Btrfs: hooks for qgroup to record delayed refs
[linux-2.6-block.git] / fs / btrfs / extent-tree.c
CommitLineData
6cbd5570
CM
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 */
ec6b910f 18#include <linux/sched.h>
edbd8d4e 19#include <linux/pagemap.h>
ec44a35c 20#include <linux/writeback.h>
21af804c 21#include <linux/blkdev.h>
b7a9f29f 22#include <linux/sort.h>
4184ea7f 23#include <linux/rcupdate.h>
817d52f8 24#include <linux/kthread.h>
5a0e3ad6 25#include <linux/slab.h>
dff51cd1 26#include <linux/ratelimit.h>
4b4e25f2 27#include "compat.h"
74493f7a 28#include "hash.h"
fec577fb
CM
29#include "ctree.h"
30#include "disk-io.h"
31#include "print-tree.h"
e089f05c 32#include "transaction.h"
0b86a832 33#include "volumes.h"
925baedd 34#include "locking.h"
fa9c0d79 35#include "free-space-cache.h"
fec577fb 36
709c0486
AJ
37#undef SCRAMBLE_DELAYED_REFS
38
9e622d6b
MX
39/*
40 * control flags for do_chunk_alloc's force field
0e4f8f88
CM
41 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
42 * if we really need one.
43 *
0e4f8f88
CM
44 * CHUNK_ALLOC_LIMITED means to only try and allocate one
45 * if we have very few chunks already allocated. This is
46 * used as part of the clustering code to help make sure
47 * we have a good pool of storage to cluster in, without
48 * filling the FS with empty chunks
49 *
9e622d6b
MX
50 * CHUNK_ALLOC_FORCE means it must try to allocate one
51 *
0e4f8f88
CM
52 */
53enum {
54 CHUNK_ALLOC_NO_FORCE = 0,
9e622d6b
MX
55 CHUNK_ALLOC_LIMITED = 1,
56 CHUNK_ALLOC_FORCE = 2,
0e4f8f88
CM
57};
58
fb25e914
JB
59/*
60 * Control how reservations are dealt with.
61 *
62 * RESERVE_FREE - freeing a reservation.
63 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
64 * ENOSPC accounting
65 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
66 * bytes_may_use as the ENOSPC accounting is done elsewhere
67 */
68enum {
69 RESERVE_FREE = 0,
70 RESERVE_ALLOC = 1,
71 RESERVE_ALLOC_NO_ACCOUNT = 2,
72};
73
f3465ca4
JB
74static int update_block_group(struct btrfs_trans_handle *trans,
75 struct btrfs_root *root,
f0486c68 76 u64 bytenr, u64 num_bytes, int alloc);
5d4f98a2
YZ
77static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
78 struct btrfs_root *root,
79 u64 bytenr, u64 num_bytes, u64 parent,
80 u64 root_objectid, u64 owner_objectid,
81 u64 owner_offset, int refs_to_drop,
82 struct btrfs_delayed_extent_op *extra_op);
83static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
84 struct extent_buffer *leaf,
85 struct btrfs_extent_item *ei);
86static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
87 struct btrfs_root *root,
88 u64 parent, u64 root_objectid,
89 u64 flags, u64 owner, u64 offset,
90 struct btrfs_key *ins, int ref_mod);
91static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
92 struct btrfs_root *root,
93 u64 parent, u64 root_objectid,
94 u64 flags, struct btrfs_disk_key *key,
95 int level, struct btrfs_key *ins);
6a63209f
JB
96static int do_chunk_alloc(struct btrfs_trans_handle *trans,
97 struct btrfs_root *extent_root, u64 alloc_bytes,
98 u64 flags, int force);
11833d66
YZ
99static int find_next_key(struct btrfs_path *path, int level,
100 struct btrfs_key *key);
9ed74f2d
JB
101static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
102 int dump_block_groups);
fb25e914
JB
103static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
104 u64 num_bytes, int reserve);
6a63209f 105
817d52f8
JB
106static noinline int
107block_group_cache_done(struct btrfs_block_group_cache *cache)
108{
109 smp_mb();
110 return cache->cached == BTRFS_CACHE_FINISHED;
111}
112
0f9dd46c
JB
113static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
114{
115 return (cache->flags & bits) == bits;
116}
117
62a45b60 118static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
11dfe35a
JB
119{
120 atomic_inc(&cache->count);
121}
122
123void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
124{
f0486c68
YZ
125 if (atomic_dec_and_test(&cache->count)) {
126 WARN_ON(cache->pinned > 0);
127 WARN_ON(cache->reserved > 0);
34d52cb6 128 kfree(cache->free_space_ctl);
11dfe35a 129 kfree(cache);
f0486c68 130 }
11dfe35a
JB
131}
132
0f9dd46c
JB
133/*
134 * this adds the block group to the fs_info rb tree for the block group
135 * cache
136 */
b2950863 137static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
0f9dd46c
JB
138 struct btrfs_block_group_cache *block_group)
139{
140 struct rb_node **p;
141 struct rb_node *parent = NULL;
142 struct btrfs_block_group_cache *cache;
143
144 spin_lock(&info->block_group_cache_lock);
145 p = &info->block_group_cache_tree.rb_node;
146
147 while (*p) {
148 parent = *p;
149 cache = rb_entry(parent, struct btrfs_block_group_cache,
150 cache_node);
151 if (block_group->key.objectid < cache->key.objectid) {
152 p = &(*p)->rb_left;
153 } else if (block_group->key.objectid > cache->key.objectid) {
154 p = &(*p)->rb_right;
155 } else {
156 spin_unlock(&info->block_group_cache_lock);
157 return -EEXIST;
158 }
159 }
160
161 rb_link_node(&block_group->cache_node, parent, p);
162 rb_insert_color(&block_group->cache_node,
163 &info->block_group_cache_tree);
164 spin_unlock(&info->block_group_cache_lock);
165
166 return 0;
167}
168
169/*
170 * This will return the block group at or after bytenr if contains is 0, else
171 * it will return the block group that contains the bytenr
172 */
173static struct btrfs_block_group_cache *
174block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
175 int contains)
176{
177 struct btrfs_block_group_cache *cache, *ret = NULL;
178 struct rb_node *n;
179 u64 end, start;
180
181 spin_lock(&info->block_group_cache_lock);
182 n = info->block_group_cache_tree.rb_node;
183
184 while (n) {
185 cache = rb_entry(n, struct btrfs_block_group_cache,
186 cache_node);
187 end = cache->key.objectid + cache->key.offset - 1;
188 start = cache->key.objectid;
189
190 if (bytenr < start) {
191 if (!contains && (!ret || start < ret->key.objectid))
192 ret = cache;
193 n = n->rb_left;
194 } else if (bytenr > start) {
195 if (contains && bytenr <= end) {
196 ret = cache;
197 break;
198 }
199 n = n->rb_right;
200 } else {
201 ret = cache;
202 break;
203 }
204 }
d2fb3437 205 if (ret)
11dfe35a 206 btrfs_get_block_group(ret);
0f9dd46c
JB
207 spin_unlock(&info->block_group_cache_lock);
208
209 return ret;
210}
211
11833d66
YZ
212static int add_excluded_extent(struct btrfs_root *root,
213 u64 start, u64 num_bytes)
817d52f8 214{
11833d66
YZ
215 u64 end = start + num_bytes - 1;
216 set_extent_bits(&root->fs_info->freed_extents[0],
217 start, end, EXTENT_UPTODATE, GFP_NOFS);
218 set_extent_bits(&root->fs_info->freed_extents[1],
219 start, end, EXTENT_UPTODATE, GFP_NOFS);
220 return 0;
221}
817d52f8 222
11833d66
YZ
223static void free_excluded_extents(struct btrfs_root *root,
224 struct btrfs_block_group_cache *cache)
225{
226 u64 start, end;
817d52f8 227
11833d66
YZ
228 start = cache->key.objectid;
229 end = start + cache->key.offset - 1;
230
231 clear_extent_bits(&root->fs_info->freed_extents[0],
232 start, end, EXTENT_UPTODATE, GFP_NOFS);
233 clear_extent_bits(&root->fs_info->freed_extents[1],
234 start, end, EXTENT_UPTODATE, GFP_NOFS);
817d52f8
JB
235}
236
11833d66
YZ
237static int exclude_super_stripes(struct btrfs_root *root,
238 struct btrfs_block_group_cache *cache)
817d52f8 239{
817d52f8
JB
240 u64 bytenr;
241 u64 *logical;
242 int stripe_len;
243 int i, nr, ret;
244
06b2331f
YZ
245 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
246 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
247 cache->bytes_super += stripe_len;
248 ret = add_excluded_extent(root, cache->key.objectid,
249 stripe_len);
79787eaa 250 BUG_ON(ret); /* -ENOMEM */
06b2331f
YZ
251 }
252
817d52f8
JB
253 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
254 bytenr = btrfs_sb_offset(i);
255 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
256 cache->key.objectid, bytenr,
257 0, &logical, &nr, &stripe_len);
79787eaa 258 BUG_ON(ret); /* -ENOMEM */
11833d66 259
817d52f8 260 while (nr--) {
1b2da372 261 cache->bytes_super += stripe_len;
11833d66
YZ
262 ret = add_excluded_extent(root, logical[nr],
263 stripe_len);
79787eaa 264 BUG_ON(ret); /* -ENOMEM */
817d52f8 265 }
11833d66 266
817d52f8
JB
267 kfree(logical);
268 }
817d52f8
JB
269 return 0;
270}
271
11833d66
YZ
272static struct btrfs_caching_control *
273get_caching_control(struct btrfs_block_group_cache *cache)
274{
275 struct btrfs_caching_control *ctl;
276
277 spin_lock(&cache->lock);
278 if (cache->cached != BTRFS_CACHE_STARTED) {
279 spin_unlock(&cache->lock);
280 return NULL;
281 }
282
dde5abee
JB
283 /* We're loading it the fast way, so we don't have a caching_ctl. */
284 if (!cache->caching_ctl) {
285 spin_unlock(&cache->lock);
11833d66
YZ
286 return NULL;
287 }
288
289 ctl = cache->caching_ctl;
290 atomic_inc(&ctl->count);
291 spin_unlock(&cache->lock);
292 return ctl;
293}
294
295static void put_caching_control(struct btrfs_caching_control *ctl)
296{
297 if (atomic_dec_and_test(&ctl->count))
298 kfree(ctl);
299}
300
0f9dd46c
JB
301/*
302 * this is only called by cache_block_group, since we could have freed extents
303 * we need to check the pinned_extents for any extents that can't be used yet
304 * since their free space will be released as soon as the transaction commits.
305 */
817d52f8 306static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
0f9dd46c
JB
307 struct btrfs_fs_info *info, u64 start, u64 end)
308{
817d52f8 309 u64 extent_start, extent_end, size, total_added = 0;
0f9dd46c
JB
310 int ret;
311
312 while (start < end) {
11833d66 313 ret = find_first_extent_bit(info->pinned_extents, start,
0f9dd46c 314 &extent_start, &extent_end,
11833d66 315 EXTENT_DIRTY | EXTENT_UPTODATE);
0f9dd46c
JB
316 if (ret)
317 break;
318
06b2331f 319 if (extent_start <= start) {
0f9dd46c
JB
320 start = extent_end + 1;
321 } else if (extent_start > start && extent_start < end) {
322 size = extent_start - start;
817d52f8 323 total_added += size;
ea6a478e
JB
324 ret = btrfs_add_free_space(block_group, start,
325 size);
79787eaa 326 BUG_ON(ret); /* -ENOMEM or logic error */
0f9dd46c
JB
327 start = extent_end + 1;
328 } else {
329 break;
330 }
331 }
332
333 if (start < end) {
334 size = end - start;
817d52f8 335 total_added += size;
ea6a478e 336 ret = btrfs_add_free_space(block_group, start, size);
79787eaa 337 BUG_ON(ret); /* -ENOMEM or logic error */
0f9dd46c
JB
338 }
339
817d52f8 340 return total_added;
0f9dd46c
JB
341}
342
bab39bf9 343static noinline void caching_thread(struct btrfs_work *work)
e37c9e69 344{
bab39bf9
JB
345 struct btrfs_block_group_cache *block_group;
346 struct btrfs_fs_info *fs_info;
347 struct btrfs_caching_control *caching_ctl;
348 struct btrfs_root *extent_root;
e37c9e69 349 struct btrfs_path *path;
5f39d397 350 struct extent_buffer *leaf;
11833d66 351 struct btrfs_key key;
817d52f8 352 u64 total_found = 0;
11833d66
YZ
353 u64 last = 0;
354 u32 nritems;
355 int ret = 0;
f510cfec 356
bab39bf9
JB
357 caching_ctl = container_of(work, struct btrfs_caching_control, work);
358 block_group = caching_ctl->block_group;
359 fs_info = block_group->fs_info;
360 extent_root = fs_info->extent_root;
361
e37c9e69
CM
362 path = btrfs_alloc_path();
363 if (!path)
bab39bf9 364 goto out;
7d7d6068 365
817d52f8 366 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
11833d66 367
5cd57b2c 368 /*
817d52f8
JB
369 * We don't want to deadlock with somebody trying to allocate a new
370 * extent for the extent root while also trying to search the extent
371 * root to add free space. So we skip locking and search the commit
372 * root, since its read-only
5cd57b2c
CM
373 */
374 path->skip_locking = 1;
817d52f8 375 path->search_commit_root = 1;
026fd317 376 path->reada = 1;
817d52f8 377
e4404d6e 378 key.objectid = last;
e37c9e69 379 key.offset = 0;
11833d66 380 key.type = BTRFS_EXTENT_ITEM_KEY;
013f1b12 381again:
11833d66 382 mutex_lock(&caching_ctl->mutex);
013f1b12
CM
383 /* need to make sure the commit_root doesn't disappear */
384 down_read(&fs_info->extent_commit_sem);
385
11833d66 386 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
e37c9e69 387 if (ret < 0)
ef8bbdfe 388 goto err;
a512bbf8 389
11833d66
YZ
390 leaf = path->nodes[0];
391 nritems = btrfs_header_nritems(leaf);
392
d397712b 393 while (1) {
7841cb28 394 if (btrfs_fs_closing(fs_info) > 1) {
f25784b3 395 last = (u64)-1;
817d52f8 396 break;
f25784b3 397 }
817d52f8 398
11833d66
YZ
399 if (path->slots[0] < nritems) {
400 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
401 } else {
402 ret = find_next_key(path, 0, &key);
403 if (ret)
e37c9e69 404 break;
817d52f8 405
589d8ade
JB
406 if (need_resched() ||
407 btrfs_next_leaf(extent_root, path)) {
408 caching_ctl->progress = last;
ff5714cc 409 btrfs_release_path(path);
589d8ade
JB
410 up_read(&fs_info->extent_commit_sem);
411 mutex_unlock(&caching_ctl->mutex);
11833d66 412 cond_resched();
589d8ade
JB
413 goto again;
414 }
415 leaf = path->nodes[0];
416 nritems = btrfs_header_nritems(leaf);
417 continue;
11833d66 418 }
817d52f8 419
11833d66
YZ
420 if (key.objectid < block_group->key.objectid) {
421 path->slots[0]++;
817d52f8 422 continue;
e37c9e69 423 }
0f9dd46c 424
e37c9e69 425 if (key.objectid >= block_group->key.objectid +
0f9dd46c 426 block_group->key.offset)
e37c9e69 427 break;
7d7d6068 428
11833d66 429 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
817d52f8
JB
430 total_found += add_new_free_space(block_group,
431 fs_info, last,
432 key.objectid);
7d7d6068 433 last = key.objectid + key.offset;
817d52f8 434
11833d66
YZ
435 if (total_found > (1024 * 1024 * 2)) {
436 total_found = 0;
437 wake_up(&caching_ctl->wait);
438 }
817d52f8 439 }
e37c9e69
CM
440 path->slots[0]++;
441 }
817d52f8 442 ret = 0;
e37c9e69 443
817d52f8
JB
444 total_found += add_new_free_space(block_group, fs_info, last,
445 block_group->key.objectid +
446 block_group->key.offset);
11833d66 447 caching_ctl->progress = (u64)-1;
817d52f8
JB
448
449 spin_lock(&block_group->lock);
11833d66 450 block_group->caching_ctl = NULL;
817d52f8
JB
451 block_group->cached = BTRFS_CACHE_FINISHED;
452 spin_unlock(&block_group->lock);
0f9dd46c 453
54aa1f4d 454err:
e37c9e69 455 btrfs_free_path(path);
276e680d 456 up_read(&fs_info->extent_commit_sem);
817d52f8 457
11833d66
YZ
458 free_excluded_extents(extent_root, block_group);
459
460 mutex_unlock(&caching_ctl->mutex);
bab39bf9 461out:
11833d66
YZ
462 wake_up(&caching_ctl->wait);
463
464 put_caching_control(caching_ctl);
11dfe35a 465 btrfs_put_block_group(block_group);
817d52f8
JB
466}
467
9d66e233
JB
468static int cache_block_group(struct btrfs_block_group_cache *cache,
469 struct btrfs_trans_handle *trans,
b8399dee 470 struct btrfs_root *root,
9d66e233 471 int load_cache_only)
817d52f8 472{
291c7d2f 473 DEFINE_WAIT(wait);
11833d66
YZ
474 struct btrfs_fs_info *fs_info = cache->fs_info;
475 struct btrfs_caching_control *caching_ctl;
817d52f8
JB
476 int ret = 0;
477
291c7d2f 478 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
79787eaa
JM
479 if (!caching_ctl)
480 return -ENOMEM;
291c7d2f
JB
481
482 INIT_LIST_HEAD(&caching_ctl->list);
483 mutex_init(&caching_ctl->mutex);
484 init_waitqueue_head(&caching_ctl->wait);
485 caching_ctl->block_group = cache;
486 caching_ctl->progress = cache->key.objectid;
487 atomic_set(&caching_ctl->count, 1);
488 caching_ctl->work.func = caching_thread;
489
490 spin_lock(&cache->lock);
491 /*
492 * This should be a rare occasion, but this could happen I think in the
493 * case where one thread starts to load the space cache info, and then
494 * some other thread starts a transaction commit which tries to do an
495 * allocation while the other thread is still loading the space cache
496 * info. The previous loop should have kept us from choosing this block
497 * group, but if we've moved to the state where we will wait on caching
498 * block groups we need to first check if we're doing a fast load here,
499 * so we can wait for it to finish, otherwise we could end up allocating
500 * from a block group who's cache gets evicted for one reason or
501 * another.
502 */
503 while (cache->cached == BTRFS_CACHE_FAST) {
504 struct btrfs_caching_control *ctl;
505
506 ctl = cache->caching_ctl;
507 atomic_inc(&ctl->count);
508 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
509 spin_unlock(&cache->lock);
510
511 schedule();
512
513 finish_wait(&ctl->wait, &wait);
514 put_caching_control(ctl);
515 spin_lock(&cache->lock);
516 }
517
518 if (cache->cached != BTRFS_CACHE_NO) {
519 spin_unlock(&cache->lock);
520 kfree(caching_ctl);
11833d66 521 return 0;
291c7d2f
JB
522 }
523 WARN_ON(cache->caching_ctl);
524 cache->caching_ctl = caching_ctl;
525 cache->cached = BTRFS_CACHE_FAST;
526 spin_unlock(&cache->lock);
11833d66 527
9d66e233
JB
528 /*
529 * We can't do the read from on-disk cache during a commit since we need
b8399dee
JB
530 * to have the normal tree locking. Also if we are currently trying to
531 * allocate blocks for the tree root we can't do the fast caching since
532 * we likely hold important locks.
9d66e233 533 */
d53ba474 534 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
9d66e233
JB
535 ret = load_free_space_cache(fs_info, cache);
536
537 spin_lock(&cache->lock);
538 if (ret == 1) {
291c7d2f 539 cache->caching_ctl = NULL;
9d66e233
JB
540 cache->cached = BTRFS_CACHE_FINISHED;
541 cache->last_byte_to_unpin = (u64)-1;
542 } else {
291c7d2f
JB
543 if (load_cache_only) {
544 cache->caching_ctl = NULL;
545 cache->cached = BTRFS_CACHE_NO;
546 } else {
547 cache->cached = BTRFS_CACHE_STARTED;
548 }
9d66e233
JB
549 }
550 spin_unlock(&cache->lock);
291c7d2f 551 wake_up(&caching_ctl->wait);
3c14874a 552 if (ret == 1) {
291c7d2f 553 put_caching_control(caching_ctl);
3c14874a 554 free_excluded_extents(fs_info->extent_root, cache);
9d66e233 555 return 0;
3c14874a 556 }
291c7d2f
JB
557 } else {
558 /*
559 * We are not going to do the fast caching, set cached to the
560 * appropriate value and wakeup any waiters.
561 */
562 spin_lock(&cache->lock);
563 if (load_cache_only) {
564 cache->caching_ctl = NULL;
565 cache->cached = BTRFS_CACHE_NO;
566 } else {
567 cache->cached = BTRFS_CACHE_STARTED;
568 }
569 spin_unlock(&cache->lock);
570 wake_up(&caching_ctl->wait);
9d66e233
JB
571 }
572
291c7d2f
JB
573 if (load_cache_only) {
574 put_caching_control(caching_ctl);
11833d66 575 return 0;
817d52f8 576 }
817d52f8 577
11833d66 578 down_write(&fs_info->extent_commit_sem);
291c7d2f 579 atomic_inc(&caching_ctl->count);
11833d66
YZ
580 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
581 up_write(&fs_info->extent_commit_sem);
582
11dfe35a 583 btrfs_get_block_group(cache);
11833d66 584
bab39bf9 585 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
817d52f8 586
ef8bbdfe 587 return ret;
e37c9e69
CM
588}
589
0f9dd46c
JB
590/*
591 * return the block group that starts at or after bytenr
592 */
d397712b
CM
593static struct btrfs_block_group_cache *
594btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
0ef3e66b 595{
0f9dd46c 596 struct btrfs_block_group_cache *cache;
0ef3e66b 597
0f9dd46c 598 cache = block_group_cache_tree_search(info, bytenr, 0);
0ef3e66b 599
0f9dd46c 600 return cache;
0ef3e66b
CM
601}
602
0f9dd46c 603/*
9f55684c 604 * return the block group that contains the given bytenr
0f9dd46c 605 */
d397712b
CM
606struct btrfs_block_group_cache *btrfs_lookup_block_group(
607 struct btrfs_fs_info *info,
608 u64 bytenr)
be744175 609{
0f9dd46c 610 struct btrfs_block_group_cache *cache;
be744175 611
0f9dd46c 612 cache = block_group_cache_tree_search(info, bytenr, 1);
96b5179d 613
0f9dd46c 614 return cache;
be744175 615}
0b86a832 616
0f9dd46c
JB
617static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
618 u64 flags)
6324fbf3 619{
0f9dd46c 620 struct list_head *head = &info->space_info;
0f9dd46c 621 struct btrfs_space_info *found;
4184ea7f 622
52ba6929 623 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
b742bb82 624
4184ea7f
CM
625 rcu_read_lock();
626 list_for_each_entry_rcu(found, head, list) {
67377734 627 if (found->flags & flags) {
4184ea7f 628 rcu_read_unlock();
0f9dd46c 629 return found;
4184ea7f 630 }
0f9dd46c 631 }
4184ea7f 632 rcu_read_unlock();
0f9dd46c 633 return NULL;
6324fbf3
CM
634}
635
4184ea7f
CM
636/*
637 * after adding space to the filesystem, we need to clear the full flags
638 * on all the space infos.
639 */
640void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
641{
642 struct list_head *head = &info->space_info;
643 struct btrfs_space_info *found;
644
645 rcu_read_lock();
646 list_for_each_entry_rcu(found, head, list)
647 found->full = 0;
648 rcu_read_unlock();
649}
650
80eb234a
JB
651static u64 div_factor(u64 num, int factor)
652{
653 if (factor == 10)
654 return num;
655 num *= factor;
656 do_div(num, 10);
657 return num;
658}
659
e5bc2458
CM
660static u64 div_factor_fine(u64 num, int factor)
661{
662 if (factor == 100)
663 return num;
664 num *= factor;
665 do_div(num, 100);
666 return num;
667}
668
d2fb3437
YZ
669u64 btrfs_find_block_group(struct btrfs_root *root,
670 u64 search_start, u64 search_hint, int owner)
cd1bc465 671{
96b5179d 672 struct btrfs_block_group_cache *cache;
cd1bc465 673 u64 used;
d2fb3437
YZ
674 u64 last = max(search_hint, search_start);
675 u64 group_start = 0;
31f3c99b 676 int full_search = 0;
d2fb3437 677 int factor = 9;
0ef3e66b 678 int wrapped = 0;
31f3c99b 679again:
e8569813
ZY
680 while (1) {
681 cache = btrfs_lookup_first_block_group(root->fs_info, last);
0f9dd46c
JB
682 if (!cache)
683 break;
96b5179d 684
c286ac48 685 spin_lock(&cache->lock);
96b5179d
CM
686 last = cache->key.objectid + cache->key.offset;
687 used = btrfs_block_group_used(&cache->item);
688
d2fb3437
YZ
689 if ((full_search || !cache->ro) &&
690 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
e8569813 691 if (used + cache->pinned + cache->reserved <
d2fb3437
YZ
692 div_factor(cache->key.offset, factor)) {
693 group_start = cache->key.objectid;
c286ac48 694 spin_unlock(&cache->lock);
fa9c0d79 695 btrfs_put_block_group(cache);
8790d502
CM
696 goto found;
697 }
6324fbf3 698 }
c286ac48 699 spin_unlock(&cache->lock);
fa9c0d79 700 btrfs_put_block_group(cache);
de428b63 701 cond_resched();
cd1bc465 702 }
0ef3e66b
CM
703 if (!wrapped) {
704 last = search_start;
705 wrapped = 1;
706 goto again;
707 }
708 if (!full_search && factor < 10) {
be744175 709 last = search_start;
31f3c99b 710 full_search = 1;
0ef3e66b 711 factor = 10;
31f3c99b
CM
712 goto again;
713 }
be744175 714found:
d2fb3437 715 return group_start;
925baedd 716}
0f9dd46c 717
e02119d5 718/* simple helper to search for an existing extent at a given offset */
31840ae1 719int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
e02119d5
CM
720{
721 int ret;
722 struct btrfs_key key;
31840ae1 723 struct btrfs_path *path;
e02119d5 724
31840ae1 725 path = btrfs_alloc_path();
d8926bb3
MF
726 if (!path)
727 return -ENOMEM;
728
e02119d5
CM
729 key.objectid = start;
730 key.offset = len;
731 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
732 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
733 0, 0);
31840ae1 734 btrfs_free_path(path);
7bb86316
CM
735 return ret;
736}
737
a22285a6
YZ
738/*
739 * helper function to lookup reference count and flags of extent.
740 *
741 * the head node for delayed ref is used to store the sum of all the
742 * reference count modifications queued up in the rbtree. the head
743 * node may also store the extent flags to set. This way you can check
744 * to see what the reference count and extent flags would be if all of
745 * the delayed refs are not processed.
746 */
747int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
748 struct btrfs_root *root, u64 bytenr,
749 u64 num_bytes, u64 *refs, u64 *flags)
750{
751 struct btrfs_delayed_ref_head *head;
752 struct btrfs_delayed_ref_root *delayed_refs;
753 struct btrfs_path *path;
754 struct btrfs_extent_item *ei;
755 struct extent_buffer *leaf;
756 struct btrfs_key key;
757 u32 item_size;
758 u64 num_refs;
759 u64 extent_flags;
760 int ret;
761
762 path = btrfs_alloc_path();
763 if (!path)
764 return -ENOMEM;
765
766 key.objectid = bytenr;
767 key.type = BTRFS_EXTENT_ITEM_KEY;
768 key.offset = num_bytes;
769 if (!trans) {
770 path->skip_locking = 1;
771 path->search_commit_root = 1;
772 }
773again:
774 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
775 &key, path, 0, 0);
776 if (ret < 0)
777 goto out_free;
778
779 if (ret == 0) {
780 leaf = path->nodes[0];
781 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
782 if (item_size >= sizeof(*ei)) {
783 ei = btrfs_item_ptr(leaf, path->slots[0],
784 struct btrfs_extent_item);
785 num_refs = btrfs_extent_refs(leaf, ei);
786 extent_flags = btrfs_extent_flags(leaf, ei);
787 } else {
788#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
789 struct btrfs_extent_item_v0 *ei0;
790 BUG_ON(item_size != sizeof(*ei0));
791 ei0 = btrfs_item_ptr(leaf, path->slots[0],
792 struct btrfs_extent_item_v0);
793 num_refs = btrfs_extent_refs_v0(leaf, ei0);
794 /* FIXME: this isn't correct for data */
795 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
796#else
797 BUG();
798#endif
799 }
800 BUG_ON(num_refs == 0);
801 } else {
802 num_refs = 0;
803 extent_flags = 0;
804 ret = 0;
805 }
806
807 if (!trans)
808 goto out;
809
810 delayed_refs = &trans->transaction->delayed_refs;
811 spin_lock(&delayed_refs->lock);
812 head = btrfs_find_delayed_ref_head(trans, bytenr);
813 if (head) {
814 if (!mutex_trylock(&head->mutex)) {
815 atomic_inc(&head->node.refs);
816 spin_unlock(&delayed_refs->lock);
817
b3b4aa74 818 btrfs_release_path(path);
a22285a6 819
8cc33e5c
DS
820 /*
821 * Mutex was contended, block until it's released and try
822 * again
823 */
a22285a6
YZ
824 mutex_lock(&head->mutex);
825 mutex_unlock(&head->mutex);
826 btrfs_put_delayed_ref(&head->node);
827 goto again;
828 }
829 if (head->extent_op && head->extent_op->update_flags)
830 extent_flags |= head->extent_op->flags_to_set;
831 else
832 BUG_ON(num_refs == 0);
833
834 num_refs += head->node.ref_mod;
835 mutex_unlock(&head->mutex);
836 }
837 spin_unlock(&delayed_refs->lock);
838out:
839 WARN_ON(num_refs == 0);
840 if (refs)
841 *refs = num_refs;
842 if (flags)
843 *flags = extent_flags;
844out_free:
845 btrfs_free_path(path);
846 return ret;
847}
848
d8d5f3e1
CM
849/*
850 * Back reference rules. Back refs have three main goals:
851 *
852 * 1) differentiate between all holders of references to an extent so that
853 * when a reference is dropped we can make sure it was a valid reference
854 * before freeing the extent.
855 *
856 * 2) Provide enough information to quickly find the holders of an extent
857 * if we notice a given block is corrupted or bad.
858 *
859 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
860 * maintenance. This is actually the same as #2, but with a slightly
861 * different use case.
862 *
5d4f98a2
YZ
863 * There are two kinds of back refs. The implicit back refs is optimized
864 * for pointers in non-shared tree blocks. For a given pointer in a block,
865 * back refs of this kind provide information about the block's owner tree
866 * and the pointer's key. These information allow us to find the block by
867 * b-tree searching. The full back refs is for pointers in tree blocks not
868 * referenced by their owner trees. The location of tree block is recorded
869 * in the back refs. Actually the full back refs is generic, and can be
870 * used in all cases the implicit back refs is used. The major shortcoming
871 * of the full back refs is its overhead. Every time a tree block gets
872 * COWed, we have to update back refs entry for all pointers in it.
873 *
874 * For a newly allocated tree block, we use implicit back refs for
875 * pointers in it. This means most tree related operations only involve
876 * implicit back refs. For a tree block created in old transaction, the
877 * only way to drop a reference to it is COW it. So we can detect the
878 * event that tree block loses its owner tree's reference and do the
879 * back refs conversion.
880 *
881 * When a tree block is COW'd through a tree, there are four cases:
882 *
883 * The reference count of the block is one and the tree is the block's
884 * owner tree. Nothing to do in this case.
885 *
886 * The reference count of the block is one and the tree is not the
887 * block's owner tree. In this case, full back refs is used for pointers
888 * in the block. Remove these full back refs, add implicit back refs for
889 * every pointers in the new block.
890 *
891 * The reference count of the block is greater than one and the tree is
892 * the block's owner tree. In this case, implicit back refs is used for
893 * pointers in the block. Add full back refs for every pointers in the
894 * block, increase lower level extents' reference counts. The original
895 * implicit back refs are entailed to the new block.
896 *
897 * The reference count of the block is greater than one and the tree is
898 * not the block's owner tree. Add implicit back refs for every pointer in
899 * the new block, increase lower level extents' reference count.
900 *
901 * Back Reference Key composing:
902 *
903 * The key objectid corresponds to the first byte in the extent,
904 * The key type is used to differentiate between types of back refs.
905 * There are different meanings of the key offset for different types
906 * of back refs.
907 *
d8d5f3e1
CM
908 * File extents can be referenced by:
909 *
910 * - multiple snapshots, subvolumes, or different generations in one subvol
31840ae1 911 * - different files inside a single subvolume
d8d5f3e1
CM
912 * - different offsets inside a file (bookend extents in file.c)
913 *
5d4f98a2 914 * The extent ref structure for the implicit back refs has fields for:
d8d5f3e1
CM
915 *
916 * - Objectid of the subvolume root
d8d5f3e1 917 * - objectid of the file holding the reference
5d4f98a2
YZ
918 * - original offset in the file
919 * - how many bookend extents
d8d5f3e1 920 *
5d4f98a2
YZ
921 * The key offset for the implicit back refs is hash of the first
922 * three fields.
d8d5f3e1 923 *
5d4f98a2 924 * The extent ref structure for the full back refs has field for:
d8d5f3e1 925 *
5d4f98a2 926 * - number of pointers in the tree leaf
d8d5f3e1 927 *
5d4f98a2
YZ
928 * The key offset for the implicit back refs is the first byte of
929 * the tree leaf
d8d5f3e1 930 *
5d4f98a2
YZ
931 * When a file extent is allocated, The implicit back refs is used.
