs;
1272 int ret;
1273
1274 key.objectid = bytenr;
1275 if (parent) {
1276 key.type = BTRFS_SHARED_DATA_REF_KEY;
1277 key.offset = parent;
1278 size = sizeof(struct btrfs_shared_data_ref);
1279 } else {
1280 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1281 key.offset = hash_extent_data_ref(root_objectid,
1282 owner, offset);
1283 size = sizeof(struct btrfs_extent_data_ref);
1284 }
1285
1286 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1287 if (ret && ret != -EEXIST)
1288 goto fail;
1289
1290 leaf = path->nodes[0];
1291 if (parent) {
1292 struct btrfs_shared_data_ref *ref;
1293 ref = btrfs_item_ptr(leaf, path->slots[0],
1294 struct btrfs_shared_data_ref);
1295 if (ret == 0) {
1296 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1297 } else {
1298 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1299 num_refs += refs_to_add;
1300 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1301 }
1302 } else {
1303 struct btrfs_extent_data_ref *ref;
1304 while (ret == -EEXIST) {
1305 ref = btrfs_item_ptr(leaf, path->slots[0],
1306 struct btrfs_extent_data_ref);
1307 if (match_extent_data_ref(leaf, ref, root_objectid,
1308 owner, offset))
1309 break;
1310 btrfs_release_path(path);
1311 key.offset++;
1312 ret = btrfs_insert_empty_item(trans, root, path, &key,
1313 size);
1314 if (ret && ret != -EEXIST)
1315 goto fail;
1316
1317 leaf = path->nodes[0];
1318 }
1319 ref = btrfs_item_ptr(leaf, path->slots[0],
1320 struct btrfs_extent_data_ref);
1321 if (ret == 0) {
1322 btrfs_set_extent_data_ref_root(leaf, ref,
1323 root_objectid);
1324 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1325 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1326 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1327 } else {
1328 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1329 num_refs += refs_to_add;
1330 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1331 }
1332 }
1333 btrfs_mark_buffer_dirty(leaf);
1334 ret = 0;
1335 fail:
1336 btrfs_release_path(path);
1337 return ret;
1338 }
1339
1340 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1341 struct btrfs_root *root,
1342 struct btrfs_path *path,
1343 int refs_to_drop, int *last_ref)
1344 {
1345 struct btrfs_key key;
1346 struct btrfs_extent_data_ref *ref1 = NULL;
1347 struct btrfs_shared_data_ref *ref2 = NULL;
1348 struct extent_buffer *leaf;
1349 u32 num_refs = 0;
1350 int ret = 0;
1351
1352 leaf = path->nodes[0];
1353 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1354
1355 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1356 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1357 struct btrfs_extent_data_ref);
1358 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1359 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1360 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1361 struct btrfs_shared_data_ref);
1362 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1363 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1364 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1365 struct btrfs_extent_ref_v0 *ref0;
1366 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1367 struct btrfs_extent_ref_v0);
1368 num_refs = btrfs_ref_count_v0(leaf, ref0);
1369 #endif
1370 } else {
1371 BUG();
1372 }
1373
1374 BUG_ON(num_refs < refs_to_drop);
1375 num_refs -= refs_to_drop;
1376
1377 if (num_refs == 0) {
1378 ret = btrfs_del_item(trans, root, path);
1379 *last_ref = 1;
1380 } else {
1381 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1382 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1383 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1384 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1385 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1386 else {
1387 struct btrfs_extent_ref_v0 *ref0;
1388 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1389 struct btrfs_extent_ref_v0);
1390 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1391 }
1392 #endif
1393 btrfs_mark_buffer_dirty(leaf);
1394 }
1395 return ret;
1396 }
1397
1398 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
1399 struct btrfs_extent_inline_ref *iref)
1400 {
1401 struct btrfs_key key;
1402 struct extent_buffer *leaf;
1403 struct btrfs_extent_data_ref *ref1;
1404 struct btrfs_shared_data_ref *ref2;
1405 u32 num_refs = 0;
1406
1407 leaf = path->nodes[0];
1408 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1409 if (iref) {
1410 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1411 BTRFS_EXTENT_DATA_REF_KEY) {
1412 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1413 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1414 } else {
1415 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1416 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1417 }
1418 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1419 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1420 struct btrfs_extent_data_ref);
1421 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1422 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1423 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1424 struct btrfs_shared_data_ref);
1425 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1426 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1427 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1428 struct btrfs_extent_ref_v0 *ref0;
1429 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1430 struct btrfs_extent_ref_v0);
1431 num_refs = btrfs_ref_count_v0(leaf, ref0);
1432 #endif
1433 } else {
1434 WARN_ON(1);
1435 }
1436 return num_refs;
1437 }
1438
1439 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1440 struct btrfs_root *root,
1441 struct btrfs_path *path,
1442 u64 bytenr, u64 parent,
1443 u64 root_objectid)
1444 {
1445 struct btrfs_key key;
1446 int ret;
1447
1448 key.objectid = bytenr;
1449 if (parent) {
1450 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1451 key.offset = parent;
1452 } else {
1453 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1454 key.offset = root_objectid;
1455 }
1456
1457 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1458 if (ret > 0)
1459 ret = -ENOENT;
1460 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1461 if (ret == -ENOENT && parent) {
1462 btrfs_release_path(path);
1463 key.type = BTRFS_EXTENT_REF_V0_KEY;
1464 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1465 if (ret > 0)
1466 ret = -ENOENT;
1467 }
1468 #endif
1469 return ret;
1470 }
1471
1472 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1473 struct btrfs_root *root,
1474 struct btrfs_path *path,
1475 u64 bytenr, u64 parent,
1476 u64 root_objectid)
1477 {
1478 struct btrfs_key key;
1479 int ret;
1480
1481 key.objectid = bytenr;
1482 if (parent) {
1483 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1484 key.offset = parent;
1485 } else {
1486 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1487 key.offset = root_objectid;
1488 }
1489
1490 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1491 btrfs_release_path(path);
1492 return ret;
1493 }
1494
1495 static inline int extent_ref_type(u64 parent, u64 owner)
1496 {
1497 int type;
1498 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1499 if (parent > 0)
1500 type = BTRFS_SHARED_BLOCK_REF_KEY;
1501 else
1502 type = BTRFS_TREE_BLOCK_REF_KEY;
1503 } else {
1504 if (parent > 0)
1505 type = BTRFS_SHARED_DATA_REF_KEY;
1506 else
1507 type = BTRFS_EXTENT_DATA_REF_KEY;
1508 }
1509 return type;
1510 }
1511
1512 static int find_next_key(struct btrfs_path *path, int level,
1513 struct btrfs_key *key)
1514
1515 {
1516 for (; level < BTRFS_MAX_LEVEL; level++) {
1517 if (!path->nodes[level])
1518 break;
1519 if (path->slots[level] + 1 >=
1520 btrfs_header_nritems(path->nodes[level]))
1521 continue;
1522 if (level == 0)
1523 btrfs_item_key_to_cpu(path->nodes[level], key,
1524 path->slots[level] + 1);
1525 else
1526 btrfs_node_key_to_cpu(path->nodes[level], key,
1527 path->slots[level] + 1);
1528 return 0;
1529 }
1530 return 1;
1531 }
1532
1533 /*
1534 * look for inline back ref. if back ref is found, *ref_ret is set
1535 * to the address of inline back ref, and 0 is returned.
1536 *
1537 * if back ref isn't found, *ref_ret is set to the address where it
1538 * should be inserted, and -ENOENT is returned.
1539 *
1540 * if insert is true and there are too many inline back refs, the path
1541 * points to the extent item, and -EAGAIN is returned.
1542 *
1543 * NOTE: inline back refs are ordered in the same way that back ref
1544 * items in the tree are ordered.
1545 */
1546 static noinline_for_stack
1547 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1548 struct btrfs_root *root,
1549 struct btrfs_path *path,
1550 struct btrfs_extent_inline_ref **ref_ret,
1551 u64 bytenr, u64 num_bytes,
1552 u64 parent, u64 root_objectid,
1553 u64 owner, u64 offset, int insert)
1554 {
1555 struct btrfs_key key;
1556 struct extent_buffer *leaf;
1557 struct btrfs_extent_item *ei;
1558 struct btrfs_extent_inline_ref *iref;
1559 u64 flags;
1560 u64 item_size;
1561 unsigned long ptr;
1562 unsigned long end;
1563 int extra_size;
1564 int type;
1565 int want;
1566 int ret;
1567 int err = 0;
1568 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1569 SKINNY_METADATA);
1570
1571 key.objectid = bytenr;
1572 key.type = BTRFS_EXTENT_ITEM_KEY;
1573 key.offset = num_bytes;
1574
1575 want = extent_ref_type(parent, owner);
1576 if (insert) {
1577 extra_size = btrfs_extent_inline_ref_size(want);
1578 path->keep_locks = 1;
1579 } else
1580 extra_size = -1;
1581
1582 /*
1583 * Owner is our parent level, so we can just add one to get the level
1584 * for the block we are interested in.
1585 */
1586 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1587 key.type = BTRFS_METADATA_ITEM_KEY;
1588 key.offset = owner;
1589 }
1590
1591 again:
1592 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1593 if (ret < 0) {
1594 err = ret;
1595 goto out;
1596 }
1597
1598 /*
1599 * We may be a newly converted file system which still has the old fat
1600 * extent entries for metadata, so try and see if we have one of those.
1601 */
1602 if (ret > 0 && skinny_metadata) {
1603 skinny_metadata = false;
1604 if (path->slots[0]) {
1605 path->slots[0]--;
1606 btrfs_item_key_to_cpu(path->nodes[0], &key,
1607 path->slots[0]);
1608 if (key.objectid == bytenr &&
1609 key.type == BTRFS_EXTENT_ITEM_KEY &&
1610 key.offset == num_bytes)
1611 ret = 0;
1612 }
1613 if (ret) {
1614 key.objectid = bytenr;
1615 key.type = BTRFS_EXTENT_ITEM_KEY;
1616 key.offset = num_bytes;
1617 btrfs_release_path(path);
1618 goto again;
1619 }
1620 }
1621
1622 if (ret && !insert) {
1623 err = -ENOENT;
1624 goto out;
1625 } else if (WARN_ON(ret)) {
1626 err = -EIO;
1627 goto out;
1628 }
1629
1630 leaf = path->nodes[0];
1631 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1632 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1633 if (item_size < sizeof(*ei)) {
1634 if (!insert) {
1635 err = -ENOENT;
1636 goto out;
1637 }
1638 ret = convert_extent_item_v0(trans, root, path, owner,
1639 extra_size);
1640 if (ret < 0) {
1641 err = ret;
1642 goto out;
1643 }
1644 leaf = path->nodes[0];
1645 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1646 }
1647 #endif
1648 BUG_ON(item_size < sizeof(*ei));
1649
1650 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1651 flags = btrfs_extent_flags(leaf, ei);
1652
1653 ptr = (unsigned long)(ei + 1);
1654 end = (unsigned long)ei + item_size;
1655
1656 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1657 ptr += sizeof(struct btrfs_tree_block_info);
1658 BUG_ON(ptr > end);
1659 }
1660
1661 err = -ENOENT;
1662 while (1) {
1663 if (ptr >= end) {
1664 WARN_ON(ptr > end);
1665 break;
1666 }
1667 iref = (struct btrfs_extent_inline_ref *)ptr;
1668 type = btrfs_extent_inline_ref_type(leaf, iref);
1669 if (want < type)
1670 break;
1671 if (want > type) {
1672 ptr += btrfs_extent_inline_ref_size(type);
1673 continue;
1674 }
1675
1676 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1677 struct btrfs_extent_data_ref *dref;
1678 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1679 if (match_extent_data_ref(leaf, dref, root_objectid,
1680 owner, offset)) {
1681 err = 0;
1682 break;
1683 }
1684 if (hash_extent_data_ref_item(leaf, dref) <
1685 hash_extent_data_ref(root_objectid, owner, offset))
1686 break;
1687 } else {
1688 u64 ref_offset;
1689 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1690 if (parent > 0) {
1691 if (parent == ref_offset) {
1692 err = 0;
1693 break;
1694 }
1695 if (ref_offset < parent)
1696 break;
1697 } else {
1698 if (root_objectid == ref_offset) {
1699 err = 0;
1700 break;
1701 }
1702 if (ref_offset < root_objectid)
1703 break;
1704 }
1705 }
1706 ptr += btrfs_extent_inline_ref_size(type);
1707 }
1708 if (err == -ENOENT && insert) {
1709 if (item_size + extra_size >=
1710 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1711 err = -EAGAIN;
1712 goto out;
1713 }
1714 /*
1715 * To add new inline back ref, we have to make sure
1716 * there is no corresponding back ref item.
1717 * For simplicity, we just do not add new inline back
1718 * ref if there is any kind of item for this block
1719 */
1720 if (find_next_key(path, 0, &key) == 0 &&
1721 key.objectid == bytenr &&
1722 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1723 err = -EAGAIN;
1724 goto out;
1725 }
1726 }
1727 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1728 out:
1729 if (insert) {
1730 path->keep_locks = 0;
1731 btrfs_unlock_up_safe(path, 1);
1732 }
1733 return err;
1734 }
1735
1736 /*
1737 * helper to add new inline back ref
1738 */
1739 static noinline_for_stack
1740 void setup_inline_extent_backref(struct btrfs_root *root,
1741 struct btrfs_path *path,
1742 struct btrfs_extent_inline_ref *iref,
1743 u64 parent, u64 root_objectid,
1744 u64 owner, u64 offset, int refs_to_add,
1745 struct btrfs_delayed_extent_op *extent_op)
1746 {
1747 struct extent_buffer *leaf;
1748 struct btrfs_extent_item *ei;
1749 unsigned long ptr;
1750 unsigned long end;
1751 unsigned long item_offset;
1752 u64 refs;
1753 int size;
1754 int type;
1755
1756 leaf = path->nodes[0];
1757 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1758 item_offset = (unsigned long)iref - (unsigned long)ei;
1759
1760 type = extent_ref_type(parent, owner);
1761 size = btrfs_extent_inline_ref_size(type);
1762
1763 btrfs_extend_item(root, path, size);
1764
1765 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1766 refs = btrfs_extent_refs(leaf, ei);
1767 refs += refs_to_add;
1768 btrfs_set_extent_refs(leaf, ei, refs);
1769 if (extent_op)
1770 __run_delayed_extent_op(extent_op, leaf, ei);
1771
1772 ptr = (unsigned long)ei + item_offset;
1773 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1774 if (ptr < end - size)
1775 memmove_extent_buffer(leaf, ptr + size, ptr,
1776 end - size - ptr);
1777
1778 iref = (struct btrfs_extent_inline_ref *)ptr;
1779 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1780 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1781 struct btrfs_extent_data_ref *dref;
1782 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1783 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1784 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1785 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1786 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1787 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1788 struct btrfs_shared_data_ref *sref;
1789 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1790 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1791 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1792 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1793 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1794 } else {
1795 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1796 }
1797 btrfs_mark_buffer_dirty(leaf);
1798 }
1799
1800 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1801 struct btrfs_root *root,
1802 struct btrfs_path *path,
1803 struct btrfs_extent_inline_ref **ref_ret,
1804 u64 bytenr, u64 num_bytes, u64 parent,
1805 u64 root_objectid, u64 owner, u64 offset)
1806 {
1807 int ret;
1808
1809 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1810 bytenr, num_bytes, parent,
1811 root_objectid, owner, offset, 0);
1812 if (ret != -ENOENT)
1813 return ret;
1814
1815 btrfs_release_path(path);
1816 *ref_ret = NULL;
1817
1818 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1819 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1820 root_objectid);
1821 } else {
1822 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1823 root_objectid, owner, offset);
1824 }
1825 return ret;
1826 }
1827
1828 /*
1829 * helper to update/remove inline back ref
1830 */
1831 static noinline_for_stack
1832 void update_inline_extent_backref(struct btrfs_root *root,
1833 struct btrfs_path *path,
1834 struct btrfs_extent_inline_ref *iref,
1835 int refs_to_mod,
1836 struct btrfs_delayed_extent_op *extent_op,
1837 int *last_ref)
1838 {
1839 struct extent_buffer *leaf;
1840 struct btrfs_extent_item *ei;
1841 struct btrfs_extent_data_ref *dref = NULL;
1842 struct btrfs_shared_data_ref *sref = NULL;
1843 unsigned long ptr;
1844 unsigned long end;
1845 u32 item_size;
1846 int size;
1847 int type;
1848 u64 refs;
1849
1850 leaf = path->nodes[0];
1851 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1852 refs = btrfs_extent_refs(leaf, ei);
1853 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1854 refs += refs_to_mod;
1855 btrfs_set_extent_refs(leaf, ei, refs);
1856 if (extent_op)
1857 __run_delayed_extent_op(extent_op, leaf, ei);
1858
1859 type = btrfs_extent_inline_ref_type(leaf, iref);
1860
1861 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1862 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1863 refs = btrfs_extent_data_ref_count(leaf, dref);
1864 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1865 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1866 refs = btrfs_shared_data_ref_count(leaf, sref);
1867 } else {
1868 refs = 1;
1869 BUG_ON(refs_to_mod != -1);
1870 }
1871
1872 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1873 refs += refs_to_mod;
1874
1875 if (refs > 0) {
1876 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1877 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1878 else
1879 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1880 } else {
1881 *last_ref = 1;
1882 size = btrfs_extent_inline_ref_size(type);
1883 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1884 ptr = (unsigned long)iref;
1885 end = (unsigned long)ei + item_size;
1886 if (ptr + size < end)
1887 memmove_extent_buffer(leaf, ptr, ptr + size,
1888 end - ptr - size);
1889 item_size -= size;
1890 btrfs_truncate_item(root, path, item_size, 1);
1891 }
1892 btrfs_mark_buffer_dirty(leaf);
1893 }
1894
1895 static noinline_for_stack
1896 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1897 struct btrfs_root *root,
1898 struct btrfs_path *path,
1899 u64 bytenr, u64 num_bytes, u64 parent,
1900 u64 root_objectid, u64 owner,
1901 u64 offset, int refs_to_add,
1902 struct btrfs_delayed_extent_op *extent_op)
1903 {
1904 struct btrfs_extent_inline_ref *iref;
1905 int ret;
1906
1907 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1908 bytenr, num_bytes, parent,
1909 root_objectid, owner, offset, 1);
1910 if (ret == 0) {
1911 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1912 update_inline_extent_backref(root, path, iref,
1913 refs_to_add, extent_op, NULL);
1914 } else if (ret == -ENOENT) {
1915 setup_inline_extent_backref(root, path, iref, parent,
1916 root_objectid, owner, offset,
1917 refs_to_add, extent_op);
1918 ret = 0;
1919 }
1920 return ret;
1921 }
1922
1923 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1924 struct btrfs_root *root,
1925 struct btrfs_path *path,
1926 u64 bytenr, u64 parent, u64 root_objectid,
1927 u64 owner, u64 offset, int refs_to_add)
1928 {
1929 int ret;
1930 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1931 BUG_ON(refs_to_add != 1);
1932 ret = insert_tree_block_ref(trans, root, path, bytenr,
1933 parent, root_objectid);
1934 } else {
1935 ret = insert_extent_data_ref(trans, root, path, bytenr,
1936 parent, root_objectid,
1937 owner, offset, refs_to_add);
1938 }
1939 return ret;
1940 }
1941
1942 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1943 struct btrfs_root *root,
1944 struct btrfs_path *path,
1945 struct btrfs_extent_inline_ref *iref,
1946 int refs_to_drop, int is_data, int *last_ref)
1947 {
1948 int ret = 0;
1949
1950 BUG_ON(!is_data && refs_to_drop != 1);
1951 if (iref) {
1952 update_inline_extent_backref(root, path, iref,
1953 -refs_to_drop, NULL, last_ref);
1954 } else if (is_data) {
1955 ret = remove_extent_data_ref(trans, root, path, refs_to_drop,
1956 last_ref);
1957 } else {
1958 *last_ref = 1;
1959 ret = btrfs_del_item(trans, root, path);
1960 }
1961 return ret;
1962 }
1963
1964 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
1965 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1966 u64 *discarded_bytes)
1967 {
1968 int j, ret = 0;
1969 u64 bytes_left, end;
1970 u64 aligned_start = ALIGN(start, 1 << 9);
1971
1972 if (WARN_ON(start != aligned_start)) {
1973 len -= aligned_start - start;
1974 len = round_down(len, 1 << 9);
1975 start = aligned_start;
1976 }
1977
1978 *discarded_bytes = 0;
1979
1980 if (!len)
1981 return 0;
1982
1983 end = start + len;
1984 bytes_left = len;
1985
1986 /* Skip any superblocks on this device. */
1987 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1988 u64 sb_start = btrfs_sb_offset(j);
1989 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
1990 u64 size = sb_start - start;
1991
1992 if (!in_range(sb_start, start, bytes_left) &&
1993 !in_range(sb_end, start, bytes_left) &&
1994 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
1995 continue;
1996
1997 /*
1998 * Superblock spans beginning of range. Adjust start and
1999 * try again.
2000 */
2001 if (sb_start <= start) {
2002 start += sb_end - start;
2003 if (start > end) {
2004 bytes_left = 0;
2005 break;
2006 }
2007 bytes_left = end - start;
2008 continue;
2009 }
2010
2011 if (size) {
2012 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
2013 GFP_NOFS, 0);
2014 if (!ret)
2015 *discarded_bytes += size;
2016 else if (ret != -EOPNOTSUPP)
2017 return ret;
2018 }
2019
2020 start = sb_end;
2021 if (start > end) {
2022 bytes_left = 0;
2023 break;
2024 }
2025 bytes_left = end - start;
2026 }
2027
2028 if (bytes_left) {
2029 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
2030 GFP_NOFS, 0);
2031 if (!ret)
2032 *discarded_bytes += bytes_left;
2033 }
2034 return ret;
2035 }
2036
2037 int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
2038 u64 num_bytes, u64 *actual_bytes)
2039 {
2040 int ret;
2041 u64 discarded_bytes = 0;
2042 struct btrfs_bio *bbio = NULL;
2043
2044
2045 /* Tell the block device(s) that the sectors can be discarded */
2046 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
2047 bytenr, &num_bytes, &bbio, 0);
2048 /* Error condition is -ENOMEM */
2049 if (!ret) {
2050 struct btrfs_bio_stripe *stripe = bbio->stripes;
2051 int i;
2052
2053
2054 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
2055 u64 bytes;
2056 if (!stripe->dev->can_discard)
2057 continue;
2058
2059 ret = btrfs_issue_discard(stripe->dev->bdev,
2060 stripe->physical,
2061 stripe->length,
2062 &bytes);
2063 if (!ret)
2064 discarded_bytes += bytes;
2065 else if (ret != -EOPNOTSUPP)
2066 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2067
2068 /*
2069 * Just in case we get back EOPNOTSUPP for some reason,
2070 * just ignore the return value so we don't screw up
2071 * people calling discard_extent.
2072 */
2073 ret = 0;
2074 }
2075 btrfs_put_bbio(bbio);
2076 }
2077
2078 if (actual_bytes)
2079 *actual_bytes = discarded_bytes;
2080
2081
2082 if (ret == -EOPNOTSUPP)
2083 ret = 0;
2084 return ret;
2085 }
2086
2087 /* Can return -ENOMEM */
2088 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2089 struct btrfs_root *root,
2090 u64 bytenr, u64 num_bytes, u64 parent,
2091 u64 root_objectid, u64 owner, u64 offset)
2092 {
2093 int ret;
2094 struct btrfs_fs_info *fs_info = root->fs_info;
2095
2096 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2097 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2098
2099 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2100 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
2101 num_bytes,
2102 parent, root_objectid, (int)owner,
2103 BTRFS_ADD_DELAYED_REF, NULL);
2104 } else {
2105 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
2106 num_bytes, parent, root_objectid,
2107 owner, offset, 0,
2108 BTRFS_ADD_DELAYED_REF, NULL);
2109 }
2110 return ret;
2111 }
2112
2113 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2114 struct btrfs_root *root,
2115 struct btrfs_delayed_ref_node *node,
2116 u64 parent, u64 root_objectid,
2117 u64 owner, u64 offset, int refs_to_add,
2118 struct btrfs_delayed_extent_op *extent_op)
2119 {
2120 struct btrfs_fs_info *fs_info = root->fs_info;
2121 struct btrfs_path *path;
2122 struct extent_buffer *leaf;
2123 struct btrfs_extent_item *item;
2124 struct btrfs_key key;
2125 u64 bytenr = node->bytenr;
2126 u64 num_bytes = node->num_bytes;
2127 u64 refs;
2128 int ret;
2129
2130 path = btrfs_alloc_path();
2131 if (!path)
2132 return -ENOMEM;
2133
2134 path->reada = READA_FORWARD;
2135 path->leave_spinning = 1;
2136 /* this will setup the path even if it fails to insert the back ref */
2137 ret = insert_inline_extent_backref(trans, fs_info->extent_root, path,
2138 bytenr, num_bytes, parent,
2139 root_objectid, owner, offset,
2140 refs_to_add, extent_op);
2141 if ((ret < 0 && ret != -EAGAIN) || !ret)
2142 goto out;
2143
2144 /*
2145 * Ok we had -EAGAIN which means we didn't have space to insert and
2146 * inline extent ref, so just update the reference count and add a
2147 * normal backref.
2148 */
2149 leaf = path->nodes[0];
2150 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2151 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2152 refs = btrfs_extent_refs(leaf, item);
2153 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2154 if (extent_op)
2155 __run_delayed_extent_op(extent_op, leaf, item);
2156
2157 btrfs_mark_buffer_dirty(leaf);
2158 btrfs_release_path(path);
2159
2160 path->reada = READA_FORWARD;
2161 path->leave_spinning = 1;
2162 /* now insert the actual backref */
2163 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2164 path, bytenr, parent, root_objectid,
2165 owner, offset, refs_to_add);
2166 if (ret)
2167 btrfs_abort_transaction(trans, root, ret);
2168 out:
2169 btrfs_free_path(path);
2170 return ret;
2171 }
2172
2173 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2174 struct btrfs_root *root,
2175 struct btrfs_delayed_ref_node *node,
2176 struct btrfs_delayed_extent_op *extent_op,
2177 int insert_reserved)
2178 {
2179 int ret = 0;
2180 struct btrfs_delayed_data_ref *ref;
2181 struct btrfs_key ins;
2182 u64 parent = 0;
2183 u64 ref_root = 0;
2184 u64 flags = 0;
2185
2186 ins.objectid = node->bytenr;
2187 ins.offset = node->num_bytes;
2188 ins.type = BTRFS_EXTENT_ITEM_KEY;
2189
2190 ref = btrfs_delayed_node_to_data_ref(node);
2191 trace_run_delayed_data_ref(node, ref, node->action);
2192
2193 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2194 parent = ref->parent;
2195 ref_root = ref->root;
2196
2197 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2198 if (extent_op)
2199 flags |= extent_op->flags_to_set;
2200 ret = alloc_reserved_file_extent(trans, root,
2201 parent, ref_root, flags,
2202 ref->objectid, ref->offset,
2203 &ins, node->ref_mod);
2204 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2205 ret = __btrfs_inc_extent_ref(trans, root, node, parent,
2206 ref_root, ref->objectid,
2207 ref->offset, node->ref_mod,
2208 extent_op);
2209 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2210 ret = __btrfs_free_extent(trans, root, node, parent,
2211 ref_root, ref->objectid,
2212 ref->offset, node->ref_mod,
2213 extent_op);
2214 } else {
2215 BUG();
2216 }
2217 return ret;
2218 }
2219
2220 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2221 struct extent_buffer *leaf,
2222 struct btrfs_extent_item *ei)
2223 {
2224 u64 flags = btrfs_extent_flags(leaf, ei);
2225 if (extent_op->update_flags) {
2226 flags |= extent_op->flags_to_set;
2227 btrfs_set_extent_flags(leaf, ei, flags);
2228 }
2229
2230 if (extent_op->update_key) {
2231 struct btrfs_tree_block_info *bi;
2232 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2233 bi = (struct btrfs_tree_block_info *)(ei + 1);
2234 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2235 }
2236 }
2237
2238 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2239 struct btrfs_root *root,
2240 struct btrfs_delayed_ref_node *node,
2241 struct btrfs_delayed_extent_op *extent_op)
2242 {
2243 struct btrfs_key key;
2244 struct btrfs_path *path;
2245 struct btrfs_extent_item *ei;
2246 struct extent_buffer *leaf;
2247 u32 item_size;
2248 int ret;
2249 int err = 0;
2250 int metadata = !extent_op->is_data;
2251
2252 if (trans->aborted)
2253 return 0;
2254
2255 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2256 metadata = 0;
2257
2258 path = btrfs_alloc_path();
2259 if (!path)
2260 return -ENOMEM;
2261
2262 key.objectid = node->bytenr;
2263
2264 if (metadata) {
2265 key.type = BTRFS_METADATA_ITEM_KEY;
2266 key.offset = extent_op->level;
2267 } else {
2268 key.type = BTRFS_EXTENT_ITEM_KEY;
2269 key.offset = node->num_bytes;
2270 }
2271
2272 again:
2273 path->reada = READA_FORWARD;
2274 path->leave_spinning = 1;
2275 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2276 path, 0, 1);
2277 if (ret < 0) {
2278 err = ret;
2279 goto out;
2280 }
2281 if (ret > 0) {
2282 if (metadata) {
2283 if (path->slots[0] > 0) {
2284 path->slots[0]--;
2285 btrfs_item_key_to_cpu(path->nodes[0], &key,
2286 path->slots[0]);
2287 if (key.objectid == node->bytenr &&
2288 key.type == BTRFS_EXTENT_ITEM_KEY &&
2289 key.offset == node->num_bytes)
2290 ret = 0;
2291 }
2292 if (ret > 0) {
2293 btrfs_release_path(path);
2294 metadata = 0;
2295
2296 key.objectid = node->bytenr;
2297 key.offset = node->num_bytes;
2298 key.type = BTRFS_EXTENT_ITEM_KEY;
2299 goto again;
2300 }
2301 } else {
2302 err = -EIO;
2303 goto out;
2304 }
2305 }
2306
2307 leaf = path->nodes[0];
2308 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2309 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2310 if (item_size < sizeof(*ei)) {
2311 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2312 path, (u64)-1, 0);
2313 if (ret < 0) {
2314 err = ret;
2315 goto out;
2316 }
2317 leaf = path->nodes[0];
2318 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2319 }
2320 #endif
2321 BUG_ON(item_size < sizeof(*ei));
2322 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2323 __run_delayed_extent_op(extent_op, leaf, ei);
2324
2325 btrfs_mark_buffer_dirty(leaf);
2326 out:
2327 btrfs_free_path(path);
2328 return err;
2329 }
2330
2331 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2332 struct btrfs_root *root,
2333 struct btrfs_delayed_ref_node *node,
2334 struct btrfs_delayed_extent_op *extent_op,
2335 int insert_reserved)
2336 {
2337 int ret = 0;
2338 struct btrfs_delayed_tree_ref *ref;
2339 struct btrfs_key ins;
2340 u64 parent = 0;
2341 u64 ref_root = 0;
2342 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2343 SKINNY_METADATA);
2344
2345 ref = btrfs_delayed_node_to_tree_ref(node);
2346 trace_run_delayed_tree_ref(node, ref, node->action);
2347
2348 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2349 parent = ref->parent;
2350 ref_root = ref->root;
2351
2352 ins.objectid = node->bytenr;
2353 if (skinny_metadata) {
2354 ins.offset = ref->level;
2355 ins.type = BTRFS_METADATA_ITEM_KEY;
2356 } else {
2357 ins.offset = node->num_bytes;
2358 ins.type = BTRFS_EXTENT_ITEM_KEY;
2359 }
2360
2361 BUG_ON(node->ref_mod != 1);
2362 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2363 BUG_ON(!extent_op || !extent_op->update_flags);
2364 ret = alloc_reserved_tree_block(trans, root,
2365 parent, ref_root,
2366 extent_op->flags_to_set,
2367 &extent_op->key,
2368 ref->level, &ins);
2369 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2370 ret = __btrfs_inc_extent_ref(trans, root, node,
2371 parent, ref_root,
2372 ref->level, 0, 1,
2373 extent_op);
2374 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2375 ret = __btrfs_free_extent(trans, root, node,
2376 parent, ref_root,
2377 ref->level, 0, 1, extent_op);
2378 } else {
2379 BUG();
2380 }
2381 return ret;
2382 }
2383
2384 /* helper function to actually process a single delayed ref entry */
2385 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2386 struct btrfs_root *root,
2387 struct btrfs_delayed_ref_node *node,
2388 struct btrfs_delayed_extent_op *extent_op,
2389 int insert_reserved)
2390 {
2391 int ret = 0;
2392
2393 if (trans->aborted) {
2394 if (insert_reserved)
2395 btrfs_pin_extent(root, node->bytenr,
2396 node->num_bytes, 1);
2397 return 0;
2398 }
2399
2400 if (btrfs_delayed_ref_is_head(node)) {
2401 struct btrfs_delayed_ref_head *head;
2402 /*
2403 * we've hit the end of the chain and we were supposed
2404 * to insert this extent into the tree. But, it got
2405 * deleted before we ever needed to insert it, so all
2406 * we have to do is clean up the accounting
2407 */
2408 BUG_ON(extent_op);
2409 head = btrfs_delayed_node_to_head(node);
2410 trace_run_delayed_ref_head(node, head, node->action);
2411
2412 if (insert_reserved) {
2413 btrfs_pin_extent(root, node->bytenr,
2414 node->num_bytes, 1);
2415 if (head->is_data) {
2416 ret = btrfs_del_csums(trans, root,
2417 node->bytenr,
2418 node->num_bytes);
2419 }
2420 }
2421
2422 /* Also free its reserved qgroup space */
2423 btrfs_qgroup_free_delayed_ref(root->fs_info,
2424 head->qgroup_ref_root,
2425 head->qgroup_reserved);
2426 return ret;
2427 }
2428
2429 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2430 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2431 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2432 insert_reserved);
2433 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2434 node->type == BTRFS_SHARED_DATA_REF_KEY)
2435 ret = run_delayed_data_ref(trans, root, node, extent_op,
2436 insert_reserved);
2437 else
2438 BUG();
2439 return ret;
2440 }
2441
2442 static inline struct btrfs_delayed_ref_node *
2443 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2444 {
2445 struct btrfs_delayed_ref_node *ref;
2446
2447 if (list_empty(&head->ref_list))
2448 return NULL;
2449
2450 /*
2451 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2452 * This is to prevent a ref count from going down to zero, which deletes
2453 * the extent item from the extent tree, when there still are references
2454 * to add, which would fail because they would not find the extent item.
2455 */
2456 list_for_each_entry(ref, &head->ref_list, list) {
2457 if (ref->action == BTRFS_ADD_DELAYED_REF)
2458 return ref;
2459 }
2460
2461 return list_entry(head->ref_list.next, struct btrfs_delayed_ref_node,
2462 list);
2463 }
2464
2465 /*
2466 * Returns 0 on success or if called with an already aborted transaction.
2467 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2468 */
2469 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2470 struct btrfs_root *root,
2471 unsigned long nr)
2472 {
2473 struct btrfs_delayed_ref_root *delayed_refs;
2474 struct btrfs_delayed_ref_node *ref;
2475 struct btrfs_delayed_ref_head *locked_ref = NULL;
2476 struct btrfs_delayed_extent_op *extent_op;
2477 struct btrfs_fs_info *fs_info = root->fs_info;
2478 ktime_t start = ktime_get();
2479 int ret;
2480 unsigned long count = 0;
2481 unsigned long actual_count = 0;
2482 int must_insert_reserved = 0;
2483
2484 delayed_refs = &trans->transaction->delayed_refs;
2485 while (1) {
2486 if (!locked_ref) {
2487 if (count >= nr)
2488 break;
2489
2490 spin_lock(&delayed_refs->lock);
2491 locked_ref = btrfs_select_ref_head(trans);
2492 if (!locked_ref) {
2493 spin_unlock(&delayed_refs->lock);
2494 break;
2495 }
2496
2497 /* grab the lock that says we are going to process
2498 * all the refs for this head */
2499 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2500 spin_unlock(&delayed_refs->lock);
2501 /*
2502 * we may have dropped the spin lock to get the head
2503 * mutex lock, and that might have given someone else
2504 * time to free the head. If that's true, it has been
2505 * removed from our list and we can move on.
2506 */
2507 if (ret == -EAGAIN) {
2508 locked_ref = NULL;
2509 count++;
2510 continue;
2511 }
2512 }
2513
2514 /*
2515 * We need to try and merge add/drops of the same ref since we
2516 * can run into issues with relocate dropping the implicit ref
2517 * and then it being added back again before the drop can
2518 * finish. If we merged anything we need to re-loop so we can
2519 * get a good ref.
2520 * Or we can get node references of the same type that weren't
2521 * merged when created due to bumps in the tree mod seq, and
2522 * we need to merge them to prevent adding an inline extent
2523 * backref before dropping it (triggering a BUG_ON at
2524 * insert_inline_extent_backref()).
2525 */
2526 spin_lock(&locked_ref->lock);
2527 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2528 locked_ref);
2529
2530 /*
2531 * locked_ref is the head node, so we have to go one
2532 * node back for any delayed ref updates
2533 */
2534 ref = select_delayed_ref(locked_ref);
2535
2536 if (ref && ref->seq &&
2537 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2538 spin_unlock(&locked_ref->lock);
2539 btrfs_delayed_ref_unlock(locked_ref);
2540 spin_lock(&delayed_refs->lock);
2541 locked_ref->processing = 0;
2542 delayed_refs->num_heads_ready++;
2543 spin_unlock(&delayed_refs->lock);
2544 locked_ref = NULL;
2545 cond_resched();
2546 count++;
2547 continue;
2548 }
2549
2550 /*
2551 * record the must insert reserved flag before we
2552 * drop the spin lock.
