?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
net: qlge: remove superfluous statement
[linux-2.6/btrfs-unstable.git] / fs / btrfs / extent-tree.c
blobb480fd55577480d11b7bdf3440eac2cb954af785
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);
235 set_extent_bits(&root->fs_info->freed_extents[1],
236 start, end, EXTENT_UPTODATE);
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);
250 clear_extent_bits(&root->fs_info->freed_extents[1],
251 start, end, EXTENT_UPTODATE);
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 COWed 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 /*
2046 * Avoid races with device replace and make sure our bbio has devices
2047 * associated to its stripes that don't go away while we are discarding.
2048 */
2049 btrfs_bio_counter_inc_blocked(root->fs_info);
2050 /* Tell the block device(s) that the sectors can be discarded */
2051 ret = btrfs_map_block(root->fs_info, REQ_OP_DISCARD,
2052 bytenr, &num_bytes, &bbio, 0);
2053 /* Error condition is -ENOMEM */
2054 if (!ret) {
2055 struct btrfs_bio_stripe *stripe = bbio->stripes;
2056 int i;
2057
2058
2059 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
2060 u64 bytes;
2061 if (!stripe->dev->can_discard)
2062 continue;
2063
2064 ret = btrfs_issue_discard(stripe->dev->bdev,
2065 stripe->physical,
2066 stripe->length,
2067 &bytes);
2068 if (!ret)
2069 discarded_bytes += bytes;
2070 else if (ret != -EOPNOTSUPP)
2071 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2072
2073 /*
2074 * Just in case we get back EOPNOTSUPP for some reason,
2075 * just ignore the return value so we don't screw up
2076 * people calling discard_extent.
2077 */
2078 ret = 0;
2079 }
2080 btrfs_put_bbio(bbio);
2081 }
2082 btrfs_bio_counter_dec(root->fs_info);
2083
2084 if (actual_bytes)
2085 *actual_bytes = discarded_bytes;
2086
2087
2088 if (ret == -EOPNOTSUPP)
2089 ret = 0;
2090 return ret;
2091 }
2092
2093 /* Can return -ENOMEM */
2094 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2095 struct btrfs_root *root,
2096 u64 bytenr, u64 num_bytes, u64 parent,
2097 u64 root_objectid, u64 owner, u64 offset)
2098 {
2099 int ret;
2100 struct btrfs_fs_info *fs_info = root->fs_info;
2101
2102 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2103 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2104
2105 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2106 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
2107 num_bytes,
2108 parent, root_objectid, (int)owner,
2109 BTRFS_ADD_DELAYED_REF, NULL);
2110 } else {
2111 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
2112 num_bytes, parent, root_objectid,
2113 owner, offset, 0,
2114 BTRFS_ADD_DELAYED_REF, NULL);
2115 }
2116 return ret;
2117 }
2118
2119 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2120 struct btrfs_root *root,
2121 struct btrfs_delayed_ref_node *node,
2122 u64 parent, u64 root_objectid,
2123 u64 owner, u64 offset, int refs_to_add,
2124 struct btrfs_delayed_extent_op *extent_op)
2125 {
2126 struct btrfs_fs_info *fs_info = root->fs_info;
2127 struct btrfs_path *path;
2128 struct extent_buffer *leaf;
2129 struct btrfs_extent_item *item;
2130 struct btrfs_key key;
2131 u64 bytenr = node->bytenr;
2132 u64 num_bytes = node->num_bytes;
2133 u64 refs;
2134 int ret;
2135
2136 path = btrfs_alloc_path();
2137 if (!path)
2138 return -ENOMEM;
2139
2140 path->reada = READA_FORWARD;
2141 path->leave_spinning = 1;
2142 /* this will setup the path even if it fails to insert the back ref */
2143 ret = insert_inline_extent_backref(trans, fs_info->extent_root, path,
2144 bytenr, num_bytes, parent,
2145 root_objectid, owner, offset,
2146 refs_to_add, extent_op);
2147 if ((ret < 0 && ret != -EAGAIN) || !ret)
2148 goto out;
2149
2150 /*
2151 * Ok we had -EAGAIN which means we didn't have space to insert and
2152 * inline extent ref, so just update the reference count and add a
2153 * normal backref.
2154 */
2155 leaf = path->nodes[0];
2156 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2157 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2158 refs = btrfs_extent_refs(leaf, item);
2159 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2160 if (extent_op)
2161 __run_delayed_extent_op(extent_op, leaf, item);
2162
2163 btrfs_mark_buffer_dirty(leaf);
2164 btrfs_release_path(path);
2165
2166 path->reada = READA_FORWARD;
2167 path->leave_spinning = 1;
2168 /* now insert the actual backref */
2169 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2170 path, bytenr, parent, root_objectid,
2171 owner, offset, refs_to_add);
2172 if (ret)
2173 btrfs_abort_transaction(trans, root, ret);
2174 out:
2175 btrfs_free_path(path);
2176 return ret;
2177 }
2178
2179 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2180 struct btrfs_root *root,
2181 struct btrfs_delayed_ref_node *node,
2182 struct btrfs_delayed_extent_op *extent_op,
2183 int insert_reserved)
2184 {
2185 int ret = 0;
2186 struct btrfs_delayed_data_ref *ref;
2187 struct btrfs_key ins;
2188 u64 parent = 0;
2189 u64 ref_root = 0;
2190 u64 flags = 0;
2191
2192 ins.objectid = node->bytenr;
2193 ins.offset = node->num_bytes;
2194 ins.type = BTRFS_EXTENT_ITEM_KEY;
2195
2196 ref = btrfs_delayed_node_to_data_ref(node);
2197 trace_run_delayed_data_ref(node, ref, node->action);
2198
2199 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2200 parent = ref->parent;
2201 ref_root = ref->root;
2202
2203 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2204 if (extent_op)
2205 flags |= extent_op->flags_to_set;
2206 ret = alloc_reserved_file_extent(trans, root,
2207 parent, ref_root, flags,
2208 ref->objectid, ref->offset,
2209 &ins, node->ref_mod);
2210 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2211 ret = __btrfs_inc_extent_ref(trans, root, node, parent,
2212 ref_root, ref->objectid,
2213 ref->offset, node->ref_mod,
2214 extent_op);
2215 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2216 ret = __btrfs_free_extent(trans, root, node, parent,
2217 ref_root, ref->objectid,
2218 ref->offset, node->ref_mod,
2219 extent_op);
2220 } else {
2221 BUG();
2222 }
2223 return ret;
2224 }
2225
2226 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2227 struct extent_buffer *leaf,
2228 struct btrfs_extent_item *ei)
2229 {
2230 u64 flags = btrfs_extent_flags(leaf, ei);
2231 if (extent_op->update_flags) {
2232 flags |= extent_op->flags_to_set;
2233 btrfs_set_extent_flags(leaf, ei, flags);
2234 }
2235
2236 if (extent_op->update_key) {
2237 struct btrfs_tree_block_info *bi;
2238 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2239 bi = (struct btrfs_tree_block_info *)(ei + 1);
2240 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2241 }
2242 }
2243
2244 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2245 struct btrfs_root *root,
2246 struct btrfs_delayed_ref_node *node,
2247 struct btrfs_delayed_extent_op *extent_op)
2248 {
2249 struct btrfs_key key;
2250 struct btrfs_path *path;
2251 struct btrfs_extent_item *ei;
2252 struct extent_buffer *leaf;
2253 u32 item_size;
2254 int ret;
2255 int err = 0;
2256 int metadata = !extent_op->is_data;
2257
2258 if (trans->aborted)
2259 return 0;
2260
2261 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2262 metadata = 0;
2263
2264 path = btrfs_alloc_path();
2265 if (!path)
2266 return -ENOMEM;
2267
2268 key.objectid = node->bytenr;
2269
2270 if (metadata) {
2271 key.type = BTRFS_METADATA_ITEM_KEY;
2272 key.offset = extent_op->level;
2273 } else {
2274 key.type = BTRFS_EXTENT_ITEM_KEY;
2275 key.offset = node->num_bytes;
2276 }
2277
2278 again:
2279 path->reada = READA_FORWARD;
2280 path->leave_spinning = 1;
2281 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2282 path, 0, 1);
2283 if (ret < 0) {
2284 err = ret;
2285 goto out;
2286 }
2287 if (ret > 0) {
2288 if (metadata) {
2289 if (path->slots[0] > 0) {
2290 path->slots[0]--;
2291 btrfs_item_key_to_cpu(path->nodes[0], &key,
2292 path->slots[0]);
2293 if (key.objectid == node->bytenr &&
2294 key.type == BTRFS_EXTENT_ITEM_KEY &&
2295 key.offset == node->num_bytes)
2296 ret = 0;
2297 }
2298 if (ret > 0) {
2299 btrfs_release_path(path);
2300 metadata = 0;
2301
2302 key.objectid = node->bytenr;
2303 key.offset = node->num_bytes;
2304 key.type = BTRFS_EXTENT_ITEM_KEY;
2305 goto again;
2306 }
2307 } else {
2308 err = -EIO;
2309 goto out;
2310 }
2311 }
2312
2313 leaf = path->nodes[0];
2314 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2315 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2316 if (item_size < sizeof(*ei)) {
2317 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2318 path, (u64)-1, 0);
2319 if (ret < 0) {
2320 err = ret;
2321 goto out;
2322 }
2323 leaf = path->nodes[0];
2324 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2325 }
2326 #endif
2327 BUG_ON(item_size < sizeof(*ei));
2328 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2329 __run_delayed_extent_op(extent_op, leaf, ei);
2330
2331 btrfs_mark_buffer_dirty(leaf);
2332 out:
2333 btrfs_free_path(path);
2334 return err;
2335 }
2336
2337 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2338 struct btrfs_root *root,
2339 struct btrfs_delayed_ref_node *node,
2340 struct btrfs_delayed_extent_op *extent_op,
2341 int insert_reserved)
2342 {
2343 int ret = 0;
2344 struct btrfs_delayed_tree_ref *ref;
2345 struct btrfs_key ins;
2346 u64 parent = 0;
2347 u64 ref_root = 0;
2348 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2349 SKINNY_METADATA);
2350
2351 ref = btrfs_delayed_node_to_tree_ref(node);
2352 trace_run_delayed_tree_ref(node, ref, node->action);
2353
2354 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2355 parent = ref->parent;
2356 ref_root = ref->root;
2357
2358 ins.objectid = node->bytenr;
2359 if (skinny_metadata) {
2360 ins.offset = ref->level;
2361 ins.type = BTRFS_METADATA_ITEM_KEY;
2362 } else {
2363 ins.offset = node->num_bytes;
2364 ins.type = BTRFS_EXTENT_ITEM_KEY;
2365 }
2366
2367 BUG_ON(node->ref_mod != 1);
2368 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2369 BUG_ON(!extent_op || !extent_op->update_flags);
2370 ret = alloc_reserved_tree_block(trans, root,
2371 parent, ref_root,
2372 extent_op->flags_to_set,
2373 &extent_op->key,
2374 ref->level, &ins);
2375 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2376 ret = __btrfs_inc_extent_ref(trans, root, node,
2377 parent, ref_root,
2378 ref->level, 0, 1,
2379 extent_op);
2380 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2381 ret = __btrfs_free_extent(trans, root, node,
2382 parent, ref_root,
2383 ref->level, 0, 1, extent_op);
2384 } else {
2385 BUG();
2386 }
2387 return ret;
2388 }
2389
2390 /* helper function to actually process a single delayed ref entry */
2391 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2392 struct btrfs_root *root,
2393 struct btrfs_delayed_ref_node *node,
2394 struct btrfs_delayed_extent_op *extent_op,
2395 int insert_reserved)
2396 {
2397 int ret = 0;
2398
2399 if (trans->aborted) {
2400 if (insert_reserved)
2401 btrfs_pin_extent(root, node->bytenr,
2402 node->num_bytes, 1);
2403 return 0;
2404 }
2405
2406 if (btrfs_delayed_ref_is_head(node)) {
2407 struct btrfs_delayed_ref_head *head;
2408 /*
2409 * we've hit the end of the chain and we were supposed
2410 * to insert this extent into the tree. But, it got
2411 * deleted before we ever needed to insert it, so all
2412 * we have to do is clean up the accounting
2413 */
2414 BUG_ON(extent_op);
2415 head = btrfs_delayed_node_to_head(node);
2416 trace_run_delayed_ref_head(node, head, node->action);
2417
2418 if (insert_reserved) {
2419 btrfs_pin_extent(root, node->bytenr,
2420 node->num_bytes, 1);
2421 if (head->is_data) {
2422 ret = btrfs_del_csums(trans, root,
2423 node->bytenr,
2424 node->num_bytes);
2425 }
2426 }
2427
2428 /* Also free its reserved qgroup space */
2429 btrfs_qgroup_free_delayed_ref(root->fs_info,
2430 head->qgroup_ref_root,
2431 head->qgroup_reserved);
2432 return ret;
2433 }
2434
2435 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2436 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2437 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2438 insert_reserved);
2439 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2440 node->type == BTRFS_SHARED_DATA_REF_KEY)
2441 ret = run_delayed_data_ref(trans, root, node, extent_op,
2442 insert_reserved);
2443 else
2444 BUG();
2445 return ret;
2446 }
2447
2448 static inline struct btrfs_delayed_ref_node *
2449 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2450 {
2451 struct btrfs_delayed_ref_node *ref;
2452
2453 if (list_empty(&head->ref_list))
2454 return NULL;
2455
2456 /*
2457 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2458 * This is to prevent a ref count from going down to zero, which deletes
2459 * the extent item from the extent tree, when there still are references
2460 * to add, which would fail because they would not find the extent item.
2461 */
2462 list_for_each_entry(ref, &head->ref_list, list) {
2463 if (ref->action == BTRFS_ADD_DELAYED_REF)
2464 return ref;
2465 }
2466
2467 return list_entry(head->ref_list.next, struct btrfs_delayed_ref_node,
2468 list);
2469 }
2470
2471 /*
2472 * Returns 0 on success or if called with an already aborted transaction.
2473 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2474 */
2475 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2476 struct btrfs_root *root,
2477 unsigned long nr)
2478 {
2479 struct btrfs_delayed_ref_root *delayed_refs;
2480 struct btrfs_delayed_ref_node *ref;
2481 struct btrfs_delayed_ref_head *locked_ref = NULL;
2482 struct btrfs_delayed_extent_op *extent_op;
2483 struct btrfs_fs_info *fs_info = root->fs_info;
2484 ktime_t start = ktime_get();
2485 int ret;
2486 unsigned long count = 0;
2487 unsigned long actual_count = 0;
2488 int must_insert_reserved = 0;
2489
2490 delayed_refs = &trans->transaction->delayed_refs;
2491 while (1) {
2492 if (!locked_ref) {
2493 if (count >= nr)
2494 break;
2495
2496 spin_lock(&delayed_refs->lock);
2497 locked_ref = btrfs_select_ref_head(trans);
2498 if (!locked_ref) {
2499 spin_unlock(&delayed_refs->lock);
2500 break;
2501 }
2502
2503 /* grab the lock that says we are going to process
2504 * all the refs for this head */
2505 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2506 spin_unlock(&delayed_refs->lock);
2507 /*
2508 * we may have dropped the spin lock to get the head
2509 * mutex lock, and that might have given someone else
2510 * time to free the head. If that's true, it has been
2511 * removed from our list and we can move on.
2512 */
2513 if (ret == -EAGAIN) {
2514 locked_ref = NULL;
2515 count++;
2516 continue;
2517 }
2518 }
2519
2520 /*
2521 * We need to try and merge add/drops of the same ref since we
2522 * can run into issues with relocate dropping the implicit ref
2523 * and then it being added back again before the drop can
2524 * finish. If we merged anything we need to re-loop so we can
2525 * get a good ref.
2526 * Or we can get node references of the same type that weren't
2527 * merged when created due to bumps in the tree mod seq, and
2528 * we need to merge them to prevent adding an inline extent
2529 * backref before dropping it (triggering a BUG_ON at
2530 * insert_inline_extent_backref()).
2531 */
2532 spin_lock(&locked_ref->lock);
2533 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2534 locked_ref);
2535
2536 /*
2537 * locked_ref is the head node, so we have to go one
2538 * node back for any delayed ref updates
2539 */
2540 ref = select_delayed_ref(locked_ref);
2541
2542 if (ref && ref->seq &&
2543 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2544 spin_unlock(&locked_ref->lock);
2545 btrfs_delayed_ref_unlock(locked_ref);
2546 spin_lock(&delayed_refs->lock);
2547 locked_ref->processing = 0;
2548 delayed_refs->num_heads_ready++;
2549 spin_unlock(&delayed_refs->lock);
2550 locked_ref = NULL;
2551 cond_resched();
2552 count++;
2553 continue;
2554 }
2555
2556 /*
2557 * record the must insert reserved flag before we
2558 * drop the spin lock.
2559 */
2560 must_insert_reserved = locked_ref->must_insert_reserved;
2561 locked_ref->must_insert_reserved = 0;
2562
2563 extent_op = locked_ref->extent_op;
2564 locked_ref->extent_op = NULL;
2565
2566 if (!ref) {
2567
2568
2569 /* All delayed refs have been processed, Go ahead
2570 * and send the head node to run_one_delayed_ref,
2571 * so that any accounting fixes can happen
2572 */
2573 ref = &locked_ref->node;
2574
2575 if (extent_op && must_insert_reserved) {
2576 btrfs_free_delayed_extent_op(extent_op);
2577 extent_op = NULL;
2578 }
2579
2580 if (extent_op) {
2581 spin_unlock(&locked_ref->lock);
2582 ret = run_delayed_extent_op(trans, root,
2583 ref, extent_op);
2584 btrfs_free_delayed_extent_op(extent_op);
2585
2586 if (ret) {
2587 /*
2588 * Need to reset must_insert_reserved if
2589 * there was an error so the abort stuff
2590 * can cleanup the reserved space
2591 * properly.
2592 */
2593 if (must_insert_reserved)
2594 locked_ref->must_insert_reserved = 1;
2595 locked_ref->processing = 0;
2596 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2597 btrfs_delayed_ref_unlock(locked_ref);
2598 return ret;
2599 }
2600 continue;
2601 }
2602
2603 /*
2604 * Need to drop our head ref lock and re-acquire the
2605 * delayed ref lock and then re-check to make sure
2606 * nobody got added.
2607 */
2608 spin_unlock(&locked_ref->lock);
2609 spin_lock(&delayed_refs->lock);
2610 spin_lock(&locked_ref->lock);
2611 if (!list_empty(&locked_ref->ref_list) ||
2612 locked_ref->extent_op) {
2613 spin_unlock(&locked_ref->lock);
2614 spin_unlock(&delayed_refs->lock);
2615 continue;
2616 }
2617 ref->in_tree = 0;
2618 delayed_refs->num_heads--;
2619 rb_erase(&locked_ref->href_node,
2620 &delayed_refs->href_root);
2621 spin_unlock(&delayed_refs->lock);
2622 } else {
2623 actual_count++;
2624 ref->in_tree = 0;
2625 list_del(&ref->list);
2626 }
2627 atomic_dec(&delayed_refs->num_entries);
2628
2629 if (!btrfs_delayed_ref_is_head(ref)) {
2630 /*
2631 * when we play the delayed ref, also correct the
2632 * ref_mod on head
2633 */
2634 switch (ref->action) {
2635 case BTRFS_ADD_DELAYED_REF:
2636 case BTRFS_ADD_DELAYED_EXTENT:
2637 locked_ref->node.ref_mod -= ref->ref_mod;
2638 break;
2639 case BTRFS_DROP_DELAYED_REF:
2640 locked_ref->node.ref_mod += ref->ref_mod;
2641 break;
2642 default:
2643 WARN_ON(1);
2644 }
2645 }
2646 spin_unlock(&locked_ref->lock);
2647
2648 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2649 must_insert_reserved);
2650
2651 btrfs_free_delayed_extent_op(extent_op);
2652 if (ret) {
2653 locked_ref->processing = 0;
2654 btrfs_delayed_ref_unlock(locked_ref);
2655 btrfs_put_delayed_ref(ref);
2656 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2657 return ret;
2658 }
2659
2660 /*
2661 * If this node is a head, that means all the refs in this head
2662 * have been dealt with, and we will pick the next head to deal
2663 * with, so we must unlock the head and drop it from the cluster
2664 * list before we release it.
2665 */
2666 if (btrfs_delayed_ref_is_head(ref)) {
2667 if (locked_ref->is_data &&
2668 locked_ref->total_ref_mod < 0) {
2669 spin_lock(&delayed_refs->lock);
2670 delayed_refs->pending_csums -= ref->num_bytes;
2671 spin_unlock(&delayed_refs->lock);
2672 }
2673 btrfs_delayed_ref_unlock(locked_ref);
2674 locked_ref = NULL;
2675 }
2676 btrfs_put_delayed_ref(ref);
2677 count++;
2678 cond_resched();
2679 }
2680
2681 /*
2682 * We don't want to include ref heads since we can have empty ref heads
2683 * and those will drastically skew our runtime down since we just do
2684 * accounting, no actual extent tree updates.
2685 */
2686 if (actual_count > 0) {
2687 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2688 u64 avg;
2689
2690 /*
2691 * We weigh the current average higher than our current runtime
2692 * to avoid large swings in the average.
2693 */
2694 spin_lock(&delayed_refs->lock);
2695 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2696 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2697 spin_unlock(&delayed_refs->lock);
2698 }
2699 return 0;
2700 }
2701
2702 #ifdef SCRAMBLE_DELAYED_REFS
2703 /*
2704 * Normally delayed refs get processed in ascending bytenr order. This
2705 * correlates in most cases to the order added. To expose dependencies on this
2706 * order, we start to process the tree in the middle instead of the beginning
2707 */
2708 static u64 find_middle(struct rb_root *root)
2709 {
2710 struct rb_node *n = root->rb_node;
2711 struct btrfs_delayed_ref_node *entry;
2712 int alt = 1;
2713 u64 middle;
2714 u64 first = 0, last = 0;
2715
2716 n = rb_first(root);
2717 if (n) {
2718 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2719 first = entry->bytenr;
2720 }
2721 n = rb_last(root);
2722 if (n) {
2723 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2724 last = entry->bytenr;
2725 }
2726 n = root->rb_node;
2727
2728 while (n) {
2729 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2730 WARN_ON(!entry->in_tree);
2731
2732 middle = entry->bytenr;
2733
2734 if (alt)
2735 n = n->rb_left;
2736 else
2737 n = n->rb_right;
2738
2739 alt = 1 - alt;
2740 }
2741 return middle;
2742 }
2743 #endif
2744
2745 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2746 {
2747 u64 num_bytes;
2748
2749 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2750 sizeof(struct btrfs_extent_inline_ref));
2751 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2752 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2753
2754 /*
2755 * We don't ever fill up leaves all the way so multiply by 2 just to be
2756 * closer to what we're really going to want to use.
2757 */
2758 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2759 }
2760
2761 /*
2762 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2763 * would require to store the csums for that many bytes.
2764 */
2765 u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes)
2766 {
2767 u64 csum_size;
2768 u64 num_csums_per_leaf;
2769 u64 num_csums;
2770
2771 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
2772 num_csums_per_leaf = div64_u64(csum_size,
2773 (u64)btrfs_super_csum_size(root->fs_info->super_copy));
2774 num_csums = div64_u64(csum_bytes, root->sectorsize);
2775 num_csums += num_csums_per_leaf - 1;
2776 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2777 return num_csums;
2778 }
2779
2780 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2781 struct btrfs_root *root)
2782 {
2783 struct btrfs_block_rsv *global_rsv;
2784 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2785 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2786 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2787 u64 num_bytes, num_dirty_bgs_bytes;
2788 int ret = 0;
2789
2790 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2791 num_heads = heads_to_leaves(root, num_heads);
2792 if (num_heads > 1)
2793 num_bytes += (num_heads - 1) * root->nodesize;
2794 num_bytes <<= 1;
2795 num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize;
2796 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root,
2797 num_dirty_bgs);
2798 global_rsv = &root->fs_info->global_block_rsv;
2799
2800 /*
2801 * If we can't allocate any more chunks lets make sure we have _lots_ of
2802 * wiggle room since running delayed refs can create more delayed refs.
2803 */
2804 if (global_rsv->space_info->full) {
2805 num_dirty_bgs_bytes <<= 1;
2806 num_bytes <<= 1;
2807 }
2808
2809 spin_lock(&global_rsv->lock);
2810 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2811 ret = 1;
2812 spin_unlock(&global_rsv->lock);
2813 return ret;
2814 }
2815
2816 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2817 struct btrfs_root *root)
2818 {
2819 struct btrfs_fs_info *fs_info = root->fs_info;
2820 u64 num_entries =
2821 atomic_read(&trans->transaction->delayed_refs.num_entries);
2822 u64 avg_runtime;
2823 u64 val;
2824
2825 smp_mb();
2826 avg_runtime = fs_info->avg_delayed_ref_runtime;
2827 val = num_entries * avg_runtime;
2828 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2829 return 1;
2830 if (val >= NSEC_PER_SEC / 2)
2831 return 2;
2832
2833 return btrfs_check_space_for_delayed_refs(trans, root);
2834 }
2835
2836 struct async_delayed_refs {
2837 struct btrfs_root *root;
2838 u64 transid;
2839 int count;
2840 int error;
2841 int sync;
2842 struct completion wait;
2843 struct btrfs_work work;
2844 };
2845
2846 static void delayed_ref_async_start(struct btrfs_work *work)
2847 {
2848 struct async_delayed_refs *async;
2849 struct btrfs_trans_handle *trans;
2850 int ret;
2851
2852 async = container_of(work, struct async_delayed_refs, work);
2853
2854 /* if the commit is already started, we don't need to wait here */
2855 if (btrfs_transaction_blocked(async->root->fs_info))
2856 goto done;
2857
2858 trans = btrfs_join_transaction(async->root);
2859 if (IS_ERR(trans)) {
2860 async->error = PTR_ERR(trans);
2861 goto done;
2862 }
2863
2864 /*
2865 * trans->sync means that when we call end_transaction, we won't
2866 * wait on delayed refs
2867 */
2868 trans->sync = true;
2869
2870 /* Don't bother flushing if we got into a different transaction */
2871 if (trans->transid > async->transid)
2872 goto end;
2873
2874 ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2875 if (ret)
2876 async->error = ret;
2877 end:
2878 ret = btrfs_end_transaction(trans, async->root);
2879 if (ret && !async->error)
2880 async->error = ret;
2881 done:
2882 if (async->sync)
2883 complete(&async->wait);
2884 else
2885 kfree(async);
2886 }
2887
2888 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2889 unsigned long count, u64 transid, int wait)
2890 {
2891 struct async_delayed_refs *async;
2892 int ret;
2893
2894 async = kmalloc(sizeof(*async), GFP_NOFS);
2895 if (!async)
2896 return -ENOMEM;
2897
2898 async->root = root->fs_info->tree_root;
2899 async->count = count;
2900 async->error = 0;
2901 async->transid = transid;
2902 if (wait)
2903 async->sync = 1;
2904 else
2905 async->sync = 0;
2906 init_completion(&async->wait);
2907
2908 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2909 delayed_ref_async_start, NULL, NULL);
2910
2911 btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2912
2913 if (wait) {
2914 wait_for_completion(&async->wait);
2915 ret = async->error;
2916 kfree(async);
2917 return ret;
2918 }
2919 return 0;
2920 }
2921
2922 /*
2923 * this starts processing the delayed reference count updates and
2924 * extent insertions we have queued up so far. count can be
2925 * 0, which means to process everything in the tree at the start
2926 * of the run (but not newly added entries), or it can be some target
2927 * number you'd like to process.
2928 *
2929 * Returns 0 on success or if called with an aborted transaction
2930 * Returns <0 on error and aborts the transaction
2931 */
2932 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2933 struct btrfs_root *root, unsigned long count)
2934 {
2935 struct rb_node *node;
2936 struct btrfs_delayed_ref_root *delayed_refs;
2937 struct btrfs_delayed_ref_head *head;
2938 int ret;
2939 int run_all = count == (unsigned long)-1;
2940 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
2941
2942 /* We'll clean this up in btrfs_cleanup_transaction */
2943 if (trans->aborted)
2944 return 0;
2945
2946 if (root->fs_info->creating_free_space_tree)
2947 return 0;
2948
2949 if (root == root->fs_info->extent_root)
2950 root = root->fs_info->tree_root;
2951
2952 delayed_refs = &trans->transaction->delayed_refs;
2953 if (count == 0)
2954 count = atomic_read(&delayed_refs->num_entries) * 2;
2955
2956 again:
2957 #ifdef SCRAMBLE_DELAYED_REFS
2958 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2959 #endif
2960 trans->can_flush_pending_bgs = false;
2961 ret = __btrfs_run_delayed_refs(trans, root, count);
2962 if (ret < 0) {
2963 btrfs_abort_transaction(trans, root, ret);
2964 return ret;
2965 }
2966
2967 if (run_all) {
2968 if (!list_empty(&trans->new_bgs))
2969 btrfs_create_pending_block_groups(trans, root);
2970
2971 spin_lock(&delayed_refs->lock);
2972 node = rb_first(&delayed_refs->href_root);
2973 if (!node) {
2974 spin_unlock(&delayed_refs->lock);
2975 goto out;
2976 }
2977 count = (unsigned long)-1;
2978
2979 while (node) {
2980 head = rb_entry(node, struct btrfs_delayed_ref_head,
2981 href_node);
2982 if (btrfs_delayed_ref_is_head(&head->node)) {
2983 struct btrfs_delayed_ref_node *ref;
2984
2985 ref = &head->node;
2986 atomic_inc(&ref->refs);
2987
2988 spin_unlock(&delayed_refs->lock);
2989 /*
2990 * Mutex was contended, block until it's
2991 * released and try again
2992 */
2993 mutex_lock(&head->mutex);
2994 mutex_unlock(&head->mutex);
2995
2996 btrfs_put_delayed_ref(ref);
2997 cond_resched();
2998 goto again;
2999 } else {
3000 WARN_ON(1);
3001 }
3002 node = rb_next(node);
3003 }
3004 spin_unlock(&delayed_refs->lock);
3005 cond_resched();
3006 goto again;
3007 }
3008 out:
3009 assert_qgroups_uptodate(trans);
3010 trans->can_flush_pending_bgs = can_flush_pending_bgs;
3011 return 0;
3012 }
3013
3014 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
3015 struct btrfs_root *root,
3016 u64 bytenr, u64 num_bytes, u64 flags,
3017 int level, int is_data)
3018 {
3019 struct btrfs_delayed_extent_op *extent_op;
3020 int ret;
3021
3022 extent_op = btrfs_alloc_delayed_extent_op();
3023 if (!extent_op)
3024 return -ENOMEM;
3025
3026 extent_op->flags_to_set = flags;
3027 extent_op->update_flags = true;
3028 extent_op->update_key = false;
3029 extent_op->is_data = is_data ? true : false;
3030 extent_op->level = level;
3031
3032 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
3033 num_bytes, extent_op);
3034 if (ret)
3035 btrfs_free_delayed_extent_op(extent_op);
3036 return ret;
3037 }
3038
3039 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
3040 struct btrfs_root *root,
3041 struct btrfs_path *path,
3042 u64 objectid, u64 offset, u64 bytenr)
3043 {
3044 struct btrfs_delayed_ref_head *head;
3045 struct btrfs_delayed_ref_node *ref;
3046 struct btrfs_delayed_data_ref *data_ref;
3047 struct btrfs_delayed_ref_root *delayed_refs;
3048 int ret = 0;
3049
3050 delayed_refs = &trans->transaction->delayed_refs;
3051 spin_lock(&delayed_refs->lock);
3052 head = btrfs_find_delayed_ref_head(trans, bytenr);
3053 if (!head) {
3054 spin_unlock(&delayed_refs->lock);
3055 return 0;
3056 }
3057
3058 if (!mutex_trylock(&head->mutex)) {
3059 atomic_inc(&head->node.refs);
3060 spin_unlock(&delayed_refs->lock);
3061
3062 btrfs_release_path(path);
3063
3064 /*
3065 * Mutex was contended, block until it's released and let
3066 * caller try again
3067 */
3068 mutex_lock(&head->mutex);
3069 mutex_unlock(&head->mutex);
3070 btrfs_put_delayed_ref(&head->node);
3071 return -EAGAIN;
3072 }
3073 spin_unlock(&delayed_refs->lock);
3074
3075 spin_lock(&head->lock);
3076 list_for_each_entry(ref, &head->ref_list, list) {
3077 /* If it's a shared ref we know a cross reference exists */
3078 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3079 ret = 1;
3080 break;
3081 }
3082
3083 data_ref = btrfs_delayed_node_to_data_ref(ref);
3084
3085 /*
3086 * If our ref doesn't match the one we're currently looking at
3087 * then we have a cross reference.
