1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
27 #include "compat.h"
28 #include "hash.h"
29 #include "ctree.h"
30 #include "disk-io.h"
31 #include "print-tree.h"
32 #include "transaction.h"
33 #include "volumes.h"
34 #include "locking.h"
35 #include "free-space-cache.h"
36 #include "math.h"
38 #undef SCRAMBLE_DELAYED_REFS
40 /*
41 * control flags for do_chunk_alloc's force field
42 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
43 * if we really need one.
44 *
45 * CHUNK_ALLOC_LIMITED means to only try and allocate one
46 * if we have very few chunks already allocated. This is
47 * used as part of the clustering code to help make sure
48 * we have a good pool of storage to cluster in, without
49 * filling the FS with empty chunks
50 *
51 * CHUNK_ALLOC_FORCE means it must try to allocate one
52 *
53 */
54 enum {
55 CHUNK_ALLOC_NO_FORCE = 0,
56 CHUNK_ALLOC_LIMITED = 1,
57 CHUNK_ALLOC_FORCE = 2,
58 };
60 /*
61 * Control how reservations are dealt with.
62 *
63 * RESERVE_FREE - freeing a reservation.
64 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
65 * ENOSPC accounting
66 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
67 * bytes_may_use as the ENOSPC accounting is done elsewhere
68 */
69 enum {
70 RESERVE_FREE = 0,
71 RESERVE_ALLOC = 1,
72 RESERVE_ALLOC_NO_ACCOUNT = 2,
73 };
75 static int update_block_group(struct btrfs_trans_handle *trans,
76 struct btrfs_root *root,
77 u64 bytenr, u64 num_bytes, int alloc);
78 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
79 struct btrfs_root *root,
80 u64 bytenr, u64 num_bytes, u64 parent,
81 u64 root_objectid, u64 owner_objectid,
82 u64 owner_offset, int refs_to_drop,
83 struct btrfs_delayed_extent_op *extra_op);
84 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
85 struct extent_buffer *leaf,
86 struct btrfs_extent_item *ei);
87 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
88 struct btrfs_root *root,
89 u64 parent, u64 root_objectid,
90 u64 flags, u64 owner, u64 offset,
91 struct btrfs_key *ins, int ref_mod);
92 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
93 struct btrfs_root *root,
94 u64 parent, u64 root_objectid,
95 u64 flags, struct btrfs_disk_key *key,
96 int level, struct btrfs_key *ins);
97 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
98 struct btrfs_root *extent_root, u64 flags,
99 int force);
100 static int find_next_key(struct btrfs_path *path, int level,
101 struct btrfs_key *key);
102 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
103 int dump_block_groups);
104 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
105 u64 num_bytes, int reserve);
107 static noinline int
108 block_group_cache_done(struct btrfs_block_group_cache *cache)
109 {
110 smp_mb();
111 return cache->cached == BTRFS_CACHE_FINISHED;
112 }
114 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
115 {
116 return (cache->flags & bits) == bits;
117 }
119 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
120 {
121 atomic_inc(&cache->count);
122 }
124 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
125 {
126 if (atomic_dec_and_test(&cache->count)) {
127 WARN_ON(cache->pinned > 0);
128 WARN_ON(cache->reserved > 0);
129 kfree(cache->free_space_ctl);
130 kfree(cache);
131 }
132 }
134 /*
135 * this adds the block group to the fs_info rb tree for the block group
136 * cache
137 */
138 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
139 struct btrfs_block_group_cache *block_group)
140 {
141 struct rb_node **p;
142 struct rb_node *parent = NULL;
143 struct btrfs_block_group_cache *cache;
145 spin_lock(&info->block_group_cache_lock);
146 p = &info->block_group_cache_tree.rb_node;
148 while (*p) {
149 parent = *p;
150 cache = rb_entry(parent, struct btrfs_block_group_cache,
151 cache_node);
152 if (block_group->key.objectid < cache->key.objectid) {
153 p = &(*p)->rb_left;
154 } else if (block_group->key.objectid > cache->key.objectid) {
155 p = &(*p)->rb_right;
156 } else {
157 spin_unlock(&info->block_group_cache_lock);
158 return -EEXIST;
159 }
160 }
162 rb_link_node(&block_group->cache_node, parent, p);
163 rb_insert_color(&block_group->cache_node,
164 &info->block_group_cache_tree);
165 spin_unlock(&info->block_group_cache_lock);
167 return 0;
168 }
170 /*
171 * This will return the block group at or after bytenr if contains is 0, else
172 * it will return the block group that contains the bytenr
173 */
174 static struct btrfs_block_group_cache *
175 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
176 int contains)
177 {
178 struct btrfs_block_group_cache *cache, *ret = NULL;
179 struct rb_node *n;
180 u64 end, start;
182 spin_lock(&info->block_group_cache_lock);
183 n = info->block_group_cache_tree.rb_node;
185 while (n) {
186 cache = rb_entry(n, struct btrfs_block_group_cache,
187 cache_node);
188 end = cache->key.objectid + cache->key.offset - 1;
189 start = cache->key.objectid;
191 if (bytenr < start) {
192 if (!contains && (!ret || start < ret->key.objectid))
193 ret = cache;
194 n = n->rb_left;
195 } else if (bytenr > start) {
196 if (contains && bytenr <= end) {
197 ret = cache;
198 break;
199 }
200 n = n->rb_right;
201 } else {
202 ret = cache;
203 break;
204 }
205 }
206 if (ret)
207 btrfs_get_block_group(ret);
208 spin_unlock(&info->block_group_cache_lock);
210 return ret;
211 }
213 static int add_excluded_extent(struct btrfs_root *root,
214 u64 start, u64 num_bytes)
215 {
216 u64 end = start + num_bytes - 1;
217 set_extent_bits(&root->fs_info->freed_extents[0],
218 start, end, EXTENT_UPTODATE, GFP_NOFS);
219 set_extent_bits(&root->fs_info->freed_extents[1],
220 start, end, EXTENT_UPTODATE, GFP_NOFS);
221 return 0;
222 }
224 static void free_excluded_extents(struct btrfs_root *root,
225 struct btrfs_block_group_cache *cache)
226 {
227 u64 start, end;
229 start = cache->key.objectid;
230 end = start + cache->key.offset - 1;
232 clear_extent_bits(&root->fs_info->freed_extents[0],
233 start, end, EXTENT_UPTODATE, GFP_NOFS);
234 clear_extent_bits(&root->fs_info->freed_extents[1],
235 start, end, EXTENT_UPTODATE, GFP_NOFS);
236 }
238 static int exclude_super_stripes(struct btrfs_root *root,
239 struct btrfs_block_group_cache *cache)
240 {
241 u64 bytenr;
242 u64 *logical;
243 int stripe_len;
244 int i, nr, ret;
246 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
247 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
248 cache->bytes_super += stripe_len;
249 ret = add_excluded_extent(root, cache->key.objectid,
250 stripe_len);
251 BUG_ON(ret); /* -ENOMEM */
252 }
254 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
255 bytenr = btrfs_sb_offset(i);
256 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
257 cache->key.objectid, bytenr,
258 0, &logical, &nr, &stripe_len);
259 BUG_ON(ret); /* -ENOMEM */
261 while (nr--) {
262 cache->bytes_super += stripe_len;
263 ret = add_excluded_extent(root, logical[nr],
264 stripe_len);
265 BUG_ON(ret); /* -ENOMEM */
266 }
268 kfree(logical);
269 }
270 return 0;
271 }
273 static struct btrfs_caching_control *
274 get_caching_control(struct btrfs_block_group_cache *cache)
275 {
276 struct btrfs_caching_control *ctl;
278 spin_lock(&cache->lock);
279 if (cache->cached != BTRFS_CACHE_STARTED) {
280 spin_unlock(&cache->lock);
281 return NULL;
282 }
284 /* We're loading it the fast way, so we don't have a caching_ctl. */
285 if (!cache->caching_ctl) {
286 spin_unlock(&cache->lock);
287 return NULL;
288 }
290 ctl = cache->caching_ctl;
291 atomic_inc(&ctl->count);
292 spin_unlock(&cache->lock);
293 return ctl;
294 }
296 static void put_caching_control(struct btrfs_caching_control *ctl)
297 {
298 if (atomic_dec_and_test(&ctl->count))
299 kfree(ctl);
300 }
302 /*
303 * this is only called by cache_block_group, since we could have freed extents
304 * we need to check the pinned_extents for any extents that can't be used yet
305 * since their free space will be released as soon as the transaction commits.
306 */
307 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
308 struct btrfs_fs_info *info, u64 start, u64 end)
309 {
310 u64 extent_start, extent_end, size, total_added = 0;
311 int ret;
313 while (start < end) {
314 ret = find_first_extent_bit(info->pinned_extents, start,
315 &extent_start, &extent_end,
316 EXTENT_DIRTY | EXTENT_UPTODATE,
317 NULL);
318 if (ret)
319 break;
321 if (extent_start <= start) {
322 start = extent_end + 1;
323 } else if (extent_start > start && extent_start < end) {
324 size = extent_start - start;
325 total_added += size;
326 ret = btrfs_add_free_space(block_group, start,
327 size);
328 BUG_ON(ret); /* -ENOMEM or logic error */
329 start = extent_end + 1;
330 } else {
331 break;
332 }
333 }
335 if (start < end) {
336 size = end - start;
337 total_added += size;
338 ret = btrfs_add_free_space(block_group, start, size);
339 BUG_ON(ret); /* -ENOMEM or logic error */
340 }
342 return total_added;
343 }
345 static noinline void caching_thread(struct btrfs_work *work)
346 {
347 struct btrfs_block_group_cache *block_group;
348 struct btrfs_fs_info *fs_info;
349 struct btrfs_caching_control *caching_ctl;
350 struct btrfs_root *extent_root;
351 struct btrfs_path *path;
352 struct extent_buffer *leaf;
353 struct btrfs_key key;
354 u64 total_found = 0;
355 u64 last = 0;
356 u32 nritems;
357 int ret = 0;
359 caching_ctl = container_of(work, struct btrfs_caching_control, work);
360 block_group = caching_ctl->block_group;
361 fs_info = block_group->fs_info;
362 extent_root = fs_info->extent_root;
364 path = btrfs_alloc_path();
365 if (!path)
366 goto out;
368 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
370 /*
371 * We don't want to deadlock with somebody trying to allocate a new
372 * extent for the extent root while also trying to search the extent
373 * root to add free space. So we skip locking and search the commit
374 * root, since its read-only
375 */
376 path->skip_locking = 1;
377 path->search_commit_root = 1;
378 path->reada = 1;
380 key.objectid = last;
381 key.offset = 0;
382 key.type = BTRFS_EXTENT_ITEM_KEY;
383 again:
384 mutex_lock(&caching_ctl->mutex);
385 /* need to make sure the commit_root doesn't disappear */
386 down_read(&fs_info->extent_commit_sem);
388 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
389 if (ret < 0)
390 goto err;
392 leaf = path->nodes[0];
393 nritems = btrfs_header_nritems(leaf);
395 while (1) {
396 if (btrfs_fs_closing(fs_info) > 1) {
397 last = (u64)-1;
398 break;
399 }
401 if (path->slots[0] < nritems) {
402 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
403 } else {
404 ret = find_next_key(path, 0, &key);
405 if (ret)
406 break;
408 if (need_resched() ||
409 btrfs_next_leaf(extent_root, path)) {
410 caching_ctl->progress = last;
411 btrfs_release_path(path);
412 up_read(&fs_info->extent_commit_sem);
413 mutex_unlock(&caching_ctl->mutex);
414 cond_resched();
415 goto again;
416 }
417 leaf = path->nodes[0];
418 nritems = btrfs_header_nritems(leaf);
419 continue;
420 }
422 if (key.objectid < block_group->key.objectid) {
423 path->slots[0]++;
424 continue;
425 }
427 if (key.objectid >= block_group->key.objectid +
428 block_group->key.offset)
429 break;
431 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
432 total_found += add_new_free_space(block_group,
433 fs_info, last,
434 key.objectid);
435 last = key.objectid + key.offset;
437 if (total_found > (1024 * 1024 * 2)) {
438 total_found = 0;
439 wake_up(&caching_ctl->wait);
440 }
441 }
442 path->slots[0]++;
443 }
444 ret = 0;
446 total_found += add_new_free_space(block_group, fs_info, last,
447 block_group->key.objectid +
448 block_group->key.offset);
449 caching_ctl->progress = (u64)-1;
451 spin_lock(&block_group->lock);
452 block_group->caching_ctl = NULL;
453 block_group->cached = BTRFS_CACHE_FINISHED;
454 spin_unlock(&block_group->lock);
456 err:
457 btrfs_free_path(path);
458 up_read(&fs_info->extent_commit_sem);
460 free_excluded_extents(extent_root, block_group);
462 mutex_unlock(&caching_ctl->mutex);
463 out:
464 wake_up(&caching_ctl->wait);
466 put_caching_control(caching_ctl);
467 btrfs_put_block_group(block_group);
468 }
470 static int cache_block_group(struct btrfs_block_group_cache *cache,
471 struct btrfs_trans_handle *trans,
472 struct btrfs_root *root,
473 int load_cache_only)
474 {
475 DEFINE_WAIT(wait);
476 struct btrfs_fs_info *fs_info = cache->fs_info;
477 struct btrfs_caching_control *caching_ctl;
478 int ret = 0;
480 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
481 if (!caching_ctl)
482 return -ENOMEM;
484 INIT_LIST_HEAD(&caching_ctl->list);
485 mutex_init(&caching_ctl->mutex);
486 init_waitqueue_head(&caching_ctl->wait);
487 caching_ctl->block_group = cache;
488 caching_ctl->progress = cache->key.objectid;
489 atomic_set(&caching_ctl->count, 1);
490 caching_ctl->work.func = caching_thread;
492 spin_lock(&cache->lock);
493 /*
494 * This should be a rare occasion, but this could happen I think in the
495 * case where one thread starts to load the space cache info, and then
496 * some other thread starts a transaction commit which tries to do an
497 * allocation while the other thread is still loading the space cache
498 * info. The previous loop should have kept us from choosing this block
499 * group, but if we've moved to the state where we will wait on caching
500 * block groups we need to first check if we're doing a fast load here,
501 * so we can wait for it to finish, otherwise we could end up allocating
502 * from a block group who's cache gets evicted for one reason or
503 * another.
504 */
505 while (cache->cached == BTRFS_CACHE_FAST) {
506 struct btrfs_caching_control *ctl;
508 ctl = cache->caching_ctl;
509 atomic_inc(&ctl->count);
510 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
511 spin_unlock(&cache->lock);
513 schedule();
515 finish_wait(&ctl->wait, &wait);
516 put_caching_control(ctl);
517 spin_lock(&cache->lock);
518 }
520 if (cache->cached != BTRFS_CACHE_NO) {
521 spin_unlock(&cache->lock);
522 kfree(caching_ctl);
523 return 0;
524 }
525 WARN_ON(cache->caching_ctl);
526 cache->caching_ctl = caching_ctl;
527 cache->cached = BTRFS_CACHE_FAST;
528 spin_unlock(&cache->lock);
530 /*
531 * We can't do the read from on-disk cache during a commit since we need
532 * to have the normal tree locking. Also if we are currently trying to
533 * allocate blocks for the tree root we can't do the fast caching since
534 * we likely hold important locks.
535 */
536 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
537 ret = load_free_space_cache(fs_info, cache);
539 spin_lock(&cache->lock);
540 if (ret == 1) {
541 cache->caching_ctl = NULL;
542 cache->cached = BTRFS_CACHE_FINISHED;
543 cache->last_byte_to_unpin = (u64)-1;
544 } else {
545 if (load_cache_only) {
546 cache->caching_ctl = NULL;
547 cache->cached = BTRFS_CACHE_NO;
548 } else {
549 cache->cached = BTRFS_CACHE_STARTED;
550 }
551 }
552 spin_unlock(&cache->lock);
553 wake_up(&caching_ctl->wait);
554 if (ret == 1) {
555 put_caching_control(caching_ctl);
556 free_excluded_extents(fs_info->extent_root, cache);
557 return 0;
558 }
559 } else {
560 /*
561 * We are not going to do the fast caching, set cached to the
562 * appropriate value and wakeup any waiters.
563 */
564 spin_lock(&cache->lock);
565 if (load_cache_only) {
566 cache->caching_ctl = NULL;
567 cache->cached = BTRFS_CACHE_NO;
568 } else {
569 cache->cached = BTRFS_CACHE_STARTED;
570 }
571 spin_unlock(&cache->lock);
572 wake_up(&caching_ctl->wait);
573 }
575 if (load_cache_only) {
576 put_caching_control(caching_ctl);
577 return 0;
578 }
580 down_write(&fs_info->extent_commit_sem);
581 atomic_inc(&caching_ctl->count);
582 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
583 up_write(&fs_info->extent_commit_sem);
585 btrfs_get_block_group(cache);
587 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
589 return ret;
590 }
592 /*
593 * return the block group that starts at or after bytenr
594 */
595 static struct btrfs_block_group_cache *
596 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
597 {
598 struct btrfs_block_group_cache *cache;
600 cache = block_group_cache_tree_search(info, bytenr, 0);
602 return cache;
603 }
605 /*
606 * return the block group that contains the given bytenr
607 */
608 struct btrfs_block_group_cache *btrfs_lookup_block_group(
609 struct btrfs_fs_info *info,
610 u64 bytenr)
611 {
612 struct btrfs_block_group_cache *cache;
614 cache = block_group_cache_tree_search(info, bytenr, 1);
616 return cache;
617 }
619 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
620 u64 flags)
621 {
622 struct list_head *head = &info->space_info;
623 struct btrfs_space_info *found;
625 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
627 rcu_read_lock();
628 list_for_each_entry_rcu(found, head, list) {
629 if (found->flags & flags) {
630 rcu_read_unlock();
631 return found;
632 }
633 }
634 rcu_read_unlock();
635 return NULL;
636 }
638 /*
639 * after adding space to the filesystem, we need to clear the full flags
640 * on all the space infos.
641 */
642 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
643 {
644 struct list_head *head = &info->space_info;
645 struct btrfs_space_info *found;
647 rcu_read_lock();
648 list_for_each_entry_rcu(found, head, list)
649 found->full = 0;
650 rcu_read_unlock();
651 }
653 u64 btrfs_find_block_group(struct btrfs_root *root,
654 u64 search_start, u64 search_hint, int owner)
655 {
656 struct btrfs_block_group_cache *cache;
657 u64 used;
658 u64 last = max(search_hint, search_start);
659 u64 group_start = 0;
660 int full_search = 0;
661 int factor = 9;
662 int wrapped = 0;
663 again:
664 while (1) {
665 cache = btrfs_lookup_first_block_group(root->fs_info, last);
666 if (!cache)
667 break;
669 spin_lock(&cache->lock);
670 last = cache->key.objectid + cache->key.offset;
671 used = btrfs_block_group_used(&cache->item);
673 if ((full_search || !cache->ro) &&
674 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
675 if (used + cache->pinned + cache->reserved <
676 div_factor(cache->key.offset, factor)) {
677 group_start = cache->key.objectid;
678 spin_unlock(&cache->lock);
679 btrfs_put_block_group(cache);
680 goto found;
681 }
682 }
683 spin_unlock(&cache->lock);
684 btrfs_put_block_group(cache);
685 cond_resched();
686 }
687 if (!wrapped) {
688 last = search_start;
689 wrapped = 1;
690 goto again;
691 }
692 if (!full_search && factor < 10) {
693 last = search_start;
694 full_search = 1;
695 factor = 10;
696 goto again;
697 }
698 found:
699 return group_start;
700 }
702 /* simple helper to search for an existing extent at a given offset */
703 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
704 {
705 int ret;
706 struct btrfs_key key;
707 struct btrfs_path *path;
709 path = btrfs_alloc_path();
710 if (!path)
711 return -ENOMEM;
713 key.objectid = start;
714 key.offset = len;
715 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
716 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
717 0, 0);
718 btrfs_free_path(path);
719 return ret;
720 }
722 /*
723 * helper function to lookup reference count and flags of extent.
724 *
725 * the head node for delayed ref is used to store the sum of all the
726 * reference count modifications queued up in the rbtree. the head
727 * node may also store the extent flags to set. This way you can check
728 * to see what the reference count and extent flags would be if all of
729 * the delayed refs are not processed.
730 */
731 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
732 struct btrfs_root *root, u64 bytenr,
733 u64 num_bytes, u64 *refs, u64 *flags)
734 {
735 struct btrfs_delayed_ref_head *head;
736 struct btrfs_delayed_ref_root *delayed_refs;
737 struct btrfs_path *path;
738 struct btrfs_extent_item *ei;
739 struct extent_buffer *leaf;
740 struct btrfs_key key;
741 u32 item_size;
742 u64 num_refs;
743 u64 extent_flags;
744 int ret;
746 path = btrfs_alloc_path();
747 if (!path)
748 return -ENOMEM;
750 key.objectid = bytenr;
751 key.type = BTRFS_EXTENT_ITEM_KEY;
752 key.offset = num_bytes;
753 if (!trans) {
754 path->skip_locking = 1;
755 path->search_commit_root = 1;
756 }
757 again:
758 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
759 &key, path, 0, 0);
760 if (ret < 0)
761 goto out_free;
763 if (ret == 0) {
764 leaf = path->nodes[0];
765 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
766 if (item_size >= sizeof(*ei)) {
767 ei = btrfs_item_ptr(leaf, path->slots[0],
768 struct btrfs_extent_item);
769 num_refs = btrfs_extent_refs(leaf, ei);
770 extent_flags = btrfs_extent_flags(leaf, ei);
771 } else {
772 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
773 struct btrfs_extent_item_v0 *ei0;
774 BUG_ON(item_size != sizeof(*ei0));
775 ei0 = btrfs_item_ptr(leaf, path->slots[0],
776 struct btrfs_extent_item_v0);
777 num_refs = btrfs_extent_refs_v0(leaf, ei0);
778 /* FIXME: this isn't correct for data */
779 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
780 #else
781 BUG();
782 #endif
783 }
784 BUG_ON(num_refs == 0);
785 } else {
786 num_refs = 0;
787 extent_flags = 0;
788 ret = 0;
789 }
791 if (!trans)
792 goto out;
794 delayed_refs = &trans->transaction->delayed_refs;
795 spin_lock(&delayed_refs->lock);
796 head = btrfs_find_delayed_ref_head(trans, bytenr);
797 if (head) {
798 if (!mutex_trylock(&head->mutex)) {
799 atomic_inc(&head->node.refs);
800 spin_unlock(&delayed_refs->lock);
802 btrfs_release_path(path);
804 /*
805 * Mutex was contended, block until it's released and try
806 * again
807 */
808 mutex_lock(&head->mutex);
809 mutex_unlock(&head->mutex);
810 btrfs_put_delayed_ref(&head->node);
811 goto again;
812 }
813 if (head->extent_op && head->extent_op->update_flags)
814 extent_flags |= head->extent_op->flags_to_set;
815 else
816 BUG_ON(num_refs == 0);
818 num_refs += head->node.ref_mod;
819 mutex_unlock(&head->mutex);
820 }
821 spin_unlock(&delayed_refs->lock);
822 out:
823 WARN_ON(num_refs == 0);
824 if (refs)
825 *refs = num_refs;
826 if (flags)
827 *flags = extent_flags;
828 out_free:
829 btrfs_free_path(path);
830 return ret;
831 }
833 /*
834 * Back reference rules. Back refs have three main goals:
835 *
836 * 1) differentiate between all holders of references to an extent so that
837 * when a reference is dropped we can make sure it was a valid reference
838 * before freeing the extent.
839 *
840 * 2) Provide enough information to quickly find the holders of an extent
841 * if we notice a given block is corrupted or bad.
842 *
843 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
844 * maintenance. This is actually the same as #2, but with a slightly
845 * different use case.
846 *
847 * There are two kinds of back refs. The implicit back refs is optimized
848 * for pointers in non-shared tree blocks. For a given pointer in a block,
849 * back refs of this kind provide information about the block's owner tree
850 * and the pointer's key. These information allow us to find the block by
851 * b-tree searching. The full back refs is for pointers in tree blocks not
852 * referenced by their owner trees. The location of tree block is recorded
853 * in the back refs. Actually the full back refs is generic, and can be
854 * used in all cases the implicit back refs is used. The major shortcoming
855 * of the full back refs is its overhead. Every time a tree block gets
856 * COWed, we have to update back refs entry for all pointers in it.
857 *
858 * For a newly allocated tree block, we use implicit back refs for
859 * pointers in it. This means most tree related operations only involve
860 * implicit back refs. For a tree block created in old transaction, the
861 * only way to drop a reference to it is COW it. So we can detect the
862 * event that tree block loses its owner tree's reference and do the
863 * back refs conversion.
864 *
865 * When a tree block is COW'd through a tree, there are four cases:
866 *
867 * The reference count of the block is one and the tree is the block's
868 * owner tree. Nothing to do in this case.
869 *
870 * The reference count of the block is one and the tree is not the
871 * block's owner tree. In this case, full back refs is used for pointers
872 * in the block. Remove these full back refs, add implicit back refs for
873 * every pointers in the new block.
874 *
875 * The reference count of the block is greater than one and the tree is
876 * the block's owner tree. In this case, implicit back refs is used for
877 * pointers in the block. Add full back refs for every pointers in the
878 * block, increase lower level extents' reference counts. The original
879 * implicit back refs are entailed to the new block.
880 *
881 * The reference count of the block is greater than one and the tree is
882 * not the block's owner tree. Add implicit back refs for every pointer in
883 * the new block, increase lower level extents' reference count.
884 *
885 * Back Reference Key composing:
886 *
887 * The key objectid corresponds to the first byte in the extent,
888 * The key type is used to differentiate between types of back refs.
889 * There are different meanings of the key offset for different types
890 * of back refs.
891 *
892 * File extents can be referenced by:
893 *
894 * - multiple snapshots, subvolumes, or different generations in one subvol
895 * - different files inside a single subvolume
896 * - different offsets inside a file (bookend extents in file.c)
897 *
898 * The extent ref structure for the implicit back refs has fields for:
899 *
900 * - Objectid of the subvolume root
901 * - objectid of the file holding the reference
902 * - original offset in the file
903 * - how many bookend extents
904 *
905 * The key offset for the implicit back refs is hash of the first
906 * three fields.
907 *
908 * The extent ref structure for the full back refs has field for:
909 *
910 * - number of pointers in the tree leaf
911 *
912 * The key offset for the implicit back refs is the first byte of
913 * the tree leaf
914 *
915 * When a file extent is allocated, The implicit back refs is used.
916 * the fields are filled in:
917 *
918 * (root_key.objectid, inode objectid, offset in file, 1)
919 *
920 * When a file extent is removed file truncation, we find the
921 * corresponding implicit back refs and check the following fields:
922 *
923 * (btrfs_header_owner(leaf), inode objectid, offset in file)
924 *
925 * Btree extents can be referenced by:
926 *
927 * - Different subvolumes
928 *
929 * Both the implicit back refs and the full back refs for tree blocks
930 * only consist of key. The key offset for the implicit back refs is
931 * objectid of block's owner tree. The key offset for the full back refs
932 * is the first byte of parent block.
933 *
934 * When implicit back refs is used, information about the lowest key and
935 * level of the tree block are required. These information are stored in
936 * tree block info structure.