932 * the fields are filled in:
d8d5f3e1 933 *
5d4f98a2 934 * (root_key.objectid, inode objectid, offset in file, 1)
d8d5f3e1 935 *
5d4f98a2
YZ
936 * When a file extent is removed file truncation, we find the
937 * corresponding implicit back refs and check the following fields:
d8d5f3e1 938 *
5d4f98a2 939 * (btrfs_header_owner(leaf), inode objectid, offset in file)
d8d5f3e1 940 *
5d4f98a2 941 * Btree extents can be referenced by:
d8d5f3e1 942 *
5d4f98a2 943 * - Different subvolumes
d8d5f3e1 944 *
5d4f98a2
YZ
945 * Both the implicit back refs and the full back refs for tree blocks
946 * only consist of key. The key offset for the implicit back refs is
947 * objectid of block's owner tree. The key offset for the full back refs
948 * is the first byte of parent block.
d8d5f3e1 949 *
5d4f98a2
YZ
950 * When implicit back refs is used, information about the lowest key and
951 * level of the tree block are required. These information are stored in
952 * tree block info structure.
d8d5f3e1 953 */
31840ae1 954
5d4f98a2
YZ
955#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
956static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
957 struct btrfs_root *root,
958 struct btrfs_path *path,
959 u64 owner, u32 extra_size)
7bb86316 960{
5d4f98a2
YZ
961 struct btrfs_extent_item *item;
962 struct btrfs_extent_item_v0 *ei0;
963 struct btrfs_extent_ref_v0 *ref0;
964 struct btrfs_tree_block_info *bi;
965 struct extent_buffer *leaf;
7bb86316 966 struct btrfs_key key;
5d4f98a2
YZ
967 struct btrfs_key found_key;
968 u32 new_size = sizeof(*item);
969 u64 refs;
970 int ret;
971
972 leaf = path->nodes[0];
973 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
974
975 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
976 ei0 = btrfs_item_ptr(leaf, path->slots[0],
977 struct btrfs_extent_item_v0);
978 refs = btrfs_extent_refs_v0(leaf, ei0);
979
980 if (owner == (u64)-1) {
981 while (1) {
982 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
983 ret = btrfs_next_leaf(root, path);
984 if (ret < 0)
985 return ret;
79787eaa 986 BUG_ON(ret > 0); /* Corruption */
5d4f98a2
YZ
987 leaf = path->nodes[0];
988 }
989 btrfs_item_key_to_cpu(leaf, &found_key,
990 path->slots[0]);
991 BUG_ON(key.objectid != found_key.objectid);
992 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
993 path->slots[0]++;
994 continue;
995 }
996 ref0 = btrfs_item_ptr(leaf, path->slots[0],
997 struct btrfs_extent_ref_v0);
998 owner = btrfs_ref_objectid_v0(leaf, ref0);
999 break;
1000 }
1001 }
b3b4aa74 1002 btrfs_release_path(path);
5d4f98a2
YZ
1003
1004 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1005 new_size += sizeof(*bi);
1006
1007 new_size -= sizeof(*ei0);
1008 ret = btrfs_search_slot(trans, root, &key, path,
1009 new_size + extra_size, 1);
1010 if (ret < 0)
1011 return ret;
79787eaa 1012 BUG_ON(ret); /* Corruption */
5d4f98a2 1013
143bede5 1014 btrfs_extend_item(trans, root, path, new_size);
5d4f98a2
YZ
1015
1016 leaf = path->nodes[0];
1017 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1018 btrfs_set_extent_refs(leaf, item, refs);
1019 /* FIXME: get real generation */
1020 btrfs_set_extent_generation(leaf, item, 0);
1021 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1022 btrfs_set_extent_flags(leaf, item,
1023 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1024 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1025 bi = (struct btrfs_tree_block_info *)(item + 1);
1026 /* FIXME: get first key of the block */
1027 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1028 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1029 } else {
1030 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1031 }
1032 btrfs_mark_buffer_dirty(leaf);
1033 return 0;
1034}
1035#endif
1036
1037static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1038{
1039 u32 high_crc = ~(u32)0;
1040 u32 low_crc = ~(u32)0;
1041 __le64 lenum;
1042
1043 lenum = cpu_to_le64(root_objectid);
163e783e 1044 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
5d4f98a2 1045 lenum = cpu_to_le64(owner);
163e783e 1046 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
5d4f98a2 1047 lenum = cpu_to_le64(offset);
163e783e 1048 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
5d4f98a2
YZ
1049
1050 return ((u64)high_crc << 31) ^ (u64)low_crc;
1051}
1052
1053static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1054 struct btrfs_extent_data_ref *ref)
1055{
1056 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1057 btrfs_extent_data_ref_objectid(leaf, ref),
1058 btrfs_extent_data_ref_offset(leaf, ref));
1059}
1060
1061static int match_extent_data_ref(struct extent_buffer *leaf,
1062 struct btrfs_extent_data_ref *ref,
1063 u64 root_objectid, u64 owner, u64 offset)
1064{
1065 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1066 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1067 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1068 return 0;
1069 return 1;
1070}
1071
1072static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1073 struct btrfs_root *root,
1074 struct btrfs_path *path,
1075 u64 bytenr, u64 parent,
1076 u64 root_objectid,
1077 u64 owner, u64 offset)
1078{
1079 struct btrfs_key key;
1080 struct btrfs_extent_data_ref *ref;
31840ae1 1081 struct extent_buffer *leaf;
5d4f98a2 1082 u32 nritems;
74493f7a 1083 int ret;
5d4f98a2
YZ
1084 int recow;
1085 int err = -ENOENT;
74493f7a 1086
31840ae1 1087 key.objectid = bytenr;
5d4f98a2
YZ
1088 if (parent) {
1089 key.type = BTRFS_SHARED_DATA_REF_KEY;
1090 key.offset = parent;
1091 } else {
1092 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1093 key.offset = hash_extent_data_ref(root_objectid,
1094 owner, offset);
1095 }
1096again:
1097 recow = 0;
1098 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1099 if (ret < 0) {
1100 err = ret;
1101 goto fail;
1102 }
31840ae1 1103
5d4f98a2
YZ
1104 if (parent) {
1105 if (!ret)
1106 return 0;
1107#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1108 key.type = BTRFS_EXTENT_REF_V0_KEY;
b3b4aa74 1109 btrfs_release_path(path);
5d4f98a2
YZ
1110 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1111 if (ret < 0) {
1112 err = ret;
1113 goto fail;
1114 }
1115 if (!ret)
1116 return 0;
1117#endif
1118 goto fail;
31840ae1
ZY
1119 }
1120
1121 leaf = path->nodes[0];
5d4f98a2
YZ
1122 nritems = btrfs_header_nritems(leaf);
1123 while (1) {
1124 if (path->slots[0] >= nritems) {
1125 ret = btrfs_next_leaf(root, path);
1126 if (ret < 0)
1127 err = ret;
1128 if (ret)
1129 goto fail;
1130
1131 leaf = path->nodes[0];
1132 nritems = btrfs_header_nritems(leaf);
1133 recow = 1;
1134 }
1135
1136 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1137 if (key.objectid != bytenr ||
1138 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1139 goto fail;
1140
1141 ref = btrfs_item_ptr(leaf, path->slots[0],
1142 struct btrfs_extent_data_ref);
1143
1144 if (match_extent_data_ref(leaf, ref, root_objectid,
1145 owner, offset)) {
1146 if (recow) {
b3b4aa74 1147 btrfs_release_path(path);
5d4f98a2
YZ
1148 goto again;
1149 }
1150 err = 0;
1151 break;
1152 }
1153 path->slots[0]++;
31840ae1 1154 }
5d4f98a2
YZ
1155fail:
1156 return err;
31840ae1
ZY
1157}
1158
5d4f98a2
YZ
1159static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1160 struct btrfs_root *root,
1161 struct btrfs_path *path,
1162 u64 bytenr, u64 parent,
1163 u64 root_objectid, u64 owner,
1164 u64 offset, int refs_to_add)
31840ae1
ZY
1165{
1166 struct btrfs_key key;
1167 struct extent_buffer *leaf;
5d4f98a2 1168 u32 size;
31840ae1
ZY
1169 u32 num_refs;
1170 int ret;
74493f7a 1171
74493f7a 1172 key.objectid = bytenr;
5d4f98a2
YZ
1173 if (parent) {
1174 key.type = BTRFS_SHARED_DATA_REF_KEY;
1175 key.offset = parent;
1176 size = sizeof(struct btrfs_shared_data_ref);
1177 } else {
1178 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1179 key.offset = hash_extent_data_ref(root_objectid,
1180 owner, offset);
1181 size = sizeof(struct btrfs_extent_data_ref);
1182 }
74493f7a 1183
5d4f98a2
YZ
1184 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1185 if (ret && ret != -EEXIST)
1186 goto fail;
1187
1188 leaf = path->nodes[0];
1189 if (parent) {
1190 struct btrfs_shared_data_ref *ref;
31840ae1 1191 ref = btrfs_item_ptr(leaf, path->slots[0],
5d4f98a2
YZ
1192 struct btrfs_shared_data_ref);
1193 if (ret == 0) {
1194 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1195 } else {
1196 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1197 num_refs += refs_to_add;
1198 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
31840ae1 1199 }
5d4f98a2
YZ
1200 } else {
1201 struct btrfs_extent_data_ref *ref;
1202 while (ret == -EEXIST) {
1203 ref = btrfs_item_ptr(leaf, path->slots[0],
1204 struct btrfs_extent_data_ref);
1205 if (match_extent_data_ref(leaf, ref, root_objectid,
1206 owner, offset))
1207 break;
b3b4aa74 1208 btrfs_release_path(path);
5d4f98a2
YZ
1209 key.offset++;
1210 ret = btrfs_insert_empty_item(trans, root, path, &key,
1211 size);
1212 if (ret && ret != -EEXIST)
1213 goto fail;
31840ae1 1214
5d4f98a2
YZ
1215 leaf = path->nodes[0];
1216 }
1217 ref = btrfs_item_ptr(leaf, path->slots[0],
1218 struct btrfs_extent_data_ref);
1219 if (ret == 0) {
1220 btrfs_set_extent_data_ref_root(leaf, ref,
1221 root_objectid);
1222 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1223 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1224 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1225 } else {
1226 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1227 num_refs += refs_to_add;
1228 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
31840ae1 1229 }
31840ae1 1230 }
5d4f98a2
YZ
1231 btrfs_mark_buffer_dirty(leaf);
1232 ret = 0;
1233fail:
b3b4aa74 1234 btrfs_release_path(path);
7bb86316 1235 return ret;
74493f7a
CM
1236}
1237
5d4f98a2
YZ
1238static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1239 struct btrfs_root *root,
1240 struct btrfs_path *path,
1241 int refs_to_drop)
31840ae1 1242{
5d4f98a2
YZ
1243 struct btrfs_key key;
1244 struct btrfs_extent_data_ref *ref1 = NULL;
1245 struct btrfs_shared_data_ref *ref2 = NULL;
31840ae1 1246 struct extent_buffer *leaf;
5d4f98a2 1247 u32 num_refs = 0;
31840ae1
ZY
1248 int ret = 0;
1249
1250 leaf = path->nodes[0];
5d4f98a2
YZ
1251 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1252
1253 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1254 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1255 struct btrfs_extent_data_ref);
1256 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1257 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1258 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1259 struct btrfs_shared_data_ref);
1260 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1261#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1262 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1263 struct btrfs_extent_ref_v0 *ref0;
1264 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1265 struct btrfs_extent_ref_v0);
1266 num_refs = btrfs_ref_count_v0(leaf, ref0);
1267#endif
1268 } else {
1269 BUG();
1270 }
1271
56bec294
CM
1272 BUG_ON(num_refs < refs_to_drop);
1273 num_refs -= refs_to_drop;
5d4f98a2 1274
31840ae1
ZY
1275 if (num_refs == 0) {
1276 ret = btrfs_del_item(trans, root, path);
1277 } else {
5d4f98a2
YZ
1278 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1279 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1280 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1281 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1282#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1283 else {
1284 struct btrfs_extent_ref_v0 *ref0;
1285 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1286 struct btrfs_extent_ref_v0);
1287 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1288 }
1289#endif
31840ae1
ZY
1290 btrfs_mark_buffer_dirty(leaf);
1291 }
31840ae1
ZY
1292 return ret;
1293}
1294
5d4f98a2
YZ
1295static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1296 struct btrfs_path *path,
1297 struct btrfs_extent_inline_ref *iref)
15916de8 1298{
5d4f98a2
YZ
1299 struct btrfs_key key;
1300 struct extent_buffer *leaf;
1301 struct btrfs_extent_data_ref *ref1;
1302 struct btrfs_shared_data_ref *ref2;
1303 u32 num_refs = 0;
1304
1305 leaf = path->nodes[0];
1306 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1307 if (iref) {
1308 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1309 BTRFS_EXTENT_DATA_REF_KEY) {
1310 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1311 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1312 } else {
1313 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1314 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1315 }
1316 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1317 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1318 struct btrfs_extent_data_ref);
1319 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1320 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1321 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1322 struct btrfs_shared_data_ref);
1323 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1324#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1325 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1326 struct btrfs_extent_ref_v0 *ref0;
1327 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1328 struct btrfs_extent_ref_v0);
1329 num_refs = btrfs_ref_count_v0(leaf, ref0);
4b4e25f2 1330#endif
5d4f98a2
YZ
1331 } else {
1332 WARN_ON(1);
1333 }
1334 return num_refs;
1335}
15916de8 1336
5d4f98a2
YZ
1337static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1338 struct btrfs_root *root,
1339 struct btrfs_path *path,
1340 u64 bytenr, u64 parent,
1341 u64 root_objectid)
1f3c79a2 1342{
5d4f98a2 1343 struct btrfs_key key;
1f3c79a2 1344 int ret;
1f3c79a2 1345
5d4f98a2
YZ
1346 key.objectid = bytenr;
1347 if (parent) {
1348 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1349 key.offset = parent;
1350 } else {
1351 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1352 key.offset = root_objectid;
1f3c79a2
LH
1353 }
1354
5d4f98a2
YZ
1355 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1356 if (ret > 0)
1357 ret = -ENOENT;
1358#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1359 if (ret == -ENOENT && parent) {
b3b4aa74 1360 btrfs_release_path(path);
5d4f98a2
YZ
1361 key.type = BTRFS_EXTENT_REF_V0_KEY;
1362 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1363 if (ret > 0)
1364 ret = -ENOENT;
1365 }
1f3c79a2 1366#endif
5d4f98a2 1367 return ret;
1f3c79a2
LH
1368}
1369
5d4f98a2
YZ
1370static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1371 struct btrfs_root *root,
1372 struct btrfs_path *path,
1373 u64 bytenr, u64 parent,
1374 u64 root_objectid)
31840ae1 1375{
5d4f98a2 1376 struct btrfs_key key;
31840ae1 1377 int ret;
31840ae1 1378
5d4f98a2
YZ
1379 key.objectid = bytenr;
1380 if (parent) {
1381 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1382 key.offset = parent;
1383 } else {
1384 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1385 key.offset = root_objectid;
1386 }
1387
1388 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
b3b4aa74 1389 btrfs_release_path(path);
31840ae1
ZY
1390 return ret;
1391}
1392
5d4f98a2 1393static inline int extent_ref_type(u64 parent, u64 owner)
31840ae1 1394{
5d4f98a2
YZ
1395 int type;
1396 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1397 if (parent > 0)
1398 type = BTRFS_SHARED_BLOCK_REF_KEY;
1399 else
1400 type = BTRFS_TREE_BLOCK_REF_KEY;
1401 } else {
1402 if (parent > 0)
1403 type = BTRFS_SHARED_DATA_REF_KEY;
1404 else
1405 type = BTRFS_EXTENT_DATA_REF_KEY;
1406 }
1407 return type;
31840ae1 1408}
56bec294 1409
2c47e605
YZ
1410static int find_next_key(struct btrfs_path *path, int level,
1411 struct btrfs_key *key)
56bec294 1412
02217ed2 1413{
2c47e605 1414 for (; level < BTRFS_MAX_LEVEL; level++) {
5d4f98a2
YZ
1415 if (!path->nodes[level])
1416 break;
5d4f98a2
YZ
1417 if (path->slots[level] + 1 >=
1418 btrfs_header_nritems(path->nodes[level]))
1419 continue;
1420 if (level == 0)
1421 btrfs_item_key_to_cpu(path->nodes[level], key,
1422 path->slots[level] + 1);
1423 else
1424 btrfs_node_key_to_cpu(path->nodes[level], key,
1425 path->slots[level] + 1);
1426 return 0;
1427 }
1428 return 1;
1429}
037e6390 1430
5d4f98a2
YZ
1431/*
1432 * look for inline back ref. if back ref is found, *ref_ret is set
1433 * to the address of inline back ref, and 0 is returned.
1434 *
1435 * if back ref isn't found, *ref_ret is set to the address where it
1436 * should be inserted, and -ENOENT is returned.
1437 *
1438 * if insert is true and there are too many inline back refs, the path
1439 * points to the extent item, and -EAGAIN is returned.
1440 *
1441 * NOTE: inline back refs are ordered in the same way that back ref
1442 * items in the tree are ordered.
1443 */
1444static noinline_for_stack
1445int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1446 struct btrfs_root *root,
1447 struct btrfs_path *path,
1448 struct btrfs_extent_inline_ref **ref_ret,
1449 u64 bytenr, u64 num_bytes,
1450 u64 parent, u64 root_objectid,
1451 u64 owner, u64 offset, int insert)
1452{
1453 struct btrfs_key key;
1454 struct extent_buffer *leaf;
1455 struct btrfs_extent_item *ei;
1456 struct btrfs_extent_inline_ref *iref;
1457 u64 flags;
1458 u64 item_size;
1459 unsigned long ptr;
1460 unsigned long end;
1461 int extra_size;
1462 int type;
1463 int want;
1464 int ret;
1465 int err = 0;
26b8003f 1466
db94535d 1467 key.objectid = bytenr;
31840ae1 1468 key.type = BTRFS_EXTENT_ITEM_KEY;
56bec294 1469 key.offset = num_bytes;
31840ae1 1470
5d4f98a2
YZ
1471 want = extent_ref_type(parent, owner);
1472 if (insert) {
1473 extra_size = btrfs_extent_inline_ref_size(want);
85d4198e 1474 path->keep_locks = 1;
5d4f98a2
YZ
1475 } else
1476 extra_size = -1;
1477 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
b9473439 1478 if (ret < 0) {
5d4f98a2
YZ
1479 err = ret;
1480 goto out;
1481 }
79787eaa
JM
1482 if (ret && !insert) {
1483 err = -ENOENT;
1484 goto out;
1485 }
1486 BUG_ON(ret); /* Corruption */
5d4f98a2
YZ
1487
1488 leaf = path->nodes[0];
1489 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1490#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1491 if (item_size < sizeof(*ei)) {
1492 if (!insert) {
1493 err = -ENOENT;
1494 goto out;
1495 }
1496 ret = convert_extent_item_v0(trans, root, path, owner,
1497 extra_size);
1498 if (ret < 0) {
1499 err = ret;
1500 goto out;
1501 }
1502 leaf = path->nodes[0];
1503 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1504 }
1505#endif
1506 BUG_ON(item_size < sizeof(*ei));
1507
5d4f98a2
YZ
1508 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1509 flags = btrfs_extent_flags(leaf, ei);
1510
1511 ptr = (unsigned long)(ei + 1);
1512 end = (unsigned long)ei + item_size;
1513
1514 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1515 ptr += sizeof(struct btrfs_tree_block_info);
1516 BUG_ON(ptr > end);
1517 } else {
1518 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1519 }
1520
1521 err = -ENOENT;
1522 while (1) {
1523 if (ptr >= end) {
1524 WARN_ON(ptr > end);
1525 break;
1526 }
1527 iref = (struct btrfs_extent_inline_ref *)ptr;
1528 type = btrfs_extent_inline_ref_type(leaf, iref);
1529 if (want < type)
1530 break;
1531 if (want > type) {
1532 ptr += btrfs_extent_inline_ref_size(type);
1533 continue;
1534 }
1535
1536 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1537 struct btrfs_extent_data_ref *dref;
1538 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1539 if (match_extent_data_ref(leaf, dref, root_objectid,
1540 owner, offset)) {
1541 err = 0;
1542 break;
1543 }
1544 if (hash_extent_data_ref_item(leaf, dref) <
1545 hash_extent_data_ref(root_objectid, owner, offset))
1546 break;
1547 } else {
1548 u64 ref_offset;
1549 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1550 if (parent > 0) {
1551 if (parent == ref_offset) {
1552 err = 0;
1553 break;
1554 }
1555 if (ref_offset < parent)
1556 break;
1557 } else {
1558 if (root_objectid == ref_offset) {
1559 err = 0;
1560 break;
1561 }
1562 if (ref_offset < root_objectid)
1563 break;
1564 }
1565 }
1566 ptr += btrfs_extent_inline_ref_size(type);
1567 }
1568 if (err == -ENOENT && insert) {
1569 if (item_size + extra_size >=
1570 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1571 err = -EAGAIN;
1572 goto out;
1573 }
1574 /*
1575 * To add new inline back ref, we have to make sure
1576 * there is no corresponding back ref item.
1577 * For simplicity, we just do not add new inline back
1578 * ref if there is any kind of item for this block
1579 */
2c47e605
YZ
1580 if (find_next_key(path, 0, &key) == 0 &&
1581 key.objectid == bytenr &&
85d4198e 1582 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
5d4f98a2
YZ
1583 err = -EAGAIN;
1584 goto out;
1585 }
1586 }
1587 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1588out:
85d4198e 1589 if (insert) {
5d4f98a2
YZ
1590 path->keep_locks = 0;
1591 btrfs_unlock_up_safe(path, 1);
1592 }
1593 return err;
1594}
1595
1596/*
1597 * helper to add new inline back ref
1598 */
1599static noinline_for_stack
143bede5
JM
1600void setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1601 struct btrfs_root *root,
1602 struct btrfs_path *path,
1603 struct btrfs_extent_inline_ref *iref,
1604 u64 parent, u64 root_objectid,
1605 u64 owner, u64 offset, int refs_to_add,
1606 struct btrfs_delayed_extent_op *extent_op)
5d4f98a2
YZ
1607{
1608 struct extent_buffer *leaf;
1609 struct btrfs_extent_item *ei;
1610 unsigned long ptr;
1611 unsigned long end;
1612 unsigned long item_offset;
1613 u64 refs;
1614 int size;
1615 int type;
5d4f98a2
YZ
1616
1617 leaf = path->nodes[0];
1618 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1619 item_offset = (unsigned long)iref - (unsigned long)ei;
1620
1621 type = extent_ref_type(parent, owner);
1622 size = btrfs_extent_inline_ref_size(type);
1623
143bede5 1624 btrfs_extend_item(trans, root, path, size);
5d4f98a2
YZ
1625
1626 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1627 refs = btrfs_extent_refs(leaf, ei);
1628 refs += refs_to_add;
1629 btrfs_set_extent_refs(leaf, ei, refs);
1630 if (extent_op)
1631 __run_delayed_extent_op(extent_op, leaf, ei);
1632
1633 ptr = (unsigned long)ei + item_offset;
1634 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1635 if (ptr < end - size)
1636 memmove_extent_buffer(leaf, ptr + size, ptr,
1637 end - size - ptr);
1638
1639 iref = (struct btrfs_extent_inline_ref *)ptr;
1640 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1641 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1642 struct btrfs_extent_data_ref *dref;
1643 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1644 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1645 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1646 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1647 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1648 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1649 struct btrfs_shared_data_ref *sref;
1650 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1651 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1652 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1653 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1654 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1655 } else {
1656 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1657 }
1658 btrfs_mark_buffer_dirty(leaf);
5d4f98a2
YZ
1659}
1660
1661static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1662 struct btrfs_root *root,
1663 struct btrfs_path *path,
1664 struct btrfs_extent_inline_ref **ref_ret,
1665 u64 bytenr, u64 num_bytes, u64 parent,
1666 u64 root_objectid, u64 owner, u64 offset)
1667{
1668 int ret;
1669
1670 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1671 bytenr, num_bytes, parent,
1672 root_objectid, owner, offset, 0);
1673 if (ret != -ENOENT)
54aa1f4d 1674 return ret;
5d4f98a2 1675
b3b4aa74 1676 btrfs_release_path(path);
5d4f98a2
YZ
1677 *ref_ret = NULL;
1678
1679 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1680 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1681 root_objectid);
1682 } else {
1683 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1684 root_objectid, owner, offset);
b9473439 1685 }
5d4f98a2
YZ
1686 return ret;
1687}
31840ae1 1688
5d4f98a2
YZ
1689/*
1690 * helper to update/remove inline back ref
1691 */
1692static noinline_for_stack
143bede5
JM
1693void update_inline_extent_backref(struct btrfs_trans_handle *trans,
1694 struct btrfs_root *root,
1695 struct btrfs_path *path,
1696 struct btrfs_extent_inline_ref *iref,
1697 int refs_to_mod,
1698 struct btrfs_delayed_extent_op *extent_op)
5d4f98a2
YZ
1699{
1700 struct extent_buffer *leaf;
1701 struct btrfs_extent_item *ei;
1702 struct btrfs_extent_data_ref *dref = NULL;
1703 struct btrfs_shared_data_ref *sref = NULL;
1704 unsigned long ptr;
1705 unsigned long end;
1706 u32 item_size;
1707 int size;
1708 int type;
5d4f98a2
YZ
1709 u64 refs;
1710
1711 leaf = path->nodes[0];
1712 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1713 refs = btrfs_extent_refs(leaf, ei);
1714 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1715 refs += refs_to_mod;
1716 btrfs_set_extent_refs(leaf, ei, refs);
1717 if (extent_op)
1718 __run_delayed_extent_op(extent_op, leaf, ei);
1719
1720 type = btrfs_extent_inline_ref_type(leaf, iref);
1721
1722 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1723 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1724 refs = btrfs_extent_data_ref_count(leaf, dref);
1725 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1726 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1727 refs = btrfs_shared_data_ref_count(leaf, sref);
1728 } else {
1729 refs = 1;
1730 BUG_ON(refs_to_mod != -1);
56bec294 1731 }
31840ae1 1732
5d4f98a2
YZ
1733 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1734 refs += refs_to_mod;
1735
1736 if (refs > 0) {
1737 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1738 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1739 else
1740 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1741 } else {
1742 size = btrfs_extent_inline_ref_size(type);
1743 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1744 ptr = (unsigned long)iref;
1745 end = (unsigned long)ei + item_size;
1746 if (ptr + size < end)
1747 memmove_extent_buffer(leaf, ptr, ptr + size,
1748 end - ptr - size);
1749 item_size -= size;
143bede5 1750 btrfs_truncate_item(trans, root, path, item_size, 1);
5d4f98a2
YZ
1751 }
1752 btrfs_mark_buffer_dirty(leaf);
5d4f98a2
YZ
1753}
1754
1755static noinline_for_stack
1756int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1757 struct btrfs_root *root,
1758 struct btrfs_path *path,
1759 u64 bytenr, u64 num_bytes, u64 parent,
1760 u64 root_objectid, u64 owner,
1761 u64 offset, int refs_to_add,
1762 struct btrfs_delayed_extent_op *extent_op)
1763{
1764 struct btrfs_extent_inline_ref *iref;
1765 int ret;
1766
1767 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1768 bytenr, num_bytes, parent,
1769 root_objectid, owner, offset, 1);
1770 if (ret == 0) {
1771 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
143bede5
JM
1772 update_inline_extent_backref(trans, root, path, iref,
1773 refs_to_add, extent_op);
5d4f98a2 1774 } else if (ret == -ENOENT) {
143bede5
JM
1775 setup_inline_extent_backref(trans, root, path, iref, parent,
1776 root_objectid, owner, offset,
1777 refs_to_add, extent_op);
1778 ret = 0;
771ed689 1779 }
5d4f98a2
YZ
1780 return ret;
1781}
31840ae1 1782
5d4f98a2
YZ
1783static int insert_extent_backref(struct btrfs_trans_handle *trans,
1784 struct btrfs_root *root,
1785 struct btrfs_path *path,
1786 u64 bytenr, u64 parent, u64 root_objectid,
1787 u64 owner, u64 offset, int refs_to_add)
1788{
1789 int ret;
1790 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1791 BUG_ON(refs_to_add != 1);
1792 ret = insert_tree_block_ref(trans, root, path, bytenr,
1793 parent, root_objectid);
1794 } else {
1795 ret = insert_extent_data_ref(trans, root, path, bytenr,
1796 parent, root_objectid,
1797 owner, offset, refs_to_add);
1798 }
1799 return ret;
1800}
56bec294 1801
5d4f98a2
YZ
1802static int remove_extent_backref(struct btrfs_trans_handle *trans,
1803 struct btrfs_root *root,
1804 struct btrfs_path *path,
1805 struct btrfs_extent_inline_ref *iref,
1806 int refs_to_drop, int is_data)
1807{
143bede5 1808 int ret = 0;
b9473439 1809
5d4f98a2
YZ
1810 BUG_ON(!is_data && refs_to_drop != 1);
1811 if (iref) {
143bede5
JM
1812 update_inline_extent_backref(trans, root, path, iref,
1813 -refs_to_drop, NULL);
5d4f98a2
YZ
1814 } else if (is_data) {
1815 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1816 } else {
1817 ret = btrfs_del_item(trans, root, path);
1818 }
1819 return ret;
1820}
1821
5378e607 1822static int btrfs_issue_discard(struct block_device *bdev,
5d4f98a2
YZ
1823 u64 start, u64 len)
1824{
5378e607 1825 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
5d4f98a2 1826}
5d4f98a2
YZ
1827
1828static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
5378e607 1829 u64 num_bytes, u64 *actual_bytes)
5d4f98a2 1830{
5d4f98a2 1831 int ret;
5378e607 1832 u64 discarded_bytes = 0;
a1d3c478 1833 struct btrfs_bio *bbio = NULL;
5d4f98a2 1834
e244a0ae 1835
5d4f98a2 1836 /* Tell the block device(s) that the sectors can be discarded */
5378e607 1837 ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD,
a1d3c478 1838 bytenr, &num_bytes, &bbio, 0);
79787eaa 1839 /* Error condition is -ENOMEM */
5d4f98a2 1840 if (!ret) {
a1d3c478 1841 struct btrfs_bio_stripe *stripe = bbio->stripes;
5d4f98a2
YZ
1842 int i;
1843
5d4f98a2 1844
a1d3c478 1845 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
d5e2003c
JB
1846 if (!stripe->dev->can_discard)
1847 continue;
1848
5378e607
LD
1849 ret = btrfs_issue_discard(stripe->dev->bdev,
1850 stripe->physical,
1851 stripe->length);
1852 if (!ret)
1853 discarded_bytes += stripe->length;
1854 else if (ret != -EOPNOTSUPP)
79787eaa 1855 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
d5e2003c
JB
1856
1857 /*
1858 * Just in case we get back EOPNOTSUPP for some reason,
1859 * just ignore the return value so we don't screw up
1860 * people calling discard_extent.
1861 */
1862 ret = 0;
5d4f98a2 1863 }
a1d3c478 1864 kfree(bbio);
5d4f98a2 1865 }
5378e607
LD
1866
1867 if (actual_bytes)
1868 *actual_bytes = discarded_bytes;
1869
5d4f98a2
YZ
1870
1871 return ret;
5d4f98a2
YZ
1872}
1873
79787eaa 1874/* Can return -ENOMEM */
5d4f98a2
YZ
1875int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1876 struct btrfs_root *root,
1877 u64 bytenr, u64 num_bytes, u64 parent,
66d7e7f0 1878 u64 root_objectid, u64 owner, u64 offset, int for_cow)
5d4f98a2
YZ
1879{
1880 int ret;
66d7e7f0
AJ
1881 struct btrfs_fs_info *fs_info = root->fs_info;
1882
5d4f98a2
YZ
1883 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1884 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1885
1886 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
66d7e7f0
AJ
1887 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1888 num_bytes,
5d4f98a2 1889 parent, root_objectid, (int)owner,
66d7e7f0 1890 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
5d4f98a2 1891 } else {
66d7e7f0
AJ
1892 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1893 num_bytes,
5d4f98a2 1894 parent, root_objectid, owner, offset,
66d7e7f0 1895 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
5d4f98a2
YZ
1896 }
1897 return ret;
1898}
1899
1900static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1901 struct btrfs_root *root,
1902 u64 bytenr, u64 num_bytes,
1903 u64 parent, u64 root_objectid,
1904 u64 owner, u64 offset, int refs_to_add,
1905 struct btrfs_delayed_extent_op *extent_op)
1906{
1907 struct btrfs_path *path;
1908 struct extent_buffer *leaf;
1909 struct btrfs_extent_item *item;
1910 u64 refs;
1911 int ret;
1912 int err = 0;
1913
1914 path = btrfs_alloc_path();
1915 if (!path)
1916 return -ENOMEM;
1917
1918 path->reada = 1;
1919 path->leave_spinning = 1;
1920 /* this will setup the path even if it fails to insert the back ref */
1921 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1922 path, bytenr, num_bytes, parent,
1923 root_objectid, owner, offset,
1924 refs_to_add, extent_op);
1925 if (ret == 0)
1926 goto out;
1927
1928 if (ret != -EAGAIN) {
1929 err = ret;
1930 goto out;
1931 }
1932
1933 leaf = path->nodes[0];
1934 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1935 refs = btrfs_extent_refs(leaf, item);
1936 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1937 if (extent_op)
1938 __run_delayed_extent_op(extent_op, leaf, item);
56bec294 1939
5d4f98a2 1940 btrfs_mark_buffer_dirty(leaf);
b3b4aa74 1941 btrfs_release_path(path);
56bec294
CM
1942
1943 path->reada = 1;
b9473439
CM
1944 path->leave_spinning = 1;
1945
56bec294
CM
1946 /* now insert the actual backref */
1947 ret = insert_extent_backref(trans, root->fs_info->extent_root,
5d4f98a2
YZ
1948 path, bytenr, parent, root_objectid,
1949 owner, offset, refs_to_add);
79787eaa
JM
1950 if (ret)
1951 btrfs_abort_transaction(trans, root, ret);
5d4f98a2 1952out:
56bec294 1953 btrfs_free_path(path);
5d4f98a2 1954 return err;
56bec294
CM
1955}
1956
5d4f98a2
YZ
1957static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1958 struct btrfs_root *root,
1959 struct btrfs_delayed_ref_node *node,
1960 struct btrfs_delayed_extent_op *extent_op,
1961 int insert_reserved)
56bec294 1962{
5d4f98a2
YZ
1963 int ret = 0;
1964 struct btrfs_delayed_data_ref *ref;
1965 struct btrfs_key ins;
1966 u64 parent = 0;
1967 u64 ref_root = 0;
1968 u64 flags = 0;
1969
1970 ins.objectid = node->bytenr;
1971 ins.offset = node->num_bytes;
1972 ins.type = BTRFS_EXTENT_ITEM_KEY;
1973
1974 ref = btrfs_delayed_node_to_data_ref(node);
1975 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1976 parent = ref->parent;
1977 else
1978 ref_root = ref->root;
1979
1980 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1981 if (extent_op) {
1982 BUG_ON(extent_op->update_key);
1983 flags |= extent_op->flags_to_set;
1984 }
1985 ret = alloc_reserved_file_extent(trans, root,
1986 parent, ref_root, flags,
1987 ref->objectid, ref->offset,
1988 &ins, node->ref_mod);
5d4f98a2
YZ
1989 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1990 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1991 node->num_bytes, parent,
1992 ref_root, ref->objectid,
1993 ref->offset, node->ref_mod,
1994 extent_op);
1995 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1996 ret = __btrfs_free_extent(trans, root, node->bytenr,
1997 node->num_bytes, parent,
1998 ref_root, ref->objectid,
1999 ref->offset, node->ref_mod,
2000 extent_op);
2001 } else {
2002 BUG();
2003 }
2004 return ret;
2005}
2006
2007static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2008 struct extent_buffer *leaf,
2009 struct btrfs_extent_item *ei)
2010{
2011 u64 flags = btrfs_extent_flags(leaf, ei);
2012 if (extent_op->update_flags) {
2013 flags |= extent_op->flags_to_set;
2014 btrfs_set_extent_flags(leaf, ei, flags);
2015 }
2016
2017 if (extent_op->update_key) {
2018 struct btrfs_tree_block_info *bi;
2019 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2020 bi = (struct btrfs_tree_block_info *)(ei + 1);
2021 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2022 }
2023}
2024
2025static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2026 struct btrfs_root *root,
2027 struct btrfs_delayed_ref_node *node,
2028 struct btrfs_delayed_extent_op *extent_op)
2029{
2030 struct btrfs_key key;
2031 struct btrfs_path *path;
2032 struct btrfs_extent_item *ei;
2033 struct extent_buffer *leaf;
2034 u32 item_size;
56bec294 2035 int ret;
5d4f98a2
YZ
2036 int err = 0;
2037
79787eaa
JM
2038 if (trans->aborted)
2039 return 0;
2040
5d4f98a2
YZ
2041 path = btrfs_alloc_path();
2042 if (!path)
2043 return -ENOMEM;
2044
2045 key.objectid = node->bytenr;
2046 key.type = BTRFS_EXTENT_ITEM_KEY;
2047 key.offset = node->num_bytes;
2048
2049 path->reada = 1;
2050 path->leave_spinning = 1;
2051 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2052 path, 0, 1);
2053 if (ret < 0) {
2054 err = ret;
2055 goto out;
2056 }
2057 if (ret > 0) {
2058 err = -EIO;
2059 goto out;
2060 }
2061
2062 leaf = path->nodes[0];
2063 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2064#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2065 if (item_size < sizeof(*ei)) {
2066 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2067 path, (u64)-1, 0);
2068 if (ret < 0) {
2069 err = ret;
2070 goto out;
2071 }
2072 leaf = path->nodes[0];
2073 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2074 }
2075#endif
2076 BUG_ON(item_size < sizeof(*ei));
2077 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2078 __run_delayed_extent_op(extent_op, leaf, ei);
56bec294 2079
5d4f98a2
YZ
2080 btrfs_mark_buffer_dirty(leaf);
2081out:
2082 btrfs_free_path(path);
2083 return err;
56bec294
CM
2084}
2085
5d4f98a2
YZ
2086static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2087 struct btrfs_root *root,
2088 struct btrfs_delayed_ref_node *node,
2089 struct btrfs_delayed_extent_op *extent_op,
2090 int insert_reserved)
56bec294
CM
2091{
2092 int ret = 0;
5d4f98a2
YZ
2093 struct btrfs_delayed_tree_ref *ref;
2094 struct btrfs_key ins;
2095 u64 parent = 0;
2096 u64 ref_root = 0;
56bec294 2097
5d4f98a2
YZ
2098 ins.objectid = node->bytenr;
2099 ins.offset = node->num_bytes;
2100 ins.type = BTRFS_EXTENT_ITEM_KEY;
56bec294 2101
5d4f98a2
YZ
2102 ref = btrfs_delayed_node_to_tree_ref(node);
2103 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2104 parent = ref->parent;
2105 else
2106 ref_root = ref->root;
2107
2108 BUG_ON(node->ref_mod != 1);
2109 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2110 BUG_ON(!extent_op || !extent_op->update_flags ||
2111 !extent_op->update_key);
2112 ret = alloc_reserved_tree_block(trans, root,
2113 parent, ref_root,
2114 extent_op->flags_to_set,
2115 &extent_op->key,
2116 ref->level, &ins);
5d4f98a2
YZ
2117 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2118 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2119 node->num_bytes, parent, ref_root,
2120 ref->level, 0, 1, extent_op);
2121 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2122 ret = __btrfs_free_extent(trans, root, node->bytenr,
2123 node->num_bytes, parent, ref_root,
2124 ref->level, 0, 1, extent_op);
2125 } else {
2126 BUG();
2127 }
56bec294
CM
2128 return ret;
2129}
2130
2131/* helper function to actually process a single delayed ref entry */
5d4f98a2
YZ
2132static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2133 struct btrfs_root *root,
2134 struct btrfs_delayed_ref_node *node,
2135 struct btrfs_delayed_extent_op *extent_op,
2136 int insert_reserved)
56bec294 2137{
79787eaa
JM
2138 int ret = 0;
2139
2140 if (trans->aborted)
2141 return 0;
2142
5d4f98a2 2143 if (btrfs_delayed_ref_is_head(node)) {
56bec294
CM
2144 struct btrfs_delayed_ref_head *head;
2145 /*
2146 * we've hit the end of the chain and we were supposed
2147 * to insert this extent into the tree. But, it got
2148 * deleted before we ever needed to insert it, so all
2149 * we have to do is clean up the accounting
2150 */
5d4f98a2
YZ
2151 BUG_ON(extent_op);
2152 head = btrfs_delayed_node_to_head(node);
56bec294 2153 if (insert_reserved) {
f0486c68
YZ
2154 btrfs_pin_extent(root, node->bytenr,
2155 node->num_bytes, 1);
5d4f98a2
YZ
2156 if (head->is_data) {
2157 ret = btrfs_del_csums(trans, root,
2158 node->bytenr,
2159 node->num_bytes);
5d4f98a2 2160 }
56bec294 2161 }
56bec294 2162 mutex_unlock(&head->mutex);
79787eaa 2163 return ret;
56bec294
CM
2164 }
2165
5d4f98a2
YZ
2166 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2167 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2168 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2169 insert_reserved);
2170 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2171 node->type == BTRFS_SHARED_DATA_REF_KEY)
2172 ret = run_delayed_data_ref(trans, root, node, extent_op,
2173 insert_reserved);
2174 else
2175 BUG();
2176 return ret;
56bec294
CM
2177}
2178
2179static noinline struct btrfs_delayed_ref_node *
2180select_delayed_ref(struct btrfs_delayed_ref_head *head)
2181{
2182 struct rb_node *node;
2183 struct btrfs_delayed_ref_node *ref;
2184 int action = BTRFS_ADD_DELAYED_REF;
2185again:
2186 /*
2187 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2188 * this prevents ref count from going down to zero when
2189 * there still are pending delayed ref.
2190 */
2191 node = rb_prev(&head->node.rb_node);
2192 while (1) {
2193 if (!node)
2194 break;
2195 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2196 rb_node);
2197 if (ref->bytenr != head->node.bytenr)
2198 break;
5d4f98a2 2199 if (ref->action == action)
56bec294
CM
2200 return ref;
2201 node = rb_prev(node);
2202 }
2203 if (action == BTRFS_ADD_DELAYED_REF) {
2204 action = BTRFS_DROP_DELAYED_REF;
2205 goto again;
2206 }
2207 return NULL;
2208}
2209
79787eaa
JM
2210/*
2211 * Returns 0 on success or if called with an already aborted transaction.
