Linux 3.12.28
[linux/fpc-iii.git] / fs / btrfs / extent-tree.c
blob63ee604efa6c93f7b9eb76ed47fd6bb4ed38b310
1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
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.
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.
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.
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 "compat.h"
29 #include "hash.h"
30 #include "ctree.h"
31 #include "disk-io.h"
32 #include "print-tree.h"
33 #include "transaction.h"
34 #include "volumes.h"
35 #include "raid56.h"
36 #include "locking.h"
37 #include "free-space-cache.h"
38 #include "math.h"
40 #undef SCRAMBLE_DELAYED_REFS
43 * control flags for do_chunk_alloc's force field
44 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
45 * if we really need one.
47 * CHUNK_ALLOC_LIMITED means to only try and allocate one
48 * if we have very few chunks already allocated. This is
49 * used as part of the clustering code to help make sure
50 * we have a good pool of storage to cluster in, without
51 * filling the FS with empty chunks
53 * CHUNK_ALLOC_FORCE means it must try to allocate one
56 enum {
57 CHUNK_ALLOC_NO_FORCE = 0,
58 CHUNK_ALLOC_LIMITED = 1,
59 CHUNK_ALLOC_FORCE = 2,
63 * Control how reservations are dealt with.
65 * RESERVE_FREE - freeing a reservation.
66 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
67 * ENOSPC accounting
68 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
69 * bytes_may_use as the ENOSPC accounting is done elsewhere
71 enum {
72 RESERVE_FREE = 0,
73 RESERVE_ALLOC = 1,
74 RESERVE_ALLOC_NO_ACCOUNT = 2,
77 static int update_block_group(struct btrfs_root *root,
78 u64 bytenr, u64 num_bytes, int alloc);
79 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
80 struct btrfs_root *root,
81 u64 bytenr, u64 num_bytes, u64 parent,
82 u64 root_objectid, u64 owner_objectid,
83 u64 owner_offset, int refs_to_drop,
84 struct btrfs_delayed_extent_op *extra_op);
85 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
86 struct extent_buffer *leaf,
87 struct btrfs_extent_item *ei);
88 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
89 struct btrfs_root *root,
90 u64 parent, u64 root_objectid,
91 u64 flags, u64 owner, u64 offset,
92 struct btrfs_key *ins, int ref_mod);
93 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
94 struct btrfs_root *root,
95 u64 parent, u64 root_objectid,
96 u64 flags, struct btrfs_disk_key *key,
97 int level, struct btrfs_key *ins);
98 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
99 struct btrfs_root *extent_root, u64 flags,
100 int force);
101 static int find_next_key(struct btrfs_path *path, int level,
102 struct btrfs_key *key);
103 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
104 int dump_block_groups);
105 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
106 u64 num_bytes, int reserve);
107 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
108 u64 num_bytes);
109 int btrfs_pin_extent(struct btrfs_root *root,
110 u64 bytenr, u64 num_bytes, int reserved);
112 static noinline int
113 block_group_cache_done(struct btrfs_block_group_cache *cache)
115 smp_mb();
116 return cache->cached == BTRFS_CACHE_FINISHED ||
117 cache->cached == BTRFS_CACHE_ERROR;
120 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
122 return (cache->flags & bits) == bits;
125 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
127 atomic_inc(&cache->count);
130 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
132 if (atomic_dec_and_test(&cache->count)) {
133 WARN_ON(cache->pinned > 0);
134 WARN_ON(cache->reserved > 0);
135 kfree(cache->free_space_ctl);
136 kfree(cache);
141 * this adds the block group to the fs_info rb tree for the block group
142 * cache
144 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
145 struct btrfs_block_group_cache *block_group)
147 struct rb_node **p;
148 struct rb_node *parent = NULL;
149 struct btrfs_block_group_cache *cache;
151 spin_lock(&info->block_group_cache_lock);
152 p = &info->block_group_cache_tree.rb_node;
154 while (*p) {
155 parent = *p;
156 cache = rb_entry(parent, struct btrfs_block_group_cache,
157 cache_node);
158 if (block_group->key.objectid < cache->key.objectid) {
159 p = &(*p)->rb_left;
160 } else if (block_group->key.objectid > cache->key.objectid) {
161 p = &(*p)->rb_right;
162 } else {
163 spin_unlock(&info->block_group_cache_lock);
164 return -EEXIST;
168 rb_link_node(&block_group->cache_node, parent, p);
169 rb_insert_color(&block_group->cache_node,
170 &info->block_group_cache_tree);
172 if (info->first_logical_byte > block_group->key.objectid)
173 info->first_logical_byte = block_group->key.objectid;
175 spin_unlock(&info->block_group_cache_lock);
177 return 0;
181 * This will return the block group at or after bytenr if contains is 0, else
182 * it will return the block group that contains the bytenr
184 static struct btrfs_block_group_cache *
185 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
186 int contains)
188 struct btrfs_block_group_cache *cache, *ret = NULL;
189 struct rb_node *n;
190 u64 end, start;
192 spin_lock(&info->block_group_cache_lock);
193 n = info->block_group_cache_tree.rb_node;
195 while (n) {
196 cache = rb_entry(n, struct btrfs_block_group_cache,
197 cache_node);
198 end = cache->key.objectid + cache->key.offset - 1;
199 start = cache->key.objectid;
201 if (bytenr < start) {
202 if (!contains && (!ret || start < ret->key.objectid))
203 ret = cache;
204 n = n->rb_left;
205 } else if (bytenr > start) {
206 if (contains && bytenr <= end) {
207 ret = cache;
208 break;
210 n = n->rb_right;
211 } else {
212 ret = cache;
213 break;
216 if (ret) {
217 btrfs_get_block_group(ret);
218 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
219 info->first_logical_byte = ret->key.objectid;
221 spin_unlock(&info->block_group_cache_lock);
223 return ret;
226 static int add_excluded_extent(struct btrfs_root *root,
227 u64 start, u64 num_bytes)
229 u64 end = start + num_bytes - 1;
230 set_extent_bits(&root->fs_info->freed_extents[0],
231 start, end, EXTENT_UPTODATE, GFP_NOFS);
232 set_extent_bits(&root->fs_info->freed_extents[1],
233 start, end, EXTENT_UPTODATE, GFP_NOFS);
234 return 0;
237 static void free_excluded_extents(struct btrfs_root *root,
238 struct btrfs_block_group_cache *cache)
240 u64 start, end;
242 start = cache->key.objectid;
243 end = start + cache->key.offset - 1;
245 clear_extent_bits(&root->fs_info->freed_extents[0],
246 start, end, EXTENT_UPTODATE, GFP_NOFS);
247 clear_extent_bits(&root->fs_info->freed_extents[1],
248 start, end, EXTENT_UPTODATE, GFP_NOFS);
251 static int exclude_super_stripes(struct btrfs_root *root,
252 struct btrfs_block_group_cache *cache)
254 u64 bytenr;
255 u64 *logical;
256 int stripe_len;
257 int i, nr, ret;
259 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
260 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
261 cache->bytes_super += stripe_len;
262 ret = add_excluded_extent(root, cache->key.objectid,
263 stripe_len);
264 if (ret)
265 return ret;
268 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
269 bytenr = btrfs_sb_offset(i);
270 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
271 cache->key.objectid, bytenr,
272 0, &logical, &nr, &stripe_len);
273 if (ret)
274 return ret;
276 while (nr--) {
277 u64 start, len;
279 if (logical[nr] > cache->key.objectid +
280 cache->key.offset)
281 continue;
283 if (logical[nr] + stripe_len <= cache->key.objectid)
284 continue;
286 start = logical[nr];
287 if (start < cache->key.objectid) {
288 start = cache->key.objectid;
289 len = (logical[nr] + stripe_len) - start;
290 } else {
291 len = min_t(u64, stripe_len,
292 cache->key.objectid +
293 cache->key.offset - start);
296 cache->bytes_super += len;
297 ret = add_excluded_extent(root, start, len);
298 if (ret) {
299 kfree(logical);
300 return ret;
304 kfree(logical);
306 return 0;
309 static struct btrfs_caching_control *
310 get_caching_control(struct btrfs_block_group_cache *cache)
312 struct btrfs_caching_control *ctl;
314 spin_lock(&cache->lock);
315 if (cache->cached != BTRFS_CACHE_STARTED) {
316 spin_unlock(&cache->lock);
317 return NULL;
320 /* We're loading it the fast way, so we don't have a caching_ctl. */
321 if (!cache->caching_ctl) {
322 spin_unlock(&cache->lock);
323 return NULL;
326 ctl = cache->caching_ctl;
327 atomic_inc(&ctl->count);
328 spin_unlock(&cache->lock);
329 return ctl;
332 static void put_caching_control(struct btrfs_caching_control *ctl)
334 if (atomic_dec_and_test(&ctl->count))
335 kfree(ctl);
339 * this is only called by cache_block_group, since we could have freed extents
340 * we need to check the pinned_extents for any extents that can't be used yet
341 * since their free space will be released as soon as the transaction commits.
343 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
344 struct btrfs_fs_info *info, u64 start, u64 end)
346 u64 extent_start, extent_end, size, total_added = 0;
347 int ret;
349 while (start < end) {
350 ret = find_first_extent_bit(info->pinned_extents, start,
351 &extent_start, &extent_end,
352 EXTENT_DIRTY | EXTENT_UPTODATE,
353 NULL);
354 if (ret)
355 break;
357 if (extent_start <= start) {
358 start = extent_end + 1;
359 } else if (extent_start > start && extent_start < end) {
360 size = extent_start - start;
361 total_added += size;
362 ret = btrfs_add_free_space(block_group, start,
363 size);
364 BUG_ON(ret); /* -ENOMEM or logic error */
365 start = extent_end + 1;
366 } else {
367 break;
371 if (start < end) {
372 size = end - start;
373 total_added += size;
374 ret = btrfs_add_free_space(block_group, start, size);
375 BUG_ON(ret); /* -ENOMEM or logic error */
378 return total_added;
381 static noinline void caching_thread(struct btrfs_work *work)
383 struct btrfs_block_group_cache *block_group;
384 struct btrfs_fs_info *fs_info;
385 struct btrfs_caching_control *caching_ctl;
386 struct btrfs_root *extent_root;
387 struct btrfs_path *path;
388 struct extent_buffer *leaf;
389 struct btrfs_key key;
390 u64 total_found = 0;
391 u64 last = 0;
392 u32 nritems;
393 int ret = -ENOMEM;
395 caching_ctl = container_of(work, struct btrfs_caching_control, work);
396 block_group = caching_ctl->block_group;
397 fs_info = block_group->fs_info;
398 extent_root = fs_info->extent_root;
400 path = btrfs_alloc_path();
401 if (!path)
402 goto out;
404 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
407 * We don't want to deadlock with somebody trying to allocate a new
408 * extent for the extent root while also trying to search the extent
409 * root to add free space. So we skip locking and search the commit
410 * root, since its read-only
412 path->skip_locking = 1;
413 path->search_commit_root = 1;
414 path->reada = 1;
416 key.objectid = last;
417 key.offset = 0;
418 key.type = BTRFS_EXTENT_ITEM_KEY;
419 again:
420 mutex_lock(&caching_ctl->mutex);
421 /* need to make sure the commit_root doesn't disappear */
422 down_read(&fs_info->extent_commit_sem);
424 next:
425 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
426 if (ret < 0)
427 goto err;
429 leaf = path->nodes[0];
430 nritems = btrfs_header_nritems(leaf);
432 while (1) {
433 if (btrfs_fs_closing(fs_info) > 1) {
434 last = (u64)-1;
435 break;
438 if (path->slots[0] < nritems) {
439 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
440 } else {
441 ret = find_next_key(path, 0, &key);
442 if (ret)
443 break;
445 if (need_resched()) {
446 caching_ctl->progress = last;
447 btrfs_release_path(path);
448 up_read(&fs_info->extent_commit_sem);
449 mutex_unlock(&caching_ctl->mutex);
450 cond_resched();
451 goto again;
454 ret = btrfs_next_leaf(extent_root, path);
455 if (ret < 0)
456 goto err;
457 if (ret)
458 break;
459 leaf = path->nodes[0];
460 nritems = btrfs_header_nritems(leaf);
461 continue;
464 if (key.objectid < last) {
465 key.objectid = last;
466 key.offset = 0;
467 key.type = BTRFS_EXTENT_ITEM_KEY;
469 caching_ctl->progress = last;
470 btrfs_release_path(path);
471 goto next;
474 if (key.objectid < block_group->key.objectid) {
475 path->slots[0]++;
476 continue;
479 if (key.objectid >= block_group->key.objectid +
480 block_group->key.offset)
481 break;
483 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
484 key.type == BTRFS_METADATA_ITEM_KEY) {
485 total_found += add_new_free_space(block_group,
486 fs_info, last,
487 key.objectid);
488 if (key.type == BTRFS_METADATA_ITEM_KEY)
489 last = key.objectid +
490 fs_info->tree_root->leafsize;
491 else
492 last = key.objectid + key.offset;
494 if (total_found > (1024 * 1024 * 2)) {
495 total_found = 0;
496 wake_up(&caching_ctl->wait);
499 path->slots[0]++;
501 ret = 0;
503 total_found += add_new_free_space(block_group, fs_info, last,
504 block_group->key.objectid +
505 block_group->key.offset);
506 caching_ctl->progress = (u64)-1;
508 spin_lock(&block_group->lock);
509 block_group->caching_ctl = NULL;
510 block_group->cached = BTRFS_CACHE_FINISHED;
511 spin_unlock(&block_group->lock);
513 err:
514 btrfs_free_path(path);
515 up_read(&fs_info->extent_commit_sem);
517 free_excluded_extents(extent_root, block_group);
519 mutex_unlock(&caching_ctl->mutex);
520 out:
521 if (ret) {
522 spin_lock(&block_group->lock);
523 block_group->caching_ctl = NULL;
524 block_group->cached = BTRFS_CACHE_ERROR;
525 spin_unlock(&block_group->lock);
527 wake_up(&caching_ctl->wait);
529 put_caching_control(caching_ctl);
530 btrfs_put_block_group(block_group);
533 static int cache_block_group(struct btrfs_block_group_cache *cache,
534 int load_cache_only)
536 DEFINE_WAIT(wait);
537 struct btrfs_fs_info *fs_info = cache->fs_info;
538 struct btrfs_caching_control *caching_ctl;
539 int ret = 0;
541 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
542 if (!caching_ctl)
543 return -ENOMEM;
545 INIT_LIST_HEAD(&caching_ctl->list);
546 mutex_init(&caching_ctl->mutex);
547 init_waitqueue_head(&caching_ctl->wait);
548 caching_ctl->block_group = cache;
549 caching_ctl->progress = cache->key.objectid;
550 atomic_set(&caching_ctl->count, 1);
551 caching_ctl->work.func = caching_thread;
553 spin_lock(&cache->lock);
555 * This should be a rare occasion, but this could happen I think in the
556 * case where one thread starts to load the space cache info, and then
557 * some other thread starts a transaction commit which tries to do an
558 * allocation while the other thread is still loading the space cache
559 * info. The previous loop should have kept us from choosing this block
560 * group, but if we've moved to the state where we will wait on caching
561 * block groups we need to first check if we're doing a fast load here,
562 * so we can wait for it to finish, otherwise we could end up allocating
563 * from a block group who's cache gets evicted for one reason or
564 * another.
566 while (cache->cached == BTRFS_CACHE_FAST) {
567 struct btrfs_caching_control *ctl;
569 ctl = cache->caching_ctl;
570 atomic_inc(&ctl->count);
571 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
572 spin_unlock(&cache->lock);
574 schedule();
576 finish_wait(&ctl->wait, &wait);
577 put_caching_control(ctl);
578 spin_lock(&cache->lock);
581 if (cache->cached != BTRFS_CACHE_NO) {
582 spin_unlock(&cache->lock);
583 kfree(caching_ctl);
584 return 0;
586 WARN_ON(cache->caching_ctl);
587 cache->caching_ctl = caching_ctl;
588 cache->cached = BTRFS_CACHE_FAST;
589 spin_unlock(&cache->lock);
591 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
592 ret = load_free_space_cache(fs_info, cache);
594 spin_lock(&cache->lock);
595 if (ret == 1) {
596 cache->caching_ctl = NULL;
597 cache->cached = BTRFS_CACHE_FINISHED;
598 cache->last_byte_to_unpin = (u64)-1;
599 } else {
600 if (load_cache_only) {
601 cache->caching_ctl = NULL;
602 cache->cached = BTRFS_CACHE_NO;
603 } else {
604 cache->cached = BTRFS_CACHE_STARTED;
607 spin_unlock(&cache->lock);
608 wake_up(&caching_ctl->wait);
609 if (ret == 1) {
610 put_caching_control(caching_ctl);
611 free_excluded_extents(fs_info->extent_root, cache);
612 return 0;
614 } else {
616 * We are not going to do the fast caching, set cached to the
617 * appropriate value and wakeup any waiters.
619 spin_lock(&cache->lock);
620 if (load_cache_only) {
621 cache->caching_ctl = NULL;
622 cache->cached = BTRFS_CACHE_NO;
623 } else {
624 cache->cached = BTRFS_CACHE_STARTED;
626 spin_unlock(&cache->lock);
627 wake_up(&caching_ctl->wait);
630 if (load_cache_only) {
631 put_caching_control(caching_ctl);
632 return 0;
635 down_write(&fs_info->extent_commit_sem);
636 atomic_inc(&caching_ctl->count);
637 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
638 up_write(&fs_info->extent_commit_sem);
640 btrfs_get_block_group(cache);
642 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
644 return ret;
648 * return the block group that starts at or after bytenr
650 static struct btrfs_block_group_cache *
651 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
653 struct btrfs_block_group_cache *cache;
655 cache = block_group_cache_tree_search(info, bytenr, 0);
657 return cache;
661 * return the block group that contains the given bytenr
663 struct btrfs_block_group_cache *btrfs_lookup_block_group(
664 struct btrfs_fs_info *info,
665 u64 bytenr)
667 struct btrfs_block_group_cache *cache;
669 cache = block_group_cache_tree_search(info, bytenr, 1);
671 return cache;
674 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
675 u64 flags)
677 struct list_head *head = &info->space_info;
678 struct btrfs_space_info *found;
680 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
682 rcu_read_lock();
683 list_for_each_entry_rcu(found, head, list) {
684 if (found->flags & flags) {
685 rcu_read_unlock();
686 return found;
689 rcu_read_unlock();
690 return NULL;
694 * after adding space to the filesystem, we need to clear the full flags
695 * on all the space infos.
697 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
699 struct list_head *head = &info->space_info;
700 struct btrfs_space_info *found;
702 rcu_read_lock();
703 list_for_each_entry_rcu(found, head, list)
704 found->full = 0;
705 rcu_read_unlock();
708 /* simple helper to search for an existing extent at a given offset */
709 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
711 int ret;
712 struct btrfs_key key;
713 struct btrfs_path *path;
715 path = btrfs_alloc_path();
716 if (!path)
717 return -ENOMEM;
719 key.objectid = start;
720 key.offset = len;
721 key.type = BTRFS_EXTENT_ITEM_KEY;
722 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
723 0, 0);
724 if (ret > 0) {
725 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
726 if (key.objectid == start &&
727 key.type == BTRFS_METADATA_ITEM_KEY)
728 ret = 0;
730 btrfs_free_path(path);
731 return ret;
735 * helper function to lookup reference count and flags of a tree block.
737 * the head node for delayed ref is used to store the sum of all the
738 * reference count modifications queued up in the rbtree. the head
739 * node may also store the extent flags to set. This way you can check
740 * to see what the reference count and extent flags would be if all of
741 * the delayed refs are not processed.
743 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
744 struct btrfs_root *root, u64 bytenr,
745 u64 offset, int metadata, u64 *refs, u64 *flags)
747 struct btrfs_delayed_ref_head *head;
748 struct btrfs_delayed_ref_root *delayed_refs;
749 struct btrfs_path *path;
750 struct btrfs_extent_item *ei;
751 struct extent_buffer *leaf;
752 struct btrfs_key key;
753 u32 item_size;
754 u64 num_refs;
755 u64 extent_flags;
756 int ret;
759 * If we don't have skinny metadata, don't bother doing anything
760 * different
762 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
763 offset = root->leafsize;
764 metadata = 0;
767 path = btrfs_alloc_path();
768 if (!path)
769 return -ENOMEM;
771 if (metadata) {
772 key.objectid = bytenr;
773 key.type = BTRFS_METADATA_ITEM_KEY;
774 key.offset = offset;
775 } else {
776 key.objectid = bytenr;
777 key.type = BTRFS_EXTENT_ITEM_KEY;
778 key.offset = offset;
781 if (!trans) {
782 path->skip_locking = 1;
783 path->search_commit_root = 1;
785 again:
786 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
787 &key, path, 0, 0);
788 if (ret < 0)
789 goto out_free;
791 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
792 metadata = 0;
793 if (path->slots[0]) {
794 path->slots[0]--;
795 btrfs_item_key_to_cpu(path->nodes[0], &key,
796 path->slots[0]);
797 if (key.objectid == bytenr &&
798 key.type == BTRFS_EXTENT_ITEM_KEY &&
799 key.offset == root->leafsize)
800 ret = 0;
802 if (ret) {
803 key.objectid = bytenr;
804 key.type = BTRFS_EXTENT_ITEM_KEY;
805 key.offset = root->leafsize;
806 btrfs_release_path(path);
807 goto again;
811 if (ret == 0) {
812 leaf = path->nodes[0];
813 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
814 if (item_size >= sizeof(*ei)) {
815 ei = btrfs_item_ptr(leaf, path->slots[0],
816 struct btrfs_extent_item);
817 num_refs = btrfs_extent_refs(leaf, ei);
818 extent_flags = btrfs_extent_flags(leaf, ei);
819 } else {
820 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
821 struct btrfs_extent_item_v0 *ei0;
822 BUG_ON(item_size != sizeof(*ei0));
823 ei0 = btrfs_item_ptr(leaf, path->slots[0],
824 struct btrfs_extent_item_v0);
825 num_refs = btrfs_extent_refs_v0(leaf, ei0);
826 /* FIXME: this isn't correct for data */
827 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
828 #else
829 BUG();
830 #endif
832 BUG_ON(num_refs == 0);
833 } else {
834 num_refs = 0;
835 extent_flags = 0;
836 ret = 0;
839 if (!trans)
840 goto out;
842 delayed_refs = &trans->transaction->delayed_refs;
843 spin_lock(&delayed_refs->lock);
844 head = btrfs_find_delayed_ref_head(trans, bytenr);
845 if (head) {
846 if (!mutex_trylock(&head->mutex)) {
847 atomic_inc(&head->node.refs);
848 spin_unlock(&delayed_refs->lock);
850 btrfs_release_path(path);
853 * Mutex was contended, block until it's released and try
854 * again
856 mutex_lock(&head->mutex);
857 mutex_unlock(&head->mutex);
858 btrfs_put_delayed_ref(&head->node);
859 goto again;
861 if (head->extent_op && head->extent_op->update_flags)
862 extent_flags |= head->extent_op->flags_to_set;
863 else
864 BUG_ON(num_refs == 0);
866 num_refs += head->node.ref_mod;
867 mutex_unlock(&head->mutex);
869 spin_unlock(&delayed_refs->lock);
870 out:
871 WARN_ON(num_refs == 0);
872 if (refs)
873 *refs = num_refs;
874 if (flags)
875 *flags = extent_flags;
876 out_free:
877 btrfs_free_path(path);
878 return ret;
882 * Back reference rules. Back refs have three main goals:
884 * 1) differentiate between all holders of references to an extent so that
885 * when a reference is dropped we can make sure it was a valid reference
886 * before freeing the extent.
888 * 2) Provide enough information to quickly find the holders of an extent
889 * if we notice a given block is corrupted or bad.
891 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
892 * maintenance. This is actually the same as #2, but with a slightly
893 * different use case.
895 * There are two kinds of back refs. The implicit back refs is optimized
896 * for pointers in non-shared tree blocks. For a given pointer in a block,
897 * back refs of this kind provide information about the block's owner tree
898 * and the pointer's key. These information allow us to find the block by
899 * b-tree searching. The full back refs is for pointers in tree blocks not
900 * referenced by their owner trees. The location of tree block is recorded
901 * in the back refs. Actually the full back refs is generic, and can be
902 * used in all cases the implicit back refs is used. The major shortcoming
903 * of the full back refs is its overhead. Every time a tree block gets
904 * COWed, we have to update back refs entry for all pointers in it.
906 * For a newly allocated tree block, we use implicit back refs for
907 * pointers in it. This means most tree related operations only involve
908 * implicit back refs. For a tree block created in old transaction, the
909 * only way to drop a reference to it is COW it. So we can detect the
910 * event that tree block loses its owner tree's reference and do the
911 * back refs conversion.
913 * When a tree block is COW'd through a tree, there are four cases:
915 * The reference count of the block is one and the tree is the block's
916 * owner tree. Nothing to do in this case.
918 * The reference count of the block is one and the tree is not the
919 * block's owner tree. In this case, full back refs is used for pointers
920 * in the block. Remove these full back refs, add implicit back refs for
921 * every pointers in the new block.
923 * The reference count of the block is greater than one and the tree is
924 * the block's owner tree. In this case, implicit back refs is used for
925 * pointers in the block. Add full back refs for every pointers in the
926 * block, increase lower level extents' reference counts. The original
927 * implicit back refs are entailed to the new block.
929 * The reference count of the block is greater than one and the tree is
930 * not the block's owner tree. Add implicit back refs for every pointer in
931 * the new block, increase lower level extents' reference count.
933 * Back Reference Key composing:
935 * The key objectid corresponds to the first byte in the extent,
936 * The key type is used to differentiate between types of back refs.
937 * There are different meanings of the key offset for different types
938 * of back refs.
940 * File extents can be referenced by:
942 * - multiple snapshots, subvolumes, or different generations in one subvol
943 * - different files inside a single subvolume
944 * - different offsets inside a file (bookend extents in file.c)
946 * The extent ref structure for the implicit back refs has fields for:
948 * - Objectid of the subvolume root
949 * - objectid of the file holding the reference
950 * - original offset in the file
951 * - how many bookend extents
953 * The key offset for the implicit back refs is hash of the first
954 * three fields.
956 * The extent ref structure for the full back refs has field for:
958 * - number of pointers in the tree leaf
960 * The key offset for the implicit back refs is the first byte of
961 * the tree leaf
963 * When a file extent is allocated, The implicit back refs is used.
964 * the fields are filled in:
966 * (root_key.objectid, inode objectid, offset in file, 1)
968 * When a file extent is removed file truncation, we find the
969 * corresponding implicit back refs and check the following fields:
971 * (btrfs_header_owner(leaf), inode objectid, offset in file)
973 * Btree extents can be referenced by:
975 * - Different subvolumes
977 * Both the implicit back refs and the full back refs for tree blocks
978 * only consist of key. The key offset for the implicit back refs is
979 * objectid of block's owner tree. The key offset for the full back refs
980 * is the first byte of parent block.
982 * When implicit back refs is used, information about the lowest key and
983 * level of the tree block are required. These information are stored in
984 * tree block info structure.