937 */
939 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
940 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
941 struct btrfs_root *root,
942 struct btrfs_path *path,
943 u64 owner, u32 extra_size)
944 {
945 struct btrfs_extent_item *item;
946 struct btrfs_extent_item_v0 *ei0;
947 struct btrfs_extent_ref_v0 *ref0;
948 struct btrfs_tree_block_info *bi;
949 struct extent_buffer *leaf;
950 struct btrfs_key key;
951 struct btrfs_key found_key;
952 u32 new_size = sizeof(*item);
953 u64 refs;
954 int ret;
956 leaf = path->nodes[0];
957 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
959 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
960 ei0 = btrfs_item_ptr(leaf, path->slots[0],
961 struct btrfs_extent_item_v0);
962 refs = btrfs_extent_refs_v0(leaf, ei0);
964 if (owner == (u64)-1) {
965 while (1) {
966 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
967 ret = btrfs_next_leaf(root, path);
968 if (ret < 0)
969 return ret;
970 BUG_ON(ret > 0); /* Corruption */
971 leaf = path->nodes[0];
972 }
973 btrfs_item_key_to_cpu(leaf, &found_key,
974 path->slots[0]);
975 BUG_ON(key.objectid != found_key.objectid);
976 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
977 path->slots[0]++;
978 continue;
979 }
980 ref0 = btrfs_item_ptr(leaf, path->slots[0],
981 struct btrfs_extent_ref_v0);
982 owner = btrfs_ref_objectid_v0(leaf, ref0);
983 break;
984 }
985 }
986 btrfs_release_path(path);
988 if (owner < BTRFS_FIRST_FREE_OBJECTID)
989 new_size += sizeof(*bi);
991 new_size -= sizeof(*ei0);
992 ret = btrfs_search_slot(trans, root, &key, path,
993 new_size + extra_size, 1);
994 if (ret < 0)
995 return ret;
996 BUG_ON(ret); /* Corruption */
998 btrfs_extend_item(trans, root, path, new_size);
1000 leaf = path->nodes[0];
1001 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1002 btrfs_set_extent_refs(leaf, item, refs);
1003 /* FIXME: get real generation */
1004 btrfs_set_extent_generation(leaf, item, 0);
1005 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1006 btrfs_set_extent_flags(leaf, item,
1007 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1008 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1009 bi = (struct btrfs_tree_block_info *)(item + 1);
1010 /* FIXME: get first key of the block */
1011 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1012 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1013 } else {
1014 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1015 }
1016 btrfs_mark_buffer_dirty(leaf);
1017 return 0;
1018 }
1019 #endif
1021 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1022 {
1023 u32 high_crc = ~(u32)0;
1024 u32 low_crc = ~(u32)0;
1025 __le64 lenum;
1027 lenum = cpu_to_le64(root_objectid);
1028 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1029 lenum = cpu_to_le64(owner);
1030 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1031 lenum = cpu_to_le64(offset);
1032 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1034 return ((u64)high_crc << 31) ^ (u64)low_crc;
1035 }
1037 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1038 struct btrfs_extent_data_ref *ref)
1039 {
1040 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1041 btrfs_extent_data_ref_objectid(leaf, ref),
1042 btrfs_extent_data_ref_offset(leaf, ref));
1043 }
1045 static int match_extent_data_ref(struct extent_buffer *leaf,
1046 struct btrfs_extent_data_ref *ref,
1047 u64 root_objectid, u64 owner, u64 offset)
1048 {
1049 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1050 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1051 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1052 return 0;
1053 return 1;
1054 }
1056 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1057 struct btrfs_root *root,
1058 struct btrfs_path *path,
1059 u64 bytenr, u64 parent,
1060 u64 root_objectid,
1061 u64 owner, u64 offset)
1062 {
1063 struct btrfs_key key;
1064 struct btrfs_extent_data_ref *ref;
1065 struct extent_buffer *leaf;
1066 u32 nritems;
1067 int ret;
1068 int recow;
1069 int err = -ENOENT;
1071 key.objectid = bytenr;
1072 if (parent) {
1073 key.type = BTRFS_SHARED_DATA_REF_KEY;
1074 key.offset = parent;
1075 } else {
1076 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1077 key.offset = hash_extent_data_ref(root_objectid,
1078 owner, offset);
1079 }
1080 again:
1081 recow = 0;
1082 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1083 if (ret < 0) {
1084 err = ret;
1085 goto fail;
1086 }
1088 if (parent) {
1089 if (!ret)
1090 return 0;
1091 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1092 key.type = BTRFS_EXTENT_REF_V0_KEY;
1093 btrfs_release_path(path);
1094 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1095 if (ret < 0) {
1096 err = ret;
1097 goto fail;
1098 }
1099 if (!ret)
1100 return 0;
1101 #endif
1102 goto fail;
1103 }
1105 leaf = path->nodes[0];
1106 nritems = btrfs_header_nritems(leaf);
1107 while (1) {
1108 if (path->slots[0] >= nritems) {
1109 ret = btrfs_next_leaf(root, path);
1110 if (ret < 0)
1111 err = ret;
1112 if (ret)
1113 goto fail;
1115 leaf = path->nodes[0];
1116 nritems = btrfs_header_nritems(leaf);
1117 recow = 1;
1118 }
1120 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1121 if (key.objectid != bytenr ||
1122 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1123 goto fail;
1125 ref = btrfs_item_ptr(leaf, path->slots[0],
1126 struct btrfs_extent_data_ref);
1128 if (match_extent_data_ref(leaf, ref, root_objectid,
1129 owner, offset)) {
1130 if (recow) {
1131 btrfs_release_path(path);
1132 goto again;
1133 }
1134 err = 0;
1135 break;
1136 }
1137 path->slots[0]++;
1138 }
1139 fail:
1140 return err;
1141 }
1143 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1144 struct btrfs_root *root,
1145 struct btrfs_path *path,
1146 u64 bytenr, u64 parent,
1147 u64 root_objectid, u64 owner,
1148 u64 offset, int refs_to_add)
1149 {
1150 struct btrfs_key key;
1151 struct extent_buffer *leaf;
1152 u32 size;
1153 u32 num_refs;
1154 int ret;
1156 key.objectid = bytenr;
1157 if (parent) {
1158 key.type = BTRFS_SHARED_DATA_REF_KEY;
1159 key.offset = parent;
1160 size = sizeof(struct btrfs_shared_data_ref);
1161 } else {
1162 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1163 key.offset = hash_extent_data_ref(root_objectid,
1164 owner, offset);
1165 size = sizeof(struct btrfs_extent_data_ref);
1166 }
1168 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1169 if (ret && ret != -EEXIST)
1170 goto fail;
1172 leaf = path->nodes[0];
1173 if (parent) {
1174 struct btrfs_shared_data_ref *ref;
1175 ref = btrfs_item_ptr(leaf, path->slots[0],
1176 struct btrfs_shared_data_ref);
1177 if (ret == 0) {
1178 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1179 } else {
1180 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1181 num_refs += refs_to_add;
1182 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1183 }
1184 } else {
1185 struct btrfs_extent_data_ref *ref;
1186 while (ret == -EEXIST) {
1187 ref = btrfs_item_ptr(leaf, path->slots[0],
1188 struct btrfs_extent_data_ref);
1189 if (match_extent_data_ref(leaf, ref, root_objectid,
1190 owner, offset))
1191 break;
1192 btrfs_release_path(path);
1193 key.offset++;
1194 ret = btrfs_insert_empty_item(trans, root, path, &key,
1195 size);
1196 if (ret && ret != -EEXIST)
1197 goto fail;
1199 leaf = path->nodes[0];
1200 }
1201 ref = btrfs_item_ptr(leaf, path->slots[0],
1202 struct btrfs_extent_data_ref);
1203 if (ret == 0) {
1204 btrfs_set_extent_data_ref_root(leaf, ref,
1205 root_objectid);
1206 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1207 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1208 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1209 } else {
1210 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1211 num_refs += refs_to_add;
1212 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1213 }
1214 }
1215 btrfs_mark_buffer_dirty(leaf);
1216 ret = 0;
1217 fail:
1218 btrfs_release_path(path);
1219 return ret;
1220 }
1222 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1223 struct btrfs_root *root,
1224 struct btrfs_path *path,
1225 int refs_to_drop)
1226 {
1227 struct btrfs_key key;
1228 struct btrfs_extent_data_ref *ref1 = NULL;
1229 struct btrfs_shared_data_ref *ref2 = NULL;
1230 struct extent_buffer *leaf;
1231 u32 num_refs = 0;
1232 int ret = 0;
1234 leaf = path->nodes[0];
1235 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1237 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1238 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1239 struct btrfs_extent_data_ref);
1240 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1241 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1242 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1243 struct btrfs_shared_data_ref);
1244 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1245 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1246 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1247 struct btrfs_extent_ref_v0 *ref0;
1248 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1249 struct btrfs_extent_ref_v0);
1250 num_refs = btrfs_ref_count_v0(leaf, ref0);
1251 #endif
1252 } else {
1253 BUG();
1254 }
1256 BUG_ON(num_refs < refs_to_drop);
1257 num_refs -= refs_to_drop;
1259 if (num_refs == 0) {
1260 ret = btrfs_del_item(trans, root, path);
1261 } else {
1262 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1263 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1264 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1265 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1266 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1267 else {
1268 struct btrfs_extent_ref_v0 *ref0;
1269 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1270 struct btrfs_extent_ref_v0);
1271 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1272 }
1273 #endif
1274 btrfs_mark_buffer_dirty(leaf);
1275 }
1276 return ret;
1277 }
1279 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1280 struct btrfs_path *path,
1281 struct btrfs_extent_inline_ref *iref)
1282 {
1283 struct btrfs_key key;
1284 struct extent_buffer *leaf;
1285 struct btrfs_extent_data_ref *ref1;
1286 struct btrfs_shared_data_ref *ref2;
1287 u32 num_refs = 0;
1289 leaf = path->nodes[0];
1290 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1291 if (iref) {
1292 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1293 BTRFS_EXTENT_DATA_REF_KEY) {
1294 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1295 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1296 } else {
1297 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1298 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1299 }
1300 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1301 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1302 struct btrfs_extent_data_ref);
1303 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1304 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1305 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1306 struct btrfs_shared_data_ref);
1307 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1308 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1309 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1310 struct btrfs_extent_ref_v0 *ref0;
1311 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1312 struct btrfs_extent_ref_v0);
1313 num_refs = btrfs_ref_count_v0(leaf, ref0);
1314 #endif
1315 } else {
1316 WARN_ON(1);
1317 }
1318 return num_refs;
1319 }
1321 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1322 struct btrfs_root *root,
1323 struct btrfs_path *path,
1324 u64 bytenr, u64 parent,
1325 u64 root_objectid)
1326 {
1327 struct btrfs_key key;
1328 int ret;
1330 key.objectid = bytenr;
1331 if (parent) {
1332 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1333 key.offset = parent;
1334 } else {
1335 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1336 key.offset = root_objectid;
1337 }
1339 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1340 if (ret > 0)
1341 ret = -ENOENT;
1342 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1343 if (ret == -ENOENT && parent) {
1344 btrfs_release_path(path);
1345 key.type = BTRFS_EXTENT_REF_V0_KEY;
1346 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1347 if (ret > 0)
1348 ret = -ENOENT;
1349 }
1350 #endif
1351 return ret;
1352 }
1354 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1355 struct btrfs_root *root,
1356 struct btrfs_path *path,
1357 u64 bytenr, u64 parent,
1358 u64 root_objectid)
1359 {
1360 struct btrfs_key key;
1361 int ret;
1363 key.objectid = bytenr;
1364 if (parent) {
1365 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1366 key.offset = parent;
1367 } else {
1368 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1369 key.offset = root_objectid;
1370 }
1372 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1373 btrfs_release_path(path);
1374 return ret;
1375 }
1377 static inline int extent_ref_type(u64 parent, u64 owner)
1378 {
1379 int type;
1380 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1381 if (parent > 0)
1382 type = BTRFS_SHARED_BLOCK_REF_KEY;
1383 else
1384 type = BTRFS_TREE_BLOCK_REF_KEY;
1385 } else {
1386 if (parent > 0)
1387 type = BTRFS_SHARED_DATA_REF_KEY;
1388 else
1389 type = BTRFS_EXTENT_DATA_REF_KEY;
1390 }
1391 return type;
1392 }
1394 static int find_next_key(struct btrfs_path *path, int level,
1395 struct btrfs_key *key)
1397 {
1398 for (; level < BTRFS_MAX_LEVEL; level++) {
1399 if (!path->nodes[level])
1400 break;
1401 if (path->slots[level] + 1 >=
1402 btrfs_header_nritems(path->nodes[level]))
1403 continue;
1404 if (level == 0)
1405 btrfs_item_key_to_cpu(path->nodes[level], key,
1406 path->slots[level] + 1);
1407 else
1408 btrfs_node_key_to_cpu(path->nodes[level], key,
1409 path->slots[level] + 1);
1410 return 0;
1411 }
1412 return 1;
1413 }
1415 /*
1416 * look for inline back ref. if back ref is found, *ref_ret is set
1417 * to the address of inline back ref, and 0 is returned.
1418 *
1419 * if back ref isn't found, *ref_ret is set to the address where it
1420 * should be inserted, and -ENOENT is returned.
1421 *
1422 * if insert is true and there are too many inline back refs, the path
1423 * points to the extent item, and -EAGAIN is returned.
1424 *
1425 * NOTE: inline back refs are ordered in the same way that back ref
1426 * items in the tree are ordered.
1427 */
1428 static noinline_for_stack
1429 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1430 struct btrfs_root *root,
1431 struct btrfs_path *path,
1432 struct btrfs_extent_inline_ref **ref_ret,
1433 u64 bytenr, u64 num_bytes,
1434 u64 parent, u64 root_objectid,
1435 u64 owner, u64 offset, int insert)
1436 {
1437 struct btrfs_key key;
1438 struct extent_buffer *leaf;
1439 struct btrfs_extent_item *ei;
1440 struct btrfs_extent_inline_ref *iref;
1441 u64 flags;
1442 u64 item_size;
1443 unsigned long ptr;
1444 unsigned long end;
1445 int extra_size;
1446 int type;
1447 int want;
1448 int ret;
1449 int err = 0;
1451 key.objectid = bytenr;
1452 key.type = BTRFS_EXTENT_ITEM_KEY;
1453 key.offset = num_bytes;
1455 want = extent_ref_type(parent, owner);
1456 if (insert) {
1457 extra_size = btrfs_extent_inline_ref_size(want);
1458 path->keep_locks = 1;
1459 } else
1460 extra_size = -1;
1461 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1462 if (ret < 0) {
1463 err = ret;
1464 goto out;
1465 }
1466 if (ret && !insert) {
1467 err = -ENOENT;
1468 goto out;
1469 }
1470 BUG_ON(ret); /* Corruption */
1472 leaf = path->nodes[0];
1473 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1474 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1475 if (item_size < sizeof(*ei)) {
1476 if (!insert) {
1477 err = -ENOENT;
1478 goto out;
1479 }
1480 ret = convert_extent_item_v0(trans, root, path, owner,
1481 extra_size);
1482 if (ret < 0) {
1483 err = ret;
1484 goto out;
1485 }
1486 leaf = path->nodes[0];
1487 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1488 }
1489 #endif
1490 BUG_ON(item_size < sizeof(*ei));
1492 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1493 flags = btrfs_extent_flags(leaf, ei);
1495 ptr = (unsigned long)(ei + 1);
1496 end = (unsigned long)ei + item_size;
1498 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1499 ptr += sizeof(struct btrfs_tree_block_info);
1500 BUG_ON(ptr > end);
1501 } else {
1502 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1503 }
1505 err = -ENOENT;
1506 while (1) {
1507 if (ptr >= end) {
1508 WARN_ON(ptr > end);
1509 break;
1510 }
1511 iref = (struct btrfs_extent_inline_ref *)ptr;
1512 type = btrfs_extent_inline_ref_type(leaf, iref);
1513 if (want < type)
1514 break;
1515 if (want > type) {
1516 ptr += btrfs_extent_inline_ref_size(type);
1517 continue;
1518 }
1520 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1521 struct btrfs_extent_data_ref *dref;
1522 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1523 if (match_extent_data_ref(leaf, dref, root_objectid,
1524 owner, offset)) {
1525 err = 0;
1526 break;
1527 }
1528 if (hash_extent_data_ref_item(leaf, dref) <
1529 hash_extent_data_ref(root_objectid, owner, offset))
1530 break;
1531 } else {
1532 u64 ref_offset;
1533 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1534 if (parent > 0) {
1535 if (parent == ref_offset) {
1536 err = 0;
1537 break;
1538 }
1539 if (ref_offset < parent)
1540 break;
1541 } else {
1542 if (root_objectid == ref_offset) {
1543 err = 0;
1544 break;
1545 }
1546 if (ref_offset < root_objectid)
1547 break;
1548 }
1549 }
1550 ptr += btrfs_extent_inline_ref_size(type);
1551 }
1552 if (err == -ENOENT && insert) {
1553 if (item_size + extra_size >=
1554 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1555 err = -EAGAIN;
1556 goto out;
1557 }
1558 /*
1559 * To add new inline back ref, we have to make sure
1560 * there is no corresponding back ref item.
1561 * For simplicity, we just do not add new inline back
1562 * ref if there is any kind of item for this block
1563 */
1564 if (find_next_key(path, 0, &key) == 0 &&
1565 key.objectid == bytenr &&
1566 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1567 err = -EAGAIN;
1568 goto out;
1569 }
1570 }
1571 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1572 out:
1573 if (insert) {
1574 path->keep_locks = 0;
1575 btrfs_unlock_up_safe(path, 1);
1576 }
1577 return err;
1578 }
1580 /*
1581 * helper to add new inline back ref
1582 */
1583 static noinline_for_stack
1584 void setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1585 struct btrfs_root *root,
1586 struct btrfs_path *path,
1587 struct btrfs_extent_inline_ref *iref,
1588 u64 parent, u64 root_objectid,
1589 u64 owner, u64 offset, int refs_to_add,
1590 struct btrfs_delayed_extent_op *extent_op)
1591 {
1592 struct extent_buffer *leaf;
1593 struct btrfs_extent_item *ei;
1594 unsigned long ptr;
1595 unsigned long end;
1596 unsigned long item_offset;
1597 u64 refs;
1598 int size;
1599 int type;
1601 leaf = path->nodes[0];
1602 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1603 item_offset = (unsigned long)iref - (unsigned long)ei;
1605 type = extent_ref_type(parent, owner);
1606 size = btrfs_extent_inline_ref_size(type);
1608 btrfs_extend_item(trans, root, path, size);
1610 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1611 refs = btrfs_extent_refs(leaf, ei);
1612 refs += refs_to_add;
1613 btrfs_set_extent_refs(leaf, ei, refs);
1614 if (extent_op)
1615 __run_delayed_extent_op(extent_op, leaf, ei);
1617 ptr = (unsigned long)ei + item_offset;
1618 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1619 if (ptr < end - size)
1620 memmove_extent_buffer(leaf, ptr + size, ptr,
1621 end - size - ptr);
1623 iref = (struct btrfs_extent_inline_ref *)ptr;
1624 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1625 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1626 struct btrfs_extent_data_ref *dref;
1627 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1628 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1629 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1630 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1631 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1632 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1633 struct btrfs_shared_data_ref *sref;
1634 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1635 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1636 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1637 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1638 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1639 } else {
1640 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1641 }
1642 btrfs_mark_buffer_dirty(leaf);
1643 }
1645 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1646 struct btrfs_root *root,
1647 struct btrfs_path *path,
1648 struct btrfs_extent_inline_ref **ref_ret,
1649 u64 bytenr, u64 num_bytes, u64 parent,
1650 u64 root_objectid, u64 owner, u64 offset)
1651 {
1652 int ret;
1654 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1655 bytenr, num_bytes, parent,
1656 root_objectid, owner, offset, 0);
1657 if (ret != -ENOENT)
1658 return ret;
1660 btrfs_release_path(path);
1661 *ref_ret = NULL;
1663 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1664 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1665 root_objectid);
1666 } else {
1667 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1668 root_objectid, owner, offset);
1669 }
1670 return ret;
1671 }
1673 /*
1674 * helper to update/remove inline back ref
1675 */
1676 static noinline_for_stack
1677 void update_inline_extent_backref(struct btrfs_trans_handle *trans,
1678 struct btrfs_root *root,
1679 struct btrfs_path *path,
1680 struct btrfs_extent_inline_ref *iref,
1681 int refs_to_mod,
1682 struct btrfs_delayed_extent_op *extent_op)
1683 {
1684 struct extent_buffer *leaf;
1685 struct btrfs_extent_item *ei;
1686 struct btrfs_extent_data_ref *dref = NULL;
1687 struct btrfs_shared_data_ref *sref = NULL;
1688 unsigned long ptr;
1689 unsigned long end;
1690 u32 item_size;
1691 int size;
1692 int type;
1693 u64 refs;
1695 leaf = path->nodes[0];
1696 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1697 refs = btrfs_extent_refs(leaf, ei);
1698 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1699 refs += refs_to_mod;
1700 btrfs_set_extent_refs(leaf, ei, refs);
1701 if (extent_op)
1702 __run_delayed_extent_op(extent_op, leaf, ei);
1704 type = btrfs_extent_inline_ref_type(leaf, iref);
1706 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1707 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1708 refs = btrfs_extent_data_ref_count(leaf, dref);
1709 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1710 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1711 refs = btrfs_shared_data_ref_count(leaf, sref);
1712 } else {
1713 refs = 1;
1714 BUG_ON(refs_to_mod != -1);
1715 }
1717 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1718 refs += refs_to_mod;
1720 if (refs > 0) {
1721 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1722 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1723 else
1724 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1725 } else {
1726 size = btrfs_extent_inline_ref_size(type);
1727 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1728 ptr = (unsigned long)iref;
1729 end = (unsigned long)ei + item_size;
1730 if (ptr + size < end)
1731 memmove_extent_buffer(leaf, ptr, ptr + size,
1732 end - ptr - size);
1733 item_size -= size;
1734 btrfs_truncate_item(trans, root, path, item_size, 1);
1735 }
1736 btrfs_mark_buffer_dirty(leaf);
1737 }
1739 static noinline_for_stack
1740 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1741 struct btrfs_root *root,
1742 struct btrfs_path *path,
1743 u64 bytenr, u64 num_bytes, u64 parent,
1744 u64 root_objectid, u64 owner,
1745 u64 offset, int refs_to_add,
1746 struct btrfs_delayed_extent_op *extent_op)
1747 {
1748 struct btrfs_extent_inline_ref *iref;
1749 int ret;
1751 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1752 bytenr, num_bytes, parent,
1753 root_objectid, owner, offset, 1);
1754 if (ret == 0) {
1755 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1756 update_inline_extent_backref(trans, root, path, iref,
1757 refs_to_add, extent_op);
1758 } else if (ret == -ENOENT) {
1759 setup_inline_extent_backref(trans, root, path, iref, parent,
1760 root_objectid, owner, offset,
1761 refs_to_add, extent_op);
1762 ret = 0;
1763 }
1764 return ret;
1765 }
1767 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1768 struct btrfs_root *root,
1769 struct btrfs_path *path,
1770 u64 bytenr, u64 parent, u64 root_objectid,
1771 u64 owner, u64 offset, int refs_to_add)
1772 {
1773 int ret;
1774 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1775 BUG_ON(refs_to_add != 1);
1776 ret = insert_tree_block_ref(trans, root, path, bytenr,
1777 parent, root_objectid);
1778 } else {
1779 ret = insert_extent_data_ref(trans, root, path, bytenr,
1780 parent, root_objectid,
1781 owner, offset, refs_to_add);
1782 }
1783 return ret;
1784 }
1786 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1787 struct btrfs_root *root,
1788 struct btrfs_path *path,
1789 struct btrfs_extent_inline_ref *iref,
1790 int refs_to_drop, int is_data)
1791 {
1792 int ret = 0;
1794 BUG_ON(!is_data && refs_to_drop != 1);
1795 if (iref) {
1796 update_inline_extent_backref(trans, root, path, iref,
1797 -refs_to_drop, NULL);
1798 } else if (is_data) {
1799 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1800 } else {
1801 ret = btrfs_del_item(trans, root, path);
1802 }
1803 return ret;
1804 }
1806 static int btrfs_issue_discard(struct block_device *bdev,
1807 u64 start, u64 len)
1808 {
1809 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1810 }
1812 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1813 u64 num_bytes, u64 *actual_bytes)
1814 {
1815 int ret;
1816 u64 discarded_bytes = 0;
1817 struct btrfs_bio *bbio = NULL;
1820 /* Tell the block device(s) that the sectors can be discarded */
1821 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1822 bytenr, &num_bytes, &bbio, 0);
1823 /* Error condition is -ENOMEM */
1824 if (!ret) {
1825 struct btrfs_bio_stripe *stripe = bbio->stripes;
1826 int i;
1829 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1830 if (!stripe->dev->can_discard)
1831 continue;
1833 ret = btrfs_issue_discard(stripe->dev->bdev,
1834 stripe->physical,
1835 stripe->length);
1836 if (!ret)
1837 discarded_bytes += stripe->length;
1838 else if (ret != -EOPNOTSUPP)
1839 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1841 /*
1842 * Just in case we get back EOPNOTSUPP for some reason,
1843 * just ignore the return value so we don't screw up
1844 * people calling discard_extent.
1845 */
1846 ret = 0;
1847 }
1848 kfree(bbio);
1849 }
1851 if (actual_bytes)
1852 *actual_bytes = discarded_bytes;
1855 return ret;
1856 }
1858 /* Can return -ENOMEM */
1859 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1860 struct btrfs_root *root,
1861 u64 bytenr, u64 num_bytes, u64 parent,
1862 u64 root_objectid, u64 owner, u64 offset, int for_cow)
1863 {
1864 int ret;
1865 struct btrfs_fs_info *fs_info = root->fs_info;
1867 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1868 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1870 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1871 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1872 num_bytes,
1873 parent, root_objectid, (int)owner,
1874 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1875 } else {
1876 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1877 num_bytes,
1878 parent, root_objectid, owner, offset,
1879 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1880 }
1881 return ret;
1882 }
1884 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1885 struct btrfs_root *root,
1886 u64 bytenr, u64 num_bytes,
1887 u64 parent, u64 root_objectid,
1888 u64 owner, u64 offset, int refs_to_add,
1889 struct btrfs_delayed_extent_op *extent_op)
1890 {
1891 struct btrfs_path *path;
1892 struct extent_buffer *leaf;
1893 struct btrfs_extent_item *item;
1894 u64 refs;
1895 int ret;
1896 int err = 0;
1898 path = btrfs_alloc_path();
1899 if (!path)
1900 return -ENOMEM;
1902 path->reada = 1;
1903 path->leave_spinning = 1;
1904 /* this will setup the path even if it fails to insert the back ref */
1905 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1906 path, bytenr, num_bytes, parent,
1907 root_objectid, owner, offset,
1908 refs_to_add, extent_op);
1909 if (ret == 0)
1910 goto out;
1912 if (ret != -EAGAIN) {
1913 err = ret;
1914 goto out;
1915 }
1917 leaf = path->nodes[0];
1918 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1919 refs = btrfs_extent_refs(leaf, item);
1920 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1921 if (extent_op)
1922 __run_delayed_extent_op(extent_op, leaf, item);
1924 btrfs_mark_buffer_dirty(leaf);
1925 btrfs_release_path(path);
1927 path->reada = 1;
1928 path->leave_spinning = 1;
1930 /* now insert the actual backref */
1931 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1932 path, bytenr, parent, root_objectid,
1933 owner, offset, refs_to_add);
1934 if (ret)
1935 btrfs_abort_transaction(trans, root, ret);
1936 out:
1937 btrfs_free_path(path);
1938 return err;
1939 }
1941 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1942 struct btrfs_root *root,
1943 struct btrfs_delayed_ref_node *node,
1944 struct btrfs_delayed_extent_op *extent_op,
1945 int insert_reserved)
1946 {
1947 int ret = 0;
1948 struct btrfs_delayed_data_ref *ref;
1949 struct btrfs_key ins;
1950 u64 parent = 0;
1951 u64 ref_root = 0;
1952 u64 flags = 0;
1954 ins.objectid = node->bytenr;
1955 ins.offset = node->num_bytes;
1956 ins.type = BTRFS_EXTENT_ITEM_KEY;
1958 ref = btrfs_delayed_node_to_data_ref(node);
1959 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1960 parent = ref->parent;
1961 else
1962 ref_root = ref->root;
1964 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1965 if (extent_op) {
1966 BUG_ON(extent_op->update_key);
1967 flags |= extent_op->flags_to_set;
1968 }
1969 ret = alloc_reserved_file_extent(trans, root,
1970 parent, ref_root, flags,
1971 ref->objectid, ref->offset,
1972 &ins, node->ref_mod);
1973 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1974 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1975 node->num_bytes, parent,
1976 ref_root, ref->objectid,
1977 ref->offset, node->ref_mod,
1978 extent_op);
1979 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1980 ret = __btrfs_free_extent(trans, root, node->bytenr,
1981 node->num_bytes, parent,
1982 ref_root, ref->objectid,
1983 ref->offset, node->ref_mod,
1984 extent_op);
1985 } else {
1986 BUG();
1987 }
1988 return ret;
1989 }
1991 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1992 struct extent_buffer *leaf,
1993 struct btrfs_extent_item *ei)
1994 {
1995 u64 flags = btrfs_extent_flags(leaf, ei);
1996 if (extent_op->update_flags) {
1997 flags |= extent_op->flags_to_set;
1998 btrfs_set_extent_flags(leaf, ei, flags);
1999 }
2001 if (extent_op->update_key) {
2002 struct btrfs_tree_block_info *bi;
2003 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2004 bi = (struct btrfs_tree_block_info *)(ei + 1);
2005 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2006 }
2007 }
2009 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2010 struct btrfs_root *root,
2011 struct btrfs_delayed_ref_node *node,
2012 struct btrfs_delayed_extent_op *extent_op)
2013 {
2014 struct btrfs_key key;
2015 struct btrfs_path *path;
2016 struct btrfs_extent_item *ei;
2017 struct extent_buffer *leaf;
2018 u32 item_size;
2019 int ret;
2020 int err = 0;
2022 if (trans->aborted)
2023 return 0;
2025 path = btrfs_alloc_path();
2026 if (!path)
2027 return -ENOMEM;
2029 key.objectid = node->bytenr;
2030 key.type = BTRFS_EXTENT_ITEM_KEY;
2031 key.offset = node->num_bytes;
2033 path->reada = 1;
2034 path->leave_spinning = 1;
2035 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2036 path, 0, 1);
2037 if (ret < 0) {
2038 err = ret;
2039 goto out;
2040 }
2041 if (ret > 0) {
2042 err = -EIO;
2043 goto out;
2044 }
2046 leaf = path->nodes[0];
2047 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2048 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2049 if (item_size < sizeof(*ei)) {
2050 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2051 path, (u64)-1, 0);
2052 if (ret < 0) {
2053 err = ret;
2054 goto out;
2055 }
2056 leaf = path->nodes[0];
2057 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2058 }
2059 #endif
2060 BUG_ON(item_size < sizeof(*ei));
2061 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2062 __run_delayed_extent_op(extent_op, leaf, ei);
2064 btrfs_mark_buffer_dirty(leaf);
2065 out:
2066 btrfs_free_path(path);
2067 return err;
2068 }
2070 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2071 struct btrfs_root *root,
2072 struct btrfs_delayed_ref_node *node,
2073 struct btrfs_delayed_extent_op *extent_op,
2074 int insert_reserved)
2075 {
2076 int ret = 0;
2077 struct btrfs_delayed_tree_ref *ref;
2078 struct btrfs_key ins;
2079 u64 parent = 0;
2080 u64 ref_root = 0;
2082 ins.objectid = node->bytenr;
2083 ins.offset = node->num_bytes;
2084 ins.type = BTRFS_EXTENT_ITEM_KEY;
2086 ref = btrfs_delayed_node_to_tree_ref(node);
2087 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2088 parent = ref->parent;
2089 else
2090 ref_root = ref->root;
2092 BUG_ON(node->ref_mod != 1);
2093 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2094 BUG_ON(!extent_op || !extent_op->update_flags ||
2095 !extent_op->update_key);
2096 ret = alloc_reserved_tree_block(trans, root,
2097 parent, ref_root,
2098 extent_op->flags_to_set,
2099 &extent_op->key,
2100 ref->level, &ins);
2101 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2102 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2103 node->num_bytes, parent, ref_root,
2104 ref->level, 0, 1, extent_op);
2105 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2106 ret = __btrfs_free_extent(trans, root, node->bytenr,
2107 node->num_bytes, parent, ref_root,
2108 ref->level, 0, 1, extent_op);
2109 } else {
2110 BUG();
2111 }
2112 return ret;
2113 }
2115 /* helper function to actually process a single delayed ref entry */
2116 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2117 struct btrfs_root *root,
2118 struct btrfs_delayed_ref_node *node,
2119 struct btrfs_delayed_extent_op *extent_op,
2120 int insert_reserved)
2121 {
2122 int ret = 0;
2124 if (trans->aborted)
2125 return 0;
2127 if (btrfs_delayed_ref_is_head(node)) {
2128 struct btrfs_delayed_ref_head *head;
2129 /*
2130 * we've hit the end of the chain and we were supposed
2131 * to insert this extent into the tree. But, it got
2132 * deleted before we ever needed to insert it, so all
2133 * we have to do is clean up the accounting
2134 */
2135 BUG_ON(extent_op);
2136 head = btrfs_delayed_node_to_head(node);
2137 if (insert_reserved) {
2138 btrfs_pin_extent(root, node->bytenr,
2139 node->num_bytes, 1);
2140 if (head->is_data) {
2141 ret = btrfs_del_csums(trans, root,
2142 node->bytenr,
2143 node->num_bytes);
2144 }
2145 }
2146 mutex_unlock(&head->mutex);
2147 return ret;
2148 }
2150 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2151 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2152 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2153 insert_reserved);
2154 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2155 node->type == BTRFS_SHARED_DATA_REF_KEY)
2156 ret = run_delayed_data_ref(trans, root, node, extent_op,
2157 insert_reserved);
2158 else
2159 BUG();
2160 return ret;
2161 }
2163 static noinline struct btrfs_delayed_ref_node *
2164 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2165 {
2166 struct rb_node *node;
2167 struct btrfs_delayed_ref_node *ref;
2168 int action = BTRFS_ADD_DELAYED_REF;
2169 again:
2170 /*
2171 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2172 * this prevents ref count from going down to zero when
2173 * there still are pending delayed ref.