3088 */
3089 if (data_ref->root != root->root_key.objectid ||
3090 data_ref->objectid != objectid ||
3091 data_ref->offset != offset) {
3092 ret = 1;
3093 break;
3094 }
3095 }
3096 spin_unlock(&head->lock);
3097 mutex_unlock(&head->mutex);
3098 return ret;
3099 }
3100
3101 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
3102 struct btrfs_root *root,
3103 struct btrfs_path *path,
3104 u64 objectid, u64 offset, u64 bytenr)
3105 {
3106 struct btrfs_root *extent_root = root->fs_info->extent_root;
3107 struct extent_buffer *leaf;
3108 struct btrfs_extent_data_ref *ref;
3109 struct btrfs_extent_inline_ref *iref;
3110 struct btrfs_extent_item *ei;
3111 struct btrfs_key key;
3112 u32 item_size;
3113 int ret;
3114
3115 key.objectid = bytenr;
3116 key.offset = (u64)-1;
3117 key.type = BTRFS_EXTENT_ITEM_KEY;
3118
3119 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3120 if (ret < 0)
3121 goto out;
3122 BUG_ON(ret == 0); /* Corruption */
3123
3124 ret = -ENOENT;
3125 if (path->slots[0] == 0)
3126 goto out;
3127
3128 path->slots[0]--;
3129 leaf = path->nodes[0];
3130 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3131
3132 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3133 goto out;
3134
3135 ret = 1;
3136 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3137 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3138 if (item_size < sizeof(*ei)) {
3139 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3140 goto out;
3141 }
3142 #endif
3143 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3144
3145 if (item_size != sizeof(*ei) +
3146 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3147 goto out;
3148
3149 if (btrfs_extent_generation(leaf, ei) <=
3150 btrfs_root_last_snapshot(&root->root_item))
3151 goto out;
3152
3153 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3154 if (btrfs_extent_inline_ref_type(leaf, iref) !=
3155 BTRFS_EXTENT_DATA_REF_KEY)
3156 goto out;
3157
3158 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3159 if (btrfs_extent_refs(leaf, ei) !=
3160 btrfs_extent_data_ref_count(leaf, ref) ||
3161 btrfs_extent_data_ref_root(leaf, ref) !=
3162 root->root_key.objectid ||
3163 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3164 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3165 goto out;
3166
3167 ret = 0;
3168 out:
3169 return ret;
3170 }
3171
3172 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
3173 struct btrfs_root *root,
3174 u64 objectid, u64 offset, u64 bytenr)
3175 {
3176 struct btrfs_path *path;
3177 int ret;
3178 int ret2;
3179
3180 path = btrfs_alloc_path();
3181 if (!path)
3182 return -ENOENT;
3183
3184 do {
3185 ret = check_committed_ref(trans, root, path, objectid,
3186 offset, bytenr);
3187 if (ret && ret != -ENOENT)
3188 goto out;
3189
3190 ret2 = check_delayed_ref(trans, root, path, objectid,
3191 offset, bytenr);
3192 } while (ret2 == -EAGAIN);
3193
3194 if (ret2 && ret2 != -ENOENT) {
3195 ret = ret2;
3196 goto out;
3197 }
3198
3199 if (ret != -ENOENT || ret2 != -ENOENT)
3200 ret = 0;
3201 out:
3202 btrfs_free_path(path);
3203 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3204 WARN_ON(ret > 0);
3205 return ret;
3206 }
3207
3208 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3209 struct btrfs_root *root,
3210 struct extent_buffer *buf,
3211 int full_backref, int inc)
3212 {
3213 u64 bytenr;
3214 u64 num_bytes;
3215 u64 parent;
3216 u64 ref_root;
3217 u32 nritems;
3218 struct btrfs_key key;
3219 struct btrfs_file_extent_item *fi;
3220 int i;
3221 int level;
3222 int ret = 0;
3223 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3224 u64, u64, u64, u64, u64, u64);
3225
3226
3227 if (btrfs_test_is_dummy_root(root))
3228 return 0;
3229
3230 ref_root = btrfs_header_owner(buf);
3231 nritems = btrfs_header_nritems(buf);
3232 level = btrfs_header_level(buf);
3233
3234 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3235 return 0;
3236
3237 if (inc)
3238 process_func = btrfs_inc_extent_ref;
3239 else
3240 process_func = btrfs_free_extent;
3241
3242 if (full_backref)
3243 parent = buf->start;
3244 else
3245 parent = 0;
3246
3247 for (i = 0; i < nritems; i++) {
3248 if (level == 0) {
3249 btrfs_item_key_to_cpu(buf, &key, i);
3250 if (key.type != BTRFS_EXTENT_DATA_KEY)
3251 continue;
3252 fi = btrfs_item_ptr(buf, i,
3253 struct btrfs_file_extent_item);
3254 if (btrfs_file_extent_type(buf, fi) ==
3255 BTRFS_FILE_EXTENT_INLINE)
3256 continue;
3257 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3258 if (bytenr == 0)
3259 continue;
3260
3261 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3262 key.offset -= btrfs_file_extent_offset(buf, fi);
3263 ret = process_func(trans, root, bytenr, num_bytes,
3264 parent, ref_root, key.objectid,
3265 key.offset);
3266 if (ret)
3267 goto fail;
3268 } else {
3269 bytenr = btrfs_node_blockptr(buf, i);
3270 num_bytes = root->nodesize;
3271 ret = process_func(trans, root, bytenr, num_bytes,
3272 parent, ref_root, level - 1, 0);
3273 if (ret)
3274 goto fail;
3275 }
3276 }
3277 return 0;
3278 fail:
3279 return ret;
3280 }
3281
3282 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3283 struct extent_buffer *buf, int full_backref)
3284 {
3285 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3286 }
3287
3288 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3289 struct extent_buffer *buf, int full_backref)
3290 {
3291 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3292 }
3293
3294 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3295 struct btrfs_root *root,
3296 struct btrfs_path *path,
3297 struct btrfs_block_group_cache *cache)
3298 {
3299 int ret;
3300 struct btrfs_root *extent_root = root->fs_info->extent_root;
3301 unsigned long bi;
3302 struct extent_buffer *leaf;
3303
3304 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3305 if (ret) {
3306 if (ret > 0)
3307 ret = -ENOENT;
3308 goto fail;
3309 }
3310
3311 leaf = path->nodes[0];
3312 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3313 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3314 btrfs_mark_buffer_dirty(leaf);
3315 fail:
3316 btrfs_release_path(path);
3317 return ret;
3318
3319 }
3320
3321 static struct btrfs_block_group_cache *
3322 next_block_group(struct btrfs_root *root,
3323 struct btrfs_block_group_cache *cache)
3324 {
3325 struct rb_node *node;
3326
3327 spin_lock(&root->fs_info->block_group_cache_lock);
3328
3329 /* If our block group was removed, we need a full search. */
3330 if (RB_EMPTY_NODE(&cache->cache_node)) {
3331 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3332
3333 spin_unlock(&root->fs_info->block_group_cache_lock);
3334 btrfs_put_block_group(cache);
3335 cache = btrfs_lookup_first_block_group(root->fs_info,
3336 next_bytenr);
3337 return cache;
3338 }
3339 node = rb_next(&cache->cache_node);
3340 btrfs_put_block_group(cache);
3341 if (node) {
3342 cache = rb_entry(node, struct btrfs_block_group_cache,
3343 cache_node);
3344 btrfs_get_block_group(cache);
3345 } else
3346 cache = NULL;
3347 spin_unlock(&root->fs_info->block_group_cache_lock);
3348 return cache;
3349 }
3350
3351 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3352 struct btrfs_trans_handle *trans,
3353 struct btrfs_path *path)
3354 {
3355 struct btrfs_root *root = block_group->fs_info->tree_root;
3356 struct inode *inode = NULL;
3357 u64 alloc_hint = 0;
3358 int dcs = BTRFS_DC_ERROR;
3359 u64 num_pages = 0;
3360 int retries = 0;
3361 int ret = 0;
3362
3363 /*
3364 * If this block group is smaller than 100 megs don't bother caching the
3365 * block group.
3366 */
3367 if (block_group->key.offset < (100 * SZ_1M)) {
3368 spin_lock(&block_group->lock);
3369 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3370 spin_unlock(&block_group->lock);
3371 return 0;
3372 }
3373
3374 if (trans->aborted)
3375 return 0;
3376 again:
3377 inode = lookup_free_space_inode(root, block_group, path);
3378 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3379 ret = PTR_ERR(inode);
3380 btrfs_release_path(path);
3381 goto out;
3382 }
3383
3384 if (IS_ERR(inode)) {
3385 BUG_ON(retries);
3386 retries++;
3387
3388 if (block_group->ro)
3389 goto out_free;
3390
3391 ret = create_free_space_inode(root, trans, block_group, path);
3392 if (ret)
3393 goto out_free;
3394 goto again;
3395 }
3396
3397 /* We've already setup this transaction, go ahead and exit */
3398 if (block_group->cache_generation == trans->transid &&
3399 i_size_read(inode)) {
3400 dcs = BTRFS_DC_SETUP;
3401 goto out_put;
3402 }
3403
3404 /*
3405 * We want to set the generation to 0, that way if anything goes wrong
3406 * from here on out we know not to trust this cache when we load up next
3407 * time.
3408 */
3409 BTRFS_I(inode)->generation = 0;
3410 ret = btrfs_update_inode(trans, root, inode);
3411 if (ret) {
3412 /*
3413 * So theoretically we could recover from this, simply set the
3414 * super cache generation to 0 so we know to invalidate the
3415 * cache, but then we'd have to keep track of the block groups
3416 * that fail this way so we know we _have_ to reset this cache
3417 * before the next commit or risk reading stale cache. So to
3418 * limit our exposure to horrible edge cases lets just abort the
3419 * transaction, this only happens in really bad situations
3420 * anyway.
3421 */
3422 btrfs_abort_transaction(trans, root, ret);
3423 goto out_put;
3424 }
3425 WARN_ON(ret);
3426
3427 if (i_size_read(inode) > 0) {
3428 ret = btrfs_check_trunc_cache_free_space(root,
3429 &root->fs_info->global_block_rsv);
3430 if (ret)
3431 goto out_put;
3432
3433 ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode);
3434 if (ret)
3435 goto out_put;
3436 }
3437
3438 spin_lock(&block_group->lock);
3439 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3440 !btrfs_test_opt(root, SPACE_CACHE)) {
3441 /*
3442 * don't bother trying to write stuff out _if_
3443 * a) we're not cached,
3444 * b) we're with nospace_cache mount option.
3445 */
3446 dcs = BTRFS_DC_WRITTEN;
3447 spin_unlock(&block_group->lock);
3448 goto out_put;
3449 }
3450 spin_unlock(&block_group->lock);
3451
3452 /*
3453 * We hit an ENOSPC when setting up the cache in this transaction, just
3454 * skip doing the setup, we've already cleared the cache so we're safe.
3455 */
3456 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3457 ret = -ENOSPC;
3458 goto out_put;
3459 }
3460
3461 /*
3462 * Try to preallocate enough space based on how big the block group is.
3463 * Keep in mind this has to include any pinned space which could end up
3464 * taking up quite a bit since it's not folded into the other space
3465 * cache.
3466 */
3467 num_pages = div_u64(block_group->key.offset, SZ_256M);
3468 if (!num_pages)
3469 num_pages = 1;
3470
3471 num_pages *= 16;
3472 num_pages *= PAGE_SIZE;
3473
3474 ret = btrfs_check_data_free_space(inode, 0, num_pages);
3475 if (ret)
3476 goto out_put;
3477
3478 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3479 num_pages, num_pages,
3480 &alloc_hint);
3481 /*
3482 * Our cache requires contiguous chunks so that we don't modify a bunch
3483 * of metadata or split extents when writing the cache out, which means
3484 * we can enospc if we are heavily fragmented in addition to just normal
3485 * out of space conditions. So if we hit this just skip setting up any
3486 * other block groups for this transaction, maybe we'll unpin enough
3487 * space the next time around.
3488 */
3489 if (!ret)
3490 dcs = BTRFS_DC_SETUP;
3491 else if (ret == -ENOSPC)
3492 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3493 btrfs_free_reserved_data_space(inode, 0, num_pages);
3494
3495 out_put:
3496 iput(inode);
3497 out_free:
3498 btrfs_release_path(path);
3499 out:
3500 spin_lock(&block_group->lock);
3501 if (!ret && dcs == BTRFS_DC_SETUP)
3502 block_group->cache_generation = trans->transid;
3503 block_group->disk_cache_state = dcs;
3504 spin_unlock(&block_group->lock);
3505
3506 return ret;
3507 }
3508
3509 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3510 struct btrfs_root *root)
3511 {
3512 struct btrfs_block_group_cache *cache, *tmp;
3513 struct btrfs_transaction *cur_trans = trans->transaction;
3514 struct btrfs_path *path;
3515
3516 if (list_empty(&cur_trans->dirty_bgs) ||
3517 !btrfs_test_opt(root, SPACE_CACHE))
3518 return 0;
3519
3520 path = btrfs_alloc_path();
3521 if (!path)
3522 return -ENOMEM;
3523
3524 /* Could add new block groups, use _safe just in case */
3525 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3526 dirty_list) {
3527 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3528 cache_save_setup(cache, trans, path);
3529 }
3530
3531 btrfs_free_path(path);
3532 return 0;
3533 }
3534
3535 /*
3536 * transaction commit does final block group cache writeback during a
3537 * critical section where nothing is allowed to change the FS. This is
3538 * required in order for the cache to actually match the block group,
3539 * but can introduce a lot of latency into the commit.
3540 *
3541 * So, btrfs_start_dirty_block_groups is here to kick off block group
3542 * cache IO. There's a chance we'll have to redo some of it if the
3543 * block group changes again during the commit, but it greatly reduces
3544 * the commit latency by getting rid of the easy block groups while
3545 * we're still allowing others to join the commit.
3546 */
3547 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3548 struct btrfs_root *root)
3549 {
3550 struct btrfs_block_group_cache *cache;
3551 struct btrfs_transaction *cur_trans = trans->transaction;
3552 int ret = 0;
3553 int should_put;
3554 struct btrfs_path *path = NULL;
3555 LIST_HEAD(dirty);
3556 struct list_head *io = &cur_trans->io_bgs;
3557 int num_started = 0;
3558 int loops = 0;
3559
3560 spin_lock(&cur_trans->dirty_bgs_lock);
3561 if (list_empty(&cur_trans->dirty_bgs)) {
3562 spin_unlock(&cur_trans->dirty_bgs_lock);
3563 return 0;
3564 }
3565 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3566 spin_unlock(&cur_trans->dirty_bgs_lock);
3567
3568 again:
3569 /*
3570 * make sure all the block groups on our dirty list actually
3571 * exist
3572 */
3573 btrfs_create_pending_block_groups(trans, root);
3574
3575 if (!path) {
3576 path = btrfs_alloc_path();
3577 if (!path)
3578 return -ENOMEM;
3579 }
3580
3581 /*
3582 * cache_write_mutex is here only to save us from balance or automatic
3583 * removal of empty block groups deleting this block group while we are
3584 * writing out the cache
3585 */
3586 mutex_lock(&trans->transaction->cache_write_mutex);
3587 while (!list_empty(&dirty)) {
3588 cache = list_first_entry(&dirty,
3589 struct btrfs_block_group_cache,
3590 dirty_list);
3591 /*
3592 * this can happen if something re-dirties a block
3593 * group that is already under IO. Just wait for it to
3594 * finish and then do it all again
3595 */
3596 if (!list_empty(&cache->io_list)) {
3597 list_del_init(&cache->io_list);
3598 btrfs_wait_cache_io(root, trans, cache,
3599 &cache->io_ctl, path,
3600 cache->key.objectid);
3601 btrfs_put_block_group(cache);
3602 }
3603
3604
3605 /*
3606 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3607 * if it should update the cache_state. Don't delete
3608 * until after we wait.
3609 *
3610 * Since we're not running in the commit critical section
3611 * we need the dirty_bgs_lock to protect from update_block_group
3612 */
3613 spin_lock(&cur_trans->dirty_bgs_lock);
3614 list_del_init(&cache->dirty_list);
3615 spin_unlock(&cur_trans->dirty_bgs_lock);
3616
3617 should_put = 1;
3618
3619 cache_save_setup(cache, trans, path);
3620
3621 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3622 cache->io_ctl.inode = NULL;
3623 ret = btrfs_write_out_cache(root, trans, cache, path);
3624 if (ret == 0 && cache->io_ctl.inode) {
3625 num_started++;
3626 should_put = 0;
3627
3628 /*
3629 * the cache_write_mutex is protecting
3630 * the io_list
3631 */
3632 list_add_tail(&cache->io_list, io);
3633 } else {
3634 /*
3635 * if we failed to write the cache, the
3636 * generation will be bad and life goes on
3637 */
3638 ret = 0;
3639 }
3640 }
3641 if (!ret) {
3642 ret = write_one_cache_group(trans, root, path, cache);
3643 /*
3644 * Our block group might still be attached to the list
3645 * of new block groups in the transaction handle of some
3646 * other task (struct btrfs_trans_handle->new_bgs). This
3647 * means its block group item isn't yet in the extent
3648 * tree. If this happens ignore the error, as we will
3649 * try again later in the critical section of the
3650 * transaction commit.
3651 */
3652 if (ret == -ENOENT) {
3653 ret = 0;
3654 spin_lock(&cur_trans->dirty_bgs_lock);
3655 if (list_empty(&cache->dirty_list)) {
3656 list_add_tail(&cache->dirty_list,
3657 &cur_trans->dirty_bgs);
3658 btrfs_get_block_group(cache);
3659 }
3660 spin_unlock(&cur_trans->dirty_bgs_lock);
3661 } else if (ret) {
3662 btrfs_abort_transaction(trans, root, ret);
3663 }
3664 }
3665
3666 /* if its not on the io list, we need to put the block group */
3667 if (should_put)
3668 btrfs_put_block_group(cache);
3669
3670 if (ret)
3671 break;
3672
3673 /*
3674 * Avoid blocking other tasks for too long. It might even save
3675 * us from writing caches for block groups that are going to be
3676 * removed.
3677 */
3678 mutex_unlock(&trans->transaction->cache_write_mutex);
3679 mutex_lock(&trans->transaction->cache_write_mutex);
3680 }
3681 mutex_unlock(&trans->transaction->cache_write_mutex);
3682
3683 /*
3684 * go through delayed refs for all the stuff we've just kicked off
3685 * and then loop back (just once)
3686 */
3687 ret = btrfs_run_delayed_refs(trans, root, 0);
3688 if (!ret && loops == 0) {
3689 loops++;
3690 spin_lock(&cur_trans->dirty_bgs_lock);
3691 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3692 /*
3693 * dirty_bgs_lock protects us from concurrent block group
3694 * deletes too (not just cache_write_mutex).
3695 */
3696 if (!list_empty(&dirty)) {
3697 spin_unlock(&cur_trans->dirty_bgs_lock);
3698 goto again;
3699 }
3700 spin_unlock(&cur_trans->dirty_bgs_lock);
3701 }
3702
3703 btrfs_free_path(path);
3704 return ret;
3705 }
3706
3707 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3708 struct btrfs_root *root)
3709 {
3710 struct btrfs_block_group_cache *cache;
3711 struct btrfs_transaction *cur_trans = trans->transaction;
3712 int ret = 0;
3713 int should_put;
3714 struct btrfs_path *path;
3715 struct list_head *io = &cur_trans->io_bgs;
3716 int num_started = 0;
3717
3718 path = btrfs_alloc_path();
3719 if (!path)
3720 return -ENOMEM;
3721
3722 /*
3723 * Even though we are in the critical section of the transaction commit,
3724 * we can still have concurrent tasks adding elements to this
3725 * transaction's list of dirty block groups. These tasks correspond to
3726 * endio free space workers started when writeback finishes for a
3727 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3728 * allocate new block groups as a result of COWing nodes of the root
3729 * tree when updating the free space inode. The writeback for the space
3730 * caches is triggered by an earlier call to
3731 * btrfs_start_dirty_block_groups() and iterations of the following
3732 * loop.
3733 * Also we want to do the cache_save_setup first and then run the
3734 * delayed refs to make sure we have the best chance at doing this all
3735 * in one shot.
3736 */
3737 spin_lock(&cur_trans->dirty_bgs_lock);
3738 while (!list_empty(&cur_trans->dirty_bgs)) {
3739 cache = list_first_entry(&cur_trans->dirty_bgs,
3740 struct btrfs_block_group_cache,
3741 dirty_list);
3742
3743 /*
3744 * this can happen if cache_save_setup re-dirties a block
3745 * group that is already under IO. Just wait for it to
3746 * finish and then do it all again
3747 */
3748 if (!list_empty(&cache->io_list)) {
3749 spin_unlock(&cur_trans->dirty_bgs_lock);
3750 list_del_init(&cache->io_list);
3751 btrfs_wait_cache_io(root, trans, cache,
3752 &cache->io_ctl, path,
3753 cache->key.objectid);
3754 btrfs_put_block_group(cache);
3755 spin_lock(&cur_trans->dirty_bgs_lock);
3756 }
3757
3758 /*
3759 * don't remove from the dirty list until after we've waited
3760 * on any pending IO
3761 */
3762 list_del_init(&cache->dirty_list);
3763 spin_unlock(&cur_trans->dirty_bgs_lock);
3764 should_put = 1;
3765
3766 cache_save_setup(cache, trans, path);
3767
3768 if (!ret)
3769 ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1);
3770
3771 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3772 cache->io_ctl.inode = NULL;
3773 ret = btrfs_write_out_cache(root, trans, cache, path);
3774 if (ret == 0 && cache->io_ctl.inode) {
3775 num_started++;
3776 should_put = 0;
3777 list_add_tail(&cache->io_list, io);
3778 } else {
3779 /*
3780 * if we failed to write the cache, the
3781 * generation will be bad and life goes on
3782 */
3783 ret = 0;
3784 }
3785 }
3786 if (!ret) {
3787 ret = write_one_cache_group(trans, root, path, cache);
3788 /*
3789 * One of the free space endio workers might have
3790 * created a new block group while updating a free space
3791 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3792 * and hasn't released its transaction handle yet, in
3793 * which case the new block group is still attached to
3794 * its transaction handle and its creation has not
3795 * finished yet (no block group item in the extent tree
3796 * yet, etc). If this is the case, wait for all free
3797 * space endio workers to finish and retry. This is a
3798 * a very rare case so no need for a more efficient and
3799 * complex approach.
3800 */
3801 if (ret == -ENOENT) {
3802 wait_event(cur_trans->writer_wait,
3803 atomic_read(&cur_trans->num_writers) == 1);
3804 ret = write_one_cache_group(trans, root, path,
3805 cache);
3806 }
3807 if (ret)
3808 btrfs_abort_transaction(trans, root, ret);
3809 }
3810
3811 /* if its not on the io list, we need to put the block group */
3812 if (should_put)
3813 btrfs_put_block_group(cache);
3814 spin_lock(&cur_trans->dirty_bgs_lock);
3815 }
3816 spin_unlock(&cur_trans->dirty_bgs_lock);
3817
3818 while (!list_empty(io)) {
3819 cache = list_first_entry(io, struct btrfs_block_group_cache,
3820 io_list);
3821 list_del_init(&cache->io_list);
3822 btrfs_wait_cache_io(root, trans, cache,
3823 &cache->io_ctl, path, cache->key.objectid);
3824 btrfs_put_block_group(cache);
3825 }
3826
3827 btrfs_free_path(path);
3828 return ret;
3829 }
3830
3831 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3832 {
3833 struct btrfs_block_group_cache *block_group;
3834 int readonly = 0;
3835
3836 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3837 if (!block_group || block_group->ro)
3838 readonly = 1;
3839 if (block_group)
3840 btrfs_put_block_group(block_group);
3841 return readonly;
3842 }
3843
3844 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3845 {
3846 struct btrfs_block_group_cache *bg;
3847 bool ret = true;
3848
3849 bg = btrfs_lookup_block_group(fs_info, bytenr);
3850 if (!bg)
3851 return false;
3852
3853 spin_lock(&bg->lock);
3854 if (bg->ro)
3855 ret = false;
3856 else
3857 atomic_inc(&bg->nocow_writers);
3858 spin_unlock(&bg->lock);
3859
3860 /* no put on block group, done by btrfs_dec_nocow_writers */
3861 if (!ret)
3862 btrfs_put_block_group(bg);
3863
3864 return ret;
3865
3866 }
3867
3868 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3869 {
3870 struct btrfs_block_group_cache *bg;
3871
3872 bg = btrfs_lookup_block_group(fs_info, bytenr);
3873 ASSERT(bg);
3874 if (atomic_dec_and_test(&bg->nocow_writers))
3875 wake_up_atomic_t(&bg->nocow_writers);
3876 /*
3877 * Once for our lookup and once for the lookup done by a previous call
3878 * to btrfs_inc_nocow_writers()
3879 */
3880 btrfs_put_block_group(bg);
3881 btrfs_put_block_group(bg);
3882 }
3883
3884 static int btrfs_wait_nocow_writers_atomic_t(atomic_t *a)
3885 {
3886 schedule();
3887 return 0;
3888 }
3889
3890 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
3891 {
3892 wait_on_atomic_t(&bg->nocow_writers,
3893 btrfs_wait_nocow_writers_atomic_t,
3894 TASK_UNINTERRUPTIBLE);
3895 }
3896
3897 static const char *alloc_name(u64 flags)
3898 {
3899 switch (flags) {
3900 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3901 return "mixed";
3902 case BTRFS_BLOCK_GROUP_METADATA:
3903 return "metadata";
3904 case BTRFS_BLOCK_GROUP_DATA:
3905 return "data";
3906 case BTRFS_BLOCK_GROUP_SYSTEM:
3907 return "system";
3908 default:
3909 WARN_ON(1);
3910 return "invalid-combination";
3911 };
3912 }
3913
3914 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3915 u64 total_bytes, u64 bytes_used,
3916 struct btrfs_space_info **space_info)
3917 {
3918 struct btrfs_space_info *found;
3919 int i;
3920 int factor;
3921 int ret;
3922
3923 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3924 BTRFS_BLOCK_GROUP_RAID10))
3925 factor = 2;
3926 else
3927 factor = 1;
3928
3929 found = __find_space_info(info, flags);
3930 if (found) {
3931 spin_lock(&found->lock);
3932 found->total_bytes += total_bytes;
3933 found->disk_total += total_bytes * factor;
3934 found->bytes_used += bytes_used;
3935 found->disk_used += bytes_used * factor;
3936 if (total_bytes > 0)
3937 found->full = 0;
3938 spin_unlock(&found->lock);
3939 *space_info = found;
3940 return 0;
3941 }
3942 found = kzalloc(sizeof(*found), GFP_NOFS);
3943 if (!found)
3944 return -ENOMEM;
3945
3946 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3947 if (ret) {
3948 kfree(found);
3949 return ret;
3950 }
3951
3952 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3953 INIT_LIST_HEAD(&found->block_groups[i]);
3954 init_rwsem(&found->groups_sem);
3955 spin_lock_init(&found->lock);
3956 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3957 found->total_bytes = total_bytes;
3958 found->disk_total = total_bytes * factor;
3959 found->bytes_used = bytes_used;
3960 found->disk_used = bytes_used * factor;
3961 found->bytes_pinned = 0;
3962 found->bytes_reserved = 0;
3963 found->bytes_readonly = 0;
3964 found->bytes_may_use = 0;
3965 found->full = 0;
3966 found->max_extent_size = 0;
3967 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3968 found->chunk_alloc = 0;
3969 found->flush = 0;
3970 init_waitqueue_head(&found->wait);
3971 INIT_LIST_HEAD(&found->ro_bgs);
3972
3973 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3974 info->space_info_kobj, "%s",
3975 alloc_name(found->flags));
3976 if (ret) {
3977 kfree(found);
3978 return ret;
3979 }
3980
3981 *space_info = found;
3982 list_add_rcu(&found->list, &info->space_info);
3983 if (flags & BTRFS_BLOCK_GROUP_DATA)
3984 info->data_sinfo = found;
3985
3986 return ret;
3987 }
3988
3989 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3990 {
3991 u64 extra_flags = chunk_to_extended(flags) &
3992 BTRFS_EXTENDED_PROFILE_MASK;
3993
3994 write_seqlock(&fs_info->profiles_lock);
3995 if (flags & BTRFS_BLOCK_GROUP_DATA)
3996 fs_info->avail_data_alloc_bits |= extra_flags;
3997 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3998 fs_info->avail_metadata_alloc_bits |= extra_flags;
3999 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4000 fs_info->avail_system_alloc_bits |= extra_flags;
4001 write_sequnlock(&fs_info->profiles_lock);
4002 }
4003
4004 /*
4005 * returns target flags in extended format or 0 if restripe for this
4006 * chunk_type is not in progress
4007 *
4008 * should be called with either volume_mutex or balance_lock held
4009 */
4010 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
4011 {
4012 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4013 u64 target = 0;
4014
4015 if (!bctl)
4016 return 0;
4017
4018 if (flags & BTRFS_BLOCK_GROUP_DATA &&
4019 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4020 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
4021 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
4022 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4023 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
4024 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
4025 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4026 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
4027 }
4028
4029 return target;
4030 }
4031
4032 /*
4033 * @flags: available profiles in extended format (see ctree.h)
4034 *
4035 * Returns reduced profile in chunk format. If profile changing is in
4036 * progress (either running or paused) picks the target profile (if it's
4037 * already available), otherwise falls back to plain reducing.
4038 */
4039 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
4040 {
4041 u64 num_devices = root->fs_info->fs_devices->rw_devices;
4042 u64 target;
4043 u64 raid_type;
4044 u64 allowed = 0;
4045
4046 /*
4047 * see if restripe for this chunk_type is in progress, if so
4048 * try to reduce to the target profile
4049 */
4050 spin_lock(&root->fs_info->balance_lock);
4051 target = get_restripe_target(root->fs_info, flags);
4052 if (target) {
4053 /* pick target profile only if it's already available */
4054 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
4055 spin_unlock(&root->fs_info->balance_lock);
4056 return extended_to_chunk(target);
4057 }
4058 }
4059 spin_unlock(&root->fs_info->balance_lock);
4060
4061 /* First, mask out the RAID levels which aren't possible */
4062 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4063 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
4064 allowed |= btrfs_raid_group[raid_type];
4065 }
4066 allowed &= flags;
4067
4068 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
4069 allowed = BTRFS_BLOCK_GROUP_RAID6;
4070 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4071 allowed = BTRFS_BLOCK_GROUP_RAID5;
4072 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4073 allowed = BTRFS_BLOCK_GROUP_RAID10;
4074 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4075 allowed = BTRFS_BLOCK_GROUP_RAID1;
4076 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4077 allowed = BTRFS_BLOCK_GROUP_RAID0;
4078
4079 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4080
4081 return extended_to_chunk(flags | allowed);
4082 }
4083
4084 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
4085 {
4086 unsigned seq;
4087 u64 flags;
4088
4089 do {
4090 flags = orig_flags;
4091 seq = read_seqbegin(&root->fs_info->profiles_lock);
4092
4093 if (flags & BTRFS_BLOCK_GROUP_DATA)
4094 flags |= root->fs_info->avail_data_alloc_bits;
4095 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4096 flags |= root->fs_info->avail_system_alloc_bits;
4097 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4098 flags |= root->fs_info->avail_metadata_alloc_bits;
4099 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
4100
4101 return btrfs_reduce_alloc_profile(root, flags);
4102 }
4103
4104 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
4105 {
4106 u64 flags;
4107 u64 ret;
4108
4109 if (data)
4110 flags = BTRFS_BLOCK_GROUP_DATA;
4111 else if (root == root->fs_info->chunk_root)
4112 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4113 else
4114 flags = BTRFS_BLOCK_GROUP_METADATA;
4115
4116 ret = get_alloc_profile(root, flags);
4117 return ret;
4118 }
4119
4120 int btrfs_alloc_data_chunk_ondemand(struct inode *inode, u64 bytes)
4121 {
4122 struct btrfs_space_info *data_sinfo;
4123 struct btrfs_root *root = BTRFS_I(inode)->root;
4124 struct btrfs_fs_info *fs_info = root->fs_info;
4125 u64 used;
4126 int ret = 0;
4127 int need_commit = 2;
4128 int have_pinned_space;
4129
4130 /* make sure bytes are sectorsize aligned */
4131 bytes = ALIGN(bytes, root->sectorsize);
4132
4133 if (btrfs_is_free_space_inode(inode)) {
4134 need_commit = 0;
4135 ASSERT(current->journal_info);
4136 }
4137
4138 data_sinfo = fs_info->data_sinfo;
4139 if (!data_sinfo)
4140 goto alloc;
4141
4142 again:
4143 /* make sure we have enough space to handle the data first */
4144 spin_lock(&data_sinfo->lock);
4145 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
4146 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
4147 data_sinfo->bytes_may_use;
4148
4149 if (used + bytes > data_sinfo->total_bytes) {
4150 struct btrfs_trans_handle *trans;
4151
4152 /*
4153 * if we don't have enough free bytes in this space then we need
4154 * to alloc a new chunk.
4155 */
4156 if (!data_sinfo->full) {
4157 u64 alloc_target;
4158
4159 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4160 spin_unlock(&data_sinfo->lock);
4161 alloc:
4162 alloc_target = btrfs_get_alloc_profile(root, 1);
4163 /*
4164 * It is ugly that we don't call nolock join
4165 * transaction for the free space inode case here.
4166 * But it is safe because we only do the data space
4167 * reservation for the free space cache in the
4168 * transaction context, the common join transaction
4169 * just increase the counter of the current transaction
4170 * handler, doesn't try to acquire the trans_lock of
4171 * the fs.
4172 */
4173 trans = btrfs_join_transaction(root);
4174 if (IS_ERR(trans))
4175 return PTR_ERR(trans);
4176
4177 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4178 alloc_target,
4179 CHUNK_ALLOC_NO_FORCE);
4180 btrfs_end_transaction(trans, root);
4181 if (ret < 0) {
4182 if (ret != -ENOSPC)
4183 return ret;
4184 else {
4185 have_pinned_space = 1;
4186 goto commit_trans;
4187 }
4188 }
4189
4190 if (!data_sinfo)
4191 data_sinfo = fs_info->data_sinfo;
4192
4193 goto again;
4194 }
4195
4196 /*
4197 * If we don't have enough pinned space to deal with this
4198 * allocation, and no removed chunk in current transaction,
4199 * don't bother committing the transaction.
4200 */
4201 have_pinned_space = percpu_counter_compare(
4202 &data_sinfo->total_bytes_pinned,
4203 used + bytes - data_sinfo->total_bytes);
4204 spin_unlock(&data_sinfo->lock);
4205
4206 /* commit the current transaction and try again */
4207 commit_trans:
4208 if (need_commit &&
4209 !atomic_read(&root->fs_info->open_ioctl_trans)) {
4210 need_commit--;
4211
4212 if (need_commit > 0) {
4213 btrfs_start_delalloc_roots(fs_info, 0, -1);
4214 btrfs_wait_ordered_roots(fs_info, -1, 0, (u64)-1);
4215 }
4216
4217 trans = btrfs_join_transaction(root);
4218 if (IS_ERR(trans))
4219 return PTR_ERR(trans);
4220 if (have_pinned_space >= 0 ||
4221 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4222 &trans->transaction->flags) ||
4223 need_commit > 0) {
4224 ret = btrfs_commit_transaction(trans, root);
4225 if (ret)
4226 return ret;
4227 /*
4228 * The cleaner kthread might still be doing iput
4229 * operations. Wait for it to finish so that
4230 * more space is released.
4231 */
4232 mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex);
4233 mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex);
4234 goto again;
4235 } else {
4236 btrfs_end_transaction(trans, root);
4237 }
4238 }
4239
4240 trace_btrfs_space_reservation(root->fs_info,
4241 "space_info:enospc",
4242 data_sinfo->flags, bytes, 1);
4243 return -ENOSPC;
4244 }
4245 data_sinfo->bytes_may_use += bytes;
4246 trace_btrfs_space_reservation(root->fs_info, "space_info",
4247 data_sinfo->flags, bytes, 1);
4248 spin_unlock(&data_sinfo->lock);
4249
4250 return ret;
4251 }
4252
4253 /*
4254 * New check_data_free_space() with ability for precious data reservation
4255 * Will replace old btrfs_check_data_free_space(), but for patch split,
4256 * add a new function first and then replace it.
4257 */
4258 int btrfs_check_data_free_space(struct inode *inode, u64 start, u64 len)
4259 {
4260 struct btrfs_root *root = BTRFS_I(inode)->root;
4261 int ret;
4262
4263 /* align the range */
4264 len = round_up(start + len, root->sectorsize) -
4265 round_down(start, root->sectorsize);
4266 start = round_down(start, root->sectorsize);
4267
4268 ret = btrfs_alloc_data_chunk_ondemand(inode, len);
4269 if (ret < 0)
4270 return ret;
4271
4272 /*
4273 * Use new btrfs_qgroup_reserve_data to reserve precious data space
4274 *
4275 * TODO: Find a good method to avoid reserve data space for NOCOW
4276 * range, but don't impact performance on quota disable case.
4277 */
4278 ret = btrfs_qgroup_reserve_data(inode, start, len);
4279 return ret;
4280 }
4281
4282 /*
4283 * Called if we need to clear a data reservation for this inode
4284 * Normally in a error case.
4285 *
4286 * This one will *NOT* use accurate qgroup reserved space API, just for case
4287 * which we can't sleep and is sure it won't affect qgroup reserved space.
4288 * Like clear_bit_hook().
4289 */
4290 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4291 u64 len)
4292 {
4293 struct btrfs_root *root = BTRFS_I(inode)->root;
4294 struct btrfs_space_info *data_sinfo;
4295
4296 /* Make sure the range is aligned to sectorsize */
4297 len = round_up(start + len, root->sectorsize) -
4298 round_down(start, root->sectorsize);
4299 start = round_down(start, root->sectorsize);
4300
4301 data_sinfo = root->fs_info->data_sinfo;
4302 spin_lock(&data_sinfo->lock);
4303 if (WARN_ON(data_sinfo->bytes_may_use < len))
4304 data_sinfo->bytes_may_use = 0;
4305 else
4306 data_sinfo->bytes_may_use -= len;
4307 trace_btrfs_space_reservation(root->fs_info, "space_info",
4308 data_sinfo->flags, len, 0);
4309 spin_unlock(&data_sinfo->lock);
4310 }
4311
4312 /*
4313 * Called if we need to clear a data reservation for this inode
4314 * Normally in a error case.