987 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
988 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
989 struct btrfs_root *root,
990 struct btrfs_path *path,
991 u64 owner, u32 extra_size)
993 struct btrfs_extent_item *item;
994 struct btrfs_extent_item_v0 *ei0;
995 struct btrfs_extent_ref_v0 *ref0;
996 struct btrfs_tree_block_info *bi;
997 struct extent_buffer *leaf;
998 struct btrfs_key key;
999 struct btrfs_key found_key;
1000 u32 new_size = sizeof(*item);
1001 u64 refs;
1002 int ret;
1004 leaf = path->nodes[0];
1005 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1007 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1008 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1009 struct btrfs_extent_item_v0);
1010 refs = btrfs_extent_refs_v0(leaf, ei0);
1012 if (owner == (u64)-1) {
1013 while (1) {
1014 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1015 ret = btrfs_next_leaf(root, path);
1016 if (ret < 0)
1017 return ret;
1018 BUG_ON(ret > 0); /* Corruption */
1019 leaf = path->nodes[0];
1021 btrfs_item_key_to_cpu(leaf, &found_key,
1022 path->slots[0]);
1023 BUG_ON(key.objectid != found_key.objectid);
1024 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1025 path->slots[0]++;
1026 continue;
1028 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1029 struct btrfs_extent_ref_v0);
1030 owner = btrfs_ref_objectid_v0(leaf, ref0);
1031 break;
1034 btrfs_release_path(path);
1036 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1037 new_size += sizeof(*bi);
1039 new_size -= sizeof(*ei0);
1040 ret = btrfs_search_slot(trans, root, &key, path,
1041 new_size + extra_size, 1);
1042 if (ret < 0)
1043 return ret;
1044 BUG_ON(ret); /* Corruption */
1046 btrfs_extend_item(root, path, new_size);
1048 leaf = path->nodes[0];
1049 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1050 btrfs_set_extent_refs(leaf, item, refs);
1051 /* FIXME: get real generation */
1052 btrfs_set_extent_generation(leaf, item, 0);
1053 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1054 btrfs_set_extent_flags(leaf, item,
1055 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1056 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1057 bi = (struct btrfs_tree_block_info *)(item + 1);
1058 /* FIXME: get first key of the block */
1059 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1060 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1061 } else {
1062 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1064 btrfs_mark_buffer_dirty(leaf);
1065 return 0;
1067 #endif
1069 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1071 u32 high_crc = ~(u32)0;
1072 u32 low_crc = ~(u32)0;
1073 __le64 lenum;
1075 lenum = cpu_to_le64(root_objectid);
1076 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1077 lenum = cpu_to_le64(owner);
1078 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1079 lenum = cpu_to_le64(offset);
1080 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1082 return ((u64)high_crc << 31) ^ (u64)low_crc;
1085 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1086 struct btrfs_extent_data_ref *ref)
1088 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1089 btrfs_extent_data_ref_objectid(leaf, ref),
1090 btrfs_extent_data_ref_offset(leaf, ref));
1093 static int match_extent_data_ref(struct extent_buffer *leaf,
1094 struct btrfs_extent_data_ref *ref,
1095 u64 root_objectid, u64 owner, u64 offset)
1097 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1098 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1099 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1100 return 0;
1101 return 1;
1104 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1105 struct btrfs_root *root,
1106 struct btrfs_path *path,
1107 u64 bytenr, u64 parent,
1108 u64 root_objectid,
1109 u64 owner, u64 offset)
1111 struct btrfs_key key;
1112 struct btrfs_extent_data_ref *ref;
1113 struct extent_buffer *leaf;
1114 u32 nritems;
1115 int ret;
1116 int recow;
1117 int err = -ENOENT;
1119 key.objectid = bytenr;
1120 if (parent) {
1121 key.type = BTRFS_SHARED_DATA_REF_KEY;
1122 key.offset = parent;
1123 } else {
1124 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1125 key.offset = hash_extent_data_ref(root_objectid,
1126 owner, offset);
1128 again:
1129 recow = 0;
1130 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1131 if (ret < 0) {
1132 err = ret;
1133 goto fail;
1136 if (parent) {
1137 if (!ret)
1138 return 0;
1139 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1140 key.type = BTRFS_EXTENT_REF_V0_KEY;
1141 btrfs_release_path(path);
1142 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1143 if (ret < 0) {
1144 err = ret;
1145 goto fail;
1147 if (!ret)
1148 return 0;
1149 #endif
1150 goto fail;
1153 leaf = path->nodes[0];
1154 nritems = btrfs_header_nritems(leaf);
1155 while (1) {
1156 if (path->slots[0] >= nritems) {
1157 ret = btrfs_next_leaf(root, path);
1158 if (ret < 0)
1159 err = ret;
1160 if (ret)
1161 goto fail;
1163 leaf = path->nodes[0];
1164 nritems = btrfs_header_nritems(leaf);
1165 recow = 1;
1168 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1169 if (key.objectid != bytenr ||
1170 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1171 goto fail;
1173 ref = btrfs_item_ptr(leaf, path->slots[0],
1174 struct btrfs_extent_data_ref);
1176 if (match_extent_data_ref(leaf, ref, root_objectid,
1177 owner, offset)) {
1178 if (recow) {
1179 btrfs_release_path(path);
1180 goto again;
1182 err = 0;
1183 break;
1185 path->slots[0]++;
1187 fail:
1188 return err;
1191 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1192 struct btrfs_root *root,
1193 struct btrfs_path *path,
1194 u64 bytenr, u64 parent,
1195 u64 root_objectid, u64 owner,
1196 u64 offset, int refs_to_add)
1198 struct btrfs_key key;
1199 struct extent_buffer *leaf;
1200 u32 size;
1201 u32 num_refs;
1202 int ret;
1204 key.objectid = bytenr;
1205 if (parent) {
1206 key.type = BTRFS_SHARED_DATA_REF_KEY;
1207 key.offset = parent;
1208 size = sizeof(struct btrfs_shared_data_ref);
1209 } else {
1210 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1211 key.offset = hash_extent_data_ref(root_objectid,
1212 owner, offset);
1213 size = sizeof(struct btrfs_extent_data_ref);
1216 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1217 if (ret && ret != -EEXIST)
1218 goto fail;
1220 leaf = path->nodes[0];
1221 if (parent) {
1222 struct btrfs_shared_data_ref *ref;
1223 ref = btrfs_item_ptr(leaf, path->slots[0],
1224 struct btrfs_shared_data_ref);
1225 if (ret == 0) {
1226 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1227 } else {
1228 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1229 num_refs += refs_to_add;
1230 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1232 } else {
1233 struct btrfs_extent_data_ref *ref;
1234 while (ret == -EEXIST) {
1235 ref = btrfs_item_ptr(leaf, path->slots[0],
1236 struct btrfs_extent_data_ref);
1237 if (match_extent_data_ref(leaf, ref, root_objectid,
1238 owner, offset))
1239 break;
1240 btrfs_release_path(path);
1241 key.offset++;
1242 ret = btrfs_insert_empty_item(trans, root, path, &key,
1243 size);
1244 if (ret && ret != -EEXIST)
1245 goto fail;
1247 leaf = path->nodes[0];
1249 ref = btrfs_item_ptr(leaf, path->slots[0],
1250 struct btrfs_extent_data_ref);
1251 if (ret == 0) {
1252 btrfs_set_extent_data_ref_root(leaf, ref,
1253 root_objectid);
1254 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1255 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1256 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1257 } else {
1258 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1259 num_refs += refs_to_add;
1260 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1263 btrfs_mark_buffer_dirty(leaf);
1264 ret = 0;
1265 fail:
1266 btrfs_release_path(path);
1267 return ret;
1270 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1271 struct btrfs_root *root,
1272 struct btrfs_path *path,
1273 int refs_to_drop)
1275 struct btrfs_key key;
1276 struct btrfs_extent_data_ref *ref1 = NULL;
1277 struct btrfs_shared_data_ref *ref2 = NULL;
1278 struct extent_buffer *leaf;
1279 u32 num_refs = 0;
1280 int ret = 0;
1282 leaf = path->nodes[0];
1283 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1285 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1286 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1287 struct btrfs_extent_data_ref);
1288 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1289 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1290 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1291 struct btrfs_shared_data_ref);
1292 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1293 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1294 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1295 struct btrfs_extent_ref_v0 *ref0;
1296 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1297 struct btrfs_extent_ref_v0);
1298 num_refs = btrfs_ref_count_v0(leaf, ref0);
1299 #endif
1300 } else {
1301 BUG();
1304 BUG_ON(num_refs < refs_to_drop);
1305 num_refs -= refs_to_drop;
1307 if (num_refs == 0) {
1308 ret = btrfs_del_item(trans, root, path);
1309 } else {
1310 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1311 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1312 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1313 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1314 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1315 else {
1316 struct btrfs_extent_ref_v0 *ref0;
1317 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1318 struct btrfs_extent_ref_v0);
1319 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1321 #endif
1322 btrfs_mark_buffer_dirty(leaf);
1324 return ret;
1327 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1328 struct btrfs_path *path,
1329 struct btrfs_extent_inline_ref *iref)
1331 struct btrfs_key key;
1332 struct extent_buffer *leaf;
1333 struct btrfs_extent_data_ref *ref1;
1334 struct btrfs_shared_data_ref *ref2;
1335 u32 num_refs = 0;
1337 leaf = path->nodes[0];
1338 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1339 if (iref) {
1340 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1341 BTRFS_EXTENT_DATA_REF_KEY) {
1342 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1343 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1344 } else {
1345 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1346 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1348 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1349 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1350 struct btrfs_extent_data_ref);
1351 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1352 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1353 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1354 struct btrfs_shared_data_ref);
1355 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1356 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1357 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1358 struct btrfs_extent_ref_v0 *ref0;
1359 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1360 struct btrfs_extent_ref_v0);
1361 num_refs = btrfs_ref_count_v0(leaf, ref0);
1362 #endif
1363 } else {
1364 WARN_ON(1);
1366 return num_refs;
1369 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1370 struct btrfs_root *root,
1371 struct btrfs_path *path,
1372 u64 bytenr, u64 parent,
1373 u64 root_objectid)
1375 struct btrfs_key key;
1376 int ret;
1378 key.objectid = bytenr;
1379 if (parent) {
1380 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1381 key.offset = parent;
1382 } else {
1383 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1384 key.offset = root_objectid;
1387 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1388 if (ret > 0)
1389 ret = -ENOENT;
1390 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1391 if (ret == -ENOENT && parent) {
1392 btrfs_release_path(path);
1393 key.type = BTRFS_EXTENT_REF_V0_KEY;
1394 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1395 if (ret > 0)
1396 ret = -ENOENT;
1398 #endif
1399 return ret;
1402 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1403 struct btrfs_root *root,
1404 struct btrfs_path *path,
1405 u64 bytenr, u64 parent,
1406 u64 root_objectid)
1408 struct btrfs_key key;
1409 int ret;
1411 key.objectid = bytenr;
1412 if (parent) {
1413 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1414 key.offset = parent;
1415 } else {
1416 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1417 key.offset = root_objectid;
1420 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1421 btrfs_release_path(path);
1422 return ret;
1425 static inline int extent_ref_type(u64 parent, u64 owner)
1427 int type;
1428 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1429 if (parent > 0)
1430 type = BTRFS_SHARED_BLOCK_REF_KEY;
1431 else
1432 type = BTRFS_TREE_BLOCK_REF_KEY;
1433 } else {
1434 if (parent > 0)
1435 type = BTRFS_SHARED_DATA_REF_KEY;
1436 else
1437 type = BTRFS_EXTENT_DATA_REF_KEY;
1439 return type;
1442 static int find_next_key(struct btrfs_path *path, int level,
1443 struct btrfs_key *key)
1446 for (; level < BTRFS_MAX_LEVEL; level++) {
1447 if (!path->nodes[level])
1448 break;
1449 if (path->slots[level] + 1 >=
1450 btrfs_header_nritems(path->nodes[level]))
1451 continue;
1452 if (level == 0)
1453 btrfs_item_key_to_cpu(path->nodes[level], key,
1454 path->slots[level] + 1);
1455 else
1456 btrfs_node_key_to_cpu(path->nodes[level], key,
1457 path->slots[level] + 1);
1458 return 0;
1460 return 1;
1464 * look for inline back ref. if back ref is found, *ref_ret is set
1465 * to the address of inline back ref, and 0 is returned.
1467 * if back ref isn't found, *ref_ret is set to the address where it
1468 * should be inserted, and -ENOENT is returned.
1470 * if insert is true and there are too many inline back refs, the path
1471 * points to the extent item, and -EAGAIN is returned.
1473 * NOTE: inline back refs are ordered in the same way that back ref
1474 * items in the tree are ordered.
1476 static noinline_for_stack
1477 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1478 struct btrfs_root *root,
1479 struct btrfs_path *path,
1480 struct btrfs_extent_inline_ref **ref_ret,
1481 u64 bytenr, u64 num_bytes,
1482 u64 parent, u64 root_objectid,
1483 u64 owner, u64 offset, int insert)
1485 struct btrfs_key key;
1486 struct extent_buffer *leaf;
1487 struct btrfs_extent_item *ei;
1488 struct btrfs_extent_inline_ref *iref;
1489 u64 flags;
1490 u64 item_size;
1491 unsigned long ptr;
1492 unsigned long end;
1493 int extra_size;
1494 int type;
1495 int want;
1496 int ret;
1497 int err = 0;
1498 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1499 SKINNY_METADATA);
1501 key.objectid = bytenr;
1502 key.type = BTRFS_EXTENT_ITEM_KEY;
1503 key.offset = num_bytes;
1505 want = extent_ref_type(parent, owner);
1506 if (insert) {
1507 extra_size = btrfs_extent_inline_ref_size(want);
1508 path->keep_locks = 1;
1509 } else
1510 extra_size = -1;
1513 * Owner is our parent level, so we can just add one to get the level
1514 * for the block we are interested in.
1516 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1517 key.type = BTRFS_METADATA_ITEM_KEY;
1518 key.offset = owner;
1521 again:
1522 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1523 if (ret < 0) {
1524 err = ret;
1525 goto out;
1529 * We may be a newly converted file system which still has the old fat
1530 * extent entries for metadata, so try and see if we have one of those.
1532 if (ret > 0 && skinny_metadata) {
1533 skinny_metadata = false;
1534 if (path->slots[0]) {
1535 path->slots[0]--;
1536 btrfs_item_key_to_cpu(path->nodes[0], &key,
1537 path->slots[0]);
1538 if (key.objectid == bytenr &&
1539 key.type == BTRFS_EXTENT_ITEM_KEY &&
1540 key.offset == num_bytes)
1541 ret = 0;
1543 if (ret) {
1544 key.type = BTRFS_EXTENT_ITEM_KEY;
1545 key.offset = num_bytes;
1546 btrfs_release_path(path);
1547 goto again;
1551 if (ret && !insert) {
1552 err = -ENOENT;
1553 goto out;
1554 } else if (ret) {
1555 err = -EIO;
1556 WARN_ON(1);
1557 goto out;
1560 leaf = path->nodes[0];
1561 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1562 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1563 if (item_size < sizeof(*ei)) {
1564 if (!insert) {
1565 err = -ENOENT;
1566 goto out;
1568 ret = convert_extent_item_v0(trans, root, path, owner,
1569 extra_size);
1570 if (ret < 0) {
1571 err = ret;
1572 goto out;
1574 leaf = path->nodes[0];
1575 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1577 #endif
1578 BUG_ON(item_size < sizeof(*ei));
1580 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1581 flags = btrfs_extent_flags(leaf, ei);
1583 ptr = (unsigned long)(ei + 1);
1584 end = (unsigned long)ei + item_size;
1586 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1587 ptr += sizeof(struct btrfs_tree_block_info);
1588 BUG_ON(ptr > end);
1591 err = -ENOENT;
1592 while (1) {
1593 if (ptr >= end) {
1594 WARN_ON(ptr > end);
1595 break;
1597 iref = (struct btrfs_extent_inline_ref *)ptr;
1598 type = btrfs_extent_inline_ref_type(leaf, iref);
1599 if (want < type)
1600 break;
1601 if (want > type) {
1602 ptr += btrfs_extent_inline_ref_size(type);
1603 continue;
1606 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1607 struct btrfs_extent_data_ref *dref;
1608 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1609 if (match_extent_data_ref(leaf, dref, root_objectid,
1610 owner, offset)) {
1611 err = 0;
1612 break;
1614 if (hash_extent_data_ref_item(leaf, dref) <
1615 hash_extent_data_ref(root_objectid, owner, offset))
1616 break;
1617 } else {
1618 u64 ref_offset;
1619 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1620 if (parent > 0) {
1621 if (parent == ref_offset) {
1622 err = 0;
1623 break;
1625 if (ref_offset < parent)
1626 break;
1627 } else {
1628 if (root_objectid == ref_offset) {
1629 err = 0;
1630 break;
1632 if (ref_offset < root_objectid)
1633 break;
1636 ptr += btrfs_extent_inline_ref_size(type);
1638 if (err == -ENOENT && insert) {
1639 if (item_size + extra_size >=
1640 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1641 err = -EAGAIN;
1642 goto out;
1645 * To add new inline back ref, we have to make sure
1646 * there is no corresponding back ref item.
1647 * For simplicity, we just do not add new inline back
1648 * ref if there is any kind of item for this block
1650 if (find_next_key(path, 0, &key) == 0 &&
1651 key.objectid == bytenr &&
1652 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1653 err = -EAGAIN;
1654 goto out;
1657 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1658 out:
1659 if (insert) {
1660 path->keep_locks = 0;
1661 btrfs_unlock_up_safe(path, 1);
1663 return err;
1667 * helper to add new inline back ref
1669 static noinline_for_stack
1670 void setup_inline_extent_backref(struct btrfs_root *root,
1671 struct btrfs_path *path,
1672 struct btrfs_extent_inline_ref *iref,
1673 u64 parent, u64 root_objectid,
1674 u64 owner, u64 offset, int refs_to_add,
1675 struct btrfs_delayed_extent_op *extent_op)
1677 struct extent_buffer *leaf;
1678 struct btrfs_extent_item *ei;
1679 unsigned long ptr;
1680 unsigned long end;
1681 unsigned long item_offset;
1682 u64 refs;
1683 int size;
1684 int type;
1686 leaf = path->nodes[0];
1687 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1688 item_offset = (unsigned long)iref - (unsigned long)ei;
1690 type = extent_ref_type(parent, owner);
1691 size = btrfs_extent_inline_ref_size(type);
1693 btrfs_extend_item(root, path, size);
1695 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1696 refs = btrfs_extent_refs(leaf, ei);
1697 refs += refs_to_add;
1698 btrfs_set_extent_refs(leaf, ei, refs);
1699 if (extent_op)
1700 __run_delayed_extent_op(extent_op, leaf, ei);
1702 ptr = (unsigned long)ei + item_offset;
1703 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1704 if (ptr < end - size)
1705 memmove_extent_buffer(leaf, ptr + size, ptr,
1706 end - size - ptr);
1708 iref = (struct btrfs_extent_inline_ref *)ptr;
1709 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1710 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1711 struct btrfs_extent_data_ref *dref;
1712 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1713 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1714 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1715 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1716 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1717 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1718 struct btrfs_shared_data_ref *sref;
1719 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1720 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1721 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1722 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1723 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1724 } else {
1725 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1727 btrfs_mark_buffer_dirty(leaf);
1730 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1731 struct btrfs_root *root,
1732 struct btrfs_path *path,
1733 struct btrfs_extent_inline_ref **ref_ret,
1734 u64 bytenr, u64 num_bytes, u64 parent,
1735 u64 root_objectid, u64 owner, u64 offset)
1737 int ret;
1739 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1740 bytenr, num_bytes, parent,
1741 root_objectid, owner, offset, 0);
1742 if (ret != -ENOENT)
1743 return ret;
1745 btrfs_release_path(path);
1746 *ref_ret = NULL;
1748 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1749 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1750 root_objectid);
1751 } else {
1752 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1753 root_objectid, owner, offset);
1755 return ret;
1759 * helper to update/remove inline back ref
1761 static noinline_for_stack
1762 void update_inline_extent_backref(struct btrfs_root *root,
1763 struct btrfs_path *path,
1764 struct btrfs_extent_inline_ref *iref,
1765 int refs_to_mod,
1766 struct btrfs_delayed_extent_op *extent_op)
1768 struct extent_buffer *leaf;
1769 struct btrfs_extent_item *ei;
1770 struct btrfs_extent_data_ref *dref = NULL;
1771 struct btrfs_shared_data_ref *sref = NULL;
1772 unsigned long ptr;
1773 unsigned long end;
1774 u32 item_size;
1775 int size;
1776 int type;
1777 u64 refs;
1779 leaf = path->nodes[0];
1780 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1781 refs = btrfs_extent_refs(leaf, ei);
1782 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1783 refs += refs_to_mod;
1784 btrfs_set_extent_refs(leaf, ei, refs);
1785 if (extent_op)
1786 __run_delayed_extent_op(extent_op, leaf, ei);
1788 type = btrfs_extent_inline_ref_type(leaf, iref);
1790 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1791 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1792 refs = btrfs_extent_data_ref_count(leaf, dref);
1793 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1794 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1795 refs = btrfs_shared_data_ref_count(leaf, sref);
1796 } else {
1797 refs = 1;
1798 BUG_ON(refs_to_mod != -1);
1801 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1802 refs += refs_to_mod;
1804 if (refs > 0) {
1805 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1806 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1807 else
1808 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1809 } else {
1810 size = btrfs_extent_inline_ref_size(type);
1811 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1812 ptr = (unsigned long)iref;
1813 end = (unsigned long)ei + item_size;
1814 if (ptr + size < end)
1815 memmove_extent_buffer(leaf, ptr, ptr + size,
1816 end - ptr - size);
1817 item_size -= size;
1818 btrfs_truncate_item(root, path, item_size, 1);
1820 btrfs_mark_buffer_dirty(leaf);
1823 static noinline_for_stack
1824 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1825 struct btrfs_root *root,
1826 struct btrfs_path *path,
1827 u64 bytenr, u64 num_bytes, u64 parent,
1828 u64 root_objectid, u64 owner,
1829 u64 offset, int refs_to_add,
1830 struct btrfs_delayed_extent_op *extent_op)
1832 struct btrfs_extent_inline_ref *iref;
1833 int ret;
1835 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1836 bytenr, num_bytes, parent,
1837 root_objectid, owner, offset, 1);
1838 if (ret == 0) {
1839 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1840 update_inline_extent_backref(root, path, iref,
1841 refs_to_add, extent_op);
1842 } else if (ret == -ENOENT) {
1843 setup_inline_extent_backref(root, path, iref, parent,
1844 root_objectid, owner, offset,
1845 refs_to_add, extent_op);
1846 ret = 0;
1848 return ret;
1851 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1852 struct btrfs_root *root,
1853 struct btrfs_path *path,
1854 u64 bytenr, u64 parent, u64 root_objectid,
1855 u64 owner, u64 offset, int refs_to_add)
1857 int ret;
1858 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1859 BUG_ON(refs_to_add != 1);
1860 ret = insert_tree_block_ref(trans, root, path, bytenr,
1861 parent, root_objectid);
1862 } else {
1863 ret = insert_extent_data_ref(trans, root, path, bytenr,
1864 parent, root_objectid,
1865 owner, offset, refs_to_add);
1867 return ret;
1870 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1871 struct btrfs_root *root,
1872 struct btrfs_path *path,
1873 struct btrfs_extent_inline_ref *iref,
1874 int refs_to_drop, int is_data)
1876 int ret = 0;
1878 BUG_ON(!is_data && refs_to_drop != 1);
1879 if (iref) {
1880 update_inline_extent_backref(root, path, iref,
1881 -refs_to_drop, NULL);
1882 } else if (is_data) {
1883 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1884 } else {
1885 ret = btrfs_del_item(trans, root, path);
1887 return ret;
1890 static int btrfs_issue_discard(struct block_device *bdev,
1891 u64 start, u64 len)
1893 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1896 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1897 u64 num_bytes, u64 *actual_bytes)
1899 int ret;
1900 u64 discarded_bytes = 0;
1901 struct btrfs_bio *bbio = NULL;
1904 /* Tell the block device(s) that the sectors can be discarded */
1905 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1906 bytenr, &num_bytes, &bbio, 0);
1907 /* Error condition is -ENOMEM */
1908 if (!ret) {
1909 struct btrfs_bio_stripe *stripe = bbio->stripes;
1910 int i;
1913 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1914 if (!stripe->dev->can_discard)
1915 continue;
1917 ret = btrfs_issue_discard(stripe->dev->bdev,
1918 stripe->physical,
1919 stripe->length);
1920 if (!ret)
1921 discarded_bytes += stripe->length;
1922 else if (ret != -EOPNOTSUPP)
1923 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1926 * Just in case we get back EOPNOTSUPP for some reason,
1927 * just ignore the return value so we don't screw up
1928 * people calling discard_extent.
1930 ret = 0;
1932 kfree(bbio);
1935 if (actual_bytes)
1936 *actual_bytes = discarded_bytes;
1939 if (ret == -EOPNOTSUPP)
1940 ret = 0;
1941 return ret;
1944 /* Can return -ENOMEM */
1945 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1946 struct btrfs_root *root,
1947 u64 bytenr, u64 num_bytes, u64 parent,
1948 u64 root_objectid, u64 owner, u64 offset, int for_cow)
1950 int ret;
1951 struct btrfs_fs_info *fs_info = root->fs_info;
1953 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1954 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1956 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1957 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1958 num_bytes,
1959 parent, root_objectid, (int)owner,
1960 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1961 } else {
1962 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1963 num_bytes,
1964 parent, root_objectid, owner, offset,
1965 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1967 return ret;
1970 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1971 struct btrfs_root *root,
1972 u64 bytenr, u64 num_bytes,
1973 u64 parent, u64 root_objectid,
1974 u64 owner, u64 offset, int refs_to_add,
1975 struct btrfs_delayed_extent_op *extent_op)
1977 struct btrfs_path *path;
1978 struct extent_buffer *leaf;
1979 struct btrfs_extent_item *item;
1980 u64 refs;
1981 int ret;
1982 int err = 0;
1984 path = btrfs_alloc_path();
1985 if (!path)
1986 return -ENOMEM;
1988 path->reada = 1;
1989 path->leave_spinning = 1;
1990 /* this will setup the path even if it fails to insert the back ref */
1991 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1992 path, bytenr, num_bytes, parent,
1993 root_objectid, owner, offset,
1994 refs_to_add, extent_op);
1995 if (ret == 0)
1996 goto out;
1998 if (ret != -EAGAIN) {
1999 err = ret;
2000 goto out;
2003 leaf = path->nodes[0];
2004 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2005 refs = btrfs_extent_refs(leaf, item);
2006 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2007 if (extent_op)
2008 __run_delayed_extent_op(extent_op, leaf, item);
2010 btrfs_mark_buffer_dirty(leaf);
2011 btrfs_release_path(path);
2013 path->reada = 1;
2014 path->leave_spinning = 1;
2016 /* now insert the actual backref */
2017 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2018 path, bytenr, parent, root_objectid,
2019 owner, offset, refs_to_add);
2020 if (ret)
2021 btrfs_abort_transaction(trans, root, ret);
2022 out:
2023 btrfs_free_path(path);
2024 return err;
2027 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2028 struct btrfs_root *root,
2029 struct btrfs_delayed_ref_node *node,
2030 struct btrfs_delayed_extent_op *extent_op,
2031 int insert_reserved)
2033 int ret = 0;
2034 struct btrfs_delayed_data_ref *ref;
2035 struct btrfs_key ins;
2036 u64 parent = 0;
2037 u64 ref_root = 0;
2038 u64 flags = 0;
2040 ins.objectid = node->bytenr;
2041 ins.offset = node->num_bytes;
2042 ins.type = BTRFS_EXTENT_ITEM_KEY;
2044 ref = btrfs_delayed_node_to_data_ref(node);
2045 trace_run_delayed_data_ref(node, ref, node->action);
2047 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2048 parent = ref->parent;
2049 else
2050 ref_root = ref->root;
2052 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2053 if (extent_op)
2054 flags |= extent_op->flags_to_set;
2055 ret = alloc_reserved_file_extent(trans, root,
2056 parent, ref_root, flags,
2057 ref->objectid, ref->offset,
2058 &ins, node->ref_mod);
2059 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2060 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2061 node->num_bytes, parent,
2062 ref_root, ref->objectid,
2063 ref->offset, node->ref_mod,
2064 extent_op);
2065 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2066 ret = __btrfs_free_extent(trans, root, node->bytenr,
2067 node->num_bytes, parent,
2068 ref_root, ref->objectid,
2069 ref->offset, node->ref_mod,
2070 extent_op);
2071 } else {
2072 BUG();
2074 return ret;
2077 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2078 struct extent_buffer *leaf,
2079 struct btrfs_extent_item *ei)
2081 u64 flags = btrfs_extent_flags(leaf, ei);
2082 if (extent_op->update_flags) {
2083 flags |= extent_op->flags_to_set;
2084 btrfs_set_extent_flags(leaf, ei, flags);
2087 if (extent_op->update_key) {
2088 struct btrfs_tree_block_info *bi;
2089 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2090 bi = (struct btrfs_tree_block_info *)(ei + 1);
2091 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2095 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2096 struct btrfs_root *root,
2097 struct btrfs_delayed_ref_node *node,
2098 struct btrfs_delayed_extent_op *extent_op)
2100 struct btrfs_key key;
2101 struct btrfs_path *path;
2102 struct btrfs_extent_item *ei;
2103 struct extent_buffer *leaf;
2104 u32 item_size;
2105 int ret;
2106 int err = 0;
2107 int metadata = !extent_op->is_data;
2109 if (trans->aborted)
2110 return 0;
2112 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2113 metadata = 0;
2115 path = btrfs_alloc_path();
2116 if (!path)
2117 return -ENOMEM;
2119 key.objectid = node->bytenr;
2121 if (metadata) {
2122 key.type = BTRFS_METADATA_ITEM_KEY;
2123 key.offset = extent_op->level;
2124 } else {
2125 key.type = BTRFS_EXTENT_ITEM_KEY;
2126 key.offset = node->num_bytes;
2129 again:
2130 path->reada = 1;
2131 path->leave_spinning = 1;
2132 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2133 path, 0, 1);
2134 if (ret < 0) {
2135 err = ret;
2136 goto out;
2138 if (ret > 0) {
2139 if (metadata) {
2140 btrfs_release_path(path);
2141 metadata = 0;
2143 key.offset = node->num_bytes;
2144 key.type = BTRFS_EXTENT_ITEM_KEY;
2145 goto again;
2147 err = -EIO;
2148 goto out;
2151 leaf = path->nodes[0];
2152 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2153 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2154 if (item_size < sizeof(*ei)) {
2155 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2156 path, (u64)-1, 0);
2157 if (ret < 0) {
2158 err = ret;
2159 goto out;
2161 leaf = path->nodes[0];
2162 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2164 #endif
2165 BUG_ON(item_size < sizeof(*ei));
2166 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2167 __run_delayed_extent_op(extent_op, leaf, ei);
2169 btrfs_mark_buffer_dirty(leaf);
2170 out:
2171 btrfs_free_path(path);
2172 return err;
2175 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2176 struct btrfs_root *root,
2177 struct btrfs_delayed_ref_node *node,
2178 struct btrfs_delayed_extent_op *extent_op,
2179 int insert_reserved)
2181 int ret = 0;
2182 struct btrfs_delayed_tree_ref *ref;
2183 struct btrfs_key ins;
2184 u64 parent = 0;
2185 u64 ref_root = 0;
2186 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2187 SKINNY_METADATA);
2189 ref = btrfs_delayed_node_to_tree_ref(node);
2190 trace_run_delayed_tree_ref(node, ref, node->action);
2192 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2193 parent = ref->parent;
2194 else
2195 ref_root = ref->root;
2197 ins.objectid = node->bytenr;
2198 if (skinny_metadata) {
2199 ins.offset = ref->level;
2200 ins.type = BTRFS_METADATA_ITEM_KEY;
2201 } else {
2202 ins.offset = node->num_bytes;
2203 ins.type = BTRFS_EXTENT_ITEM_KEY;
2206 BUG_ON(node->ref_mod != 1);
2207 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2208 BUG_ON(!extent_op || !extent_op->update_flags);
2209 ret = alloc_reserved_tree_block(trans, root,
2210 parent, ref_root,
2211 extent_op->flags_to_set,
2212 &extent_op->key,
2213 ref->level, &ins);
2214 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2215 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2216 node->num_bytes, parent, ref_root,
2217 ref->level, 0, 1, extent_op);
2218 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2219 ret = __btrfs_free_extent(trans, root, node->bytenr,
2220 node->num_bytes, parent, ref_root,
2221 ref->level, 0, 1, extent_op);
2222 } else {
2223 BUG();
2225 return ret;
2228 /* helper function to actually process a single delayed ref entry */
2229 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2230 struct btrfs_root *root,
2231 struct btrfs_delayed_ref_node *node,
2232 struct btrfs_delayed_extent_op *extent_op,
2233 int insert_reserved)
2235 int ret = 0;
2237 if (trans->aborted)
2238 return 0;
2240 if (btrfs_delayed_ref_is_head(node)) {
2241 struct btrfs_delayed_ref_head *head;
2243 * we've hit the end of the chain and we were supposed
2244 * to insert this extent into the tree. But, it got
2245 * deleted before we ever needed to insert it, so all
2246 * we have to do is clean up the accounting
2248 BUG_ON(extent_op);
2249 head = btrfs_delayed_node_to_head(node);
2250 trace_run_delayed_ref_head(node, head, node->action);
2252 if (insert_reserved) {
2253 btrfs_pin_extent(root, node->bytenr,
2254 node->num_bytes, 1);
2255 if (head->is_data) {
2256 ret = btrfs_del_csums(trans, root,
2257 node->bytenr,
2258 node->num_bytes);
2261 return ret;
2264 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2265 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2266 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2267 insert_reserved);
2268 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2269 node->type == BTRFS_SHARED_DATA_REF_KEY)
2270 ret = run_delayed_data_ref(trans, root, node, extent_op,
2271 insert_reserved);
2272 else
2273 BUG();
2274 return ret;
2277 static noinline struct btrfs_delayed_ref_node *
2278 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2280 struct rb_node *node;
2281 struct btrfs_delayed_ref_node *ref;
2282 int action = BTRFS_ADD_DELAYED_REF;
2283 again:
2285 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2286 * this prevents ref count from going down to zero when
2287 * there still are pending delayed ref.
2289 node = rb_prev(&head->node.rb_node);
2290 while (1) {
2291 if (!node)
2292 break;
2293 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2294 rb_node);
2295 if (ref->bytenr != head->node.bytenr)
2296 break;
2297 if (ref->action == action)
2298 return ref;
2299 node = rb_prev(node);
2301 if (action == BTRFS_ADD_DELAYED_REF) {
2302 action = BTRFS_DROP_DELAYED_REF;
2303 goto again;
2305 return NULL;
2309 * Returns 0 on success or if called with an already aborted transaction.