2174 */
2175 node = rb_prev(&head->node.rb_node);
2176 while (1) {
2177 if (!node)
2178 break;
2179 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2180 rb_node);
2181 if (ref->bytenr != head->node.bytenr)
2182 break;
2183 if (ref->action == action)
2184 return ref;
2185 node = rb_prev(node);
2186 }
2187 if (action == BTRFS_ADD_DELAYED_REF) {
2188 action = BTRFS_DROP_DELAYED_REF;
2189 goto again;
2190 }
2191 return NULL;
2192 }
2194 /*
2195 * Returns 0 on success or if called with an already aborted transaction.
2196 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2197 */
2198 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2199 struct btrfs_root *root,
2200 struct list_head *cluster)
2201 {
2202 struct btrfs_delayed_ref_root *delayed_refs;
2203 struct btrfs_delayed_ref_node *ref;
2204 struct btrfs_delayed_ref_head *locked_ref = NULL;
2205 struct btrfs_delayed_extent_op *extent_op;
2206 struct btrfs_fs_info *fs_info = root->fs_info;
2207 int ret;
2208 int count = 0;
2209 int must_insert_reserved = 0;
2211 delayed_refs = &trans->transaction->delayed_refs;
2212 while (1) {
2213 if (!locked_ref) {
2214 /* pick a new head ref from the cluster list */
2215 if (list_empty(cluster))
2216 break;
2218 locked_ref = list_entry(cluster->next,
2219 struct btrfs_delayed_ref_head, cluster);
2221 /* grab the lock that says we are going to process
2222 * all the refs for this head */
2223 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2225 /*
2226 * we may have dropped the spin lock to get the head
2227 * mutex lock, and that might have given someone else
2228 * time to free the head. If that's true, it has been
2229 * removed from our list and we can move on.
2230 */
2231 if (ret == -EAGAIN) {
2232 locked_ref = NULL;
2233 count++;
2234 continue;
2235 }
2236 }
2238 /*
2239 * We need to try and merge add/drops of the same ref since we
2240 * can run into issues with relocate dropping the implicit ref
2241 * and then it being added back again before the drop can
2242 * finish. If we merged anything we need to re-loop so we can
2243 * get a good ref.
2244 */
2245 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2246 locked_ref);
2248 /*
2249 * locked_ref is the head node, so we have to go one
2250 * node back for any delayed ref updates
2251 */
2252 ref = select_delayed_ref(locked_ref);
2254 if (ref && ref->seq &&
2255 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2256 /*
2257 * there are still refs with lower seq numbers in the
2258 * process of being added. Don't run this ref yet.
2259 */
2260 list_del_init(&locked_ref->cluster);
2261 mutex_unlock(&locked_ref->mutex);
2262 locked_ref = NULL;
2263 delayed_refs->num_heads_ready++;
2264 spin_unlock(&delayed_refs->lock);
2265 cond_resched();
2266 spin_lock(&delayed_refs->lock);
2267 continue;
2268 }
2270 /*
2271 * record the must insert reserved flag before we
2272 * drop the spin lock.
2273 */
2274 must_insert_reserved = locked_ref->must_insert_reserved;
2275 locked_ref->must_insert_reserved = 0;
2277 extent_op = locked_ref->extent_op;
2278 locked_ref->extent_op = NULL;
2280 if (!ref) {
2281 /* All delayed refs have been processed, Go ahead
2282 * and send the head node to run_one_delayed_ref,
2283 * so that any accounting fixes can happen
2284 */
2285 ref = &locked_ref->node;
2287 if (extent_op && must_insert_reserved) {
2288 kfree(extent_op);
2289 extent_op = NULL;
2290 }
2292 if (extent_op) {
2293 spin_unlock(&delayed_refs->lock);
2295 ret = run_delayed_extent_op(trans, root,
2296 ref, extent_op);
2297 kfree(extent_op);
2299 if (ret) {
2300 list_del_init(&locked_ref->cluster);
2301 mutex_unlock(&locked_ref->mutex);
2303 printk(KERN_DEBUG "btrfs: run_delayed_extent_op returned %d\n", ret);
2304 spin_lock(&delayed_refs->lock);
2305 return ret;
2306 }
2308 goto next;
2309 }
2311 list_del_init(&locked_ref->cluster);
2312 locked_ref = NULL;
2313 }
2315 ref->in_tree = 0;
2316 rb_erase(&ref->rb_node, &delayed_refs->root);
2317 delayed_refs->num_entries--;
2318 if (locked_ref) {
2319 /*
2320 * when we play the delayed ref, also correct the
2321 * ref_mod on head
2322 */
2323 switch (ref->action) {
2324 case BTRFS_ADD_DELAYED_REF:
2325 case BTRFS_ADD_DELAYED_EXTENT:
2326 locked_ref->node.ref_mod -= ref->ref_mod;
2327 break;
2328 case BTRFS_DROP_DELAYED_REF:
2329 locked_ref->node.ref_mod += ref->ref_mod;
2330 break;
2331 default:
2332 WARN_ON(1);
2333 }
2334 }
2335 spin_unlock(&delayed_refs->lock);
2337 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2338 must_insert_reserved);
2340 btrfs_put_delayed_ref(ref);
2341 kfree(extent_op);
2342 count++;
2344 if (ret) {
2345 if (locked_ref) {
2346 list_del_init(&locked_ref->cluster);
2347 mutex_unlock(&locked_ref->mutex);
2348 }
2349 printk(KERN_DEBUG "btrfs: run_one_delayed_ref returned %d\n", ret);
2350 spin_lock(&delayed_refs->lock);
2351 return ret;
2352 }
2354 next:
2355 cond_resched();
2356 spin_lock(&delayed_refs->lock);
2357 }
2358 return count;
2359 }
2361 #ifdef SCRAMBLE_DELAYED_REFS
2362 /*
2363 * Normally delayed refs get processed in ascending bytenr order. This
2364 * correlates in most cases to the order added. To expose dependencies on this
2365 * order, we start to process the tree in the middle instead of the beginning
2366 */
2367 static u64 find_middle(struct rb_root *root)
2368 {
2369 struct rb_node *n = root->rb_node;
2370 struct btrfs_delayed_ref_node *entry;
2371 int alt = 1;
2372 u64 middle;
2373 u64 first = 0, last = 0;
2375 n = rb_first(root);
2376 if (n) {
2377 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2378 first = entry->bytenr;
2379 }
2380 n = rb_last(root);
2381 if (n) {
2382 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2383 last = entry->bytenr;
2384 }
2385 n = root->rb_node;
2387 while (n) {
2388 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2389 WARN_ON(!entry->in_tree);
2391 middle = entry->bytenr;
2393 if (alt)
2394 n = n->rb_left;
2395 else
2396 n = n->rb_right;
2398 alt = 1 - alt;
2399 }
2400 return middle;
2401 }
2402 #endif
2404 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2405 struct btrfs_fs_info *fs_info)
2406 {
2407 struct qgroup_update *qgroup_update;
2408 int ret = 0;
2410 if (list_empty(&trans->qgroup_ref_list) !=
2411 !trans->delayed_ref_elem.seq) {
2412 /* list without seq or seq without list */
2413 printk(KERN_ERR "btrfs: qgroup accounting update error, list is%s empty, seq is %llu\n",
2414 list_empty(&trans->qgroup_ref_list) ? "" : " not",
2415 trans->delayed_ref_elem.seq);
2416 BUG();
2417 }
2419 if (!trans->delayed_ref_elem.seq)
2420 return 0;
2422 while (!list_empty(&trans->qgroup_ref_list)) {
2423 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2424 struct qgroup_update, list);
2425 list_del(&qgroup_update->list);
2426 if (!ret)
2427 ret = btrfs_qgroup_account_ref(
2428 trans, fs_info, qgroup_update->node,
2429 qgroup_update->extent_op);
2430 kfree(qgroup_update);
2431 }
2433 btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2435 return ret;
2436 }
2438 /*
2439 * this starts processing the delayed reference count updates and
2440 * extent insertions we have queued up so far. count can be
2441 * 0, which means to process everything in the tree at the start
2442 * of the run (but not newly added entries), or it can be some target
2443 * number you'd like to process.
2444 *
2445 * Returns 0 on success or if called with an aborted transaction
2446 * Returns <0 on error and aborts the transaction
2447 */
2448 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2449 struct btrfs_root *root, unsigned long count)
2450 {
2451 struct rb_node *node;
2452 struct btrfs_delayed_ref_root *delayed_refs;
2453 struct btrfs_delayed_ref_node *ref;
2454 struct list_head cluster;
2455 int ret;
2456 u64 delayed_start;
2457 int run_all = count == (unsigned long)-1;
2458 int run_most = 0;
2459 int loops;
2461 /* We'll clean this up in btrfs_cleanup_transaction */
2462 if (trans->aborted)
2463 return 0;
2465 if (root == root->fs_info->extent_root)
2466 root = root->fs_info->tree_root;
2468 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2470 delayed_refs = &trans->transaction->delayed_refs;
2471 INIT_LIST_HEAD(&cluster);
2472 again:
2473 loops = 0;
2474 spin_lock(&delayed_refs->lock);
2476 #ifdef SCRAMBLE_DELAYED_REFS
2477 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2478 #endif
2480 if (count == 0) {
2481 count = delayed_refs->num_entries * 2;
2482 run_most = 1;
2483 }
2484 while (1) {
2485 if (!(run_all || run_most) &&
2486 delayed_refs->num_heads_ready < 64)
2487 break;
2489 /*
2490 * go find something we can process in the rbtree. We start at
2491 * the beginning of the tree, and then build a cluster
2492 * of refs to process starting at the first one we are able to
2493 * lock
2494 */
2495 delayed_start = delayed_refs->run_delayed_start;
2496 ret = btrfs_find_ref_cluster(trans, &cluster,
2497 delayed_refs->run_delayed_start);
2498 if (ret)
2499 break;
2501 ret = run_clustered_refs(trans, root, &cluster);
2502 if (ret < 0) {
2503 spin_unlock(&delayed_refs->lock);
2504 btrfs_abort_transaction(trans, root, ret);
2505 return ret;
2506 }
2508 count -= min_t(unsigned long, ret, count);
2510 if (count == 0)
2511 break;
2513 if (delayed_start >= delayed_refs->run_delayed_start) {
2514 if (loops == 0) {
2515 /*
2516 * btrfs_find_ref_cluster looped. let's do one
2517 * more cycle. if we don't run any delayed ref
2518 * during that cycle (because we can't because
2519 * all of them are blocked), bail out.
2520 */
2521 loops = 1;
2522 } else {
2523 /*
2524 * no runnable refs left, stop trying
2525 */
2526 BUG_ON(run_all);
2527 break;
2528 }
2529 }
2530 if (ret) {
2531 /* refs were run, let's reset staleness detection */
2532 loops = 0;
2533 }
2534 }
2536 if (run_all) {
2537 if (!list_empty(&trans->new_bgs)) {
2538 spin_unlock(&delayed_refs->lock);
2539 btrfs_create_pending_block_groups(trans, root);
2540 spin_lock(&delayed_refs->lock);
2541 }
2543 node = rb_first(&delayed_refs->root);
2544 if (!node)
2545 goto out;
2546 count = (unsigned long)-1;
2548 while (node) {
2549 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2550 rb_node);
2551 if (btrfs_delayed_ref_is_head(ref)) {
2552 struct btrfs_delayed_ref_head *head;
2554 head = btrfs_delayed_node_to_head(ref);
2555 atomic_inc(&ref->refs);
2557 spin_unlock(&delayed_refs->lock);
2558 /*
2559 * Mutex was contended, block until it's
2560 * released and try again
2561 */
2562 mutex_lock(&head->mutex);
2563 mutex_unlock(&head->mutex);
2565 btrfs_put_delayed_ref(ref);
2566 cond_resched();
2567 goto again;
2568 }
2569 node = rb_next(node);
2570 }
2571 spin_unlock(&delayed_refs->lock);
2572 schedule_timeout(1);
2573 goto again;
2574 }
2575 out:
2576 spin_unlock(&delayed_refs->lock);
2577 assert_qgroups_uptodate(trans);
2578 return 0;
2579 }
2581 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2582 struct btrfs_root *root,
2583 u64 bytenr, u64 num_bytes, u64 flags,
2584 int is_data)
2585 {
2586 struct btrfs_delayed_extent_op *extent_op;
2587 int ret;
2589 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2590 if (!extent_op)
2591 return -ENOMEM;
2593 extent_op->flags_to_set = flags;
2594 extent_op->update_flags = 1;
2595 extent_op->update_key = 0;
2596 extent_op->is_data = is_data ? 1 : 0;
2598 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2599 num_bytes, extent_op);
2600 if (ret)
2601 kfree(extent_op);
2602 return ret;
2603 }
2605 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2606 struct btrfs_root *root,
2607 struct btrfs_path *path,
2608 u64 objectid, u64 offset, u64 bytenr)
2609 {
2610 struct btrfs_delayed_ref_head *head;
2611 struct btrfs_delayed_ref_node *ref;
2612 struct btrfs_delayed_data_ref *data_ref;
2613 struct btrfs_delayed_ref_root *delayed_refs;
2614 struct rb_node *node;
2615 int ret = 0;
2617 ret = -ENOENT;
2618 delayed_refs = &trans->transaction->delayed_refs;
2619 spin_lock(&delayed_refs->lock);
2620 head = btrfs_find_delayed_ref_head(trans, bytenr);
2621 if (!head)
2622 goto out;
2624 if (!mutex_trylock(&head->mutex)) {
2625 atomic_inc(&head->node.refs);
2626 spin_unlock(&delayed_refs->lock);
2628 btrfs_release_path(path);
2630 /*
2631 * Mutex was contended, block until it's released and let
2632 * caller try again
2633 */
2634 mutex_lock(&head->mutex);
2635 mutex_unlock(&head->mutex);
2636 btrfs_put_delayed_ref(&head->node);
2637 return -EAGAIN;
2638 }
2640 node = rb_prev(&head->node.rb_node);
2641 if (!node)
2642 goto out_unlock;
2644 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2646 if (ref->bytenr != bytenr)
2647 goto out_unlock;
2649 ret = 1;
2650 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2651 goto out_unlock;
2653 data_ref = btrfs_delayed_node_to_data_ref(ref);
2655 node = rb_prev(node);
2656 if (node) {
2657 int seq = ref->seq;
2659 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2660 if (ref->bytenr == bytenr && ref->seq == seq)
2661 goto out_unlock;
2662 }
2664 if (data_ref->root != root->root_key.objectid ||
2665 data_ref->objectid != objectid || data_ref->offset != offset)
2666 goto out_unlock;
2668 ret = 0;
2669 out_unlock:
2670 mutex_unlock(&head->mutex);
2671 out:
2672 spin_unlock(&delayed_refs->lock);
2673 return ret;
2674 }
2676 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2677 struct btrfs_root *root,
2678 struct btrfs_path *path,
2679 u64 objectid, u64 offset, u64 bytenr)
2680 {
2681 struct btrfs_root *extent_root = root->fs_info->extent_root;
2682 struct extent_buffer *leaf;
2683 struct btrfs_extent_data_ref *ref;
2684 struct btrfs_extent_inline_ref *iref;
2685 struct btrfs_extent_item *ei;
2686 struct btrfs_key key;
2687 u32 item_size;
2688 int ret;
2690 key.objectid = bytenr;
2691 key.offset = (u64)-1;
2692 key.type = BTRFS_EXTENT_ITEM_KEY;
2694 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2695 if (ret < 0)
2696 goto out;
2697 BUG_ON(ret == 0); /* Corruption */
2699 ret = -ENOENT;
2700 if (path->slots[0] == 0)
2701 goto out;
2703 path->slots[0]--;
2704 leaf = path->nodes[0];
2705 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2707 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2708 goto out;
2710 ret = 1;
2711 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2712 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2713 if (item_size < sizeof(*ei)) {
2714 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2715 goto out;
2716 }
2717 #endif
2718 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2720 if (item_size != sizeof(*ei) +
2721 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2722 goto out;
2724 if (btrfs_extent_generation(leaf, ei) <=
2725 btrfs_root_last_snapshot(&root->root_item))
2726 goto out;
2728 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2729 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2730 BTRFS_EXTENT_DATA_REF_KEY)
2731 goto out;
2733 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2734 if (btrfs_extent_refs(leaf, ei) !=
2735 btrfs_extent_data_ref_count(leaf, ref) ||
2736 btrfs_extent_data_ref_root(leaf, ref) !=
2737 root->root_key.objectid ||
2738 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2739 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2740 goto out;
2742 ret = 0;
2743 out:
2744 return ret;
2745 }
2747 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2748 struct btrfs_root *root,
2749 u64 objectid, u64 offset, u64 bytenr)
2750 {
2751 struct btrfs_path *path;
2752 int ret;
2753 int ret2;
2755 path = btrfs_alloc_path();
2756 if (!path)
2757 return -ENOENT;
2759 do {
2760 ret = check_committed_ref(trans, root, path, objectid,
2761 offset, bytenr);
2762 if (ret && ret != -ENOENT)
2763 goto out;
2765 ret2 = check_delayed_ref(trans, root, path, objectid,
2766 offset, bytenr);
2767 } while (ret2 == -EAGAIN);
2769 if (ret2 && ret2 != -ENOENT) {
2770 ret = ret2;
2771 goto out;
2772 }
2774 if (ret != -ENOENT || ret2 != -ENOENT)
2775 ret = 0;
2776 out:
2777 btrfs_free_path(path);
2778 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2779 WARN_ON(ret > 0);
2780 return ret;
2781 }
2783 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2784 struct btrfs_root *root,
2785 struct extent_buffer *buf,
2786 int full_backref, int inc, int for_cow)
2787 {
2788 u64 bytenr;
2789 u64 num_bytes;
2790 u64 parent;
2791 u64 ref_root;
2792 u32 nritems;
2793 struct btrfs_key key;
2794 struct btrfs_file_extent_item *fi;
2795 int i;
2796 int level;
2797 int ret = 0;
2798 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2799 u64, u64, u64, u64, u64, u64, int);
2801 ref_root = btrfs_header_owner(buf);
2802 nritems = btrfs_header_nritems(buf);
2803 level = btrfs_header_level(buf);
2805 if (!root->ref_cows && level == 0)
2806 return 0;
2808 if (inc)
2809 process_func = btrfs_inc_extent_ref;
2810 else
2811 process_func = btrfs_free_extent;
2813 if (full_backref)
2814 parent = buf->start;
2815 else
2816 parent = 0;
2818 for (i = 0; i < nritems; i++) {
2819 if (level == 0) {
2820 btrfs_item_key_to_cpu(buf, &key, i);
2821 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2822 continue;
2823 fi = btrfs_item_ptr(buf, i,
2824 struct btrfs_file_extent_item);
2825 if (btrfs_file_extent_type(buf, fi) ==
2826 BTRFS_FILE_EXTENT_INLINE)
2827 continue;
2828 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2829 if (bytenr == 0)
2830 continue;
2832 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2833 key.offset -= btrfs_file_extent_offset(buf, fi);
2834 ret = process_func(trans, root, bytenr, num_bytes,
2835 parent, ref_root, key.objectid,
2836 key.offset, for_cow);
2837 if (ret)
2838 goto fail;
2839 } else {
2840 bytenr = btrfs_node_blockptr(buf, i);
2841 num_bytes = btrfs_level_size(root, level - 1);
2842 ret = process_func(trans, root, bytenr, num_bytes,
2843 parent, ref_root, level - 1, 0,
2844 for_cow);
2845 if (ret)
2846 goto fail;
2847 }
2848 }
2849 return 0;
2850 fail:
2851 return ret;
2852 }
2854 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2855 struct extent_buffer *buf, int full_backref, int for_cow)
2856 {
2857 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
2858 }
2860 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2861 struct extent_buffer *buf, int full_backref, int for_cow)
2862 {
2863 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
2864 }
2866 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2867 struct btrfs_root *root,
2868 struct btrfs_path *path,
2869 struct btrfs_block_group_cache *cache)
2870 {
2871 int ret;
2872 struct btrfs_root *extent_root = root->fs_info->extent_root;
2873 unsigned long bi;
2874 struct extent_buffer *leaf;
2876 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2877 if (ret < 0)
2878 goto fail;
2879 BUG_ON(ret); /* Corruption */
2881 leaf = path->nodes[0];
2882 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2883 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2884 btrfs_mark_buffer_dirty(leaf);
2885 btrfs_release_path(path);
2886 fail:
2887 if (ret) {
2888 btrfs_abort_transaction(trans, root, ret);
2889 return ret;
2890 }
2891 return 0;
2893 }
2895 static struct btrfs_block_group_cache *
2896 next_block_group(struct btrfs_root *root,
2897 struct btrfs_block_group_cache *cache)
2898 {
2899 struct rb_node *node;
2900 spin_lock(&root->fs_info->block_group_cache_lock);
2901 node = rb_next(&cache->cache_node);
2902 btrfs_put_block_group(cache);
2903 if (node) {
2904 cache = rb_entry(node, struct btrfs_block_group_cache,
2905 cache_node);
2906 btrfs_get_block_group(cache);
2907 } else
2908 cache = NULL;
2909 spin_unlock(&root->fs_info->block_group_cache_lock);
2910 return cache;
2911 }
2913 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2914 struct btrfs_trans_handle *trans,
2915 struct btrfs_path *path)
2916 {
2917 struct btrfs_root *root = block_group->fs_info->tree_root;
2918 struct inode *inode = NULL;
2919 u64 alloc_hint = 0;
2920 int dcs = BTRFS_DC_ERROR;
2921 int num_pages = 0;
2922 int retries = 0;
2923 int ret = 0;
2925 /*
2926 * If this block group is smaller than 100 megs don't bother caching the
2927 * block group.
2928 */
2929 if (block_group->key.offset < (100 * 1024 * 1024)) {
2930 spin_lock(&block_group->lock);
2931 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2932 spin_unlock(&block_group->lock);
2933 return 0;
2934 }
2936 again:
2937 inode = lookup_free_space_inode(root, block_group, path);
2938 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2939 ret = PTR_ERR(inode);
2940 btrfs_release_path(path);
2941 goto out;
2942 }
2944 if (IS_ERR(inode)) {
2945 BUG_ON(retries);
2946 retries++;
2948 if (block_group->ro)
2949 goto out_free;
2951 ret = create_free_space_inode(root, trans, block_group, path);
2952 if (ret)
2953 goto out_free;
2954 goto again;
2955 }
2957 /* We've already setup this transaction, go ahead and exit */
2958 if (block_group->cache_generation == trans->transid &&
2959 i_size_read(inode)) {
2960 dcs = BTRFS_DC_SETUP;
2961 goto out_put;
2962 }
2964 /*
2965 * We want to set the generation to 0, that way if anything goes wrong
2966 * from here on out we know not to trust this cache when we load up next
2967 * time.
2968 */
2969 BTRFS_I(inode)->generation = 0;
2970 ret = btrfs_update_inode(trans, root, inode);
2971 WARN_ON(ret);
2973 if (i_size_read(inode) > 0) {
2974 ret = btrfs_truncate_free_space_cache(root, trans, path,
2975 inode);
2976 if (ret)
2977 goto out_put;
2978 }
2980 spin_lock(&block_group->lock);
2981 if (block_group->cached != BTRFS_CACHE_FINISHED ||
2982 !btrfs_test_opt(root, SPACE_CACHE)) {
2983 /*
2984 * don't bother trying to write stuff out _if_
2985 * a) we're not cached,
2986 * b) we're with nospace_cache mount option.
2987 */
2988 dcs = BTRFS_DC_WRITTEN;
2989 spin_unlock(&block_group->lock);
2990 goto out_put;
2991 }
2992 spin_unlock(&block_group->lock);
2994 /*
2995 * Try to preallocate enough space based on how big the block group is.
2996 * Keep in mind this has to include any pinned space which could end up
2997 * taking up quite a bit since it's not folded into the other space
2998 * cache.
2999 */
3000 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3001 if (!num_pages)
3002 num_pages = 1;
3004 num_pages *= 16;
3005 num_pages *= PAGE_CACHE_SIZE;
3007 ret = btrfs_check_data_free_space(inode, num_pages);
3008 if (ret)
3009 goto out_put;
3011 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3012 num_pages, num_pages,
3013 &alloc_hint);
3014 if (!ret)
3015 dcs = BTRFS_DC_SETUP;
3016 btrfs_free_reserved_data_space(inode, num_pages);
3018 out_put:
3019 iput(inode);
3020 out_free:
3021 btrfs_release_path(path);
3022 out:
3023 spin_lock(&block_group->lock);
3024 if (!ret && dcs == BTRFS_DC_SETUP)
3025 block_group->cache_generation = trans->transid;
3026 block_group->disk_cache_state = dcs;
3027 spin_unlock(&block_group->lock);
3029 return ret;
3030 }
3032 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3033 struct btrfs_root *root)
3034 {
3035 struct btrfs_block_group_cache *cache;
3036 int err = 0;
3037 struct btrfs_path *path;
3038 u64 last = 0;
3040 path = btrfs_alloc_path();
3041 if (!path)
3042 return -ENOMEM;
3044 again:
3045 while (1) {
3046 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3047 while (cache) {
3048 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3049 break;
3050 cache = next_block_group(root, cache);
3051 }
3052 if (!cache) {
3053 if (last == 0)
3054 break;
3055 last = 0;
3056 continue;
3057 }
3058 err = cache_save_setup(cache, trans, path);
3059 last = cache->key.objectid + cache->key.offset;
3060 btrfs_put_block_group(cache);
3061 }
3063 while (1) {
3064 if (last == 0) {
3065 err = btrfs_run_delayed_refs(trans, root,
3066 (unsigned long)-1);
3067 if (err) /* File system offline */
3068 goto out;
3069 }
3071 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3072 while (cache) {
3073 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3074 btrfs_put_block_group(cache);
3075 goto again;
3076 }
3078 if (cache->dirty)
3079 break;
3080 cache = next_block_group(root, cache);
3081 }
3082 if (!cache) {
3083 if (last == 0)
3084 break;
3085 last = 0;
3086 continue;
3087 }
3089 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3090 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3091 cache->dirty = 0;
3092 last = cache->key.objectid + cache->key.offset;
3094 err = write_one_cache_group(trans, root, path, cache);
3095 if (err) /* File system offline */
3096 goto out;
3098 btrfs_put_block_group(cache);
3099 }
3101 while (1) {
3102 /*
3103 * I don't think this is needed since we're just marking our
3104 * preallocated extent as written, but just in case it can't
3105 * hurt.
3106 */
3107 if (last == 0) {
3108 err = btrfs_run_delayed_refs(trans, root,
3109 (unsigned long)-1);
3110 if (err) /* File system offline */
3111 goto out;
3112 }
3114 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3115 while (cache) {
3116 /*
3117 * Really this shouldn't happen, but it could if we
3118 * couldn't write the entire preallocated extent and
3119 * splitting the extent resulted in a new block.
3120 */
3121 if (cache->dirty) {
3122 btrfs_put_block_group(cache);
3123 goto again;
3124 }
3125 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3126 break;
3127 cache = next_block_group(root, cache);
3128 }
3129 if (!cache) {
3130 if (last == 0)
3131 break;
3132 last = 0;
3133 continue;
3134 }
3136 err = btrfs_write_out_cache(root, trans, cache, path);
3138 /*
3139 * If we didn't have an error then the cache state is still
3140 * NEED_WRITE, so we can set it to WRITTEN.