2212 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2213 */
c3e69d58
CM
2214static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2215 struct btrfs_root *root,
2216 struct list_head *cluster)
56bec294 2217{
56bec294
CM
2218 struct btrfs_delayed_ref_root *delayed_refs;
2219 struct btrfs_delayed_ref_node *ref;
2220 struct btrfs_delayed_ref_head *locked_ref = NULL;
5d4f98a2 2221 struct btrfs_delayed_extent_op *extent_op;
097b8a7c 2222 struct btrfs_fs_info *fs_info = root->fs_info;
56bec294 2223 int ret;
c3e69d58 2224 int count = 0;
56bec294 2225 int must_insert_reserved = 0;
56bec294
CM
2226
2227 delayed_refs = &trans->transaction->delayed_refs;
56bec294
CM
2228 while (1) {
2229 if (!locked_ref) {
c3e69d58
CM
2230 /* pick a new head ref from the cluster list */
2231 if (list_empty(cluster))
56bec294 2232 break;
56bec294 2233
c3e69d58
CM
2234 locked_ref = list_entry(cluster->next,
2235 struct btrfs_delayed_ref_head, cluster);
2236
2237 /* grab the lock that says we are going to process
2238 * all the refs for this head */
2239 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2240
2241 /*
2242 * we may have dropped the spin lock to get the head
2243 * mutex lock, and that might have given someone else
2244 * time to free the head. If that's true, it has been
2245 * removed from our list and we can move on.
2246 */
2247 if (ret == -EAGAIN) {
2248 locked_ref = NULL;
2249 count++;
2250 continue;
56bec294
CM
2251 }
2252 }
a28ec197 2253
d1270cd9
AJ
2254 /*
2255 * locked_ref is the head node, so we have to go one
2256 * node back for any delayed ref updates
2257 */
2258 ref = select_delayed_ref(locked_ref);
2259
2260 if (ref && ref->seq &&
097b8a7c 2261 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
d1270cd9
AJ
2262 /*
2263 * there are still refs with lower seq numbers in the
2264 * process of being added. Don't run this ref yet.
2265 */
2266 list_del_init(&locked_ref->cluster);
2267 mutex_unlock(&locked_ref->mutex);
2268 locked_ref = NULL;
2269 delayed_refs->num_heads_ready++;
2270 spin_unlock(&delayed_refs->lock);
2271 cond_resched();
2272 spin_lock(&delayed_refs->lock);
2273 continue;
2274 }
2275
56bec294
CM
2276 /*
2277 * record the must insert reserved flag before we
2278 * drop the spin lock.
2279 */
2280 must_insert_reserved = locked_ref->must_insert_reserved;
2281 locked_ref->must_insert_reserved = 0;
7bb86316 2282
5d4f98a2
YZ
2283 extent_op = locked_ref->extent_op;
2284 locked_ref->extent_op = NULL;
2285
56bec294
CM
2286 if (!ref) {
2287 /* All delayed refs have been processed, Go ahead
2288 * and send the head node to run_one_delayed_ref,
2289 * so that any accounting fixes can happen
2290 */
2291 ref = &locked_ref->node;
5d4f98a2
YZ
2292
2293 if (extent_op && must_insert_reserved) {
2294 kfree(extent_op);
2295 extent_op = NULL;
2296 }
2297
2298 if (extent_op) {
2299 spin_unlock(&delayed_refs->lock);
2300
2301 ret = run_delayed_extent_op(trans, root,
2302 ref, extent_op);
5d4f98a2
YZ
2303 kfree(extent_op);
2304
79787eaa
JM
2305 if (ret) {
2306 printk(KERN_DEBUG "btrfs: run_delayed_extent_op returned %d\n", ret);
253beebd 2307 spin_lock(&delayed_refs->lock);
79787eaa
JM
2308 return ret;
2309 }
2310
203bf287 2311 goto next;
5d4f98a2
YZ
2312 }
2313
c3e69d58 2314 list_del_init(&locked_ref->cluster);
56bec294
CM
2315 locked_ref = NULL;
2316 }
02217ed2 2317
56bec294
CM
2318 ref->in_tree = 0;
2319 rb_erase(&ref->rb_node, &delayed_refs->root);
2320 delayed_refs->num_entries--;
a168650c
JS
2321 /*
2322 * we modified num_entries, but as we're currently running
2323 * delayed refs, skip
2324 * wake_up(&delayed_refs->seq_wait);
2325 * here.
2326 */
56bec294 2327 spin_unlock(&delayed_refs->lock);
925baedd 2328
5d4f98a2 2329 ret = run_one_delayed_ref(trans, root, ref, extent_op,
56bec294 2330 must_insert_reserved);
eb099670 2331
5d4f98a2
YZ
2332 btrfs_put_delayed_ref(ref);
2333 kfree(extent_op);
c3e69d58 2334 count++;
79787eaa
JM
2335
2336 if (ret) {
2337 printk(KERN_DEBUG "btrfs: run_one_delayed_ref returned %d\n", ret);
253beebd 2338 spin_lock(&delayed_refs->lock);
79787eaa
JM
2339 return ret;
2340 }
2341
203bf287 2342next:
097b8a7c 2343 do_chunk_alloc(trans, fs_info->extent_root,
203bf287
CM
2344 2 * 1024 * 1024,
2345 btrfs_get_alloc_profile(root, 0),
2346 CHUNK_ALLOC_NO_FORCE);
c3e69d58
CM
2347 cond_resched();
2348 spin_lock(&delayed_refs->lock);
2349 }
2350 return count;
2351}
2352
097b8a7c
JS
2353static void wait_for_more_refs(struct btrfs_fs_info *fs_info,
2354 struct btrfs_delayed_ref_root *delayed_refs,
8ca78f3e
JS
2355 unsigned long num_refs,
2356 struct list_head *first_seq)
a168650c 2357{
a168650c
JS
2358 spin_unlock(&delayed_refs->lock);
2359 pr_debug("waiting for more refs (num %ld, first %p)\n",
2360 num_refs, first_seq);
097b8a7c 2361 wait_event(fs_info->tree_mod_seq_wait,
a168650c 2362 num_refs != delayed_refs->num_entries ||
097b8a7c 2363 fs_info->tree_mod_seq_list.next != first_seq);
a168650c 2364 pr_debug("done waiting for more refs (num %ld, first %p)\n",
097b8a7c 2365 delayed_refs->num_entries, fs_info->tree_mod_seq_list.next);
a168650c
JS
2366 spin_lock(&delayed_refs->lock);
2367}
2368
709c0486
AJ
2369#ifdef SCRAMBLE_DELAYED_REFS
2370/*
2371 * Normally delayed refs get processed in ascending bytenr order. This
2372 * correlates in most cases to the order added. To expose dependencies on this
2373 * order, we start to process the tree in the middle instead of the beginning
2374 */
2375static u64 find_middle(struct rb_root *root)
2376{
2377 struct rb_node *n = root->rb_node;
2378 struct btrfs_delayed_ref_node *entry;
2379 int alt = 1;
2380 u64 middle;
2381 u64 first = 0, last = 0;
2382
2383 n = rb_first(root);
2384 if (n) {
2385 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2386 first = entry->bytenr;
2387 }
2388 n = rb_last(root);
2389 if (n) {
2390 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2391 last = entry->bytenr;
2392 }
2393 n = root->rb_node;
2394
2395 while (n) {
2396 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2397 WARN_ON(!entry->in_tree);
2398
2399 middle = entry->bytenr;
2400
2401 if (alt)
2402 n = n->rb_left;
2403 else
2404 n = n->rb_right;
2405
2406 alt = 1 - alt;
2407 }
2408 return middle;
2409}
2410#endif
2411
bed92eae
AJ
2412int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2413 struct btrfs_fs_info *fs_info)
2414{
2415 struct qgroup_update *qgroup_update;
2416 int ret = 0;
2417
2418 if (list_empty(&trans->qgroup_ref_list) !=
2419 !trans->delayed_ref_elem.seq) {
2420 /* list without seq or seq without list */
2421 printk(KERN_ERR "btrfs: qgroup accounting update error, list is%s empty, seq is %llu\n",
2422 list_empty(&trans->qgroup_ref_list) ? "" : " not",
2423 trans->delayed_ref_elem.seq);
2424 BUG();
2425 }
2426
2427 if (!trans->delayed_ref_elem.seq)
2428 return 0;
2429
2430 while (!list_empty(&trans->qgroup_ref_list)) {
2431 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2432 struct qgroup_update, list);
2433 list_del(&qgroup_update->list);
2434 if (!ret)
2435 ret = btrfs_qgroup_account_ref(
2436 trans, fs_info, qgroup_update->node,
2437 qgroup_update->extent_op);
2438 kfree(qgroup_update);
2439 }
2440
2441 btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2442
2443 return ret;
2444}
2445
c3e69d58
CM
2446/*
2447 * this starts processing the delayed reference count updates and
2448 * extent insertions we have queued up so far. count can be
2449 * 0, which means to process everything in the tree at the start
2450 * of the run (but not newly added entries), or it can be some target
2451 * number you'd like to process.
79787eaa
JM
2452 *
2453 * Returns 0 on success or if called with an aborted transaction
2454 * Returns <0 on error and aborts the transaction
c3e69d58
CM
2455 */
2456int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2457 struct btrfs_root *root, unsigned long count)
2458{
2459 struct rb_node *node;
2460 struct btrfs_delayed_ref_root *delayed_refs;
2461 struct btrfs_delayed_ref_node *ref;
2462 struct list_head cluster;
8ca78f3e 2463 struct list_head *first_seq = NULL;
c3e69d58 2464 int ret;
a168650c 2465 u64 delayed_start;
c3e69d58
CM
2466 int run_all = count == (unsigned long)-1;
2467 int run_most = 0;
a168650c
JS
2468 unsigned long num_refs = 0;
2469 int consider_waiting;
c3e69d58 2470
79787eaa
JM
2471 /* We'll clean this up in btrfs_cleanup_transaction */
2472 if (trans->aborted)
2473 return 0;
2474
c3e69d58
CM
2475 if (root == root->fs_info->extent_root)
2476 root = root->fs_info->tree_root;
2477
203bf287
CM
2478 do_chunk_alloc(trans, root->fs_info->extent_root,
2479 2 * 1024 * 1024, btrfs_get_alloc_profile(root, 0),
2480 CHUNK_ALLOC_NO_FORCE);
2481
edf39272
JS
2482 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2483
c3e69d58
CM
2484 delayed_refs = &trans->transaction->delayed_refs;
2485 INIT_LIST_HEAD(&cluster);
2486again:
a168650c 2487 consider_waiting = 0;
c3e69d58 2488 spin_lock(&delayed_refs->lock);
097b8a7c 2489
709c0486
AJ
2490#ifdef SCRAMBLE_DELAYED_REFS
2491 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2492#endif
2493
c3e69d58
CM
2494 if (count == 0) {
2495 count = delayed_refs->num_entries * 2;
2496 run_most = 1;
2497 }
2498 while (1) {
2499 if (!(run_all || run_most) &&
2500 delayed_refs->num_heads_ready < 64)
2501 break;
eb099670 2502
56bec294 2503 /*
c3e69d58
CM
2504 * go find something we can process in the rbtree. We start at
2505 * the beginning of the tree, and then build a cluster
2506 * of refs to process starting at the first one we are able to
2507 * lock
56bec294 2508 */
a168650c 2509 delayed_start = delayed_refs->run_delayed_start;
c3e69d58
CM
2510 ret = btrfs_find_ref_cluster(trans, &cluster,
2511 delayed_refs->run_delayed_start);
2512 if (ret)
56bec294
CM
2513 break;
2514
a168650c
JS
2515 if (delayed_start >= delayed_refs->run_delayed_start) {
2516 if (consider_waiting == 0) {
2517 /*
2518 * btrfs_find_ref_cluster looped. let's do one
2519 * more cycle. if we don't run any delayed ref
2520 * during that cycle (because we can't because
2521 * all of them are blocked) and if the number of
2522 * refs doesn't change, we avoid busy waiting.
2523 */
2524 consider_waiting = 1;
2525 num_refs = delayed_refs->num_entries;
8ca78f3e 2526 first_seq = root->fs_info->tree_mod_seq_list.next;
a168650c 2527 } else {
097b8a7c 2528 wait_for_more_refs(root->fs_info, delayed_refs,
8ca78f3e 2529 num_refs, first_seq);
a168650c
JS
2530 /*
2531 * after waiting, things have changed. we
2532 * dropped the lock and someone else might have
2533 * run some refs, built new clusters and so on.
2534 * therefore, we restart staleness detection.
2535 */
2536 consider_waiting = 0;
2537 }
2538 }
2539
c3e69d58 2540 ret = run_clustered_refs(trans, root, &cluster);
79787eaa
JM
2541 if (ret < 0) {
2542 spin_unlock(&delayed_refs->lock);
2543 btrfs_abort_transaction(trans, root, ret);
2544 return ret;
2545 }
c3e69d58
CM
2546
2547 count -= min_t(unsigned long, ret, count);
2548
2549 if (count == 0)
2550 break;
a168650c
JS
2551
2552 if (ret || delayed_refs->run_delayed_start == 0) {
2553 /* refs were run, let's reset staleness detection */
2554 consider_waiting = 0;
2555 }
eb099670 2556 }
c3e69d58 2557
56bec294 2558 if (run_all) {
56bec294 2559 node = rb_first(&delayed_refs->root);
c3e69d58 2560 if (!node)
56bec294 2561 goto out;
c3e69d58 2562 count = (unsigned long)-1;
e9d0b13b 2563
56bec294
CM
2564 while (node) {
2565 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2566 rb_node);
2567 if (btrfs_delayed_ref_is_head(ref)) {
2568 struct btrfs_delayed_ref_head *head;
5caf2a00 2569
56bec294
CM
2570 head = btrfs_delayed_node_to_head(ref);
2571 atomic_inc(&ref->refs);
2572
2573 spin_unlock(&delayed_refs->lock);
8cc33e5c
DS
2574 /*
2575 * Mutex was contended, block until it's
2576 * released and try again
2577 */
56bec294
CM
2578 mutex_lock(&head->mutex);
2579 mutex_unlock(&head->mutex);
2580
2581 btrfs_put_delayed_ref(ref);
1887be66 2582 cond_resched();
56bec294
CM
2583 goto again;
2584 }
2585 node = rb_next(node);
2586 }
2587 spin_unlock(&delayed_refs->lock);
56bec294
CM
2588 schedule_timeout(1);
2589 goto again;
5f39d397 2590 }
54aa1f4d 2591out:
c3e69d58 2592 spin_unlock(&delayed_refs->lock);
edf39272 2593 assert_qgroups_uptodate(trans);
a28ec197
CM
2594 return 0;
2595}
2596
5d4f98a2
YZ
2597int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2598 struct btrfs_root *root,
2599 u64 bytenr, u64 num_bytes, u64 flags,
2600 int is_data)
2601{
2602 struct btrfs_delayed_extent_op *extent_op;
2603 int ret;
2604
2605 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2606 if (!extent_op)
2607 return -ENOMEM;
2608
2609 extent_op->flags_to_set = flags;
2610 extent_op->update_flags = 1;
2611 extent_op->update_key = 0;
2612 extent_op->is_data = is_data ? 1 : 0;
2613
66d7e7f0
AJ
2614 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2615 num_bytes, extent_op);
5d4f98a2
YZ
2616 if (ret)
2617 kfree(extent_op);
2618 return ret;
2619}
2620
2621static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2622 struct btrfs_root *root,
2623 struct btrfs_path *path,
2624 u64 objectid, u64 offset, u64 bytenr)
2625{
2626 struct btrfs_delayed_ref_head *head;
2627 struct btrfs_delayed_ref_node *ref;
2628 struct btrfs_delayed_data_ref *data_ref;
2629 struct btrfs_delayed_ref_root *delayed_refs;
2630 struct rb_node *node;
2631 int ret = 0;
2632
2633 ret = -ENOENT;
2634 delayed_refs = &trans->transaction->delayed_refs;
2635 spin_lock(&delayed_refs->lock);
2636 head = btrfs_find_delayed_ref_head(trans, bytenr);
2637 if (!head)
2638 goto out;
2639
2640 if (!mutex_trylock(&head->mutex)) {
2641 atomic_inc(&head->node.refs);
2642 spin_unlock(&delayed_refs->lock);
2643
b3b4aa74 2644 btrfs_release_path(path);
5d4f98a2 2645
8cc33e5c
DS
2646 /*
2647 * Mutex was contended, block until it's released and let
2648 * caller try again
2649 */
5d4f98a2
YZ
2650 mutex_lock(&head->mutex);
2651 mutex_unlock(&head->mutex);
2652 btrfs_put_delayed_ref(&head->node);
2653 return -EAGAIN;
2654 }
2655
2656 node = rb_prev(&head->node.rb_node);
2657 if (!node)
2658 goto out_unlock;
2659
2660 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2661
2662 if (ref->bytenr != bytenr)
2663 goto out_unlock;
2664
2665 ret = 1;
2666 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2667 goto out_unlock;
2668
2669 data_ref = btrfs_delayed_node_to_data_ref(ref);
2670
2671 node = rb_prev(node);
2672 if (node) {
2673 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2674 if (ref->bytenr == bytenr)
2675 goto out_unlock;
2676 }
2677
2678 if (data_ref->root != root->root_key.objectid ||
2679 data_ref->objectid != objectid || data_ref->offset != offset)
2680 goto out_unlock;
2681
2682 ret = 0;
2683out_unlock:
2684 mutex_unlock(&head->mutex);
2685out:
2686 spin_unlock(&delayed_refs->lock);
2687 return ret;
2688}
2689
2690static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2691 struct btrfs_root *root,
2692 struct btrfs_path *path,
2693 u64 objectid, u64 offset, u64 bytenr)
be20aa9d
CM
2694{
2695 struct btrfs_root *extent_root = root->fs_info->extent_root;
f321e491 2696 struct extent_buffer *leaf;
5d4f98a2
YZ
2697 struct btrfs_extent_data_ref *ref;
2698 struct btrfs_extent_inline_ref *iref;
2699 struct btrfs_extent_item *ei;
f321e491 2700 struct btrfs_key key;
5d4f98a2 2701 u32 item_size;
be20aa9d 2702 int ret;
925baedd 2703
be20aa9d 2704 key.objectid = bytenr;
31840ae1 2705 key.offset = (u64)-1;
f321e491 2706 key.type = BTRFS_EXTENT_ITEM_KEY;
be20aa9d 2707
be20aa9d
CM
2708 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2709 if (ret < 0)
2710 goto out;
79787eaa 2711 BUG_ON(ret == 0); /* Corruption */
80ff3856
YZ
2712
2713 ret = -ENOENT;
2714 if (path->slots[0] == 0)
31840ae1 2715 goto out;
be20aa9d 2716
31840ae1 2717 path->slots[0]--;
f321e491 2718 leaf = path->nodes[0];
5d4f98a2 2719 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
be20aa9d 2720
5d4f98a2 2721 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
be20aa9d 2722 goto out;
f321e491 2723
5d4f98a2
YZ
2724 ret = 1;
2725 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2726#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2727 if (item_size < sizeof(*ei)) {
2728 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2729 goto out;
2730 }
2731#endif
2732 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
bd09835d 2733
5d4f98a2
YZ
2734 if (item_size != sizeof(*ei) +
2735 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2736 goto out;
be20aa9d 2737
5d4f98a2
YZ
2738 if (btrfs_extent_generation(leaf, ei) <=
2739 btrfs_root_last_snapshot(&root->root_item))
2740 goto out;
2741
2742 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2743 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2744 BTRFS_EXTENT_DATA_REF_KEY)
2745 goto out;
2746
2747 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2748 if (btrfs_extent_refs(leaf, ei) !=
2749 btrfs_extent_data_ref_count(leaf, ref) ||
2750 btrfs_extent_data_ref_root(leaf, ref) !=
2751 root->root_key.objectid ||
2752 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2753 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2754 goto out;
2755
2756 ret = 0;
2757out:
2758 return ret;
2759}
2760
2761int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2762 struct btrfs_root *root,
2763 u64 objectid, u64 offset, u64 bytenr)
2764{
2765 struct btrfs_path *path;
2766 int ret;
2767 int ret2;
2768
2769 path = btrfs_alloc_path();
2770 if (!path)
2771 return -ENOENT;
2772
2773 do {
2774 ret = check_committed_ref(trans, root, path, objectid,
2775 offset, bytenr);
2776 if (ret && ret != -ENOENT)
f321e491 2777 goto out;
80ff3856 2778
5d4f98a2
YZ
2779 ret2 = check_delayed_ref(trans, root, path, objectid,
2780 offset, bytenr);
2781 } while (ret2 == -EAGAIN);
2782
2783 if (ret2 && ret2 != -ENOENT) {
2784 ret = ret2;
2785 goto out;
f321e491 2786 }
5d4f98a2
YZ
2787
2788 if (ret != -ENOENT || ret2 != -ENOENT)
2789 ret = 0;
be20aa9d 2790out:
80ff3856 2791 btrfs_free_path(path);
f0486c68
YZ
2792 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2793 WARN_ON(ret > 0);
f321e491 2794 return ret;
be20aa9d 2795}
c5739bba 2796
5d4f98a2 2797static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
b7a9f29f 2798 struct btrfs_root *root,
5d4f98a2 2799 struct extent_buffer *buf,
66d7e7f0 2800 int full_backref, int inc, int for_cow)
31840ae1
ZY
2801{
2802 u64 bytenr;
5d4f98a2
YZ
2803 u64 num_bytes;
2804 u64 parent;
31840ae1 2805 u64 ref_root;
31840ae1 2806 u32 nritems;
31840ae1
ZY
2807 struct btrfs_key key;
2808 struct btrfs_file_extent_item *fi;
2809 int i;
2810 int level;
2811 int ret = 0;
31840ae1 2812 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
66d7e7f0 2813 u64, u64, u64, u64, u64, u64, int);
31840ae1
ZY
2814
2815 ref_root = btrfs_header_owner(buf);
31840ae1
ZY
2816 nritems = btrfs_header_nritems(buf);
2817 level = btrfs_header_level(buf);
2818
5d4f98a2
YZ
2819 if (!root->ref_cows && level == 0)
2820 return 0;
31840ae1 2821
5d4f98a2
YZ
2822 if (inc)
2823 process_func = btrfs_inc_extent_ref;
2824 else
2825 process_func = btrfs_free_extent;
31840ae1 2826
5d4f98a2
YZ
2827 if (full_backref)
2828 parent = buf->start;
2829 else
2830 parent = 0;
2831
2832 for (i = 0; i < nritems; i++) {
31840ae1 2833 if (level == 0) {
5d4f98a2 2834 btrfs_item_key_to_cpu(buf, &key, i);
31840ae1
ZY
2835 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2836 continue;
5d4f98a2 2837 fi = btrfs_item_ptr(buf, i,
31840ae1
ZY
2838 struct btrfs_file_extent_item);
2839 if (btrfs_file_extent_type(buf, fi) ==
2840 BTRFS_FILE_EXTENT_INLINE)
2841 continue;
2842 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2843 if (bytenr == 0)
2844 continue;
5d4f98a2
YZ
2845
2846 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2847 key.offset -= btrfs_file_extent_offset(buf, fi);
2848 ret = process_func(trans, root, bytenr, num_bytes,
2849 parent, ref_root, key.objectid,
66d7e7f0 2850 key.offset, for_cow);
31840ae1
ZY
2851 if (ret)
2852 goto fail;
2853 } else {
5d4f98a2
YZ
2854 bytenr = btrfs_node_blockptr(buf, i);
2855 num_bytes = btrfs_level_size(root, level - 1);
2856 ret = process_func(trans, root, bytenr, num_bytes,
66d7e7f0
AJ
2857 parent, ref_root, level - 1, 0,
2858 for_cow);
31840ae1
ZY
2859 if (ret)
2860 goto fail;
2861 }
2862 }
2863 return 0;
2864fail:
5d4f98a2
YZ
2865 return ret;
2866}
2867
2868int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
66d7e7f0 2869 struct extent_buffer *buf, int full_backref, int for_cow)
5d4f98a2 2870{
66d7e7f0 2871 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
5d4f98a2
YZ
2872}
2873
2874int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
66d7e7f0 2875 struct extent_buffer *buf, int full_backref, int for_cow)
5d4f98a2 2876{
66d7e7f0 2877 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
31840ae1
ZY
2878}
2879
9078a3e1
CM
2880static int write_one_cache_group(struct btrfs_trans_handle *trans,
2881 struct btrfs_root *root,
2882 struct btrfs_path *path,
2883 struct btrfs_block_group_cache *cache)
2884{
2885 int ret;
9078a3e1 2886 struct btrfs_root *extent_root = root->fs_info->extent_root;
5f39d397
CM
2887 unsigned long bi;
2888 struct extent_buffer *leaf;
9078a3e1 2889
9078a3e1 2890 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
54aa1f4d
CM
2891 if (ret < 0)
2892 goto fail;
79787eaa 2893 BUG_ON(ret); /* Corruption */
5f39d397
CM
2894
2895 leaf = path->nodes[0];
2896 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2897 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2898 btrfs_mark_buffer_dirty(leaf);
b3b4aa74 2899 btrfs_release_path(path);
54aa1f4d 2900fail:
79787eaa
JM
2901 if (ret) {
2902 btrfs_abort_transaction(trans, root, ret);
9078a3e1 2903 return ret;
79787eaa 2904 }
9078a3e1
CM
2905 return 0;
2906
2907}
2908
4a8c9a62
YZ
2909static struct btrfs_block_group_cache *
2910next_block_group(struct btrfs_root *root,
2911 struct btrfs_block_group_cache *cache)
2912{
2913 struct rb_node *node;
2914 spin_lock(&root->fs_info->block_group_cache_lock);
2915 node = rb_next(&cache->cache_node);
2916 btrfs_put_block_group(cache);
2917 if (node) {
2918 cache = rb_entry(node, struct btrfs_block_group_cache,
2919 cache_node);
11dfe35a 2920 btrfs_get_block_group(cache);
4a8c9a62
YZ
2921 } else
2922 cache = NULL;
2923 spin_unlock(&root->fs_info->block_group_cache_lock);
2924 return cache;
2925}
2926
0af3d00b
JB
2927static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2928 struct btrfs_trans_handle *trans,
2929 struct btrfs_path *path)
2930{
2931 struct btrfs_root *root = block_group->fs_info->tree_root;
2932 struct inode *inode = NULL;
2933 u64 alloc_hint = 0;
2b20982e 2934 int dcs = BTRFS_DC_ERROR;
0af3d00b
JB
2935 int num_pages = 0;
2936 int retries = 0;
2937 int ret = 0;
2938
2939 /*
2940 * If this block group is smaller than 100 megs don't bother caching the
2941 * block group.
2942 */
2943 if (block_group->key.offset < (100 * 1024 * 1024)) {
2944 spin_lock(&block_group->lock);
2945 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2946 spin_unlock(&block_group->lock);
2947 return 0;
2948 }
2949
2950again:
2951 inode = lookup_free_space_inode(root, block_group, path);
2952 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2953 ret = PTR_ERR(inode);
b3b4aa74 2954 btrfs_release_path(path);
0af3d00b
JB
2955 goto out;
2956 }
2957
2958 if (IS_ERR(inode)) {
2959 BUG_ON(retries);
2960 retries++;
2961
2962 if (block_group->ro)
2963 goto out_free;
2964
2965 ret = create_free_space_inode(root, trans, block_group, path);
2966 if (ret)
2967 goto out_free;
2968 goto again;
2969 }
2970
5b0e95bf
JB
2971 /* We've already setup this transaction, go ahead and exit */
2972 if (block_group->cache_generation == trans->transid &&
2973 i_size_read(inode)) {
2974 dcs = BTRFS_DC_SETUP;
2975 goto out_put;
2976 }
2977
0af3d00b
JB
2978 /*
2979 * We want to set the generation to 0, that way if anything goes wrong
2980 * from here on out we know not to trust this cache when we load up next
2981 * time.
2982 */
2983 BTRFS_I(inode)->generation = 0;
2984 ret = btrfs_update_inode(trans, root, inode);
2985 WARN_ON(ret);
2986
2987 if (i_size_read(inode) > 0) {
2988 ret = btrfs_truncate_free_space_cache(root, trans, path,
2989 inode);
2990 if (ret)
2991 goto out_put;
2992 }
2993
2994 spin_lock(&block_group->lock);
2995 if (block_group->cached != BTRFS_CACHE_FINISHED) {
2b20982e
JB
2996 /* We're not cached, don't bother trying to write stuff out */
2997 dcs = BTRFS_DC_WRITTEN;
0af3d00b
JB
2998 spin_unlock(&block_group->lock);
2999 goto out_put;
3000 }
3001 spin_unlock(&block_group->lock);
3002
3003 num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
3004 if (!num_pages)
3005 num_pages = 1;
3006
3007 /*
3008 * Just to make absolutely sure we have enough space, we're going to
3009 * preallocate 12 pages worth of space for each block group. In
3010 * practice we ought to use at most 8, but we need extra space so we can
3011 * add our header and have a terminator between the extents and the
3012 * bitmaps.
3013 */
3014 num_pages *= 16;
3015 num_pages *= PAGE_CACHE_SIZE;
3016
3017 ret = btrfs_check_data_free_space(inode, num_pages);
3018 if (ret)
3019 goto out_put;
3020
3021 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3022 num_pages, num_pages,
3023 &alloc_hint);
2b20982e
JB
3024 if (!ret)
3025 dcs = BTRFS_DC_SETUP;
0af3d00b 3026 btrfs_free_reserved_data_space(inode, num_pages);
c09544e0 3027
0af3d00b
JB
3028out_put:
3029 iput(inode);
3030out_free:
b3b4aa74 3031 btrfs_release_path(path);
0af3d00b
JB
3032out:
3033 spin_lock(&block_group->lock);
e65cbb94 3034 if (!ret && dcs == BTRFS_DC_SETUP)
5b0e95bf 3035 block_group->cache_generation = trans->transid;
2b20982e 3036 block_group->disk_cache_state = dcs;
0af3d00b
JB
3037 spin_unlock(&block_group->lock);
3038
3039 return ret;
3040}
3041
96b5179d
CM
3042int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3043 struct btrfs_root *root)
9078a3e1 3044{
4a8c9a62 3045 struct btrfs_block_group_cache *cache;
9078a3e1 3046 int err = 0;
9078a3e1 3047 struct btrfs_path *path;
96b5179d 3048 u64 last = 0;
9078a3e1
CM
3049
3050 path = btrfs_alloc_path();
3051 if (!path)
3052 return -ENOMEM;
3053
0af3d00b
JB
3054again:
3055 while (1) {
3056 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3057 while (cache) {
3058 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3059 break;
3060 cache = next_block_group(root, cache);
3061 }
3062 if (!cache) {
3063 if (last == 0)
3064 break;
3065 last = 0;
3066 continue;
3067 }
3068 err = cache_save_setup(cache, trans, path);
3069 last = cache->key.objectid + cache->key.offset;
3070 btrfs_put_block_group(cache);
3071 }
3072
d397712b 3073 while (1) {
4a8c9a62
YZ
3074 if (last == 0) {
3075 err = btrfs_run_delayed_refs(trans, root,
3076 (unsigned long)-1);
79787eaa
JM
3077 if (err) /* File system offline */
3078 goto out;
0f9dd46c 3079 }
54aa1f4d 3080
4a8c9a62
YZ
3081 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3082 while (cache) {
0af3d00b
JB
3083 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3084 btrfs_put_block_group(cache);
3085 goto again;
3086 }
3087
4a8c9a62
YZ
3088 if (cache->dirty)
3089 break;
3090 cache = next_block_group(root, cache);
3091 }
3092 if (!cache) {
3093 if (last == 0)
3094 break;
3095 last = 0;
3096 continue;
3097 }
0f9dd46c 3098
0cb59c99
JB
3099 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3100 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
e8569813 3101 cache->dirty = 0;
4a8c9a62 3102 last = cache->key.objectid + cache->key.offset;
0f9dd46c 3103
4a8c9a62 3104 err = write_one_cache_group(trans, root, path, cache);
79787eaa
JM
3105 if (err) /* File system offline */
3106 goto out;
3107
4a8c9a62 3108 btrfs_put_block_group(cache);
9078a3e1 3109 }
4a8c9a62 3110
0cb59c99
JB
3111 while (1) {
3112 /*
3113 * I don't think this is needed since we're just marking our
3114 * preallocated extent as written, but just in case it can't
3115 * hurt.
3116 */
3117 if (last == 0) {
3118 err = btrfs_run_delayed_refs(trans, root,
3119 (unsigned long)-1);
79787eaa
JM
3120 if (err) /* File system offline */
3121 goto out;
0cb59c99
JB
3122 }
3123
3124 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3125 while (cache) {
3126 /*
3127 * Really this shouldn't happen, but it could if we
3128 * couldn't write the entire preallocated extent and
3129 * splitting the extent resulted in a new block.
3130 */
3131 if (cache->dirty) {
3132 btrfs_put_block_group(cache);
3133 goto again;
3134 }
3135 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3136 break;
3137 cache = next_block_group(root, cache);
3138 }
3139 if (!cache) {
3140 if (last == 0)
3141 break;
3142 last = 0;
3143 continue;
3144 }
3145
79787eaa 3146 err = btrfs_write_out_cache(root, trans, cache, path);
0cb59c99
JB
3147
3148 /*
3149 * If we didn't have an error then the cache state is still
3150 * NEED_WRITE, so we can set it to WRITTEN.
3151 */
79787eaa 3152 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
0cb59c99
JB
3153 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3154 last = cache->key.objectid + cache->key.offset;
3155 btrfs_put_block_group(cache);
3156 }
79787eaa 3157out:
0cb59c99 3158
9078a3e1 3159 btrfs_free_path(path);
79787eaa 3160 return err;
9078a3e1
CM
3161}
3162
d2fb3437
YZ
3163int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3164{
3165 struct btrfs_block_group_cache *block_group;
3166 int readonly = 0;
3167
3168 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3169 if (!block_group || block_group->ro)
3170 readonly = 1;
3171 if (block_group)
fa9c0d79 3172 btrfs_put_block_group(block_group);
d2fb3437
YZ
3173 return readonly;
3174}
3175
593060d7
CM
3176static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3177 u64 total_bytes, u64 bytes_used,
3178 struct btrfs_space_info **space_info)
3179{
3180 struct btrfs_space_info *found;
b742bb82
YZ
3181 int i;
3182 int factor;
3183
3184 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3185 BTRFS_BLOCK_GROUP_RAID10))
3186 factor = 2;
3187 else
3188 factor = 1;
593060d7
CM
3189
3190 found = __find_space_info(info, flags);
3191 if (found) {
25179201 3192 spin_lock(&found->lock);
593060d7 3193 found->total_bytes += total_bytes;
89a55897 3194 found->disk_total += total_bytes * factor;
593060d7 3195 found->bytes_used += bytes_used;
b742bb82 3196 found->disk_used += bytes_used * factor;
8f18cf13 3197 found->full = 0;
25179201 3198 spin_unlock(&found->lock);
593060d7
CM
3199 *space_info = found;
3200 return 0;
3201 }
c146afad 3202 found = kzalloc(sizeof(*found), GFP_NOFS);
593060d7
CM
3203 if (!found)
3204 return -ENOMEM;
3205
b742bb82
YZ
3206 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3207 INIT_LIST_HEAD(&found->block_groups[i]);
80eb234a 3208 init_rwsem(&found->groups_sem);
0f9dd46c 3209 spin_lock_init(&found->lock);
52ba6929 3210 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
593060d7 3211 found->total_bytes = total_bytes;
89a55897 3212 found->disk_total = total_bytes * factor;
593060d7 3213 found->bytes_used = bytes_used;
b742bb82 3214 found->disk_used = bytes_used * factor;
593060d7 3215 found->bytes_pinned = 0;
e8569813 3216 found->bytes_reserved = 0;
c146afad 3217 found->bytes_readonly = 0;
f0486c68 3218 found->bytes_may_use = 0;
593060d7 3219 found->full = 0;
0e4f8f88 3220 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
6d74119f 3221 found->chunk_alloc = 0;
fdb5effd
JB
3222 found->flush = 0;
3223 init_waitqueue_head(&found->wait);
593060d7 3224 *space_info = found;
4184ea7f 3225 list_add_rcu(&found->list, &info->space_info);
593060d7
CM
3226 return 0;
3227}
3228
8790d502
CM
3229static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3230{
899c81ea
ID
3231 u64 extra_flags = chunk_to_extended(flags) &
3232 BTRFS_EXTENDED_PROFILE_MASK;
a46d11a8
ID
3233
3234 if (flags & BTRFS_BLOCK_GROUP_DATA)
3235 fs_info->avail_data_alloc_bits |= extra_flags;
3236 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3237 fs_info->avail_metadata_alloc_bits |= extra_flags;
3238 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3239 fs_info->avail_system_alloc_bits |= extra_flags;
8790d502 3240}
593060d7 3241
fc67c450
ID
3242/*
3243 * returns target flags in extended format or 0 if restripe for this
3244 * chunk_type is not in progress
c6664b42
ID
3245 *
3246 * should be called with either volume_mutex or balance_lock held
fc67c450
ID
3247 */
3248static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3249{
3250 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3251 u64 target = 0;
3252
fc67c450
ID
3253 if (!bctl)
3254 return 0;
3255
3256 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3257 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3258 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3259 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3260 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3261 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3262 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3263 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3264 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3265 }
3266
3267 return target;
3268}
3269
a46d11a8
ID
3270/*
3271 * @flags: available profiles in extended format (see ctree.h)
3272 *
e4d8ec0f
ID
3273 * Returns reduced profile in chunk format. If profile changing is in
3274 * progress (either running or paused) picks the target profile (if it's
3275 * already available), otherwise falls back to plain reducing.
a46d11a8 3276 */
2b82032c 3277u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
ec44a35c 3278{
cd02dca5
CM
3279 /*
3280 * we add in the count of missing devices because we want
3281 * to make sure that any RAID levels on a degraded FS
3282 * continue to be honored.