2310 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2312 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2313 struct btrfs_root *root,
2314 struct list_head *cluster)
2316 struct btrfs_delayed_ref_root *delayed_refs;
2317 struct btrfs_delayed_ref_node *ref;
2318 struct btrfs_delayed_ref_head *locked_ref = NULL;
2319 struct btrfs_delayed_extent_op *extent_op;
2320 struct btrfs_fs_info *fs_info = root->fs_info;
2321 int ret;
2322 int count = 0;
2323 int must_insert_reserved = 0;
2325 delayed_refs = &trans->transaction->delayed_refs;
2326 while (1) {
2327 if (!locked_ref) {
2328 /* pick a new head ref from the cluster list */
2329 if (list_empty(cluster))
2330 break;
2332 locked_ref = list_entry(cluster->next,
2333 struct btrfs_delayed_ref_head, cluster);
2335 /* grab the lock that says we are going to process
2336 * all the refs for this head */
2337 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2340 * we may have dropped the spin lock to get the head
2341 * mutex lock, and that might have given someone else
2342 * time to free the head. If that's true, it has been
2343 * removed from our list and we can move on.
2345 if (ret == -EAGAIN) {
2346 locked_ref = NULL;
2347 count++;
2348 continue;
2353 * We need to try and merge add/drops of the same ref since we
2354 * can run into issues with relocate dropping the implicit ref
2355 * and then it being added back again before the drop can
2356 * finish. If we merged anything we need to re-loop so we can
2357 * get a good ref.
2359 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2360 locked_ref);
2363 * locked_ref is the head node, so we have to go one
2364 * node back for any delayed ref updates
2366 ref = select_delayed_ref(locked_ref);
2368 if (ref && ref->seq &&
2369 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2371 * there are still refs with lower seq numbers in the
2372 * process of being added. Don't run this ref yet.
2374 list_del_init(&locked_ref->cluster);
2375 btrfs_delayed_ref_unlock(locked_ref);
2376 locked_ref = NULL;
2377 delayed_refs->num_heads_ready++;
2378 spin_unlock(&delayed_refs->lock);
2379 cond_resched();
2380 spin_lock(&delayed_refs->lock);
2381 continue;
2385 * record the must insert reserved flag before we
2386 * drop the spin lock.
2388 must_insert_reserved = locked_ref->must_insert_reserved;
2389 locked_ref->must_insert_reserved = 0;
2391 extent_op = locked_ref->extent_op;
2392 locked_ref->extent_op = NULL;
2394 if (!ref) {
2395 /* All delayed refs have been processed, Go ahead
2396 * and send the head node to run_one_delayed_ref,
2397 * so that any accounting fixes can happen
2399 ref = &locked_ref->node;
2401 if (extent_op && must_insert_reserved) {
2402 btrfs_free_delayed_extent_op(extent_op);
2403 extent_op = NULL;
2406 if (extent_op) {
2407 spin_unlock(&delayed_refs->lock);
2409 ret = run_delayed_extent_op(trans, root,
2410 ref, extent_op);
2411 btrfs_free_delayed_extent_op(extent_op);
2413 if (ret) {
2414 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2415 spin_lock(&delayed_refs->lock);
2416 btrfs_delayed_ref_unlock(locked_ref);
2417 return ret;
2420 goto next;
2424 ref->in_tree = 0;
2425 rb_erase(&ref->rb_node, &delayed_refs->root);
2426 delayed_refs->num_entries--;
2427 if (!btrfs_delayed_ref_is_head(ref)) {
2429 * when we play the delayed ref, also correct the
2430 * ref_mod on head
2432 switch (ref->action) {
2433 case BTRFS_ADD_DELAYED_REF:
2434 case BTRFS_ADD_DELAYED_EXTENT:
2435 locked_ref->node.ref_mod -= ref->ref_mod;
2436 break;
2437 case BTRFS_DROP_DELAYED_REF:
2438 locked_ref->node.ref_mod += ref->ref_mod;
2439 break;
2440 default:
2441 WARN_ON(1);
2443 } else {
2444 list_del_init(&locked_ref->cluster);
2446 spin_unlock(&delayed_refs->lock);
2448 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2449 must_insert_reserved);
2451 btrfs_free_delayed_extent_op(extent_op);
2452 if (ret) {
2453 btrfs_delayed_ref_unlock(locked_ref);
2454 btrfs_put_delayed_ref(ref);
2455 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2456 spin_lock(&delayed_refs->lock);
2457 return ret;
2461 * If this node is a head, that means all the refs in this head
2462 * have been dealt with, and we will pick the next head to deal
2463 * with, so we must unlock the head and drop it from the cluster
2464 * list before we release it.
2466 if (btrfs_delayed_ref_is_head(ref)) {
2467 btrfs_delayed_ref_unlock(locked_ref);
2468 locked_ref = NULL;
2470 btrfs_put_delayed_ref(ref);
2471 count++;
2472 next:
2473 cond_resched();
2474 spin_lock(&delayed_refs->lock);
2476 return count;
2479 #ifdef SCRAMBLE_DELAYED_REFS
2481 * Normally delayed refs get processed in ascending bytenr order. This
2482 * correlates in most cases to the order added. To expose dependencies on this
2483 * order, we start to process the tree in the middle instead of the beginning
2485 static u64 find_middle(struct rb_root *root)
2487 struct rb_node *n = root->rb_node;
2488 struct btrfs_delayed_ref_node *entry;
2489 int alt = 1;
2490 u64 middle;
2491 u64 first = 0, last = 0;
2493 n = rb_first(root);
2494 if (n) {
2495 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2496 first = entry->bytenr;
2498 n = rb_last(root);
2499 if (n) {
2500 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2501 last = entry->bytenr;
2503 n = root->rb_node;
2505 while (n) {
2506 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2507 WARN_ON(!entry->in_tree);
2509 middle = entry->bytenr;
2511 if (alt)
2512 n = n->rb_left;
2513 else
2514 n = n->rb_right;
2516 alt = 1 - alt;
2518 return middle;
2520 #endif
2522 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2523 struct btrfs_fs_info *fs_info)
2525 struct qgroup_update *qgroup_update;
2526 int ret = 0;
2528 if (list_empty(&trans->qgroup_ref_list) !=
2529 !trans->delayed_ref_elem.seq) {
2530 /* list without seq or seq without list */
2531 btrfs_err(fs_info,
2532 "qgroup accounting update error, list is%s empty, seq is %#x.%x",
2533 list_empty(&trans->qgroup_ref_list) ? "" : " not",
2534 (u32)(trans->delayed_ref_elem.seq >> 32),
2535 (u32)trans->delayed_ref_elem.seq);
2536 BUG();
2539 if (!trans->delayed_ref_elem.seq)
2540 return 0;
2542 while (!list_empty(&trans->qgroup_ref_list)) {
2543 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2544 struct qgroup_update, list);
2545 list_del(&qgroup_update->list);
2546 if (!ret)
2547 ret = btrfs_qgroup_account_ref(
2548 trans, fs_info, qgroup_update->node,
2549 qgroup_update->extent_op);
2550 kfree(qgroup_update);
2553 btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2555 return ret;
2558 static int refs_newer(struct btrfs_delayed_ref_root *delayed_refs, int seq,
2559 int count)
2561 int val = atomic_read(&delayed_refs->ref_seq);
2563 if (val < seq || val >= seq + count)
2564 return 1;
2565 return 0;
2568 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2570 u64 num_bytes;
2572 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2573 sizeof(struct btrfs_extent_inline_ref));
2574 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2575 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2578 * We don't ever fill up leaves all the way so multiply by 2 just to be
2579 * closer to what we're really going to want to ouse.
2581 return div64_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2584 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2585 struct btrfs_root *root)
2587 struct btrfs_block_rsv *global_rsv;
2588 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2589 u64 num_bytes;
2590 int ret = 0;
2592 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2593 num_heads = heads_to_leaves(root, num_heads);
2594 if (num_heads > 1)
2595 num_bytes += (num_heads - 1) * root->leafsize;
2596 num_bytes <<= 1;
2597 global_rsv = &root->fs_info->global_block_rsv;
2600 * If we can't allocate any more chunks lets make sure we have _lots_ of
2601 * wiggle room since running delayed refs can create more delayed refs.
2603 if (global_rsv->space_info->full)
2604 num_bytes <<= 1;
2606 spin_lock(&global_rsv->lock);
2607 if (global_rsv->reserved <= num_bytes)
2608 ret = 1;
2609 spin_unlock(&global_rsv->lock);
2610 return ret;
2614 * this starts processing the delayed reference count updates and
2615 * extent insertions we have queued up so far. count can be
2616 * 0, which means to process everything in the tree at the start
2617 * of the run (but not newly added entries), or it can be some target
2618 * number you'd like to process.
2620 * Returns 0 on success or if called with an aborted transaction
2621 * Returns <0 on error and aborts the transaction
2623 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2624 struct btrfs_root *root, unsigned long count)
2626 struct rb_node *node;
2627 struct btrfs_delayed_ref_root *delayed_refs;
2628 struct btrfs_delayed_ref_node *ref;
2629 struct list_head cluster;
2630 int ret;
2631 u64 delayed_start;
2632 int run_all = count == (unsigned long)-1;
2633 int run_most = 0;
2634 int loops;
2636 /* We'll clean this up in btrfs_cleanup_transaction */
2637 if (trans->aborted)
2638 return 0;
2640 if (root == root->fs_info->extent_root)
2641 root = root->fs_info->tree_root;
2643 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2645 delayed_refs = &trans->transaction->delayed_refs;
2646 INIT_LIST_HEAD(&cluster);
2647 if (count == 0) {
2648 count = delayed_refs->num_entries * 2;
2649 run_most = 1;
2652 if (!run_all && !run_most) {
2653 int old;
2654 int seq = atomic_read(&delayed_refs->ref_seq);
2656 progress:
2657 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2658 if (old) {
2659 DEFINE_WAIT(__wait);
2660 if (delayed_refs->flushing ||
2661 !btrfs_should_throttle_delayed_refs(trans, root))
2662 return 0;
2664 prepare_to_wait(&delayed_refs->wait, &__wait,
2665 TASK_UNINTERRUPTIBLE);
2667 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2668 if (old) {
2669 schedule();
2670 finish_wait(&delayed_refs->wait, &__wait);
2672 if (!refs_newer(delayed_refs, seq, 256))
2673 goto progress;
2674 else
2675 return 0;
2676 } else {
2677 finish_wait(&delayed_refs->wait, &__wait);
2678 goto again;
2682 } else {
2683 atomic_inc(&delayed_refs->procs_running_refs);
2686 again:
2687 loops = 0;
2688 spin_lock(&delayed_refs->lock);
2690 #ifdef SCRAMBLE_DELAYED_REFS
2691 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2692 #endif
2694 while (1) {
2695 if (!(run_all || run_most) &&
2696 !btrfs_should_throttle_delayed_refs(trans, root))
2697 break;
2700 * go find something we can process in the rbtree. We start at
2701 * the beginning of the tree, and then build a cluster
2702 * of refs to process starting at the first one we are able to
2703 * lock
2705 delayed_start = delayed_refs->run_delayed_start;
2706 ret = btrfs_find_ref_cluster(trans, &cluster,
2707 delayed_refs->run_delayed_start);
2708 if (ret)
2709 break;
2711 ret = run_clustered_refs(trans, root, &cluster);
2712 if (ret < 0) {
2713 btrfs_release_ref_cluster(&cluster);
2714 spin_unlock(&delayed_refs->lock);
2715 btrfs_abort_transaction(trans, root, ret);
2716 atomic_dec(&delayed_refs->procs_running_refs);
2717 wake_up(&delayed_refs->wait);
2718 return ret;
2721 atomic_add(ret, &delayed_refs->ref_seq);
2723 count -= min_t(unsigned long, ret, count);
2725 if (count == 0)
2726 break;
2728 if (delayed_start >= delayed_refs->run_delayed_start) {
2729 if (loops == 0) {
2731 * btrfs_find_ref_cluster looped. let's do one
2732 * more cycle. if we don't run any delayed ref
2733 * during that cycle (because we can't because
2734 * all of them are blocked), bail out.
2736 loops = 1;
2737 } else {
2739 * no runnable refs left, stop trying
2741 BUG_ON(run_all);
2742 break;
2745 if (ret) {
2746 /* refs were run, let's reset staleness detection */
2747 loops = 0;
2751 if (run_all) {
2752 if (!list_empty(&trans->new_bgs)) {
2753 spin_unlock(&delayed_refs->lock);
2754 btrfs_create_pending_block_groups(trans, root);
2755 spin_lock(&delayed_refs->lock);
2758 node = rb_first(&delayed_refs->root);
2759 if (!node)
2760 goto out;
2761 count = (unsigned long)-1;
2763 while (node) {
2764 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2765 rb_node);
2766 if (btrfs_delayed_ref_is_head(ref)) {
2767 struct btrfs_delayed_ref_head *head;
2769 head = btrfs_delayed_node_to_head(ref);
2770 atomic_inc(&ref->refs);
2772 spin_unlock(&delayed_refs->lock);
2774 * Mutex was contended, block until it's
2775 * released and try again
2777 mutex_lock(&head->mutex);
2778 mutex_unlock(&head->mutex);
2780 btrfs_put_delayed_ref(ref);
2781 cond_resched();
2782 goto again;
2784 node = rb_next(node);
2786 spin_unlock(&delayed_refs->lock);
2787 schedule_timeout(1);
2788 goto again;
2790 out:
2791 atomic_dec(&delayed_refs->procs_running_refs);
2792 smp_mb();
2793 if (waitqueue_active(&delayed_refs->wait))
2794 wake_up(&delayed_refs->wait);
2796 spin_unlock(&delayed_refs->lock);
2797 assert_qgroups_uptodate(trans);
2798 return 0;
2801 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2802 struct btrfs_root *root,
2803 u64 bytenr, u64 num_bytes, u64 flags,
2804 int level, int is_data)
2806 struct btrfs_delayed_extent_op *extent_op;
2807 int ret;
2809 extent_op = btrfs_alloc_delayed_extent_op();
2810 if (!extent_op)
2811 return -ENOMEM;
2813 extent_op->flags_to_set = flags;
2814 extent_op->update_flags = 1;
2815 extent_op->update_key = 0;
2816 extent_op->is_data = is_data ? 1 : 0;
2817 extent_op->level = level;
2819 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2820 num_bytes, extent_op);
2821 if (ret)
2822 btrfs_free_delayed_extent_op(extent_op);
2823 return ret;
2826 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2827 struct btrfs_root *root,
2828 struct btrfs_path *path,
2829 u64 objectid, u64 offset, u64 bytenr)
2831 struct btrfs_delayed_ref_head *head;
2832 struct btrfs_delayed_ref_node *ref;
2833 struct btrfs_delayed_data_ref *data_ref;
2834 struct btrfs_delayed_ref_root *delayed_refs;
2835 struct rb_node *node;
2836 int ret = 0;
2838 ret = -ENOENT;
2839 delayed_refs = &trans->transaction->delayed_refs;
2840 spin_lock(&delayed_refs->lock);
2841 head = btrfs_find_delayed_ref_head(trans, bytenr);
2842 if (!head)
2843 goto out;
2845 if (!mutex_trylock(&head->mutex)) {
2846 atomic_inc(&head->node.refs);
2847 spin_unlock(&delayed_refs->lock);
2849 btrfs_release_path(path);
2852 * Mutex was contended, block until it's released and let
2853 * caller try again
2855 mutex_lock(&head->mutex);
2856 mutex_unlock(&head->mutex);
2857 btrfs_put_delayed_ref(&head->node);
2858 return -EAGAIN;
2861 node = rb_prev(&head->node.rb_node);
2862 if (!node)
2863 goto out_unlock;
2865 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2867 if (ref->bytenr != bytenr)
2868 goto out_unlock;
2870 ret = 1;
2871 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2872 goto out_unlock;
2874 data_ref = btrfs_delayed_node_to_data_ref(ref);
2876 node = rb_prev(node);
2877 if (node) {
2878 int seq = ref->seq;
2880 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2881 if (ref->bytenr == bytenr && ref->seq == seq)
2882 goto out_unlock;
2885 if (data_ref->root != root->root_key.objectid ||
2886 data_ref->objectid != objectid || data_ref->offset != offset)
2887 goto out_unlock;
2889 ret = 0;
2890 out_unlock:
2891 mutex_unlock(&head->mutex);
2892 out:
2893 spin_unlock(&delayed_refs->lock);
2894 return ret;
2897 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2898 struct btrfs_root *root,
2899 struct btrfs_path *path,
2900 u64 objectid, u64 offset, u64 bytenr)
2902 struct btrfs_root *extent_root = root->fs_info->extent_root;
2903 struct extent_buffer *leaf;
2904 struct btrfs_extent_data_ref *ref;
2905 struct btrfs_extent_inline_ref *iref;
2906 struct btrfs_extent_item *ei;
2907 struct btrfs_key key;
2908 u32 item_size;
2909 int ret;
2911 key.objectid = bytenr;
2912 key.offset = (u64)-1;
2913 key.type = BTRFS_EXTENT_ITEM_KEY;
2915 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2916 if (ret < 0)
2917 goto out;
2918 BUG_ON(ret == 0); /* Corruption */
2920 ret = -ENOENT;
2921 if (path->slots[0] == 0)
2922 goto out;
2924 path->slots[0]--;
2925 leaf = path->nodes[0];
2926 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2928 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2929 goto out;
2931 ret = 1;
2932 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2933 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2934 if (item_size < sizeof(*ei)) {
2935 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2936 goto out;
2938 #endif
2939 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2941 if (item_size != sizeof(*ei) +
2942 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2943 goto out;
2945 if (btrfs_extent_generation(leaf, ei) <=
2946 btrfs_root_last_snapshot(&root->root_item))
2947 goto out;
2949 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2950 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2951 BTRFS_EXTENT_DATA_REF_KEY)
2952 goto out;
2954 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2955 if (btrfs_extent_refs(leaf, ei) !=
2956 btrfs_extent_data_ref_count(leaf, ref) ||
2957 btrfs_extent_data_ref_root(leaf, ref) !=
2958 root->root_key.objectid ||
2959 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2960 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2961 goto out;
2963 ret = 0;
2964 out:
2965 return ret;
2968 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2969 struct btrfs_root *root,
2970 u64 objectid, u64 offset, u64 bytenr)
2972 struct btrfs_path *path;
2973 int ret;
2974 int ret2;
2976 path = btrfs_alloc_path();
2977 if (!path)
2978 return -ENOENT;
2980 do {
2981 ret = check_committed_ref(trans, root, path, objectid,
2982 offset, bytenr);
2983 if (ret && ret != -ENOENT)
2984 goto out;
2986 ret2 = check_delayed_ref(trans, root, path, objectid,
2987 offset, bytenr);
2988 } while (ret2 == -EAGAIN);
2990 if (ret2 && ret2 != -ENOENT) {
2991 ret = ret2;
2992 goto out;
2995 if (ret != -ENOENT || ret2 != -ENOENT)
2996 ret = 0;
2997 out:
2998 btrfs_free_path(path);
2999 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3000 WARN_ON(ret > 0);
3001 return ret;
3004 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3005 struct btrfs_root *root,
3006 struct extent_buffer *buf,
3007 int full_backref, int inc, int for_cow)
3009 u64 bytenr;
3010 u64 num_bytes;
3011 u64 parent;
3012 u64 ref_root;
3013 u32 nritems;
3014 struct btrfs_key key;
3015 struct btrfs_file_extent_item *fi;
3016 int i;
3017 int level;
3018 int ret = 0;
3019 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3020 u64, u64, u64, u64, u64, u64, int);
3022 ref_root = btrfs_header_owner(buf);
3023 nritems = btrfs_header_nritems(buf);
3024 level = btrfs_header_level(buf);
3026 if (!root->ref_cows && level == 0)
3027 return 0;
3029 if (inc)
3030 process_func = btrfs_inc_extent_ref;
3031 else
3032 process_func = btrfs_free_extent;
3034 if (full_backref)
3035 parent = buf->start;
3036 else
3037 parent = 0;
3039 for (i = 0; i < nritems; i++) {
3040 if (level == 0) {
3041 btrfs_item_key_to_cpu(buf, &key, i);
3042 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3043 continue;
3044 fi = btrfs_item_ptr(buf, i,
3045 struct btrfs_file_extent_item);
3046 if (btrfs_file_extent_type(buf, fi) ==
3047 BTRFS_FILE_EXTENT_INLINE)
3048 continue;
3049 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3050 if (bytenr == 0)
3051 continue;
3053 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3054 key.offset -= btrfs_file_extent_offset(buf, fi);
3055 ret = process_func(trans, root, bytenr, num_bytes,
3056 parent, ref_root, key.objectid,
3057 key.offset, for_cow);
3058 if (ret)
3059 goto fail;
3060 } else {
3061 bytenr = btrfs_node_blockptr(buf, i);
3062 num_bytes = btrfs_level_size(root, level - 1);
3063 ret = process_func(trans, root, bytenr, num_bytes,
3064 parent, ref_root, level - 1, 0,
3065 for_cow);
3066 if (ret)
3067 goto fail;
3070 return 0;
3071 fail:
3072 return ret;
3075 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3076 struct extent_buffer *buf, int full_backref, int for_cow)
3078 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
3081 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3082 struct extent_buffer *buf, int full_backref, int for_cow)
3084 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
3087 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3088 struct btrfs_root *root,
3089 struct btrfs_path *path,
3090 struct btrfs_block_group_cache *cache)
3092 int ret;
3093 struct btrfs_root *extent_root = root->fs_info->extent_root;
3094 unsigned long bi;
3095 struct extent_buffer *leaf;
3097 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3098 if (ret < 0)
3099 goto fail;
3100 BUG_ON(ret); /* Corruption */
3102 leaf = path->nodes[0];
3103 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3104 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3105 btrfs_mark_buffer_dirty(leaf);
3106 btrfs_release_path(path);
3107 fail:
3108 if (ret) {
3109 btrfs_abort_transaction(trans, root, ret);
3110 return ret;
3112 return 0;
3116 static struct btrfs_block_group_cache *
3117 next_block_group(struct btrfs_root *root,
3118 struct btrfs_block_group_cache *cache)
3120 struct rb_node *node;
3121 spin_lock(&root->fs_info->block_group_cache_lock);
3122 node = rb_next(&cache->cache_node);
3123 btrfs_put_block_group(cache);
3124 if (node) {
3125 cache = rb_entry(node, struct btrfs_block_group_cache,
3126 cache_node);
3127 btrfs_get_block_group(cache);
3128 } else
3129 cache = NULL;
3130 spin_unlock(&root->fs_info->block_group_cache_lock);
3131 return cache;
3134 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3135 struct btrfs_trans_handle *trans,
3136 struct btrfs_path *path)
3138 struct btrfs_root *root = block_group->fs_info->tree_root;
3139 struct inode *inode = NULL;
3140 u64 alloc_hint = 0;
3141 int dcs = BTRFS_DC_ERROR;
3142 int num_pages = 0;
3143 int retries = 0;
3144 int ret = 0;
3147 * If this block group is smaller than 100 megs don't bother caching the
3148 * block group.
3150 if (block_group->key.offset < (100 * 1024 * 1024)) {
3151 spin_lock(&block_group->lock);
3152 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3153 spin_unlock(&block_group->lock);
3154 return 0;
3157 again:
3158 inode = lookup_free_space_inode(root, block_group, path);
3159 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3160 ret = PTR_ERR(inode);
3161 btrfs_release_path(path);
3162 goto out;
3165 if (IS_ERR(inode)) {
3166 BUG_ON(retries);
3167 retries++;
3169 if (block_group->ro)
3170 goto out_free;
3172 ret = create_free_space_inode(root, trans, block_group, path);
3173 if (ret)
3174 goto out_free;
3175 goto again;
3178 /* We've already setup this transaction, go ahead and exit */
3179 if (block_group->cache_generation == trans->transid &&
3180 i_size_read(inode)) {
3181 dcs = BTRFS_DC_SETUP;
3182 goto out_put;
3186 * We want to set the generation to 0, that way if anything goes wrong
3187 * from here on out we know not to trust this cache when we load up next
3188 * time.
3190 BTRFS_I(inode)->generation = 0;
3191 ret = btrfs_update_inode(trans, root, inode);
3192 WARN_ON(ret);
3194 if (i_size_read(inode) > 0) {
3195 ret = btrfs_check_trunc_cache_free_space(root,
3196 &root->fs_info->global_block_rsv);
3197 if (ret)
3198 goto out_put;
3200 ret = btrfs_truncate_free_space_cache(root, trans, path,
3201 inode);
3202 if (ret)
3203 goto out_put;
3206 spin_lock(&block_group->lock);
3207 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3208 !btrfs_test_opt(root, SPACE_CACHE)) {
3210 * don't bother trying to write stuff out _if_
3211 * a) we're not cached,
3212 * b) we're with nospace_cache mount option.
3214 dcs = BTRFS_DC_WRITTEN;
3215 spin_unlock(&block_group->lock);
3216 goto out_put;
3218 spin_unlock(&block_group->lock);
3221 * Try to preallocate enough space based on how big the block group is.
3222 * Keep in mind this has to include any pinned space which could end up
3223 * taking up quite a bit since it's not folded into the other space
3224 * cache.
3226 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3227 if (!num_pages)
3228 num_pages = 1;
3230 num_pages *= 16;
3231 num_pages *= PAGE_CACHE_SIZE;
3233 ret = btrfs_check_data_free_space(inode, num_pages);
3234 if (ret)
3235 goto out_put;
3237 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3238 num_pages, num_pages,
3239 &alloc_hint);
3240 if (!ret)
3241 dcs = BTRFS_DC_SETUP;
3242 btrfs_free_reserved_data_space(inode, num_pages);
3244 out_put:
3245 iput(inode);
3246 out_free:
3247 btrfs_release_path(path);
3248 out:
3249 spin_lock(&block_group->lock);
3250 if (!ret && dcs == BTRFS_DC_SETUP)
3251 block_group->cache_generation = trans->transid;
3252 block_group->disk_cache_state = dcs;
3253 spin_unlock(&block_group->lock);
3255 return ret;
3258 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3259 struct btrfs_root *root)
3261 struct btrfs_block_group_cache *cache;
3262 int err = 0;
3263 struct btrfs_path *path;
3264 u64 last = 0;
3266 path = btrfs_alloc_path();
3267 if (!path)
3268 return -ENOMEM;
3270 again:
3271 while (1) {
3272 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3273 while (cache) {
3274 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3275 break;
3276 cache = next_block_group(root, cache);
3278 if (!cache) {
3279 if (last == 0)
3280 break;
3281 last = 0;
3282 continue;
3284 err = cache_save_setup(cache, trans, path);
3285 last = cache->key.objectid + cache->key.offset;
3286 btrfs_put_block_group(cache);
3289 while (1) {
3290 if (last == 0) {
3291 err = btrfs_run_delayed_refs(trans, root,
3292 (unsigned long)-1);
3293 if (err) /* File system offline */
3294 goto out;
3297 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3298 while (cache) {
3299 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3300 btrfs_put_block_group(cache);
3301 goto again;
3304 if (cache->dirty)
3305 break;
3306 cache = next_block_group(root, cache);
3308 if (!cache) {
3309 if (last == 0)
3310 break;
3311 last = 0;
3312 continue;
3315 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3316 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3317 cache->dirty = 0;
3318 last = cache->key.objectid + cache->key.offset;
3320 err = write_one_cache_group(trans, root, path, cache);
3321 btrfs_put_block_group(cache);
3322 if (err) /* File system offline */
3323 goto out;
3326 while (1) {
3328 * I don't think this is needed since we're just marking our
3329 * preallocated extent as written, but just in case it can't
3330 * hurt.
3332 if (last == 0) {
3333 err = btrfs_run_delayed_refs(trans, root,
3334 (unsigned long)-1);
3335 if (err) /* File system offline */
3336 goto out;
3339 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3340 while (cache) {
3342 * Really this shouldn't happen, but it could if we
3343 * couldn't write the entire preallocated extent and
3344 * splitting the extent resulted in a new block.
3346 if (cache->dirty) {
3347 btrfs_put_block_group(cache);
3348 goto again;
3350 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3351 break;
3352 cache = next_block_group(root, cache);
3354 if (!cache) {
3355 if (last == 0)
3356 break;
3357 last = 0;
3358 continue;
3361 err = btrfs_write_out_cache(root, trans, cache, path);
3364 * If we didn't have an error then the cache state is still
3365 * NEED_WRITE, so we can set it to WRITTEN.
3367 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3368 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3369 last = cache->key.objectid + cache->key.offset;
3370 btrfs_put_block_group(cache);
3372 out:
3374 btrfs_free_path(path);
3375 return err;
3378 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3380 struct btrfs_block_group_cache *block_group;
3381 int readonly = 0;
3383 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3384 if (!block_group || block_group->ro)
3385 readonly = 1;
3386 if (block_group)
3387 btrfs_put_block_group(block_group);
3388 return readonly;
3391 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3392 u64 total_bytes, u64 bytes_used,
3393 struct btrfs_space_info **space_info)
3395 struct btrfs_space_info *found;
3396 int i;
3397 int factor;
3398 int ret;
3400 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3401 BTRFS_BLOCK_GROUP_RAID10))
3402 factor = 2;
3403 else
3404 factor = 1;
3406 found = __find_space_info(info, flags);
3407 if (found) {
3408 spin_lock(&found->lock);
3409 found->total_bytes += total_bytes;
3410 found->disk_total += total_bytes * factor;
3411 found->bytes_used += bytes_used;
3412 found->disk_used += bytes_used * factor;
3413 found->full = 0;
3414 spin_unlock(&found->lock);
3415 *space_info = found;
3416 return 0;
3418 found = kzalloc(sizeof(*found), GFP_NOFS);
3419 if (!found)
3420 return -ENOMEM;
3422 ret = percpu_counter_init(&found->total_bytes_pinned, 0);
3423 if (ret) {
3424 kfree(found);
3425 return ret;
3428 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3429 INIT_LIST_HEAD(&found->block_groups[i]);
3430 init_rwsem(&found->groups_sem);
3431 spin_lock_init(&found->lock);
3432 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3433 found->total_bytes = total_bytes;
3434 found->disk_total = total_bytes * factor;
3435 found->bytes_used = bytes_used;
3436 found->disk_used = bytes_used * factor;
3437 found->bytes_pinned = 0;
3438 found->bytes_reserved = 0;
3439 found->bytes_readonly = 0;
3440 found->bytes_may_use = 0;
3441 found->full = 0;
3442 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3443 found->chunk_alloc = 0;
3444 found->flush = 0;
3445 init_waitqueue_head(&found->wait);
3446 *space_info = found;
3447 list_add_rcu(&found->list, &info->space_info);
3448 if (flags & BTRFS_BLOCK_GROUP_DATA)
3449 info->data_sinfo = found;
3450 return 0;
3453 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3455 u64 extra_flags = chunk_to_extended(flags) &
3456 BTRFS_EXTENDED_PROFILE_MASK;
3458 write_seqlock(&fs_info->profiles_lock);
3459 if (flags & BTRFS_BLOCK_GROUP_DATA)
3460 fs_info->avail_data_alloc_bits |= extra_flags;
3461 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3462 fs_info->avail_metadata_alloc_bits |= extra_flags;
3463 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3464 fs_info->avail_system_alloc_bits |= extra_flags;
3465 write_sequnlock(&fs_info->profiles_lock);
3469 * returns target flags in extended format or 0 if restripe for this
3470 * chunk_type is not in progress
3472 * should be called with either volume_mutex or balance_lock held
3474 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3476 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3477 u64 target = 0;
3479 if (!bctl)
3480 return 0;
3482 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3483 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3484 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3485 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3486 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3487 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3488 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3489 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3490 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3493 return target;
3497 * @flags: available profiles in extended format (see ctree.h)
3499 * Returns reduced profile in chunk format. If profile changing is in
3500 * progress (either running or paused) picks the target profile (if it's
3501 * already available), otherwise falls back to plain reducing.