3141 */
3142 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3143 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3144 last = cache->key.objectid + cache->key.offset;
3145 btrfs_put_block_group(cache);
3146 }
3147 out:
3149 btrfs_free_path(path);
3150 return err;
3151 }
3153 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3154 {
3155 struct btrfs_block_group_cache *block_group;
3156 int readonly = 0;
3158 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3159 if (!block_group || block_group->ro)
3160 readonly = 1;
3161 if (block_group)
3162 btrfs_put_block_group(block_group);
3163 return readonly;
3164 }
3166 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3167 u64 total_bytes, u64 bytes_used,
3168 struct btrfs_space_info **space_info)
3169 {
3170 struct btrfs_space_info *found;
3171 int i;
3172 int factor;
3174 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3175 BTRFS_BLOCK_GROUP_RAID10))
3176 factor = 2;
3177 else
3178 factor = 1;
3180 found = __find_space_info(info, flags);
3181 if (found) {
3182 spin_lock(&found->lock);
3183 found->total_bytes += total_bytes;
3184 found->disk_total += total_bytes * factor;
3185 found->bytes_used += bytes_used;
3186 found->disk_used += bytes_used * factor;
3187 found->full = 0;
3188 spin_unlock(&found->lock);
3189 *space_info = found;
3190 return 0;
3191 }
3192 found = kzalloc(sizeof(*found), GFP_NOFS);
3193 if (!found)
3194 return -ENOMEM;
3196 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3197 INIT_LIST_HEAD(&found->block_groups[i]);
3198 init_rwsem(&found->groups_sem);
3199 spin_lock_init(&found->lock);
3200 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3201 found->total_bytes = total_bytes;
3202 found->disk_total = total_bytes * factor;
3203 found->bytes_used = bytes_used;
3204 found->disk_used = bytes_used * factor;
3205 found->bytes_pinned = 0;
3206 found->bytes_reserved = 0;
3207 found->bytes_readonly = 0;
3208 found->bytes_may_use = 0;
3209 found->full = 0;
3210 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3211 found->chunk_alloc = 0;
3212 found->flush = 0;
3213 init_waitqueue_head(&found->wait);
3214 *space_info = found;
3215 list_add_rcu(&found->list, &info->space_info);
3216 if (flags & BTRFS_BLOCK_GROUP_DATA)
3217 info->data_sinfo = found;
3218 return 0;
3219 }
3221 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3222 {
3223 u64 extra_flags = chunk_to_extended(flags) &
3224 BTRFS_EXTENDED_PROFILE_MASK;
3226 if (flags & BTRFS_BLOCK_GROUP_DATA)
3227 fs_info->avail_data_alloc_bits |= extra_flags;
3228 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3229 fs_info->avail_metadata_alloc_bits |= extra_flags;
3230 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3231 fs_info->avail_system_alloc_bits |= extra_flags;
3232 }
3234 /*
3235 * returns target flags in extended format or 0 if restripe for this
3236 * chunk_type is not in progress
3237 *
3238 * should be called with either volume_mutex or balance_lock held
3239 */
3240 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3241 {
3242 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3243 u64 target = 0;
3245 if (!bctl)
3246 return 0;
3248 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3249 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3250 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3251 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3252 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3253 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3254 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3255 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3256 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3257 }
3259 return target;
3260 }
3262 /*
3263 * @flags: available profiles in extended format (see ctree.h)
3264 *
3265 * Returns reduced profile in chunk format. If profile changing is in
3266 * progress (either running or paused) picks the target profile (if it's
3267 * already available), otherwise falls back to plain reducing.
3268 */
3269 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3270 {
3271 /*
3272 * we add in the count of missing devices because we want
3273 * to make sure that any RAID levels on a degraded FS
3274 * continue to be honored.
3275 */
3276 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3277 root->fs_info->fs_devices->missing_devices;
3278 u64 target;
3280 /*
3281 * see if restripe for this chunk_type is in progress, if so
3282 * try to reduce to the target profile
3283 */
3284 spin_lock(&root->fs_info->balance_lock);
3285 target = get_restripe_target(root->fs_info, flags);
3286 if (target) {
3287 /* pick target profile only if it's already available */
3288 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3289 spin_unlock(&root->fs_info->balance_lock);
3290 return extended_to_chunk(target);
3291 }
3292 }
3293 spin_unlock(&root->fs_info->balance_lock);
3295 if (num_devices == 1)
3296 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3297 if (num_devices < 4)
3298 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3300 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3301 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3302 BTRFS_BLOCK_GROUP_RAID10))) {
3303 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3304 }
3306 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3307 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3308 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3309 }
3311 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3312 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3313 (flags & BTRFS_BLOCK_GROUP_RAID10) |
3314 (flags & BTRFS_BLOCK_GROUP_DUP))) {
3315 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3316 }
3318 return extended_to_chunk(flags);
3319 }
3321 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3322 {
3323 if (flags & BTRFS_BLOCK_GROUP_DATA)
3324 flags |= root->fs_info->avail_data_alloc_bits;
3325 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3326 flags |= root->fs_info->avail_system_alloc_bits;
3327 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3328 flags |= root->fs_info->avail_metadata_alloc_bits;
3330 return btrfs_reduce_alloc_profile(root, flags);
3331 }
3333 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3334 {
3335 u64 flags;
3337 if (data)
3338 flags = BTRFS_BLOCK_GROUP_DATA;
3339 else if (root == root->fs_info->chunk_root)
3340 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3341 else
3342 flags = BTRFS_BLOCK_GROUP_METADATA;
3344 return get_alloc_profile(root, flags);
3345 }
3347 /*
3348 * This will check the space that the inode allocates from to make sure we have
3349 * enough space for bytes.
3350 */
3351 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3352 {
3353 struct btrfs_space_info *data_sinfo;
3354 struct btrfs_root *root = BTRFS_I(inode)->root;
3355 struct btrfs_fs_info *fs_info = root->fs_info;
3356 u64 used;
3357 int ret = 0, committed = 0, alloc_chunk = 1;
3359 /* make sure bytes are sectorsize aligned */
3360 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3362 if (root == root->fs_info->tree_root ||
3363 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3364 alloc_chunk = 0;
3365 committed = 1;
3366 }
3368 data_sinfo = fs_info->data_sinfo;
3369 if (!data_sinfo)
3370 goto alloc;
3372 again:
3373 /* make sure we have enough space to handle the data first */
3374 spin_lock(&data_sinfo->lock);
3375 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3376 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3377 data_sinfo->bytes_may_use;
3379 if (used + bytes > data_sinfo->total_bytes) {
3380 struct btrfs_trans_handle *trans;
3382 /*
3383 * if we don't have enough free bytes in this space then we need
3384 * to alloc a new chunk.
3385 */
3386 if (!data_sinfo->full && alloc_chunk) {
3387 u64 alloc_target;
3389 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3390 spin_unlock(&data_sinfo->lock);
3391 alloc:
3392 alloc_target = btrfs_get_alloc_profile(root, 1);
3393 trans = btrfs_join_transaction(root);
3394 if (IS_ERR(trans))
3395 return PTR_ERR(trans);
3397 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3398 alloc_target,
3399 CHUNK_ALLOC_NO_FORCE);
3400 btrfs_end_transaction(trans, root);
3401 if (ret < 0) {
3402 if (ret != -ENOSPC)
3403 return ret;
3404 else
3405 goto commit_trans;
3406 }
3408 if (!data_sinfo)
3409 data_sinfo = fs_info->data_sinfo;
3411 goto again;
3412 }
3414 /*
3415 * If we have less pinned bytes than we want to allocate then
3416 * don't bother committing the transaction, it won't help us.
3417 */
3418 if (data_sinfo->bytes_pinned < bytes)
3419 committed = 1;
3420 spin_unlock(&data_sinfo->lock);
3422 /* commit the current transaction and try again */
3423 commit_trans:
3424 if (!committed &&
3425 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3426 committed = 1;
3427 trans = btrfs_join_transaction(root);
3428 if (IS_ERR(trans))
3429 return PTR_ERR(trans);
3430 ret = btrfs_commit_transaction(trans, root);
3431 if (ret)
3432 return ret;
3433 goto again;
3434 }
3436 return -ENOSPC;
3437 }
3438 data_sinfo->bytes_may_use += bytes;
3439 trace_btrfs_space_reservation(root->fs_info, "space_info",
3440 data_sinfo->flags, bytes, 1);
3441 spin_unlock(&data_sinfo->lock);
3443 return 0;
3444 }
3446 /*
3447 * Called if we need to clear a data reservation for this inode.
3448 */
3449 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3450 {
3451 struct btrfs_root *root = BTRFS_I(inode)->root;
3452 struct btrfs_space_info *data_sinfo;
3454 /* make sure bytes are sectorsize aligned */
3455 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3457 data_sinfo = root->fs_info->data_sinfo;
3458 spin_lock(&data_sinfo->lock);
3459 data_sinfo->bytes_may_use -= bytes;
3460 trace_btrfs_space_reservation(root->fs_info, "space_info",
3461 data_sinfo->flags, bytes, 0);
3462 spin_unlock(&data_sinfo->lock);
3463 }
3465 static void force_metadata_allocation(struct btrfs_fs_info *info)
3466 {
3467 struct list_head *head = &info->space_info;
3468 struct btrfs_space_info *found;
3470 rcu_read_lock();
3471 list_for_each_entry_rcu(found, head, list) {
3472 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3473 found->force_alloc = CHUNK_ALLOC_FORCE;
3474 }
3475 rcu_read_unlock();
3476 }
3478 static int should_alloc_chunk(struct btrfs_root *root,
3479 struct btrfs_space_info *sinfo, int force)
3480 {
3481 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3482 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3483 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3484 u64 thresh;
3486 if (force == CHUNK_ALLOC_FORCE)
3487 return 1;
3489 /*
3490 * We need to take into account the global rsv because for all intents
3491 * and purposes it's used space. Don't worry about locking the
3492 * global_rsv, it doesn't change except when the transaction commits.
3493 */
3494 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3495 num_allocated += global_rsv->size;
3497 /*
3498 * in limited mode, we want to have some free space up to
3499 * about 1% of the FS size.
3500 */
3501 if (force == CHUNK_ALLOC_LIMITED) {
3502 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3503 thresh = max_t(u64, 64 * 1024 * 1024,
3504 div_factor_fine(thresh, 1));
3506 if (num_bytes - num_allocated < thresh)
3507 return 1;
3508 }
3510 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3511 return 0;
3512 return 1;
3513 }
3515 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3516 {
3517 u64 num_dev;
3519 if (type & BTRFS_BLOCK_GROUP_RAID10 ||
3520 type & BTRFS_BLOCK_GROUP_RAID0)
3521 num_dev = root->fs_info->fs_devices->rw_devices;
3522 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3523 num_dev = 2;
3524 else
3525 num_dev = 1; /* DUP or single */
3527 /* metadata for updaing devices and chunk tree */
3528 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3529 }
3531 static void check_system_chunk(struct btrfs_trans_handle *trans,
3532 struct btrfs_root *root, u64 type)
3533 {
3534 struct btrfs_space_info *info;
3535 u64 left;
3536 u64 thresh;
3538 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3539 spin_lock(&info->lock);
3540 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3541 info->bytes_reserved - info->bytes_readonly;
3542 spin_unlock(&info->lock);
3544 thresh = get_system_chunk_thresh(root, type);
3545 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3546 printk(KERN_INFO "left=%llu, need=%llu, flags=%llu\n",
3547 left, thresh, type);
3548 dump_space_info(info, 0, 0);
3549 }
3551 if (left < thresh) {
3552 u64 flags;
3554 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3555 btrfs_alloc_chunk(trans, root, flags);
3556 }
3557 }
3559 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3560 struct btrfs_root *extent_root, u64 flags, int force)
3561 {
3562 struct btrfs_space_info *space_info;
3563 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3564 int wait_for_alloc = 0;
3565 int ret = 0;
3567 space_info = __find_space_info(extent_root->fs_info, flags);
3568 if (!space_info) {
3569 ret = update_space_info(extent_root->fs_info, flags,
3570 0, 0, &space_info);
3571 BUG_ON(ret); /* -ENOMEM */
3572 }
3573 BUG_ON(!space_info); /* Logic error */
3575 again:
3576 spin_lock(&space_info->lock);
3577 if (force < space_info->force_alloc)
3578 force = space_info->force_alloc;
3579 if (space_info->full) {
3580 spin_unlock(&space_info->lock);
3581 return 0;
3582 }
3584 if (!should_alloc_chunk(extent_root, space_info, force)) {
3585 spin_unlock(&space_info->lock);
3586 return 0;
3587 } else if (space_info->chunk_alloc) {
3588 wait_for_alloc = 1;
3589 } else {
3590 space_info->chunk_alloc = 1;
3591 }
3593 spin_unlock(&space_info->lock);
3595 mutex_lock(&fs_info->chunk_mutex);
3597 /*
3598 * The chunk_mutex is held throughout the entirety of a chunk
3599 * allocation, so once we've acquired the chunk_mutex we know that the
3600 * other guy is done and we need to recheck and see if we should
3601 * allocate.
3602 */
3603 if (wait_for_alloc) {
3604 mutex_unlock(&fs_info->chunk_mutex);
3605 wait_for_alloc = 0;
3606 goto again;
3607 }
3609 /*
3610 * If we have mixed data/metadata chunks we want to make sure we keep
3611 * allocating mixed chunks instead of individual chunks.
3612 */
3613 if (btrfs_mixed_space_info(space_info))
3614 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3616 /*
3617 * if we're doing a data chunk, go ahead and make sure that
3618 * we keep a reasonable number of metadata chunks allocated in the
3619 * FS as well.
3620 */
3621 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3622 fs_info->data_chunk_allocations++;
3623 if (!(fs_info->data_chunk_allocations %
3624 fs_info->metadata_ratio))
3625 force_metadata_allocation(fs_info);
3626 }
3628 /*
3629 * Check if we have enough space in SYSTEM chunk because we may need
3630 * to update devices.
3631 */
3632 check_system_chunk(trans, extent_root, flags);
3634 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3635 if (ret < 0 && ret != -ENOSPC)
3636 goto out;
3638 spin_lock(&space_info->lock);
3639 if (ret)
3640 space_info->full = 1;
3641 else
3642 ret = 1;
3644 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3645 space_info->chunk_alloc = 0;
3646 spin_unlock(&space_info->lock);
3647 out:
3648 mutex_unlock(&fs_info->chunk_mutex);
3649 return ret;
3650 }
3652 static int can_overcommit(struct btrfs_root *root,
3653 struct btrfs_space_info *space_info, u64 bytes,
3654 enum btrfs_reserve_flush_enum flush)
3655 {
3656 u64 profile = btrfs_get_alloc_profile(root, 0);
3657 u64 avail;
3658 u64 used;
3660 used = space_info->bytes_used + space_info->bytes_reserved +
3661 space_info->bytes_pinned + space_info->bytes_readonly +
3662 space_info->bytes_may_use;
3664 spin_lock(&root->fs_info->free_chunk_lock);
3665 avail = root->fs_info->free_chunk_space;
3666 spin_unlock(&root->fs_info->free_chunk_lock);
3668 /*
3669 * If we have dup, raid1 or raid10 then only half of the free
3670 * space is actually useable.
3671 */
3672 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3673 BTRFS_BLOCK_GROUP_RAID1 |
3674 BTRFS_BLOCK_GROUP_RAID10))
3675 avail >>= 1;
3677 /*
3678 * If we aren't flushing all things, let us overcommit up to
3679 * 1/2th of the space. If we can flush, don't let us overcommit
3680 * too much, let it overcommit up to 1/8 of the space.
3681 */
3682 if (flush == BTRFS_RESERVE_FLUSH_ALL)
3683 avail >>= 3;
3684 else
3685 avail >>= 1;
3687 if (used + bytes < space_info->total_bytes + avail)
3688 return 1;
3689 return 0;
3690 }
3692 static int writeback_inodes_sb_nr_if_idle_safe(struct super_block *sb,
3693 unsigned long nr_pages,
3694 enum wb_reason reason)
3695 {
3696 if (!writeback_in_progress(sb->s_bdi) &&
3697 down_read_trylock(&sb->s_umount)) {
3698 writeback_inodes_sb_nr(sb, nr_pages, reason);
3699 up_read(&sb->s_umount);
3700 return 1;
3701 }
3703 return 0;
3704 }
3706 /*
3707 * shrink metadata reservation for delalloc
3708 */
3709 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
3710 bool wait_ordered)
3711 {
3712 struct btrfs_block_rsv *block_rsv;
3713 struct btrfs_space_info *space_info;
3714 struct btrfs_trans_handle *trans;
3715 u64 delalloc_bytes;
3716 u64 max_reclaim;
3717 long time_left;
3718 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3719 int loops = 0;
3720 enum btrfs_reserve_flush_enum flush;
3722 trans = (struct btrfs_trans_handle *)current->journal_info;
3723 block_rsv = &root->fs_info->delalloc_block_rsv;
3724 space_info = block_rsv->space_info;
3726 smp_mb();
3727 delalloc_bytes = root->fs_info->delalloc_bytes;
3728 if (delalloc_bytes == 0) {
3729 if (trans)
3730 return;
3731 btrfs_wait_ordered_extents(root, 0);
3732 return;
3733 }
3735 while (delalloc_bytes && loops < 3) {
3736 max_reclaim = min(delalloc_bytes, to_reclaim);
3737 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
3738 writeback_inodes_sb_nr_if_idle_safe(root->fs_info->sb,
3739 nr_pages,
3740 WB_REASON_FS_FREE_SPACE);
3742 /*
3743 * We need to wait for the async pages to actually start before
3744 * we do anything.
3745 */
3746 wait_event(root->fs_info->async_submit_wait,
3747 !atomic_read(&root->fs_info->async_delalloc_pages));
3749 if (!trans)
3750 flush = BTRFS_RESERVE_FLUSH_ALL;
3751 else
3752 flush = BTRFS_RESERVE_NO_FLUSH;
3753 spin_lock(&space_info->lock);
3754 if (can_overcommit(root, space_info, orig, flush)) {
3755 spin_unlock(&space_info->lock);
3756 break;
3757 }
3758 spin_unlock(&space_info->lock);
3760 loops++;
3761 if (wait_ordered && !trans) {
3762 btrfs_wait_ordered_extents(root, 0);
3763 } else {
3764 time_left = schedule_timeout_killable(1);
3765 if (time_left)
3766 break;
3767 }
3768 smp_mb();
3769 delalloc_bytes = root->fs_info->delalloc_bytes;
3770 }
3771 }
3773 /**
3774 * maybe_commit_transaction - possibly commit the transaction if its ok to
3775 * @root - the root we're allocating for
3776 * @bytes - the number of bytes we want to reserve
3777 * @force - force the commit
3778 *
3779 * This will check to make sure that committing the transaction will actually
3780 * get us somewhere and then commit the transaction if it does. Otherwise it
3781 * will return -ENOSPC.
3782 */
3783 static int may_commit_transaction(struct btrfs_root *root,
3784 struct btrfs_space_info *space_info,
3785 u64 bytes, int force)
3786 {
3787 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
3788 struct btrfs_trans_handle *trans;
3790 trans = (struct btrfs_trans_handle *)current->journal_info;
3791 if (trans)
3792 return -EAGAIN;
3794 if (force)
3795 goto commit;
3797 /* See if there is enough pinned space to make this reservation */
3798 spin_lock(&space_info->lock);
3799 if (space_info->bytes_pinned >= bytes) {
3800 spin_unlock(&space_info->lock);
3801 goto commit;
3802 }
3803 spin_unlock(&space_info->lock);
3805 /*
3806 * See if there is some space in the delayed insertion reservation for
3807 * this reservation.
3808 */
3809 if (space_info != delayed_rsv->space_info)
3810 return -ENOSPC;
3812 spin_lock(&space_info->lock);
3813 spin_lock(&delayed_rsv->lock);
3814 if (space_info->bytes_pinned + delayed_rsv->size < bytes) {
3815 spin_unlock(&delayed_rsv->lock);
3816 spin_unlock(&space_info->lock);
3817 return -ENOSPC;
3818 }
3819 spin_unlock(&delayed_rsv->lock);
3820 spin_unlock(&space_info->lock);
3822 commit:
3823 trans = btrfs_join_transaction(root);
3824 if (IS_ERR(trans))
3825 return -ENOSPC;
3827 return btrfs_commit_transaction(trans, root);
3828 }
3830 enum flush_state {
3831 FLUSH_DELAYED_ITEMS_NR = 1,
3832 FLUSH_DELAYED_ITEMS = 2,
3833 FLUSH_DELALLOC = 3,
3834 FLUSH_DELALLOC_WAIT = 4,
3835 ALLOC_CHUNK = 5,
3836 COMMIT_TRANS = 6,
3837 };
3839 static int flush_space(struct btrfs_root *root,
3840 struct btrfs_space_info *space_info, u64 num_bytes,
3841 u64 orig_bytes, int state)
3842 {
3843 struct btrfs_trans_handle *trans;
3844 int nr;
3845 int ret = 0;
3847 switch (state) {
3848 case FLUSH_DELAYED_ITEMS_NR:
3849 case FLUSH_DELAYED_ITEMS:
3850 if (state == FLUSH_DELAYED_ITEMS_NR) {
3851 u64 bytes = btrfs_calc_trans_metadata_size(root, 1);
3853 nr = (int)div64_u64(num_bytes, bytes);
3854 if (!nr)
3855 nr = 1;
3856 nr *= 2;
3857 } else {
3858 nr = -1;
3859 }
3860 trans = btrfs_join_transaction(root);
3861 if (IS_ERR(trans)) {
3862 ret = PTR_ERR(trans);
3863 break;
3864 }
3865 ret = btrfs_run_delayed_items_nr(trans, root, nr);
3866 btrfs_end_transaction(trans, root);
3867 break;
3868 case FLUSH_DELALLOC:
3869 case FLUSH_DELALLOC_WAIT:
3870 shrink_delalloc(root, num_bytes, orig_bytes,
3871 state == FLUSH_DELALLOC_WAIT);
3872 break;
3873 case ALLOC_CHUNK:
3874 trans = btrfs_join_transaction(root);
3875 if (IS_ERR(trans)) {
3876 ret = PTR_ERR(trans);
3877 break;
3878 }
3879 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3880 btrfs_get_alloc_profile(root, 0),
3881 CHUNK_ALLOC_NO_FORCE);
3882 btrfs_end_transaction(trans, root);
3883 if (ret == -ENOSPC)
3884 ret = 0;
3885 break;
3886 case COMMIT_TRANS:
3887 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
3888 break;
3889 default:
3890 ret = -ENOSPC;
3891 break;
3892 }
3894 return ret;
3895 }
3896 /**
3897 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3898 * @root - the root we're allocating for
3899 * @block_rsv - the block_rsv we're allocating for
3900 * @orig_bytes - the number of bytes we want
3901 * @flush - whether or not we can flush to make our reservation
3902 *
3903 * This will reserve orgi_bytes number of bytes from the space info associated
3904 * with the block_rsv. If there is not enough space it will make an attempt to
3905 * flush out space to make room. It will do this by flushing delalloc if
3906 * possible or committing the transaction. If flush is 0 then no attempts to
3907 * regain reservations will be made and this will fail if there is not enough
3908 * space already.
3909 */
3910 static int reserve_metadata_bytes(struct btrfs_root *root,
3911 struct btrfs_block_rsv *block_rsv,
3912 u64 orig_bytes,
3913 enum btrfs_reserve_flush_enum flush)
3914 {
3915 struct btrfs_space_info *space_info = block_rsv->space_info;
3916 u64 used;
3917 u64 num_bytes = orig_bytes;
3918 int flush_state = FLUSH_DELAYED_ITEMS_NR;
3919 int ret = 0;
3920 bool flushing = false;
3922 again:
3923 ret = 0;
3924 spin_lock(&space_info->lock);
3925 /*
3926 * We only want to wait if somebody other than us is flushing and we
3927 * are actually allowed to flush all things.
3928 */
3929 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
3930 space_info->flush) {
3931 spin_unlock(&space_info->lock);
3932 /*
3933 * If we have a trans handle we can't wait because the flusher
3934 * may have to commit the transaction, which would mean we would
3935 * deadlock since we are waiting for the flusher to finish, but
3936 * hold the current transaction open.
3937 */
3938 if (current->journal_info)
3939 return -EAGAIN;
3940 ret = wait_event_killable(space_info->wait, !space_info->flush);
3941 /* Must have been killed, return */
3942 if (ret)
3943 return -EINTR;
3945 spin_lock(&space_info->lock);
3946 }
3948 ret = -ENOSPC;
3949 used = space_info->bytes_used + space_info->bytes_reserved +
3950 space_info->bytes_pinned + space_info->bytes_readonly +
3951 space_info->bytes_may_use;
3953 /*
3954 * The idea here is that we've not already over-reserved the block group
3955 * then we can go ahead and save our reservation first and then start
3956 * flushing if we need to. Otherwise if we've already overcommitted
3957 * lets start flushing stuff first and then come back and try to make
3958 * our reservation.
3959 */
3960 if (used <= space_info->total_bytes) {
3961 if (used + orig_bytes <= space_info->total_bytes) {
3962 space_info->bytes_may_use += orig_bytes;
3963 trace_btrfs_space_reservation(root->fs_info,
3964 "space_info", space_info->flags, orig_bytes, 1);
3965 ret = 0;
3966 } else {
3967 /*
3968 * Ok set num_bytes to orig_bytes since we aren't
3969 * overocmmitted, this way we only try and reclaim what
3970 * we need.
3971 */
3972 num_bytes = orig_bytes;
3973 }
3974 } else {
3975 /*
3976 * Ok we're over committed, set num_bytes to the overcommitted
3977 * amount plus the amount of bytes that we need for this
3978 * reservation.
3979 */
3980 num_bytes = used - space_info->total_bytes +
3981 (orig_bytes * 2);
3982 }
3984 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
3985 space_info->bytes_may_use += orig_bytes;
3986 trace_btrfs_space_reservation(root->fs_info, "space_info",
3987 space_info->flags, orig_bytes,
3988 1);
3989 ret = 0;
3990 }
3992 /*
3993 * Couldn't make our reservation, save our place so while we're trying
3994 * to reclaim space we can actually use it instead of somebody else
3995 * stealing it from us.
3996 *
3997 * We make the other tasks wait for the flush only when we can flush
3998 * all things.
3999 */
4000 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4001 flushing = true;
4002 space_info->flush = 1;
4003 }
4005 spin_unlock(&space_info->lock);
4007 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4008 goto out;
4010 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4011 flush_state);
4012 flush_state++;
4014 /*
4015 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4016 * would happen. So skip delalloc flush.
4017 */
4018 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4019 (flush_state == FLUSH_DELALLOC ||
4020 flush_state == FLUSH_DELALLOC_WAIT))
4021 flush_state = ALLOC_CHUNK;
4023 if (!ret)
4024 goto again;
4025 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4026 flush_state < COMMIT_TRANS)
4027 goto again;
4028 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4029 flush_state <= COMMIT_TRANS)
4030 goto again;
4032 out:
4033 if (flushing) {
4034 spin_lock(&space_info->lock);
4035 space_info->flush = 0;
4036 wake_up_all(&space_info->wait);
4037 spin_unlock(&space_info->lock);
4038 }
4039 return ret;
4040 }
4042 static struct btrfs_block_rsv *get_block_rsv(
4043 const struct btrfs_trans_handle *trans,
4044 const struct btrfs_root *root)
4045 {
4046 struct btrfs_block_rsv *block_rsv = NULL;
4048 if (root->ref_cows)
4049 block_rsv = trans->block_rsv;
4051 if (root == root->fs_info->csum_root && trans->adding_csums)
4052 block_rsv = trans->block_rsv;
4054 if (!block_rsv)
4055 block_rsv = root->block_rsv;
4057 if (!block_rsv)
4058 block_rsv = &root->fs_info->empty_block_rsv;
4060 return block_rsv;
4061 }
4063 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4064 u64 num_bytes)
4065 {
4066 int ret = -ENOSPC;
4067 spin_lock(&block_rsv->lock);
4068 if (block_rsv->reserved >= num_bytes) {
4069 block_rsv->reserved -= num_bytes;
4070 if (block_rsv->reserved < block_rsv->size)
4071 block_rsv->full = 0;
4072 ret = 0;
4073 }
4074 spin_unlock(&block_rsv->lock);
4075 return ret;
4076 }
4078 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4079 u64 num_bytes, int update_size)
4080 {
4081 spin_lock(&block_rsv->lock);
4082 block_rsv->reserved += num_bytes;
4083 if (update_size)
4084 block_rsv->size += num_bytes;
4085 else if (block_rsv->reserved >= block_rsv->size)
4086 block_rsv->full = 1;
4087 spin_unlock(&block_rsv->lock);
4088 }
4090 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4091 struct btrfs_block_rsv *block_rsv,
4092 struct btrfs_block_rsv *dest, u64 num_bytes)
4093 {
4094 struct btrfs_space_info *space_info = block_rsv->space_info;
4096 spin_lock(&block_rsv->lock);
4097 if (num_bytes == (u64)-1)
4098 num_bytes = block_rsv->size;
4099 block_rsv->size -= num_bytes;
4100 if (block_rsv->reserved >= block_rsv->size) {
4101 num_bytes = block_rsv->reserved - block_rsv->size;
4102 block_rsv->reserved = block_rsv->size;
4103 block_rsv->full = 1;
4104 } else {
4105 num_bytes = 0;
4106 }
4107 spin_unlock(&block_rsv->lock);
4109 if (num_bytes > 0) {
4110 if (dest) {
4111 spin_lock(&dest->lock);
4112 if (!dest->full) {
4113 u64 bytes_to_add;
4115 bytes_to_add = dest->size - dest->reserved;
4116 bytes_to_add = min(num_bytes, bytes_to_add);
4117 dest->reserved += bytes_to_add;
4118 if (dest->reserved >= dest->size)
4119 dest->full = 1;
4120 num_bytes -= bytes_to_add;
4121 }
4122 spin_unlock(&dest->lock);
4123 }
4124 if (num_bytes) {
4125 spin_lock(&space_info->lock);
4126 space_info->bytes_may_use -= num_bytes;
4127 trace_btrfs_space_reservation(fs_info, "space_info",
4128 space_info->flags, num_bytes, 0);
4129 space_info->reservation_progress++;
4130 spin_unlock(&space_info->lock);
4131 }
4132 }
4133 }
4135 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4136 struct btrfs_block_rsv *dst, u64 num_bytes)
4137 {
4138 int ret;
4140 ret = block_rsv_use_bytes(src, num_bytes);
4141 if (ret)
4142 return ret;
4144 block_rsv_add_bytes(dst, num_bytes, 1);
4145 return 0;
4146 }
4148 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4149 {
4150 memset(rsv, 0, sizeof(*rsv));
4151 spin_lock_init(&rsv->lock);
4152 rsv->type = type;
4153 }
4155 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4156 unsigned short type)
4157 {
4158 struct btrfs_block_rsv *block_rsv;
4159 struct btrfs_fs_info *fs_info = root->fs_info;
4161 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4162 if (!block_rsv)
4163 return NULL;
4165 btrfs_init_block_rsv(block_rsv, type);
4166 block_rsv->space_info = __find_space_info(fs_info,
4167 BTRFS_BLOCK_GROUP_METADATA);
4168 return block_rsv;
4169 }
4171 void btrfs_free_block_rsv(struct btrfs_root *root,
4172 struct btrfs_block_rsv *rsv)
4173 {
4174 if (!rsv)
4175 return;
4176 btrfs_block_rsv_release(root, rsv, (u64)-1);
4177 kfree(rsv);
4178 }
4180 int btrfs_block_rsv_add(struct btrfs_root *root,
4181 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4182 enum btrfs_reserve_flush_enum flush)
4183 {
4184 int ret;
4186 if (num_bytes == 0)
4187 return 0;
4189 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4190 if (!ret) {
4191 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4192 return 0;
4193 }
4195 return ret;
4196 }
4198 int btrfs_block_rsv_check(struct btrfs_root *root,
4199 struct btrfs_block_rsv *block_rsv, int min_factor)
4200 {
4201 u64 num_bytes = 0;
4202 int ret = -ENOSPC;
4204 if (!block_rsv)
4205 return 0;
4207 spin_lock(&block_rsv->lock);
4208 num_bytes = div_factor(block_rsv->size, min_factor);
4209 if (block_rsv->reserved >= num_bytes)
4210 ret = 0;
4211 spin_unlock(&block_rsv->lock);
4213 return ret;
4214 }
4216 int btrfs_block_rsv_refill(struct btrfs_root *root,
4217 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4218 enum btrfs_reserve_flush_enum flush)
4219 {
4220 u64 num_bytes = 0;
4221 int ret = -ENOSPC;
4223 if (!block_rsv)
4224 return 0;
4226 spin_lock(&block_rsv->lock);
4227 num_bytes = min_reserved;
4228 if (block_rsv->reserved >= num_bytes)
4229 ret = 0;
4230 else
4231 num_bytes -= block_rsv->reserved;
4232 spin_unlock(&block_rsv->lock);
4234 if (!ret)
4235 return 0;
4237 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4238 if (!ret) {
4239 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4240 return 0;
4241 }
4243 return ret;
4244 }
4246 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4247 struct btrfs_block_rsv *dst_rsv,
4248 u64 num_bytes)
4249 {
4250 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4251 }
4253 void btrfs_block_rsv_release(struct btrfs_root *root,
4254 struct btrfs_block_rsv *block_rsv,
4255 u64 num_bytes)
4256 {
4257 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4258 if (global_rsv->full || global_rsv == block_rsv ||
4259 block_rsv->space_info != global_rsv->space_info)
4260 global_rsv = NULL;
4261 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4262 num_bytes);
4263 }
4265 /*
4266 * helper to calculate size of global block reservation.