4315 *
4316 * This one will handle the per-inode data rsv map for accurate reserved
4317 * space framework.
4318 */
4319 void btrfs_free_reserved_data_space(struct inode *inode, u64 start, u64 len)
4320 {
4321 btrfs_free_reserved_data_space_noquota(inode, start, len);
4322 btrfs_qgroup_free_data(inode, start, len);
4323 }
4324
4325 static void force_metadata_allocation(struct btrfs_fs_info *info)
4326 {
4327 struct list_head *head = &info->space_info;
4328 struct btrfs_space_info *found;
4329
4330 rcu_read_lock();
4331 list_for_each_entry_rcu(found, head, list) {
4332 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4333 found->force_alloc = CHUNK_ALLOC_FORCE;
4334 }
4335 rcu_read_unlock();
4336 }
4337
4338 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4339 {
4340 return (global->size << 1);
4341 }
4342
4343 static int should_alloc_chunk(struct btrfs_root *root,
4344 struct btrfs_space_info *sinfo, int force)
4345 {
4346 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4347 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
4348 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
4349 u64 thresh;
4350
4351 if (force == CHUNK_ALLOC_FORCE)
4352 return 1;
4353
4354 /*
4355 * We need to take into account the global rsv because for all intents
4356 * and purposes it's used space. Don't worry about locking the
4357 * global_rsv, it doesn't change except when the transaction commits.
4358 */
4359 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4360 num_allocated += calc_global_rsv_need_space(global_rsv);
4361
4362 /*
4363 * in limited mode, we want to have some free space up to
4364 * about 1% of the FS size.
4365 */
4366 if (force == CHUNK_ALLOC_LIMITED) {
4367 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
4368 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4369
4370 if (num_bytes - num_allocated < thresh)
4371 return 1;
4372 }
4373
4374 if (num_allocated + SZ_2M < div_factor(num_bytes, 8))
4375 return 0;
4376 return 1;
4377 }
4378
4379 static u64 get_profile_num_devs(struct btrfs_root *root, u64 type)
4380 {
4381 u64 num_dev;
4382
4383 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4384 BTRFS_BLOCK_GROUP_RAID0 |
4385 BTRFS_BLOCK_GROUP_RAID5 |
4386 BTRFS_BLOCK_GROUP_RAID6))
4387 num_dev = root->fs_info->fs_devices->rw_devices;
4388 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4389 num_dev = 2;
4390 else
4391 num_dev = 1; /* DUP or single */
4392
4393 return num_dev;
4394 }
4395
4396 /*
4397 * If @is_allocation is true, reserve space in the system space info necessary
4398 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4399 * removing a chunk.
4400 */
4401 void check_system_chunk(struct btrfs_trans_handle *trans,
4402 struct btrfs_root *root,
4403 u64 type)
4404 {
4405 struct btrfs_space_info *info;
4406 u64 left;
4407 u64 thresh;
4408 int ret = 0;
4409 u64 num_devs;
4410
4411 /*
4412 * Needed because we can end up allocating a system chunk and for an
4413 * atomic and race free space reservation in the chunk block reserve.
4414 */
4415 ASSERT(mutex_is_locked(&root->fs_info->chunk_mutex));
4416
4417 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4418 spin_lock(&info->lock);
4419 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
4420 info->bytes_reserved - info->bytes_readonly -
4421 info->bytes_may_use;
4422 spin_unlock(&info->lock);
4423
4424 num_devs = get_profile_num_devs(root, type);
4425
4426 /* num_devs device items to update and 1 chunk item to add or remove */
4427 thresh = btrfs_calc_trunc_metadata_size(root, num_devs) +
4428 btrfs_calc_trans_metadata_size(root, 1);
4429
4430 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
4431 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
4432 left, thresh, type);
4433 dump_space_info(info, 0, 0);
4434 }
4435
4436 if (left < thresh) {
4437 u64 flags;
4438
4439 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
4440 /*
4441 * Ignore failure to create system chunk. We might end up not
4442 * needing it, as we might not need to COW all nodes/leafs from
4443 * the paths we visit in the chunk tree (they were already COWed
4444 * or created in the current transaction for example).
4445 */
4446 ret = btrfs_alloc_chunk(trans, root, flags);
4447 }
4448
4449 if (!ret) {
4450 ret = btrfs_block_rsv_add(root->fs_info->chunk_root,
4451 &root->fs_info->chunk_block_rsv,
4452 thresh, BTRFS_RESERVE_NO_FLUSH);
4453 if (!ret)
4454 trans->chunk_bytes_reserved += thresh;
4455 }
4456 }
4457
4458 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4459 struct btrfs_root *extent_root, u64 flags, int force)
4460 {
4461 struct btrfs_space_info *space_info;
4462 struct btrfs_fs_info *fs_info = extent_root->fs_info;
4463 int wait_for_alloc = 0;
4464 int ret = 0;
4465
4466 /* Don't re-enter if we're already allocating a chunk */
4467 if (trans->allocating_chunk)
4468 return -ENOSPC;
4469
4470 space_info = __find_space_info(extent_root->fs_info, flags);
4471 if (!space_info) {
4472 ret = update_space_info(extent_root->fs_info, flags,
4473 0, 0, &space_info);
4474 BUG_ON(ret); /* -ENOMEM */
4475 }
4476 BUG_ON(!space_info); /* Logic error */
4477
4478 again:
4479 spin_lock(&space_info->lock);
4480 if (force < space_info->force_alloc)
4481 force = space_info->force_alloc;
4482 if (space_info->full) {
4483 if (should_alloc_chunk(extent_root, space_info, force))
4484 ret = -ENOSPC;
4485 else
4486 ret = 0;
4487 spin_unlock(&space_info->lock);
4488 return ret;
4489 }
4490
4491 if (!should_alloc_chunk(extent_root, space_info, force)) {
4492 spin_unlock(&space_info->lock);
4493 return 0;
4494 } else if (space_info->chunk_alloc) {
4495 wait_for_alloc = 1;
4496 } else {
4497 space_info->chunk_alloc = 1;
4498 }
4499
4500 spin_unlock(&space_info->lock);
4501
4502 mutex_lock(&fs_info->chunk_mutex);
4503
4504 /*
4505 * The chunk_mutex is held throughout the entirety of a chunk
4506 * allocation, so once we've acquired the chunk_mutex we know that the
4507 * other guy is done and we need to recheck and see if we should
4508 * allocate.
4509 */
4510 if (wait_for_alloc) {
4511 mutex_unlock(&fs_info->chunk_mutex);
4512 wait_for_alloc = 0;
4513 goto again;
4514 }
4515
4516 trans->allocating_chunk = true;
4517
4518 /*
4519 * If we have mixed data/metadata chunks we want to make sure we keep
4520 * allocating mixed chunks instead of individual chunks.
4521 */
4522 if (btrfs_mixed_space_info(space_info))
4523 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4524
4525 /*
4526 * if we're doing a data chunk, go ahead and make sure that
4527 * we keep a reasonable number of metadata chunks allocated in the
4528 * FS as well.
4529 */
4530 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4531 fs_info->data_chunk_allocations++;
4532 if (!(fs_info->data_chunk_allocations %
4533 fs_info->metadata_ratio))
4534 force_metadata_allocation(fs_info);
4535 }
4536
4537 /*
4538 * Check if we have enough space in SYSTEM chunk because we may need
4539 * to update devices.
4540 */
4541 check_system_chunk(trans, extent_root, flags);
4542
4543 ret = btrfs_alloc_chunk(trans, extent_root, flags);
4544 trans->allocating_chunk = false;
4545
4546 spin_lock(&space_info->lock);
4547 if (ret < 0 && ret != -ENOSPC)
4548 goto out;
4549 if (ret)
4550 space_info->full = 1;
4551 else
4552 ret = 1;
4553
4554 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4555 out:
4556 space_info->chunk_alloc = 0;
4557 spin_unlock(&space_info->lock);
4558 mutex_unlock(&fs_info->chunk_mutex);
4559 /*
4560 * When we allocate a new chunk we reserve space in the chunk block
4561 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4562 * add new nodes/leafs to it if we end up needing to do it when
4563 * inserting the chunk item and updating device items as part of the
4564 * second phase of chunk allocation, performed by
4565 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4566 * large number of new block groups to create in our transaction
4567 * handle's new_bgs list to avoid exhausting the chunk block reserve
4568 * in extreme cases - like having a single transaction create many new
4569 * block groups when starting to write out the free space caches of all
4570 * the block groups that were made dirty during the lifetime of the
4571 * transaction.
4572 */
4573 if (trans->can_flush_pending_bgs &&
4574 trans->chunk_bytes_reserved >= (u64)SZ_2M) {
4575 btrfs_create_pending_block_groups(trans, trans->root);
4576 btrfs_trans_release_chunk_metadata(trans);
4577 }
4578 return ret;
4579 }
4580
4581 static int can_overcommit(struct btrfs_root *root,
4582 struct btrfs_space_info *space_info, u64 bytes,
4583 enum btrfs_reserve_flush_enum flush)
4584 {
4585 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4586 u64 profile = btrfs_get_alloc_profile(root, 0);
4587 u64 space_size;
4588 u64 avail;
4589 u64 used;
4590
4591 used = space_info->bytes_used + space_info->bytes_reserved +
4592 space_info->bytes_pinned + space_info->bytes_readonly;
4593
4594 /*
4595 * We only want to allow over committing if we have lots of actual space
4596 * free, but if we don't have enough space to handle the global reserve
4597 * space then we could end up having a real enospc problem when trying
4598 * to allocate a chunk or some other such important allocation.
4599 */
4600 spin_lock(&global_rsv->lock);
4601 space_size = calc_global_rsv_need_space(global_rsv);
4602 spin_unlock(&global_rsv->lock);
4603 if (used + space_size >= space_info->total_bytes)
4604 return 0;
4605
4606 used += space_info->bytes_may_use;
4607
4608 spin_lock(&root->fs_info->free_chunk_lock);
4609 avail = root->fs_info->free_chunk_space;
4610 spin_unlock(&root->fs_info->free_chunk_lock);
4611
4612 /*
4613 * If we have dup, raid1 or raid10 then only half of the free
4614 * space is actually useable. For raid56, the space info used
4615 * doesn't include the parity drive, so we don't have to
4616 * change the math
4617 */
4618 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4619 BTRFS_BLOCK_GROUP_RAID1 |
4620 BTRFS_BLOCK_GROUP_RAID10))
4621 avail >>= 1;
4622
4623 /*
4624 * If we aren't flushing all things, let us overcommit up to
4625 * 1/2th of the space. If we can flush, don't let us overcommit
4626 * too much, let it overcommit up to 1/8 of the space.
4627 */
4628 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4629 avail >>= 3;
4630 else
4631 avail >>= 1;
4632
4633 if (used + bytes < space_info->total_bytes + avail)
4634 return 1;
4635 return 0;
4636 }
4637
4638 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4639 unsigned long nr_pages, int nr_items)
4640 {
4641 struct super_block *sb = root->fs_info->sb;
4642
4643 if (down_read_trylock(&sb->s_umount)) {
4644 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4645 up_read(&sb->s_umount);
4646 } else {
4647 /*
4648 * We needn't worry the filesystem going from r/w to r/o though
4649 * we don't acquire ->s_umount mutex, because the filesystem
4650 * should guarantee the delalloc inodes list be empty after
4651 * the filesystem is readonly(all dirty pages are written to
4652 * the disk).
4653 */
4654 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4655 if (!current->journal_info)
4656 btrfs_wait_ordered_roots(root->fs_info, nr_items,
4657 0, (u64)-1);
4658 }
4659 }
4660
4661 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4662 {
4663 u64 bytes;
4664 int nr;
4665
4666 bytes = btrfs_calc_trans_metadata_size(root, 1);
4667 nr = (int)div64_u64(to_reclaim, bytes);
4668 if (!nr)
4669 nr = 1;
4670 return nr;
4671 }
4672
4673 #define EXTENT_SIZE_PER_ITEM SZ_256K
4674
4675 /*
4676 * shrink metadata reservation for delalloc
4677 */
4678 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4679 bool wait_ordered)
4680 {
4681 struct btrfs_block_rsv *block_rsv;
4682 struct btrfs_space_info *space_info;
4683 struct btrfs_trans_handle *trans;
4684 u64 delalloc_bytes;
4685 u64 max_reclaim;
4686 long time_left;
4687 unsigned long nr_pages;
4688 int loops;
4689 int items;
4690 enum btrfs_reserve_flush_enum flush;
4691
4692 /* Calc the number of the pages we need flush for space reservation */
4693 items = calc_reclaim_items_nr(root, to_reclaim);
4694 to_reclaim = (u64)items * EXTENT_SIZE_PER_ITEM;
4695
4696 trans = (struct btrfs_trans_handle *)current->journal_info;
4697 block_rsv = &root->fs_info->delalloc_block_rsv;
4698 space_info = block_rsv->space_info;
4699
4700 delalloc_bytes = percpu_counter_sum_positive(
4701 &root->fs_info->delalloc_bytes);
4702 if (delalloc_bytes == 0) {
4703 if (trans)
4704 return;
4705 if (wait_ordered)
4706 btrfs_wait_ordered_roots(root->fs_info, items,
4707 0, (u64)-1);
4708 return;
4709 }
4710
4711 loops = 0;
4712 while (delalloc_bytes && loops < 3) {
4713 max_reclaim = min(delalloc_bytes, to_reclaim);
4714 nr_pages = max_reclaim >> PAGE_SHIFT;
4715 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4716 /*
4717 * We need to wait for the async pages to actually start before
4718 * we do anything.
4719 */
4720 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4721 if (!max_reclaim)
4722 goto skip_async;
4723
4724 if (max_reclaim <= nr_pages)
4725 max_reclaim = 0;
4726 else
4727 max_reclaim -= nr_pages;
4728
4729 wait_event(root->fs_info->async_submit_wait,
4730 atomic_read(&root->fs_info->async_delalloc_pages) <=
4731 (int)max_reclaim);
4732 skip_async:
4733 if (!trans)
4734 flush = BTRFS_RESERVE_FLUSH_ALL;
4735 else
4736 flush = BTRFS_RESERVE_NO_FLUSH;
4737 spin_lock(&space_info->lock);
4738 if (can_overcommit(root, space_info, orig, flush)) {
4739 spin_unlock(&space_info->lock);
4740 break;
4741 }
4742 spin_unlock(&space_info->lock);
4743
4744 loops++;
4745 if (wait_ordered && !trans) {
4746 btrfs_wait_ordered_roots(root->fs_info, items,
4747 0, (u64)-1);
4748 } else {
4749 time_left = schedule_timeout_killable(1);
4750 if (time_left)
4751 break;
4752 }
4753 delalloc_bytes = percpu_counter_sum_positive(
4754 &root->fs_info->delalloc_bytes);
4755 }
4756 }
4757
4758 /**
4759 * maybe_commit_transaction - possibly commit the transaction if its ok to
4760 * @root - the root we're allocating for
4761 * @bytes - the number of bytes we want to reserve
4762 * @force - force the commit
4763 *
4764 * This will check to make sure that committing the transaction will actually
4765 * get us somewhere and then commit the transaction if it does. Otherwise it
4766 * will return -ENOSPC.
4767 */
4768 static int may_commit_transaction(struct btrfs_root *root,
4769 struct btrfs_space_info *space_info,
4770 u64 bytes, int force)
4771 {
4772 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4773 struct btrfs_trans_handle *trans;
4774
4775 trans = (struct btrfs_trans_handle *)current->journal_info;
4776 if (trans)
4777 return -EAGAIN;
4778
4779 if (force)
4780 goto commit;
4781
4782 /* See if there is enough pinned space to make this reservation */
4783 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4784 bytes) >= 0)
4785 goto commit;
4786
4787 /*
4788 * See if there is some space in the delayed insertion reservation for
4789 * this reservation.
4790 */
4791 if (space_info != delayed_rsv->space_info)
4792 return -ENOSPC;
4793
4794 spin_lock(&delayed_rsv->lock);
4795 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4796 bytes - delayed_rsv->size) >= 0) {
4797 spin_unlock(&delayed_rsv->lock);
4798 return -ENOSPC;
4799 }
4800 spin_unlock(&delayed_rsv->lock);
4801
4802 commit:
4803 trans = btrfs_join_transaction(root);
4804 if (IS_ERR(trans))
4805 return -ENOSPC;
4806
4807 return btrfs_commit_transaction(trans, root);
4808 }
4809
4810 enum flush_state {
4811 FLUSH_DELAYED_ITEMS_NR = 1,
4812 FLUSH_DELAYED_ITEMS = 2,
4813 FLUSH_DELALLOC = 3,
4814 FLUSH_DELALLOC_WAIT = 4,
4815 ALLOC_CHUNK = 5,
4816 COMMIT_TRANS = 6,
4817 };
4818
4819 static int flush_space(struct btrfs_root *root,
4820 struct btrfs_space_info *space_info, u64 num_bytes,
4821 u64 orig_bytes, int state)
4822 {
4823 struct btrfs_trans_handle *trans;
4824 int nr;
4825 int ret = 0;
4826
4827 switch (state) {
4828 case FLUSH_DELAYED_ITEMS_NR:
4829 case FLUSH_DELAYED_ITEMS:
4830 if (state == FLUSH_DELAYED_ITEMS_NR)
4831 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4832 else
4833 nr = -1;
4834
4835 trans = btrfs_join_transaction(root);
4836 if (IS_ERR(trans)) {
4837 ret = PTR_ERR(trans);
4838 break;
4839 }
4840 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4841 btrfs_end_transaction(trans, root);
4842 break;
4843 case FLUSH_DELALLOC:
4844 case FLUSH_DELALLOC_WAIT:
4845 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4846 state == FLUSH_DELALLOC_WAIT);
4847 break;
4848 case ALLOC_CHUNK:
4849 trans = btrfs_join_transaction(root);
4850 if (IS_ERR(trans)) {
4851 ret = PTR_ERR(trans);
4852 break;
4853 }
4854 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4855 btrfs_get_alloc_profile(root, 0),
4856 CHUNK_ALLOC_NO_FORCE);
4857 btrfs_end_transaction(trans, root);
4858 if (ret == -ENOSPC)
4859 ret = 0;
4860 break;
4861 case COMMIT_TRANS:
4862 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4863 break;
4864 default:
4865 ret = -ENOSPC;
4866 break;
4867 }
4868
4869 return ret;
4870 }
4871
4872 static inline u64
4873 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4874 struct btrfs_space_info *space_info)
4875 {
4876 u64 used;
4877 u64 expected;
4878 u64 to_reclaim;
4879
4880 to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
4881 spin_lock(&space_info->lock);
4882 if (can_overcommit(root, space_info, to_reclaim,
4883 BTRFS_RESERVE_FLUSH_ALL)) {
4884 to_reclaim = 0;
4885 goto out;
4886 }
4887
4888 used = space_info->bytes_used + space_info->bytes_reserved +
4889 space_info->bytes_pinned + space_info->bytes_readonly +
4890 space_info->bytes_may_use;
4891 if (can_overcommit(root, space_info, SZ_1M, BTRFS_RESERVE_FLUSH_ALL))
4892 expected = div_factor_fine(space_info->total_bytes, 95);
4893 else
4894 expected = div_factor_fine(space_info->total_bytes, 90);
4895
4896 if (used > expected)
4897 to_reclaim = used - expected;
4898 else
4899 to_reclaim = 0;
4900 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4901 space_info->bytes_reserved);
4902 out:
4903 spin_unlock(&space_info->lock);
4904
4905 return to_reclaim;
4906 }
4907
4908 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4909 struct btrfs_fs_info *fs_info, u64 used)
4910 {
4911 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4912
4913 /* If we're just plain full then async reclaim just slows us down. */
4914 if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
4915 return 0;
4916
4917 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
4918 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4919 }
4920
4921 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4922 struct btrfs_fs_info *fs_info,
4923 int flush_state)
4924 {
4925 u64 used;
4926
4927 spin_lock(&space_info->lock);
4928 /*
4929 * We run out of space and have not got any free space via flush_space,
4930 * so don't bother doing async reclaim.
4931 */
4932 if (flush_state > COMMIT_TRANS && space_info->full) {
4933 spin_unlock(&space_info->lock);
4934 return 0;
4935 }
4936
4937 used = space_info->bytes_used + space_info->bytes_reserved +
4938 space_info->bytes_pinned + space_info->bytes_readonly +
4939 space_info->bytes_may_use;
4940 if (need_do_async_reclaim(space_info, fs_info, used)) {
4941 spin_unlock(&space_info->lock);
4942 return 1;
4943 }
4944 spin_unlock(&space_info->lock);
4945
4946 return 0;
4947 }
4948
4949 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4950 {
4951 struct btrfs_fs_info *fs_info;
4952 struct btrfs_space_info *space_info;
4953 u64 to_reclaim;
4954 int flush_state;
4955
4956 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4957 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4958
4959 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4960 space_info);
4961 if (!to_reclaim)
4962 return;
4963
4964 flush_state = FLUSH_DELAYED_ITEMS_NR;
4965 do {
4966 flush_space(fs_info->fs_root, space_info, to_reclaim,
4967 to_reclaim, flush_state);
4968 flush_state++;
4969 if (!btrfs_need_do_async_reclaim(space_info, fs_info,
4970 flush_state))
4971 return;
4972 } while (flush_state < COMMIT_TRANS);
4973 }
4974
4975 void btrfs_init_async_reclaim_work(struct work_struct *work)
4976 {
4977 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4978 }
4979
4980 /**
4981 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4982 * @root - the root we're allocating for
4983 * @block_rsv - the block_rsv we're allocating for
4984 * @orig_bytes - the number of bytes we want
4985 * @flush - whether or not we can flush to make our reservation
4986 *
4987 * This will reserve orig_bytes number of bytes from the space info associated
4988 * with the block_rsv. If there is not enough space it will make an attempt to
4989 * flush out space to make room. It will do this by flushing delalloc if
4990 * possible or committing the transaction. If flush is 0 then no attempts to
4991 * regain reservations will be made and this will fail if there is not enough
4992 * space already.
4993 */
4994 static int reserve_metadata_bytes(struct btrfs_root *root,
4995 struct btrfs_block_rsv *block_rsv,
4996 u64 orig_bytes,
4997 enum btrfs_reserve_flush_enum flush)
4998 {
4999 struct btrfs_space_info *space_info = block_rsv->space_info;
5000 u64 used;
5001 u64 num_bytes = orig_bytes;
5002 int flush_state = FLUSH_DELAYED_ITEMS_NR;
5003 int ret = 0;
5004 bool flushing = false;
5005
5006 again:
5007 ret = 0;
5008 spin_lock(&space_info->lock);
5009 /*
5010 * We only want to wait if somebody other than us is flushing and we
5011 * are actually allowed to flush all things.
5012 */
5013 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
5014 space_info->flush) {
5015 spin_unlock(&space_info->lock);
5016 /*
5017 * If we have a trans handle we can't wait because the flusher
5018 * may have to commit the transaction, which would mean we would
5019 * deadlock since we are waiting for the flusher to finish, but
5020 * hold the current transaction open.
5021 */
5022 if (current->journal_info)
5023 return -EAGAIN;
5024 ret = wait_event_killable(space_info->wait, !space_info->flush);
5025 /* Must have been killed, return */
5026 if (ret)
5027 return -EINTR;
5028
5029 spin_lock(&space_info->lock);
5030 }
5031
5032 ret = -ENOSPC;
5033 used = space_info->bytes_used + space_info->bytes_reserved +
5034 space_info->bytes_pinned + space_info->bytes_readonly +
5035 space_info->bytes_may_use;
5036
5037 /*
5038 * The idea here is that we've not already over-reserved the block group
5039 * then we can go ahead and save our reservation first and then start
5040 * flushing if we need to. Otherwise if we've already overcommitted
5041 * lets start flushing stuff first and then come back and try to make
5042 * our reservation.
5043 */
5044 if (used <= space_info->total_bytes) {
5045 if (used + orig_bytes <= space_info->total_bytes) {
5046 space_info->bytes_may_use += orig_bytes;
5047 trace_btrfs_space_reservation(root->fs_info,
5048 "space_info", space_info->flags, orig_bytes, 1);
5049 ret = 0;
5050 } else {
5051 /*
5052 * Ok set num_bytes to orig_bytes since we aren't
5053 * overocmmitted, this way we only try and reclaim what
5054 * we need.
5055 */
5056 num_bytes = orig_bytes;
5057 }
5058 } else {
5059 /*
5060 * Ok we're over committed, set num_bytes to the overcommitted
5061 * amount plus the amount of bytes that we need for this
5062 * reservation.
5063 */
5064 num_bytes = used - space_info->total_bytes +
5065 (orig_bytes * 2);
5066 }
5067
5068 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
5069 space_info->bytes_may_use += orig_bytes;
5070 trace_btrfs_space_reservation(root->fs_info, "space_info",
5071 space_info->flags, orig_bytes,
5072 1);
5073 ret = 0;
5074 }
5075
5076 /*
5077 * Couldn't make our reservation, save our place so while we're trying
5078 * to reclaim space we can actually use it instead of somebody else
5079 * stealing it from us.
5080 *
5081 * We make the other tasks wait for the flush only when we can flush
5082 * all things.
5083 */
5084 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5085 flushing = true;
5086 space_info->flush = 1;
5087 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5088 used += orig_bytes;
5089 /*
5090 * We will do the space reservation dance during log replay,
5091 * which means we won't have fs_info->fs_root set, so don't do
5092 * the async reclaim as we will panic.
5093 */
5094 if (!root->fs_info->log_root_recovering &&
5095 need_do_async_reclaim(space_info, root->fs_info, used) &&
5096 !work_busy(&root->fs_info->async_reclaim_work))
5097 queue_work(system_unbound_wq,
5098 &root->fs_info->async_reclaim_work);
5099 }
5100 spin_unlock(&space_info->lock);
5101
5102 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5103 goto out;
5104
5105 ret = flush_space(root, space_info, num_bytes, orig_bytes,
5106 flush_state);
5107 flush_state++;
5108
5109 /*
5110 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
5111 * would happen. So skip delalloc flush.
5112 */
5113 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
5114 (flush_state == FLUSH_DELALLOC ||
5115 flush_state == FLUSH_DELALLOC_WAIT))
5116 flush_state = ALLOC_CHUNK;
5117
5118 if (!ret)
5119 goto again;
5120 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
5121 flush_state < COMMIT_TRANS)
5122 goto again;
5123 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
5124 flush_state <= COMMIT_TRANS)
5125 goto again;
5126
5127 out:
5128 if (ret == -ENOSPC &&
5129 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5130 struct btrfs_block_rsv *global_rsv =
5131 &root->fs_info->global_block_rsv;
5132
5133 if (block_rsv != global_rsv &&
5134 !block_rsv_use_bytes(global_rsv, orig_bytes))
5135 ret = 0;
5136 }
5137 if (ret == -ENOSPC)
5138 trace_btrfs_space_reservation(root->fs_info,
5139 "space_info:enospc",
5140 space_info->flags, orig_bytes, 1);
5141 if (flushing) {
5142 spin_lock(&space_info->lock);
5143 space_info->flush = 0;
5144 wake_up_all(&space_info->wait);
5145 spin_unlock(&space_info->lock);
5146 }
5147 return ret;
5148 }
5149
5150 static struct btrfs_block_rsv *get_block_rsv(
5151 const struct btrfs_trans_handle *trans,
5152 const struct btrfs_root *root)
5153 {
5154 struct btrfs_block_rsv *block_rsv = NULL;
5155
5156 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5157 (root == root->fs_info->csum_root && trans->adding_csums) ||
5158 (root == root->fs_info->uuid_root))
5159 block_rsv = trans->block_rsv;
5160
5161 if (!block_rsv)
5162 block_rsv = root->block_rsv;
5163
5164 if (!block_rsv)
5165 block_rsv = &root->fs_info->empty_block_rsv;
5166
5167 return block_rsv;
5168 }
5169
5170 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5171 u64 num_bytes)
5172 {
5173 int ret = -ENOSPC;
5174 spin_lock(&block_rsv->lock);
5175 if (block_rsv->reserved >= num_bytes) {
5176 block_rsv->reserved -= num_bytes;
5177 if (block_rsv->reserved < block_rsv->size)
5178 block_rsv->full = 0;
5179 ret = 0;
5180 }
5181 spin_unlock(&block_rsv->lock);
5182 return ret;
5183 }
5184
5185 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5186 u64 num_bytes, int update_size)
5187 {
5188 spin_lock(&block_rsv->lock);
5189 block_rsv->reserved += num_bytes;
5190 if (update_size)
5191 block_rsv->size += num_bytes;
5192 else if (block_rsv->reserved >= block_rsv->size)
5193 block_rsv->full = 1;
5194 spin_unlock(&block_rsv->lock);
5195 }
5196
5197 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5198 struct btrfs_block_rsv *dest, u64 num_bytes,
5199 int min_factor)
5200 {
5201 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5202 u64 min_bytes;
5203
5204 if (global_rsv->space_info != dest->space_info)
5205 return -ENOSPC;
5206
5207 spin_lock(&global_rsv->lock);
5208 min_bytes = div_factor(global_rsv->size, min_factor);
5209 if (global_rsv->reserved < min_bytes + num_bytes) {
5210 spin_unlock(&global_rsv->lock);
5211 return -ENOSPC;
5212 }
5213 global_rsv->reserved -= num_bytes;
5214 if (global_rsv->reserved < global_rsv->size)
5215 global_rsv->full = 0;
5216 spin_unlock(&global_rsv->lock);
5217
5218 block_rsv_add_bytes(dest, num_bytes, 1);
5219 return 0;
5220 }
5221
5222 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5223 struct btrfs_block_rsv *block_rsv,
5224 struct btrfs_block_rsv *dest, u64 num_bytes)
5225 {
5226 struct btrfs_space_info *space_info = block_rsv->space_info;
5227
5228 spin_lock(&block_rsv->lock);
5229 if (num_bytes == (u64)-1)
5230 num_bytes = block_rsv->size;
5231 block_rsv->size -= num_bytes;
5232 if (block_rsv->reserved >= block_rsv->size) {
5233 num_bytes = block_rsv->reserved - block_rsv->size;
5234 block_rsv->reserved = block_rsv->size;
5235 block_rsv->full = 1;
5236 } else {
5237 num_bytes = 0;
5238 }
5239 spin_unlock(&block_rsv->lock);
5240
5241 if (num_bytes > 0) {
5242 if (dest) {
5243 spin_lock(&dest->lock);
5244 if (!dest->full) {
5245 u64 bytes_to_add;
5246
5247 bytes_to_add = dest->size - dest->reserved;
5248 bytes_to_add = min(num_bytes, bytes_to_add);
5249 dest->reserved += bytes_to_add;
5250 if (dest->reserved >= dest->size)
5251 dest->full = 1;
5252 num_bytes -= bytes_to_add;
5253 }
5254 spin_unlock(&dest->lock);
5255 }
5256 if (num_bytes) {
5257 spin_lock(&space_info->lock);
5258 space_info->bytes_may_use -= num_bytes;
5259 trace_btrfs_space_reservation(fs_info, "space_info",
5260 space_info->flags, num_bytes, 0);
5261 spin_unlock(&space_info->lock);
5262 }
5263 }
5264 }
5265
5266 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
5267 struct btrfs_block_rsv *dst, u64 num_bytes)
5268 {
5269 int ret;
5270
5271 ret = block_rsv_use_bytes(src, num_bytes);
5272 if (ret)
5273 return ret;
5274
5275 block_rsv_add_bytes(dst, num_bytes, 1);
5276 return 0;
5277 }
5278
5279 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5280 {
5281 memset(rsv, 0, sizeof(*rsv));
5282 spin_lock_init(&rsv->lock);
5283 rsv->type = type;
5284 }
5285
5286 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
5287 unsigned short type)
5288 {
5289 struct btrfs_block_rsv *block_rsv;
5290 struct btrfs_fs_info *fs_info = root->fs_info;
5291
5292 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5293 if (!block_rsv)
5294 return NULL;
5295
5296 btrfs_init_block_rsv(block_rsv, type);
5297 block_rsv->space_info = __find_space_info(fs_info,
5298 BTRFS_BLOCK_GROUP_METADATA);
5299 return block_rsv;
5300 }
5301
5302 void btrfs_free_block_rsv(struct btrfs_root *root,
5303 struct btrfs_block_rsv *rsv)
5304 {
5305 if (!rsv)
5306 return;
5307 btrfs_block_rsv_release(root, rsv, (u64)-1);
5308 kfree(rsv);
5309 }
5310
5311 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5312 {
5313 kfree(rsv);
5314 }
5315
5316 int btrfs_block_rsv_add(struct btrfs_root *root,
5317 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5318 enum btrfs_reserve_flush_enum flush)
5319 {
5320 int ret;
5321
5322 if (num_bytes == 0)
5323 return 0;
5324
5325 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5326 if (!ret) {
5327 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5328 return 0;
5329 }
5330
5331 return ret;
5332 }
5333
5334 int btrfs_block_rsv_check(struct btrfs_root *root,
5335 struct btrfs_block_rsv *block_rsv, int min_factor)
5336 {
5337 u64 num_bytes = 0;
5338 int ret = -ENOSPC;
5339
5340 if (!block_rsv)
5341 return 0;
5342
5343 spin_lock(&block_rsv->lock);
5344 num_bytes = div_factor(block_rsv->size, min_factor);
5345 if (block_rsv->reserved >= num_bytes)
5346 ret = 0;
5347 spin_unlock(&block_rsv->lock);
5348
5349 return ret;
5350 }
5351
5352 int btrfs_block_rsv_refill(struct btrfs_root *root,
5353 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5354 enum btrfs_reserve_flush_enum flush)
5355 {
5356 u64 num_bytes = 0;
5357 int ret = -ENOSPC;
5358
5359 if (!block_rsv)
5360 return 0;
5361
5362 spin_lock(&block_rsv->lock);
5363 num_bytes = min_reserved;
5364 if (block_rsv->reserved >= num_bytes)
5365 ret = 0;
5366 else
5367 num_bytes -= block_rsv->reserved;
5368 spin_unlock(&block_rsv->lock);
5369
5370 if (!ret)
5371 return 0;
5372
5373 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5374 if (!ret) {
5375 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5376 return 0;
5377 }
5378
5379 return ret;
5380 }
5381
5382 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
5383 struct btrfs_block_rsv *dst_rsv,
5384 u64 num_bytes)
5385 {
5386 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5387 }
5388
5389 void btrfs_block_rsv_release(struct btrfs_root *root,
5390 struct btrfs_block_rsv *block_rsv,
5391 u64 num_bytes)
5392 {
5393 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5394 if (global_rsv == block_rsv ||
5395 block_rsv->space_info != global_rsv->space_info)
5396 global_rsv = NULL;
5397 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
5398 num_bytes);
5399 }
5400
5401 /*
5402 * helper to calculate size of global block reservation.