2553 */
2554 must_insert_reserved = locked_ref->must_insert_reserved;
2555 locked_ref->must_insert_reserved = 0;
2556
2557 extent_op = locked_ref->extent_op;
2558 locked_ref->extent_op = NULL;
2559
2560 if (!ref) {
2561
2562
2563 /* All delayed refs have been processed, Go ahead
2564 * and send the head node to run_one_delayed_ref,
2565 * so that any accounting fixes can happen
2566 */
2567 ref = &locked_ref->node;
2568
2569 if (extent_op && must_insert_reserved) {
2570 btrfs_free_delayed_extent_op(extent_op);
2571 extent_op = NULL;
2572 }
2573
2574 if (extent_op) {
2575 spin_unlock(&locked_ref->lock);
2576 ret = run_delayed_extent_op(trans, root,
2577 ref, extent_op);
2578 btrfs_free_delayed_extent_op(extent_op);
2579
2580 if (ret) {
2581 /*
2582 * Need to reset must_insert_reserved if
2583 * there was an error so the abort stuff
2584 * can cleanup the reserved space
2585 * properly.
2586 */
2587 if (must_insert_reserved)
2588 locked_ref->must_insert_reserved = 1;
2589 locked_ref->processing = 0;
2590 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2591 btrfs_delayed_ref_unlock(locked_ref);
2592 return ret;
2593 }
2594 continue;
2595 }
2596
2597 /*
2598 * Need to drop our head ref lock and re-aqcuire the
2599 * delayed ref lock and then re-check to make sure
2600 * nobody got added.
2601 */
2602 spin_unlock(&locked_ref->lock);
2603 spin_lock(&delayed_refs->lock);
2604 spin_lock(&locked_ref->lock);
2605 if (!list_empty(&locked_ref->ref_list) ||
2606 locked_ref->extent_op) {
2607 spin_unlock(&locked_ref->lock);
2608 spin_unlock(&delayed_refs->lock);
2609 continue;
2610 }
2611 ref->in_tree = 0;
2612 delayed_refs->num_heads--;
2613 rb_erase(&locked_ref->href_node,
2614 &delayed_refs->href_root);
2615 spin_unlock(&delayed_refs->lock);
2616 } else {
2617 actual_count++;
2618 ref->in_tree = 0;
2619 list_del(&ref->list);
2620 }
2621 atomic_dec(&delayed_refs->num_entries);
2622
2623 if (!btrfs_delayed_ref_is_head(ref)) {
2624 /*
2625 * when we play the delayed ref, also correct the
2626 * ref_mod on head
2627 */
2628 switch (ref->action) {
2629 case BTRFS_ADD_DELAYED_REF:
2630 case BTRFS_ADD_DELAYED_EXTENT:
2631 locked_ref->node.ref_mod -= ref->ref_mod;
2632 break;
2633 case BTRFS_DROP_DELAYED_REF:
2634 locked_ref->node.ref_mod += ref->ref_mod;
2635 break;
2636 default:
2637 WARN_ON(1);
2638 }
2639 }
2640 spin_unlock(&locked_ref->lock);
2641
2642 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2643 must_insert_reserved);
2644
2645 btrfs_free_delayed_extent_op(extent_op);
2646 if (ret) {
2647 locked_ref->processing = 0;
2648 btrfs_delayed_ref_unlock(locked_ref);
2649 btrfs_put_delayed_ref(ref);
2650 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2651 return ret;
2652 }
2653
2654 /*
2655 * If this node is a head, that means all the refs in this head
2656 * have been dealt with, and we will pick the next head to deal
2657 * with, so we must unlock the head and drop it from the cluster
2658 * list before we release it.
2659 */
2660 if (btrfs_delayed_ref_is_head(ref)) {
2661 if (locked_ref->is_data &&
2662 locked_ref->total_ref_mod < 0) {
2663 spin_lock(&delayed_refs->lock);
2664 delayed_refs->pending_csums -= ref->num_bytes;
2665 spin_unlock(&delayed_refs->lock);
2666 }
2667 btrfs_delayed_ref_unlock(locked_ref);
2668 locked_ref = NULL;
2669 }
2670 btrfs_put_delayed_ref(ref);
2671 count++;
2672 cond_resched();
2673 }
2674
2675 /*
2676 * We don't want to include ref heads since we can have empty ref heads
2677 * and those will drastically skew our runtime down since we just do
2678 * accounting, no actual extent tree updates.
2679 */
2680 if (actual_count > 0) {
2681 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2682 u64 avg;
2683
2684 /*
2685 * We weigh the current average higher than our current runtime
2686 * to avoid large swings in the average.
2687 */
2688 spin_lock(&delayed_refs->lock);
2689 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2690 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2691 spin_unlock(&delayed_refs->lock);
2692 }
2693 return 0;
2694 }
2695
2696 #ifdef SCRAMBLE_DELAYED_REFS
2697 /*
2698 * Normally delayed refs get processed in ascending bytenr order. This
2699 * correlates in most cases to the order added. To expose dependencies on this
2700 * order, we start to process the tree in the middle instead of the beginning
2701 */
2702 static u64 find_middle(struct rb_root *root)
2703 {
2704 struct rb_node *n = root->rb_node;
2705 struct btrfs_delayed_ref_node *entry;
2706 int alt = 1;
2707 u64 middle;
2708 u64 first = 0, last = 0;
2709
2710 n = rb_first(root);
2711 if (n) {
2712 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2713 first = entry->bytenr;
2714 }
2715 n = rb_last(root);
2716 if (n) {
2717 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2718 last = entry->bytenr;
2719 }
2720 n = root->rb_node;
2721
2722 while (n) {
2723 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2724 WARN_ON(!entry->in_tree);
2725
2726 middle = entry->bytenr;
2727
2728 if (alt)
2729 n = n->rb_left;
2730 else
2731 n = n->rb_right;
2732
2733 alt = 1 - alt;
2734 }
2735 return middle;
2736 }
2737 #endif
2738
2739 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2740 {
2741 u64 num_bytes;
2742
2743 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2744 sizeof(struct btrfs_extent_inline_ref));
2745 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2746 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2747
2748 /*
2749 * We don't ever fill up leaves all the way so multiply by 2 just to be
2750 * closer to what we're really going to want to ouse.
2751 */
2752 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2753 }
2754
2755 /*
2756 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2757 * would require to store the csums for that many bytes.
2758 */
2759 u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes)
2760 {
2761 u64 csum_size;
2762 u64 num_csums_per_leaf;
2763 u64 num_csums;
2764
2765 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
2766 num_csums_per_leaf = div64_u64(csum_size,
2767 (u64)btrfs_super_csum_size(root->fs_info->super_copy));
2768 num_csums = div64_u64(csum_bytes, root->sectorsize);
2769 num_csums += num_csums_per_leaf - 1;
2770 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2771 return num_csums;
2772 }
2773
2774 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2775 struct btrfs_root *root)
2776 {
2777 struct btrfs_block_rsv *global_rsv;
2778 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2779 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2780 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2781 u64 num_bytes, num_dirty_bgs_bytes;
2782 int ret = 0;
2783
2784 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2785 num_heads = heads_to_leaves(root, num_heads);
2786 if (num_heads > 1)
2787 num_bytes += (num_heads - 1) * root->nodesize;
2788 num_bytes <<= 1;
2789 num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize;
2790 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root,
2791 num_dirty_bgs);
2792 global_rsv = &root->fs_info->global_block_rsv;
2793
2794 /*
2795 * If we can't allocate any more chunks lets make sure we have _lots_ of
2796 * wiggle room since running delayed refs can create more delayed refs.
2797 */
2798 if (global_rsv->space_info->full) {
2799 num_dirty_bgs_bytes <<= 1;
2800 num_bytes <<= 1;
2801 }
2802
2803 spin_lock(&global_rsv->lock);
2804 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2805 ret = 1;
2806 spin_unlock(&global_rsv->lock);
2807 return ret;
2808 }
2809
2810 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2811 struct btrfs_root *root)
2812 {
2813 struct btrfs_fs_info *fs_info = root->fs_info;
2814 u64 num_entries =
2815 atomic_read(&trans->transaction->delayed_refs.num_entries);
2816 u64 avg_runtime;
2817 u64 val;
2818
2819 smp_mb();
2820 avg_runtime = fs_info->avg_delayed_ref_runtime;
2821 val = num_entries * avg_runtime;
2822 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2823 return 1;
2824 if (val >= NSEC_PER_SEC / 2)
2825 return 2;
2826
2827 return btrfs_check_space_for_delayed_refs(trans, root);
2828 }
2829
2830 struct async_delayed_refs {
2831 struct btrfs_root *root;
2832 int count;
2833 int error;
2834 int sync;
2835 struct completion wait;
2836 struct btrfs_work work;
2837 };
2838
2839 static void delayed_ref_async_start(struct btrfs_work *work)
2840 {
2841 struct async_delayed_refs *async;
2842 struct btrfs_trans_handle *trans;
2843 int ret;
2844
2845 async = container_of(work, struct async_delayed_refs, work);
2846
2847 trans = btrfs_join_transaction(async->root);
2848 if (IS_ERR(trans)) {
2849 async->error = PTR_ERR(trans);
2850 goto done;
2851 }
2852
2853 /*
2854 * trans->sync means that when we call end_transaciton, we won't
2855 * wait on delayed refs
2856 */
2857 trans->sync = true;
2858 ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2859 if (ret)
2860 async->error = ret;
2861
2862 ret = btrfs_end_transaction(trans, async->root);
2863 if (ret && !async->error)
2864 async->error = ret;
2865 done:
2866 if (async->sync)
2867 complete(&async->wait);
2868 else
2869 kfree(async);
2870 }
2871
2872 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2873 unsigned long count, int wait)
2874 {
2875 struct async_delayed_refs *async;
2876 int ret;
2877
2878 async = kmalloc(sizeof(*async), GFP_NOFS);
2879 if (!async)
2880 return -ENOMEM;
2881
2882 async->root = root->fs_info->tree_root;
2883 async->count = count;
2884 async->error = 0;
2885 if (wait)
2886 async->sync = 1;
2887 else
2888 async->sync = 0;
2889 init_completion(&async->wait);
2890
2891 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2892 delayed_ref_async_start, NULL, NULL);
2893
2894 btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2895
2896 if (wait) {
2897 wait_for_completion(&async->wait);
2898 ret = async->error;
2899 kfree(async);
2900 return ret;
2901 }
2902 return 0;
2903 }
2904
2905 /*
2906 * this starts processing the delayed reference count updates and
2907 * extent insertions we have queued up so far. count can be
2908 * 0, which means to process everything in the tree at the start
2909 * of the run (but not newly added entries), or it can be some target
2910 * number you'd like to process.
2911 *
2912 * Returns 0 on success or if called with an aborted transaction
2913 * Returns <0 on error and aborts the transaction
2914 */
2915 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2916 struct btrfs_root *root, unsigned long count)
2917 {
2918 struct rb_node *node;
2919 struct btrfs_delayed_ref_root *delayed_refs;
2920 struct btrfs_delayed_ref_head *head;
2921 int ret;
2922 int run_all = count == (unsigned long)-1;
2923 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
2924
2925 /* We'll clean this up in btrfs_cleanup_transaction */
2926 if (trans->aborted)
2927 return 0;
2928
2929 if (root->fs_info->creating_free_space_tree)
2930 return 0;
2931
2932 if (root == root->fs_info->extent_root)
2933 root = root->fs_info->tree_root;
2934
2935 delayed_refs = &trans->transaction->delayed_refs;
2936 if (count == 0)
2937 count = atomic_read(&delayed_refs->num_entries) * 2;
2938
2939 again:
2940 #ifdef SCRAMBLE_DELAYED_REFS
2941 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2942 #endif
2943 trans->can_flush_pending_bgs = false;
2944 ret = __btrfs_run_delayed_refs(trans, root, count);
2945 if (ret < 0) {
2946 btrfs_abort_transaction(trans, root, ret);
2947 return ret;
2948 }
2949
2950 if (run_all) {
2951 if (!list_empty(&trans->new_bgs))
2952 btrfs_create_pending_block_groups(trans, root);
2953
2954 spin_lock(&delayed_refs->lock);
2955 node = rb_first(&delayed_refs->href_root);
2956 if (!node) {
2957 spin_unlock(&delayed_refs->lock);
2958 goto out;
2959 }
2960 count = (unsigned long)-1;
2961
2962 while (node) {
2963 head = rb_entry(node, struct btrfs_delayed_ref_head,
2964 href_node);
2965 if (btrfs_delayed_ref_is_head(&head->node)) {
2966 struct btrfs_delayed_ref_node *ref;
2967
2968 ref = &head->node;
2969 atomic_inc(&ref->refs);
2970
2971 spin_unlock(&delayed_refs->lock);
2972 /*
2973 * Mutex was contended, block until it's
2974 * released and try again
2975 */
2976 mutex_lock(&head->mutex);
2977 mutex_unlock(&head->mutex);
2978
2979 btrfs_put_delayed_ref(ref);
2980 cond_resched();
2981 goto again;
2982 } else {
2983 WARN_ON(1);
2984 }
2985 node = rb_next(node);
2986 }
2987 spin_unlock(&delayed_refs->lock);
2988 cond_resched();
2989 goto again;
2990 }
2991 out:
2992 assert_qgroups_uptodate(trans);
2993 trans->can_flush_pending_bgs = can_flush_pending_bgs;
2994 return 0;
2995 }
2996
2997 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2998 struct btrfs_root *root,
2999 u64 bytenr, u64 num_bytes, u64 flags,
3000 int level, int is_data)
3001 {
3002 struct btrfs_delayed_extent_op *extent_op;
3003 int ret;
3004
3005 extent_op = btrfs_alloc_delayed_extent_op();
3006 if (!extent_op)
3007 return -ENOMEM;
3008
3009 extent_op->flags_to_set = flags;
3010 extent_op->update_flags = true;
3011 extent_op->update_key = false;
3012 extent_op->is_data = is_data ? true : false;
3013 extent_op->level = level;
3014
3015 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
3016 num_bytes, extent_op);
3017 if (ret)
3018 btrfs_free_delayed_extent_op(extent_op);
3019 return ret;
3020 }
3021
3022 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
3023 struct btrfs_root *root,
3024 struct btrfs_path *path,
3025 u64 objectid, u64 offset, u64 bytenr)
3026 {
3027 struct btrfs_delayed_ref_head *head;
3028 struct btrfs_delayed_ref_node *ref;
3029 struct btrfs_delayed_data_ref *data_ref;
3030 struct btrfs_delayed_ref_root *delayed_refs;
3031 int ret = 0;
3032
3033 delayed_refs = &trans->transaction->delayed_refs;
3034 spin_lock(&delayed_refs->lock);
3035 head = btrfs_find_delayed_ref_head(trans, bytenr);
3036 if (!head) {
3037 spin_unlock(&delayed_refs->lock);
3038 return 0;
3039 }
3040
3041 if (!mutex_trylock(&head->mutex)) {
3042 atomic_inc(&head->node.refs);
3043 spin_unlock(&delayed_refs->lock);
3044
3045 btrfs_release_path(path);
3046
3047 /*
3048 * Mutex was contended, block until it's released and let
3049 * caller try again
3050 */
3051 mutex_lock(&head->mutex);
3052 mutex_unlock(&head->mutex);
3053 btrfs_put_delayed_ref(&head->node);
3054 return -EAGAIN;
3055 }
3056 spin_unlock(&delayed_refs->lock);
3057
3058 spin_lock(&head->lock);
3059 list_for_each_entry(ref, &head->ref_list, list) {
3060 /* If it's a shared ref we know a cross reference exists */
3061 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3062 ret = 1;
3063 break;
3064 }
3065
3066 data_ref = btrfs_delayed_node_to_data_ref(ref);
3067
3068 /*
3069 * If our ref doesn't match the one we're currently looking at
3070 * then we have a cross reference.
3071 */
3072 if (data_ref->root != root->root_key.objectid ||
3073 data_ref->objectid != objectid ||
3074 data_ref->offset != offset) {
3075 ret = 1;
3076 break;
3077 }
3078 }
3079 spin_unlock(&head->lock);
3080 mutex_unlock(&head->mutex);
3081 return ret;
3082 }
3083
3084 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
3085 struct btrfs_root *root,
3086 struct btrfs_path *path,
3087 u64 objectid, u64 offset, u64 bytenr)
3088 {
3089 struct btrfs_root *extent_root = root->fs_info->extent_root;
3090 struct extent_buffer *leaf;
3091 struct btrfs_extent_data_ref *ref;
3092 struct btrfs_extent_inline_ref *iref;
3093 struct btrfs_extent_item *ei;
3094 struct btrfs_key key;
3095 u32 item_size;
3096 int ret;
3097
3098 key.objectid = bytenr;
3099 key.offset = (u64)-1;
3100 key.type = BTRFS_EXTENT_ITEM_KEY;
3101
3102 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3103 if (ret < 0)
3104 goto out;
3105 BUG_ON(ret == 0); /* Corruption */
3106
3107 ret = -ENOENT;
3108 if (path->slots[0] == 0)
3109 goto out;
3110
3111 path->slots[0]--;
3112 leaf = path->nodes[0];
3113 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3114
3115 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3116 goto out;
3117
3118 ret = 1;
3119 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3120 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3121 if (item_size < sizeof(*ei)) {
3122 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3123 goto out;
3124 }
3125 #endif
3126 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3127
3128 if (item_size != sizeof(*ei) +
3129 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3130 goto out;
3131
3132 if (btrfs_extent_generation(leaf, ei) <=
3133 btrfs_root_last_snapshot(&root->root_item))
3134 goto out;
3135
3136 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3137 if (btrfs_extent_inline_ref_type(leaf, iref) !=
3138 BTRFS_EXTENT_DATA_REF_KEY)
3139 goto out;
3140
3141 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3142 if (btrfs_extent_refs(leaf, ei) !=
3143 btrfs_extent_data_ref_count(leaf, ref) ||
3144 btrfs_extent_data_ref_root(leaf, ref) !=
3145 root->root_key.objectid ||
3146 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3147 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3148 goto out;
3149
3150 ret = 0;
3151 out:
3152 return ret;
3153 }
3154
3155 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
3156 struct btrfs_root *root,
3157 u64 objectid, u64 offset, u64 bytenr)
3158 {
3159 struct btrfs_path *path;
3160 int ret;
3161 int ret2;
3162
3163 path = btrfs_alloc_path();
3164 if (!path)
3165 return -ENOENT;
3166
3167 do {
3168 ret = check_committed_ref(trans, root, path, objectid,
3169 offset, bytenr);
3170 if (ret && ret != -ENOENT)
3171 goto out;
3172
3173 ret2 = check_delayed_ref(trans, root, path, objectid,
3174 offset, bytenr);
3175 } while (ret2 == -EAGAIN);
3176
3177 if (ret2 && ret2 != -ENOENT) {
3178 ret = ret2;
3179 goto out;
3180 }
3181
3182 if (ret != -ENOENT || ret2 != -ENOENT)
3183 ret = 0;
3184 out:
3185 btrfs_free_path(path);
3186 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3187 WARN_ON(ret > 0);
3188 return ret;
3189 }
3190
3191 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3192 struct btrfs_root *root,
3193 struct extent_buffer *buf,
3194 int full_backref, int inc)
3195 {
3196 u64 bytenr;
3197 u64 num_bytes;
3198 u64 parent;
3199 u64 ref_root;
3200 u32 nritems;
3201 struct btrfs_key key;
3202 struct btrfs_file_extent_item *fi;
3203 int i;
3204 int level;
3205 int ret = 0;
3206 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3207 u64, u64, u64, u64, u64, u64);
3208
3209
3210 if (btrfs_test_is_dummy_root(root))
3211 return 0;
3212
3213 ref_root = btrfs_header_owner(buf);
3214 nritems = btrfs_header_nritems(buf);
3215 level = btrfs_header_level(buf);
3216
3217 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3218 return 0;
3219
3220 if (inc)
3221 process_func = btrfs_inc_extent_ref;
3222 else
3223 process_func = btrfs_free_extent;
3224
3225 if (full_backref)
3226 parent = buf->start;
3227 else
3228 parent = 0;
3229
3230 for (i = 0; i < nritems; i++) {
3231 if (level == 0) {
3232 btrfs_item_key_to_cpu(buf, &key, i);
3233 if (key.type != BTRFS_EXTENT_DATA_KEY)
3234 continue;
3235 fi = btrfs_item_ptr(buf, i,
3236 struct btrfs_file_extent_item);
3237 if (btrfs_file_extent_type(buf, fi) ==
3238 BTRFS_FILE_EXTENT_INLINE)
3239 continue;
3240 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3241 if (bytenr == 0)
3242 continue;
3243
3244 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3245 key.offset -= btrfs_file_extent_offset(buf, fi);
3246 ret = process_func(trans, root, bytenr, num_bytes,
3247 parent, ref_root, key.objectid,
3248 key.offset);
3249 if (ret)
3250 goto fail;
3251 } else {
3252 bytenr = btrfs_node_blockptr(buf, i);
3253 num_bytes = root->nodesize;
3254 ret = process_func(trans, root, bytenr, num_bytes,
3255 parent, ref_root, level - 1, 0);
3256 if (ret)
3257 goto fail;
3258 }
3259 }
3260 return 0;
3261 fail:
3262 return ret;
3263 }
3264
3265 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3266 struct extent_buffer *buf, int full_backref)
3267 {
3268 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3269 }
3270
3271 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3272 struct extent_buffer *buf, int full_backref)
3273 {
3274 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3275 }
3276
3277 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3278 struct btrfs_root *root,
3279 struct btrfs_path *path,
3280 struct btrfs_block_group_cache *cache)
3281 {
3282 int ret;
3283 struct btrfs_root *extent_root = root->fs_info->extent_root;
3284 unsigned long bi;
3285 struct extent_buffer *leaf;
3286
3287 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3288 if (ret) {
3289 if (ret > 0)
3290 ret = -ENOENT;
3291 goto fail;
3292 }
3293
3294 leaf = path->nodes[0];
3295 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3296 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3297 btrfs_mark_buffer_dirty(leaf);
3298 fail:
3299 btrfs_release_path(path);
3300 return ret;
3301
3302 }
3303
3304 static struct btrfs_block_group_cache *
3305 next_block_group(struct btrfs_root *root,
3306 struct btrfs_block_group_cache *cache)
3307 {
3308 struct rb_node *node;
3309
3310 spin_lock(&root->fs_info->block_group_cache_lock);
3311
3312 /* If our block group was removed, we need a full search. */
3313 if (RB_EMPTY_NODE(&cache->cache_node)) {
3314 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3315
3316 spin_unlock(&root->fs_info->block_group_cache_lock);
3317 btrfs_put_block_group(cache);
3318 cache = btrfs_lookup_first_block_group(root->fs_info,
3319 next_bytenr);
3320 return cache;
3321 }
3322 node = rb_next(&cache->cache_node);
3323 btrfs_put_block_group(cache);
3324 if (node) {
3325 cache = rb_entry(node, struct btrfs_block_group_cache,
3326 cache_node);
3327 btrfs_get_block_group(cache);
3328 } else
3329 cache = NULL;
3330 spin_unlock(&root->fs_info->block_group_cache_lock);
3331 return cache;
3332 }
3333
3334 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3335 struct btrfs_trans_handle *trans,
3336 struct btrfs_path *path)
3337 {
3338 struct btrfs_root *root = block_group->fs_info->tree_root;
3339 struct inode *inode = NULL;
3340 u64 alloc_hint = 0;
3341 int dcs = BTRFS_DC_ERROR;
3342 u64 num_pages = 0;
3343 int retries = 0;
3344 int ret = 0;
3345
3346 /*
3347 * If this block group is smaller than 100 megs don't bother caching the
3348 * block group.
3349 */
3350 if (block_group->key.offset < (100 * SZ_1M)) {
3351 spin_lock(&block_group->lock);
3352 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3353 spin_unlock(&block_group->lock);
3354 return 0;
3355 }
3356
3357 if (trans->aborted)
3358 return 0;
3359 again:
3360 inode = lookup_free_space_inode(root, block_group, path);
3361 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3362 ret = PTR_ERR(inode);
3363 btrfs_release_path(path);
3364 goto out;
3365 }
3366
3367 if (IS_ERR(inode)) {
3368 BUG_ON(retries);
3369 retries++;
3370
3371 if (block_group->ro)
3372 goto out_free;
3373
3374 ret = create_free_space_inode(root, trans, block_group, path);
3375 if (ret)
3376 goto out_free;
3377 goto again;
3378 }
3379
3380 /* We've already setup this transaction, go ahead and exit */
3381 if (block_group->cache_generation == trans->transid &&
3382 i_size_read(inode)) {
3383 dcs = BTRFS_DC_SETUP;
3384 goto out_put;
3385 }
3386
3387 /*
3388 * We want to set the generation to 0, that way if anything goes wrong
3389 * from here on out we know not to trust this cache when we load up next
3390 * time.
3391 */
3392 BTRFS_I(inode)->generation = 0;
3393 ret = btrfs_update_inode(trans, root, inode);
3394 if (ret) {
3395 /*
3396 * So theoretically we could recover from this, simply set the
3397 * super cache generation to 0 so we know to invalidate the
3398 * cache, but then we'd have to keep track of the block groups
3399 * that fail this way so we know we _have_ to reset this cache
3400 * before the next commit or risk reading stale cache. So to
3401 * limit our exposure to horrible edge cases lets just abort the
3402 * transaction, this only happens in really bad situations
3403 * anyway.
3404 */
3405 btrfs_abort_transaction(trans, root, ret);
3406 goto out_put;
3407 }
3408 WARN_ON(ret);
3409
3410 if (i_size_read(inode) > 0) {
3411 ret = btrfs_check_trunc_cache_free_space(root,
3412 &root->fs_info->global_block_rsv);
3413 if (ret)
3414 goto out_put;
3415
3416 ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode);
3417 if (ret)
3418 goto out_put;
3419 }
3420
3421 spin_lock(&block_group->lock);
3422 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3423 !btrfs_test_opt(root, SPACE_CACHE)) {
3424 /*
3425 * don't bother trying to write stuff out _if_
3426 * a) we're not cached,
3427 * b) we're with nospace_cache mount option.
3428 */
3429 dcs = BTRFS_DC_WRITTEN;
3430 spin_unlock(&block_group->lock);
3431 goto out_put;
3432 }
3433 spin_unlock(&block_group->lock);
3434
3435 /*
3436 * We hit an ENOSPC when setting up the cache in this transaction, just
3437 * skip doing the setup, we've already cleared the cache so we're safe.
3438 */
3439 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3440 ret = -ENOSPC;
3441 goto out_put;
3442 }
3443
3444 /*
3445 * Try to preallocate enough space based on how big the block group is.
3446 * Keep in mind this has to include any pinned space which could end up
3447 * taking up quite a bit since it's not folded into the other space
3448 * cache.
3449 */
3450 num_pages = div_u64(block_group->key.offset, SZ_256M);
3451 if (!num_pages)
3452 num_pages = 1;
3453
3454 num_pages *= 16;
3455 num_pages *= PAGE_SIZE;
3456
3457 ret = btrfs_check_data_free_space(inode, 0, num_pages);
3458 if (ret)
3459 goto out_put;
3460
3461 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3462 num_pages, num_pages,
3463 &alloc_hint);
3464 /*
3465 * Our cache requires contiguous chunks so that we don't modify a bunch
3466 * of metadata or split extents when writing the cache out, which means
3467 * we can enospc if we are heavily fragmented in addition to just normal
3468 * out of space conditions. So if we hit this just skip setting up any
3469 * other block groups for this transaction, maybe we'll unpin enough
3470 * space the next time around.
3471 */
3472 if (!ret)
3473 dcs = BTRFS_DC_SETUP;
3474 else if (ret == -ENOSPC)
3475 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3476 btrfs_free_reserved_data_space(inode, 0, num_pages);
3477
3478 out_put:
3479 iput(inode);
3480 out_free:
3481 btrfs_release_path(path);
3482 out:
3483 spin_lock(&block_group->lock);
3484 if (!ret && dcs == BTRFS_DC_SETUP)
3485 block_group->cache_generation = trans->transid;
3486 block_group->disk_cache_state = dcs;
3487 spin_unlock(&block_group->lock);
3488
3489 return ret;
3490 }
3491
3492 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3493 struct btrfs_root *root)
3494 {
3495 struct btrfs_block_group_cache *cache, *tmp;
3496 struct btrfs_transaction *cur_trans = trans->transaction;
3497 struct btrfs_path *path;
3498
3499 if (list_empty(&cur_trans->dirty_bgs) ||
3500 !btrfs_test_opt(root, SPACE_CACHE))
3501 return 0;
3502
3503 path = btrfs_alloc_path();
3504 if (!path)
3505 return -ENOMEM;
3506
3507 /* Could add new block groups, use _safe just in case */
3508 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3509 dirty_list) {
3510 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3511 cache_save_setup(cache, trans, path);
3512 }
3513
3514 btrfs_free_path(path);
3515 return 0;
3516 }
3517
3518 /*
3519 * transaction commit does final block group cache writeback during a
3520 * critical section where nothing is allowed to change the FS. This is
3521 * required in order for the cache to actually match the block group,
3522 * but can introduce a lot of latency into the commit.
3523 *
3524 * So, btrfs_start_dirty_block_groups is here to kick off block group
3525 * cache IO. There's a chance we'll have to redo some of it if the
3526 * block group changes again during the commit, but it greatly reduces
3527 * the commit latency by getting rid of the easy block groups while
3528 * we're still allowing others to join the commit.
3529 */
3530 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3531 struct btrfs_root *root)
3532 {
3533 struct btrfs_block_group_cache *cache;
3534 struct btrfs_transaction *cur_trans = trans->transaction;
3535 int ret = 0;
3536 int should_put;
3537 struct btrfs_path *path = NULL;
3538 LIST_HEAD(dirty);
3539 struct list_head *io = &cur_trans->io_bgs;
3540 int num_started = 0;
3541 int loops = 0;
3542
3543 spin_lock(&cur_trans->dirty_bgs_lock);
3544 if (list_empty(&cur_trans->dirty_bgs)) {
3545 spin_unlock(&cur_trans->dirty_bgs_lock);
3546 return 0;
3547 }
3548 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3549 spin_unlock(&cur_trans->dirty_bgs_lock);
3550
3551 again:
3552 /*
3553 * make sure all the block groups on our dirty list actually
3554 * exist
3555 */
3556 btrfs_create_pending_block_groups(trans, root);
3557
3558 if (!path) {
3559 path = btrfs_alloc_path();
3560 if (!path)
3561 return -ENOMEM;
3562 }
3563
3564 /*
3565 * cache_write_mutex is here only to save us from balance or automatic
3566 * removal of empty block groups deleting this block group while we are
3567 * writing out the cache
3568 */
3569 mutex_lock(&trans->transaction->cache_write_mutex);
3570 while (!list_empty(&dirty)) {
3571 cache = list_first_entry(&dirty,
3572 struct btrfs_block_group_cache,
3573 dirty_list);
3574 /*
3575 * this can happen if something re-dirties a block
3576 * group that is already under IO. Just wait for it to
3577 * finish and then do it all again
3578 */
3579 if (!list_empty(&cache->io_list)) {
3580 list_del_init(&cache->io_list);
3581 btrfs_wait_cache_io(root, trans, cache,
3582 &cache->io_ctl, path,
3583 cache->key.objectid);
3584 btrfs_put_block_group(cache);
3585 }
3586
3587
3588 /*
3589 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3590 * if it should update the cache_state. Don't delete
3591 * until after we wait.
3592 *
3593 * Since we're not running in the commit critical section
3594 * we need the dirty_bgs_lock to protect from update_block_group
3595 */
3596 spin_lock(&cur_trans->dirty_bgs_lock);
3597 list_del_init(&cache->dirty_list);
3598 spin_unlock(&cur_trans->dirty_bgs_lock);
3599
3600 should_put = 1;
3601
3602 cache_save_setup(cache, trans, path);
3603
3604 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3605 cache->io_ctl.inode = NULL;
3606 ret = btrfs_write_out_cache(root, trans, cache, path);
3607 if (ret == 0 && cache->io_ctl.inode) {
3608 num_started++;
3609 should_put = 0;
3610
3611 /*
3612 * the cache_write_mutex is protecting
3613 * the io_list
3614 */
3615 list_add_tail(&cache->io_list, io);
3616 } else {
3617 /*
3618 * if we failed to write the cache, the
3619 * generation will be bad and life goes on
3620 */
3621 ret = 0;
3622 }
3623 }
3624 if (!ret) {
3625 ret = write_one_cache_group(trans, root, path, cache);
3626 /*
3627 * Our block group might still be attached to the list
3628 * of new block groups in the transaction handle of some
3629 * other task (struct btrfs_trans_handle->new_bgs). This
3630 * means its block group item isn't yet in the extent
3631 * tree. If this happens ignore the error, as we will
3632 * try again later in the critical section of the
3633 * transaction commit.
3634 */
3635 if (ret == -ENOENT) {
3636 ret = 0;
3637 spin_lock(&cur_trans->dirty_bgs_lock);
3638 if (list_empty(&cache->dirty_list)) {
3639 list_add_tail(&cache->dirty_list,
3640 &cur_trans->dirty_bgs);
3641 btrfs_get_block_group(cache);
3642 }
3643 spin_unlock(&cur_trans->dirty_bgs_lock);
3644 } else if (ret) {
3645 btrfs_abort_transaction(trans, root, ret);
3646 }
3647 }
3648
3649 /* if its not on the io list, we need to put the block group */
3650 if (should_put)
3651 btrfs_put_block_group(cache);
3652
3653 if (ret)
3654 break;
3655
3656 /*
3657 * Avoid blocking other tasks for too long. It might even save
3658 * us from writing caches for block groups that are going to be
3659 * removed.
3660 */
3661 mutex_unlock(&trans->transaction->cache_write_mutex);
3662 mutex_lock(&trans->transaction->cache_write_mutex);
3663 }
3664 mutex_unlock(&trans->transaction->cache_write_mutex);
3665
3666 /*
3667 * go through delayed refs for all the stuff we've just kicked off
3668 * and then loop back (just once)
3669 */
3670 ret = btrfs_run_delayed_refs(trans, root, 0);
3671 if (!ret && loops == 0) {
3672 loops++;
3673 spin_lock(&cur_trans->dirty_bgs_lock);
3674 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3675 /*
3676 * dirty_bgs_lock protects us from concurrent block group
3677 * deletes too (not just cache_write_mutex).
3678 */
3679 if (!list_empty(&dirty)) {
3680 spin_unlock(&cur_trans->dirty_bgs_lock);
3681 goto again;
3682 }
3683 spin_unlock(&cur_trans->dirty_bgs_lock);
3684 }
3685
3686 btrfs_free_path(path);
3687 return ret;
3688 }
3689
3690 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3691 struct btrfs_root *root)
3692 {
3693 struct btrfs_block_group_cache *cache;
3694 struct btrfs_transaction *cur_trans = trans->transaction;
3695 int ret = 0;
3696 int should_put;
3697 struct btrfs_path *path;
3698 struct list_head *io = &cur_trans->io_bgs;
3699 int num_started = 0;
3700
3701 path = btrfs_alloc_path();
3702 if (!path)
3703 return -ENOMEM;
3704
3705 /*
3706 * Even though we are in the critical section of the transaction commit,
3707 * we can still have concurrent tasks adding elements to this
3708 * transaction's list of dirty block groups. These tasks correspond to
3709 * endio free space workers started when writeback finishes for a
3710 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3711 * allocate new block groups as a result of COWing nodes of the root
3712 * tree when updating the free space inode. The writeback for the space
3713 * caches is triggered by an earlier call to
3714 * btrfs_start_dirty_block_groups() and iterations of the following
3715 * loop.
3716 * Also we want to do the cache_save_setup first and then run the
3717 * delayed refs to make sure we have the best chance at doing this all
3718 * in one shot.
3719 */
3720 spin_lock(&cur_trans->dirty_bgs_lock);
3721 while (!list_empty(&cur_trans->dirty_bgs)) {
3722 cache = list_first_entry(&cur_trans->dirty_bgs,
3723 struct btrfs_block_group_cache,
3724 dirty_list);
3725
3726 /*
3727 * this can happen if cache_save_setup re-dirties a block
3728 * group that is already under IO. Just wait for it to
3729 * finish and then do it all again
3730 */
3731 if (!list_empty(&cache->io_list)) {
3732 spin_unlock(&cur_trans->dirty_bgs_lock);
3733 list_del_init(&cache->io_list);
3734 btrfs_wait_cache_io(root, trans, cache,
3735 &cache->io_ctl, path,
3736 cache->key.objectid);
3737 btrfs_put_block_group(cache);
3738 spin_lock(&cur_trans->dirty_bgs_lock);
3739 }
3740
3741 /*
3742 * don't remove from the dirty list until after we've waited
3743 * on any pending IO
3744 */
3745 list_del_init(&cache->dirty_list);
3746 spin_unlock(&cur_trans->dirty_bgs_lock);
3747 should_put = 1;
3748
3749 cache_save_setup(cache, trans, path);
3750
3751 if (!ret)
3752 ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1);
3753
3754 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3755 cache->io_ctl.inode = NULL;
3756 ret = btrfs_write_out_cache(root, trans, cache, path);
3757 if (ret == 0 && cache->io_ctl.inode) {
3758 num_started++;
3759 should_put = 0;
3760 list_add_tail(&cache->io_list, io);
3761 } else {
3762 /*
3763 * if we failed to write the cache, the
3764 * generation will be bad and life goes on
3765 */
3766 ret = 0;
3767 }
3768 }
3769 if (!ret) {
3770 ret = write_one_cache_group(trans, root, path, cache);
3771 /*
3772 * One of the free space endio workers might have
3773 * created a new block group while updating a free space
3774 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3775 * and hasn't released its transaction handle yet, in
3776 * which case the new block group is still attached to
3777 * its transaction handle and its creation has not
3778 * finished yet (no block group item in the extent tree
3779 * yet, etc). If this is the case, wait for all free
3780 * space endio workers to finish and retry. This is a
3781 * a very rare case so no need for a more efficient and
3782 * complex approach.