3503 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3506 * we add in the count of missing devices because we want
3507 * to make sure that any RAID levels on a degraded FS
3508 * continue to be honored.
3510 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3511 root->fs_info->fs_devices->missing_devices;
3512 u64 target;
3513 u64 tmp;
3516 * see if restripe for this chunk_type is in progress, if so
3517 * try to reduce to the target profile
3519 spin_lock(&root->fs_info->balance_lock);
3520 target = get_restripe_target(root->fs_info, flags);
3521 if (target) {
3522 /* pick target profile only if it's already available */
3523 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3524 spin_unlock(&root->fs_info->balance_lock);
3525 return extended_to_chunk(target);
3528 spin_unlock(&root->fs_info->balance_lock);
3530 /* First, mask out the RAID levels which aren't possible */
3531 if (num_devices == 1)
3532 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3533 BTRFS_BLOCK_GROUP_RAID5);
3534 if (num_devices < 3)
3535 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3536 if (num_devices < 4)
3537 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3539 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3540 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3541 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3542 flags &= ~tmp;
3544 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3545 tmp = BTRFS_BLOCK_GROUP_RAID6;
3546 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3547 tmp = BTRFS_BLOCK_GROUP_RAID5;
3548 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3549 tmp = BTRFS_BLOCK_GROUP_RAID10;
3550 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3551 tmp = BTRFS_BLOCK_GROUP_RAID1;
3552 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3553 tmp = BTRFS_BLOCK_GROUP_RAID0;
3555 return extended_to_chunk(flags | tmp);
3558 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3560 unsigned seq;
3562 do {
3563 seq = read_seqbegin(&root->fs_info->profiles_lock);
3565 if (flags & BTRFS_BLOCK_GROUP_DATA)
3566 flags |= root->fs_info->avail_data_alloc_bits;
3567 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3568 flags |= root->fs_info->avail_system_alloc_bits;
3569 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3570 flags |= root->fs_info->avail_metadata_alloc_bits;
3571 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3573 return btrfs_reduce_alloc_profile(root, flags);
3576 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3578 u64 flags;
3579 u64 ret;
3581 if (data)
3582 flags = BTRFS_BLOCK_GROUP_DATA;
3583 else if (root == root->fs_info->chunk_root)
3584 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3585 else
3586 flags = BTRFS_BLOCK_GROUP_METADATA;
3588 ret = get_alloc_profile(root, flags);
3589 return ret;
3593 * This will check the space that the inode allocates from to make sure we have
3594 * enough space for bytes.
3596 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3598 struct btrfs_space_info *data_sinfo;
3599 struct btrfs_root *root = BTRFS_I(inode)->root;
3600 struct btrfs_fs_info *fs_info = root->fs_info;
3601 u64 used;
3602 int ret = 0, committed = 0, alloc_chunk = 1;
3604 /* make sure bytes are sectorsize aligned */
3605 bytes = ALIGN(bytes, root->sectorsize);
3607 if (root == root->fs_info->tree_root ||
3608 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3609 alloc_chunk = 0;
3610 committed = 1;
3613 data_sinfo = fs_info->data_sinfo;
3614 if (!data_sinfo)
3615 goto alloc;
3617 again:
3618 /* make sure we have enough space to handle the data first */
3619 spin_lock(&data_sinfo->lock);
3620 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3621 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3622 data_sinfo->bytes_may_use;
3624 if (used + bytes > data_sinfo->total_bytes) {
3625 struct btrfs_trans_handle *trans;
3628 * if we don't have enough free bytes in this space then we need
3629 * to alloc a new chunk.
3631 if (!data_sinfo->full && alloc_chunk) {
3632 u64 alloc_target;
3634 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3635 spin_unlock(&data_sinfo->lock);
3636 alloc:
3637 alloc_target = btrfs_get_alloc_profile(root, 1);
3638 trans = btrfs_join_transaction(root);
3639 if (IS_ERR(trans))
3640 return PTR_ERR(trans);
3642 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3643 alloc_target,
3644 CHUNK_ALLOC_NO_FORCE);
3645 btrfs_end_transaction(trans, root);
3646 if (ret < 0) {
3647 if (ret != -ENOSPC)
3648 return ret;
3649 else
3650 goto commit_trans;
3653 if (!data_sinfo)
3654 data_sinfo = fs_info->data_sinfo;
3656 goto again;
3660 * If we don't have enough pinned space to deal with this
3661 * allocation don't bother committing the transaction.
3663 if (percpu_counter_compare(&data_sinfo->total_bytes_pinned,
3664 bytes) < 0)
3665 committed = 1;
3666 spin_unlock(&data_sinfo->lock);
3668 /* commit the current transaction and try again */
3669 commit_trans:
3670 if (!committed &&
3671 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3672 committed = 1;
3674 trans = btrfs_join_transaction(root);
3675 if (IS_ERR(trans))
3676 return PTR_ERR(trans);
3677 ret = btrfs_commit_transaction(trans, root);
3678 if (ret)
3679 return ret;
3680 goto again;
3683 return -ENOSPC;
3685 data_sinfo->bytes_may_use += bytes;
3686 trace_btrfs_space_reservation(root->fs_info, "space_info",
3687 data_sinfo->flags, bytes, 1);
3688 spin_unlock(&data_sinfo->lock);
3690 return 0;
3694 * Called if we need to clear a data reservation for this inode.
3696 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3698 struct btrfs_root *root = BTRFS_I(inode)->root;
3699 struct btrfs_space_info *data_sinfo;
3701 /* make sure bytes are sectorsize aligned */
3702 bytes = ALIGN(bytes, root->sectorsize);
3704 data_sinfo = root->fs_info->data_sinfo;
3705 spin_lock(&data_sinfo->lock);
3706 WARN_ON(data_sinfo->bytes_may_use < bytes);
3707 data_sinfo->bytes_may_use -= bytes;
3708 trace_btrfs_space_reservation(root->fs_info, "space_info",
3709 data_sinfo->flags, bytes, 0);
3710 spin_unlock(&data_sinfo->lock);
3713 static void force_metadata_allocation(struct btrfs_fs_info *info)
3715 struct list_head *head = &info->space_info;
3716 struct btrfs_space_info *found;
3718 rcu_read_lock();
3719 list_for_each_entry_rcu(found, head, list) {
3720 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3721 found->force_alloc = CHUNK_ALLOC_FORCE;
3723 rcu_read_unlock();
3726 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
3728 return (global->size << 1);
3731 static int should_alloc_chunk(struct btrfs_root *root,
3732 struct btrfs_space_info *sinfo, int force)
3734 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3735 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3736 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3737 u64 thresh;
3739 if (force == CHUNK_ALLOC_FORCE)
3740 return 1;
3743 * We need to take into account the global rsv because for all intents
3744 * and purposes it's used space. Don't worry about locking the
3745 * global_rsv, it doesn't change except when the transaction commits.
3747 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3748 num_allocated += calc_global_rsv_need_space(global_rsv);
3751 * in limited mode, we want to have some free space up to
3752 * about 1% of the FS size.
3754 if (force == CHUNK_ALLOC_LIMITED) {
3755 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3756 thresh = max_t(u64, 64 * 1024 * 1024,
3757 div_factor_fine(thresh, 1));
3759 if (num_bytes - num_allocated < thresh)
3760 return 1;
3763 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3764 return 0;
3765 return 1;
3768 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3770 u64 num_dev;
3772 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3773 BTRFS_BLOCK_GROUP_RAID0 |
3774 BTRFS_BLOCK_GROUP_RAID5 |
3775 BTRFS_BLOCK_GROUP_RAID6))
3776 num_dev = root->fs_info->fs_devices->rw_devices;
3777 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3778 num_dev = 2;
3779 else
3780 num_dev = 1; /* DUP or single */
3782 /* metadata for updaing devices and chunk tree */
3783 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3786 static void check_system_chunk(struct btrfs_trans_handle *trans,
3787 struct btrfs_root *root, u64 type)
3789 struct btrfs_space_info *info;
3790 u64 left;
3791 u64 thresh;
3793 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3794 spin_lock(&info->lock);
3795 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3796 info->bytes_reserved - info->bytes_readonly;
3797 spin_unlock(&info->lock);
3799 thresh = get_system_chunk_thresh(root, type);
3800 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3801 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
3802 left, thresh, type);
3803 dump_space_info(info, 0, 0);
3806 if (left < thresh) {
3807 u64 flags;
3809 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3810 btrfs_alloc_chunk(trans, root, flags);
3814 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3815 struct btrfs_root *extent_root, u64 flags, int force)
3817 struct btrfs_space_info *space_info;
3818 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3819 int wait_for_alloc = 0;
3820 int ret = 0;
3822 /* Don't re-enter if we're already allocating a chunk */
3823 if (trans->allocating_chunk)
3824 return -ENOSPC;
3826 space_info = __find_space_info(extent_root->fs_info, flags);
3827 if (!space_info) {
3828 ret = update_space_info(extent_root->fs_info, flags,
3829 0, 0, &space_info);
3830 BUG_ON(ret); /* -ENOMEM */
3832 BUG_ON(!space_info); /* Logic error */
3834 again:
3835 spin_lock(&space_info->lock);
3836 if (force < space_info->force_alloc)
3837 force = space_info->force_alloc;
3838 if (space_info->full) {
3839 if (should_alloc_chunk(extent_root, space_info, force))
3840 ret = -ENOSPC;
3841 else
3842 ret = 0;
3843 spin_unlock(&space_info->lock);
3844 return ret;
3847 if (!should_alloc_chunk(extent_root, space_info, force)) {
3848 spin_unlock(&space_info->lock);
3849 return 0;
3850 } else if (space_info->chunk_alloc) {
3851 wait_for_alloc = 1;
3852 } else {
3853 space_info->chunk_alloc = 1;
3856 spin_unlock(&space_info->lock);
3858 mutex_lock(&fs_info->chunk_mutex);
3861 * The chunk_mutex is held throughout the entirety of a chunk
3862 * allocation, so once we've acquired the chunk_mutex we know that the
3863 * other guy is done and we need to recheck and see if we should
3864 * allocate.
3866 if (wait_for_alloc) {
3867 mutex_unlock(&fs_info->chunk_mutex);
3868 wait_for_alloc = 0;
3869 goto again;
3872 trans->allocating_chunk = true;
3875 * If we have mixed data/metadata chunks we want to make sure we keep
3876 * allocating mixed chunks instead of individual chunks.
3878 if (btrfs_mixed_space_info(space_info))
3879 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3882 * if we're doing a data chunk, go ahead and make sure that
3883 * we keep a reasonable number of metadata chunks allocated in the
3884 * FS as well.
3886 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3887 fs_info->data_chunk_allocations++;
3888 if (!(fs_info->data_chunk_allocations %
3889 fs_info->metadata_ratio))
3890 force_metadata_allocation(fs_info);
3894 * Check if we have enough space in SYSTEM chunk because we may need
3895 * to update devices.
3897 check_system_chunk(trans, extent_root, flags);
3899 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3900 trans->allocating_chunk = false;
3902 spin_lock(&space_info->lock);
3903 if (ret < 0 && ret != -ENOSPC)
3904 goto out;
3905 if (ret)
3906 space_info->full = 1;
3907 else
3908 ret = 1;
3910 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3911 out:
3912 space_info->chunk_alloc = 0;
3913 spin_unlock(&space_info->lock);
3914 mutex_unlock(&fs_info->chunk_mutex);
3915 return ret;
3918 static int can_overcommit(struct btrfs_root *root,
3919 struct btrfs_space_info *space_info, u64 bytes,
3920 enum btrfs_reserve_flush_enum flush)
3922 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3923 u64 profile = btrfs_get_alloc_profile(root, 0);
3924 u64 space_size;
3925 u64 avail;
3926 u64 used;
3928 used = space_info->bytes_used + space_info->bytes_reserved +
3929 space_info->bytes_pinned + space_info->bytes_readonly;
3932 * We only want to allow over committing if we have lots of actual space
3933 * free, but if we don't have enough space to handle the global reserve
3934 * space then we could end up having a real enospc problem when trying
3935 * to allocate a chunk or some other such important allocation.
3937 spin_lock(&global_rsv->lock);
3938 space_size = calc_global_rsv_need_space(global_rsv);
3939 spin_unlock(&global_rsv->lock);
3940 if (used + space_size >= space_info->total_bytes)
3941 return 0;
3943 used += space_info->bytes_may_use;
3945 spin_lock(&root->fs_info->free_chunk_lock);
3946 avail = root->fs_info->free_chunk_space;
3947 spin_unlock(&root->fs_info->free_chunk_lock);
3950 * If we have dup, raid1 or raid10 then only half of the free
3951 * space is actually useable. For raid56, the space info used
3952 * doesn't include the parity drive, so we don't have to
3953 * change the math
3955 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3956 BTRFS_BLOCK_GROUP_RAID1 |
3957 BTRFS_BLOCK_GROUP_RAID10))
3958 avail >>= 1;
3961 * If we aren't flushing all things, let us overcommit up to
3962 * 1/2th of the space. If we can flush, don't let us overcommit
3963 * too much, let it overcommit up to 1/8 of the space.
3965 if (flush == BTRFS_RESERVE_FLUSH_ALL)
3966 avail >>= 3;
3967 else
3968 avail >>= 1;
3970 if (used + bytes < space_info->total_bytes + avail)
3971 return 1;
3972 return 0;
3975 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
3976 unsigned long nr_pages)
3978 struct super_block *sb = root->fs_info->sb;
3980 if (down_read_trylock(&sb->s_umount)) {
3981 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
3982 up_read(&sb->s_umount);
3983 } else {
3985 * We needn't worry the filesystem going from r/w to r/o though
3986 * we don't acquire ->s_umount mutex, because the filesystem
3987 * should guarantee the delalloc inodes list be empty after
3988 * the filesystem is readonly(all dirty pages are written to
3989 * the disk).
3991 btrfs_start_all_delalloc_inodes(root->fs_info, 0);
3992 if (!current->journal_info)
3993 btrfs_wait_all_ordered_extents(root->fs_info);
3998 * shrink metadata reservation for delalloc
4000 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4001 bool wait_ordered)
4003 struct btrfs_block_rsv *block_rsv;
4004 struct btrfs_space_info *space_info;
4005 struct btrfs_trans_handle *trans;
4006 u64 delalloc_bytes;
4007 u64 max_reclaim;
4008 long time_left;
4009 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
4010 int loops = 0;
4011 enum btrfs_reserve_flush_enum flush;
4013 trans = (struct btrfs_trans_handle *)current->journal_info;
4014 block_rsv = &root->fs_info->delalloc_block_rsv;
4015 space_info = block_rsv->space_info;
4017 smp_mb();
4018 delalloc_bytes = percpu_counter_sum_positive(
4019 &root->fs_info->delalloc_bytes);
4020 if (delalloc_bytes == 0) {
4021 if (trans)
4022 return;
4023 btrfs_wait_all_ordered_extents(root->fs_info);
4024 return;
4027 while (delalloc_bytes && loops < 3) {
4028 max_reclaim = min(delalloc_bytes, to_reclaim);
4029 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4030 btrfs_writeback_inodes_sb_nr(root, nr_pages);
4032 * We need to wait for the async pages to actually start before
4033 * we do anything.
4035 wait_event(root->fs_info->async_submit_wait,
4036 !atomic_read(&root->fs_info->async_delalloc_pages));
4038 if (!trans)
4039 flush = BTRFS_RESERVE_FLUSH_ALL;
4040 else
4041 flush = BTRFS_RESERVE_NO_FLUSH;
4042 spin_lock(&space_info->lock);
4043 if (can_overcommit(root, space_info, orig, flush)) {
4044 spin_unlock(&space_info->lock);
4045 break;
4047 spin_unlock(&space_info->lock);
4049 loops++;
4050 if (wait_ordered && !trans) {
4051 btrfs_wait_all_ordered_extents(root->fs_info);
4052 } else {
4053 time_left = schedule_timeout_killable(1);
4054 if (time_left)
4055 break;
4057 smp_mb();
4058 delalloc_bytes = percpu_counter_sum_positive(
4059 &root->fs_info->delalloc_bytes);
4064 * maybe_commit_transaction - possibly commit the transaction if its ok to
4065 * @root - the root we're allocating for
4066 * @bytes - the number of bytes we want to reserve
4067 * @force - force the commit
4069 * This will check to make sure that committing the transaction will actually
4070 * get us somewhere and then commit the transaction if it does. Otherwise it
4071 * will return -ENOSPC.
4073 static int may_commit_transaction(struct btrfs_root *root,
4074 struct btrfs_space_info *space_info,
4075 u64 bytes, int force)
4077 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4078 struct btrfs_trans_handle *trans;
4080 trans = (struct btrfs_trans_handle *)current->journal_info;
4081 if (trans)
4082 return -EAGAIN;
4084 if (force)
4085 goto commit;
4087 /* See if there is enough pinned space to make this reservation */
4088 spin_lock(&space_info->lock);
4089 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4090 bytes) >= 0) {
4091 spin_unlock(&space_info->lock);
4092 goto commit;
4094 spin_unlock(&space_info->lock);
4097 * See if there is some space in the delayed insertion reservation for
4098 * this reservation.
4100 if (space_info != delayed_rsv->space_info)
4101 return -ENOSPC;
4103 spin_lock(&space_info->lock);
4104 spin_lock(&delayed_rsv->lock);
4105 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4106 bytes - delayed_rsv->size) >= 0) {
4107 spin_unlock(&delayed_rsv->lock);
4108 spin_unlock(&space_info->lock);
4109 return -ENOSPC;
4111 spin_unlock(&delayed_rsv->lock);
4112 spin_unlock(&space_info->lock);
4114 commit:
4115 trans = btrfs_join_transaction(root);
4116 if (IS_ERR(trans))
4117 return -ENOSPC;
4119 return btrfs_commit_transaction(trans, root);
4122 enum flush_state {
4123 FLUSH_DELAYED_ITEMS_NR = 1,
4124 FLUSH_DELAYED_ITEMS = 2,
4125 FLUSH_DELALLOC = 3,
4126 FLUSH_DELALLOC_WAIT = 4,
4127 ALLOC_CHUNK = 5,
4128 COMMIT_TRANS = 6,
4131 static int flush_space(struct btrfs_root *root,
4132 struct btrfs_space_info *space_info, u64 num_bytes,
4133 u64 orig_bytes, int state)
4135 struct btrfs_trans_handle *trans;
4136 int nr;
4137 int ret = 0;
4139 switch (state) {
4140 case FLUSH_DELAYED_ITEMS_NR:
4141 case FLUSH_DELAYED_ITEMS:
4142 if (state == FLUSH_DELAYED_ITEMS_NR) {
4143 u64 bytes = btrfs_calc_trans_metadata_size(root, 1);
4145 nr = (int)div64_u64(num_bytes, bytes);
4146 if (!nr)
4147 nr = 1;
4148 nr *= 2;
4149 } else {
4150 nr = -1;
4152 trans = btrfs_join_transaction(root);
4153 if (IS_ERR(trans)) {
4154 ret = PTR_ERR(trans);
4155 break;
4157 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4158 btrfs_end_transaction(trans, root);
4159 break;
4160 case FLUSH_DELALLOC:
4161 case FLUSH_DELALLOC_WAIT:
4162 shrink_delalloc(root, num_bytes, orig_bytes,
4163 state == FLUSH_DELALLOC_WAIT);
4164 break;
4165 case ALLOC_CHUNK:
4166 trans = btrfs_join_transaction(root);
4167 if (IS_ERR(trans)) {
4168 ret = PTR_ERR(trans);
4169 break;
4171 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4172 btrfs_get_alloc_profile(root, 0),
4173 CHUNK_ALLOC_NO_FORCE);
4174 btrfs_end_transaction(trans, root);
4175 if (ret == -ENOSPC)
4176 ret = 0;
4177 break;
4178 case COMMIT_TRANS:
4179 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4180 break;
4181 default:
4182 ret = -ENOSPC;
4183 break;
4186 return ret;
4189 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4190 * @root - the root we're allocating for
4191 * @block_rsv - the block_rsv we're allocating for
4192 * @orig_bytes - the number of bytes we want
4193 * @flush - whether or not we can flush to make our reservation
4195 * This will reserve orgi_bytes number of bytes from the space info associated
4196 * with the block_rsv. If there is not enough space it will make an attempt to
4197 * flush out space to make room. It will do this by flushing delalloc if
4198 * possible or committing the transaction. If flush is 0 then no attempts to
4199 * regain reservations will be made and this will fail if there is not enough
4200 * space already.
4202 static int reserve_metadata_bytes(struct btrfs_root *root,
4203 struct btrfs_block_rsv *block_rsv,
4204 u64 orig_bytes,
4205 enum btrfs_reserve_flush_enum flush)
4207 struct btrfs_space_info *space_info = block_rsv->space_info;
4208 u64 used;
4209 u64 num_bytes = orig_bytes;
4210 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4211 int ret = 0;
4212 bool flushing = false;
4214 again:
4215 ret = 0;
4216 spin_lock(&space_info->lock);
4218 * We only want to wait if somebody other than us is flushing and we
4219 * are actually allowed to flush all things.
4221 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4222 space_info->flush) {
4223 spin_unlock(&space_info->lock);
4225 * If we have a trans handle we can't wait because the flusher
4226 * may have to commit the transaction, which would mean we would
4227 * deadlock since we are waiting for the flusher to finish, but
4228 * hold the current transaction open.
4230 if (current->journal_info)
4231 return -EAGAIN;
4232 ret = wait_event_killable(space_info->wait, !space_info->flush);
4233 /* Must have been killed, return */
4234 if (ret)
4235 return -EINTR;
4237 spin_lock(&space_info->lock);
4240 ret = -ENOSPC;
4241 used = space_info->bytes_used + space_info->bytes_reserved +
4242 space_info->bytes_pinned + space_info->bytes_readonly +
4243 space_info->bytes_may_use;
4246 * The idea here is that we've not already over-reserved the block group
4247 * then we can go ahead and save our reservation first and then start
4248 * flushing if we need to. Otherwise if we've already overcommitted
4249 * lets start flushing stuff first and then come back and try to make
4250 * our reservation.
4252 if (used <= space_info->total_bytes) {
4253 if (used + orig_bytes <= space_info->total_bytes) {
4254 space_info->bytes_may_use += orig_bytes;
4255 trace_btrfs_space_reservation(root->fs_info,
4256 "space_info", space_info->flags, orig_bytes, 1);
4257 ret = 0;
4258 } else {
4260 * Ok set num_bytes to orig_bytes since we aren't
4261 * overocmmitted, this way we only try and reclaim what
4262 * we need.
4264 num_bytes = orig_bytes;
4266 } else {
4268 * Ok we're over committed, set num_bytes to the overcommitted
4269 * amount plus the amount of bytes that we need for this
4270 * reservation.
4272 num_bytes = used - space_info->total_bytes +
4273 (orig_bytes * 2);
4276 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4277 space_info->bytes_may_use += orig_bytes;
4278 trace_btrfs_space_reservation(root->fs_info, "space_info",
4279 space_info->flags, orig_bytes,
4281 ret = 0;
4285 * Couldn't make our reservation, save our place so while we're trying
4286 * to reclaim space we can actually use it instead of somebody else
4287 * stealing it from us.
4289 * We make the other tasks wait for the flush only when we can flush
4290 * all things.
4292 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4293 flushing = true;
4294 space_info->flush = 1;
4297 spin_unlock(&space_info->lock);
4299 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4300 goto out;
4302 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4303 flush_state);
4304 flush_state++;
4307 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4308 * would happen. So skip delalloc flush.
4310 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4311 (flush_state == FLUSH_DELALLOC ||
4312 flush_state == FLUSH_DELALLOC_WAIT))
4313 flush_state = ALLOC_CHUNK;
4315 if (!ret)
4316 goto again;
4317 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4318 flush_state < COMMIT_TRANS)
4319 goto again;
4320 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4321 flush_state <= COMMIT_TRANS)
4322 goto again;
4324 out:
4325 if (ret == -ENOSPC &&
4326 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4327 struct btrfs_block_rsv *global_rsv =
4328 &root->fs_info->global_block_rsv;
4330 if (block_rsv != global_rsv &&
4331 !block_rsv_use_bytes(global_rsv, orig_bytes))
4332 ret = 0;
4334 if (flushing) {
4335 spin_lock(&space_info->lock);
4336 space_info->flush = 0;
4337 wake_up_all(&space_info->wait);
4338 spin_unlock(&space_info->lock);
4340 return ret;
4343 static struct btrfs_block_rsv *get_block_rsv(
4344 const struct btrfs_trans_handle *trans,
4345 const struct btrfs_root *root)
4347 struct btrfs_block_rsv *block_rsv = NULL;
4349 if (root->ref_cows)
4350 block_rsv = trans->block_rsv;
4352 if (root == root->fs_info->csum_root && trans->adding_csums)
4353 block_rsv = trans->block_rsv;
4355 if (root == root->fs_info->uuid_root)
4356 block_rsv = trans->block_rsv;
4358 if (!block_rsv)
4359 block_rsv = root->block_rsv;
4361 if (!block_rsv)
4362 block_rsv = &root->fs_info->empty_block_rsv;
4364 return block_rsv;
4367 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4368 u64 num_bytes)
4370 int ret = -ENOSPC;
4371 spin_lock(&block_rsv->lock);
4372 if (block_rsv->reserved >= num_bytes) {
4373 block_rsv->reserved -= num_bytes;
4374 if (block_rsv->reserved < block_rsv->size)
4375 block_rsv->full = 0;
4376 ret = 0;
4378 spin_unlock(&block_rsv->lock);
4379 return ret;
4382 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4383 u64 num_bytes, int update_size)
4385 spin_lock(&block_rsv->lock);
4386 block_rsv->reserved += num_bytes;
4387 if (update_size)
4388 block_rsv->size += num_bytes;
4389 else if (block_rsv->reserved >= block_rsv->size)
4390 block_rsv->full = 1;
4391 spin_unlock(&block_rsv->lock);
4394 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4395 struct btrfs_block_rsv *dest, u64 num_bytes,
4396 int min_factor)
4398 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4399 u64 min_bytes;
4401 if (global_rsv->space_info != dest->space_info)
4402 return -ENOSPC;
4404 spin_lock(&global_rsv->lock);
4405 min_bytes = div_factor(global_rsv->size, min_factor);
4406 if (global_rsv->reserved < min_bytes + num_bytes) {
4407 spin_unlock(&global_rsv->lock);
4408 return -ENOSPC;
4410 global_rsv->reserved -= num_bytes;
4411 if (global_rsv->reserved < global_rsv->size)
4412 global_rsv->full = 0;
4413 spin_unlock(&global_rsv->lock);
4415 block_rsv_add_bytes(dest, num_bytes, 1);
4416 return 0;
4419 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4420 struct btrfs_block_rsv *block_rsv,
4421 struct btrfs_block_rsv *dest, u64 num_bytes)
4423 struct btrfs_space_info *space_info = block_rsv->space_info;
4425 spin_lock(&block_rsv->lock);
4426 if (num_bytes == (u64)-1)
4427 num_bytes = block_rsv->size;
4428 block_rsv->size -= num_bytes;
4429 if (block_rsv->reserved >= block_rsv->size) {
4430 num_bytes = block_rsv->reserved - block_rsv->size;
4431 block_rsv->reserved = block_rsv->size;
4432 block_rsv->full = 1;
4433 } else {
4434 num_bytes = 0;
4436 spin_unlock(&block_rsv->lock);
4438 if (num_bytes > 0) {
4439 if (dest) {
4440 spin_lock(&dest->lock);
4441 if (!dest->full) {
4442 u64 bytes_to_add;
4444 bytes_to_add = dest->size - dest->reserved;
4445 bytes_to_add = min(num_bytes, bytes_to_add);
4446 dest->reserved += bytes_to_add;
4447 if (dest->reserved >= dest->size)
4448 dest->full = 1;
4449 num_bytes -= bytes_to_add;
4451 spin_unlock(&dest->lock);
4453 if (num_bytes) {
4454 spin_lock(&space_info->lock);
4455 space_info->bytes_may_use -= num_bytes;
4456 trace_btrfs_space_reservation(fs_info, "space_info",
4457 space_info->flags, num_bytes, 0);
4458 spin_unlock(&space_info->lock);
4463 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4464 struct btrfs_block_rsv *dst, u64 num_bytes)
4466 int ret;
4468 ret = block_rsv_use_bytes(src, num_bytes);
4469 if (ret)
4470 return ret;
4472 block_rsv_add_bytes(dst, num_bytes, 1);
4473 return 0;
4476 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4478 memset(rsv, 0, sizeof(*rsv));
4479 spin_lock_init(&rsv->lock);
4480 rsv->type = type;
4483 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4484 unsigned short type)
4486 struct btrfs_block_rsv *block_rsv;
4487 struct btrfs_fs_info *fs_info = root->fs_info;
4489 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4490 if (!block_rsv)
4491 return NULL;
4493 btrfs_init_block_rsv(block_rsv, type);
4494 block_rsv->space_info = __find_space_info(fs_info,
4495 BTRFS_BLOCK_GROUP_METADATA);
4496 return block_rsv;
4499 void btrfs_free_block_rsv(struct btrfs_root *root,
4500 struct btrfs_block_rsv *rsv)
4502 if (!rsv)
4503 return;
4504 btrfs_block_rsv_release(root, rsv, (u64)-1);
4505 kfree(rsv);
4508 int btrfs_block_rsv_add(struct btrfs_root *root,
4509 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4510 enum btrfs_reserve_flush_enum flush)
4512 int ret;
4514 if (num_bytes == 0)
4515 return 0;
4517 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4518 if (!ret) {
4519 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4520 return 0;
4523 return ret;
4526 int btrfs_block_rsv_check(struct btrfs_root *root,
4527 struct btrfs_block_rsv *block_rsv, int min_factor)
4529 u64 num_bytes = 0;
4530 int ret = -ENOSPC;
4532 if (!block_rsv)
4533 return 0;
4535 spin_lock(&block_rsv->lock);
4536 num_bytes = div_factor(block_rsv->size, min_factor);
4537 if (block_rsv->reserved >= num_bytes)
4538 ret = 0;
4539 spin_unlock(&block_rsv->lock);
4541 return ret;
4544 int btrfs_block_rsv_refill(struct btrfs_root *root,
4545 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4546 enum btrfs_reserve_flush_enum flush)
4548 u64 num_bytes = 0;
4549 int ret = -ENOSPC;
4551 if (!block_rsv)
4552 return 0;
4554 spin_lock(&block_rsv->lock);
4555 num_bytes = min_reserved;
4556 if (block_rsv->reserved >= num_bytes)
4557 ret = 0;
4558 else
4559 num_bytes -= block_rsv->reserved;
4560 spin_unlock(&block_rsv->lock);
4562 if (!ret)
4563 return 0;
4565 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4566 if (!ret) {
4567 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4568 return 0;
4571 return ret;
4574 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4575 struct btrfs_block_rsv *dst_rsv,
4576 u64 num_bytes)
4578 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4581 void btrfs_block_rsv_release(struct btrfs_root *root,
4582 struct btrfs_block_rsv *block_rsv,
4583 u64 num_bytes)
4585 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4586 if (global_rsv->full || global_rsv == block_rsv ||
4587 block_rsv->space_info != global_rsv->space_info)
4588 global_rsv = NULL;
4589 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4590 num_bytes);
4594 * helper to calculate size of global block reservation.
