Commit | Line | Data |
---|---|---|
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> |
b150a4f1 | 27 | #include <linux/percpu_counter.h> |
74493f7a | 28 | #include "hash.h" |
995946dd | 29 | #include "tree-log.h" |
fec577fb CM |
30 | #include "disk-io.h" |
31 | #include "print-tree.h" | |
0b86a832 | 32 | #include "volumes.h" |
53b381b3 | 33 | #include "raid56.h" |
925baedd | 34 | #include "locking.h" |
fa9c0d79 | 35 | #include "free-space-cache.h" |
3fed40cc | 36 | #include "math.h" |
6ab0a202 | 37 | #include "sysfs.h" |
fcebe456 | 38 | #include "qgroup.h" |
fec577fb | 39 | |
709c0486 AJ |
40 | #undef SCRAMBLE_DELAYED_REFS |
41 | ||
9e622d6b MX |
42 | /* |
43 | * control flags for do_chunk_alloc's force field | |
0e4f8f88 CM |
44 | * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk |
45 | * if we really need one. | |
46 | * | |
0e4f8f88 CM |
47 | * CHUNK_ALLOC_LIMITED means to only try and allocate one |
48 | * if we have very few chunks already allocated. This is | |
49 | * used as part of the clustering code to help make sure | |
50 | * we have a good pool of storage to cluster in, without | |
51 | * filling the FS with empty chunks | |
52 | * | |
9e622d6b MX |
53 | * CHUNK_ALLOC_FORCE means it must try to allocate one |
54 | * | |
0e4f8f88 CM |
55 | */ |
56 | enum { | |
57 | CHUNK_ALLOC_NO_FORCE = 0, | |
9e622d6b MX |
58 | CHUNK_ALLOC_LIMITED = 1, |
59 | CHUNK_ALLOC_FORCE = 2, | |
0e4f8f88 CM |
60 | }; |
61 | ||
fb25e914 JB |
62 | /* |
63 | * Control how reservations are dealt with. | |
64 | * | |
65 | * RESERVE_FREE - freeing a reservation. | |
66 | * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for | |
67 | * ENOSPC accounting | |
68 | * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update | |
69 | * bytes_may_use as the ENOSPC accounting is done elsewhere | |
70 | */ | |
71 | enum { | |
72 | RESERVE_FREE = 0, | |
73 | RESERVE_ALLOC = 1, | |
74 | RESERVE_ALLOC_NO_ACCOUNT = 2, | |
75 | }; | |
76 | ||
c53d613e | 77 | static int update_block_group(struct btrfs_root *root, |
f0486c68 | 78 | u64 bytenr, u64 num_bytes, int alloc); |
5d4f98a2 YZ |
79 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
80 | struct btrfs_root *root, | |
81 | u64 bytenr, u64 num_bytes, u64 parent, | |
82 | u64 root_objectid, u64 owner_objectid, | |
83 | u64 owner_offset, int refs_to_drop, | |
fcebe456 JB |
84 | struct btrfs_delayed_extent_op *extra_op, |
85 | int no_quota); | |
5d4f98a2 YZ |
86 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, |
87 | struct extent_buffer *leaf, | |
88 | struct btrfs_extent_item *ei); | |
89 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
90 | struct btrfs_root *root, | |
91 | u64 parent, u64 root_objectid, | |
92 | u64 flags, u64 owner, u64 offset, | |
93 | struct btrfs_key *ins, int ref_mod); | |
94 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
95 | struct btrfs_root *root, | |
96 | u64 parent, u64 root_objectid, | |
97 | u64 flags, struct btrfs_disk_key *key, | |
fcebe456 JB |
98 | int level, struct btrfs_key *ins, |
99 | int no_quota); | |
6a63209f | 100 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
698d0082 JB |
101 | struct btrfs_root *extent_root, u64 flags, |
102 | int force); | |
11833d66 YZ |
103 | static int find_next_key(struct btrfs_path *path, int level, |
104 | struct btrfs_key *key); | |
9ed74f2d JB |
105 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
106 | int dump_block_groups); | |
fb25e914 | 107 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
e570fd27 MX |
108 | u64 num_bytes, int reserve, |
109 | int delalloc); | |
5d80366e JB |
110 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, |
111 | u64 num_bytes); | |
48a3b636 ES |
112 | int btrfs_pin_extent(struct btrfs_root *root, |
113 | u64 bytenr, u64 num_bytes, int reserved); | |
6a63209f | 114 | |
817d52f8 JB |
115 | static noinline int |
116 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
117 | { | |
118 | smp_mb(); | |
36cce922 JB |
119 | return cache->cached == BTRFS_CACHE_FINISHED || |
120 | cache->cached == BTRFS_CACHE_ERROR; | |
817d52f8 JB |
121 | } |
122 | ||
0f9dd46c JB |
123 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
124 | { | |
125 | return (cache->flags & bits) == bits; | |
126 | } | |
127 | ||
62a45b60 | 128 | static void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
11dfe35a JB |
129 | { |
130 | atomic_inc(&cache->count); | |
131 | } | |
132 | ||
133 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) | |
134 | { | |
f0486c68 YZ |
135 | if (atomic_dec_and_test(&cache->count)) { |
136 | WARN_ON(cache->pinned > 0); | |
137 | WARN_ON(cache->reserved > 0); | |
34d52cb6 | 138 | kfree(cache->free_space_ctl); |
11dfe35a | 139 | kfree(cache); |
f0486c68 | 140 | } |
11dfe35a JB |
141 | } |
142 | ||
0f9dd46c JB |
143 | /* |
144 | * this adds the block group to the fs_info rb tree for the block group | |
145 | * cache | |
146 | */ | |
b2950863 | 147 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
148 | struct btrfs_block_group_cache *block_group) |
149 | { | |
150 | struct rb_node **p; | |
151 | struct rb_node *parent = NULL; | |
152 | struct btrfs_block_group_cache *cache; | |
153 | ||
154 | spin_lock(&info->block_group_cache_lock); | |
155 | p = &info->block_group_cache_tree.rb_node; | |
156 | ||
157 | while (*p) { | |
158 | parent = *p; | |
159 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
160 | cache_node); | |
161 | if (block_group->key.objectid < cache->key.objectid) { | |
162 | p = &(*p)->rb_left; | |
163 | } else if (block_group->key.objectid > cache->key.objectid) { | |
164 | p = &(*p)->rb_right; | |
165 | } else { | |
166 | spin_unlock(&info->block_group_cache_lock); | |
167 | return -EEXIST; | |
168 | } | |
169 | } | |
170 | ||
171 | rb_link_node(&block_group->cache_node, parent, p); | |
172 | rb_insert_color(&block_group->cache_node, | |
173 | &info->block_group_cache_tree); | |
a1897fdd LB |
174 | |
175 | if (info->first_logical_byte > block_group->key.objectid) | |
176 | info->first_logical_byte = block_group->key.objectid; | |
177 | ||
0f9dd46c JB |
178 | spin_unlock(&info->block_group_cache_lock); |
179 | ||
180 | return 0; | |
181 | } | |
182 | ||
183 | /* | |
184 | * This will return the block group at or after bytenr if contains is 0, else | |
185 | * it will return the block group that contains the bytenr | |
186 | */ | |
187 | static struct btrfs_block_group_cache * | |
188 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
189 | int contains) | |
190 | { | |
191 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
192 | struct rb_node *n; | |
193 | u64 end, start; | |
194 | ||
195 | spin_lock(&info->block_group_cache_lock); | |
196 | n = info->block_group_cache_tree.rb_node; | |
197 | ||
198 | while (n) { | |
199 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
200 | cache_node); | |
201 | end = cache->key.objectid + cache->key.offset - 1; | |
202 | start = cache->key.objectid; | |
203 | ||
204 | if (bytenr < start) { | |
205 | if (!contains && (!ret || start < ret->key.objectid)) | |
206 | ret = cache; | |
207 | n = n->rb_left; | |
208 | } else if (bytenr > start) { | |
209 | if (contains && bytenr <= end) { | |
210 | ret = cache; | |
211 | break; | |
212 | } | |
213 | n = n->rb_right; | |
214 | } else { | |
215 | ret = cache; | |
216 | break; | |
217 | } | |
218 | } | |
a1897fdd | 219 | if (ret) { |
11dfe35a | 220 | btrfs_get_block_group(ret); |
a1897fdd LB |
221 | if (bytenr == 0 && info->first_logical_byte > ret->key.objectid) |
222 | info->first_logical_byte = ret->key.objectid; | |
223 | } | |
0f9dd46c JB |
224 | spin_unlock(&info->block_group_cache_lock); |
225 | ||
226 | return ret; | |
227 | } | |
228 | ||
11833d66 YZ |
229 | static int add_excluded_extent(struct btrfs_root *root, |
230 | u64 start, u64 num_bytes) | |
817d52f8 | 231 | { |
11833d66 YZ |
232 | u64 end = start + num_bytes - 1; |
233 | set_extent_bits(&root->fs_info->freed_extents[0], | |
234 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
235 | set_extent_bits(&root->fs_info->freed_extents[1], | |
236 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
237 | return 0; | |
238 | } | |
817d52f8 | 239 | |
11833d66 YZ |
240 | static void free_excluded_extents(struct btrfs_root *root, |
241 | struct btrfs_block_group_cache *cache) | |
242 | { | |
243 | u64 start, end; | |
817d52f8 | 244 | |
11833d66 YZ |
245 | start = cache->key.objectid; |
246 | end = start + cache->key.offset - 1; | |
247 | ||
248 | clear_extent_bits(&root->fs_info->freed_extents[0], | |
249 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
250 | clear_extent_bits(&root->fs_info->freed_extents[1], | |
251 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
817d52f8 JB |
252 | } |
253 | ||
11833d66 YZ |
254 | static int exclude_super_stripes(struct btrfs_root *root, |
255 | struct btrfs_block_group_cache *cache) | |
817d52f8 | 256 | { |
817d52f8 JB |
257 | u64 bytenr; |
258 | u64 *logical; | |
259 | int stripe_len; | |
260 | int i, nr, ret; | |
261 | ||
06b2331f YZ |
262 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
263 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
264 | cache->bytes_super += stripe_len; | |
265 | ret = add_excluded_extent(root, cache->key.objectid, | |
266 | stripe_len); | |
835d974f JB |
267 | if (ret) |
268 | return ret; | |
06b2331f YZ |
269 | } |
270 | ||
817d52f8 JB |
271 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
272 | bytenr = btrfs_sb_offset(i); | |
273 | ret = btrfs_rmap_block(&root->fs_info->mapping_tree, | |
274 | cache->key.objectid, bytenr, | |
275 | 0, &logical, &nr, &stripe_len); | |
835d974f JB |
276 | if (ret) |
277 | return ret; | |
11833d66 | 278 | |
817d52f8 | 279 | while (nr--) { |
51bf5f0b JB |
280 | u64 start, len; |
281 | ||
282 | if (logical[nr] > cache->key.objectid + | |
283 | cache->key.offset) | |
284 | continue; | |
285 | ||
286 | if (logical[nr] + stripe_len <= cache->key.objectid) | |
287 | continue; | |
288 | ||
289 | start = logical[nr]; | |
290 | if (start < cache->key.objectid) { | |
291 | start = cache->key.objectid; | |
292 | len = (logical[nr] + stripe_len) - start; | |
293 | } else { | |
294 | len = min_t(u64, stripe_len, | |
295 | cache->key.objectid + | |
296 | cache->key.offset - start); | |
297 | } | |
298 | ||
299 | cache->bytes_super += len; | |
300 | ret = add_excluded_extent(root, start, len); | |
835d974f JB |
301 | if (ret) { |
302 | kfree(logical); | |
303 | return ret; | |
304 | } | |
817d52f8 | 305 | } |
11833d66 | 306 | |
817d52f8 JB |
307 | kfree(logical); |
308 | } | |
817d52f8 JB |
309 | return 0; |
310 | } | |
311 | ||
11833d66 YZ |
312 | static struct btrfs_caching_control * |
313 | get_caching_control(struct btrfs_block_group_cache *cache) | |
314 | { | |
315 | struct btrfs_caching_control *ctl; | |
316 | ||
317 | spin_lock(&cache->lock); | |
318 | if (cache->cached != BTRFS_CACHE_STARTED) { | |
319 | spin_unlock(&cache->lock); | |
320 | return NULL; | |
321 | } | |
322 | ||
dde5abee JB |
323 | /* We're loading it the fast way, so we don't have a caching_ctl. */ |
324 | if (!cache->caching_ctl) { | |
325 | spin_unlock(&cache->lock); | |
11833d66 YZ |
326 | return NULL; |
327 | } | |
328 | ||
329 | ctl = cache->caching_ctl; | |
330 | atomic_inc(&ctl->count); | |
331 | spin_unlock(&cache->lock); | |
332 | return ctl; | |
333 | } | |
334 | ||
335 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
336 | { | |
337 | if (atomic_dec_and_test(&ctl->count)) | |
338 | kfree(ctl); | |
339 | } | |
340 | ||
0f9dd46c JB |
341 | /* |
342 | * this is only called by cache_block_group, since we could have freed extents | |
343 | * we need to check the pinned_extents for any extents that can't be used yet | |
344 | * since their free space will be released as soon as the transaction commits. | |
345 | */ | |
817d52f8 | 346 | static u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
0f9dd46c JB |
347 | struct btrfs_fs_info *info, u64 start, u64 end) |
348 | { | |
817d52f8 | 349 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
350 | int ret; |
351 | ||
352 | while (start < end) { | |
11833d66 | 353 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 354 | &extent_start, &extent_end, |
e6138876 JB |
355 | EXTENT_DIRTY | EXTENT_UPTODATE, |
356 | NULL); | |
0f9dd46c JB |
357 | if (ret) |
358 | break; | |
359 | ||
06b2331f | 360 | if (extent_start <= start) { |
0f9dd46c JB |
361 | start = extent_end + 1; |
362 | } else if (extent_start > start && extent_start < end) { | |
363 | size = extent_start - start; | |
817d52f8 | 364 | total_added += size; |
ea6a478e JB |
365 | ret = btrfs_add_free_space(block_group, start, |
366 | size); | |
79787eaa | 367 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
368 | start = extent_end + 1; |
369 | } else { | |
370 | break; | |
371 | } | |
372 | } | |
373 | ||
374 | if (start < end) { | |
375 | size = end - start; | |
817d52f8 | 376 | total_added += size; |
ea6a478e | 377 | ret = btrfs_add_free_space(block_group, start, size); |
79787eaa | 378 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
379 | } |
380 | ||
817d52f8 | 381 | return total_added; |
0f9dd46c JB |
382 | } |
383 | ||
d458b054 | 384 | static noinline void caching_thread(struct btrfs_work *work) |