3283 */
3284 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3285 root->fs_info->fs_devices->missing_devices;
fc67c450 3286 u64 target;
a061fc8d 3287
fc67c450
ID
3288 /*
3289 * see if restripe for this chunk_type is in progress, if so
3290 * try to reduce to the target profile
3291 */
e4d8ec0f 3292 spin_lock(&root->fs_info->balance_lock);
fc67c450
ID
3293 target = get_restripe_target(root->fs_info, flags);
3294 if (target) {
3295 /* pick target profile only if it's already available */
3296 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
e4d8ec0f 3297 spin_unlock(&root->fs_info->balance_lock);
fc67c450 3298 return extended_to_chunk(target);
e4d8ec0f
ID
3299 }
3300 }
3301 spin_unlock(&root->fs_info->balance_lock);
3302
a061fc8d
CM
3303 if (num_devices == 1)
3304 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3305 if (num_devices < 4)
3306 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3307
ec44a35c
CM
3308 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3309 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
a061fc8d 3310 BTRFS_BLOCK_GROUP_RAID10))) {
ec44a35c 3311 flags &= ~BTRFS_BLOCK_GROUP_DUP;
a061fc8d 3312 }
ec44a35c
CM
3313
3314 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
a061fc8d 3315 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
ec44a35c 3316 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
a061fc8d 3317 }
ec44a35c
CM
3318
3319 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3320 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3321 (flags & BTRFS_BLOCK_GROUP_RAID10) |
a46d11a8 3322 (flags & BTRFS_BLOCK_GROUP_DUP))) {
ec44a35c 3323 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
a46d11a8
ID
3324 }
3325
899c81ea 3326 return extended_to_chunk(flags);
ec44a35c
CM
3327}
3328
b742bb82 3329static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
6a63209f 3330{
b742bb82 3331 if (flags & BTRFS_BLOCK_GROUP_DATA)
6fef8df1 3332 flags |= root->fs_info->avail_data_alloc_bits;
b742bb82 3333 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
6fef8df1 3334 flags |= root->fs_info->avail_system_alloc_bits;
b742bb82 3335 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
6fef8df1
ID
3336 flags |= root->fs_info->avail_metadata_alloc_bits;
3337
b742bb82 3338 return btrfs_reduce_alloc_profile(root, flags);
6a63209f
JB
3339}
3340
6d07bcec 3341u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
9ed74f2d 3342{
b742bb82 3343 u64 flags;
9ed74f2d 3344
b742bb82
YZ
3345 if (data)
3346 flags = BTRFS_BLOCK_GROUP_DATA;
3347 else if (root == root->fs_info->chunk_root)
3348 flags = BTRFS_BLOCK_GROUP_SYSTEM;
9ed74f2d 3349 else
b742bb82 3350 flags = BTRFS_BLOCK_GROUP_METADATA;
9ed74f2d 3351
b742bb82 3352 return get_alloc_profile(root, flags);
6a63209f 3353}
9ed74f2d 3354
6a63209f
JB
3355void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3356{
6a63209f 3357 BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
f0486c68 3358 BTRFS_BLOCK_GROUP_DATA);
9ed74f2d
JB
3359}
3360
6a63209f 3361/*
6a63209f
JB
3362 * This will check the space that the inode allocates from to make sure we have
3363 * enough space for bytes.
6a63209f 3364 */
0ca1f7ce 3365int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
6a63209f 3366{
6a63209f 3367 struct btrfs_space_info *data_sinfo;
0ca1f7ce 3368 struct btrfs_root *root = BTRFS_I(inode)->root;
ab6e2410 3369 u64 used;
0af3d00b 3370 int ret = 0, committed = 0, alloc_chunk = 1;
6a63209f 3371
6a63209f
JB
3372 /* make sure bytes are sectorsize aligned */
3373 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
6a63209f 3374
82d5902d
LZ
3375 if (root == root->fs_info->tree_root ||
3376 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
0af3d00b
JB
3377 alloc_chunk = 0;
3378 committed = 1;
3379 }
3380
6a63209f 3381 data_sinfo = BTRFS_I(inode)->space_info;
33b4d47f
CM
3382 if (!data_sinfo)
3383 goto alloc;
9ed74f2d 3384
6a63209f
JB
3385again:
3386 /* make sure we have enough space to handle the data first */
3387 spin_lock(&data_sinfo->lock);
8929ecfa
YZ
3388 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3389 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3390 data_sinfo->bytes_may_use;
ab6e2410
JB
3391
3392 if (used + bytes > data_sinfo->total_bytes) {
4e06bdd6 3393 struct btrfs_trans_handle *trans;
9ed74f2d 3394
6a63209f
JB
3395 /*
3396 * if we don't have enough free bytes in this space then we need
3397 * to alloc a new chunk.
3398 */
0af3d00b 3399 if (!data_sinfo->full && alloc_chunk) {
6a63209f 3400 u64 alloc_target;
9ed74f2d 3401
0e4f8f88 3402 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
6a63209f 3403 spin_unlock(&data_sinfo->lock);
33b4d47f 3404alloc:
6a63209f 3405 alloc_target = btrfs_get_alloc_profile(root, 1);
7a7eaa40 3406 trans = btrfs_join_transaction(root);
a22285a6
YZ
3407 if (IS_ERR(trans))
3408 return PTR_ERR(trans);
9ed74f2d 3409
6a63209f
JB
3410 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3411 bytes + 2 * 1024 * 1024,
0e4f8f88
CM
3412 alloc_target,
3413 CHUNK_ALLOC_NO_FORCE);
6a63209f 3414 btrfs_end_transaction(trans, root);
d52a5b5f
MX
3415 if (ret < 0) {
3416 if (ret != -ENOSPC)
3417 return ret;
3418 else
3419 goto commit_trans;
3420 }
9ed74f2d 3421
33b4d47f
CM
3422 if (!data_sinfo) {
3423 btrfs_set_inode_space_info(root, inode);
3424 data_sinfo = BTRFS_I(inode)->space_info;
3425 }
6a63209f
JB
3426 goto again;
3427 }
f2bb8f5c
JB
3428
3429 /*
3430 * If we have less pinned bytes than we want to allocate then
3431 * don't bother committing the transaction, it won't help us.
3432 */
3433 if (data_sinfo->bytes_pinned < bytes)
3434 committed = 1;
6a63209f 3435 spin_unlock(&data_sinfo->lock);
6a63209f 3436
4e06bdd6 3437 /* commit the current transaction and try again */
d52a5b5f 3438commit_trans:
a4abeea4
JB
3439 if (!committed &&
3440 !atomic_read(&root->fs_info->open_ioctl_trans)) {
4e06bdd6 3441 committed = 1;
7a7eaa40 3442 trans = btrfs_join_transaction(root);
a22285a6
YZ
3443 if (IS_ERR(trans))
3444 return PTR_ERR(trans);
4e06bdd6
JB
3445 ret = btrfs_commit_transaction(trans, root);
3446 if (ret)
3447 return ret;
3448 goto again;
3449 }
9ed74f2d 3450
6a63209f
JB
3451 return -ENOSPC;
3452 }
3453 data_sinfo->bytes_may_use += bytes;
8c2a3ca2 3454 trace_btrfs_space_reservation(root->fs_info, "space_info",
2bcc0328 3455 data_sinfo->flags, bytes, 1);
6a63209f 3456 spin_unlock(&data_sinfo->lock);
6a63209f 3457
9ed74f2d 3458 return 0;
9ed74f2d 3459}
6a63209f 3460
6a63209f 3461/*
fb25e914 3462 * Called if we need to clear a data reservation for this inode.
6a63209f 3463 */
0ca1f7ce 3464void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
e3ccfa98 3465{
0ca1f7ce 3466 struct btrfs_root *root = BTRFS_I(inode)->root;
6a63209f 3467 struct btrfs_space_info *data_sinfo;
e3ccfa98 3468
6a63209f
JB
3469 /* make sure bytes are sectorsize aligned */
3470 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
e3ccfa98 3471
6a63209f
JB
3472 data_sinfo = BTRFS_I(inode)->space_info;
3473 spin_lock(&data_sinfo->lock);
3474 data_sinfo->bytes_may_use -= bytes;
8c2a3ca2 3475 trace_btrfs_space_reservation(root->fs_info, "space_info",
2bcc0328 3476 data_sinfo->flags, bytes, 0);
6a63209f 3477 spin_unlock(&data_sinfo->lock);
e3ccfa98
JB
3478}
3479
97e728d4 3480static void force_metadata_allocation(struct btrfs_fs_info *info)
e3ccfa98 3481{
97e728d4
JB
3482 struct list_head *head = &info->space_info;
3483 struct btrfs_space_info *found;
e3ccfa98 3484
97e728d4
JB
3485 rcu_read_lock();
3486 list_for_each_entry_rcu(found, head, list) {
3487 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
0e4f8f88 3488 found->force_alloc = CHUNK_ALLOC_FORCE;
e3ccfa98 3489 }
97e728d4 3490 rcu_read_unlock();
e3ccfa98
JB
3491}
3492
e5bc2458 3493static int should_alloc_chunk(struct btrfs_root *root,
0e4f8f88
CM
3494 struct btrfs_space_info *sinfo, u64 alloc_bytes,
3495 int force)
32c00aff 3496{
fb25e914 3497 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
424499db 3498 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
0e4f8f88 3499 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
e5bc2458 3500 u64 thresh;
e3ccfa98 3501
0e4f8f88
CM
3502 if (force == CHUNK_ALLOC_FORCE)
3503 return 1;
3504
fb25e914
JB
3505 /*
3506 * We need to take into account the global rsv because for all intents
3507 * and purposes it's used space. Don't worry about locking the
3508 * global_rsv, it doesn't change except when the transaction commits.
3509 */
3510 num_allocated += global_rsv->size;
3511
0e4f8f88
CM
3512 /*
3513 * in limited mode, we want to have some free space up to
3514 * about 1% of the FS size.
3515 */
3516 if (force == CHUNK_ALLOC_LIMITED) {
6c41761f 3517 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
0e4f8f88
CM
3518 thresh = max_t(u64, 64 * 1024 * 1024,
3519 div_factor_fine(thresh, 1));
3520
3521 if (num_bytes - num_allocated < thresh)
3522 return 1;
3523 }
6c41761f 3524 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
0e4f8f88 3525
cf1d72c9
CM
3526 /* 256MB or 2% of the FS */
3527 thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 2));
96bdc7dc
CM
3528 /* system chunks need a much small threshold */
3529 if (sinfo->flags & BTRFS_BLOCK_GROUP_SYSTEM)
3530 thresh = 32 * 1024 * 1024;
e5bc2458 3531
cf1d72c9 3532 if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 8))
14ed0ca6 3533 return 0;
424499db 3534 return 1;
32c00aff
JB
3535}
3536
15d1ff81
LB
3537static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3538{
3539 u64 num_dev;
3540
3541 if (type & BTRFS_BLOCK_GROUP_RAID10 ||
3542 type & BTRFS_BLOCK_GROUP_RAID0)
3543 num_dev = root->fs_info->fs_devices->rw_devices;
3544 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3545 num_dev = 2;
3546 else
3547 num_dev = 1; /* DUP or single */
3548
3549 /* metadata for updaing devices and chunk tree */
3550 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3551}
3552
3553static void check_system_chunk(struct btrfs_trans_handle *trans,
3554 struct btrfs_root *root, u64 type)
3555{
3556 struct btrfs_space_info *info;
3557 u64 left;
3558 u64 thresh;
3559
3560 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3561 spin_lock(&info->lock);
3562 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3563 info->bytes_reserved - info->bytes_readonly;
3564 spin_unlock(&info->lock);
3565
3566 thresh = get_system_chunk_thresh(root, type);
3567 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3568 printk(KERN_INFO "left=%llu, need=%llu, flags=%llu\n",
3569 left, thresh, type);
3570 dump_space_info(info, 0, 0);
3571 }
3572
3573 if (left < thresh) {
3574 u64 flags;
3575
3576 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3577 btrfs_alloc_chunk(trans, root, flags);
3578 }
3579}
3580
6324fbf3
CM
3581static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3582 struct btrfs_root *extent_root, u64 alloc_bytes,
0ef3e66b 3583 u64 flags, int force)
9ed74f2d 3584{
6324fbf3 3585 struct btrfs_space_info *space_info;
97e728d4 3586 struct btrfs_fs_info *fs_info = extent_root->fs_info;
6d74119f 3587 int wait_for_alloc = 0;
9ed74f2d 3588 int ret = 0;
9ed74f2d 3589
6324fbf3 3590 space_info = __find_space_info(extent_root->fs_info, flags);
593060d7
CM
3591 if (!space_info) {
3592 ret = update_space_info(extent_root->fs_info, flags,
3593 0, 0, &space_info);
79787eaa 3594 BUG_ON(ret); /* -ENOMEM */
9ed74f2d 3595 }
79787eaa 3596 BUG_ON(!space_info); /* Logic error */
9ed74f2d 3597
6d74119f 3598again:
25179201 3599 spin_lock(&space_info->lock);
9e622d6b 3600 if (force < space_info->force_alloc)
0e4f8f88 3601 force = space_info->force_alloc;
25179201
JB
3602 if (space_info->full) {
3603 spin_unlock(&space_info->lock);
6d74119f 3604 return 0;
9ed74f2d
JB
3605 }
3606
0e4f8f88 3607 if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
25179201 3608 spin_unlock(&space_info->lock);
6d74119f
JB
3609 return 0;
3610 } else if (space_info->chunk_alloc) {
3611 wait_for_alloc = 1;
3612 } else {
3613 space_info->chunk_alloc = 1;
9ed74f2d 3614 }
0e4f8f88 3615
25179201 3616 spin_unlock(&space_info->lock);
9ed74f2d 3617
6d74119f
JB
3618 mutex_lock(&fs_info->chunk_mutex);
3619
3620 /*
3621 * The chunk_mutex is held throughout the entirety of a chunk
3622 * allocation, so once we've acquired the chunk_mutex we know that the
3623 * other guy is done and we need to recheck and see if we should
3624 * allocate.
3625 */
3626 if (wait_for_alloc) {
3627 mutex_unlock(&fs_info->chunk_mutex);
3628 wait_for_alloc = 0;
3629 goto again;
3630 }
3631
67377734
JB
3632 /*
3633 * If we have mixed data/metadata chunks we want to make sure we keep
3634 * allocating mixed chunks instead of individual chunks.
3635 */
3636 if (btrfs_mixed_space_info(space_info))
3637 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3638
97e728d4
JB
3639 /*
3640 * if we're doing a data chunk, go ahead and make sure that
3641 * we keep a reasonable number of metadata chunks allocated in the
3642 * FS as well.
3643 */
9ed74f2d 3644 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
97e728d4
JB
3645 fs_info->data_chunk_allocations++;
3646 if (!(fs_info->data_chunk_allocations %
3647 fs_info->metadata_ratio))
3648 force_metadata_allocation(fs_info);
9ed74f2d
JB
3649 }
3650
15d1ff81
LB
3651 /*
3652 * Check if we have enough space in SYSTEM chunk because we may need
3653 * to update devices.
3654 */
3655 check_system_chunk(trans, extent_root, flags);
3656
2b82032c 3657 ret = btrfs_alloc_chunk(trans, extent_root, flags);
92b8e897
MF
3658 if (ret < 0 && ret != -ENOSPC)
3659 goto out;
3660
9ed74f2d 3661 spin_lock(&space_info->lock);
9ed74f2d 3662 if (ret)
6324fbf3 3663 space_info->full = 1;
424499db
YZ
3664 else
3665 ret = 1;
6d74119f 3666
0e4f8f88 3667 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
6d74119f 3668 space_info->chunk_alloc = 0;
9ed74f2d 3669 spin_unlock(&space_info->lock);
92b8e897 3670out:
a25c75d5 3671 mutex_unlock(&fs_info->chunk_mutex);
0f9dd46c 3672 return ret;
6324fbf3 3673}
9ed74f2d 3674
9ed74f2d 3675/*
5da9d01b 3676 * shrink metadata reservation for delalloc
9ed74f2d 3677 */
663350ac 3678static int shrink_delalloc(struct btrfs_root *root, u64 to_reclaim,
f104d044 3679 bool wait_ordered)
5da9d01b 3680{
0ca1f7ce 3681 struct btrfs_block_rsv *block_rsv;
0019f10d 3682 struct btrfs_space_info *space_info;
663350ac 3683 struct btrfs_trans_handle *trans;
5da9d01b
YZ
3684 u64 reserved;
3685 u64 max_reclaim;
3686 u64 reclaimed = 0;
b1953bce 3687 long time_left;
877da174 3688 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
b1953bce 3689 int loops = 0;
36e39c40 3690 unsigned long progress;
5da9d01b 3691
663350ac 3692 trans = (struct btrfs_trans_handle *)current->journal_info;
0ca1f7ce 3693 block_rsv = &root->fs_info->delalloc_block_rsv;
0019f10d 3694 space_info = block_rsv->space_info;
bf9022e0
CM
3695
3696 smp_mb();
fb25e914 3697 reserved = space_info->bytes_may_use;
36e39c40 3698 progress = space_info->reservation_progress;
5da9d01b
YZ
3699
3700 if (reserved == 0)
3701 return 0;
c4f675cd 3702
fdb5effd
JB
3703 smp_mb();
3704 if (root->fs_info->delalloc_bytes == 0) {
3705 if (trans)
3706 return 0;
3707 btrfs_wait_ordered_extents(root, 0, 0);
3708 return 0;
3709 }
3710
5da9d01b 3711 max_reclaim = min(reserved, to_reclaim);
877da174
JB
3712 nr_pages = max_t(unsigned long, nr_pages,
3713 max_reclaim >> PAGE_CACHE_SHIFT);
b1953bce 3714 while (loops < 1024) {
bf9022e0
CM
3715 /* have the flusher threads jump in and do some IO */
3716 smp_mb();
3717 nr_pages = min_t(unsigned long, nr_pages,
3718 root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
0e175a18
CW
3719 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages,
3720 WB_REASON_FS_FREE_SPACE);
5da9d01b 3721
0019f10d 3722 spin_lock(&space_info->lock);
fb25e914
JB
3723 if (reserved > space_info->bytes_may_use)
3724 reclaimed += reserved - space_info->bytes_may_use;
3725 reserved = space_info->bytes_may_use;
0019f10d 3726 spin_unlock(&space_info->lock);
5da9d01b 3727
36e39c40
CM
3728 loops++;
3729
5da9d01b
YZ
3730 if (reserved == 0 || reclaimed >= max_reclaim)
3731 break;
3732
3733 if (trans && trans->transaction->blocked)
3734 return -EAGAIN;
bf9022e0 3735
f104d044
JB
3736 if (wait_ordered && !trans) {
3737 btrfs_wait_ordered_extents(root, 0, 0);
3738 } else {
3739 time_left = schedule_timeout_interruptible(1);
b1953bce 3740
f104d044
JB
3741 /* We were interrupted, exit */
3742 if (time_left)
3743 break;
3744 }
b1953bce 3745
36e39c40
CM
3746 /* we've kicked the IO a few times, if anything has been freed,
3747 * exit. There is no sense in looping here for a long time
3748 * when we really need to commit the transaction, or there are
3749 * just too many writers without enough free space
3750 */
3751
3752 if (loops > 3) {
3753 smp_mb();
3754 if (progress != space_info->reservation_progress)
3755 break;
3756 }
bf9022e0 3757
5da9d01b 3758 }
f104d044 3759
5da9d01b
YZ
3760 return reclaimed >= to_reclaim;
3761}
3762
663350ac
JB
3763/**
3764 * maybe_commit_transaction - possibly commit the transaction if its ok to
3765 * @root - the root we're allocating for
3766 * @bytes - the number of bytes we want to reserve
3767 * @force - force the commit
8bb8ab2e 3768 *
663350ac
JB
3769 * This will check to make sure that committing the transaction will actually
3770 * get us somewhere and then commit the transaction if it does. Otherwise it
3771 * will return -ENOSPC.
8bb8ab2e 3772 */
663350ac
JB
3773static int may_commit_transaction(struct btrfs_root *root,
3774 struct btrfs_space_info *space_info,
3775 u64 bytes, int force)
3776{
3777 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
3778 struct btrfs_trans_handle *trans;
3779
3780 trans = (struct btrfs_trans_handle *)current->journal_info;
3781 if (trans)
3782 return -EAGAIN;
3783
3784 if (force)
3785 goto commit;
3786
3787 /* See if there is enough pinned space to make this reservation */
3788 spin_lock(&space_info->lock);
3789 if (space_info->bytes_pinned >= bytes) {
3790 spin_unlock(&space_info->lock);
3791 goto commit;
3792 }
3793 spin_unlock(&space_info->lock);
3794
3795 /*
3796 * See if there is some space in the delayed insertion reservation for
3797 * this reservation.
3798 */
3799 if (space_info != delayed_rsv->space_info)
3800 return -ENOSPC;
3801
d9b0218f 3802 spin_lock(&space_info->lock);
663350ac 3803 spin_lock(&delayed_rsv->lock);
d9b0218f 3804 if (space_info->bytes_pinned + delayed_rsv->size < bytes) {
663350ac 3805 spin_unlock(&delayed_rsv->lock);
d9b0218f 3806 spin_unlock(&space_info->lock);
663350ac
JB
3807 return -ENOSPC;
3808 }
3809 spin_unlock(&delayed_rsv->lock);
d9b0218f 3810 spin_unlock(&space_info->lock);
663350ac
JB
3811
3812commit:
3813 trans = btrfs_join_transaction(root);
3814 if (IS_ERR(trans))
3815 return -ENOSPC;
3816
3817 return btrfs_commit_transaction(trans, root);
3818}
3819
4a92b1b8
JB
3820/**
3821 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3822 * @root - the root we're allocating for
3823 * @block_rsv - the block_rsv we're allocating for
3824 * @orig_bytes - the number of bytes we want
3825 * @flush - wether or not we can flush to make our reservation
8bb8ab2e 3826 *
4a92b1b8
JB
3827 * This will reserve orgi_bytes number of bytes from the space info associated
3828 * with the block_rsv. If there is not enough space it will make an attempt to
3829 * flush out space to make room. It will do this by flushing delalloc if
3830 * possible or committing the transaction. If flush is 0 then no attempts to
3831 * regain reservations will be made and this will fail if there is not enough
3832 * space already.
8bb8ab2e 3833 */
4a92b1b8 3834static int reserve_metadata_bytes(struct btrfs_root *root,
8bb8ab2e
JB
3835 struct btrfs_block_rsv *block_rsv,
3836 u64 orig_bytes, int flush)
9ed74f2d 3837{
f0486c68 3838 struct btrfs_space_info *space_info = block_rsv->space_info;
2bf64758 3839 u64 used;
8bb8ab2e
JB
3840 u64 num_bytes = orig_bytes;
3841 int retries = 0;
3842 int ret = 0;
38227933 3843 bool committed = false;
fdb5effd 3844 bool flushing = false;
f104d044 3845 bool wait_ordered = false;
9ed74f2d 3846
8bb8ab2e 3847again:
fdb5effd 3848 ret = 0;
8bb8ab2e 3849 spin_lock(&space_info->lock);
fdb5effd
JB
3850 /*
3851 * We only want to wait if somebody other than us is flushing and we are
3852 * actually alloed to flush.
3853 */
3854 while (flush && !flushing && space_info->flush) {
3855 spin_unlock(&space_info->lock);
3856 /*
3857 * If we have a trans handle we can't wait because the flusher
3858 * may have to commit the transaction, which would mean we would
3859 * deadlock since we are waiting for the flusher to finish, but
3860 * hold the current transaction open.
3861 */
663350ac 3862 if (current->journal_info)
fdb5effd 3863 return -EAGAIN;
b9688bb8
AJ
3864 ret = wait_event_killable(space_info->wait, !space_info->flush);
3865 /* Must have been killed, return */
3866 if (ret)
fdb5effd
JB
3867 return -EINTR;
3868
3869 spin_lock(&space_info->lock);
3870 }
3871
3872 ret = -ENOSPC;
2bf64758
JB
3873 used = space_info->bytes_used + space_info->bytes_reserved +
3874 space_info->bytes_pinned + space_info->bytes_readonly +
3875 space_info->bytes_may_use;
9ed74f2d 3876
8bb8ab2e
JB
3877 /*
3878 * The idea here is that we've not already over-reserved the block group
3879 * then we can go ahead and save our reservation first and then start
3880 * flushing if we need to. Otherwise if we've already overcommitted
3881 * lets start flushing stuff first and then come back and try to make
3882 * our reservation.
3883 */
2bf64758
JB
3884 if (used <= space_info->total_bytes) {
3885 if (used + orig_bytes <= space_info->total_bytes) {
fb25e914 3886 space_info->bytes_may_use += orig_bytes;
8c2a3ca2 3887 trace_btrfs_space_reservation(root->fs_info,
2bcc0328 3888 "space_info", space_info->flags, orig_bytes, 1);
8bb8ab2e
JB
3889 ret = 0;
3890 } else {
3891 /*
3892 * Ok set num_bytes to orig_bytes since we aren't
3893 * overocmmitted, this way we only try and reclaim what
3894 * we need.
3895 */
3896 num_bytes = orig_bytes;
3897 }
3898 } else {
3899 /*
3900 * Ok we're over committed, set num_bytes to the overcommitted
3901 * amount plus the amount of bytes that we need for this
3902 * reservation.
3903 */
f104d044 3904 wait_ordered = true;
2bf64758 3905 num_bytes = used - space_info->total_bytes +
8bb8ab2e
JB
3906 (orig_bytes * (retries + 1));
3907 }
9ed74f2d 3908
36ba022a 3909 if (ret) {
2bf64758
JB
3910 u64 profile = btrfs_get_alloc_profile(root, 0);
3911 u64 avail;
3912
7e355b83
JB
3913 /*
3914 * If we have a lot of space that's pinned, don't bother doing
3915 * the overcommit dance yet and just commit the transaction.
3916 */
3917 avail = (space_info->total_bytes - space_info->bytes_used) * 8;
3918 do_div(avail, 10);
663350ac 3919 if (space_info->bytes_pinned >= avail && flush && !committed) {
7e355b83
JB
3920 space_info->flush = 1;
3921 flushing = true;
3922 spin_unlock(&space_info->lock);
663350ac
JB
3923 ret = may_commit_transaction(root, space_info,
3924 orig_bytes, 1);
3925 if (ret)
3926 goto out;
3927 committed = true;
3928 goto again;
7e355b83
JB
3929 }
3930
2bf64758
JB
3931 spin_lock(&root->fs_info->free_chunk_lock);
3932 avail = root->fs_info->free_chunk_space;
3933
3934 /*
3935 * If we have dup, raid1 or raid10 then only half of the free
3936 * space is actually useable.
3937 */
3938 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3939 BTRFS_BLOCK_GROUP_RAID1 |
3940 BTRFS_BLOCK_GROUP_RAID10))
3941 avail >>= 1;
3942
3943 /*
3944 * If we aren't flushing don't let us overcommit too much, say
3945 * 1/8th of the space. If we can flush, let it overcommit up to
3946 * 1/2 of the space.
3947 */
3948 if (flush)
3949 avail >>= 3;
3950 else
3951 avail >>= 1;
3952 spin_unlock(&root->fs_info->free_chunk_lock);
3953
9a82ca65 3954 if (used + num_bytes < space_info->total_bytes + avail) {
2bf64758 3955 space_info->bytes_may_use += orig_bytes;
8c2a3ca2 3956 trace_btrfs_space_reservation(root->fs_info,
2bcc0328 3957 "space_info", space_info->flags, orig_bytes, 1);
2bf64758 3958 ret = 0;
f104d044
JB
3959 } else {
3960 wait_ordered = true;
2bf64758
JB
3961 }
3962 }
3963
8bb8ab2e
JB
3964 /*
3965 * Couldn't make our reservation, save our place so while we're trying
3966 * to reclaim space we can actually use it instead of somebody else
3967 * stealing it from us.
3968 */
fdb5effd
JB
3969 if (ret && flush) {
3970 flushing = true;
3971 space_info->flush = 1;
8bb8ab2e 3972 }
9ed74f2d 3973
f0486c68 3974 spin_unlock(&space_info->lock);
9ed74f2d 3975
fdb5effd 3976 if (!ret || !flush)
8bb8ab2e 3977 goto out;
f0486c68 3978
8bb8ab2e
JB
3979 /*
3980 * We do synchronous shrinking since we don't actually unreserve
3981 * metadata until after the IO is completed.
3982 */
663350ac 3983 ret = shrink_delalloc(root, num_bytes, wait_ordered);
fdb5effd 3984 if (ret < 0)
8bb8ab2e 3985 goto out;
f0486c68 3986
75c195a2
CM
3987 ret = 0;
3988
8bb8ab2e
JB
3989 /*
3990 * So if we were overcommitted it's possible that somebody else flushed
3991 * out enough space and we simply didn't have enough space to reclaim,
3992 * so go back around and try again.
3993 */
3994 if (retries < 2) {
f104d044 3995 wait_ordered = true;
8bb8ab2e
JB
3996 retries++;
3997 goto again;
3998 }
f0486c68 3999
8bb8ab2e 4000 ret = -ENOSPC;
75c195a2
CM
4001 if (committed)
4002 goto out;
4003
663350ac 4004 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
38227933 4005 if (!ret) {
38227933 4006 committed = true;
8bb8ab2e 4007 goto again;
38227933 4008 }
8bb8ab2e
JB
4009
4010out:
fdb5effd 4011 if (flushing) {
8bb8ab2e 4012 spin_lock(&space_info->lock);
fdb5effd
JB
4013 space_info->flush = 0;
4014 wake_up_all(&space_info->wait);
8bb8ab2e 4015 spin_unlock(&space_info->lock);
f0486c68 4016 }
f0486c68
YZ
4017 return ret;
4018}
4019
79787eaa
JM
4020static struct btrfs_block_rsv *get_block_rsv(
4021 const struct btrfs_trans_handle *trans,
4022 const struct btrfs_root *root)
f0486c68 4023{
4c13d758
JB
4024 struct btrfs_block_rsv *block_rsv = NULL;
4025
4026 if (root->ref_cows || root == root->fs_info->csum_root)
f0486c68 4027 block_rsv = trans->block_rsv;
4c13d758
JB
4028
4029 if (!block_rsv)
f0486c68
YZ
4030 block_rsv = root->block_rsv;
4031
4032 if (!block_rsv)
4033 block_rsv = &root->fs_info->empty_block_rsv;
4034
4035 return block_rsv;
4036}
4037
4038static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4039 u64 num_bytes)
4040{
4041 int ret = -ENOSPC;
4042 spin_lock(&block_rsv->lock);
4043 if (block_rsv->reserved >= num_bytes) {
4044 block_rsv->reserved -= num_bytes;
4045 if (block_rsv->reserved < block_rsv->size)
4046 block_rsv->full = 0;
4047 ret = 0;
4048 }
4049 spin_unlock(&block_rsv->lock);
4050 return ret;
4051}
4052
4053static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4054 u64 num_bytes, int update_size)
4055{
4056 spin_lock(&block_rsv->lock);
4057 block_rsv->reserved += num_bytes;
4058 if (update_size)
4059 block_rsv->size += num_bytes;
4060 else if (block_rsv->reserved >= block_rsv->size)
4061 block_rsv->full = 1;
4062 spin_unlock(&block_rsv->lock);
4063}
4064
8c2a3ca2
JB
4065static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4066 struct btrfs_block_rsv *block_rsv,
62a45b60 4067 struct btrfs_block_rsv *dest, u64 num_bytes)
f0486c68
YZ
4068{
4069 struct btrfs_space_info *space_info = block_rsv->space_info;
4070
4071 spin_lock(&block_rsv->lock);
4072 if (num_bytes == (u64)-1)
4073 num_bytes = block_rsv->size;
4074 block_rsv->size -= num_bytes;
4075 if (block_rsv->reserved >= block_rsv->size) {
4076 num_bytes = block_rsv->reserved - block_rsv->size;
4077 block_rsv->reserved = block_rsv->size;
4078 block_rsv->full = 1;
4079 } else {
4080 num_bytes = 0;
4081 }
4082 spin_unlock(&block_rsv->lock);
4083
4084 if (num_bytes > 0) {
4085 if (dest) {
e9e22899
JB
4086 spin_lock(&dest->lock);
4087 if (!dest->full) {
4088 u64 bytes_to_add;
4089
4090 bytes_to_add = dest->size - dest->reserved;
4091 bytes_to_add = min(num_bytes, bytes_to_add);
4092 dest->reserved += bytes_to_add;
4093 if (dest->reserved >= dest->size)
4094 dest->full = 1;
4095 num_bytes -= bytes_to_add;
4096 }
4097 spin_unlock(&dest->lock);
4098 }
4099 if (num_bytes) {
f0486c68 4100 spin_lock(&space_info->lock);
fb25e914 4101 space_info->bytes_may_use -= num_bytes;
8c2a3ca2 4102 trace_btrfs_space_reservation(fs_info, "space_info",
2bcc0328 4103 space_info->flags, num_bytes, 0);
36e39c40 4104 space_info->reservation_progress++;
f0486c68 4105 spin_unlock(&space_info->lock);
4e06bdd6 4106 }
9ed74f2d 4107 }
f0486c68 4108}
4e06bdd6 4109
f0486c68
YZ
4110static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4111 struct btrfs_block_rsv *dst, u64 num_bytes)
4112{
4113 int ret;
9ed74f2d 4114
f0486c68
YZ
4115 ret = block_rsv_use_bytes(src, num_bytes);
4116 if (ret)
4117 return ret;
9ed74f2d 4118
f0486c68 4119 block_rsv_add_bytes(dst, num_bytes, 1);
9ed74f2d
JB
4120 return 0;
4121}
4122
f0486c68 4123void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
9ed74f2d 4124{
f0486c68
YZ
4125 memset(rsv, 0, sizeof(*rsv));
4126 spin_lock_init(&rsv->lock);
f0486c68
YZ
4127}
4128
4129struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
4130{
4131 struct btrfs_block_rsv *block_rsv;
4132 struct btrfs_fs_info *fs_info = root->fs_info;
9ed74f2d 4133
f0486c68
YZ
4134 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4135 if (!block_rsv)
4136 return NULL;
9ed74f2d 4137
f0486c68 4138 btrfs_init_block_rsv(block_rsv);
f0486c68
YZ
4139 block_rsv->space_info = __find_space_info(fs_info,
4140 BTRFS_BLOCK_GROUP_METADATA);
f0486c68
YZ
4141 return block_rsv;
4142}
9ed74f2d 4143
f0486c68
YZ
4144void btrfs_free_block_rsv(struct btrfs_root *root,
4145 struct btrfs_block_rsv *rsv)
4146{
dabdb640
JB
4147 btrfs_block_rsv_release(root, rsv, (u64)-1);
4148 kfree(rsv);
9ed74f2d
JB
4149}
4150
61b520a9
MX
4151static inline int __block_rsv_add(struct btrfs_root *root,
4152 struct btrfs_block_rsv *block_rsv,
4153 u64 num_bytes, int flush)
9ed74f2d 4154{
f0486c68 4155 int ret;
9ed74f2d 4156
f0486c68
YZ
4157 if (num_bytes == 0)
4158 return 0;
8bb8ab2e 4159
61b520a9 4160 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
f0486c68
YZ
4161 if (!ret) {
4162 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4163 return 0;
4164 }
9ed74f2d 4165
f0486c68 4166 return ret;
f0486c68 4167}
9ed74f2d 4168
61b520a9
MX
4169int btrfs_block_rsv_add(struct btrfs_root *root,
4170 struct btrfs_block_rsv *block_rsv,
4171 u64 num_bytes)
4172{
4173 return __block_rsv_add(root, block_rsv, num_bytes, 1);
4174}
4175
c06a0e12
JB
4176int btrfs_block_rsv_add_noflush(struct btrfs_root *root,
4177 struct btrfs_block_rsv *block_rsv,
4178 u64 num_bytes)
f0486c68 4179{
61b520a9 4180 return __block_rsv_add(root, block_rsv, num_bytes, 0);
f0486c68 4181}
9ed74f2d 4182
4a92b1b8 4183int btrfs_block_rsv_check(struct btrfs_root *root,
36ba022a 4184 struct btrfs_block_rsv *block_rsv, int min_factor)
f0486c68
YZ
4185{
4186 u64 num_bytes = 0;
f0486c68 4187 int ret = -ENOSPC;
9ed74f2d 4188
f0486c68
YZ
4189 if (!block_rsv)
4190 return 0;
9ed74f2d 4191
f0486c68 4192 spin_lock(&block_rsv->lock);
36ba022a
JB
4193 num_bytes = div_factor(block_rsv->size, min_factor);
4194 if (block_rsv->reserved >= num_bytes)
4195 ret = 0;
4196 spin_unlock(&block_rsv->lock);
9ed74f2d 4197
36ba022a
JB
4198 return ret;
4199}
4200
aa38a711
MX
4201static inline int __btrfs_block_rsv_refill(struct btrfs_root *root,
4202 struct btrfs_block_rsv *block_rsv,
4203 u64 min_reserved, int flush)
36ba022a
JB
4204{
4205 u64 num_bytes = 0;
4206 int ret = -ENOSPC;
4207
4208 if (!block_rsv)
4209 return 0;
4210
4211 spin_lock(&block_rsv->lock);
4212 num_bytes = min_reserved;
13553e52 4213 if (block_rsv->reserved >= num_bytes)
f0486c68 4214 ret = 0;
13553e52 4215 else
f0486c68 4216 num_bytes -= block_rsv->reserved;
f0486c68 4217 spin_unlock(&block_rsv->lock);
13553e52 4218
f0486c68
YZ
4219 if (!ret)
4220 return 0;
4221
aa38a711 4222 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
dabdb640
JB
4223 if (!ret) {
4224 block_rsv_add_bytes(block_rsv, num_bytes, 0);
f0486c68 4225 return 0;
6a63209f 4226 }
9ed74f2d 4227
13553e52 4228 return ret;
f0486c68
YZ
4229}
4230
aa38a711
MX
4231int btrfs_block_rsv_refill(struct btrfs_root *root,
4232 struct btrfs_block_rsv *block_rsv,
4233 u64 min_reserved)
4234{
4235 return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 1);
4236}
4237
4238int btrfs_block_rsv_refill_noflush(struct btrfs_root *root,
4239 struct btrfs_block_rsv *block_rsv,
4240 u64 min_reserved)
4241{
4242 return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 0);
4243}
4244
f0486c68
YZ
4245int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4246 struct btrfs_block_rsv *dst_rsv,
4247 u64 num_bytes)
4248{
4249 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4250}
4251
4252void btrfs_block_rsv_release(struct btrfs_root *root,
4253 struct btrfs_block_rsv *block_rsv,
4254 u64 num_bytes)
4255{
4256 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4257 if (global_rsv->full || global_rsv == block_rsv ||
4258 block_rsv->space_info != global_rsv->space_info)
4259 global_rsv = NULL;
8c2a3ca2
JB
4260 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4261 num_bytes);
6a63209f
JB
4262}
4263
4264/*
8929ecfa
YZ
4265 * helper to calculate size of global block reservation.