4267 * the desired value is sum of space used by extent tree,
4268 * checksum tree and root tree
4269 */
4270 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4271 {
4272 struct btrfs_space_info *sinfo;
4273 u64 num_bytes;
4274 u64 meta_used;
4275 u64 data_used;
4276 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4278 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4279 spin_lock(&sinfo->lock);
4280 data_used = sinfo->bytes_used;
4281 spin_unlock(&sinfo->lock);
4283 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4284 spin_lock(&sinfo->lock);
4285 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4286 data_used = 0;
4287 meta_used = sinfo->bytes_used;
4288 spin_unlock(&sinfo->lock);
4290 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4291 csum_size * 2;
4292 num_bytes += div64_u64(data_used + meta_used, 50);
4294 if (num_bytes * 3 > meta_used)
4295 num_bytes = div64_u64(meta_used, 3);
4297 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4298 }
4300 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4301 {
4302 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4303 struct btrfs_space_info *sinfo = block_rsv->space_info;
4304 u64 num_bytes;
4306 num_bytes = calc_global_metadata_size(fs_info);
4308 spin_lock(&sinfo->lock);
4309 spin_lock(&block_rsv->lock);
4311 block_rsv->size = num_bytes;
4313 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4314 sinfo->bytes_reserved + sinfo->bytes_readonly +
4315 sinfo->bytes_may_use;
4317 if (sinfo->total_bytes > num_bytes) {
4318 num_bytes = sinfo->total_bytes - num_bytes;
4319 block_rsv->reserved += num_bytes;
4320 sinfo->bytes_may_use += num_bytes;
4321 trace_btrfs_space_reservation(fs_info, "space_info",
4322 sinfo->flags, num_bytes, 1);
4323 }
4325 if (block_rsv->reserved >= block_rsv->size) {
4326 num_bytes = block_rsv->reserved - block_rsv->size;
4327 sinfo->bytes_may_use -= num_bytes;
4328 trace_btrfs_space_reservation(fs_info, "space_info",
4329 sinfo->flags, num_bytes, 0);
4330 sinfo->reservation_progress++;
4331 block_rsv->reserved = block_rsv->size;
4332 block_rsv->full = 1;
4333 }
4335 spin_unlock(&block_rsv->lock);
4336 spin_unlock(&sinfo->lock);
4337 }
4339 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4340 {
4341 struct btrfs_space_info *space_info;
4343 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4344 fs_info->chunk_block_rsv.space_info = space_info;
4346 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4347 fs_info->global_block_rsv.space_info = space_info;
4348 fs_info->delalloc_block_rsv.space_info = space_info;
4349 fs_info->trans_block_rsv.space_info = space_info;
4350 fs_info->empty_block_rsv.space_info = space_info;
4351 fs_info->delayed_block_rsv.space_info = space_info;
4353 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4354 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4355 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4356 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4357 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4359 update_global_block_rsv(fs_info);
4360 }
4362 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4363 {
4364 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4365 (u64)-1);
4366 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4367 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4368 WARN_ON(fs_info->trans_block_rsv.size > 0);
4369 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4370 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4371 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4372 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4373 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4374 }
4376 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4377 struct btrfs_root *root)
4378 {
4379 if (!trans->block_rsv)
4380 return;
4382 if (!trans->bytes_reserved)
4383 return;
4385 trace_btrfs_space_reservation(root->fs_info, "transaction",
4386 trans->transid, trans->bytes_reserved, 0);
4387 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4388 trans->bytes_reserved = 0;
4389 }
4391 /* Can only return 0 or -ENOSPC */
4392 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4393 struct inode *inode)
4394 {
4395 struct btrfs_root *root = BTRFS_I(inode)->root;
4396 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4397 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4399 /*
4400 * We need to hold space in order to delete our orphan item once we've
4401 * added it, so this takes the reservation so we can release it later
4402 * when we are truly done with the orphan item.
4403 */
4404 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4405 trace_btrfs_space_reservation(root->fs_info, "orphan",
4406 btrfs_ino(inode), num_bytes, 1);
4407 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4408 }
4410 void btrfs_orphan_release_metadata(struct inode *inode)
4411 {
4412 struct btrfs_root *root = BTRFS_I(inode)->root;
4413 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4414 trace_btrfs_space_reservation(root->fs_info, "orphan",
4415 btrfs_ino(inode), num_bytes, 0);
4416 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4417 }
4419 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4420 struct btrfs_pending_snapshot *pending)
4421 {
4422 struct btrfs_root *root = pending->root;
4423 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4424 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4425 /*
4426 * two for root back/forward refs, two for directory entries,
4427 * one for root of the snapshot and one for parent inode.
4428 */
4429 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 6);
4430 dst_rsv->space_info = src_rsv->space_info;
4431 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4432 }
4434 /**
4435 * drop_outstanding_extent - drop an outstanding extent
4436 * @inode: the inode we're dropping the extent for
4437 *
4438 * This is called when we are freeing up an outstanding extent, either called
4439 * after an error or after an extent is written. This will return the number of
4440 * reserved extents that need to be freed. This must be called with
4441 * BTRFS_I(inode)->lock held.
4442 */
4443 static unsigned drop_outstanding_extent(struct inode *inode)
4444 {
4445 unsigned drop_inode_space = 0;
4446 unsigned dropped_extents = 0;
4448 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4449 BTRFS_I(inode)->outstanding_extents--;
4451 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4452 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4453 &BTRFS_I(inode)->runtime_flags))
4454 drop_inode_space = 1;
4456 /*
4457 * If we have more or the same amount of outsanding extents than we have
4458 * reserved then we need to leave the reserved extents count alone.
4459 */
4460 if (BTRFS_I(inode)->outstanding_extents >=
4461 BTRFS_I(inode)->reserved_extents)
4462 return drop_inode_space;
4464 dropped_extents = BTRFS_I(inode)->reserved_extents -
4465 BTRFS_I(inode)->outstanding_extents;
4466 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4467 return dropped_extents + drop_inode_space;
4468 }
4470 /**
4471 * calc_csum_metadata_size - return the amount of metada space that must be
4472 * reserved/free'd for the given bytes.
4473 * @inode: the inode we're manipulating
4474 * @num_bytes: the number of bytes in question
4475 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4476 *
4477 * This adjusts the number of csum_bytes in the inode and then returns the
4478 * correct amount of metadata that must either be reserved or freed. We
4479 * calculate how many checksums we can fit into one leaf and then divide the
4480 * number of bytes that will need to be checksumed by this value to figure out
4481 * how many checksums will be required. If we are adding bytes then the number
4482 * may go up and we will return the number of additional bytes that must be
4483 * reserved. If it is going down we will return the number of bytes that must
4484 * be freed.
4485 *
4486 * This must be called with BTRFS_I(inode)->lock held.
4487 */
4488 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4489 int reserve)
4490 {
4491 struct btrfs_root *root = BTRFS_I(inode)->root;
4492 u64 csum_size;
4493 int num_csums_per_leaf;
4494 int num_csums;
4495 int old_csums;
4497 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4498 BTRFS_I(inode)->csum_bytes == 0)
4499 return 0;
4501 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4502 if (reserve)
4503 BTRFS_I(inode)->csum_bytes += num_bytes;
4504 else
4505 BTRFS_I(inode)->csum_bytes -= num_bytes;
4506 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4507 num_csums_per_leaf = (int)div64_u64(csum_size,
4508 sizeof(struct btrfs_csum_item) +
4509 sizeof(struct btrfs_disk_key));
4510 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4511 num_csums = num_csums + num_csums_per_leaf - 1;
4512 num_csums = num_csums / num_csums_per_leaf;
4514 old_csums = old_csums + num_csums_per_leaf - 1;
4515 old_csums = old_csums / num_csums_per_leaf;
4517 /* No change, no need to reserve more */
4518 if (old_csums == num_csums)
4519 return 0;
4521 if (reserve)
4522 return btrfs_calc_trans_metadata_size(root,
4523 num_csums - old_csums);
4525 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4526 }
4528 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4529 {
4530 struct btrfs_root *root = BTRFS_I(inode)->root;
4531 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4532 u64 to_reserve = 0;
4533 u64 csum_bytes;
4534 unsigned nr_extents = 0;
4535 int extra_reserve = 0;
4536 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
4537 int ret = 0;
4538 bool delalloc_lock = true;
4540 /* If we are a free space inode we need to not flush since we will be in
4541 * the middle of a transaction commit. We also don't need the delalloc
4542 * mutex since we won't race with anybody. We need this mostly to make
4543 * lockdep shut its filthy mouth.
4544 */
4545 if (btrfs_is_free_space_inode(inode)) {
4546 flush = BTRFS_RESERVE_NO_FLUSH;
4547 delalloc_lock = false;
4548 }
4550 if (flush != BTRFS_RESERVE_NO_FLUSH &&
4551 btrfs_transaction_in_commit(root->fs_info))
4552 schedule_timeout(1);
4554 if (delalloc_lock)
4555 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4557 num_bytes = ALIGN(num_bytes, root->sectorsize);
4559 spin_lock(&BTRFS_I(inode)->lock);
4560 BTRFS_I(inode)->outstanding_extents++;
4562 if (BTRFS_I(inode)->outstanding_extents >
4563 BTRFS_I(inode)->reserved_extents)
4564 nr_extents = BTRFS_I(inode)->outstanding_extents -
4565 BTRFS_I(inode)->reserved_extents;
4567 /*
4568 * Add an item to reserve for updating the inode when we complete the
4569 * delalloc io.
4570 */
4571 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4572 &BTRFS_I(inode)->runtime_flags)) {
4573 nr_extents++;
4574 extra_reserve = 1;
4575 }
4577 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4578 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4579 csum_bytes = BTRFS_I(inode)->csum_bytes;
4580 spin_unlock(&BTRFS_I(inode)->lock);
4582 if (root->fs_info->quota_enabled)
4583 ret = btrfs_qgroup_reserve(root, num_bytes +
4584 nr_extents * root->leafsize);
4586 /*
4587 * ret != 0 here means the qgroup reservation failed, we go straight to
4588 * the shared error handling then.
4589 */
4590 if (ret == 0)
4591 ret = reserve_metadata_bytes(root, block_rsv,
4592 to_reserve, flush);
4594 if (ret) {
4595 u64 to_free = 0;
4596 unsigned dropped;
4598 spin_lock(&BTRFS_I(inode)->lock);
4599 dropped = drop_outstanding_extent(inode);
4600 /*
4601 * If the inodes csum_bytes is the same as the original
4602 * csum_bytes then we know we haven't raced with any free()ers
4603 * so we can just reduce our inodes csum bytes and carry on.
4604 * Otherwise we have to do the normal free thing to account for
4605 * the case that the free side didn't free up its reserve
4606 * because of this outstanding reservation.
4607 */
4608 if (BTRFS_I(inode)->csum_bytes == csum_bytes)
4609 calc_csum_metadata_size(inode, num_bytes, 0);
4610 else
4611 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4612 spin_unlock(&BTRFS_I(inode)->lock);
4613 if (dropped)
4614 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4616 if (to_free) {
4617 btrfs_block_rsv_release(root, block_rsv, to_free);
4618 trace_btrfs_space_reservation(root->fs_info,
4619 "delalloc",
4620 btrfs_ino(inode),
4621 to_free, 0);
4622 }
4623 if (root->fs_info->quota_enabled) {
4624 btrfs_qgroup_free(root, num_bytes +
4625 nr_extents * root->leafsize);
4626 }
4627 if (delalloc_lock)
4628 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4629 return ret;
4630 }
4632 spin_lock(&BTRFS_I(inode)->lock);
4633 if (extra_reserve) {
4634 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4635 &BTRFS_I(inode)->runtime_flags);
4636 nr_extents--;
4637 }
4638 BTRFS_I(inode)->reserved_extents += nr_extents;
4639 spin_unlock(&BTRFS_I(inode)->lock);
4641 if (delalloc_lock)
4642 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4644 if (to_reserve)
4645 trace_btrfs_space_reservation(root->fs_info,"delalloc",
4646 btrfs_ino(inode), to_reserve, 1);
4647 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4649 return 0;
4650 }
4652 /**
4653 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4654 * @inode: the inode to release the reservation for
4655 * @num_bytes: the number of bytes we're releasing
4656 *
4657 * This will release the metadata reservation for an inode. This can be called
4658 * once we complete IO for a given set of bytes to release their metadata
4659 * reservations.
4660 */
4661 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4662 {
4663 struct btrfs_root *root = BTRFS_I(inode)->root;
4664 u64 to_free = 0;
4665 unsigned dropped;
4667 num_bytes = ALIGN(num_bytes, root->sectorsize);
4668 spin_lock(&BTRFS_I(inode)->lock);
4669 dropped = drop_outstanding_extent(inode);
4671 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4672 spin_unlock(&BTRFS_I(inode)->lock);
4673 if (dropped > 0)
4674 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4676 trace_btrfs_space_reservation(root->fs_info, "delalloc",
4677 btrfs_ino(inode), to_free, 0);
4678 if (root->fs_info->quota_enabled) {
4679 btrfs_qgroup_free(root, num_bytes +
4680 dropped * root->leafsize);
4681 }
4683 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4684 to_free);
4685 }
4687 /**
4688 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4689 * @inode: inode we're writing to
4690 * @num_bytes: the number of bytes we want to allocate
4691 *
4692 * This will do the following things
4693 *
4694 * o reserve space in the data space info for num_bytes
4695 * o reserve space in the metadata space info based on number of outstanding
4696 * extents and how much csums will be needed
4697 * o add to the inodes ->delalloc_bytes
4698 * o add it to the fs_info's delalloc inodes list.
4699 *
4700 * This will return 0 for success and -ENOSPC if there is no space left.
4701 */
4702 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4703 {
4704 int ret;
4706 ret = btrfs_check_data_free_space(inode, num_bytes);
4707 if (ret)
4708 return ret;
4710 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4711 if (ret) {
4712 btrfs_free_reserved_data_space(inode, num_bytes);
4713 return ret;
4714 }
4716 return 0;
4717 }
4719 /**
4720 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4721 * @inode: inode we're releasing space for
4722 * @num_bytes: the number of bytes we want to free up
4723 *
4724 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4725 * called in the case that we don't need the metadata AND data reservations
4726 * anymore. So if there is an error or we insert an inline extent.
4727 *
4728 * This function will release the metadata space that was not used and will
4729 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4730 * list if there are no delalloc bytes left.
4731 */
4732 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4733 {
4734 btrfs_delalloc_release_metadata(inode, num_bytes);
4735 btrfs_free_reserved_data_space(inode, num_bytes);
4736 }
4738 static int update_block_group(struct btrfs_trans_handle *trans,
4739 struct btrfs_root *root,
4740 u64 bytenr, u64 num_bytes, int alloc)
4741 {
4742 struct btrfs_block_group_cache *cache = NULL;
4743 struct btrfs_fs_info *info = root->fs_info;
4744 u64 total = num_bytes;
4745 u64 old_val;
4746 u64 byte_in_group;
4747 int factor;
4749 /* block accounting for super block */
4750 spin_lock(&info->delalloc_lock);
4751 old_val = btrfs_super_bytes_used(info->super_copy);
4752 if (alloc)
4753 old_val += num_bytes;
4754 else
4755 old_val -= num_bytes;
4756 btrfs_set_super_bytes_used(info->super_copy, old_val);
4757 spin_unlock(&info->delalloc_lock);
4759 while (total) {
4760 cache = btrfs_lookup_block_group(info, bytenr);
4761 if (!cache)
4762 return -ENOENT;
4763 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4764 BTRFS_BLOCK_GROUP_RAID1 |
4765 BTRFS_BLOCK_GROUP_RAID10))
4766 factor = 2;
4767 else
4768 factor = 1;
4769 /*
4770 * If this block group has free space cache written out, we
4771 * need to make sure to load it if we are removing space. This
4772 * is because we need the unpinning stage to actually add the
4773 * space back to the block group, otherwise we will leak space.
4774 */
4775 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4776 cache_block_group(cache, trans, NULL, 1);
4778 byte_in_group = bytenr - cache->key.objectid;
4779 WARN_ON(byte_in_group > cache->key.offset);
4781 spin_lock(&cache->space_info->lock);
4782 spin_lock(&cache->lock);
4784 if (btrfs_test_opt(root, SPACE_CACHE) &&
4785 cache->disk_cache_state < BTRFS_DC_CLEAR)
4786 cache->disk_cache_state = BTRFS_DC_CLEAR;
4788 cache->dirty = 1;
4789 old_val = btrfs_block_group_used(&cache->item);
4790 num_bytes = min(total, cache->key.offset - byte_in_group);
4791 if (alloc) {
4792 old_val += num_bytes;
4793 btrfs_set_block_group_used(&cache->item, old_val);
4794 cache->reserved -= num_bytes;
4795 cache->space_info->bytes_reserved -= num_bytes;
4796 cache->space_info->bytes_used += num_bytes;
4797 cache->space_info->disk_used += num_bytes * factor;
4798 spin_unlock(&cache->lock);
4799 spin_unlock(&cache->space_info->lock);
4800 } else {
4801 old_val -= num_bytes;
4802 btrfs_set_block_group_used(&cache->item, old_val);
4803 cache->pinned += num_bytes;
4804 cache->space_info->bytes_pinned += num_bytes;
4805 cache->space_info->bytes_used -= num_bytes;
4806 cache->space_info->disk_used -= num_bytes * factor;
4807 spin_unlock(&cache->lock);
4808 spin_unlock(&cache->space_info->lock);
4810 set_extent_dirty(info->pinned_extents,
4811 bytenr, bytenr + num_bytes - 1,
4812 GFP_NOFS | __GFP_NOFAIL);
4813 }
4814 btrfs_put_block_group(cache);
4815 total -= num_bytes;
4816 bytenr += num_bytes;
4817 }
4818 return 0;
4819 }
4821 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4822 {
4823 struct btrfs_block_group_cache *cache;
4824 u64 bytenr;
4826 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4827 if (!cache)
4828 return 0;
4830 bytenr = cache->key.objectid;
4831 btrfs_put_block_group(cache);
4833 return bytenr;
4834 }
4836 static int pin_down_extent(struct btrfs_root *root,
4837 struct btrfs_block_group_cache *cache,
4838 u64 bytenr, u64 num_bytes, int reserved)
4839 {
4840 spin_lock(&cache->space_info->lock);
4841 spin_lock(&cache->lock);
4842 cache->pinned += num_bytes;
4843 cache->space_info->bytes_pinned += num_bytes;
4844 if (reserved) {
4845 cache->reserved -= num_bytes;
4846 cache->space_info->bytes_reserved -= num_bytes;
4847 }
4848 spin_unlock(&cache->lock);
4849 spin_unlock(&cache->space_info->lock);
4851 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4852 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4853 return 0;
4854 }
4856 /*
4857 * this function must be called within transaction
4858 */
4859 int btrfs_pin_extent(struct btrfs_root *root,
4860 u64 bytenr, u64 num_bytes, int reserved)
4861 {
4862 struct btrfs_block_group_cache *cache;
4864 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4865 BUG_ON(!cache); /* Logic error */
4867 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4869 btrfs_put_block_group(cache);
4870 return 0;
4871 }
4873 /*
4874 * this function must be called within transaction
4875 */
4876 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
4877 struct btrfs_root *root,
4878 u64 bytenr, u64 num_bytes)
4879 {
4880 struct btrfs_block_group_cache *cache;
4882 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4883 BUG_ON(!cache); /* Logic error */
4885 /*
4886 * pull in the free space cache (if any) so that our pin
4887 * removes the free space from the cache. We have load_only set
4888 * to one because the slow code to read in the free extents does check
4889 * the pinned extents.
4890 */
4891 cache_block_group(cache, trans, root, 1);
4893 pin_down_extent(root, cache, bytenr, num_bytes, 0);
4895 /* remove us from the free space cache (if we're there at all) */
4896 btrfs_remove_free_space(cache, bytenr, num_bytes);
4897 btrfs_put_block_group(cache);
4898 return 0;
4899 }
4901 /**
4902 * btrfs_update_reserved_bytes - update the block_group and space info counters
4903 * @cache: The cache we are manipulating
4904 * @num_bytes: The number of bytes in question
4905 * @reserve: One of the reservation enums
4906 *
4907 * This is called by the allocator when it reserves space, or by somebody who is
4908 * freeing space that was never actually used on disk. For example if you
4909 * reserve some space for a new leaf in transaction A and before transaction A
4910 * commits you free that leaf, you call this with reserve set to 0 in order to
4911 * clear the reservation.
4912 *
4913 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4914 * ENOSPC accounting. For data we handle the reservation through clearing the
4915 * delalloc bits in the io_tree. We have to do this since we could end up
4916 * allocating less disk space for the amount of data we have reserved in the
4917 * case of compression.
4918 *
4919 * If this is a reservation and the block group has become read only we cannot
4920 * make the reservation and return -EAGAIN, otherwise this function always
4921 * succeeds.
4922 */
4923 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4924 u64 num_bytes, int reserve)
4925 {
4926 struct btrfs_space_info *space_info = cache->space_info;
4927 int ret = 0;
4929 spin_lock(&space_info->lock);
4930 spin_lock(&cache->lock);
4931 if (reserve != RESERVE_FREE) {
4932 if (cache->ro) {
4933 ret = -EAGAIN;
4934 } else {
4935 cache->reserved += num_bytes;
4936 space_info->bytes_reserved += num_bytes;
4937 if (reserve == RESERVE_ALLOC) {
4938 trace_btrfs_space_reservation(cache->fs_info,
4939 "space_info", space_info->flags,
4940 num_bytes, 0);
4941 space_info->bytes_may_use -= num_bytes;
4942 }
4943 }
4944 } else {
4945 if (cache->ro)
4946 space_info->bytes_readonly += num_bytes;
4947 cache->reserved -= num_bytes;
4948 space_info->bytes_reserved -= num_bytes;
4949 space_info->reservation_progress++;
4950 }
4951 spin_unlock(&cache->lock);
4952 spin_unlock(&space_info->lock);
4953 return ret;
4954 }
4956 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4957 struct btrfs_root *root)
4958 {
4959 struct btrfs_fs_info *fs_info = root->fs_info;
4960 struct btrfs_caching_control *next;
4961 struct btrfs_caching_control *caching_ctl;
4962 struct btrfs_block_group_cache *cache;
4964 down_write(&fs_info->extent_commit_sem);
4966 list_for_each_entry_safe(caching_ctl, next,
4967 &fs_info->caching_block_groups, list) {
4968 cache = caching_ctl->block_group;
4969 if (block_group_cache_done(cache)) {
4970 cache->last_byte_to_unpin = (u64)-1;
4971 list_del_init(&caching_ctl->list);
4972 put_caching_control(caching_ctl);
4973 } else {
4974 cache->last_byte_to_unpin = caching_ctl->progress;
4975 }
4976 }
4978 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4979 fs_info->pinned_extents = &fs_info->freed_extents[1];
4980 else
4981 fs_info->pinned_extents = &fs_info->freed_extents[0];
4983 up_write(&fs_info->extent_commit_sem);
4985 update_global_block_rsv(fs_info);
4986 }
4988 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4989 {
4990 struct btrfs_fs_info *fs_info = root->fs_info;
4991 struct btrfs_block_group_cache *cache = NULL;
4992 struct btrfs_space_info *space_info;
4993 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4994 u64 len;
4995 bool readonly;
4997 while (start <= end) {
4998 readonly = false;
4999 if (!cache ||
5000 start >= cache->key.objectid + cache->key.offset) {
5001 if (cache)
5002 btrfs_put_block_group(cache);
5003 cache = btrfs_lookup_block_group(fs_info, start);
5004 BUG_ON(!cache); /* Logic error */
5005 }
5007 len = cache->key.objectid + cache->key.offset - start;
5008 len = min(len, end + 1 - start);
5010 if (start < cache->last_byte_to_unpin) {
5011 len = min(len, cache->last_byte_to_unpin - start);
5012 btrfs_add_free_space(cache, start, len);
5013 }
5015 start += len;
5016 space_info = cache->space_info;
5018 spin_lock(&space_info->lock);
5019 spin_lock(&cache->lock);
5020 cache->pinned -= len;
5021 space_info->bytes_pinned -= len;
5022 if (cache->ro) {
5023 space_info->bytes_readonly += len;
5024 readonly = true;
5025 }
5026 spin_unlock(&cache->lock);
5027 if (!readonly && global_rsv->space_info == space_info) {
5028 spin_lock(&global_rsv->lock);
5029 if (!global_rsv->full) {
5030 len = min(len, global_rsv->size -
5031 global_rsv->reserved);
5032 global_rsv->reserved += len;
5033 space_info->bytes_may_use += len;
5034 if (global_rsv->reserved >= global_rsv->size)
5035 global_rsv->full = 1;
5036 }
5037 spin_unlock(&global_rsv->lock);
5038 }
5039 spin_unlock(&space_info->lock);
5040 }
5042 if (cache)
5043 btrfs_put_block_group(cache);
5044 return 0;
5045 }
5047 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5048 struct btrfs_root *root)
5049 {
5050 struct btrfs_fs_info *fs_info = root->fs_info;
5051 struct extent_io_tree *unpin;
5052 u64 start;
5053 u64 end;
5054 int ret;
5056 if (trans->aborted)
5057 return 0;
5059 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5060 unpin = &fs_info->freed_extents[1];
5061 else
5062 unpin = &fs_info->freed_extents[0];
5064 while (1) {
5065 ret = find_first_extent_bit(unpin, 0, &start, &end,
5066 EXTENT_DIRTY, NULL);
5067 if (ret)
5068 break;
5070 if (btrfs_test_opt(root, DISCARD))
5071 ret = btrfs_discard_extent(root, start,
5072 end + 1 - start, NULL);
5074 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5075 unpin_extent_range(root, start, end);
5076 cond_resched();
5077 }
5079 return 0;
5080 }
5082 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5083 struct btrfs_root *root,
5084 u64 bytenr, u64 num_bytes, u64 parent,
5085 u64 root_objectid, u64 owner_objectid,
5086 u64 owner_offset, int refs_to_drop,
5087 struct btrfs_delayed_extent_op *extent_op)
5088 {
5089 struct btrfs_key key;
5090 struct btrfs_path *path;
5091 struct btrfs_fs_info *info = root->fs_info;
5092 struct btrfs_root *extent_root = info->extent_root;
5093 struct extent_buffer *leaf;
5094 struct btrfs_extent_item *ei;
5095 struct btrfs_extent_inline_ref *iref;
5096 int ret;
5097 int is_data;
5098 int extent_slot = 0;
5099 int found_extent = 0;
5100 int num_to_del = 1;
5101 u32 item_size;
5102 u64 refs;
5104 path = btrfs_alloc_path();
5105 if (!path)
5106 return -ENOMEM;
5108 path->reada = 1;
5109 path->leave_spinning = 1;
5111 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5112 BUG_ON(!is_data && refs_to_drop != 1);
5114 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5115 bytenr, num_bytes, parent,
5116 root_objectid, owner_objectid,
5117 owner_offset);
5118 if (ret == 0) {
5119 extent_slot = path->slots[0];
5120 while (extent_slot >= 0) {
5121 btrfs_item_key_to_cpu(path->nodes[0], &key,
5122 extent_slot);
5123 if (key.objectid != bytenr)
5124 break;
5125 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5126 key.offset == num_bytes) {
5127 found_extent = 1;
5128 break;
5129 }
5130 if (path->slots[0] - extent_slot > 5)
5131 break;
5132 extent_slot--;
5133 }
5134 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5135 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5136 if (found_extent && item_size < sizeof(*ei))
5137 found_extent = 0;
5138 #endif
5139 if (!found_extent) {
5140 BUG_ON(iref);
5141 ret = remove_extent_backref(trans, extent_root, path,
5142 NULL, refs_to_drop,
5143 is_data);
5144 if (ret) {
5145 btrfs_abort_transaction(trans, extent_root, ret);
5146 goto out;
5147 }
5148 btrfs_release_path(path);
5149 path->leave_spinning = 1;
5151 key.objectid = bytenr;
5152 key.type = BTRFS_EXTENT_ITEM_KEY;
5153 key.offset = num_bytes;
5155 ret = btrfs_search_slot(trans, extent_root,
5156 &key, path, -1, 1);
5157 if (ret) {
5158 printk(KERN_ERR "umm, got %d back from search"
5159 ", was looking for %llu\n", ret,
5160 (unsigned long long)bytenr);
5161 if (ret > 0)
5162 btrfs_print_leaf(extent_root,
5163 path->nodes[0]);
5164 }
5165 if (ret < 0) {
5166 btrfs_abort_transaction(trans, extent_root, ret);
5167 goto out;
5168 }
5169 extent_slot = path->slots[0];
5170 }
5171 } else if (ret == -ENOENT) {
5172 btrfs_print_leaf(extent_root, path->nodes[0]);
5173 WARN_ON(1);
5174 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
5175 "parent %llu root %llu owner %llu offset %llu\n",
5176 (unsigned long long)bytenr,
5177 (unsigned long long)parent,
5178 (unsigned long long)root_objectid,
5179 (unsigned long long)owner_objectid,
5180 (unsigned long long)owner_offset);
5181 } else {
5182 btrfs_abort_transaction(trans, extent_root, ret);
5183 goto out;
5184 }
5186 leaf = path->nodes[0];
5187 item_size = btrfs_item_size_nr(leaf, extent_slot);
5188 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5189 if (item_size < sizeof(*ei)) {
5190 BUG_ON(found_extent || extent_slot != path->slots[0]);
5191 ret = convert_extent_item_v0(trans, extent_root, path,
5192 owner_objectid, 0);
5193 if (ret < 0) {
5194 btrfs_abort_transaction(trans, extent_root, ret);
5195 goto out;
5196 }
5198 btrfs_release_path(path);
5199 path->leave_spinning = 1;
5201 key.objectid = bytenr;
5202 key.type = BTRFS_EXTENT_ITEM_KEY;
5203 key.offset = num_bytes;
5205 ret = btrfs_search_slot(trans, extent_root, &key, path,
5206 -1, 1);
5207 if (ret) {
5208 printk(KERN_ERR "umm, got %d back from search"
5209 ", was looking for %llu\n", ret,
5210 (unsigned long long)bytenr);
5211 btrfs_print_leaf(extent_root, path->nodes[0]);
5212 }
5213 if (ret < 0) {
5214 btrfs_abort_transaction(trans, extent_root, ret);
5215 goto out;
5216 }
5218 extent_slot = path->slots[0];
5219 leaf = path->nodes[0];
5220 item_size = btrfs_item_size_nr(leaf, extent_slot);
5221 }
5222 #endif
5223 BUG_ON(item_size < sizeof(*ei));
5224 ei = btrfs_item_ptr(leaf, extent_slot,
5225 struct btrfs_extent_item);
5226 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5227 struct btrfs_tree_block_info *bi;
5228 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5229 bi = (struct btrfs_tree_block_info *)(ei + 1);
5230 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5231 }
5233 refs = btrfs_extent_refs(leaf, ei);
5234 BUG_ON(refs < refs_to_drop);
5235 refs -= refs_to_drop;
5237 if (refs > 0) {
5238 if (extent_op)
5239 __run_delayed_extent_op(extent_op, leaf, ei);
5240 /*
5241 * In the case of inline back ref, reference count will
5242 * be updated by remove_extent_backref
5243 */
5244 if (iref) {
5245 BUG_ON(!found_extent);
5246 } else {
5247 btrfs_set_extent_refs(leaf, ei, refs);
5248 btrfs_mark_buffer_dirty(leaf);
5249 }
5250 if (found_extent) {
5251 ret = remove_extent_backref(trans, extent_root, path,
5252 iref, refs_to_drop,
5253 is_data);
5254 if (ret) {
5255 btrfs_abort_transaction(trans, extent_root, ret);
5256 goto out;
5257 }
5258 }
5259 } else {
5260 if (found_extent) {
5261 BUG_ON(is_data && refs_to_drop !=
5262 extent_data_ref_count(root, path, iref));
5263 if (iref) {
5264 BUG_ON(path->slots[0] != extent_slot);
5265 } else {
5266 BUG_ON(path->slots[0] != extent_slot + 1);
5267 path->slots[0] = extent_slot;
5268 num_to_del = 2;
5269 }
5270 }
5272 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5273 num_to_del);
5274 if (ret) {
5275 btrfs_abort_transaction(trans, extent_root, ret);
5276 goto out;
5277 }
5278 btrfs_release_path(path);
5280 if (is_data) {
5281 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5282 if (ret) {
5283 btrfs_abort_transaction(trans, extent_root, ret);
5284 goto out;
5285 }
5286 }
5288 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
5289 if (ret) {
5290 btrfs_abort_transaction(trans, extent_root, ret);
5291 goto out;
5292 }
5293 }
5294 out:
5295 btrfs_free_path(path);
5296 return ret;
5297 }
5299 /*
5300 * when we free an block, it is possible (and likely) that we free the last
5301 * delayed ref for that extent as well. This searches the delayed ref tree for
5302 * a given extent, and if there are no other delayed refs to be processed, it
5303 * removes it from the tree.