e37c9e69 | 385 | { |
bab39bf9 JB |
386 | struct btrfs_block_group_cache *block_group; |
387 | struct btrfs_fs_info *fs_info; | |
388 | struct btrfs_caching_control *caching_ctl; | |
389 | struct btrfs_root *extent_root; | |
e37c9e69 | 390 | struct btrfs_path *path; |
5f39d397 | 391 | struct extent_buffer *leaf; |
11833d66 | 392 | struct btrfs_key key; |
817d52f8 | 393 | u64 total_found = 0; |
11833d66 YZ |
394 | u64 last = 0; |
395 | u32 nritems; | |
36cce922 | 396 | int ret = -ENOMEM; |
f510cfec | 397 | |
bab39bf9 JB |
398 | caching_ctl = container_of(work, struct btrfs_caching_control, work); |
399 | block_group = caching_ctl->block_group; | |
400 | fs_info = block_group->fs_info; | |
401 | extent_root = fs_info->extent_root; | |
402 | ||
e37c9e69 CM |
403 | path = btrfs_alloc_path(); |
404 | if (!path) | |
bab39bf9 | 405 | goto out; |
7d7d6068 | 406 | |
817d52f8 | 407 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 408 | |
5cd57b2c | 409 | /* |
817d52f8 JB |
410 | * We don't want to deadlock with somebody trying to allocate a new |
411 | * extent for the extent root while also trying to search the extent | |
412 | * root to add free space. So we skip locking and search the commit | |
413 | * root, since its read-only | |
5cd57b2c CM |
414 | */ |
415 | path->skip_locking = 1; | |
817d52f8 | 416 | path->search_commit_root = 1; |
026fd317 | 417 | path->reada = 1; |
817d52f8 | 418 | |
e4404d6e | 419 | key.objectid = last; |
e37c9e69 | 420 | key.offset = 0; |
11833d66 | 421 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 422 | again: |
11833d66 | 423 | mutex_lock(&caching_ctl->mutex); |
013f1b12 | 424 | /* need to make sure the commit_root doesn't disappear */ |
9e351cc8 | 425 | down_read(&fs_info->commit_root_sem); |
013f1b12 | 426 | |
52ee28d2 | 427 | next: |
11833d66 | 428 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 429 | if (ret < 0) |
ef8bbdfe | 430 | goto err; |
a512bbf8 | 431 | |
11833d66 YZ |
432 | leaf = path->nodes[0]; |
433 | nritems = btrfs_header_nritems(leaf); | |
434 | ||
d397712b | 435 | while (1) { |
7841cb28 | 436 | if (btrfs_fs_closing(fs_info) > 1) { |
f25784b3 | 437 | last = (u64)-1; |
817d52f8 | 438 | break; |
f25784b3 | 439 | } |
817d52f8 | 440 | |
11833d66 YZ |
441 | if (path->slots[0] < nritems) { |
442 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
443 | } else { | |
444 | ret = find_next_key(path, 0, &key); | |
445 | if (ret) | |
e37c9e69 | 446 | break; |
817d52f8 | 447 | |
c9ea7b24 | 448 | if (need_resched() || |
9e351cc8 | 449 | rwsem_is_contended(&fs_info->commit_root_sem)) { |
589d8ade | 450 | caching_ctl->progress = last; |
ff5714cc | 451 | btrfs_release_path(path); |
9e351cc8 | 452 | up_read(&fs_info->commit_root_sem); |
589d8ade | 453 | mutex_unlock(&caching_ctl->mutex); |
11833d66 | 454 | cond_resched(); |
589d8ade JB |
455 | goto again; |
456 | } | |
0a3896d0 JB |
457 | |
458 | ret = btrfs_next_leaf(extent_root, path); | |
459 | if (ret < 0) | |
460 | goto err; | |
461 | if (ret) | |
462 | break; | |
589d8ade JB |
463 | leaf = path->nodes[0]; |
464 | nritems = btrfs_header_nritems(leaf); | |
465 | continue; | |
11833d66 | 466 | } |
817d52f8 | 467 | |
52ee28d2 LB |
468 | if (key.objectid < last) { |
469 | key.objectid = last; | |
470 | key.offset = 0; | |
471 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
472 | ||
473 | caching_ctl->progress = last; | |
474 | btrfs_release_path(path); | |
475 | goto next; | |
476 | } | |
477 | ||
11833d66 YZ |
478 | if (key.objectid < block_group->key.objectid) { |
479 | path->slots[0]++; | |
817d52f8 | 480 | continue; |
e37c9e69 | 481 | } |
0f9dd46c | 482 | |
e37c9e69 | 483 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 484 | block_group->key.offset) |
e37c9e69 | 485 | break; |
7d7d6068 | 486 | |
3173a18f JB |
487 | if (key.type == BTRFS_EXTENT_ITEM_KEY || |
488 | key.type == BTRFS_METADATA_ITEM_KEY) { | |
817d52f8 JB |
489 | total_found += add_new_free_space(block_group, |
490 | fs_info, last, | |
491 | key.objectid); | |
3173a18f JB |
492 | if (key.type == BTRFS_METADATA_ITEM_KEY) |
493 | last = key.objectid + | |
494 | fs_info->tree_root->leafsize; | |
495 | else | |
496 | last = key.objectid + key.offset; | |
817d52f8 | 497 | |
11833d66 YZ |
498 | if (total_found > (1024 * 1024 * 2)) { |
499 | total_found = 0; | |
500 | wake_up(&caching_ctl->wait); | |
501 | } | |
817d52f8 | 502 | } |
e37c9e69 CM |
503 | path->slots[0]++; |
504 | } | |
817d52f8 | 505 | ret = 0; |
e37c9e69 | 506 | |
817d52f8 JB |
507 | total_found += add_new_free_space(block_group, fs_info, last, |
508 | block_group->key.objectid + | |
509 | block_group->key.offset); | |
11833d66 | 510 | caching_ctl->progress = (u64)-1; |
817d52f8 JB |
511 | |
512 | spin_lock(&block_group->lock); | |
11833d66 | 513 | block_group->caching_ctl = NULL; |
817d52f8 JB |
514 | block_group->cached = BTRFS_CACHE_FINISHED; |
515 | spin_unlock(&block_group->lock); | |
0f9dd46c | 516 | |
54aa1f4d | 517 | err: |
e37c9e69 | 518 | btrfs_free_path(path); |
9e351cc8 | 519 | up_read(&fs_info->commit_root_sem); |
817d52f8 | 520 | |
11833d66 YZ |
521 | free_excluded_extents(extent_root, block_group); |
522 | ||
523 | mutex_unlock(&caching_ctl->mutex); | |
bab39bf9 | 524 | out: |
36cce922 JB |
525 | if (ret) { |
526 | spin_lock(&block_group->lock); | |
527 | block_group->caching_ctl = NULL; | |
528 | block_group->cached = BTRFS_CACHE_ERROR; | |
529 | spin_unlock(&block_group->lock); | |
530 | } | |
11833d66 YZ |
531 | wake_up(&caching_ctl->wait); |
532 | ||
533 | put_caching_control(caching_ctl); | |
11dfe35a | 534 | btrfs_put_block_group(block_group); |
817d52f8 JB |
535 | } |
536 | ||
9d66e233 | 537 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
9d66e233 | 538 | int load_cache_only) |
817d52f8 | 539 | { |
291c7d2f | 540 | DEFINE_WAIT(wait); |
11833d66 YZ |
541 | struct btrfs_fs_info *fs_info = cache->fs_info; |
542 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
543 | int ret = 0; |
544 | ||
291c7d2f | 545 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
79787eaa JM |
546 | if (!caching_ctl) |
547 | return -ENOMEM; | |
291c7d2f JB |
548 | |
549 | INIT_LIST_HEAD(&caching_ctl->list); | |
550 | mutex_init(&caching_ctl->mutex); | |
551 | init_waitqueue_head(&caching_ctl->wait); | |
552 | caching_ctl->block_group = cache; | |
553 | caching_ctl->progress = cache->key.objectid; | |
554 | atomic_set(&caching_ctl->count, 1); | |
e66f0bb1 | 555 | btrfs_init_work(&caching_ctl->work, caching_thread, NULL, NULL); |
291c7d2f JB |
556 | |
557 | spin_lock(&cache->lock); | |
558 | /* | |
559 | * This should be a rare occasion, but this could happen I think in the | |
560 | * case where one thread starts to load the space cache info, and then | |
561 | * some other thread starts a transaction commit which tries to do an | |
562 | * allocation while the other thread is still loading the space cache | |
563 | * info. The previous loop should have kept us from choosing this block | |
564 | * group, but if we've moved to the state where we will wait on caching | |
565 | * block groups we need to first check if we're doing a fast load here, | |
566 | * so we can wait for it to finish, otherwise we could end up allocating | |
567 | * from a block group who's cache gets evicted for one reason or | |
568 | * another. | |
569 | */ | |
570 | while (cache->cached == BTRFS_CACHE_FAST) { | |
571 | struct btrfs_caching_control *ctl; | |
572 | ||
573 | ctl = cache->caching_ctl; | |
574 | atomic_inc(&ctl->count); | |
575 | prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); | |
576 | spin_unlock(&cache->lock); | |
577 | ||
578 | schedule(); | |
579 | ||
580 | finish_wait(&ctl->wait, &wait); | |
581 | put_caching_control(ctl); | |
582 | spin_lock(&cache->lock); | |
583 | } | |
584 | ||
585 | if (cache->cached != BTRFS_CACHE_NO) { | |
586 | spin_unlock(&cache->lock); | |
587 | kfree(caching_ctl); | |
11833d66 | 588 | return 0; |
291c7d2f JB |
589 | } |
590 | WARN_ON(cache->caching_ctl); | |
591 | cache->caching_ctl = caching_ctl; | |
592 | cache->cached = BTRFS_CACHE_FAST; | |
593 | spin_unlock(&cache->lock); | |
11833d66 | 594 | |
d53ba474 | 595 | if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) { |
9d66e233 JB |
596 | ret = load_free_space_cache(fs_info, cache); |
597 | ||
598 | spin_lock(&cache->lock); | |
599 | if (ret == 1) { | |
291c7d2f | 600 | cache->caching_ctl = NULL; |
9d66e233 JB |
601 | cache->cached = BTRFS_CACHE_FINISHED; |
602 | cache->last_byte_to_unpin = (u64)-1; | |
603 | } else { | |
291c7d2f JB |
604 | if (load_cache_only) { |
605 | cache->caching_ctl = NULL; | |
606 | cache->cached = BTRFS_CACHE_NO; | |
607 | } else { | |
608 | cache->cached = BTRFS_CACHE_STARTED; | |
609 | } | |
9d66e233 JB |
610 | } |
611 | spin_unlock(&cache->lock); | |
291c7d2f | 612 | wake_up(&caching_ctl->wait); |
3c14874a | 613 | if (ret == 1) { |
291c7d2f | 614 | put_caching_control(caching_ctl); |
3c14874a | 615 | free_excluded_extents(fs_info->extent_root, cache); |
9d66e233 | 616 | return 0; |
3c14874a | 617 | } |
291c7d2f JB |
618 | } else { |
619 | /* | |
620 | * We are not going to do the fast caching, set cached to the | |
621 | * appropriate value and wakeup any waiters. | |
622 | */ | |
623 | spin_lock(&cache->lock); | |
624 | if (load_cache_only) { | |
625 | cache->caching_ctl = NULL; | |
626 | cache->cached = BTRFS_CACHE_NO; | |
627 | } else { | |
628 | cache->cached = BTRFS_CACHE_STARTED; | |
629 | } | |
630 | spin_unlock(&cache->lock); | |
631 | wake_up(&caching_ctl->wait); | |
9d66e233 JB |
632 | } |
633 | ||
291c7d2f JB |
634 | if (load_cache_only) { |
635 | put_caching_control(caching_ctl); | |
11833d66 | 636 | return 0; |
817d52f8 | 637 | } |
817d52f8 | 638 | |
9e351cc8 | 639 | down_write(&fs_info->commit_root_sem); |
291c7d2f | 640 | atomic_inc(&caching_ctl->count); |
11833d66 | 641 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); |
9e351cc8 | 642 | up_write(&fs_info->commit_root_sem); |
11833d66 | 643 | |
11dfe35a | 644 | btrfs_get_block_group(cache); |
11833d66 | 645 | |
e66f0bb1 | 646 | btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work); |
817d52f8 | 647 | |
ef8bbdfe | 648 | return ret; |
e37c9e69 CM |
649 | } |
650 | ||
0f9dd46c JB |
651 | /* |
652 | * return the block group that starts at or after bytenr | |
653 | */ | |
d397712b CM |
654 | static struct btrfs_block_group_cache * |
655 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 656 | { |
0f9dd46c | 657 | struct btrfs_block_group_cache *cache; |
0ef3e66b | 658 | |
0f9dd46c | 659 | cache = block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b | 660 | |
0f9dd46c | 661 | return cache; |
0ef3e66b CM |
662 | } |
663 | ||
0f9dd46c | 664 | /* |
9f55684c | 665 | * return the block group that contains the given bytenr |
0f9dd46c | 666 | */ |
d397712b CM |
667 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
668 | struct btrfs_fs_info *info, | |
669 | u64 bytenr) | |
be744175 | 670 | { |
0f9dd46c | 671 | struct btrfs_block_group_cache *cache; |
be744175 | 672 | |
0f9dd46c | 673 | cache = block_group_cache_tree_search(info, bytenr, 1); |
96b5179d | 674 | |
0f9dd46c | 675 | return cache; |
be744175 | 676 | } |
0b86a832 | 677 | |
0f9dd46c JB |
678 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
679 | u64 flags) | |
6324fbf3 | 680 | { |
0f9dd46c | 681 | struct list_head *head = &info->space_info; |
0f9dd46c | 682 | struct btrfs_space_info *found; |
4184ea7f | 683 | |
52ba6929 | 684 | flags &= BTRFS_BLOCK_GROUP_TYPE_MASK; |
b742bb82 | 685 | |
4184ea7f CM |
686 | rcu_read_lock(); |
687 | list_for_each_entry_rcu(found, head, list) { | |
67377734 | 688 | if (found->flags & flags) { |
4184ea7f | 689 | rcu_read_unlock(); |
0f9dd46c | 690 | return found; |
4184ea7f | 691 | } |
0f9dd46c | 692 | } |
4184ea7f | 693 | rcu_read_unlock(); |
0f9dd46c | 694 | return NULL; |
6324fbf3 CM |
695 | } |
696 | ||
4184ea7f CM |
697 | /* |
698 | * after adding space to the filesystem, we need to clear the full flags | |
699 | * on all the space infos. | |
700 | */ | |
701 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
702 | { | |
703 | struct list_head *head = &info->space_info; | |
704 | struct btrfs_space_info *found; | |
705 | ||
706 | rcu_read_lock(); | |
707 | list_for_each_entry_rcu(found, head, list) | |
708 | found->full = 0; | |
709 | rcu_read_unlock(); | |
710 | } | |
711 | ||
e02119d5 | 712 | /* simple helper to search for an existing extent at a given offset */ |
31840ae1 | 713 | int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len) |
e02119d5 CM |
714 | { |
715 | int ret; | |
716 | struct btrfs_key key; | |
31840ae1 | 717 | struct btrfs_path *path; |
e02119d5 | 718 | |
31840ae1 | 719 | path = btrfs_alloc_path(); |
d8926bb3 MF |
720 | if (!path) |
721 | return -ENOMEM; | |
722 | ||
e02119d5 CM |
723 | key.objectid = start; |
724 | key.offset = len; | |
3173a18f | 725 | key.type = BTRFS_EXTENT_ITEM_KEY; |
e02119d5 CM |
726 | ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path, |
727 | 0, 0); | |
3173a18f JB |
728 | if (ret > 0) { |
729 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
730 | if (key.objectid == start && | |
731 | key.type == BTRFS_METADATA_ITEM_KEY) | |
732 | ret = 0; | |