5403 * the desired value is sum of space used by extent tree,
5404 * checksum tree and root tree
5405 */
5406 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
5407 {
5408 struct btrfs_space_info *sinfo;
5409 u64 num_bytes;
5410 u64 meta_used;
5411 u64 data_used;
5412 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
5413
5414 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
5415 spin_lock(&sinfo->lock);
5416 data_used = sinfo->bytes_used;
5417 spin_unlock(&sinfo->lock);
5418
5419 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5420 spin_lock(&sinfo->lock);
5421 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
5422 data_used = 0;
5423 meta_used = sinfo->bytes_used;
5424 spin_unlock(&sinfo->lock);
5425
5426 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
5427 csum_size * 2;
5428 num_bytes += div_u64(data_used + meta_used, 50);
5429
5430 if (num_bytes * 3 > meta_used)
5431 num_bytes = div_u64(meta_used, 3);
5432
5433 return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
5434 }
5435
5436 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5437 {
5438 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5439 struct btrfs_space_info *sinfo = block_rsv->space_info;
5440 u64 num_bytes;
5441
5442 num_bytes = calc_global_metadata_size(fs_info);
5443
5444 spin_lock(&sinfo->lock);
5445 spin_lock(&block_rsv->lock);
5446
5447 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5448
5449 if (block_rsv->reserved < block_rsv->size) {
5450 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
5451 sinfo->bytes_reserved + sinfo->bytes_readonly +
5452 sinfo->bytes_may_use;
5453 if (sinfo->total_bytes > num_bytes) {
5454 num_bytes = sinfo->total_bytes - num_bytes;
5455 num_bytes = min(num_bytes,
5456 block_rsv->size - block_rsv->reserved);
5457 block_rsv->reserved += num_bytes;
5458 sinfo->bytes_may_use += num_bytes;
5459 trace_btrfs_space_reservation(fs_info, "space_info",
5460 sinfo->flags, num_bytes,
5461 1);
5462 }
5463 } else if (block_rsv->reserved > block_rsv->size) {
5464 num_bytes = block_rsv->reserved - block_rsv->size;
5465 sinfo->bytes_may_use -= num_bytes;
5466 trace_btrfs_space_reservation(fs_info, "space_info",
5467 sinfo->flags, num_bytes, 0);
5468 block_rsv->reserved = block_rsv->size;
5469 }
5470
5471 if (block_rsv->reserved == block_rsv->size)
5472 block_rsv->full = 1;
5473 else
5474 block_rsv->full = 0;
5475
5476 spin_unlock(&block_rsv->lock);
5477 spin_unlock(&sinfo->lock);
5478 }
5479
5480 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5481 {
5482 struct btrfs_space_info *space_info;
5483
5484 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5485 fs_info->chunk_block_rsv.space_info = space_info;
5486
5487 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5488 fs_info->global_block_rsv.space_info = space_info;
5489 fs_info->delalloc_block_rsv.space_info = space_info;
5490 fs_info->trans_block_rsv.space_info = space_info;
5491 fs_info->empty_block_rsv.space_info = space_info;
5492 fs_info->delayed_block_rsv.space_info = space_info;
5493
5494 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5495 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5496 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5497 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5498 if (fs_info->quota_root)
5499 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5500 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5501
5502 update_global_block_rsv(fs_info);
5503 }
5504
5505 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5506 {
5507 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5508 (u64)-1);
5509 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5510 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5511 WARN_ON(fs_info->trans_block_rsv.size > 0);
5512 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5513 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5514 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5515 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5516 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5517 }
5518
5519 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5520 struct btrfs_root *root)
5521 {
5522 if (!trans->block_rsv)
5523 return;
5524
5525 if (!trans->bytes_reserved)
5526 return;
5527
5528 trace_btrfs_space_reservation(root->fs_info, "transaction",
5529 trans->transid, trans->bytes_reserved, 0);
5530 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
5531 trans->bytes_reserved = 0;
5532 }
5533
5534 /*
5535 * To be called after all the new block groups attached to the transaction
5536 * handle have been created (btrfs_create_pending_block_groups()).
5537 */
5538 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5539 {
5540 struct btrfs_fs_info *fs_info = trans->root->fs_info;
5541
5542 if (!trans->chunk_bytes_reserved)
5543 return;
5544
5545 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5546
5547 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5548 trans->chunk_bytes_reserved);
5549 trans->chunk_bytes_reserved = 0;
5550 }
5551
5552 /* Can only return 0 or -ENOSPC */
5553 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5554 struct inode *inode)
5555 {
5556 struct btrfs_root *root = BTRFS_I(inode)->root;
5557 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
5558 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5559
5560 /*
5561 * We need to hold space in order to delete our orphan item once we've
5562 * added it, so this takes the reservation so we can release it later
5563 * when we are truly done with the orphan item.
5564 */
5565 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5566 trace_btrfs_space_reservation(root->fs_info, "orphan",
5567 btrfs_ino(inode), num_bytes, 1);
5568 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5569 }
5570
5571 void btrfs_orphan_release_metadata(struct inode *inode)
5572 {
5573 struct btrfs_root *root = BTRFS_I(inode)->root;
5574 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5575 trace_btrfs_space_reservation(root->fs_info, "orphan",
5576 btrfs_ino(inode), num_bytes, 0);
5577 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
5578 }
5579
5580 /*
5581 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5582 * root: the root of the parent directory
5583 * rsv: block reservation
5584 * items: the number of items that we need do reservation
5585 * qgroup_reserved: used to return the reserved size in qgroup
5586 *
5587 * This function is used to reserve the space for snapshot/subvolume
5588 * creation and deletion. Those operations are different with the
5589 * common file/directory operations, they change two fs/file trees
5590 * and root tree, the number of items that the qgroup reserves is
5591 * different with the free space reservation. So we can not use
5592 * the space reservation mechanism in start_transaction().
5593 */
5594 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5595 struct btrfs_block_rsv *rsv,
5596 int items,
5597 u64 *qgroup_reserved,
5598 bool use_global_rsv)
5599 {
5600 u64 num_bytes;
5601 int ret;
5602 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5603
5604 if (root->fs_info->quota_enabled) {
5605 /* One for parent inode, two for dir entries */
5606 num_bytes = 3 * root->nodesize;
5607 ret = btrfs_qgroup_reserve_meta(root, num_bytes);
5608 if (ret)
5609 return ret;
5610 } else {
5611 num_bytes = 0;
5612 }
5613
5614 *qgroup_reserved = num_bytes;
5615
5616 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5617 rsv->space_info = __find_space_info(root->fs_info,
5618 BTRFS_BLOCK_GROUP_METADATA);
5619 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5620 BTRFS_RESERVE_FLUSH_ALL);
5621
5622 if (ret == -ENOSPC && use_global_rsv)
5623 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
5624
5625 if (ret && *qgroup_reserved)
5626 btrfs_qgroup_free_meta(root, *qgroup_reserved);
5627
5628 return ret;
5629 }
5630
5631 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5632 struct btrfs_block_rsv *rsv,
5633 u64 qgroup_reserved)
5634 {
5635 btrfs_block_rsv_release(root, rsv, (u64)-1);
5636 }
5637
5638 /**
5639 * drop_outstanding_extent - drop an outstanding extent
5640 * @inode: the inode we're dropping the extent for
5641 * @num_bytes: the number of bytes we're releasing.
5642 *
5643 * This is called when we are freeing up an outstanding extent, either called
5644 * after an error or after an extent is written. This will return the number of
5645 * reserved extents that need to be freed. This must be called with
5646 * BTRFS_I(inode)->lock held.
5647 */
5648 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5649 {
5650 unsigned drop_inode_space = 0;
5651 unsigned dropped_extents = 0;
5652 unsigned num_extents = 0;
5653
5654 num_extents = (unsigned)div64_u64(num_bytes +
5655 BTRFS_MAX_EXTENT_SIZE - 1,
5656 BTRFS_MAX_EXTENT_SIZE);
5657 ASSERT(num_extents);
5658 ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5659 BTRFS_I(inode)->outstanding_extents -= num_extents;
5660
5661 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5662 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5663 &BTRFS_I(inode)->runtime_flags))
5664 drop_inode_space = 1;
5665
5666 /*
5667 * If we have more or the same amount of outstanding extents than we have
5668 * reserved then we need to leave the reserved extents count alone.
5669 */
5670 if (BTRFS_I(inode)->outstanding_extents >=
5671 BTRFS_I(inode)->reserved_extents)
5672 return drop_inode_space;
5673
5674 dropped_extents = BTRFS_I(inode)->reserved_extents -
5675 BTRFS_I(inode)->outstanding_extents;
5676 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5677 return dropped_extents + drop_inode_space;
5678 }
5679
5680 /**
5681 * calc_csum_metadata_size - return the amount of metadata space that must be
5682 * reserved/freed for the given bytes.
5683 * @inode: the inode we're manipulating
5684 * @num_bytes: the number of bytes in question
5685 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5686 *
5687 * This adjusts the number of csum_bytes in the inode and then returns the
5688 * correct amount of metadata that must either be reserved or freed. We
5689 * calculate how many checksums we can fit into one leaf and then divide the
5690 * number of bytes that will need to be checksumed by this value to figure out
5691 * how many checksums will be required. If we are adding bytes then the number
5692 * may go up and we will return the number of additional bytes that must be
5693 * reserved. If it is going down we will return the number of bytes that must
5694 * be freed.
5695 *
5696 * This must be called with BTRFS_I(inode)->lock held.
5697 */
5698 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5699 int reserve)
5700 {
5701 struct btrfs_root *root = BTRFS_I(inode)->root;
5702 u64 old_csums, num_csums;
5703
5704 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5705 BTRFS_I(inode)->csum_bytes == 0)
5706 return 0;
5707
5708 old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5709 if (reserve)
5710 BTRFS_I(inode)->csum_bytes += num_bytes;
5711 else
5712 BTRFS_I(inode)->csum_bytes -= num_bytes;
5713 num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5714
5715 /* No change, no need to reserve more */
5716 if (old_csums == num_csums)
5717 return 0;
5718
5719 if (reserve)
5720 return btrfs_calc_trans_metadata_size(root,
5721 num_csums - old_csums);
5722
5723 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5724 }
5725
5726 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5727 {
5728 struct btrfs_root *root = BTRFS_I(inode)->root;
5729 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5730 u64 to_reserve = 0;
5731 u64 csum_bytes;
5732 unsigned nr_extents = 0;
5733 int extra_reserve = 0;
5734 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5735 int ret = 0;
5736 bool delalloc_lock = true;
5737 u64 to_free = 0;
5738 unsigned dropped;
5739
5740 /* If we are a free space inode we need to not flush since we will be in
5741 * the middle of a transaction commit. We also don't need the delalloc
5742 * mutex since we won't race with anybody. We need this mostly to make
5743 * lockdep shut its filthy mouth.
5744 */
5745 if (btrfs_is_free_space_inode(inode)) {
5746 flush = BTRFS_RESERVE_NO_FLUSH;
5747 delalloc_lock = false;
5748 }
5749
5750 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5751 btrfs_transaction_in_commit(root->fs_info))
5752 schedule_timeout(1);
5753
5754 if (delalloc_lock)
5755 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5756
5757 num_bytes = ALIGN(num_bytes, root->sectorsize);
5758
5759 spin_lock(&BTRFS_I(inode)->lock);
5760 nr_extents = (unsigned)div64_u64(num_bytes +
5761 BTRFS_MAX_EXTENT_SIZE - 1,
5762 BTRFS_MAX_EXTENT_SIZE);
5763 BTRFS_I(inode)->outstanding_extents += nr_extents;
5764 nr_extents = 0;
5765
5766 if (BTRFS_I(inode)->outstanding_extents >
5767 BTRFS_I(inode)->reserved_extents)
5768 nr_extents = BTRFS_I(inode)->outstanding_extents -
5769 BTRFS_I(inode)->reserved_extents;
5770
5771 /*
5772 * Add an item to reserve for updating the inode when we complete the
5773 * delalloc io.
5774 */
5775 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5776 &BTRFS_I(inode)->runtime_flags)) {
5777 nr_extents++;
5778 extra_reserve = 1;
5779 }
5780
5781 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5782 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5783 csum_bytes = BTRFS_I(inode)->csum_bytes;
5784 spin_unlock(&BTRFS_I(inode)->lock);
5785
5786 if (root->fs_info->quota_enabled) {
5787 ret = btrfs_qgroup_reserve_meta(root,
5788 nr_extents * root->nodesize);
5789 if (ret)
5790 goto out_fail;
5791 }
5792
5793 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5794 if (unlikely(ret)) {
5795 btrfs_qgroup_free_meta(root, nr_extents * root->nodesize);
5796 goto out_fail;
5797 }
5798
5799 spin_lock(&BTRFS_I(inode)->lock);
5800 if (extra_reserve) {
5801 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5802 &BTRFS_I(inode)->runtime_flags);
5803 nr_extents--;
5804 }
5805 BTRFS_I(inode)->reserved_extents += nr_extents;
5806 spin_unlock(&BTRFS_I(inode)->lock);
5807
5808 if (delalloc_lock)
5809 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5810
5811 if (to_reserve)
5812 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5813 btrfs_ino(inode), to_reserve, 1);
5814 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5815
5816 return 0;
5817
5818 out_fail:
5819 spin_lock(&BTRFS_I(inode)->lock);
5820 dropped = drop_outstanding_extent(inode, num_bytes);
5821 /*
5822 * If the inodes csum_bytes is the same as the original
5823 * csum_bytes then we know we haven't raced with any free()ers
5824 * so we can just reduce our inodes csum bytes and carry on.
5825 */
5826 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5827 calc_csum_metadata_size(inode, num_bytes, 0);
5828 } else {
5829 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5830 u64 bytes;
5831
5832 /*
5833 * This is tricky, but first we need to figure out how much we
5834 * freed from any free-ers that occurred during this
5835 * reservation, so we reset ->csum_bytes to the csum_bytes
5836 * before we dropped our lock, and then call the free for the
5837 * number of bytes that were freed while we were trying our
5838 * reservation.
5839 */
5840 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5841 BTRFS_I(inode)->csum_bytes = csum_bytes;
5842 to_free = calc_csum_metadata_size(inode, bytes, 0);
5843
5844
5845 /*
5846 * Now we need to see how much we would have freed had we not
5847 * been making this reservation and our ->csum_bytes were not
5848 * artificially inflated.
5849 */
5850 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5851 bytes = csum_bytes - orig_csum_bytes;
5852 bytes = calc_csum_metadata_size(inode, bytes, 0);
5853
5854 /*
5855 * Now reset ->csum_bytes to what it should be. If bytes is
5856 * more than to_free then we would have freed more space had we
5857 * not had an artificially high ->csum_bytes, so we need to free
5858 * the remainder. If bytes is the same or less then we don't
5859 * need to do anything, the other free-ers did the correct
5860 * thing.
5861 */
5862 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5863 if (bytes > to_free)
5864 to_free = bytes - to_free;
5865 else
5866 to_free = 0;
5867 }
5868 spin_unlock(&BTRFS_I(inode)->lock);
5869 if (dropped)
5870 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5871
5872 if (to_free) {
5873 btrfs_block_rsv_release(root, block_rsv, to_free);
5874 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5875 btrfs_ino(inode), to_free, 0);
5876 }
5877 if (delalloc_lock)
5878 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5879 return ret;
5880 }
5881
5882 /**
5883 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5884 * @inode: the inode to release the reservation for
5885 * @num_bytes: the number of bytes we're releasing
5886 *
5887 * This will release the metadata reservation for an inode. This can be called
5888 * once we complete IO for a given set of bytes to release their metadata
5889 * reservations.
5890 */
5891 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5892 {
5893 struct btrfs_root *root = BTRFS_I(inode)->root;
5894 u64 to_free = 0;
5895 unsigned dropped;
5896
5897 num_bytes = ALIGN(num_bytes, root->sectorsize);
5898 spin_lock(&BTRFS_I(inode)->lock);
5899 dropped = drop_outstanding_extent(inode, num_bytes);
5900
5901 if (num_bytes)
5902 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5903 spin_unlock(&BTRFS_I(inode)->lock);
5904 if (dropped > 0)
5905 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5906
5907 if (btrfs_test_is_dummy_root(root))
5908 return;
5909
5910 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5911 btrfs_ino(inode), to_free, 0);
5912
5913 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5914 to_free);
5915 }
5916
5917 /**
5918 * btrfs_delalloc_reserve_space - reserve data and metadata space for
5919 * delalloc
5920 * @inode: inode we're writing to
5921 * @start: start range we are writing to
5922 * @len: how long the range we are writing to
5923 *
5924 * TODO: This function will finally replace old btrfs_delalloc_reserve_space()
5925 *
5926 * This will do the following things
5927 *
5928 * o reserve space in data space info for num bytes
5929 * and reserve precious corresponding qgroup space
5930 * (Done in check_data_free_space)
5931 *
5932 * o reserve space for metadata space, based on the number of outstanding
5933 * extents and how much csums will be needed
5934 * also reserve metadata space in a per root over-reserve method.
5935 * o add to the inodes->delalloc_bytes
5936 * o add it to the fs_info's delalloc inodes list.
5937 * (Above 3 all done in delalloc_reserve_metadata)
5938 *
5939 * Return 0 for success
5940 * Return <0 for error(-ENOSPC or -EQUOT)
5941 */
5942 int btrfs_delalloc_reserve_space(struct inode *inode, u64 start, u64 len)
5943 {
5944 int ret;
5945
5946 ret = btrfs_check_data_free_space(inode, start, len);
5947 if (ret < 0)
5948 return ret;
5949 ret = btrfs_delalloc_reserve_metadata(inode, len);
5950 if (ret < 0)
5951 btrfs_free_reserved_data_space(inode, start, len);
5952 return ret;
5953 }
5954
5955 /**
5956 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5957 * @inode: inode we're releasing space for
5958 * @start: start position of the space already reserved
5959 * @len: the len of the space already reserved
5960 *
5961 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5962 * called in the case that we don't need the metadata AND data reservations
5963 * anymore. So if there is an error or we insert an inline extent.
5964 *
5965 * This function will release the metadata space that was not used and will
5966 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5967 * list if there are no delalloc bytes left.
5968 * Also it will handle the qgroup reserved space.
5969 */
5970 void btrfs_delalloc_release_space(struct inode *inode, u64 start, u64 len)
5971 {
5972 btrfs_delalloc_release_metadata(inode, len);
5973 btrfs_free_reserved_data_space(inode, start, len);
5974 }
5975
5976 static int update_block_group(struct btrfs_trans_handle *trans,
5977 struct btrfs_root *root, u64 bytenr,
5978 u64 num_bytes, int alloc)
5979 {
5980 struct btrfs_block_group_cache *cache = NULL;
5981 struct btrfs_fs_info *info = root->fs_info;
5982 u64 total = num_bytes;
5983 u64 old_val;
5984 u64 byte_in_group;
5985 int factor;
5986
5987 /* block accounting for super block */
5988 spin_lock(&info->delalloc_root_lock);
5989 old_val = btrfs_super_bytes_used(info->super_copy);
5990 if (alloc)
5991 old_val += num_bytes;
5992 else
5993 old_val -= num_bytes;
5994 btrfs_set_super_bytes_used(info->super_copy, old_val);
5995 spin_unlock(&info->delalloc_root_lock);
5996
5997 while (total) {
5998 cache = btrfs_lookup_block_group(info, bytenr);
5999 if (!cache)
6000 return -ENOENT;
6001 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
6002 BTRFS_BLOCK_GROUP_RAID1 |
6003 BTRFS_BLOCK_GROUP_RAID10))
6004 factor = 2;
6005 else
6006 factor = 1;
6007 /*
6008 * If this block group has free space cache written out, we
6009 * need to make sure to load it if we are removing space. This
6010 * is because we need the unpinning stage to actually add the
6011 * space back to the block group, otherwise we will leak space.
6012 */
6013 if (!alloc && cache->cached == BTRFS_CACHE_NO)
6014 cache_block_group(cache, 1);
6015
6016 byte_in_group = bytenr - cache->key.objectid;
6017 WARN_ON(byte_in_group > cache->key.offset);
6018
6019 spin_lock(&cache->space_info->lock);
6020 spin_lock(&cache->lock);
6021
6022 if (btrfs_test_opt(root, SPACE_CACHE) &&
6023 cache->disk_cache_state < BTRFS_DC_CLEAR)
6024 cache->disk_cache_state = BTRFS_DC_CLEAR;
6025
6026 old_val = btrfs_block_group_used(&cache->item);
6027 num_bytes = min(total, cache->key.offset - byte_in_group);
6028 if (alloc) {
6029 old_val += num_bytes;
6030 btrfs_set_block_group_used(&cache->item, old_val);
6031 cache->reserved -= num_bytes;
6032 cache->space_info->bytes_reserved -= num_bytes;
6033 cache->space_info->bytes_used += num_bytes;
6034 cache->space_info->disk_used += num_bytes * factor;
6035 spin_unlock(&cache->lock);
6036 spin_unlock(&cache->space_info->lock);
6037 } else {
6038 old_val -= num_bytes;
6039 btrfs_set_block_group_used(&cache->item, old_val);
6040 cache->pinned += num_bytes;
6041 cache->space_info->bytes_pinned += num_bytes;
6042 cache->space_info->bytes_used -= num_bytes;
6043 cache->space_info->disk_used -= num_bytes * factor;
6044 spin_unlock(&cache->lock);
6045 spin_unlock(&cache->space_info->lock);
6046
6047 set_extent_dirty(info->pinned_extents,
6048 bytenr, bytenr + num_bytes - 1,
6049 GFP_NOFS | __GFP_NOFAIL);
6050 }
6051
6052 spin_lock(&trans->transaction->dirty_bgs_lock);
6053 if (list_empty(&cache->dirty_list)) {
6054 list_add_tail(&cache->dirty_list,
6055 &trans->transaction->dirty_bgs);
6056 trans->transaction->num_dirty_bgs++;
6057 btrfs_get_block_group(cache);
6058 }
6059 spin_unlock(&trans->transaction->dirty_bgs_lock);
6060
6061 /*
6062 * No longer have used bytes in this block group, queue it for
6063 * deletion. We do this after adding the block group to the
6064 * dirty list to avoid races between cleaner kthread and space
6065 * cache writeout.
6066 */
6067 if (!alloc && old_val == 0) {
6068 spin_lock(&info->unused_bgs_lock);
6069 if (list_empty(&cache->bg_list)) {
6070 btrfs_get_block_group(cache);
6071 list_add_tail(&cache->bg_list,
6072 &info->unused_bgs);
6073 }
6074 spin_unlock(&info->unused_bgs_lock);
6075 }
6076
6077 btrfs_put_block_group(cache);
6078 total -= num_bytes;
6079 bytenr += num_bytes;
6080 }
6081 return 0;
6082 }
6083
6084 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
6085 {
6086 struct btrfs_block_group_cache *cache;
6087 u64 bytenr;
6088
6089 spin_lock(&root->fs_info->block_group_cache_lock);
6090 bytenr = root->fs_info->first_logical_byte;
6091 spin_unlock(&root->fs_info->block_group_cache_lock);
6092
6093 if (bytenr < (u64)-1)
6094 return bytenr;
6095
6096 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
6097 if (!cache)
6098 return 0;
6099
6100 bytenr = cache->key.objectid;
6101 btrfs_put_block_group(cache);
6102
6103 return bytenr;
6104 }
6105
6106 static int pin_down_extent(struct btrfs_root *root,
6107 struct btrfs_block_group_cache *cache,
6108 u64 bytenr, u64 num_bytes, int reserved)
6109 {
6110 spin_lock(&cache->space_info->lock);
6111 spin_lock(&cache->lock);
6112 cache->pinned += num_bytes;
6113 cache->space_info->bytes_pinned += num_bytes;
6114 if (reserved) {
6115 cache->reserved -= num_bytes;
6116 cache->space_info->bytes_reserved -= num_bytes;
6117 }
6118 spin_unlock(&cache->lock);
6119 spin_unlock(&cache->space_info->lock);
6120
6121 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
6122 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6123 if (reserved)
6124 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
6125 return 0;
6126 }
6127
6128 /*
6129 * this function must be called within transaction
6130 */
6131 int btrfs_pin_extent(struct btrfs_root *root,
6132 u64 bytenr, u64 num_bytes, int reserved)
6133 {
6134 struct btrfs_block_group_cache *cache;
6135
6136 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
6137 BUG_ON(!cache); /* Logic error */
6138
6139 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
6140
6141 btrfs_put_block_group(cache);
6142 return 0;
6143 }
6144
6145 /*
6146 * this function must be called within transaction
6147 */
6148 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
6149 u64 bytenr, u64 num_bytes)
6150 {
6151 struct btrfs_block_group_cache *cache;
6152 int ret;
6153
6154 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
6155 if (!cache)
6156 return -EINVAL;
6157
6158 /*
6159 * pull in the free space cache (if any) so that our pin
6160 * removes the free space from the cache. We have load_only set
6161 * to one because the slow code to read in the free extents does check
6162 * the pinned extents.
6163 */
6164 cache_block_group(cache, 1);
6165
6166 pin_down_extent(root, cache, bytenr, num_bytes, 0);
6167
6168 /* remove us from the free space cache (if we're there at all) */
6169 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6170 btrfs_put_block_group(cache);
6171 return ret;
6172 }
6173
6174 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
6175 {
6176 int ret;
6177 struct btrfs_block_group_cache *block_group;
6178 struct btrfs_caching_control *caching_ctl;
6179
6180 block_group = btrfs_lookup_block_group(root->fs_info, start);
6181 if (!block_group)
6182 return -EINVAL;
6183
6184 cache_block_group(block_group, 0);
6185 caching_ctl = get_caching_control(block_group);
6186
6187 if (!caching_ctl) {
6188 /* Logic error */
6189 BUG_ON(!block_group_cache_done(block_group));
6190 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6191 } else {
6192 mutex_lock(&caching_ctl->mutex);
6193
6194 if (start >= caching_ctl->progress) {
6195 ret = add_excluded_extent(root, start, num_bytes);
6196 } else if (start + num_bytes <= caching_ctl->progress) {
6197 ret = btrfs_remove_free_space(block_group,
6198 start, num_bytes);
6199 } else {
6200 num_bytes = caching_ctl->progress - start;
6201 ret = btrfs_remove_free_space(block_group,
6202 start, num_bytes);
6203 if (ret)
6204 goto out_lock;
6205
6206 num_bytes = (start + num_bytes) -
6207 caching_ctl->progress;
6208 start = caching_ctl->progress;
6209 ret = add_excluded_extent(root, start, num_bytes);
6210 }
6211 out_lock:
6212 mutex_unlock(&caching_ctl->mutex);
6213 put_caching_control(caching_ctl);
6214 }
6215 btrfs_put_block_group(block_group);
6216 return ret;
6217 }
6218
6219 int btrfs_exclude_logged_extents(struct btrfs_root *log,
6220 struct extent_buffer *eb)
6221 {
6222 struct btrfs_file_extent_item *item;
6223 struct btrfs_key key;
6224 int found_type;
6225 int i;
6226
6227 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
6228 return 0;
6229
6230 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6231 btrfs_item_key_to_cpu(eb, &key, i);
6232 if (key.type != BTRFS_EXTENT_DATA_KEY)
6233 continue;
6234 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6235 found_type = btrfs_file_extent_type(eb, item);
6236 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6237 continue;
6238 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6239 continue;
6240 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6241 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6242 __exclude_logged_extent(log, key.objectid, key.offset);
6243 }
6244
6245 return 0;
6246 }
6247
6248 static void
6249 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
6250 {
6251 atomic_inc(&bg->reservations);
6252 }
6253
6254 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
6255 const u64 start)
6256 {
6257 struct btrfs_block_group_cache *bg;
6258
6259 bg = btrfs_lookup_block_group(fs_info, start);
6260 ASSERT(bg);
6261 if (atomic_dec_and_test(&bg->reservations))
6262 wake_up_atomic_t(&bg->reservations);
6263 btrfs_put_block_group(bg);
6264 }
6265
6266 static int btrfs_wait_bg_reservations_atomic_t(atomic_t *a)
6267 {
6268 schedule();
6269 return 0;
6270 }
6271
6272 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg)
6273 {
6274 struct btrfs_space_info *space_info = bg->space_info;
6275
6276 ASSERT(bg->ro);
6277
6278 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
6279 return;
6280
6281 /*
6282 * Our block group is read only but before we set it to read only,
6283 * some task might have had allocated an extent from it already, but it
6284 * has not yet created a respective ordered extent (and added it to a
6285 * root's list of ordered extents).
6286 * Therefore wait for any task currently allocating extents, since the
6287 * block group's reservations counter is incremented while a read lock
6288 * on the groups' semaphore is held and decremented after releasing
6289 * the read access on that semaphore and creating the ordered extent.
6290 */
6291 down_write(&space_info->groups_sem);
6292 up_write(&space_info->groups_sem);
6293
6294 wait_on_atomic_t(&bg->reservations,
6295 btrfs_wait_bg_reservations_atomic_t,
6296 TASK_UNINTERRUPTIBLE);
6297 }
6298
6299 /**
6300 * btrfs_update_reserved_bytes - update the block_group and space info counters
6301 * @cache: The cache we are manipulating
6302 * @num_bytes: The number of bytes in question
6303 * @reserve: One of the reservation enums
6304 * @delalloc: The blocks are allocated for the delalloc write
6305 *
6306 * This is called by the allocator when it reserves space, or by somebody who is
6307 * freeing space that was never actually used on disk. For example if you
6308 * reserve some space for a new leaf in transaction A and before transaction A
6309 * commits you free that leaf, you call this with reserve set to 0 in order to
6310 * clear the reservation.
6311 *
6312 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
6313 * ENOSPC accounting. For data we handle the reservation through clearing the
6314 * delalloc bits in the io_tree. We have to do this since we could end up
6315 * allocating less disk space for the amount of data we have reserved in the
6316 * case of compression.
6317 *
6318 * If this is a reservation and the block group has become read only we cannot
6319 * make the reservation and return -EAGAIN, otherwise this function always
6320 * succeeds.
6321 */
6322 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
6323 u64 num_bytes, int reserve, int delalloc)
6324 {
6325 struct btrfs_space_info *space_info = cache->space_info;
6326 int ret = 0;
6327
6328 spin_lock(&space_info->lock);
6329 spin_lock(&cache->lock);
6330 if (reserve != RESERVE_FREE) {
6331 if (cache->ro) {
6332 ret = -EAGAIN;
6333 } else {
6334 cache->reserved += num_bytes;
6335 space_info->bytes_reserved += num_bytes;
6336 if (reserve == RESERVE_ALLOC) {
6337 trace_btrfs_space_reservation(cache->fs_info,
6338 "space_info", space_info->flags,
6339 num_bytes, 0);
6340 space_info->bytes_may_use -= num_bytes;
6341 }
6342
6343 if (delalloc)
6344 cache->delalloc_bytes += num_bytes;
6345 }
6346 } else {
6347 if (cache->ro)
6348 space_info->bytes_readonly += num_bytes;
6349 cache->reserved -= num_bytes;
6350 space_info->bytes_reserved -= num_bytes;
6351
6352 if (delalloc)
6353 cache->delalloc_bytes -= num_bytes;
6354 }
6355 spin_unlock(&cache->lock);
6356 spin_unlock(&space_info->lock);
6357 return ret;
6358 }
6359
6360 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
6361 struct btrfs_root *root)
6362 {
6363 struct btrfs_fs_info *fs_info = root->fs_info;
6364 struct btrfs_caching_control *next;
6365 struct btrfs_caching_control *caching_ctl;
6366 struct btrfs_block_group_cache *cache;
6367
6368 down_write(&fs_info->commit_root_sem);
6369
6370 list_for_each_entry_safe(caching_ctl, next,
6371 &fs_info->caching_block_groups, list) {
6372 cache = caching_ctl->block_group;
6373 if (block_group_cache_done(cache)) {
6374 cache->last_byte_to_unpin = (u64)-1;
6375 list_del_init(&caching_ctl->list);
6376 put_caching_control(caching_ctl);
6377 } else {
6378 cache->last_byte_to_unpin = caching_ctl->progress;
6379 }
6380 }
6381
6382 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6383 fs_info->pinned_extents = &fs_info->freed_extents[1];
6384 else
6385 fs_info->pinned_extents = &fs_info->freed_extents[0];
6386
6387 up_write(&fs_info->commit_root_sem);
6388
6389 update_global_block_rsv(fs_info);
6390 }
6391
6392 /*
6393 * Returns the free cluster for the given space info and sets empty_cluster to
6394 * what it should be based on the mount options.
6395 */
6396 static struct btrfs_free_cluster *
6397 fetch_cluster_info(struct btrfs_root *root, struct btrfs_space_info *space_info,
6398 u64 *empty_cluster)
6399 {
6400 struct btrfs_free_cluster *ret = NULL;
6401 bool ssd = btrfs_test_opt(root, SSD);
6402
6403 *empty_cluster = 0;
6404 if (btrfs_mixed_space_info(space_info))
6405 return ret;
6406
6407 if (ssd)
6408 *empty_cluster = SZ_2M;
6409 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6410 ret = &root->fs_info->meta_alloc_cluster;
6411 if (!ssd)
6412 *empty_cluster = SZ_64K;
6413 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && ssd) {
6414 ret = &root->fs_info->data_alloc_cluster;
6415 }
6416
6417 return ret;
6418 }
6419
6420 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
6421 const bool return_free_space)
6422 {
6423 struct btrfs_fs_info *fs_info = root->fs_info;
6424 struct btrfs_block_group_cache *cache = NULL;
6425 struct btrfs_space_info *space_info;
6426 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6427 struct btrfs_free_cluster *cluster = NULL;
6428 u64 len;
6429 u64 total_unpinned = 0;
6430 u64 empty_cluster = 0;
6431 bool readonly;
6432
6433 while (start <= end) {
6434 readonly = false;
6435 if (!cache ||
6436 start >= cache->key.objectid + cache->key.offset) {
6437 if (cache)
6438 btrfs_put_block_group(cache);
6439 total_unpinned = 0;
6440 cache = btrfs_lookup_block_group(fs_info, start);
6441 BUG_ON(!cache); /* Logic error */
6442
6443 cluster = fetch_cluster_info(root,
6444 cache->space_info,
6445 &empty_cluster);
6446 empty_cluster <<= 1;
6447 }
6448
6449 len = cache->key.objectid + cache->key.offset - start;
6450 len = min(len, end + 1 - start);
6451
6452 if (start < cache->last_byte_to_unpin) {
6453 len = min(len, cache->last_byte_to_unpin - start);
6454 if (return_free_space)
6455 btrfs_add_free_space(cache, start, len);
6456 }
6457
6458 start += len;
6459 total_unpinned += len;
6460 space_info = cache->space_info;
6461
6462 /*
6463 * If this space cluster has been marked as fragmented and we've
6464 * unpinned enough in this block group to potentially allow a
6465 * cluster to be created inside of it go ahead and clear the
6466 * fragmented check.