4595 * the desired value is sum of space used by extent tree,
4596 * checksum tree and root tree
4598 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4600 struct btrfs_space_info *sinfo;
4601 u64 num_bytes;
4602 u64 meta_used;
4603 u64 data_used;
4604 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4606 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4607 spin_lock(&sinfo->lock);
4608 data_used = sinfo->bytes_used;
4609 spin_unlock(&sinfo->lock);
4611 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4612 spin_lock(&sinfo->lock);
4613 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4614 data_used = 0;
4615 meta_used = sinfo->bytes_used;
4616 spin_unlock(&sinfo->lock);
4618 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4619 csum_size * 2;
4620 num_bytes += div64_u64(data_used + meta_used, 50);
4622 if (num_bytes * 3 > meta_used)
4623 num_bytes = div64_u64(meta_used, 3);
4625 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4628 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4630 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4631 struct btrfs_space_info *sinfo = block_rsv->space_info;
4632 u64 num_bytes;
4634 num_bytes = calc_global_metadata_size(fs_info);
4636 spin_lock(&sinfo->lock);
4637 spin_lock(&block_rsv->lock);
4639 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
4641 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4642 sinfo->bytes_reserved + sinfo->bytes_readonly +
4643 sinfo->bytes_may_use;
4645 if (sinfo->total_bytes > num_bytes) {
4646 num_bytes = sinfo->total_bytes - num_bytes;
4647 block_rsv->reserved += num_bytes;
4648 sinfo->bytes_may_use += num_bytes;
4649 trace_btrfs_space_reservation(fs_info, "space_info",
4650 sinfo->flags, num_bytes, 1);
4653 if (block_rsv->reserved >= block_rsv->size) {
4654 num_bytes = block_rsv->reserved - block_rsv->size;
4655 sinfo->bytes_may_use -= num_bytes;
4656 trace_btrfs_space_reservation(fs_info, "space_info",
4657 sinfo->flags, num_bytes, 0);
4658 block_rsv->reserved = block_rsv->size;
4659 block_rsv->full = 1;
4662 spin_unlock(&block_rsv->lock);
4663 spin_unlock(&sinfo->lock);
4666 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4668 struct btrfs_space_info *space_info;
4670 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4671 fs_info->chunk_block_rsv.space_info = space_info;
4673 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4674 fs_info->global_block_rsv.space_info = space_info;
4675 fs_info->delalloc_block_rsv.space_info = space_info;
4676 fs_info->trans_block_rsv.space_info = space_info;
4677 fs_info->empty_block_rsv.space_info = space_info;
4678 fs_info->delayed_block_rsv.space_info = space_info;
4680 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4681 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4682 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4683 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4684 if (fs_info->quota_root)
4685 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
4686 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4688 update_global_block_rsv(fs_info);
4691 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4693 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4694 (u64)-1);
4695 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4696 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4697 WARN_ON(fs_info->trans_block_rsv.size > 0);
4698 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4699 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4700 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4701 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4702 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4705 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4706 struct btrfs_root *root)
4708 if (!trans->block_rsv)
4709 return;
4711 if (!trans->bytes_reserved)
4712 return;
4714 trace_btrfs_space_reservation(root->fs_info, "transaction",
4715 trans->transid, trans->bytes_reserved, 0);
4716 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4717 trans->bytes_reserved = 0;
4720 /* Can only return 0 or -ENOSPC */
4721 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4722 struct inode *inode)
4724 struct btrfs_root *root = BTRFS_I(inode)->root;
4725 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4726 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4729 * We need to hold space in order to delete our orphan item once we've
4730 * added it, so this takes the reservation so we can release it later
4731 * when we are truly done with the orphan item.
4733 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4734 trace_btrfs_space_reservation(root->fs_info, "orphan",
4735 btrfs_ino(inode), num_bytes, 1);
4736 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4739 void btrfs_orphan_release_metadata(struct inode *inode)
4741 struct btrfs_root *root = BTRFS_I(inode)->root;
4742 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4743 trace_btrfs_space_reservation(root->fs_info, "orphan",
4744 btrfs_ino(inode), num_bytes, 0);
4745 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4749 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4750 * root: the root of the parent directory
4751 * rsv: block reservation
4752 * items: the number of items that we need do reservation
4753 * qgroup_reserved: used to return the reserved size in qgroup
4755 * This function is used to reserve the space for snapshot/subvolume
4756 * creation and deletion. Those operations are different with the
4757 * common file/directory operations, they change two fs/file trees
4758 * and root tree, the number of items that the qgroup reserves is
4759 * different with the free space reservation. So we can not use
4760 * the space reseravtion mechanism in start_transaction().
4762 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4763 struct btrfs_block_rsv *rsv,
4764 int items,
4765 u64 *qgroup_reserved,
4766 bool use_global_rsv)
4768 u64 num_bytes;
4769 int ret;
4770 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4772 if (root->fs_info->quota_enabled) {
4773 /* One for parent inode, two for dir entries */
4774 num_bytes = 3 * root->leafsize;
4775 ret = btrfs_qgroup_reserve(root, num_bytes);
4776 if (ret)
4777 return ret;
4778 } else {
4779 num_bytes = 0;
4782 *qgroup_reserved = num_bytes;
4784 num_bytes = btrfs_calc_trans_metadata_size(root, items);
4785 rsv->space_info = __find_space_info(root->fs_info,
4786 BTRFS_BLOCK_GROUP_METADATA);
4787 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
4788 BTRFS_RESERVE_FLUSH_ALL);
4790 if (ret == -ENOSPC && use_global_rsv)
4791 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
4793 if (ret) {
4794 if (*qgroup_reserved)
4795 btrfs_qgroup_free(root, *qgroup_reserved);
4798 return ret;
4801 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
4802 struct btrfs_block_rsv *rsv,
4803 u64 qgroup_reserved)
4805 btrfs_block_rsv_release(root, rsv, (u64)-1);
4806 if (qgroup_reserved)
4807 btrfs_qgroup_free(root, qgroup_reserved);
4811 * drop_outstanding_extent - drop an outstanding extent
4812 * @inode: the inode we're dropping the extent for
4814 * This is called when we are freeing up an outstanding extent, either called
4815 * after an error or after an extent is written. This will return the number of
4816 * reserved extents that need to be freed. This must be called with
4817 * BTRFS_I(inode)->lock held.
4819 static unsigned drop_outstanding_extent(struct inode *inode)
4821 unsigned drop_inode_space = 0;
4822 unsigned dropped_extents = 0;
4824 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4825 BTRFS_I(inode)->outstanding_extents--;
4827 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4828 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4829 &BTRFS_I(inode)->runtime_flags))
4830 drop_inode_space = 1;
4833 * If we have more or the same amount of outsanding extents than we have
4834 * reserved then we need to leave the reserved extents count alone.
4836 if (BTRFS_I(inode)->outstanding_extents >=
4837 BTRFS_I(inode)->reserved_extents)
4838 return drop_inode_space;
4840 dropped_extents = BTRFS_I(inode)->reserved_extents -
4841 BTRFS_I(inode)->outstanding_extents;
4842 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4843 return dropped_extents + drop_inode_space;
4847 * calc_csum_metadata_size - return the amount of metada space that must be
4848 * reserved/free'd for the given bytes.
4849 * @inode: the inode we're manipulating
4850 * @num_bytes: the number of bytes in question
4851 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4853 * This adjusts the number of csum_bytes in the inode and then returns the
4854 * correct amount of metadata that must either be reserved or freed. We
4855 * calculate how many checksums we can fit into one leaf and then divide the
4856 * number of bytes that will need to be checksumed by this value to figure out
4857 * how many checksums will be required. If we are adding bytes then the number
4858 * may go up and we will return the number of additional bytes that must be
4859 * reserved. If it is going down we will return the number of bytes that must
4860 * be freed.
4862 * This must be called with BTRFS_I(inode)->lock held.
4864 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4865 int reserve)
4867 struct btrfs_root *root = BTRFS_I(inode)->root;
4868 u64 csum_size;
4869 int num_csums_per_leaf;
4870 int num_csums;
4871 int old_csums;
4873 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4874 BTRFS_I(inode)->csum_bytes == 0)
4875 return 0;
4877 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4878 if (reserve)
4879 BTRFS_I(inode)->csum_bytes += num_bytes;
4880 else
4881 BTRFS_I(inode)->csum_bytes -= num_bytes;
4882 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4883 num_csums_per_leaf = (int)div64_u64(csum_size,
4884 sizeof(struct btrfs_csum_item) +
4885 sizeof(struct btrfs_disk_key));
4886 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4887 num_csums = num_csums + num_csums_per_leaf - 1;
4888 num_csums = num_csums / num_csums_per_leaf;
4890 old_csums = old_csums + num_csums_per_leaf - 1;
4891 old_csums = old_csums / num_csums_per_leaf;
4893 /* No change, no need to reserve more */
4894 if (old_csums == num_csums)
4895 return 0;
4897 if (reserve)
4898 return btrfs_calc_trans_metadata_size(root,
4899 num_csums - old_csums);
4901 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4904 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4906 struct btrfs_root *root = BTRFS_I(inode)->root;
4907 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4908 u64 to_reserve = 0;
4909 u64 csum_bytes;
4910 unsigned nr_extents = 0;
4911 int extra_reserve = 0;
4912 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
4913 int ret = 0;
4914 bool delalloc_lock = true;
4915 u64 to_free = 0;
4916 unsigned dropped;
4918 /* If we are a free space inode we need to not flush since we will be in
4919 * the middle of a transaction commit. We also don't need the delalloc
4920 * mutex since we won't race with anybody. We need this mostly to make
4921 * lockdep shut its filthy mouth.
4923 if (btrfs_is_free_space_inode(inode)) {
4924 flush = BTRFS_RESERVE_NO_FLUSH;
4925 delalloc_lock = false;
4928 if (flush != BTRFS_RESERVE_NO_FLUSH &&
4929 btrfs_transaction_in_commit(root->fs_info))
4930 schedule_timeout(1);
4932 if (delalloc_lock)
4933 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4935 num_bytes = ALIGN(num_bytes, root->sectorsize);
4937 spin_lock(&BTRFS_I(inode)->lock);
4938 BTRFS_I(inode)->outstanding_extents++;
4940 if (BTRFS_I(inode)->outstanding_extents >
4941 BTRFS_I(inode)->reserved_extents)
4942 nr_extents = BTRFS_I(inode)->outstanding_extents -
4943 BTRFS_I(inode)->reserved_extents;
4946 * Add an item to reserve for updating the inode when we complete the
4947 * delalloc io.
4949 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4950 &BTRFS_I(inode)->runtime_flags)) {
4951 nr_extents++;
4952 extra_reserve = 1;
4955 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4956 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4957 csum_bytes = BTRFS_I(inode)->csum_bytes;
4958 spin_unlock(&BTRFS_I(inode)->lock);
4960 if (root->fs_info->quota_enabled) {
4961 ret = btrfs_qgroup_reserve(root, num_bytes +
4962 nr_extents * root->leafsize);
4963 if (ret)
4964 goto out_fail;
4967 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4968 if (unlikely(ret)) {
4969 if (root->fs_info->quota_enabled)
4970 btrfs_qgroup_free(root, num_bytes +
4971 nr_extents * root->leafsize);
4972 goto out_fail;
4975 spin_lock(&BTRFS_I(inode)->lock);
4976 if (extra_reserve) {
4977 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4978 &BTRFS_I(inode)->runtime_flags);
4979 nr_extents--;
4981 BTRFS_I(inode)->reserved_extents += nr_extents;
4982 spin_unlock(&BTRFS_I(inode)->lock);
4984 if (delalloc_lock)
4985 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4987 if (to_reserve)
4988 trace_btrfs_space_reservation(root->fs_info,"delalloc",
4989 btrfs_ino(inode), to_reserve, 1);
4990 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4992 return 0;
4994 out_fail:
4995 spin_lock(&BTRFS_I(inode)->lock);
4996 dropped = drop_outstanding_extent(inode);
4998 * If the inodes csum_bytes is the same as the original
4999 * csum_bytes then we know we haven't raced with any free()ers
5000 * so we can just reduce our inodes csum bytes and carry on.
5002 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5003 calc_csum_metadata_size(inode, num_bytes, 0);
5004 } else {
5005 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5006 u64 bytes;
5009 * This is tricky, but first we need to figure out how much we
5010 * free'd from any free-ers that occured during this
5011 * reservation, so we reset ->csum_bytes to the csum_bytes
5012 * before we dropped our lock, and then call the free for the
5013 * number of bytes that were freed while we were trying our
5014 * reservation.
5016 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5017 BTRFS_I(inode)->csum_bytes = csum_bytes;
5018 to_free = calc_csum_metadata_size(inode, bytes, 0);
5022 * Now we need to see how much we would have freed had we not
5023 * been making this reservation and our ->csum_bytes were not
5024 * artificially inflated.
5026 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5027 bytes = csum_bytes - orig_csum_bytes;
5028 bytes = calc_csum_metadata_size(inode, bytes, 0);
5031 * Now reset ->csum_bytes to what it should be. If bytes is
5032 * more than to_free then we would have free'd more space had we
5033 * not had an artificially high ->csum_bytes, so we need to free
5034 * the remainder. If bytes is the same or less then we don't
5035 * need to do anything, the other free-ers did the correct
5036 * thing.
5038 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5039 if (bytes > to_free)
5040 to_free = bytes - to_free;
5041 else
5042 to_free = 0;
5044 spin_unlock(&BTRFS_I(inode)->lock);
5045 if (dropped)
5046 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5048 if (to_free) {
5049 btrfs_block_rsv_release(root, block_rsv, to_free);
5050 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5051 btrfs_ino(inode), to_free, 0);
5053 if (delalloc_lock)
5054 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5055 return ret;
5059 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5060 * @inode: the inode to release the reservation for
5061 * @num_bytes: the number of bytes we're releasing
5063 * This will release the metadata reservation for an inode. This can be called
5064 * once we complete IO for a given set of bytes to release their metadata
5065 * reservations.
5067 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5069 struct btrfs_root *root = BTRFS_I(inode)->root;
5070 u64 to_free = 0;
5071 unsigned dropped;
5073 num_bytes = ALIGN(num_bytes, root->sectorsize);
5074 spin_lock(&BTRFS_I(inode)->lock);
5075 dropped = drop_outstanding_extent(inode);
5077 if (num_bytes)
5078 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5079 spin_unlock(&BTRFS_I(inode)->lock);
5080 if (dropped > 0)
5081 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5083 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5084 btrfs_ino(inode), to_free, 0);
5085 if (root->fs_info->quota_enabled) {
5086 btrfs_qgroup_free(root, num_bytes +
5087 dropped * root->leafsize);
5090 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5091 to_free);
5095 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5096 * @inode: inode we're writing to
5097 * @num_bytes: the number of bytes we want to allocate
5099 * This will do the following things
5101 * o reserve space in the data space info for num_bytes
5102 * o reserve space in the metadata space info based on number of outstanding
5103 * extents and how much csums will be needed
5104 * o add to the inodes ->delalloc_bytes
5105 * o add it to the fs_info's delalloc inodes list.
5107 * This will return 0 for success and -ENOSPC if there is no space left.
5109 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5111 int ret;
5113 ret = btrfs_check_data_free_space(inode, num_bytes);
5114 if (ret)
5115 return ret;
5117 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5118 if (ret) {
5119 btrfs_free_reserved_data_space(inode, num_bytes);
5120 return ret;
5123 return 0;
5127 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5128 * @inode: inode we're releasing space for
5129 * @num_bytes: the number of bytes we want to free up
5131 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5132 * called in the case that we don't need the metadata AND data reservations
5133 * anymore. So if there is an error or we insert an inline extent.
5135 * This function will release the metadata space that was not used and will
5136 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5137 * list if there are no delalloc bytes left.
5139 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5141 btrfs_delalloc_release_metadata(inode, num_bytes);
5142 btrfs_free_reserved_data_space(inode, num_bytes);
5145 static int update_block_group(struct btrfs_root *root,
5146 u64 bytenr, u64 num_bytes, int alloc)
5148 struct btrfs_block_group_cache *cache = NULL;
5149 struct btrfs_fs_info *info = root->fs_info;
5150 u64 total = num_bytes;
5151 u64 old_val;
5152 u64 byte_in_group;
5153 int factor;
5155 /* block accounting for super block */
5156 spin_lock(&info->delalloc_root_lock);
5157 old_val = btrfs_super_bytes_used(info->super_copy);
5158 if (alloc)
5159 old_val += num_bytes;
5160 else
5161 old_val -= num_bytes;
5162 btrfs_set_super_bytes_used(info->super_copy, old_val);
5163 spin_unlock(&info->delalloc_root_lock);
5165 while (total) {
5166 cache = btrfs_lookup_block_group(info, bytenr);
5167 if (!cache)
5168 return -ENOENT;
5169 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5170 BTRFS_BLOCK_GROUP_RAID1 |
5171 BTRFS_BLOCK_GROUP_RAID10))
5172 factor = 2;
5173 else
5174 factor = 1;
5176 * If this block group has free space cache written out, we
5177 * need to make sure to load it if we are removing space. This
5178 * is because we need the unpinning stage to actually add the
5179 * space back to the block group, otherwise we will leak space.
5181 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5182 cache_block_group(cache, 1);
5184 byte_in_group = bytenr - cache->key.objectid;
5185 WARN_ON(byte_in_group > cache->key.offset);
5187 spin_lock(&cache->space_info->lock);
5188 spin_lock(&cache->lock);
5190 if (btrfs_test_opt(root, SPACE_CACHE) &&
5191 cache->disk_cache_state < BTRFS_DC_CLEAR)
5192 cache->disk_cache_state = BTRFS_DC_CLEAR;
5194 cache->dirty = 1;
5195 old_val = btrfs_block_group_used(&cache->item);
5196 num_bytes = min(total, cache->key.offset - byte_in_group);
5197 if (alloc) {
5198 old_val += num_bytes;
5199 btrfs_set_block_group_used(&cache->item, old_val);
5200 cache->reserved -= num_bytes;
5201 cache->space_info->bytes_reserved -= num_bytes;
5202 cache->space_info->bytes_used += num_bytes;
5203 cache->space_info->disk_used += num_bytes * factor;
5204 spin_unlock(&cache->lock);
5205 spin_unlock(&cache->space_info->lock);
5206 } else {
5207 old_val -= num_bytes;
5208 btrfs_set_block_group_used(&cache->item, old_val);
5209 cache->pinned += num_bytes;
5210 cache->space_info->bytes_pinned += num_bytes;
5211 cache->space_info->bytes_used -= num_bytes;
5212 cache->space_info->disk_used -= num_bytes * factor;
5213 spin_unlock(&cache->lock);
5214 spin_unlock(&cache->space_info->lock);
5216 set_extent_dirty(info->pinned_extents,
5217 bytenr, bytenr + num_bytes - 1,
5218 GFP_NOFS | __GFP_NOFAIL);
5220 btrfs_put_block_group(cache);
5221 total -= num_bytes;
5222 bytenr += num_bytes;
5224 return 0;
5227 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5229 struct btrfs_block_group_cache *cache;
5230 u64 bytenr;
5232 spin_lock(&root->fs_info->block_group_cache_lock);
5233 bytenr = root->fs_info->first_logical_byte;
5234 spin_unlock(&root->fs_info->block_group_cache_lock);
5236 if (bytenr < (u64)-1)
5237 return bytenr;
5239 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5240 if (!cache)
5241 return 0;
5243 bytenr = cache->key.objectid;
5244 btrfs_put_block_group(cache);
5246 return bytenr;
5249 static int pin_down_extent(struct btrfs_root *root,
5250 struct btrfs_block_group_cache *cache,
5251 u64 bytenr, u64 num_bytes, int reserved)
5253 spin_lock(&cache->space_info->lock);
5254 spin_lock(&cache->lock);
5255 cache->pinned += num_bytes;
5256 cache->space_info->bytes_pinned += num_bytes;
5257 if (reserved) {
5258 cache->reserved -= num_bytes;
5259 cache->space_info->bytes_reserved -= num_bytes;
5261 spin_unlock(&cache->lock);
5262 spin_unlock(&cache->space_info->lock);
5264 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5265 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5266 return 0;
5270 * this function must be called within transaction
5272 int btrfs_pin_extent(struct btrfs_root *root,
5273 u64 bytenr, u64 num_bytes, int reserved)
5275 struct btrfs_block_group_cache *cache;
5277 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5278 BUG_ON(!cache); /* Logic error */
5280 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5282 btrfs_put_block_group(cache);
5283 return 0;
5287 * this function must be called within transaction
5289 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5290 u64 bytenr, u64 num_bytes)
5292 struct btrfs_block_group_cache *cache;
5293 int ret;
5295 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5296 if (!cache)
5297 return -EINVAL;
5300 * pull in the free space cache (if any) so that our pin
5301 * removes the free space from the cache. We have load_only set
5302 * to one because the slow code to read in the free extents does check
5303 * the pinned extents.
5305 cache_block_group(cache, 1);
5307 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5309 /* remove us from the free space cache (if we're there at all) */
5310 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5311 btrfs_put_block_group(cache);
5312 return ret;
5315 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5317 int ret;
5318 struct btrfs_block_group_cache *block_group;
5319 struct btrfs_caching_control *caching_ctl;
5321 block_group = btrfs_lookup_block_group(root->fs_info, start);
5322 if (!block_group)
5323 return -EINVAL;
5325 cache_block_group(block_group, 0);
5326 caching_ctl = get_caching_control(block_group);
5328 if (!caching_ctl) {
5329 /* Logic error */
5330 BUG_ON(!block_group_cache_done(block_group));
5331 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5332 } else {
5333 mutex_lock(&caching_ctl->mutex);
5335 if (start >= caching_ctl->progress) {
5336 ret = add_excluded_extent(root, start, num_bytes);
5337 } else if (start + num_bytes <= caching_ctl->progress) {
5338 ret = btrfs_remove_free_space(block_group,
5339 start, num_bytes);
5340 } else {
5341 num_bytes = caching_ctl->progress - start;
5342 ret = btrfs_remove_free_space(block_group,
5343 start, num_bytes);
5344 if (ret)
5345 goto out_lock;
5347 num_bytes = (start + num_bytes) -
5348 caching_ctl->progress;
5349 start = caching_ctl->progress;
5350 ret = add_excluded_extent(root, start, num_bytes);
5352 out_lock:
5353 mutex_unlock(&caching_ctl->mutex);
5354 put_caching_control(caching_ctl);
5356 btrfs_put_block_group(block_group);
5357 return ret;
5360 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5361 struct extent_buffer *eb)
5363 struct btrfs_file_extent_item *item;
5364 struct btrfs_key key;
5365 int found_type;
5366 int i;
5368 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5369 return 0;
5371 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5372 btrfs_item_key_to_cpu(eb, &key, i);
5373 if (key.type != BTRFS_EXTENT_DATA_KEY)
5374 continue;
5375 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5376 found_type = btrfs_file_extent_type(eb, item);
5377 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5378 continue;
5379 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5380 continue;
5381 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5382 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5383 __exclude_logged_extent(log, key.objectid, key.offset);
5386 return 0;
5390 * btrfs_update_reserved_bytes - update the block_group and space info counters
5391 * @cache: The cache we are manipulating
5392 * @num_bytes: The number of bytes in question
5393 * @reserve: One of the reservation enums
5395 * This is called by the allocator when it reserves space, or by somebody who is
5396 * freeing space that was never actually used on disk. For example if you
5397 * reserve some space for a new leaf in transaction A and before transaction A
5398 * commits you free that leaf, you call this with reserve set to 0 in order to
5399 * clear the reservation.
5401 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5402 * ENOSPC accounting. For data we handle the reservation through clearing the
5403 * delalloc bits in the io_tree. We have to do this since we could end up
5404 * allocating less disk space for the amount of data we have reserved in the
5405 * case of compression.
5407 * If this is a reservation and the block group has become read only we cannot
5408 * make the reservation and return -EAGAIN, otherwise this function always
5409 * succeeds.
5411 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5412 u64 num_bytes, int reserve)
5414 struct btrfs_space_info *space_info = cache->space_info;
5415 int ret = 0;
5417 spin_lock(&space_info->lock);
5418 spin_lock(&cache->lock);
5419 if (reserve != RESERVE_FREE) {
5420 if (cache->ro) {
5421 ret = -EAGAIN;
5422 } else {
5423 cache->reserved += num_bytes;
5424 space_info->bytes_reserved += num_bytes;
5425 if (reserve == RESERVE_ALLOC) {
5426 trace_btrfs_space_reservation(cache->fs_info,
5427 "space_info", space_info->flags,
5428 num_bytes, 0);
5429 space_info->bytes_may_use -= num_bytes;
5432 } else {
5433 if (cache->ro)
5434 space_info->bytes_readonly += num_bytes;
5435 cache->reserved -= num_bytes;
5436 space_info->bytes_reserved -= num_bytes;
5438 spin_unlock(&cache->lock);
5439 spin_unlock(&space_info->lock);
5440 return ret;
5443 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5444 struct btrfs_root *root)
5446 struct btrfs_fs_info *fs_info = root->fs_info;
5447 struct btrfs_caching_control *next;
5448 struct btrfs_caching_control *caching_ctl;
5449 struct btrfs_block_group_cache *cache;
5450 struct btrfs_space_info *space_info;
5452 down_write(&fs_info->extent_commit_sem);
5454 list_for_each_entry_safe(caching_ctl, next,
5455 &fs_info->caching_block_groups, list) {
5456 cache = caching_ctl->block_group;
5457 if (block_group_cache_done(cache)) {
5458 cache->last_byte_to_unpin = (u64)-1;
5459 list_del_init(&caching_ctl->list);
5460 put_caching_control(caching_ctl);
5461 } else {
5462 cache->last_byte_to_unpin = caching_ctl->progress;
5466 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5467 fs_info->pinned_extents = &fs_info->freed_extents[1];
5468 else
5469 fs_info->pinned_extents = &fs_info->freed_extents[0];
5471 up_write(&fs_info->extent_commit_sem);
5473 list_for_each_entry_rcu(space_info, &fs_info->space_info, list)
5474 percpu_counter_set(&space_info->total_bytes_pinned, 0);
5476 update_global_block_rsv(fs_info);
5479 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
5481 struct btrfs_fs_info *fs_info = root->fs_info;
5482 struct btrfs_block_group_cache *cache = NULL;
5483 struct btrfs_space_info *space_info;
5484 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5485 u64 len;
5486 bool readonly;
5488 while (start <= end) {
5489 readonly = false;
5490 if (!cache ||
5491 start >= cache->key.objectid + cache->key.offset) {
5492 if (cache)
5493 btrfs_put_block_group(cache);
5494 cache = btrfs_lookup_block_group(fs_info, start);
5495 BUG_ON(!cache); /* Logic error */
5498 len = cache->key.objectid + cache->key.offset - start;
5499 len = min(len, end + 1 - start);
5501 if (start < cache->last_byte_to_unpin) {
5502 len = min(len, cache->last_byte_to_unpin - start);
5503 btrfs_add_free_space(cache, start, len);
5506 start += len;
5507 space_info = cache->space_info;
5509 spin_lock(&space_info->lock);
5510 spin_lock(&cache->lock);
5511 cache->pinned -= len;
5512 space_info->bytes_pinned -= len;
5513 if (cache->ro) {
5514 space_info->bytes_readonly += len;
5515 readonly = true;
5517 spin_unlock(&cache->lock);
5518 if (!readonly && global_rsv->space_info == space_info) {
5519 spin_lock(&global_rsv->lock);
5520 if (!global_rsv->full) {
5521 len = min(len, global_rsv->size -
5522 global_rsv->reserved);
5523 global_rsv->reserved += len;
5524 space_info->bytes_may_use += len;
5525 if (global_rsv->reserved >= global_rsv->size)
5526 global_rsv->full = 1;
5528 spin_unlock(&global_rsv->lock);
5530 spin_unlock(&space_info->lock);
5533 if (cache)
5534 btrfs_put_block_group(cache);
5535 return 0;
5538 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5539 struct btrfs_root *root)
5541 struct btrfs_fs_info *fs_info = root->fs_info;
5542 struct extent_io_tree *unpin;
5543 u64 start;
5544 u64 end;
5545 int ret;
5547 if (trans->aborted)
5548 return 0;
5550 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5551 unpin = &fs_info->freed_extents[1];
5552 else
5553 unpin = &fs_info->freed_extents[0];
5555 while (1) {
5556 ret = find_first_extent_bit(unpin, 0, &start, &end,
5557 EXTENT_DIRTY, NULL);
5558 if (ret)
5559 break;
5561 if (btrfs_test_opt(root, DISCARD))
5562 ret = btrfs_discard_extent(root, start,
5563 end + 1 - start, NULL);
5565 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5566 unpin_extent_range(root, start, end);
5567 cond_resched();
5570 return 0;
5573 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
5574 u64 owner, u64 root_objectid)
5576 struct btrfs_space_info *space_info;
5577 u64 flags;
5579 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5580 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
5581 flags = BTRFS_BLOCK_GROUP_SYSTEM;
5582 else
5583 flags = BTRFS_BLOCK_GROUP_METADATA;
5584 } else {
5585 flags = BTRFS_BLOCK_GROUP_DATA;
5588 space_info = __find_space_info(fs_info, flags);
5589 BUG_ON(!space_info); /* Logic bug */
5590 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
5594 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5595 struct btrfs_root *root,
5596 u64 bytenr, u64 num_bytes, u64 parent,
5597 u64 root_objectid, u64 owner_objectid,
5598 u64 owner_offset, int refs_to_drop,
5599 struct btrfs_delayed_extent_op *extent_op)
5601 struct btrfs_key key;
5602 struct btrfs_path *path;
5603 struct btrfs_fs_info *info = root->fs_info;
5604 struct btrfs_root *extent_root = info->extent_root;
5605 struct extent_buffer *leaf;
5606 struct btrfs_extent_item *ei;
5607 struct btrfs_extent_inline_ref *iref;
5608 int ret;
5609 int is_data;
5610 int extent_slot = 0;
5611 int found_extent = 0;
5612 int num_to_del = 1;
5613 u32 item_size;
5614 u64 refs;
5615 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
5616 SKINNY_METADATA);
5618 path = btrfs_alloc_path();
5619 if (!path)
5620 return -ENOMEM;
5622 path->reada = 1;
5623 path->leave_spinning = 1;
5625 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5626 BUG_ON(!is_data && refs_to_drop != 1);
5628 if (is_data)
5629 skinny_metadata = 0;
5631 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5632 bytenr, num_bytes, parent,
5633 root_objectid, owner_objectid,
5634 owner_offset);
5635 if (ret == 0) {
5636 extent_slot = path->slots[0];
5637 while (extent_slot >= 0) {
5638 btrfs_item_key_to_cpu(path->nodes[0], &key,
5639 extent_slot);
5640 if (key.objectid != bytenr)
5641 break;
5642 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5643 key.offset == num_bytes) {
5644 found_extent = 1;
5645 break;
5647 if (key.type == BTRFS_METADATA_ITEM_KEY &&
5648 key.offset == owner_objectid) {
5649 found_extent = 1;
5650 break;
5652 if (path->slots[0] - extent_slot > 5)
5653 break;
5654 extent_slot--;
5656 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5657 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5658 if (found_extent && item_size < sizeof(*ei))
5659 found_extent = 0;
5660 #endif
5661 if (!found_extent) {
5662 BUG_ON(iref);
5663 ret = remove_extent_backref(trans, extent_root, path,
5664 NULL, refs_to_drop,
5665 is_data);
5666 if (ret) {
5667 btrfs_abort_transaction(trans, extent_root, ret);
5668 goto out;
5670 btrfs_release_path(path);
5671 path->leave_spinning = 1;
5673 key.objectid = bytenr;
5674 key.type = BTRFS_EXTENT_ITEM_KEY;
5675 key.offset = num_bytes;
5677 if (!is_data && skinny_metadata) {
5678 key.type = BTRFS_METADATA_ITEM_KEY;
5679 key.offset = owner_objectid;
5682 ret = btrfs_search_slot(trans, extent_root,
5683 &key, path, -1, 1);
5684 if (ret > 0 && skinny_metadata && path->slots[0]) {
5686 * Couldn't find our skinny metadata item,
5687 * see if we have ye olde extent item.