733 | } | |
31840ae1 | 734 | btrfs_free_path(path); |
7bb86316 CM |
735 | return ret; |
736 | } | |
737 | ||
a22285a6 | 738 | /* |
3173a18f | 739 | * helper function to lookup reference count and flags of a tree block. |
a22285a6 YZ |
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 | */ | |
747 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
748 | struct btrfs_root *root, u64 bytenr, | |
3173a18f | 749 | u64 offset, int metadata, u64 *refs, u64 *flags) |
a22285a6 YZ |
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 | ||
3173a18f JB |
762 | /* |
763 | * If we don't have skinny metadata, don't bother doing anything | |
764 | * different | |
765 | */ | |
766 | if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) { | |
767 | offset = root->leafsize; | |
768 | metadata = 0; | |
769 | } | |
770 | ||
a22285a6 YZ |
771 | path = btrfs_alloc_path(); |
772 | if (!path) | |
773 | return -ENOMEM; | |
774 | ||
a22285a6 YZ |
775 | if (!trans) { |
776 | path->skip_locking = 1; | |
777 | path->search_commit_root = 1; | |
778 | } | |
639eefc8 FDBM |
779 | |
780 | search_again: | |
781 | key.objectid = bytenr; | |
782 | key.offset = offset; | |
783 | if (metadata) | |
784 | key.type = BTRFS_METADATA_ITEM_KEY; | |
785 | else | |
786 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
787 | ||
a22285a6 YZ |
788 | again: |
789 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, | |
790 | &key, path, 0, 0); | |
791 | if (ret < 0) | |
792 | goto out_free; | |
793 | ||
3173a18f | 794 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { |
74be9510 FDBM |
795 | if (path->slots[0]) { |
796 | path->slots[0]--; | |
797 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
798 | path->slots[0]); | |
799 | if (key.objectid == bytenr && | |
800 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
801 | key.offset == root->leafsize) | |
802 | ret = 0; | |
803 | } | |
804 | if (ret) { | |
805 | key.objectid = bytenr; | |
806 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
807 | key.offset = root->leafsize; | |
808 | btrfs_release_path(path); | |
809 | goto again; | |
810 | } | |
3173a18f JB |
811 | } |
812 | ||
a22285a6 YZ |
813 | if (ret == 0) { |
814 | leaf = path->nodes[0]; | |
815 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
816 | if (item_size >= sizeof(*ei)) { | |
817 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
818 | struct btrfs_extent_item); | |
819 | num_refs = btrfs_extent_refs(leaf, ei); | |
820 | extent_flags = btrfs_extent_flags(leaf, ei); | |
821 | } else { | |
822 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
823 | struct btrfs_extent_item_v0 *ei0; | |
824 | BUG_ON(item_size != sizeof(*ei0)); | |
825 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
826 | struct btrfs_extent_item_v0); | |
827 | num_refs = btrfs_extent_refs_v0(leaf, ei0); | |
828 | /* FIXME: this isn't correct for data */ | |
829 | extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
830 | #else | |
831 | BUG(); | |
832 | #endif | |
833 | } | |
834 | BUG_ON(num_refs == 0); | |
835 | } else { | |
836 | num_refs = 0; | |
837 | extent_flags = 0; | |
838 | ret = 0; | |
839 | } | |
840 | ||
841 | if (!trans) | |
842 | goto out; | |
843 | ||
844 | delayed_refs = &trans->transaction->delayed_refs; | |
845 | spin_lock(&delayed_refs->lock); | |
846 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
847 | if (head) { | |
848 | if (!mutex_trylock(&head->mutex)) { | |
849 | atomic_inc(&head->node.refs); | |
850 | spin_unlock(&delayed_refs->lock); | |
851 | ||
b3b4aa74 | 852 | btrfs_release_path(path); |
a22285a6 | 853 | |
8cc33e5c DS |
854 | /* |
855 | * Mutex was contended, block until it's released and try | |
856 | * again | |
857 | */ | |
a22285a6 YZ |
858 | mutex_lock(&head->mutex); |
859 | mutex_unlock(&head->mutex); | |
860 | btrfs_put_delayed_ref(&head->node); | |
639eefc8 | 861 | goto search_again; |
a22285a6 | 862 | } |
d7df2c79 | 863 | spin_lock(&head->lock); |
a22285a6 YZ |
864 | if (head->extent_op && head->extent_op->update_flags) |
865 | extent_flags |= head->extent_op->flags_to_set; | |
866 | else | |
867 | BUG_ON(num_refs == 0); | |
868 | ||
869 | num_refs += head->node.ref_mod; | |
d7df2c79 | 870 | spin_unlock(&head->lock); |
a22285a6 YZ |
871 | mutex_unlock(&head->mutex); |
872 | } | |
873 | spin_unlock(&delayed_refs->lock); | |
874 | out: | |
875 | WARN_ON(num_refs == 0); | |
876 | if (refs) | |
877 | *refs = num_refs; | |
878 | if (flags) | |
879 | *flags = extent_flags; | |
880 | out_free: | |
881 | btrfs_free_path(path); | |
882 | return ret; | |
883 | } | |
884 | ||
d8d5f3e1 CM |
885 | /* |
886 | * Back reference rules. Back refs have three main goals: | |
887 | * | |
888 | * 1) differentiate between all holders of references to an extent so that | |
889 | * when a reference is dropped we can make sure it was a valid reference | |
890 | * before freeing the extent. | |
891 | * | |
892 | * 2) Provide enough information to quickly find the holders of an extent | |
893 | * if we notice a given block is corrupted or bad. | |
894 | * | |
895 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
896 | * maintenance. This is actually the same as #2, but with a slightly | |
897 | * different use case. | |
898 | * | |
5d4f98a2 YZ |
899 | * There are two kinds of back refs. The implicit back refs is optimized |
900 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
901 | * back refs of this kind provide information about the block's owner tree | |
902 | * and the pointer's key. These information allow us to find the block by | |
903 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
904 | * referenced by their owner trees. The location of tree block is recorded | |
905 | * in the back refs. Actually the full back refs is generic, and can be | |
906 | * used in all cases the implicit back refs is used. The major shortcoming | |
907 | * of the full back refs is its overhead. Every time a tree block gets | |
908 | * COWed, we have to update back refs entry for all pointers in it. | |
909 | * | |
910 | * For a newly allocated tree block, we use implicit back refs for | |
911 | * pointers in it. This means most tree related operations only involve | |
912 | * implicit back refs. For a tree block created in old transaction, the | |
913 | * only way to drop a reference to it is COW it. So we can detect the | |
914 | * event that tree block loses its owner tree's reference and do the | |
915 | * back refs conversion. | |
916 | * | |
917 | * When a tree block is COW'd through a tree, there are four cases: | |
918 | * | |
919 | * The reference count of the block is one and the tree is the block's | |
920 | * owner tree. Nothing to do in this case. | |
921 | * | |
922 | * The reference count of the block is one and the tree is not the | |
923 | * block's owner tree. In this case, full back refs is used for pointers | |
924 | * in the block. Remove these full back refs, add implicit back refs for | |
925 | * every pointers in the new block. | |
926 | * | |
927 | * The reference count of the block is greater than one and the tree is | |
928 | * the block's owner tree. In this case, implicit back refs is used for | |
929 | * pointers in the block. Add full back refs for every pointers in the | |
930 | * block, increase lower level extents' reference counts. The original | |
931 | * implicit back refs are entailed to the new block. | |
932 | * | |
933 | * The reference count of the block is greater than one and the tree is | |
934 | * not the block's owner tree. Add implicit back refs for every pointer in | |
935 | * the new block, increase lower level extents' reference count. | |
936 | * | |
937 | * Back Reference Key composing: | |
938 | * | |
939 | * The key objectid corresponds to the first byte in the extent, | |
940 | * The key type is used to differentiate between types of back refs. | |
941 | * There are different meanings of the key offset for different types | |
942 | * of back refs. | |
943 | * | |
d8d5f3e1 CM |
944 | * File extents can be referenced by: |
945 | * | |
946 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 947 | * - different files inside a single subvolume |
d8d5f3e1 CM |
948 | * - different offsets inside a file (bookend extents in file.c) |
949 | * | |
5d4f98a2 | 950 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
951 | * |
952 | * - Objectid of the subvolume root | |
d8d5f3e1 | 953 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
954 | * - original offset in the file |
955 | * - how many bookend extents | |
d8d5f3e1 | 956 | * |
5d4f98a2 YZ |
957 | * The key offset for the implicit back refs is hash of the first |
958 | * three fields. | |
d8d5f3e1 | 959 | * |
5d4f98a2 | 960 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 961 | * |
5d4f98a2 | 962 | * - number of pointers in the tree leaf |
d8d5f3e1 | 963 | * |
5d4f98a2 YZ |
964 | * The key offset for the implicit back refs is the first byte of |
965 | * the tree leaf | |
d8d5f3e1 | 966 | * |
5d4f98a2 YZ |
967 | * When a file extent is allocated, The implicit back refs is used. |
968 | * the fields are filled in: | |
d8d5f3e1 | 969 | * |
5d4f98a2 | 970 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 971 | * |
5d4f98a2 YZ |
972 | * When a file extent is removed file truncation, we find the |
973 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 974 | * |
5d4f98a2 | 975 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 976 | * |
5d4f98a2 | 977 | * Btree extents can be referenced by: |
d8d5f3e1 | 978 | * |
5d4f98a2 | 979 | * - Different subvolumes |
d8d5f3e1 | 980 | * |
5d4f98a2 YZ |
981 | * Both the implicit back refs and the full back refs for tree blocks |
982 | * only consist of key. The key offset for the implicit back refs is | |
983 | * objectid of block's owner tree. The key offset for the full back refs | |
984 | * is the first byte of parent block. | |
d8d5f3e1 | 985 | * |
5d4f98a2 YZ |
986 | * When implicit back refs is used, information about the lowest key and |
987 | * level of the tree block are required. These information are stored in | |
988 | * tree block info structure. | |
d8d5f3e1 | 989 | */ |
31840ae1 | 990 | |
5d4f98a2 YZ |
991 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
992 | static int convert_extent_item_v0(struct btrfs_trans_handle *trans, | |
993 | struct btrfs_root *root, | |
994 | struct btrfs_path *path, | |
995 | u64 owner, u32 extra_size) | |
7bb86316 | 996 | { |
5d4f98a2 YZ |
997 | struct btrfs_extent_item *item; |
998 | struct btrfs_extent_item_v0 *ei0; | |
999 | struct btrfs_extent_ref_v0 *ref0; | |
1000 | struct btrfs_tree_block_info *bi; | |
1001 | struct extent_buffer *leaf; | |
7bb86316 | 1002 | struct btrfs_key key; |
5d4f98a2 YZ |
1003 | struct btrfs_key found_key; |
1004 | u32 new_size = sizeof(*item); | |
1005 | u64 refs; | |
1006 | int ret; | |
1007 | ||
1008 | leaf = path->nodes[0]; | |
1009 | BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); | |
1010 | ||
1011 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1012 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
1013 | struct btrfs_extent_item_v0); | |
1014 | refs = btrfs_extent_refs_v0(leaf, ei0); | |
1015 | ||
1016 | if (owner == (u64)-1) { | |
1017 | while (1) { | |
1018 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1019 | ret = btrfs_next_leaf(root, path); | |
1020 | if (ret < 0) | |
1021 | return ret; | |
79787eaa | 1022 | BUG_ON(ret > 0); /* Corruption */ |
5d4f98a2 YZ |
1023 | leaf = path->nodes[0]; |
1024 | } | |
1025 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1026 | path->slots[0]); | |
1027 | BUG_ON(key.objectid != found_key.objectid); | |
1028 | if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { | |
1029 | path->slots[0]++; | |
1030 | continue; | |
1031 | } | |
1032 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1033 | struct btrfs_extent_ref_v0); | |
1034 | owner = btrfs_ref_objectid_v0(leaf, ref0); | |
1035 | break; | |
1036 | } | |
1037 | } | |
b3b4aa74 | 1038 | btrfs_release_path(path); |
5d4f98a2 YZ |
1039 | |
1040 | if (owner < BTRFS_FIRST_FREE_OBJECTID) | |
1041 | new_size += sizeof(*bi); | |
1042 | ||
1043 | new_size -= sizeof(*ei0); | |
1044 | ret = btrfs_search_slot(trans, root, &key, path, | |
1045 | new_size + extra_size, 1); | |
1046 | if (ret < 0) | |
1047 | return ret; | |
79787eaa | 1048 | BUG_ON(ret); /* Corruption */ |
5d4f98a2 | 1049 | |
4b90c680 | 1050 | btrfs_extend_item(root, path, new_size); |
5d4f98a2 YZ |
1051 | |
1052 | leaf = path->nodes[0]; | |
1053 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1054 | btrfs_set_extent_refs(leaf, item, refs); | |
1055 | /* FIXME: get real generation */ | |
1056 | btrfs_set_extent_generation(leaf, item, 0); | |
1057 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1058 | btrfs_set_extent_flags(leaf, item, | |
1059 | BTRFS_EXTENT_FLAG_TREE_BLOCK | | |
1060 | BTRFS_BLOCK_FLAG_FULL_BACKREF); | |
1061 | bi = (struct btrfs_tree_block_info *)(item + 1); | |
1062 | /* FIXME: get first key of the block */ | |
1063 | memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi)); | |
1064 | btrfs_set_tree_block_level(leaf, bi, (int)owner); | |
1065 | } else { | |
1066 | btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); | |