4266 * the desired value is sum of space used by extent tree,
4267 * checksum tree and root tree
6a63209f 4268 */
8929ecfa 4269static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
6a63209f 4270{
8929ecfa
YZ
4271 struct btrfs_space_info *sinfo;
4272 u64 num_bytes;
4273 u64 meta_used;
4274 u64 data_used;
6c41761f 4275 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
6a63209f 4276
8929ecfa
YZ
4277 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4278 spin_lock(&sinfo->lock);
4279 data_used = sinfo->bytes_used;
4280 spin_unlock(&sinfo->lock);
33b4d47f 4281
8929ecfa
YZ
4282 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4283 spin_lock(&sinfo->lock);
6d48755d
JB
4284 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4285 data_used = 0;
8929ecfa
YZ
4286 meta_used = sinfo->bytes_used;
4287 spin_unlock(&sinfo->lock);
ab6e2410 4288
8929ecfa
YZ
4289 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4290 csum_size * 2;
4291 num_bytes += div64_u64(data_used + meta_used, 50);
4e06bdd6 4292
8929ecfa 4293 if (num_bytes * 3 > meta_used)
8e62c2de 4294 num_bytes = div64_u64(meta_used, 3);
ab6e2410 4295
8929ecfa
YZ
4296 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4297}
6a63209f 4298
8929ecfa
YZ
4299static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4300{
4301 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4302 struct btrfs_space_info *sinfo = block_rsv->space_info;
4303 u64 num_bytes;
6a63209f 4304
8929ecfa 4305 num_bytes = calc_global_metadata_size(fs_info);
33b4d47f 4306
8929ecfa 4307 spin_lock(&sinfo->lock);
1f699d38 4308 spin_lock(&block_rsv->lock);
4e06bdd6 4309
8929ecfa 4310 block_rsv->size = num_bytes;
4e06bdd6 4311
8929ecfa 4312 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
6d48755d
JB
4313 sinfo->bytes_reserved + sinfo->bytes_readonly +
4314 sinfo->bytes_may_use;
8929ecfa
YZ
4315
4316 if (sinfo->total_bytes > num_bytes) {
4317 num_bytes = sinfo->total_bytes - num_bytes;
4318 block_rsv->reserved += num_bytes;
fb25e914 4319 sinfo->bytes_may_use += num_bytes;
8c2a3ca2 4320 trace_btrfs_space_reservation(fs_info, "space_info",
2bcc0328 4321 sinfo->flags, num_bytes, 1);
6a63209f 4322 }
6a63209f 4323
8929ecfa
YZ
4324 if (block_rsv->reserved >= block_rsv->size) {
4325 num_bytes = block_rsv->reserved - block_rsv->size;
fb25e914 4326 sinfo->bytes_may_use -= num_bytes;
8c2a3ca2 4327 trace_btrfs_space_reservation(fs_info, "space_info",
2bcc0328 4328 sinfo->flags, num_bytes, 0);
36e39c40 4329 sinfo->reservation_progress++;
8929ecfa
YZ
4330 block_rsv->reserved = block_rsv->size;
4331 block_rsv->full = 1;
4332 }
182608c8 4333
8929ecfa 4334 spin_unlock(&block_rsv->lock);
1f699d38 4335 spin_unlock(&sinfo->lock);
6a63209f
JB
4336}
4337
f0486c68 4338static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
6a63209f 4339{
f0486c68 4340 struct btrfs_space_info *space_info;
6a63209f 4341
f0486c68
YZ
4342 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4343 fs_info->chunk_block_rsv.space_info = space_info;
6a63209f 4344
f0486c68 4345 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
8929ecfa 4346 fs_info->global_block_rsv.space_info = space_info;
8929ecfa 4347 fs_info->delalloc_block_rsv.space_info = space_info;
f0486c68
YZ
4348 fs_info->trans_block_rsv.space_info = space_info;
4349 fs_info->empty_block_rsv.space_info = space_info;
6d668dda 4350 fs_info->delayed_block_rsv.space_info = space_info;
f0486c68 4351
8929ecfa
YZ
4352 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4353 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4354 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4355 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
f0486c68 4356 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
8929ecfa 4357
8929ecfa 4358 update_global_block_rsv(fs_info);
6a63209f
JB
4359}
4360
8929ecfa 4361static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
6a63209f 4362{
8c2a3ca2
JB
4363 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4364 (u64)-1);
8929ecfa
YZ
4365 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4366 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4367 WARN_ON(fs_info->trans_block_rsv.size > 0);
4368 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4369 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4370 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
6d668dda
JB
4371 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4372 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
fcb80c2a
JB
4373}
4374
a22285a6
YZ
4375void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4376 struct btrfs_root *root)
6a63209f 4377{
a22285a6
YZ
4378 if (!trans->bytes_reserved)
4379 return;
6a63209f 4380
e77266e4 4381 trace_btrfs_space_reservation(root->fs_info, "transaction",
2bcc0328 4382 trans->transid, trans->bytes_reserved, 0);
b24e03db 4383 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
a22285a6
YZ
4384 trans->bytes_reserved = 0;
4385}
6a63209f 4386
79787eaa 4387/* Can only return 0 or -ENOSPC */
d68fc57b
YZ
4388int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4389 struct inode *inode)
4390{
4391 struct btrfs_root *root = BTRFS_I(inode)->root;
4392 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4393 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4394
4395 /*
fcb80c2a
JB
4396 * We need to hold space in order to delete our orphan item once we've
4397 * added it, so this takes the reservation so we can release it later
4398 * when we are truly done with the orphan item.
d68fc57b 4399 */
ff5714cc 4400 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
8c2a3ca2
JB
4401 trace_btrfs_space_reservation(root->fs_info, "orphan",
4402 btrfs_ino(inode), num_bytes, 1);
d68fc57b 4403 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
6a63209f
JB
4404}
4405
d68fc57b 4406void btrfs_orphan_release_metadata(struct inode *inode)
97e728d4 4407{
d68fc57b 4408 struct btrfs_root *root = BTRFS_I(inode)->root;
ff5714cc 4409 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
8c2a3ca2
JB
4410 trace_btrfs_space_reservation(root->fs_info, "orphan",
4411 btrfs_ino(inode), num_bytes, 0);
d68fc57b
YZ
4412 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4413}
97e728d4 4414
a22285a6
YZ
4415int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4416 struct btrfs_pending_snapshot *pending)
4417{
4418 struct btrfs_root *root = pending->root;
4419 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4420 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4421 /*
4422 * two for root back/forward refs, two for directory entries
4423 * and one for root of the snapshot.
4424 */
16cdcec7 4425 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
a22285a6
YZ
4426 dst_rsv->space_info = src_rsv->space_info;
4427 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
97e728d4
JB
4428}
4429
7709cde3
JB
4430/**
4431 * drop_outstanding_extent - drop an outstanding extent
4432 * @inode: the inode we're dropping the extent for
4433 *
4434 * This is called when we are freeing up an outstanding extent, either called
4435 * after an error or after an extent is written. This will return the number of
4436 * reserved extents that need to be freed. This must be called with
4437 * BTRFS_I(inode)->lock held.
4438 */
9e0baf60
JB
4439static unsigned drop_outstanding_extent(struct inode *inode)
4440{
7fd2ae21 4441 unsigned drop_inode_space = 0;
9e0baf60
JB
4442 unsigned dropped_extents = 0;
4443
9e0baf60
JB
4444 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4445 BTRFS_I(inode)->outstanding_extents--;
4446
7fd2ae21 4447 if (BTRFS_I(inode)->outstanding_extents == 0 &&
72ac3c0d
JB
4448 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4449 &BTRFS_I(inode)->runtime_flags))
7fd2ae21 4450 drop_inode_space = 1;
7fd2ae21 4451
9e0baf60
JB
4452 /*
4453 * If we have more or the same amount of outsanding extents than we have
4454 * reserved then we need to leave the reserved extents count alone.
4455 */
4456 if (BTRFS_I(inode)->outstanding_extents >=
4457 BTRFS_I(inode)->reserved_extents)
7fd2ae21 4458 return drop_inode_space;
9e0baf60
JB
4459
4460 dropped_extents = BTRFS_I(inode)->reserved_extents -
4461 BTRFS_I(inode)->outstanding_extents;
4462 BTRFS_I(inode)->reserved_extents -= dropped_extents;
7fd2ae21 4463 return dropped_extents + drop_inode_space;
9e0baf60
JB
4464}
4465
7709cde3
JB
4466/**
4467 * calc_csum_metadata_size - return the amount of metada space that must be
4468 * reserved/free'd for the given bytes.
4469 * @inode: the inode we're manipulating
4470 * @num_bytes: the number of bytes in question
4471 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4472 *
4473 * This adjusts the number of csum_bytes in the inode and then returns the
4474 * correct amount of metadata that must either be reserved or freed. We
4475 * calculate how many checksums we can fit into one leaf and then divide the
4476 * number of bytes that will need to be checksumed by this value to figure out
4477 * how many checksums will be required. If we are adding bytes then the number
4478 * may go up and we will return the number of additional bytes that must be
4479 * reserved. If it is going down we will return the number of bytes that must
4480 * be freed.
4481 *
4482 * This must be called with BTRFS_I(inode)->lock held.
4483 */
4484static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4485 int reserve)
6324fbf3 4486{
7709cde3
JB
4487 struct btrfs_root *root = BTRFS_I(inode)->root;
4488 u64 csum_size;
4489 int num_csums_per_leaf;
4490 int num_csums;
4491 int old_csums;
4492
4493 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4494 BTRFS_I(inode)->csum_bytes == 0)
4495 return 0;
4496
4497 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4498 if (reserve)
4499 BTRFS_I(inode)->csum_bytes += num_bytes;
4500 else
4501 BTRFS_I(inode)->csum_bytes -= num_bytes;
4502 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4503 num_csums_per_leaf = (int)div64_u64(csum_size,
4504 sizeof(struct btrfs_csum_item) +
4505 sizeof(struct btrfs_disk_key));
4506 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4507 num_csums = num_csums + num_csums_per_leaf - 1;
4508 num_csums = num_csums / num_csums_per_leaf;
4509
4510 old_csums = old_csums + num_csums_per_leaf - 1;
4511 old_csums = old_csums / num_csums_per_leaf;
4512
4513 /* No change, no need to reserve more */
4514 if (old_csums == num_csums)
4515 return 0;
4516
4517 if (reserve)
4518 return btrfs_calc_trans_metadata_size(root,
4519 num_csums - old_csums);
4520
4521 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
0ca1f7ce 4522}
c146afad 4523
0ca1f7ce
YZ
4524int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4525{
4526 struct btrfs_root *root = BTRFS_I(inode)->root;
4527 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
9e0baf60 4528 u64 to_reserve = 0;
660d3f6c 4529 u64 csum_bytes;
9e0baf60 4530 unsigned nr_extents = 0;
660d3f6c 4531 int extra_reserve = 0;
c09544e0 4532 int flush = 1;
0ca1f7ce 4533 int ret;
6324fbf3 4534
660d3f6c 4535 /* Need to be holding the i_mutex here if we aren't free space cache */
c09544e0
JB
4536 if (btrfs_is_free_space_inode(root, inode))
4537 flush = 0;
4538
4539 if (flush && btrfs_transaction_in_commit(root->fs_info))
0ca1f7ce 4540 schedule_timeout(1);
ec44a35c 4541
f248679e 4542 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
0ca1f7ce 4543 num_bytes = ALIGN(num_bytes, root->sectorsize);
8bb8ab2e 4544
9e0baf60
JB
4545 spin_lock(&BTRFS_I(inode)->lock);
4546 BTRFS_I(inode)->outstanding_extents++;
4547
4548 if (BTRFS_I(inode)->outstanding_extents >
660d3f6c 4549 BTRFS_I(inode)->reserved_extents)
9e0baf60
JB
4550 nr_extents = BTRFS_I(inode)->outstanding_extents -
4551 BTRFS_I(inode)->reserved_extents;
57a45ced 4552
7fd2ae21
JB
4553 /*
4554 * Add an item to reserve for updating the inode when we complete the
4555 * delalloc io.
4556 */
72ac3c0d
JB
4557 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4558 &BTRFS_I(inode)->runtime_flags)) {
7fd2ae21 4559 nr_extents++;
660d3f6c 4560 extra_reserve = 1;
593060d7 4561 }
7fd2ae21
JB
4562
4563 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
7709cde3 4564 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
660d3f6c 4565 csum_bytes = BTRFS_I(inode)->csum_bytes;
9e0baf60 4566 spin_unlock(&BTRFS_I(inode)->lock);
57a45ced 4567
36ba022a 4568 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
9e0baf60 4569 if (ret) {
7ed49f18 4570 u64 to_free = 0;
9e0baf60 4571 unsigned dropped;
7ed49f18 4572
7709cde3 4573 spin_lock(&BTRFS_I(inode)->lock);
9e0baf60 4574 dropped = drop_outstanding_extent(inode);
9e0baf60 4575 /*
660d3f6c
JB
4576 * If the inodes csum_bytes is the same as the original
4577 * csum_bytes then we know we haven't raced with any free()ers
4578 * so we can just reduce our inodes csum bytes and carry on.
4579 * Otherwise we have to do the normal free thing to account for
4580 * the case that the free side didn't free up its reserve
4581 * because of this outstanding reservation.
9e0baf60 4582 */
660d3f6c
JB
4583 if (BTRFS_I(inode)->csum_bytes == csum_bytes)
4584 calc_csum_metadata_size(inode, num_bytes, 0);
4585 else
4586 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4587 spin_unlock(&BTRFS_I(inode)->lock);
4588 if (dropped)
4589 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4590
8c2a3ca2 4591 if (to_free) {
7ed49f18 4592 btrfs_block_rsv_release(root, block_rsv, to_free);
8c2a3ca2
JB
4593 trace_btrfs_space_reservation(root->fs_info,
4594 "delalloc",
4595 btrfs_ino(inode),
4596 to_free, 0);
4597 }
f248679e 4598 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
0ca1f7ce 4599 return ret;
9e0baf60 4600 }
25179201 4601
660d3f6c
JB
4602 spin_lock(&BTRFS_I(inode)->lock);
4603 if (extra_reserve) {
72ac3c0d
JB
4604 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4605 &BTRFS_I(inode)->runtime_flags);
660d3f6c
JB
4606 nr_extents--;
4607 }
4608 BTRFS_I(inode)->reserved_extents += nr_extents;
4609 spin_unlock(&BTRFS_I(inode)->lock);
f248679e 4610 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
660d3f6c 4611
8c2a3ca2
JB
4612 if (to_reserve)
4613 trace_btrfs_space_reservation(root->fs_info,"delalloc",
4614 btrfs_ino(inode), to_reserve, 1);
0ca1f7ce
YZ
4615 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4616
0ca1f7ce
YZ
4617 return 0;
4618}
4619
7709cde3
JB
4620/**
4621 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4622 * @inode: the inode to release the reservation for
4623 * @num_bytes: the number of bytes we're releasing
4624 *
4625 * This will release the metadata reservation for an inode. This can be called
4626 * once we complete IO for a given set of bytes to release their metadata
4627 * reservations.
4628 */
0ca1f7ce
YZ
4629void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4630{
4631 struct btrfs_root *root = BTRFS_I(inode)->root;
9e0baf60
JB
4632 u64 to_free = 0;
4633 unsigned dropped;
0ca1f7ce
YZ
4634
4635 num_bytes = ALIGN(num_bytes, root->sectorsize);
7709cde3 4636 spin_lock(&BTRFS_I(inode)->lock);
9e0baf60 4637 dropped = drop_outstanding_extent(inode);
97e728d4 4638
7709cde3
JB
4639 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4640 spin_unlock(&BTRFS_I(inode)->lock);
9e0baf60
JB
4641 if (dropped > 0)
4642 to_free += btrfs_calc_trans_metadata_size(root, dropped);
0ca1f7ce 4643
8c2a3ca2
JB
4644 trace_btrfs_space_reservation(root->fs_info, "delalloc",
4645 btrfs_ino(inode), to_free, 0);
0ca1f7ce
YZ
4646 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4647 to_free);
4648}
4649
7709cde3
JB
4650/**
4651 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4652 * @inode: inode we're writing to
4653 * @num_bytes: the number of bytes we want to allocate
4654 *
4655 * This will do the following things
4656 *
4657 * o reserve space in the data space info for num_bytes
4658 * o reserve space in the metadata space info based on number of outstanding
4659 * extents and how much csums will be needed
4660 * o add to the inodes ->delalloc_bytes
4661 * o add it to the fs_info's delalloc inodes list.
4662 *
4663 * This will return 0 for success and -ENOSPC if there is no space left.
4664 */
0ca1f7ce
YZ
4665int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4666{
4667 int ret;
4668
4669 ret = btrfs_check_data_free_space(inode, num_bytes);
d397712b 4670 if (ret)
0ca1f7ce
YZ
4671 return ret;
4672
4673 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4674 if (ret) {
4675 btrfs_free_reserved_data_space(inode, num_bytes);
4676 return ret;
4677 }
4678
4679 return 0;
4680}
4681
7709cde3
JB
4682/**
4683 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4684 * @inode: inode we're releasing space for
4685 * @num_bytes: the number of bytes we want to free up
4686 *
4687 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4688 * called in the case that we don't need the metadata AND data reservations
4689 * anymore. So if there is an error or we insert an inline extent.
4690 *
4691 * This function will release the metadata space that was not used and will
4692 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4693 * list if there are no delalloc bytes left.
4694 */
0ca1f7ce
YZ
4695void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4696{
4697 btrfs_delalloc_release_metadata(inode, num_bytes);
4698 btrfs_free_reserved_data_space(inode, num_bytes);
6324fbf3
CM
4699}
4700
9078a3e1
CM
4701static int update_block_group(struct btrfs_trans_handle *trans,
4702 struct btrfs_root *root,
f0486c68 4703 u64 bytenr, u64 num_bytes, int alloc)
9078a3e1 4704{
0af3d00b 4705 struct btrfs_block_group_cache *cache = NULL;
9078a3e1 4706 struct btrfs_fs_info *info = root->fs_info;
db94535d 4707 u64 total = num_bytes;
9078a3e1 4708 u64 old_val;
db94535d 4709 u64 byte_in_group;
0af3d00b 4710 int factor;
3e1ad54f 4711
5d4f98a2
YZ
4712 /* block accounting for super block */
4713 spin_lock(&info->delalloc_lock);
6c41761f 4714 old_val = btrfs_super_bytes_used(info->super_copy);
5d4f98a2
YZ
4715 if (alloc)
4716 old_val += num_bytes;
4717 else
4718 old_val -= num_bytes;
6c41761f 4719 btrfs_set_super_bytes_used(info->super_copy, old_val);
5d4f98a2
YZ
4720 spin_unlock(&info->delalloc_lock);
4721
d397712b 4722 while (total) {
db94535d 4723 cache = btrfs_lookup_block_group(info, bytenr);
f3465ca4 4724 if (!cache)
79787eaa 4725 return -ENOENT;
b742bb82
YZ
4726 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4727 BTRFS_BLOCK_GROUP_RAID1 |
4728 BTRFS_BLOCK_GROUP_RAID10))
4729 factor = 2;
4730 else
4731 factor = 1;
9d66e233
JB
4732 /*
4733 * If this block group has free space cache written out, we
4734 * need to make sure to load it if we are removing space. This
4735 * is because we need the unpinning stage to actually add the
4736 * space back to the block group, otherwise we will leak space.
4737 */
4738 if (!alloc && cache->cached == BTRFS_CACHE_NO)
b8399dee 4739 cache_block_group(cache, trans, NULL, 1);
0af3d00b 4740
db94535d
CM
4741 byte_in_group = bytenr - cache->key.objectid;
4742 WARN_ON(byte_in_group > cache->key.offset);
9078a3e1 4743
25179201 4744 spin_lock(&cache->space_info->lock);
c286ac48 4745 spin_lock(&cache->lock);
0af3d00b 4746
73bc1876 4747 if (btrfs_test_opt(root, SPACE_CACHE) &&
0af3d00b
JB
4748 cache->disk_cache_state < BTRFS_DC_CLEAR)
4749 cache->disk_cache_state = BTRFS_DC_CLEAR;
4750
0f9dd46c 4751 cache->dirty = 1;
9078a3e1 4752 old_val = btrfs_block_group_used(&cache->item);
db94535d 4753 num_bytes = min(total, cache->key.offset - byte_in_group);
cd1bc465 4754 if (alloc) {
db94535d 4755 old_val += num_bytes;
11833d66
YZ
4756 btrfs_set_block_group_used(&cache->item, old_val);
4757 cache->reserved -= num_bytes;
11833d66 4758 cache->space_info->bytes_reserved -= num_bytes;
b742bb82
YZ
4759 cache->space_info->bytes_used += num_bytes;
4760 cache->space_info->disk_used += num_bytes * factor;
c286ac48 4761 spin_unlock(&cache->lock);
25179201 4762 spin_unlock(&cache->space_info->lock);
cd1bc465 4763 } else {
db94535d 4764 old_val -= num_bytes;
c286ac48 4765 btrfs_set_block_group_used(&cache->item, old_val);
f0486c68
YZ
4766 cache->pinned += num_bytes;
4767 cache->space_info->bytes_pinned += num_bytes;
6324fbf3 4768 cache->space_info->bytes_used -= num_bytes;
b742bb82 4769 cache->space_info->disk_used -= num_bytes * factor;
c286ac48 4770 spin_unlock(&cache->lock);
25179201 4771 spin_unlock(&cache->space_info->lock);
1f3c79a2 4772
f0486c68
YZ
4773 set_extent_dirty(info->pinned_extents,
4774 bytenr, bytenr + num_bytes - 1,
4775 GFP_NOFS | __GFP_NOFAIL);
cd1bc465 4776 }
fa9c0d79 4777 btrfs_put_block_group(cache);
db94535d
CM
4778 total -= num_bytes;
4779 bytenr += num_bytes;
9078a3e1
CM
4780 }
4781 return 0;
4782}
6324fbf3 4783
a061fc8d
CM
4784static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4785{
0f9dd46c 4786 struct btrfs_block_group_cache *cache;
d2fb3437 4787 u64 bytenr;
0f9dd46c
JB
4788
4789 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4790 if (!cache)
a061fc8d 4791 return 0;
0f9dd46c 4792
d2fb3437 4793 bytenr = cache->key.objectid;
fa9c0d79 4794 btrfs_put_block_group(cache);
d2fb3437
YZ
4795
4796 return bytenr;
a061fc8d
CM
4797}
4798
f0486c68
YZ
4799static int pin_down_extent(struct btrfs_root *root,
4800 struct btrfs_block_group_cache *cache,
4801 u64 bytenr, u64 num_bytes, int reserved)
324ae4df 4802{
11833d66
YZ
4803 spin_lock(&cache->space_info->lock);
4804 spin_lock(&cache->lock);
4805 cache->pinned += num_bytes;
4806 cache->space_info->bytes_pinned += num_bytes;
4807 if (reserved) {
4808 cache->reserved -= num_bytes;
4809 cache->space_info->bytes_reserved -= num_bytes;
4810 }
4811 spin_unlock(&cache->lock);
4812 spin_unlock(&cache->space_info->lock);
68b38550 4813
f0486c68
YZ
4814 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4815 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4816 return 0;
4817}
68b38550 4818
f0486c68
YZ
4819/*
4820 * this function must be called within transaction
4821 */
4822int btrfs_pin_extent(struct btrfs_root *root,
4823 u64 bytenr, u64 num_bytes, int reserved)
4824{
4825 struct btrfs_block_group_cache *cache;
68b38550 4826
f0486c68 4827 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
79787eaa 4828 BUG_ON(!cache); /* Logic error */
f0486c68
YZ
4829
4830 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4831
4832 btrfs_put_block_group(cache);
11833d66
YZ
4833 return 0;
4834}
4835
f0486c68 4836/*
e688b725
CM
4837 * this function must be called within transaction
4838 */
4839int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
4840 struct btrfs_root *root,
4841 u64 bytenr, u64 num_bytes)
4842{
4843 struct btrfs_block_group_cache *cache;
4844
4845 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
79787eaa 4846 BUG_ON(!cache); /* Logic error */
e688b725
CM
4847
4848 /*
4849 * pull in the free space cache (if any) so that our pin
4850 * removes the free space from the cache. We have load_only set
4851 * to one because the slow code to read in the free extents does check
4852 * the pinned extents.
4853 */
4854 cache_block_group(cache, trans, root, 1);
4855
4856 pin_down_extent(root, cache, bytenr, num_bytes, 0);
4857
4858 /* remove us from the free space cache (if we're there at all) */
4859 btrfs_remove_free_space(cache, bytenr, num_bytes);
4860 btrfs_put_block_group(cache);
4861 return 0;
4862}
4863
fb25e914
JB
4864/**
4865 * btrfs_update_reserved_bytes - update the block_group and space info counters
4866 * @cache: The cache we are manipulating
4867 * @num_bytes: The number of bytes in question
4868 * @reserve: One of the reservation enums
4869 *
4870 * This is called by the allocator when it reserves space, or by somebody who is
4871 * freeing space that was never actually used on disk. For example if you
4872 * reserve some space for a new leaf in transaction A and before transaction A
4873 * commits you free that leaf, you call this with reserve set to 0 in order to
4874 * clear the reservation.
4875 *
4876 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4877 * ENOSPC accounting. For data we handle the reservation through clearing the
4878 * delalloc bits in the io_tree. We have to do this since we could end up
4879 * allocating less disk space for the amount of data we have reserved in the
4880 * case of compression.
4881 *
4882 * If this is a reservation and the block group has become read only we cannot
4883 * make the reservation and return -EAGAIN, otherwise this function always
4884 * succeeds.
f0486c68 4885 */
fb25e914
JB
4886static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4887 u64 num_bytes, int reserve)
11833d66 4888{
fb25e914 4889 struct btrfs_space_info *space_info = cache->space_info;
f0486c68 4890 int ret = 0;
79787eaa 4891
fb25e914
JB
4892 spin_lock(&space_info->lock);
4893 spin_lock(&cache->lock);
4894 if (reserve != RESERVE_FREE) {
f0486c68
YZ
4895 if (cache->ro) {
4896 ret = -EAGAIN;
4897 } else {
fb25e914
JB
4898 cache->reserved += num_bytes;
4899 space_info->bytes_reserved += num_bytes;
4900 if (reserve == RESERVE_ALLOC) {
8c2a3ca2 4901 trace_btrfs_space_reservation(cache->fs_info,
2bcc0328
LB
4902 "space_info", space_info->flags,
4903 num_bytes, 0);
fb25e914
JB
4904 space_info->bytes_may_use -= num_bytes;
4905 }
f0486c68 4906 }
fb25e914
JB
4907 } else {
4908 if (cache->ro)
4909 space_info->bytes_readonly += num_bytes;
4910 cache->reserved -= num_bytes;
4911 space_info->bytes_reserved -= num_bytes;
4912 space_info->reservation_progress++;
324ae4df 4913 }
fb25e914
JB
4914 spin_unlock(&cache->lock);
4915 spin_unlock(&space_info->lock);
f0486c68 4916 return ret;
324ae4df 4917}
9078a3e1 4918
143bede5 4919void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
11833d66 4920 struct btrfs_root *root)
e8569813 4921{
e8569813 4922 struct btrfs_fs_info *fs_info = root->fs_info;
11833d66
YZ
4923 struct btrfs_caching_control *next;
4924 struct btrfs_caching_control *caching_ctl;
4925 struct btrfs_block_group_cache *cache;
e8569813 4926
11833d66 4927 down_write(&fs_info->extent_commit_sem);
25179201 4928
11833d66
YZ
4929 list_for_each_entry_safe(caching_ctl, next,
4930 &fs_info->caching_block_groups, list) {
4931 cache = caching_ctl->block_group;
4932 if (block_group_cache_done(cache)) {
4933 cache->last_byte_to_unpin = (u64)-1;
4934 list_del_init(&caching_ctl->list);
4935 put_caching_control(caching_ctl);
e8569813 4936 } else {
11833d66 4937 cache->last_byte_to_unpin = caching_ctl->progress;
e8569813 4938 }
e8569813 4939 }
11833d66
YZ
4940
4941 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4942 fs_info->pinned_extents = &fs_info->freed_extents[1];
4943 else
4944 fs_info->pinned_extents = &fs_info->freed_extents[0];
4945
4946 up_write(&fs_info->extent_commit_sem);
8929ecfa
YZ
4947
4948 update_global_block_rsv(fs_info);
e8569813
ZY
4949}
4950
11833d66 4951static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
ccd467d6 4952{
11833d66
YZ
4953 struct btrfs_fs_info *fs_info = root->fs_info;
4954 struct btrfs_block_group_cache *cache = NULL;
4955 u64 len;
ccd467d6 4956
11833d66
YZ
4957 while (start <= end) {
4958 if (!cache ||
4959 start >= cache->key.objectid + cache->key.offset) {
4960 if (cache)
4961 btrfs_put_block_group(cache);
4962 cache = btrfs_lookup_block_group(fs_info, start);
79787eaa 4963 BUG_ON(!cache); /* Logic error */
11833d66
YZ
4964 }
4965
4966 len = cache->key.objectid + cache->key.offset - start;
4967 len = min(len, end + 1 - start);
4968
4969 if (start < cache->last_byte_to_unpin) {
4970 len = min(len, cache->last_byte_to_unpin - start);
4971 btrfs_add_free_space(cache, start, len);
4972 }
4973
f0486c68
YZ
4974 start += len;
4975
11833d66
YZ
4976 spin_lock(&cache->space_info->lock);
4977 spin_lock(&cache->lock);
4978 cache->pinned -= len;
4979 cache->space_info->bytes_pinned -= len;
37be25bc 4980 if (cache->ro)
f0486c68 4981 cache->space_info->bytes_readonly += len;
11833d66
YZ
4982 spin_unlock(&cache->lock);
4983 spin_unlock(&cache->space_info->lock);
ccd467d6 4984 }
11833d66
YZ
4985
4986 if (cache)
4987 btrfs_put_block_group(cache);
ccd467d6
CM
4988 return 0;
4989}
4990
4991int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
11833d66 4992 struct btrfs_root *root)
a28ec197 4993{
11833d66
YZ
4994 struct btrfs_fs_info *fs_info = root->fs_info;
4995 struct extent_io_tree *unpin;
1a5bc167
CM
4996 u64 start;
4997 u64 end;
a28ec197 4998 int ret;
a28ec197 4999
79787eaa
JM
5000 if (trans->aborted)
5001 return 0;
5002
11833d66
YZ
5003 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5004 unpin = &fs_info->freed_extents[1];
5005 else
5006 unpin = &fs_info->freed_extents[0];
5007
d397712b 5008 while (1) {
1a5bc167
CM
5009 ret = find_first_extent_bit(unpin, 0, &start, &end,
5010 EXTENT_DIRTY);
5011 if (ret)
a28ec197 5012 break;
1f3c79a2 5013
5378e607
LD
5014 if (btrfs_test_opt(root, DISCARD))
5015 ret = btrfs_discard_extent(root, start,
5016 end + 1 - start, NULL);
1f3c79a2 5017
1a5bc167 5018 clear_extent_dirty(unpin, start, end, GFP_NOFS);
11833d66 5019 unpin_extent_range(root, start, end);
b9473439 5020 cond_resched();
a28ec197 5021 }
817d52f8 5022
e20d96d6
CM
5023 return 0;
5024}
5025
5d4f98a2
YZ
5026static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5027 struct btrfs_root *root,
5028 u64 bytenr, u64 num_bytes, u64 parent,
5029 u64 root_objectid, u64 owner_objectid,
5030 u64 owner_offset, int refs_to_drop,
5031 struct btrfs_delayed_extent_op *extent_op)
a28ec197 5032{
e2fa7227 5033 struct btrfs_key key;
5d4f98a2 5034 struct btrfs_path *path;
1261ec42
CM
5035 struct btrfs_fs_info *info = root->fs_info;
5036 struct btrfs_root *extent_root = info->extent_root;
5f39d397 5037 struct extent_buffer *leaf;
5d4f98a2
YZ
5038 struct btrfs_extent_item *ei;
5039 struct btrfs_extent_inline_ref *iref;
a28ec197 5040 int ret;
5d4f98a2 5041 int is_data;
952fccac
CM
5042 int extent_slot = 0;
5043 int found_extent = 0;
5044 int num_to_del = 1;
5d4f98a2
YZ
5045 u32 item_size;
5046 u64 refs;
037e6390 5047
5caf2a00 5048 path = btrfs_alloc_path();
54aa1f4d
CM
5049 if (!path)
5050 return -ENOMEM;
5f26f772 5051
3c12ac72 5052 path->reada = 1;
b9473439 5053 path->leave_spinning = 1;
5d4f98a2
YZ
5054
5055 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5056 BUG_ON(!is_data && refs_to_drop != 1);
5057
5058 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5059 bytenr, num_bytes, parent,
5060 root_objectid, owner_objectid,
5061 owner_offset);
7bb86316 5062 if (ret == 0) {
952fccac 5063 extent_slot = path->slots[0];
5d4f98a2
YZ
5064 while (extent_slot >= 0) {
5065 btrfs_item_key_to_cpu(path->nodes[0], &key,
952fccac 5066 extent_slot);
5d4f98a2 5067 if (key.objectid != bytenr)
952fccac 5068 break;
5d4f98a2
YZ
5069 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5070 key.offset == num_bytes) {
952fccac
CM
5071 found_extent = 1;
5072 break;
5073 }
5074 if (path->slots[0] - extent_slot > 5)
5075 break;
5d4f98a2 5076 extent_slot--;
952fccac 5077 }
5d4f98a2
YZ
5078#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5079 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5080 if (found_extent && item_size < sizeof(*ei))
5081 found_extent = 0;
5082#endif
31840ae1 5083 if (!found_extent) {
5d4f98a2 5084 BUG_ON(iref);
56bec294 5085 ret = remove_extent_backref(trans, extent_root, path,
5d4f98a2
YZ
5086 NULL, refs_to_drop,
5087 is_data);
79787eaa
JM
5088 if (ret)
5089 goto abort;
b3b4aa74 5090 btrfs_release_path(path);
b9473439 5091 path->leave_spinning = 1;
5d4f98a2
YZ
5092
5093 key.objectid = bytenr;
5094 key.type = BTRFS_EXTENT_ITEM_KEY;
5095 key.offset = num_bytes;
5096
31840ae1
ZY
5097 ret = btrfs_search_slot(trans, extent_root,
5098 &key, path, -1, 1);
f3465ca4
JB
5099 if (ret) {
5100 printk(KERN_ERR "umm, got %d back from search"
d397712b
CM
5101 ", was looking for %llu\n", ret,
5102 (unsigned long long)bytenr);
b783e62d
JB
5103 if (ret > 0)
5104 btrfs_print_leaf(extent_root,
5105 path->nodes[0]);
f3465ca4 5106 }
79787eaa
JM
5107 if (ret < 0)
5108 goto abort;
31840ae1
ZY
5109 extent_slot = path->slots[0];
5110 }
79787eaa 5111 } else if (ret == -ENOENT) {
7bb86316
CM
5112 btrfs_print_leaf(extent_root, path->nodes[0]);
5113 WARN_ON(1);
d397712b 5114 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
5d4f98a2 5115 "parent %llu root %llu owner %llu offset %llu\n",
d397712b 5116 (unsigned long long)bytenr,
56bec294 5117 (unsigned long long)parent,
d397712b 5118 (unsigned long long)root_objectid,
5d4f98a2
YZ
5119 (unsigned long long)owner_objectid,
5120 (unsigned long long)owner_offset);
79787eaa
JM
5121 } else {
5122 goto abort;
7bb86316 5123 }
5f39d397
CM
5124
5125 leaf = path->nodes[0];
5d4f98a2
YZ
5126 item_size = btrfs_item_size_nr(leaf, extent_slot);
5127#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5128 if (item_size < sizeof(*ei)) {
5129 BUG_ON(found_extent || extent_slot != path->slots[0]);
5130 ret = convert_extent_item_v0(trans, extent_root, path,
5131 owner_objectid, 0);
79787eaa
JM
5132 if (ret < 0)
5133 goto abort;
5d4f98a2 5134
b3b4aa74 5135 btrfs_release_path(path);
5d4f98a2
YZ
5136 path->leave_spinning = 1;
5137
5138 key.objectid = bytenr;
5139 key.type = BTRFS_EXTENT_ITEM_KEY;
5140 key.offset = num_bytes;
5141
5142 ret = btrfs_search_slot(trans, extent_root, &key, path,
5143 -1, 1);
5144 if (ret) {
5145 printk(KERN_ERR "umm, got %d back from search"
5146 ", was looking for %llu\n", ret,
5147 (unsigned long long)bytenr);
5148 btrfs_print_leaf(extent_root, path->nodes[0]);
5149 }
79787eaa
JM
5150 if (ret < 0)
5151 goto abort;
5d4f98a2
YZ
5152 extent_slot = path->slots[0];
5153 leaf = path->nodes[0];
5154 item_size = btrfs_item_size_nr(leaf, extent_slot);
5155 }
5156#endif
5157 BUG_ON(item_size < sizeof(*ei));
952fccac 5158 ei = btrfs_item_ptr(leaf, extent_slot,
123abc88 5159 struct btrfs_extent_item);
5d4f98a2
YZ
5160 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5161 struct btrfs_tree_block_info *bi;
5162 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5163 bi = (struct btrfs_tree_block_info *)(ei + 1);
5164 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5165 }
56bec294 5166
5d4f98a2 5167 refs = btrfs_extent_refs(leaf, ei);
56bec294
CM
5168 BUG_ON(refs < refs_to_drop);
5169 refs -= refs_to_drop;
5f39d397 5170
5d4f98a2
YZ
5171 if (refs > 0) {
5172 if (extent_op)
5173 __run_delayed_extent_op(extent_op, leaf, ei);
5174 /*
5175 * In the case of inline back ref, reference count will
5176 * be updated by remove_extent_backref
952fccac 5177 */
5d4f98a2
YZ
5178 if (iref) {
5179 BUG_ON(!found_extent);
5180 } else {
5181 btrfs_set_extent_refs(leaf, ei, refs);
5182 btrfs_mark_buffer_dirty(leaf);
5183 }
5184 if (found_extent) {
5185 ret = remove_extent_backref(trans, extent_root, path,
5186 iref, refs_to_drop,
5187 is_data);
79787eaa
JM
5188 if (ret)
5189 goto abort;
952fccac 5190 }
5d4f98a2 5191 } else {
5d4f98a2
YZ
5192 if (found_extent) {
5193 BUG_ON(is_data && refs_to_drop !=
5194 extent_data_ref_count(root, path, iref));
5195 if (iref) {
5196 BUG_ON(path->slots[0] != extent_slot);
5197 } else {
5198 BUG_ON(path->slots[0] != extent_slot + 1);
5199 path->slots[0] = extent_slot;
5200 num_to_del = 2;
5201 }
78fae27e 5202 }
b9473439 5203
952fccac
CM
5204 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5205 num_to_del);
79787eaa
JM
5206 if (ret)
5207 goto abort;
b3b4aa74 5208 btrfs_release_path(path);
21af804c 5209
5d4f98a2 5210 if (is_data) {
459931ec 5211 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
79787eaa
JM
5212 if (ret)
5213 goto abort;
459931ec
CM
5214 }
5215
f0486c68 5216 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
79787eaa
JM
5217 if (ret)
5218 goto abort;
a28ec197 5219 }
79787eaa 5220out:
5caf2a00 5221 btrfs_free_path(path);
a28ec197 5222 return ret;
79787eaa
JM
5223
5224abort:
5225 btrfs_abort_transaction(trans, extent_root, ret);
5226 goto out;
a28ec197
CM
5227}
5228
1887be66 5229/*
f0486c68 5230 * when we free an block, it is possible (and likely) that we free the last
1887be66
CM
5231 * delayed ref for that extent as well. This searches the delayed ref tree for
5232 * a given extent, and if there are no other delayed refs to be processed, it
5233 * removes it from the tree.