3783 */
3784 if (ret == -ENOENT) {
3785 wait_event(cur_trans->writer_wait,
3786 atomic_read(&cur_trans->num_writers) == 1);
3787 ret = write_one_cache_group(trans, root, path,
3788 cache);
3789 }
3790 if (ret)
3791 btrfs_abort_transaction(trans, root, ret);
3792 }
3793
3794 /* if its not on the io list, we need to put the block group */
3795 if (should_put)
3796 btrfs_put_block_group(cache);
3797 spin_lock(&cur_trans->dirty_bgs_lock);
3798 }
3799 spin_unlock(&cur_trans->dirty_bgs_lock);
3800
3801 while (!list_empty(io)) {
3802 cache = list_first_entry(io, struct btrfs_block_group_cache,
3803 io_list);
3804 list_del_init(&cache->io_list);
3805 btrfs_wait_cache_io(root, trans, cache,
3806 &cache->io_ctl, path, cache->key.objectid);
3807 btrfs_put_block_group(cache);
3808 }
3809
3810 btrfs_free_path(path);
3811 return ret;
3812 }
3813
3814 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3815 {
3816 struct btrfs_block_group_cache *block_group;
3817 int readonly = 0;
3818
3819 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3820 if (!block_group || block_group->ro)
3821 readonly = 1;
3822 if (block_group)
3823 btrfs_put_block_group(block_group);
3824 return readonly;
3825 }
3826
3827 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3828 {
3829 struct btrfs_block_group_cache *bg;
3830 bool ret = true;
3831
3832 bg = btrfs_lookup_block_group(fs_info, bytenr);
3833 if (!bg)
3834 return false;
3835
3836 spin_lock(&bg->lock);
3837 if (bg->ro)
3838 ret = false;
3839 else
3840 atomic_inc(&bg->nocow_writers);
3841 spin_unlock(&bg->lock);
3842
3843 /* no put on block group, done by btrfs_dec_nocow_writers */
3844 if (!ret)
3845 btrfs_put_block_group(bg);
3846
3847 return ret;
3848
3849 }
3850
3851 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3852 {
3853 struct btrfs_block_group_cache *bg;
3854
3855 bg = btrfs_lookup_block_group(fs_info, bytenr);
3856 ASSERT(bg);
3857 if (atomic_dec_and_test(&bg->nocow_writers))
3858 wake_up_atomic_t(&bg->nocow_writers);
3859 /*
3860 * Once for our lookup and once for the lookup done by a previous call
3861 * to btrfs_inc_nocow_writers()
3862 */
3863 btrfs_put_block_group(bg);
3864 btrfs_put_block_group(bg);
3865 }
3866
3867 static int btrfs_wait_nocow_writers_atomic_t(atomic_t *a)
3868 {
3869 schedule();
3870 return 0;
3871 }
3872
3873 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
3874 {
3875 wait_on_atomic_t(&bg->nocow_writers,
3876 btrfs_wait_nocow_writers_atomic_t,
3877 TASK_UNINTERRUPTIBLE);
3878 }
3879
3880 static const char *alloc_name(u64 flags)
3881 {
3882 switch (flags) {
3883 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3884 return "mixed";
3885 case BTRFS_BLOCK_GROUP_METADATA:
3886 return "metadata";
3887 case BTRFS_BLOCK_GROUP_DATA:
3888 return "data";
3889 case BTRFS_BLOCK_GROUP_SYSTEM:
3890 return "system";
3891 default:
3892 WARN_ON(1);
3893 return "invalid-combination";
3894 };
3895 }
3896
3897 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3898 u64 total_bytes, u64 bytes_used,
3899 struct btrfs_space_info **space_info)
3900 {
3901 struct btrfs_space_info *found;
3902 int i;
3903 int factor;
3904 int ret;
3905
3906 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3907 BTRFS_BLOCK_GROUP_RAID10))
3908 factor = 2;
3909 else
3910 factor = 1;
3911
3912 found = __find_space_info(info, flags);
3913 if (found) {
3914 spin_lock(&found->lock);
3915 found->total_bytes += total_bytes;
3916 found->disk_total += total_bytes * factor;
3917 found->bytes_used += bytes_used;
3918 found->disk_used += bytes_used * factor;
3919 if (total_bytes > 0)
3920 found->full = 0;
3921 spin_unlock(&found->lock);
3922 *space_info = found;
3923 return 0;
3924 }
3925 found = kzalloc(sizeof(*found), GFP_NOFS);
3926 if (!found)
3927 return -ENOMEM;
3928
3929 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3930 if (ret) {
3931 kfree(found);
3932 return ret;
3933 }
3934
3935 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3936 INIT_LIST_HEAD(&found->block_groups[i]);
3937 init_rwsem(&found->groups_sem);
3938 spin_lock_init(&found->lock);
3939 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3940 found->total_bytes = total_bytes;
3941 found->disk_total = total_bytes * factor;
3942 found->bytes_used = bytes_used;
3943 found->disk_used = bytes_used * factor;
3944 found->bytes_pinned = 0;
3945 found->bytes_reserved = 0;
3946 found->bytes_readonly = 0;
3947 found->bytes_may_use = 0;
3948 found->full = 0;
3949 found->max_extent_size = 0;
3950 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3951 found->chunk_alloc = 0;
3952 found->flush = 0;
3953 init_waitqueue_head(&found->wait);
3954 INIT_LIST_HEAD(&found->ro_bgs);
3955
3956 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3957 info->space_info_kobj, "%s",
3958 alloc_name(found->flags));
3959 if (ret) {
3960 kfree(found);
3961 return ret;
3962 }
3963
3964 *space_info = found;
3965 list_add_rcu(&found->list, &info->space_info);
3966 if (flags & BTRFS_BLOCK_GROUP_DATA)
3967 info->data_sinfo = found;
3968
3969 return ret;
3970 }
3971
3972 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3973 {
3974 u64 extra_flags = chunk_to_extended(flags) &
3975 BTRFS_EXTENDED_PROFILE_MASK;
3976
3977 write_seqlock(&fs_info->profiles_lock);
3978 if (flags & BTRFS_BLOCK_GROUP_DATA)
3979 fs_info->avail_data_alloc_bits |= extra_flags;
3980 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3981 fs_info->avail_metadata_alloc_bits |= extra_flags;
3982 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3983 fs_info->avail_system_alloc_bits |= extra_flags;
3984 write_sequnlock(&fs_info->profiles_lock);
3985 }
3986
3987 /*
3988 * returns target flags in extended format or 0 if restripe for this
3989 * chunk_type is not in progress
3990 *
3991 * should be called with either volume_mutex or balance_lock held
3992 */
3993 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3994 {
3995 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3996 u64 target = 0;
3997
3998 if (!bctl)
3999 return 0;
4000
4001 if (flags & BTRFS_BLOCK_GROUP_DATA &&
4002 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4003 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
4004 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
4005 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4006 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
4007 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
4008 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4009 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
4010 }
4011
4012 return target;
4013 }
4014
4015 /*
4016 * @flags: available profiles in extended format (see ctree.h)
4017 *
4018 * Returns reduced profile in chunk format. If profile changing is in
4019 * progress (either running or paused) picks the target profile (if it's
4020 * already available), otherwise falls back to plain reducing.
4021 */
4022 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
4023 {
4024 u64 num_devices = root->fs_info->fs_devices->rw_devices;
4025 u64 target;
4026 u64 raid_type;
4027 u64 allowed = 0;
4028
4029 /*
4030 * see if restripe for this chunk_type is in progress, if so
4031 * try to reduce to the target profile
4032 */
4033 spin_lock(&root->fs_info->balance_lock);
4034 target = get_restripe_target(root->fs_info, flags);
4035 if (target) {
4036 /* pick target profile only if it's already available */
4037 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
4038 spin_unlock(&root->fs_info->balance_lock);
4039 return extended_to_chunk(target);
4040 }
4041 }
4042 spin_unlock(&root->fs_info->balance_lock);
4043
4044 /* First, mask out the RAID levels which aren't possible */
4045 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4046 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
4047 allowed |= btrfs_raid_group[raid_type];
4048 }
4049 allowed &= flags;
4050
4051 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
4052 allowed = BTRFS_BLOCK_GROUP_RAID6;
4053 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4054 allowed = BTRFS_BLOCK_GROUP_RAID5;
4055 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4056 allowed = BTRFS_BLOCK_GROUP_RAID10;
4057 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4058 allowed = BTRFS_BLOCK_GROUP_RAID1;
4059 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4060 allowed = BTRFS_BLOCK_GROUP_RAID0;
4061
4062 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4063
4064 return extended_to_chunk(flags | allowed);
4065 }
4066
4067 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
4068 {
4069 unsigned seq;
4070 u64 flags;
4071
4072 do {
4073 flags = orig_flags;
4074 seq = read_seqbegin(&root->fs_info->profiles_lock);
4075
4076 if (flags & BTRFS_BLOCK_GROUP_DATA)
4077 flags |= root->fs_info->avail_data_alloc_bits;
4078 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4079 flags |= root->fs_info->avail_system_alloc_bits;
4080 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4081 flags |= root->fs_info->avail_metadata_alloc_bits;
4082 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
4083
4084 return btrfs_reduce_alloc_profile(root, flags);
4085 }
4086
4087 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
4088 {
4089 u64 flags;
4090 u64 ret;
4091
4092 if (data)
4093 flags = BTRFS_BLOCK_GROUP_DATA;
4094 else if (root == root->fs_info->chunk_root)
4095 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4096 else
4097 flags = BTRFS_BLOCK_GROUP_METADATA;
4098
4099 ret = get_alloc_profile(root, flags);
4100 return ret;
4101 }
4102
4103 int btrfs_alloc_data_chunk_ondemand(struct inode *inode, u64 bytes)
4104 {
4105 struct btrfs_space_info *data_sinfo;
4106 struct btrfs_root *root = BTRFS_I(inode)->root;
4107 struct btrfs_fs_info *fs_info = root->fs_info;
4108 u64 used;
4109 int ret = 0;
4110 int need_commit = 2;
4111 int have_pinned_space;
4112
4113 /* make sure bytes are sectorsize aligned */
4114 bytes = ALIGN(bytes, root->sectorsize);
4115
4116 if (btrfs_is_free_space_inode(inode)) {
4117 need_commit = 0;
4118 ASSERT(current->journal_info);
4119 }
4120
4121 data_sinfo = fs_info->data_sinfo;
4122 if (!data_sinfo)
4123 goto alloc;
4124
4125 again:
4126 /* make sure we have enough space to handle the data first */
4127 spin_lock(&data_sinfo->lock);
4128 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
4129 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
4130 data_sinfo->bytes_may_use;
4131
4132 if (used + bytes > data_sinfo->total_bytes) {
4133 struct btrfs_trans_handle *trans;
4134
4135 /*
4136 * if we don't have enough free bytes in this space then we need
4137 * to alloc a new chunk.
4138 */
4139 if (!data_sinfo->full) {
4140 u64 alloc_target;
4141
4142 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4143 spin_unlock(&data_sinfo->lock);
4144 alloc:
4145 alloc_target = btrfs_get_alloc_profile(root, 1);
4146 /*
4147 * It is ugly that we don't call nolock join
4148 * transaction for the free space inode case here.
4149 * But it is safe because we only do the data space
4150 * reservation for the free space cache in the
4151 * transaction context, the common join transaction
4152 * just increase the counter of the current transaction
4153 * handler, doesn't try to acquire the trans_lock of
4154 * the fs.
4155 */
4156 trans = btrfs_join_transaction(root);
4157 if (IS_ERR(trans))
4158 return PTR_ERR(trans);
4159
4160 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4161 alloc_target,
4162 CHUNK_ALLOC_NO_FORCE);
4163 btrfs_end_transaction(trans, root);
4164 if (ret < 0) {
4165 if (ret != -ENOSPC)
4166 return ret;
4167 else {
4168 have_pinned_space = 1;
4169 goto commit_trans;
4170 }
4171 }
4172
4173 if (!data_sinfo)
4174 data_sinfo = fs_info->data_sinfo;
4175
4176 goto again;
4177 }
4178
4179 /*
4180 * If we don't have enough pinned space to deal with this
4181 * allocation, and no removed chunk in current transaction,
4182 * don't bother committing the transaction.
4183 */
4184 have_pinned_space = percpu_counter_compare(
4185 &data_sinfo->total_bytes_pinned,
4186 used + bytes - data_sinfo->total_bytes);
4187 spin_unlock(&data_sinfo->lock);
4188
4189 /* commit the current transaction and try again */
4190 commit_trans:
4191 if (need_commit &&
4192 !atomic_read(&root->fs_info->open_ioctl_trans)) {
4193 need_commit--;
4194
4195 if (need_commit > 0) {
4196 btrfs_start_delalloc_roots(fs_info, 0, -1);
4197 btrfs_wait_ordered_roots(fs_info, -1, 0, (u64)-1);
4198 }
4199
4200 trans = btrfs_join_transaction(root);
4201 if (IS_ERR(trans))
4202 return PTR_ERR(trans);
4203 if (have_pinned_space >= 0 ||
4204 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4205 &trans->transaction->flags) ||
4206 need_commit > 0) {
4207 ret = btrfs_commit_transaction(trans, root);
4208 if (ret)
4209 return ret;
4210 /*
4211 * The cleaner kthread might still be doing iput
4212 * operations. Wait for it to finish so that
4213 * more space is released.
4214 */
4215 mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex);
4216 mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex);
4217 goto again;
4218 } else {
4219 btrfs_end_transaction(trans, root);
4220 }
4221 }
4222
4223 trace_btrfs_space_reservation(root->fs_info,
4224 "space_info:enospc",
4225 data_sinfo->flags, bytes, 1);
4226 return -ENOSPC;
4227 }
4228 data_sinfo->bytes_may_use += bytes;
4229 trace_btrfs_space_reservation(root->fs_info, "space_info",
4230 data_sinfo->flags, bytes, 1);
4231 spin_unlock(&data_sinfo->lock);
4232
4233 return ret;
4234 }
4235
4236 /*
4237 * New check_data_free_space() with ability for precious data reservation
4238 * Will replace old btrfs_check_data_free_space(), but for patch split,
4239 * add a new function first and then replace it.
4240 */
4241 int btrfs_check_data_free_space(struct inode *inode, u64 start, u64 len)
4242 {
4243 struct btrfs_root *root = BTRFS_I(inode)->root;
4244 int ret;
4245
4246 /* align the range */
4247 len = round_up(start + len, root->sectorsize) -
4248 round_down(start, root->sectorsize);
4249 start = round_down(start, root->sectorsize);
4250
4251 ret = btrfs_alloc_data_chunk_ondemand(inode, len);
4252 if (ret < 0)
4253 return ret;
4254
4255 /*
4256 * Use new btrfs_qgroup_reserve_data to reserve precious data space
4257 *
4258 * TODO: Find a good method to avoid reserve data space for NOCOW
4259 * range, but don't impact performance on quota disable case.
4260 */
4261 ret = btrfs_qgroup_reserve_data(inode, start, len);
4262 return ret;
4263 }
4264
4265 /*
4266 * Called if we need to clear a data reservation for this inode
4267 * Normally in a error case.
4268 *
4269 * This one will *NOT* use accurate qgroup reserved space API, just for case
4270 * which we can't sleep and is sure it won't affect qgroup reserved space.
4271 * Like clear_bit_hook().
4272 */
4273 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4274 u64 len)
4275 {
4276 struct btrfs_root *root = BTRFS_I(inode)->root;
4277 struct btrfs_space_info *data_sinfo;
4278
4279 /* Make sure the range is aligned to sectorsize */
4280 len = round_up(start + len, root->sectorsize) -
4281 round_down(start, root->sectorsize);
4282 start = round_down(start, root->sectorsize);
4283
4284 data_sinfo = root->fs_info->data_sinfo;
4285 spin_lock(&data_sinfo->lock);
4286 if (WARN_ON(data_sinfo->bytes_may_use < len))
4287 data_sinfo->bytes_may_use = 0;
4288 else
4289 data_sinfo->bytes_may_use -= len;
4290 trace_btrfs_space_reservation(root->fs_info, "space_info",
4291 data_sinfo->flags, len, 0);
4292 spin_unlock(&data_sinfo->lock);
4293 }
4294
4295 /*
4296 * Called if we need to clear a data reservation for this inode
4297 * Normally in a error case.
4298 *
4299 * This one will handle the per-indoe data rsv map for accurate reserved
4300 * space framework.
4301 */
4302 void btrfs_free_reserved_data_space(struct inode *inode, u64 start, u64 len)
4303 {
4304 btrfs_free_reserved_data_space_noquota(inode, start, len);
4305 btrfs_qgroup_free_data(inode, start, len);
4306 }
4307
4308 static void force_metadata_allocation(struct btrfs_fs_info *info)
4309 {
4310 struct list_head *head = &info->space_info;
4311 struct btrfs_space_info *found;
4312
4313 rcu_read_lock();
4314 list_for_each_entry_rcu(found, head, list) {
4315 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4316 found->force_alloc = CHUNK_ALLOC_FORCE;
4317 }
4318 rcu_read_unlock();
4319 }
4320
4321 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4322 {
4323 return (global->size << 1);
4324 }
4325
4326 static int should_alloc_chunk(struct btrfs_root *root,
4327 struct btrfs_space_info *sinfo, int force)
4328 {
4329 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4330 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
4331 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
4332 u64 thresh;
4333
4334 if (force == CHUNK_ALLOC_FORCE)
4335 return 1;
4336
4337 /*
4338 * We need to take into account the global rsv because for all intents
4339 * and purposes it's used space. Don't worry about locking the
4340 * global_rsv, it doesn't change except when the transaction commits.
4341 */
4342 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4343 num_allocated += calc_global_rsv_need_space(global_rsv);
4344
4345 /*
4346 * in limited mode, we want to have some free space up to
4347 * about 1% of the FS size.
4348 */
4349 if (force == CHUNK_ALLOC_LIMITED) {
4350 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
4351 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4352
4353 if (num_bytes - num_allocated < thresh)
4354 return 1;
4355 }
4356
4357 if (num_allocated + SZ_2M < div_factor(num_bytes, 8))
4358 return 0;
4359 return 1;
4360 }
4361
4362 static u64 get_profile_num_devs(struct btrfs_root *root, u64 type)
4363 {
4364 u64 num_dev;
4365
4366 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4367 BTRFS_BLOCK_GROUP_RAID0 |
4368 BTRFS_BLOCK_GROUP_RAID5 |
4369 BTRFS_BLOCK_GROUP_RAID6))
4370 num_dev = root->fs_info->fs_devices->rw_devices;
4371 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4372 num_dev = 2;
4373 else
4374 num_dev = 1; /* DUP or single */
4375
4376 return num_dev;
4377 }
4378
4379 /*
4380 * If @is_allocation is true, reserve space in the system space info necessary
4381 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4382 * removing a chunk.
4383 */
4384 void check_system_chunk(struct btrfs_trans_handle *trans,
4385 struct btrfs_root *root,
4386 u64 type)
4387 {
4388 struct btrfs_space_info *info;
4389 u64 left;
4390 u64 thresh;
4391 int ret = 0;
4392 u64 num_devs;
4393
4394 /*
4395 * Needed because we can end up allocating a system chunk and for an
4396 * atomic and race free space reservation in the chunk block reserve.
4397 */
4398 ASSERT(mutex_is_locked(&root->fs_info->chunk_mutex));
4399
4400 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4401 spin_lock(&info->lock);
4402 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
4403 info->bytes_reserved - info->bytes_readonly -
4404 info->bytes_may_use;
4405 spin_unlock(&info->lock);
4406
4407 num_devs = get_profile_num_devs(root, type);
4408
4409 /* num_devs device items to update and 1 chunk item to add or remove */
4410 thresh = btrfs_calc_trunc_metadata_size(root, num_devs) +
4411 btrfs_calc_trans_metadata_size(root, 1);
4412
4413 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
4414 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
4415 left, thresh, type);
4416 dump_space_info(info, 0, 0);
4417 }
4418
4419 if (left < thresh) {
4420 u64 flags;
4421
4422 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
4423 /*
4424 * Ignore failure to create system chunk. We might end up not
4425 * needing it, as we might not need to COW all nodes/leafs from
4426 * the paths we visit in the chunk tree (they were already COWed
4427 * or created in the current transaction for example).
4428 */
4429 ret = btrfs_alloc_chunk(trans, root, flags);
4430 }
4431
4432 if (!ret) {
4433 ret = btrfs_block_rsv_add(root->fs_info->chunk_root,
4434 &root->fs_info->chunk_block_rsv,
4435 thresh, BTRFS_RESERVE_NO_FLUSH);
4436 if (!ret)
4437 trans->chunk_bytes_reserved += thresh;
4438 }
4439 }
4440
4441 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4442 struct btrfs_root *extent_root, u64 flags, int force)
4443 {
4444 struct btrfs_space_info *space_info;
4445 struct btrfs_fs_info *fs_info = extent_root->fs_info;
4446 int wait_for_alloc = 0;
4447 int ret = 0;
4448
4449 /* Don't re-enter if we're already allocating a chunk */
4450 if (trans->allocating_chunk)
4451 return -ENOSPC;
4452
4453 space_info = __find_space_info(extent_root->fs_info, flags);
4454 if (!space_info) {
4455 ret = update_space_info(extent_root->fs_info, flags,
4456 0, 0, &space_info);
4457 BUG_ON(ret); /* -ENOMEM */
4458 }
4459 BUG_ON(!space_info); /* Logic error */
4460
4461 again:
4462 spin_lock(&space_info->lock);
4463 if (force < space_info->force_alloc)
4464 force = space_info->force_alloc;
4465 if (space_info->full) {
4466 if (should_alloc_chunk(extent_root, space_info, force))
4467 ret = -ENOSPC;
4468 else
4469 ret = 0;
4470 spin_unlock(&space_info->lock);
4471 return ret;
4472 }
4473
4474 if (!should_alloc_chunk(extent_root, space_info, force)) {
4475 spin_unlock(&space_info->lock);
4476 return 0;
4477 } else if (space_info->chunk_alloc) {
4478 wait_for_alloc = 1;
4479 } else {
4480 space_info->chunk_alloc = 1;
4481 }
4482
4483 spin_unlock(&space_info->lock);
4484
4485 mutex_lock(&fs_info->chunk_mutex);
4486
4487 /*
4488 * The chunk_mutex is held throughout the entirety of a chunk
4489 * allocation, so once we've acquired the chunk_mutex we know that the
4490 * other guy is done and we need to recheck and see if we should
4491 * allocate.
4492 */
4493 if (wait_for_alloc) {
4494 mutex_unlock(&fs_info->chunk_mutex);
4495 wait_for_alloc = 0;
4496 goto again;
4497 }
4498
4499 trans->allocating_chunk = true;
4500
4501 /*
4502 * If we have mixed data/metadata chunks we want to make sure we keep
4503 * allocating mixed chunks instead of individual chunks.
4504 */
4505 if (btrfs_mixed_space_info(space_info))
4506 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4507
4508 /*
4509 * if we're doing a data chunk, go ahead and make sure that
4510 * we keep a reasonable number of metadata chunks allocated in the
4511 * FS as well.
4512 */
4513 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4514 fs_info->data_chunk_allocations++;
4515 if (!(fs_info->data_chunk_allocations %
4516 fs_info->metadata_ratio))
4517 force_metadata_allocation(fs_info);
4518 }
4519
4520 /*
4521 * Check if we have enough space in SYSTEM chunk because we may need
4522 * to update devices.
4523 */
4524 check_system_chunk(trans, extent_root, flags);
4525
4526 ret = btrfs_alloc_chunk(trans, extent_root, flags);
4527 trans->allocating_chunk = false;
4528
4529 spin_lock(&space_info->lock);
4530 if (ret < 0 && ret != -ENOSPC)
4531 goto out;
4532 if (ret)
4533 space_info->full = 1;
4534 else
4535 ret = 1;
4536
4537 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4538 out:
4539 space_info->chunk_alloc = 0;
4540 spin_unlock(&space_info->lock);
4541 mutex_unlock(&fs_info->chunk_mutex);
4542 /*
4543 * When we allocate a new chunk we reserve space in the chunk block
4544 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4545 * add new nodes/leafs to it if we end up needing to do it when
4546 * inserting the chunk item and updating device items as part of the
4547 * second phase of chunk allocation, performed by
4548 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4549 * large number of new block groups to create in our transaction
4550 * handle's new_bgs list to avoid exhausting the chunk block reserve
4551 * in extreme cases - like having a single transaction create many new
4552 * block groups when starting to write out the free space caches of all
4553 * the block groups that were made dirty during the lifetime of the
4554 * transaction.
4555 */
4556 if (trans->can_flush_pending_bgs &&
4557 trans->chunk_bytes_reserved >= (u64)SZ_2M) {
4558 btrfs_create_pending_block_groups(trans, trans->root);
4559 btrfs_trans_release_chunk_metadata(trans);
4560 }
4561 return ret;
4562 }
4563
4564 static int can_overcommit(struct btrfs_root *root,
4565 struct btrfs_space_info *space_info, u64 bytes,
4566 enum btrfs_reserve_flush_enum flush)
4567 {
4568 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4569 u64 profile = btrfs_get_alloc_profile(root, 0);
4570 u64 space_size;
4571 u64 avail;
4572 u64 used;
4573
4574 used = space_info->bytes_used + space_info->bytes_reserved +
4575 space_info->bytes_pinned + space_info->bytes_readonly;
4576
4577 /*
4578 * We only want to allow over committing if we have lots of actual space
4579 * free, but if we don't have enough space to handle the global reserve
4580 * space then we could end up having a real enospc problem when trying
4581 * to allocate a chunk or some other such important allocation.
4582 */
4583 spin_lock(&global_rsv->lock);
4584 space_size = calc_global_rsv_need_space(global_rsv);
4585 spin_unlock(&global_rsv->lock);
4586 if (used + space_size >= space_info->total_bytes)
4587 return 0;
4588
4589 used += space_info->bytes_may_use;
4590
4591 spin_lock(&root->fs_info->free_chunk_lock);
4592 avail = root->fs_info->free_chunk_space;
4593 spin_unlock(&root->fs_info->free_chunk_lock);
4594
4595 /*
4596 * If we have dup, raid1 or raid10 then only half of the free
4597 * space is actually useable. For raid56, the space info used
4598 * doesn't include the parity drive, so we don't have to
4599 * change the math
4600 */
4601 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4602 BTRFS_BLOCK_GROUP_RAID1 |
4603 BTRFS_BLOCK_GROUP_RAID10))
4604 avail >>= 1;
4605
4606 /*
4607 * If we aren't flushing all things, let us overcommit up to
4608 * 1/2th of the space. If we can flush, don't let us overcommit
4609 * too much, let it overcommit up to 1/8 of the space.
4610 */
4611 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4612 avail >>= 3;
4613 else
4614 avail >>= 1;
4615
4616 if (used + bytes < space_info->total_bytes + avail)
4617 return 1;
4618 return 0;
4619 }
4620
4621 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4622 unsigned long nr_pages, int nr_items)
4623 {
4624 struct super_block *sb = root->fs_info->sb;
4625
4626 if (down_read_trylock(&sb->s_umount)) {
4627 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4628 up_read(&sb->s_umount);
4629 } else {
4630 /*
4631 * We needn't worry the filesystem going from r/w to r/o though
4632 * we don't acquire ->s_umount mutex, because the filesystem
4633 * should guarantee the delalloc inodes list be empty after
4634 * the filesystem is readonly(all dirty pages are written to
4635 * the disk).
4636 */
4637 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4638 if (!current->journal_info)
4639 btrfs_wait_ordered_roots(root->fs_info, nr_items,
4640 0, (u64)-1);
4641 }
4642 }
4643
4644 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4645 {
4646 u64 bytes;
4647 int nr;
4648
4649 bytes = btrfs_calc_trans_metadata_size(root, 1);
4650 nr = (int)div64_u64(to_reclaim, bytes);
4651 if (!nr)
4652 nr = 1;
4653 return nr;
4654 }
4655
4656 #define EXTENT_SIZE_PER_ITEM SZ_256K
4657
4658 /*
4659 * shrink metadata reservation for delalloc
4660 */
4661 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4662 bool wait_ordered)
4663 {
4664 struct btrfs_block_rsv *block_rsv;
4665 struct btrfs_space_info *space_info;
4666 struct btrfs_trans_handle *trans;
4667 u64 delalloc_bytes;
4668 u64 max_reclaim;
4669 long time_left;
4670 unsigned long nr_pages;
4671 int loops;
4672 int items;
4673 enum btrfs_reserve_flush_enum flush;
4674
4675 /* Calc the number of the pages we need flush for space reservation */
4676 items = calc_reclaim_items_nr(root, to_reclaim);
4677 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4678
4679 trans = (struct btrfs_trans_handle *)current->journal_info;
4680 block_rsv = &root->fs_info->delalloc_block_rsv;
4681 space_info = block_rsv->space_info;
4682
4683 delalloc_bytes = percpu_counter_sum_positive(
4684 &root->fs_info->delalloc_bytes);
4685 if (delalloc_bytes == 0) {
4686 if (trans)
4687 return;
4688 if (wait_ordered)
4689 btrfs_wait_ordered_roots(root->fs_info, items,
4690 0, (u64)-1);
4691 return;
4692 }
4693
4694 loops = 0;
4695 while (delalloc_bytes && loops < 3) {
4696 max_reclaim = min(delalloc_bytes, to_reclaim);
4697 nr_pages = max_reclaim >> PAGE_SHIFT;
4698 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4699 /*
4700 * We need to wait for the async pages to actually start before
4701 * we do anything.
4702 */
4703 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4704 if (!max_reclaim)
4705 goto skip_async;
4706
4707 if (max_reclaim <= nr_pages)
4708 max_reclaim = 0;
4709 else
4710 max_reclaim -= nr_pages;
4711
4712 wait_event(root->fs_info->async_submit_wait,
4713 atomic_read(&root->fs_info->async_delalloc_pages) <=
4714 (int)max_reclaim);
4715 skip_async:
4716 if (!trans)
4717 flush = BTRFS_RESERVE_FLUSH_ALL;
4718 else
4719 flush = BTRFS_RESERVE_NO_FLUSH;
4720 spin_lock(&space_info->lock);
4721 if (can_overcommit(root, space_info, orig, flush)) {
4722 spin_unlock(&space_info->lock);
4723 break;
4724 }
4725 spin_unlock(&space_info->lock);
4726
4727 loops++;
4728 if (wait_ordered && !trans) {
4729 btrfs_wait_ordered_roots(root->fs_info, items,
4730 0, (u64)-1);
4731 } else {
4732 time_left = schedule_timeout_killable(1);
4733 if (time_left)
4734 break;
4735 }
4736 delalloc_bytes = percpu_counter_sum_positive(
4737 &root->fs_info->delalloc_bytes);
4738 }
4739 }
4740
4741 /**
4742 * maybe_commit_transaction - possibly commit the transaction if its ok to
4743 * @root - the root we're allocating for
4744 * @bytes - the number of bytes we want to reserve
4745 * @force - force the commit
4746 *
4747 * This will check to make sure that committing the transaction will actually
4748 * get us somewhere and then commit the transaction if it does. Otherwise it
4749 * will return -ENOSPC.
4750 */
4751 static int may_commit_transaction(struct btrfs_root *root,
4752 struct btrfs_space_info *space_info,
4753 u64 bytes, int force)
4754 {
4755 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4756 struct btrfs_trans_handle *trans;
4757
4758 trans = (struct btrfs_trans_handle *)current->journal_info;
4759 if (trans)
4760 return -EAGAIN;
4761
4762 if (force)
4763 goto commit;
4764
4765 /* See if there is enough pinned space to make this reservation */
4766 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4767 bytes) >= 0)
4768 goto commit;
4769
4770 /*
4771 * See if there is some space in the delayed insertion reservation for
4772 * this reservation.
4773 */
4774 if (space_info != delayed_rsv->space_info)
4775 return -ENOSPC;
4776
4777 spin_lock(&delayed_rsv->lock);
4778 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4779 bytes - delayed_rsv->size) >= 0) {
4780 spin_unlock(&delayed_rsv->lock);
4781 return -ENOSPC;
4782 }
4783 spin_unlock(&delayed_rsv->lock);
4784
4785 commit:
4786 trans = btrfs_join_transaction(root);
4787 if (IS_ERR(trans))
4788 return -ENOSPC;
4789
4790 return btrfs_commit_transaction(trans, root);
4791 }
4792
4793 enum flush_state {
4794 FLUSH_DELAYED_ITEMS_NR = 1,
4795 FLUSH_DELAYED_ITEMS = 2,
4796 FLUSH_DELALLOC = 3,
4797 FLUSH_DELALLOC_WAIT = 4,
4798 ALLOC_CHUNK = 5,
4799 COMMIT_TRANS = 6,
4800 };
4801
4802 static int flush_space(struct btrfs_root *root,
4803 struct btrfs_space_info *space_info, u64 num_bytes,
4804 u64 orig_bytes, int state)
4805 {
4806 struct btrfs_trans_handle *trans;
4807 int nr;
4808 int ret = 0;
4809
4810 switch (state) {
4811 case FLUSH_DELAYED_ITEMS_NR:
4812 case FLUSH_DELAYED_ITEMS:
4813 if (state == FLUSH_DELAYED_ITEMS_NR)
4814 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4815 else
4816 nr = -1;
4817
4818 trans = btrfs_join_transaction(root);
4819 if (IS_ERR(trans)) {
4820 ret = PTR_ERR(trans);
4821 break;
4822 }
4823 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4824 btrfs_end_transaction(trans, root);
4825 break;
4826 case FLUSH_DELALLOC:
4827 case FLUSH_DELALLOC_WAIT:
4828 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4829 state == FLUSH_DELALLOC_WAIT);
4830 break;
4831 case ALLOC_CHUNK:
4832 trans = btrfs_join_transaction(root);
4833 if (IS_ERR(trans)) {
4834 ret = PTR_ERR(trans);
4835 break;
4836 }
4837 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4838 btrfs_get_alloc_profile(root, 0),
4839 CHUNK_ALLOC_NO_FORCE);
4840 btrfs_end_transaction(trans, root);
4841 if (ret == -ENOSPC)
4842 ret = 0;
4843 break;
4844 case COMMIT_TRANS:
4845 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4846 break;
4847 default:
4848 ret = -ENOSPC;
4849 break;
4850 }
4851
4852 return ret;
4853 }
4854
4855 static inline u64
4856 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4857 struct btrfs_space_info *space_info)
4858 {
4859 u64 used;
4860 u64 expected;
4861 u64 to_reclaim;
4862
4863 to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
4864 spin_lock(&space_info->lock);
4865 if (can_overcommit(root, space_info, to_reclaim,
4866 BTRFS_RESERVE_FLUSH_ALL)) {
4867 to_reclaim = 0;
4868 goto out;
4869 }
4870
4871 used = space_info->bytes_used + space_info->bytes_reserved +
4872 space_info->bytes_pinned + space_info->bytes_readonly +
4873 space_info->bytes_may_use;
4874 if (can_overcommit(root, space_info, SZ_1M, BTRFS_RESERVE_FLUSH_ALL))
4875 expected = div_factor_fine(space_info->total_bytes, 95);
4876 else
4877 expected = div_factor_fine(space_info->total_bytes, 90);
4878
4879 if (used > expected)
4880 to_reclaim = used - expected;
4881 else
4882 to_reclaim = 0;
4883 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4884 space_info->bytes_reserved);
4885 out:
4886 spin_unlock(&space_info->lock);
4887
4888 return to_reclaim;
4889 }
4890
4891 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4892 struct btrfs_fs_info *fs_info, u64 used)
4893 {
4894 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4895
4896 /* If we're just plain full then async reclaim just slows us down. */
4897 if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
4898 return 0;
4899
4900 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
4901 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4902 }
4903
4904 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4905 struct btrfs_fs_info *fs_info,
4906 int flush_state)
4907 {
4908 u64 used;
4909
4910 spin_lock(&space_info->lock);
4911 /*
4912 * We run out of space and have not got any free space via flush_space,
4913 * so don't bother doing async reclaim.
4914 */
4915 if (flush_state > COMMIT_TRANS && space_info->full) {
4916 spin_unlock(&space_info->lock);
4917 return 0;
4918 }
4919
4920 used = space_info->bytes_used + space_info->bytes_reserved +
4921 space_info->bytes_pinned + space_info->bytes_readonly +
4922 space_info->bytes_may_use;
4923 if (need_do_async_reclaim(space_info, fs_info, used)) {
4924 spin_unlock(&space_info->lock);
4925 return 1;
4926 }
4927 spin_unlock(&space_info->lock);
4928
4929 return 0;
4930 }
4931
4932 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4933 {
4934 struct btrfs_fs_info *fs_info;
4935 struct btrfs_space_info *space_info;
4936 u64 to_reclaim;
4937 int flush_state;
4938
4939 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4940 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4941
4942 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4943 space_info);
4944 if (!to_reclaim)
4945 return;
4946
4947 flush_state = FLUSH_DELAYED_ITEMS_NR;
4948 do {
4949 flush_space(fs_info->fs_root, space_info, to_reclaim,
4950 to_reclaim, flush_state);
4951 flush_state++;
4952 if (!btrfs_need_do_async_reclaim(space_info, fs_info,
4953 flush_state))
4954 return;
4955 } while (flush_state < COMMIT_TRANS);
4956 }
4957
4958 void btrfs_init_async_reclaim_work(struct work_struct *work)
4959 {
4960 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4961 }
4962
4963 /**
4964 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4965 * @root - the root we're allocating for
4966 * @block_rsv - the block_rsv we're allocating for
4967 * @orig_bytes - the number of bytes we want
4968 * @flush - whether or not we can flush to make our reservation
4969 *
4970 * This will reserve orgi_bytes number of bytes from the space info associated
4971 * with the block_rsv. If there is not enough space it will make an attempt to
4972 * flush out space to make room. It will do this by flushing delalloc if
4973 * possible or committing the transaction. If flush is 0 then no attempts to
4974 * regain reservations will be made and this will fail if there is not enough
4975 * space already.