1067 | } | |
1068 | btrfs_mark_buffer_dirty(leaf); | |
1069 | return 0; | |
1070 | } | |
1071 | #endif | |
1072 | ||
1073 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) | |
1074 | { | |
1075 | u32 high_crc = ~(u32)0; | |
1076 | u32 low_crc = ~(u32)0; | |
1077 | __le64 lenum; | |
1078 | ||
1079 | lenum = cpu_to_le64(root_objectid); | |
14a958e6 | 1080 | high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1081 | lenum = cpu_to_le64(owner); |
14a958e6 | 1082 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1083 | lenum = cpu_to_le64(offset); |
14a958e6 | 1084 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
1085 | |
1086 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
1087 | } | |
1088 | ||
1089 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
1090 | struct btrfs_extent_data_ref *ref) | |
1091 | { | |
1092 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
1093 | btrfs_extent_data_ref_objectid(leaf, ref), | |
1094 | btrfs_extent_data_ref_offset(leaf, ref)); | |
1095 | } | |
1096 | ||
1097 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1098 | struct btrfs_extent_data_ref *ref, | |
1099 | u64 root_objectid, u64 owner, u64 offset) | |
1100 | { | |
1101 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1102 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1103 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1104 | return 0; | |
1105 | return 1; | |
1106 | } | |
1107 | ||
1108 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
1109 | struct btrfs_root *root, | |
1110 | struct btrfs_path *path, | |
1111 | u64 bytenr, u64 parent, | |
1112 | u64 root_objectid, | |
1113 | u64 owner, u64 offset) | |
1114 | { | |
1115 | struct btrfs_key key; | |
1116 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1117 | struct extent_buffer *leaf; |
5d4f98a2 | 1118 | u32 nritems; |
74493f7a | 1119 | int ret; |
5d4f98a2 YZ |
1120 | int recow; |
1121 | int err = -ENOENT; | |
74493f7a | 1122 | |
31840ae1 | 1123 | key.objectid = bytenr; |
5d4f98a2 YZ |
1124 | if (parent) { |
1125 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1126 | key.offset = parent; | |
1127 | } else { | |
1128 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1129 | key.offset = hash_extent_data_ref(root_objectid, | |
1130 | owner, offset); | |
1131 | } | |
1132 | again: | |
1133 | recow = 0; | |
1134 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1135 | if (ret < 0) { | |
1136 | err = ret; | |
1137 | goto fail; | |
1138 | } | |
31840ae1 | 1139 | |
5d4f98a2 YZ |
1140 | if (parent) { |
1141 | if (!ret) | |
1142 | return 0; | |
1143 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1144 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
b3b4aa74 | 1145 | btrfs_release_path(path); |
5d4f98a2 YZ |
1146 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1147 | if (ret < 0) { | |
1148 | err = ret; | |
1149 | goto fail; | |
1150 | } | |
1151 | if (!ret) | |
1152 | return 0; | |
1153 | #endif | |
1154 | goto fail; | |
31840ae1 ZY |
1155 | } |
1156 | ||
1157 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1158 | nritems = btrfs_header_nritems(leaf); |
1159 | while (1) { | |
1160 | if (path->slots[0] >= nritems) { | |
1161 | ret = btrfs_next_leaf(root, path); | |
1162 | if (ret < 0) | |
1163 | err = ret; | |
1164 | if (ret) | |
1165 | goto fail; | |
1166 | ||
1167 | leaf = path->nodes[0]; | |
1168 | nritems = btrfs_header_nritems(leaf); | |
1169 | recow = 1; | |
1170 | } | |
1171 | ||
1172 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1173 | if (key.objectid != bytenr || | |
1174 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1175 | goto fail; | |
1176 | ||
1177 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1178 | struct btrfs_extent_data_ref); | |
1179 | ||
1180 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1181 | owner, offset)) { | |
1182 | if (recow) { | |
b3b4aa74 | 1183 | btrfs_release_path(path); |
5d4f98a2 YZ |
1184 | goto again; |
1185 | } | |
1186 | err = 0; | |
1187 | break; | |
1188 | } | |
1189 | path->slots[0]++; | |
31840ae1 | 1190 | } |
5d4f98a2 YZ |
1191 | fail: |
1192 | return err; | |
31840ae1 ZY |
1193 | } |
1194 | ||
5d4f98a2 YZ |
1195 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
1196 | struct btrfs_root *root, | |
1197 | struct btrfs_path *path, | |
1198 | u64 bytenr, u64 parent, | |
1199 | u64 root_objectid, u64 owner, | |
1200 | u64 offset, int refs_to_add) | |
31840ae1 ZY |
1201 | { |
1202 | struct btrfs_key key; | |
1203 | struct extent_buffer *leaf; | |
5d4f98a2 | 1204 | u32 size; |
31840ae1 ZY |
1205 | u32 num_refs; |
1206 | int ret; | |
74493f7a | 1207 | |
74493f7a | 1208 | key.objectid = bytenr; |
5d4f98a2 YZ |
1209 | if (parent) { |
1210 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1211 | key.offset = parent; | |
1212 | size = sizeof(struct btrfs_shared_data_ref); | |
1213 | } else { | |
1214 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1215 | key.offset = hash_extent_data_ref(root_objectid, | |
1216 | owner, offset); | |
1217 | size = sizeof(struct btrfs_extent_data_ref); | |
1218 | } | |
74493f7a | 1219 | |
5d4f98a2 YZ |
1220 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1221 | if (ret && ret != -EEXIST) | |
1222 | goto fail; | |
1223 | ||
1224 | leaf = path->nodes[0]; | |
1225 | if (parent) { | |
1226 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1227 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1228 | struct btrfs_shared_data_ref); |
1229 | if (ret == 0) { | |
1230 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1231 | } else { | |
1232 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1233 | num_refs += refs_to_add; | |
1234 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1235 | } |
5d4f98a2 YZ |
1236 | } else { |
1237 | struct btrfs_extent_data_ref *ref; | |
1238 | while (ret == -EEXIST) { | |
1239 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1240 | struct btrfs_extent_data_ref); | |
1241 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1242 | owner, offset)) | |
1243 | break; | |
b3b4aa74 | 1244 | btrfs_release_path(path); |
5d4f98a2 YZ |
1245 | key.offset++; |
1246 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1247 | size); | |
1248 | if (ret && ret != -EEXIST) | |
1249 | goto fail; | |
31840ae1 | 1250 | |
5d4f98a2 YZ |
1251 | leaf = path->nodes[0]; |
1252 | } | |
1253 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1254 | struct btrfs_extent_data_ref); | |
1255 | if (ret == 0) { | |
1256 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1257 | root_objectid); | |
1258 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1259 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1260 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1261 | } else { | |
1262 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1263 | num_refs += refs_to_add; | |
1264 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1265 | } |
31840ae1 | 1266 | } |
5d4f98a2 YZ |
1267 | btrfs_mark_buffer_dirty(leaf); |
1268 | ret = 0; | |
1269 | fail: | |
b3b4aa74 | 1270 | btrfs_release_path(path); |
7bb86316 | 1271 | return ret; |
74493f7a CM |
1272 | } |
1273 | ||
5d4f98a2 YZ |
1274 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
1275 | struct btrfs_root *root, | |
1276 | struct btrfs_path *path, | |
fcebe456 | 1277 | int refs_to_drop, int *last_ref) |
31840ae1 | 1278 | { |
5d4f98a2 YZ |
1279 | struct btrfs_key key; |
1280 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1281 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1282 | struct extent_buffer *leaf; |
5d4f98a2 | 1283 | u32 num_refs = 0; |
31840ae1 ZY |
1284 | int ret = 0; |
1285 | ||
1286 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1287 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1288 | ||
1289 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1290 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1291 | struct btrfs_extent_data_ref); | |
1292 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1293 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1294 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1295 | struct btrfs_shared_data_ref); | |
1296 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1297 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1298 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1299 | struct btrfs_extent_ref_v0 *ref0; | |
1300 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1301 | struct btrfs_extent_ref_v0); | |
1302 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
1303 | #endif | |
1304 | } else { | |
1305 | BUG(); | |
1306 | } | |
1307 | ||
56bec294 CM |
1308 | BUG_ON(num_refs < refs_to_drop); |
1309 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1310 | |
31840ae1 ZY |
1311 | if (num_refs == 0) { |
1312 | ret = btrfs_del_item(trans, root, path); | |
fcebe456 | 1313 | *last_ref = 1; |
31840ae1 | 1314 | } else { |
5d4f98a2 YZ |
1315 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1316 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1317 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1318 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
1319 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1320 | else { | |
1321 | struct btrfs_extent_ref_v0 *ref0; | |
1322 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1323 | struct btrfs_extent_ref_v0); | |
1324 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); | |
1325 | } | |
1326 | #endif | |
31840ae1 ZY |
1327 | btrfs_mark_buffer_dirty(leaf); |
1328 | } | |
31840ae1 ZY |
1329 | return ret; |
1330 | } | |
1331 | ||
5d4f98a2 YZ |
1332 | static noinline u32 extent_data_ref_count(struct btrfs_root *root, |
1333 | struct btrfs_path *path, | |
1334 | struct btrfs_extent_inline_ref *iref) | |
15916de8 | 1335 | { |
5d4f98a2 YZ |
1336 | struct btrfs_key key; |
1337 | struct extent_buffer *leaf; | |
1338 | struct btrfs_extent_data_ref *ref1; | |
1339 | struct btrfs_shared_data_ref *ref2; | |
1340 | u32 num_refs = 0; | |
1341 | ||
1342 | leaf = path->nodes[0]; | |
1343 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1344 | if (iref) { | |
1345 | if (btrfs_extent_inline_ref_type(leaf, iref) == | |
1346 | BTRFS_EXTENT_DATA_REF_KEY) { | |
1347 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1348 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1349 | } else { | |
1350 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1351 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1352 | } | |
1353 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1354 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1355 | struct btrfs_extent_data_ref); | |
1356 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1357 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1358 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1359 | struct btrfs_shared_data_ref); | |
1360 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1361 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1362 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1363 | struct btrfs_extent_ref_v0 *ref0; | |
1364 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1365 | struct btrfs_extent_ref_v0); | |
1366 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
4b4e25f2 | 1367 | #endif |
5d4f98a2 YZ |
1368 | } else { |
1369 | WARN_ON(1); | |
1370 | } | |
1371 | return num_refs; | |
1372 | } | |
15916de8 | 1373 | |
5d4f98a2 YZ |
1374 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
1375 | struct btrfs_root *root, | |
1376 | struct btrfs_path *path, | |
1377 | u64 bytenr, u64 parent, | |
1378 | u64 root_objectid) | |
1f3c79a2 | 1379 | { |
5d4f98a2 | 1380 | struct btrfs_key key; |
1f3c79a2 | 1381 | int ret; |
1f3c79a2 | 1382 | |
5d4f98a2 YZ |
1383 | key.objectid = bytenr; |
1384 | if (parent) { | |
1385 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1386 | key.offset = parent; | |
1387 | } else { | |
1388 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1389 | key.offset = root_objectid; | |
1f3c79a2 LH |
1390 | } |
1391 | ||
5d4f98a2 YZ |
1392 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1393 | if (ret > 0) | |
1394 | ret = -ENOENT; | |
1395 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1396 | if (ret == -ENOENT && parent) { | |
b3b4aa74 | 1397 | btrfs_release_path(path); |
5d4f98a2 YZ |
1398 | key.type = BTRFS_EXTENT_REF_V0_KEY; |
1399 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1400 | if (ret > 0) | |
1401 | ret = -ENOENT; | |
1402 | } | |
1f3c79a2 | 1403 | #endif |
5d4f98a2 | 1404 | return ret; |
1f3c79a2 LH |
1405 | } |
1406 | ||
5d4f98a2 YZ |
1407 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
1408 | struct btrfs_root *root, | |
1409 | struct btrfs_path *path, | |
1410 | u64 bytenr, u64 parent, | |
1411 | u64 root_objectid) | |
31840ae1 | 1412 | { |
5d4f98a2 | 1413 | struct btrfs_key key; |
31840ae1 | 1414 | int ret; |
31840ae1 | 1415 | |
5d4f98a2 YZ |
1416 | key.objectid = bytenr; |
1417 | if (parent) { | |
1418 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1419 | key.offset = parent; | |
1420 | } else { | |
1421 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1422 | key.offset = root_objectid; | |
1423 | } | |
1424 | ||