5304 */
5305 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5306 struct btrfs_root *root, u64 bytenr)
5307 {
5308 struct btrfs_delayed_ref_head *head;
5309 struct btrfs_delayed_ref_root *delayed_refs;
5310 struct btrfs_delayed_ref_node *ref;
5311 struct rb_node *node;
5312 int ret = 0;
5314 delayed_refs = &trans->transaction->delayed_refs;
5315 spin_lock(&delayed_refs->lock);
5316 head = btrfs_find_delayed_ref_head(trans, bytenr);
5317 if (!head)
5318 goto out;
5320 node = rb_prev(&head->node.rb_node);
5321 if (!node)
5322 goto out;
5324 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5326 /* there are still entries for this ref, we can't drop it */
5327 if (ref->bytenr == bytenr)
5328 goto out;
5330 if (head->extent_op) {
5331 if (!head->must_insert_reserved)
5332 goto out;
5333 kfree(head->extent_op);
5334 head->extent_op = NULL;
5335 }
5337 /*
5338 * waiting for the lock here would deadlock. If someone else has it
5339 * locked they are already in the process of dropping it anyway
5340 */
5341 if (!mutex_trylock(&head->mutex))
5342 goto out;
5344 /*
5345 * at this point we have a head with no other entries. Go
5346 * ahead and process it.
5347 */
5348 head->node.in_tree = 0;
5349 rb_erase(&head->node.rb_node, &delayed_refs->root);
5351 delayed_refs->num_entries--;
5353 /*
5354 * we don't take a ref on the node because we're removing it from the
5355 * tree, so we just steal the ref the tree was holding.
5356 */
5357 delayed_refs->num_heads--;
5358 if (list_empty(&head->cluster))
5359 delayed_refs->num_heads_ready--;
5361 list_del_init(&head->cluster);
5362 spin_unlock(&delayed_refs->lock);
5364 BUG_ON(head->extent_op);
5365 if (head->must_insert_reserved)
5366 ret = 1;
5368 mutex_unlock(&head->mutex);
5369 btrfs_put_delayed_ref(&head->node);
5370 return ret;
5371 out:
5372 spin_unlock(&delayed_refs->lock);
5373 return 0;
5374 }
5376 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5377 struct btrfs_root *root,
5378 struct extent_buffer *buf,
5379 u64 parent, int last_ref)
5380 {
5381 struct btrfs_block_group_cache *cache = NULL;
5382 int ret;
5384 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5385 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5386 buf->start, buf->len,
5387 parent, root->root_key.objectid,
5388 btrfs_header_level(buf),
5389 BTRFS_DROP_DELAYED_REF, NULL, 0);
5390 BUG_ON(ret); /* -ENOMEM */
5391 }
5393 if (!last_ref)
5394 return;
5396 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5398 if (btrfs_header_generation(buf) == trans->transid) {
5399 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5400 ret = check_ref_cleanup(trans, root, buf->start);
5401 if (!ret)
5402 goto out;
5403 }
5405 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5406 pin_down_extent(root, cache, buf->start, buf->len, 1);
5407 goto out;
5408 }
5410 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5412 btrfs_add_free_space(cache, buf->start, buf->len);
5413 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5414 }
5415 out:
5416 /*
5417 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5418 * anymore.
5419 */
5420 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5421 btrfs_put_block_group(cache);
5422 }
5424 /* Can return -ENOMEM */
5425 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5426 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5427 u64 owner, u64 offset, int for_cow)
5428 {
5429 int ret;
5430 struct btrfs_fs_info *fs_info = root->fs_info;
5432 /*
5433 * tree log blocks never actually go into the extent allocation
5434 * tree, just update pinning info and exit early.
5435 */
5436 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5437 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5438 /* unlocks the pinned mutex */
5439 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5440 ret = 0;
5441 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5442 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5443 num_bytes,
5444 parent, root_objectid, (int)owner,
5445 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5446 } else {
5447 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5448 num_bytes,
5449 parent, root_objectid, owner,
5450 offset, BTRFS_DROP_DELAYED_REF,
5451 NULL, for_cow);
5452 }
5453 return ret;
5454 }
5456 static u64 stripe_align(struct btrfs_root *root, u64 val)
5457 {
5458 u64 mask = ((u64)root->stripesize - 1);
5459 u64 ret = (val + mask) & ~mask;
5460 return ret;
5461 }
5463 /*
5464 * when we wait for progress in the block group caching, its because
5465 * our allocation attempt failed at least once. So, we must sleep
5466 * and let some progress happen before we try again.
5467 *
5468 * This function will sleep at least once waiting for new free space to
5469 * show up, and then it will check the block group free space numbers
5470 * for our min num_bytes. Another option is to have it go ahead
5471 * and look in the rbtree for a free extent of a given size, but this
5472 * is a good start.
5473 */
5474 static noinline int
5475 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5476 u64 num_bytes)
5477 {
5478 struct btrfs_caching_control *caching_ctl;
5479 DEFINE_WAIT(wait);
5481 caching_ctl = get_caching_control(cache);
5482 if (!caching_ctl)
5483 return 0;
5485 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5486 (cache->free_space_ctl->free_space >= num_bytes));
5488 put_caching_control(caching_ctl);
5489 return 0;
5490 }
5492 static noinline int
5493 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5494 {
5495 struct btrfs_caching_control *caching_ctl;
5496 DEFINE_WAIT(wait);
5498 caching_ctl = get_caching_control(cache);
5499 if (!caching_ctl)
5500 return 0;
5502 wait_event(caching_ctl->wait, block_group_cache_done(cache));
5504 put_caching_control(caching_ctl);
5505 return 0;
5506 }
5508 int __get_raid_index(u64 flags)
5509 {
5510 int index;
5512 if (flags & BTRFS_BLOCK_GROUP_RAID10)
5513 index = 0;
5514 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
5515 index = 1;
5516 else if (flags & BTRFS_BLOCK_GROUP_DUP)
5517 index = 2;
5518 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
5519 index = 3;
5520 else
5521 index = 4;
5523 return index;
5524 }
5526 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5527 {
5528 return __get_raid_index(cache->flags);
5529 }
5531 enum btrfs_loop_type {
5532 LOOP_CACHING_NOWAIT = 0,
5533 LOOP_CACHING_WAIT = 1,
5534 LOOP_ALLOC_CHUNK = 2,
5535 LOOP_NO_EMPTY_SIZE = 3,
5536 };
5538 /*
5539 * walks the btree of allocated extents and find a hole of a given size.
5540 * The key ins is changed to record the hole:
5541 * ins->objectid == block start
5542 * ins->flags = BTRFS_EXTENT_ITEM_KEY
5543 * ins->offset == number of blocks
5544 * Any available blocks before search_start are skipped.
5545 */
5546 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5547 struct btrfs_root *orig_root,
5548 u64 num_bytes, u64 empty_size,
5549 u64 hint_byte, struct btrfs_key *ins,
5550 u64 data)
5551 {
5552 int ret = 0;
5553 struct btrfs_root *root = orig_root->fs_info->extent_root;
5554 struct btrfs_free_cluster *last_ptr = NULL;
5555 struct btrfs_block_group_cache *block_group = NULL;
5556 struct btrfs_block_group_cache *used_block_group;
5557 u64 search_start = 0;
5558 int empty_cluster = 2 * 1024 * 1024;
5559 struct btrfs_space_info *space_info;
5560 int loop = 0;
5561 int index = __get_raid_index(data);
5562 int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5563 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5564 bool found_uncached_bg = false;
5565 bool failed_cluster_refill = false;
5566 bool failed_alloc = false;
5567 bool use_cluster = true;
5568 bool have_caching_bg = false;
5570 WARN_ON(num_bytes < root->sectorsize);
5571 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5572 ins->objectid = 0;
5573 ins->offset = 0;
5575 trace_find_free_extent(orig_root, num_bytes, empty_size, data);
5577 space_info = __find_space_info(root->fs_info, data);
5578 if (!space_info) {
5579 printk(KERN_ERR "No space info for %llu\n", data);
5580 return -ENOSPC;
5581 }
5583 /*
5584 * If the space info is for both data and metadata it means we have a
5585 * small filesystem and we can't use the clustering stuff.
5586 */
5587 if (btrfs_mixed_space_info(space_info))
5588 use_cluster = false;
5590 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5591 last_ptr = &root->fs_info->meta_alloc_cluster;
5592 if (!btrfs_test_opt(root, SSD))
5593 empty_cluster = 64 * 1024;
5594 }
5596 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5597 btrfs_test_opt(root, SSD)) {
5598 last_ptr = &root->fs_info->data_alloc_cluster;
5599 }
5601 if (last_ptr) {
5602 spin_lock(&last_ptr->lock);
5603 if (last_ptr->block_group)
5604 hint_byte = last_ptr->window_start;
5605 spin_unlock(&last_ptr->lock);
5606 }
5608 search_start = max(search_start, first_logical_byte(root, 0));
5609 search_start = max(search_start, hint_byte);
5611 if (!last_ptr)
5612 empty_cluster = 0;
5614 if (search_start == hint_byte) {
5615 block_group = btrfs_lookup_block_group(root->fs_info,
5616 search_start);
5617 used_block_group = block_group;
5618 /*
5619 * we don't want to use the block group if it doesn't match our
5620 * allocation bits, or if its not cached.
5621 *
5622 * However if we are re-searching with an ideal block group
5623 * picked out then we don't care that the block group is cached.
5624 */
5625 if (block_group && block_group_bits(block_group, data) &&
5626 block_group->cached != BTRFS_CACHE_NO) {
5627 down_read(&space_info->groups_sem);
5628 if (list_empty(&block_group->list) ||
5629 block_group->ro) {
5630 /*
5631 * someone is removing this block group,
5632 * we can't jump into the have_block_group
5633 * target because our list pointers are not
5634 * valid
5635 */
5636 btrfs_put_block_group(block_group);
5637 up_read(&space_info->groups_sem);
5638 } else {
5639 index = get_block_group_index(block_group);
5640 goto have_block_group;
5641 }
5642 } else if (block_group) {
5643 btrfs_put_block_group(block_group);
5644 }
5645 }
5646 search:
5647 have_caching_bg = false;
5648 down_read(&space_info->groups_sem);
5649 list_for_each_entry(block_group, &space_info->block_groups[index],
5650 list) {
5651 u64 offset;
5652 int cached;
5654 used_block_group = block_group;
5655 btrfs_get_block_group(block_group);
5656 search_start = block_group->key.objectid;
5658 /*
5659 * this can happen if we end up cycling through all the
5660 * raid types, but we want to make sure we only allocate
5661 * for the proper type.
5662 */
5663 if (!block_group_bits(block_group, data)) {
5664 u64 extra = BTRFS_BLOCK_GROUP_DUP |
5665 BTRFS_BLOCK_GROUP_RAID1 |
5666 BTRFS_BLOCK_GROUP_RAID10;
5668 /*
5669 * if they asked for extra copies and this block group
5670 * doesn't provide them, bail. This does allow us to
5671 * fill raid0 from raid1.
5672 */
5673 if ((data & extra) && !(block_group->flags & extra))
5674 goto loop;
5675 }
5677 have_block_group:
5678 cached = block_group_cache_done(block_group);
5679 if (unlikely(!cached)) {
5680 found_uncached_bg = true;
5681 ret = cache_block_group(block_group, trans,
5682 orig_root, 0);
5683 BUG_ON(ret < 0);
5684 ret = 0;
5685 }
5687 if (unlikely(block_group->ro))
5688 goto loop;
5690 /*
5691 * Ok we want to try and use the cluster allocator, so
5692 * lets look there
5693 */
5694 if (last_ptr) {
5695 /*
5696 * the refill lock keeps out other
5697 * people trying to start a new cluster
5698 */
5699 spin_lock(&last_ptr->refill_lock);
5700 used_block_group = last_ptr->block_group;
5701 if (used_block_group != block_group &&
5702 (!used_block_group ||
5703 used_block_group->ro ||
5704 !block_group_bits(used_block_group, data))) {
5705 used_block_group = block_group;
5706 goto refill_cluster;
5707 }
5709 if (used_block_group != block_group)
5710 btrfs_get_block_group(used_block_group);
5712 offset = btrfs_alloc_from_cluster(used_block_group,
5713 last_ptr, num_bytes, used_block_group->key.objectid);
5714 if (offset) {
5715 /* we have a block, we're done */
5716 spin_unlock(&last_ptr->refill_lock);
5717 trace_btrfs_reserve_extent_cluster(root,
5718 block_group, search_start, num_bytes);
5719 goto checks;
5720 }
5722 WARN_ON(last_ptr->block_group != used_block_group);
5723 if (used_block_group != block_group) {
5724 btrfs_put_block_group(used_block_group);
5725 used_block_group = block_group;
5726 }
5727 refill_cluster:
5728 BUG_ON(used_block_group != block_group);
5729 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
5730 * set up a new clusters, so lets just skip it
5731 * and let the allocator find whatever block
5732 * it can find. If we reach this point, we
5733 * will have tried the cluster allocator
5734 * plenty of times and not have found
5735 * anything, so we are likely way too
5736 * fragmented for the clustering stuff to find
5737 * anything.
5738 *
5739 * However, if the cluster is taken from the
5740 * current block group, release the cluster
5741 * first, so that we stand a better chance of
5742 * succeeding in the unclustered
5743 * allocation. */
5744 if (loop >= LOOP_NO_EMPTY_SIZE &&
5745 last_ptr->block_group != block_group) {
5746 spin_unlock(&last_ptr->refill_lock);
5747 goto unclustered_alloc;
5748 }
5750 /*
5751 * this cluster didn't work out, free it and
5752 * start over
5753 */
5754 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5756 if (loop >= LOOP_NO_EMPTY_SIZE) {
5757 spin_unlock(&last_ptr->refill_lock);
5758 goto unclustered_alloc;
5759 }
5761 /* allocate a cluster in this block group */
5762 ret = btrfs_find_space_cluster(trans, root,
5763 block_group, last_ptr,
5764 search_start, num_bytes,
5765 empty_cluster + empty_size);
5766 if (ret == 0) {
5767 /*
5768 * now pull our allocation out of this
5769 * cluster
5770 */
5771 offset = btrfs_alloc_from_cluster(block_group,
5772 last_ptr, num_bytes,
5773 search_start);
5774 if (offset) {
5775 /* we found one, proceed */
5776 spin_unlock(&last_ptr->refill_lock);
5777 trace_btrfs_reserve_extent_cluster(root,
5778 block_group, search_start,
5779 num_bytes);
5780 goto checks;
5781 }
5782 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5783 && !failed_cluster_refill) {
5784 spin_unlock(&last_ptr->refill_lock);
5786 failed_cluster_refill = true;
5787 wait_block_group_cache_progress(block_group,
5788 num_bytes + empty_cluster + empty_size);
5789 goto have_block_group;
5790 }
5792 /*
5793 * at this point we either didn't find a cluster
5794 * or we weren't able to allocate a block from our
5795 * cluster. Free the cluster we've been trying
5796 * to use, and go to the next block group
5797 */
5798 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5799 spin_unlock(&last_ptr->refill_lock);
5800 goto loop;
5801 }
5803 unclustered_alloc:
5804 spin_lock(&block_group->free_space_ctl->tree_lock);
5805 if (cached &&
5806 block_group->free_space_ctl->free_space <
5807 num_bytes + empty_cluster + empty_size) {
5808 spin_unlock(&block_group->free_space_ctl->tree_lock);
5809 goto loop;
5810 }
5811 spin_unlock(&block_group->free_space_ctl->tree_lock);
5813 offset = btrfs_find_space_for_alloc(block_group, search_start,
5814 num_bytes, empty_size);
5815 /*
5816 * If we didn't find a chunk, and we haven't failed on this
5817 * block group before, and this block group is in the middle of
5818 * caching and we are ok with waiting, then go ahead and wait
5819 * for progress to be made, and set failed_alloc to true.
5820 *
5821 * If failed_alloc is true then we've already waited on this
5822 * block group once and should move on to the next block group.
5823 */
5824 if (!offset && !failed_alloc && !cached &&
5825 loop > LOOP_CACHING_NOWAIT) {
5826 wait_block_group_cache_progress(block_group,
5827 num_bytes + empty_size);
5828 failed_alloc = true;
5829 goto have_block_group;
5830 } else if (!offset) {
5831 if (!cached)
5832 have_caching_bg = true;
5833 goto loop;
5834 }
5835 checks:
5836 search_start = stripe_align(root, offset);
5838 /* move on to the next group */
5839 if (search_start + num_bytes >
5840 used_block_group->key.objectid + used_block_group->key.offset) {
5841 btrfs_add_free_space(used_block_group, offset, num_bytes);
5842 goto loop;
5843 }
5845 if (offset < search_start)
5846 btrfs_add_free_space(used_block_group, offset,
5847 search_start - offset);
5848 BUG_ON(offset > search_start);
5850 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
5851 alloc_type);
5852 if (ret == -EAGAIN) {
5853 btrfs_add_free_space(used_block_group, offset, num_bytes);
5854 goto loop;
5855 }
5857 /* we are all good, lets return */
5858 ins->objectid = search_start;
5859 ins->offset = num_bytes;
5861 trace_btrfs_reserve_extent(orig_root, block_group,
5862 search_start, num_bytes);
5863 if (used_block_group != block_group)
5864 btrfs_put_block_group(used_block_group);
5865 btrfs_put_block_group(block_group);
5866 break;
5867 loop:
5868 failed_cluster_refill = false;
5869 failed_alloc = false;
5870 BUG_ON(index != get_block_group_index(block_group));
5871 if (used_block_group != block_group)
5872 btrfs_put_block_group(used_block_group);
5873 btrfs_put_block_group(block_group);
5874 }
5875 up_read(&space_info->groups_sem);
5877 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
5878 goto search;
5880 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5881 goto search;
5883 /*
5884 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5885 * caching kthreads as we move along
5886 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5887 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5888 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5889 * again
5890 */
5891 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5892 index = 0;
5893 loop++;
5894 if (loop == LOOP_ALLOC_CHUNK) {
5895 ret = do_chunk_alloc(trans, root, data,
5896 CHUNK_ALLOC_FORCE);
5897 /*
5898 * Do not bail out on ENOSPC since we
5899 * can do more things.
5900 */
5901 if (ret < 0 && ret != -ENOSPC) {
5902 btrfs_abort_transaction(trans,
5903 root, ret);
5904 goto out;
5905 }
5906 }
5908 if (loop == LOOP_NO_EMPTY_SIZE) {
5909 empty_size = 0;
5910 empty_cluster = 0;
5911 }
5913 goto search;
5914 } else if (!ins->objectid) {
5915 ret = -ENOSPC;
5916 } else if (ins->objectid) {
5917 ret = 0;
5918 }
5919 out:
5921 return ret;
5922 }
5924 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5925 int dump_block_groups)
5926 {
5927 struct btrfs_block_group_cache *cache;
5928 int index = 0;
5930 spin_lock(&info->lock);
5931 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5932 (unsigned long long)info->flags,
5933 (unsigned long long)(info->total_bytes - info->bytes_used -
5934 info->bytes_pinned - info->bytes_reserved -
5935 info->bytes_readonly),
5936 (info->full) ? "" : "not ");
5937 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5938 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5939 (unsigned long long)info->total_bytes,
5940 (unsigned long long)info->bytes_used,
5941 (unsigned long long)info->bytes_pinned,
5942 (unsigned long long)info->bytes_reserved,
5943 (unsigned long long)info->bytes_may_use,
5944 (unsigned long long)info->bytes_readonly);
5945 spin_unlock(&info->lock);
5947 if (!dump_block_groups)
5948 return;
5950 down_read(&info->groups_sem);
5951 again:
5952 list_for_each_entry(cache, &info->block_groups[index], list) {
5953 spin_lock(&cache->lock);
5954 printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
5955 (unsigned long long)cache->key.objectid,
5956 (unsigned long long)cache->key.offset,
5957 (unsigned long long)btrfs_block_group_used(&cache->item),
5958 (unsigned long long)cache->pinned,
5959 (unsigned long long)cache->reserved,
5960 cache->ro ? "[readonly]" : "");
5961 btrfs_dump_free_space(cache, bytes);
5962 spin_unlock(&cache->lock);
5963 }
5964 if (++index < BTRFS_NR_RAID_TYPES)
5965 goto again;
5966 up_read(&info->groups_sem);
5967 }
5969 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5970 struct btrfs_root *root,
5971 u64 num_bytes, u64 min_alloc_size,
5972 u64 empty_size, u64 hint_byte,
5973 struct btrfs_key *ins, u64 data)
5974 {
5975 bool final_tried = false;
5976 int ret;
5978 data = btrfs_get_alloc_profile(root, data);
5979 again:
5980 WARN_ON(num_bytes < root->sectorsize);
5981 ret = find_free_extent(trans, root, num_bytes, empty_size,
5982 hint_byte, ins, data);
5984 if (ret == -ENOSPC) {
5985 if (!final_tried) {
5986 num_bytes = num_bytes >> 1;
5987 num_bytes = num_bytes & ~(root->sectorsize - 1);
5988 num_bytes = max(num_bytes, min_alloc_size);
5989 if (num_bytes == min_alloc_size)
5990 final_tried = true;
5991 goto again;
5992 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
5993 struct btrfs_space_info *sinfo;
5995 sinfo = __find_space_info(root->fs_info, data);
5996 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5997 "wanted %llu\n", (unsigned long long)data,
5998 (unsigned long long)num_bytes);
5999 if (sinfo)
6000 dump_space_info(sinfo, num_bytes, 1);
6001 }
6002 }
6004 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
6006 return ret;
6007 }
6009 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6010 u64 start, u64 len, int pin)
6011 {
6012 struct btrfs_block_group_cache *cache;
6013 int ret = 0;
6015 cache = btrfs_lookup_block_group(root->fs_info, start);
6016 if (!cache) {
6017 printk(KERN_ERR "Unable to find block group for %llu\n",
6018 (unsigned long long)start);
6019 return -ENOSPC;
6020 }
6022 if (btrfs_test_opt(root, DISCARD))
6023 ret = btrfs_discard_extent(root, start, len, NULL);
6025 if (pin)
6026 pin_down_extent(root, cache, start, len, 1);
6027 else {
6028 btrfs_add_free_space(cache, start, len);
6029 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6030 }
6031 btrfs_put_block_group(cache);
6033 trace_btrfs_reserved_extent_free(root, start, len);
6035 return ret;
6036 }
6038 int btrfs_free_reserved_extent(struct btrfs_root *root,
6039 u64 start, u64 len)
6040 {
6041 return __btrfs_free_reserved_extent(root, start, len, 0);
6042 }
6044 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6045 u64 start, u64 len)
6046 {
6047 return __btrfs_free_reserved_extent(root, start, len, 1);
6048 }
6050 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6051 struct btrfs_root *root,
6052 u64 parent, u64 root_objectid,
6053 u64 flags, u64 owner, u64 offset,
6054 struct btrfs_key *ins, int ref_mod)
6055 {
6056 int ret;
6057 struct btrfs_fs_info *fs_info = root->fs_info;
6058 struct btrfs_extent_item *extent_item;
6059 struct btrfs_extent_inline_ref *iref;
6060 struct btrfs_path *path;
6061 struct extent_buffer *leaf;
6062 int type;
6063 u32 size;
6065 if (parent > 0)
6066 type = BTRFS_SHARED_DATA_REF_KEY;
6067 else
6068 type = BTRFS_EXTENT_DATA_REF_KEY;
6070 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6072 path = btrfs_alloc_path();
6073 if (!path)
6074 return -ENOMEM;
6076 path->leave_spinning = 1;
6077 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6078 ins, size);
6079 if (ret) {
6080 btrfs_free_path(path);
6081 return ret;
6082 }
6084 leaf = path->nodes[0];
6085 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6086 struct btrfs_extent_item);
6087 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6088 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6089 btrfs_set_extent_flags(leaf, extent_item,
6090 flags | BTRFS_EXTENT_FLAG_DATA);
6092 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6093 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6094 if (parent > 0) {
6095 struct btrfs_shared_data_ref *ref;
6096 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6097 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6098 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6099 } else {
6100 struct btrfs_extent_data_ref *ref;
6101 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6102 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6103 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6104 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6105 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6106 }
6108 btrfs_mark_buffer_dirty(path->nodes[0]);
6109 btrfs_free_path(path);
6111 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
6112 if (ret) { /* -ENOENT, logic error */
6113 printk(KERN_ERR "btrfs update block group failed for %llu "
6114 "%llu\n", (unsigned long long)ins->objectid,
6115 (unsigned long long)ins->offset);
6116 BUG();
6117 }
6118 return ret;
6119 }
6121 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6122 struct btrfs_root *root,
6123 u64 parent, u64 root_objectid,
6124 u64 flags, struct btrfs_disk_key *key,
6125 int level, struct btrfs_key *ins)
6126 {
6127 int ret;
6128 struct btrfs_fs_info *fs_info = root->fs_info;
6129 struct btrfs_extent_item *extent_item;
6130 struct btrfs_tree_block_info *block_info;
6131 struct btrfs_extent_inline_ref *iref;
6132 struct btrfs_path *path;
6133 struct extent_buffer *leaf;
6134 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
6136 path = btrfs_alloc_path();
6137 if (!path)
6138 return -ENOMEM;
6140 path->leave_spinning = 1;
6141 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6142 ins, size);
6143 if (ret) {
6144 btrfs_free_path(path);
6145 return ret;
6146 }
6148 leaf = path->nodes[0];
6149 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6150 struct btrfs_extent_item);
6151 btrfs_set_extent_refs(leaf, extent_item, 1);
6152 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6153 btrfs_set_extent_flags(leaf, extent_item,
6154 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6155 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6157 btrfs_set_tree_block_key(leaf, block_info, key);
6158 btrfs_set_tree_block_level(leaf, block_info, level);
6160 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6161 if (parent > 0) {
6162 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6163 btrfs_set_extent_inline_ref_type(leaf, iref,
6164 BTRFS_SHARED_BLOCK_REF_KEY);
6165 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6166 } else {
6167 btrfs_set_extent_inline_ref_type(leaf, iref,
6168 BTRFS_TREE_BLOCK_REF_KEY);
6169 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6170 }
6172 btrfs_mark_buffer_dirty(leaf);
6173 btrfs_free_path(path);
6175 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
6176 if (ret) { /* -ENOENT, logic error */
6177 printk(KERN_ERR "btrfs update block group failed for %llu "
6178 "%llu\n", (unsigned long long)ins->objectid,
6179 (unsigned long long)ins->offset);
6180 BUG();
6181 }
6182 return ret;
6183 }
6185 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6186 struct btrfs_root *root,
6187 u64 root_objectid, u64 owner,
6188 u64 offset, struct btrfs_key *ins)
6189 {
6190 int ret;
6192 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6194 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6195 ins->offset, 0,
6196 root_objectid, owner, offset,
6197 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6198 return ret;
6199 }
6201 /*
6202 * this is used by the tree logging recovery code. It records that
6203 * an extent has been allocated and makes sure to clear the free
6204 * space cache bits as well
6205 */
6206 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6207 struct btrfs_root *root,
6208 u64 root_objectid, u64 owner, u64 offset,
6209 struct btrfs_key *ins)
6210 {
6211 int ret;
6212 struct btrfs_block_group_cache *block_group;
6213 struct btrfs_caching_control *caching_ctl;
6214 u64 start = ins->objectid;
6215 u64 num_bytes = ins->offset;
6217 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6218 cache_block_group(block_group, trans, NULL, 0);
6219 caching_ctl = get_caching_control(block_group);
6221 if (!caching_ctl) {
6222 BUG_ON(!block_group_cache_done(block_group));
6223 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6224 BUG_ON(ret); /* -ENOMEM */
6225 } else {
6226 mutex_lock(&caching_ctl->mutex);
6228 if (start >= caching_ctl->progress) {
6229 ret = add_excluded_extent(root, start, num_bytes);
6230 BUG_ON(ret); /* -ENOMEM */
6231 } else if (start + num_bytes <= caching_ctl->progress) {
6232 ret = btrfs_remove_free_space(block_group,
6233 start, num_bytes);
6234 BUG_ON(ret); /* -ENOMEM */
6235 } else {
6236 num_bytes = caching_ctl->progress - start;
6237 ret = btrfs_remove_free_space(block_group,
6238 start, num_bytes);
6239 BUG_ON(ret); /* -ENOMEM */
6241 start = caching_ctl->progress;
6242 num_bytes = ins->objectid + ins->offset -
6243 caching_ctl->progress;
6244 ret = add_excluded_extent(root, start, num_bytes);
6245 BUG_ON(ret); /* -ENOMEM */
6246 }
6248 mutex_unlock(&caching_ctl->mutex);
6249 put_caching_control(caching_ctl);
6250 }
6252 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6253 RESERVE_ALLOC_NO_ACCOUNT);
6254 BUG_ON(ret); /* logic error */
6255 btrfs_put_block_group(block_group);
6256 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6257 0, owner, offset, ins, 1);
6258 return ret;
6259 }
6261 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
6262 struct btrfs_root *root,
6263 u64 bytenr, u32 blocksize,
6264 int level)
6265 {
6266 struct extent_buffer *buf;
6268 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6269 if (!buf)
6270 return ERR_PTR(-ENOMEM);
6271 btrfs_set_header_generation(buf, trans->transid);
6272 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6273 btrfs_tree_lock(buf);
6274 clean_tree_block(trans, root, buf);
6275 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6277 btrfs_set_lock_blocking(buf);
6278 btrfs_set_buffer_uptodate(buf);
6280 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6281 /*
6282 * we allow two log transactions at a time, use different
6283 * EXENT bit to differentiate dirty pages.