6467 */
6468 if (cluster && cluster->fragmented &&
6469 total_unpinned > empty_cluster) {
6470 spin_lock(&cluster->lock);
6471 cluster->fragmented = 0;
6472 spin_unlock(&cluster->lock);
6473 }
6474
6475 spin_lock(&space_info->lock);
6476 spin_lock(&cache->lock);
6477 cache->pinned -= len;
6478 space_info->bytes_pinned -= len;
6479 space_info->max_extent_size = 0;
6480 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6481 if (cache->ro) {
6482 space_info->bytes_readonly += len;
6483 readonly = true;
6484 }
6485 spin_unlock(&cache->lock);
6486 if (!readonly && global_rsv->space_info == space_info) {
6487 spin_lock(&global_rsv->lock);
6488 if (!global_rsv->full) {
6489 len = min(len, global_rsv->size -
6490 global_rsv->reserved);
6491 global_rsv->reserved += len;
6492 space_info->bytes_may_use += len;
6493 if (global_rsv->reserved >= global_rsv->size)
6494 global_rsv->full = 1;
6495 }
6496 spin_unlock(&global_rsv->lock);
6497 }
6498 spin_unlock(&space_info->lock);
6499 }
6500
6501 if (cache)
6502 btrfs_put_block_group(cache);
6503 return 0;
6504 }
6505
6506 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6507 struct btrfs_root *root)
6508 {
6509 struct btrfs_fs_info *fs_info = root->fs_info;
6510 struct btrfs_block_group_cache *block_group, *tmp;
6511 struct list_head *deleted_bgs;
6512 struct extent_io_tree *unpin;
6513 u64 start;
6514 u64 end;
6515 int ret;
6516
6517 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6518 unpin = &fs_info->freed_extents[1];
6519 else
6520 unpin = &fs_info->freed_extents[0];
6521
6522 while (!trans->aborted) {
6523 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6524 ret = find_first_extent_bit(unpin, 0, &start, &end,
6525 EXTENT_DIRTY, NULL);
6526 if (ret) {
6527 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6528 break;
6529 }
6530
6531 if (btrfs_test_opt(root, DISCARD))
6532 ret = btrfs_discard_extent(root, start,
6533 end + 1 - start, NULL);
6534
6535 clear_extent_dirty(unpin, start, end);
6536 unpin_extent_range(root, start, end, true);
6537 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6538 cond_resched();
6539 }
6540
6541 /*
6542 * Transaction is finished. We don't need the lock anymore. We
6543 * do need to clean up the block groups in case of a transaction
6544 * abort.
6545 */
6546 deleted_bgs = &trans->transaction->deleted_bgs;
6547 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6548 u64 trimmed = 0;
6549
6550 ret = -EROFS;
6551 if (!trans->aborted)
6552 ret = btrfs_discard_extent(root,
6553 block_group->key.objectid,
6554 block_group->key.offset,
6555 &trimmed);
6556
6557 list_del_init(&block_group->bg_list);
6558 btrfs_put_block_group_trimming(block_group);
6559 btrfs_put_block_group(block_group);
6560
6561 if (ret) {
6562 const char *errstr = btrfs_decode_error(ret);
6563 btrfs_warn(fs_info,
6564 "Discard failed while removing blockgroup: errno=%d %s\n",
6565 ret, errstr);
6566 }
6567 }
6568
6569 return 0;
6570 }
6571
6572 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
6573 u64 owner, u64 root_objectid)
6574 {
6575 struct btrfs_space_info *space_info;
6576 u64 flags;
6577
6578 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6579 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
6580 flags = BTRFS_BLOCK_GROUP_SYSTEM;
6581 else
6582 flags = BTRFS_BLOCK_GROUP_METADATA;
6583 } else {
6584 flags = BTRFS_BLOCK_GROUP_DATA;
6585 }
6586
6587 space_info = __find_space_info(fs_info, flags);
6588 BUG_ON(!space_info); /* Logic bug */
6589 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
6590 }
6591
6592
6593 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6594 struct btrfs_root *root,
6595 struct btrfs_delayed_ref_node *node, u64 parent,
6596 u64 root_objectid, u64 owner_objectid,
6597 u64 owner_offset, int refs_to_drop,
6598 struct btrfs_delayed_extent_op *extent_op)
6599 {
6600 struct btrfs_key key;
6601 struct btrfs_path *path;
6602 struct btrfs_fs_info *info = root->fs_info;
6603 struct btrfs_root *extent_root = info->extent_root;
6604 struct extent_buffer *leaf;
6605 struct btrfs_extent_item *ei;
6606 struct btrfs_extent_inline_ref *iref;
6607 int ret;
6608 int is_data;
6609 int extent_slot = 0;
6610 int found_extent = 0;
6611 int num_to_del = 1;
6612 u32 item_size;
6613 u64 refs;
6614 u64 bytenr = node->bytenr;
6615 u64 num_bytes = node->num_bytes;
6616 int last_ref = 0;
6617 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6618 SKINNY_METADATA);
6619
6620 path = btrfs_alloc_path();
6621 if (!path)
6622 return -ENOMEM;
6623
6624 path->reada = READA_FORWARD;
6625 path->leave_spinning = 1;
6626
6627 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6628 BUG_ON(!is_data && refs_to_drop != 1);
6629
6630 if (is_data)
6631 skinny_metadata = 0;
6632
6633 ret = lookup_extent_backref(trans, extent_root, path, &iref,
6634 bytenr, num_bytes, parent,
6635 root_objectid, owner_objectid,
6636 owner_offset);
6637 if (ret == 0) {
6638 extent_slot = path->slots[0];
6639 while (extent_slot >= 0) {
6640 btrfs_item_key_to_cpu(path->nodes[0], &key,
6641 extent_slot);
6642 if (key.objectid != bytenr)
6643 break;
6644 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6645 key.offset == num_bytes) {
6646 found_extent = 1;
6647 break;
6648 }
6649 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6650 key.offset == owner_objectid) {
6651 found_extent = 1;
6652 break;
6653 }
6654 if (path->slots[0] - extent_slot > 5)
6655 break;
6656 extent_slot--;
6657 }
6658 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6659 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6660 if (found_extent && item_size < sizeof(*ei))
6661 found_extent = 0;
6662 #endif
6663 if (!found_extent) {
6664 BUG_ON(iref);
6665 ret = remove_extent_backref(trans, extent_root, path,
6666 NULL, refs_to_drop,
6667 is_data, &last_ref);
6668 if (ret) {
6669 btrfs_abort_transaction(trans, extent_root, ret);
6670 goto out;
6671 }
6672 btrfs_release_path(path);
6673 path->leave_spinning = 1;
6674
6675 key.objectid = bytenr;
6676 key.type = BTRFS_EXTENT_ITEM_KEY;
6677 key.offset = num_bytes;
6678
6679 if (!is_data && skinny_metadata) {
6680 key.type = BTRFS_METADATA_ITEM_KEY;
6681 key.offset = owner_objectid;
6682 }
6683
6684 ret = btrfs_search_slot(trans, extent_root,
6685 &key, path, -1, 1);
6686 if (ret > 0 && skinny_metadata && path->slots[0]) {
6687 /*
6688 * Couldn't find our skinny metadata item,
6689 * see if we have ye olde extent item.
6690 */
6691 path->slots[0]--;
6692 btrfs_item_key_to_cpu(path->nodes[0], &key,
6693 path->slots[0]);
6694 if (key.objectid == bytenr &&
6695 key.type == BTRFS_EXTENT_ITEM_KEY &&
6696 key.offset == num_bytes)
6697 ret = 0;
6698 }
6699
6700 if (ret > 0 && skinny_metadata) {
6701 skinny_metadata = false;
6702 key.objectid = bytenr;
6703 key.type = BTRFS_EXTENT_ITEM_KEY;
6704 key.offset = num_bytes;
6705 btrfs_release_path(path);
6706 ret = btrfs_search_slot(trans, extent_root,
6707 &key, path, -1, 1);
6708 }
6709
6710 if (ret) {
6711 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6712 ret, bytenr);
6713 if (ret > 0)
6714 btrfs_print_leaf(extent_root,
6715 path->nodes[0]);
6716 }
6717 if (ret < 0) {
6718 btrfs_abort_transaction(trans, extent_root, ret);
6719 goto out;
6720 }
6721 extent_slot = path->slots[0];
6722 }
6723 } else if (WARN_ON(ret == -ENOENT)) {
6724 btrfs_print_leaf(extent_root, path->nodes[0]);
6725 btrfs_err(info,
6726 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6727 bytenr, parent, root_objectid, owner_objectid,
6728 owner_offset);
6729 btrfs_abort_transaction(trans, extent_root, ret);
6730 goto out;
6731 } else {
6732 btrfs_abort_transaction(trans, extent_root, ret);
6733 goto out;
6734 }
6735
6736 leaf = path->nodes[0];
6737 item_size = btrfs_item_size_nr(leaf, extent_slot);
6738 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6739 if (item_size < sizeof(*ei)) {
6740 BUG_ON(found_extent || extent_slot != path->slots[0]);
6741 ret = convert_extent_item_v0(trans, extent_root, path,
6742 owner_objectid, 0);
6743 if (ret < 0) {
6744 btrfs_abort_transaction(trans, extent_root, ret);
6745 goto out;
6746 }
6747
6748 btrfs_release_path(path);
6749 path->leave_spinning = 1;
6750
6751 key.objectid = bytenr;
6752 key.type = BTRFS_EXTENT_ITEM_KEY;
6753 key.offset = num_bytes;
6754
6755 ret = btrfs_search_slot(trans, extent_root, &key, path,
6756 -1, 1);
6757 if (ret) {
6758 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6759 ret, bytenr);
6760 btrfs_print_leaf(extent_root, path->nodes[0]);
6761 }
6762 if (ret < 0) {
6763 btrfs_abort_transaction(trans, extent_root, ret);
6764 goto out;
6765 }
6766
6767 extent_slot = path->slots[0];
6768 leaf = path->nodes[0];
6769 item_size = btrfs_item_size_nr(leaf, extent_slot);
6770 }
6771 #endif
6772 BUG_ON(item_size < sizeof(*ei));
6773 ei = btrfs_item_ptr(leaf, extent_slot,
6774 struct btrfs_extent_item);
6775 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6776 key.type == BTRFS_EXTENT_ITEM_KEY) {
6777 struct btrfs_tree_block_info *bi;
6778 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6779 bi = (struct btrfs_tree_block_info *)(ei + 1);
6780 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6781 }
6782
6783 refs = btrfs_extent_refs(leaf, ei);
6784 if (refs < refs_to_drop) {
6785 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
6786 "for bytenr %Lu", refs_to_drop, refs, bytenr);
6787 ret = -EINVAL;
6788 btrfs_abort_transaction(trans, extent_root, ret);
6789 goto out;
6790 }
6791 refs -= refs_to_drop;
6792
6793 if (refs > 0) {
6794 if (extent_op)
6795 __run_delayed_extent_op(extent_op, leaf, ei);
6796 /*
6797 * In the case of inline back ref, reference count will
6798 * be updated by remove_extent_backref
6799 */
6800 if (iref) {
6801 BUG_ON(!found_extent);
6802 } else {
6803 btrfs_set_extent_refs(leaf, ei, refs);
6804 btrfs_mark_buffer_dirty(leaf);
6805 }
6806 if (found_extent) {
6807 ret = remove_extent_backref(trans, extent_root, path,
6808 iref, refs_to_drop,
6809 is_data, &last_ref);
6810 if (ret) {
6811 btrfs_abort_transaction(trans, extent_root, ret);
6812 goto out;
6813 }
6814 }
6815 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
6816 root_objectid);
6817 } else {
6818 if (found_extent) {
6819 BUG_ON(is_data && refs_to_drop !=
6820 extent_data_ref_count(path, iref));
6821 if (iref) {
6822 BUG_ON(path->slots[0] != extent_slot);
6823 } else {
6824 BUG_ON(path->slots[0] != extent_slot + 1);
6825 path->slots[0] = extent_slot;
6826 num_to_del = 2;
6827 }
6828 }
6829
6830 last_ref = 1;
6831 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6832 num_to_del);
6833 if (ret) {
6834 btrfs_abort_transaction(trans, extent_root, ret);
6835 goto out;
6836 }
6837 btrfs_release_path(path);
6838
6839 if (is_data) {
6840 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
6841 if (ret) {
6842 btrfs_abort_transaction(trans, extent_root, ret);
6843 goto out;
6844 }
6845 }
6846
6847 ret = add_to_free_space_tree(trans, root->fs_info, bytenr,
6848 num_bytes);
6849 if (ret) {
6850 btrfs_abort_transaction(trans, extent_root, ret);
6851 goto out;
6852 }
6853
6854 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
6855 if (ret) {
6856 btrfs_abort_transaction(trans, extent_root, ret);
6857 goto out;
6858 }
6859 }
6860 btrfs_release_path(path);
6861
6862 out:
6863 btrfs_free_path(path);
6864 return ret;
6865 }
6866
6867 /*
6868 * when we free an block, it is possible (and likely) that we free the last
6869 * delayed ref for that extent as well. This searches the delayed ref tree for
6870 * a given extent, and if there are no other delayed refs to be processed, it
6871 * removes it from the tree.
6872 */
6873 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6874 struct btrfs_root *root, u64 bytenr)
6875 {
6876 struct btrfs_delayed_ref_head *head;
6877 struct btrfs_delayed_ref_root *delayed_refs;
6878 int ret = 0;
6879
6880 delayed_refs = &trans->transaction->delayed_refs;
6881 spin_lock(&delayed_refs->lock);
6882 head = btrfs_find_delayed_ref_head(trans, bytenr);
6883 if (!head)
6884 goto out_delayed_unlock;
6885
6886 spin_lock(&head->lock);
6887 if (!list_empty(&head->ref_list))
6888 goto out;
6889
6890 if (head->extent_op) {
6891 if (!head->must_insert_reserved)
6892 goto out;
6893 btrfs_free_delayed_extent_op(head->extent_op);
6894 head->extent_op = NULL;
6895 }
6896
6897 /*
6898 * waiting for the lock here would deadlock. If someone else has it
6899 * locked they are already in the process of dropping it anyway
6900 */
6901 if (!mutex_trylock(&head->mutex))
6902 goto out;
6903
6904 /*
6905 * at this point we have a head with no other entries. Go
6906 * ahead and process it.
6907 */
6908 head->node.in_tree = 0;
6909 rb_erase(&head->href_node, &delayed_refs->href_root);
6910
6911 atomic_dec(&delayed_refs->num_entries);
6912
6913 /*
6914 * we don't take a ref on the node because we're removing it from the
6915 * tree, so we just steal the ref the tree was holding.
6916 */
6917 delayed_refs->num_heads--;
6918 if (head->processing == 0)
6919 delayed_refs->num_heads_ready--;
6920 head->processing = 0;
6921 spin_unlock(&head->lock);
6922 spin_unlock(&delayed_refs->lock);
6923
6924 BUG_ON(head->extent_op);
6925 if (head->must_insert_reserved)
6926 ret = 1;
6927
6928 mutex_unlock(&head->mutex);
6929 btrfs_put_delayed_ref(&head->node);
6930 return ret;
6931 out:
6932 spin_unlock(&head->lock);
6933
6934 out_delayed_unlock:
6935 spin_unlock(&delayed_refs->lock);
6936 return 0;
6937 }
6938
6939 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6940 struct btrfs_root *root,
6941 struct extent_buffer *buf,
6942 u64 parent, int last_ref)
6943 {
6944 int pin = 1;
6945 int ret;
6946
6947 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6948 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6949 buf->start, buf->len,
6950 parent, root->root_key.objectid,
6951 btrfs_header_level(buf),
6952 BTRFS_DROP_DELAYED_REF, NULL);
6953 BUG_ON(ret); /* -ENOMEM */
6954 }
6955
6956 if (!last_ref)
6957 return;
6958
6959 if (btrfs_header_generation(buf) == trans->transid) {
6960 struct btrfs_block_group_cache *cache;
6961
6962 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6963 ret = check_ref_cleanup(trans, root, buf->start);
6964 if (!ret)
6965 goto out;
6966 }
6967
6968 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6969
6970 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6971 pin_down_extent(root, cache, buf->start, buf->len, 1);
6972 btrfs_put_block_group(cache);
6973 goto out;
6974 }
6975
6976 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6977
6978 btrfs_add_free_space(cache, buf->start, buf->len);
6979 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
6980 btrfs_put_block_group(cache);
6981 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6982 pin = 0;
6983 }
6984 out:
6985 if (pin)
6986 add_pinned_bytes(root->fs_info, buf->len,
6987 btrfs_header_level(buf),
6988 root->root_key.objectid);
6989
6990 /*
6991 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6992 * anymore.
6993 */
6994 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6995 }
6996
6997 /* Can return -ENOMEM */
6998 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6999 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
7000 u64 owner, u64 offset)
7001 {
7002 int ret;
7003 struct btrfs_fs_info *fs_info = root->fs_info;
7004
7005 if (btrfs_test_is_dummy_root(root))
7006 return 0;
7007
7008 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
7009
7010 /*
7011 * tree log blocks never actually go into the extent allocation
7012 * tree, just update pinning info and exit early.
7013 */
7014 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
7015 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
7016 /* unlocks the pinned mutex */
7017 btrfs_pin_extent(root, bytenr, num_bytes, 1);
7018 ret = 0;
7019 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
7020 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
7021 num_bytes,
7022 parent, root_objectid, (int)owner,
7023 BTRFS_DROP_DELAYED_REF, NULL);
7024 } else {
7025 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
7026 num_bytes,
7027 parent, root_objectid, owner,
7028 offset, 0,
7029 BTRFS_DROP_DELAYED_REF, NULL);
7030 }
7031 return ret;
7032 }
7033
7034 /*
7035 * when we wait for progress in the block group caching, its because
7036 * our allocation attempt failed at least once. So, we must sleep
7037 * and let some progress happen before we try again.
7038 *
7039 * This function will sleep at least once waiting for new free space to
7040 * show up, and then it will check the block group free space numbers
7041 * for our min num_bytes. Another option is to have it go ahead
7042 * and look in the rbtree for a free extent of a given size, but this
7043 * is a good start.
7044 *
7045 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7046 * any of the information in this block group.
7047 */
7048 static noinline void
7049 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
7050 u64 num_bytes)
7051 {
7052 struct btrfs_caching_control *caching_ctl;
7053
7054 caching_ctl = get_caching_control(cache);
7055 if (!caching_ctl)
7056 return;
7057
7058 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
7059 (cache->free_space_ctl->free_space >= num_bytes));
7060
7061 put_caching_control(caching_ctl);
7062 }
7063
7064 static noinline int
7065 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
7066 {
7067 struct btrfs_caching_control *caching_ctl;
7068 int ret = 0;
7069
7070 caching_ctl = get_caching_control(cache);
7071 if (!caching_ctl)
7072 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
7073
7074 wait_event(caching_ctl->wait, block_group_cache_done(cache));
7075 if (cache->cached == BTRFS_CACHE_ERROR)
7076 ret = -EIO;
7077 put_caching_control(caching_ctl);
7078 return ret;
7079 }
7080
7081 int __get_raid_index(u64 flags)
7082 {
7083 if (flags & BTRFS_BLOCK_GROUP_RAID10)
7084 return BTRFS_RAID_RAID10;
7085 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
7086 return BTRFS_RAID_RAID1;
7087 else if (flags & BTRFS_BLOCK_GROUP_DUP)
7088 return BTRFS_RAID_DUP;
7089 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
7090 return BTRFS_RAID_RAID0;
7091 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
7092 return BTRFS_RAID_RAID5;
7093 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
7094 return BTRFS_RAID_RAID6;
7095
7096 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
7097 }
7098
7099 int get_block_group_index(struct btrfs_block_group_cache *cache)
7100 {
7101 return __get_raid_index(cache->flags);
7102 }
7103
7104 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
7105 [BTRFS_RAID_RAID10] = "raid10",
7106 [BTRFS_RAID_RAID1] = "raid1",
7107 [BTRFS_RAID_DUP] = "dup",
7108 [BTRFS_RAID_RAID0] = "raid0",
7109 [BTRFS_RAID_SINGLE] = "single",
7110 [BTRFS_RAID_RAID5] = "raid5",
7111 [BTRFS_RAID_RAID6] = "raid6",
7112 };
7113
7114 static const char *get_raid_name(enum btrfs_raid_types type)
7115 {
7116 if (type >= BTRFS_NR_RAID_TYPES)
7117 return NULL;
7118
7119 return btrfs_raid_type_names[type];
7120 }
7121
7122 enum btrfs_loop_type {
7123 LOOP_CACHING_NOWAIT = 0,
7124 LOOP_CACHING_WAIT = 1,
7125 LOOP_ALLOC_CHUNK = 2,
7126 LOOP_NO_EMPTY_SIZE = 3,
7127 };
7128
7129 static inline void
7130 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7131 int delalloc)
7132 {
7133 if (delalloc)
7134 down_read(&cache->data_rwsem);
7135 }
7136
7137 static inline void
7138 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7139 int delalloc)
7140 {
7141 btrfs_get_block_group(cache);
7142 if (delalloc)
7143 down_read(&cache->data_rwsem);
7144 }
7145
7146 static struct btrfs_block_group_cache *
7147 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7148 struct btrfs_free_cluster *cluster,
7149 int delalloc)
7150 {
7151 struct btrfs_block_group_cache *used_bg = NULL;
7152
7153 spin_lock(&cluster->refill_lock);
7154 while (1) {
7155 used_bg = cluster->block_group;
7156 if (!used_bg)
7157 return NULL;
7158
7159 if (used_bg == block_group)
7160 return used_bg;
7161
7162 btrfs_get_block_group(used_bg);
7163
7164 if (!delalloc)
7165 return used_bg;
7166
7167 if (down_read_trylock(&used_bg->data_rwsem))
7168 return used_bg;
7169
7170 spin_unlock(&cluster->refill_lock);
7171
7172 down_read(&used_bg->data_rwsem);
7173
7174 spin_lock(&cluster->refill_lock);
7175 if (used_bg == cluster->block_group)
7176 return used_bg;
7177
7178 up_read(&used_bg->data_rwsem);
7179 btrfs_put_block_group(used_bg);
7180 }
7181 }
7182
7183 static inline void
7184 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7185 int delalloc)
7186 {
7187 if (delalloc)
7188 up_read(&cache->data_rwsem);
7189 btrfs_put_block_group(cache);
7190 }
7191
7192 /*
7193 * walks the btree of allocated extents and find a hole of a given size.
7194 * The key ins is changed to record the hole:
7195 * ins->objectid == start position
7196 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7197 * ins->offset == the size of the hole.
7198 * Any available blocks before search_start are skipped.
7199 *
7200 * If there is no suitable free space, we will record the max size of
7201 * the free space extent currently.
7202 */
7203 static noinline int find_free_extent(struct btrfs_root *orig_root,
7204 u64 num_bytes, u64 empty_size,
7205 u64 hint_byte, struct btrfs_key *ins,
7206 u64 flags, int delalloc)
7207 {
7208 int ret = 0;
7209 struct btrfs_root *root = orig_root->fs_info->extent_root;
7210 struct btrfs_free_cluster *last_ptr = NULL;
7211 struct btrfs_block_group_cache *block_group = NULL;
7212 u64 search_start = 0;
7213 u64 max_extent_size = 0;
7214 u64 empty_cluster = 0;
7215 struct btrfs_space_info *space_info;
7216 int loop = 0;
7217 int index = __get_raid_index(flags);
7218 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
7219 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
7220 bool failed_cluster_refill = false;
7221 bool failed_alloc = false;
7222 bool use_cluster = true;
7223 bool have_caching_bg = false;
7224 bool orig_have_caching_bg = false;
7225 bool full_search = false;
7226
7227 WARN_ON(num_bytes < root->sectorsize);
7228 ins->type = BTRFS_EXTENT_ITEM_KEY;
7229 ins->objectid = 0;
7230 ins->offset = 0;
7231
7232 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
7233
7234 space_info = __find_space_info(root->fs_info, flags);
7235 if (!space_info) {
7236 btrfs_err(root->fs_info, "No space info for %llu", flags);
7237 return -ENOSPC;
7238 }
7239
7240 /*
7241 * If our free space is heavily fragmented we may not be able to make
7242 * big contiguous allocations, so instead of doing the expensive search
7243 * for free space, simply return ENOSPC with our max_extent_size so we
7244 * can go ahead and search for a more manageable chunk.
7245 *
7246 * If our max_extent_size is large enough for our allocation simply
7247 * disable clustering since we will likely not be able to find enough
7248 * space to create a cluster and induce latency trying.
7249 */
7250 if (unlikely(space_info->max_extent_size)) {
7251 spin_lock(&space_info->lock);
7252 if (space_info->max_extent_size &&
7253 num_bytes > space_info->max_extent_size) {
7254 ins->offset = space_info->max_extent_size;
7255 spin_unlock(&space_info->lock);
7256 return -ENOSPC;
7257 } else if (space_info->max_extent_size) {
7258 use_cluster = false;
7259 }
7260 spin_unlock(&space_info->lock);
7261 }
7262
7263 last_ptr = fetch_cluster_info(orig_root, space_info, &empty_cluster);
7264 if (last_ptr) {
7265 spin_lock(&last_ptr->lock);
7266 if (last_ptr->block_group)
7267 hint_byte = last_ptr->window_start;
7268 if (last_ptr->fragmented) {
7269 /*
7270 * We still set window_start so we can keep track of the
7271 * last place we found an allocation to try and save
7272 * some time.
7273 */
7274 hint_byte = last_ptr->window_start;
7275 use_cluster = false;
7276 }
7277 spin_unlock(&last_ptr->lock);
7278 }
7279
7280 search_start = max(search_start, first_logical_byte(root, 0));
7281 search_start = max(search_start, hint_byte);
7282 if (search_start == hint_byte) {
7283 block_group = btrfs_lookup_block_group(root->fs_info,
7284 search_start);
7285 /*
7286 * we don't want to use the block group if it doesn't match our
7287 * allocation bits, or if its not cached.
7288 *
7289 * However if we are re-searching with an ideal block group
7290 * picked out then we don't care that the block group is cached.
7291 */
7292 if (block_group && block_group_bits(block_group, flags) &&
7293 block_group->cached != BTRFS_CACHE_NO) {
7294 down_read(&space_info->groups_sem);
7295 if (list_empty(&block_group->list) ||
7296 block_group->ro) {
7297 /*
7298 * someone is removing this block group,
7299 * we can't jump into the have_block_group
7300 * target because our list pointers are not
7301 * valid
7302 */
7303 btrfs_put_block_group(block_group);
7304 up_read(&space_info->groups_sem);
7305 } else {
7306 index = get_block_group_index(block_group);
7307 btrfs_lock_block_group(block_group, delalloc);
7308 goto have_block_group;
7309 }
7310 } else if (block_group) {
7311 btrfs_put_block_group(block_group);
7312 }
7313 }
7314 search:
7315 have_caching_bg = false;
7316 if (index == 0 || index == __get_raid_index(flags))
7317 full_search = true;
7318 down_read(&space_info->groups_sem);
7319 list_for_each_entry(block_group, &space_info->block_groups[index],
7320 list) {
7321 u64 offset;
7322 int cached;
7323
7324 btrfs_grab_block_group(block_group, delalloc);
7325 search_start = block_group->key.objectid;
7326
7327 /*
7328 * this can happen if we end up cycling through all the
7329 * raid types, but we want to make sure we only allocate
7330 * for the proper type.
7331 */
7332 if (!block_group_bits(block_group, flags)) {
7333 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7334 BTRFS_BLOCK_GROUP_RAID1 |
7335 BTRFS_BLOCK_GROUP_RAID5 |
7336 BTRFS_BLOCK_GROUP_RAID6 |
7337 BTRFS_BLOCK_GROUP_RAID10;
7338
7339 /*
7340 * if they asked for extra copies and this block group
7341 * doesn't provide them, bail. This does allow us to
7342 * fill raid0 from raid1.
7343 */
7344 if ((flags & extra) && !(block_group->flags & extra))
7345 goto loop;
7346 }
7347
7348 have_block_group:
7349 cached = block_group_cache_done(block_group);
7350 if (unlikely(!cached)) {
7351 have_caching_bg = true;
7352 ret = cache_block_group(block_group, 0);
7353 BUG_ON(ret < 0);
7354 ret = 0;
7355 }
7356
7357 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7358 goto loop;
7359 if (unlikely(block_group->ro))
7360 goto loop;
7361
7362 /*
7363 * Ok we want to try and use the cluster allocator, so
7364 * lets look there
7365 */
7366 if (last_ptr && use_cluster) {
7367 struct btrfs_block_group_cache *used_block_group;
7368 unsigned long aligned_cluster;
7369 /*
7370 * the refill lock keeps out other
7371 * people trying to start a new cluster
7372 */
7373 used_block_group = btrfs_lock_cluster(block_group,
7374 last_ptr,
7375 delalloc);
7376 if (!used_block_group)
7377 goto refill_cluster;
7378
7379 if (used_block_group != block_group &&
7380 (used_block_group->ro ||
7381 !block_group_bits(used_block_group, flags)))
7382 goto release_cluster;
7383
7384 offset = btrfs_alloc_from_cluster(used_block_group,
7385 last_ptr,
7386 num_bytes,
7387 used_block_group->key.objectid,
7388 &max_extent_size);
7389 if (offset) {
7390 /* we have a block, we're done */
7391 spin_unlock(&last_ptr->refill_lock);
7392 trace_btrfs_reserve_extent_cluster(root,
7393 used_block_group,
7394 search_start, num_bytes);
7395 if (used_block_group != block_group) {
7396 btrfs_release_block_group(block_group,
7397 delalloc);
7398 block_group = used_block_group;
7399 }
7400 goto checks;
7401 }
7402
7403 WARN_ON(last_ptr->block_group != used_block_group);
7404 release_cluster:
7405 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7406 * set up a new clusters, so lets just skip it
7407 * and let the allocator find whatever block
7408 * it can find. If we reach this point, we
7409 * will have tried the cluster allocator
7410 * plenty of times and not have found
7411 * anything, so we are likely way too
7412 * fragmented for the clustering stuff to find
7413 * anything.
7414 *
7415 * However, if the cluster is taken from the
7416 * current block group, release the cluster
7417 * first, so that we stand a better chance of
7418 * succeeding in the unclustered
7419 * allocation. */
7420 if (loop >= LOOP_NO_EMPTY_SIZE &&
7421 used_block_group != block_group) {
7422 spin_unlock(&last_ptr->refill_lock);
7423 btrfs_release_block_group(used_block_group,
7424 delalloc);
7425 goto unclustered_alloc;
7426 }
7427
7428 /*
7429 * this cluster didn't work out, free it and
7430 * start over
7431 */
7432 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7433
7434 if (used_block_group != block_group)
7435 btrfs_release_block_group(used_block_group,
7436 delalloc);
7437 refill_cluster:
7438 if (loop >= LOOP_NO_EMPTY_SIZE) {
7439 spin_unlock(&last_ptr->refill_lock);
7440 goto unclustered_alloc;
7441 }
7442
7443 aligned_cluster = max_t(unsigned long,
7444 empty_cluster + empty_size,
7445 block_group->full_stripe_len);
7446
7447 /* allocate a cluster in this block group */
7448 ret = btrfs_find_space_cluster(root, block_group,
7449 last_ptr, search_start,
7450 num_bytes,
7451 aligned_cluster);
7452 if (ret == 0) {
7453 /*
7454 * now pull our allocation out of this
7455 * cluster
7456 */
7457 offset = btrfs_alloc_from_cluster(block_group,
7458 last_ptr,
7459 num_bytes,
7460 search_start,
7461 &max_extent_size);
7462 if (offset) {
7463 /* we found one, proceed */
7464 spin_unlock(&last_ptr->refill_lock);
7465 trace_btrfs_reserve_extent_cluster(root,
7466 block_group, search_start,
7467 num_bytes);
7468 goto checks;
7469 }
7470 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7471 && !failed_cluster_refill) {
7472 spin_unlock(&last_ptr->refill_lock);
7473
7474 failed_cluster_refill = true;
7475 wait_block_group_cache_progress(block_group,
7476 num_bytes + empty_cluster + empty_size);
7477 goto have_block_group;
7478 }
7479
7480 /*
7481 * at this point we either didn't find a cluster
7482 * or we weren't able to allocate a block from our
7483 * cluster. Free the cluster we've been trying
7484 * to use, and go to the next block group
7485 */
7486 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7487 spin_unlock(&last_ptr->refill_lock);
7488 goto loop;
7489 }
7490
7491 unclustered_alloc:
7492 /*
7493 * We are doing an unclustered alloc, set the fragmented flag so
7494 * we don't bother trying to setup a cluster again until we get
7495 * more space.
7496 */
7497 if (unlikely(last_ptr)) {
7498 spin_lock(&last_ptr->lock);
7499 last_ptr->fragmented = 1;
7500 spin_unlock(&last_ptr->lock);
7501 }
7502 spin_lock(&block_group->free_space_ctl->tree_lock);
7503 if (cached &&
7504 block_group->free_space_ctl->free_space <
7505 num_bytes + empty_cluster + empty_size) {
7506 if (block_group->free_space_ctl->free_space >
7507 max_extent_size)
7508 max_extent_size =
7509 block_group->free_space_ctl->free_space;
7510 spin_unlock(&block_group->free_space_ctl->tree_lock);
7511 goto loop;
7512 }
7513 spin_unlock(&block_group->free_space_ctl->tree_lock);
7514
7515 offset = btrfs_find_space_for_alloc(block_group, search_start,
7516 num_bytes, empty_size,
7517 &max_extent_size);
7518 /*
7519 * If we didn't find a chunk, and we haven't failed on this
7520 * block group before, and this block group is in the middle of
7521 * caching and we are ok with waiting, then go ahead and wait
7522 * for progress to be made, and set failed_alloc to true.
7523 *
7524 * If failed_alloc is true then we've already waited on this
7525 * block group once and should move on to the next block group.
7526 */
7527 if (!offset && !failed_alloc && !cached &&
7528 loop > LOOP_CACHING_NOWAIT) {
7529 wait_block_group_cache_progress(block_group,
7530 num_bytes + empty_size);
7531 failed_alloc = true;
7532 goto have_block_group;
7533 } else if (!offset) {
7534 goto loop;
7535 }
7536 checks:
7537 search_start = ALIGN(offset, root->stripesize);
7538
7539 /* move on to the next group */
7540 if (search_start + num_bytes >
7541 block_group->key.objectid + block_group->key.offset) {
7542 btrfs_add_free_space(block_group, offset, num_bytes);
7543 goto loop;
7544 }
7545
7546 if (offset < search_start)
7547 btrfs_add_free_space(block_group, offset,
7548 search_start - offset);
7549 BUG_ON(offset > search_start);
7550
7551 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
7552 alloc_type, delalloc);
7553 if (ret == -EAGAIN) {
7554 btrfs_add_free_space(block_group, offset, num_bytes);
7555 goto loop;
7556 }
7557 btrfs_inc_block_group_reservations(block_group);
7558
7559 /* we are all good, lets return */
7560 ins->objectid = search_start;
7561 ins->offset = num_bytes;
7562
7563 trace_btrfs_reserve_extent(orig_root, block_group,
7564 search_start, num_bytes);
7565 btrfs_release_block_group(block_group, delalloc);
7566 break;
7567 loop:
7568 failed_cluster_refill = false;
7569 failed_alloc = false;
7570 BUG_ON(index != get_block_group_index(block_group));
7571 btrfs_release_block_group(block_group, delalloc);
7572 }
7573 up_read(&space_info->groups_sem);
7574
7575 if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg
7576 && !orig_have_caching_bg)
7577 orig_have_caching_bg = true;
7578
7579 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7580 goto search;
7581
7582 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7583 goto search;
7584
7585 /*
7586 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7587 * caching kthreads as we move along
7588 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7589 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7590 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7591 * again
7592 */
7593 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7594 index = 0;
7595 if (loop == LOOP_CACHING_NOWAIT) {
7596 /*
7597 * We want to skip the LOOP_CACHING_WAIT step if we
7598 * don't have any uncached bgs and we've already done a
7599 * full search through.