1425 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
b3b4aa74 | 1426 | btrfs_release_path(path); |
31840ae1 ZY |
1427 | return ret; |
1428 | } | |
1429 | ||
5d4f98a2 | 1430 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1431 | { |
5d4f98a2 YZ |
1432 | int type; |
1433 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1434 | if (parent > 0) | |
1435 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1436 | else | |
1437 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1438 | } else { | |
1439 | if (parent > 0) | |
1440 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1441 | else | |
1442 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1443 | } | |
1444 | return type; | |
31840ae1 | 1445 | } |
56bec294 | 1446 | |
2c47e605 YZ |
1447 | static int find_next_key(struct btrfs_path *path, int level, |
1448 | struct btrfs_key *key) | |
56bec294 | 1449 | |
02217ed2 | 1450 | { |
2c47e605 | 1451 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1452 | if (!path->nodes[level]) |
1453 | break; | |
5d4f98a2 YZ |
1454 | if (path->slots[level] + 1 >= |
1455 | btrfs_header_nritems(path->nodes[level])) | |
1456 | continue; | |
1457 | if (level == 0) | |
1458 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1459 | path->slots[level] + 1); | |
1460 | else | |
1461 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1462 | path->slots[level] + 1); | |
1463 | return 0; | |
1464 | } | |
1465 | return 1; | |
1466 | } | |
037e6390 | 1467 | |
5d4f98a2 YZ |
1468 | /* |
1469 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1470 | * to the address of inline back ref, and 0 is returned. | |
1471 | * | |
1472 | * if back ref isn't found, *ref_ret is set to the address where it | |
1473 | * should be inserted, and -ENOENT is returned. | |
1474 | * | |
1475 | * if insert is true and there are too many inline back refs, the path | |
1476 | * points to the extent item, and -EAGAIN is returned. | |
1477 | * | |
1478 | * NOTE: inline back refs are ordered in the same way that back ref | |
1479 | * items in the tree are ordered. | |
1480 | */ | |
1481 | static noinline_for_stack | |
1482 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1483 | struct btrfs_root *root, | |
1484 | struct btrfs_path *path, | |
1485 | struct btrfs_extent_inline_ref **ref_ret, | |
1486 | u64 bytenr, u64 num_bytes, | |
1487 | u64 parent, u64 root_objectid, | |
1488 | u64 owner, u64 offset, int insert) | |
1489 | { | |
1490 | struct btrfs_key key; | |
1491 | struct extent_buffer *leaf; | |
1492 | struct btrfs_extent_item *ei; | |
1493 | struct btrfs_extent_inline_ref *iref; | |
1494 | u64 flags; | |
1495 | u64 item_size; | |
1496 | unsigned long ptr; | |
1497 | unsigned long end; | |
1498 | int extra_size; | |
1499 | int type; | |
1500 | int want; | |
1501 | int ret; | |
1502 | int err = 0; | |
3173a18f JB |
1503 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
1504 | SKINNY_METADATA); | |
26b8003f | 1505 | |
db94535d | 1506 | key.objectid = bytenr; |
31840ae1 | 1507 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1508 | key.offset = num_bytes; |
31840ae1 | 1509 | |
5d4f98a2 YZ |
1510 | want = extent_ref_type(parent, owner); |
1511 | if (insert) { | |
1512 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1513 | path->keep_locks = 1; |
5d4f98a2 YZ |
1514 | } else |
1515 | extra_size = -1; | |
3173a18f JB |
1516 | |
1517 | /* | |
1518 | * Owner is our parent level, so we can just add one to get the level | |
1519 | * for the block we are interested in. | |
1520 | */ | |
1521 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1522 | key.type = BTRFS_METADATA_ITEM_KEY; | |
1523 | key.offset = owner; | |
1524 | } | |
1525 | ||
1526 | again: | |
5d4f98a2 | 1527 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
b9473439 | 1528 | if (ret < 0) { |
5d4f98a2 YZ |
1529 | err = ret; |
1530 | goto out; | |
1531 | } | |
3173a18f JB |
1532 | |
1533 | /* | |
1534 | * We may be a newly converted file system which still has the old fat | |
1535 | * extent entries for metadata, so try and see if we have one of those. | |
1536 | */ | |
1537 | if (ret > 0 && skinny_metadata) { | |
1538 | skinny_metadata = false; | |
1539 | if (path->slots[0]) { | |
1540 | path->slots[0]--; | |
1541 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1542 | path->slots[0]); | |
1543 | if (key.objectid == bytenr && | |
1544 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
1545 | key.offset == num_bytes) | |
1546 | ret = 0; | |
1547 | } | |
1548 | if (ret) { | |
9ce49a0b | 1549 | key.objectid = bytenr; |
3173a18f JB |
1550 | key.type = BTRFS_EXTENT_ITEM_KEY; |
1551 | key.offset = num_bytes; | |
1552 | btrfs_release_path(path); | |
1553 | goto again; | |
1554 | } | |
1555 | } | |
1556 | ||
79787eaa JM |
1557 | if (ret && !insert) { |
1558 | err = -ENOENT; | |
1559 | goto out; | |
fae7f21c | 1560 | } else if (WARN_ON(ret)) { |
492104c8 | 1561 | err = -EIO; |
492104c8 | 1562 | goto out; |
79787eaa | 1563 | } |
5d4f98a2 YZ |
1564 | |
1565 | leaf = path->nodes[0]; | |
1566 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1567 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1568 | if (item_size < sizeof(*ei)) { | |
1569 | if (!insert) { | |
1570 | err = -ENOENT; | |
1571 | goto out; | |
1572 | } | |
1573 | ret = convert_extent_item_v0(trans, root, path, owner, | |
1574 | extra_size); | |
1575 | if (ret < 0) { | |
1576 | err = ret; | |
1577 | goto out; | |
1578 | } | |
1579 | leaf = path->nodes[0]; | |
1580 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1581 | } | |
1582 | #endif | |
1583 | BUG_ON(item_size < sizeof(*ei)); | |
1584 | ||
5d4f98a2 YZ |
1585 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1586 | flags = btrfs_extent_flags(leaf, ei); | |
1587 | ||
1588 | ptr = (unsigned long)(ei + 1); | |
1589 | end = (unsigned long)ei + item_size; | |
1590 | ||
3173a18f | 1591 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
5d4f98a2 YZ |
1592 | ptr += sizeof(struct btrfs_tree_block_info); |
1593 | BUG_ON(ptr > end); | |
5d4f98a2 YZ |
1594 | } |
1595 | ||
1596 | err = -ENOENT; | |
1597 | while (1) { | |
1598 | if (ptr >= end) { | |
1599 | WARN_ON(ptr > end); | |
1600 | break; | |
1601 | } | |
1602 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1603 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1604 | if (want < type) | |
1605 | break; | |
1606 | if (want > type) { | |
1607 | ptr += btrfs_extent_inline_ref_size(type); | |
1608 | continue; | |
1609 | } | |
1610 | ||
1611 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1612 | struct btrfs_extent_data_ref *dref; | |
1613 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1614 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1615 | owner, offset)) { | |
1616 | err = 0; | |
1617 | break; | |
1618 | } | |
1619 | if (hash_extent_data_ref_item(leaf, dref) < | |
1620 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1621 | break; | |
1622 | } else { | |
1623 | u64 ref_offset; | |
1624 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1625 | if (parent > 0) { | |
1626 | if (parent == ref_offset) { | |
1627 | err = 0; | |
1628 | break; | |
1629 | } | |
1630 | if (ref_offset < parent) | |
1631 | break; | |
1632 | } else { | |
1633 | if (root_objectid == ref_offset) { | |
1634 | err = 0; | |
1635 | break; | |
1636 | } | |
1637 | if (ref_offset < root_objectid) | |
1638 | break; | |
1639 | } | |
1640 | } | |
1641 | ptr += btrfs_extent_inline_ref_size(type); | |
1642 | } | |
1643 | if (err == -ENOENT && insert) { | |
1644 | if (item_size + extra_size >= | |
1645 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1646 | err = -EAGAIN; | |
1647 | goto out; | |
1648 | } | |
1649 | /* | |
1650 | * To add new inline back ref, we have to make sure | |
1651 | * there is no corresponding back ref item. | |
1652 | * For simplicity, we just do not add new inline back | |
1653 | * ref if there is any kind of item for this block | |
1654 | */ | |
2c47e605 YZ |
1655 | if (find_next_key(path, 0, &key) == 0 && |
1656 | key.objectid == bytenr && | |
85d4198e | 1657 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1658 | err = -EAGAIN; |
1659 | goto out; | |
1660 | } | |
1661 | } | |
1662 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1663 | out: | |
85d4198e | 1664 | if (insert) { |
5d4f98a2 YZ |
1665 | path->keep_locks = 0; |
1666 | btrfs_unlock_up_safe(path, 1); | |
1667 | } | |
1668 | return err; | |
1669 | } | |
1670 | ||
1671 | /* | |
1672 | * helper to add new inline back ref | |
1673 | */ | |
1674 | static noinline_for_stack | |
fd279fae | 1675 | void setup_inline_extent_backref(struct btrfs_root *root, |
143bede5 JM |
1676 | struct btrfs_path *path, |
1677 | struct btrfs_extent_inline_ref *iref, | |
1678 | u64 parent, u64 root_objectid, | |
1679 | u64 owner, u64 offset, int refs_to_add, | |
1680 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1681 | { |
1682 | struct extent_buffer *leaf; | |
1683 | struct btrfs_extent_item *ei; | |
1684 | unsigned long ptr; | |
1685 | unsigned long end; | |
1686 | unsigned long item_offset; | |
1687 | u64 refs; | |
1688 | int size; | |
1689 | int type; | |
5d4f98a2 YZ |
1690 | |
1691 | leaf = path->nodes[0]; | |
1692 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1693 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1694 | ||
1695 | type = extent_ref_type(parent, owner); | |
1696 | size = btrfs_extent_inline_ref_size(type); | |
1697 | ||
4b90c680 | 1698 | btrfs_extend_item(root, path, size); |
5d4f98a2 YZ |
1699 | |
1700 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1701 | refs = btrfs_extent_refs(leaf, ei); | |
1702 | refs += refs_to_add; | |
1703 | btrfs_set_extent_refs(leaf, ei, refs); | |
1704 | if (extent_op) | |
1705 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1706 | ||
1707 | ptr = (unsigned long)ei + item_offset; | |
1708 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1709 | if (ptr < end - size) | |
1710 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1711 | end - size - ptr); | |
1712 | ||
1713 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1714 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1715 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1716 | struct btrfs_extent_data_ref *dref; | |
1717 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1718 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1719 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1720 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1721 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1722 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1723 | struct btrfs_shared_data_ref *sref; | |
1724 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1725 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1726 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1727 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1728 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1729 | } else { | |
1730 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1731 | } | |
1732 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1733 | } |
1734 | ||
1735 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
1736 | struct btrfs_root *root, | |
1737 | struct btrfs_path *path, | |
1738 | struct btrfs_extent_inline_ref **ref_ret, | |
1739 | u64 bytenr, u64 num_bytes, u64 parent, | |
1740 | u64 root_objectid, u64 owner, u64 offset) | |
1741 | { | |
1742 | int ret; | |
1743 | ||
1744 | ret = lookup_inline_extent_backref(trans, root, path, ref_ret, | |
1745 | bytenr, num_bytes, parent, | |
1746 | root_objectid, owner, offset, 0); | |
1747 | if (ret != -ENOENT) | |
54aa1f4d | 1748 | return ret; |
5d4f98a2 | 1749 | |
b3b4aa74 | 1750 | btrfs_release_path(path); |
5d4f98a2 YZ |
1751 | *ref_ret = NULL; |
1752 | ||
1753 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1754 | ret = lookup_tree_block_ref(trans, root, path, bytenr, parent, | |
1755 | root_objectid); | |
1756 | } else { | |
1757 | ret = lookup_extent_data_ref(trans, root, path, bytenr, parent, | |
1758 | root_objectid, owner, offset); | |
b9473439 | 1759 | } |
5d4f98a2 YZ |
1760 | return ret; |
1761 | } | |
31840ae1 | 1762 | |
5d4f98a2 YZ |
1763 | /* |
1764 | * helper to update/remove inline back ref | |
1765 | */ | |
1766 | static noinline_for_stack | |
afe5fea7 | 1767 | void update_inline_extent_backref(struct btrfs_root *root, |
143bede5 JM |
1768 | struct btrfs_path *path, |
1769 | struct btrfs_extent_inline_ref *iref, | |
1770 | int refs_to_mod, | |
fcebe456 JB |
1771 | struct btrfs_delayed_extent_op *extent_op, |
1772 | int *last_ref) | |
5d4f98a2 YZ |
1773 | { |
1774 | struct extent_buffer *leaf; | |
1775 | struct btrfs_extent_item *ei; | |
1776 | struct btrfs_extent_data_ref *dref = NULL; | |
1777 | struct btrfs_shared_data_ref *sref = NULL; | |
1778 | unsigned long ptr; | |
1779 | unsigned long end; | |
1780 | u32 item_size; | |
1781 | int size; | |
1782 | int type; | |
5d4f98a2 YZ |
1783 | u64 refs; |
1784 | ||
1785 | leaf = path->nodes[0]; | |
1786 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1787 | refs = btrfs_extent_refs(leaf, ei); | |
1788 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1789 | refs += refs_to_mod; | |
1790 | btrfs_set_extent_refs(leaf, ei, refs); | |