6284 */
6285 if (root->log_transid % 2 == 0)
6286 set_extent_dirty(&root->dirty_log_pages, buf->start,
6287 buf->start + buf->len - 1, GFP_NOFS);
6288 else
6289 set_extent_new(&root->dirty_log_pages, buf->start,
6290 buf->start + buf->len - 1, GFP_NOFS);
6291 } else {
6292 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6293 buf->start + buf->len - 1, GFP_NOFS);
6294 }
6295 trans->blocks_used++;
6296 /* this returns a buffer locked for blocking */
6297 return buf;
6298 }
6300 static struct btrfs_block_rsv *
6301 use_block_rsv(struct btrfs_trans_handle *trans,
6302 struct btrfs_root *root, u32 blocksize)
6303 {
6304 struct btrfs_block_rsv *block_rsv;
6305 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6306 int ret;
6308 block_rsv = get_block_rsv(trans, root);
6310 if (block_rsv->size == 0) {
6311 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6312 BTRFS_RESERVE_NO_FLUSH);
6313 /*
6314 * If we couldn't reserve metadata bytes try and use some from
6315 * the global reserve.
6316 */
6317 if (ret && block_rsv != global_rsv) {
6318 ret = block_rsv_use_bytes(global_rsv, blocksize);
6319 if (!ret)
6320 return global_rsv;
6321 return ERR_PTR(ret);
6322 } else if (ret) {
6323 return ERR_PTR(ret);
6324 }
6325 return block_rsv;
6326 }
6328 ret = block_rsv_use_bytes(block_rsv, blocksize);
6329 if (!ret)
6330 return block_rsv;
6331 if (ret && !block_rsv->failfast) {
6332 static DEFINE_RATELIMIT_STATE(_rs,
6333 DEFAULT_RATELIMIT_INTERVAL,
6334 /*DEFAULT_RATELIMIT_BURST*/ 2);
6335 if (__ratelimit(&_rs))
6336 WARN(1, KERN_DEBUG "btrfs: block rsv returned %d\n",
6337 ret);
6338 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6339 BTRFS_RESERVE_NO_FLUSH);
6340 if (!ret) {
6341 return block_rsv;
6342 } else if (ret && block_rsv != global_rsv) {
6343 ret = block_rsv_use_bytes(global_rsv, blocksize);
6344 if (!ret)
6345 return global_rsv;
6346 }
6347 }
6349 return ERR_PTR(-ENOSPC);
6350 }
6352 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6353 struct btrfs_block_rsv *block_rsv, u32 blocksize)
6354 {
6355 block_rsv_add_bytes(block_rsv, blocksize, 0);
6356 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6357 }
6359 /*
6360 * finds a free extent and does all the dirty work required for allocation
6361 * returns the key for the extent through ins, and a tree buffer for
6362 * the first block of the extent through buf.
6363 *
6364 * returns the tree buffer or NULL.
6365 */
6366 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6367 struct btrfs_root *root, u32 blocksize,
6368 u64 parent, u64 root_objectid,
6369 struct btrfs_disk_key *key, int level,
6370 u64 hint, u64 empty_size)
6371 {
6372 struct btrfs_key ins;
6373 struct btrfs_block_rsv *block_rsv;
6374 struct extent_buffer *buf;
6375 u64 flags = 0;
6376 int ret;
6379 block_rsv = use_block_rsv(trans, root, blocksize);
6380 if (IS_ERR(block_rsv))
6381 return ERR_CAST(block_rsv);
6383 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6384 empty_size, hint, &ins, 0);
6385 if (ret) {
6386 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
6387 return ERR_PTR(ret);
6388 }
6390 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6391 blocksize, level);
6392 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
6394 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6395 if (parent == 0)
6396 parent = ins.objectid;
6397 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6398 } else
6399 BUG_ON(parent > 0);
6401 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6402 struct btrfs_delayed_extent_op *extent_op;
6403 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
6404 BUG_ON(!extent_op); /* -ENOMEM */
6405 if (key)
6406 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6407 else
6408 memset(&extent_op->key, 0, sizeof(extent_op->key));
6409 extent_op->flags_to_set = flags;
6410 extent_op->update_key = 1;
6411 extent_op->update_flags = 1;
6412 extent_op->is_data = 0;
6414 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6415 ins.objectid,
6416 ins.offset, parent, root_objectid,
6417 level, BTRFS_ADD_DELAYED_EXTENT,
6418 extent_op, 0);
6419 BUG_ON(ret); /* -ENOMEM */
6420 }
6421 return buf;
6422 }
6424 struct walk_control {
6425 u64 refs[BTRFS_MAX_LEVEL];
6426 u64 flags[BTRFS_MAX_LEVEL];
6427 struct btrfs_key update_progress;
6428 int stage;
6429 int level;
6430 int shared_level;
6431 int update_ref;
6432 int keep_locks;
6433 int reada_slot;
6434 int reada_count;
6435 int for_reloc;
6436 };
6438 #define DROP_REFERENCE 1
6439 #define UPDATE_BACKREF 2
6441 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6442 struct btrfs_root *root,
6443 struct walk_control *wc,
6444 struct btrfs_path *path)
6445 {
6446 u64 bytenr;
6447 u64 generation;
6448 u64 refs;
6449 u64 flags;
6450 u32 nritems;
6451 u32 blocksize;
6452 struct btrfs_key key;
6453 struct extent_buffer *eb;
6454 int ret;
6455 int slot;
6456 int nread = 0;
6458 if (path->slots[wc->level] < wc->reada_slot) {
6459 wc->reada_count = wc->reada_count * 2 / 3;
6460 wc->reada_count = max(wc->reada_count, 2);
6461 } else {
6462 wc->reada_count = wc->reada_count * 3 / 2;
6463 wc->reada_count = min_t(int, wc->reada_count,
6464 BTRFS_NODEPTRS_PER_BLOCK(root));
6465 }
6467 eb = path->nodes[wc->level];
6468 nritems = btrfs_header_nritems(eb);
6469 blocksize = btrfs_level_size(root, wc->level - 1);
6471 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6472 if (nread >= wc->reada_count)
6473 break;
6475 cond_resched();
6476 bytenr = btrfs_node_blockptr(eb, slot);
6477 generation = btrfs_node_ptr_generation(eb, slot);
6479 if (slot == path->slots[wc->level])
6480 goto reada;
6482 if (wc->stage == UPDATE_BACKREF &&
6483 generation <= root->root_key.offset)
6484 continue;
6486 /* We don't lock the tree block, it's OK to be racy here */
6487 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6488 &refs, &flags);
6489 /* We don't care about errors in readahead. */
6490 if (ret < 0)
6491 continue;
6492 BUG_ON(refs == 0);
6494 if (wc->stage == DROP_REFERENCE) {
6495 if (refs == 1)
6496 goto reada;
6498 if (wc->level == 1 &&
6499 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6500 continue;
6501 if (!wc->update_ref ||
6502 generation <= root->root_key.offset)
6503 continue;
6504 btrfs_node_key_to_cpu(eb, &key, slot);
6505 ret = btrfs_comp_cpu_keys(&key,
6506 &wc->update_progress);
6507 if (ret < 0)
6508 continue;
6509 } else {
6510 if (wc->level == 1 &&
6511 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6512 continue;
6513 }
6514 reada:
6515 ret = readahead_tree_block(root, bytenr, blocksize,
6516 generation);
6517 if (ret)
6518 break;
6519 nread++;
6520 }
6521 wc->reada_slot = slot;
6522 }
6524 /*
6525 * hepler to process tree block while walking down the tree.
6526 *
6527 * when wc->stage == UPDATE_BACKREF, this function updates
6528 * back refs for pointers in the block.
6529 *
6530 * NOTE: return value 1 means we should stop walking down.
6531 */
6532 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6533 struct btrfs_root *root,
6534 struct btrfs_path *path,
6535 struct walk_control *wc, int lookup_info)
6536 {
6537 int level = wc->level;
6538 struct extent_buffer *eb = path->nodes[level];
6539 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6540 int ret;
6542 if (wc->stage == UPDATE_BACKREF &&
6543 btrfs_header_owner(eb) != root->root_key.objectid)
6544 return 1;
6546 /*
6547 * when reference count of tree block is 1, it won't increase
6548 * again. once full backref flag is set, we never clear it.
6549 */
6550 if (lookup_info &&
6551 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6552 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6553 BUG_ON(!path->locks[level]);
6554 ret = btrfs_lookup_extent_info(trans, root,
6555 eb->start, eb->len,
6556 &wc->refs[level],
6557 &wc->flags[level]);
6558 BUG_ON(ret == -ENOMEM);
6559 if (ret)
6560 return ret;
6561 BUG_ON(wc->refs[level] == 0);
6562 }
6564 if (wc->stage == DROP_REFERENCE) {
6565 if (wc->refs[level] > 1)
6566 return 1;
6568 if (path->locks[level] && !wc->keep_locks) {
6569 btrfs_tree_unlock_rw(eb, path->locks[level]);
6570 path->locks[level] = 0;
6571 }
6572 return 0;
6573 }
6575 /* wc->stage == UPDATE_BACKREF */
6576 if (!(wc->flags[level] & flag)) {
6577 BUG_ON(!path->locks[level]);
6578 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
6579 BUG_ON(ret); /* -ENOMEM */
6580 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
6581 BUG_ON(ret); /* -ENOMEM */
6582 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6583 eb->len, flag, 0);
6584 BUG_ON(ret); /* -ENOMEM */
6585 wc->flags[level] |= flag;
6586 }
6588 /*
6589 * the block is shared by multiple trees, so it's not good to
6590 * keep the tree lock
6591 */
6592 if (path->locks[level] && level > 0) {
6593 btrfs_tree_unlock_rw(eb, path->locks[level]);
6594 path->locks[level] = 0;
6595 }
6596 return 0;
6597 }
6599 /*
6600 * hepler to process tree block pointer.
6601 *
6602 * when wc->stage == DROP_REFERENCE, this function checks
6603 * reference count of the block pointed to. if the block
6604 * is shared and we need update back refs for the subtree
6605 * rooted at the block, this function changes wc->stage to
6606 * UPDATE_BACKREF. if the block is shared and there is no
6607 * need to update back, this function drops the reference
6608 * to the block.
6609 *
6610 * NOTE: return value 1 means we should stop walking down.
6611 */
6612 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6613 struct btrfs_root *root,
6614 struct btrfs_path *path,
6615 struct walk_control *wc, int *lookup_info)
6616 {
6617 u64 bytenr;
6618 u64 generation;
6619 u64 parent;
6620 u32 blocksize;
6621 struct btrfs_key key;
6622 struct extent_buffer *next;
6623 int level = wc->level;
6624 int reada = 0;
6625 int ret = 0;
6627 generation = btrfs_node_ptr_generation(path->nodes[level],
6628 path->slots[level]);
6629 /*
6630 * if the lower level block was created before the snapshot
6631 * was created, we know there is no need to update back refs
6632 * for the subtree
6633 */
6634 if (wc->stage == UPDATE_BACKREF &&
6635 generation <= root->root_key.offset) {
6636 *lookup_info = 1;
6637 return 1;
6638 }
6640 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6641 blocksize = btrfs_level_size(root, level - 1);
6643 next = btrfs_find_tree_block(root, bytenr, blocksize);
6644 if (!next) {
6645 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6646 if (!next)
6647 return -ENOMEM;
6648 reada = 1;
6649 }
6650 btrfs_tree_lock(next);
6651 btrfs_set_lock_blocking(next);
6653 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6654 &wc->refs[level - 1],
6655 &wc->flags[level - 1]);
6656 if (ret < 0) {
6657 btrfs_tree_unlock(next);
6658 return ret;
6659 }
6661 BUG_ON(wc->refs[level - 1] == 0);
6662 *lookup_info = 0;
6664 if (wc->stage == DROP_REFERENCE) {
6665 if (wc->refs[level - 1] > 1) {
6666 if (level == 1 &&
6667 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6668 goto skip;
6670 if (!wc->update_ref ||
6671 generation <= root->root_key.offset)
6672 goto skip;
6674 btrfs_node_key_to_cpu(path->nodes[level], &key,
6675 path->slots[level]);
6676 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6677 if (ret < 0)
6678 goto skip;
6680 wc->stage = UPDATE_BACKREF;
6681 wc->shared_level = level - 1;
6682 }
6683 } else {
6684 if (level == 1 &&
6685 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6686 goto skip;
6687 }
6689 if (!btrfs_buffer_uptodate(next, generation, 0)) {
6690 btrfs_tree_unlock(next);
6691 free_extent_buffer(next);
6692 next = NULL;
6693 *lookup_info = 1;
6694 }
6696 if (!next) {
6697 if (reada && level == 1)
6698 reada_walk_down(trans, root, wc, path);
6699 next = read_tree_block(root, bytenr, blocksize, generation);
6700 if (!next)
6701 return -EIO;
6702 btrfs_tree_lock(next);
6703 btrfs_set_lock_blocking(next);
6704 }
6706 level--;
6707 BUG_ON(level != btrfs_header_level(next));
6708 path->nodes[level] = next;
6709 path->slots[level] = 0;
6710 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6711 wc->level = level;
6712 if (wc->level == 1)
6713 wc->reada_slot = 0;
6714 return 0;
6715 skip:
6716 wc->refs[level - 1] = 0;
6717 wc->flags[level - 1] = 0;
6718 if (wc->stage == DROP_REFERENCE) {
6719 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6720 parent = path->nodes[level]->start;
6721 } else {
6722 BUG_ON(root->root_key.objectid !=
6723 btrfs_header_owner(path->nodes[level]));
6724 parent = 0;
6725 }
6727 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6728 root->root_key.objectid, level - 1, 0, 0);
6729 BUG_ON(ret); /* -ENOMEM */
6730 }
6731 btrfs_tree_unlock(next);
6732 free_extent_buffer(next);
6733 *lookup_info = 1;
6734 return 1;
6735 }
6737 /*
6738 * hepler to process tree block while walking up the tree.
6739 *
6740 * when wc->stage == DROP_REFERENCE, this function drops
6741 * reference count on the block.
6742 *
6743 * when wc->stage == UPDATE_BACKREF, this function changes
6744 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6745 * to UPDATE_BACKREF previously while processing the block.
6746 *
6747 * NOTE: return value 1 means we should stop walking up.
6748 */
6749 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6750 struct btrfs_root *root,
6751 struct btrfs_path *path,
6752 struct walk_control *wc)
6753 {
6754 int ret;
6755 int level = wc->level;
6756 struct extent_buffer *eb = path->nodes[level];
6757 u64 parent = 0;
6759 if (wc->stage == UPDATE_BACKREF) {
6760 BUG_ON(wc->shared_level < level);
6761 if (level < wc->shared_level)
6762 goto out;
6764 ret = find_next_key(path, level + 1, &wc->update_progress);
6765 if (ret > 0)
6766 wc->update_ref = 0;
6768 wc->stage = DROP_REFERENCE;
6769 wc->shared_level = -1;
6770 path->slots[level] = 0;
6772 /*
6773 * check reference count again if the block isn't locked.
6774 * we should start walking down the tree again if reference
6775 * count is one.
6776 */
6777 if (!path->locks[level]) {
6778 BUG_ON(level == 0);
6779 btrfs_tree_lock(eb);
6780 btrfs_set_lock_blocking(eb);
6781 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6783 ret = btrfs_lookup_extent_info(trans, root,
6784 eb->start, eb->len,
6785 &wc->refs[level],
6786 &wc->flags[level]);
6787 if (ret < 0) {
6788 btrfs_tree_unlock_rw(eb, path->locks[level]);
6789 path->locks[level] = 0;
6790 return ret;
6791 }
6792 BUG_ON(wc->refs[level] == 0);
6793 if (wc->refs[level] == 1) {
6794 btrfs_tree_unlock_rw(eb, path->locks[level]);
6795 path->locks[level] = 0;
6796 return 1;
6797 }
6798 }
6799 }
6801 /* wc->stage == DROP_REFERENCE */
6802 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6804 if (wc->refs[level] == 1) {
6805 if (level == 0) {
6806 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6807 ret = btrfs_dec_ref(trans, root, eb, 1,
6808 wc->for_reloc);
6809 else
6810 ret = btrfs_dec_ref(trans, root, eb, 0,
6811 wc->for_reloc);
6812 BUG_ON(ret); /* -ENOMEM */
6813 }
6814 /* make block locked assertion in clean_tree_block happy */
6815 if (!path->locks[level] &&
6816 btrfs_header_generation(eb) == trans->transid) {
6817 btrfs_tree_lock(eb);
6818 btrfs_set_lock_blocking(eb);
6819 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6820 }
6821 clean_tree_block(trans, root, eb);
6822 }
6824 if (eb == root->node) {
6825 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6826 parent = eb->start;
6827 else
6828 BUG_ON(root->root_key.objectid !=
6829 btrfs_header_owner(eb));
6830 } else {
6831 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6832 parent = path->nodes[level + 1]->start;
6833 else
6834 BUG_ON(root->root_key.objectid !=
6835 btrfs_header_owner(path->nodes[level + 1]));
6836 }
6838 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6839 out:
6840 wc->refs[level] = 0;
6841 wc->flags[level] = 0;
6842 return 0;
6843 }
6845 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6846 struct btrfs_root *root,
6847 struct btrfs_path *path,
6848 struct walk_control *wc)
6849 {
6850 int level = wc->level;
6851 int lookup_info = 1;
6852 int ret;
6854 while (level >= 0) {
6855 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6856 if (ret > 0)
6857 break;
6859 if (level == 0)
6860 break;
6862 if (path->slots[level] >=
6863 btrfs_header_nritems(path->nodes[level]))
6864 break;
6866 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6867 if (ret > 0) {
6868 path->slots[level]++;
6869 continue;
6870 } else if (ret < 0)
6871 return ret;
6872 level = wc->level;
6873 }
6874 return 0;
6875 }
6877 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6878 struct btrfs_root *root,
6879 struct btrfs_path *path,
6880 struct walk_control *wc, int max_level)
6881 {
6882 int level = wc->level;
6883 int ret;
6885 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6886 while (level < max_level && path->nodes[level]) {
6887 wc->level = level;
6888 if (path->slots[level] + 1 <
6889 btrfs_header_nritems(path->nodes[level])) {
6890 path->slots[level]++;
6891 return 0;
6892 } else {
6893 ret = walk_up_proc(trans, root, path, wc);
6894 if (ret > 0)
6895 return 0;
6897 if (path->locks[level]) {
6898 btrfs_tree_unlock_rw(path->nodes[level],
6899 path->locks[level]);
6900 path->locks[level] = 0;
6901 }
6902 free_extent_buffer(path->nodes[level]);
6903 path->nodes[level] = NULL;
6904 level++;
6905 }
6906 }
6907 return 1;
6908 }
6910 /*
6911 * drop a subvolume tree.
6912 *
6913 * this function traverses the tree freeing any blocks that only
6914 * referenced by the tree.
6915 *
6916 * when a shared tree block is found. this function decreases its
6917 * reference count by one. if update_ref is true, this function
6918 * also make sure backrefs for the shared block and all lower level
6919 * blocks are properly updated.
6920 */
6921 int btrfs_drop_snapshot(struct btrfs_root *root,
6922 struct btrfs_block_rsv *block_rsv, int update_ref,
6923 int for_reloc)
6924 {
6925 struct btrfs_path *path;
6926 struct btrfs_trans_handle *trans;
6927 struct btrfs_root *tree_root = root->fs_info->tree_root;
6928 struct btrfs_root_item *root_item = &root->root_item;
6929 struct walk_control *wc;
6930 struct btrfs_key key;
6931 int err = 0;
6932 int ret;
6933 int level;
6935 path = btrfs_alloc_path();
6936 if (!path) {
6937 err = -ENOMEM;
6938 goto out;
6939 }
6941 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6942 if (!wc) {
6943 btrfs_free_path(path);
6944 err = -ENOMEM;
6945 goto out;
6946 }
6948 trans = btrfs_start_transaction(tree_root, 0);
6949 if (IS_ERR(trans)) {
6950 err = PTR_ERR(trans);
6951 goto out_free;
6952 }
6954 if (block_rsv)
6955 trans->block_rsv = block_rsv;
6957 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6958 level = btrfs_header_level(root->node);
6959 path->nodes[level] = btrfs_lock_root_node(root);
6960 btrfs_set_lock_blocking(path->nodes[level]);
6961 path->slots[level] = 0;
6962 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6963 memset(&wc->update_progress, 0,
6964 sizeof(wc->update_progress));
6965 } else {
6966 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6967 memcpy(&wc->update_progress, &key,
6968 sizeof(wc->update_progress));
6970 level = root_item->drop_level;
6971 BUG_ON(level == 0);
6972 path->lowest_level = level;
6973 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6974 path->lowest_level = 0;
6975 if (ret < 0) {
6976 err = ret;
6977 goto out_end_trans;
6978 }
6979 WARN_ON(ret > 0);
6981 /*
6982 * unlock our path, this is safe because only this
6983 * function is allowed to delete this snapshot
6984 */
6985 btrfs_unlock_up_safe(path, 0);
6987 level = btrfs_header_level(root->node);
6988 while (1) {
6989 btrfs_tree_lock(path->nodes[level]);
6990 btrfs_set_lock_blocking(path->nodes[level]);
6992 ret = btrfs_lookup_extent_info(trans, root,
6993 path->nodes[level]->start,
6994 path->nodes[level]->len,
6995 &wc->refs[level],
6996 &wc->flags[level]);
6997 if (ret < 0) {
6998 err = ret;
6999 goto out_end_trans;
7000 }
7001 BUG_ON(wc->refs[level] == 0);
7003 if (level == root_item->drop_level)
7004 break;
7006 btrfs_tree_unlock(path->nodes[level]);
7007 WARN_ON(wc->refs[level] != 1);
7008 level--;
7009 }
7010 }
7012 wc->level = level;
7013 wc->shared_level = -1;
7014 wc->stage = DROP_REFERENCE;
7015 wc->update_ref = update_ref;
7016 wc->keep_locks = 0;
7017 wc->for_reloc = for_reloc;
7018 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7020 while (1) {
7021 ret = walk_down_tree(trans, root, path, wc);
7022 if (ret < 0) {
7023 err = ret;
7024 break;
7025 }
7027 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7028 if (ret < 0) {
7029 err = ret;
7030 break;
7031 }
7033 if (ret > 0) {
7034 BUG_ON(wc->stage != DROP_REFERENCE);
7035 break;
7036 }
7038 if (wc->stage == DROP_REFERENCE) {
7039 level = wc->level;
7040 btrfs_node_key(path->nodes[level],
7041 &root_item->drop_progress,
7042 path->slots[level]);
7043 root_item->drop_level = level;
7044 }
7046 BUG_ON(wc->level == 0);
7047 if (btrfs_should_end_transaction(trans, tree_root)) {
7048 ret = btrfs_update_root(trans, tree_root,
7049 &root->root_key,
7050 root_item);
7051 if (ret) {
7052 btrfs_abort_transaction(trans, tree_root, ret);
7053 err = ret;
7054 goto out_end_trans;
7055 }
7057 btrfs_end_transaction_throttle(trans, tree_root);
7058 trans = btrfs_start_transaction(tree_root, 0);
7059 if (IS_ERR(trans)) {
7060 err = PTR_ERR(trans);
7061 goto out_free;
7062 }
7063 if (block_rsv)
7064 trans->block_rsv = block_rsv;
7065 }
7066 }
7067 btrfs_release_path(path);
7068 if (err)
7069 goto out_end_trans;
7071 ret = btrfs_del_root(trans, tree_root, &root->root_key);
7072 if (ret) {
7073 btrfs_abort_transaction(trans, tree_root, ret);
7074 goto out_end_trans;
7075 }
7077 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7078 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
7079 NULL, NULL);
7080 if (ret < 0) {
7081 btrfs_abort_transaction(trans, tree_root, ret);
7082 err = ret;
7083 goto out_end_trans;
7084 } else if (ret > 0) {
7085 /* if we fail to delete the orphan item this time
7086 * around, it'll get picked up the next time.
7087 *
7088 * The most common failure here is just -ENOENT.
7089 */
7090 btrfs_del_orphan_item(trans, tree_root,
7091 root->root_key.objectid);
7092 }
7093 }
7095 if (root->in_radix) {
7096 btrfs_free_fs_root(tree_root->fs_info, root);
7097 } else {
7098 free_extent_buffer(root->node);
7099 free_extent_buffer(root->commit_root);
7100 kfree(root);
7101 }
7102 out_end_trans:
7103 btrfs_end_transaction_throttle(trans, tree_root);
7104 out_free:
7105 kfree(wc);
7106 btrfs_free_path(path);
7107 out:
7108 if (err)
7109 btrfs_std_error(root->fs_info, err);
7110 return err;
7111 }
7113 /*
7114 * drop subtree rooted at tree block 'node'.
7115 *
7116 * NOTE: this function will unlock and release tree block 'node'
7117 * only used by relocation code
7118 */
7119 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7120 struct btrfs_root *root,
7121 struct extent_buffer *node,
7122 struct extent_buffer *parent)
7123 {
7124 struct btrfs_path *path;
7125 struct walk_control *wc;
7126 int level;
7127 int parent_level;
7128 int ret = 0;
7129 int wret;
7131 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7133 path = btrfs_alloc_path();
7134 if (!path)
7135 return -ENOMEM;
7137 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7138 if (!wc) {
7139 btrfs_free_path(path);
7140 return -ENOMEM;
7141 }
7143 btrfs_assert_tree_locked(parent);
7144 parent_level = btrfs_header_level(parent);
7145 extent_buffer_get(parent);
7146 path->nodes[parent_level] = parent;
7147 path->slots[parent_level] = btrfs_header_nritems(parent);
7149 btrfs_assert_tree_locked(node);
7150 level = btrfs_header_level(node);
7151 path->nodes[level] = node;
7152 path->slots[level] = 0;
7153 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7155 wc->refs[parent_level] = 1;
7156 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7157 wc->level = level;
7158 wc->shared_level = -1;
7159 wc->stage = DROP_REFERENCE;
7160 wc->update_ref = 0;
7161 wc->keep_locks = 1;
7162 wc->for_reloc = 1;
7163 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7165 while (1) {
7166 wret = walk_down_tree(trans, root, path, wc);
7167 if (wret < 0) {
7168 ret = wret;
7169 break;
7170 }
7172 wret = walk_up_tree(trans, root, path, wc, parent_level);
7173 if (wret < 0)
7174 ret = wret;
7175 if (wret != 0)
7176 break;
7177 }
7179 kfree(wc);
7180 btrfs_free_path(path);
7181 return ret;
7182 }
7184 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7185 {
7186 u64 num_devices;
7187 u64 stripped;
7189 /*
7190 * if restripe for this chunk_type is on pick target profile and
7191 * return, otherwise do the usual balance
7192 */
7193 stripped = get_restripe_target(root->fs_info, flags);
7194 if (stripped)
7195 return extended_to_chunk(stripped);
7197 /*
7198 * we add in the count of missing devices because we want
7199 * to make sure that any RAID levels on a degraded FS
7200 * continue to be honored.