5689 path->slots[0]--;
5690 btrfs_item_key_to_cpu(path->nodes[0], &key,
5691 path->slots[0]);
5692 if (key.objectid == bytenr &&
5693 key.type == BTRFS_EXTENT_ITEM_KEY &&
5694 key.offset == num_bytes)
5695 ret = 0;
5698 if (ret > 0 && skinny_metadata) {
5699 skinny_metadata = false;
5700 key.type = BTRFS_EXTENT_ITEM_KEY;
5701 key.offset = num_bytes;
5702 btrfs_release_path(path);
5703 ret = btrfs_search_slot(trans, extent_root,
5704 &key, path, -1, 1);
5707 if (ret) {
5708 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5709 ret, bytenr);
5710 if (ret > 0)
5711 btrfs_print_leaf(extent_root,
5712 path->nodes[0]);
5714 if (ret < 0) {
5715 btrfs_abort_transaction(trans, extent_root, ret);
5716 goto out;
5718 extent_slot = path->slots[0];
5720 } else if (ret == -ENOENT) {
5721 btrfs_print_leaf(extent_root, path->nodes[0]);
5722 WARN_ON(1);
5723 btrfs_err(info,
5724 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
5725 bytenr, parent, root_objectid, owner_objectid,
5726 owner_offset);
5727 } else {
5728 btrfs_abort_transaction(trans, extent_root, ret);
5729 goto out;
5732 leaf = path->nodes[0];
5733 item_size = btrfs_item_size_nr(leaf, extent_slot);
5734 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5735 if (item_size < sizeof(*ei)) {
5736 BUG_ON(found_extent || extent_slot != path->slots[0]);
5737 ret = convert_extent_item_v0(trans, extent_root, path,
5738 owner_objectid, 0);
5739 if (ret < 0) {
5740 btrfs_abort_transaction(trans, extent_root, ret);
5741 goto out;
5744 btrfs_release_path(path);
5745 path->leave_spinning = 1;
5747 key.objectid = bytenr;
5748 key.type = BTRFS_EXTENT_ITEM_KEY;
5749 key.offset = num_bytes;
5751 ret = btrfs_search_slot(trans, extent_root, &key, path,
5752 -1, 1);
5753 if (ret) {
5754 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5755 ret, bytenr);
5756 btrfs_print_leaf(extent_root, path->nodes[0]);
5758 if (ret < 0) {
5759 btrfs_abort_transaction(trans, extent_root, ret);
5760 goto out;
5763 extent_slot = path->slots[0];
5764 leaf = path->nodes[0];
5765 item_size = btrfs_item_size_nr(leaf, extent_slot);
5767 #endif
5768 BUG_ON(item_size < sizeof(*ei));
5769 ei = btrfs_item_ptr(leaf, extent_slot,
5770 struct btrfs_extent_item);
5771 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
5772 key.type == BTRFS_EXTENT_ITEM_KEY) {
5773 struct btrfs_tree_block_info *bi;
5774 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5775 bi = (struct btrfs_tree_block_info *)(ei + 1);
5776 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5779 refs = btrfs_extent_refs(leaf, ei);
5780 if (refs < refs_to_drop) {
5781 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
5782 "for bytenr %Lu\n", refs_to_drop, refs, bytenr);
5783 ret = -EINVAL;
5784 btrfs_abort_transaction(trans, extent_root, ret);
5785 goto out;
5787 refs -= refs_to_drop;
5789 if (refs > 0) {
5790 if (extent_op)
5791 __run_delayed_extent_op(extent_op, leaf, ei);
5793 * In the case of inline back ref, reference count will
5794 * be updated by remove_extent_backref
5796 if (iref) {
5797 BUG_ON(!found_extent);
5798 } else {
5799 btrfs_set_extent_refs(leaf, ei, refs);
5800 btrfs_mark_buffer_dirty(leaf);
5802 if (found_extent) {
5803 ret = remove_extent_backref(trans, extent_root, path,
5804 iref, refs_to_drop,
5805 is_data);
5806 if (ret) {
5807 btrfs_abort_transaction(trans, extent_root, ret);
5808 goto out;
5811 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
5812 root_objectid);
5813 } else {
5814 if (found_extent) {
5815 BUG_ON(is_data && refs_to_drop !=
5816 extent_data_ref_count(root, path, iref));
5817 if (iref) {
5818 BUG_ON(path->slots[0] != extent_slot);
5819 } else {
5820 BUG_ON(path->slots[0] != extent_slot + 1);
5821 path->slots[0] = extent_slot;
5822 num_to_del = 2;
5826 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5827 num_to_del);
5828 if (ret) {
5829 btrfs_abort_transaction(trans, extent_root, ret);
5830 goto out;
5832 btrfs_release_path(path);
5834 if (is_data) {
5835 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5836 if (ret) {
5837 btrfs_abort_transaction(trans, extent_root, ret);
5838 goto out;
5842 ret = update_block_group(root, bytenr, num_bytes, 0);
5843 if (ret) {
5844 btrfs_abort_transaction(trans, extent_root, ret);
5845 goto out;
5848 out:
5849 btrfs_free_path(path);
5850 return ret;
5854 * when we free an block, it is possible (and likely) that we free the last
5855 * delayed ref for that extent as well. This searches the delayed ref tree for
5856 * a given extent, and if there are no other delayed refs to be processed, it
5857 * removes it from the tree.
5859 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5860 struct btrfs_root *root, u64 bytenr)
5862 struct btrfs_delayed_ref_head *head;
5863 struct btrfs_delayed_ref_root *delayed_refs;
5864 struct btrfs_delayed_ref_node *ref;
5865 struct rb_node *node;
5866 int ret = 0;
5868 delayed_refs = &trans->transaction->delayed_refs;
5869 spin_lock(&delayed_refs->lock);
5870 head = btrfs_find_delayed_ref_head(trans, bytenr);
5871 if (!head)
5872 goto out;
5874 node = rb_prev(&head->node.rb_node);
5875 if (!node)
5876 goto out;
5878 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5880 /* there are still entries for this ref, we can't drop it */
5881 if (ref->bytenr == bytenr)
5882 goto out;
5884 if (head->extent_op) {
5885 if (!head->must_insert_reserved)
5886 goto out;
5887 btrfs_free_delayed_extent_op(head->extent_op);
5888 head->extent_op = NULL;
5892 * waiting for the lock here would deadlock. If someone else has it
5893 * locked they are already in the process of dropping it anyway
5895 if (!mutex_trylock(&head->mutex))
5896 goto out;
5899 * at this point we have a head with no other entries. Go
5900 * ahead and process it.
5902 head->node.in_tree = 0;
5903 rb_erase(&head->node.rb_node, &delayed_refs->root);
5905 delayed_refs->num_entries--;
5908 * we don't take a ref on the node because we're removing it from the
5909 * tree, so we just steal the ref the tree was holding.
5911 delayed_refs->num_heads--;
5912 if (list_empty(&head->cluster))
5913 delayed_refs->num_heads_ready--;
5915 list_del_init(&head->cluster);
5916 spin_unlock(&delayed_refs->lock);
5918 BUG_ON(head->extent_op);
5919 if (head->must_insert_reserved)
5920 ret = 1;
5922 mutex_unlock(&head->mutex);
5923 btrfs_put_delayed_ref(&head->node);
5924 return ret;
5925 out:
5926 spin_unlock(&delayed_refs->lock);
5927 return 0;
5930 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5931 struct btrfs_root *root,
5932 struct extent_buffer *buf,
5933 u64 parent, int last_ref)
5935 struct btrfs_block_group_cache *cache = NULL;
5936 int pin = 1;
5937 int ret;
5939 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5940 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5941 buf->start, buf->len,
5942 parent, root->root_key.objectid,
5943 btrfs_header_level(buf),
5944 BTRFS_DROP_DELAYED_REF, NULL, 0);
5945 BUG_ON(ret); /* -ENOMEM */
5948 if (!last_ref)
5949 return;
5951 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5953 if (btrfs_header_generation(buf) == trans->transid) {
5954 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5955 ret = check_ref_cleanup(trans, root, buf->start);
5956 if (!ret)
5957 goto out;
5960 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5961 pin_down_extent(root, cache, buf->start, buf->len, 1);
5962 goto out;
5965 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5967 btrfs_add_free_space(cache, buf->start, buf->len);
5968 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5969 pin = 0;
5971 out:
5972 if (pin)
5973 add_pinned_bytes(root->fs_info, buf->len,
5974 btrfs_header_level(buf),
5975 root->root_key.objectid);
5978 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5979 * anymore.
5981 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5982 btrfs_put_block_group(cache);
5985 /* Can return -ENOMEM */
5986 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5987 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5988 u64 owner, u64 offset, int for_cow)
5990 int ret;
5991 struct btrfs_fs_info *fs_info = root->fs_info;
5993 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
5996 * tree log blocks never actually go into the extent allocation
5997 * tree, just update pinning info and exit early.
5999 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6000 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6001 /* unlocks the pinned mutex */
6002 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6003 ret = 0;
6004 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6005 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6006 num_bytes,
6007 parent, root_objectid, (int)owner,
6008 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
6009 } else {
6010 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6011 num_bytes,
6012 parent, root_objectid, owner,
6013 offset, BTRFS_DROP_DELAYED_REF,
6014 NULL, for_cow);
6016 return ret;
6019 static u64 stripe_align(struct btrfs_root *root,
6020 struct btrfs_block_group_cache *cache,
6021 u64 val, u64 num_bytes)
6023 u64 ret = ALIGN(val, root->stripesize);
6024 return ret;
6028 * when we wait for progress in the block group caching, its because
6029 * our allocation attempt failed at least once. So, we must sleep
6030 * and let some progress happen before we try again.
6032 * This function will sleep at least once waiting for new free space to
6033 * show up, and then it will check the block group free space numbers
6034 * for our min num_bytes. Another option is to have it go ahead
6035 * and look in the rbtree for a free extent of a given size, but this
6036 * is a good start.
6038 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6039 * any of the information in this block group.
6041 static noinline void
6042 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6043 u64 num_bytes)
6045 struct btrfs_caching_control *caching_ctl;
6047 caching_ctl = get_caching_control(cache);
6048 if (!caching_ctl)
6049 return;
6051 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6052 (cache->free_space_ctl->free_space >= num_bytes));
6054 put_caching_control(caching_ctl);
6057 static noinline int
6058 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6060 struct btrfs_caching_control *caching_ctl;
6061 int ret = 0;
6063 caching_ctl = get_caching_control(cache);
6064 if (!caching_ctl)
6065 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6067 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6068 if (cache->cached == BTRFS_CACHE_ERROR)
6069 ret = -EIO;
6070 put_caching_control(caching_ctl);
6071 return ret;
6074 int __get_raid_index(u64 flags)
6076 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6077 return BTRFS_RAID_RAID10;
6078 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6079 return BTRFS_RAID_RAID1;
6080 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6081 return BTRFS_RAID_DUP;
6082 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6083 return BTRFS_RAID_RAID0;
6084 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6085 return BTRFS_RAID_RAID5;
6086 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6087 return BTRFS_RAID_RAID6;
6089 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6092 static int get_block_group_index(struct btrfs_block_group_cache *cache)
6094 return __get_raid_index(cache->flags);
6097 enum btrfs_loop_type {
6098 LOOP_CACHING_NOWAIT = 0,
6099 LOOP_CACHING_WAIT = 1,
6100 LOOP_ALLOC_CHUNK = 2,
6101 LOOP_NO_EMPTY_SIZE = 3,
6105 * walks the btree of allocated extents and find a hole of a given size.
6106 * The key ins is changed to record the hole:
6107 * ins->objectid == start position
6108 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6109 * ins->offset == the size of the hole.
6110 * Any available blocks before search_start are skipped.
6112 * If there is no suitable free space, we will record the max size of
6113 * the free space extent currently.
6115 static noinline int find_free_extent(struct btrfs_root *orig_root,
6116 u64 num_bytes, u64 empty_size,
6117 u64 hint_byte, struct btrfs_key *ins,
6118 u64 flags)
6120 int ret = 0;
6121 struct btrfs_root *root = orig_root->fs_info->extent_root;
6122 struct btrfs_free_cluster *last_ptr = NULL;
6123 struct btrfs_block_group_cache *block_group = NULL;
6124 struct btrfs_block_group_cache *used_block_group;
6125 u64 search_start = 0;
6126 u64 max_extent_size = 0;
6127 int empty_cluster = 2 * 1024 * 1024;
6128 struct btrfs_space_info *space_info;
6129 int loop = 0;
6130 int index = __get_raid_index(flags);
6131 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6132 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6133 bool found_uncached_bg = false;
6134 bool failed_cluster_refill = false;
6135 bool failed_alloc = false;
6136 bool use_cluster = true;
6137 bool have_caching_bg = false;
6139 WARN_ON(num_bytes < root->sectorsize);
6140 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
6141 ins->objectid = 0;
6142 ins->offset = 0;
6144 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6146 space_info = __find_space_info(root->fs_info, flags);
6147 if (!space_info) {
6148 btrfs_err(root->fs_info, "No space info for %llu", flags);
6149 return -ENOSPC;
6153 * If the space info is for both data and metadata it means we have a
6154 * small filesystem and we can't use the clustering stuff.
6156 if (btrfs_mixed_space_info(space_info))
6157 use_cluster = false;
6159 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6160 last_ptr = &root->fs_info->meta_alloc_cluster;
6161 if (!btrfs_test_opt(root, SSD))
6162 empty_cluster = 64 * 1024;
6165 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6166 btrfs_test_opt(root, SSD)) {
6167 last_ptr = &root->fs_info->data_alloc_cluster;
6170 if (last_ptr) {
6171 spin_lock(&last_ptr->lock);
6172 if (last_ptr->block_group)
6173 hint_byte = last_ptr->window_start;
6174 spin_unlock(&last_ptr->lock);
6177 search_start = max(search_start, first_logical_byte(root, 0));
6178 search_start = max(search_start, hint_byte);
6180 if (!last_ptr)
6181 empty_cluster = 0;
6183 if (search_start == hint_byte) {
6184 block_group = btrfs_lookup_block_group(root->fs_info,
6185 search_start);
6186 used_block_group = block_group;
6188 * we don't want to use the block group if it doesn't match our
6189 * allocation bits, or if its not cached.
6191 * However if we are re-searching with an ideal block group
6192 * picked out then we don't care that the block group is cached.
6194 if (block_group && block_group_bits(block_group, flags) &&
6195 block_group->cached != BTRFS_CACHE_NO) {
6196 down_read(&space_info->groups_sem);
6197 if (list_empty(&block_group->list) ||
6198 block_group->ro) {
6200 * someone is removing this block group,
6201 * we can't jump into the have_block_group
6202 * target because our list pointers are not
6203 * valid
6205 btrfs_put_block_group(block_group);
6206 up_read(&space_info->groups_sem);
6207 } else {
6208 index = get_block_group_index(block_group);
6209 goto have_block_group;
6211 } else if (block_group) {
6212 btrfs_put_block_group(block_group);
6215 search:
6216 have_caching_bg = false;
6217 down_read(&space_info->groups_sem);
6218 list_for_each_entry(block_group, &space_info->block_groups[index],
6219 list) {
6220 u64 offset;
6221 int cached;
6223 used_block_group = block_group;
6224 btrfs_get_block_group(block_group);
6225 search_start = block_group->key.objectid;
6228 * this can happen if we end up cycling through all the
6229 * raid types, but we want to make sure we only allocate
6230 * for the proper type.
6232 if (!block_group_bits(block_group, flags)) {
6233 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6234 BTRFS_BLOCK_GROUP_RAID1 |
6235 BTRFS_BLOCK_GROUP_RAID5 |
6236 BTRFS_BLOCK_GROUP_RAID6 |
6237 BTRFS_BLOCK_GROUP_RAID10;
6240 * if they asked for extra copies and this block group
6241 * doesn't provide them, bail. This does allow us to
6242 * fill raid0 from raid1.
6244 if ((flags & extra) && !(block_group->flags & extra))
6245 goto loop;
6248 have_block_group:
6249 cached = block_group_cache_done(block_group);
6250 if (unlikely(!cached)) {
6251 found_uncached_bg = true;
6252 ret = cache_block_group(block_group, 0);
6253 BUG_ON(ret < 0);
6254 ret = 0;
6257 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6258 goto loop;
6259 if (unlikely(block_group->ro))
6260 goto loop;
6263 * Ok we want to try and use the cluster allocator, so
6264 * lets look there
6266 if (last_ptr) {
6267 unsigned long aligned_cluster;
6269 * the refill lock keeps out other
6270 * people trying to start a new cluster
6272 spin_lock(&last_ptr->refill_lock);
6273 used_block_group = last_ptr->block_group;
6274 if (used_block_group != block_group &&
6275 (!used_block_group ||
6276 used_block_group->ro ||
6277 !block_group_bits(used_block_group, flags))) {
6278 used_block_group = block_group;
6279 goto refill_cluster;
6282 if (used_block_group != block_group)
6283 btrfs_get_block_group(used_block_group);
6285 offset = btrfs_alloc_from_cluster(used_block_group,
6286 last_ptr,
6287 num_bytes,
6288 used_block_group->key.objectid,
6289 &max_extent_size);
6290 if (offset) {
6291 /* we have a block, we're done */
6292 spin_unlock(&last_ptr->refill_lock);
6293 trace_btrfs_reserve_extent_cluster(root,
6294 block_group, search_start, num_bytes);
6295 goto checks;
6298 WARN_ON(last_ptr->block_group != used_block_group);
6299 if (used_block_group != block_group) {
6300 btrfs_put_block_group(used_block_group);
6301 used_block_group = block_group;
6303 refill_cluster:
6304 BUG_ON(used_block_group != block_group);
6305 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6306 * set up a new clusters, so lets just skip it
6307 * and let the allocator find whatever block
6308 * it can find. If we reach this point, we
6309 * will have tried the cluster allocator
6310 * plenty of times and not have found
6311 * anything, so we are likely way too
6312 * fragmented for the clustering stuff to find
6313 * anything.
6315 * However, if the cluster is taken from the
6316 * current block group, release the cluster
6317 * first, so that we stand a better chance of
6318 * succeeding in the unclustered
6319 * allocation. */
6320 if (loop >= LOOP_NO_EMPTY_SIZE &&
6321 last_ptr->block_group != block_group) {
6322 spin_unlock(&last_ptr->refill_lock);
6323 goto unclustered_alloc;
6327 * this cluster didn't work out, free it and
6328 * start over
6330 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6332 if (loop >= LOOP_NO_EMPTY_SIZE) {
6333 spin_unlock(&last_ptr->refill_lock);
6334 goto unclustered_alloc;
6337 aligned_cluster = max_t(unsigned long,
6338 empty_cluster + empty_size,
6339 block_group->full_stripe_len);
6341 /* allocate a cluster in this block group */
6342 ret = btrfs_find_space_cluster(root, block_group,
6343 last_ptr, search_start,
6344 num_bytes,
6345 aligned_cluster);
6346 if (ret == 0) {
6348 * now pull our allocation out of this
6349 * cluster
6351 offset = btrfs_alloc_from_cluster(block_group,
6352 last_ptr,
6353 num_bytes,
6354 search_start,
6355 &max_extent_size);
6356 if (offset) {
6357 /* we found one, proceed */
6358 spin_unlock(&last_ptr->refill_lock);
6359 trace_btrfs_reserve_extent_cluster(root,
6360 block_group, search_start,
6361 num_bytes);
6362 goto checks;
6364 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6365 && !failed_cluster_refill) {
6366 spin_unlock(&last_ptr->refill_lock);
6368 failed_cluster_refill = true;
6369 wait_block_group_cache_progress(block_group,
6370 num_bytes + empty_cluster + empty_size);
6371 goto have_block_group;
6375 * at this point we either didn't find a cluster
6376 * or we weren't able to allocate a block from our
6377 * cluster. Free the cluster we've been trying
6378 * to use, and go to the next block group
6380 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6381 spin_unlock(&last_ptr->refill_lock);
6382 goto loop;
6385 unclustered_alloc:
6386 spin_lock(&block_group->free_space_ctl->tree_lock);
6387 if (cached &&
6388 block_group->free_space_ctl->free_space <
6389 num_bytes + empty_cluster + empty_size) {
6390 if (block_group->free_space_ctl->free_space >
6391 max_extent_size)
6392 max_extent_size =
6393 block_group->free_space_ctl->free_space;
6394 spin_unlock(&block_group->free_space_ctl->tree_lock);
6395 goto loop;
6397 spin_unlock(&block_group->free_space_ctl->tree_lock);
6399 offset = btrfs_find_space_for_alloc(block_group, search_start,
6400 num_bytes, empty_size,
6401 &max_extent_size);
6403 * If we didn't find a chunk, and we haven't failed on this
6404 * block group before, and this block group is in the middle of
6405 * caching and we are ok with waiting, then go ahead and wait
6406 * for progress to be made, and set failed_alloc to true.
6408 * If failed_alloc is true then we've already waited on this
6409 * block group once and should move on to the next block group.
6411 if (!offset && !failed_alloc && !cached &&
6412 loop > LOOP_CACHING_NOWAIT) {
6413 wait_block_group_cache_progress(block_group,
6414 num_bytes + empty_size);
6415 failed_alloc = true;
6416 goto have_block_group;
6417 } else if (!offset) {
6418 if (!cached)
6419 have_caching_bg = true;
6420 goto loop;
6422 checks:
6423 search_start = stripe_align(root, used_block_group,
6424 offset, num_bytes);
6426 /* move on to the next group */
6427 if (search_start + num_bytes >
6428 used_block_group->key.objectid + used_block_group->key.offset) {
6429 btrfs_add_free_space(used_block_group, offset, num_bytes);
6430 goto loop;
6433 if (offset < search_start)
6434 btrfs_add_free_space(used_block_group, offset,
6435 search_start - offset);
6436 BUG_ON(offset > search_start);
6438 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
6439 alloc_type);
6440 if (ret == -EAGAIN) {
6441 btrfs_add_free_space(used_block_group, offset, num_bytes);
6442 goto loop;
6445 /* we are all good, lets return */
6446 ins->objectid = search_start;
6447 ins->offset = num_bytes;
6449 trace_btrfs_reserve_extent(orig_root, block_group,
6450 search_start, num_bytes);
6451 if (used_block_group != block_group)
6452 btrfs_put_block_group(used_block_group);
6453 btrfs_put_block_group(block_group);
6454 break;
6455 loop:
6456 failed_cluster_refill = false;
6457 failed_alloc = false;
6458 BUG_ON(index != get_block_group_index(block_group));
6459 if (used_block_group != block_group)
6460 btrfs_put_block_group(used_block_group);
6461 btrfs_put_block_group(block_group);
6463 up_read(&space_info->groups_sem);
6465 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
6466 goto search;
6468 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
6469 goto search;
6472 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6473 * caching kthreads as we move along
6474 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6475 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6476 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6477 * again
6479 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
6480 index = 0;
6481 loop++;
6482 if (loop == LOOP_ALLOC_CHUNK) {
6483 struct btrfs_trans_handle *trans;
6485 trans = btrfs_join_transaction(root);
6486 if (IS_ERR(trans)) {
6487 ret = PTR_ERR(trans);
6488 goto out;
6491 ret = do_chunk_alloc(trans, root, flags,
6492 CHUNK_ALLOC_FORCE);
6494 * Do not bail out on ENOSPC since we
6495 * can do more things.
6497 if (ret < 0 && ret != -ENOSPC)
6498 btrfs_abort_transaction(trans,
6499 root, ret);
6500 else
6501 ret = 0;
6502 btrfs_end_transaction(trans, root);
6503 if (ret)
6504 goto out;
6507 if (loop == LOOP_NO_EMPTY_SIZE) {
6508 empty_size = 0;
6509 empty_cluster = 0;
6512 goto search;
6513 } else if (!ins->objectid) {
6514 ret = -ENOSPC;
6515 } else if (ins->objectid) {
6516 ret = 0;
6518 out:
6519 if (ret == -ENOSPC)
6520 ins->offset = max_extent_size;
6521 return ret;
6524 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6525 int dump_block_groups)
6527 struct btrfs_block_group_cache *cache;
6528 int index = 0;
6530 spin_lock(&info->lock);
6531 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
6532 info->flags,
6533 info->total_bytes - info->bytes_used - info->bytes_pinned -
6534 info->bytes_reserved - info->bytes_readonly,
6535 (info->full) ? "" : "not ");
6536 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
6537 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6538 info->total_bytes, info->bytes_used, info->bytes_pinned,
6539 info->bytes_reserved, info->bytes_may_use,
6540 info->bytes_readonly);
6541 spin_unlock(&info->lock);
6543 if (!dump_block_groups)
6544 return;
6546 down_read(&info->groups_sem);
6547 again:
6548 list_for_each_entry(cache, &info->block_groups[index], list) {
6549 spin_lock(&cache->lock);
6550 printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
6551 cache->key.objectid, cache->key.offset,
6552 btrfs_block_group_used(&cache->item), cache->pinned,
6553 cache->reserved, cache->ro ? "[readonly]" : "");
6554 btrfs_dump_free_space(cache, bytes);
6555 spin_unlock(&cache->lock);
6557 if (++index < BTRFS_NR_RAID_TYPES)
6558 goto again;
6559 up_read(&info->groups_sem);
6562 int btrfs_reserve_extent(struct btrfs_root *root,
6563 u64 num_bytes, u64 min_alloc_size,
6564 u64 empty_size, u64 hint_byte,
6565 struct btrfs_key *ins, int is_data)
6567 bool final_tried = false;
6568 u64 flags;
6569 int ret;
6571 flags = btrfs_get_alloc_profile(root, is_data);
6572 again:
6573 WARN_ON(num_bytes < root->sectorsize);
6574 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
6575 flags);
6577 if (ret == -ENOSPC) {
6578 if (!final_tried && ins->offset) {
6579 num_bytes = min(num_bytes >> 1, ins->offset);
6580 num_bytes = round_down(num_bytes, root->sectorsize);
6581 num_bytes = max(num_bytes, min_alloc_size);
6582 if (num_bytes == min_alloc_size)
6583 final_tried = true;
6584 goto again;
6585 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6586 struct btrfs_space_info *sinfo;
6588 sinfo = __find_space_info(root->fs_info, flags);
6589 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
6590 flags, num_bytes);
6591 if (sinfo)
6592 dump_space_info(sinfo, num_bytes, 1);
6596 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
6598 return ret;
6601 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6602 u64 start, u64 len, int pin)
6604 struct btrfs_block_group_cache *cache;
6605 int ret = 0;
6607 cache = btrfs_lookup_block_group(root->fs_info, start);
6608 if (!cache) {
6609 btrfs_err(root->fs_info, "Unable to find block group for %llu",
6610 start);
6611 return -ENOSPC;
6614 if (btrfs_test_opt(root, DISCARD))
6615 ret = btrfs_discard_extent(root, start, len, NULL);
6617 if (pin)
6618 pin_down_extent(root, cache, start, len, 1);
6619 else {
6620 btrfs_add_free_space(cache, start, len);
6621 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6623 btrfs_put_block_group(cache);
6625 trace_btrfs_reserved_extent_free(root, start, len);
6627 return ret;
6630 int btrfs_free_reserved_extent(struct btrfs_root *root,
6631 u64 start, u64 len)
6633 return __btrfs_free_reserved_extent(root, start, len, 0);
6636 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6637 u64 start, u64 len)
6639 return __btrfs_free_reserved_extent(root, start, len, 1);
6642 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6643 struct btrfs_root *root,
6644 u64 parent, u64 root_objectid,
6645 u64 flags, u64 owner, u64 offset,
6646 struct btrfs_key *ins, int ref_mod)
6648 int ret;
6649 struct btrfs_fs_info *fs_info = root->fs_info;
6650 struct btrfs_extent_item *extent_item;
6651 struct btrfs_extent_inline_ref *iref;
6652 struct btrfs_path *path;
6653 struct extent_buffer *leaf;
6654 int type;
6655 u32 size;
6657 if (parent > 0)
6658 type = BTRFS_SHARED_DATA_REF_KEY;
6659 else
6660 type = BTRFS_EXTENT_DATA_REF_KEY;
6662 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6664 path = btrfs_alloc_path();
6665 if (!path)
6666 return -ENOMEM;
6668 path->leave_spinning = 1;
6669 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6670 ins, size);
6671 if (ret) {
6672 btrfs_free_path(path);
6673 return ret;
6676 leaf = path->nodes[0];
6677 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6678 struct btrfs_extent_item);
6679 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6680 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6681 btrfs_set_extent_flags(leaf, extent_item,
6682 flags | BTRFS_EXTENT_FLAG_DATA);
6684 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6685 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6686 if (parent > 0) {
6687 struct btrfs_shared_data_ref *ref;
6688 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6689 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6690 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6691 } else {
6692 struct btrfs_extent_data_ref *ref;
6693 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6694 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6695 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6696 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6697 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6700 btrfs_mark_buffer_dirty(path->nodes[0]);
6701 btrfs_free_path(path);
6703 ret = update_block_group(root, ins->objectid, ins->offset, 1);
6704 if (ret) { /* -ENOENT, logic error */
6705 btrfs_err(fs_info, "update block group failed for %llu %llu",
6706 ins->objectid, ins->offset);
6707 BUG();
6709 return ret;
6712 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6713 struct btrfs_root *root,
6714 u64 parent, u64 root_objectid,
6715 u64 flags, struct btrfs_disk_key *key,
6716 int level, struct btrfs_key *ins)
6718 int ret;
6719 struct btrfs_fs_info *fs_info = root->fs_info;
6720 struct btrfs_extent_item *extent_item;
6721 struct btrfs_tree_block_info *block_info;
6722 struct btrfs_extent_inline_ref *iref;
6723 struct btrfs_path *path;
6724 struct extent_buffer *leaf;
6725 u32 size = sizeof(*extent_item) + sizeof(*iref);
6726 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6727 SKINNY_METADATA);
6729 if (!skinny_metadata)
6730 size += sizeof(*block_info);
6732 path = btrfs_alloc_path();
6733 if (!path)
6734 return -ENOMEM;
6736 path->leave_spinning = 1;
6737 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6738 ins, size);
6739 if (ret) {
6740 btrfs_free_path(path);
6741 return ret;
6744 leaf = path->nodes[0];
6745 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6746 struct btrfs_extent_item);
6747 btrfs_set_extent_refs(leaf, extent_item, 1);
6748 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6749 btrfs_set_extent_flags(leaf, extent_item,
6750 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6752 if (skinny_metadata) {
6753 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6754 } else {
6755 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6756 btrfs_set_tree_block_key(leaf, block_info, key);
6757 btrfs_set_tree_block_level(leaf, block_info, level);
6758 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6761 if (parent > 0) {
6762 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6763 btrfs_set_extent_inline_ref_type(leaf, iref,
6764 BTRFS_SHARED_BLOCK_REF_KEY);
6765 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6766 } else {
6767 btrfs_set_extent_inline_ref_type(leaf, iref,
6768 BTRFS_TREE_BLOCK_REF_KEY);
6769 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6772 btrfs_mark_buffer_dirty(leaf);
6773 btrfs_free_path(path);
6775 ret = update_block_group(root, ins->objectid, root->leafsize, 1);
6776 if (ret) { /* -ENOENT, logic error */
6777 btrfs_err(fs_info, "update block group failed for %llu %llu",
6778 ins->objectid, ins->offset);
6779 BUG();
6781 return ret;
6784 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6785 struct btrfs_root *root,
6786 u64 root_objectid, u64 owner,
6787 u64 offset, struct btrfs_key *ins)
6789 int ret;
6791 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6793 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6794 ins->offset, 0,
6795 root_objectid, owner, offset,
6796 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6797 return ret;
6801 * this is used by the tree logging recovery code. It records that
6802 * an extent has been allocated and makes sure to clear the free
6803 * space cache bits as well
6805 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6806 struct btrfs_root *root,
6807 u64 root_objectid, u64 owner, u64 offset,
6808 struct btrfs_key *ins)
6810 int ret;
6811 struct btrfs_block_group_cache *block_group;
6814 * Mixed block groups will exclude before processing the log so we only
6815 * need to do the exlude dance if this fs isn't mixed.