1791 | if (extent_op) | |
1792 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1793 | ||
1794 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1795 | ||
1796 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1797 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1798 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1799 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1800 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1801 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1802 | } else { | |
1803 | refs = 1; | |
1804 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1805 | } |
31840ae1 | 1806 | |
5d4f98a2 YZ |
1807 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1808 | refs += refs_to_mod; | |
1809 | ||
1810 | if (refs > 0) { | |
1811 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1812 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1813 | else | |
1814 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1815 | } else { | |
fcebe456 | 1816 | *last_ref = 1; |
5d4f98a2 YZ |
1817 | size = btrfs_extent_inline_ref_size(type); |
1818 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1819 | ptr = (unsigned long)iref; | |
1820 | end = (unsigned long)ei + item_size; | |
1821 | if (ptr + size < end) | |
1822 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1823 | end - ptr - size); | |
1824 | item_size -= size; | |
afe5fea7 | 1825 | btrfs_truncate_item(root, path, item_size, 1); |
5d4f98a2 YZ |
1826 | } |
1827 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1828 | } |
1829 | ||
1830 | static noinline_for_stack | |
1831 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1832 | struct btrfs_root *root, | |
1833 | struct btrfs_path *path, | |
1834 | u64 bytenr, u64 num_bytes, u64 parent, | |
1835 | u64 root_objectid, u64 owner, | |
1836 | u64 offset, int refs_to_add, | |
1837 | struct btrfs_delayed_extent_op *extent_op) | |
1838 | { | |
1839 | struct btrfs_extent_inline_ref *iref; | |
1840 | int ret; | |
1841 | ||
1842 | ret = lookup_inline_extent_backref(trans, root, path, &iref, | |
1843 | bytenr, num_bytes, parent, | |
1844 | root_objectid, owner, offset, 1); | |
1845 | if (ret == 0) { | |
1846 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
afe5fea7 | 1847 | update_inline_extent_backref(root, path, iref, |
fcebe456 | 1848 | refs_to_add, extent_op, NULL); |
5d4f98a2 | 1849 | } else if (ret == -ENOENT) { |
fd279fae | 1850 | setup_inline_extent_backref(root, path, iref, parent, |
143bede5 JM |
1851 | root_objectid, owner, offset, |
1852 | refs_to_add, extent_op); | |
1853 | ret = 0; | |
771ed689 | 1854 | } |
5d4f98a2 YZ |
1855 | return ret; |
1856 | } | |
31840ae1 | 1857 | |
5d4f98a2 YZ |
1858 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
1859 | struct btrfs_root *root, | |
1860 | struct btrfs_path *path, | |
1861 | u64 bytenr, u64 parent, u64 root_objectid, | |
1862 | u64 owner, u64 offset, int refs_to_add) | |
1863 | { | |
1864 | int ret; | |
1865 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1866 | BUG_ON(refs_to_add != 1); | |
1867 | ret = insert_tree_block_ref(trans, root, path, bytenr, | |
1868 | parent, root_objectid); | |
1869 | } else { | |
1870 | ret = insert_extent_data_ref(trans, root, path, bytenr, | |
1871 | parent, root_objectid, | |
1872 | owner, offset, refs_to_add); | |
1873 | } | |
1874 | return ret; | |
1875 | } | |
56bec294 | 1876 | |
5d4f98a2 YZ |
1877 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
1878 | struct btrfs_root *root, | |
1879 | struct btrfs_path *path, | |
1880 | struct btrfs_extent_inline_ref *iref, | |
fcebe456 | 1881 | int refs_to_drop, int is_data, int *last_ref) |
5d4f98a2 | 1882 | { |
143bede5 | 1883 | int ret = 0; |
b9473439 | 1884 | |
5d4f98a2 YZ |
1885 | BUG_ON(!is_data && refs_to_drop != 1); |
1886 | if (iref) { | |
afe5fea7 | 1887 | update_inline_extent_backref(root, path, iref, |
fcebe456 | 1888 | -refs_to_drop, NULL, last_ref); |
5d4f98a2 | 1889 | } else if (is_data) { |
fcebe456 JB |
1890 | ret = remove_extent_data_ref(trans, root, path, refs_to_drop, |
1891 | last_ref); | |
5d4f98a2 | 1892 | } else { |
fcebe456 | 1893 | *last_ref = 1; |
5d4f98a2 YZ |
1894 | ret = btrfs_del_item(trans, root, path); |
1895 | } | |
1896 | return ret; | |
1897 | } | |
1898 | ||
5378e607 | 1899 | static int btrfs_issue_discard(struct block_device *bdev, |
5d4f98a2 YZ |
1900 | u64 start, u64 len) |
1901 | { | |
5378e607 | 1902 | return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0); |
5d4f98a2 | 1903 | } |
5d4f98a2 YZ |
1904 | |
1905 | static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr, | |
5378e607 | 1906 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 1907 | { |
5d4f98a2 | 1908 | int ret; |
5378e607 | 1909 | u64 discarded_bytes = 0; |
a1d3c478 | 1910 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 1911 | |
e244a0ae | 1912 | |
5d4f98a2 | 1913 | /* Tell the block device(s) that the sectors can be discarded */ |
3ec706c8 | 1914 | ret = btrfs_map_block(root->fs_info, REQ_DISCARD, |
a1d3c478 | 1915 | bytenr, &num_bytes, &bbio, 0); |
79787eaa | 1916 | /* Error condition is -ENOMEM */ |
5d4f98a2 | 1917 | if (!ret) { |
a1d3c478 | 1918 | struct btrfs_bio_stripe *stripe = bbio->stripes; |
5d4f98a2 YZ |
1919 | int i; |
1920 | ||
5d4f98a2 | 1921 | |
a1d3c478 | 1922 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d5e2003c JB |
1923 | if (!stripe->dev->can_discard) |
1924 | continue; | |
1925 | ||
5378e607 LD |
1926 | ret = btrfs_issue_discard(stripe->dev->bdev, |
1927 | stripe->physical, | |
1928 | stripe->length); | |
1929 | if (!ret) | |
1930 | discarded_bytes += stripe->length; | |
1931 | else if (ret != -EOPNOTSUPP) | |
79787eaa | 1932 | break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */ |
d5e2003c JB |
1933 | |
1934 | /* | |
1935 | * Just in case we get back EOPNOTSUPP for some reason, | |
1936 | * just ignore the return value so we don't screw up | |
1937 | * people calling discard_extent. | |
1938 | */ | |
1939 | ret = 0; | |
5d4f98a2 | 1940 | } |
a1d3c478 | 1941 | kfree(bbio); |
5d4f98a2 | 1942 | } |
5378e607 LD |
1943 | |
1944 | if (actual_bytes) | |
1945 | *actual_bytes = discarded_bytes; | |
1946 | ||
5d4f98a2 | 1947 | |
53b381b3 DW |
1948 | if (ret == -EOPNOTSUPP) |
1949 | ret = 0; | |
5d4f98a2 | 1950 | return ret; |
5d4f98a2 YZ |
1951 | } |
1952 | ||
79787eaa | 1953 | /* Can return -ENOMEM */ |
5d4f98a2 YZ |
1954 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
1955 | struct btrfs_root *root, | |
1956 | u64 bytenr, u64 num_bytes, u64 parent, | |
fcebe456 JB |
1957 | u64 root_objectid, u64 owner, u64 offset, |
1958 | int no_quota) | |
5d4f98a2 YZ |
1959 | { |
1960 | int ret; | |
66d7e7f0 AJ |
1961 | struct btrfs_fs_info *fs_info = root->fs_info; |
1962 | ||
5d4f98a2 YZ |
1963 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && |
1964 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
1965 | ||
1966 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
66d7e7f0 AJ |
1967 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
1968 | num_bytes, | |
5d4f98a2 | 1969 | parent, root_objectid, (int)owner, |
fcebe456 | 1970 | BTRFS_ADD_DELAYED_REF, NULL, no_quota); |
5d4f98a2 | 1971 | } else { |
66d7e7f0 AJ |
1972 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
1973 | num_bytes, | |
5d4f98a2 | 1974 | parent, root_objectid, owner, offset, |
fcebe456 | 1975 | BTRFS_ADD_DELAYED_REF, NULL, no_quota); |
5d4f98a2 YZ |
1976 | } |
1977 | return ret; | |
1978 | } | |
1979 | ||
1980 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
1981 | struct btrfs_root *root, | |
1982 | u64 bytenr, u64 num_bytes, | |
1983 | u64 parent, u64 root_objectid, | |
1984 | u64 owner, u64 offset, int refs_to_add, | |
fcebe456 | 1985 | int no_quota, |
5d4f98a2 YZ |
1986 | struct btrfs_delayed_extent_op *extent_op) |
1987 | { | |
fcebe456 | 1988 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
1989 | struct btrfs_path *path; |
1990 | struct extent_buffer *leaf; | |
1991 | struct btrfs_extent_item *item; | |
fcebe456 | 1992 | struct btrfs_key key; |
5d4f98a2 YZ |
1993 | u64 refs; |
1994 | int ret; | |
fcebe456 | 1995 | enum btrfs_qgroup_operation_type type = BTRFS_QGROUP_OPER_ADD_EXCL; |
5d4f98a2 YZ |
1996 | |
1997 | path = btrfs_alloc_path(); | |
1998 | if (!path) | |
1999 | return -ENOMEM; | |
2000 | ||
fcebe456 JB |
2001 | if (!is_fstree(root_objectid) || !root->fs_info->quota_enabled) |
2002 | no_quota = 1; | |
2003 | ||
5d4f98a2 YZ |
2004 | path->reada = 1; |
2005 | path->leave_spinning = 1; | |
2006 | /* this will setup the path even if it fails to insert the back ref */ | |
fcebe456 JB |
2007 | ret = insert_inline_extent_backref(trans, fs_info->extent_root, path, |
2008 | bytenr, num_bytes, parent, | |
5d4f98a2 YZ |
2009 | root_objectid, owner, offset, |
2010 | refs_to_add, extent_op); | |
fcebe456 | 2011 | if ((ret < 0 && ret != -EAGAIN) || (!ret && no_quota)) |
5d4f98a2 | 2012 | goto out; |
fcebe456 JB |
2013 | /* |
2014 | * Ok we were able to insert an inline extent and it appears to be a new | |
2015 | * reference, deal with the qgroup accounting. | |
2016 | */ | |
2017 | if (!ret && !no_quota) { | |
2018 | ASSERT(root->fs_info->quota_enabled); | |
2019 | leaf = path->nodes[0]; | |
2020 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2021 | item = btrfs_item_ptr(leaf, path->slots[0], | |
2022 | struct btrfs_extent_item); | |
2023 | if (btrfs_extent_refs(leaf, item) > (u64)refs_to_add) | |
2024 | type = BTRFS_QGROUP_OPER_ADD_SHARED; | |
2025 | btrfs_release_path(path); | |
5d4f98a2 | 2026 | |
fcebe456 JB |
2027 | ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid, |
2028 | bytenr, num_bytes, type, 0); | |
2029 | goto out; | |
2030 | } | |
2031 | ||
2032 | /* | |
2033 | * Ok we had -EAGAIN which means we didn't have space to insert and | |
2034 | * inline extent ref, so just update the reference count and add a | |
2035 | * normal backref. | |
2036 | */ | |
5d4f98a2 | 2037 | leaf = path->nodes[0]; |
fcebe456 | 2038 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
5d4f98a2 YZ |
2039 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
2040 | refs = btrfs_extent_refs(leaf, item); | |
fcebe456 JB |
2041 | if (refs) |
2042 | type = BTRFS_QGROUP_OPER_ADD_SHARED; | |
5d4f98a2 YZ |
2043 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); |
2044 | if (extent_op) | |
2045 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 2046 | |
5d4f98a2 | 2047 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 2048 | btrfs_release_path(path); |
56bec294 | 2049 | |
fcebe456 JB |
2050 | if (!no_quota) { |
2051 | ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid, | |
2052 | bytenr, num_bytes, type, 0); | |
2053 | if (ret) | |
2054 | goto out; | |
2055 | } | |
2056 | ||
56bec294 | 2057 | path->reada = 1; |
b9473439 | 2058 | path->leave_spinning = 1; |
56bec294 CM |
2059 | /* now insert the actual backref */ |
2060 | ret = insert_extent_backref(trans, root->fs_info->extent_root, | |
5d4f98a2 YZ |
2061 | path, bytenr, parent, root_objectid, |
2062 | owner, offset, refs_to_add); | |
79787eaa JM |
2063 | if (ret) |
2064 | btrfs_abort_transaction(trans, root, ret); | |
5d4f98a2 | 2065 | out: |
56bec294 | 2066 | btrfs_free_path(path); |
30d133fc | 2067 | return ret; |
56bec294 CM |
2068 | } |
2069 | ||
5d4f98a2 YZ |
2070 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
2071 | struct btrfs_root *root, | |
2072 | struct btrfs_delayed_ref_node *node, | |
2073 | struct btrfs_delayed_extent_op *extent_op, | |
2074 | int insert_reserved) | |
56bec294 | 2075 | { |
5d4f98a2 YZ |
2076 | int ret = 0; |
2077 | struct btrfs_delayed_data_ref *ref; | |
2078 | struct btrfs_key ins; | |
2079 | u64 parent = 0; | |
2080 | u64 ref_root = 0; | |
2081 | u64 flags = 0; | |
2082 | ||
2083 | ins.objectid = node->bytenr; | |
2084 | ins.offset = node->num_bytes; | |
2085 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2086 | ||
2087 | ref = btrfs_delayed_node_to_data_ref(node); | |
599c75ec LB |
2088 | trace_run_delayed_data_ref(node, ref, node->action); |
2089 | ||
5d4f98a2 YZ |
2090 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
2091 | parent = ref->parent; | |
fcebe456 | 2092 | ref_root = ref->root; |
5d4f98a2 YZ |
2093 | |
2094 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2095 | if (extent_op) |
5d4f98a2 | 2096 | flags |= extent_op->flags_to_set; |
5d4f98a2 YZ |
2097 | ret = alloc_reserved_file_extent(trans, root, |
2098 | parent, ref_root, flags, | |
2099 | ref->objectid, ref->offset, | |
2100 | &ins, node->ref_mod); | |
5d4f98a2 YZ |
2101 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2102 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
2103 | node->num_bytes, parent, | |
2104 | ref_root, ref->objectid, | |
2105 | ref->offset, node->ref_mod, | |
fcebe456 | 2106 | node->no_quota, extent_op); |
5d4f98a2 YZ |
2107 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
2108 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
2109 | node->num_bytes, parent, | |
2110 | ref_root, ref->objectid, | |
2111 | ref->offset, node->ref_mod, | |
fcebe456 | 2112 | extent_op, node->no_quota); |
5d4f98a2 YZ |
2113 | } else { |
2114 | BUG(); | |
2115 | } | |
2116 | return ret; | |
2117 | } | |
2118 | ||