5234 */
5235static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5236 struct btrfs_root *root, u64 bytenr)
5237{
5238 struct btrfs_delayed_ref_head *head;
5239 struct btrfs_delayed_ref_root *delayed_refs;
5240 struct btrfs_delayed_ref_node *ref;
5241 struct rb_node *node;
f0486c68 5242 int ret = 0;
1887be66
CM
5243
5244 delayed_refs = &trans->transaction->delayed_refs;
5245 spin_lock(&delayed_refs->lock);
5246 head = btrfs_find_delayed_ref_head(trans, bytenr);
5247 if (!head)
5248 goto out;
5249
5250 node = rb_prev(&head->node.rb_node);
5251 if (!node)
5252 goto out;
5253
5254 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5255
5256 /* there are still entries for this ref, we can't drop it */
5257 if (ref->bytenr == bytenr)
5258 goto out;
5259
5d4f98a2
YZ
5260 if (head->extent_op) {
5261 if (!head->must_insert_reserved)
5262 goto out;
5263 kfree(head->extent_op);
5264 head->extent_op = NULL;
5265 }
5266
1887be66
CM
5267 /*
5268 * waiting for the lock here would deadlock. If someone else has it
5269 * locked they are already in the process of dropping it anyway
5270 */
5271 if (!mutex_trylock(&head->mutex))
5272 goto out;
5273
5274 /*
5275 * at this point we have a head with no other entries. Go
5276 * ahead and process it.
5277 */
5278 head->node.in_tree = 0;
5279 rb_erase(&head->node.rb_node, &delayed_refs->root);
c3e69d58 5280
1887be66 5281 delayed_refs->num_entries--;
097b8a7c
JS
5282 if (waitqueue_active(&root->fs_info->tree_mod_seq_wait))
5283 wake_up(&root->fs_info->tree_mod_seq_wait);
1887be66
CM
5284
5285 /*
5286 * we don't take a ref on the node because we're removing it from the
5287 * tree, so we just steal the ref the tree was holding.
5288 */
c3e69d58
CM
5289 delayed_refs->num_heads--;
5290 if (list_empty(&head->cluster))
5291 delayed_refs->num_heads_ready--;
5292
5293 list_del_init(&head->cluster);
1887be66
CM
5294 spin_unlock(&delayed_refs->lock);
5295
f0486c68
YZ
5296 BUG_ON(head->extent_op);
5297 if (head->must_insert_reserved)
5298 ret = 1;
5299
5300 mutex_unlock(&head->mutex);
1887be66 5301 btrfs_put_delayed_ref(&head->node);
f0486c68 5302 return ret;
1887be66
CM
5303out:
5304 spin_unlock(&delayed_refs->lock);
5305 return 0;
5306}
5307
f0486c68
YZ
5308void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5309 struct btrfs_root *root,
5310 struct extent_buffer *buf,
5581a51a 5311 u64 parent, int last_ref)
f0486c68 5312{
f0486c68
YZ
5313 struct btrfs_block_group_cache *cache = NULL;
5314 int ret;
5315
5316 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
66d7e7f0
AJ
5317 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5318 buf->start, buf->len,
5319 parent, root->root_key.objectid,
5320 btrfs_header_level(buf),
5581a51a 5321 BTRFS_DROP_DELAYED_REF, NULL, 0);
79787eaa 5322 BUG_ON(ret); /* -ENOMEM */
f0486c68
YZ
5323 }
5324
5325 if (!last_ref)
5326 return;
5327
f0486c68 5328 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
f0486c68
YZ
5329
5330 if (btrfs_header_generation(buf) == trans->transid) {
5331 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5332 ret = check_ref_cleanup(trans, root, buf->start);
5333 if (!ret)
37be25bc 5334 goto out;
f0486c68
YZ
5335 }
5336
5337 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5338 pin_down_extent(root, cache, buf->start, buf->len, 1);
37be25bc 5339 goto out;
f0486c68
YZ
5340 }
5341
5342 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5343
5344 btrfs_add_free_space(cache, buf->start, buf->len);
fb25e914 5345 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
f0486c68
YZ
5346 }
5347out:
a826d6dc
JB
5348 /*
5349 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5350 * anymore.
5351 */
5352 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
f0486c68
YZ
5353 btrfs_put_block_group(cache);
5354}
5355
79787eaa 5356/* Can return -ENOMEM */
66d7e7f0
AJ
5357int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5358 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5359 u64 owner, u64 offset, int for_cow)
925baedd
CM
5360{
5361 int ret;
66d7e7f0 5362 struct btrfs_fs_info *fs_info = root->fs_info;
925baedd 5363
56bec294
CM
5364 /*
5365 * tree log blocks never actually go into the extent allocation
5366 * tree, just update pinning info and exit early.
56bec294 5367 */
5d4f98a2
YZ
5368 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5369 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
b9473439 5370 /* unlocks the pinned mutex */
11833d66 5371 btrfs_pin_extent(root, bytenr, num_bytes, 1);
56bec294 5372 ret = 0;
5d4f98a2 5373 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
66d7e7f0
AJ
5374 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5375 num_bytes,
5d4f98a2 5376 parent, root_objectid, (int)owner,
66d7e7f0 5377 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5d4f98a2 5378 } else {
66d7e7f0
AJ
5379 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5380 num_bytes,
5381 parent, root_objectid, owner,
5382 offset, BTRFS_DROP_DELAYED_REF,
5383 NULL, for_cow);
56bec294 5384 }
925baedd
CM
5385 return ret;
5386}
5387
87ee04eb
CM
5388static u64 stripe_align(struct btrfs_root *root, u64 val)
5389{
5390 u64 mask = ((u64)root->stripesize - 1);
5391 u64 ret = (val + mask) & ~mask;
5392 return ret;
5393}
5394
817d52f8
JB
5395/*
5396 * when we wait for progress in the block group caching, its because
5397 * our allocation attempt failed at least once. So, we must sleep
5398 * and let some progress happen before we try again.
5399 *
5400 * This function will sleep at least once waiting for new free space to
5401 * show up, and then it will check the block group free space numbers
5402 * for our min num_bytes. Another option is to have it go ahead
5403 * and look in the rbtree for a free extent of a given size, but this
5404 * is a good start.
5405 */
5406static noinline int
5407wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5408 u64 num_bytes)
5409{
11833d66 5410 struct btrfs_caching_control *caching_ctl;
817d52f8
JB
5411 DEFINE_WAIT(wait);
5412
11833d66
YZ
5413 caching_ctl = get_caching_control(cache);
5414 if (!caching_ctl)
817d52f8 5415 return 0;
817d52f8 5416
11833d66 5417 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
34d52cb6 5418 (cache->free_space_ctl->free_space >= num_bytes));
11833d66
YZ
5419
5420 put_caching_control(caching_ctl);
5421 return 0;
5422}
5423
5424static noinline int
5425wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5426{
5427 struct btrfs_caching_control *caching_ctl;
5428 DEFINE_WAIT(wait);
5429
5430 caching_ctl = get_caching_control(cache);
5431 if (!caching_ctl)
5432 return 0;
5433
5434 wait_event(caching_ctl->wait, block_group_cache_done(cache));
5435
5436 put_caching_control(caching_ctl);
817d52f8
JB
5437 return 0;
5438}
5439
7738a53a 5440static int __get_block_group_index(u64 flags)
b742bb82
YZ
5441{
5442 int index;
7738a53a
ID
5443
5444 if (flags & BTRFS_BLOCK_GROUP_RAID10)
b742bb82 5445 index = 0;
7738a53a 5446 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
b742bb82 5447 index = 1;
7738a53a 5448 else if (flags & BTRFS_BLOCK_GROUP_DUP)
b742bb82 5449 index = 2;
7738a53a 5450 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
b742bb82
YZ
5451 index = 3;
5452 else
5453 index = 4;
7738a53a 5454
b742bb82
YZ
5455 return index;
5456}
5457
7738a53a
ID
5458static int get_block_group_index(struct btrfs_block_group_cache *cache)
5459{
5460 return __get_block_group_index(cache->flags);
5461}
5462
817d52f8 5463enum btrfs_loop_type {
285ff5af
JB
5464 LOOP_CACHING_NOWAIT = 0,
5465 LOOP_CACHING_WAIT = 1,
5466 LOOP_ALLOC_CHUNK = 2,
5467 LOOP_NO_EMPTY_SIZE = 3,
817d52f8
JB
5468};
5469
fec577fb
CM
5470/*
5471 * walks the btree of allocated extents and find a hole of a given size.
5472 * The key ins is changed to record the hole:
5473 * ins->objectid == block start
62e2749e 5474 * ins->flags = BTRFS_EXTENT_ITEM_KEY
fec577fb
CM
5475 * ins->offset == number of blocks
5476 * Any available blocks before search_start are skipped.
5477 */
d397712b 5478static noinline int find_free_extent(struct btrfs_trans_handle *trans,
98ed5174
CM
5479 struct btrfs_root *orig_root,
5480 u64 num_bytes, u64 empty_size,
98ed5174 5481 u64 hint_byte, struct btrfs_key *ins,
e0f54067 5482 u64 data)
fec577fb 5483{
80eb234a 5484 int ret = 0;
d397712b 5485 struct btrfs_root *root = orig_root->fs_info->extent_root;
fa9c0d79 5486 struct btrfs_free_cluster *last_ptr = NULL;
80eb234a 5487 struct btrfs_block_group_cache *block_group = NULL;
274bd4fb 5488 struct btrfs_block_group_cache *used_block_group;
81c9ad23 5489 u64 search_start = 0;
239b14b3 5490 int empty_cluster = 2 * 1024 * 1024;
0ef3e66b 5491 int allowed_chunk_alloc = 0;
ccf0e725 5492 int done_chunk_alloc = 0;
80eb234a 5493 struct btrfs_space_info *space_info;
fa9c0d79 5494 int loop = 0;
f0486c68 5495 int index = 0;
fb25e914
JB
5496 int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5497 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
817d52f8 5498 bool found_uncached_bg = false;
0a24325e 5499 bool failed_cluster_refill = false;
1cdda9b8 5500 bool failed_alloc = false;
67377734 5501 bool use_cluster = true;
60d2adbb 5502 bool have_caching_bg = false;
fec577fb 5503
db94535d 5504 WARN_ON(num_bytes < root->sectorsize);
b1a4d965 5505 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
80eb234a
JB
5506 ins->objectid = 0;
5507 ins->offset = 0;
b1a4d965 5508
3f7de037
JB
5509 trace_find_free_extent(orig_root, num_bytes, empty_size, data);
5510
2552d17e 5511 space_info = __find_space_info(root->fs_info, data);
1b1d1f66 5512 if (!space_info) {
e0f54067 5513 printk(KERN_ERR "No space info for %llu\n", data);
1b1d1f66
JB
5514 return -ENOSPC;
5515 }
2552d17e 5516
67377734
JB
5517 /*
5518 * If the space info is for both data and metadata it means we have a
5519 * small filesystem and we can't use the clustering stuff.
5520 */
5521 if (btrfs_mixed_space_info(space_info))
5522 use_cluster = false;
5523
0ef3e66b
CM
5524 if (orig_root->ref_cows || empty_size)
5525 allowed_chunk_alloc = 1;
5526
67377734 5527 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
fa9c0d79 5528 last_ptr = &root->fs_info->meta_alloc_cluster;
536ac8ae
CM
5529 if (!btrfs_test_opt(root, SSD))
5530 empty_cluster = 64 * 1024;
239b14b3
CM
5531 }
5532
67377734
JB
5533 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5534 btrfs_test_opt(root, SSD)) {
fa9c0d79
CM
5535 last_ptr = &root->fs_info->data_alloc_cluster;
5536 }
0f9dd46c 5537
239b14b3 5538 if (last_ptr) {
fa9c0d79
CM
5539 spin_lock(&last_ptr->lock);
5540 if (last_ptr->block_group)
5541 hint_byte = last_ptr->window_start;
5542 spin_unlock(&last_ptr->lock);
239b14b3 5543 }
fa9c0d79 5544
a061fc8d 5545 search_start = max(search_start, first_logical_byte(root, 0));
239b14b3 5546 search_start = max(search_start, hint_byte);
0b86a832 5547
817d52f8 5548 if (!last_ptr)
fa9c0d79 5549 empty_cluster = 0;
fa9c0d79 5550
2552d17e 5551 if (search_start == hint_byte) {
2552d17e
JB
5552 block_group = btrfs_lookup_block_group(root->fs_info,
5553 search_start);
274bd4fb 5554 used_block_group = block_group;
817d52f8
JB
5555 /*
5556 * we don't want to use the block group if it doesn't match our
5557 * allocation bits, or if its not cached.
ccf0e725
JB
5558 *
5559 * However if we are re-searching with an ideal block group
5560 * picked out then we don't care that the block group is cached.
817d52f8
JB
5561 */
5562 if (block_group && block_group_bits(block_group, data) &&
285ff5af 5563 block_group->cached != BTRFS_CACHE_NO) {
2552d17e 5564 down_read(&space_info->groups_sem);
44fb5511
CM
5565 if (list_empty(&block_group->list) ||
5566 block_group->ro) {
5567 /*
5568 * someone is removing this block group,
5569 * we can't jump into the have_block_group
5570 * target because our list pointers are not
5571 * valid
5572 */
5573 btrfs_put_block_group(block_group);
5574 up_read(&space_info->groups_sem);
ccf0e725 5575 } else {
b742bb82 5576 index = get_block_group_index(block_group);
44fb5511 5577 goto have_block_group;
ccf0e725 5578 }
2552d17e 5579 } else if (block_group) {
fa9c0d79 5580 btrfs_put_block_group(block_group);
2552d17e 5581 }
42e70e7a 5582 }
2552d17e 5583search:
60d2adbb 5584 have_caching_bg = false;
80eb234a 5585 down_read(&space_info->groups_sem);
b742bb82
YZ
5586 list_for_each_entry(block_group, &space_info->block_groups[index],
5587 list) {
6226cb0a 5588 u64 offset;
817d52f8 5589 int cached;
8a1413a2 5590
274bd4fb 5591 used_block_group = block_group;
11dfe35a 5592 btrfs_get_block_group(block_group);
2552d17e 5593 search_start = block_group->key.objectid;
42e70e7a 5594
83a50de9
CM
5595 /*
5596 * this can happen if we end up cycling through all the
5597 * raid types, but we want to make sure we only allocate
5598 * for the proper type.
5599 */
5600 if (!block_group_bits(block_group, data)) {
5601 u64 extra = BTRFS_BLOCK_GROUP_DUP |
5602 BTRFS_BLOCK_GROUP_RAID1 |
5603 BTRFS_BLOCK_GROUP_RAID10;
5604
5605 /*
5606 * if they asked for extra copies and this block group
5607 * doesn't provide them, bail. This does allow us to
5608 * fill raid0 from raid1.
5609 */
5610 if ((data & extra) && !(block_group->flags & extra))
5611 goto loop;
5612 }
5613
2552d17e 5614have_block_group:
291c7d2f
JB
5615 cached = block_group_cache_done(block_group);
5616 if (unlikely(!cached)) {
291c7d2f 5617 found_uncached_bg = true;
b8399dee 5618 ret = cache_block_group(block_group, trans,
285ff5af 5619 orig_root, 0);
1d4284bd
CM
5620 BUG_ON(ret < 0);
5621 ret = 0;
817d52f8
JB
5622 }
5623
ea6a478e 5624 if (unlikely(block_group->ro))
2552d17e 5625 goto loop;
0f9dd46c 5626
0a24325e 5627 /*
062c05c4
AO
5628 * Ok we want to try and use the cluster allocator, so
5629 * lets look there
0a24325e 5630 */
062c05c4 5631 if (last_ptr) {
fa9c0d79
CM
5632 /*
5633 * the refill lock keeps out other
5634 * people trying to start a new cluster
5635 */
5636 spin_lock(&last_ptr->refill_lock);
274bd4fb
AO
5637 used_block_group = last_ptr->block_group;
5638 if (used_block_group != block_group &&
5639 (!used_block_group ||
5640 used_block_group->ro ||
5641 !block_group_bits(used_block_group, data))) {
5642 used_block_group = block_group;
44fb5511 5643 goto refill_cluster;
274bd4fb
AO
5644 }
5645
5646 if (used_block_group != block_group)
5647 btrfs_get_block_group(used_block_group);
44fb5511 5648
274bd4fb
AO
5649 offset = btrfs_alloc_from_cluster(used_block_group,
5650 last_ptr, num_bytes, used_block_group->key.objectid);
fa9c0d79
CM
5651 if (offset) {
5652 /* we have a block, we're done */
5653 spin_unlock(&last_ptr->refill_lock);
3f7de037
JB
5654 trace_btrfs_reserve_extent_cluster(root,
5655 block_group, search_start, num_bytes);
fa9c0d79
CM
5656 goto checks;
5657 }
5658
274bd4fb
AO
5659 WARN_ON(last_ptr->block_group != used_block_group);
5660 if (used_block_group != block_group) {
5661 btrfs_put_block_group(used_block_group);
5662 used_block_group = block_group;
fa9c0d79 5663 }
44fb5511 5664refill_cluster:
274bd4fb 5665 BUG_ON(used_block_group != block_group);
062c05c4
AO
5666 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
5667 * set up a new clusters, so lets just skip it
5668 * and let the allocator find whatever block
5669 * it can find. If we reach this point, we
5670 * will have tried the cluster allocator
5671 * plenty of times and not have found
5672 * anything, so we are likely way too
5673 * fragmented for the clustering stuff to find
a5f6f719
AO
5674 * anything.
5675 *
5676 * However, if the cluster is taken from the
5677 * current block group, release the cluster
5678 * first, so that we stand a better chance of
5679 * succeeding in the unclustered
5680 * allocation. */
5681 if (loop >= LOOP_NO_EMPTY_SIZE &&
5682 last_ptr->block_group != block_group) {
062c05c4
AO
5683 spin_unlock(&last_ptr->refill_lock);
5684 goto unclustered_alloc;
5685 }
5686
fa9c0d79
CM
5687 /*
5688 * this cluster didn't work out, free it and
5689 * start over
5690 */
5691 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5692
a5f6f719
AO
5693 if (loop >= LOOP_NO_EMPTY_SIZE) {
5694 spin_unlock(&last_ptr->refill_lock);
5695 goto unclustered_alloc;
5696 }
5697
fa9c0d79 5698 /* allocate a cluster in this block group */
451d7585 5699 ret = btrfs_find_space_cluster(trans, root,
fa9c0d79 5700 block_group, last_ptr,
1b22bad7 5701 search_start, num_bytes,
fa9c0d79
CM
5702 empty_cluster + empty_size);
5703 if (ret == 0) {
5704 /*
5705 * now pull our allocation out of this
5706 * cluster
5707 */
5708 offset = btrfs_alloc_from_cluster(block_group,
5709 last_ptr, num_bytes,
5710 search_start);
5711 if (offset) {
5712 /* we found one, proceed */
5713 spin_unlock(&last_ptr->refill_lock);
3f7de037
JB
5714 trace_btrfs_reserve_extent_cluster(root,
5715 block_group, search_start,
5716 num_bytes);
fa9c0d79
CM
5717 goto checks;
5718 }
0a24325e
JB
5719 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5720 && !failed_cluster_refill) {
817d52f8
JB
5721 spin_unlock(&last_ptr->refill_lock);
5722
0a24325e 5723 failed_cluster_refill = true;
817d52f8
JB
5724 wait_block_group_cache_progress(block_group,
5725 num_bytes + empty_cluster + empty_size);
5726 goto have_block_group;
fa9c0d79 5727 }
817d52f8 5728
fa9c0d79
CM
5729 /*
5730 * at this point we either didn't find a cluster
5731 * or we weren't able to allocate a block from our
5732 * cluster. Free the cluster we've been trying
5733 * to use, and go to the next block group
5734 */
0a24325e 5735 btrfs_return_cluster_to_free_space(NULL, last_ptr);
fa9c0d79 5736 spin_unlock(&last_ptr->refill_lock);
0a24325e 5737 goto loop;
fa9c0d79
CM
5738 }
5739
062c05c4 5740unclustered_alloc:
a5f6f719
AO
5741 spin_lock(&block_group->free_space_ctl->tree_lock);
5742 if (cached &&
5743 block_group->free_space_ctl->free_space <
5744 num_bytes + empty_cluster + empty_size) {
5745 spin_unlock(&block_group->free_space_ctl->tree_lock);
5746 goto loop;
5747 }
5748 spin_unlock(&block_group->free_space_ctl->tree_lock);
5749
6226cb0a
JB
5750 offset = btrfs_find_space_for_alloc(block_group, search_start,
5751 num_bytes, empty_size);
1cdda9b8
JB
5752 /*
5753 * If we didn't find a chunk, and we haven't failed on this
5754 * block group before, and this block group is in the middle of
5755 * caching and we are ok with waiting, then go ahead and wait
5756 * for progress to be made, and set failed_alloc to true.
5757 *
5758 * If failed_alloc is true then we've already waited on this
5759 * block group once and should move on to the next block group.
5760 */
5761 if (!offset && !failed_alloc && !cached &&
5762 loop > LOOP_CACHING_NOWAIT) {
817d52f8 5763 wait_block_group_cache_progress(block_group,
1cdda9b8
JB
5764 num_bytes + empty_size);
5765 failed_alloc = true;
817d52f8 5766 goto have_block_group;
1cdda9b8 5767 } else if (!offset) {
60d2adbb
MX
5768 if (!cached)
5769 have_caching_bg = true;
1cdda9b8 5770 goto loop;
817d52f8 5771 }
fa9c0d79 5772checks:
6226cb0a 5773 search_start = stripe_align(root, offset);
25179201 5774
2552d17e
JB
5775 /* move on to the next group */
5776 if (search_start + num_bytes >
274bd4fb
AO
5777 used_block_group->key.objectid + used_block_group->key.offset) {
5778 btrfs_add_free_space(used_block_group, offset, num_bytes);
2552d17e 5779 goto loop;
6226cb0a 5780 }
f5a31e16 5781
f0486c68 5782 if (offset < search_start)
274bd4fb 5783 btrfs_add_free_space(used_block_group, offset,
f0486c68
YZ
5784 search_start - offset);
5785 BUG_ON(offset > search_start);
2552d17e 5786
274bd4fb 5787 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
fb25e914 5788 alloc_type);
f0486c68 5789 if (ret == -EAGAIN) {
274bd4fb 5790 btrfs_add_free_space(used_block_group, offset, num_bytes);
2552d17e 5791 goto loop;
0f9dd46c 5792 }
0b86a832 5793
f0486c68 5794 /* we are all good, lets return */
2552d17e
JB
5795 ins->objectid = search_start;
5796 ins->offset = num_bytes;
d2fb3437 5797
3f7de037
JB
5798 trace_btrfs_reserve_extent(orig_root, block_group,
5799 search_start, num_bytes);
6226cb0a 5800 if (offset < search_start)
274bd4fb 5801 btrfs_add_free_space(used_block_group, offset,
6226cb0a
JB
5802 search_start - offset);
5803 BUG_ON(offset > search_start);
274bd4fb
AO
5804 if (used_block_group != block_group)
5805 btrfs_put_block_group(used_block_group);
d82a6f1d 5806 btrfs_put_block_group(block_group);
2552d17e
JB
5807 break;
5808loop:
0a24325e 5809 failed_cluster_refill = false;
1cdda9b8 5810 failed_alloc = false;
b742bb82 5811 BUG_ON(index != get_block_group_index(block_group));
274bd4fb
AO
5812 if (used_block_group != block_group)
5813 btrfs_put_block_group(used_block_group);
fa9c0d79 5814 btrfs_put_block_group(block_group);
2552d17e
JB
5815 }
5816 up_read(&space_info->groups_sem);
5817
60d2adbb
MX
5818 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
5819 goto search;
5820
b742bb82
YZ
5821 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5822 goto search;
5823
285ff5af 5824 /*
ccf0e725
JB
5825 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5826 * caching kthreads as we move along
817d52f8
JB
5827 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5828 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5829 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5830 * again
fa9c0d79 5831 */
723bda20 5832 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
b742bb82 5833 index = 0;
723bda20 5834 loop++;
817d52f8 5835 if (loop == LOOP_ALLOC_CHUNK) {
723bda20
JB
5836 if (allowed_chunk_alloc) {
5837 ret = do_chunk_alloc(trans, root, num_bytes +
5838 2 * 1024 * 1024, data,
5839 CHUNK_ALLOC_LIMITED);
79787eaa
JM
5840 if (ret < 0) {
5841 btrfs_abort_transaction(trans,
5842 root, ret);
5843 goto out;
5844 }
723bda20
JB
5845 allowed_chunk_alloc = 0;
5846 if (ret == 1)
5847 done_chunk_alloc = 1;
5848 } else if (!done_chunk_alloc &&
5849 space_info->force_alloc ==
5850 CHUNK_ALLOC_NO_FORCE) {
5851 space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5852 }
2552d17e 5853
723bda20
JB
5854 /*
5855 * We didn't allocate a chunk, go ahead and drop the
5856 * empty size and loop again.
5857 */
5858 if (!done_chunk_alloc)
5859 loop = LOOP_NO_EMPTY_SIZE;
2552d17e
JB
5860 }
5861
723bda20
JB
5862 if (loop == LOOP_NO_EMPTY_SIZE) {
5863 empty_size = 0;
5864 empty_cluster = 0;
fa9c0d79 5865 }
723bda20
JB
5866
5867 goto search;
2552d17e
JB
5868 } else if (!ins->objectid) {
5869 ret = -ENOSPC;
d82a6f1d 5870 } else if (ins->objectid) {
80eb234a 5871 ret = 0;
be744175 5872 }
79787eaa 5873out:
be744175 5874
0f70abe2 5875 return ret;
fec577fb 5876}
ec44a35c 5877
9ed74f2d
JB
5878static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5879 int dump_block_groups)
0f9dd46c
JB
5880{
5881 struct btrfs_block_group_cache *cache;
b742bb82 5882 int index = 0;
0f9dd46c 5883
9ed74f2d 5884 spin_lock(&info->lock);
fb25e914
JB
5885 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5886 (unsigned long long)info->flags,
d397712b 5887 (unsigned long long)(info->total_bytes - info->bytes_used -
9ed74f2d 5888 info->bytes_pinned - info->bytes_reserved -
8929ecfa 5889 info->bytes_readonly),
d397712b 5890 (info->full) ? "" : "not ");
8929ecfa
YZ
5891 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5892 "reserved=%llu, may_use=%llu, readonly=%llu\n",
21380931 5893 (unsigned long long)info->total_bytes,
8929ecfa 5894 (unsigned long long)info->bytes_used,
21380931 5895 (unsigned long long)info->bytes_pinned,
8929ecfa 5896 (unsigned long long)info->bytes_reserved,
21380931 5897 (unsigned long long)info->bytes_may_use,
8929ecfa 5898 (unsigned long long)info->bytes_readonly);
9ed74f2d
JB
5899 spin_unlock(&info->lock);
5900
5901 if (!dump_block_groups)
5902 return;
0f9dd46c 5903
80eb234a 5904 down_read(&info->groups_sem);
b742bb82
YZ
5905again:
5906 list_for_each_entry(cache, &info->block_groups[index], list) {
0f9dd46c 5907 spin_lock(&cache->lock);
d397712b
CM
5908 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5909 "%llu pinned %llu reserved\n",
5910 (unsigned long long)cache->key.objectid,
5911 (unsigned long long)cache->key.offset,
5912 (unsigned long long)btrfs_block_group_used(&cache->item),
5913 (unsigned long long)cache->pinned,
5914 (unsigned long long)cache->reserved);
0f9dd46c
JB
5915 btrfs_dump_free_space(cache, bytes);
5916 spin_unlock(&cache->lock);
5917 }
b742bb82
YZ
5918 if (++index < BTRFS_NR_RAID_TYPES)
5919 goto again;
80eb234a 5920 up_read(&info->groups_sem);
0f9dd46c 5921}
e8569813 5922
11833d66
YZ
5923int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5924 struct btrfs_root *root,
5925 u64 num_bytes, u64 min_alloc_size,
5926 u64 empty_size, u64 hint_byte,
81c9ad23 5927 struct btrfs_key *ins, u64 data)
fec577fb 5928{
9e622d6b 5929 bool final_tried = false;
fec577fb 5930 int ret;
925baedd 5931
6a63209f 5932 data = btrfs_get_alloc_profile(root, data);
98d20f67 5933again:
0ef3e66b
CM
5934 /*
5935 * the only place that sets empty_size is btrfs_realloc_node, which
5936 * is not called recursively on allocations
5937 */
79787eaa 5938 if (empty_size || root->ref_cows) {
6324fbf3 5939 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
0e4f8f88
CM
5940 num_bytes + 2 * 1024 * 1024, data,
5941 CHUNK_ALLOC_NO_FORCE);
79787eaa
JM
5942 if (ret < 0 && ret != -ENOSPC) {
5943 btrfs_abort_transaction(trans, root, ret);
5944 return ret;
5945 }
5946 }
0b86a832 5947
db94535d
CM
5948 WARN_ON(num_bytes < root->sectorsize);
5949 ret = find_free_extent(trans, root, num_bytes, empty_size,
81c9ad23 5950 hint_byte, ins, data);
3b951516 5951
9e622d6b
MX
5952 if (ret == -ENOSPC) {
5953 if (!final_tried) {
5954 num_bytes = num_bytes >> 1;
5955 num_bytes = num_bytes & ~(root->sectorsize - 1);
5956 num_bytes = max(num_bytes, min_alloc_size);
79787eaa 5957 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
9e622d6b 5958 num_bytes, data, CHUNK_ALLOC_FORCE);
79787eaa
JM
5959 if (ret < 0 && ret != -ENOSPC) {
5960 btrfs_abort_transaction(trans, root, ret);
5961 return ret;
5962 }
9e622d6b
MX
5963 if (num_bytes == min_alloc_size)
5964 final_tried = true;
5965 goto again;
5966 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
5967 struct btrfs_space_info *sinfo;
5968
5969 sinfo = __find_space_info(root->fs_info, data);
5970 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5971 "wanted %llu\n", (unsigned long long)data,
5972 (unsigned long long)num_bytes);
53804280
JM
5973 if (sinfo)
5974 dump_space_info(sinfo, num_bytes, 1);
9e622d6b 5975 }
925baedd 5976 }
0f9dd46c 5977
1abe9b8a 5978 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5979
0f9dd46c 5980 return ret;
e6dcd2dc
CM
5981}
5982
e688b725
CM
5983static int __btrfs_free_reserved_extent(struct btrfs_root *root,
5984 u64 start, u64 len, int pin)
65b51a00 5985{
0f9dd46c 5986 struct btrfs_block_group_cache *cache;
1f3c79a2 5987 int ret = 0;
0f9dd46c 5988
0f9dd46c
JB
5989 cache = btrfs_lookup_block_group(root->fs_info, start);
5990 if (!cache) {
d397712b
CM
5991 printk(KERN_ERR "Unable to find block group for %llu\n",
5992 (unsigned long long)start);
0f9dd46c
JB
5993 return -ENOSPC;
5994 }
1f3c79a2 5995
5378e607
LD
5996 if (btrfs_test_opt(root, DISCARD))
5997 ret = btrfs_discard_extent(root, start, len, NULL);
1f3c79a2 5998
e688b725
CM
5999 if (pin)
6000 pin_down_extent(root, cache, start, len, 1);
6001 else {
6002 btrfs_add_free_space(cache, start, len);
6003 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6004 }
fa9c0d79 6005 btrfs_put_block_group(cache);
817d52f8 6006
1abe9b8a 6007 trace_btrfs_reserved_extent_free(root, start, len);
6008
e6dcd2dc
CM
6009 return ret;
6010}
6011
e688b725
CM
6012int btrfs_free_reserved_extent(struct btrfs_root *root,
6013 u64 start, u64 len)
6014{
6015 return __btrfs_free_reserved_extent(root, start, len, 0);
6016}
6017
6018int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6019 u64 start, u64 len)
6020{
6021 return __btrfs_free_reserved_extent(root, start, len, 1);
6022}
6023
5d4f98a2
YZ
6024static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6025 struct btrfs_root *root,
6026 u64 parent, u64 root_objectid,
6027 u64 flags, u64 owner, u64 offset,
6028 struct btrfs_key *ins, int ref_mod)
e6dcd2dc
CM
6029{
6030 int ret;
5d4f98a2 6031 struct btrfs_fs_info *fs_info = root->fs_info;
e6dcd2dc 6032 struct btrfs_extent_item *extent_item;
5d4f98a2 6033 struct btrfs_extent_inline_ref *iref;
e6dcd2dc 6034 struct btrfs_path *path;
5d4f98a2
YZ
6035 struct extent_buffer *leaf;
6036 int type;
6037 u32 size;
26b8003f 6038
5d4f98a2
YZ
6039 if (parent > 0)
6040 type = BTRFS_SHARED_DATA_REF_KEY;
6041 else
6042 type = BTRFS_EXTENT_DATA_REF_KEY;
58176a96 6043
5d4f98a2 6044 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7bb86316
CM
6045
6046 path = btrfs_alloc_path();
db5b493a
TI
6047 if (!path)
6048 return -ENOMEM;
47e4bb98 6049
b9473439 6050 path->leave_spinning = 1;
5d4f98a2
YZ
6051 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6052 ins, size);
79787eaa
JM
6053 if (ret) {
6054 btrfs_free_path(path);
6055 return ret;
6056 }
0f9dd46c 6057
5d4f98a2
YZ
6058 leaf = path->nodes[0];
6059 extent_item = btrfs_item_ptr(leaf, path->slots[0],
47e4bb98 6060 struct btrfs_extent_item);
5d4f98a2
YZ
6061 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6062 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6063 btrfs_set_extent_flags(leaf, extent_item,
6064 flags | BTRFS_EXTENT_FLAG_DATA);
6065
6066 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6067 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6068 if (parent > 0) {
6069 struct btrfs_shared_data_ref *ref;
6070 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6071 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6072 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6073 } else {
6074 struct btrfs_extent_data_ref *ref;
6075 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6076 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6077 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6078 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6079 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6080 }
47e4bb98
CM
6081
6082 btrfs_mark_buffer_dirty(path->nodes[0]);
7bb86316 6083 btrfs_free_path(path);
f510cfec 6084
f0486c68 6085 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
79787eaa 6086 if (ret) { /* -ENOENT, logic error */
d397712b
CM
6087 printk(KERN_ERR "btrfs update block group failed for %llu "
6088 "%llu\n", (unsigned long long)ins->objectid,
6089 (unsigned long long)ins->offset);
f5947066
CM
6090 BUG();
6091 }
e6dcd2dc
CM
6092 return ret;
6093}
6094
5d4f98a2
YZ
6095static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6096 struct btrfs_root *root,
6097 u64 parent, u64 root_objectid,
6098 u64 flags, struct btrfs_disk_key *key,
6099 int level, struct btrfs_key *ins)
e6dcd2dc
CM
6100{
6101 int ret;
5d4f98a2
YZ
6102 struct btrfs_fs_info *fs_info = root->fs_info;
6103 struct btrfs_extent_item *extent_item;
6104 struct btrfs_tree_block_info *block_info;
6105 struct btrfs_extent_inline_ref *iref;
6106 struct btrfs_path *path;
6107 struct extent_buffer *leaf;
6108 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
1c2308f8 6109
5d4f98a2 6110 path = btrfs_alloc_path();
d8926bb3
MF
6111 if (!path)
6112 return -ENOMEM;
56bec294 6113
5d4f98a2
YZ
6114 path->leave_spinning = 1;
6115 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6116 ins, size);
79787eaa
JM
6117 if (ret) {
6118 btrfs_free_path(path);
6119 return ret;
6120 }
5d4f98a2
YZ
6121
6122 leaf = path->nodes[0];
6123 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6124 struct btrfs_extent_item);
6125 btrfs_set_extent_refs(leaf, extent_item, 1);
6126 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6127 btrfs_set_extent_flags(leaf, extent_item,
6128 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6129 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6130
6131 btrfs_set_tree_block_key(leaf, block_info, key);
6132 btrfs_set_tree_block_level(leaf, block_info, level);
6133
6134 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6135 if (parent > 0) {
6136 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6137 btrfs_set_extent_inline_ref_type(leaf, iref,
6138 BTRFS_SHARED_BLOCK_REF_KEY);
6139 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6140 } else {
6141 btrfs_set_extent_inline_ref_type(leaf, iref,
6142 BTRFS_TREE_BLOCK_REF_KEY);
6143 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6144 }
6145
6146 btrfs_mark_buffer_dirty(leaf);
6147 btrfs_free_path(path);
6148
f0486c68 6149 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
79787eaa 6150 if (ret) { /* -ENOENT, logic error */
5d4f98a2
YZ
6151 printk(KERN_ERR "btrfs update block group failed for %llu "
6152 "%llu\n", (unsigned long long)ins->objectid,
6153 (unsigned long long)ins->offset);
6154 BUG();
6155 }
6156 return ret;
6157}
6158
6159int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6160 struct btrfs_root *root,
6161 u64 root_objectid, u64 owner,
6162 u64 offset, struct btrfs_key *ins)
6163{
6164 int ret;
6165
6166 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6167
66d7e7f0
AJ
6168 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6169 ins->offset, 0,
6170 root_objectid, owner, offset,
6171 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
e6dcd2dc
CM
6172 return ret;
6173}
e02119d5
CM
6174
6175/*
6176 * this is used by the tree logging recovery code. It records that
6177 * an extent has been allocated and makes sure to clear the free
6178 * space cache bits as well
6179 */
5d4f98a2
YZ
6180int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6181 struct btrfs_root *root,
6182 u64 root_objectid, u64 owner, u64 offset,
6183 struct btrfs_key *ins)
e02119d5
CM
6184{
6185 int ret;
6186 struct btrfs_block_group_cache *block_group;
11833d66
YZ
6187 struct btrfs_caching_control *caching_ctl;
6188 u64 start = ins->objectid;
6189 u64 num_bytes = ins->offset;
e02119d5 6190
e02119d5 6191 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
b8399dee 6192 cache_block_group(block_group, trans, NULL, 0);
11833d66 6193 caching_ctl = get_caching_control(block_group);
e02119d5 6194
11833d66
YZ
6195 if (!caching_ctl) {
6196 BUG_ON(!block_group_cache_done(block_group));
6197 ret = btrfs_remove_free_space(block_group, start, num_bytes);
79787eaa 6198 BUG_ON(ret); /* -ENOMEM */
11833d66
YZ
6199 } else {
6200 mutex_lock(&caching_ctl->mutex);
6201
6202 if (start >= caching_ctl->progress) {
6203 ret = add_excluded_extent(root, start, num_bytes);
79787eaa 6204 BUG_ON(ret); /* -ENOMEM */
11833d66
YZ
6205 } else if (start + num_bytes <= caching_ctl->progress) {
6206 ret = btrfs_remove_free_space(block_group,
6207 start, num_bytes);
79787eaa 6208 BUG_ON(ret); /* -ENOMEM */
11833d66
YZ
6209 } else {
6210 num_bytes = caching_ctl->progress - start;
6211 ret = btrfs_remove_free_space(block_group,
6212 start, num_bytes);
79787eaa 6213 BUG_ON(ret); /* -ENOMEM */
11833d66
YZ
6214
6215 start = caching_ctl->progress;
6216 num_bytes = ins->objectid + ins->offset -
6217 caching_ctl->progress;
6218 ret = add_excluded_extent(root, start, num_bytes);
79787eaa 6219 BUG_ON(ret); /* -ENOMEM */
11833d66
YZ
6220 }
6221
6222 mutex_unlock(&caching_ctl->mutex);
6223 put_caching_control(caching_ctl);
6224 }
6225
fb25e914
JB
6226 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6227 RESERVE_ALLOC_NO_ACCOUNT);
79787eaa 6228 BUG_ON(ret); /* logic error */
fa9c0d79 6229 btrfs_put_block_group(block_group);
5d4f98a2
YZ
6230 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6231 0, owner, offset, ins, 1);
e02119d5
CM
6232 return ret;
6233}
6234
65b51a00
CM
6235struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
6236 struct btrfs_root *root,
4008c04a
CM
6237 u64 bytenr, u32 blocksize,
6238 int level)
65b51a00
CM
6239{
6240 struct extent_buffer *buf;
6241
6242 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6243 if (!buf)
6244 return ERR_PTR(-ENOMEM);
6245 btrfs_set_header_generation(buf, trans->transid);
85d4e461 6246 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
65b51a00
CM
6247 btrfs_tree_lock(buf);
6248 clean_tree_block(trans, root, buf);
3083ee2e 6249 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
b4ce94de
CM
6250
6251 btrfs_set_lock_blocking(buf);
65b51a00 6252 btrfs_set_buffer_uptodate(buf);
b4ce94de 6253
d0c803c4 6254 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8cef4e16
YZ
6255 /*
6256 * we allow two log transactions at a time, use different
6257 * EXENT bit to differentiate dirty pages.