7600 */
7601 if (orig_have_caching_bg || !full_search)
7602 loop = LOOP_CACHING_WAIT;
7603 else
7604 loop = LOOP_ALLOC_CHUNK;
7605 } else {
7606 loop++;
7607 }
7608
7609 if (loop == LOOP_ALLOC_CHUNK) {
7610 struct btrfs_trans_handle *trans;
7611 int exist = 0;
7612
7613 trans = current->journal_info;
7614 if (trans)
7615 exist = 1;
7616 else
7617 trans = btrfs_join_transaction(root);
7618
7619 if (IS_ERR(trans)) {
7620 ret = PTR_ERR(trans);
7621 goto out;
7622 }
7623
7624 ret = do_chunk_alloc(trans, root, flags,
7625 CHUNK_ALLOC_FORCE);
7626
7627 /*
7628 * If we can't allocate a new chunk we've already looped
7629 * through at least once, move on to the NO_EMPTY_SIZE
7630 * case.
7631 */
7632 if (ret == -ENOSPC)
7633 loop = LOOP_NO_EMPTY_SIZE;
7634
7635 /*
7636 * Do not bail out on ENOSPC since we
7637 * can do more things.
7638 */
7639 if (ret < 0 && ret != -ENOSPC)
7640 btrfs_abort_transaction(trans,
7641 root, ret);
7642 else
7643 ret = 0;
7644 if (!exist)
7645 btrfs_end_transaction(trans, root);
7646 if (ret)
7647 goto out;
7648 }
7649
7650 if (loop == LOOP_NO_EMPTY_SIZE) {
7651 /*
7652 * Don't loop again if we already have no empty_size and
7653 * no empty_cluster.
7654 */
7655 if (empty_size == 0 &&
7656 empty_cluster == 0) {
7657 ret = -ENOSPC;
7658 goto out;
7659 }
7660 empty_size = 0;
7661 empty_cluster = 0;
7662 }
7663
7664 goto search;
7665 } else if (!ins->objectid) {
7666 ret = -ENOSPC;
7667 } else if (ins->objectid) {
7668 if (!use_cluster && last_ptr) {
7669 spin_lock(&last_ptr->lock);
7670 last_ptr->window_start = ins->objectid;
7671 spin_unlock(&last_ptr->lock);
7672 }
7673 ret = 0;
7674 }
7675 out:
7676 if (ret == -ENOSPC) {
7677 spin_lock(&space_info->lock);
7678 space_info->max_extent_size = max_extent_size;
7679 spin_unlock(&space_info->lock);
7680 ins->offset = max_extent_size;
7681 }
7682 return ret;
7683 }
7684
7685 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
7686 int dump_block_groups)
7687 {
7688 struct btrfs_block_group_cache *cache;
7689 int index = 0;
7690
7691 spin_lock(&info->lock);
7692 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
7693 info->flags,
7694 info->total_bytes - info->bytes_used - info->bytes_pinned -
7695 info->bytes_reserved - info->bytes_readonly,
7696 (info->full) ? "" : "not ");
7697 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7698 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7699 info->total_bytes, info->bytes_used, info->bytes_pinned,
7700 info->bytes_reserved, info->bytes_may_use,
7701 info->bytes_readonly);
7702 spin_unlock(&info->lock);
7703
7704 if (!dump_block_groups)
7705 return;
7706
7707 down_read(&info->groups_sem);
7708 again:
7709 list_for_each_entry(cache, &info->block_groups[index], list) {
7710 spin_lock(&cache->lock);
7711 printk(KERN_INFO "BTRFS: "
7712 "block group %llu has %llu bytes, "
7713 "%llu used %llu pinned %llu reserved %s\n",
7714 cache->key.objectid, cache->key.offset,
7715 btrfs_block_group_used(&cache->item), cache->pinned,
7716 cache->reserved, cache->ro ? "[readonly]" : "");
7717 btrfs_dump_free_space(cache, bytes);
7718 spin_unlock(&cache->lock);
7719 }
7720 if (++index < BTRFS_NR_RAID_TYPES)
7721 goto again;
7722 up_read(&info->groups_sem);
7723 }
7724
7725 int btrfs_reserve_extent(struct btrfs_root *root,
7726 u64 num_bytes, u64 min_alloc_size,
7727 u64 empty_size, u64 hint_byte,
7728 struct btrfs_key *ins, int is_data, int delalloc)
7729 {
7730 bool final_tried = num_bytes == min_alloc_size;
7731 u64 flags;
7732 int ret;
7733
7734 flags = btrfs_get_alloc_profile(root, is_data);
7735 again:
7736 WARN_ON(num_bytes < root->sectorsize);
7737 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
7738 flags, delalloc);
7739 if (!ret && !is_data) {
7740 btrfs_dec_block_group_reservations(root->fs_info,
7741 ins->objectid);
7742 } else if (ret == -ENOSPC) {
7743 if (!final_tried && ins->offset) {
7744 num_bytes = min(num_bytes >> 1, ins->offset);
7745 num_bytes = round_down(num_bytes, root->sectorsize);
7746 num_bytes = max(num_bytes, min_alloc_size);
7747 if (num_bytes == min_alloc_size)
7748 final_tried = true;
7749 goto again;
7750 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7751 struct btrfs_space_info *sinfo;
7752
7753 sinfo = __find_space_info(root->fs_info, flags);
7754 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
7755 flags, num_bytes);
7756 if (sinfo)
7757 dump_space_info(sinfo, num_bytes, 1);
7758 }
7759 }
7760
7761 return ret;
7762 }
7763
7764 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
7765 u64 start, u64 len,
7766 int pin, int delalloc)
7767 {
7768 struct btrfs_block_group_cache *cache;
7769 int ret = 0;
7770
7771 cache = btrfs_lookup_block_group(root->fs_info, start);
7772 if (!cache) {
7773 btrfs_err(root->fs_info, "Unable to find block group for %llu",
7774 start);
7775 return -ENOSPC;
7776 }
7777
7778 if (pin)
7779 pin_down_extent(root, cache, start, len, 1);
7780 else {
7781 if (btrfs_test_opt(root, DISCARD))
7782 ret = btrfs_discard_extent(root, start, len, NULL);
7783 btrfs_add_free_space(cache, start, len);
7784 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
7785 }
7786
7787 btrfs_put_block_group(cache);
7788
7789 trace_btrfs_reserved_extent_free(root, start, len);
7790
7791 return ret;
7792 }
7793
7794 int btrfs_free_reserved_extent(struct btrfs_root *root,
7795 u64 start, u64 len, int delalloc)
7796 {
7797 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
7798 }
7799
7800 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
7801 u64 start, u64 len)
7802 {
7803 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
7804 }
7805
7806 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7807 struct btrfs_root *root,
7808 u64 parent, u64 root_objectid,
7809 u64 flags, u64 owner, u64 offset,
7810 struct btrfs_key *ins, int ref_mod)
7811 {
7812 int ret;
7813 struct btrfs_fs_info *fs_info = root->fs_info;
7814 struct btrfs_extent_item *extent_item;
7815 struct btrfs_extent_inline_ref *iref;
7816 struct btrfs_path *path;
7817 struct extent_buffer *leaf;
7818 int type;
7819 u32 size;
7820
7821 if (parent > 0)
7822 type = BTRFS_SHARED_DATA_REF_KEY;
7823 else
7824 type = BTRFS_EXTENT_DATA_REF_KEY;
7825
7826 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7827
7828 path = btrfs_alloc_path();
7829 if (!path)
7830 return -ENOMEM;
7831
7832 path->leave_spinning = 1;
7833 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7834 ins, size);
7835 if (ret) {
7836 btrfs_free_path(path);
7837 return ret;
7838 }
7839
7840 leaf = path->nodes[0];
7841 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7842 struct btrfs_extent_item);
7843 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
7844 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7845 btrfs_set_extent_flags(leaf, extent_item,
7846 flags | BTRFS_EXTENT_FLAG_DATA);
7847
7848 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7849 btrfs_set_extent_inline_ref_type(leaf, iref, type);
7850 if (parent > 0) {
7851 struct btrfs_shared_data_ref *ref;
7852 ref = (struct btrfs_shared_data_ref *)(iref + 1);
7853 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7854 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
7855 } else {
7856 struct btrfs_extent_data_ref *ref;
7857 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
7858 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
7859 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
7860 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
7861 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
7862 }
7863
7864 btrfs_mark_buffer_dirty(path->nodes[0]);
7865 btrfs_free_path(path);
7866
7867 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
7868 ins->offset);
7869 if (ret)
7870 return ret;
7871
7872 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
7873 if (ret) { /* -ENOENT, logic error */
7874 btrfs_err(fs_info, "update block group failed for %llu %llu",
7875 ins->objectid, ins->offset);
7876 BUG();
7877 }
7878 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
7879 return ret;
7880 }
7881
7882 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
7883 struct btrfs_root *root,
7884 u64 parent, u64 root_objectid,
7885 u64 flags, struct btrfs_disk_key *key,
7886 int level, struct btrfs_key *ins)
7887 {
7888 int ret;
7889 struct btrfs_fs_info *fs_info = root->fs_info;
7890 struct btrfs_extent_item *extent_item;
7891 struct btrfs_tree_block_info *block_info;
7892 struct btrfs_extent_inline_ref *iref;
7893 struct btrfs_path *path;
7894 struct extent_buffer *leaf;
7895 u32 size = sizeof(*extent_item) + sizeof(*iref);
7896 u64 num_bytes = ins->offset;
7897 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7898 SKINNY_METADATA);
7899
7900 if (!skinny_metadata)
7901 size += sizeof(*block_info);
7902
7903 path = btrfs_alloc_path();
7904 if (!path) {
7905 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7906 root->nodesize);
7907 return -ENOMEM;
7908 }
7909
7910 path->leave_spinning = 1;
7911 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7912 ins, size);
7913 if (ret) {
7914 btrfs_free_path(path);
7915 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7916 root->nodesize);
7917 return ret;
7918 }
7919
7920 leaf = path->nodes[0];
7921 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7922 struct btrfs_extent_item);
7923 btrfs_set_extent_refs(leaf, extent_item, 1);
7924 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7925 btrfs_set_extent_flags(leaf, extent_item,
7926 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
7927
7928 if (skinny_metadata) {
7929 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7930 num_bytes = root->nodesize;
7931 } else {
7932 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
7933 btrfs_set_tree_block_key(leaf, block_info, key);
7934 btrfs_set_tree_block_level(leaf, block_info, level);
7935 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
7936 }
7937
7938 if (parent > 0) {
7939 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
7940 btrfs_set_extent_inline_ref_type(leaf, iref,
7941 BTRFS_SHARED_BLOCK_REF_KEY);
7942 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7943 } else {
7944 btrfs_set_extent_inline_ref_type(leaf, iref,
7945 BTRFS_TREE_BLOCK_REF_KEY);
7946 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
7947 }
7948
7949 btrfs_mark_buffer_dirty(leaf);
7950 btrfs_free_path(path);
7951
7952 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
7953 num_bytes);
7954 if (ret)
7955 return ret;
7956
7957 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
7958 1);
7959 if (ret) { /* -ENOENT, logic error */
7960 btrfs_err(fs_info, "update block group failed for %llu %llu",
7961 ins->objectid, ins->offset);
7962 BUG();
7963 }
7964
7965 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
7966 return ret;
7967 }
7968
7969 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7970 struct btrfs_root *root,
7971 u64 root_objectid, u64 owner,
7972 u64 offset, u64 ram_bytes,
7973 struct btrfs_key *ins)
7974 {
7975 int ret;
7976
7977 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
7978
7979 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
7980 ins->offset, 0,
7981 root_objectid, owner, offset,
7982 ram_bytes, BTRFS_ADD_DELAYED_EXTENT,
7983 NULL);
7984 return ret;
7985 }
7986
7987 /*
7988 * this is used by the tree logging recovery code. It records that
7989 * an extent has been allocated and makes sure to clear the free
7990 * space cache bits as well
7991 */
7992 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
7993 struct btrfs_root *root,
7994 u64 root_objectid, u64 owner, u64 offset,
7995 struct btrfs_key *ins)
7996 {
7997 int ret;
7998 struct btrfs_block_group_cache *block_group;
7999
8000 /*
8001 * Mixed block groups will exclude before processing the log so we only
8002 * need to do the exclude dance if this fs isn't mixed.
8003 */
8004 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
8005 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
8006 if (ret)
8007 return ret;
8008 }
8009
8010 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
8011 if (!block_group)
8012 return -EINVAL;
8013
8014 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
8015 RESERVE_ALLOC_NO_ACCOUNT, 0);
8016 BUG_ON(ret); /* logic error */
8017 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
8018 0, owner, offset, ins, 1);
8019 btrfs_put_block_group(block_group);
8020 return ret;
8021 }
8022
8023 static struct extent_buffer *
8024 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
8025 u64 bytenr, int level)
8026 {
8027 struct extent_buffer *buf;
8028
8029 buf = btrfs_find_create_tree_block(root, bytenr);
8030 if (IS_ERR(buf))
8031 return buf;
8032
8033 btrfs_set_header_generation(buf, trans->transid);
8034 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
8035 btrfs_tree_lock(buf);
8036 clean_tree_block(trans, root->fs_info, buf);
8037 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
8038
8039 btrfs_set_lock_blocking(buf);
8040 set_extent_buffer_uptodate(buf);
8041
8042 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8043 buf->log_index = root->log_transid % 2;
8044 /*
8045 * we allow two log transactions at a time, use different
8046 * EXENT bit to differentiate dirty pages.
8047 */
8048 if (buf->log_index == 0)
8049 set_extent_dirty(&root->dirty_log_pages, buf->start,
8050 buf->start + buf->len - 1, GFP_NOFS);
8051 else
8052 set_extent_new(&root->dirty_log_pages, buf->start,
8053 buf->start + buf->len - 1);
8054 } else {
8055 buf->log_index = -1;
8056 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
8057 buf->start + buf->len - 1, GFP_NOFS);
8058 }
8059 trans->dirty = true;
8060 /* this returns a buffer locked for blocking */
8061 return buf;
8062 }
8063
8064 static struct btrfs_block_rsv *
8065 use_block_rsv(struct btrfs_trans_handle *trans,
8066 struct btrfs_root *root, u32 blocksize)
8067 {
8068 struct btrfs_block_rsv *block_rsv;
8069 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
8070 int ret;
8071 bool global_updated = false;
8072
8073 block_rsv = get_block_rsv(trans, root);
8074
8075 if (unlikely(block_rsv->size == 0))
8076 goto try_reserve;
8077 again:
8078 ret = block_rsv_use_bytes(block_rsv, blocksize);
8079 if (!ret)
8080 return block_rsv;
8081
8082 if (block_rsv->failfast)
8083 return ERR_PTR(ret);
8084
8085 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
8086 global_updated = true;
8087 update_global_block_rsv(root->fs_info);
8088 goto again;
8089 }
8090
8091 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
8092 static DEFINE_RATELIMIT_STATE(_rs,
8093 DEFAULT_RATELIMIT_INTERVAL * 10,
8094 /*DEFAULT_RATELIMIT_BURST*/ 1);
8095 if (__ratelimit(&_rs))
8096 WARN(1, KERN_DEBUG
8097 "BTRFS: block rsv returned %d\n", ret);
8098 }
8099 try_reserve:
8100 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
8101 BTRFS_RESERVE_NO_FLUSH);
8102 if (!ret)
8103 return block_rsv;
8104 /*
8105 * If we couldn't reserve metadata bytes try and use some from
8106 * the global reserve if its space type is the same as the global
8107 * reservation.
8108 */
8109 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
8110 block_rsv->space_info == global_rsv->space_info) {
8111 ret = block_rsv_use_bytes(global_rsv, blocksize);
8112 if (!ret)
8113 return global_rsv;
8114 }
8115 return ERR_PTR(ret);
8116 }
8117
8118 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
8119 struct btrfs_block_rsv *block_rsv, u32 blocksize)
8120 {
8121 block_rsv_add_bytes(block_rsv, blocksize, 0);
8122 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
8123 }
8124
8125 /*
8126 * finds a free extent and does all the dirty work required for allocation
8127 * returns the tree buffer or an ERR_PTR on error.
8128 */
8129 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8130 struct btrfs_root *root,
8131 u64 parent, u64 root_objectid,
8132 struct btrfs_disk_key *key, int level,
8133 u64 hint, u64 empty_size)
8134 {
8135 struct btrfs_key ins;
8136 struct btrfs_block_rsv *block_rsv;
8137 struct extent_buffer *buf;
8138 struct btrfs_delayed_extent_op *extent_op;
8139 u64 flags = 0;
8140 int ret;
8141 u32 blocksize = root->nodesize;
8142 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
8143 SKINNY_METADATA);
8144
8145 if (btrfs_test_is_dummy_root(root)) {
8146 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8147 level);
8148 if (!IS_ERR(buf))
8149 root->alloc_bytenr += blocksize;
8150 return buf;
8151 }
8152
8153 block_rsv = use_block_rsv(trans, root, blocksize);
8154 if (IS_ERR(block_rsv))
8155 return ERR_CAST(block_rsv);
8156
8157 ret = btrfs_reserve_extent(root, blocksize, blocksize,
8158 empty_size, hint, &ins, 0, 0);
8159 if (ret)
8160 goto out_unuse;
8161
8162 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
8163 if (IS_ERR(buf)) {
8164 ret = PTR_ERR(buf);
8165 goto out_free_reserved;
8166 }
8167
8168 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8169 if (parent == 0)
8170 parent = ins.objectid;
8171 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8172 } else
8173 BUG_ON(parent > 0);
8174
8175 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8176 extent_op = btrfs_alloc_delayed_extent_op();
8177 if (!extent_op) {
8178 ret = -ENOMEM;
8179 goto out_free_buf;
8180 }
8181 if (key)
8182 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8183 else
8184 memset(&extent_op->key, 0, sizeof(extent_op->key));
8185 extent_op->flags_to_set = flags;
8186 extent_op->update_key = skinny_metadata ? false : true;
8187 extent_op->update_flags = true;
8188 extent_op->is_data = false;
8189 extent_op->level = level;
8190
8191 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
8192 ins.objectid, ins.offset,
8193 parent, root_objectid, level,
8194 BTRFS_ADD_DELAYED_EXTENT,
8195 extent_op);
8196 if (ret)
8197 goto out_free_delayed;
8198 }
8199 return buf;
8200
8201 out_free_delayed:
8202 btrfs_free_delayed_extent_op(extent_op);
8203 out_free_buf:
8204 free_extent_buffer(buf);
8205 out_free_reserved:
8206 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0);
8207 out_unuse:
8208 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
8209 return ERR_PTR(ret);
8210 }
8211
8212 struct walk_control {
8213 u64 refs[BTRFS_MAX_LEVEL];
8214 u64 flags[BTRFS_MAX_LEVEL];
8215 struct btrfs_key update_progress;
8216 int stage;
8217 int level;
8218 int shared_level;
8219 int update_ref;
8220 int keep_locks;
8221 int reada_slot;
8222 int reada_count;
8223 int for_reloc;
8224 };
8225
8226 #define DROP_REFERENCE 1
8227 #define UPDATE_BACKREF 2
8228
8229 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8230 struct btrfs_root *root,
8231 struct walk_control *wc,
8232 struct btrfs_path *path)
8233 {
8234 u64 bytenr;
8235 u64 generation;
8236 u64 refs;
8237 u64 flags;
8238 u32 nritems;
8239 u32 blocksize;
8240 struct btrfs_key key;
8241 struct extent_buffer *eb;
8242 int ret;
8243 int slot;
8244 int nread = 0;
8245
8246 if (path->slots[wc->level] < wc->reada_slot) {
8247 wc->reada_count = wc->reada_count * 2 / 3;
8248 wc->reada_count = max(wc->reada_count, 2);
8249 } else {
8250 wc->reada_count = wc->reada_count * 3 / 2;
8251 wc->reada_count = min_t(int, wc->reada_count,
8252 BTRFS_NODEPTRS_PER_BLOCK(root));
8253 }
8254
8255 eb = path->nodes[wc->level];
8256 nritems = btrfs_header_nritems(eb);
8257 blocksize = root->nodesize;
8258
8259 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8260 if (nread >= wc->reada_count)
8261 break;
8262
8263 cond_resched();
8264 bytenr = btrfs_node_blockptr(eb, slot);
8265 generation = btrfs_node_ptr_generation(eb, slot);
8266
8267 if (slot == path->slots[wc->level])
8268 goto reada;
8269
8270 if (wc->stage == UPDATE_BACKREF &&
8271 generation <= root->root_key.offset)
8272 continue;
8273
8274 /* We don't lock the tree block, it's OK to be racy here */
8275 ret = btrfs_lookup_extent_info(trans, root, bytenr,
8276 wc->level - 1, 1, &refs,
8277 &flags);
8278 /* We don't care about errors in readahead. */
8279 if (ret < 0)
8280 continue;
8281 BUG_ON(refs == 0);
8282
8283 if (wc->stage == DROP_REFERENCE) {
8284 if (refs == 1)
8285 goto reada;
8286
8287 if (wc->level == 1 &&
8288 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8289 continue;
8290 if (!wc->update_ref ||
8291 generation <= root->root_key.offset)
8292 continue;
8293 btrfs_node_key_to_cpu(eb, &key, slot);
8294 ret = btrfs_comp_cpu_keys(&key,
8295 &wc->update_progress);
8296 if (ret < 0)
8297 continue;
8298 } else {
8299 if (wc->level == 1 &&
8300 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8301 continue;
8302 }
8303 reada:
8304 readahead_tree_block(root, bytenr);
8305 nread++;
8306 }
8307 wc->reada_slot = slot;
8308 }
8309
8310 /*
8311 * These may not be seen by the usual inc/dec ref code so we have to
8312 * add them here.
8313 */
8314 static int record_one_subtree_extent(struct btrfs_trans_handle *trans,
8315 struct btrfs_root *root, u64 bytenr,
8316 u64 num_bytes)
8317 {
8318 struct btrfs_qgroup_extent_record *qrecord;
8319 struct btrfs_delayed_ref_root *delayed_refs;
8320
8321 qrecord = kmalloc(sizeof(*qrecord), GFP_NOFS);
8322 if (!qrecord)
8323 return -ENOMEM;
8324
8325 qrecord->bytenr = bytenr;
8326 qrecord->num_bytes = num_bytes;
8327 qrecord->old_roots = NULL;
8328
8329 delayed_refs = &trans->transaction->delayed_refs;
8330 spin_lock(&delayed_refs->lock);
8331 if (btrfs_qgroup_insert_dirty_extent(delayed_refs, qrecord))
8332 kfree(qrecord);
8333 spin_unlock(&delayed_refs->lock);
8334
8335 return 0;
8336 }
8337
8338 static int account_leaf_items(struct btrfs_trans_handle *trans,
8339 struct btrfs_root *root,
8340 struct extent_buffer *eb)
8341 {
8342 int nr = btrfs_header_nritems(eb);
8343 int i, extent_type, ret;
8344 struct btrfs_key key;
8345 struct btrfs_file_extent_item *fi;
8346 u64 bytenr, num_bytes;
8347
8348 /* We can be called directly from walk_up_proc() */
8349 if (!root->fs_info->quota_enabled)
8350 return 0;
8351
8352 for (i = 0; i < nr; i++) {
8353 btrfs_item_key_to_cpu(eb, &key, i);
8354
8355 if (key.type != BTRFS_EXTENT_DATA_KEY)
8356 continue;
8357
8358 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
8359 /* filter out non qgroup-accountable extents */
8360 extent_type = btrfs_file_extent_type(eb, fi);
8361
8362 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
8363 continue;
8364
8365 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
8366 if (!bytenr)
8367 continue;
8368
8369 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
8370
8371 ret = record_one_subtree_extent(trans, root, bytenr, num_bytes);
8372 if (ret)
8373 return ret;
8374 }
8375 return 0;
8376 }
8377
8378 /*
8379 * Walk up the tree from the bottom, freeing leaves and any interior
8380 * nodes which have had all slots visited. If a node (leaf or
8381 * interior) is freed, the node above it will have it's slot
8382 * incremented. The root node will never be freed.
8383 *
8384 * At the end of this function, we should have a path which has all
8385 * slots incremented to the next position for a search. If we need to
8386 * read a new node it will be NULL and the node above it will have the
8387 * correct slot selected for a later read.
8388 *
8389 * If we increment the root nodes slot counter past the number of
8390 * elements, 1 is returned to signal completion of the search.
8391 */
8392 static int adjust_slots_upwards(struct btrfs_root *root,
8393 struct btrfs_path *path, int root_level)
8394 {
8395 int level = 0;
8396 int nr, slot;
8397 struct extent_buffer *eb;
8398
8399 if (root_level == 0)
8400 return 1;
8401
8402 while (level <= root_level) {
8403 eb = path->nodes[level];
8404 nr = btrfs_header_nritems(eb);
8405 path->slots[level]++;
8406 slot = path->slots[level];
8407 if (slot >= nr || level == 0) {
8408 /*
8409 * Don't free the root - we will detect this
8410 * condition after our loop and return a
8411 * positive value for caller to stop walking the tree.
8412 */
8413 if (level != root_level) {
8414 btrfs_tree_unlock_rw(eb, path->locks[level]);
8415 path->locks[level] = 0;
8416
8417 free_extent_buffer(eb);
8418 path->nodes[level] = NULL;
8419 path->slots[level] = 0;
8420 }
8421 } else {
8422 /*
8423 * We have a valid slot to walk back down
8424 * from. Stop here so caller can process these
8425 * new nodes.
8426 */
8427 break;
8428 }
8429
8430 level++;
8431 }
8432
8433 eb = path->nodes[root_level];
8434 if (path->slots[root_level] >= btrfs_header_nritems(eb))
8435 return 1;
8436
8437 return 0;
8438 }
8439
8440 /*
8441 * root_eb is the subtree root and is locked before this function is called.
8442 */
8443 static int account_shared_subtree(struct btrfs_trans_handle *trans,
8444 struct btrfs_root *root,
8445 struct extent_buffer *root_eb,
8446 u64 root_gen,
8447 int root_level)
8448 {
8449 int ret = 0;
8450 int level;
8451 struct extent_buffer *eb = root_eb;
8452 struct btrfs_path *path = NULL;
8453
8454 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
8455 BUG_ON(root_eb == NULL);
8456
8457 if (!root->fs_info->quota_enabled)
8458 return 0;
8459
8460 if (!extent_buffer_uptodate(root_eb)) {
8461 ret = btrfs_read_buffer(root_eb, root_gen);
8462 if (ret)
8463 goto out;
8464 }
8465
8466 if (root_level == 0) {
8467 ret = account_leaf_items(trans, root, root_eb);
8468 goto out;
8469 }
8470
8471 path = btrfs_alloc_path();
8472 if (!path)
8473 return -ENOMEM;
8474
8475 /*
8476 * Walk down the tree. Missing extent blocks are filled in as
8477 * we go. Metadata is accounted every time we read a new
8478 * extent block.
8479 *
8480 * When we reach a leaf, we account for file extent items in it,
8481 * walk back up the tree (adjusting slot pointers as we go)
8482 * and restart the search process.
8483 */
8484 extent_buffer_get(root_eb); /* For path */
8485 path->nodes[root_level] = root_eb;
8486 path->slots[root_level] = 0;
8487 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
8488 walk_down:
8489 level = root_level;
8490 while (level >= 0) {
8491 if (path->nodes[level] == NULL) {
8492 int parent_slot;
8493 u64 child_gen;
8494 u64 child_bytenr;
8495
8496 /* We need to get child blockptr/gen from
8497 * parent before we can read it. */
8498 eb = path->nodes[level + 1];
8499 parent_slot = path->slots[level + 1];
8500 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
8501 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
8502
8503 eb = read_tree_block(root, child_bytenr, child_gen);
8504 if (IS_ERR(eb)) {
8505 ret = PTR_ERR(eb);
8506 goto out;
8507 } else if (!extent_buffer_uptodate(eb)) {
8508 free_extent_buffer(eb);
8509 ret = -EIO;
8510 goto out;
8511 }
8512
8513 path->nodes[level] = eb;
8514 path->slots[level] = 0;
8515
8516 btrfs_tree_read_lock(eb);
8517 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
8518 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
8519
8520 ret = record_one_subtree_extent(trans, root, child_bytenr,
8521 root->nodesize);
8522 if (ret)
8523 goto out;
8524 }
8525
8526 if (level == 0) {
8527 ret = account_leaf_items(trans, root, path->nodes[level]);
8528 if (ret)
8529 goto out;
8530
8531 /* Nonzero return here means we completed our search */
8532 ret = adjust_slots_upwards(root, path, root_level);
8533 if (ret)
8534 break;
8535
8536 /* Restart search with new slots */
8537 goto walk_down;
8538 }
8539
8540 level--;
8541 }
8542
8543 ret = 0;
8544 out:
8545 btrfs_free_path(path);
8546
8547 return ret;
8548 }
8549
8550 /*
8551 * helper to process tree block while walking down the tree.
8552 *
8553 * when wc->stage == UPDATE_BACKREF, this function updates
8554 * back refs for pointers in the block.
8555 *
8556 * NOTE: return value 1 means we should stop walking down.
8557 */
8558 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8559 struct btrfs_root *root,
8560 struct btrfs_path *path,
8561 struct walk_control *wc, int lookup_info)
8562 {
8563 int level = wc->level;
8564 struct extent_buffer *eb = path->nodes[level];
8565 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8566 int ret;
8567
8568 if (wc->stage == UPDATE_BACKREF &&
8569 btrfs_header_owner(eb) != root->root_key.objectid)
8570 return 1;
8571
8572 /*
8573 * when reference count of tree block is 1, it won't increase
8574 * again. once full backref flag is set, we never clear it.
8575 */
8576 if (lookup_info &&
8577 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8578 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8579 BUG_ON(!path->locks[level]);
8580 ret = btrfs_lookup_extent_info(trans, root,
8581 eb->start, level, 1,
8582 &wc->refs[level],
8583 &wc->flags[level]);
8584 BUG_ON(ret == -ENOMEM);
8585 if (ret)
8586 return ret;
8587 BUG_ON(wc->refs[level] == 0);
8588 }
8589
8590 if (wc->stage == DROP_REFERENCE) {
8591 if (wc->refs[level] > 1)
8592 return 1;
8593
8594 if (path->locks[level] && !wc->keep_locks) {
8595 btrfs_tree_unlock_rw(eb, path->locks[level]);
8596 path->locks[level] = 0;
8597 }
8598 return 0;
8599 }
8600
8601 /* wc->stage == UPDATE_BACKREF */
8602 if (!(wc->flags[level] & flag)) {
8603 BUG_ON(!path->locks[level]);
8604 ret = btrfs_inc_ref(trans, root, eb, 1);
8605 BUG_ON(ret); /* -ENOMEM */
8606 ret = btrfs_dec_ref(trans, root, eb, 0);
8607 BUG_ON(ret); /* -ENOMEM */
8608 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
8609 eb->len, flag,
8610 btrfs_header_level(eb), 0);
8611 BUG_ON(ret); /* -ENOMEM */
8612 wc->flags[level] |= flag;
8613 }
8614
8615 /*
8616 * the block is shared by multiple trees, so it's not good to
8617 * keep the tree lock
8618 */
8619 if (path->locks[level] && level > 0) {
8620 btrfs_tree_unlock_rw(eb, path->locks[level]);
8621 path->locks[level] = 0;
8622 }
8623 return 0;
8624 }
8625
8626 /*
8627 * helper to process tree block pointer.
8628 *
8629 * when wc->stage == DROP_REFERENCE, this function checks
8630 * reference count of the block pointed to. if the block
8631 * is shared and we need update back refs for the subtree
8632 * rooted at the block, this function changes wc->stage to
8633 * UPDATE_BACKREF. if the block is shared and there is no
8634 * need to update back, this function drops the reference
8635 * to the block.
8636 *
8637 * NOTE: return value 1 means we should stop walking down.
8638 */
8639 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8640 struct btrfs_root *root,
8641 struct btrfs_path *path,
8642 struct walk_control *wc, int *lookup_info)
8643 {
8644 u64 bytenr;
8645 u64 generation;
8646 u64 parent;
8647 u32 blocksize;
8648 struct btrfs_key key;
8649 struct extent_buffer *next;
8650 int level = wc->level;
8651 int reada = 0;
8652 int ret = 0;
8653 bool need_account = false;
8654
8655 generation = btrfs_node_ptr_generation(path->nodes[level],
8656 path->slots[level]);
8657 /*
8658 * if the lower level block was created before the snapshot
8659 * was created, we know there is no need to update back refs
8660 * for the subtree
8661 */
8662 if (wc->stage == UPDATE_BACKREF &&
8663 generation <= root->root_key.offset) {
8664 *lookup_info = 1;
8665 return 1;
8666 }
8667
8668 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8669 blocksize = root->nodesize;
8670
8671 next = btrfs_find_tree_block(root->fs_info, bytenr);
8672 if (!next) {
8673 next = btrfs_find_create_tree_block(root, bytenr);
8674 if (IS_ERR(next))
8675 return PTR_ERR(next);
8676
8677 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8678 level - 1);
8679 reada = 1;
8680 }
8681 btrfs_tree_lock(next);
8682 btrfs_set_lock_blocking(next);
8683
8684 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
8685 &wc->refs[level - 1],
8686 &wc->flags[level - 1]);
8687 if (ret < 0) {
8688 btrfs_tree_unlock(next);
8689 return ret;
8690 }
8691
8692 if (unlikely(wc->refs[level - 1] == 0)) {
8693 btrfs_err(root->fs_info, "Missing references.");
8694 BUG();
8695 }
8696 *lookup_info = 0;
8697
8698 if (wc->stage == DROP_REFERENCE) {
8699 if (wc->refs[level - 1] > 1) {
8700 need_account = true;
8701 if (level == 1 &&
8702 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8703 goto skip;
8704
8705 if (!wc->update_ref ||
8706 generation <= root->root_key.offset)
8707 goto skip;
8708
8709 btrfs_node_key_to_cpu(path->nodes[level], &key,
8710 path->slots[level]);
8711 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8712 if (ret < 0)
8713 goto skip;
8714
8715 wc->stage = UPDATE_BACKREF;
8716 wc->shared_level = level - 1;
8717 }
8718 } else {
8719 if (level == 1 &&
8720 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8721 goto skip;
8722 }
8723
8724 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8725 btrfs_tree_unlock(next);
8726 free_extent_buffer(next);
8727 next = NULL;
8728 *lookup_info = 1;
8729 }
8730
8731 if (!next) {
8732 if (reada && level == 1)
8733 reada_walk_down(trans, root, wc, path);
8734 next = read_tree_block(root, bytenr, generation);
8735 if (IS_ERR(next)) {
8736 return PTR_ERR(next);
8737 } else if (!extent_buffer_uptodate(next)) {
8738 free_extent_buffer(next);
8739 return -EIO;
8740 }
8741 btrfs_tree_lock(next);
8742 btrfs_set_lock_blocking(next);
8743 }
8744
8745 level--;
8746 BUG_ON(level != btrfs_header_level(next));
8747 path->nodes[level] = next;
8748 path->slots[level] = 0;
8749 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8750 wc->level = level;
8751 if (wc->level == 1)
8752 wc->reada_slot = 0;
8753 return 0;
8754 skip:
8755 wc->refs[level - 1] = 0;
8756 wc->flags[level - 1] = 0;
8757 if (wc->stage == DROP_REFERENCE) {
8758 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8759 parent = path->nodes[level]->start;
8760 } else {
8761 BUG_ON(root->root_key.objectid !=
8762 btrfs_header_owner(path->nodes[level]));
8763 parent = 0;
8764 }
8765
8766 if (need_account) {
8767 ret = account_shared_subtree(trans, root, next,
8768 generation, level - 1);
8769 if (ret) {
8770 btrfs_err_rl(root->fs_info,
8771 "Error "
8772 "%d accounting shared subtree. Quota "
8773 "is out of sync, rescan required.",
8774 ret);
8775 }
8776 }
8777 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
8778 root->root_key.objectid, level - 1, 0);
8779 BUG_ON(ret); /* -ENOMEM */
8780 }
8781 btrfs_tree_unlock(next);
8782 free_extent_buffer(next);
8783 *lookup_info = 1;
8784 return 1;
8785 }
8786
8787 /*
8788 * helper to process tree block while walking up the tree.