2119 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
2120 | struct extent_buffer *leaf, | |
2121 | struct btrfs_extent_item *ei) | |
2122 | { | |
2123 | u64 flags = btrfs_extent_flags(leaf, ei); | |
2124 | if (extent_op->update_flags) { | |
2125 | flags |= extent_op->flags_to_set; | |
2126 | btrfs_set_extent_flags(leaf, ei, flags); | |
2127 | } | |
2128 | ||
2129 | if (extent_op->update_key) { | |
2130 | struct btrfs_tree_block_info *bi; | |
2131 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
2132 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
2133 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
2134 | } | |
2135 | } | |
2136 | ||
2137 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
2138 | struct btrfs_root *root, | |
2139 | struct btrfs_delayed_ref_node *node, | |
2140 | struct btrfs_delayed_extent_op *extent_op) | |
2141 | { | |
2142 | struct btrfs_key key; | |
2143 | struct btrfs_path *path; | |
2144 | struct btrfs_extent_item *ei; | |
2145 | struct extent_buffer *leaf; | |
2146 | u32 item_size; | |
56bec294 | 2147 | int ret; |
5d4f98a2 | 2148 | int err = 0; |
b1c79e09 | 2149 | int metadata = !extent_op->is_data; |
5d4f98a2 | 2150 | |
79787eaa JM |
2151 | if (trans->aborted) |
2152 | return 0; | |
2153 | ||
3173a18f JB |
2154 | if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) |
2155 | metadata = 0; | |
2156 | ||
5d4f98a2 YZ |
2157 | path = btrfs_alloc_path(); |
2158 | if (!path) | |
2159 | return -ENOMEM; | |
2160 | ||
2161 | key.objectid = node->bytenr; | |
5d4f98a2 | 2162 | |
3173a18f | 2163 | if (metadata) { |
3173a18f | 2164 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 2165 | key.offset = extent_op->level; |
3173a18f JB |
2166 | } else { |
2167 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2168 | key.offset = node->num_bytes; | |
2169 | } | |
2170 | ||
2171 | again: | |
5d4f98a2 YZ |
2172 | path->reada = 1; |
2173 | path->leave_spinning = 1; | |
2174 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, | |
2175 | path, 0, 1); | |
2176 | if (ret < 0) { | |
2177 | err = ret; | |
2178 | goto out; | |
2179 | } | |
2180 | if (ret > 0) { | |
3173a18f | 2181 | if (metadata) { |
55994887 FDBM |
2182 | if (path->slots[0] > 0) { |
2183 | path->slots[0]--; | |
2184 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
2185 | path->slots[0]); | |
2186 | if (key.objectid == node->bytenr && | |
2187 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
2188 | key.offset == node->num_bytes) | |
2189 | ret = 0; | |
2190 | } | |
2191 | if (ret > 0) { | |
2192 | btrfs_release_path(path); | |
2193 | metadata = 0; | |
3173a18f | 2194 | |
55994887 FDBM |
2195 | key.objectid = node->bytenr; |
2196 | key.offset = node->num_bytes; | |
2197 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2198 | goto again; | |
2199 | } | |
2200 | } else { | |
2201 | err = -EIO; | |
2202 | goto out; | |
3173a18f | 2203 | } |
5d4f98a2 YZ |
2204 | } |
2205 | ||
2206 | leaf = path->nodes[0]; | |
2207 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2208 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2209 | if (item_size < sizeof(*ei)) { | |
2210 | ret = convert_extent_item_v0(trans, root->fs_info->extent_root, | |
2211 | path, (u64)-1, 0); | |
2212 | if (ret < 0) { | |
2213 | err = ret; | |
2214 | goto out; | |
2215 | } | |
2216 | leaf = path->nodes[0]; | |
2217 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2218 | } | |
2219 | #endif | |
2220 | BUG_ON(item_size < sizeof(*ei)); | |
2221 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
2222 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2223 | |
5d4f98a2 YZ |
2224 | btrfs_mark_buffer_dirty(leaf); |
2225 | out: | |
2226 | btrfs_free_path(path); | |
2227 | return err; | |
56bec294 CM |
2228 | } |
2229 | ||
5d4f98a2 YZ |
2230 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
2231 | struct btrfs_root *root, | |
2232 | struct btrfs_delayed_ref_node *node, | |
2233 | struct btrfs_delayed_extent_op *extent_op, | |
2234 | int insert_reserved) | |
56bec294 CM |
2235 | { |
2236 | int ret = 0; | |
5d4f98a2 YZ |
2237 | struct btrfs_delayed_tree_ref *ref; |
2238 | struct btrfs_key ins; | |
2239 | u64 parent = 0; | |
2240 | u64 ref_root = 0; | |
3173a18f JB |
2241 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
2242 | SKINNY_METADATA); | |
56bec294 | 2243 | |
5d4f98a2 | 2244 | ref = btrfs_delayed_node_to_tree_ref(node); |
599c75ec LB |
2245 | trace_run_delayed_tree_ref(node, ref, node->action); |
2246 | ||
5d4f98a2 YZ |
2247 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
2248 | parent = ref->parent; | |
fcebe456 | 2249 | ref_root = ref->root; |
5d4f98a2 | 2250 | |
3173a18f JB |
2251 | ins.objectid = node->bytenr; |
2252 | if (skinny_metadata) { | |
2253 | ins.offset = ref->level; | |
2254 | ins.type = BTRFS_METADATA_ITEM_KEY; | |
2255 | } else { | |
2256 | ins.offset = node->num_bytes; | |
2257 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2258 | } | |
2259 | ||
5d4f98a2 YZ |
2260 | BUG_ON(node->ref_mod != 1); |
2261 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2262 | BUG_ON(!extent_op || !extent_op->update_flags); |
5d4f98a2 YZ |
2263 | ret = alloc_reserved_tree_block(trans, root, |
2264 | parent, ref_root, | |
2265 | extent_op->flags_to_set, | |
2266 | &extent_op->key, | |
fcebe456 JB |
2267 | ref->level, &ins, |
2268 | node->no_quota); | |
5d4f98a2 YZ |
2269 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2270 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
2271 | node->num_bytes, parent, ref_root, | |
fcebe456 JB |
2272 | ref->level, 0, 1, node->no_quota, |
2273 | extent_op); | |
5d4f98a2 YZ |
2274 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
2275 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
2276 | node->num_bytes, parent, ref_root, | |
fcebe456 JB |
2277 | ref->level, 0, 1, extent_op, |
2278 | node->no_quota); | |
5d4f98a2 YZ |
2279 | } else { |
2280 | BUG(); | |
2281 | } | |
56bec294 CM |
2282 | return ret; |
2283 | } | |
2284 | ||
2285 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 YZ |
2286 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
2287 | struct btrfs_root *root, | |
2288 | struct btrfs_delayed_ref_node *node, | |
2289 | struct btrfs_delayed_extent_op *extent_op, | |
2290 | int insert_reserved) | |
56bec294 | 2291 | { |
79787eaa JM |
2292 | int ret = 0; |
2293 | ||
857cc2fc JB |
2294 | if (trans->aborted) { |
2295 | if (insert_reserved) | |
2296 | btrfs_pin_extent(root, node->bytenr, | |
2297 | node->num_bytes, 1); | |
79787eaa | 2298 | return 0; |
857cc2fc | 2299 | } |
79787eaa | 2300 | |
5d4f98a2 | 2301 | if (btrfs_delayed_ref_is_head(node)) { |
56bec294 CM |
2302 | struct btrfs_delayed_ref_head *head; |
2303 | /* | |
2304 | * we've hit the end of the chain and we were supposed | |
2305 | * to insert this extent into the tree. But, it got | |
2306 | * deleted before we ever needed to insert it, so all | |
2307 | * we have to do is clean up the accounting | |
2308 | */ | |
5d4f98a2 YZ |
2309 | BUG_ON(extent_op); |
2310 | head = btrfs_delayed_node_to_head(node); | |
599c75ec LB |
2311 | trace_run_delayed_ref_head(node, head, node->action); |
2312 | ||
56bec294 | 2313 | if (insert_reserved) { |
f0486c68 YZ |
2314 | btrfs_pin_extent(root, node->bytenr, |
2315 | node->num_bytes, 1); | |
5d4f98a2 YZ |
2316 | if (head->is_data) { |
2317 | ret = btrfs_del_csums(trans, root, | |
2318 | node->bytenr, | |
2319 | node->num_bytes); | |
5d4f98a2 | 2320 | } |
56bec294 | 2321 | } |
79787eaa | 2322 | return ret; |
56bec294 CM |
2323 | } |
2324 | ||
5d4f98a2 YZ |
2325 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2326 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2327 | ret = run_delayed_tree_ref(trans, root, node, extent_op, | |
2328 | insert_reserved); | |
2329 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2330 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2331 | ret = run_delayed_data_ref(trans, root, node, extent_op, | |
2332 | insert_reserved); | |
2333 | else | |
2334 | BUG(); | |
2335 | return ret; | |
56bec294 CM |
2336 | } |
2337 | ||
2338 | static noinline struct btrfs_delayed_ref_node * | |
2339 | select_delayed_ref(struct btrfs_delayed_ref_head *head) | |
2340 | { | |
2341 | struct rb_node *node; | |
d7df2c79 JB |
2342 | struct btrfs_delayed_ref_node *ref, *last = NULL;; |
2343 | ||
56bec294 CM |
2344 | /* |
2345 | * select delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2346 | * this prevents ref count from going down to zero when | |
2347 | * there still are pending delayed ref. | |
2348 | */ | |
d7df2c79 JB |
2349 | node = rb_first(&head->ref_root); |
2350 | while (node) { | |
56bec294 CM |
2351 | ref = rb_entry(node, struct btrfs_delayed_ref_node, |
2352 | rb_node); | |
d7df2c79 | 2353 | if (ref->action == BTRFS_ADD_DELAYED_REF) |
56bec294 | 2354 | return ref; |
d7df2c79 JB |
2355 | else if (last == NULL) |
2356 | last = ref; | |
2357 | node = rb_next(node); | |
56bec294 | 2358 | } |
d7df2c79 | 2359 | return last; |
56bec294 CM |
2360 | } |
2361 | ||
79787eaa JM |
2362 | /* |
2363 | * Returns 0 on success or if called with an already aborted transaction. | |
2364 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
2365 | */ | |
d7df2c79 JB |
2366 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
2367 | struct btrfs_root *root, | |
2368 | unsigned long nr) | |
56bec294 | 2369 | { |
56bec294 CM |
2370 | struct btrfs_delayed_ref_root *delayed_refs; |
2371 | struct btrfs_delayed_ref_node *ref; | |
2372 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
5d4f98a2 | 2373 | struct btrfs_delayed_extent_op *extent_op; |
097b8a7c | 2374 | struct btrfs_fs_info *fs_info = root->fs_info; |
0a2b2a84 | 2375 | ktime_t start = ktime_get(); |
56bec294 | 2376 | int ret; |
d7df2c79 | 2377 | unsigned long count = 0; |
0a2b2a84 | 2378 | unsigned long actual_count = 0; |
56bec294 | 2379 | int must_insert_reserved = 0; |
56bec294 CM |
2380 | |
2381 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
2382 | while (1) { |
2383 | if (!locked_ref) { | |
d7df2c79 | 2384 | if (count >= nr) |
56bec294 | 2385 | break; |
56bec294 | 2386 | |
d7df2c79 JB |
2387 | spin_lock(&delayed_refs->lock); |
2388 | locked_ref = btrfs_select_ref_head(trans); | |
2389 | if (!locked_ref) { | |
2390 | spin_unlock(&delayed_refs->lock); | |
2391 | break; | |
2392 | } | |
c3e69d58 CM |
2393 | |
2394 | /* grab the lock that says we are going to process | |
2395 | * all the refs for this head */ | |
2396 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
d7df2c79 | 2397 | spin_unlock(&delayed_refs->lock); |
c3e69d58 CM |
2398 | /* |
2399 | * we may have dropped the spin lock to get the head | |
2400 | * mutex lock, and that might have given someone else | |
2401 | * time to free the head. If that's true, it has been | |
2402 | * removed from our list and we can move on. | |
2403 | */ | |
2404 | if (ret == -EAGAIN) { | |
2405 | locked_ref = NULL; | |
2406 | count++; | |
2407 | continue; | |
56bec294 CM |
2408 | } |
2409 | } | |
a28ec197 | 2410 | |
ae1e206b JB |
2411 | /* |
2412 | * We need to try and merge add/drops of the same ref since we | |
2413 | * can run into issues with relocate dropping the implicit ref | |
2414 | * and then it being added back again before the drop can | |
2415 | * finish. If we merged anything we need to re-loop so we can | |
2416 | * get a good ref. | |
2417 | */ | |
d7df2c79 | 2418 | spin_lock(&locked_ref->lock); |
ae1e206b JB |
2419 | btrfs_merge_delayed_refs(trans, fs_info, delayed_refs, |
2420 | locked_ref); | |
2421 | ||
d1270cd9 AJ |
2422 | /* |
2423 | * locked_ref is the head node, so we have to go one | |
2424 | * node back for any delayed ref updates | |
2425 | */ | |
2426 | ref = select_delayed_ref(locked_ref); | |
2427 | ||
2428 | if (ref && ref->seq && | |
097b8a7c | 2429 | btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) { |
d7df2c79 | 2430 | spin_unlock(&locked_ref->lock); |
093486c4 | 2431 | btrfs_delayed_ref_unlock(locked_ref); |
d7df2c79 JB |
2432 | spin_lock(&delayed_refs->lock); |
2433 | locked_ref->processing = 0; | |
d1270cd9 AJ |
2434 | delayed_refs->num_heads_ready++; |
2435 | spin_unlock(&delayed_refs->lock); | |
d7df2c79 | 2436 | locked_ref = NULL; |
d1270cd9 | 2437 | cond_resched(); |
27a377db | 2438 | count++; |
d1270cd9 AJ |
2439 | continue; |
2440 | } | |
2441 | ||
56bec294 CM |
2442 | /* |
2443 | * record the must insert reserved flag before we | |
2444 | * drop the spin lock. | |
2445 | */ | |
2446 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2447 | locked_ref->must_insert_reserved = 0; | |
7bb86316 | 2448 | |
5d4f98a2 YZ |
2449 | extent_op = locked_ref->extent_op; |
2450 | locked_ref->extent_op = NULL; | |
2451 | ||
56bec294 | 2452 | if (!ref) { |
d7df2c79 JB |
2453 | |
2454 | ||
56bec294 CM |
2455 | /* All delayed refs have been processed, Go ahead |
2456 | * and send the head node to run_one_delayed_ref, | |
2457 | * so that any accounting fixes can happen | |
2458 | */ | |
2459 | ref = &locked_ref->node; | |
5d4f98a2 YZ |
2460 | |
2461 | if (extent_op && must_insert_reserved) { | |
78a6184a | 2462 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2463 | extent_op = NULL; |
2464 | } | |
2465 | ||
2466 | if (extent_op) { | |