6258 */
6259 if (root->log_transid % 2 == 0)
6260 set_extent_dirty(&root->dirty_log_pages, buf->start,
6261 buf->start + buf->len - 1, GFP_NOFS);
6262 else
6263 set_extent_new(&root->dirty_log_pages, buf->start,
6264 buf->start + buf->len - 1, GFP_NOFS);
d0c803c4
CM
6265 } else {
6266 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
65b51a00 6267 buf->start + buf->len - 1, GFP_NOFS);
d0c803c4 6268 }
65b51a00 6269 trans->blocks_used++;
b4ce94de 6270 /* this returns a buffer locked for blocking */
65b51a00
CM
6271 return buf;
6272}
6273
f0486c68
YZ
6274static struct btrfs_block_rsv *
6275use_block_rsv(struct btrfs_trans_handle *trans,
6276 struct btrfs_root *root, u32 blocksize)
6277{
6278 struct btrfs_block_rsv *block_rsv;
68a82277 6279 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
f0486c68
YZ
6280 int ret;
6281
6282 block_rsv = get_block_rsv(trans, root);
6283
6284 if (block_rsv->size == 0) {
36ba022a 6285 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
68a82277
JB
6286 /*
6287 * If we couldn't reserve metadata bytes try and use some from
6288 * the global reserve.
6289 */
6290 if (ret && block_rsv != global_rsv) {
6291 ret = block_rsv_use_bytes(global_rsv, blocksize);
6292 if (!ret)
6293 return global_rsv;
f0486c68 6294 return ERR_PTR(ret);
68a82277 6295 } else if (ret) {
f0486c68 6296 return ERR_PTR(ret);
68a82277 6297 }
f0486c68
YZ
6298 return block_rsv;
6299 }
6300
6301 ret = block_rsv_use_bytes(block_rsv, blocksize);
6302 if (!ret)
6303 return block_rsv;
68a82277 6304 if (ret) {
dff51cd1
DS
6305 static DEFINE_RATELIMIT_STATE(_rs,
6306 DEFAULT_RATELIMIT_INTERVAL,
6307 /*DEFAULT_RATELIMIT_BURST*/ 2);
6308 if (__ratelimit(&_rs)) {
6309 printk(KERN_DEBUG "btrfs: block rsv returned %d\n", ret);
6310 WARN_ON(1);
6311 }
36ba022a 6312 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
68a82277 6313 if (!ret) {
68a82277
JB
6314 return block_rsv;
6315 } else if (ret && block_rsv != global_rsv) {
6316 ret = block_rsv_use_bytes(global_rsv, blocksize);
6317 if (!ret)
6318 return global_rsv;
6319 }
6320 }
f0486c68 6321
f0486c68
YZ
6322 return ERR_PTR(-ENOSPC);
6323}
6324
8c2a3ca2
JB
6325static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6326 struct btrfs_block_rsv *block_rsv, u32 blocksize)
f0486c68
YZ
6327{
6328 block_rsv_add_bytes(block_rsv, blocksize, 0);
8c2a3ca2 6329 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
f0486c68
YZ
6330}
6331
fec577fb 6332/*
f0486c68
YZ
6333 * finds a free extent and does all the dirty work required for allocation
6334 * returns the key for the extent through ins, and a tree buffer for
6335 * the first block of the extent through buf.
6336 *
fec577fb
CM
6337 * returns the tree buffer or NULL.
6338 */
5f39d397 6339struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5d4f98a2
YZ
6340 struct btrfs_root *root, u32 blocksize,
6341 u64 parent, u64 root_objectid,
6342 struct btrfs_disk_key *key, int level,
5581a51a 6343 u64 hint, u64 empty_size)
fec577fb 6344{
e2fa7227 6345 struct btrfs_key ins;
f0486c68 6346 struct btrfs_block_rsv *block_rsv;
5f39d397 6347 struct extent_buffer *buf;
f0486c68
YZ
6348 u64 flags = 0;
6349 int ret;
6350
fec577fb 6351
f0486c68
YZ
6352 block_rsv = use_block_rsv(trans, root, blocksize);
6353 if (IS_ERR(block_rsv))
6354 return ERR_CAST(block_rsv);
6355
6356 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
81c9ad23 6357 empty_size, hint, &ins, 0);
fec577fb 6358 if (ret) {
8c2a3ca2 6359 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
54aa1f4d 6360 return ERR_PTR(ret);
fec577fb 6361 }
55c69072 6362
4008c04a
CM
6363 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6364 blocksize, level);
79787eaa 6365 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
f0486c68
YZ
6366
6367 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6368 if (parent == 0)
6369 parent = ins.objectid;
6370 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6371 } else
6372 BUG_ON(parent > 0);
6373
6374 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6375 struct btrfs_delayed_extent_op *extent_op;
6376 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
79787eaa 6377 BUG_ON(!extent_op); /* -ENOMEM */
f0486c68
YZ
6378 if (key)
6379 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6380 else
6381 memset(&extent_op->key, 0, sizeof(extent_op->key));
6382 extent_op->flags_to_set = flags;
6383 extent_op->update_key = 1;
6384 extent_op->update_flags = 1;
6385 extent_op->is_data = 0;
6386
66d7e7f0
AJ
6387 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6388 ins.objectid,
f0486c68
YZ
6389 ins.offset, parent, root_objectid,
6390 level, BTRFS_ADD_DELAYED_EXTENT,
5581a51a 6391 extent_op, 0);
79787eaa 6392 BUG_ON(ret); /* -ENOMEM */
f0486c68 6393 }
fec577fb
CM
6394 return buf;
6395}
a28ec197 6396
2c47e605
YZ
6397struct walk_control {
6398 u64 refs[BTRFS_MAX_LEVEL];
6399 u64 flags[BTRFS_MAX_LEVEL];
6400 struct btrfs_key update_progress;
6401 int stage;
6402 int level;
6403 int shared_level;
6404 int update_ref;
6405 int keep_locks;
1c4850e2
YZ
6406 int reada_slot;
6407 int reada_count;
66d7e7f0 6408 int for_reloc;
2c47e605
YZ
6409};
6410
6411#define DROP_REFERENCE 1
6412#define UPDATE_BACKREF 2
6413
1c4850e2
YZ
6414static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6415 struct btrfs_root *root,
6416 struct walk_control *wc,
6417 struct btrfs_path *path)
6407bf6d 6418{
1c4850e2
YZ
6419 u64 bytenr;
6420 u64 generation;
6421 u64 refs;
94fcca9f 6422 u64 flags;
5d4f98a2 6423 u32 nritems;
1c4850e2
YZ
6424 u32 blocksize;
6425 struct btrfs_key key;
6426 struct extent_buffer *eb;
6407bf6d 6427 int ret;
1c4850e2
YZ
6428 int slot;
6429 int nread = 0;
6407bf6d 6430
1c4850e2
YZ
6431 if (path->slots[wc->level] < wc->reada_slot) {
6432 wc->reada_count = wc->reada_count * 2 / 3;
6433 wc->reada_count = max(wc->reada_count, 2);
6434 } else {
6435 wc->reada_count = wc->reada_count * 3 / 2;
6436 wc->reada_count = min_t(int, wc->reada_count,
6437 BTRFS_NODEPTRS_PER_BLOCK(root));
6438 }
7bb86316 6439
1c4850e2
YZ
6440 eb = path->nodes[wc->level];
6441 nritems = btrfs_header_nritems(eb);
6442 blocksize = btrfs_level_size(root, wc->level - 1);
bd56b302 6443
1c4850e2
YZ
6444 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6445 if (nread >= wc->reada_count)
6446 break;
bd56b302 6447
2dd3e67b 6448 cond_resched();
1c4850e2
YZ
6449 bytenr = btrfs_node_blockptr(eb, slot);
6450 generation = btrfs_node_ptr_generation(eb, slot);
2dd3e67b 6451
1c4850e2
YZ
6452 if (slot == path->slots[wc->level])
6453 goto reada;
5d4f98a2 6454
1c4850e2
YZ
6455 if (wc->stage == UPDATE_BACKREF &&
6456 generation <= root->root_key.offset)
bd56b302
CM
6457 continue;
6458
94fcca9f
YZ
6459 /* We don't lock the tree block, it's OK to be racy here */
6460 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6461 &refs, &flags);
79787eaa
JM
6462 /* We don't care about errors in readahead. */
6463 if (ret < 0)
6464 continue;
94fcca9f
YZ
6465 BUG_ON(refs == 0);
6466
1c4850e2 6467 if (wc->stage == DROP_REFERENCE) {
1c4850e2
YZ
6468 if (refs == 1)
6469 goto reada;
bd56b302 6470
94fcca9f
YZ
6471 if (wc->level == 1 &&
6472 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6473 continue;
1c4850e2
YZ
6474 if (!wc->update_ref ||
6475 generation <= root->root_key.offset)
6476 continue;
6477 btrfs_node_key_to_cpu(eb, &key, slot);
6478 ret = btrfs_comp_cpu_keys(&key,
6479 &wc->update_progress);
6480 if (ret < 0)
6481 continue;
94fcca9f
YZ
6482 } else {
6483 if (wc->level == 1 &&
6484 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6485 continue;
6407bf6d 6486 }
1c4850e2
YZ
6487reada:
6488 ret = readahead_tree_block(root, bytenr, blocksize,
6489 generation);
6490 if (ret)
bd56b302 6491 break;
1c4850e2 6492 nread++;
20524f02 6493 }
1c4850e2 6494 wc->reada_slot = slot;
20524f02 6495}
2c47e605 6496
f82d02d9 6497/*
2c47e605
YZ
6498 * hepler to process tree block while walking down the tree.
6499 *
2c47e605
YZ
6500 * when wc->stage == UPDATE_BACKREF, this function updates
6501 * back refs for pointers in the block.
6502 *
6503 * NOTE: return value 1 means we should stop walking down.
f82d02d9 6504 */
2c47e605 6505static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5d4f98a2 6506 struct btrfs_root *root,
2c47e605 6507 struct btrfs_path *path,
94fcca9f 6508 struct walk_control *wc, int lookup_info)
f82d02d9 6509{
2c47e605
YZ
6510 int level = wc->level;
6511 struct extent_buffer *eb = path->nodes[level];
2c47e605 6512 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
f82d02d9
YZ
6513 int ret;
6514
2c47e605
YZ
6515 if (wc->stage == UPDATE_BACKREF &&
6516 btrfs_header_owner(eb) != root->root_key.objectid)
6517 return 1;
f82d02d9 6518
2c47e605
YZ
6519 /*
6520 * when reference count of tree block is 1, it won't increase
6521 * again. once full backref flag is set, we never clear it.
6522 */
94fcca9f
YZ
6523 if (lookup_info &&
6524 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6525 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
2c47e605
YZ
6526 BUG_ON(!path->locks[level]);
6527 ret = btrfs_lookup_extent_info(trans, root,
6528 eb->start, eb->len,
6529 &wc->refs[level],
6530 &wc->flags[level]);
79787eaa
JM
6531 BUG_ON(ret == -ENOMEM);
6532 if (ret)
6533 return ret;
2c47e605
YZ
6534 BUG_ON(wc->refs[level] == 0);
6535 }
5d4f98a2 6536
2c47e605
YZ
6537 if (wc->stage == DROP_REFERENCE) {
6538 if (wc->refs[level] > 1)
6539 return 1;
f82d02d9 6540
2c47e605 6541 if (path->locks[level] && !wc->keep_locks) {
bd681513 6542 btrfs_tree_unlock_rw(eb, path->locks[level]);
2c47e605
YZ
6543 path->locks[level] = 0;
6544 }
6545 return 0;
6546 }
f82d02d9 6547
2c47e605
YZ
6548 /* wc->stage == UPDATE_BACKREF */
6549 if (!(wc->flags[level] & flag)) {
6550 BUG_ON(!path->locks[level]);
66d7e7f0 6551 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
79787eaa 6552 BUG_ON(ret); /* -ENOMEM */
66d7e7f0 6553 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
79787eaa 6554 BUG_ON(ret); /* -ENOMEM */
2c47e605
YZ
6555 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6556 eb->len, flag, 0);
79787eaa 6557 BUG_ON(ret); /* -ENOMEM */
2c47e605
YZ
6558 wc->flags[level] |= flag;
6559 }
6560
6561 /*
6562 * the block is shared by multiple trees, so it's not good to
6563 * keep the tree lock
6564 */
6565 if (path->locks[level] && level > 0) {
bd681513 6566 btrfs_tree_unlock_rw(eb, path->locks[level]);
2c47e605
YZ
6567 path->locks[level] = 0;
6568 }
6569 return 0;
6570}
6571
1c4850e2
YZ
6572/*
6573 * hepler to process tree block pointer.
6574 *
6575 * when wc->stage == DROP_REFERENCE, this function checks
6576 * reference count of the block pointed to. if the block
6577 * is shared and we need update back refs for the subtree
6578 * rooted at the block, this function changes wc->stage to
6579 * UPDATE_BACKREF. if the block is shared and there is no
6580 * need to update back, this function drops the reference
6581 * to the block.
6582 *
6583 * NOTE: return value 1 means we should stop walking down.
6584 */
6585static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6586 struct btrfs_root *root,
6587 struct btrfs_path *path,
94fcca9f 6588 struct walk_control *wc, int *lookup_info)
1c4850e2
YZ
6589{
6590 u64 bytenr;
6591 u64 generation;
6592 u64 parent;
6593 u32 blocksize;
6594 struct btrfs_key key;
6595 struct extent_buffer *next;
6596 int level = wc->level;
6597 int reada = 0;
6598 int ret = 0;
6599
6600 generation = btrfs_node_ptr_generation(path->nodes[level],
6601 path->slots[level]);
6602 /*
6603 * if the lower level block was created before the snapshot
6604 * was created, we know there is no need to update back refs
6605 * for the subtree
6606 */
6607 if (wc->stage == UPDATE_BACKREF &&
94fcca9f
YZ
6608 generation <= root->root_key.offset) {
6609 *lookup_info = 1;
1c4850e2 6610 return 1;
94fcca9f 6611 }
1c4850e2
YZ
6612
6613 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6614 blocksize = btrfs_level_size(root, level - 1);
6615
6616 next = btrfs_find_tree_block(root, bytenr, blocksize);
6617 if (!next) {
6618 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
90d2c51d
MX
6619 if (!next)
6620 return -ENOMEM;
1c4850e2
YZ
6621 reada = 1;
6622 }
6623 btrfs_tree_lock(next);
6624 btrfs_set_lock_blocking(next);
6625
94fcca9f
YZ
6626 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6627 &wc->refs[level - 1],
6628 &wc->flags[level - 1]);
79787eaa
JM
6629 if (ret < 0) {
6630 btrfs_tree_unlock(next);
6631 return ret;
6632 }
6633
94fcca9f
YZ
6634 BUG_ON(wc->refs[level - 1] == 0);
6635 *lookup_info = 0;
1c4850e2 6636
94fcca9f 6637 if (wc->stage == DROP_REFERENCE) {
1c4850e2 6638 if (wc->refs[level - 1] > 1) {
94fcca9f
YZ
6639 if (level == 1 &&
6640 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6641 goto skip;
6642
1c4850e2
YZ
6643 if (!wc->update_ref ||
6644 generation <= root->root_key.offset)
6645 goto skip;
6646
6647 btrfs_node_key_to_cpu(path->nodes[level], &key,
6648 path->slots[level]);
6649 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6650 if (ret < 0)
6651 goto skip;
6652
6653 wc->stage = UPDATE_BACKREF;
6654 wc->shared_level = level - 1;
6655 }
94fcca9f
YZ
6656 } else {
6657 if (level == 1 &&
6658 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6659 goto skip;
1c4850e2
YZ
6660 }
6661
b9fab919 6662 if (!btrfs_buffer_uptodate(next, generation, 0)) {
1c4850e2
YZ
6663 btrfs_tree_unlock(next);
6664 free_extent_buffer(next);
6665 next = NULL;
94fcca9f 6666 *lookup_info = 1;
1c4850e2
YZ
6667 }
6668
6669 if (!next) {
6670 if (reada && level == 1)
6671 reada_walk_down(trans, root, wc, path);
6672 next = read_tree_block(root, bytenr, blocksize, generation);
97d9a8a4
TI
6673 if (!next)
6674 return -EIO;
1c4850e2
YZ
6675 btrfs_tree_lock(next);
6676 btrfs_set_lock_blocking(next);
6677 }
6678
6679 level--;
6680 BUG_ON(level != btrfs_header_level(next));
6681 path->nodes[level] = next;
6682 path->slots[level] = 0;
bd681513 6683 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
1c4850e2
YZ
6684 wc->level = level;
6685 if (wc->level == 1)
6686 wc->reada_slot = 0;
6687 return 0;
6688skip:
6689 wc->refs[level - 1] = 0;
6690 wc->flags[level - 1] = 0;
94fcca9f
YZ
6691 if (wc->stage == DROP_REFERENCE) {
6692 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6693 parent = path->nodes[level]->start;
6694 } else {
6695 BUG_ON(root->root_key.objectid !=
6696 btrfs_header_owner(path->nodes[level]));
6697 parent = 0;
6698 }
1c4850e2 6699
94fcca9f 6700 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
66d7e7f0 6701 root->root_key.objectid, level - 1, 0, 0);
79787eaa 6702 BUG_ON(ret); /* -ENOMEM */
1c4850e2 6703 }
1c4850e2
YZ
6704 btrfs_tree_unlock(next);
6705 free_extent_buffer(next);
94fcca9f 6706 *lookup_info = 1;
1c4850e2
YZ
6707 return 1;
6708}
6709
2c47e605
YZ
6710/*
6711 * hepler to process tree block while walking up the tree.
6712 *
6713 * when wc->stage == DROP_REFERENCE, this function drops
6714 * reference count on the block.
6715 *
6716 * when wc->stage == UPDATE_BACKREF, this function changes
6717 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6718 * to UPDATE_BACKREF previously while processing the block.
6719 *
6720 * NOTE: return value 1 means we should stop walking up.
6721 */
6722static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6723 struct btrfs_root *root,
6724 struct btrfs_path *path,
6725 struct walk_control *wc)
6726{
f0486c68 6727 int ret;
2c47e605
YZ
6728 int level = wc->level;
6729 struct extent_buffer *eb = path->nodes[level];
6730 u64 parent = 0;
6731
6732 if (wc->stage == UPDATE_BACKREF) {
6733 BUG_ON(wc->shared_level < level);
6734 if (level < wc->shared_level)
6735 goto out;
6736
2c47e605
YZ
6737 ret = find_next_key(path, level + 1, &wc->update_progress);
6738 if (ret > 0)
6739 wc->update_ref = 0;
6740
6741 wc->stage = DROP_REFERENCE;
6742 wc->shared_level = -1;
6743 path->slots[level] = 0;
6744
6745 /*
6746 * check reference count again if the block isn't locked.
6747 * we should start walking down the tree again if reference
6748 * count is one.
6749 */
6750 if (!path->locks[level]) {
6751 BUG_ON(level == 0);
6752 btrfs_tree_lock(eb);
6753 btrfs_set_lock_blocking(eb);
bd681513 6754 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
2c47e605
YZ
6755
6756 ret = btrfs_lookup_extent_info(trans, root,
6757 eb->start, eb->len,
6758 &wc->refs[level],
6759 &wc->flags[level]);
79787eaa
JM
6760 if (ret < 0) {
6761 btrfs_tree_unlock_rw(eb, path->locks[level]);
6762 return ret;
6763 }
2c47e605
YZ
6764 BUG_ON(wc->refs[level] == 0);
6765 if (wc->refs[level] == 1) {
bd681513 6766 btrfs_tree_unlock_rw(eb, path->locks[level]);
2c47e605
YZ
6767 return 1;
6768 }
f82d02d9 6769 }
2c47e605 6770 }
f82d02d9 6771
2c47e605
YZ
6772 /* wc->stage == DROP_REFERENCE */
6773 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5d4f98a2 6774
2c47e605
YZ
6775 if (wc->refs[level] == 1) {
6776 if (level == 0) {
6777 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
66d7e7f0
AJ
6778 ret = btrfs_dec_ref(trans, root, eb, 1,
6779 wc->for_reloc);
2c47e605 6780 else
66d7e7f0
AJ
6781 ret = btrfs_dec_ref(trans, root, eb, 0,
6782 wc->for_reloc);
79787eaa 6783 BUG_ON(ret); /* -ENOMEM */
2c47e605
YZ
6784 }
6785 /* make block locked assertion in clean_tree_block happy */
6786 if (!path->locks[level] &&
6787 btrfs_header_generation(eb) == trans->transid) {
6788 btrfs_tree_lock(eb);
6789 btrfs_set_lock_blocking(eb);
bd681513 6790 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
2c47e605
YZ
6791 }
6792 clean_tree_block(trans, root, eb);
6793 }
6794
6795 if (eb == root->node) {
6796 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6797 parent = eb->start;
6798 else
6799 BUG_ON(root->root_key.objectid !=
6800 btrfs_header_owner(eb));
6801 } else {
6802 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6803 parent = path->nodes[level + 1]->start;
6804 else
6805 BUG_ON(root->root_key.objectid !=
6806 btrfs_header_owner(path->nodes[level + 1]));
f82d02d9 6807 }
f82d02d9 6808
5581a51a 6809 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
2c47e605
YZ
6810out:
6811 wc->refs[level] = 0;
6812 wc->flags[level] = 0;
f0486c68 6813 return 0;
2c47e605
YZ
6814}
6815
6816static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6817 struct btrfs_root *root,
6818 struct btrfs_path *path,
6819 struct walk_control *wc)
6820{
2c47e605 6821 int level = wc->level;
94fcca9f 6822 int lookup_info = 1;
2c47e605
YZ
6823 int ret;
6824
6825 while (level >= 0) {
94fcca9f 6826 ret = walk_down_proc(trans, root, path, wc, lookup_info);
2c47e605
YZ
6827 if (ret > 0)
6828 break;
6829
6830 if (level == 0)
6831 break;
6832
7a7965f8
YZ
6833 if (path->slots[level] >=
6834 btrfs_header_nritems(path->nodes[level]))
6835 break;
6836
94fcca9f 6837 ret = do_walk_down(trans, root, path, wc, &lookup_info);
1c4850e2
YZ
6838 if (ret > 0) {
6839 path->slots[level]++;
6840 continue;
90d2c51d
MX
6841 } else if (ret < 0)
6842 return ret;
1c4850e2 6843 level = wc->level;
f82d02d9 6844 }
f82d02d9
YZ
6845 return 0;
6846}
6847
d397712b 6848static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
98ed5174 6849 struct btrfs_root *root,
f82d02d9 6850 struct btrfs_path *path,
2c47e605 6851 struct walk_control *wc, int max_level)
20524f02 6852{
2c47e605 6853 int level = wc->level;
20524f02 6854 int ret;
9f3a7427 6855
2c47e605
YZ
6856 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6857 while (level < max_level && path->nodes[level]) {
6858 wc->level = level;
6859 if (path->slots[level] + 1 <
6860 btrfs_header_nritems(path->nodes[level])) {
6861 path->slots[level]++;
20524f02
CM
6862 return 0;
6863 } else {
2c47e605
YZ
6864 ret = walk_up_proc(trans, root, path, wc);
6865 if (ret > 0)
6866 return 0;
bd56b302 6867
2c47e605 6868 if (path->locks[level]) {
bd681513
CM
6869 btrfs_tree_unlock_rw(path->nodes[level],
6870 path->locks[level]);
2c47e605 6871 path->locks[level] = 0;
f82d02d9 6872 }
2c47e605
YZ
6873 free_extent_buffer(path->nodes[level]);
6874 path->nodes[level] = NULL;
6875 level++;
20524f02
CM
6876 }
6877 }
6878 return 1;
6879}
6880
9aca1d51 6881/*
2c47e605
YZ
6882 * drop a subvolume tree.
6883 *
6884 * this function traverses the tree freeing any blocks that only
6885 * referenced by the tree.
6886 *
6887 * when a shared tree block is found. this function decreases its
6888 * reference count by one. if update_ref is true, this function
6889 * also make sure backrefs for the shared block and all lower level
6890 * blocks are properly updated.
9aca1d51 6891 */
2c536799 6892int btrfs_drop_snapshot(struct btrfs_root *root,
66d7e7f0
AJ
6893 struct btrfs_block_rsv *block_rsv, int update_ref,
6894 int for_reloc)
20524f02 6895{
5caf2a00 6896 struct btrfs_path *path;
2c47e605
YZ
6897 struct btrfs_trans_handle *trans;
6898 struct btrfs_root *tree_root = root->fs_info->tree_root;
9f3a7427 6899 struct btrfs_root_item *root_item = &root->root_item;
2c47e605
YZ
6900 struct walk_control *wc;
6901 struct btrfs_key key;
6902 int err = 0;
6903 int ret;
6904 int level;
20524f02 6905
5caf2a00 6906 path = btrfs_alloc_path();
cb1b69f4
TI
6907 if (!path) {
6908 err = -ENOMEM;
6909 goto out;
6910 }
20524f02 6911
2c47e605 6912 wc = kzalloc(sizeof(*wc), GFP_NOFS);
38a1a919
MF
6913 if (!wc) {
6914 btrfs_free_path(path);
cb1b69f4
TI
6915 err = -ENOMEM;
6916 goto out;
38a1a919 6917 }
2c47e605 6918
a22285a6 6919 trans = btrfs_start_transaction(tree_root, 0);
79787eaa
JM
6920 if (IS_ERR(trans)) {
6921 err = PTR_ERR(trans);
6922 goto out_free;
6923 }
98d5dc13 6924
3fd0a558
YZ
6925 if (block_rsv)
6926 trans->block_rsv = block_rsv;
2c47e605 6927
9f3a7427 6928 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2c47e605 6929 level = btrfs_header_level(root->node);
5d4f98a2
YZ
6930 path->nodes[level] = btrfs_lock_root_node(root);
6931 btrfs_set_lock_blocking(path->nodes[level]);
9f3a7427 6932 path->slots[level] = 0;
bd681513 6933 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
2c47e605
YZ
6934 memset(&wc->update_progress, 0,
6935 sizeof(wc->update_progress));
9f3a7427 6936 } else {
9f3a7427 6937 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2c47e605
YZ
6938 memcpy(&wc->update_progress, &key,
6939 sizeof(wc->update_progress));
6940
6702ed49 6941 level = root_item->drop_level;
2c47e605 6942 BUG_ON(level == 0);
6702ed49 6943 path->lowest_level = level;
2c47e605
YZ
6944 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6945 path->lowest_level = 0;
6946 if (ret < 0) {
6947 err = ret;
79787eaa 6948 goto out_end_trans;
9f3a7427 6949 }
1c4850e2 6950 WARN_ON(ret > 0);
2c47e605 6951
7d9eb12c
CM
6952 /*
6953 * unlock our path, this is safe because only this
6954 * function is allowed to delete this snapshot
6955 */
5d4f98a2 6956 btrfs_unlock_up_safe(path, 0);
2c47e605
YZ
6957
6958 level = btrfs_header_level(root->node);
6959 while (1) {
6960 btrfs_tree_lock(path->nodes[level]);
6961 btrfs_set_lock_blocking(path->nodes[level]);
6962
6963 ret = btrfs_lookup_extent_info(trans, root,
6964 path->nodes[level]->start,
6965 path->nodes[level]->len,
6966 &wc->refs[level],
6967 &wc->flags[level]);
79787eaa
JM
6968 if (ret < 0) {
6969 err = ret;
6970 goto out_end_trans;
6971 }
2c47e605
YZ
6972 BUG_ON(wc->refs[level] == 0);
6973
6974 if (level == root_item->drop_level)
6975 break;
6976
6977 btrfs_tree_unlock(path->nodes[level]);
6978 WARN_ON(wc->refs[level] != 1);
6979 level--;
6980 }
9f3a7427 6981 }
2c47e605
YZ
6982
6983 wc->level = level;
6984 wc->shared_level = -1;
6985 wc->stage = DROP_REFERENCE;
6986 wc->update_ref = update_ref;
6987 wc->keep_locks = 0;
66d7e7f0 6988 wc->for_reloc = for_reloc;
1c4850e2 6989 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
2c47e605 6990
d397712b 6991 while (1) {
2c47e605
YZ
6992 ret = walk_down_tree(trans, root, path, wc);
6993 if (ret < 0) {
6994 err = ret;
20524f02 6995 break;
2c47e605 6996 }
9aca1d51 6997
2c47e605
YZ
6998 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6999 if (ret < 0) {
7000 err = ret;
20524f02 7001 break;
2c47e605
YZ
7002 }
7003
7004 if (ret > 0) {
7005 BUG_ON(wc->stage != DROP_REFERENCE);
e7a84565
CM
7006 break;
7007 }
2c47e605
YZ
7008
7009 if (wc->stage == DROP_REFERENCE) {
7010 level = wc->level;
7011 btrfs_node_key(path->nodes[level],
7012 &root_item->drop_progress,
7013 path->slots[level]);
7014 root_item->drop_level = level;
7015 }
7016
7017 BUG_ON(wc->level == 0);
3fd0a558 7018 if (btrfs_should_end_transaction(trans, tree_root)) {
2c47e605
YZ
7019 ret = btrfs_update_root(trans, tree_root,
7020 &root->root_key,
7021 root_item);
79787eaa
JM
7022 if (ret) {
7023 btrfs_abort_transaction(trans, tree_root, ret);
7024 err = ret;
7025 goto out_end_trans;
7026 }
2c47e605 7027
3fd0a558 7028 btrfs_end_transaction_throttle(trans, tree_root);
a22285a6 7029 trans = btrfs_start_transaction(tree_root, 0);
79787eaa
JM
7030 if (IS_ERR(trans)) {
7031 err = PTR_ERR(trans);
7032 goto out_free;
7033 }
3fd0a558
YZ
7034 if (block_rsv)
7035 trans->block_rsv = block_rsv;
c3e69d58 7036 }
20524f02 7037 }
b3b4aa74 7038 btrfs_release_path(path);
79787eaa
JM
7039 if (err)
7040 goto out_end_trans;
2c47e605
YZ
7041
7042 ret = btrfs_del_root(trans, tree_root, &root->root_key);
79787eaa
JM
7043 if (ret) {
7044 btrfs_abort_transaction(trans, tree_root, ret);
7045 goto out_end_trans;
7046 }
2c47e605 7047
76dda93c
YZ
7048 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7049 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
7050 NULL, NULL);
79787eaa
JM
7051 if (ret < 0) {
7052 btrfs_abort_transaction(trans, tree_root, ret);
7053 err = ret;
7054 goto out_end_trans;
7055 } else if (ret > 0) {
84cd948c
JB
7056 /* if we fail to delete the orphan item this time
7057 * around, it'll get picked up the next time.
7058 *
7059 * The most common failure here is just -ENOENT.
7060 */
7061 btrfs_del_orphan_item(trans, tree_root,
7062 root->root_key.objectid);
76dda93c
YZ
7063 }
7064 }
7065
7066 if (root->in_radix) {
7067 btrfs_free_fs_root(tree_root->fs_info, root);
7068 } else {
7069 free_extent_buffer(root->node);
7070 free_extent_buffer(root->commit_root);
7071 kfree(root);
7072 }
79787eaa 7073out_end_trans:
3fd0a558 7074 btrfs_end_transaction_throttle(trans, tree_root);
79787eaa 7075out_free:
2c47e605 7076 kfree(wc);
5caf2a00 7077 btrfs_free_path(path);
cb1b69f4
TI
7078out:
7079 if (err)
7080 btrfs_std_error(root->fs_info, err);
2c536799 7081 return err;
20524f02 7082}
9078a3e1 7083
2c47e605
YZ
7084/*
7085 * drop subtree rooted at tree block 'node'.