8789 *
8790 * when wc->stage == DROP_REFERENCE, this function drops
8791 * reference count on the block.
8792 *
8793 * when wc->stage == UPDATE_BACKREF, this function changes
8794 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8795 * to UPDATE_BACKREF previously while processing the block.
8796 *
8797 * NOTE: return value 1 means we should stop walking up.
8798 */
8799 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8800 struct btrfs_root *root,
8801 struct btrfs_path *path,
8802 struct walk_control *wc)
8803 {
8804 int ret;
8805 int level = wc->level;
8806 struct extent_buffer *eb = path->nodes[level];
8807 u64 parent = 0;
8808
8809 if (wc->stage == UPDATE_BACKREF) {
8810 BUG_ON(wc->shared_level < level);
8811 if (level < wc->shared_level)
8812 goto out;
8813
8814 ret = find_next_key(path, level + 1, &wc->update_progress);
8815 if (ret > 0)
8816 wc->update_ref = 0;
8817
8818 wc->stage = DROP_REFERENCE;
8819 wc->shared_level = -1;
8820 path->slots[level] = 0;
8821
8822 /*
8823 * check reference count again if the block isn't locked.
8824 * we should start walking down the tree again if reference
8825 * count is one.
8826 */
8827 if (!path->locks[level]) {
8828 BUG_ON(level == 0);
8829 btrfs_tree_lock(eb);
8830 btrfs_set_lock_blocking(eb);
8831 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8832
8833 ret = btrfs_lookup_extent_info(trans, root,
8834 eb->start, level, 1,
8835 &wc->refs[level],
8836 &wc->flags[level]);
8837 if (ret < 0) {
8838 btrfs_tree_unlock_rw(eb, path->locks[level]);
8839 path->locks[level] = 0;
8840 return ret;
8841 }
8842 BUG_ON(wc->refs[level] == 0);
8843 if (wc->refs[level] == 1) {
8844 btrfs_tree_unlock_rw(eb, path->locks[level]);
8845 path->locks[level] = 0;
8846 return 1;
8847 }
8848 }
8849 }
8850
8851 /* wc->stage == DROP_REFERENCE */
8852 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8853
8854 if (wc->refs[level] == 1) {
8855 if (level == 0) {
8856 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8857 ret = btrfs_dec_ref(trans, root, eb, 1);
8858 else
8859 ret = btrfs_dec_ref(trans, root, eb, 0);
8860 BUG_ON(ret); /* -ENOMEM */
8861 ret = account_leaf_items(trans, root, eb);
8862 if (ret) {
8863 btrfs_err_rl(root->fs_info,
8864 "error "
8865 "%d accounting leaf items. Quota "
8866 "is out of sync, rescan required.",
8867 ret);
8868 }
8869 }
8870 /* make block locked assertion in clean_tree_block happy */
8871 if (!path->locks[level] &&
8872 btrfs_header_generation(eb) == trans->transid) {
8873 btrfs_tree_lock(eb);
8874 btrfs_set_lock_blocking(eb);
8875 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8876 }
8877 clean_tree_block(trans, root->fs_info, eb);
8878 }
8879
8880 if (eb == root->node) {
8881 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8882 parent = eb->start;
8883 else
8884 BUG_ON(root->root_key.objectid !=
8885 btrfs_header_owner(eb));
8886 } else {
8887 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8888 parent = path->nodes[level + 1]->start;
8889 else
8890 BUG_ON(root->root_key.objectid !=
8891 btrfs_header_owner(path->nodes[level + 1]));
8892 }
8893
8894 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8895 out:
8896 wc->refs[level] = 0;
8897 wc->flags[level] = 0;
8898 return 0;
8899 }
8900
8901 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8902 struct btrfs_root *root,
8903 struct btrfs_path *path,
8904 struct walk_control *wc)
8905 {
8906 int level = wc->level;
8907 int lookup_info = 1;
8908 int ret;
8909
8910 while (level >= 0) {
8911 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8912 if (ret > 0)
8913 break;
8914
8915 if (level == 0)
8916 break;
8917
8918 if (path->slots[level] >=
8919 btrfs_header_nritems(path->nodes[level]))
8920 break;
8921
8922 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8923 if (ret > 0) {
8924 path->slots[level]++;
8925 continue;
8926 } else if (ret < 0)
8927 return ret;
8928 level = wc->level;
8929 }
8930 return 0;
8931 }
8932
8933 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8934 struct btrfs_root *root,
8935 struct btrfs_path *path,
8936 struct walk_control *wc, int max_level)
8937 {
8938 int level = wc->level;
8939 int ret;
8940
8941 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8942 while (level < max_level && path->nodes[level]) {
8943 wc->level = level;
8944 if (path->slots[level] + 1 <
8945 btrfs_header_nritems(path->nodes[level])) {
8946 path->slots[level]++;
8947 return 0;
8948 } else {
8949 ret = walk_up_proc(trans, root, path, wc);
8950 if (ret > 0)
8951 return 0;
8952
8953 if (path->locks[level]) {
8954 btrfs_tree_unlock_rw(path->nodes[level],
8955 path->locks[level]);
8956 path->locks[level] = 0;
8957 }
8958 free_extent_buffer(path->nodes[level]);
8959 path->nodes[level] = NULL;
8960 level++;
8961 }
8962 }
8963 return 1;
8964 }
8965
8966 /*
8967 * drop a subvolume tree.
8968 *
8969 * this function traverses the tree freeing any blocks that only
8970 * referenced by the tree.
8971 *
8972 * when a shared tree block is found. this function decreases its
8973 * reference count by one. if update_ref is true, this function
8974 * also make sure backrefs for the shared block and all lower level
8975 * blocks are properly updated.
8976 *
8977 * If called with for_reloc == 0, may exit early with -EAGAIN
8978 */
8979 int btrfs_drop_snapshot(struct btrfs_root *root,
8980 struct btrfs_block_rsv *block_rsv, int update_ref,
8981 int for_reloc)
8982 {
8983 struct btrfs_path *path;
8984 struct btrfs_trans_handle *trans;
8985 struct btrfs_root *tree_root = root->fs_info->tree_root;
8986 struct btrfs_root_item *root_item = &root->root_item;
8987 struct walk_control *wc;
8988 struct btrfs_key key;
8989 int err = 0;
8990 int ret;
8991 int level;
8992 bool root_dropped = false;
8993
8994 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
8995
8996 path = btrfs_alloc_path();
8997 if (!path) {
8998 err = -ENOMEM;
8999 goto out;
9000 }
9001
9002 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9003 if (!wc) {
9004 btrfs_free_path(path);
9005 err = -ENOMEM;
9006 goto out;
9007 }
9008
9009 trans = btrfs_start_transaction(tree_root, 0);
9010 if (IS_ERR(trans)) {
9011 err = PTR_ERR(trans);
9012 goto out_free;
9013 }
9014
9015 if (block_rsv)
9016 trans->block_rsv = block_rsv;
9017
9018 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
9019 level = btrfs_header_level(root->node);
9020 path->nodes[level] = btrfs_lock_root_node(root);
9021 btrfs_set_lock_blocking(path->nodes[level]);
9022 path->slots[level] = 0;
9023 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9024 memset(&wc->update_progress, 0,
9025 sizeof(wc->update_progress));
9026 } else {
9027 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
9028 memcpy(&wc->update_progress, &key,
9029 sizeof(wc->update_progress));
9030
9031 level = root_item->drop_level;
9032 BUG_ON(level == 0);
9033 path->lowest_level = level;
9034 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
9035 path->lowest_level = 0;
9036 if (ret < 0) {
9037 err = ret;
9038 goto out_end_trans;
9039 }
9040 WARN_ON(ret > 0);
9041
9042 /*
9043 * unlock our path, this is safe because only this
9044 * function is allowed to delete this snapshot
9045 */
9046 btrfs_unlock_up_safe(path, 0);
9047
9048 level = btrfs_header_level(root->node);
9049 while (1) {
9050 btrfs_tree_lock(path->nodes[level]);
9051 btrfs_set_lock_blocking(path->nodes[level]);
9052 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9053
9054 ret = btrfs_lookup_extent_info(trans, root,
9055 path->nodes[level]->start,
9056 level, 1, &wc->refs[level],
9057 &wc->flags[level]);
9058 if (ret < 0) {
9059 err = ret;
9060 goto out_end_trans;
9061 }
9062 BUG_ON(wc->refs[level] == 0);
9063
9064 if (level == root_item->drop_level)
9065 break;
9066
9067 btrfs_tree_unlock(path->nodes[level]);
9068 path->locks[level] = 0;
9069 WARN_ON(wc->refs[level] != 1);
9070 level--;
9071 }
9072 }
9073
9074 wc->level = level;
9075 wc->shared_level = -1;
9076 wc->stage = DROP_REFERENCE;
9077 wc->update_ref = update_ref;
9078 wc->keep_locks = 0;
9079 wc->for_reloc = for_reloc;
9080 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
9081
9082 while (1) {
9083
9084 ret = walk_down_tree(trans, root, path, wc);
9085 if (ret < 0) {
9086 err = ret;
9087 break;
9088 }
9089
9090 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
9091 if (ret < 0) {
9092 err = ret;
9093 break;
9094 }
9095
9096 if (ret > 0) {
9097 BUG_ON(wc->stage != DROP_REFERENCE);
9098 break;
9099 }
9100
9101 if (wc->stage == DROP_REFERENCE) {
9102 level = wc->level;
9103 btrfs_node_key(path->nodes[level],
9104 &root_item->drop_progress,
9105 path->slots[level]);
9106 root_item->drop_level = level;
9107 }
9108
9109 BUG_ON(wc->level == 0);
9110 if (btrfs_should_end_transaction(trans, tree_root) ||
9111 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
9112 ret = btrfs_update_root(trans, tree_root,
9113 &root->root_key,
9114 root_item);
9115 if (ret) {
9116 btrfs_abort_transaction(trans, tree_root, ret);
9117 err = ret;
9118 goto out_end_trans;
9119 }
9120
9121 btrfs_end_transaction_throttle(trans, tree_root);
9122 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
9123 pr_debug("BTRFS: drop snapshot early exit\n");
9124 err = -EAGAIN;
9125 goto out_free;
9126 }
9127
9128 trans = btrfs_start_transaction(tree_root, 0);
9129 if (IS_ERR(trans)) {
9130 err = PTR_ERR(trans);
9131 goto out_free;
9132 }
9133 if (block_rsv)
9134 trans->block_rsv = block_rsv;
9135 }
9136 }
9137 btrfs_release_path(path);
9138 if (err)
9139 goto out_end_trans;
9140
9141 ret = btrfs_del_root(trans, tree_root, &root->root_key);
9142 if (ret) {
9143 btrfs_abort_transaction(trans, tree_root, ret);
9144 goto out_end_trans;
9145 }
9146
9147 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9148 ret = btrfs_find_root(tree_root, &root->root_key, path,
9149 NULL, NULL);
9150 if (ret < 0) {
9151 btrfs_abort_transaction(trans, tree_root, ret);
9152 err = ret;
9153 goto out_end_trans;
9154 } else if (ret > 0) {
9155 /* if we fail to delete the orphan item this time
9156 * around, it'll get picked up the next time.
9157 *
9158 * The most common failure here is just -ENOENT.
9159 */
9160 btrfs_del_orphan_item(trans, tree_root,
9161 root->root_key.objectid);
9162 }
9163 }
9164
9165 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9166 btrfs_add_dropped_root(trans, root);
9167 } else {
9168 free_extent_buffer(root->node);
9169 free_extent_buffer(root->commit_root);
9170 btrfs_put_fs_root(root);
9171 }
9172 root_dropped = true;
9173 out_end_trans:
9174 btrfs_end_transaction_throttle(trans, tree_root);
9175 out_free:
9176 kfree(wc);
9177 btrfs_free_path(path);
9178 out:
9179 /*
9180 * So if we need to stop dropping the snapshot for whatever reason we
9181 * need to make sure to add it back to the dead root list so that we
9182 * keep trying to do the work later. This also cleans up roots if we
9183 * don't have it in the radix (like when we recover after a power fail
9184 * or unmount) so we don't leak memory.
9185 */
9186 if (!for_reloc && root_dropped == false)
9187 btrfs_add_dead_root(root);
9188 if (err && err != -EAGAIN)
9189 btrfs_handle_fs_error(root->fs_info, err, NULL);
9190 return err;
9191 }
9192
9193 /*
9194 * drop subtree rooted at tree block 'node'.
9195 *
9196 * NOTE: this function will unlock and release tree block 'node'
9197 * only used by relocation code
9198 */
9199 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9200 struct btrfs_root *root,
9201 struct extent_buffer *node,
9202 struct extent_buffer *parent)
9203 {
9204 struct btrfs_path *path;
9205 struct walk_control *wc;
9206 int level;
9207 int parent_level;
9208 int ret = 0;
9209 int wret;
9210
9211 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9212
9213 path = btrfs_alloc_path();
9214 if (!path)
9215 return -ENOMEM;
9216
9217 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9218 if (!wc) {
9219 btrfs_free_path(path);
9220 return -ENOMEM;
9221 }
9222
9223 btrfs_assert_tree_locked(parent);
9224 parent_level = btrfs_header_level(parent);
9225 extent_buffer_get(parent);
9226 path->nodes[parent_level] = parent;
9227 path->slots[parent_level] = btrfs_header_nritems(parent);
9228
9229 btrfs_assert_tree_locked(node);
9230 level = btrfs_header_level(node);
9231 path->nodes[level] = node;
9232 path->slots[level] = 0;
9233 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9234
9235 wc->refs[parent_level] = 1;
9236 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9237 wc->level = level;
9238 wc->shared_level = -1;
9239 wc->stage = DROP_REFERENCE;
9240 wc->update_ref = 0;
9241 wc->keep_locks = 1;
9242 wc->for_reloc = 1;
9243 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
9244
9245 while (1) {
9246 wret = walk_down_tree(trans, root, path, wc);
9247 if (wret < 0) {
9248 ret = wret;
9249 break;
9250 }
9251
9252 wret = walk_up_tree(trans, root, path, wc, parent_level);
9253 if (wret < 0)
9254 ret = wret;
9255 if (wret != 0)
9256 break;
9257 }
9258
9259 kfree(wc);
9260 btrfs_free_path(path);
9261 return ret;
9262 }
9263
9264 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
9265 {
9266 u64 num_devices;
9267 u64 stripped;
9268
9269 /*
9270 * if restripe for this chunk_type is on pick target profile and
9271 * return, otherwise do the usual balance
9272 */
9273 stripped = get_restripe_target(root->fs_info, flags);
9274 if (stripped)
9275 return extended_to_chunk(stripped);
9276
9277 num_devices = root->fs_info->fs_devices->rw_devices;
9278
9279 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9280 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9281 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9282
9283 if (num_devices == 1) {
9284 stripped |= BTRFS_BLOCK_GROUP_DUP;
9285 stripped = flags & ~stripped;
9286
9287 /* turn raid0 into single device chunks */
9288 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9289 return stripped;
9290
9291 /* turn mirroring into duplication */
9292 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9293 BTRFS_BLOCK_GROUP_RAID10))
9294 return stripped | BTRFS_BLOCK_GROUP_DUP;
9295 } else {
9296 /* they already had raid on here, just return */
9297 if (flags & stripped)
9298 return flags;
9299
9300 stripped |= BTRFS_BLOCK_GROUP_DUP;
9301 stripped = flags & ~stripped;
9302
9303 /* switch duplicated blocks with raid1 */
9304 if (flags & BTRFS_BLOCK_GROUP_DUP)
9305 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9306
9307 /* this is drive concat, leave it alone */
9308 }
9309
9310 return flags;
9311 }
9312
9313 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9314 {
9315 struct btrfs_space_info *sinfo = cache->space_info;
9316 u64 num_bytes;
9317 u64 min_allocable_bytes;
9318 int ret = -ENOSPC;
9319
9320 /*
9321 * We need some metadata space and system metadata space for
9322 * allocating chunks in some corner cases until we force to set
9323 * it to be readonly.
9324 */
9325 if ((sinfo->flags &
9326 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9327 !force)
9328 min_allocable_bytes = SZ_1M;
9329 else
9330 min_allocable_bytes = 0;
9331
9332 spin_lock(&sinfo->lock);
9333 spin_lock(&cache->lock);
9334
9335 if (cache->ro) {
9336 cache->ro++;
9337 ret = 0;
9338 goto out;
9339 }
9340
9341 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9342 cache->bytes_super - btrfs_block_group_used(&cache->item);
9343
9344 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
9345 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
9346 min_allocable_bytes <= sinfo->total_bytes) {
9347 sinfo->bytes_readonly += num_bytes;
9348 cache->ro++;
9349 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9350 ret = 0;
9351 }
9352 out:
9353 spin_unlock(&cache->lock);
9354 spin_unlock(&sinfo->lock);
9355 return ret;
9356 }
9357
9358 int btrfs_inc_block_group_ro(struct btrfs_root *root,
9359 struct btrfs_block_group_cache *cache)
9360
9361 {
9362 struct btrfs_trans_handle *trans;
9363 u64 alloc_flags;
9364 int ret;
9365
9366 again:
9367 trans = btrfs_join_transaction(root);
9368 if (IS_ERR(trans))
9369 return PTR_ERR(trans);
9370
9371 /*
9372 * we're not allowed to set block groups readonly after the dirty
9373 * block groups cache has started writing. If it already started,
9374 * back off and let this transaction commit
9375 */
9376 mutex_lock(&root->fs_info->ro_block_group_mutex);
9377 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9378 u64 transid = trans->transid;
9379
9380 mutex_unlock(&root->fs_info->ro_block_group_mutex);
9381 btrfs_end_transaction(trans, root);
9382
9383 ret = btrfs_wait_for_commit(root, transid);
9384 if (ret)
9385 return ret;
9386 goto again;
9387 }
9388
9389 /*
9390 * if we are changing raid levels, try to allocate a corresponding
9391 * block group with the new raid level.
9392 */
9393 alloc_flags = update_block_group_flags(root, cache->flags);
9394 if (alloc_flags != cache->flags) {
9395 ret = do_chunk_alloc(trans, root, alloc_flags,
9396 CHUNK_ALLOC_FORCE);
9397 /*
9398 * ENOSPC is allowed here, we may have enough space
9399 * already allocated at the new raid level to
9400 * carry on
9401 */
9402 if (ret == -ENOSPC)
9403 ret = 0;
9404 if (ret < 0)
9405 goto out;
9406 }
9407
9408 ret = inc_block_group_ro(cache, 0);
9409 if (!ret)
9410 goto out;
9411 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
9412 ret = do_chunk_alloc(trans, root, alloc_flags,
9413 CHUNK_ALLOC_FORCE);
9414 if (ret < 0)
9415 goto out;
9416 ret = inc_block_group_ro(cache, 0);
9417 out:
9418 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9419 alloc_flags = update_block_group_flags(root, cache->flags);
9420 lock_chunks(root->fs_info->chunk_root);
9421 check_system_chunk(trans, root, alloc_flags);
9422 unlock_chunks(root->fs_info->chunk_root);
9423 }
9424 mutex_unlock(&root->fs_info->ro_block_group_mutex);
9425
9426 btrfs_end_transaction(trans, root);
9427 return ret;
9428 }
9429
9430 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
9431 struct btrfs_root *root, u64 type)
9432 {
9433 u64 alloc_flags = get_alloc_profile(root, type);
9434 return do_chunk_alloc(trans, root, alloc_flags,
9435 CHUNK_ALLOC_FORCE);
9436 }
9437
9438 /*
9439 * helper to account the unused space of all the readonly block group in the
9440 * space_info. takes mirrors into account.
9441 */
9442 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9443 {
9444 struct btrfs_block_group_cache *block_group;
9445 u64 free_bytes = 0;
9446 int factor;
9447
9448 /* It's df, we don't care if it's racy */
9449 if (list_empty(&sinfo->ro_bgs))
9450 return 0;
9451
9452 spin_lock(&sinfo->lock);
9453 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9454 spin_lock(&block_group->lock);
9455
9456 if (!block_group->ro) {
9457 spin_unlock(&block_group->lock);
9458 continue;
9459 }
9460
9461 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
9462 BTRFS_BLOCK_GROUP_RAID10 |
9463 BTRFS_BLOCK_GROUP_DUP))
9464 factor = 2;
9465 else
9466 factor = 1;
9467
9468 free_bytes += (block_group->key.offset -
9469 btrfs_block_group_used(&block_group->item)) *
9470 factor;
9471
9472 spin_unlock(&block_group->lock);
9473 }
9474 spin_unlock(&sinfo->lock);
9475
9476 return free_bytes;
9477 }
9478
9479 void btrfs_dec_block_group_ro(struct btrfs_root *root,
9480 struct btrfs_block_group_cache *cache)
9481 {
9482 struct btrfs_space_info *sinfo = cache->space_info;
9483 u64 num_bytes;
9484
9485 BUG_ON(!cache->ro);
9486
9487 spin_lock(&sinfo->lock);
9488 spin_lock(&cache->lock);
9489 if (!--cache->ro) {
9490 num_bytes = cache->key.offset - cache->reserved -
9491 cache->pinned - cache->bytes_super -
9492 btrfs_block_group_used(&cache->item);
9493 sinfo->bytes_readonly -= num_bytes;
9494 list_del_init(&cache->ro_list);
9495 }
9496 spin_unlock(&cache->lock);
9497 spin_unlock(&sinfo->lock);
9498 }
9499
9500 /*
9501 * checks to see if its even possible to relocate this block group.
9502 *
9503 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9504 * ok to go ahead and try.
9505 */
9506 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
9507 {
9508 struct btrfs_block_group_cache *block_group;
9509 struct btrfs_space_info *space_info;
9510 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
9511 struct btrfs_device *device;
9512 struct btrfs_trans_handle *trans;
9513 u64 min_free;
9514 u64 dev_min = 1;
9515 u64 dev_nr = 0;
9516 u64 target;
9517 int debug;
9518 int index;
9519 int full = 0;
9520 int ret = 0;
9521
9522 debug = btrfs_test_opt(root, ENOSPC_DEBUG);
9523
9524 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
9525
9526 /* odd, couldn't find the block group, leave it alone */
9527 if (!block_group) {
9528 if (debug)
9529 btrfs_warn(root->fs_info,
9530 "can't find block group for bytenr %llu",
9531 bytenr);
9532 return -1;
9533 }
9534
9535 min_free = btrfs_block_group_used(&block_group->item);
9536
9537 /* no bytes used, we're good */
9538 if (!min_free)
9539 goto out;
9540
9541 space_info = block_group->space_info;
9542 spin_lock(&space_info->lock);
9543
9544 full = space_info->full;
9545
9546 /*
9547 * if this is the last block group we have in this space, we can't
9548 * relocate it unless we're able to allocate a new chunk below.
9549 *
9550 * Otherwise, we need to make sure we have room in the space to handle
9551 * all of the extents from this block group. If we can, we're good
9552 */
9553 if ((space_info->total_bytes != block_group->key.offset) &&
9554 (space_info->bytes_used + space_info->bytes_reserved +
9555 space_info->bytes_pinned + space_info->bytes_readonly +
9556 min_free < space_info->total_bytes)) {
9557 spin_unlock(&space_info->lock);
9558 goto out;
9559 }
9560 spin_unlock(&space_info->lock);
9561
9562 /*
9563 * ok we don't have enough space, but maybe we have free space on our
9564 * devices to allocate new chunks for relocation, so loop through our
9565 * alloc devices and guess if we have enough space. if this block
9566 * group is going to be restriped, run checks against the target
9567 * profile instead of the current one.
9568 */
9569 ret = -1;
9570
9571 /*
9572 * index:
9573 * 0: raid10
9574 * 1: raid1
9575 * 2: dup
9576 * 3: raid0
9577 * 4: single
9578 */
9579 target = get_restripe_target(root->fs_info, block_group->flags);
9580 if (target) {
9581 index = __get_raid_index(extended_to_chunk(target));
9582 } else {
9583 /*
9584 * this is just a balance, so if we were marked as full
9585 * we know there is no space for a new chunk
9586 */
9587 if (full) {
9588 if (debug)
9589 btrfs_warn(root->fs_info,
9590 "no space to alloc new chunk for block group %llu",
9591 block_group->key.objectid);
9592 goto out;
9593 }
9594
9595 index = get_block_group_index(block_group);
9596 }
9597
9598 if (index == BTRFS_RAID_RAID10) {
9599 dev_min = 4;
9600 /* Divide by 2 */
9601 min_free >>= 1;
9602 } else if (index == BTRFS_RAID_RAID1) {
9603 dev_min = 2;
9604 } else if (index == BTRFS_RAID_DUP) {
9605 /* Multiply by 2 */
9606 min_free <<= 1;
9607 } else if (index == BTRFS_RAID_RAID0) {
9608 dev_min = fs_devices->rw_devices;
9609 min_free = div64_u64(min_free, dev_min);
9610 }
9611
9612 /* We need to do this so that we can look at pending chunks */
9613 trans = btrfs_join_transaction(root);
9614 if (IS_ERR(trans)) {
9615 ret = PTR_ERR(trans);
9616 goto out;
9617 }
9618
9619 mutex_lock(&root->fs_info->chunk_mutex);
9620 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9621 u64 dev_offset;
9622
9623 /*
9624 * check to make sure we can actually find a chunk with enough
9625 * space to fit our block group in.
9626 */
9627 if (device->total_bytes > device->bytes_used + min_free &&
9628 !device->is_tgtdev_for_dev_replace) {
9629 ret = find_free_dev_extent(trans, device, min_free,
9630 &dev_offset, NULL);
9631 if (!ret)
9632 dev_nr++;
9633
9634 if (dev_nr >= dev_min)
9635 break;
9636
9637 ret = -1;
9638 }
9639 }
9640 if (debug && ret == -1)
9641 btrfs_warn(root->fs_info,
9642 "no space to allocate a new chunk for block group %llu",
9643 block_group->key.objectid);
9644 mutex_unlock(&root->fs_info->chunk_mutex);
9645 btrfs_end_transaction(trans, root);
9646 out:
9647 btrfs_put_block_group(block_group);
9648 return ret;
9649 }
9650
9651 static int find_first_block_group(struct btrfs_root *root,
9652 struct btrfs_path *path, struct btrfs_key *key)
9653 {
9654 int ret = 0;
9655 struct btrfs_key found_key;
9656 struct extent_buffer *leaf;
9657 int slot;
9658
9659 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9660 if (ret < 0)
9661 goto out;
9662
9663 while (1) {
9664 slot = path->slots[0];
9665 leaf = path->nodes[0];
9666 if (slot >= btrfs_header_nritems(leaf)) {
9667 ret = btrfs_next_leaf(root, path);
9668 if (ret == 0)
9669 continue;
9670 if (ret < 0)
9671 goto out;
9672 break;
9673 }
9674 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9675
9676 if (found_key.objectid >= key->objectid &&
9677 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9678 ret = 0;
9679 goto out;
9680 }
9681 path->slots[0]++;
9682 }
9683 out:
9684 return ret;
9685 }
9686
9687 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9688 {
9689 struct btrfs_block_group_cache *block_group;
9690 u64 last = 0;
9691
9692 while (1) {
9693 struct inode *inode;
9694
9695 block_group = btrfs_lookup_first_block_group(info, last);
9696 while (block_group) {
9697 spin_lock(&block_group->lock);
9698 if (block_group->iref)
9699 break;
9700 spin_unlock(&block_group->lock);
9701 block_group = next_block_group(info->tree_root,
9702 block_group);
9703 }
9704 if (!block_group) {
9705 if (last == 0)
9706 break;
9707 last = 0;
9708 continue;
9709 }
9710
9711 inode = block_group->inode;
9712 block_group->iref = 0;
9713 block_group->inode = NULL;
9714 spin_unlock(&block_group->lock);
9715 iput(inode);
9716 last = block_group->key.objectid + block_group->key.offset;
9717 btrfs_put_block_group(block_group);
9718 }
9719 }
9720
9721 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9722 {
9723 struct btrfs_block_group_cache *block_group;
9724 struct btrfs_space_info *space_info;
9725 struct btrfs_caching_control *caching_ctl;
9726 struct rb_node *n;
9727
9728 down_write(&info->commit_root_sem);
9729 while (!list_empty(&info->caching_block_groups)) {
9730 caching_ctl = list_entry(info->caching_block_groups.next,
9731 struct btrfs_caching_control, list);
9732 list_del(&caching_ctl->list);
9733 put_caching_control(caching_ctl);
9734 }
9735 up_write(&info->commit_root_sem);
9736
9737 spin_lock(&info->unused_bgs_lock);
9738 while (!list_empty(&info->unused_bgs)) {
9739 block_group = list_first_entry(&info->unused_bgs,
9740 struct btrfs_block_group_cache,
9741 bg_list);
9742 list_del_init(&block_group->bg_list);
9743 btrfs_put_block_group(block_group);
9744 }
9745 spin_unlock(&info->unused_bgs_lock);
9746
9747 spin_lock(&info->block_group_cache_lock);
9748 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9749 block_group = rb_entry(n, struct btrfs_block_group_cache,
9750 cache_node);
9751 rb_erase(&block_group->cache_node,
9752 &info->block_group_cache_tree);
9753 RB_CLEAR_NODE(&block_group->cache_node);
9754 spin_unlock(&info->block_group_cache_lock);
9755
9756 down_write(&block_group->space_info->groups_sem);
9757 list_del(&block_group->list);
9758 up_write(&block_group->space_info->groups_sem);
9759
9760 if (block_group->cached == BTRFS_CACHE_STARTED)
9761 wait_block_group_cache_done(block_group);
9762
9763 /*
9764 * We haven't cached this block group, which means we could
9765 * possibly have excluded extents on this block group.
9766 */
9767 if (block_group->cached == BTRFS_CACHE_NO ||
9768 block_group->cached == BTRFS_CACHE_ERROR)
9769 free_excluded_extents(info->extent_root, block_group);
9770
9771 btrfs_remove_free_space_cache(block_group);
9772 btrfs_put_block_group(block_group);
9773
9774 spin_lock(&info->block_group_cache_lock);
9775 }
9776 spin_unlock(&info->block_group_cache_lock);
9777
9778 /* now that all the block groups are freed, go through and
9779 * free all the space_info structs. This is only called during
9780 * the final stages of unmount, and so we know nobody is
9781 * using them. We call synchronize_rcu() once before we start,
9782 * just to be on the safe side.
9783 */
9784 synchronize_rcu();
9785
9786 release_global_block_rsv(info);
9787
9788 while (!list_empty(&info->space_info)) {
9789 int i;
9790
9791 space_info = list_entry(info->space_info.next,
9792 struct btrfs_space_info,
9793 list);
9794 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
9795 if (WARN_ON(space_info->bytes_pinned > 0 ||
9796 space_info->bytes_reserved > 0 ||
9797 space_info->bytes_may_use > 0)) {
9798 dump_space_info(space_info, 0, 0);
9799 }
9800 }
9801 list_del(&space_info->list);
9802 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9803 struct kobject *kobj;
9804 kobj = space_info->block_group_kobjs[i];
9805 space_info->block_group_kobjs[i] = NULL;
9806 if (kobj) {
9807 kobject_del(kobj);
9808 kobject_put(kobj);
9809 }
9810 }
9811 kobject_del(&space_info->kobj);
9812 kobject_put(&space_info->kobj);
9813 }
9814 return 0;
9815 }
9816
9817 static void __link_block_group(struct btrfs_space_info *space_info,
9818 struct btrfs_block_group_cache *cache)
9819 {
9820 int index = get_block_group_index(cache);
9821 bool first = false;
9822
9823 down_write(&space_info->groups_sem);
9824 if (list_empty(&space_info->block_groups[index]))
9825 first = true;
9826 list_add_tail(&cache->list, &space_info->block_groups[index]);
9827 up_write(&space_info->groups_sem);
9828
9829 if (first) {
9830 struct raid_kobject *rkobj;
9831 int ret;
9832
9833 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9834 if (!rkobj)
9835 goto out_err;
9836 rkobj->raid_type = index;
9837 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9838 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9839 "%s", get_raid_name(index));
9840 if (ret) {
9841 kobject_put(&rkobj->kobj);
9842 goto out_err;
9843 }
9844 space_info->block_group_kobjs[index] = &rkobj->kobj;
9845 }
9846
9847 return;
9848 out_err:
9849 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
9850 }
9851
9852 static struct btrfs_block_group_cache *
9853 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
9854 {
9855 struct btrfs_block_group_cache *cache;
9856
9857 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9858 if (!cache)
9859 return NULL;
9860
9861 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9862 GFP_NOFS);
9863 if (!cache->free_space_ctl) {
9864 kfree(cache);
9865 return NULL;
9866 }
9867
9868 cache->key.objectid = start;
9869 cache->key.offset = size;
9870 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9871
9872 cache->sectorsize = root->sectorsize;
9873 cache->fs_info = root->fs_info;
9874 cache->full_stripe_len = btrfs_full_stripe_len(root,
9875 &root->fs_info->mapping_tree,
9876 start);
9877 set_free_space_tree_thresholds(cache);
9878
9879 atomic_set(&cache->count, 1);
9880 spin_lock_init(&cache->lock);
9881 init_rwsem(&cache->data_rwsem);
9882 INIT_LIST_HEAD(&cache->list);
9883 INIT_LIST_HEAD(&cache->cluster_list);
9884 INIT_LIST_HEAD(&cache->bg_list);
9885 INIT_LIST_HEAD(&cache->ro_list);
9886 INIT_LIST_HEAD(&cache->dirty_list);
9887 INIT_LIST_HEAD(&cache->io_list);
9888 btrfs_init_free_space_ctl(cache);
9889 atomic_set(&cache->trimming, 0);
9890 mutex_init(&cache->free_space_lock);
9891
9892 return cache;
9893 }
9894
9895 int btrfs_read_block_groups(struct btrfs_root *root)
9896 {
9897 struct btrfs_path *path;
9898 int ret;
9899 struct btrfs_block_group_cache *cache;
9900 struct btrfs_fs_info *info = root->fs_info;
9901 struct btrfs_space_info *space_info;
9902 struct btrfs_key key;
9903 struct btrfs_key found_key;
9904 struct extent_buffer *leaf;
9905 int need_clear = 0;
9906 u64 cache_gen;
9907
9908 root = info->extent_root;
9909 key.objectid = 0;
9910 key.offset = 0;
9911 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9912 path = btrfs_alloc_path();
9913 if (!path)
9914 return -ENOMEM;
9915 path->reada = READA_FORWARD;
9916
9917 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
9918 if (btrfs_test_opt(root, SPACE_CACHE) &&
9919 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
9920 need_clear = 1;
9921 if (btrfs_test_opt(root, CLEAR_CACHE))
9922 need_clear = 1;
9923
9924 while (1) {
9925 ret = find_first_block_group(root, path, &key);
9926 if (ret > 0)
9927 break;
9928 if (ret != 0)
9929 goto error;
9930
9931 leaf = path->nodes[0];
9932 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9933
9934 cache = btrfs_create_block_group_cache(root, found_key.objectid,
9935 found_key.offset);
9936 if (!cache) {
9937 ret = -ENOMEM;
9938 goto error;
9939 }
9940
9941 if (need_clear) {
9942 /*
9943 * When we mount with old space cache, we need to
9944 * set BTRFS_DC_CLEAR and set dirty flag.