4976 */
4977 static int reserve_metadata_bytes(struct btrfs_root *root,
4978 struct btrfs_block_rsv *block_rsv,
4979 u64 orig_bytes,
4980 enum btrfs_reserve_flush_enum flush)
4981 {
4982 struct btrfs_space_info *space_info = block_rsv->space_info;
4983 u64 used;
4984 u64 num_bytes = orig_bytes;
4985 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4986 int ret = 0;
4987 bool flushing = false;
4988
4989 again:
4990 ret = 0;
4991 spin_lock(&space_info->lock);
4992 /*
4993 * We only want to wait if somebody other than us is flushing and we
4994 * are actually allowed to flush all things.
4995 */
4996 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4997 space_info->flush) {
4998 spin_unlock(&space_info->lock);
4999 /*
5000 * If we have a trans handle we can't wait because the flusher
5001 * may have to commit the transaction, which would mean we would
5002 * deadlock since we are waiting for the flusher to finish, but
5003 * hold the current transaction open.
5004 */
5005 if (current->journal_info)
5006 return -EAGAIN;
5007 ret = wait_event_killable(space_info->wait, !space_info->flush);
5008 /* Must have been killed, return */
5009 if (ret)
5010 return -EINTR;
5011
5012 spin_lock(&space_info->lock);
5013 }
5014
5015 ret = -ENOSPC;
5016 used = space_info->bytes_used + space_info->bytes_reserved +
5017 space_info->bytes_pinned + space_info->bytes_readonly +
5018 space_info->bytes_may_use;
5019
5020 /*
5021 * The idea here is that we've not already over-reserved the block group
5022 * then we can go ahead and save our reservation first and then start
5023 * flushing if we need to. Otherwise if we've already overcommitted
5024 * lets start flushing stuff first and then come back and try to make
5025 * our reservation.
5026 */
5027 if (used <= space_info->total_bytes) {
5028 if (used + orig_bytes <= space_info->total_bytes) {
5029 space_info->bytes_may_use += orig_bytes;
5030 trace_btrfs_space_reservation(root->fs_info,
5031 "space_info", space_info->flags, orig_bytes, 1);
5032 ret = 0;
5033 } else {
5034 /*
5035 * Ok set num_bytes to orig_bytes since we aren't
5036 * overocmmitted, this way we only try and reclaim what
5037 * we need.
5038 */
5039 num_bytes = orig_bytes;
5040 }
5041 } else {
5042 /*
5043 * Ok we're over committed, set num_bytes to the overcommitted
5044 * amount plus the amount of bytes that we need for this
5045 * reservation.
5046 */
5047 num_bytes = used - space_info->total_bytes +
5048 (orig_bytes * 2);
5049 }
5050
5051 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
5052 space_info->bytes_may_use += orig_bytes;
5053 trace_btrfs_space_reservation(root->fs_info, "space_info",
5054 space_info->flags, orig_bytes,
5055 1);
5056 ret = 0;
5057 }
5058
5059 /*
5060 * Couldn't make our reservation, save our place so while we're trying
5061 * to reclaim space we can actually use it instead of somebody else
5062 * stealing it from us.
5063 *
5064 * We make the other tasks wait for the flush only when we can flush
5065 * all things.
5066 */
5067 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5068 flushing = true;
5069 space_info->flush = 1;
5070 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5071 used += orig_bytes;
5072 /*
5073 * We will do the space reservation dance during log replay,
5074 * which means we won't have fs_info->fs_root set, so don't do
5075 * the async reclaim as we will panic.
5076 */
5077 if (!root->fs_info->log_root_recovering &&
5078 need_do_async_reclaim(space_info, root->fs_info, used) &&
5079 !work_busy(&root->fs_info->async_reclaim_work))
5080 queue_work(system_unbound_wq,
5081 &root->fs_info->async_reclaim_work);
5082 }
5083 spin_unlock(&space_info->lock);
5084
5085 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5086 goto out;
5087
5088 ret = flush_space(root, space_info, num_bytes, orig_bytes,
5089 flush_state);
5090 flush_state++;
5091
5092 /*
5093 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
5094 * would happen. So skip delalloc flush.
5095 */
5096 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
5097 (flush_state == FLUSH_DELALLOC ||
5098 flush_state == FLUSH_DELALLOC_WAIT))
5099 flush_state = ALLOC_CHUNK;
5100
5101 if (!ret)
5102 goto again;
5103 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
5104 flush_state < COMMIT_TRANS)
5105 goto again;
5106 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
5107 flush_state <= COMMIT_TRANS)
5108 goto again;
5109
5110 out:
5111 if (ret == -ENOSPC &&
5112 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5113 struct btrfs_block_rsv *global_rsv =
5114 &root->fs_info->global_block_rsv;
5115
5116 if (block_rsv != global_rsv &&
5117 !block_rsv_use_bytes(global_rsv, orig_bytes))
5118 ret = 0;
5119 }
5120 if (ret == -ENOSPC)
5121 trace_btrfs_space_reservation(root->fs_info,
5122 "space_info:enospc",
5123 space_info->flags, orig_bytes, 1);
5124 if (flushing) {
5125 spin_lock(&space_info->lock);
5126 space_info->flush = 0;
5127 wake_up_all(&space_info->wait);
5128 spin_unlock(&space_info->lock);
5129 }
5130 return ret;
5131 }
5132
5133 static struct btrfs_block_rsv *get_block_rsv(
5134 const struct btrfs_trans_handle *trans,
5135 const struct btrfs_root *root)
5136 {
5137 struct btrfs_block_rsv *block_rsv = NULL;
5138
5139 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5140 (root == root->fs_info->csum_root && trans->adding_csums) ||
5141 (root == root->fs_info->uuid_root))
5142 block_rsv = trans->block_rsv;
5143
5144 if (!block_rsv)
5145 block_rsv = root->block_rsv;
5146
5147 if (!block_rsv)
5148 block_rsv = &root->fs_info->empty_block_rsv;
5149
5150 return block_rsv;
5151 }
5152
5153 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5154 u64 num_bytes)
5155 {
5156 int ret = -ENOSPC;
5157 spin_lock(&block_rsv->lock);
5158 if (block_rsv->reserved >= num_bytes) {
5159 block_rsv->reserved -= num_bytes;
5160 if (block_rsv->reserved < block_rsv->size)
5161 block_rsv->full = 0;
5162 ret = 0;
5163 }
5164 spin_unlock(&block_rsv->lock);
5165 return ret;
5166 }
5167
5168 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5169 u64 num_bytes, int update_size)
5170 {
5171 spin_lock(&block_rsv->lock);
5172 block_rsv->reserved += num_bytes;
5173 if (update_size)
5174 block_rsv->size += num_bytes;
5175 else if (block_rsv->reserved >= block_rsv->size)
5176 block_rsv->full = 1;
5177 spin_unlock(&block_rsv->lock);
5178 }
5179
5180 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5181 struct btrfs_block_rsv *dest, u64 num_bytes,
5182 int min_factor)
5183 {
5184 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5185 u64 min_bytes;
5186
5187 if (global_rsv->space_info != dest->space_info)
5188 return -ENOSPC;
5189
5190 spin_lock(&global_rsv->lock);
5191 min_bytes = div_factor(global_rsv->size, min_factor);
5192 if (global_rsv->reserved < min_bytes + num_bytes) {
5193 spin_unlock(&global_rsv->lock);
5194 return -ENOSPC;
5195 }
5196 global_rsv->reserved -= num_bytes;
5197 if (global_rsv->reserved < global_rsv->size)
5198 global_rsv->full = 0;
5199 spin_unlock(&global_rsv->lock);
5200
5201 block_rsv_add_bytes(dest, num_bytes, 1);
5202 return 0;
5203 }
5204
5205 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5206 struct btrfs_block_rsv *block_rsv,
5207 struct btrfs_block_rsv *dest, u64 num_bytes)
5208 {
5209 struct btrfs_space_info *space_info = block_rsv->space_info;
5210
5211 spin_lock(&block_rsv->lock);
5212 if (num_bytes == (u64)-1)
5213 num_bytes = block_rsv->size;
5214 block_rsv->size -= num_bytes;
5215 if (block_rsv->reserved >= block_rsv->size) {
5216 num_bytes = block_rsv->reserved - block_rsv->size;
5217 block_rsv->reserved = block_rsv->size;
5218 block_rsv->full = 1;
5219 } else {
5220 num_bytes = 0;
5221 }
5222 spin_unlock(&block_rsv->lock);
5223
5224 if (num_bytes > 0) {
5225 if (dest) {
5226 spin_lock(&dest->lock);
5227 if (!dest->full) {
5228 u64 bytes_to_add;
5229
5230 bytes_to_add = dest->size - dest->reserved;
5231 bytes_to_add = min(num_bytes, bytes_to_add);
5232 dest->reserved += bytes_to_add;
5233 if (dest->reserved >= dest->size)
5234 dest->full = 1;
5235 num_bytes -= bytes_to_add;
5236 }
5237 spin_unlock(&dest->lock);
5238 }
5239 if (num_bytes) {
5240 spin_lock(&space_info->lock);
5241 space_info->bytes_may_use -= num_bytes;
5242 trace_btrfs_space_reservation(fs_info, "space_info",
5243 space_info->flags, num_bytes, 0);
5244 spin_unlock(&space_info->lock);
5245 }
5246 }
5247 }
5248
5249 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
5250 struct btrfs_block_rsv *dst, u64 num_bytes)
5251 {
5252 int ret;
5253
5254 ret = block_rsv_use_bytes(src, num_bytes);
5255 if (ret)
5256 return ret;
5257
5258 block_rsv_add_bytes(dst, num_bytes, 1);
5259 return 0;
5260 }
5261
5262 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5263 {
5264 memset(rsv, 0, sizeof(*rsv));
5265 spin_lock_init(&rsv->lock);
5266 rsv->type = type;
5267 }
5268
5269 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
5270 unsigned short type)
5271 {
5272 struct btrfs_block_rsv *block_rsv;
5273 struct btrfs_fs_info *fs_info = root->fs_info;
5274
5275 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5276 if (!block_rsv)
5277 return NULL;
5278
5279 btrfs_init_block_rsv(block_rsv, type);
5280 block_rsv->space_info = __find_space_info(fs_info,
5281 BTRFS_BLOCK_GROUP_METADATA);
5282 return block_rsv;
5283 }
5284
5285 void btrfs_free_block_rsv(struct btrfs_root *root,
5286 struct btrfs_block_rsv *rsv)
5287 {
5288 if (!rsv)
5289 return;
5290 btrfs_block_rsv_release(root, rsv, (u64)-1);
5291 kfree(rsv);
5292 }
5293
5294 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5295 {
5296 kfree(rsv);
5297 }
5298
5299 int btrfs_block_rsv_add(struct btrfs_root *root,
5300 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5301 enum btrfs_reserve_flush_enum flush)
5302 {
5303 int ret;
5304
5305 if (num_bytes == 0)
5306 return 0;
5307
5308 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5309 if (!ret) {
5310 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5311 return 0;
5312 }
5313
5314 return ret;
5315 }
5316
5317 int btrfs_block_rsv_check(struct btrfs_root *root,
5318 struct btrfs_block_rsv *block_rsv, int min_factor)
5319 {
5320 u64 num_bytes = 0;
5321 int ret = -ENOSPC;
5322
5323 if (!block_rsv)
5324 return 0;
5325
5326 spin_lock(&block_rsv->lock);
5327 num_bytes = div_factor(block_rsv->size, min_factor);
5328 if (block_rsv->reserved >= num_bytes)
5329 ret = 0;
5330 spin_unlock(&block_rsv->lock);
5331
5332 return ret;
5333 }
5334
5335 int btrfs_block_rsv_refill(struct btrfs_root *root,
5336 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5337 enum btrfs_reserve_flush_enum flush)
5338 {
5339 u64 num_bytes = 0;
5340 int ret = -ENOSPC;
5341
5342 if (!block_rsv)
5343 return 0;
5344
5345 spin_lock(&block_rsv->lock);
5346 num_bytes = min_reserved;
5347 if (block_rsv->reserved >= num_bytes)
5348 ret = 0;
5349 else
5350 num_bytes -= block_rsv->reserved;
5351 spin_unlock(&block_rsv->lock);
5352
5353 if (!ret)
5354 return 0;
5355
5356 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5357 if (!ret) {
5358 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5359 return 0;
5360 }
5361
5362 return ret;
5363 }
5364
5365 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
5366 struct btrfs_block_rsv *dst_rsv,
5367 u64 num_bytes)
5368 {
5369 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5370 }
5371
5372 void btrfs_block_rsv_release(struct btrfs_root *root,
5373 struct btrfs_block_rsv *block_rsv,
5374 u64 num_bytes)
5375 {
5376 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5377 if (global_rsv == block_rsv ||
5378 block_rsv->space_info != global_rsv->space_info)
5379 global_rsv = NULL;
5380 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
5381 num_bytes);
5382 }
5383
5384 /*
5385 * helper to calculate size of global block reservation.
5386 * the desired value is sum of space used by extent tree,
5387 * checksum tree and root tree
5388 */
5389 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
5390 {
5391 struct btrfs_space_info *sinfo;
5392 u64 num_bytes;
5393 u64 meta_used;
5394 u64 data_used;
5395 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
5396
5397 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
5398 spin_lock(&sinfo->lock);
5399 data_used = sinfo->bytes_used;
5400 spin_unlock(&sinfo->lock);
5401
5402 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5403 spin_lock(&sinfo->lock);
5404 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
5405 data_used = 0;
5406 meta_used = sinfo->bytes_used;
5407 spin_unlock(&sinfo->lock);
5408
5409 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
5410 csum_size * 2;
5411 num_bytes += div_u64(data_used + meta_used, 50);
5412
5413 if (num_bytes * 3 > meta_used)
5414 num_bytes = div_u64(meta_used, 3);
5415
5416 return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
5417 }
5418
5419 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5420 {
5421 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5422 struct btrfs_space_info *sinfo = block_rsv->space_info;
5423 u64 num_bytes;
5424
5425 num_bytes = calc_global_metadata_size(fs_info);
5426
5427 spin_lock(&sinfo->lock);
5428 spin_lock(&block_rsv->lock);
5429
5430 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5431
5432 if (block_rsv->reserved < block_rsv->size) {
5433 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
5434 sinfo->bytes_reserved + sinfo->bytes_readonly +
5435 sinfo->bytes_may_use;
5436 if (sinfo->total_bytes > num_bytes) {
5437 num_bytes = sinfo->total_bytes - num_bytes;
5438 num_bytes = min(num_bytes,
5439 block_rsv->size - block_rsv->reserved);
5440 block_rsv->reserved += num_bytes;
5441 sinfo->bytes_may_use += num_bytes;
5442 trace_btrfs_space_reservation(fs_info, "space_info",
5443 sinfo->flags, num_bytes,
5444 1);
5445 }
5446 } else if (block_rsv->reserved > block_rsv->size) {
5447 num_bytes = block_rsv->reserved - block_rsv->size;
5448 sinfo->bytes_may_use -= num_bytes;
5449 trace_btrfs_space_reservation(fs_info, "space_info",
5450 sinfo->flags, num_bytes, 0);
5451 block_rsv->reserved = block_rsv->size;
5452 }
5453
5454 if (block_rsv->reserved == block_rsv->size)
5455 block_rsv->full = 1;
5456 else
5457 block_rsv->full = 0;
5458
5459 spin_unlock(&block_rsv->lock);
5460 spin_unlock(&sinfo->lock);
5461 }
5462
5463 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5464 {
5465 struct btrfs_space_info *space_info;
5466
5467 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5468 fs_info->chunk_block_rsv.space_info = space_info;
5469
5470 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5471 fs_info->global_block_rsv.space_info = space_info;
5472 fs_info->delalloc_block_rsv.space_info = space_info;
5473 fs_info->trans_block_rsv.space_info = space_info;
5474 fs_info->empty_block_rsv.space_info = space_info;
5475 fs_info->delayed_block_rsv.space_info = space_info;
5476
5477 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5478 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5479 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5480 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5481 if (fs_info->quota_root)
5482 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5483 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5484
5485 update_global_block_rsv(fs_info);
5486 }
5487
5488 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5489 {
5490 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5491 (u64)-1);
5492 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5493 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5494 WARN_ON(fs_info->trans_block_rsv.size > 0);
5495 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5496 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5497 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5498 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5499 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5500 }
5501
5502 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5503 struct btrfs_root *root)
5504 {
5505 if (!trans->block_rsv)
5506 return;
5507
5508 if (!trans->bytes_reserved)
5509 return;
5510
5511 trace_btrfs_space_reservation(root->fs_info, "transaction",
5512 trans->transid, trans->bytes_reserved, 0);
5513 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
5514 trans->bytes_reserved = 0;
5515 }
5516
5517 /*
5518 * To be called after all the new block groups attached to the transaction
5519 * handle have been created (btrfs_create_pending_block_groups()).
5520 */
5521 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5522 {
5523 struct btrfs_fs_info *fs_info = trans->root->fs_info;
5524
5525 if (!trans->chunk_bytes_reserved)
5526 return;
5527
5528 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5529
5530 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5531 trans->chunk_bytes_reserved);
5532 trans->chunk_bytes_reserved = 0;
5533 }
5534
5535 /* Can only return 0 or -ENOSPC */
5536 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5537 struct inode *inode)
5538 {
5539 struct btrfs_root *root = BTRFS_I(inode)->root;
5540 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
5541 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5542
5543 /*
5544 * We need to hold space in order to delete our orphan item once we've
5545 * added it, so this takes the reservation so we can release it later
5546 * when we are truly done with the orphan item.
5547 */
5548 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5549 trace_btrfs_space_reservation(root->fs_info, "orphan",
5550 btrfs_ino(inode), num_bytes, 1);
5551 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5552 }
5553
5554 void btrfs_orphan_release_metadata(struct inode *inode)
5555 {
5556 struct btrfs_root *root = BTRFS_I(inode)->root;
5557 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5558 trace_btrfs_space_reservation(root->fs_info, "orphan",
5559 btrfs_ino(inode), num_bytes, 0);
5560 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
5561 }
5562
5563 /*
5564 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5565 * root: the root of the parent directory
5566 * rsv: block reservation
5567 * items: the number of items that we need do reservation
5568 * qgroup_reserved: used to return the reserved size in qgroup
5569 *
5570 * This function is used to reserve the space for snapshot/subvolume
5571 * creation and deletion. Those operations are different with the
5572 * common file/directory operations, they change two fs/file trees
5573 * and root tree, the number of items that the qgroup reserves is
5574 * different with the free space reservation. So we can not use
5575 * the space reseravtion mechanism in start_transaction().
5576 */
5577 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5578 struct btrfs_block_rsv *rsv,
5579 int items,
5580 u64 *qgroup_reserved,
5581 bool use_global_rsv)
5582 {
5583 u64 num_bytes;
5584 int ret;
5585 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5586
5587 if (root->fs_info->quota_enabled) {
5588 /* One for parent inode, two for dir entries */
5589 num_bytes = 3 * root->nodesize;
5590 ret = btrfs_qgroup_reserve_meta(root, num_bytes);
5591 if (ret)
5592 return ret;
5593 } else {
5594 num_bytes = 0;
5595 }
5596
5597 *qgroup_reserved = num_bytes;
5598
5599 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5600 rsv->space_info = __find_space_info(root->fs_info,
5601 BTRFS_BLOCK_GROUP_METADATA);
5602 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5603 BTRFS_RESERVE_FLUSH_ALL);
5604
5605 if (ret == -ENOSPC && use_global_rsv)
5606 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
5607
5608 if (ret && *qgroup_reserved)
5609 btrfs_qgroup_free_meta(root, *qgroup_reserved);
5610
5611 return ret;
5612 }
5613
5614 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5615 struct btrfs_block_rsv *rsv,
5616 u64 qgroup_reserved)
5617 {
5618 btrfs_block_rsv_release(root, rsv, (u64)-1);
5619 }
5620
5621 /**
5622 * drop_outstanding_extent - drop an outstanding extent
5623 * @inode: the inode we're dropping the extent for
5624 * @num_bytes: the number of bytes we're relaseing.
5625 *
5626 * This is called when we are freeing up an outstanding extent, either called
5627 * after an error or after an extent is written. This will return the number of
5628 * reserved extents that need to be freed. This must be called with
5629 * BTRFS_I(inode)->lock held.
5630 */
5631 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5632 {
5633 unsigned drop_inode_space = 0;
5634 unsigned dropped_extents = 0;
5635 unsigned num_extents = 0;
5636
5637 num_extents = (unsigned)div64_u64(num_bytes +
5638 BTRFS_MAX_EXTENT_SIZE - 1,
5639 BTRFS_MAX_EXTENT_SIZE);
5640 ASSERT(num_extents);
5641 ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5642 BTRFS_I(inode)->outstanding_extents -= num_extents;
5643
5644 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5645 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5646 &BTRFS_I(inode)->runtime_flags))
5647 drop_inode_space = 1;
5648
5649 /*
5650 * If we have more or the same amount of outsanding extents than we have
5651 * reserved then we need to leave the reserved extents count alone.
5652 */
5653 if (BTRFS_I(inode)->outstanding_extents >=
5654 BTRFS_I(inode)->reserved_extents)
5655 return drop_inode_space;
5656
5657 dropped_extents = BTRFS_I(inode)->reserved_extents -
5658 BTRFS_I(inode)->outstanding_extents;
5659 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5660 return dropped_extents + drop_inode_space;
5661 }
5662
5663 /**
5664 * calc_csum_metadata_size - return the amount of metada space that must be
5665 * reserved/free'd for the given bytes.
5666 * @inode: the inode we're manipulating
5667 * @num_bytes: the number of bytes in question
5668 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5669 *
5670 * This adjusts the number of csum_bytes in the inode and then returns the
5671 * correct amount of metadata that must either be reserved or freed. We
5672 * calculate how many checksums we can fit into one leaf and then divide the
5673 * number of bytes that will need to be checksumed by this value to figure out
5674 * how many checksums will be required. If we are adding bytes then the number
5675 * may go up and we will return the number of additional bytes that must be
5676 * reserved. If it is going down we will return the number of bytes that must
5677 * be freed.
5678 *
5679 * This must be called with BTRFS_I(inode)->lock held.
5680 */
5681 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5682 int reserve)
5683 {
5684 struct btrfs_root *root = BTRFS_I(inode)->root;
5685 u64 old_csums, num_csums;
5686
5687 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5688 BTRFS_I(inode)->csum_bytes == 0)
5689 return 0;
5690
5691 old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5692 if (reserve)
5693 BTRFS_I(inode)->csum_bytes += num_bytes;
5694 else
5695 BTRFS_I(inode)->csum_bytes -= num_bytes;
5696 num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5697
5698 /* No change, no need to reserve more */
5699 if (old_csums == num_csums)
5700 return 0;
5701
5702 if (reserve)
5703 return btrfs_calc_trans_metadata_size(root,
5704 num_csums - old_csums);
5705
5706 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5707 }
5708
5709 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5710 {
5711 struct btrfs_root *root = BTRFS_I(inode)->root;
5712 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5713 u64 to_reserve = 0;
5714 u64 csum_bytes;
5715 unsigned nr_extents = 0;
5716 int extra_reserve = 0;
5717 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5718 int ret = 0;
5719 bool delalloc_lock = true;
5720 u64 to_free = 0;
5721 unsigned dropped;
5722
5723 /* If we are a free space inode we need to not flush since we will be in
5724 * the middle of a transaction commit. We also don't need the delalloc
5725 * mutex since we won't race with anybody. We need this mostly to make
5726 * lockdep shut its filthy mouth.
5727 */
5728 if (btrfs_is_free_space_inode(inode)) {
5729 flush = BTRFS_RESERVE_NO_FLUSH;
5730 delalloc_lock = false;
5731 }
5732
5733 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5734 btrfs_transaction_in_commit(root->fs_info))
5735 schedule_timeout(1);
5736
5737 if (delalloc_lock)
5738 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5739
5740 num_bytes = ALIGN(num_bytes, root->sectorsize);
5741
5742 spin_lock(&BTRFS_I(inode)->lock);
5743 nr_extents = (unsigned)div64_u64(num_bytes +
5744 BTRFS_MAX_EXTENT_SIZE - 1,
5745 BTRFS_MAX_EXTENT_SIZE);
5746 BTRFS_I(inode)->outstanding_extents += nr_extents;
5747 nr_extents = 0;
5748
5749 if (BTRFS_I(inode)->outstanding_extents >
5750 BTRFS_I(inode)->reserved_extents)
5751 nr_extents = BTRFS_I(inode)->outstanding_extents -
5752 BTRFS_I(inode)->reserved_extents;
5753
5754 /*
5755 * Add an item to reserve for updating the inode when we complete the
5756 * delalloc io.
5757 */
5758 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5759 &BTRFS_I(inode)->runtime_flags)) {
5760 nr_extents++;
5761 extra_reserve = 1;
5762 }
5763
5764 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5765 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5766 csum_bytes = BTRFS_I(inode)->csum_bytes;
5767 spin_unlock(&BTRFS_I(inode)->lock);
5768
5769 if (root->fs_info->quota_enabled) {
5770 ret = btrfs_qgroup_reserve_meta(root,
5771 nr_extents * root->nodesize);
5772 if (ret)
5773 goto out_fail;
5774 }
5775
5776 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5777 if (unlikely(ret)) {
5778 btrfs_qgroup_free_meta(root, nr_extents * root->nodesize);
5779 goto out_fail;
5780 }
5781
5782 spin_lock(&BTRFS_I(inode)->lock);
5783 if (extra_reserve) {
5784 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5785 &BTRFS_I(inode)->runtime_flags);
5786 nr_extents--;
5787 }
5788 BTRFS_I(inode)->reserved_extents += nr_extents;
5789 spin_unlock(&BTRFS_I(inode)->lock);
5790
5791 if (delalloc_lock)
5792 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5793
5794 if (to_reserve)
5795 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5796 btrfs_ino(inode), to_reserve, 1);
5797 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5798
5799 return 0;
5800
5801 out_fail:
5802 spin_lock(&BTRFS_I(inode)->lock);
5803 dropped = drop_outstanding_extent(inode, num_bytes);
5804 /*
5805 * If the inodes csum_bytes is the same as the original
5806 * csum_bytes then we know we haven't raced with any free()ers
5807 * so we can just reduce our inodes csum bytes and carry on.
5808 */
5809 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5810 calc_csum_metadata_size(inode, num_bytes, 0);
5811 } else {
5812 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5813 u64 bytes;
5814
5815 /*
5816 * This is tricky, but first we need to figure out how much we
5817 * free'd from any free-ers that occurred during this
5818 * reservation, so we reset ->csum_bytes to the csum_bytes
5819 * before we dropped our lock, and then call the free for the
5820 * number of bytes that were freed while we were trying our
5821 * reservation.
5822 */
5823 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5824 BTRFS_I(inode)->csum_bytes = csum_bytes;
5825 to_free = calc_csum_metadata_size(inode, bytes, 0);
5826
5827
5828 /*
5829 * Now we need to see how much we would have freed had we not
5830 * been making this reservation and our ->csum_bytes were not
5831 * artificially inflated.
5832 */
5833 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5834 bytes = csum_bytes - orig_csum_bytes;
5835 bytes = calc_csum_metadata_size(inode, bytes, 0);
5836
5837 /*
5838 * Now reset ->csum_bytes to what it should be. If bytes is
5839 * more than to_free then we would have free'd more space had we
5840 * not had an artificially high ->csum_bytes, so we need to free
5841 * the remainder. If bytes is the same or less then we don't
5842 * need to do anything, the other free-ers did the correct
5843 * thing.
5844 */
5845 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5846 if (bytes > to_free)
5847 to_free = bytes - to_free;
5848 else
5849 to_free = 0;
5850 }
5851 spin_unlock(&BTRFS_I(inode)->lock);
5852 if (dropped)
5853 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5854
5855 if (to_free) {
5856 btrfs_block_rsv_release(root, block_rsv, to_free);
5857 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5858 btrfs_ino(inode), to_free, 0);
5859 }
5860 if (delalloc_lock)
5861 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5862 return ret;
5863 }
5864
5865 /**
5866 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5867 * @inode: the inode to release the reservation for
5868 * @num_bytes: the number of bytes we're releasing
5869 *
5870 * This will release the metadata reservation for an inode. This can be called
5871 * once we complete IO for a given set of bytes to release their metadata
5872 * reservations.
5873 */
5874 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5875 {
5876 struct btrfs_root *root = BTRFS_I(inode)->root;
5877 u64 to_free = 0;
5878 unsigned dropped;
5879
5880 num_bytes = ALIGN(num_bytes, root->sectorsize);
5881 spin_lock(&BTRFS_I(inode)->lock);
5882 dropped = drop_outstanding_extent(inode, num_bytes);
5883
5884 if (num_bytes)
5885 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5886 spin_unlock(&BTRFS_I(inode)->lock);
5887 if (dropped > 0)
5888 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5889
5890 if (btrfs_test_is_dummy_root(root))
5891 return;
5892
5893 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5894 btrfs_ino(inode), to_free, 0);
5895
5896 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5897 to_free);
5898 }
5899
5900 /**
5901 * btrfs_delalloc_reserve_space - reserve data and metadata space for
5902 * delalloc
5903 * @inode: inode we're writing to
5904 * @start: start range we are writing to
5905 * @len: how long the range we are writing to
5906 *
5907 * TODO: This function will finally replace old btrfs_delalloc_reserve_space()
5908 *
5909 * This will do the following things
5910 *
5911 * o reserve space in data space info for num bytes
5912 * and reserve precious corresponding qgroup space
5913 * (Done in check_data_free_space)
5914 *
5915 * o reserve space for metadata space, based on the number of outstanding
5916 * extents and how much csums will be needed
5917 * also reserve metadata space in a per root over-reserve method.
5918 * o add to the inodes->delalloc_bytes
5919 * o add it to the fs_info's delalloc inodes list.
5920 * (Above 3 all done in delalloc_reserve_metadata)
5921 *
5922 * Return 0 for success
5923 * Return <0 for error(-ENOSPC or -EQUOT)
5924 */
5925 int btrfs_delalloc_reserve_space(struct inode *inode, u64 start, u64 len)
5926 {
5927 int ret;
5928
5929 ret = btrfs_check_data_free_space(inode, start, len);
5930 if (ret < 0)
5931 return ret;
5932 ret = btrfs_delalloc_reserve_metadata(inode, len);
5933 if (ret < 0)
5934 btrfs_free_reserved_data_space(inode, start, len);
5935 return ret;
5936 }
5937
5938 /**
5939 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5940 * @inode: inode we're releasing space for
5941 * @start: start position of the space already reserved
5942 * @len: the len of the space already reserved
5943 *
5944 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5945 * called in the case that we don't need the metadata AND data reservations
5946 * anymore. So if there is an error or we insert an inline extent.
5947 *
5948 * This function will release the metadata space that was not used and will
5949 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5950 * list if there are no delalloc bytes left.
5951 * Also it will handle the qgroup reserved space.
5952 */
5953 void btrfs_delalloc_release_space(struct inode *inode, u64 start, u64 len)
5954 {
5955 btrfs_delalloc_release_metadata(inode, len);
5956 btrfs_free_reserved_data_space(inode, start, len);
5957 }
5958
5959 static int update_block_group(struct btrfs_trans_handle *trans,
5960 struct btrfs_root *root, u64 bytenr,
5961 u64 num_bytes, int alloc)
5962 {
5963 struct btrfs_block_group_cache *cache = NULL;
5964 struct btrfs_fs_info *info = root->fs_info;
5965 u64 total = num_bytes;
5966 u64 old_val;
5967 u64 byte_in_group;
5968 int factor;
5969
5970 /* block accounting for super block */
5971 spin_lock(&info->delalloc_root_lock);
5972 old_val = btrfs_super_bytes_used(info->super_copy);
5973 if (alloc)
5974 old_val += num_bytes;
5975 else
5976 old_val -= num_bytes;
5977 btrfs_set_super_bytes_used(info->super_copy, old_val);
5978 spin_unlock(&info->delalloc_root_lock);
5979
5980 while (total) {
5981 cache = btrfs_lookup_block_group(info, bytenr);
5982 if (!cache)
5983 return -ENOENT;
5984 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5985 BTRFS_BLOCK_GROUP_RAID1 |
5986 BTRFS_BLOCK_GROUP_RAID10))
5987 factor = 2;
5988 else
5989 factor = 1;
5990 /*
5991 * If this block group has free space cache written out, we
5992 * need to make sure to load it if we are removing space. This
5993 * is because we need the unpinning stage to actually add the
5994 * space back to the block group, otherwise we will leak space.
5995 */
5996 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5997 cache_block_group(cache, 1);
5998
5999 byte_in_group = bytenr - cache->key.objectid;
6000 WARN_ON(byte_in_group > cache->key.offset);
6001
6002 spin_lock(&cache->space_info->lock);
6003 spin_lock(&cache->lock);
6004
6005 if (btrfs_test_opt(root, SPACE_CACHE) &&
6006 cache->disk_cache_state < BTRFS_DC_CLEAR)
6007 cache->disk_cache_state = BTRFS_DC_CLEAR;
6008
6009 old_val = btrfs_block_group_used(&cache->item);
6010 num_bytes = min(total, cache->key.offset - byte_in_group);
6011 if (alloc) {
6012 old_val += num_bytes;
6013 btrfs_set_block_group_used(&cache->item, old_val);
6014 cache->reserved -= num_bytes;
6015 cache->space_info->bytes_reserved -= num_bytes;
6016 cache->space_info->bytes_used += num_bytes;
6017 cache->space_info->disk_used += num_bytes * factor;
6018 spin_unlock(&cache->lock);
6019 spin_unlock(&cache->space_info->lock);
6020 } else {
6021 old_val -= num_bytes;
6022 btrfs_set_block_group_used(&cache->item, old_val);
6023 cache->pinned += num_bytes;
6024 cache->space_info->bytes_pinned += num_bytes;
6025 cache->space_info->bytes_used -= num_bytes;
6026 cache->space_info->disk_used -= num_bytes * factor;
6027 spin_unlock(&cache->lock);
6028 spin_unlock(&cache->space_info->lock);
6029
6030 set_extent_dirty(info->pinned_extents,
6031 bytenr, bytenr + num_bytes - 1,
6032 GFP_NOFS | __GFP_NOFAIL);
6033 }
6034
6035 spin_lock(&trans->transaction->dirty_bgs_lock);
6036 if (list_empty(&cache->dirty_list)) {
6037 list_add_tail(&cache->dirty_list,
6038 &trans->transaction->dirty_bgs);
6039 trans->transaction->num_dirty_bgs++;
6040 btrfs_get_block_group(cache);
6041 }
6042 spin_unlock(&trans->transaction->dirty_bgs_lock);
6043
6044 /*
6045 * No longer have used bytes in this block group, queue it for
6046 * deletion. We do this after adding the block group to the
6047 * dirty list to avoid races between cleaner kthread and space
6048 * cache writeout.
6049 */
6050 if (!alloc && old_val == 0) {
6051 spin_lock(&info->unused_bgs_lock);
6052 if (list_empty(&cache->bg_list)) {
6053 btrfs_get_block_group(cache);
6054 list_add_tail(&cache->bg_list,
6055 &info->unused_bgs);
6056 }
6057 spin_unlock(&info->unused_bgs_lock);
6058 }
6059
6060 btrfs_put_block_group(cache);
6061 total -= num_bytes;
6062 bytenr += num_bytes;
6063 }
6064 return 0;
6065 }
6066
6067 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
6068 {
6069 struct btrfs_block_group_cache *cache;
6070 u64 bytenr;
6071
6072 spin_lock(&root->fs_info->block_group_cache_lock);
6073 bytenr = root->fs_info->first_logical_byte;
6074 spin_unlock(&root->fs_info->block_group_cache_lock);
6075
6076 if (bytenr < (u64)-1)
6077 return bytenr;
6078
6079 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
6080 if (!cache)
6081 return 0;
6082
6083 bytenr = cache->key.objectid;
6084 btrfs_put_block_group(cache);
6085
6086 return bytenr;
6087 }
6088
6089 static int pin_down_extent(struct btrfs_root *root,
6090 struct btrfs_block_group_cache *cache,
6091 u64 bytenr, u64 num_bytes, int reserved)
6092 {
6093 spin_lock(&cache->space_info->lock);
6094 spin_lock(&cache->lock);
6095 cache->pinned += num_bytes;
6096 cache->space_info->bytes_pinned += num_bytes;
6097 if (reserved) {
6098 cache->reserved -= num_bytes;
6099 cache->space_info->bytes_reserved -= num_bytes;
6100 }
6101 spin_unlock(&cache->lock);
6102 spin_unlock(&cache->space_info->lock);
6103
6104 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
6105 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6106 if (reserved)
6107 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
6108 return 0;
6109 }
6110
6111 /*
6112 * this function must be called within transaction
6113 */
6114 int btrfs_pin_extent(struct btrfs_root *root,
6115 u64 bytenr, u64 num_bytes, int reserved)
6116 {
6117 struct btrfs_block_group_cache *cache;
6118
6119 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
6120 BUG_ON(!cache); /* Logic error */
6121
6122 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
6123
6124 btrfs_put_block_group(cache);
6125 return 0;
6126 }
6127
6128 /*
6129 * this function must be called within transaction
6130 */
6131 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
6132 u64 bytenr, u64 num_bytes)
6133 {
6134 struct btrfs_block_group_cache *cache;
6135 int ret;
6136
6137 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
6138 if (!cache)
6139 return -EINVAL;
6140
6141 /*
6142 * pull in the free space cache (if any) so that our pin
6143 * removes the free space from the cache. We have load_only set
6144 * to one because the slow code to read in the free extents does check
6145 * the pinned extents.