7086 *
7087 * NOTE: this function will unlock and release tree block 'node'
66d7e7f0 7088 * only used by relocation code
2c47e605 7089 */
f82d02d9
YZ
7090int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7091 struct btrfs_root *root,
7092 struct extent_buffer *node,
7093 struct extent_buffer *parent)
7094{
7095 struct btrfs_path *path;
2c47e605 7096 struct walk_control *wc;
f82d02d9
YZ
7097 int level;
7098 int parent_level;
7099 int ret = 0;
7100 int wret;
7101
2c47e605
YZ
7102 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7103
f82d02d9 7104 path = btrfs_alloc_path();
db5b493a
TI
7105 if (!path)
7106 return -ENOMEM;
f82d02d9 7107
2c47e605 7108 wc = kzalloc(sizeof(*wc), GFP_NOFS);
db5b493a
TI
7109 if (!wc) {
7110 btrfs_free_path(path);
7111 return -ENOMEM;
7112 }
2c47e605 7113
b9447ef8 7114 btrfs_assert_tree_locked(parent);
f82d02d9
YZ
7115 parent_level = btrfs_header_level(parent);
7116 extent_buffer_get(parent);
7117 path->nodes[parent_level] = parent;
7118 path->slots[parent_level] = btrfs_header_nritems(parent);
7119
b9447ef8 7120 btrfs_assert_tree_locked(node);
f82d02d9 7121 level = btrfs_header_level(node);
f82d02d9
YZ
7122 path->nodes[level] = node;
7123 path->slots[level] = 0;
bd681513 7124 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
2c47e605
YZ
7125
7126 wc->refs[parent_level] = 1;
7127 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7128 wc->level = level;
7129 wc->shared_level = -1;
7130 wc->stage = DROP_REFERENCE;
7131 wc->update_ref = 0;
7132 wc->keep_locks = 1;
66d7e7f0 7133 wc->for_reloc = 1;
1c4850e2 7134 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
f82d02d9
YZ
7135
7136 while (1) {
2c47e605
YZ
7137 wret = walk_down_tree(trans, root, path, wc);
7138 if (wret < 0) {
f82d02d9 7139 ret = wret;
f82d02d9 7140 break;
2c47e605 7141 }
f82d02d9 7142
2c47e605 7143 wret = walk_up_tree(trans, root, path, wc, parent_level);
f82d02d9
YZ
7144 if (wret < 0)
7145 ret = wret;
7146 if (wret != 0)
7147 break;
7148 }
7149
2c47e605 7150 kfree(wc);
f82d02d9
YZ
7151 btrfs_free_path(path);
7152 return ret;
7153}
7154
ec44a35c
CM
7155static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7156{
7157 u64 num_devices;
fc67c450 7158 u64 stripped;
e4d8ec0f 7159
fc67c450
ID
7160 /*
7161 * if restripe for this chunk_type is on pick target profile and
7162 * return, otherwise do the usual balance
7163 */
7164 stripped = get_restripe_target(root->fs_info, flags);
7165 if (stripped)
7166 return extended_to_chunk(stripped);
e4d8ec0f 7167
cd02dca5
CM
7168 /*
7169 * we add in the count of missing devices because we want
7170 * to make sure that any RAID levels on a degraded FS
7171 * continue to be honored.
7172 */
7173 num_devices = root->fs_info->fs_devices->rw_devices +
7174 root->fs_info->fs_devices->missing_devices;
7175
fc67c450
ID
7176 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7177 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7178
ec44a35c
CM
7179 if (num_devices == 1) {
7180 stripped |= BTRFS_BLOCK_GROUP_DUP;
7181 stripped = flags & ~stripped;
7182
7183 /* turn raid0 into single device chunks */
7184 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7185 return stripped;
7186
7187 /* turn mirroring into duplication */
7188 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7189 BTRFS_BLOCK_GROUP_RAID10))
7190 return stripped | BTRFS_BLOCK_GROUP_DUP;
ec44a35c
CM
7191 } else {
7192 /* they already had raid on here, just return */
ec44a35c
CM
7193 if (flags & stripped)
7194 return flags;
7195
7196 stripped |= BTRFS_BLOCK_GROUP_DUP;
7197 stripped = flags & ~stripped;
7198
7199 /* switch duplicated blocks with raid1 */
7200 if (flags & BTRFS_BLOCK_GROUP_DUP)
7201 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7202
e3176ca2 7203 /* this is drive concat, leave it alone */
ec44a35c 7204 }
e3176ca2 7205
ec44a35c
CM
7206 return flags;
7207}
7208
199c36ea 7209static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
0ef3e66b 7210{
f0486c68
YZ
7211 struct btrfs_space_info *sinfo = cache->space_info;
7212 u64 num_bytes;
199c36ea 7213 u64 min_allocable_bytes;
f0486c68 7214 int ret = -ENOSPC;
0ef3e66b 7215
c286ac48 7216
199c36ea
MX
7217 /*
7218 * We need some metadata space and system metadata space for
7219 * allocating chunks in some corner cases until we force to set
7220 * it to be readonly.
7221 */
7222 if ((sinfo->flags &
7223 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7224 !force)
7225 min_allocable_bytes = 1 * 1024 * 1024;
7226 else
7227 min_allocable_bytes = 0;
7228
f0486c68
YZ
7229 spin_lock(&sinfo->lock);
7230 spin_lock(&cache->lock);
61cfea9b
W
7231
7232 if (cache->ro) {
7233 ret = 0;
7234 goto out;
7235 }
7236
f0486c68
YZ
7237 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7238 cache->bytes_super - btrfs_block_group_used(&cache->item);
7239
7240 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
37be25bc
JB
7241 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7242 min_allocable_bytes <= sinfo->total_bytes) {
f0486c68 7243 sinfo->bytes_readonly += num_bytes;
f0486c68
YZ
7244 cache->ro = 1;
7245 ret = 0;
7246 }
61cfea9b 7247out:
f0486c68
YZ
7248 spin_unlock(&cache->lock);
7249 spin_unlock(&sinfo->lock);
7250 return ret;
7251}
7d9eb12c 7252
f0486c68
YZ
7253int btrfs_set_block_group_ro(struct btrfs_root *root,
7254 struct btrfs_block_group_cache *cache)
c286ac48 7255
f0486c68
YZ
7256{
7257 struct btrfs_trans_handle *trans;
7258 u64 alloc_flags;
7259 int ret;
7d9eb12c 7260
f0486c68 7261 BUG_ON(cache->ro);
0ef3e66b 7262
ff5714cc 7263 trans = btrfs_join_transaction(root);
79787eaa
JM
7264 if (IS_ERR(trans))
7265 return PTR_ERR(trans);
5d4f98a2 7266
f0486c68 7267 alloc_flags = update_block_group_flags(root, cache->flags);
79787eaa
JM
7268 if (alloc_flags != cache->flags) {
7269 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7270 CHUNK_ALLOC_FORCE);
7271 if (ret < 0)
7272 goto out;
7273 }
5d4f98a2 7274
199c36ea 7275 ret = set_block_group_ro(cache, 0);
f0486c68
YZ
7276 if (!ret)
7277 goto out;
7278 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
0e4f8f88
CM
7279 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7280 CHUNK_ALLOC_FORCE);
f0486c68
YZ
7281 if (ret < 0)
7282 goto out;
199c36ea 7283 ret = set_block_group_ro(cache, 0);
f0486c68
YZ
7284out:
7285 btrfs_end_transaction(trans, root);
7286 return ret;
7287}
5d4f98a2 7288
c87f08ca
CM
7289int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7290 struct btrfs_root *root, u64 type)
7291{
7292 u64 alloc_flags = get_alloc_profile(root, type);
0e4f8f88
CM
7293 return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7294 CHUNK_ALLOC_FORCE);
c87f08ca
CM
7295}
7296
6d07bcec
MX
7297/*
7298 * helper to account the unused space of all the readonly block group in the
7299 * list. takes mirrors into account.
7300 */
7301static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7302{
7303 struct btrfs_block_group_cache *block_group;
7304 u64 free_bytes = 0;
7305 int factor;
7306
7307 list_for_each_entry(block_group, groups_list, list) {
7308 spin_lock(&block_group->lock);
7309
7310 if (!block_group->ro) {
7311 spin_unlock(&block_group->lock);
7312 continue;
7313 }
7314
7315 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7316 BTRFS_BLOCK_GROUP_RAID10 |
7317 BTRFS_BLOCK_GROUP_DUP))
7318 factor = 2;
7319 else
7320 factor = 1;
7321
7322 free_bytes += (block_group->key.offset -
7323 btrfs_block_group_used(&block_group->item)) *
7324 factor;
7325
7326 spin_unlock(&block_group->lock);
7327 }
7328
7329 return free_bytes;
7330}
7331
7332/*
7333 * helper to account the unused space of all the readonly block group in the
7334 * space_info. takes mirrors into account.
7335 */
7336u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7337{
7338 int i;
7339 u64 free_bytes = 0;
7340
7341 spin_lock(&sinfo->lock);
7342
7343 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7344 if (!list_empty(&sinfo->block_groups[i]))
7345 free_bytes += __btrfs_get_ro_block_group_free_space(
7346 &sinfo->block_groups[i]);
7347
7348 spin_unlock(&sinfo->lock);
7349
7350 return free_bytes;
7351}
7352
143bede5 7353void btrfs_set_block_group_rw(struct btrfs_root *root,
f0486c68 7354 struct btrfs_block_group_cache *cache)
5d4f98a2 7355{
f0486c68
YZ
7356 struct btrfs_space_info *sinfo = cache->space_info;
7357 u64 num_bytes;
7358
7359 BUG_ON(!cache->ro);
7360
7361 spin_lock(&sinfo->lock);
7362 spin_lock(&cache->lock);
7363 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7364 cache->bytes_super - btrfs_block_group_used(&cache->item);
7365 sinfo->bytes_readonly -= num_bytes;
7366 cache->ro = 0;
7367 spin_unlock(&cache->lock);
7368 spin_unlock(&sinfo->lock);
5d4f98a2
YZ
7369}
7370
ba1bf481
JB
7371/*
7372 * checks to see if its even possible to relocate this block group.
7373 *
7374 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7375 * ok to go ahead and try.
7376 */
7377int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
1a40e23b 7378{
ba1bf481
JB
7379 struct btrfs_block_group_cache *block_group;
7380 struct btrfs_space_info *space_info;
7381 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7382 struct btrfs_device *device;
cdcb725c 7383 u64 min_free;
6719db6a
JB
7384 u64 dev_min = 1;
7385 u64 dev_nr = 0;
4a5e98f5 7386 u64 target;
cdcb725c 7387 int index;
ba1bf481
JB
7388 int full = 0;
7389 int ret = 0;
1a40e23b 7390
ba1bf481 7391 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
1a40e23b 7392
ba1bf481
JB
7393 /* odd, couldn't find the block group, leave it alone */
7394 if (!block_group)
7395 return -1;
1a40e23b 7396
cdcb725c 7397 min_free = btrfs_block_group_used(&block_group->item);
7398
ba1bf481 7399 /* no bytes used, we're good */
cdcb725c 7400 if (!min_free)
1a40e23b
ZY
7401 goto out;
7402
ba1bf481
JB
7403 space_info = block_group->space_info;
7404 spin_lock(&space_info->lock);
17d217fe 7405
ba1bf481 7406 full = space_info->full;
17d217fe 7407
ba1bf481
JB
7408 /*
7409 * if this is the last block group we have in this space, we can't
7ce618db
CM
7410 * relocate it unless we're able to allocate a new chunk below.
7411 *
7412 * Otherwise, we need to make sure we have room in the space to handle
7413 * all of the extents from this block group. If we can, we're good
ba1bf481 7414 */
7ce618db 7415 if ((space_info->total_bytes != block_group->key.offset) &&
cdcb725c 7416 (space_info->bytes_used + space_info->bytes_reserved +
7417 space_info->bytes_pinned + space_info->bytes_readonly +
7418 min_free < space_info->total_bytes)) {
ba1bf481
JB
7419 spin_unlock(&space_info->lock);
7420 goto out;
17d217fe 7421 }
ba1bf481 7422 spin_unlock(&space_info->lock);
ea8c2819 7423
ba1bf481
JB
7424 /*
7425 * ok we don't have enough space, but maybe we have free space on our
7426 * devices to allocate new chunks for relocation, so loop through our
4a5e98f5
ID
7427 * alloc devices and guess if we have enough space. if this block
7428 * group is going to be restriped, run checks against the target
7429 * profile instead of the current one.
ba1bf481
JB
7430 */
7431 ret = -1;
ea8c2819 7432
cdcb725c 7433 /*
7434 * index:
7435 * 0: raid10
7436 * 1: raid1
7437 * 2: dup
7438 * 3: raid0
7439 * 4: single
7440 */
4a5e98f5
ID
7441 target = get_restripe_target(root->fs_info, block_group->flags);
7442 if (target) {
7443 index = __get_block_group_index(extended_to_chunk(target));
7444 } else {
7445 /*
7446 * this is just a balance, so if we were marked as full
7447 * we know there is no space for a new chunk
7448 */
7449 if (full)
7450 goto out;
7451
7452 index = get_block_group_index(block_group);
7453 }
7454
cdcb725c 7455 if (index == 0) {
7456 dev_min = 4;
6719db6a
JB
7457 /* Divide by 2 */
7458 min_free >>= 1;
cdcb725c 7459 } else if (index == 1) {
7460 dev_min = 2;
7461 } else if (index == 2) {
6719db6a
JB
7462 /* Multiply by 2 */
7463 min_free <<= 1;
cdcb725c 7464 } else if (index == 3) {
7465 dev_min = fs_devices->rw_devices;
6719db6a 7466 do_div(min_free, dev_min);
cdcb725c 7467 }
7468
ba1bf481
JB
7469 mutex_lock(&root->fs_info->chunk_mutex);
7470 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7bfc837d 7471 u64 dev_offset;
56bec294 7472
ba1bf481
JB
7473 /*
7474 * check to make sure we can actually find a chunk with enough
7475 * space to fit our block group in.
7476 */
7477 if (device->total_bytes > device->bytes_used + min_free) {
125ccb0a 7478 ret = find_free_dev_extent(device, min_free,
7bfc837d 7479 &dev_offset, NULL);
ba1bf481 7480 if (!ret)
cdcb725c 7481 dev_nr++;
7482
7483 if (dev_nr >= dev_min)
73e48b27 7484 break;
cdcb725c 7485
ba1bf481 7486 ret = -1;
725c8463 7487 }
edbd8d4e 7488 }
ba1bf481 7489 mutex_unlock(&root->fs_info->chunk_mutex);
edbd8d4e 7490out:
ba1bf481 7491 btrfs_put_block_group(block_group);
edbd8d4e
CM
7492 return ret;
7493}
7494
b2950863
CH
7495static int find_first_block_group(struct btrfs_root *root,
7496 struct btrfs_path *path, struct btrfs_key *key)
0b86a832 7497{
925baedd 7498 int ret = 0;
0b86a832
CM
7499 struct btrfs_key found_key;
7500 struct extent_buffer *leaf;
7501 int slot;
edbd8d4e 7502
0b86a832
CM
7503 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7504 if (ret < 0)
925baedd
CM
7505 goto out;
7506
d397712b 7507 while (1) {
0b86a832 7508 slot = path->slots[0];
edbd8d4e 7509 leaf = path->nodes[0];
0b86a832
CM
7510 if (slot >= btrfs_header_nritems(leaf)) {
7511 ret = btrfs_next_leaf(root, path);
7512 if (ret == 0)
7513 continue;
7514 if (ret < 0)
925baedd 7515 goto out;
0b86a832 7516 break;
edbd8d4e 7517 }
0b86a832 7518 btrfs_item_key_to_cpu(leaf, &found_key, slot);
edbd8d4e 7519
0b86a832 7520 if (found_key.objectid >= key->objectid &&
925baedd
CM
7521 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7522 ret = 0;
7523 goto out;
7524 }
0b86a832 7525 path->slots[0]++;
edbd8d4e 7526 }
925baedd 7527out:
0b86a832 7528 return ret;
edbd8d4e
CM
7529}
7530
0af3d00b
JB
7531void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7532{
7533 struct btrfs_block_group_cache *block_group;
7534 u64 last = 0;
7535
7536 while (1) {
7537 struct inode *inode;
7538
7539 block_group = btrfs_lookup_first_block_group(info, last);
7540 while (block_group) {
7541 spin_lock(&block_group->lock);
7542 if (block_group->iref)
7543 break;
7544 spin_unlock(&block_group->lock);
7545 block_group = next_block_group(info->tree_root,
7546 block_group);
7547 }
7548 if (!block_group) {
7549 if (last == 0)
7550 break;
7551 last = 0;
7552 continue;
7553 }
7554
7555 inode = block_group->inode;
7556 block_group->iref = 0;
7557 block_group->inode = NULL;
7558 spin_unlock(&block_group->lock);
7559 iput(inode);
7560 last = block_group->key.objectid + block_group->key.offset;
7561 btrfs_put_block_group(block_group);
7562 }
7563}
7564
1a40e23b
ZY
7565int btrfs_free_block_groups(struct btrfs_fs_info *info)
7566{
7567 struct btrfs_block_group_cache *block_group;
4184ea7f 7568 struct btrfs_space_info *space_info;
11833d66 7569 struct btrfs_caching_control *caching_ctl;
1a40e23b
ZY
7570 struct rb_node *n;
7571
11833d66
YZ
7572 down_write(&info->extent_commit_sem);
7573 while (!list_empty(&info->caching_block_groups)) {
7574 caching_ctl = list_entry(info->caching_block_groups.next,
7575 struct btrfs_caching_control, list);
7576 list_del(&caching_ctl->list);
7577 put_caching_control(caching_ctl);
7578 }
7579 up_write(&info->extent_commit_sem);
7580
1a40e23b
ZY
7581 spin_lock(&info->block_group_cache_lock);
7582 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7583 block_group = rb_entry(n, struct btrfs_block_group_cache,
7584 cache_node);
1a40e23b
ZY
7585 rb_erase(&block_group->cache_node,
7586 &info->block_group_cache_tree);
d899e052
YZ
7587 spin_unlock(&info->block_group_cache_lock);
7588
80eb234a 7589 down_write(&block_group->space_info->groups_sem);
1a40e23b 7590 list_del(&block_group->list);
80eb234a 7591 up_write(&block_group->space_info->groups_sem);
d2fb3437 7592
817d52f8 7593 if (block_group->cached == BTRFS_CACHE_STARTED)
11833d66 7594 wait_block_group_cache_done(block_group);
817d52f8 7595
3c14874a
JB
7596 /*
7597 * We haven't cached this block group, which means we could
7598 * possibly have excluded extents on this block group.
7599 */
7600 if (block_group->cached == BTRFS_CACHE_NO)
7601 free_excluded_extents(info->extent_root, block_group);
7602
817d52f8 7603 btrfs_remove_free_space_cache(block_group);
11dfe35a 7604 btrfs_put_block_group(block_group);
d899e052
YZ
7605
7606 spin_lock(&info->block_group_cache_lock);
1a40e23b
ZY
7607 }
7608 spin_unlock(&info->block_group_cache_lock);
4184ea7f
CM
7609
7610 /* now that all the block groups are freed, go through and
7611 * free all the space_info structs. This is only called during
7612 * the final stages of unmount, and so we know nobody is
7613 * using them. We call synchronize_rcu() once before we start,
7614 * just to be on the safe side.
7615 */
7616 synchronize_rcu();
7617
8929ecfa
YZ
7618 release_global_block_rsv(info);
7619
4184ea7f
CM
7620 while(!list_empty(&info->space_info)) {
7621 space_info = list_entry(info->space_info.next,
7622 struct btrfs_space_info,
7623 list);
f0486c68 7624 if (space_info->bytes_pinned > 0 ||
fb25e914
JB
7625 space_info->bytes_reserved > 0 ||
7626 space_info->bytes_may_use > 0) {
f0486c68
YZ
7627 WARN_ON(1);
7628 dump_space_info(space_info, 0, 0);
7629 }
4184ea7f
CM
7630 list_del(&space_info->list);
7631 kfree(space_info);
7632 }
1a40e23b
ZY
7633 return 0;
7634}
7635
b742bb82
YZ
7636static void __link_block_group(struct btrfs_space_info *space_info,
7637 struct btrfs_block_group_cache *cache)
7638{
7639 int index = get_block_group_index(cache);
7640
7641 down_write(&space_info->groups_sem);
7642 list_add_tail(&cache->list, &space_info->block_groups[index]);
7643 up_write(&space_info->groups_sem);
7644}
7645
9078a3e1
CM
7646int btrfs_read_block_groups(struct btrfs_root *root)
7647{
7648 struct btrfs_path *path;
7649 int ret;
9078a3e1 7650 struct btrfs_block_group_cache *cache;
be744175 7651 struct btrfs_fs_info *info = root->fs_info;
6324fbf3 7652 struct btrfs_space_info *space_info;
9078a3e1
CM
7653 struct btrfs_key key;
7654 struct btrfs_key found_key;
5f39d397 7655 struct extent_buffer *leaf;
0af3d00b
JB
7656 int need_clear = 0;
7657 u64 cache_gen;
96b5179d 7658
be744175 7659 root = info->extent_root;
9078a3e1 7660 key.objectid = 0;
0b86a832 7661 key.offset = 0;
9078a3e1 7662 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
9078a3e1
CM
7663 path = btrfs_alloc_path();
7664 if (!path)
7665 return -ENOMEM;
026fd317 7666 path->reada = 1;
9078a3e1 7667
6c41761f 7668 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
73bc1876 7669 if (btrfs_test_opt(root, SPACE_CACHE) &&
6c41761f 7670 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
0af3d00b 7671 need_clear = 1;
88c2ba3b
JB
7672 if (btrfs_test_opt(root, CLEAR_CACHE))
7673 need_clear = 1;
0af3d00b 7674
d397712b 7675 while (1) {
0b86a832 7676 ret = find_first_block_group(root, path, &key);
b742bb82
YZ
7677 if (ret > 0)
7678 break;
0b86a832
CM
7679 if (ret != 0)
7680 goto error;
5f39d397
CM
7681 leaf = path->nodes[0];
7682 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8f18cf13 7683 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9078a3e1 7684 if (!cache) {
0b86a832 7685 ret = -ENOMEM;
f0486c68 7686 goto error;
9078a3e1 7687 }
34d52cb6
LZ
7688 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7689 GFP_NOFS);
7690 if (!cache->free_space_ctl) {
7691 kfree(cache);
7692 ret = -ENOMEM;
7693 goto error;
7694 }
3e1ad54f 7695
d2fb3437 7696 atomic_set(&cache->count, 1);
c286ac48 7697 spin_lock_init(&cache->lock);
817d52f8 7698 cache->fs_info = info;
0f9dd46c 7699 INIT_LIST_HEAD(&cache->list);
fa9c0d79 7700 INIT_LIST_HEAD(&cache->cluster_list);
96303081 7701
0af3d00b
JB
7702 if (need_clear)
7703 cache->disk_cache_state = BTRFS_DC_CLEAR;
7704
5f39d397
CM
7705 read_extent_buffer(leaf, &cache->item,
7706 btrfs_item_ptr_offset(leaf, path->slots[0]),
7707 sizeof(cache->item));
9078a3e1 7708 memcpy(&cache->key, &found_key, sizeof(found_key));
0b86a832 7709
9078a3e1 7710 key.objectid = found_key.objectid + found_key.offset;
b3b4aa74 7711 btrfs_release_path(path);
0b86a832 7712 cache->flags = btrfs_block_group_flags(&cache->item);
817d52f8
JB
7713 cache->sectorsize = root->sectorsize;
7714
34d52cb6
LZ
7715 btrfs_init_free_space_ctl(cache);
7716
3c14874a
JB
7717 /*
7718 * We need to exclude the super stripes now so that the space
7719 * info has super bytes accounted for, otherwise we'll think
7720 * we have more space than we actually do.
7721 */
7722 exclude_super_stripes(root, cache);
7723
817d52f8
JB
7724 /*
7725 * check for two cases, either we are full, and therefore
7726 * don't need to bother with the caching work since we won't
7727 * find any space, or we are empty, and we can just add all
7728 * the space in and be done with it. This saves us _alot_ of
7729 * time, particularly in the full case.
7730 */
7731 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
11833d66 7732 cache->last_byte_to_unpin = (u64)-1;
817d52f8 7733 cache->cached = BTRFS_CACHE_FINISHED;
1b2da372 7734 free_excluded_extents(root, cache);
817d52f8 7735 } else if (btrfs_block_group_used(&cache->item) == 0) {
11833d66 7736 cache->last_byte_to_unpin = (u64)-1;
817d52f8
JB
7737 cache->cached = BTRFS_CACHE_FINISHED;
7738 add_new_free_space(cache, root->fs_info,
7739 found_key.objectid,
7740 found_key.objectid +
7741 found_key.offset);
11833d66 7742 free_excluded_extents(root, cache);
817d52f8 7743 }
96b5179d 7744
6324fbf3
CM
7745 ret = update_space_info(info, cache->flags, found_key.offset,
7746 btrfs_block_group_used(&cache->item),
7747 &space_info);
79787eaa 7748 BUG_ON(ret); /* -ENOMEM */
6324fbf3 7749 cache->space_info = space_info;
1b2da372 7750 spin_lock(&cache->space_info->lock);
f0486c68 7751 cache->space_info->bytes_readonly += cache->bytes_super;
1b2da372
JB
7752 spin_unlock(&cache->space_info->lock);
7753
b742bb82 7754 __link_block_group(space_info, cache);
0f9dd46c
JB
7755
7756 ret = btrfs_add_block_group_cache(root->fs_info, cache);
79787eaa 7757 BUG_ON(ret); /* Logic error */
75ccf47d
CM
7758
7759 set_avail_alloc_bits(root->fs_info, cache->flags);
2b82032c 7760 if (btrfs_chunk_readonly(root, cache->key.objectid))
199c36ea 7761 set_block_group_ro(cache, 1);
9078a3e1 7762 }
b742bb82
YZ
7763
7764 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7765 if (!(get_alloc_profile(root, space_info->flags) &
7766 (BTRFS_BLOCK_GROUP_RAID10 |
7767 BTRFS_BLOCK_GROUP_RAID1 |
7768 BTRFS_BLOCK_GROUP_DUP)))
7769 continue;
7770 /*
7771 * avoid allocating from un-mirrored block group if there are
7772 * mirrored block groups.
7773 */
7774 list_for_each_entry(cache, &space_info->block_groups[3], list)
199c36ea 7775 set_block_group_ro(cache, 1);
b742bb82 7776 list_for_each_entry(cache, &space_info->block_groups[4], list)
199c36ea 7777 set_block_group_ro(cache, 1);
9078a3e1 7778 }
f0486c68
YZ
7779
7780 init_global_block_rsv(info);
0b86a832
CM
7781 ret = 0;
7782error:
9078a3e1 7783 btrfs_free_path(path);
0b86a832 7784 return ret;
9078a3e1 7785}
6324fbf3
CM
7786
7787int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7788 struct btrfs_root *root, u64 bytes_used,
e17cade2 7789 u64 type, u64 chunk_objectid, u64 chunk_offset,
6324fbf3
CM
7790 u64 size)
7791{
7792 int ret;
6324fbf3
CM
7793 struct btrfs_root *extent_root;
7794 struct btrfs_block_group_cache *cache;
6324fbf3
CM
7795
7796 extent_root = root->fs_info->extent_root;
6324fbf3 7797
12fcfd22 7798 root->fs_info->last_trans_log_full_commit = trans->transid;
e02119d5 7799
8f18cf13 7800 cache = kzalloc(sizeof(*cache), GFP_NOFS);
0f9dd46c
JB
7801 if (!cache)
7802 return -ENOMEM;
34d52cb6
LZ
7803 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7804 GFP_NOFS);
7805 if (!cache->free_space_ctl) {
7806 kfree(cache);
7807 return -ENOMEM;
7808 }
0f9dd46c 7809
e17cade2 7810 cache->key.objectid = chunk_offset;
6324fbf3 7811 cache->key.offset = size;
d2fb3437 7812 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
96303081 7813 cache->sectorsize = root->sectorsize;
0af3d00b 7814 cache->fs_info = root->fs_info;
96303081 7815
d2fb3437 7816 atomic_set(&cache->count, 1);
c286ac48 7817 spin_lock_init(&cache->lock);
0f9dd46c 7818 INIT_LIST_HEAD(&cache->list);
fa9c0d79 7819 INIT_LIST_HEAD(&cache->cluster_list);
0ef3e66b 7820
34d52cb6
LZ
7821 btrfs_init_free_space_ctl(cache);
7822
6324fbf3 7823 btrfs_set_block_group_used(&cache->item, bytes_used);
6324fbf3
CM
7824 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7825 cache->flags = type;
7826 btrfs_set_block_group_flags(&cache->item, type);
7827
11833d66 7828 cache->last_byte_to_unpin = (u64)-1;
817d52f8 7829 cache->cached = BTRFS_CACHE_FINISHED;
11833d66 7830 exclude_super_stripes(root, cache);
96303081 7831
817d52f8
JB
7832 add_new_free_space(cache, root->fs_info, chunk_offset,
7833 chunk_offset + size);
7834
11833d66
YZ
7835 free_excluded_extents(root, cache);
7836
6324fbf3
CM
7837 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7838 &cache->space_info);
79787eaa 7839 BUG_ON(ret); /* -ENOMEM */
c7c144db 7840 update_global_block_rsv(root->fs_info);
1b2da372
JB
7841
7842 spin_lock(&cache->space_info->lock);
f0486c68 7843 cache->space_info->bytes_readonly += cache->bytes_super;
1b2da372
JB
7844 spin_unlock(&cache->space_info->lock);
7845
b742bb82 7846 __link_block_group(cache->space_info, cache);
6324fbf3 7847
0f9dd46c 7848 ret = btrfs_add_block_group_cache(root->fs_info, cache);
79787eaa 7849 BUG_ON(ret); /* Logic error */
c286ac48 7850
6324fbf3
CM
7851 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7852 sizeof(cache->item));
79787eaa
JM
7853 if (ret) {
7854 btrfs_abort_transaction(trans, extent_root, ret);
7855 return ret;
7856 }
6324fbf3 7857
d18a2c44 7858 set_avail_alloc_bits(extent_root->fs_info, type);
925baedd 7859
6324fbf3
CM
7860 return 0;
7861}
1a40e23b 7862
10ea00f5
ID
7863static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
7864{
899c81ea
ID
7865 u64 extra_flags = chunk_to_extended(flags) &
7866 BTRFS_EXTENDED_PROFILE_MASK;
10ea00f5
ID
7867
7868 if (flags & BTRFS_BLOCK_GROUP_DATA)
7869 fs_info->avail_data_alloc_bits &= ~extra_flags;
7870 if (flags & BTRFS_BLOCK_GROUP_METADATA)
7871 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
7872 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
7873 fs_info->avail_system_alloc_bits &= ~extra_flags;
7874}
7875
1a40e23b
ZY
7876int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7877 struct btrfs_root *root, u64 group_start)
7878{
7879 struct btrfs_path *path;
7880 struct btrfs_block_group_cache *block_group;
44fb5511 7881 struct btrfs_free_cluster *cluster;
0af3d00b 7882 struct btrfs_root *tree_root = root->fs_info->tree_root;
1a40e23b 7883 struct btrfs_key key;
0af3d00b 7884 struct inode *inode;
1a40e23b 7885 int ret;
10ea00f5 7886 int index;
89a55897 7887 int factor;
1a40e23b 7888
1a40e23b
ZY
7889 root = root->fs_info->extent_root;
7890
7891 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7892 BUG_ON(!block_group);
c146afad 7893 BUG_ON(!block_group->ro);
1a40e23b 7894
9f7c43c9 7895 /*
7896 * Free the reserved super bytes from this block group before
7897 * remove it.
7898 */
7899 free_excluded_extents(root, block_group);
7900
1a40e23b 7901 memcpy(&key, &block_group->key, sizeof(key));
10ea00f5 7902 index = get_block_group_index(block_group);
89a55897
JB
7903 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7904 BTRFS_BLOCK_GROUP_RAID1 |
7905 BTRFS_BLOCK_GROUP_RAID10))
7906 factor = 2;
7907 else
7908 factor = 1;
1a40e23b 7909
44fb5511
CM
7910 /* make sure this block group isn't part of an allocation cluster */
7911 cluster = &root->fs_info->data_alloc_cluster;
7912 spin_lock(&cluster->refill_lock);
7913 btrfs_return_cluster_to_free_space(block_group, cluster);
7914 spin_unlock(&cluster->refill_lock);
7915
7916 /*
7917 * make sure this block group isn't part of a metadata
7918 * allocation cluster
7919 */
7920 cluster = &root->fs_info->meta_alloc_cluster;
7921 spin_lock(&cluster->refill_lock);
7922 btrfs_return_cluster_to_free_space(block_group, cluster);
7923 spin_unlock(&cluster->refill_lock);
7924
1a40e23b 7925 path = btrfs_alloc_path();
d8926bb3
MF
7926 if (!path) {
7927 ret = -ENOMEM;
7928 goto out;
7929 }
1a40e23b 7930
10b2f34d 7931 inode = lookup_free_space_inode(tree_root, block_group, path);
0af3d00b 7932 if (!IS_ERR(inode)) {
b532402e 7933 ret = btrfs_orphan_add(trans, inode);
79787eaa
JM
7934 if (ret) {
7935 btrfs_add_delayed_iput(inode);
7936 goto out;
7937 }
0af3d00b
JB
7938 clear_nlink(inode);
7939 /* One for the block groups ref */
7940 spin_lock(&block_group->lock);
7941 if (block_group->iref) {
7942 block_group->iref = 0;
7943 block_group->inode = NULL;
7944 spin_unlock(&block_group->lock);
7945 iput(inode);
7946 } else {
7947 spin_unlock(&block_group->lock);
7948 }
7949 /* One for our lookup ref */
455757c3 7950 btrfs_add_delayed_iput(inode);
0af3d00b
JB
7951 }
7952
7953 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
7954 key.offset = block_group->key.objectid;
7955 key.type = 0;
7956
7957 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
7958 if (ret < 0)
7959 goto out;
7960 if (ret > 0)
b3b4aa74 7961 btrfs_release_path(path);
0af3d00b
JB
7962 if (ret == 0) {
7963 ret = btrfs_del_item(trans, tree_root, path);
7964 if (ret)
7965 goto out;
b3b4aa74 7966 btrfs_release_path(path);
0af3d00b
JB
7967 }
7968
3dfdb934 7969 spin_lock(&root->fs_info->block_group_cache_lock);
1a40e23b
ZY
7970 rb_erase(&block_group->cache_node,
7971 &root->fs_info->block_group_cache_tree);
3dfdb934 7972 spin_unlock(&root->fs_info->block_group_cache_lock);
817d52f8 7973
80eb234a 7974 down_write(&block_group->space_info->groups_sem);
44fb5511
CM
7975 /*
7976 * we must use list_del_init so people can check to see if they
7977 * are still on the list after taking the semaphore
7978 */
7979 list_del_init(&block_group->list);
10ea00f5
ID
7980 if (list_empty(&block_group->space_info->block_groups[index]))
7981 clear_avail_alloc_bits(root->fs_info, block_group->flags);
80eb234a 7982 up_write(&block_group->space_info->groups_sem);
1a40e23b 7983
817d52f8 7984 if (block_group->cached == BTRFS_CACHE_STARTED)
11833d66 7985 wait_block_group_cache_done(block_group);
817d52f8
JB
7986
7987 btrfs_remove_free_space_cache(block_group);
7988
c146afad
YZ
7989 spin_lock(&block_group->space_info->lock);
7990 block_group->space_info->total_bytes -= block_group->key.offset;
7991 block_group->space_info->bytes_readonly -= block_group->key.offset;
89a55897 7992 block_group->space_info->disk_total -= block_group->key.offset * factor;
c146afad 7993 spin_unlock(&block_group->space_info->lock);
283bb197 7994
0af3d00b
JB
7995 memcpy(&key, &block_group->key, sizeof(key));
7996
283bb197 7997 btrfs_clear_space_info_full(root->fs_info);
c146afad 7998
fa9c0d79
CM
7999 btrfs_put_block_group(block_group);
8000 btrfs_put_block_group(block_group);
1a40e23b
ZY
8001
8002 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8003 if (ret > 0)
8004 ret = -EIO;
8005 if (ret < 0)
8006 goto out;
8007
8008 ret = btrfs_del_item(trans, root, path);
8009out:
8010 btrfs_free_path(path);
8011 return ret;
8012}
acce952b 8013
c59021f8 8014int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8015{
8016 struct btrfs_space_info *space_info;
1aba86d6 8017 struct btrfs_super_block *disk_super;
8018 u64 features;
8019 u64 flags;
8020 int mixed = 0;
c59021f8 8021 int ret;
8022
6c41761f 8023 disk_super = fs_info->super_copy;
1aba86d6 8024 if (!btrfs_super_root(disk_super))
8025 return 1;
c59021f8 8026
1aba86d6 8027 features = btrfs_super_incompat_flags(disk_super);
8028 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8029 mixed = 1;
c59021f8 8030
1aba86d6 8031 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8032 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
c59021f8 8033 if (ret)
1aba86d6 8034 goto out;
c59021f8 8035
1aba86d6 8036 if (mixed) {
8037 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8038 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8039 } else {
8040 flags = BTRFS_BLOCK_GROUP_METADATA;
8041 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8042 if (ret)
8043 goto out;
8044
8045 flags = BTRFS_BLOCK_GROUP_DATA;
8046 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8047 }
8048out:
c59021f8 8049 return ret;
8050}
8051
acce952b 8052int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8053{
8054 return unpin_extent_range(root, start, end);
8055}
8056
8057int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
5378e607 8058 u64 num_bytes, u64 *actual_bytes)
acce952b 8059{
5378e607 8060 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
acce952b 8061}
f7039b1d
LD
8062
8063int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8064{
8065 struct btrfs_fs_info *fs_info = root->fs_info;
8066 struct btrfs_block_group_cache *cache = NULL;
8067 u64 group_trimmed;
8068 u64 start;
8069 u64 end;
8070 u64 trimmed = 0;
2cac13e4 8071 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
f7039b1d
LD
8072 int ret = 0;
8073
2cac13e4
LB
8074 /*
8075 * try to trim all FS space, our block group may start from non-zero.
8076 */
8077 if (range->len == total_bytes)
8078 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8079 else
8080 cache = btrfs_lookup_block_group(fs_info, range->start);
f7039b1d
LD
8081
8082 while (cache) {
8083 if (cache->key.objectid >= (range->start + range->len)) {
8084 btrfs_put_block_group(cache);
8085 break;
8086 }
8087
8088 start = max(range->start, cache->key.objectid);
8089 end = min(range->start + range->len,
8090 cache->key.objectid + cache->key.offset);
8091
8092 if (end - start >= range->minlen) {
8093 if (!block_group_cache_done(cache)) {
8094 ret = cache_block_group(cache, NULL, root, 0);
8095 if (!ret)
8096 wait_block_group_cache_done(cache);
8097 }
8098 ret = btrfs_trim_block_group(cache,
8099 &group_trimmed,
8100 start,
8101 end,
8102 range->minlen);
8103
8104 trimmed += group_trimmed;
8105 if (ret) {
8106 btrfs_put_block_group(cache);
8107 break;
8108 }
8109 }
8110
8111 cache = next_block_group(fs_info->tree_root, cache);
8112 }
8113
8114 range->len = trimmed;
8115 return ret;
8116}