9945 *
9946 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9947 * truncate the old free space cache inode and
9948 * setup a new one.
9949 * b) Setting 'dirty flag' makes sure that we flush
9950 * the new space cache info onto disk.
9951 */
9952 if (btrfs_test_opt(root, SPACE_CACHE))
9953 cache->disk_cache_state = BTRFS_DC_CLEAR;
9954 }
9955
9956 read_extent_buffer(leaf, &cache->item,
9957 btrfs_item_ptr_offset(leaf, path->slots[0]),
9958 sizeof(cache->item));
9959 cache->flags = btrfs_block_group_flags(&cache->item);
9960
9961 key.objectid = found_key.objectid + found_key.offset;
9962 btrfs_release_path(path);
9963
9964 /*
9965 * We need to exclude the super stripes now so that the space
9966 * info has super bytes accounted for, otherwise we'll think
9967 * we have more space than we actually do.
9968 */
9969 ret = exclude_super_stripes(root, cache);
9970 if (ret) {
9971 /*
9972 * We may have excluded something, so call this just in
9973 * case.
9974 */
9975 free_excluded_extents(root, cache);
9976 btrfs_put_block_group(cache);
9977 goto error;
9978 }
9979
9980 /*
9981 * check for two cases, either we are full, and therefore
9982 * don't need to bother with the caching work since we won't
9983 * find any space, or we are empty, and we can just add all
9984 * the space in and be done with it. This saves us _alot_ of
9985 * time, particularly in the full case.
9986 */
9987 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
9988 cache->last_byte_to_unpin = (u64)-1;
9989 cache->cached = BTRFS_CACHE_FINISHED;
9990 free_excluded_extents(root, cache);
9991 } else if (btrfs_block_group_used(&cache->item) == 0) {
9992 cache->last_byte_to_unpin = (u64)-1;
9993 cache->cached = BTRFS_CACHE_FINISHED;
9994 add_new_free_space(cache, root->fs_info,
9995 found_key.objectid,
9996 found_key.objectid +
9997 found_key.offset);
9998 free_excluded_extents(root, cache);
9999 }
10000
10001 ret = btrfs_add_block_group_cache(root->fs_info, cache);
10002 if (ret) {
10003 btrfs_remove_free_space_cache(cache);
10004 btrfs_put_block_group(cache);
10005 goto error;
10006 }
10007
10008 ret = update_space_info(info, cache->flags, found_key.offset,
10009 btrfs_block_group_used(&cache->item),
10010 &space_info);
10011 if (ret) {
10012 btrfs_remove_free_space_cache(cache);
10013 spin_lock(&info->block_group_cache_lock);
10014 rb_erase(&cache->cache_node,
10015 &info->block_group_cache_tree);
10016 RB_CLEAR_NODE(&cache->cache_node);
10017 spin_unlock(&info->block_group_cache_lock);
10018 btrfs_put_block_group(cache);
10019 goto error;
10020 }
10021
10022 cache->space_info = space_info;
10023 spin_lock(&cache->space_info->lock);
10024 cache->space_info->bytes_readonly += cache->bytes_super;
10025 spin_unlock(&cache->space_info->lock);
10026
10027 __link_block_group(space_info, cache);
10028
10029 set_avail_alloc_bits(root->fs_info, cache->flags);
10030 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
10031 inc_block_group_ro(cache, 1);
10032 } else if (btrfs_block_group_used(&cache->item) == 0) {
10033 spin_lock(&info->unused_bgs_lock);
10034 /* Should always be true but just in case. */
10035 if (list_empty(&cache->bg_list)) {
10036 btrfs_get_block_group(cache);
10037 list_add_tail(&cache->bg_list,
10038 &info->unused_bgs);
10039 }
10040 spin_unlock(&info->unused_bgs_lock);
10041 }
10042 }
10043
10044 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
10045 if (!(get_alloc_profile(root, space_info->flags) &
10046 (BTRFS_BLOCK_GROUP_RAID10 |
10047 BTRFS_BLOCK_GROUP_RAID1 |
10048 BTRFS_BLOCK_GROUP_RAID5 |
10049 BTRFS_BLOCK_GROUP_RAID6 |
10050 BTRFS_BLOCK_GROUP_DUP)))
10051 continue;
10052 /*
10053 * avoid allocating from un-mirrored block group if there are
10054 * mirrored block groups.
10055 */
10056 list_for_each_entry(cache,
10057 &space_info->block_groups[BTRFS_RAID_RAID0],
10058 list)
10059 inc_block_group_ro(cache, 1);
10060 list_for_each_entry(cache,
10061 &space_info->block_groups[BTRFS_RAID_SINGLE],
10062 list)
10063 inc_block_group_ro(cache, 1);
10064 }
10065
10066 init_global_block_rsv(info);
10067 ret = 0;
10068 error:
10069 btrfs_free_path(path);
10070 return ret;
10071 }
10072
10073 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
10074 struct btrfs_root *root)
10075 {
10076 struct btrfs_block_group_cache *block_group, *tmp;
10077 struct btrfs_root *extent_root = root->fs_info->extent_root;
10078 struct btrfs_block_group_item item;
10079 struct btrfs_key key;
10080 int ret = 0;
10081 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
10082
10083 trans->can_flush_pending_bgs = false;
10084 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
10085 if (ret)
10086 goto next;
10087
10088 spin_lock(&block_group->lock);
10089 memcpy(&item, &block_group->item, sizeof(item));
10090 memcpy(&key, &block_group->key, sizeof(key));
10091 spin_unlock(&block_group->lock);
10092
10093 ret = btrfs_insert_item(trans, extent_root, &key, &item,
10094 sizeof(item));
10095 if (ret)
10096 btrfs_abort_transaction(trans, extent_root, ret);
10097 ret = btrfs_finish_chunk_alloc(trans, extent_root,
10098 key.objectid, key.offset);
10099 if (ret)
10100 btrfs_abort_transaction(trans, extent_root, ret);
10101 add_block_group_free_space(trans, root->fs_info, block_group);
10102 /* already aborted the transaction if it failed. */
10103 next:
10104 list_del_init(&block_group->bg_list);
10105 }
10106 trans->can_flush_pending_bgs = can_flush_pending_bgs;
10107 }
10108
10109 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
10110 struct btrfs_root *root, u64 bytes_used,
10111 u64 type, u64 chunk_objectid, u64 chunk_offset,
10112 u64 size)
10113 {
10114 int ret;
10115 struct btrfs_root *extent_root;
10116 struct btrfs_block_group_cache *cache;
10117
10118 extent_root = root->fs_info->extent_root;
10119
10120 btrfs_set_log_full_commit(root->fs_info, trans);
10121
10122 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
10123 if (!cache)
10124 return -ENOMEM;
10125
10126 btrfs_set_block_group_used(&cache->item, bytes_used);
10127 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
10128 btrfs_set_block_group_flags(&cache->item, type);
10129
10130 cache->flags = type;
10131 cache->last_byte_to_unpin = (u64)-1;
10132 cache->cached = BTRFS_CACHE_FINISHED;
10133 cache->needs_free_space = 1;
10134 ret = exclude_super_stripes(root, cache);
10135 if (ret) {
10136 /*
10137 * We may have excluded something, so call this just in
10138 * case.
10139 */
10140 free_excluded_extents(root, cache);
10141 btrfs_put_block_group(cache);
10142 return ret;
10143 }
10144
10145 add_new_free_space(cache, root->fs_info, chunk_offset,
10146 chunk_offset + size);
10147
10148 free_excluded_extents(root, cache);
10149
10150 #ifdef CONFIG_BTRFS_DEBUG
10151 if (btrfs_should_fragment_free_space(root, cache)) {
10152 u64 new_bytes_used = size - bytes_used;
10153
10154 bytes_used += new_bytes_used >> 1;
10155 fragment_free_space(root, cache);
10156 }
10157 #endif
10158 /*
10159 * Call to ensure the corresponding space_info object is created and
10160 * assigned to our block group, but don't update its counters just yet.
10161 * We want our bg to be added to the rbtree with its ->space_info set.
10162 */
10163 ret = update_space_info(root->fs_info, cache->flags, 0, 0,
10164 &cache->space_info);
10165 if (ret) {
10166 btrfs_remove_free_space_cache(cache);
10167 btrfs_put_block_group(cache);
10168 return ret;
10169 }
10170
10171 ret = btrfs_add_block_group_cache(root->fs_info, cache);
10172 if (ret) {
10173 btrfs_remove_free_space_cache(cache);
10174 btrfs_put_block_group(cache);
10175 return ret;
10176 }
10177
10178 /*
10179 * Now that our block group has its ->space_info set and is inserted in
10180 * the rbtree, update the space info's counters.
10181 */
10182 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
10183 &cache->space_info);
10184 if (ret) {
10185 btrfs_remove_free_space_cache(cache);
10186 spin_lock(&root->fs_info->block_group_cache_lock);
10187 rb_erase(&cache->cache_node,
10188 &root->fs_info->block_group_cache_tree);
10189 RB_CLEAR_NODE(&cache->cache_node);
10190 spin_unlock(&root->fs_info->block_group_cache_lock);
10191 btrfs_put_block_group(cache);
10192 return ret;
10193 }
10194 update_global_block_rsv(root->fs_info);
10195
10196 spin_lock(&cache->space_info->lock);
10197 cache->space_info->bytes_readonly += cache->bytes_super;
10198 spin_unlock(&cache->space_info->lock);
10199
10200 __link_block_group(cache->space_info, cache);
10201
10202 list_add_tail(&cache->bg_list, &trans->new_bgs);
10203
10204 set_avail_alloc_bits(extent_root->fs_info, type);
10205
10206 return 0;
10207 }
10208
10209 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10210 {
10211 u64 extra_flags = chunk_to_extended(flags) &
10212 BTRFS_EXTENDED_PROFILE_MASK;
10213
10214 write_seqlock(&fs_info->profiles_lock);
10215 if (flags & BTRFS_BLOCK_GROUP_DATA)
10216 fs_info->avail_data_alloc_bits &= ~extra_flags;
10217 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10218 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10219 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10220 fs_info->avail_system_alloc_bits &= ~extra_flags;
10221 write_sequnlock(&fs_info->profiles_lock);
10222 }
10223
10224 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10225 struct btrfs_root *root, u64 group_start,
10226 struct extent_map *em)
10227 {
10228 struct btrfs_path *path;
10229 struct btrfs_block_group_cache *block_group;
10230 struct btrfs_free_cluster *cluster;
10231 struct btrfs_root *tree_root = root->fs_info->tree_root;
10232 struct btrfs_key key;
10233 struct inode *inode;
10234 struct kobject *kobj = NULL;
10235 int ret;
10236 int index;
10237 int factor;
10238 struct btrfs_caching_control *caching_ctl = NULL;
10239 bool remove_em;
10240
10241 root = root->fs_info->extent_root;
10242
10243 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
10244 BUG_ON(!block_group);
10245 BUG_ON(!block_group->ro);
10246
10247 /*
10248 * Free the reserved super bytes from this block group before
10249 * remove it.
10250 */
10251 free_excluded_extents(root, block_group);
10252
10253 memcpy(&key, &block_group->key, sizeof(key));
10254 index = get_block_group_index(block_group);
10255 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
10256 BTRFS_BLOCK_GROUP_RAID1 |
10257 BTRFS_BLOCK_GROUP_RAID10))
10258 factor = 2;
10259 else
10260 factor = 1;
10261
10262 /* make sure this block group isn't part of an allocation cluster */
10263 cluster = &root->fs_info->data_alloc_cluster;
10264 spin_lock(&cluster->refill_lock);
10265 btrfs_return_cluster_to_free_space(block_group, cluster);
10266 spin_unlock(&cluster->refill_lock);
10267
10268 /*
10269 * make sure this block group isn't part of a metadata
10270 * allocation cluster
10271 */
10272 cluster = &root->fs_info->meta_alloc_cluster;
10273 spin_lock(&cluster->refill_lock);
10274 btrfs_return_cluster_to_free_space(block_group, cluster);
10275 spin_unlock(&cluster->refill_lock);
10276
10277 path = btrfs_alloc_path();
10278 if (!path) {
10279 ret = -ENOMEM;
10280 goto out;
10281 }
10282
10283 /*
10284 * get the inode first so any iput calls done for the io_list
10285 * aren't the final iput (no unlinks allowed now)
10286 */
10287 inode = lookup_free_space_inode(tree_root, block_group, path);
10288
10289 mutex_lock(&trans->transaction->cache_write_mutex);
10290 /*
10291 * make sure our free spache cache IO is done before remove the
10292 * free space inode
10293 */
10294 spin_lock(&trans->transaction->dirty_bgs_lock);
10295 if (!list_empty(&block_group->io_list)) {
10296 list_del_init(&block_group->io_list);
10297
10298 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10299
10300 spin_unlock(&trans->transaction->dirty_bgs_lock);
10301 btrfs_wait_cache_io(root, trans, block_group,
10302 &block_group->io_ctl, path,
10303 block_group->key.objectid);
10304 btrfs_put_block_group(block_group);
10305 spin_lock(&trans->transaction->dirty_bgs_lock);
10306 }
10307
10308 if (!list_empty(&block_group->dirty_list)) {
10309 list_del_init(&block_group->dirty_list);
10310 btrfs_put_block_group(block_group);
10311 }
10312 spin_unlock(&trans->transaction->dirty_bgs_lock);
10313 mutex_unlock(&trans->transaction->cache_write_mutex);
10314
10315 if (!IS_ERR(inode)) {
10316 ret = btrfs_orphan_add(trans, inode);
10317 if (ret) {
10318 btrfs_add_delayed_iput(inode);
10319 goto out;
10320 }
10321 clear_nlink(inode);
10322 /* One for the block groups ref */
10323 spin_lock(&block_group->lock);
10324 if (block_group->iref) {
10325 block_group->iref = 0;
10326 block_group->inode = NULL;
10327 spin_unlock(&block_group->lock);
10328 iput(inode);
10329 } else {
10330 spin_unlock(&block_group->lock);
10331 }
10332 /* One for our lookup ref */
10333 btrfs_add_delayed_iput(inode);
10334 }
10335
10336 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10337 key.offset = block_group->key.objectid;
10338 key.type = 0;
10339
10340 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10341 if (ret < 0)
10342 goto out;
10343 if (ret > 0)
10344 btrfs_release_path(path);
10345 if (ret == 0) {
10346 ret = btrfs_del_item(trans, tree_root, path);
10347 if (ret)
10348 goto out;
10349 btrfs_release_path(path);
10350 }
10351
10352 spin_lock(&root->fs_info->block_group_cache_lock);
10353 rb_erase(&block_group->cache_node,
10354 &root->fs_info->block_group_cache_tree);
10355 RB_CLEAR_NODE(&block_group->cache_node);
10356
10357 if (root->fs_info->first_logical_byte == block_group->key.objectid)
10358 root->fs_info->first_logical_byte = (u64)-1;
10359 spin_unlock(&root->fs_info->block_group_cache_lock);
10360
10361 down_write(&block_group->space_info->groups_sem);
10362 /*
10363 * we must use list_del_init so people can check to see if they
10364 * are still on the list after taking the semaphore
10365 */
10366 list_del_init(&block_group->list);
10367 if (list_empty(&block_group->space_info->block_groups[index])) {
10368 kobj = block_group->space_info->block_group_kobjs[index];
10369 block_group->space_info->block_group_kobjs[index] = NULL;
10370 clear_avail_alloc_bits(root->fs_info, block_group->flags);
10371 }
10372 up_write(&block_group->space_info->groups_sem);
10373 if (kobj) {
10374 kobject_del(kobj);
10375 kobject_put(kobj);
10376 }
10377
10378 if (block_group->has_caching_ctl)
10379 caching_ctl = get_caching_control(block_group);
10380 if (block_group->cached == BTRFS_CACHE_STARTED)
10381 wait_block_group_cache_done(block_group);
10382 if (block_group->has_caching_ctl) {
10383 down_write(&root->fs_info->commit_root_sem);
10384 if (!caching_ctl) {
10385 struct btrfs_caching_control *ctl;
10386
10387 list_for_each_entry(ctl,
10388 &root->fs_info->caching_block_groups, list)
10389 if (ctl->block_group == block_group) {
10390 caching_ctl = ctl;
10391 atomic_inc(&caching_ctl->count);
10392 break;
10393 }
10394 }
10395 if (caching_ctl)
10396 list_del_init(&caching_ctl->list);
10397 up_write(&root->fs_info->commit_root_sem);
10398 if (caching_ctl) {
10399 /* Once for the caching bgs list and once for us. */
10400 put_caching_control(caching_ctl);
10401 put_caching_control(caching_ctl);
10402 }
10403 }
10404
10405 spin_lock(&trans->transaction->dirty_bgs_lock);
10406 if (!list_empty(&block_group->dirty_list)) {
10407 WARN_ON(1);
10408 }
10409 if (!list_empty(&block_group->io_list)) {
10410 WARN_ON(1);
10411 }
10412 spin_unlock(&trans->transaction->dirty_bgs_lock);
10413 btrfs_remove_free_space_cache(block_group);
10414
10415 spin_lock(&block_group->space_info->lock);
10416 list_del_init(&block_group->ro_list);
10417
10418 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
10419 WARN_ON(block_group->space_info->total_bytes
10420 < block_group->key.offset);
10421 WARN_ON(block_group->space_info->bytes_readonly
10422 < block_group->key.offset);
10423 WARN_ON(block_group->space_info->disk_total
10424 < block_group->key.offset * factor);
10425 }
10426 block_group->space_info->total_bytes -= block_group->key.offset;
10427 block_group->space_info->bytes_readonly -= block_group->key.offset;
10428 block_group->space_info->disk_total -= block_group->key.offset * factor;
10429
10430 spin_unlock(&block_group->space_info->lock);
10431
10432 memcpy(&key, &block_group->key, sizeof(key));
10433
10434 lock_chunks(root);
10435 if (!list_empty(&em->list)) {
10436 /* We're in the transaction->pending_chunks list. */
10437 free_extent_map(em);
10438 }
10439 spin_lock(&block_group->lock);
10440 block_group->removed = 1;
10441 /*
10442 * At this point trimming can't start on this block group, because we
10443 * removed the block group from the tree fs_info->block_group_cache_tree
10444 * so no one can't find it anymore and even if someone already got this
10445 * block group before we removed it from the rbtree, they have already
10446 * incremented block_group->trimming - if they didn't, they won't find
10447 * any free space entries because we already removed them all when we
10448 * called btrfs_remove_free_space_cache().
10449 *
10450 * And we must not remove the extent map from the fs_info->mapping_tree
10451 * to prevent the same logical address range and physical device space
10452 * ranges from being reused for a new block group. This is because our
10453 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10454 * completely transactionless, so while it is trimming a range the
10455 * currently running transaction might finish and a new one start,
10456 * allowing for new block groups to be created that can reuse the same
10457 * physical device locations unless we take this special care.
10458 *
10459 * There may also be an implicit trim operation if the file system
10460 * is mounted with -odiscard. The same protections must remain
10461 * in place until the extents have been discarded completely when
10462 * the transaction commit has completed.
10463 */
10464 remove_em = (atomic_read(&block_group->trimming) == 0);
10465 /*
10466 * Make sure a trimmer task always sees the em in the pinned_chunks list
10467 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10468 * before checking block_group->removed).
10469 */
10470 if (!remove_em) {
10471 /*
10472 * Our em might be in trans->transaction->pending_chunks which
10473 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10474 * and so is the fs_info->pinned_chunks list.
10475 *
10476 * So at this point we must be holding the chunk_mutex to avoid
10477 * any races with chunk allocation (more specifically at
10478 * volumes.c:contains_pending_extent()), to ensure it always
10479 * sees the em, either in the pending_chunks list or in the
10480 * pinned_chunks list.
10481 */
10482 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
10483 }
10484 spin_unlock(&block_group->lock);
10485
10486 if (remove_em) {
10487 struct extent_map_tree *em_tree;
10488
10489 em_tree = &root->fs_info->mapping_tree.map_tree;
10490 write_lock(&em_tree->lock);
10491 /*
10492 * The em might be in the pending_chunks list, so make sure the
10493 * chunk mutex is locked, since remove_extent_mapping() will
10494 * delete us from that list.
10495 */
10496 remove_extent_mapping(em_tree, em);
10497 write_unlock(&em_tree->lock);
10498 /* once for the tree */
10499 free_extent_map(em);
10500 }
10501
10502 unlock_chunks(root);
10503
10504 ret = remove_block_group_free_space(trans, root->fs_info, block_group);
10505 if (ret)
10506 goto out;
10507
10508 btrfs_put_block_group(block_group);
10509 btrfs_put_block_group(block_group);
10510
10511 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10512 if (ret > 0)
10513 ret = -EIO;
10514 if (ret < 0)
10515 goto out;
10516
10517 ret = btrfs_del_item(trans, root, path);
10518 out:
10519 btrfs_free_path(path);
10520 return ret;
10521 }
10522
10523 struct btrfs_trans_handle *
10524 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10525 const u64 chunk_offset)
10526 {
10527 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10528 struct extent_map *em;
10529 struct map_lookup *map;
10530 unsigned int num_items;
10531
10532 read_lock(&em_tree->lock);
10533 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10534 read_unlock(&em_tree->lock);
10535 ASSERT(em && em->start == chunk_offset);
10536
10537 /*
10538 * We need to reserve 3 + N units from the metadata space info in order
10539 * to remove a block group (done at btrfs_remove_chunk() and at
10540 * btrfs_remove_block_group()), which are used for:
10541 *
10542 * 1 unit for adding the free space inode's orphan (located in the tree
10543 * of tree roots).
10544 * 1 unit for deleting the block group item (located in the extent
10545 * tree).
10546 * 1 unit for deleting the free space item (located in tree of tree
10547 * roots).
10548 * N units for deleting N device extent items corresponding to each
10549 * stripe (located in the device tree).
10550 *
10551 * In order to remove a block group we also need to reserve units in the
10552 * system space info in order to update the chunk tree (update one or
10553 * more device items and remove one chunk item), but this is done at
10554 * btrfs_remove_chunk() through a call to check_system_chunk().
10555 */
10556 map = em->map_lookup;
10557 num_items = 3 + map->num_stripes;
10558 free_extent_map(em);
10559
10560 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10561 num_items, 1);
10562 }
10563
10564 /*
10565 * Process the unused_bgs list and remove any that don't have any allocated
10566 * space inside of them.
10567 */
10568 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10569 {
10570 struct btrfs_block_group_cache *block_group;
10571 struct btrfs_space_info *space_info;
10572 struct btrfs_root *root = fs_info->extent_root;
10573 struct btrfs_trans_handle *trans;
10574 int ret = 0;
10575
10576 if (!fs_info->open)
10577 return;
10578
10579 spin_lock(&fs_info->unused_bgs_lock);
10580 while (!list_empty(&fs_info->unused_bgs)) {
10581 u64 start, end;
10582 int trimming;
10583
10584 block_group = list_first_entry(&fs_info->unused_bgs,
10585 struct btrfs_block_group_cache,
10586 bg_list);
10587 list_del_init(&block_group->bg_list);
10588
10589 space_info = block_group->space_info;
10590
10591 if (ret || btrfs_mixed_space_info(space_info)) {
10592 btrfs_put_block_group(block_group);
10593 continue;
10594 }
10595 spin_unlock(&fs_info->unused_bgs_lock);
10596
10597 mutex_lock(&fs_info->delete_unused_bgs_mutex);
10598
10599 /* Don't want to race with allocators so take the groups_sem */
10600 down_write(&space_info->groups_sem);
10601 spin_lock(&block_group->lock);
10602 if (block_group->reserved ||
10603 btrfs_block_group_used(&block_group->item) ||
10604 block_group->ro ||
10605 list_is_singular(&block_group->list)) {
10606 /*
10607 * We want to bail if we made new allocations or have
10608 * outstanding allocations in this block group. We do
10609 * the ro check in case balance is currently acting on
10610 * this block group.
10611 */
10612 spin_unlock(&block_group->lock);
10613 up_write(&space_info->groups_sem);
10614 goto next;
10615 }
10616 spin_unlock(&block_group->lock);
10617
10618 /* We don't want to force the issue, only flip if it's ok. */
10619 ret = inc_block_group_ro(block_group, 0);
10620 up_write(&space_info->groups_sem);
10621 if (ret < 0) {
10622 ret = 0;
10623 goto next;
10624 }
10625
10626 /*
10627 * Want to do this before we do anything else so we can recover
10628 * properly if we fail to join the transaction.
10629 */
10630 trans = btrfs_start_trans_remove_block_group(fs_info,
10631 block_group->key.objectid);
10632 if (IS_ERR(trans)) {
10633 btrfs_dec_block_group_ro(root, block_group);
10634 ret = PTR_ERR(trans);
10635 goto next;
10636 }
10637
10638 /*
10639 * We could have pending pinned extents for this block group,
10640 * just delete them, we don't care about them anymore.
10641 */
10642 start = block_group->key.objectid;
10643 end = start + block_group->key.offset - 1;
10644 /*
10645 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10646 * btrfs_finish_extent_commit(). If we are at transaction N,
10647 * another task might be running finish_extent_commit() for the
10648 * previous transaction N - 1, and have seen a range belonging
10649 * to the block group in freed_extents[] before we were able to
10650 * clear the whole block group range from freed_extents[]. This
10651 * means that task can lookup for the block group after we
10652 * unpinned it from freed_extents[] and removed it, leading to
10653 * a BUG_ON() at btrfs_unpin_extent_range().
10654 */
10655 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10656 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10657 EXTENT_DIRTY);
10658 if (ret) {
10659 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10660 btrfs_dec_block_group_ro(root, block_group);
10661 goto end_trans;
10662 }
10663 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10664 EXTENT_DIRTY);
10665 if (ret) {
10666 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10667 btrfs_dec_block_group_ro(root, block_group);
10668 goto end_trans;
10669 }
10670 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10671
10672 /* Reset pinned so btrfs_put_block_group doesn't complain */
10673 spin_lock(&space_info->lock);
10674 spin_lock(&block_group->lock);
10675
10676 space_info->bytes_pinned -= block_group->pinned;
10677 space_info->bytes_readonly += block_group->pinned;
10678 percpu_counter_add(&space_info->total_bytes_pinned,
10679 -block_group->pinned);
10680 block_group->pinned = 0;
10681
10682 spin_unlock(&block_group->lock);
10683 spin_unlock(&space_info->lock);
10684
10685 /* DISCARD can flip during remount */
10686 trimming = btrfs_test_opt(root, DISCARD);
10687
10688 /* Implicit trim during transaction commit. */
10689 if (trimming)
10690 btrfs_get_block_group_trimming(block_group);
10691
10692 /*
10693 * Btrfs_remove_chunk will abort the transaction if things go
10694 * horribly wrong.
10695 */
10696 ret = btrfs_remove_chunk(trans, root,
10697 block_group->key.objectid);
10698
10699 if (ret) {
10700 if (trimming)
10701 btrfs_put_block_group_trimming(block_group);
10702 goto end_trans;
10703 }
10704
10705 /*
10706 * If we're not mounted with -odiscard, we can just forget
10707 * about this block group. Otherwise we'll need to wait
10708 * until transaction commit to do the actual discard.
10709 */
10710 if (trimming) {
10711 spin_lock(&fs_info->unused_bgs_lock);
10712 /*
10713 * A concurrent scrub might have added us to the list
10714 * fs_info->unused_bgs, so use a list_move operation
10715 * to add the block group to the deleted_bgs list.
10716 */
10717 list_move(&block_group->bg_list,
10718 &trans->transaction->deleted_bgs);
10719 spin_unlock(&fs_info->unused_bgs_lock);
10720 btrfs_get_block_group(block_group);
10721 }
10722 end_trans:
10723 btrfs_end_transaction(trans, root);
10724 next:
10725 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
10726 btrfs_put_block_group(block_group);
10727 spin_lock(&fs_info->unused_bgs_lock);
10728 }
10729 spin_unlock(&fs_info->unused_bgs_lock);
10730 }
10731
10732 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10733 {
10734 struct btrfs_space_info *space_info;
10735 struct btrfs_super_block *disk_super;
10736 u64 features;
10737 u64 flags;
10738 int mixed = 0;
10739 int ret;
10740
10741 disk_super = fs_info->super_copy;
10742 if (!btrfs_super_root(disk_super))
10743 return -EINVAL;
10744
10745 features = btrfs_super_incompat_flags(disk_super);
10746 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10747 mixed = 1;
10748
10749 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10750 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10751 if (ret)
10752 goto out;
10753
10754 if (mixed) {
10755 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10756 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10757 } else {
10758 flags = BTRFS_BLOCK_GROUP_METADATA;
10759 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10760 if (ret)
10761 goto out;
10762
10763 flags = BTRFS_BLOCK_GROUP_DATA;
10764 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10765 }
10766 out:
10767 return ret;
10768 }
10769
10770 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
10771 {
10772 return unpin_extent_range(root, start, end, false);
10773 }
10774
10775 /*
10776 * It used to be that old block groups would be left around forever.
10777 * Iterating over them would be enough to trim unused space. Since we
10778 * now automatically remove them, we also need to iterate over unallocated
10779 * space.
10780 *
10781 * We don't want a transaction for this since the discard may take a
10782 * substantial amount of time. We don't require that a transaction be
10783 * running, but we do need to take a running transaction into account
10784 * to ensure that we're not discarding chunks that were released in
10785 * the current transaction.
10786 *
10787 * Holding the chunks lock will prevent other threads from allocating
10788 * or releasing chunks, but it won't prevent a running transaction
10789 * from committing and releasing the memory that the pending chunks
10790 * list head uses. For that, we need to take a reference to the
10791 * transaction.
10792 */
10793 static int btrfs_trim_free_extents(struct btrfs_device *device,
10794 u64 minlen, u64 *trimmed)
10795 {
10796 u64 start = 0, len = 0;
10797 int ret;
10798
10799 *trimmed = 0;
10800
10801 /* Not writeable = nothing to do. */
10802 if (!device->writeable)
10803 return 0;
10804
10805 /* No free space = nothing to do. */
10806 if (device->total_bytes <= device->bytes_used)
10807 return 0;
10808
10809 ret = 0;
10810
10811 while (1) {
10812 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
10813 struct btrfs_transaction *trans;
10814 u64 bytes;
10815
10816 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
10817 if (ret)
10818 return ret;
10819
10820 down_read(&fs_info->commit_root_sem);
10821
10822 spin_lock(&fs_info->trans_lock);
10823 trans = fs_info->running_transaction;
10824 if (trans)
10825 atomic_inc(&trans->use_count);
10826 spin_unlock(&fs_info->trans_lock);
10827
10828 ret = find_free_dev_extent_start(trans, device, minlen, start,
10829 &start, &len);
10830 if (trans)
10831 btrfs_put_transaction(trans);
10832
10833 if (ret) {
10834 up_read(&fs_info->commit_root_sem);
10835 mutex_unlock(&fs_info->chunk_mutex);
10836 if (ret == -ENOSPC)
10837 ret = 0;
10838 break;
10839 }
10840
10841 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
10842 up_read(&fs_info->commit_root_sem);
10843 mutex_unlock(&fs_info->chunk_mutex);
10844
10845 if (ret)
10846 break;
10847
10848 start += len;
10849 *trimmed += bytes;
10850
10851 if (fatal_signal_pending(current)) {
10852 ret = -ERESTARTSYS;
10853 break;
10854 }
10855
10856 cond_resched();
10857 }
10858
10859 return ret;
10860 }
10861
10862 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
10863 {
10864 struct btrfs_fs_info *fs_info = root->fs_info;
10865 struct btrfs_block_group_cache *cache = NULL;
10866 struct btrfs_device *device;
10867 struct list_head *devices;
10868 u64 group_trimmed;
10869 u64 start;
10870 u64 end;
10871 u64 trimmed = 0;
10872 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10873 int ret = 0;
10874
10875 /*
10876 * try to trim all FS space, our block group may start from non-zero.
10877 */
10878 if (range->len == total_bytes)
10879 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10880 else
10881 cache = btrfs_lookup_block_group(fs_info, range->start);
10882
10883 while (cache) {
10884 if (cache->key.objectid >= (range->start + range->len)) {
10885 btrfs_put_block_group(cache);
10886 break;
10887 }
10888
10889 start = max(range->start, cache->key.objectid);
10890 end = min(range->start + range->len,
10891 cache->key.objectid + cache->key.offset);
10892
10893 if (end - start >= range->minlen) {
10894 if (!block_group_cache_done(cache)) {
10895 ret = cache_block_group(cache, 0);
10896 if (ret) {
10897 btrfs_put_block_group(cache);
10898 break;
10899 }
10900 ret = wait_block_group_cache_done(cache);
10901 if (ret) {
10902 btrfs_put_block_group(cache);
10903 break;
10904 }
10905 }
10906 ret = btrfs_trim_block_group(cache,
10907 &group_trimmed,
10908 start,
10909 end,
10910 range->minlen);
10911
10912 trimmed += group_trimmed;
10913 if (ret) {
10914 btrfs_put_block_group(cache);
10915 break;
10916 }
10917 }
10918
10919 cache = next_block_group(fs_info->tree_root, cache);
10920 }
10921
10922 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
10923 devices = &root->fs_info->fs_devices->alloc_list;
10924 list_for_each_entry(device, devices, dev_alloc_list) {
10925 ret = btrfs_trim_free_extents(device, range->minlen,
10926 &group_trimmed);
10927 if (ret)
10928 break;
10929
10930 trimmed += group_trimmed;
10931 }
10932 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
10933
10934 range->len = trimmed;
10935 return ret;
10936 }
10937
10938 /*
10939 * btrfs_{start,end}_write_no_snapshoting() are similar to
10940 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10941 * data into the page cache through nocow before the subvolume is snapshoted,
10942 * but flush the data into disk after the snapshot creation, or to prevent
10943 * operations while snapshoting is ongoing and that cause the snapshot to be
10944 * inconsistent (writes followed by expanding truncates for example).
10945 */
10946 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
10947 {
10948 percpu_counter_dec(&root->subv_writers->counter);
10949 /*
10950 * Make sure counter is updated before we wake up waiters.
10951 */
10952 smp_mb();
10953 if (waitqueue_active(&root->subv_writers->wait))
10954 wake_up(&root->subv_writers->wait);
10955 }
10956
10957 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
10958 {
10959 if (atomic_read(&root->will_be_snapshoted))
10960 return 0;
10961
10962 percpu_counter_inc(&root->subv_writers->counter);
10963 /*
10964 * Make sure counter is updated before we check for snapshot creation.
10965 */
10966 smp_mb();
10967 if (atomic_read(&root->will_be_snapshoted)) {
10968 btrfs_end_write_no_snapshoting(root);
10969 return 0;
10970 }
10971 return 1;
10972 }
10973
10974 static int wait_snapshoting_atomic_t(atomic_t *a)
10975 {
10976 schedule();
10977 return 0;
10978 }
10979
10980 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
10981 {
10982 while (true) {
10983 int ret;
10984
10985 ret = btrfs_start_write_no_snapshoting(root);
10986 if (ret)
10987 break;
10988 wait_on_atomic_t(&root->will_be_snapshoted,
10989 wait_snapshoting_atomic_t,
10990 TASK_UNINTERRUPTIBLE);
10991 }
10992 }
(deprecated) Btrfs devel tree