6817 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
6818 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
6819 if (ret)
6820 return ret;
6823 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6824 if (!block_group)
6825 return -EINVAL;
6827 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6828 RESERVE_ALLOC_NO_ACCOUNT);
6829 BUG_ON(ret); /* logic error */
6830 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6831 0, owner, offset, ins, 1);
6832 btrfs_put_block_group(block_group);
6833 return ret;
6836 static struct extent_buffer *
6837 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6838 u64 bytenr, u32 blocksize, int level)
6840 struct extent_buffer *buf;
6842 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6843 if (!buf)
6844 return ERR_PTR(-ENOMEM);
6845 btrfs_set_header_generation(buf, trans->transid);
6846 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6847 btrfs_tree_lock(buf);
6848 clean_tree_block(trans, root, buf);
6849 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6851 btrfs_set_lock_blocking(buf);
6852 btrfs_set_buffer_uptodate(buf);
6854 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6856 * we allow two log transactions at a time, use different
6857 * EXENT bit to differentiate dirty pages.
6859 if (root->log_transid % 2 == 0)
6860 set_extent_dirty(&root->dirty_log_pages, buf->start,
6861 buf->start + buf->len - 1, GFP_NOFS);
6862 else
6863 set_extent_new(&root->dirty_log_pages, buf->start,
6864 buf->start + buf->len - 1, GFP_NOFS);
6865 } else {
6866 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6867 buf->start + buf->len - 1, GFP_NOFS);
6869 trans->blocks_used++;
6870 /* this returns a buffer locked for blocking */
6871 return buf;
6874 static struct btrfs_block_rsv *
6875 use_block_rsv(struct btrfs_trans_handle *trans,
6876 struct btrfs_root *root, u32 blocksize)
6878 struct btrfs_block_rsv *block_rsv;
6879 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6880 int ret;
6881 bool global_updated = false;
6883 block_rsv = get_block_rsv(trans, root);
6885 if (unlikely(block_rsv->size == 0))
6886 goto try_reserve;
6887 again:
6888 ret = block_rsv_use_bytes(block_rsv, blocksize);
6889 if (!ret)
6890 return block_rsv;
6892 if (block_rsv->failfast)
6893 return ERR_PTR(ret);
6895 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
6896 global_updated = true;
6897 update_global_block_rsv(root->fs_info);
6898 goto again;
6901 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6902 static DEFINE_RATELIMIT_STATE(_rs,
6903 DEFAULT_RATELIMIT_INTERVAL * 10,
6904 /*DEFAULT_RATELIMIT_BURST*/ 1);
6905 if (__ratelimit(&_rs))
6906 WARN(1, KERN_DEBUG
6907 "btrfs: block rsv returned %d\n", ret);
6909 try_reserve:
6910 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6911 BTRFS_RESERVE_NO_FLUSH);
6912 if (!ret)
6913 return block_rsv;
6915 * If we couldn't reserve metadata bytes try and use some from
6916 * the global reserve if its space type is the same as the global
6917 * reservation.
6919 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
6920 block_rsv->space_info == global_rsv->space_info) {
6921 ret = block_rsv_use_bytes(global_rsv, blocksize);
6922 if (!ret)
6923 return global_rsv;
6925 return ERR_PTR(ret);
6928 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6929 struct btrfs_block_rsv *block_rsv, u32 blocksize)
6931 block_rsv_add_bytes(block_rsv, blocksize, 0);
6932 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6936 * finds a free extent and does all the dirty work required for allocation
6937 * returns the key for the extent through ins, and a tree buffer for
6938 * the first block of the extent through buf.
6940 * returns the tree buffer or NULL.
6942 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6943 struct btrfs_root *root, u32 blocksize,
6944 u64 parent, u64 root_objectid,
6945 struct btrfs_disk_key *key, int level,
6946 u64 hint, u64 empty_size)
6948 struct btrfs_key ins;
6949 struct btrfs_block_rsv *block_rsv;
6950 struct extent_buffer *buf;
6951 u64 flags = 0;
6952 int ret;
6953 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6954 SKINNY_METADATA);
6956 block_rsv = use_block_rsv(trans, root, blocksize);
6957 if (IS_ERR(block_rsv))
6958 return ERR_CAST(block_rsv);
6960 ret = btrfs_reserve_extent(root, blocksize, blocksize,
6961 empty_size, hint, &ins, 0);
6962 if (ret) {
6963 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
6964 return ERR_PTR(ret);
6967 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6968 blocksize, level);
6969 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
6971 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6972 if (parent == 0)
6973 parent = ins.objectid;
6974 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6975 } else
6976 BUG_ON(parent > 0);
6978 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6979 struct btrfs_delayed_extent_op *extent_op;
6980 extent_op = btrfs_alloc_delayed_extent_op();
6981 BUG_ON(!extent_op); /* -ENOMEM */
6982 if (key)
6983 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6984 else
6985 memset(&extent_op->key, 0, sizeof(extent_op->key));
6986 extent_op->flags_to_set = flags;
6987 if (skinny_metadata)
6988 extent_op->update_key = 0;
6989 else
6990 extent_op->update_key = 1;
6991 extent_op->update_flags = 1;
6992 extent_op->is_data = 0;
6993 extent_op->level = level;
6995 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6996 ins.objectid,
6997 ins.offset, parent, root_objectid,
6998 level, BTRFS_ADD_DELAYED_EXTENT,
6999 extent_op, 0);
7000 BUG_ON(ret); /* -ENOMEM */
7002 return buf;
7005 struct walk_control {
7006 u64 refs[BTRFS_MAX_LEVEL];
7007 u64 flags[BTRFS_MAX_LEVEL];
7008 struct btrfs_key update_progress;
7009 int stage;
7010 int level;
7011 int shared_level;
7012 int update_ref;
7013 int keep_locks;
7014 int reada_slot;
7015 int reada_count;
7016 int for_reloc;
7019 #define DROP_REFERENCE 1
7020 #define UPDATE_BACKREF 2
7022 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7023 struct btrfs_root *root,
7024 struct walk_control *wc,
7025 struct btrfs_path *path)
7027 u64 bytenr;
7028 u64 generation;
7029 u64 refs;
7030 u64 flags;
7031 u32 nritems;
7032 u32 blocksize;
7033 struct btrfs_key key;
7034 struct extent_buffer *eb;
7035 int ret;
7036 int slot;
7037 int nread = 0;
7039 if (path->slots[wc->level] < wc->reada_slot) {
7040 wc->reada_count = wc->reada_count * 2 / 3;
7041 wc->reada_count = max(wc->reada_count, 2);
7042 } else {
7043 wc->reada_count = wc->reada_count * 3 / 2;
7044 wc->reada_count = min_t(int, wc->reada_count,
7045 BTRFS_NODEPTRS_PER_BLOCK(root));
7048 eb = path->nodes[wc->level];
7049 nritems = btrfs_header_nritems(eb);
7050 blocksize = btrfs_level_size(root, wc->level - 1);
7052 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7053 if (nread >= wc->reada_count)
7054 break;
7056 cond_resched();
7057 bytenr = btrfs_node_blockptr(eb, slot);
7058 generation = btrfs_node_ptr_generation(eb, slot);
7060 if (slot == path->slots[wc->level])
7061 goto reada;
7063 if (wc->stage == UPDATE_BACKREF &&
7064 generation <= root->root_key.offset)
7065 continue;
7067 /* We don't lock the tree block, it's OK to be racy here */
7068 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7069 wc->level - 1, 1, &refs,
7070 &flags);
7071 /* We don't care about errors in readahead. */
7072 if (ret < 0)
7073 continue;
7074 BUG_ON(refs == 0);
7076 if (wc->stage == DROP_REFERENCE) {
7077 if (refs == 1)
7078 goto reada;
7080 if (wc->level == 1 &&
7081 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7082 continue;
7083 if (!wc->update_ref ||
7084 generation <= root->root_key.offset)
7085 continue;
7086 btrfs_node_key_to_cpu(eb, &key, slot);
7087 ret = btrfs_comp_cpu_keys(&key,
7088 &wc->update_progress);
7089 if (ret < 0)
7090 continue;
7091 } else {
7092 if (wc->level == 1 &&
7093 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7094 continue;
7096 reada:
7097 ret = readahead_tree_block(root, bytenr, blocksize,
7098 generation);
7099 if (ret)
7100 break;
7101 nread++;
7103 wc->reada_slot = slot;
7107 * helper to process tree block while walking down the tree.
7109 * when wc->stage == UPDATE_BACKREF, this function updates
7110 * back refs for pointers in the block.
7112 * NOTE: return value 1 means we should stop walking down.
7114 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7115 struct btrfs_root *root,
7116 struct btrfs_path *path,
7117 struct walk_control *wc, int lookup_info)
7119 int level = wc->level;
7120 struct extent_buffer *eb = path->nodes[level];
7121 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7122 int ret;
7124 if (wc->stage == UPDATE_BACKREF &&
7125 btrfs_header_owner(eb) != root->root_key.objectid)
7126 return 1;
7129 * when reference count of tree block is 1, it won't increase
7130 * again. once full backref flag is set, we never clear it.
7132 if (lookup_info &&
7133 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
7134 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
7135 BUG_ON(!path->locks[level]);
7136 ret = btrfs_lookup_extent_info(trans, root,
7137 eb->start, level, 1,
7138 &wc->refs[level],
7139 &wc->flags[level]);
7140 BUG_ON(ret == -ENOMEM);
7141 if (ret)
7142 return ret;
7143 BUG_ON(wc->refs[level] == 0);
7146 if (wc->stage == DROP_REFERENCE) {
7147 if (wc->refs[level] > 1)
7148 return 1;
7150 if (path->locks[level] && !wc->keep_locks) {
7151 btrfs_tree_unlock_rw(eb, path->locks[level]);
7152 path->locks[level] = 0;
7154 return 0;
7157 /* wc->stage == UPDATE_BACKREF */
7158 if (!(wc->flags[level] & flag)) {
7159 BUG_ON(!path->locks[level]);
7160 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
7161 BUG_ON(ret); /* -ENOMEM */
7162 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
7163 BUG_ON(ret); /* -ENOMEM */
7164 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
7165 eb->len, flag,
7166 btrfs_header_level(eb), 0);
7167 BUG_ON(ret); /* -ENOMEM */
7168 wc->flags[level] |= flag;
7172 * the block is shared by multiple trees, so it's not good to
7173 * keep the tree lock
7175 if (path->locks[level] && level > 0) {
7176 btrfs_tree_unlock_rw(eb, path->locks[level]);
7177 path->locks[level] = 0;
7179 return 0;
7183 * helper to process tree block pointer.
7185 * when wc->stage == DROP_REFERENCE, this function checks
7186 * reference count of the block pointed to. if the block
7187 * is shared and we need update back refs for the subtree
7188 * rooted at the block, this function changes wc->stage to
7189 * UPDATE_BACKREF. if the block is shared and there is no
7190 * need to update back, this function drops the reference
7191 * to the block.
7193 * NOTE: return value 1 means we should stop walking down.
7195 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
7196 struct btrfs_root *root,
7197 struct btrfs_path *path,
7198 struct walk_control *wc, int *lookup_info)
7200 u64 bytenr;
7201 u64 generation;
7202 u64 parent;
7203 u32 blocksize;
7204 struct btrfs_key key;
7205 struct extent_buffer *next;
7206 int level = wc->level;
7207 int reada = 0;
7208 int ret = 0;
7210 generation = btrfs_node_ptr_generation(path->nodes[level],
7211 path->slots[level]);
7213 * if the lower level block was created before the snapshot
7214 * was created, we know there is no need to update back refs
7215 * for the subtree
7217 if (wc->stage == UPDATE_BACKREF &&
7218 generation <= root->root_key.offset) {
7219 *lookup_info = 1;
7220 return 1;
7223 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
7224 blocksize = btrfs_level_size(root, level - 1);
7226 next = btrfs_find_tree_block(root, bytenr, blocksize);
7227 if (!next) {
7228 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
7229 if (!next)
7230 return -ENOMEM;
7231 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
7232 level - 1);
7233 reada = 1;
7235 btrfs_tree_lock(next);
7236 btrfs_set_lock_blocking(next);
7238 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
7239 &wc->refs[level - 1],
7240 &wc->flags[level - 1]);
7241 if (ret < 0) {
7242 btrfs_tree_unlock(next);
7243 return ret;
7246 if (unlikely(wc->refs[level - 1] == 0)) {
7247 btrfs_err(root->fs_info, "Missing references.");
7248 BUG();
7250 *lookup_info = 0;
7252 if (wc->stage == DROP_REFERENCE) {
7253 if (wc->refs[level - 1] > 1) {
7254 if (level == 1 &&
7255 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7256 goto skip;
7258 if (!wc->update_ref ||
7259 generation <= root->root_key.offset)
7260 goto skip;
7262 btrfs_node_key_to_cpu(path->nodes[level], &key,
7263 path->slots[level]);
7264 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
7265 if (ret < 0)
7266 goto skip;
7268 wc->stage = UPDATE_BACKREF;
7269 wc->shared_level = level - 1;
7271 } else {
7272 if (level == 1 &&
7273 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7274 goto skip;
7277 if (!btrfs_buffer_uptodate(next, generation, 0)) {
7278 btrfs_tree_unlock(next);
7279 free_extent_buffer(next);
7280 next = NULL;
7281 *lookup_info = 1;
7284 if (!next) {
7285 if (reada && level == 1)
7286 reada_walk_down(trans, root, wc, path);
7287 next = read_tree_block(root, bytenr, blocksize, generation);
7288 if (!next || !extent_buffer_uptodate(next)) {
7289 free_extent_buffer(next);
7290 return -EIO;
7292 btrfs_tree_lock(next);
7293 btrfs_set_lock_blocking(next);
7296 level--;
7297 BUG_ON(level != btrfs_header_level(next));
7298 path->nodes[level] = next;
7299 path->slots[level] = 0;
7300 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7301 wc->level = level;
7302 if (wc->level == 1)
7303 wc->reada_slot = 0;
7304 return 0;
7305 skip:
7306 wc->refs[level - 1] = 0;
7307 wc->flags[level - 1] = 0;
7308 if (wc->stage == DROP_REFERENCE) {
7309 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7310 parent = path->nodes[level]->start;
7311 } else {
7312 BUG_ON(root->root_key.objectid !=
7313 btrfs_header_owner(path->nodes[level]));
7314 parent = 0;
7317 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
7318 root->root_key.objectid, level - 1, 0, 0);
7319 BUG_ON(ret); /* -ENOMEM */
7321 btrfs_tree_unlock(next);
7322 free_extent_buffer(next);
7323 *lookup_info = 1;
7324 return 1;
7328 * helper to process tree block while walking up the tree.
7330 * when wc->stage == DROP_REFERENCE, this function drops
7331 * reference count on the block.
7333 * when wc->stage == UPDATE_BACKREF, this function changes
7334 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7335 * to UPDATE_BACKREF previously while processing the block.
7337 * NOTE: return value 1 means we should stop walking up.
7339 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
7340 struct btrfs_root *root,
7341 struct btrfs_path *path,
7342 struct walk_control *wc)
7344 int ret;
7345 int level = wc->level;
7346 struct extent_buffer *eb = path->nodes[level];
7347 u64 parent = 0;
7349 if (wc->stage == UPDATE_BACKREF) {
7350 BUG_ON(wc->shared_level < level);
7351 if (level < wc->shared_level)
7352 goto out;
7354 ret = find_next_key(path, level + 1, &wc->update_progress);
7355 if (ret > 0)
7356 wc->update_ref = 0;
7358 wc->stage = DROP_REFERENCE;
7359 wc->shared_level = -1;
7360 path->slots[level] = 0;
7363 * check reference count again if the block isn't locked.
7364 * we should start walking down the tree again if reference
7365 * count is one.
7367 if (!path->locks[level]) {
7368 BUG_ON(level == 0);
7369 btrfs_tree_lock(eb);
7370 btrfs_set_lock_blocking(eb);
7371 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7373 ret = btrfs_lookup_extent_info(trans, root,
7374 eb->start, level, 1,
7375 &wc->refs[level],
7376 &wc->flags[level]);
7377 if (ret < 0) {
7378 btrfs_tree_unlock_rw(eb, path->locks[level]);
7379 path->locks[level] = 0;
7380 return ret;
7382 BUG_ON(wc->refs[level] == 0);
7383 if (wc->refs[level] == 1) {
7384 btrfs_tree_unlock_rw(eb, path->locks[level]);
7385 path->locks[level] = 0;
7386 return 1;
7391 /* wc->stage == DROP_REFERENCE */
7392 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
7394 if (wc->refs[level] == 1) {
7395 if (level == 0) {
7396 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7397 ret = btrfs_dec_ref(trans, root, eb, 1,
7398 wc->for_reloc);
7399 else
7400 ret = btrfs_dec_ref(trans, root, eb, 0,
7401 wc->for_reloc);
7402 BUG_ON(ret); /* -ENOMEM */
7404 /* make block locked assertion in clean_tree_block happy */
7405 if (!path->locks[level] &&
7406 btrfs_header_generation(eb) == trans->transid) {
7407 btrfs_tree_lock(eb);
7408 btrfs_set_lock_blocking(eb);
7409 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7411 clean_tree_block(trans, root, eb);
7414 if (eb == root->node) {
7415 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7416 parent = eb->start;
7417 else
7418 BUG_ON(root->root_key.objectid !=
7419 btrfs_header_owner(eb));
7420 } else {
7421 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7422 parent = path->nodes[level + 1]->start;
7423 else
7424 BUG_ON(root->root_key.objectid !=
7425 btrfs_header_owner(path->nodes[level + 1]));
7428 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
7429 out:
7430 wc->refs[level] = 0;
7431 wc->flags[level] = 0;
7432 return 0;
7435 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
7436 struct btrfs_root *root,
7437 struct btrfs_path *path,
7438 struct walk_control *wc)
7440 int level = wc->level;
7441 int lookup_info = 1;
7442 int ret;
7444 while (level >= 0) {
7445 ret = walk_down_proc(trans, root, path, wc, lookup_info);
7446 if (ret > 0)
7447 break;
7449 if (level == 0)
7450 break;
7452 if (path->slots[level] >=
7453 btrfs_header_nritems(path->nodes[level]))
7454 break;
7456 ret = do_walk_down(trans, root, path, wc, &lookup_info);
7457 if (ret > 0) {
7458 path->slots[level]++;
7459 continue;
7460 } else if (ret < 0)
7461 return ret;
7462 level = wc->level;
7464 return 0;
7467 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
7468 struct btrfs_root *root,
7469 struct btrfs_path *path,
7470 struct walk_control *wc, int max_level)
7472 int level = wc->level;
7473 int ret;
7475 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
7476 while (level < max_level && path->nodes[level]) {
7477 wc->level = level;
7478 if (path->slots[level] + 1 <
7479 btrfs_header_nritems(path->nodes[level])) {
7480 path->slots[level]++;
7481 return 0;
7482 } else {
7483 ret = walk_up_proc(trans, root, path, wc);
7484 if (ret > 0)
7485 return 0;
7487 if (path->locks[level]) {
7488 btrfs_tree_unlock_rw(path->nodes[level],
7489 path->locks[level]);
7490 path->locks[level] = 0;
7492 free_extent_buffer(path->nodes[level]);
7493 path->nodes[level] = NULL;
7494 level++;
7497 return 1;
7501 * drop a subvolume tree.
7503 * this function traverses the tree freeing any blocks that only
7504 * referenced by the tree.
7506 * when a shared tree block is found. this function decreases its
7507 * reference count by one. if update_ref is true, this function
7508 * also make sure backrefs for the shared block and all lower level
7509 * blocks are properly updated.
7511 * If called with for_reloc == 0, may exit early with -EAGAIN
7513 int btrfs_drop_snapshot(struct btrfs_root *root,
7514 struct btrfs_block_rsv *block_rsv, int update_ref,
7515 int for_reloc)
7517 struct btrfs_path *path;
7518 struct btrfs_trans_handle *trans;
7519 struct btrfs_root *tree_root = root->fs_info->tree_root;
7520 struct btrfs_root_item *root_item = &root->root_item;
7521 struct walk_control *wc;
7522 struct btrfs_key key;
7523 int err = 0;
7524 int ret;
7525 int level;
7526 bool root_dropped = false;
7528 path = btrfs_alloc_path();
7529 if (!path) {
7530 err = -ENOMEM;
7531 goto out;
7534 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7535 if (!wc) {
7536 btrfs_free_path(path);
7537 err = -ENOMEM;
7538 goto out;
7541 trans = btrfs_start_transaction(tree_root, 0);
7542 if (IS_ERR(trans)) {
7543 err = PTR_ERR(trans);
7544 goto out_free;
7547 if (block_rsv)
7548 trans->block_rsv = block_rsv;
7550 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
7551 level = btrfs_header_level(root->node);
7552 path->nodes[level] = btrfs_lock_root_node(root);
7553 btrfs_set_lock_blocking(path->nodes[level]);
7554 path->slots[level] = 0;
7555 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7556 memset(&wc->update_progress, 0,
7557 sizeof(wc->update_progress));
7558 } else {
7559 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
7560 memcpy(&wc->update_progress, &key,
7561 sizeof(wc->update_progress));
7563 level = root_item->drop_level;
7564 BUG_ON(level == 0);
7565 path->lowest_level = level;
7566 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7567 path->lowest_level = 0;
7568 if (ret < 0) {
7569 err = ret;
7570 goto out_end_trans;
7572 WARN_ON(ret > 0);
7575 * unlock our path, this is safe because only this
7576 * function is allowed to delete this snapshot
7578 btrfs_unlock_up_safe(path, 0);
7580 level = btrfs_header_level(root->node);
7581 while (1) {
7582 btrfs_tree_lock(path->nodes[level]);
7583 btrfs_set_lock_blocking(path->nodes[level]);
7584 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7586 ret = btrfs_lookup_extent_info(trans, root,
7587 path->nodes[level]->start,
7588 level, 1, &wc->refs[level],
7589 &wc->flags[level]);
7590 if (ret < 0) {
7591 err = ret;
7592 goto out_end_trans;
7594 BUG_ON(wc->refs[level] == 0);
7596 if (level == root_item->drop_level)
7597 break;
7599 btrfs_tree_unlock(path->nodes[level]);
7600 path->locks[level] = 0;
7601 WARN_ON(wc->refs[level] != 1);
7602 level--;
7606 wc->level = level;
7607 wc->shared_level = -1;
7608 wc->stage = DROP_REFERENCE;
7609 wc->update_ref = update_ref;
7610 wc->keep_locks = 0;
7611 wc->for_reloc = for_reloc;
7612 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7614 while (1) {
7616 ret = walk_down_tree(trans, root, path, wc);
7617 if (ret < 0) {
7618 err = ret;
7619 break;
7622 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7623 if (ret < 0) {
7624 err = ret;
7625 break;
7628 if (ret > 0) {
7629 BUG_ON(wc->stage != DROP_REFERENCE);
7630 break;
7633 if (wc->stage == DROP_REFERENCE) {
7634 level = wc->level;
7635 btrfs_node_key(path->nodes[level],
7636 &root_item->drop_progress,
7637 path->slots[level]);
7638 root_item->drop_level = level;
7641 BUG_ON(wc->level == 0);
7642 if (btrfs_should_end_transaction(trans, tree_root) ||
7643 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
7644 ret = btrfs_update_root(trans, tree_root,
7645 &root->root_key,
7646 root_item);
7647 if (ret) {
7648 btrfs_abort_transaction(trans, tree_root, ret);
7649 err = ret;
7650 goto out_end_trans;
7653 btrfs_end_transaction_throttle(trans, tree_root);
7654 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
7655 pr_debug("btrfs: drop snapshot early exit\n");
7656 err = -EAGAIN;
7657 goto out_free;
7660 trans = btrfs_start_transaction(tree_root, 0);
7661 if (IS_ERR(trans)) {
7662 err = PTR_ERR(trans);
7663 goto out_free;
7665 if (block_rsv)
7666 trans->block_rsv = block_rsv;
7669 btrfs_release_path(path);
7670 if (err)
7671 goto out_end_trans;
7673 ret = btrfs_del_root(trans, tree_root, &root->root_key);
7674 if (ret) {
7675 btrfs_abort_transaction(trans, tree_root, ret);
7676 goto out_end_trans;
7679 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7680 ret = btrfs_find_root(tree_root, &root->root_key, path,
7681 NULL, NULL);
7682 if (ret < 0) {
7683 btrfs_abort_transaction(trans, tree_root, ret);
7684 err = ret;
7685 goto out_end_trans;
7686 } else if (ret > 0) {
7687 /* if we fail to delete the orphan item this time
7688 * around, it'll get picked up the next time.
7690 * The most common failure here is just -ENOENT.
7692 btrfs_del_orphan_item(trans, tree_root,
7693 root->root_key.objectid);
7697 if (root->in_radix) {
7698 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
7699 } else {
7700 free_extent_buffer(root->node);
7701 free_extent_buffer(root->commit_root);
7702 btrfs_put_fs_root(root);
7704 root_dropped = true;
7705 out_end_trans:
7706 btrfs_end_transaction_throttle(trans, tree_root);
7707 out_free:
7708 kfree(wc);
7709 btrfs_free_path(path);
7710 out:
7712 * So if we need to stop dropping the snapshot for whatever reason we
7713 * need to make sure to add it back to the dead root list so that we
7714 * keep trying to do the work later. This also cleans up roots if we
7715 * don't have it in the radix (like when we recover after a power fail
7716 * or unmount) so we don't leak memory.
7718 if (!for_reloc && root_dropped == false)
7719 btrfs_add_dead_root(root);
7720 if (err && err != -EAGAIN)
7721 btrfs_std_error(root->fs_info, err);
7722 return err;
7726 * drop subtree rooted at tree block 'node'.
7728 * NOTE: this function will unlock and release tree block 'node'
7729 * only used by relocation code
7731 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7732 struct btrfs_root *root,
7733 struct extent_buffer *node,
7734 struct extent_buffer *parent)
7736 struct btrfs_path *path;
7737 struct walk_control *wc;
7738 int level;
7739 int parent_level;
7740 int ret = 0;
7741 int wret;
7743 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7745 path = btrfs_alloc_path();
7746 if (!path)
7747 return -ENOMEM;
7749 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7750 if (!wc) {
7751 btrfs_free_path(path);
7752 return -ENOMEM;
7755 btrfs_assert_tree_locked(parent);
7756 parent_level = btrfs_header_level(parent);
7757 extent_buffer_get(parent);
7758 path->nodes[parent_level] = parent;
7759 path->slots[parent_level] = btrfs_header_nritems(parent);
7761 btrfs_assert_tree_locked(node);
7762 level = btrfs_header_level(node);
7763 path->nodes[level] = node;
7764 path->slots[level] = 0;
7765 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7767 wc->refs[parent_level] = 1;
7768 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7769 wc->level = level;
7770 wc->shared_level = -1;
7771 wc->stage = DROP_REFERENCE;
7772 wc->update_ref = 0;
7773 wc->keep_locks = 1;
7774 wc->for_reloc = 1;
7775 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7777 while (1) {
7778 wret = walk_down_tree(trans, root, path, wc);
7779 if (wret < 0) {
7780 ret = wret;
7781 break;
7784 wret = walk_up_tree(trans, root, path, wc, parent_level);
7785 if (wret < 0)
7786 ret = wret;
7787 if (wret != 0)
7788 break;
7791 kfree(wc);
7792 btrfs_free_path(path);
7793 return ret;
7796 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7798 u64 num_devices;
7799 u64 stripped;
7802 * if restripe for this chunk_type is on pick target profile and
7803 * return, otherwise do the usual balance
7805 stripped = get_restripe_target(root->fs_info, flags);
7806 if (stripped)
7807 return extended_to_chunk(stripped);
7810 * we add in the count of missing devices because we want
7811 * to make sure that any RAID levels on a degraded FS
7812 * continue to be honored.