7201 */
7202 num_devices = root->fs_info->fs_devices->rw_devices +
7203 root->fs_info->fs_devices->missing_devices;
7205 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7206 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7208 if (num_devices == 1) {
7209 stripped |= BTRFS_BLOCK_GROUP_DUP;
7210 stripped = flags & ~stripped;
7212 /* turn raid0 into single device chunks */
7213 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7214 return stripped;
7216 /* turn mirroring into duplication */
7217 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7218 BTRFS_BLOCK_GROUP_RAID10))
7219 return stripped | BTRFS_BLOCK_GROUP_DUP;
7220 } else {
7221 /* they already had raid on here, just return */
7222 if (flags & stripped)
7223 return flags;
7225 stripped |= BTRFS_BLOCK_GROUP_DUP;
7226 stripped = flags & ~stripped;
7228 /* switch duplicated blocks with raid1 */
7229 if (flags & BTRFS_BLOCK_GROUP_DUP)
7230 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7232 /* this is drive concat, leave it alone */
7233 }
7235 return flags;
7236 }
7238 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7239 {
7240 struct btrfs_space_info *sinfo = cache->space_info;
7241 u64 num_bytes;
7242 u64 min_allocable_bytes;
7243 int ret = -ENOSPC;
7246 /*
7247 * We need some metadata space and system metadata space for
7248 * allocating chunks in some corner cases until we force to set
7249 * it to be readonly.
7250 */
7251 if ((sinfo->flags &
7252 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7253 !force)
7254 min_allocable_bytes = 1 * 1024 * 1024;
7255 else
7256 min_allocable_bytes = 0;
7258 spin_lock(&sinfo->lock);
7259 spin_lock(&cache->lock);
7261 if (cache->ro) {
7262 ret = 0;
7263 goto out;
7264 }
7266 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7267 cache->bytes_super - btrfs_block_group_used(&cache->item);
7269 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7270 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7271 min_allocable_bytes <= sinfo->total_bytes) {
7272 sinfo->bytes_readonly += num_bytes;
7273 cache->ro = 1;
7274 ret = 0;
7275 }
7276 out:
7277 spin_unlock(&cache->lock);
7278 spin_unlock(&sinfo->lock);
7279 return ret;
7280 }
7282 int btrfs_set_block_group_ro(struct btrfs_root *root,
7283 struct btrfs_block_group_cache *cache)
7285 {
7286 struct btrfs_trans_handle *trans;
7287 u64 alloc_flags;
7288 int ret;
7290 BUG_ON(cache->ro);
7292 trans = btrfs_join_transaction(root);
7293 if (IS_ERR(trans))
7294 return PTR_ERR(trans);
7296 alloc_flags = update_block_group_flags(root, cache->flags);
7297 if (alloc_flags != cache->flags) {
7298 ret = do_chunk_alloc(trans, root, alloc_flags,
7299 CHUNK_ALLOC_FORCE);
7300 if (ret < 0)
7301 goto out;
7302 }
7304 ret = set_block_group_ro(cache, 0);
7305 if (!ret)
7306 goto out;
7307 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7308 ret = do_chunk_alloc(trans, root, alloc_flags,
7309 CHUNK_ALLOC_FORCE);
7310 if (ret < 0)
7311 goto out;
7312 ret = set_block_group_ro(cache, 0);
7313 out:
7314 btrfs_end_transaction(trans, root);
7315 return ret;
7316 }
7318 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7319 struct btrfs_root *root, u64 type)
7320 {
7321 u64 alloc_flags = get_alloc_profile(root, type);
7322 return do_chunk_alloc(trans, root, alloc_flags,
7323 CHUNK_ALLOC_FORCE);
7324 }
7326 /*
7327 * helper to account the unused space of all the readonly block group in the
7328 * list. takes mirrors into account.
7329 */
7330 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7331 {
7332 struct btrfs_block_group_cache *block_group;
7333 u64 free_bytes = 0;
7334 int factor;
7336 list_for_each_entry(block_group, groups_list, list) {
7337 spin_lock(&block_group->lock);
7339 if (!block_group->ro) {
7340 spin_unlock(&block_group->lock);
7341 continue;
7342 }
7344 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7345 BTRFS_BLOCK_GROUP_RAID10 |
7346 BTRFS_BLOCK_GROUP_DUP))
7347 factor = 2;
7348 else
7349 factor = 1;
7351 free_bytes += (block_group->key.offset -
7352 btrfs_block_group_used(&block_group->item)) *
7353 factor;
7355 spin_unlock(&block_group->lock);
7356 }
7358 return free_bytes;
7359 }
7361 /*
7362 * helper to account the unused space of all the readonly block group in the
7363 * space_info. takes mirrors into account.
7364 */
7365 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7366 {
7367 int i;
7368 u64 free_bytes = 0;
7370 spin_lock(&sinfo->lock);
7372 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7373 if (!list_empty(&sinfo->block_groups[i]))
7374 free_bytes += __btrfs_get_ro_block_group_free_space(
7375 &sinfo->block_groups[i]);
7377 spin_unlock(&sinfo->lock);
7379 return free_bytes;
7380 }
7382 void btrfs_set_block_group_rw(struct btrfs_root *root,
7383 struct btrfs_block_group_cache *cache)
7384 {
7385 struct btrfs_space_info *sinfo = cache->space_info;
7386 u64 num_bytes;
7388 BUG_ON(!cache->ro);
7390 spin_lock(&sinfo->lock);
7391 spin_lock(&cache->lock);
7392 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7393 cache->bytes_super - btrfs_block_group_used(&cache->item);
7394 sinfo->bytes_readonly -= num_bytes;
7395 cache->ro = 0;
7396 spin_unlock(&cache->lock);
7397 spin_unlock(&sinfo->lock);
7398 }
7400 /*
7401 * checks to see if its even possible to relocate this block group.
7402 *
7403 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7404 * ok to go ahead and try.
7405 */
7406 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7407 {
7408 struct btrfs_block_group_cache *block_group;
7409 struct btrfs_space_info *space_info;
7410 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7411 struct btrfs_device *device;
7412 u64 min_free;
7413 u64 dev_min = 1;
7414 u64 dev_nr = 0;
7415 u64 target;
7416 int index;
7417 int full = 0;
7418 int ret = 0;
7420 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7422 /* odd, couldn't find the block group, leave it alone */
7423 if (!block_group)
7424 return -1;
7426 min_free = btrfs_block_group_used(&block_group->item);
7428 /* no bytes used, we're good */
7429 if (!min_free)
7430 goto out;
7432 space_info = block_group->space_info;
7433 spin_lock(&space_info->lock);
7435 full = space_info->full;
7437 /*
7438 * if this is the last block group we have in this space, we can't
7439 * relocate it unless we're able to allocate a new chunk below.
7440 *
7441 * Otherwise, we need to make sure we have room in the space to handle
7442 * all of the extents from this block group. If we can, we're good
7443 */
7444 if ((space_info->total_bytes != block_group->key.offset) &&
7445 (space_info->bytes_used + space_info->bytes_reserved +
7446 space_info->bytes_pinned + space_info->bytes_readonly +
7447 min_free < space_info->total_bytes)) {
7448 spin_unlock(&space_info->lock);
7449 goto out;
7450 }
7451 spin_unlock(&space_info->lock);
7453 /*
7454 * ok we don't have enough space, but maybe we have free space on our
7455 * devices to allocate new chunks for relocation, so loop through our
7456 * alloc devices and guess if we have enough space. if this block
7457 * group is going to be restriped, run checks against the target
7458 * profile instead of the current one.
7459 */
7460 ret = -1;
7462 /*
7463 * index:
7464 * 0: raid10
7465 * 1: raid1
7466 * 2: dup
7467 * 3: raid0
7468 * 4: single
7469 */
7470 target = get_restripe_target(root->fs_info, block_group->flags);
7471 if (target) {
7472 index = __get_raid_index(extended_to_chunk(target));
7473 } else {
7474 /*
7475 * this is just a balance, so if we were marked as full
7476 * we know there is no space for a new chunk
7477 */
7478 if (full)
7479 goto out;
7481 index = get_block_group_index(block_group);
7482 }
7484 if (index == 0) {
7485 dev_min = 4;
7486 /* Divide by 2 */
7487 min_free >>= 1;
7488 } else if (index == 1) {
7489 dev_min = 2;
7490 } else if (index == 2) {
7491 /* Multiply by 2 */
7492 min_free <<= 1;
7493 } else if (index == 3) {
7494 dev_min = fs_devices->rw_devices;
7495 do_div(min_free, dev_min);
7496 }
7498 mutex_lock(&root->fs_info->chunk_mutex);
7499 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7500 u64 dev_offset;
7502 /*
7503 * check to make sure we can actually find a chunk with enough
7504 * space to fit our block group in.
7505 */
7506 if (device->total_bytes > device->bytes_used + min_free &&
7507 !device->is_tgtdev_for_dev_replace) {
7508 ret = find_free_dev_extent(device, min_free,
7509 &dev_offset, NULL);
7510 if (!ret)
7511 dev_nr++;
7513 if (dev_nr >= dev_min)
7514 break;
7516 ret = -1;
7517 }
7518 }
7519 mutex_unlock(&root->fs_info->chunk_mutex);
7520 out:
7521 btrfs_put_block_group(block_group);
7522 return ret;
7523 }
7525 static int find_first_block_group(struct btrfs_root *root,
7526 struct btrfs_path *path, struct btrfs_key *key)
7527 {
7528 int ret = 0;
7529 struct btrfs_key found_key;
7530 struct extent_buffer *leaf;
7531 int slot;
7533 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7534 if (ret < 0)
7535 goto out;
7537 while (1) {
7538 slot = path->slots[0];
7539 leaf = path->nodes[0];
7540 if (slot >= btrfs_header_nritems(leaf)) {
7541 ret = btrfs_next_leaf(root, path);
7542 if (ret == 0)
7543 continue;
7544 if (ret < 0)
7545 goto out;
7546 break;
7547 }
7548 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7550 if (found_key.objectid >= key->objectid &&
7551 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7552 ret = 0;
7553 goto out;
7554 }
7555 path->slots[0]++;
7556 }
7557 out:
7558 return ret;
7559 }
7561 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7562 {
7563 struct btrfs_block_group_cache *block_group;
7564 u64 last = 0;
7566 while (1) {
7567 struct inode *inode;
7569 block_group = btrfs_lookup_first_block_group(info, last);
7570 while (block_group) {
7571 spin_lock(&block_group->lock);
7572 if (block_group->iref)
7573 break;
7574 spin_unlock(&block_group->lock);
7575 block_group = next_block_group(info->tree_root,
7576 block_group);
7577 }
7578 if (!block_group) {
7579 if (last == 0)
7580 break;
7581 last = 0;
7582 continue;
7583 }
7585 inode = block_group->inode;
7586 block_group->iref = 0;
7587 block_group->inode = NULL;
7588 spin_unlock(&block_group->lock);
7589 iput(inode);
7590 last = block_group->key.objectid + block_group->key.offset;
7591 btrfs_put_block_group(block_group);
7592 }
7593 }
7595 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7596 {
7597 struct btrfs_block_group_cache *block_group;
7598 struct btrfs_space_info *space_info;
7599 struct btrfs_caching_control *caching_ctl;
7600 struct rb_node *n;
7602 down_write(&info->extent_commit_sem);
7603 while (!list_empty(&info->caching_block_groups)) {
7604 caching_ctl = list_entry(info->caching_block_groups.next,
7605 struct btrfs_caching_control, list);
7606 list_del(&caching_ctl->list);
7607 put_caching_control(caching_ctl);
7608 }
7609 up_write(&info->extent_commit_sem);
7611 spin_lock(&info->block_group_cache_lock);
7612 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7613 block_group = rb_entry(n, struct btrfs_block_group_cache,
7614 cache_node);
7615 rb_erase(&block_group->cache_node,
7616 &info->block_group_cache_tree);
7617 spin_unlock(&info->block_group_cache_lock);
7619 down_write(&block_group->space_info->groups_sem);
7620 list_del(&block_group->list);
7621 up_write(&block_group->space_info->groups_sem);
7623 if (block_group->cached == BTRFS_CACHE_STARTED)
7624 wait_block_group_cache_done(block_group);
7626 /*
7627 * We haven't cached this block group, which means we could
7628 * possibly have excluded extents on this block group.
7629 */
7630 if (block_group->cached == BTRFS_CACHE_NO)
7631 free_excluded_extents(info->extent_root, block_group);
7633 btrfs_remove_free_space_cache(block_group);
7634 btrfs_put_block_group(block_group);
7636 spin_lock(&info->block_group_cache_lock);
7637 }
7638 spin_unlock(&info->block_group_cache_lock);
7640 /* now that all the block groups are freed, go through and
7641 * free all the space_info structs. This is only called during
7642 * the final stages of unmount, and so we know nobody is
7643 * using them. We call synchronize_rcu() once before we start,
7644 * just to be on the safe side.
7645 */
7646 synchronize_rcu();
7648 release_global_block_rsv(info);
7650 while(!list_empty(&info->space_info)) {
7651 space_info = list_entry(info->space_info.next,
7652 struct btrfs_space_info,
7653 list);
7654 if (space_info->bytes_pinned > 0 ||
7655 space_info->bytes_reserved > 0 ||
7656 space_info->bytes_may_use > 0) {
7657 WARN_ON(1);
7658 dump_space_info(space_info, 0, 0);
7659 }
7660 list_del(&space_info->list);
7661 kfree(space_info);
7662 }
7663 return 0;
7664 }
7666 static void __link_block_group(struct btrfs_space_info *space_info,
7667 struct btrfs_block_group_cache *cache)
7668 {
7669 int index = get_block_group_index(cache);
7671 down_write(&space_info->groups_sem);
7672 list_add_tail(&cache->list, &space_info->block_groups[index]);
7673 up_write(&space_info->groups_sem);
7674 }
7676 int btrfs_read_block_groups(struct btrfs_root *root)
7677 {
7678 struct btrfs_path *path;
7679 int ret;
7680 struct btrfs_block_group_cache *cache;
7681 struct btrfs_fs_info *info = root->fs_info;
7682 struct btrfs_space_info *space_info;
7683 struct btrfs_key key;
7684 struct btrfs_key found_key;
7685 struct extent_buffer *leaf;
7686 int need_clear = 0;
7687 u64 cache_gen;
7689 root = info->extent_root;
7690 key.objectid = 0;
7691 key.offset = 0;
7692 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7693 path = btrfs_alloc_path();
7694 if (!path)
7695 return -ENOMEM;
7696 path->reada = 1;
7698 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
7699 if (btrfs_test_opt(root, SPACE_CACHE) &&
7700 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
7701 need_clear = 1;
7702 if (btrfs_test_opt(root, CLEAR_CACHE))
7703 need_clear = 1;
7705 while (1) {
7706 ret = find_first_block_group(root, path, &key);
7707 if (ret > 0)
7708 break;
7709 if (ret != 0)
7710 goto error;
7711 leaf = path->nodes[0];
7712 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7713 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7714 if (!cache) {
7715 ret = -ENOMEM;
7716 goto error;
7717 }
7718 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7719 GFP_NOFS);
7720 if (!cache->free_space_ctl) {
7721 kfree(cache);
7722 ret = -ENOMEM;
7723 goto error;
7724 }
7726 atomic_set(&cache->count, 1);
7727 spin_lock_init(&cache->lock);
7728 cache->fs_info = info;
7729 INIT_LIST_HEAD(&cache->list);
7730 INIT_LIST_HEAD(&cache->cluster_list);
7732 if (need_clear) {
7733 /*
7734 * When we mount with old space cache, we need to
7735 * set BTRFS_DC_CLEAR and set dirty flag.
7736 *
7737 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
7738 * truncate the old free space cache inode and
7739 * setup a new one.
7740 * b) Setting 'dirty flag' makes sure that we flush
7741 * the new space cache info onto disk.
7742 */
7743 cache->disk_cache_state = BTRFS_DC_CLEAR;
7744 if (btrfs_test_opt(root, SPACE_CACHE))
7745 cache->dirty = 1;
7746 }
7748 read_extent_buffer(leaf, &cache->item,
7749 btrfs_item_ptr_offset(leaf, path->slots[0]),
7750 sizeof(cache->item));
7751 memcpy(&cache->key, &found_key, sizeof(found_key));
7753 key.objectid = found_key.objectid + found_key.offset;
7754 btrfs_release_path(path);
7755 cache->flags = btrfs_block_group_flags(&cache->item);
7756 cache->sectorsize = root->sectorsize;
7758 btrfs_init_free_space_ctl(cache);
7760 /*
7761 * We need to exclude the super stripes now so that the space
7762 * info has super bytes accounted for, otherwise we'll think
7763 * we have more space than we actually do.
7764 */
7765 exclude_super_stripes(root, cache);
7767 /*
7768 * check for two cases, either we are full, and therefore
7769 * don't need to bother with the caching work since we won't
7770 * find any space, or we are empty, and we can just add all
7771 * the space in and be done with it. This saves us _alot_ of
7772 * time, particularly in the full case.
7773 */
7774 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7775 cache->last_byte_to_unpin = (u64)-1;
7776 cache->cached = BTRFS_CACHE_FINISHED;
7777 free_excluded_extents(root, cache);
7778 } else if (btrfs_block_group_used(&cache->item) == 0) {
7779 cache->last_byte_to_unpin = (u64)-1;
7780 cache->cached = BTRFS_CACHE_FINISHED;
7781 add_new_free_space(cache, root->fs_info,
7782 found_key.objectid,
7783 found_key.objectid +
7784 found_key.offset);
7785 free_excluded_extents(root, cache);
7786 }
7788 ret = update_space_info(info, cache->flags, found_key.offset,
7789 btrfs_block_group_used(&cache->item),
7790 &space_info);
7791 BUG_ON(ret); /* -ENOMEM */
7792 cache->space_info = space_info;
7793 spin_lock(&cache->space_info->lock);
7794 cache->space_info->bytes_readonly += cache->bytes_super;
7795 spin_unlock(&cache->space_info->lock);
7797 __link_block_group(space_info, cache);
7799 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7800 BUG_ON(ret); /* Logic error */
7802 set_avail_alloc_bits(root->fs_info, cache->flags);
7803 if (btrfs_chunk_readonly(root, cache->key.objectid))
7804 set_block_group_ro(cache, 1);
7805 }
7807 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7808 if (!(get_alloc_profile(root, space_info->flags) &
7809 (BTRFS_BLOCK_GROUP_RAID10 |
7810 BTRFS_BLOCK_GROUP_RAID1 |
7811 BTRFS_BLOCK_GROUP_DUP)))
7812 continue;
7813 /*
7814 * avoid allocating from un-mirrored block group if there are
7815 * mirrored block groups.
7816 */
7817 list_for_each_entry(cache, &space_info->block_groups[3], list)
7818 set_block_group_ro(cache, 1);
7819 list_for_each_entry(cache, &space_info->block_groups[4], list)
7820 set_block_group_ro(cache, 1);
7821 }
7823 init_global_block_rsv(info);
7824 ret = 0;
7825 error:
7826 btrfs_free_path(path);
7827 return ret;
7828 }
7830 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
7831 struct btrfs_root *root)
7832 {
7833 struct btrfs_block_group_cache *block_group, *tmp;
7834 struct btrfs_root *extent_root = root->fs_info->extent_root;
7835 struct btrfs_block_group_item item;
7836 struct btrfs_key key;
7837 int ret = 0;
7839 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
7840 new_bg_list) {
7841 list_del_init(&block_group->new_bg_list);
7843 if (ret)
7844 continue;
7846 spin_lock(&block_group->lock);
7847 memcpy(&item, &block_group->item, sizeof(item));
7848 memcpy(&key, &block_group->key, sizeof(key));
7849 spin_unlock(&block_group->lock);
7851 ret = btrfs_insert_item(trans, extent_root, &key, &item,
7852 sizeof(item));
7853 if (ret)
7854 btrfs_abort_transaction(trans, extent_root, ret);
7855 }
7856 }
7858 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7859 struct btrfs_root *root, u64 bytes_used,
7860 u64 type, u64 chunk_objectid, u64 chunk_offset,
7861 u64 size)
7862 {
7863 int ret;
7864 struct btrfs_root *extent_root;
7865 struct btrfs_block_group_cache *cache;
7867 extent_root = root->fs_info->extent_root;
7869 root->fs_info->last_trans_log_full_commit = trans->transid;
7871 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7872 if (!cache)
7873 return -ENOMEM;
7874 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7875 GFP_NOFS);
7876 if (!cache->free_space_ctl) {
7877 kfree(cache);
7878 return -ENOMEM;
7879 }
7881 cache->key.objectid = chunk_offset;
7882 cache->key.offset = size;
7883 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7884 cache->sectorsize = root->sectorsize;
7885 cache->fs_info = root->fs_info;
7887 atomic_set(&cache->count, 1);
7888 spin_lock_init(&cache->lock);
7889 INIT_LIST_HEAD(&cache->list);
7890 INIT_LIST_HEAD(&cache->cluster_list);
7891 INIT_LIST_HEAD(&cache->new_bg_list);
7893 btrfs_init_free_space_ctl(cache);
7895 btrfs_set_block_group_used(&cache->item, bytes_used);
7896 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7897 cache->flags = type;
7898 btrfs_set_block_group_flags(&cache->item, type);
7900 cache->last_byte_to_unpin = (u64)-1;
7901 cache->cached = BTRFS_CACHE_FINISHED;
7902 exclude_super_stripes(root, cache);
7904 add_new_free_space(cache, root->fs_info, chunk_offset,
7905 chunk_offset + size);
7907 free_excluded_extents(root, cache);
7909 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7910 &cache->space_info);
7911 BUG_ON(ret); /* -ENOMEM */
7912 update_global_block_rsv(root->fs_info);
7914 spin_lock(&cache->space_info->lock);
7915 cache->space_info->bytes_readonly += cache->bytes_super;
7916 spin_unlock(&cache->space_info->lock);
7918 __link_block_group(cache->space_info, cache);
7920 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7921 BUG_ON(ret); /* Logic error */
7923 list_add_tail(&cache->new_bg_list, &trans->new_bgs);
7925 set_avail_alloc_bits(extent_root->fs_info, type);
7927 return 0;
7928 }
7930 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
7931 {
7932 u64 extra_flags = chunk_to_extended(flags) &
7933 BTRFS_EXTENDED_PROFILE_MASK;
7935 if (flags & BTRFS_BLOCK_GROUP_DATA)
7936 fs_info->avail_data_alloc_bits &= ~extra_flags;
7937 if (flags & BTRFS_BLOCK_GROUP_METADATA)
7938 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
7939 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
7940 fs_info->avail_system_alloc_bits &= ~extra_flags;
7941 }
7943 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7944 struct btrfs_root *root, u64 group_start)
7945 {
7946 struct btrfs_path *path;
7947 struct btrfs_block_group_cache *block_group;
7948 struct btrfs_free_cluster *cluster;
7949 struct btrfs_root *tree_root = root->fs_info->tree_root;
7950 struct btrfs_key key;
7951 struct inode *inode;
7952 int ret;
7953 int index;
7954 int factor;
7956 root = root->fs_info->extent_root;
7958 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7959 BUG_ON(!block_group);
7960 BUG_ON(!block_group->ro);
7962 /*
7963 * Free the reserved super bytes from this block group before
7964 * remove it.
7965 */
7966 free_excluded_extents(root, block_group);
7968 memcpy(&key, &block_group->key, sizeof(key));
7969 index = get_block_group_index(block_group);
7970 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7971 BTRFS_BLOCK_GROUP_RAID1 |
7972 BTRFS_BLOCK_GROUP_RAID10))
7973 factor = 2;
7974 else
7975 factor = 1;
7977 /* make sure this block group isn't part of an allocation cluster */
7978 cluster = &root->fs_info->data_alloc_cluster;
7979 spin_lock(&cluster->refill_lock);
7980 btrfs_return_cluster_to_free_space(block_group, cluster);
7981 spin_unlock(&cluster->refill_lock);
7983 /*
7984 * make sure this block group isn't part of a metadata
7985 * allocation cluster
7986 */
7987 cluster = &root->fs_info->meta_alloc_cluster;
7988 spin_lock(&cluster->refill_lock);
7989 btrfs_return_cluster_to_free_space(block_group, cluster);
7990 spin_unlock(&cluster->refill_lock);
7992 path = btrfs_alloc_path();
7993 if (!path) {
7994 ret = -ENOMEM;
7995 goto out;
7996 }
7998 inode = lookup_free_space_inode(tree_root, block_group, path);
7999 if (!IS_ERR(inode)) {
8000 ret = btrfs_orphan_add(trans, inode);
8001 if (ret) {
8002 btrfs_add_delayed_iput(inode);
8003 goto out;
8004 }
8005 clear_nlink(inode);
8006 /* One for the block groups ref */
8007 spin_lock(&block_group->lock);
8008 if (block_group->iref) {
8009 block_group->iref = 0;
8010 block_group->inode = NULL;
8011 spin_unlock(&block_group->lock);
8012 iput(inode);
8013 } else {
8014 spin_unlock(&block_group->lock);
8015 }
8016 /* One for our lookup ref */
8017 btrfs_add_delayed_iput(inode);
8018 }
8020 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8021 key.offset = block_group->key.objectid;
8022 key.type = 0;
8024 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8025 if (ret < 0)
8026 goto out;
8027 if (ret > 0)
8028 btrfs_release_path(path);
8029 if (ret == 0) {
8030 ret = btrfs_del_item(trans, tree_root, path);
8031 if (ret)
8032 goto out;
8033 btrfs_release_path(path);
8034 }
8036 spin_lock(&root->fs_info->block_group_cache_lock);
8037 rb_erase(&block_group->cache_node,
8038 &root->fs_info->block_group_cache_tree);
8039 spin_unlock(&root->fs_info->block_group_cache_lock);
8041 down_write(&block_group->space_info->groups_sem);
8042 /*
8043 * we must use list_del_init so people can check to see if they
8044 * are still on the list after taking the semaphore
8045 */
8046 list_del_init(&block_group->list);
8047 if (list_empty(&block_group->space_info->block_groups[index]))
8048 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8049 up_write(&block_group->space_info->groups_sem);
8051 if (block_group->cached == BTRFS_CACHE_STARTED)
8052 wait_block_group_cache_done(block_group);
8054 btrfs_remove_free_space_cache(block_group);
8056 spin_lock(&block_group->space_info->lock);
8057 block_group->space_info->total_bytes -= block_group->key.offset;
8058 block_group->space_info->bytes_readonly -= block_group->key.offset;
8059 block_group->space_info->disk_total -= block_group->key.offset * factor;
8060 spin_unlock(&block_group->space_info->lock);
8062 memcpy(&key, &block_group->key, sizeof(key));
8064 btrfs_clear_space_info_full(root->fs_info);
8066 btrfs_put_block_group(block_group);
8067 btrfs_put_block_group(block_group);
8069 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8070 if (ret > 0)
8071 ret = -EIO;
8072 if (ret < 0)
8073 goto out;
8075 ret = btrfs_del_item(trans, root, path);
8076 out:
8077 btrfs_free_path(path);
8078 return ret;
8079 }
8081 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8082 {
8083 struct btrfs_space_info *space_info;
8084 struct btrfs_super_block *disk_super;
8085 u64 features;
8086 u64 flags;
8087 int mixed = 0;
8088 int ret;
8090 disk_super = fs_info->super_copy;
8091 if (!btrfs_super_root(disk_super))
8092 return 1;
8094 features = btrfs_super_incompat_flags(disk_super);
8095 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8096 mixed = 1;
8098 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8099 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8100 if (ret)
8101 goto out;
8103 if (mixed) {
8104 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8105 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8106 } else {
8107 flags = BTRFS_BLOCK_GROUP_METADATA;
8108 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8109 if (ret)
8110 goto out;
8112 flags = BTRFS_BLOCK_GROUP_DATA;
8113 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8114 }
8115 out:
8116 return ret;
8117 }
8119 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8120 {
8121 return unpin_extent_range(root, start, end);
8122 }
8124 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8125 u64 num_bytes, u64 *actual_bytes)
8126 {
8127 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8128 }
8130 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8131 {
8132 struct btrfs_fs_info *fs_info = root->fs_info;
8133 struct btrfs_block_group_cache *cache = NULL;
8134 u64 group_trimmed;
8135 u64 start;
8136 u64 end;
8137 u64 trimmed = 0;
8138 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8139 int ret = 0;
8141 /*
8142 * try to trim all FS space, our block group may start from non-zero.
8143 */
8144 if (range->len == total_bytes)
8145 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8146 else
8147 cache = btrfs_lookup_block_group(fs_info, range->start);
8149 while (cache) {
8150 if (cache->key.objectid >= (range->start + range->len)) {
8151 btrfs_put_block_group(cache);
8152 break;
8153 }
8155 start = max(range->start, cache->key.objectid);
8156 end = min(range->start + range->len,
8157 cache->key.objectid + cache->key.offset);
8159 if (end - start >= range->minlen) {
8160 if (!block_group_cache_done(cache)) {
8161 ret = cache_block_group(cache, NULL, root, 0);
8162 if (!ret)
8163 wait_block_group_cache_done(cache);
8164 }
8165 ret = btrfs_trim_block_group(cache,
8166 &group_trimmed,
8167 start,
8168 end,
8169 range->minlen);
8171 trimmed += group_trimmed;
8172 if (ret) {
8173 btrfs_put_block_group(cache);
8174 break;
8175 }
8176 }
8178 cache = next_block_group(fs_info->tree_root, cache);
8179 }
8181 range->len = trimmed;
8182 return ret;
8183 }