6146 */
6147 cache_block_group(cache, 1);
6148
6149 pin_down_extent(root, cache, bytenr, num_bytes, 0);
6150
6151 /* remove us from the free space cache (if we're there at all) */
6152 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6153 btrfs_put_block_group(cache);
6154 return ret;
6155 }
6156
6157 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
6158 {
6159 int ret;
6160 struct btrfs_block_group_cache *block_group;
6161 struct btrfs_caching_control *caching_ctl;
6162
6163 block_group = btrfs_lookup_block_group(root->fs_info, start);
6164 if (!block_group)
6165 return -EINVAL;
6166
6167 cache_block_group(block_group, 0);
6168 caching_ctl = get_caching_control(block_group);
6169
6170 if (!caching_ctl) {
6171 /* Logic error */
6172 BUG_ON(!block_group_cache_done(block_group));
6173 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6174 } else {
6175 mutex_lock(&caching_ctl->mutex);
6176
6177 if (start >= caching_ctl->progress) {
6178 ret = add_excluded_extent(root, start, num_bytes);
6179 } else if (start + num_bytes <= caching_ctl->progress) {
6180 ret = btrfs_remove_free_space(block_group,
6181 start, num_bytes);
6182 } else {
6183 num_bytes = caching_ctl->progress - start;
6184 ret = btrfs_remove_free_space(block_group,
6185 start, num_bytes);
6186 if (ret)
6187 goto out_lock;
6188
6189 num_bytes = (start + num_bytes) -
6190 caching_ctl->progress;
6191 start = caching_ctl->progress;
6192 ret = add_excluded_extent(root, start, num_bytes);
6193 }
6194 out_lock:
6195 mutex_unlock(&caching_ctl->mutex);
6196 put_caching_control(caching_ctl);
6197 }
6198 btrfs_put_block_group(block_group);
6199 return ret;
6200 }
6201
6202 int btrfs_exclude_logged_extents(struct btrfs_root *log,
6203 struct extent_buffer *eb)
6204 {
6205 struct btrfs_file_extent_item *item;
6206 struct btrfs_key key;
6207 int found_type;
6208 int i;
6209
6210 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
6211 return 0;
6212
6213 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6214 btrfs_item_key_to_cpu(eb, &key, i);
6215 if (key.type != BTRFS_EXTENT_DATA_KEY)
6216 continue;
6217 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6218 found_type = btrfs_file_extent_type(eb, item);
6219 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6220 continue;
6221 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6222 continue;
6223 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6224 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6225 __exclude_logged_extent(log, key.objectid, key.offset);
6226 }
6227
6228 return 0;
6229 }
6230
6231 static void
6232 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
6233 {
6234 atomic_inc(&bg->reservations);
6235 }
6236
6237 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
6238 const u64 start)
6239 {
6240 struct btrfs_block_group_cache *bg;
6241
6242 bg = btrfs_lookup_block_group(fs_info, start);
6243 ASSERT(bg);
6244 if (atomic_dec_and_test(&bg->reservations))
6245 wake_up_atomic_t(&bg->reservations);
6246 btrfs_put_block_group(bg);
6247 }
6248
6249 static int btrfs_wait_bg_reservations_atomic_t(atomic_t *a)
6250 {
6251 schedule();
6252 return 0;
6253 }
6254
6255 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg)
6256 {
6257 struct btrfs_space_info *space_info = bg->space_info;
6258
6259 ASSERT(bg->ro);
6260
6261 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
6262 return;
6263
6264 /*
6265 * Our block group is read only but before we set it to read only,
6266 * some task might have had allocated an extent from it already, but it
6267 * has not yet created a respective ordered extent (and added it to a
6268 * root's list of ordered extents).
6269 * Therefore wait for any task currently allocating extents, since the
6270 * block group's reservations counter is incremented while a read lock
6271 * on the groups' semaphore is held and decremented after releasing
6272 * the read access on that semaphore and creating the ordered extent.
6273 */
6274 down_write(&space_info->groups_sem);
6275 up_write(&space_info->groups_sem);
6276
6277 wait_on_atomic_t(&bg->reservations,
6278 btrfs_wait_bg_reservations_atomic_t,
6279 TASK_UNINTERRUPTIBLE);
6280 }
6281
6282 /**
6283 * btrfs_update_reserved_bytes - update the block_group and space info counters
6284 * @cache: The cache we are manipulating
6285 * @num_bytes: The number of bytes in question
6286 * @reserve: One of the reservation enums
6287 * @delalloc: The blocks are allocated for the delalloc write
6288 *
6289 * This is called by the allocator when it reserves space, or by somebody who is
6290 * freeing space that was never actually used on disk. For example if you
6291 * reserve some space for a new leaf in transaction A and before transaction A
6292 * commits you free that leaf, you call this with reserve set to 0 in order to
6293 * clear the reservation.
6294 *
6295 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
6296 * ENOSPC accounting. For data we handle the reservation through clearing the
6297 * delalloc bits in the io_tree. We have to do this since we could end up
6298 * allocating less disk space for the amount of data we have reserved in the
6299 * case of compression.
6300 *
6301 * If this is a reservation and the block group has become read only we cannot
6302 * make the reservation and return -EAGAIN, otherwise this function always
6303 * succeeds.
6304 */
6305 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
6306 u64 num_bytes, int reserve, int delalloc)
6307 {
6308 struct btrfs_space_info *space_info = cache->space_info;
6309 int ret = 0;
6310
6311 spin_lock(&space_info->lock);
6312 spin_lock(&cache->lock);
6313 if (reserve != RESERVE_FREE) {
6314 if (cache->ro) {
6315 ret = -EAGAIN;
6316 } else {
6317 cache->reserved += num_bytes;
6318 space_info->bytes_reserved += num_bytes;
6319 if (reserve == RESERVE_ALLOC) {
6320 trace_btrfs_space_reservation(cache->fs_info,
6321 "space_info", space_info->flags,
6322 num_bytes, 0);
6323 space_info->bytes_may_use -= num_bytes;
6324 }
6325
6326 if (delalloc)
6327 cache->delalloc_bytes += num_bytes;
6328 }
6329 } else {
6330 if (cache->ro)
6331 space_info->bytes_readonly += num_bytes;
6332 cache->reserved -= num_bytes;
6333 space_info->bytes_reserved -= num_bytes;
6334
6335 if (delalloc)
6336 cache->delalloc_bytes -= num_bytes;
6337 }
6338 spin_unlock(&cache->lock);
6339 spin_unlock(&space_info->lock);
6340 return ret;
6341 }
6342
6343 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
6344 struct btrfs_root *root)
6345 {
6346 struct btrfs_fs_info *fs_info = root->fs_info;
6347 struct btrfs_caching_control *next;
6348 struct btrfs_caching_control *caching_ctl;
6349 struct btrfs_block_group_cache *cache;
6350
6351 down_write(&fs_info->commit_root_sem);
6352
6353 list_for_each_entry_safe(caching_ctl, next,
6354 &fs_info->caching_block_groups, list) {
6355 cache = caching_ctl->block_group;
6356 if (block_group_cache_done(cache)) {
6357 cache->last_byte_to_unpin = (u64)-1;
6358 list_del_init(&caching_ctl->list);
6359 put_caching_control(caching_ctl);
6360 } else {
6361 cache->last_byte_to_unpin = caching_ctl->progress;
6362 }
6363 }
6364
6365 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6366 fs_info->pinned_extents = &fs_info->freed_extents[1];
6367 else
6368 fs_info->pinned_extents = &fs_info->freed_extents[0];
6369
6370 up_write(&fs_info->commit_root_sem);
6371
6372 update_global_block_rsv(fs_info);
6373 }
6374
6375 /*
6376 * Returns the free cluster for the given space info and sets empty_cluster to
6377 * what it should be based on the mount options.
6378 */
6379 static struct btrfs_free_cluster *
6380 fetch_cluster_info(struct btrfs_root *root, struct btrfs_space_info *space_info,
6381 u64 *empty_cluster)
6382 {
6383 struct btrfs_free_cluster *ret = NULL;
6384 bool ssd = btrfs_test_opt(root, SSD);
6385
6386 *empty_cluster = 0;
6387 if (btrfs_mixed_space_info(space_info))
6388 return ret;
6389
6390 if (ssd)
6391 *empty_cluster = SZ_2M;
6392 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6393 ret = &root->fs_info->meta_alloc_cluster;
6394 if (!ssd)
6395 *empty_cluster = SZ_64K;
6396 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && ssd) {
6397 ret = &root->fs_info->data_alloc_cluster;
6398 }
6399
6400 return ret;
6401 }
6402
6403 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
6404 const bool return_free_space)
6405 {
6406 struct btrfs_fs_info *fs_info = root->fs_info;
6407 struct btrfs_block_group_cache *cache = NULL;
6408 struct btrfs_space_info *space_info;
6409 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6410 struct btrfs_free_cluster *cluster = NULL;
6411 u64 len;
6412 u64 total_unpinned = 0;
6413 u64 empty_cluster = 0;
6414 bool readonly;
6415
6416 while (start <= end) {
6417 readonly = false;
6418 if (!cache ||
6419 start >= cache->key.objectid + cache->key.offset) {
6420 if (cache)
6421 btrfs_put_block_group(cache);
6422 total_unpinned = 0;
6423 cache = btrfs_lookup_block_group(fs_info, start);
6424 BUG_ON(!cache); /* Logic error */
6425
6426 cluster = fetch_cluster_info(root,
6427 cache->space_info,
6428 &empty_cluster);
6429 empty_cluster <<= 1;
6430 }
6431
6432 len = cache->key.objectid + cache->key.offset - start;
6433 len = min(len, end + 1 - start);
6434
6435 if (start < cache->last_byte_to_unpin) {
6436 len = min(len, cache->last_byte_to_unpin - start);
6437 if (return_free_space)
6438 btrfs_add_free_space(cache, start, len);
6439 }
6440
6441 start += len;
6442 total_unpinned += len;
6443 space_info = cache->space_info;
6444
6445 /*
6446 * If this space cluster has been marked as fragmented and we've
6447 * unpinned enough in this block group to potentially allow a
6448 * cluster to be created inside of it go ahead and clear the
6449 * fragmented check.
6450 */
6451 if (cluster && cluster->fragmented &&
6452 total_unpinned > empty_cluster) {
6453 spin_lock(&cluster->lock);
6454 cluster->fragmented = 0;
6455 spin_unlock(&cluster->lock);
6456 }
6457
6458 spin_lock(&space_info->lock);
6459 spin_lock(&cache->lock);
6460 cache->pinned -= len;
6461 space_info->bytes_pinned -= len;
6462 space_info->max_extent_size = 0;
6463 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6464 if (cache->ro) {
6465 space_info->bytes_readonly += len;
6466 readonly = true;
6467 }
6468 spin_unlock(&cache->lock);
6469 if (!readonly && global_rsv->space_info == space_info) {
6470 spin_lock(&global_rsv->lock);
6471 if (!global_rsv->full) {
6472 len = min(len, global_rsv->size -
6473 global_rsv->reserved);
6474 global_rsv->reserved += len;
6475 space_info->bytes_may_use += len;
6476 if (global_rsv->reserved >= global_rsv->size)
6477 global_rsv->full = 1;
6478 }
6479 spin_unlock(&global_rsv->lock);
6480 }
6481 spin_unlock(&space_info->lock);
6482 }
6483
6484 if (cache)
6485 btrfs_put_block_group(cache);
6486 return 0;
6487 }
6488
6489 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6490 struct btrfs_root *root)
6491 {
6492 struct btrfs_fs_info *fs_info = root->fs_info;
6493 struct btrfs_block_group_cache *block_group, *tmp;
6494 struct list_head *deleted_bgs;
6495 struct extent_io_tree *unpin;
6496 u64 start;
6497 u64 end;
6498 int ret;
6499
6500 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6501 unpin = &fs_info->freed_extents[1];
6502 else
6503 unpin = &fs_info->freed_extents[0];
6504
6505 while (!trans->aborted) {
6506 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6507 ret = find_first_extent_bit(unpin, 0, &start, &end,
6508 EXTENT_DIRTY, NULL);
6509 if (ret) {
6510 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6511 break;
6512 }
6513
6514 if (btrfs_test_opt(root, DISCARD))
6515 ret = btrfs_discard_extent(root, start,
6516 end + 1 - start, NULL);
6517
6518 clear_extent_dirty(unpin, start, end, GFP_NOFS);
6519 unpin_extent_range(root, start, end, true);
6520 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6521 cond_resched();
6522 }
6523
6524 /*
6525 * Transaction is finished. We don't need the lock anymore. We
6526 * do need to clean up the block groups in case of a transaction
6527 * abort.
6528 */
6529 deleted_bgs = &trans->transaction->deleted_bgs;
6530 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6531 u64 trimmed = 0;
6532
6533 ret = -EROFS;
6534 if (!trans->aborted)
6535 ret = btrfs_discard_extent(root,
6536 block_group->key.objectid,
6537 block_group->key.offset,
6538 &trimmed);
6539
6540 list_del_init(&block_group->bg_list);
6541 btrfs_put_block_group_trimming(block_group);
6542 btrfs_put_block_group(block_group);
6543
6544 if (ret) {
6545 const char *errstr = btrfs_decode_error(ret);
6546 btrfs_warn(fs_info,
6547 "Discard failed while removing blockgroup: errno=%d %s\n",
6548 ret, errstr);
6549 }
6550 }
6551
6552 return 0;
6553 }
6554
6555 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
6556 u64 owner, u64 root_objectid)
6557 {
6558 struct btrfs_space_info *space_info;
6559 u64 flags;
6560
6561 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6562 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
6563 flags = BTRFS_BLOCK_GROUP_SYSTEM;
6564 else
6565 flags = BTRFS_BLOCK_GROUP_METADATA;
6566 } else {
6567 flags = BTRFS_BLOCK_GROUP_DATA;
6568 }
6569
6570 space_info = __find_space_info(fs_info, flags);
6571 BUG_ON(!space_info); /* Logic bug */
6572 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
6573 }
6574
6575
6576 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6577 struct btrfs_root *root,
6578 struct btrfs_delayed_ref_node *node, u64 parent,
6579 u64 root_objectid, u64 owner_objectid,
6580 u64 owner_offset, int refs_to_drop,
6581 struct btrfs_delayed_extent_op *extent_op)
6582 {
6583 struct btrfs_key key;
6584 struct btrfs_path *path;
6585 struct btrfs_fs_info *info = root->fs_info;
6586 struct btrfs_root *extent_root = info->extent_root;
6587 struct extent_buffer *leaf;
6588 struct btrfs_extent_item *ei;
6589 struct btrfs_extent_inline_ref *iref;
6590 int ret;
6591 int is_data;
6592 int extent_slot = 0;
6593 int found_extent = 0;
6594 int num_to_del = 1;
6595 u32 item_size;
6596 u64 refs;
6597 u64 bytenr = node->bytenr;
6598 u64 num_bytes = node->num_bytes;
6599 int last_ref = 0;
6600 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6601 SKINNY_METADATA);
6602
6603 path = btrfs_alloc_path();
6604 if (!path)
6605 return -ENOMEM;
6606
6607 path->reada = READA_FORWARD;
6608 path->leave_spinning = 1;
6609
6610 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6611 BUG_ON(!is_data && refs_to_drop != 1);
6612
6613 if (is_data)
6614 skinny_metadata = 0;
6615
6616 ret = lookup_extent_backref(trans, extent_root, path, &iref,
6617 bytenr, num_bytes, parent,
6618 root_objectid, owner_objectid,
6619 owner_offset);
6620 if (ret == 0) {
6621 extent_slot = path->slots[0];
6622 while (extent_slot >= 0) {
6623 btrfs_item_key_to_cpu(path->nodes[0], &key,
6624 extent_slot);
6625 if (key.objectid != bytenr)
6626 break;
6627 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6628 key.offset == num_bytes) {
6629 found_extent = 1;
6630 break;
6631 }
6632 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6633 key.offset == owner_objectid) {
6634 found_extent = 1;
6635 break;
6636 }
6637 if (path->slots[0] - extent_slot > 5)
6638 break;
6639 extent_slot--;
6640 }
6641 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6642 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6643 if (found_extent && item_size < sizeof(*ei))
6644 found_extent = 0;
6645 #endif
6646 if (!found_extent) {
6647 BUG_ON(iref);
6648 ret = remove_extent_backref(trans, extent_root, path,
6649 NULL, refs_to_drop,
6650 is_data, &last_ref);
6651 if (ret) {
6652 btrfs_abort_transaction(trans, extent_root, ret);
6653 goto out;
6654 }
6655 btrfs_release_path(path);
6656 path->leave_spinning = 1;
6657
6658 key.objectid = bytenr;
6659 key.type = BTRFS_EXTENT_ITEM_KEY;
6660 key.offset = num_bytes;
6661
6662 if (!is_data && skinny_metadata) {
6663 key.type = BTRFS_METADATA_ITEM_KEY;
6664 key.offset = owner_objectid;
6665 }
6666
6667 ret = btrfs_search_slot(trans, extent_root,
6668 &key, path, -1, 1);
6669 if (ret > 0 && skinny_metadata && path->slots[0]) {
6670 /*
6671 * Couldn't find our skinny metadata item,
6672 * see if we have ye olde extent item.
6673 */
6674 path->slots[0]--;
6675 btrfs_item_key_to_cpu(path->nodes[0], &key,
6676 path->slots[0]);
6677 if (key.objectid == bytenr &&
6678 key.type == BTRFS_EXTENT_ITEM_KEY &&
6679 key.offset == num_bytes)
6680 ret = 0;
6681 }
6682
6683 if (ret > 0 && skinny_metadata) {
6684 skinny_metadata = false;
6685 key.objectid = bytenr;
6686 key.type = BTRFS_EXTENT_ITEM_KEY;
6687 key.offset = num_bytes;
6688 btrfs_release_path(path);
6689 ret = btrfs_search_slot(trans, extent_root,
6690 &key, path, -1, 1);
6691 }
6692
6693 if (ret) {
6694 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6695 ret, bytenr);
6696 if (ret > 0)
6697 btrfs_print_leaf(extent_root,
6698 path->nodes[0]);
6699 }
6700 if (ret < 0) {
6701 btrfs_abort_transaction(trans, extent_root, ret);
6702 goto out;
6703 }
6704 extent_slot = path->slots[0];
6705 }
6706 } else if (WARN_ON(ret == -ENOENT)) {
6707 btrfs_print_leaf(extent_root, path->nodes[0]);
6708 btrfs_err(info,
6709 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6710 bytenr, parent, root_objectid, owner_objectid,
6711 owner_offset);
6712 btrfs_abort_transaction(trans, extent_root, ret);
6713 goto out;
6714 } else {
6715 btrfs_abort_transaction(trans, extent_root, ret);
6716 goto out;
6717 }
6718
6719 leaf = path->nodes[0];
6720 item_size = btrfs_item_size_nr(leaf, extent_slot);
6721 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6722 if (item_size < sizeof(*ei)) {
6723 BUG_ON(found_extent || extent_slot != path->slots[0]);
6724 ret = convert_extent_item_v0(trans, extent_root, path,
6725 owner_objectid, 0);
6726 if (ret < 0) {
6727 btrfs_abort_transaction(trans, extent_root, ret);
6728 goto out;
6729 }
6730
6731 btrfs_release_path(path);
6732 path->leave_spinning = 1;
6733
6734 key.objectid = bytenr;
6735 key.type = BTRFS_EXTENT_ITEM_KEY;
6736 key.offset = num_bytes;
6737
6738 ret = btrfs_search_slot(trans, extent_root, &key, path,
6739 -1, 1);
6740 if (ret) {
6741 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6742 ret, bytenr);
6743 btrfs_print_leaf(extent_root, path->nodes[0]);
6744 }
6745 if (ret < 0) {
6746 btrfs_abort_transaction(trans, extent_root, ret);
6747 goto out;
6748 }
6749
6750 extent_slot = path->slots[0];
6751 leaf = path->nodes[0];
6752 item_size = btrfs_item_size_nr(leaf, extent_slot);
6753 }
6754 #endif
6755 BUG_ON(item_size < sizeof(*ei));
6756 ei = btrfs_item_ptr(leaf, extent_slot,
6757 struct btrfs_extent_item);
6758 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6759 key.type == BTRFS_EXTENT_ITEM_KEY) {
6760 struct btrfs_tree_block_info *bi;
6761 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6762 bi = (struct btrfs_tree_block_info *)(ei + 1);
6763 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6764 }
6765
6766 refs = btrfs_extent_refs(leaf, ei);
6767 if (refs < refs_to_drop) {
6768 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
6769 "for bytenr %Lu", refs_to_drop, refs, bytenr);
6770 ret = -EINVAL;
6771 btrfs_abort_transaction(trans, extent_root, ret);
6772 goto out;
6773 }
6774 refs -= refs_to_drop;
6775
6776 if (refs > 0) {
6777 if (extent_op)
6778 __run_delayed_extent_op(extent_op, leaf, ei);
6779 /*
6780 * In the case of inline back ref, reference count will
6781 * be updated by remove_extent_backref
6782 */
6783 if (iref) {
6784 BUG_ON(!found_extent);
6785 } else {
6786 btrfs_set_extent_refs(leaf, ei, refs);
6787 btrfs_mark_buffer_dirty(leaf);
6788 }
6789 if (found_extent) {
6790 ret = remove_extent_backref(trans, extent_root, path,
6791 iref, refs_to_drop,
6792 is_data, &last_ref);
6793 if (ret) {
6794 btrfs_abort_transaction(trans, extent_root, ret);
6795 goto out;
6796 }
6797 }
6798 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
6799 root_objectid);
6800 } else {
6801 if (found_extent) {
6802 BUG_ON(is_data && refs_to_drop !=
6803 extent_data_ref_count(path, iref));
6804 if (iref) {
6805 BUG_ON(path->slots[0] != extent_slot);
6806 } else {
6807 BUG_ON(path->slots[0] != extent_slot + 1);
6808 path->slots[0] = extent_slot;
6809 num_to_del = 2;
6810 }
6811 }
6812
6813 last_ref = 1;
6814 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6815 num_to_del);
6816 if (ret) {
6817 btrfs_abort_transaction(trans, extent_root, ret);
6818 goto out;
6819 }
6820 btrfs_release_path(path);
6821
6822 if (is_data) {
6823 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
6824 if (ret) {
6825 btrfs_abort_transaction(trans, extent_root, ret);
6826 goto out;
6827 }
6828 }
6829
6830 ret = add_to_free_space_tree(trans, root->fs_info, bytenr,
6831 num_bytes);
6832 if (ret) {
6833 btrfs_abort_transaction(trans, extent_root, ret);
6834 goto out;
6835 }
6836
6837 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
6838 if (ret) {
6839 btrfs_abort_transaction(trans, extent_root, ret);
6840 goto out;
6841 }
6842 }
6843 btrfs_release_path(path);
6844
6845 out:
6846 btrfs_free_path(path);
6847 return ret;
6848 }
6849
6850 /*
6851 * when we free an block, it is possible (and likely) that we free the last
6852 * delayed ref for that extent as well. This searches the delayed ref tree for
6853 * a given extent, and if there are no other delayed refs to be processed, it
6854 * removes it from the tree.
6855 */
6856 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6857 struct btrfs_root *root, u64 bytenr)
6858 {
6859 struct btrfs_delayed_ref_head *head;
6860 struct btrfs_delayed_ref_root *delayed_refs;
6861 int ret = 0;
6862
6863 delayed_refs = &trans->transaction->delayed_refs;
6864 spin_lock(&delayed_refs->lock);
6865 head = btrfs_find_delayed_ref_head(trans, bytenr);
6866 if (!head)
6867 goto out_delayed_unlock;
6868
6869 spin_lock(&head->lock);
6870 if (!list_empty(&head->ref_list))
6871 goto out;
6872
6873 if (head->extent_op) {
6874 if (!head->must_insert_reserved)
6875 goto out;
6876 btrfs_free_delayed_extent_op(head->extent_op);
6877 head->extent_op = NULL;
6878 }
6879
6880 /*
6881 * waiting for the lock here would deadlock. If someone else has it
6882 * locked they are already in the process of dropping it anyway
6883 */
6884 if (!mutex_trylock(&head->mutex))
6885 goto out;
6886
6887 /*
6888 * at this point we have a head with no other entries. Go
6889 * ahead and process it.
6890 */
6891 head->node.in_tree = 0;
6892 rb_erase(&head->href_node, &delayed_refs->href_root);
6893
6894 atomic_dec(&delayed_refs->num_entries);
6895
6896 /*
6897 * we don't take a ref on the node because we're removing it from the
6898 * tree, so we just steal the ref the tree was holding.
6899 */
6900 delayed_refs->num_heads--;
6901 if (head->processing == 0)
6902 delayed_refs->num_heads_ready--;
6903 head->processing = 0;
6904 spin_unlock(&head->lock);
6905 spin_unlock(&delayed_refs->lock);
6906
6907 BUG_ON(head->extent_op);
6908 if (head->must_insert_reserved)
6909 ret = 1;
6910
6911 mutex_unlock(&head->mutex);
6912 btrfs_put_delayed_ref(&head->node);
6913 return ret;
6914 out:
6915 spin_unlock(&head->lock);
6916
6917 out_delayed_unlock:
6918 spin_unlock(&delayed_refs->lock);
6919 return 0;
6920 }
6921
6922 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6923 struct btrfs_root *root,
6924 struct extent_buffer *buf,
6925 u64 parent, int last_ref)
6926 {
6927 int pin = 1;
6928 int ret;
6929
6930 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6931 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6932 buf->start, buf->len,
6933 parent, root->root_key.objectid,
6934 btrfs_header_level(buf),
6935 BTRFS_DROP_DELAYED_REF, NULL);
6936 BUG_ON(ret); /* -ENOMEM */
6937 }
6938
6939 if (!last_ref)
6940 return;
6941
6942 if (btrfs_header_generation(buf) == trans->transid) {
6943 struct btrfs_block_group_cache *cache;
6944
6945 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6946 ret = check_ref_cleanup(trans, root, buf->start);
6947 if (!ret)
6948 goto out;
6949 }
6950
6951 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6952
6953 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6954 pin_down_extent(root, cache, buf->start, buf->len, 1);
6955 btrfs_put_block_group(cache);
6956 goto out;
6957 }
6958
6959 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6960
6961 btrfs_add_free_space(cache, buf->start, buf->len);
6962 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
6963 btrfs_put_block_group(cache);
6964 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6965 pin = 0;
6966 }
6967 out:
6968 if (pin)
6969 add_pinned_bytes(root->fs_info, buf->len,
6970 btrfs_header_level(buf),
6971 root->root_key.objectid);
6972
6973 /*
6974 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6975 * anymore.
6976 */
6977 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6978 }
6979
6980 /* Can return -ENOMEM */
6981 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6982 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6983 u64 owner, u64 offset)
6984 {
6985 int ret;
6986 struct btrfs_fs_info *fs_info = root->fs_info;
6987
6988 if (btrfs_test_is_dummy_root(root))
6989 return 0;
6990
6991 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6992
6993 /*
6994 * tree log blocks never actually go into the extent allocation
6995 * tree, just update pinning info and exit early.
6996 */
6997 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6998 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6999 /* unlocks the pinned mutex */
7000 btrfs_pin_extent(root, bytenr, num_bytes, 1);
7001 ret = 0;
7002 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
7003 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
7004 num_bytes,
7005 parent, root_objectid, (int)owner,
7006 BTRFS_DROP_DELAYED_REF, NULL);
7007 } else {
7008 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
7009 num_bytes,
7010 parent, root_objectid, owner,
7011 offset, 0,
7012 BTRFS_DROP_DELAYED_REF, NULL);
7013 }
7014 return ret;
7015 }
7016
7017 /*
7018 * when we wait for progress in the block group caching, its because
7019 * our allocation attempt failed at least once. So, we must sleep
7020 * and let some progress happen before we try again.
7021 *
7022 * This function will sleep at least once waiting for new free space to
7023 * show up, and then it will check the block group free space numbers
7024 * for our min num_bytes. Another option is to have it go ahead
7025 * and look in the rbtree for a free extent of a given size, but this
7026 * is a good start.
7027 *
7028 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7029 * any of the information in this block group.
7030 */
7031 static noinline void
7032 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
7033 u64 num_bytes)
7034 {
7035 struct btrfs_caching_control *caching_ctl;
7036
7037 caching_ctl = get_caching_control(cache);
7038 if (!caching_ctl)
7039 return;
7040
7041 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
7042 (cache->free_space_ctl->free_space >= num_bytes));
7043
7044 put_caching_control(caching_ctl);
7045 }
7046
7047 static noinline int
7048 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
7049 {
7050 struct btrfs_caching_control *caching_ctl;
7051 int ret = 0;
7052
7053 caching_ctl = get_caching_control(cache);
7054 if (!caching_ctl)
7055 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
7056
7057 wait_event(caching_ctl->wait, block_group_cache_done(cache));
7058 if (cache->cached == BTRFS_CACHE_ERROR)
7059 ret = -EIO;
7060 put_caching_control(caching_ctl);
7061 return ret;
7062 }
7063
7064 int __get_raid_index(u64 flags)
7065 {
7066 if (flags & BTRFS_BLOCK_GROUP_RAID10)
7067 return BTRFS_RAID_RAID10;
7068 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
7069 return BTRFS_RAID_RAID1;
7070 else if (flags & BTRFS_BLOCK_GROUP_DUP)
7071 return BTRFS_RAID_DUP;
7072 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
7073 return BTRFS_RAID_RAID0;
7074 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
7075 return BTRFS_RAID_RAID5;
7076 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
7077 return BTRFS_RAID_RAID6;
7078
7079 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
7080 }
7081
7082 int get_block_group_index(struct btrfs_block_group_cache *cache)
7083 {
7084 return __get_raid_index(cache->flags);
7085 }
7086
7087 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
7088 [BTRFS_RAID_RAID10] = "raid10",
7089 [BTRFS_RAID_RAID1] = "raid1",
7090 [BTRFS_RAID_DUP] = "dup",
7091 [BTRFS_RAID_RAID0] = "raid0",
7092 [BTRFS_RAID_SINGLE] = "single",
7093 [BTRFS_RAID_RAID5] = "raid5",
7094 [BTRFS_RAID_RAID6] = "raid6",
7095 };
7096
7097 static const char *get_raid_name(enum btrfs_raid_types type)
7098 {
7099 if (type >= BTRFS_NR_RAID_TYPES)
7100 return NULL;
7101
7102 return btrfs_raid_type_names[type];
7103 }
7104
7105 enum btrfs_loop_type {
7106 LOOP_CACHING_NOWAIT = 0,
7107 LOOP_CACHING_WAIT = 1,
7108 LOOP_ALLOC_CHUNK = 2,
7109 LOOP_NO_EMPTY_SIZE = 3,
7110 };
7111
7112 static inline void
7113 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7114 int delalloc)
7115 {
7116 if (delalloc)
7117 down_read(&cache->data_rwsem);
7118 }
7119
7120 static inline void
7121 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7122 int delalloc)
7123 {
7124 btrfs_get_block_group(cache);
7125 if (delalloc)
7126 down_read(&cache->data_rwsem);
7127 }
7128
7129 static struct btrfs_block_group_cache *
7130 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7131 struct btrfs_free_cluster *cluster,
7132 int delalloc)
7133 {
7134 struct btrfs_block_group_cache *used_bg = NULL;
7135 bool locked = false;
7136 again:
7137 spin_lock(&cluster->refill_lock);
7138 if (locked) {
7139 if (used_bg == cluster->block_group)
7140 return used_bg;
7141
7142 up_read(&used_bg->data_rwsem);
7143 btrfs_put_block_group(used_bg);
7144 }
7145
7146 used_bg = cluster->block_group;
7147 if (!used_bg)
7148 return NULL;
7149
7150 if (used_bg == block_group)
7151 return used_bg;
7152
7153 btrfs_get_block_group(used_bg);
7154
7155 if (!delalloc)
7156 return used_bg;
7157
7158 if (down_read_trylock(&used_bg->data_rwsem))
7159 return used_bg;
7160
7161 spin_unlock(&cluster->refill_lock);
7162 down_read(&used_bg->data_rwsem);
7163 locked = true;
7164 goto again;
7165 }
7166
7167 static inline void
7168 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7169 int delalloc)
7170 {
7171 if (delalloc)
7172 up_read(&cache->data_rwsem);
7173 btrfs_put_block_group(cache);
7174 }
7175
7176 /*
7177 * walks the btree of allocated extents and find a hole of a given size.
7178 * The key ins is changed to record the hole:
7179 * ins->objectid == start position
7180 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7181 * ins->offset == the size of the hole.
7182 * Any available blocks before search_start are skipped.
7183 *
7184 * If there is no suitable free space, we will record the max size of
7185 * the free space extent currently.
7186 */
7187 static noinline int find_free_extent(struct btrfs_root *orig_root,
7188 u64 num_bytes, u64 empty_size,
7189 u64 hint_byte, struct btrfs_key *ins,
7190 u64 flags, int delalloc)
7191 {
7192 int ret = 0;
7193 struct btrfs_root *root = orig_root->fs_info->extent_root;
7194 struct btrfs_free_cluster *last_ptr = NULL;
7195 struct btrfs_block_group_cache *block_group = NULL;
7196 u64 search_start = 0;
7197 u64 max_extent_size = 0;
7198 u64 empty_cluster = 0;
7199 struct btrfs_space_info *space_info;
7200 int loop = 0;
7201 int index = __get_raid_index(flags);
7202 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
7203 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
7204 bool failed_cluster_refill = false;
7205 bool failed_alloc = false;
7206 bool use_cluster = true;
7207 bool have_caching_bg = false;
7208 bool orig_have_caching_bg = false;
7209 bool full_search = false;
7210
7211 WARN_ON(num_bytes < root->sectorsize);
7212 ins->type = BTRFS_EXTENT_ITEM_KEY;
7213 ins->objectid = 0;
7214 ins->offset = 0;
7215
7216 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
7217
7218 space_info = __find_space_info(root->fs_info, flags);
7219 if (!space_info) {
7220 btrfs_err(root->fs_info, "No space info for %llu", flags);
7221 return -ENOSPC;
7222 }
7223
7224 /*
7225 * If our free space is heavily fragmented we may not be able to make
7226 * big contiguous allocations, so instead of doing the expensive search
7227 * for free space, simply return ENOSPC with our max_extent_size so we
7228 * can go ahead and search for a more manageable chunk.
7229 *
7230 * If our max_extent_size is large enough for our allocation simply
7231 * disable clustering since we will likely not be able to find enough
7232 * space to create a cluster and induce latency trying.
7233 */
7234 if (unlikely(space_info->max_extent_size)) {
7235 spin_lock(&space_info->lock);
7236 if (space_info->max_extent_size &&
7237 num_bytes > space_info->max_extent_size) {
7238 ins->offset = space_info->max_extent_size;
7239 spin_unlock(&space_info->lock);
7240 return -ENOSPC;
7241 } else if (space_info->max_extent_size) {
7242 use_cluster = false;
7243 }
7244 spin_unlock(&space_info->lock);
7245 }
7246
7247 last_ptr = fetch_cluster_info(orig_root, space_info, &empty_cluster);
7248 if (last_ptr) {
7249 spin_lock(&last_ptr->lock);
7250 if (last_ptr->block_group)
7251 hint_byte = last_ptr->window_start;
7252 if (last_ptr->fragmented) {
7253 /*
7254 * We still set window_start so we can keep track of the
7255 * last place we found an allocation to try and save
7256 * some time.
7257 */
7258 hint_byte = last_ptr->window_start;
7259 use_cluster = false;
7260 }
7261 spin_unlock(&last_ptr->lock);
7262 }
7263
7264 search_start = max(search_start, first_logical_byte(root, 0));
7265 search_start = max(search_start, hint_byte);
7266 if (search_start == hint_byte) {
7267 block_group = btrfs_lookup_block_group(root->fs_info,
7268 search_start);
7269 /*
7270 * we don't want to use the block group if it doesn't match our
7271 * allocation bits, or if its not cached.
7272 *
7273 * However if we are re-searching with an ideal block group
7274 * picked out then we don't care that the block group is cached.
7275 */
7276 if (block_group && block_group_bits(block_group, flags) &&
7277 block_group->cached != BTRFS_CACHE_NO) {
7278 down_read(&space_info->groups_sem);
7279 if (list_empty(&block_group->list) ||
7280 block_group->ro) {
7281 /*
7282 * someone is removing this block group,
7283 * we can't jump into the have_block_group
7284 * target because our list pointers are not
7285 * valid
7286 */
7287 btrfs_put_block_group(block_group);
7288 up_read(&space_info->groups_sem);
7289 } else {
7290 index = get_block_group_index(block_group);
7291 btrfs_lock_block_group(block_group, delalloc);
7292 goto have_block_group;
7293 }
7294 } else if (block_group) {
7295 btrfs_put_block_group(block_group);
7296 }
7297 }
7298 search:
7299 have_caching_bg = false;
7300 if (index == 0 || index == __get_raid_index(flags))
7301 full_search = true;
7302 down_read(&space_info->groups_sem);
7303 list_for_each_entry(block_group, &space_info->block_groups[index],
7304 list) {
7305 u64 offset;
7306 int cached;
7307
7308 btrfs_grab_block_group(block_group, delalloc);
7309 search_start = block_group->key.objectid;
7310
7311 /*
7312 * this can happen if we end up cycling through all the
7313 * raid types, but we want to make sure we only allocate
7314 * for the proper type.
7315 */
7316 if (!block_group_bits(block_group, flags)) {
7317 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7318 BTRFS_BLOCK_GROUP_RAID1 |
7319 BTRFS_BLOCK_GROUP_RAID5 |
7320 BTRFS_BLOCK_GROUP_RAID6 |
7321 BTRFS_BLOCK_GROUP_RAID10;
7322
7323 /*
7324 * if they asked for extra copies and this block group
7325 * doesn't provide them, bail. This does allow us to
7326 * fill raid0 from raid1.