7814 num_devices = root->fs_info->fs_devices->rw_devices +
7815 root->fs_info->fs_devices->missing_devices;
7817 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7818 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
7819 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7821 if (num_devices == 1) {
7822 stripped |= BTRFS_BLOCK_GROUP_DUP;
7823 stripped = flags & ~stripped;
7825 /* turn raid0 into single device chunks */
7826 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7827 return stripped;
7829 /* turn mirroring into duplication */
7830 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7831 BTRFS_BLOCK_GROUP_RAID10))
7832 return stripped | BTRFS_BLOCK_GROUP_DUP;
7833 } else {
7834 /* they already had raid on here, just return */
7835 if (flags & stripped)
7836 return flags;
7838 stripped |= BTRFS_BLOCK_GROUP_DUP;
7839 stripped = flags & ~stripped;
7841 /* switch duplicated blocks with raid1 */
7842 if (flags & BTRFS_BLOCK_GROUP_DUP)
7843 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7845 /* this is drive concat, leave it alone */
7848 return flags;
7851 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7853 struct btrfs_space_info *sinfo = cache->space_info;
7854 u64 num_bytes;
7855 u64 min_allocable_bytes;
7856 int ret = -ENOSPC;
7860 * We need some metadata space and system metadata space for
7861 * allocating chunks in some corner cases until we force to set
7862 * it to be readonly.
7864 if ((sinfo->flags &
7865 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7866 !force)
7867 min_allocable_bytes = 1 * 1024 * 1024;
7868 else
7869 min_allocable_bytes = 0;
7871 spin_lock(&sinfo->lock);
7872 spin_lock(&cache->lock);
7874 if (cache->ro) {
7875 ret = 0;
7876 goto out;
7879 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7880 cache->bytes_super - btrfs_block_group_used(&cache->item);
7882 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7883 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7884 min_allocable_bytes <= sinfo->total_bytes) {
7885 sinfo->bytes_readonly += num_bytes;
7886 cache->ro = 1;
7887 ret = 0;
7889 out:
7890 spin_unlock(&cache->lock);
7891 spin_unlock(&sinfo->lock);
7892 return ret;
7895 int btrfs_set_block_group_ro(struct btrfs_root *root,
7896 struct btrfs_block_group_cache *cache)
7899 struct btrfs_trans_handle *trans;
7900 u64 alloc_flags;
7901 int ret;
7903 BUG_ON(cache->ro);
7905 trans = btrfs_join_transaction(root);
7906 if (IS_ERR(trans))
7907 return PTR_ERR(trans);
7909 alloc_flags = update_block_group_flags(root, cache->flags);
7910 if (alloc_flags != cache->flags) {
7911 ret = do_chunk_alloc(trans, root, alloc_flags,
7912 CHUNK_ALLOC_FORCE);
7913 if (ret < 0)
7914 goto out;
7917 ret = set_block_group_ro(cache, 0);
7918 if (!ret)
7919 goto out;
7920 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7921 ret = do_chunk_alloc(trans, root, alloc_flags,
7922 CHUNK_ALLOC_FORCE);
7923 if (ret < 0)
7924 goto out;
7925 ret = set_block_group_ro(cache, 0);
7926 out:
7927 btrfs_end_transaction(trans, root);
7928 return ret;
7931 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7932 struct btrfs_root *root, u64 type)
7934 u64 alloc_flags = get_alloc_profile(root, type);
7935 return do_chunk_alloc(trans, root, alloc_flags,
7936 CHUNK_ALLOC_FORCE);
7940 * helper to account the unused space of all the readonly block group in the
7941 * list. takes mirrors into account.
7943 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7945 struct btrfs_block_group_cache *block_group;
7946 u64 free_bytes = 0;
7947 int factor;
7949 list_for_each_entry(block_group, groups_list, list) {
7950 spin_lock(&block_group->lock);
7952 if (!block_group->ro) {
7953 spin_unlock(&block_group->lock);
7954 continue;
7957 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7958 BTRFS_BLOCK_GROUP_RAID10 |
7959 BTRFS_BLOCK_GROUP_DUP))
7960 factor = 2;
7961 else
7962 factor = 1;
7964 free_bytes += (block_group->key.offset -
7965 btrfs_block_group_used(&block_group->item)) *
7966 factor;
7968 spin_unlock(&block_group->lock);
7971 return free_bytes;
7975 * helper to account the unused space of all the readonly block group in the
7976 * space_info. takes mirrors into account.
7978 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7980 int i;
7981 u64 free_bytes = 0;
7983 spin_lock(&sinfo->lock);
7985 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7986 if (!list_empty(&sinfo->block_groups[i]))
7987 free_bytes += __btrfs_get_ro_block_group_free_space(
7988 &sinfo->block_groups[i]);
7990 spin_unlock(&sinfo->lock);
7992 return free_bytes;
7995 void btrfs_set_block_group_rw(struct btrfs_root *root,
7996 struct btrfs_block_group_cache *cache)
7998 struct btrfs_space_info *sinfo = cache->space_info;
7999 u64 num_bytes;
8001 BUG_ON(!cache->ro);
8003 spin_lock(&sinfo->lock);
8004 spin_lock(&cache->lock);
8005 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8006 cache->bytes_super - btrfs_block_group_used(&cache->item);
8007 sinfo->bytes_readonly -= num_bytes;
8008 cache->ro = 0;
8009 spin_unlock(&cache->lock);
8010 spin_unlock(&sinfo->lock);
8014 * checks to see if its even possible to relocate this block group.
8016 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8017 * ok to go ahead and try.
8019 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8021 struct btrfs_block_group_cache *block_group;
8022 struct btrfs_space_info *space_info;
8023 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8024 struct btrfs_device *device;
8025 struct btrfs_trans_handle *trans;
8026 u64 min_free;
8027 u64 dev_min = 1;
8028 u64 dev_nr = 0;
8029 u64 target;
8030 int index;
8031 int full = 0;
8032 int ret = 0;
8034 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8036 /* odd, couldn't find the block group, leave it alone */
8037 if (!block_group)
8038 return -1;
8040 min_free = btrfs_block_group_used(&block_group->item);
8042 /* no bytes used, we're good */
8043 if (!min_free)
8044 goto out;
8046 space_info = block_group->space_info;
8047 spin_lock(&space_info->lock);
8049 full = space_info->full;
8052 * if this is the last block group we have in this space, we can't
8053 * relocate it unless we're able to allocate a new chunk below.
8055 * Otherwise, we need to make sure we have room in the space to handle
8056 * all of the extents from this block group. If we can, we're good
8058 if ((space_info->total_bytes != block_group->key.offset) &&
8059 (space_info->bytes_used + space_info->bytes_reserved +
8060 space_info->bytes_pinned + space_info->bytes_readonly +
8061 min_free < space_info->total_bytes)) {
8062 spin_unlock(&space_info->lock);
8063 goto out;
8065 spin_unlock(&space_info->lock);
8068 * ok we don't have enough space, but maybe we have free space on our
8069 * devices to allocate new chunks for relocation, so loop through our
8070 * alloc devices and guess if we have enough space. if this block
8071 * group is going to be restriped, run checks against the target
8072 * profile instead of the current one.
8074 ret = -1;
8077 * index:
8078 * 0: raid10
8079 * 1: raid1
8080 * 2: dup
8081 * 3: raid0
8082 * 4: single
8084 target = get_restripe_target(root->fs_info, block_group->flags);
8085 if (target) {
8086 index = __get_raid_index(extended_to_chunk(target));
8087 } else {
8089 * this is just a balance, so if we were marked as full
8090 * we know there is no space for a new chunk
8092 if (full)
8093 goto out;
8095 index = get_block_group_index(block_group);
8098 if (index == BTRFS_RAID_RAID10) {
8099 dev_min = 4;
8100 /* Divide by 2 */
8101 min_free >>= 1;
8102 } else if (index == BTRFS_RAID_RAID1) {
8103 dev_min = 2;
8104 } else if (index == BTRFS_RAID_DUP) {
8105 /* Multiply by 2 */
8106 min_free <<= 1;
8107 } else if (index == BTRFS_RAID_RAID0) {
8108 dev_min = fs_devices->rw_devices;
8109 do_div(min_free, dev_min);
8112 /* We need to do this so that we can look at pending chunks */
8113 trans = btrfs_join_transaction(root);
8114 if (IS_ERR(trans)) {
8115 ret = PTR_ERR(trans);
8116 goto out;
8119 mutex_lock(&root->fs_info->chunk_mutex);
8120 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8121 u64 dev_offset;
8124 * check to make sure we can actually find a chunk with enough
8125 * space to fit our block group in.
8127 if (device->total_bytes > device->bytes_used + min_free &&
8128 !device->is_tgtdev_for_dev_replace) {
8129 ret = find_free_dev_extent(trans, device, min_free,
8130 &dev_offset, NULL);
8131 if (!ret)
8132 dev_nr++;
8134 if (dev_nr >= dev_min)
8135 break;
8137 ret = -1;
8140 mutex_unlock(&root->fs_info->chunk_mutex);
8141 btrfs_end_transaction(trans, root);
8142 out:
8143 btrfs_put_block_group(block_group);
8144 return ret;
8147 static int find_first_block_group(struct btrfs_root *root,
8148 struct btrfs_path *path, struct btrfs_key *key)
8150 int ret = 0;
8151 struct btrfs_key found_key;
8152 struct extent_buffer *leaf;
8153 int slot;
8155 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8156 if (ret < 0)
8157 goto out;
8159 while (1) {
8160 slot = path->slots[0];
8161 leaf = path->nodes[0];
8162 if (slot >= btrfs_header_nritems(leaf)) {
8163 ret = btrfs_next_leaf(root, path);
8164 if (ret == 0)
8165 continue;
8166 if (ret < 0)
8167 goto out;
8168 break;
8170 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8172 if (found_key.objectid >= key->objectid &&
8173 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8174 ret = 0;
8175 goto out;
8177 path->slots[0]++;
8179 out:
8180 return ret;
8183 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8185 struct btrfs_block_group_cache *block_group;
8186 u64 last = 0;
8188 while (1) {
8189 struct inode *inode;
8191 block_group = btrfs_lookup_first_block_group(info, last);
8192 while (block_group) {
8193 spin_lock(&block_group->lock);
8194 if (block_group->iref)
8195 break;
8196 spin_unlock(&block_group->lock);
8197 block_group = next_block_group(info->tree_root,
8198 block_group);
8200 if (!block_group) {
8201 if (last == 0)
8202 break;
8203 last = 0;
8204 continue;
8207 inode = block_group->inode;
8208 block_group->iref = 0;
8209 block_group->inode = NULL;
8210 spin_unlock(&block_group->lock);
8211 iput(inode);
8212 last = block_group->key.objectid + block_group->key.offset;
8213 btrfs_put_block_group(block_group);
8217 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8219 struct btrfs_block_group_cache *block_group;
8220 struct btrfs_space_info *space_info;
8221 struct btrfs_caching_control *caching_ctl;
8222 struct rb_node *n;
8224 down_write(&info->extent_commit_sem);
8225 while (!list_empty(&info->caching_block_groups)) {
8226 caching_ctl = list_entry(info->caching_block_groups.next,
8227 struct btrfs_caching_control, list);
8228 list_del(&caching_ctl->list);
8229 put_caching_control(caching_ctl);
8231 up_write(&info->extent_commit_sem);
8233 spin_lock(&info->block_group_cache_lock);
8234 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8235 block_group = rb_entry(n, struct btrfs_block_group_cache,
8236 cache_node);
8237 rb_erase(&block_group->cache_node,
8238 &info->block_group_cache_tree);
8239 spin_unlock(&info->block_group_cache_lock);
8241 down_write(&block_group->space_info->groups_sem);
8242 list_del(&block_group->list);
8243 up_write(&block_group->space_info->groups_sem);
8245 if (block_group->cached == BTRFS_CACHE_STARTED)
8246 wait_block_group_cache_done(block_group);
8249 * We haven't cached this block group, which means we could
8250 * possibly have excluded extents on this block group.
8252 if (block_group->cached == BTRFS_CACHE_NO ||
8253 block_group->cached == BTRFS_CACHE_ERROR)
8254 free_excluded_extents(info->extent_root, block_group);
8256 btrfs_remove_free_space_cache(block_group);
8257 btrfs_put_block_group(block_group);
8259 spin_lock(&info->block_group_cache_lock);
8261 spin_unlock(&info->block_group_cache_lock);
8263 /* now that all the block groups are freed, go through and
8264 * free all the space_info structs. This is only called during
8265 * the final stages of unmount, and so we know nobody is
8266 * using them. We call synchronize_rcu() once before we start,
8267 * just to be on the safe side.
8269 synchronize_rcu();
8271 release_global_block_rsv(info);
8273 while(!list_empty(&info->space_info)) {
8274 space_info = list_entry(info->space_info.next,
8275 struct btrfs_space_info,
8276 list);
8277 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
8278 if (space_info->bytes_pinned > 0 ||
8279 space_info->bytes_reserved > 0 ||
8280 space_info->bytes_may_use > 0) {
8281 WARN_ON(1);
8282 dump_space_info(space_info, 0, 0);
8285 percpu_counter_destroy(&space_info->total_bytes_pinned);
8286 list_del(&space_info->list);
8287 kfree(space_info);
8289 return 0;
8292 static void __link_block_group(struct btrfs_space_info *space_info,
8293 struct btrfs_block_group_cache *cache)
8295 int index = get_block_group_index(cache);
8297 down_write(&space_info->groups_sem);
8298 list_add_tail(&cache->list, &space_info->block_groups[index]);
8299 up_write(&space_info->groups_sem);
8302 int btrfs_read_block_groups(struct btrfs_root *root)
8304 struct btrfs_path *path;
8305 int ret;
8306 struct btrfs_block_group_cache *cache;
8307 struct btrfs_fs_info *info = root->fs_info;
8308 struct btrfs_space_info *space_info;
8309 struct btrfs_key key;
8310 struct btrfs_key found_key;
8311 struct extent_buffer *leaf;
8312 int need_clear = 0;
8313 u64 cache_gen;
8315 root = info->extent_root;
8316 key.objectid = 0;
8317 key.offset = 0;
8318 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8319 path = btrfs_alloc_path();
8320 if (!path)
8321 return -ENOMEM;
8322 path->reada = 1;
8324 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
8325 if (btrfs_test_opt(root, SPACE_CACHE) &&
8326 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
8327 need_clear = 1;
8328 if (btrfs_test_opt(root, CLEAR_CACHE))
8329 need_clear = 1;
8331 while (1) {
8332 ret = find_first_block_group(root, path, &key);
8333 if (ret > 0)
8334 break;
8335 if (ret != 0)
8336 goto error;
8337 leaf = path->nodes[0];
8338 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8339 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8340 if (!cache) {
8341 ret = -ENOMEM;
8342 goto error;
8344 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8345 GFP_NOFS);
8346 if (!cache->free_space_ctl) {
8347 kfree(cache);
8348 ret = -ENOMEM;
8349 goto error;
8352 atomic_set(&cache->count, 1);
8353 spin_lock_init(&cache->lock);
8354 cache->fs_info = info;
8355 INIT_LIST_HEAD(&cache->list);
8356 INIT_LIST_HEAD(&cache->cluster_list);
8358 if (need_clear) {
8360 * When we mount with old space cache, we need to
8361 * set BTRFS_DC_CLEAR and set dirty flag.
8363 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
8364 * truncate the old free space cache inode and
8365 * setup a new one.
8366 * b) Setting 'dirty flag' makes sure that we flush
8367 * the new space cache info onto disk.
8369 cache->disk_cache_state = BTRFS_DC_CLEAR;
8370 if (btrfs_test_opt(root, SPACE_CACHE))
8371 cache->dirty = 1;
8374 read_extent_buffer(leaf, &cache->item,
8375 btrfs_item_ptr_offset(leaf, path->slots[0]),
8376 sizeof(cache->item));
8377 memcpy(&cache->key, &found_key, sizeof(found_key));
8379 key.objectid = found_key.objectid + found_key.offset;
8380 btrfs_release_path(path);
8381 cache->flags = btrfs_block_group_flags(&cache->item);
8382 cache->sectorsize = root->sectorsize;
8383 cache->full_stripe_len = btrfs_full_stripe_len(root,
8384 &root->fs_info->mapping_tree,
8385 found_key.objectid);
8386 btrfs_init_free_space_ctl(cache);
8389 * We need to exclude the super stripes now so that the space
8390 * info has super bytes accounted for, otherwise we'll think
8391 * we have more space than we actually do.
8393 ret = exclude_super_stripes(root, cache);
8394 if (ret) {
8396 * We may have excluded something, so call this just in
8397 * case.
8399 free_excluded_extents(root, cache);
8400 kfree(cache->free_space_ctl);
8401 kfree(cache);
8402 goto error;
8406 * check for two cases, either we are full, and therefore
8407 * don't need to bother with the caching work since we won't
8408 * find any space, or we are empty, and we can just add all
8409 * the space in and be done with it. This saves us _alot_ of
8410 * time, particularly in the full case.
8412 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8413 cache->last_byte_to_unpin = (u64)-1;
8414 cache->cached = BTRFS_CACHE_FINISHED;
8415 free_excluded_extents(root, cache);
8416 } else if (btrfs_block_group_used(&cache->item) == 0) {
8417 cache->last_byte_to_unpin = (u64)-1;
8418 cache->cached = BTRFS_CACHE_FINISHED;
8419 add_new_free_space(cache, root->fs_info,
8420 found_key.objectid,
8421 found_key.objectid +
8422 found_key.offset);
8423 free_excluded_extents(root, cache);
8426 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8427 if (ret) {
8428 btrfs_remove_free_space_cache(cache);
8429 btrfs_put_block_group(cache);
8430 goto error;
8433 ret = update_space_info(info, cache->flags, found_key.offset,
8434 btrfs_block_group_used(&cache->item),
8435 &space_info);
8436 if (ret) {
8437 btrfs_remove_free_space_cache(cache);
8438 spin_lock(&info->block_group_cache_lock);
8439 rb_erase(&cache->cache_node,
8440 &info->block_group_cache_tree);
8441 spin_unlock(&info->block_group_cache_lock);
8442 btrfs_put_block_group(cache);
8443 goto error;
8446 cache->space_info = space_info;
8447 spin_lock(&cache->space_info->lock);
8448 cache->space_info->bytes_readonly += cache->bytes_super;
8449 spin_unlock(&cache->space_info->lock);
8451 __link_block_group(space_info, cache);
8453 set_avail_alloc_bits(root->fs_info, cache->flags);
8454 if (btrfs_chunk_readonly(root, cache->key.objectid))
8455 set_block_group_ro(cache, 1);
8458 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8459 if (!(get_alloc_profile(root, space_info->flags) &
8460 (BTRFS_BLOCK_GROUP_RAID10 |
8461 BTRFS_BLOCK_GROUP_RAID1 |
8462 BTRFS_BLOCK_GROUP_RAID5 |
8463 BTRFS_BLOCK_GROUP_RAID6 |
8464 BTRFS_BLOCK_GROUP_DUP)))
8465 continue;
8467 * avoid allocating from un-mirrored block group if there are
8468 * mirrored block groups.
8470 list_for_each_entry(cache,
8471 &space_info->block_groups[BTRFS_RAID_RAID0],
8472 list)
8473 set_block_group_ro(cache, 1);
8474 list_for_each_entry(cache,
8475 &space_info->block_groups[BTRFS_RAID_SINGLE],
8476 list)
8477 set_block_group_ro(cache, 1);
8480 init_global_block_rsv(info);
8481 ret = 0;
8482 error:
8483 btrfs_free_path(path);
8484 return ret;
8487 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
8488 struct btrfs_root *root)
8490 struct btrfs_block_group_cache *block_group, *tmp;
8491 struct btrfs_root *extent_root = root->fs_info->extent_root;
8492 struct btrfs_block_group_item item;
8493 struct btrfs_key key;
8494 int ret = 0;
8496 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
8497 new_bg_list) {
8498 list_del_init(&block_group->new_bg_list);
8500 if (ret)
8501 continue;
8503 spin_lock(&block_group->lock);
8504 memcpy(&item, &block_group->item, sizeof(item));
8505 memcpy(&key, &block_group->key, sizeof(key));
8506 spin_unlock(&block_group->lock);
8508 ret = btrfs_insert_item(trans, extent_root, &key, &item,
8509 sizeof(item));
8510 if (ret)
8511 btrfs_abort_transaction(trans, extent_root, ret);
8512 ret = btrfs_finish_chunk_alloc(trans, extent_root,
8513 key.objectid, key.offset);
8514 if (ret)
8515 btrfs_abort_transaction(trans, extent_root, ret);
8519 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8520 struct btrfs_root *root, u64 bytes_used,
8521 u64 type, u64 chunk_objectid, u64 chunk_offset,
8522 u64 size)
8524 int ret;
8525 struct btrfs_root *extent_root;
8526 struct btrfs_block_group_cache *cache;
8528 extent_root = root->fs_info->extent_root;
8530 root->fs_info->last_trans_log_full_commit = trans->transid;
8532 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8533 if (!cache)
8534 return -ENOMEM;
8535 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8536 GFP_NOFS);
8537 if (!cache->free_space_ctl) {
8538 kfree(cache);
8539 return -ENOMEM;
8542 cache->key.objectid = chunk_offset;
8543 cache->key.offset = size;
8544 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8545 cache->sectorsize = root->sectorsize;
8546 cache->fs_info = root->fs_info;
8547 cache->full_stripe_len = btrfs_full_stripe_len(root,
8548 &root->fs_info->mapping_tree,
8549 chunk_offset);
8551 atomic_set(&cache->count, 1);
8552 spin_lock_init(&cache->lock);
8553 INIT_LIST_HEAD(&cache->list);
8554 INIT_LIST_HEAD(&cache->cluster_list);
8555 INIT_LIST_HEAD(&cache->new_bg_list);
8557 btrfs_init_free_space_ctl(cache);
8559 btrfs_set_block_group_used(&cache->item, bytes_used);
8560 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8561 cache->flags = type;
8562 btrfs_set_block_group_flags(&cache->item, type);
8564 cache->last_byte_to_unpin = (u64)-1;
8565 cache->cached = BTRFS_CACHE_FINISHED;
8566 ret = exclude_super_stripes(root, cache);
8567 if (ret) {
8569 * We may have excluded something, so call this just in
8570 * case.
8572 free_excluded_extents(root, cache);
8573 kfree(cache->free_space_ctl);
8574 kfree(cache);
8575 return ret;
8578 add_new_free_space(cache, root->fs_info, chunk_offset,
8579 chunk_offset + size);
8581 free_excluded_extents(root, cache);
8583 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8584 if (ret) {
8585 btrfs_remove_free_space_cache(cache);
8586 btrfs_put_block_group(cache);
8587 return ret;
8590 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8591 &cache->space_info);
8592 if (ret) {
8593 btrfs_remove_free_space_cache(cache);
8594 spin_lock(&root->fs_info->block_group_cache_lock);
8595 rb_erase(&cache->cache_node,
8596 &root->fs_info->block_group_cache_tree);
8597 spin_unlock(&root->fs_info->block_group_cache_lock);
8598 btrfs_put_block_group(cache);
8599 return ret;
8601 update_global_block_rsv(root->fs_info);
8603 spin_lock(&cache->space_info->lock);
8604 cache->space_info->bytes_readonly += cache->bytes_super;
8605 spin_unlock(&cache->space_info->lock);
8607 __link_block_group(cache->space_info, cache);
8609 list_add_tail(&cache->new_bg_list, &trans->new_bgs);
8611 set_avail_alloc_bits(extent_root->fs_info, type);
8613 return 0;
8616 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
8618 u64 extra_flags = chunk_to_extended(flags) &
8619 BTRFS_EXTENDED_PROFILE_MASK;
8621 write_seqlock(&fs_info->profiles_lock);
8622 if (flags & BTRFS_BLOCK_GROUP_DATA)
8623 fs_info->avail_data_alloc_bits &= ~extra_flags;
8624 if (flags & BTRFS_BLOCK_GROUP_METADATA)
8625 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
8626 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
8627 fs_info->avail_system_alloc_bits &= ~extra_flags;
8628 write_sequnlock(&fs_info->profiles_lock);
8631 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8632 struct btrfs_root *root, u64 group_start)
8634 struct btrfs_path *path;
8635 struct btrfs_block_group_cache *block_group;
8636 struct btrfs_free_cluster *cluster;
8637 struct btrfs_root *tree_root = root->fs_info->tree_root;
8638 struct btrfs_key key;
8639 struct inode *inode;
8640 int ret;
8641 int index;
8642 int factor;
8644 root = root->fs_info->extent_root;
8646 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8647 BUG_ON(!block_group);
8648 BUG_ON(!block_group->ro);
8651 * Free the reserved super bytes from this block group before
8652 * remove it.
8654 free_excluded_extents(root, block_group);
8656 memcpy(&key, &block_group->key, sizeof(key));
8657 index = get_block_group_index(block_group);
8658 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8659 BTRFS_BLOCK_GROUP_RAID1 |
8660 BTRFS_BLOCK_GROUP_RAID10))
8661 factor = 2;
8662 else
8663 factor = 1;
8665 /* make sure this block group isn't part of an allocation cluster */
8666 cluster = &root->fs_info->data_alloc_cluster;
8667 spin_lock(&cluster->refill_lock);
8668 btrfs_return_cluster_to_free_space(block_group, cluster);
8669 spin_unlock(&cluster->refill_lock);
8672 * make sure this block group isn't part of a metadata
8673 * allocation cluster
8675 cluster = &root->fs_info->meta_alloc_cluster;
8676 spin_lock(&cluster->refill_lock);
8677 btrfs_return_cluster_to_free_space(block_group, cluster);
8678 spin_unlock(&cluster->refill_lock);
8680 path = btrfs_alloc_path();
8681 if (!path) {
8682 ret = -ENOMEM;
8683 goto out;
8686 inode = lookup_free_space_inode(tree_root, block_group, path);
8687 if (!IS_ERR(inode)) {
8688 ret = btrfs_orphan_add(trans, inode);
8689 if (ret) {
8690 btrfs_add_delayed_iput(inode);
8691 goto out;
8693 clear_nlink(inode);
8694 /* One for the block groups ref */
8695 spin_lock(&block_group->lock);
8696 if (block_group->iref) {
8697 block_group->iref = 0;
8698 block_group->inode = NULL;
8699 spin_unlock(&block_group->lock);
8700 iput(inode);
8701 } else {
8702 spin_unlock(&block_group->lock);
8704 /* One for our lookup ref */
8705 btrfs_add_delayed_iput(inode);
8708 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8709 key.offset = block_group->key.objectid;
8710 key.type = 0;
8712 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8713 if (ret < 0)
8714 goto out;
8715 if (ret > 0)
8716 btrfs_release_path(path);
8717 if (ret == 0) {
8718 ret = btrfs_del_item(trans, tree_root, path);
8719 if (ret)
8720 goto out;
8721 btrfs_release_path(path);
8724 spin_lock(&root->fs_info->block_group_cache_lock);
8725 rb_erase(&block_group->cache_node,
8726 &root->fs_info->block_group_cache_tree);
8728 if (root->fs_info->first_logical_byte == block_group->key.objectid)
8729 root->fs_info->first_logical_byte = (u64)-1;
8730 spin_unlock(&root->fs_info->block_group_cache_lock);
8732 down_write(&block_group->space_info->groups_sem);
8734 * we must use list_del_init so people can check to see if they
8735 * are still on the list after taking the semaphore
8737 list_del_init(&block_group->list);
8738 if (list_empty(&block_group->space_info->block_groups[index]))
8739 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8740 up_write(&block_group->space_info->groups_sem);
8742 if (block_group->cached == BTRFS_CACHE_STARTED)
8743 wait_block_group_cache_done(block_group);
8745 btrfs_remove_free_space_cache(block_group);
8747 spin_lock(&block_group->space_info->lock);
8748 block_group->space_info->total_bytes -= block_group->key.offset;
8749 block_group->space_info->bytes_readonly -= block_group->key.offset;
8750 block_group->space_info->disk_total -= block_group->key.offset * factor;
8751 spin_unlock(&block_group->space_info->lock);
8753 memcpy(&key, &block_group->key, sizeof(key));
8755 btrfs_clear_space_info_full(root->fs_info);
8757 btrfs_put_block_group(block_group);
8758 btrfs_put_block_group(block_group);
8760 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8761 if (ret > 0)
8762 ret = -EIO;
8763 if (ret < 0)
8764 goto out;
8766 ret = btrfs_del_item(trans, root, path);
8767 out:
8768 btrfs_free_path(path);
8769 return ret;
8772 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8774 struct btrfs_space_info *space_info;
8775 struct btrfs_super_block *disk_super;
8776 u64 features;
8777 u64 flags;
8778 int mixed = 0;
8779 int ret;
8781 disk_super = fs_info->super_copy;
8782 if (!btrfs_super_root(disk_super))
8783 return 1;
8785 features = btrfs_super_incompat_flags(disk_super);
8786 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8787 mixed = 1;
8789 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8790 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8791 if (ret)
8792 goto out;
8794 if (mixed) {
8795 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8796 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8797 } else {
8798 flags = BTRFS_BLOCK_GROUP_METADATA;
8799 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8800 if (ret)
8801 goto out;
8803 flags = BTRFS_BLOCK_GROUP_DATA;
8804 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8806 out:
8807 return ret;
8810 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8812 return unpin_extent_range(root, start, end);
8815 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8816 u64 num_bytes, u64 *actual_bytes)
8818 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8821 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8823 struct btrfs_fs_info *fs_info = root->fs_info;
8824 struct btrfs_block_group_cache *cache = NULL;
8825 u64 group_trimmed;
8826 u64 start;
8827 u64 end;
8828 u64 trimmed = 0;
8829 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8830 int ret = 0;
8833 * try to trim all FS space, our block group may start from non-zero.
8835 if (range->len == total_bytes)
8836 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8837 else
8838 cache = btrfs_lookup_block_group(fs_info, range->start);
8840 while (cache) {
8841 if (cache->key.objectid >= (range->start + range->len)) {
8842 btrfs_put_block_group(cache);
8843 break;
8846 start = max(range->start, cache->key.objectid);
8847 end = min(range->start + range->len,
8848 cache->key.objectid + cache->key.offset);
8850 if (end - start >= range->minlen) {
8851 if (!block_group_cache_done(cache)) {
8852 ret = cache_block_group(cache, 0);
8853 if (ret) {
8854 btrfs_put_block_group(cache);
8855 break;
8857 ret = wait_block_group_cache_done(cache);
8858 if (ret) {
8859 btrfs_put_block_group(cache);
8860 break;
8863 ret = btrfs_trim_block_group(cache,
8864 &group_trimmed,
8865 start,
8866 end,
8867 range->minlen);
8869 trimmed += group_trimmed;
8870 if (ret) {
8871 btrfs_put_block_group(cache);
8872 break;
8876 cache = next_block_group(fs_info->tree_root, cache);
8879 range->len = trimmed;
8880 return ret;