7327 */
7328 if ((flags & extra) && !(block_group->flags & extra))
7329 goto loop;
7330 }
7331
7332 have_block_group:
7333 cached = block_group_cache_done(block_group);
7334 if (unlikely(!cached)) {
7335 have_caching_bg = true;
7336 ret = cache_block_group(block_group, 0);
7337 BUG_ON(ret < 0);
7338 ret = 0;
7339 }
7340
7341 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7342 goto loop;
7343 if (unlikely(block_group->ro))
7344 goto loop;
7345
7346 /*
7347 * Ok we want to try and use the cluster allocator, so
7348 * lets look there
7349 */
7350 if (last_ptr && use_cluster) {
7351 struct btrfs_block_group_cache *used_block_group;
7352 unsigned long aligned_cluster;
7353 /*
7354 * the refill lock keeps out other
7355 * people trying to start a new cluster
7356 */
7357 used_block_group = btrfs_lock_cluster(block_group,
7358 last_ptr,
7359 delalloc);
7360 if (!used_block_group)
7361 goto refill_cluster;
7362
7363 if (used_block_group != block_group &&
7364 (used_block_group->ro ||
7365 !block_group_bits(used_block_group, flags)))
7366 goto release_cluster;
7367
7368 offset = btrfs_alloc_from_cluster(used_block_group,
7369 last_ptr,
7370 num_bytes,
7371 used_block_group->key.objectid,
7372 &max_extent_size);
7373 if (offset) {
7374 /* we have a block, we're done */
7375 spin_unlock(&last_ptr->refill_lock);
7376 trace_btrfs_reserve_extent_cluster(root,
7377 used_block_group,
7378 search_start, num_bytes);
7379 if (used_block_group != block_group) {
7380 btrfs_release_block_group(block_group,
7381 delalloc);
7382 block_group = used_block_group;
7383 }
7384 goto checks;
7385 }
7386
7387 WARN_ON(last_ptr->block_group != used_block_group);
7388 release_cluster:
7389 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7390 * set up a new clusters, so lets just skip it
7391 * and let the allocator find whatever block
7392 * it can find. If we reach this point, we
7393 * will have tried the cluster allocator
7394 * plenty of times and not have found
7395 * anything, so we are likely way too
7396 * fragmented for the clustering stuff to find
7397 * anything.
7398 *
7399 * However, if the cluster is taken from the
7400 * current block group, release the cluster
7401 * first, so that we stand a better chance of
7402 * succeeding in the unclustered
7403 * allocation. */
7404 if (loop >= LOOP_NO_EMPTY_SIZE &&
7405 used_block_group != block_group) {
7406 spin_unlock(&last_ptr->refill_lock);
7407 btrfs_release_block_group(used_block_group,
7408 delalloc);
7409 goto unclustered_alloc;
7410 }
7411
7412 /*
7413 * this cluster didn't work out, free it and
7414 * start over
7415 */
7416 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7417
7418 if (used_block_group != block_group)
7419 btrfs_release_block_group(used_block_group,
7420 delalloc);
7421 refill_cluster:
7422 if (loop >= LOOP_NO_EMPTY_SIZE) {
7423 spin_unlock(&last_ptr->refill_lock);
7424 goto unclustered_alloc;
7425 }
7426
7427 aligned_cluster = max_t(unsigned long,
7428 empty_cluster + empty_size,
7429 block_group->full_stripe_len);
7430
7431 /* allocate a cluster in this block group */
7432 ret = btrfs_find_space_cluster(root, block_group,
7433 last_ptr, search_start,
7434 num_bytes,
7435 aligned_cluster);
7436 if (ret == 0) {
7437 /*
7438 * now pull our allocation out of this
7439 * cluster
7440 */
7441 offset = btrfs_alloc_from_cluster(block_group,
7442 last_ptr,
7443 num_bytes,
7444 search_start,
7445 &max_extent_size);
7446 if (offset) {
7447 /* we found one, proceed */
7448 spin_unlock(&last_ptr->refill_lock);
7449 trace_btrfs_reserve_extent_cluster(root,
7450 block_group, search_start,
7451 num_bytes);
7452 goto checks;
7453 }
7454 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7455 && !failed_cluster_refill) {
7456 spin_unlock(&last_ptr->refill_lock);
7457
7458 failed_cluster_refill = true;
7459 wait_block_group_cache_progress(block_group,
7460 num_bytes + empty_cluster + empty_size);
7461 goto have_block_group;
7462 }
7463
7464 /*
7465 * at this point we either didn't find a cluster
7466 * or we weren't able to allocate a block from our
7467 * cluster. Free the cluster we've been trying
7468 * to use, and go to the next block group
7469 */
7470 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7471 spin_unlock(&last_ptr->refill_lock);
7472 goto loop;
7473 }
7474
7475 unclustered_alloc:
7476 /*
7477 * We are doing an unclustered alloc, set the fragmented flag so
7478 * we don't bother trying to setup a cluster again until we get
7479 * more space.
7480 */
7481 if (unlikely(last_ptr)) {
7482 spin_lock(&last_ptr->lock);
7483 last_ptr->fragmented = 1;
7484 spin_unlock(&last_ptr->lock);
7485 }
7486 spin_lock(&block_group->free_space_ctl->tree_lock);
7487 if (cached &&
7488 block_group->free_space_ctl->free_space <
7489 num_bytes + empty_cluster + empty_size) {
7490 if (block_group->free_space_ctl->free_space >
7491 max_extent_size)
7492 max_extent_size =
7493 block_group->free_space_ctl->free_space;
7494 spin_unlock(&block_group->free_space_ctl->tree_lock);
7495 goto loop;
7496 }
7497 spin_unlock(&block_group->free_space_ctl->tree_lock);
7498
7499 offset = btrfs_find_space_for_alloc(block_group, search_start,
7500 num_bytes, empty_size,
7501 &max_extent_size);
7502 /*
7503 * If we didn't find a chunk, and we haven't failed on this
7504 * block group before, and this block group is in the middle of
7505 * caching and we are ok with waiting, then go ahead and wait
7506 * for progress to be made, and set failed_alloc to true.
7507 *
7508 * If failed_alloc is true then we've already waited on this
7509 * block group once and should move on to the next block group.
7510 */
7511 if (!offset && !failed_alloc && !cached &&
7512 loop > LOOP_CACHING_NOWAIT) {
7513 wait_block_group_cache_progress(block_group,
7514 num_bytes + empty_size);
7515 failed_alloc = true;
7516 goto have_block_group;
7517 } else if (!offset) {
7518 goto loop;
7519 }
7520 checks:
7521 search_start = ALIGN(offset, root->stripesize);
7522
7523 /* move on to the next group */
7524 if (search_start + num_bytes >
7525 block_group->key.objectid + block_group->key.offset) {
7526 btrfs_add_free_space(block_group, offset, num_bytes);
7527 goto loop;
7528 }
7529
7530 if (offset < search_start)
7531 btrfs_add_free_space(block_group, offset,
7532 search_start - offset);
7533 BUG_ON(offset > search_start);
7534
7535 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
7536 alloc_type, delalloc);
7537 if (ret == -EAGAIN) {
7538 btrfs_add_free_space(block_group, offset, num_bytes);
7539 goto loop;
7540 }
7541 btrfs_inc_block_group_reservations(block_group);
7542
7543 /* we are all good, lets return */
7544 ins->objectid = search_start;
7545 ins->offset = num_bytes;
7546
7547 trace_btrfs_reserve_extent(orig_root, block_group,
7548 search_start, num_bytes);
7549 btrfs_release_block_group(block_group, delalloc);
7550 break;
7551 loop:
7552 failed_cluster_refill = false;
7553 failed_alloc = false;
7554 BUG_ON(index != get_block_group_index(block_group));
7555 btrfs_release_block_group(block_group, delalloc);
7556 }
7557 up_read(&space_info->groups_sem);
7558
7559 if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg
7560 && !orig_have_caching_bg)
7561 orig_have_caching_bg = true;
7562
7563 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7564 goto search;
7565
7566 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7567 goto search;
7568
7569 /*
7570 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7571 * caching kthreads as we move along
7572 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7573 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7574 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7575 * again
7576 */
7577 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7578 index = 0;
7579 if (loop == LOOP_CACHING_NOWAIT) {
7580 /*
7581 * We want to skip the LOOP_CACHING_WAIT step if we
7582 * don't have any unached bgs and we've alrelady done a
7583 * full search through.
7584 */
7585 if (orig_have_caching_bg || !full_search)
7586 loop = LOOP_CACHING_WAIT;
7587 else
7588 loop = LOOP_ALLOC_CHUNK;
7589 } else {
7590 loop++;
7591 }
7592
7593 if (loop == LOOP_ALLOC_CHUNK) {
7594 struct btrfs_trans_handle *trans;
7595 int exist = 0;
7596
7597 trans = current->journal_info;
7598 if (trans)
7599 exist = 1;
7600 else
7601 trans = btrfs_join_transaction(root);
7602
7603 if (IS_ERR(trans)) {
7604 ret = PTR_ERR(trans);
7605 goto out;
7606 }
7607
7608 ret = do_chunk_alloc(trans, root, flags,
7609 CHUNK_ALLOC_FORCE);
7610
7611 /*
7612 * If we can't allocate a new chunk we've already looped
7613 * through at least once, move on to the NO_EMPTY_SIZE
7614 * case.
7615 */
7616 if (ret == -ENOSPC)
7617 loop = LOOP_NO_EMPTY_SIZE;
7618
7619 /*
7620 * Do not bail out on ENOSPC since we
7621 * can do more things.
7622 */
7623 if (ret < 0 && ret != -ENOSPC)
7624 btrfs_abort_transaction(trans,
7625 root, ret);
7626 else
7627 ret = 0;
7628 if (!exist)
7629 btrfs_end_transaction(trans, root);
7630 if (ret)
7631 goto out;
7632 }
7633
7634 if (loop == LOOP_NO_EMPTY_SIZE) {
7635 /*
7636 * Don't loop again if we already have no empty_size and
7637 * no empty_cluster.
7638 */
7639 if (empty_size == 0 &&
7640 empty_cluster == 0) {
7641 ret = -ENOSPC;
7642 goto out;
7643 }
7644 empty_size = 0;
7645 empty_cluster = 0;
7646 }
7647
7648 goto search;
7649 } else if (!ins->objectid) {
7650 ret = -ENOSPC;
7651 } else if (ins->objectid) {
7652 if (!use_cluster && last_ptr) {
7653 spin_lock(&last_ptr->lock);
7654 last_ptr->window_start = ins->objectid;
7655 spin_unlock(&last_ptr->lock);
7656 }
7657 ret = 0;
7658 }
7659 out:
7660 if (ret == -ENOSPC) {
7661 spin_lock(&space_info->lock);
7662 space_info->max_extent_size = max_extent_size;
7663 spin_unlock(&space_info->lock);
7664 ins->offset = max_extent_size;
7665 }
7666 return ret;
7667 }
7668
7669 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
7670 int dump_block_groups)
7671 {
7672 struct btrfs_block_group_cache *cache;
7673 int index = 0;
7674
7675 spin_lock(&info->lock);
7676 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
7677 info->flags,
7678 info->total_bytes - info->bytes_used - info->bytes_pinned -
7679 info->bytes_reserved - info->bytes_readonly,
7680 (info->full) ? "" : "not ");
7681 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7682 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7683 info->total_bytes, info->bytes_used, info->bytes_pinned,
7684 info->bytes_reserved, info->bytes_may_use,
7685 info->bytes_readonly);
7686 spin_unlock(&info->lock);
7687
7688 if (!dump_block_groups)
7689 return;
7690
7691 down_read(&info->groups_sem);
7692 again:
7693 list_for_each_entry(cache, &info->block_groups[index], list) {
7694 spin_lock(&cache->lock);
7695 printk(KERN_INFO "BTRFS: "
7696 "block group %llu has %llu bytes, "
7697 "%llu used %llu pinned %llu reserved %s\n",
7698 cache->key.objectid, cache->key.offset,
7699 btrfs_block_group_used(&cache->item), cache->pinned,
7700 cache->reserved, cache->ro ? "[readonly]" : "");
7701 btrfs_dump_free_space(cache, bytes);
7702 spin_unlock(&cache->lock);
7703 }
7704 if (++index < BTRFS_NR_RAID_TYPES)
7705 goto again;
7706 up_read(&info->groups_sem);
7707 }
7708
7709 int btrfs_reserve_extent(struct btrfs_root *root,
7710 u64 num_bytes, u64 min_alloc_size,
7711 u64 empty_size, u64 hint_byte,
7712 struct btrfs_key *ins, int is_data, int delalloc)
7713 {
7714 bool final_tried = num_bytes == min_alloc_size;
7715 u64 flags;
7716 int ret;
7717
7718 flags = btrfs_get_alloc_profile(root, is_data);
7719 again:
7720 WARN_ON(num_bytes < root->sectorsize);
7721 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
7722 flags, delalloc);
7723 if (!ret && !is_data) {
7724 btrfs_dec_block_group_reservations(root->fs_info,
7725 ins->objectid);
7726 } else if (ret == -ENOSPC) {
7727 if (!final_tried && ins->offset) {
7728 num_bytes = min(num_bytes >> 1, ins->offset);
7729 num_bytes = round_down(num_bytes, root->sectorsize);
7730 num_bytes = max(num_bytes, min_alloc_size);
7731 if (num_bytes == min_alloc_size)
7732 final_tried = true;
7733 goto again;
7734 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7735 struct btrfs_space_info *sinfo;
7736
7737 sinfo = __find_space_info(root->fs_info, flags);
7738 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
7739 flags, num_bytes);
7740 if (sinfo)
7741 dump_space_info(sinfo, num_bytes, 1);
7742 }
7743 }
7744
7745 return ret;
7746 }
7747
7748 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
7749 u64 start, u64 len,
7750 int pin, int delalloc)
7751 {
7752 struct btrfs_block_group_cache *cache;
7753 int ret = 0;
7754
7755 cache = btrfs_lookup_block_group(root->fs_info, start);
7756 if (!cache) {
7757 btrfs_err(root->fs_info, "Unable to find block group for %llu",
7758 start);
7759 return -ENOSPC;
7760 }
7761
7762 if (pin)
7763 pin_down_extent(root, cache, start, len, 1);
7764 else {
7765 if (btrfs_test_opt(root, DISCARD))
7766 ret = btrfs_discard_extent(root, start, len, NULL);
7767 btrfs_add_free_space(cache, start, len);
7768 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
7769 }
7770
7771 btrfs_put_block_group(cache);
7772
7773 trace_btrfs_reserved_extent_free(root, start, len);
7774
7775 return ret;
7776 }
7777
7778 int btrfs_free_reserved_extent(struct btrfs_root *root,
7779 u64 start, u64 len, int delalloc)
7780 {
7781 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
7782 }
7783
7784 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
7785 u64 start, u64 len)
7786 {
7787 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
7788 }
7789
7790 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7791 struct btrfs_root *root,
7792 u64 parent, u64 root_objectid,
7793 u64 flags, u64 owner, u64 offset,
7794 struct btrfs_key *ins, int ref_mod)
7795 {
7796 int ret;
7797 struct btrfs_fs_info *fs_info = root->fs_info;
7798 struct btrfs_extent_item *extent_item;
7799 struct btrfs_extent_inline_ref *iref;
7800 struct btrfs_path *path;
7801 struct extent_buffer *leaf;
7802 int type;
7803 u32 size;
7804
7805 if (parent > 0)
7806 type = BTRFS_SHARED_DATA_REF_KEY;
7807 else
7808 type = BTRFS_EXTENT_DATA_REF_KEY;
7809
7810 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7811
7812 path = btrfs_alloc_path();
7813 if (!path)
7814 return -ENOMEM;
7815
7816 path->leave_spinning = 1;
7817 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7818 ins, size);
7819 if (ret) {
7820 btrfs_free_path(path);
7821 return ret;
7822 }
7823
7824 leaf = path->nodes[0];
7825 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7826 struct btrfs_extent_item);
7827 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
7828 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7829 btrfs_set_extent_flags(leaf, extent_item,
7830 flags | BTRFS_EXTENT_FLAG_DATA);
7831
7832 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7833 btrfs_set_extent_inline_ref_type(leaf, iref, type);
7834 if (parent > 0) {
7835 struct btrfs_shared_data_ref *ref;
7836 ref = (struct btrfs_shared_data_ref *)(iref + 1);
7837 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7838 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
7839 } else {
7840 struct btrfs_extent_data_ref *ref;
7841 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
7842 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
7843 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
7844 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
7845 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
7846 }
7847
7848 btrfs_mark_buffer_dirty(path->nodes[0]);
7849 btrfs_free_path(path);
7850
7851 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
7852 ins->offset);
7853 if (ret)
7854 return ret;
7855
7856 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
7857 if (ret) { /* -ENOENT, logic error */
7858 btrfs_err(fs_info, "update block group failed for %llu %llu",
7859 ins->objectid, ins->offset);
7860 BUG();
7861 }
7862 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
7863 return ret;
7864 }
7865
7866 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
7867 struct btrfs_root *root,
7868 u64 parent, u64 root_objectid,
7869 u64 flags, struct btrfs_disk_key *key,
7870 int level, struct btrfs_key *ins)
7871 {
7872 int ret;
7873 struct btrfs_fs_info *fs_info = root->fs_info;
7874 struct btrfs_extent_item *extent_item;
7875 struct btrfs_tree_block_info *block_info;
7876 struct btrfs_extent_inline_ref *iref;
7877 struct btrfs_path *path;
7878 struct extent_buffer *leaf;
7879 u32 size = sizeof(*extent_item) + sizeof(*iref);
7880 u64 num_bytes = ins->offset;
7881 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7882 SKINNY_METADATA);
7883
7884 if (!skinny_metadata)
7885 size += sizeof(*block_info);
7886
7887 path = btrfs_alloc_path();
7888 if (!path) {
7889 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7890 root->nodesize);
7891 return -ENOMEM;
7892 }
7893
7894 path->leave_spinning = 1;
7895 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7896 ins, size);
7897 if (ret) {
7898 btrfs_free_path(path);
7899 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7900 root->nodesize);
7901 return ret;
7902 }
7903
7904 leaf = path->nodes[0];
7905 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7906 struct btrfs_extent_item);
7907 btrfs_set_extent_refs(leaf, extent_item, 1);
7908 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7909 btrfs_set_extent_flags(leaf, extent_item,
7910 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
7911
7912 if (skinny_metadata) {
7913 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7914 num_bytes = root->nodesize;
7915 } else {
7916 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
7917 btrfs_set_tree_block_key(leaf, block_info, key);
7918 btrfs_set_tree_block_level(leaf, block_info, level);
7919 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
7920 }
7921
7922 if (parent > 0) {
7923 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
7924 btrfs_set_extent_inline_ref_type(leaf, iref,
7925 BTRFS_SHARED_BLOCK_REF_KEY);
7926 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7927 } else {
7928 btrfs_set_extent_inline_ref_type(leaf, iref,
7929 BTRFS_TREE_BLOCK_REF_KEY);
7930 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
7931 }
7932
7933 btrfs_mark_buffer_dirty(leaf);
7934 btrfs_free_path(path);
7935
7936 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
7937 num_bytes);
7938 if (ret)
7939 return ret;
7940
7941 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
7942 1);
7943 if (ret) { /* -ENOENT, logic error */
7944 btrfs_err(fs_info, "update block group failed for %llu %llu",
7945 ins->objectid, ins->offset);
7946 BUG();
7947 }
7948
7949 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
7950 return ret;
7951 }
7952
7953 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7954 struct btrfs_root *root,
7955 u64 root_objectid, u64 owner,
7956 u64 offset, u64 ram_bytes,
7957 struct btrfs_key *ins)
7958 {
7959 int ret;
7960
7961 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
7962
7963 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
7964 ins->offset, 0,
7965 root_objectid, owner, offset,
7966 ram_bytes, BTRFS_ADD_DELAYED_EXTENT,
7967 NULL);
7968 return ret;
7969 }
7970
7971 /*
7972 * this is used by the tree logging recovery code. It records that
7973 * an extent has been allocated and makes sure to clear the free
7974 * space cache bits as well
7975 */
7976 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
7977 struct btrfs_root *root,
7978 u64 root_objectid, u64 owner, u64 offset,
7979 struct btrfs_key *ins)
7980 {
7981 int ret;
7982 struct btrfs_block_group_cache *block_group;
7983
7984 /*
7985 * Mixed block groups will exclude before processing the log so we only
7986 * need to do the exlude dance if this fs isn't mixed.
7987 */
7988 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
7989 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
7990 if (ret)
7991 return ret;
7992 }
7993
7994 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
7995 if (!block_group)
7996 return -EINVAL;
7997
7998 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
7999 RESERVE_ALLOC_NO_ACCOUNT, 0);
8000 BUG_ON(ret); /* logic error */
8001 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
8002 0, owner, offset, ins, 1);
8003 btrfs_put_block_group(block_group);
8004 return ret;
8005 }
8006
8007 static struct extent_buffer *
8008 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
8009 u64 bytenr, int level)
8010 {
8011 struct extent_buffer *buf;
8012
8013 buf = btrfs_find_create_tree_block(root, bytenr);
8014 if (!buf)
8015 return ERR_PTR(-ENOMEM);
8016 btrfs_set_header_generation(buf, trans->transid);
8017 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
8018 btrfs_tree_lock(buf);
8019 clean_tree_block(trans, root->fs_info, buf);
8020 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
8021
8022 btrfs_set_lock_blocking(buf);
8023 set_extent_buffer_uptodate(buf);
8024
8025 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8026 buf->log_index = root->log_transid % 2;
8027 /*
8028 * we allow two log transactions at a time, use different
8029 * EXENT bit to differentiate dirty pages.
8030 */
8031 if (buf->log_index == 0)
8032 set_extent_dirty(&root->dirty_log_pages, buf->start,
8033 buf->start + buf->len - 1, GFP_NOFS);
8034 else
8035 set_extent_new(&root->dirty_log_pages, buf->start,
8036 buf->start + buf->len - 1, GFP_NOFS);
8037 } else {
8038 buf->log_index = -1;
8039 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
8040 buf->start + buf->len - 1, GFP_NOFS);
8041 }
8042 trans->blocks_used++;
8043 /* this returns a buffer locked for blocking */
8044 return buf;
8045 }
8046
8047 static struct btrfs_block_rsv *
8048 use_block_rsv(struct btrfs_trans_handle *trans,
8049 struct btrfs_root *root, u32 blocksize)
8050 {
8051 struct btrfs_block_rsv *block_rsv;
8052 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
8053 int ret;
8054 bool global_updated = false;
8055
8056 block_rsv = get_block_rsv(trans, root);
8057
8058 if (unlikely(block_rsv->size == 0))
8059 goto try_reserve;
8060 again:
8061 ret = block_rsv_use_bytes(block_rsv, blocksize);
8062 if (!ret)
8063 return block_rsv;
8064
8065 if (block_rsv->failfast)
8066 return ERR_PTR(ret);
8067
8068 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
8069 global_updated = true;
8070 update_global_block_rsv(root->fs_info);
8071 goto again;
8072 }
8073
8074 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
8075 static DEFINE_RATELIMIT_STATE(_rs,
8076 DEFAULT_RATELIMIT_INTERVAL * 10,
8077 /*DEFAULT_RATELIMIT_BURST*/ 1);
8078 if (__ratelimit(&_rs))
8079 WARN(1, KERN_DEBUG
8080 "BTRFS: block rsv returned %d\n", ret);
8081 }
8082 try_reserve:
8083 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
8084 BTRFS_RESERVE_NO_FLUSH);
8085 if (!ret)
8086 return block_rsv;
8087 /*
8088 * If we couldn't reserve metadata bytes try and use some from
8089 * the global reserve if its space type is the same as the global
8090 * reservation.
8091 */
8092 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
8093 block_rsv->space_info == global_rsv->space_info) {
8094 ret = block_rsv_use_bytes(global_rsv, blocksize);
8095 if (!ret)
8096 return global_rsv;
8097 }
8098 return ERR_PTR(ret);
8099 }
8100
8101 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
8102 struct btrfs_block_rsv *block_rsv, u32 blocksize)
8103 {
8104 block_rsv_add_bytes(block_rsv, blocksize, 0);
8105 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
8106 }
8107
8108 /*
8109 * finds a free extent and does all the dirty work required for allocation
8110 * returns the tree buffer or an ERR_PTR on error.
8111 */
8112 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8113 struct btrfs_root *root,
8114 u64 parent, u64 root_objectid,
8115 struct btrfs_disk_key *key, int level,
8116 u64 hint, u64 empty_size)
8117 {
8118 struct btrfs_key ins;
8119 struct btrfs_block_rsv *block_rsv;
8120 struct extent_buffer *buf;
8121 struct btrfs_delayed_extent_op *extent_op;
8122 u64 flags = 0;
8123 int ret;
8124 u32 blocksize = root->nodesize;
8125 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
8126 SKINNY_METADATA);
8127
8128 if (btrfs_test_is_dummy_root(root)) {
8129 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8130 level);
8131 if (!IS_ERR(buf))
8132 root->alloc_bytenr += blocksize;
8133 return buf;
8134 }
8135
8136 block_rsv = use_block_rsv(trans, root, blocksize);
8137 if (IS_ERR(block_rsv))
8138 return ERR_CAST(block_rsv);
8139
8140 ret = btrfs_reserve_extent(root, blocksize, blocksize,
8141 empty_size, hint, &ins, 0, 0);
8142 if (ret)
8143 goto out_unuse;
8144
8145 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
8146 if (IS_ERR(buf)) {
8147 ret = PTR_ERR(buf);
8148 goto out_free_reserved;
8149 }
8150
8151 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8152 if (parent == 0)
8153 parent = ins.objectid;
8154 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8155 } else
8156 BUG_ON(parent > 0);
8157
8158 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8159 extent_op = btrfs_alloc_delayed_extent_op();
8160 if (!extent_op) {
8161 ret = -ENOMEM;
8162 goto out_free_buf;
8163 }
8164 if (key)
8165 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8166 else
8167 memset(&extent_op->key, 0, sizeof(extent_op->key));
8168 extent_op->flags_to_set = flags;
8169 extent_op->update_key = skinny_metadata ? false : true;
8170 extent_op->update_flags = true;
8171 extent_op->is_data = false;
8172 extent_op->level = level;
8173
8174 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
8175 ins.objectid, ins.offset,
8176 parent, root_objectid, level,
8177 BTRFS_ADD_DELAYED_EXTENT,
8178 extent_op);
8179 if (ret)
8180 goto out_free_delayed;
8181 }
8182 return buf;
8183
8184 out_free_delayed:
8185 btrfs_free_delayed_extent_op(extent_op);
8186 out_free_buf:
8187 free_extent_buffer(buf);
8188 out_free_reserved:
8189 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0);
8190 out_unuse:
8191 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
8192 return ERR_PTR(ret);
8193 }
8194
8195 struct walk_control {
8196 u64 refs[BTRFS_MAX_LEVEL];
8197 u64 flags[BTRFS_MAX_LEVEL];
8198 struct btrfs_key update_progress;
8199 int stage;
8200 int level;
8201 int shared_level;
8202 int update_ref;
8203 int keep_locks;
8204 int reada_slot;
8205 int reada_count;
8206 int for_reloc;
8207 };
8208
8209 #define DROP_REFERENCE 1
8210 #define UPDATE_BACKREF 2
8211
8212 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8213 struct btrfs_root *root,
8214 struct walk_control *wc,
8215 struct btrfs_path *path)
8216 {
8217 u64 bytenr;
8218 u64 generation;
8219 u64 refs;
8220 u64 flags;
8221 u32 nritems;
8222 u32 blocksize;
8223 struct btrfs_key key;
8224 struct extent_buffer *eb;
8225 int ret;
8226 int slot;
8227 int nread = 0;
8228
8229 if (path->slots[wc->level] < wc->reada_slot) {
8230 wc->reada_count = wc->reada_count * 2 / 3;
8231 wc->reada_count = max(wc->reada_count, 2);
8232 } else {
8233 wc->reada_count = wc->reada_count * 3 / 2;
8234 wc->reada_count = min_t(int, wc->reada_count,
8235 BTRFS_NODEPTRS_PER_BLOCK(root));
8236 }
8237
8238 eb = path->nodes[wc->level];
8239 nritems = btrfs_header_nritems(eb);
8240 blocksize = root->nodesize;
8241
8242 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8243 if (nread >= wc->reada_count)
8244 break;
8245
8246 cond_resched();
8247 bytenr = btrfs_node_blockptr(eb, slot);
8248 generation = btrfs_node_ptr_generation(eb, slot);
8249
8250 if (slot == path->slots[wc->level])
8251 goto reada;
8252
8253 if (wc->stage == UPDATE_BACKREF &&
8254 generation <= root->root_key.offset)
8255 continue;
8256
8257 /* We don't lock the tree block, it's OK to be racy here */
8258 ret = btrfs_lookup_extent_info(trans, root, bytenr,
8259 wc->level - 1, 1, &refs,
8260 &flags);
8261 /* We don't care about errors in readahead. */
8262 if (ret < 0)
8263 continue;
8264 BUG_ON(refs == 0);
8265
8266 if (wc->stage == DROP_REFERENCE) {
8267 if (refs == 1)
8268 goto reada;
8269
8270 if (wc->level == 1 &&
8271 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8272 continue;
8273 if (!wc->update_ref ||
8274 generation <= root->root_key.offset)
8275 continue;
8276 btrfs_node_key_to_cpu(eb, &key, slot);
8277 ret = btrfs_comp_cpu_keys(&key,
8278 &wc->update_progress);
8279 if (ret < 0)
8280 continue;
8281 } else {
8282 if (wc->level == 1 &&
8283 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8284 continue;
8285 }
8286 reada:
8287 readahead_tree_block(root, bytenr);
8288 nread++;
8289 }
8290 wc->reada_slot = slot;
8291 }
8292
8293 /*
8294 * These may not be seen by the usual inc/dec ref code so we have to
8295 * add them here.
8296 */
8297 static int record_one_subtree_extent(struct btrfs_trans_handle *trans,
8298 struct btrfs_root *root, u64 bytenr,
8299 u64 num_bytes)
8300 {
8301 struct btrfs_qgroup_extent_record *qrecord;
8302 struct btrfs_delayed_ref_root *delayed_refs;
8303
8304 qrecord = kmalloc(sizeof(*qrecord), GFP_NOFS);
8305 if (!qrecord)
8306 return -ENOMEM;
8307
8308 qrecord->bytenr = bytenr;
8309 qrecord->num_bytes = num_bytes;
8310 qrecord->old_roots = NULL;
8311
8312 delayed_refs = &trans->transaction->delayed_refs;
8313 spin_lock(&delayed_refs->lock);
8314 if (btrfs_qgroup_insert_dirty_extent(delayed_refs, qrecord))
8315 kfree(qrecord);
8316 spin_unlock(&delayed_refs->lock);
8317
8318 return 0;
8319 }
8320
8321 static int account_leaf_items(struct btrfs_trans_handle *trans,
8322 struct btrfs_root *root,
8323 struct extent_buffer *eb)
8324 {
8325 int nr = btrfs_header_nritems(eb);
8326 int i, extent_type, ret;
8327 struct btrfs_key key;
8328 struct btrfs_file_extent_item *fi;
8329 u64 bytenr, num_bytes;
8330
8331 /* We can be called directly from walk_up_proc() */
8332 if (!root->fs_info->quota_enabled)
8333 return 0;
8334
8335 for (i = 0; i < nr; i++) {
8336 btrfs_item_key_to_cpu(eb, &key, i);
8337
8338 if (key.type != BTRFS_EXTENT_DATA_KEY)
8339 continue;
8340
8341 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
8342 /* filter out non qgroup-accountable extents */
8343 extent_type = btrfs_file_extent_type(eb, fi);
8344
8345 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
8346 continue;
8347
8348 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
8349 if (!bytenr)
8350 continue;
8351
8352 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
8353
8354 ret = record_one_subtree_extent(trans, root, bytenr, num_bytes);
8355 if (ret)
8356 return ret;
8357 }
8358 return 0;
8359 }
8360
8361 /*
8362 * Walk up the tree from the bottom, freeing leaves and any interior
8363 * nodes which have had all slots visited. If a node (leaf or
8364 * interior) is freed, the node above it will have it's slot
8365 * incremented. The root node will never be freed.
8366 *
8367 * At the end of this function, we should have a path which has all
8368 * slots incremented to the next position for a search. If we need to
8369 * read a new node it will be NULL and the node above it will have the
8370 * correct slot selected for a later read.
8371 *
8372 * If we increment the root nodes slot counter past the number of
8373 * elements, 1 is returned to signal completion of the search.
8374 */
8375 static int adjust_slots_upwards(struct btrfs_root *root,
8376 struct btrfs_path *path, int root_level)
8377 {
8378 int level = 0;
8379 int nr, slot;
8380 struct extent_buffer *eb;
8381
8382 if (root_level == 0)
8383 return 1;
8384
8385 while (level <= root_level) {
8386 eb = path->nodes[level];
8387 nr = btrfs_header_nritems(eb);
8388 path->slots[level]++;
8389 slot = path->slots[level];
8390 if (slot >= nr || level == 0) {
8391 /*
8392 * Don't free the root - we will detect this
8393 * condition after our loop and return a
8394 * positive value for caller to stop walking the tree.
8395 */
8396 if (level != root_level) {
8397 btrfs_tree_unlock_rw(eb, path->locks[level]);
8398 path->locks[level] = 0;
8399
8400 free_extent_buffer(eb);
8401 path->nodes[level] = NULL;
8402 path->slots[level] = 0;
8403 }
8404 } else {
8405 /*
8406 * We have a valid slot to walk back down
8407 * from. Stop here so caller can process these
8408 * new nodes.
8409 */
8410 break;
8411 }
8412
8413 level++;
8414 }
8415
8416 eb = path->nodes[root_level];
8417 if (path->slots[root_level] >= btrfs_header_nritems(eb))
8418 return 1;
8419
8420 return 0;
8421 }
8422
8423 /*
8424 * root_eb is the subtree root and is locked before this function is called.
8425 */
8426 static int account_shared_subtree(struct btrfs_trans_handle *trans,
8427 struct btrfs_root *root,
8428 struct extent_buffer *root_eb,
8429 u64 root_gen,
8430 int root_level)
8431 {
8432 int ret = 0;
8433 int level;
8434 struct extent_buffer *eb = root_eb;
8435 struct btrfs_path *path = NULL;
8436
8437 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
8438 BUG_ON(root_eb == NULL);
8439
8440 if (!root->fs_info->quota_enabled)
8441 return 0;
8442
8443 if (!extent_buffer_uptodate(root_eb)) {
8444 ret = btrfs_read_buffer(root_eb, root_gen);
8445 if (ret)
8446 goto out;
8447 }
8448
8449 if (root_level == 0) {
8450 ret = account_leaf_items(trans, root, root_eb);
8451 goto out;
8452 }
8453
8454 path = btrfs_alloc_path();
8455 if (!path)
8456 return -ENOMEM;
8457
8458 /*
8459 * Walk down the tree. Missing extent blocks are filled in as
8460 * we go. Metadata is accounted every time we read a new
8461 * extent block.
8462 *
8463 * When we reach a leaf, we account for file extent items in it,
8464 * walk back up the tree (adjusting slot pointers as we go)
8465 * and restart the search process.
8466 */
8467 extent_buffer_get(root_eb); /* For path */
8468 path->nodes[root_level] = root_eb;
8469 path->slots[root_level] = 0;
8470 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
8471 walk_down:
8472 level = root_level;
8473 while (level >= 0) {
8474 if (path->nodes[level] == NULL) {
8475 int parent_slot;
8476 u64 child_gen;
8477 u64 child_bytenr;
8478
8479 /* We need to get child blockptr/gen from
8480 * parent before we can read it. */
8481 eb = path->nodes[level + 1];
8482 parent_slot = path->slots[level + 1];
8483 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
8484 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
8485
8486 eb = read_tree_block(root, child_bytenr, child_gen);
8487 if (IS_ERR(eb)) {
8488 ret = PTR_ERR(eb);
8489 goto out;
8490 } else if (!extent_buffer_uptodate(eb)) {
8491 free_extent_buffer(eb);
8492 ret = -EIO;
8493 goto out;
8494 }
8495
8496 path->nodes[level] = eb;
8497 path->slots[level] = 0;
8498
8499 btrfs_tree_read_lock(eb);
8500 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
8501 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
8502
8503 ret = record_one_subtree_extent(trans, root, child_bytenr,
8504 root->nodesize);
8505 if (ret)
8506 goto out;
8507 }
8508
8509 if (level == 0) {
8510 ret = account_leaf_items(trans, root, path->nodes[level]);
8511 if (ret)
8512 goto out;
8513
8514 /* Nonzero return here means we completed our search */
8515 ret = adjust_slots_upwards(root, path, root_level);
8516 if (ret)
8517 break;
8518
8519 /* Restart search with new slots */
8520 goto walk_down;
8521 }
8522
8523 level--;
8524 }
8525
8526 ret = 0;
8527 out:
8528 btrfs_free_path(path);
8529
8530 return ret;
8531 }
8532
8533 /*
8534 * helper to process tree block while walking down the tree.
8535 *
8536 * when wc->stage == UPDATE_BACKREF, this function updates
8537 * back refs for pointers in the block.
8538 *
8539 * NOTE: return value 1 means we should stop walking down.
8540 */
8541 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8542 struct btrfs_root *root,
8543 struct btrfs_path *path,
8544 struct walk_control *wc, int lookup_info)
8545 {
8546 int level = wc->level;
8547 struct extent_buffer *eb = path->nodes[level];
8548 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8549 int ret;
8550
8551 if (wc->stage == UPDATE_BACKREF &&
8552 btrfs_header_owner(eb) != root->root_key.objectid)
8553 return 1;
8554
8555 /*
8556 * when reference count of tree block is 1, it won't increase
8557 * again. once full backref flag is set, we never clear it.
8558 */
8559 if (lookup_info &&
8560 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8561 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8562 BUG_ON(!path->locks[level]);
8563 ret = btrfs_lookup_extent_info(trans, root,
8564 eb->start, level, 1,
8565 &wc->refs[level],
8566 &wc->flags[level]);
8567 BUG_ON(ret == -ENOMEM);
8568 if (ret)
8569 return ret;
8570 BUG_ON(wc->refs[level] == 0);
8571 }
8572
8573 if (wc->stage == DROP_REFERENCE) {
8574 if (wc->refs[level] > 1)
8575 return 1;
8576
8577 if (path->locks[level] && !wc->keep_locks) {
8578 btrfs_tree_unlock_rw(eb, path->locks[level]);
8579 path->locks[level] = 0;
8580 }
8581 return 0;
8582 }
8583
8584 /* wc->stage == UPDATE_BACKREF */
8585 if (!(wc->flags[level] & flag)) {
8586 BUG_ON(!path->locks[level]);
8587 ret = btrfs_inc_ref(trans, root, eb, 1);
8588 BUG_ON(ret); /* -ENOMEM */
8589 ret = btrfs_dec_ref(trans, root, eb, 0);
8590 BUG_ON(ret); /* -ENOMEM */
8591 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
8592 eb->len, flag,
8593 btrfs_header_level(eb), 0);
8594 BUG_ON(ret); /* -ENOMEM */
8595 wc->flags[level] |= flag;
8596 }
8597
8598 /*
8599 * the block is shared by multiple trees, so it's not good to
8600 * keep the tree lock
8601 */
8602 if (path->locks[level] && level > 0) {
8603 btrfs_tree_unlock_rw(eb, path->locks[level]);
8604 path->locks[level] = 0;
8605 }
8606 return 0;
8607 }
8608
8609 /*
8610 * helper to process tree block pointer.
8611 *
8612 * when wc->stage == DROP_REFERENCE, this function checks
8613 * reference count of the block pointed to. if the block
8614 * is shared and we need update back refs for the subtree
8615 * rooted at the block, this function changes wc->stage to
8616 * UPDATE_BACKREF. if the block is shared and there is no
8617 * need to update back, this function drops the reference
8618 * to the block.
8619 *
8620 * NOTE: return value 1 means we should stop walking down.
8621 */
8622 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8623 struct btrfs_root *root,
8624 struct btrfs_path *path,
8625 struct walk_control *wc, int *lookup_info)
8626 {
8627 u64 bytenr;
8628 u64 generation;
8629 u64 parent;
8630 u32 blocksize;
8631 struct btrfs_key key;
8632 struct extent_buffer *next;
8633 int level = wc->level;
8634 int reada = 0;
8635 int ret = 0;
8636 bool need_account = false;
8637
8638 generation = btrfs_node_ptr_generation(path->nodes[level],
8639 path->slots[level]);
8640 /*
8641 * if the lower level block was created before the snapshot
8642 * was created, we know there is no need to update back refs
8643 * for the subtree
8644 */
8645 if (wc->stage == UPDATE_BACKREF &&
8646 generation <= root->root_key.offset) {
8647 *lookup_info = 1;
8648 return 1;
8649 }
8650
8651 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8652 blocksize = root->nodesize;
8653
8654 next = btrfs_find_tree_block(root->fs_info, bytenr);
8655 if (!next) {
8656 next = btrfs_find_create_tree_block(root, bytenr);
8657 if (!next)
8658 return -ENOMEM;
8659 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8660 level - 1);
8661 reada = 1;
8662 }
8663 btrfs_tree_lock(next);
8664 btrfs_set_lock_blocking(next);
8665
8666 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
8667 &wc->refs[level - 1],
8668 &wc->flags[level - 1]);
8669 if (ret < 0) {
8670 btrfs_tree_unlock(next);
8671 return ret;
8672 }
8673
8674 if (unlikely(wc->refs[level - 1] == 0)) {
8675 btrfs_err(root->fs_info, "Missing references.");
8676 BUG();
8677 }
8678 *lookup_info = 0;
8679
8680 if (wc->stage == DROP_REFERENCE) {
8681 if (wc->refs[level - 1] > 1) {
8682 need_account = true;
8683 if (level == 1 &&
8684 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8685 goto skip;
8686
8687 if (!wc->update_ref ||
8688 generation <= root->root_key.offset)
8689 goto skip;
8690
8691 btrfs_node_key_to_cpu(path->nodes[level], &key,
8692 path->slots[level]);
8693 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8694 if (ret < 0)
8695 goto skip;
8696
8697 wc->stage = UPDATE_BACKREF;
8698 wc->shared_level = level - 1;
8699 }
8700 } else {
8701 if (level == 1 &&
8702 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8703 goto skip;
8704 }
8705
8706 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8707 btrfs_tree_unlock(next);
8708 free_extent_buffer(next);
8709 next = NULL;
8710 *lookup_info = 1;
8711 }
8712
8713 if (!next) {
8714 if (reada && level == 1)
8715 reada_walk_down(trans, root, wc, path);
8716 next = read_tree_block(root, bytenr, generation);
8717 if (IS_ERR(next)) {
8718 return PTR_ERR(next);
8719 } else if (!extent_buffer_uptodate(next)) {
8720 free_extent_buffer(next);
8721 return -EIO;
8722 }
8723 btrfs_tree_lock(next);
8724 btrfs_set_lock_blocking(next);
8725 }
8726
8727 level--;
8728 BUG_ON(level != btrfs_header_level(next));
8729 path->nodes[level] = next;
8730 path->slots[level] = 0;
8731 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8732 wc->level = level;
8733 if (wc->level == 1)
8734 wc->reada_slot = 0;
8735 return 0;
8736 skip:
8737 wc->refs[level - 1] = 0;
8738 wc->flags[level - 1] = 0;
8739 if (wc->stage == DROP_REFERENCE) {
8740 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8741 parent = path->nodes[level]->start;
8742 } else {
8743 BUG_ON(root->root_key.objectid !=
8744 btrfs_header_owner(path->nodes[level]));
8745 parent = 0;
8746 }
8747
8748 if (need_account) {
8749 ret = account_shared_subtree(trans, root, next,
8750 generation, level - 1);
8751 if (ret) {
8752 btrfs_err_rl(root->fs_info,
8753 "Error "
8754 "%d accounting shared subtree. Quota "
8755 "is out of sync, rescan required.",
8756 ret);
8757 }
8758 }
8759 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
8760 root->root_key.objectid, level - 1, 0);
8761 BUG_ON(ret); /* -ENOMEM */
8762 }
8763 btrfs_tree_unlock(next);
8764 free_extent_buffer(next);
8765 *lookup_info = 1;
8766 return 1;
8767 }
8768
8769 /*
8770 * helper to process tree block while walking up the tree.
8771 *
8772 * when wc->stage == DROP_REFERENCE, this function drops
8773 * reference count on the block.
8774 *
8775 * when wc->stage == UPDATE_BACKREF, this function changes
8776 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8777 * to UPDATE_BACKREF previously while processing the block.
8778 *
8779 * NOTE: return value 1 means we should stop walking up.
8780 */
8781 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8782 struct btrfs_root *root,
8783 struct btrfs_path *path,
8784 struct walk_control *wc)
8785 {
8786 int ret;
8787 int level = wc->level;
8788 struct extent_buffer *eb = path->nodes[level];
8789 u64 parent = 0;
8790
8791 if (wc->stage == UPDATE_BACKREF) {
8792 BUG_ON(wc->shared_level < level);
8793 if (level < wc->shared_level)
8794 goto out;
8795
8796 ret = find_next_key(path, level + 1, &wc->update_progress);
8797 if (ret > 0)
8798 wc->update_ref = 0;
8799
8800 wc->stage = DROP_REFERENCE;
8801 wc->shared_level = -1;
8802 path->slots[level] = 0;
8803
8804 /*
8805 * check reference count again if the block isn't locked.
8806 * we should start walking down the tree again if reference
8807 * count is one.
8808 */
8809 if (!path->locks[level]) {
8810 BUG_ON(level == 0);
8811 btrfs_tree_lock(eb);
8812 btrfs_set_lock_blocking(eb);
8813 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8814
8815 ret = btrfs_lookup_extent_info(trans, root,
8816 eb->start, level, 1,
8817 &wc->refs[level],
8818 &wc->flags[level]);
8819 if (ret < 0) {
8820 btrfs_tree_unlock_rw(eb, path->locks[level]);
8821 path->locks[level] = 0;
8822 return ret;
8823 }
8824 BUG_ON(wc->refs[level] == 0);
8825 if (wc->refs[level] == 1) {
8826 btrfs_tree_unlock_rw(eb, path->locks[level]);
8827 path->locks[level] = 0;
8828 return 1;
8829 }
8830 }
8831 }
8832
8833 /* wc->stage == DROP_REFERENCE */
8834 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8835
8836 if (wc->refs[level] == 1) {
8837 if (level == 0) {
8838 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8839 ret = btrfs_dec_ref(trans, root, eb, 1);
8840 else
8841 ret = btrfs_dec_ref(trans, root, eb, 0);
8842 BUG_ON(ret); /* -ENOMEM */
8843 ret = account_leaf_items(trans, root, eb);
8844 if (ret) {
8845 btrfs_err_rl(root->fs_info,
8846 "error "
8847 "%d accounting leaf items. Quota "
8848 "is out of sync, rescan required.",
8849 ret);
8850 }
8851 }
8852 /* make block locked assertion in clean_tree_block happy */
8853 if (!path->locks[level] &&
8854 btrfs_header_generation(eb) == trans->transid) {
8855 btrfs_tree_lock(eb);
8856 btrfs_set_lock_blocking(eb);
8857 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8858 }
8859 clean_tree_block(trans, root->fs_info, eb);
8860 }
8861
8862 if (eb == root->node) {
8863 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8864 parent = eb->start;
8865 else
8866 BUG_ON(root->root_key.objectid !=
8867 btrfs_header_owner(eb));
8868 } else {
8869 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8870 parent = path->nodes[level + 1]->start;
8871 else
8872 BUG_ON(root->root_key.objectid !=
8873 btrfs_header_owner(path->nodes[level + 1]));
8874 }
8875
8876 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8877 out:
8878 wc->refs[level] = 0;
8879 wc->flags[level] = 0;
8880 return 0;
8881 }
8882
8883 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8884 struct btrfs_root *root,
8885 struct btrfs_path *path,
8886 struct walk_control *wc)
8887 {
8888 int level = wc->level;
8889 int lookup_info = 1;
8890 int ret;
8891
8892 while (level >= 0) {
8893 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8894 if (ret > 0)
8895 break;
8896
8897 if (level == 0)
8898 break;
8899
8900 if (path->slots[level] >=
8901 btrfs_header_nritems(path->nodes[level]))
8902 break;
8903
8904 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8905 if (ret > 0) {
8906 path->slots[level]++;
8907 continue;
8908 } else if (ret < 0)
8909 return ret;
8910 level = wc->level;
8911 }
8912 return 0;
8913 }
8914
8915 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8916 struct btrfs_root *root,
8917 struct btrfs_path *path,
8918 struct walk_control *wc, int max_level)
8919 {
8920 int level = wc->level;
8921 int ret;
8922
8923 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8924 while (level < max_level && path->nodes[level]) {
8925 wc->level = level;
8926 if (path->slots[level] + 1 <
8927 btrfs_header_nritems(path->nodes[level])) {
8928 path->slots[level]++;
8929 return 0;
8930 } else {
8931 ret = walk_up_proc(trans, root, path, wc);
8932 if (ret > 0)
8933 return 0;
8934
8935 if (path->locks[level]) {
8936 btrfs_tree_unlock_rw(path->nodes[level],
8937 path->locks[level]);
8938 path->locks[level] = 0;
8939 }
8940 free_extent_buffer(path->nodes[level]);
8941 path->nodes[level] = NULL;
8942 level++;
8943 }
8944 }
8945 return 1;
8946 }
8947
8948 /*
8949 * drop a subvolume tree.
8950 *
8951 * this function traverses the tree freeing any blocks that only
8952 * referenced by the tree.
8953 *
8954 * when a shared tree block is found. this function decreases its
8955 * reference count by one. if update_ref is true, this function
8956 * also make sure backrefs for the shared block and all lower level
8957 * blocks are properly updated.
8958 *
8959 * If called with for_reloc == 0, may exit early with -EAGAIN
8960 */
8961 int btrfs_drop_snapshot(struct btrfs_root *root,
8962 struct btrfs_block_rsv *block_rsv, int update_ref,
8963 int for_reloc)
8964 {
8965 struct btrfs_path *path;
8966 struct btrfs_trans_handle *trans;
8967 struct btrfs_root *tree_root = root->fs_info->tree_root;
8968 struct btrfs_root_item *root_item = &root->root_item;
8969 struct walk_control *wc;
8970 struct btrfs_key key;
8971 int err = 0;
8972 int ret;
8973 int level;
8974 bool root_dropped = false;
8975
8976 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
8977
8978 path = btrfs_alloc_path();
8979 if (!path) {
8980 err = -ENOMEM;
8981 goto out;
8982 }
8983
8984 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8985 if (!wc) {
8986 btrfs_free_path(path);
8987 err = -ENOMEM;
8988 goto out;
8989 }
8990
8991 trans = btrfs_start_transaction(tree_root, 0);
8992 if (IS_ERR(trans)) {
8993 err = PTR_ERR(trans);
8994 goto out_free;
8995 }
8996
8997 if (block_rsv)
8998 trans->block_rsv = block_rsv;
8999
9000 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
9001 level = btrfs_header_level(root->node);
9002 path->nodes[level] = btrfs_lock_root_node(root);
9003 btrfs_set_lock_blocking(path->nodes[level]);
9004 path->slots[level] = 0;
9005 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9006 memset(&wc->update_progress, 0,
9007 sizeof(wc->update_progress));
9008 } else {
9009 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
9010 memcpy(&wc->update_progress, &key,
9011 sizeof(wc->update_progress));
9012
9013 level = root_item->drop_level;
9014 BUG_ON(level == 0);
9015 path->lowest_level = level;
9016 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
9017 path->lowest_level = 0;
9018 if (ret < 0) {
9019 err = ret;
9020 goto out_end_trans;
9021 }
9022 WARN_ON(ret > 0);
9023
9024 /*
9025 * unlock our path, this is safe because only this
9026 * function is allowed to delete this snapshot
9027 */
9028 btrfs_unlock_up_safe(path, 0);
9029
9030 level = btrfs_header_level(root->node);
9031 while (1) {
9032 btrfs_tree_lock(path->nodes[level]);
9033 btrfs_set_lock_blocking(path->nodes[level]);
9034 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9035
9036 ret = btrfs_lookup_extent_info(trans, root,
9037 path->nodes[level]->start,
9038 level, 1, &wc->refs[level],
9039 &wc->flags[level]);
9040 if (ret < 0) {
9041 err = ret;
9042 goto out_end_trans;
9043 }
9044 BUG_ON(wc->refs[level] == 0);
9045
9046 if (level == root_item->drop_level)
9047 break;
9048
9049 btrfs_tree_unlock(path->nodes[level]);
9050 path->locks[level] = 0;
9051 WARN_ON(wc->refs[level] != 1);
9052 level--;
9053 }
9054 }
9055
9056 wc->level = level;
9057 wc->shared_level = -1;
9058 wc->stage = DROP_REFERENCE;
9059 wc->update_ref = update_ref;
9060 wc->keep_locks = 0;
9061 wc->for_reloc = for_reloc;
9062 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
9063
9064 while (1) {
9065
9066 ret = walk_down_tree(trans, root, path, wc);
9067 if (ret < 0) {
9068 err = ret;
9069 break;
9070 }
9071
9072 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
9073 if (ret < 0) {
9074 err = ret;
9075 break;
9076 }
9077
9078 if (ret > 0) {
9079 BUG_ON(wc->stage != DROP_REFERENCE);
9080 break;
9081 }
9082
9083 if (wc->stage == DROP_REFERENCE) {
9084 level = wc->level;
9085 btrfs_node_key(path->nodes[level],
9086 &root_item->drop_progress,
9087 path->slots[level]);
9088 root_item->drop_level = level;
9089 }
9090
9091 BUG_ON(wc->level == 0);
9092 if (btrfs_should_end_transaction(trans, tree_root) ||
9093 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
9094 ret = btrfs_update_root(trans, tree_root,
9095 &root->root_key,
9096 root_item);
9097 if (ret) {
9098 btrfs_abort_transaction(trans, tree_root, ret);
9099 err = ret;
9100 goto out_end_trans;
9101 }
9102
9103 btrfs_end_transaction_throttle(trans, tree_root);
9104 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
9105 pr_debug("BTRFS: drop snapshot early exit\n");
9106 err = -EAGAIN;
9107 goto out_free;
9108 }
9109
9110 trans = btrfs_start_transaction(tree_root, 0);
9111 if (IS_ERR(trans)) {
9112 err = PTR_ERR(trans);
9113 goto out_free;
9114 }
9115 if (block_rsv)
9116 trans->block_rsv = block_rsv;
9117 }
9118 }
9119 btrfs_release_path(path);
9120 if (err)
9121 goto out_end_trans;
9122
9123 ret = btrfs_del_root(trans, tree_root, &root->root_key);
9124 if (ret) {
9125 btrfs_abort_transaction(trans, tree_root, ret);
9126 goto out_end_trans;
9127 }
9128
9129 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9130 ret = btrfs_find_root(tree_root, &root->root_key, path,
9131 NULL, NULL);
9132 if (ret < 0) {
9133 btrfs_abort_transaction(trans, tree_root, ret);
9134 err = ret;
9135 goto out_end_trans;
9136 } else if (ret > 0) {
9137 /* if we fail to delete the orphan item this time
9138 * around, it'll get picked up the next time.
9139 *
9140 * The most common failure here is just -ENOENT.
9141 */
9142 btrfs_del_orphan_item(trans, tree_root,
9143 root->root_key.objectid);
9144 }
9145 }
9146
9147 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9148 btrfs_add_dropped_root(trans, root);
9149 } else {
9150 free_extent_buffer(root->node);
9151 free_extent_buffer(root->commit_root);
9152 btrfs_put_fs_root(root);
9153 }
9154 root_dropped = true;
9155 out_end_trans:
9156 btrfs_end_transaction_throttle(trans, tree_root);
9157 out_free:
9158 kfree(wc);
9159 btrfs_free_path(path);
9160 out:
9161 /*
9162 * So if we need to stop dropping the snapshot for whatever reason we
9163 * need to make sure to add it back to the dead root list so that we
9164 * keep trying to do the work later. This also cleans up roots if we
9165 * don't have it in the radix (like when we recover after a power fail
9166 * or unmount) so we don't leak memory.
9167 */
9168 if (!for_reloc && root_dropped == false)
9169 btrfs_add_dead_root(root);
9170 if (err && err != -EAGAIN)
9171 btrfs_std_error(root->fs_info, err, NULL);
9172 return err;
9173 }
9174
9175 /*
9176 * drop subtree rooted at tree block 'node'.
9177 *
9178 * NOTE: this function will unlock and release tree block 'node'
9179 * only used by relocation code
9180 */
9181 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9182 struct btrfs_root *root,
9183 struct extent_buffer *node,
9184 struct extent_buffer *parent)
9185 {
9186 struct btrfs_path *path;
9187 struct walk_control *wc;
9188 int level;
9189 int parent_level;
9190 int ret = 0;
9191 int wret;
9192
9193 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9194
9195 path = btrfs_alloc_path();
9196 if (!path)
9197 return -ENOMEM;
9198
9199 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9200 if (!wc) {
9201 btrfs_free_path(path);
9202 return -ENOMEM;
9203 }
9204
9205 btrfs_assert_tree_locked(parent);
9206 parent_level = btrfs_header_level(parent);
9207 extent_buffer_get(parent);
9208 path->nodes[parent_level] = parent;
9209 path->slots[parent_level] = btrfs_header_nritems(parent);
9210
9211 btrfs_assert_tree_locked(node);
9212 level = btrfs_header_level(node);
9213 path->nodes[level] = node;
9214 path->slots[level] = 0;
9215 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9216
9217 wc->refs[parent_level] = 1;
9218 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9219 wc->level = level;
9220 wc->shared_level = -1;
9221 wc->stage = DROP_REFERENCE;
9222 wc->update_ref = 0;
9223 wc->keep_locks = 1;
9224 wc->for_reloc = 1;
9225 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
9226
9227 while (1) {
9228 wret = walk_down_tree(trans, root, path, wc);
9229 if (wret < 0) {
9230 ret = wret;
9231 break;
9232 }
9233
9234 wret = walk_up_tree(trans, root, path, wc, parent_level);
9235 if (wret < 0)
9236 ret = wret;
9237 if (wret != 0)
9238 break;
9239 }
9240
9241 kfree(wc);
9242 btrfs_free_path(path);
9243 return ret;
9244 }
9245
9246 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
9247 {
9248 u64 num_devices;
9249 u64 stripped;
9250
9251 /*
9252 * if restripe for this chunk_type is on pick target profile and
9253 * return, otherwise do the usual balance
9254 */
9255 stripped = get_restripe_target(root->fs_info, flags);
9256 if (stripped)
9257 return extended_to_chunk(stripped);
9258
9259 num_devices = root->fs_info->fs_devices->rw_devices;
9260
9261 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9262 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9263 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9264
9265 if (num_devices == 1) {
9266 stripped |= BTRFS_BLOCK_GROUP_DUP;
9267 stripped = flags & ~stripped;
9268
9269 /* turn raid0 into single device chunks */
9270 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9271 return stripped;
9272
9273 /* turn mirroring into duplication */
9274 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9275 BTRFS_BLOCK_GROUP_RAID10))
9276 return stripped | BTRFS_BLOCK_GROUP_DUP;
9277 } else {
9278 /* they already had raid on here, just return */
9279 if (flags & stripped)
9280 return flags;
9281
9282 stripped |= BTRFS_BLOCK_GROUP_DUP;
9283 stripped = flags & ~stripped;
9284
9285 /* switch duplicated blocks with raid1 */
9286 if (flags & BTRFS_BLOCK_GROUP_DUP)
9287 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9288
9289 /* this is drive concat, leave it alone */
9290 }
9291
9292 return flags;
9293 }
9294
9295 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9296 {
9297 struct btrfs_space_info *sinfo = cache->space_info;
9298 u64 num_bytes;
9299 u64 min_allocable_bytes;
9300 int ret = -ENOSPC;
9301
9302 /*
9303 * We need some metadata space and system metadata space for
9304 * allocating chunks in some corner cases until we force to set
9305 * it to be readonly.
9306 */
9307 if ((sinfo->flags &
9308 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9309 !force)
9310 min_allocable_bytes = SZ_1M;
9311 else
9312 min_allocable_bytes = 0;
9313
9314 spin_lock(&sinfo->lock);
9315 spin_lock(&cache->lock);
9316
9317 if (cache->ro) {
9318 cache->ro++;
9319 ret = 0;
9320 goto out;
9321 }
9322
9323 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9324 cache->bytes_super - btrfs_block_group_used(&cache->item);
9325
9326 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
9327 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
9328 min_allocable_bytes <= sinfo->total_bytes) {
9329 sinfo->bytes_readonly += num_bytes;
9330 cache->ro++;
9331 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9332 ret = 0;
9333 }
9334 out:
9335 spin_unlock(&cache->lock);
9336 spin_unlock(&sinfo->lock);
9337 return ret;
9338 }
9339
9340 int btrfs_inc_block_group_ro(struct btrfs_root *root,
9341 struct btrfs_block_group_cache *cache)
9342
9343 {
9344 struct btrfs_trans_handle *trans;
9345 u64 alloc_flags;
9346 int ret;
9347
9348 again:
9349 trans = btrfs_join_transaction(root);
9350 if (IS_ERR(trans))
9351 return PTR_ERR(trans);
9352
9353 /*
9354 * we're not allowed to set block groups readonly after the dirty
9355 * block groups cache has started writing. If it already started,
9356 * back off and let this transaction commit
9357 */
9358 mutex_lock(&root->fs_info->ro_block_group_mutex);
9359 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9360 u64 transid = trans->transid;
9361
9362 mutex_unlock(&root->fs_info->ro_block_group_mutex);
9363 btrfs_end_transaction(trans, root);
9364
9365 ret = btrfs_wait_for_commit(root, transid);
9366 if (ret)
9367 return ret;
9368 goto again;
9369 }
9370
9371 /*
9372 * if we are changing raid levels, try to allocate a corresponding
9373 * block group with the new raid level.
9374 */
9375 alloc_flags = update_block_group_flags(root, cache->flags);
9376 if (alloc_flags != cache->flags) {
9377 ret = do_chunk_alloc(trans, root, alloc_flags,
9378 CHUNK_ALLOC_FORCE);
9379 /*
9380 * ENOSPC is allowed here, we may have enough space
9381 * already allocated at the new raid level to
9382 * carry on
9383 */
9384 if (ret == -ENOSPC)
9385 ret = 0;
9386 if (ret < 0)
9387 goto out;
9388 }
9389
9390 ret = inc_block_group_ro(cache, 0);
9391 if (!ret)
9392 goto out;
9393 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
9394 ret = do_chunk_alloc(trans, root, alloc_flags,
9395 CHUNK_ALLOC_FORCE);
9396 if (ret < 0)
9397 goto out;
9398 ret = inc_block_group_ro(cache, 0);
9399 out:
9400 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9401 alloc_flags = update_block_group_flags(root, cache->flags);
9402 lock_chunks(root->fs_info->chunk_root);
9403 check_system_chunk(trans, root, alloc_flags);
9404 unlock_chunks(root->fs_info->chunk_root);
9405 }
9406 mutex_unlock(&root->fs_info->ro_block_group_mutex);
9407
9408 btrfs_end_transaction(trans, root);
9409 return ret;
9410 }
9411
9412 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
9413 struct btrfs_root *root, u64 type)
9414 {
9415 u64 alloc_flags = get_alloc_profile(root, type);
9416 return do_chunk_alloc(trans, root, alloc_flags,
9417 CHUNK_ALLOC_FORCE);
9418 }
9419
9420 /*
9421 * helper to account the unused space of all the readonly block group in the
9422 * space_info. takes mirrors into account.
9423 */
9424 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9425 {
9426 struct btrfs_block_group_cache *block_group;
9427 u64 free_bytes = 0;
9428 int factor;
9429
9430 /* It's df, we don't care if it's racey */
9431 if (list_empty(&sinfo->ro_bgs))
9432 return 0;
9433
9434 spin_lock(&sinfo->lock);
9435 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9436 spin_lock(&block_group->lock);
9437
9438 if (!block_group->ro) {
9439 spin_unlock(&block_group->lock);
9440 continue;
9441 }
9442
9443 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
9444 BTRFS_BLOCK_GROUP_RAID10 |
9445 BTRFS_BLOCK_GROUP_DUP))
9446 factor = 2;
9447 else
9448 factor = 1;
9449
9450 free_bytes += (block_group->key.offset -
9451 btrfs_block_group_used(&block_group->item)) *
9452 factor;
9453
9454 spin_unlock(&block_group->lock);
9455 }
9456 spin_unlock(&sinfo->lock);
9457
9458 return free_bytes;
9459 }
9460
9461 void btrfs_dec_block_group_ro(struct btrfs_root *root,
9462 struct btrfs_block_group_cache *cache)
9463 {
9464 struct btrfs_space_info *sinfo = cache->space_info;
9465 u64 num_bytes;
9466
9467 BUG_ON(!cache->ro);
9468
9469 spin_lock(&sinfo->lock);
9470 spin_lock(&cache->lock);
9471 if (!--cache->ro) {
9472 num_bytes = cache->key.offset - cache->reserved -
9473 cache->pinned - cache->bytes_super -
9474 btrfs_block_group_used(&cache->item);
9475 sinfo->bytes_readonly -= num_bytes;
9476 list_del_init(&cache->ro_list);
9477 }
9478 spin_unlock(&cache->lock);
9479 spin_unlock(&sinfo->lock);
9480 }
9481
9482 /*
9483 * checks to see if its even possible to relocate this block group.
9484 *
9485 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9486 * ok to go ahead and try.
9487 */
9488 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
9489 {
9490 struct btrfs_block_group_cache *block_group;
9491 struct btrfs_space_info *space_info;
9492 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
9493 struct btrfs_device *device;
9494 struct btrfs_trans_handle *trans;
9495 u64 min_free;
9496 u64 dev_min = 1;
9497 u64 dev_nr = 0;
9498 u64 target;
9499 int debug;
9500 int index;
9501 int full = 0;
9502 int ret = 0;
9503
9504 debug = btrfs_test_opt(root, ENOSPC_DEBUG);
9505
9506 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
9507
9508 /* odd, couldn't find the block group, leave it alone */
9509 if (!block_group) {
9510 if (debug)
9511 btrfs_warn(root->fs_info,
9512 "can't find block group for bytenr %llu",
9513 bytenr);
9514 return -1;
9515 }
9516
9517 min_free = btrfs_block_group_used(&block_group->item);
9518
9519 /* no bytes used, we're good */
9520 if (!min_free)
9521 goto out;
9522
9523 space_info = block_group->space_info;
9524 spin_lock(&space_info->lock);
9525
9526 full = space_info->full;
9527
9528 /*
9529 * if this is the last block group we have in this space, we can't
9530 * relocate it unless we're able to allocate a new chunk below.
9531 *
9532 * Otherwise, we need to make sure we have room in the space to handle
9533 * all of the extents from this block group. If we can, we're good
9534 */
9535 if ((space_info->total_bytes != block_group->key.offset) &&
9536 (space_info->bytes_used + space_info->bytes_reserved +
9537 space_info->bytes_pinned + space_info->bytes_readonly +
9538 min_free < space_info->total_bytes)) {
9539 spin_unlock(&space_info->lock);
9540 goto out;
9541 }
9542 spin_unlock(&space_info->lock);
9543
9544 /*
9545 * ok we don't have enough space, but maybe we have free space on our
9546 * devices to allocate new chunks for relocation, so loop through our
9547 * alloc devices and guess if we have enough space. if this block
9548 * group is going to be restriped, run checks against the target
9549 * profile instead of the current one.
9550 */
9551 ret = -1;
9552
9553 /*
9554 * index:
9555 * 0: raid10
9556 * 1: raid1
9557 * 2: dup
9558 * 3: raid0
9559 * 4: single
9560 */
9561 target = get_restripe_target(root->fs_info, block_group->flags);
9562 if (target) {
9563 index = __get_raid_index(extended_to_chunk(target));
9564 } else {
9565 /*
9566 * this is just a balance, so if we were marked as full
9567 * we know there is no space for a new chunk
9568 */
9569 if (full) {
9570 if (debug)
9571 btrfs_warn(root->fs_info,
9572 "no space to alloc new chunk for block group %llu",
9573 block_group->key.objectid);
9574 goto out;
9575 }
9576
9577 index = get_block_group_index(block_group);
9578 }
9579
9580 if (index == BTRFS_RAID_RAID10) {
9581 dev_min = 4;
9582 /* Divide by 2 */
9583 min_free >>= 1;
9584 } else if (index == BTRFS_RAID_RAID1) {
9585 dev_min = 2;
9586 } else if (index == BTRFS_RAID_DUP) {
9587 /* Multiply by 2 */
9588 min_free <<= 1;
9589 } else if (index == BTRFS_RAID_RAID0) {
9590 dev_min = fs_devices->rw_devices;
9591 min_free = div64_u64(min_free, dev_min);
9592 }
9593
9594 /* We need to do this so that we can look at pending chunks */
9595 trans = btrfs_join_transaction(root);
9596 if (IS_ERR(trans)) {
9597 ret = PTR_ERR(trans);
9598 goto out;
9599 }
9600
9601 mutex_lock(&root->fs_info->chunk_mutex);
9602 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9603 u64 dev_offset;
9604
9605 /*
9606 * check to make sure we can actually find a chunk with enough
9607 * space to fit our block group in.
9608 */
9609 if (device->total_bytes > device->bytes_used + min_free &&
9610 !device->is_tgtdev_for_dev_replace) {
9611 ret = find_free_dev_extent(trans, device, min_free,
9612 &dev_offset, NULL);
9613 if (!ret)
9614 dev_nr++;
9615
9616 if (dev_nr >= dev_min)
9617 break;
9618
9619 ret = -1;
9620 }
9621 }
9622 if (debug && ret == -1)
9623 btrfs_warn(root->fs_info,
9624 "no space to allocate a new chunk for block group %llu",
9625 block_group->key.objectid);
9626 mutex_unlock(&root->fs_info->chunk_mutex);
9627 btrfs_end_transaction(trans, root);
9628 out:
9629 btrfs_put_block_group(block_group);
9630 return ret;
9631 }
9632
9633 static int find_first_block_group(struct btrfs_root *root,
9634 struct btrfs_path *path, struct btrfs_key *key)
9635 {
9636 int ret = 0;
9637 struct btrfs_key found_key;
9638 struct extent_buffer *leaf;
9639 int slot;
9640
9641 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9642 if (ret < 0)
9643 goto out;
9644
9645 while (1) {
9646 slot = path->slots[0];
9647 leaf = path->nodes[0];
9648 if (slot >= btrfs_header_nritems(leaf)) {
9649 ret = btrfs_next_leaf(root, path);
9650 if (ret == 0)
9651 continue;
9652 if (ret < 0)
9653 goto out;
9654 break;
9655 }
9656 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9657
9658 if (found_key.objectid >= key->objectid &&
9659 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9660 ret = 0;
9661 goto out;
9662 }
9663 path->slots[0]++;
9664 }
9665 out:
9666 return ret;
9667 }
9668
9669 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9670 {
9671 struct btrfs_block_group_cache *block_group;
9672 u64 last = 0;
9673
9674 while (1) {
9675 struct inode *inode;
9676
9677 block_group = btrfs_lookup_first_block_group(info, last);
9678 while (block_group) {
9679 spin_lock(&block_group->lock);
9680 if (block_group->iref)
9681 break;
9682 spin_unlock(&block_group->lock);
9683 block_group = next_block_group(info->tree_root,
9684 block_group);
9685 }
9686 if (!block_group) {
9687 if (last == 0)
9688 break;
9689 last = 0;
9690 continue;
9691 }
9692
9693 inode = block_group->inode;
9694 block_group->iref = 0;
9695 block_group->inode = NULL;
9696 spin_unlock(&block_group->lock);
9697 iput(inode);
9698 last = block_group->key.objectid + block_group->key.offset;
9699 btrfs_put_block_group(block_group);
9700 }
9701 }
9702
9703 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9704 {
9705 struct btrfs_block_group_cache *block_group;
9706 struct btrfs_space_info *space_info;
9707 struct btrfs_caching_control *caching_ctl;
9708 struct rb_node *n;
9709
9710 down_write(&info->commit_root_sem);
9711 while (!list_empty(&info->caching_block_groups)) {
9712 caching_ctl = list_entry(info->caching_block_groups.next,
9713 struct btrfs_caching_control, list);
9714 list_del(&caching_ctl->list);
9715 put_caching_control(caching_ctl);
9716 }
9717 up_write(&info->commit_root_sem);
9718
9719 spin_lock(&info->unused_bgs_lock);
9720 while (!list_empty(&info->unused_bgs)) {
9721 block_group = list_first_entry(&info->unused_bgs,
9722 struct btrfs_block_group_cache,
9723 bg_list);
9724 list_del_init(&block_group->bg_list);
9725 btrfs_put_block_group(block_group);
9726 }
9727 spin_unlock(&info->unused_bgs_lock);
9728
9729 spin_lock(&info->block_group_cache_lock);
9730 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9731 block_group = rb_entry(n, struct btrfs_block_group_cache,
9732 cache_node);
9733 rb_erase(&block_group->cache_node,
9734 &info->block_group_cache_tree);
9735 RB_CLEAR_NODE(&block_group->cache_node);
9736 spin_unlock(&info->block_group_cache_lock);
9737
9738 down_write(&block_group->space_info->groups_sem);
9739 list_del(&block_group->list);
9740 up_write(&block_group->space_info->groups_sem);
9741
9742 if (block_group->cached == BTRFS_CACHE_STARTED)
9743 wait_block_group_cache_done(block_group);
9744
9745 /*
9746 * We haven't cached this block group, which means we could
9747 * possibly have excluded extents on this block group.
9748 */
9749 if (block_group->cached == BTRFS_CACHE_NO ||
9750 block_group->cached == BTRFS_CACHE_ERROR)
9751 free_excluded_extents(info->extent_root, block_group);
9752
9753 btrfs_remove_free_space_cache(block_group);
9754 btrfs_put_block_group(block_group);
9755
9756 spin_lock(&info->block_group_cache_lock);
9757 }
9758 spin_unlock(&info->block_group_cache_lock);
9759
9760 /* now that all the block groups are freed, go through and
9761 * free all the space_info structs. This is only called during
9762 * the final stages of unmount, and so we know nobody is
9763 * using them. We call synchronize_rcu() once before we start,
9764 * just to be on the safe side.
9765 */
9766 synchronize_rcu();
9767
9768 release_global_block_rsv(info);
9769
9770 while (!list_empty(&info->space_info)) {
9771 int i;
9772
9773 space_info = list_entry(info->space_info.next,
9774 struct btrfs_space_info,
9775 list);
9776 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
9777 if (WARN_ON(space_info->bytes_pinned > 0 ||
9778 space_info->bytes_reserved > 0 ||
9779 space_info->bytes_may_use > 0)) {
9780 dump_space_info(space_info, 0, 0);
9781 }
9782 }
9783 list_del(&space_info->list);
9784 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9785 struct kobject *kobj;
9786 kobj = space_info->block_group_kobjs[i];
9787 space_info->block_group_kobjs[i] = NULL;
9788 if (kobj) {
9789 kobject_del(kobj);
9790 kobject_put(kobj);
9791 }
9792 }
9793 kobject_del(&space_info->kobj);
9794 kobject_put(&space_info->kobj);
9795 }
9796 return 0;
9797 }
9798
9799 static void __link_block_group(struct btrfs_space_info *space_info,
9800 struct btrfs_block_group_cache *cache)
9801 {
9802 int index = get_block_group_index(cache);
9803 bool first = false;
9804
9805 down_write(&space_info->groups_sem);
9806 if (list_empty(&space_info->block_groups[index]))
9807 first = true;
9808 list_add_tail(&cache->list, &space_info->block_groups[index]);
9809 up_write(&space_info->groups_sem);
9810
9811 if (first) {
9812 struct raid_kobject *rkobj;
9813 int ret;
9814
9815 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9816 if (!rkobj)
9817 goto out_err;
9818 rkobj->raid_type = index;
9819 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9820 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9821 "%s", get_raid_name(index));
9822 if (ret) {
9823 kobject_put(&rkobj->kobj);
9824 goto out_err;
9825 }
9826 space_info->block_group_kobjs[index] = &rkobj->kobj;
9827 }
9828
9829 return;
9830 out_err:
9831 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
9832 }
9833
9834 static struct btrfs_block_group_cache *
9835 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
9836 {
9837 struct btrfs_block_group_cache *cache;
9838
9839 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9840 if (!cache)
9841 return NULL;
9842
9843 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9844 GFP_NOFS);
9845 if (!cache->free_space_ctl) {
9846 kfree(cache);
9847 return NULL;
9848 }
9849
9850 cache->key.objectid = start;
9851 cache->key.offset = size;
9852 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9853
9854 cache->sectorsize = root->sectorsize;
9855 cache->fs_info = root->fs_info;
9856 cache->full_stripe_len = btrfs_full_stripe_len(root,
9857 &root->fs_info->mapping_tree,
9858 start);
9859 set_free_space_tree_thresholds(cache);
9860
9861 atomic_set(&cache->count, 1);
9862 spin_lock_init(&cache->lock);
9863 init_rwsem(&cache->data_rwsem);
9864 INIT_LIST_HEAD(&cache->list);
9865 INIT_LIST_HEAD(&cache->cluster_list);
9866 INIT_LIST_HEAD(&cache->bg_list);
9867 INIT_LIST_HEAD(&cache->ro_list);
9868 INIT_LIST_HEAD(&cache->dirty_list);
9869 INIT_LIST_HEAD(&cache->io_list);
9870 btrfs_init_free_space_ctl(cache);
9871 atomic_set(&cache->trimming, 0);
9872 mutex_init(&cache->free_space_lock);
9873
9874 return cache;
9875 }
9876
9877 int btrfs_read_block_groups(struct btrfs_root *root)
9878 {
9879 struct btrfs_path *path;
9880 int ret;
9881 struct btrfs_block_group_cache *cache;
9882 struct btrfs_fs_info *info = root->fs_info;
9883 struct btrfs_space_info *space_info;
9884 struct btrfs_key key;
9885 struct btrfs_key found_key;
9886 struct extent_buffer *leaf;
9887 int need_clear = 0;
9888 u64 cache_gen;
9889
9890 root = info->extent_root;
9891 key.objectid = 0;
9892 key.offset = 0;
9893 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9894 path = btrfs_alloc_path();
9895 if (!path)
9896 return -ENOMEM;
9897 path->reada = READA_FORWARD;
9898
9899 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
9900 if (btrfs_test_opt(root, SPACE_CACHE) &&
9901 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
9902 need_clear = 1;
9903 if (btrfs_test_opt(root, CLEAR_CACHE))
9904 need_clear = 1;
9905
9906 while (1) {
9907 ret = find_first_block_group(root, path, &key);
9908 if (ret > 0)
9909 break;
9910 if (ret != 0)
9911 goto error;
9912
9913 leaf = path->nodes[0];
9914 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9915
9916 cache = btrfs_create_block_group_cache(root, found_key.objectid,
9917 found_key.offset);
9918 if (!cache) {
9919 ret = -ENOMEM;
9920 goto error;
9921 }
9922
9923 if (need_clear) {
9924 /*
9925 * When we mount with old space cache, we need to
9926 * set BTRFS_DC_CLEAR and set dirty flag.
9927 *
9928 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9929 * truncate the old free space cache inode and
9930 * setup a new one.
9931 * b) Setting 'dirty flag' makes sure that we flush
9932 * the new space cache info onto disk.
9933 */
9934 if (btrfs_test_opt(root, SPACE_CACHE))
9935 cache->disk_cache_state = BTRFS_DC_CLEAR;
9936 }
9937
9938 read_extent_buffer(leaf, &cache->item,
9939 btrfs_item_ptr_offset(leaf, path->slots[0]),
9940 sizeof(cache->item));
9941 cache->flags = btrfs_block_group_flags(&cache->item);
9942
9943 key.objectid = found_key.objectid + found_key.offset;
9944 btrfs_release_path(path);
9945
9946 /*
9947 * We need to exclude the super stripes now so that the space
9948 * info has super bytes accounted for, otherwise we'll think
9949 * we have more space than we actually do.
9950 */
9951 ret = exclude_super_stripes(root, cache);
9952 if (ret) {
9953 /*
9954 * We may have excluded something, so call this just in
9955 * case.
9956 */
9957 free_excluded_extents(root, cache);
9958 btrfs_put_block_group(cache);
9959 goto error;
9960 }
9961
9962 /*
9963 * check for two cases, either we are full, and therefore
9964 * don't need to bother with the caching work since we won't
9965 * find any space, or we are empty, and we can just add all
9966 * the space in and be done with it. This saves us _alot_ of
9967 * time, particularly in the full case.
9968 */
9969 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
9970 cache->last_byte_to_unpin = (u64)-1;
9971 cache->cached = BTRFS_CACHE_FINISHED;
9972 free_excluded_extents(root, cache);
9973 } else if (btrfs_block_group_used(&cache->item) == 0) {
9974 cache->last_byte_to_unpin = (u64)-1;
9975 cache->cached = BTRFS_CACHE_FINISHED;
9976 add_new_free_space(cache, root->fs_info,
9977 found_key.objectid,
9978 found_key.objectid +
9979 found_key.offset);
9980 free_excluded_extents(root, cache);
9981 }
9982
9983 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9984 if (ret) {
9985 btrfs_remove_free_space_cache(cache);
9986 btrfs_put_block_group(cache);
9987 goto error;
9988 }
9989
9990 ret = update_space_info(info, cache->flags, found_key.offset,
9991 btrfs_block_group_used(&cache->item),
9992 &space_info);
9993 if (ret) {
9994 btrfs_remove_free_space_cache(cache);
9995 spin_lock(&info->block_group_cache_lock);
9996 rb_erase(&cache->cache_node,
9997 &info->block_group_cache_tree);
9998 RB_CLEAR_NODE(&cache->cache_node);
9999 spin_unlock(&info->block_group_cache_lock);
10000 btrfs_put_block_group(cache);
10001 goto error;
10002 }
10003
10004 cache->space_info = space_info;
10005 spin_lock(&cache->space_info->lock);
10006 cache->space_info->bytes_readonly += cache->bytes_super;
10007 spin_unlock(&cache->space_info->lock);
10008
10009 __link_block_group(space_info, cache);
10010
10011 set_avail_alloc_bits(root->fs_info, cache->flags);
10012 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
10013 inc_block_group_ro(cache, 1);
10014 } else if (btrfs_block_group_used(&cache->item) == 0) {
10015 spin_lock(&info->unused_bgs_lock);
10016 /* Should always be true but just in case. */
10017 if (list_empty(&cache->bg_list)) {
10018 btrfs_get_block_group(cache);
10019 list_add_tail(&cache->bg_list,
10020 &info->unused_bgs);
10021 }
10022 spin_unlock(&info->unused_bgs_lock);
10023 }
10024 }
10025
10026 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
10027 if (!(get_alloc_profile(root, space_info->flags) &
10028 (BTRFS_BLOCK_GROUP_RAID10 |
10029 BTRFS_BLOCK_GROUP_RAID1 |
10030 BTRFS_BLOCK_GROUP_RAID5 |
10031 BTRFS_BLOCK_GROUP_RAID6 |
10032 BTRFS_BLOCK_GROUP_DUP)))
10033 continue;
10034 /*
10035 * avoid allocating from un-mirrored block group if there are
10036 * mirrored block groups.
10037 */
10038 list_for_each_entry(cache,
10039 &space_info->block_groups[BTRFS_RAID_RAID0],
10040 list)
10041 inc_block_group_ro(cache, 1);
10042 list_for_each_entry(cache,
10043 &space_info->block_groups[BTRFS_RAID_SINGLE],
10044 list)
10045 inc_block_group_ro(cache, 1);
10046 }
10047
10048 init_global_block_rsv(info);
10049 ret = 0;
10050 error:
10051 btrfs_free_path(path);
10052 return ret;
10053 }
10054
10055 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
10056 struct btrfs_root *root)
10057 {
10058 struct btrfs_block_group_cache *block_group, *tmp;
10059 struct btrfs_root *extent_root = root->fs_info->extent_root;
10060 struct btrfs_block_group_item item;
10061 struct btrfs_key key;
10062 int ret = 0;
10063 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
10064
10065 trans->can_flush_pending_bgs = false;
10066 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
10067 if (ret)
10068 goto next;
10069
10070 spin_lock(&block_group->lock);
10071 memcpy(&item, &block_group->item, sizeof(item));
10072 memcpy(&key, &block_group->key, sizeof(key));
10073 spin_unlock(&block_group->lock);
10074
10075 ret = btrfs_insert_item(trans, extent_root, &key, &item,
10076 sizeof(item));
10077 if (ret)
10078 btrfs_abort_transaction(trans, extent_root, ret);
10079 ret = btrfs_finish_chunk_alloc(trans, extent_root,
10080 key.objectid, key.offset);
10081 if (ret)
10082 btrfs_abort_transaction(trans, extent_root, ret);
10083 add_block_group_free_space(trans, root->fs_info, block_group);
10084 /* already aborted the transaction if it failed. */
10085 next:
10086 list_del_init(&block_group->bg_list);
10087 }
10088 trans->can_flush_pending_bgs = can_flush_pending_bgs;
10089 }
10090
10091 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
10092 struct btrfs_root *root, u64 bytes_used,
10093 u64 type, u64 chunk_objectid, u64 chunk_offset,
10094 u64 size)
10095 {
10096 int ret;
10097 struct btrfs_root *extent_root;
10098 struct btrfs_block_group_cache *cache;
10099
10100 extent_root = root->fs_info->extent_root;
10101
10102 btrfs_set_log_full_commit(root->fs_info, trans);
10103
10104 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
10105 if (!cache)
10106 return -ENOMEM;
10107
10108 btrfs_set_block_group_used(&cache->item, bytes_used);
10109 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
10110 btrfs_set_block_group_flags(&cache->item, type);
10111
10112 cache->flags = type;
10113 cache->last_byte_to_unpin = (u64)-1;
10114 cache->cached = BTRFS_CACHE_FINISHED;
10115 cache->needs_free_space = 1;
10116 ret = exclude_super_stripes(root, cache);
10117 if (ret) {
10118 /*
10119 * We may have excluded something, so call this just in
10120 * case.
10121 */
10122 free_excluded_extents(root, cache);
10123 btrfs_put_block_group(cache);
10124 return ret;
10125 }
10126
10127 add_new_free_space(cache, root->fs_info, chunk_offset,
10128 chunk_offset + size);
10129
10130 free_excluded_extents(root, cache);
10131
10132 #ifdef CONFIG_BTRFS_DEBUG
10133 if (btrfs_should_fragment_free_space(root, cache)) {
10134 u64 new_bytes_used = size - bytes_used;
10135
10136 bytes_used += new_bytes_used >> 1;
10137 fragment_free_space(root, cache);
10138 }
10139 #endif
10140 /*
10141 * Call to ensure the corresponding space_info object is created and
10142 * assigned to our block group, but don't update its counters just yet.
10143 * We want our bg to be added to the rbtree with its ->space_info set.
10144 */
10145 ret = update_space_info(root->fs_info, cache->flags, 0, 0,
10146 &cache->space_info);
10147 if (ret) {
10148 btrfs_remove_free_space_cache(cache);
10149 btrfs_put_block_group(cache);
10150 return ret;
10151 }
10152
10153 ret = btrfs_add_block_group_cache(root->fs_info, cache);
10154 if (ret) {
10155 btrfs_remove_free_space_cache(cache);
10156 btrfs_put_block_group(cache);
10157 return ret;
10158 }
10159
10160 /*
10161 * Now that our block group has its ->space_info set and is inserted in
10162 * the rbtree, update the space info's counters.
10163 */
10164 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
10165 &cache->space_info);
10166 if (ret) {
10167 btrfs_remove_free_space_cache(cache);
10168 spin_lock(&root->fs_info->block_group_cache_lock);
10169 rb_erase(&cache->cache_node,
10170 &root->fs_info->block_group_cache_tree);
10171 RB_CLEAR_NODE(&cache->cache_node);
10172 spin_unlock(&root->fs_info->block_group_cache_lock);
10173 btrfs_put_block_group(cache);
10174 return ret;
10175 }
10176 update_global_block_rsv(root->fs_info);
10177
10178 spin_lock(&cache->space_info->lock);
10179 cache->space_info->bytes_readonly += cache->bytes_super;
10180 spin_unlock(&cache->space_info->lock);
10181
10182 __link_block_group(cache->space_info, cache);
10183
10184 list_add_tail(&cache->bg_list, &trans->new_bgs);
10185
10186 set_avail_alloc_bits(extent_root->fs_info, type);
10187
10188 return 0;
10189 }
10190
10191 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10192 {
10193 u64 extra_flags = chunk_to_extended(flags) &
10194 BTRFS_EXTENDED_PROFILE_MASK;
10195
10196 write_seqlock(&fs_info->profiles_lock);
10197 if (flags & BTRFS_BLOCK_GROUP_DATA)
10198 fs_info->avail_data_alloc_bits &= ~extra_flags;
10199 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10200 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10201 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10202 fs_info->avail_system_alloc_bits &= ~extra_flags;
10203 write_sequnlock(&fs_info->profiles_lock);
10204 }
10205
10206 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10207 struct btrfs_root *root, u64 group_start,
10208 struct extent_map *em)
10209 {
10210 struct btrfs_path *path;
10211 struct btrfs_block_group_cache *block_group;
10212 struct btrfs_free_cluster *cluster;
10213 struct btrfs_root *tree_root = root->fs_info->tree_root;
10214 struct btrfs_key key;
10215 struct inode *inode;
10216 struct kobject *kobj = NULL;
10217 int ret;
10218 int index;
10219 int factor;
10220 struct btrfs_caching_control *caching_ctl = NULL;
10221 bool remove_em;
10222
10223 root = root->fs_info->extent_root;
10224
10225 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
10226 BUG_ON(!block_group);
10227 BUG_ON(!block_group->ro);
10228
10229 /*
10230 * Free the reserved super bytes from this block group before
10231 * remove it.
10232 */
10233 free_excluded_extents(root, block_group);
10234
10235 memcpy(&key, &block_group->key, sizeof(key));
10236 index = get_block_group_index(block_group);
10237 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
10238 BTRFS_BLOCK_GROUP_RAID1 |
10239 BTRFS_BLOCK_GROUP_RAID10))
10240 factor = 2;
10241 else
10242 factor = 1;
10243
10244 /* make sure this block group isn't part of an allocation cluster */
10245 cluster = &root->fs_info->data_alloc_cluster;
10246 spin_lock(&cluster->refill_lock);
10247 btrfs_return_cluster_to_free_space(block_group, cluster);
10248 spin_unlock(&cluster->refill_lock);
10249
10250 /*
10251 * make sure this block group isn't part of a metadata
10252 * allocation cluster
10253 */
10254 cluster = &root->fs_info->meta_alloc_cluster;
10255 spin_lock(&cluster->refill_lock);
10256 btrfs_return_cluster_to_free_space(block_group, cluster);
10257 spin_unlock(&cluster->refill_lock);
10258
10259 path = btrfs_alloc_path();
10260 if (!path) {
10261 ret = -ENOMEM;
10262 goto out;
10263 }
10264
10265 /*
10266 * get the inode first so any iput calls done for the io_list
10267 * aren't the final iput (no unlinks allowed now)
10268 */
10269 inode = lookup_free_space_inode(tree_root, block_group, path);
10270
10271 mutex_lock(&trans->transaction->cache_write_mutex);
10272 /*
10273 * make sure our free spache cache IO is done before remove the
10274 * free space inode
10275 */
10276 spin_lock(&trans->transaction->dirty_bgs_lock);
10277 if (!list_empty(&block_group->io_list)) {
10278 list_del_init(&block_group->io_list);
10279
10280 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10281
10282 spin_unlock(&trans->transaction->dirty_bgs_lock);
10283 btrfs_wait_cache_io(root, trans, block_group,
10284 &block_group->io_ctl, path,
10285 block_group->key.objectid);
10286 btrfs_put_block_group(block_group);
10287 spin_lock(&trans->transaction->dirty_bgs_lock);
10288 }
10289
10290 if (!list_empty(&block_group->dirty_list)) {
10291 list_del_init(&block_group->dirty_list);
10292 btrfs_put_block_group(block_group);
10293 }
10294 spin_unlock(&trans->transaction->dirty_bgs_lock);
10295 mutex_unlock(&trans->transaction->cache_write_mutex);
10296
10297 if (!IS_ERR(inode)) {
10298 ret = btrfs_orphan_add(trans, inode);
10299 if (ret) {
10300 btrfs_add_delayed_iput(inode);
10301 goto out;
10302 }
10303 clear_nlink(inode);
10304 /* One for the block groups ref */
10305 spin_lock(&block_group->lock);
10306 if (block_group->iref) {
10307 block_group->iref = 0;
10308 block_group->inode = NULL;
10309 spin_unlock(&block_group->lock);
10310 iput(inode);
10311 } else {
10312 spin_unlock(&block_group->lock);
10313 }
10314 /* One for our lookup ref */
10315 btrfs_add_delayed_iput(inode);
10316 }
10317
10318 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10319 key.offset = block_group->key.objectid;
10320 key.type = 0;
10321
10322 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10323 if (ret < 0)
10324 goto out;
10325 if (ret > 0)
10326 btrfs_release_path(path);
10327 if (ret == 0) {
10328 ret = btrfs_del_item(trans, tree_root, path);
10329 if (ret)
10330 goto out;
10331 btrfs_release_path(path);
10332 }
10333
10334 spin_lock(&root->fs_info->block_group_cache_lock);
10335 rb_erase(&block_group->cache_node,
10336 &root->fs_info->block_group_cache_tree);
10337 RB_CLEAR_NODE(&block_group->cache_node);
10338
10339 if (root->fs_info->first_logical_byte == block_group->key.objectid)
10340 root->fs_info->first_logical_byte = (u64)-1;
10341 spin_unlock(&root->fs_info->block_group_cache_lock);
10342
10343 down_write(&block_group->space_info->groups_sem);
10344 /*
10345 * we must use list_del_init so people can check to see if they
10346 * are still on the list after taking the semaphore
10347 */
10348 list_del_init(&block_group->list);
10349 if (list_empty(&block_group->space_info->block_groups[index])) {
10350 kobj = block_group->space_info->block_group_kobjs[index];
10351 block_group->space_info->block_group_kobjs[index] = NULL;
10352 clear_avail_alloc_bits(root->fs_info, block_group->flags);
10353 }
10354 up_write(&block_group->space_info->groups_sem);
10355 if (kobj) {
10356 kobject_del(kobj);
10357 kobject_put(kobj);
10358 }
10359
10360 if (block_group->has_caching_ctl)
10361 caching_ctl = get_caching_control(block_group);
10362 if (block_group->cached == BTRFS_CACHE_STARTED)
10363 wait_block_group_cache_done(block_group);
10364 if (block_group->has_caching_ctl) {
10365 down_write(&root->fs_info->commit_root_sem);
10366 if (!caching_ctl) {
10367 struct btrfs_caching_control *ctl;
10368
10369 list_for_each_entry(ctl,
10370 &root->fs_info->caching_block_groups, list)
10371 if (ctl->block_group == block_group) {
10372 caching_ctl = ctl;
10373 atomic_inc(&caching_ctl->count);
10374 break;
10375 }
10376 }
10377 if (caching_ctl)
10378 list_del_init(&caching_ctl->list);
10379 up_write(&root->fs_info->commit_root_sem);
10380 if (caching_ctl) {
10381 /* Once for the caching bgs list and once for us. */
10382 put_caching_control(caching_ctl);
10383 put_caching_control(caching_ctl);
10384 }
10385 }
10386
10387 spin_lock(&trans->transaction->dirty_bgs_lock);
10388 if (!list_empty(&block_group->dirty_list)) {
10389 WARN_ON(1);
10390 }
10391 if (!list_empty(&block_group->io_list)) {
10392 WARN_ON(1);
10393 }
10394 spin_unlock(&trans->transaction->dirty_bgs_lock);
10395 btrfs_remove_free_space_cache(block_group);
10396
10397 spin_lock(&block_group->space_info->lock);
10398 list_del_init(&block_group->ro_list);
10399
10400 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
10401 WARN_ON(block_group->space_info->total_bytes
10402 < block_group->key.offset);
10403 WARN_ON(block_group->space_info->bytes_readonly
10404 < block_group->key.offset);
10405 WARN_ON(block_group->space_info->disk_total
10406 < block_group->key.offset * factor);
10407 }
10408 block_group->space_info->total_bytes -= block_group->key.offset;
10409 block_group->space_info->bytes_readonly -= block_group->key.offset;
10410 block_group->space_info->disk_total -= block_group->key.offset * factor;
10411
10412 spin_unlock(&block_group->space_info->lock);
10413
10414 memcpy(&key, &block_group->key, sizeof(key));
10415
10416 lock_chunks(root);
10417 if (!list_empty(&em->list)) {
10418 /* We're in the transaction->pending_chunks list. */
10419 free_extent_map(em);
10420 }
10421 spin_lock(&block_group->lock);
10422 block_group->removed = 1;
10423 /*
10424 * At this point trimming can't start on this block group, because we
10425 * removed the block group from the tree fs_info->block_group_cache_tree
10426 * so no one can't find it anymore and even if someone already got this
10427 * block group before we removed it from the rbtree, they have already
10428 * incremented block_group->trimming - if they didn't, they won't find
10429 * any free space entries because we already removed them all when we
10430 * called btrfs_remove_free_space_cache().
10431 *
10432 * And we must not remove the extent map from the fs_info->mapping_tree
10433 * to prevent the same logical address range and physical device space
10434 * ranges from being reused for a new block group. This is because our
10435 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10436 * completely transactionless, so while it is trimming a range the
10437 * currently running transaction might finish and a new one start,
10438 * allowing for new block groups to be created that can reuse the same
10439 * physical device locations unless we take this special care.
10440 *
10441 * There may also be an implicit trim operation if the file system
10442 * is mounted with -odiscard. The same protections must remain
10443 * in place until the extents have been discarded completely when
10444 * the transaction commit has completed.
10445 */
10446 remove_em = (atomic_read(&block_group->trimming) == 0);
10447 /*
10448 * Make sure a trimmer task always sees the em in the pinned_chunks list
10449 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10450 * before checking block_group->removed).
10451 */
10452 if (!remove_em) {
10453 /*
10454 * Our em might be in trans->transaction->pending_chunks which
10455 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10456 * and so is the fs_info->pinned_chunks list.
10457 *
10458 * So at this point we must be holding the chunk_mutex to avoid
10459 * any races with chunk allocation (more specifically at
10460 * volumes.c:contains_pending_extent()), to ensure it always
10461 * sees the em, either in the pending_chunks list or in the
10462 * pinned_chunks list.
10463 */
10464 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
10465 }
10466 spin_unlock(&block_group->lock);
10467
10468 if (remove_em) {
10469 struct extent_map_tree *em_tree;
10470
10471 em_tree = &root->fs_info->mapping_tree.map_tree;
10472 write_lock(&em_tree->lock);
10473 /*
10474 * The em might be in the pending_chunks list, so make sure the
10475 * chunk mutex is locked, since remove_extent_mapping() will
10476 * delete us from that list.
10477 */
10478 remove_extent_mapping(em_tree, em);
10479 write_unlock(&em_tree->lock);
10480 /* once for the tree */
10481 free_extent_map(em);
10482 }
10483
10484 unlock_chunks(root);
10485
10486 ret = remove_block_group_free_space(trans, root->fs_info, block_group);
10487 if (ret)
10488 goto out;
10489
10490 btrfs_put_block_group(block_group);
10491 btrfs_put_block_group(block_group);
10492
10493 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10494 if (ret > 0)
10495 ret = -EIO;
10496 if (ret < 0)
10497 goto out;
10498
10499 ret = btrfs_del_item(trans, root, path);
10500 out:
10501 btrfs_free_path(path);
10502 return ret;
10503 }
10504
10505 struct btrfs_trans_handle *
10506 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10507 const u64 chunk_offset)
10508 {
10509 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10510 struct extent_map *em;
10511 struct map_lookup *map;
10512 unsigned int num_items;
10513
10514 read_lock(&em_tree->lock);
10515 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10516 read_unlock(&em_tree->lock);
10517 ASSERT(em && em->start == chunk_offset);
10518
10519 /*
10520 * We need to reserve 3 + N units from the metadata space info in order
10521 * to remove a block group (done at btrfs_remove_chunk() and at
10522 * btrfs_remove_block_group()), which are used for:
10523 *
10524 * 1 unit for adding the free space inode's orphan (located in the tree
10525 * of tree roots).
10526 * 1 unit for deleting the block group item (located in the extent
10527 * tree).
10528 * 1 unit for deleting the free space item (located in tree of tree
10529 * roots).
10530 * N units for deleting N device extent items corresponding to each
10531 * stripe (located in the device tree).
10532 *
10533 * In order to remove a block group we also need to reserve units in the
10534 * system space info in order to update the chunk tree (update one or
10535 * more device items and remove one chunk item), but this is done at
10536 * btrfs_remove_chunk() through a call to check_system_chunk().
10537 */
10538 map = em->map_lookup;
10539 num_items = 3 + map->num_stripes;
10540 free_extent_map(em);
10541
10542 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10543 num_items, 1);
10544 }
10545
10546 /*
10547 * Process the unused_bgs list and remove any that don't have any allocated
10548 * space inside of them.
10549 */
10550 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10551 {
10552 struct btrfs_block_group_cache *block_group;
10553 struct btrfs_space_info *space_info;
10554 struct btrfs_root *root = fs_info->extent_root;
10555 struct btrfs_trans_handle *trans;
10556 int ret = 0;
10557
10558 if (!fs_info->open)
10559 return;
10560
10561 spin_lock(&fs_info->unused_bgs_lock);
10562 while (!list_empty(&fs_info->unused_bgs)) {
10563 u64 start, end;
10564 int trimming;
10565
10566 block_group = list_first_entry(&fs_info->unused_bgs,
10567 struct btrfs_block_group_cache,
10568 bg_list);
10569 list_del_init(&block_group->bg_list);
10570
10571 space_info = block_group->space_info;
10572
10573 if (ret || btrfs_mixed_space_info(space_info)) {
10574 btrfs_put_block_group(block_group);
10575 continue;
10576 }
10577 spin_unlock(&fs_info->unused_bgs_lock);
10578
10579 mutex_lock(&fs_info->delete_unused_bgs_mutex);
10580
10581 /* Don't want to race with allocators so take the groups_sem */
10582 down_write(&space_info->groups_sem);
10583 spin_lock(&block_group->lock);
10584 if (block_group->reserved ||
10585 btrfs_block_group_used(&block_group->item) ||
10586 block_group->ro ||
10587 list_is_singular(&block_group->list)) {
10588 /*
10589 * We want to bail if we made new allocations or have
10590 * outstanding allocations in this block group. We do
10591 * the ro check in case balance is currently acting on
10592 * this block group.
10593 */
10594 spin_unlock(&block_group->lock);
10595 up_write(&space_info->groups_sem);
10596 goto next;
10597 }
10598 spin_unlock(&block_group->lock);
10599
10600 /* We don't want to force the issue, only flip if it's ok. */
10601 ret = inc_block_group_ro(block_group, 0);
10602 up_write(&space_info->groups_sem);
10603 if (ret < 0) {
10604 ret = 0;
10605 goto next;
10606 }
10607
10608 /*
10609 * Want to do this before we do anything else so we can recover
10610 * properly if we fail to join the transaction.
10611 */
10612 trans = btrfs_start_trans_remove_block_group(fs_info,
10613 block_group->key.objectid);
10614 if (IS_ERR(trans)) {
10615 btrfs_dec_block_group_ro(root, block_group);
10616 ret = PTR_ERR(trans);
10617 goto next;
10618 }
10619
10620 /*
10621 * We could have pending pinned extents for this block group,
10622 * just delete them, we don't care about them anymore.
10623 */
10624 start = block_group->key.objectid;
10625 end = start + block_group->key.offset - 1;
10626 /*
10627 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10628 * btrfs_finish_extent_commit(). If we are at transaction N,
10629 * another task might be running finish_extent_commit() for the
10630 * previous transaction N - 1, and have seen a range belonging
10631 * to the block group in freed_extents[] before we were able to
10632 * clear the whole block group range from freed_extents[]. This
10633 * means that task can lookup for the block group after we
10634 * unpinned it from freed_extents[] and removed it, leading to
10635 * a BUG_ON() at btrfs_unpin_extent_range().
10636 */
10637 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10638 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10639 EXTENT_DIRTY, GFP_NOFS);
10640 if (ret) {
10641 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10642 btrfs_dec_block_group_ro(root, block_group);
10643 goto end_trans;
10644 }
10645 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10646 EXTENT_DIRTY, GFP_NOFS);
10647 if (ret) {
10648 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10649 btrfs_dec_block_group_ro(root, block_group);
10650 goto end_trans;
10651 }
10652 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10653
10654 /* Reset pinned so btrfs_put_block_group doesn't complain */
10655 spin_lock(&space_info->lock);
10656 spin_lock(&block_group->lock);
10657
10658 space_info->bytes_pinned -= block_group->pinned;
10659 space_info->bytes_readonly += block_group->pinned;
10660 percpu_counter_add(&space_info->total_bytes_pinned,
10661 -block_group->pinned);
10662 block_group->pinned = 0;
10663
10664 spin_unlock(&block_group->lock);
10665 spin_unlock(&space_info->lock);
10666
10667 /* DISCARD can flip during remount */
10668 trimming = btrfs_test_opt(root, DISCARD);
10669
10670 /* Implicit trim during transaction commit. */
10671 if (trimming)
10672 btrfs_get_block_group_trimming(block_group);
10673
10674 /*
10675 * Btrfs_remove_chunk will abort the transaction if things go
10676 * horribly wrong.
10677 */
10678 ret = btrfs_remove_chunk(trans, root,
10679 block_group->key.objectid);
10680
10681 if (ret) {
10682 if (trimming)
10683 btrfs_put_block_group_trimming(block_group);
10684 goto end_trans;
10685 }
10686
10687 /*
10688 * If we're not mounted with -odiscard, we can just forget
10689 * about this block group. Otherwise we'll need to wait
10690 * until transaction commit to do the actual discard.
10691 */
10692 if (trimming) {
10693 spin_lock(&fs_info->unused_bgs_lock);
10694 /*
10695 * A concurrent scrub might have added us to the list
10696 * fs_info->unused_bgs, so use a list_move operation
10697 * to add the block group to the deleted_bgs list.
10698 */
10699 list_move(&block_group->bg_list,
10700 &trans->transaction->deleted_bgs);
10701 spin_unlock(&fs_info->unused_bgs_lock);
10702 btrfs_get_block_group(block_group);
10703 }
10704 end_trans:
10705 btrfs_end_transaction(trans, root);
10706 next:
10707 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
10708 btrfs_put_block_group(block_group);
10709 spin_lock(&fs_info->unused_bgs_lock);
10710 }
10711 spin_unlock(&fs_info->unused_bgs_lock);
10712 }
10713
10714 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10715 {
10716 struct btrfs_space_info *space_info;
10717 struct btrfs_super_block *disk_super;
10718 u64 features;
10719 u64 flags;
10720 int mixed = 0;
10721 int ret;
10722
10723 disk_super = fs_info->super_copy;
10724 if (!btrfs_super_root(disk_super))
10725 return -EINVAL;
10726
10727 features = btrfs_super_incompat_flags(disk_super);
10728 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10729 mixed = 1;
10730
10731 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10732 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10733 if (ret)
10734 goto out;
10735
10736 if (mixed) {
10737 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10738 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10739 } else {
10740 flags = BTRFS_BLOCK_GROUP_METADATA;
10741 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10742 if (ret)
10743 goto out;
10744
10745 flags = BTRFS_BLOCK_GROUP_DATA;
10746 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10747 }
10748 out:
10749 return ret;
10750 }
10751
10752 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
10753 {
10754 return unpin_extent_range(root, start, end, false);
10755 }
10756
10757 /*
10758 * It used to be that old block groups would be left around forever.
10759 * Iterating over them would be enough to trim unused space. Since we
10760 * now automatically remove them, we also need to iterate over unallocated
10761 * space.
10762 *
10763 * We don't want a transaction for this since the discard may take a
10764 * substantial amount of time. We don't require that a transaction be
10765 * running, but we do need to take a running transaction into account
10766 * to ensure that we're not discarding chunks that were released in
10767 * the current transaction.
10768 *
10769 * Holding the chunks lock will prevent other threads from allocating
10770 * or releasing chunks, but it won't prevent a running transaction
10771 * from committing and releasing the memory that the pending chunks
10772 * list head uses. For that, we need to take a reference to the
10773 * transaction.
10774 */
10775 static int btrfs_trim_free_extents(struct btrfs_device *device,
10776 u64 minlen, u64 *trimmed)
10777 {
10778 u64 start = 0, len = 0;
10779 int ret;
10780
10781 *trimmed = 0;
10782
10783 /* Not writeable = nothing to do. */
10784 if (!device->writeable)
10785 return 0;
10786
10787 /* No free space = nothing to do. */
10788 if (device->total_bytes <= device->bytes_used)
10789 return 0;
10790
10791 ret = 0;
10792
10793 while (1) {
10794 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
10795 struct btrfs_transaction *trans;
10796 u64 bytes;
10797
10798 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
10799 if (ret)
10800 return ret;
10801
10802 down_read(&fs_info->commit_root_sem);
10803
10804 spin_lock(&fs_info->trans_lock);
10805 trans = fs_info->running_transaction;
10806 if (trans)
10807 atomic_inc(&trans->use_count);
10808 spin_unlock(&fs_info->trans_lock);
10809
10810 ret = find_free_dev_extent_start(trans, device, minlen, start,
10811 &start, &len);
10812 if (trans)
10813 btrfs_put_transaction(trans);
10814
10815 if (ret) {
10816 up_read(&fs_info->commit_root_sem);
10817 mutex_unlock(&fs_info->chunk_mutex);
10818 if (ret == -ENOSPC)
10819 ret = 0;
10820 break;
10821 }
10822
10823 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
10824 up_read(&fs_info->commit_root_sem);
10825 mutex_unlock(&fs_info->chunk_mutex);
10826
10827 if (ret)
10828 break;
10829
10830 start += len;
10831 *trimmed += bytes;
10832
10833 if (fatal_signal_pending(current)) {
10834 ret = -ERESTARTSYS;
10835 break;
10836 }
10837
10838 cond_resched();
10839 }
10840
10841 return ret;
10842 }
10843
10844 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
10845 {
10846 struct btrfs_fs_info *fs_info = root->fs_info;
10847 struct btrfs_block_group_cache *cache = NULL;
10848 struct btrfs_device *device;
10849 struct list_head *devices;
10850 u64 group_trimmed;
10851 u64 start;
10852 u64 end;
10853 u64 trimmed = 0;
10854 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10855 int ret = 0;
10856
10857 /*
10858 * try to trim all FS space, our block group may start from non-zero.
10859 */
10860 if (range->len == total_bytes)
10861 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10862 else
10863 cache = btrfs_lookup_block_group(fs_info, range->start);
10864
10865 while (cache) {
10866 if (cache->key.objectid >= (range->start + range->len)) {
10867 btrfs_put_block_group(cache);
10868 break;
10869 }
10870
10871 start = max(range->start, cache->key.objectid);
10872 end = min(range->start + range->len,
10873 cache->key.objectid + cache->key.offset);
10874
10875 if (end - start >= range->minlen) {
10876 if (!block_group_cache_done(cache)) {
10877 ret = cache_block_group(cache, 0);
10878 if (ret) {
10879 btrfs_put_block_group(cache);
10880 break;
10881 }
10882 ret = wait_block_group_cache_done(cache);
10883 if (ret) {
10884 btrfs_put_block_group(cache);
10885 break;
10886 }
10887 }
10888 ret = btrfs_trim_block_group(cache,
10889 &group_trimmed,
10890 start,
10891 end,
10892 range->minlen);
10893
10894 trimmed += group_trimmed;
10895 if (ret) {
10896 btrfs_put_block_group(cache);
10897 break;
10898 }
10899 }
10900
10901 cache = next_block_group(fs_info->tree_root, cache);
10902 }
10903
10904 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
10905 devices = &root->fs_info->fs_devices->alloc_list;
10906 list_for_each_entry(device, devices, dev_alloc_list) {
10907 ret = btrfs_trim_free_extents(device, range->minlen,
10908 &group_trimmed);
10909 if (ret)
10910 break;
10911
10912 trimmed += group_trimmed;
10913 }
10914 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
10915
10916 range->len = trimmed;
10917 return ret;
10918 }
10919
10920 /*
10921 * btrfs_{start,end}_write_no_snapshoting() are similar to
10922 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10923 * data into the page cache through nocow before the subvolume is snapshoted,
10924 * but flush the data into disk after the snapshot creation, or to prevent
10925 * operations while snapshoting is ongoing and that cause the snapshot to be
10926 * inconsistent (writes followed by expanding truncates for example).
10927 */
10928 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
10929 {
10930 percpu_counter_dec(&root->subv_writers->counter);
10931 /*
10932 * Make sure counter is updated before we wake up waiters.
10933 */
10934 smp_mb();
10935 if (waitqueue_active(&root->subv_writers->wait))
10936 wake_up(&root->subv_writers->wait);
10937 }
10938
10939 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
10940 {
10941 if (atomic_read(&root->will_be_snapshoted))
10942 return 0;
10943
10944 percpu_counter_inc(&root->subv_writers->counter);
10945 /*
10946 * Make sure counter is updated before we check for snapshot creation.
10947 */
10948 smp_mb();
10949 if (atomic_read(&root->will_be_snapshoted)) {
10950 btrfs_end_write_no_snapshoting(root);
10951 return 0;
10952 }
10953 return 1;
10954 }
10955
10956 static int wait_snapshoting_atomic_t(atomic_t *a)
10957 {
10958 schedule();
10959 return 0;
10960 }
10961
10962 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
10963 {
10964 while (true) {
10965 int ret;
10966
10967 ret = btrfs_start_write_no_snapshoting(root);
10968 if (ret)
10969 break;
10970 wait_on_atomic_t(&root->will_be_snapshoted,
10971 wait_snapshoting_atomic_t,
10972 TASK_UNINTERRUPTIBLE);
10973 }
10974 }
Linux kernel - Deliverables
This page took 0.363307 seconds and 5 git commands to generate.