d7df2c79 | 2467 | spin_unlock(&locked_ref->lock); |
5d4f98a2 YZ |
2468 | ret = run_delayed_extent_op(trans, root, |
2469 | ref, extent_op); | |
78a6184a | 2470 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 | 2471 | |
79787eaa | 2472 | if (ret) { |
857cc2fc JB |
2473 | /* |
2474 | * Need to reset must_insert_reserved if | |
2475 | * there was an error so the abort stuff | |
2476 | * can cleanup the reserved space | |
2477 | * properly. | |
2478 | */ | |
2479 | if (must_insert_reserved) | |
2480 | locked_ref->must_insert_reserved = 1; | |
d7df2c79 | 2481 | locked_ref->processing = 0; |
c2cf52eb | 2482 | btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret); |
093486c4 | 2483 | btrfs_delayed_ref_unlock(locked_ref); |
79787eaa JM |
2484 | return ret; |
2485 | } | |
d7df2c79 | 2486 | continue; |
5d4f98a2 | 2487 | } |
02217ed2 | 2488 | |
d7df2c79 JB |
2489 | /* |
2490 | * Need to drop our head ref lock and re-aqcuire the | |
2491 | * delayed ref lock and then re-check to make sure | |
2492 | * nobody got added. | |
2493 | */ | |
2494 | spin_unlock(&locked_ref->lock); | |
2495 | spin_lock(&delayed_refs->lock); | |
2496 | spin_lock(&locked_ref->lock); | |
573a0755 JB |
2497 | if (rb_first(&locked_ref->ref_root) || |
2498 | locked_ref->extent_op) { | |
d7df2c79 JB |
2499 | spin_unlock(&locked_ref->lock); |
2500 | spin_unlock(&delayed_refs->lock); | |
2501 | continue; | |
2502 | } | |
2503 | ref->in_tree = 0; | |
2504 | delayed_refs->num_heads--; | |
c46effa6 LB |
2505 | rb_erase(&locked_ref->href_node, |
2506 | &delayed_refs->href_root); | |
d7df2c79 JB |
2507 | spin_unlock(&delayed_refs->lock); |
2508 | } else { | |
0a2b2a84 | 2509 | actual_count++; |
d7df2c79 JB |
2510 | ref->in_tree = 0; |
2511 | rb_erase(&ref->rb_node, &locked_ref->ref_root); | |
c46effa6 | 2512 | } |
d7df2c79 JB |
2513 | atomic_dec(&delayed_refs->num_entries); |
2514 | ||
093486c4 | 2515 | if (!btrfs_delayed_ref_is_head(ref)) { |
22cd2e7d AJ |
2516 | /* |
2517 | * when we play the delayed ref, also correct the | |
2518 | * ref_mod on head | |
2519 | */ | |
2520 | switch (ref->action) { | |
2521 | case BTRFS_ADD_DELAYED_REF: | |
2522 | case BTRFS_ADD_DELAYED_EXTENT: | |
2523 | locked_ref->node.ref_mod -= ref->ref_mod; | |
2524 | break; | |
2525 | case BTRFS_DROP_DELAYED_REF: | |
2526 | locked_ref->node.ref_mod += ref->ref_mod; | |
2527 | break; | |
2528 | default: | |
2529 | WARN_ON(1); | |
2530 | } | |
2531 | } | |
d7df2c79 | 2532 | spin_unlock(&locked_ref->lock); |
925baedd | 2533 | |
5d4f98a2 | 2534 | ret = run_one_delayed_ref(trans, root, ref, extent_op, |
56bec294 | 2535 | must_insert_reserved); |
eb099670 | 2536 | |
78a6184a | 2537 | btrfs_free_delayed_extent_op(extent_op); |
79787eaa | 2538 | if (ret) { |
d7df2c79 | 2539 | locked_ref->processing = 0; |
093486c4 MX |
2540 | btrfs_delayed_ref_unlock(locked_ref); |
2541 | btrfs_put_delayed_ref(ref); | |
c2cf52eb | 2542 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret); |
79787eaa JM |
2543 | return ret; |
2544 | } | |
2545 | ||
093486c4 MX |
2546 | /* |
2547 | * If this node is a head, that means all the refs in this head | |
2548 | * have been dealt with, and we will pick the next head to deal | |
2549 | * with, so we must unlock the head and drop it from the cluster | |
2550 | * list before we release it. | |
2551 | */ | |
2552 | if (btrfs_delayed_ref_is_head(ref)) { | |
093486c4 MX |
2553 | btrfs_delayed_ref_unlock(locked_ref); |
2554 | locked_ref = NULL; | |
2555 | } | |
2556 | btrfs_put_delayed_ref(ref); | |
2557 | count++; | |
c3e69d58 | 2558 | cond_resched(); |
c3e69d58 | 2559 | } |
0a2b2a84 JB |
2560 | |
2561 | /* | |
2562 | * We don't want to include ref heads since we can have empty ref heads | |
2563 | * and those will drastically skew our runtime down since we just do | |
2564 | * accounting, no actual extent tree updates. | |
2565 | */ | |
2566 | if (actual_count > 0) { | |
2567 | u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start)); | |
2568 | u64 avg; | |
2569 | ||
2570 | /* | |
2571 | * We weigh the current average higher than our current runtime | |
2572 | * to avoid large swings in the average. | |
2573 | */ | |
2574 | spin_lock(&delayed_refs->lock); | |
2575 | avg = fs_info->avg_delayed_ref_runtime * 3 + runtime; | |
2576 | avg = div64_u64(avg, 4); | |
2577 | fs_info->avg_delayed_ref_runtime = avg; | |
2578 | spin_unlock(&delayed_refs->lock); | |
2579 | } | |
d7df2c79 | 2580 | return 0; |
c3e69d58 CM |
2581 | } |
2582 | ||
709c0486 AJ |
2583 | #ifdef SCRAMBLE_DELAYED_REFS |
2584 | /* | |
2585 | * Normally delayed refs get processed in ascending bytenr order. This | |
2586 | * correlates in most cases to the order added. To expose dependencies on this | |
2587 | * order, we start to process the tree in the middle instead of the beginning | |
2588 | */ | |
2589 | static u64 find_middle(struct rb_root *root) | |
2590 | { | |
2591 | struct rb_node *n = root->rb_node; | |
2592 | struct btrfs_delayed_ref_node *entry; | |
2593 | int alt = 1; | |
2594 | u64 middle; | |
2595 | u64 first = 0, last = 0; | |
2596 | ||
2597 | n = rb_first(root); | |
2598 | if (n) { | |
2599 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2600 | first = entry->bytenr; | |
2601 | } | |
2602 | n = rb_last(root); | |
2603 | if (n) { | |
2604 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2605 | last = entry->bytenr; | |
2606 | } | |
2607 | n = root->rb_node; | |
2608 | ||
2609 | while (n) { | |
2610 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2611 | WARN_ON(!entry->in_tree); | |
2612 | ||
2613 | middle = entry->bytenr; | |
2614 | ||
2615 | if (alt) | |
2616 | n = n->rb_left; | |
2617 | else | |
2618 | n = n->rb_right; | |
2619 | ||
2620 | alt = 1 - alt; | |
2621 | } | |
2622 | return middle; | |
2623 | } | |
2624 | #endif | |
2625 | ||
1be41b78 JB |
2626 | static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads) |
2627 | { | |
2628 | u64 num_bytes; | |
2629 | ||
2630 | num_bytes = heads * (sizeof(struct btrfs_extent_item) + | |
2631 | sizeof(struct btrfs_extent_inline_ref)); | |
2632 | if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) | |
2633 | num_bytes += heads * sizeof(struct btrfs_tree_block_info); | |
2634 | ||
2635 | /* | |
2636 | * We don't ever fill up leaves all the way so multiply by 2 just to be | |
2637 | * closer to what we're really going to want to ouse. | |
2638 | */ | |
2639 | return div64_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root)); | |
2640 | } | |
2641 | ||
0a2b2a84 | 2642 | int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans, |
1be41b78 JB |
2643 | struct btrfs_root *root) |
2644 | { | |
2645 | struct btrfs_block_rsv *global_rsv; | |
2646 | u64 num_heads = trans->transaction->delayed_refs.num_heads_ready; | |
2647 | u64 num_bytes; | |
2648 | int ret = 0; | |
2649 | ||
2650 | num_bytes = btrfs_calc_trans_metadata_size(root, 1); | |
2651 | num_heads = heads_to_leaves(root, num_heads); | |
2652 | if (num_heads > 1) | |
2653 | num_bytes += (num_heads - 1) * root->leafsize; | |
2654 | num_bytes <<= 1; | |
2655 | global_rsv = &root->fs_info->global_block_rsv; | |
2656 | ||
2657 | /* | |
2658 | * If we can't allocate any more chunks lets make sure we have _lots_ of | |
2659 | * wiggle room since running delayed refs can create more delayed refs. | |
2660 | */ | |
2661 | if (global_rsv->space_info->full) | |
2662 | num_bytes <<= 1; | |
2663 | ||
2664 | spin_lock(&global_rsv->lock); | |
2665 | if (global_rsv->reserved <= num_bytes) | |
2666 | ret = 1; | |
2667 | spin_unlock(&global_rsv->lock); | |
2668 | return ret; | |
2669 | } | |
2670 | ||
0a2b2a84 JB |
2671 | int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans, |
2672 | struct btrfs_root *root) | |
2673 | { | |
2674 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2675 | u64 num_entries = | |
2676 | atomic_read(&trans->transaction->delayed_refs.num_entries); | |
2677 | u64 avg_runtime; | |
a79b7d4b | 2678 | u64 val; |
0a2b2a84 JB |
2679 | |
2680 | smp_mb(); | |
2681 | avg_runtime = fs_info->avg_delayed_ref_runtime; | |
a79b7d4b | 2682 | val = num_entries * avg_runtime; |
0a2b2a84 JB |
2683 | if (num_entries * avg_runtime >= NSEC_PER_SEC) |
2684 | return 1; | |
a79b7d4b CM |
2685 | if (val >= NSEC_PER_SEC / 2) |
2686 | return 2; | |
0a2b2a84 JB |
2687 | |
2688 | return btrfs_check_space_for_delayed_refs(trans, root); | |
2689 | } | |
2690 | ||
a79b7d4b CM |
2691 | struct async_delayed_refs { |
2692 | struct btrfs_root *root; | |
2693 | int count; | |
2694 | int error; | |
2695 | int sync; | |
2696 | struct completion wait; | |
2697 | struct btrfs_work work; | |
2698 | }; | |
2699 | ||
2700 | static void delayed_ref_async_start(struct btrfs_work *work) | |
2701 | { | |
2702 | struct async_delayed_refs *async; | |
2703 | struct btrfs_trans_handle *trans; | |
2704 | int ret; | |
2705 | ||
2706 | async = container_of(work, struct async_delayed_refs, work); | |
2707 | ||
2708 | trans = btrfs_join_transaction(async->root); | |
2709 | if (IS_ERR(trans)) { | |
2710 | async->error = PTR_ERR(trans); | |
2711 | goto done; | |
2712 | } | |
2713 | ||
2714 | /* | |
2715 | * trans->sync means that when we call end_transaciton, we won't | |
2716 | * wait on delayed refs | |
2717 | */ | |
2718 | trans->sync = true; | |
2719 | ret = btrfs_run_delayed_refs(trans, async->root, async->count); | |
2720 | if (ret) | |
2721 | async->error = ret; | |
2722 | ||
2723 | ret = btrfs_end_transaction(trans, async->root); | |
2724 | if (ret && !async->error) | |
2725 | async->error = ret; | |
2726 | done: | |
2727 | if (async->sync) | |
2728 | complete(&async->wait); | |
2729 | else | |
2730 | kfree(async); | |
2731 | } | |
2732 | ||
2733 | int btrfs_async_run_delayed_refs(struct btrfs_root *root, | |
2734 | unsigned long count, int wait) | |
2735 | { | |
2736 | struct async_delayed_refs *async; | |
2737 | int ret; | |
2738 | ||
2739 | async = kmalloc(sizeof(*async), GFP_NOFS); | |
2740 | if (!async) | |
2741 | return -ENOMEM; | |
2742 | ||
2743 | async->root = root->fs_info->tree_root; | |
2744 | async->count = count; | |
2745 | async->error = 0; | |
2746 | if (wait) | |
2747 | async->sync = 1; | |
2748 | else | |
2749 | async->sync = 0; | |
2750 | init_completion(&async->wait); | |
2751 | ||
2752 | btrfs_init_work(&async->work, delayed_ref_async_start, | |
2753 | NULL, NULL); | |
2754 | ||
2755 | btrfs_queue_work(root->fs_info->extent_workers, &async->work); | |
2756 | ||
2757 | if (wait) { | |
2758 | wait_for_completion(&async->wait); | |
2759 | ret = async->error; | |
2760 | kfree(async); | |
2761 | return ret; | |
2762 | } | |
2763 | return 0; | |
2764 | } | |
2765 | ||
c3e69d58 CM |
2766 | /* |
2767 | * this starts processing the delayed reference count updates and | |
2768 | * extent insertions we have queued up so far. count can be | |
2769 | * 0, which means to process everything in the tree at the start | |
2770 | * of the run (but not newly added entries), or it can be some target | |
2771 | * number you'd like to process. | |
79787eaa JM |
2772 | * |
2773 | * Returns 0 on success or if called with an aborted transaction | |
2774 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
2775 | */ |
2776 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
2777 | struct btrfs_root *root, unsigned long count) | |
2778 | { | |
2779 | struct rb_node *node; | |
2780 | struct btrfs_delayed_ref_root *delayed_refs; | |
c46effa6 | 2781 | struct btrfs_delayed_ref_head *head; |
c3e69d58 CM |
2782 | int ret; |
2783 | int run_all = count == (unsigned long)-1; | |
2784 | int run_most = 0; | |
2785 | ||
79787eaa JM |
2786 | /* We'll clean this up in btrfs_cleanup_transaction */ |
2787 | if (trans->aborted) | |
2788 | return 0; | |
2789 | ||
c3e69d58 CM |
2790 | if (root == root->fs_info->extent_root) |
2791 | root = root->fs_info->tree_root; | |
2792 | ||
2793 | delayed_refs = &trans->transaction->delayed_refs; | |
bb721703 | 2794 | if (count == 0) { |
d7df2c79 | 2795 | count = atomic_read(&delayed_refs->num_entries) * 2; |
bb721703 CM |
2796 | run_most = 1; |
2797 | } | |
2798 | ||
c3e69d58 | 2799 | again: |
709c0486 AJ |
2800 | #ifdef SCRAMBLE_DELAYED_REFS |
2801 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
2802 | #endif | |
d7df2c79 JB |
2803 | ret = __btrfs_run_delayed_refs(trans, root, count); |
2804 | if (ret < 0) { | |
2805 | btrfs_abort_transaction(trans, root, ret); | |
2806 | return ret; | |
eb099670 | 2807 | } |
c3e69d58 | 2808 | |
56bec294 | 2809 | if (run_all) { |
d7df2c79 | 2810 | if (!list_empty(&trans->new_bgs)) |
ea658bad | 2811 | btrfs_create_pending_block_groups(trans, root); |
ea658bad | 2812 | |
d7df2c79 | 2813 | spin_lock(&delayed_refs->lock); |
c46effa6 | 2814 | node = rb_first(&delayed_refs->href_root); |
d7df2c79 JB |
2815 | if (!node) { |
2816 | spin_unlock(&delayed_refs->lock); | |
56bec294 | 2817 | goto out; |
d7df2c79 | 2818 | } |
c3e69d58 | 2819 | count = (unsigned long)-1; |
e9d0b13b | 2820 | |
56bec294 | 2821 | while (node) { |
c46effa6 LB |
2822 | head = rb_entry(node, struct btrfs_delayed_ref_head, |
2823 | href_node); | |
2824 | if (btrfs_delayed_ref_is_head(&head->node)) { | |
2825 | struct btrfs_delayed_ref_node *ref; | |
5caf2a00 | 2826 | |
c46effa6 | 2827 | ref = &head->node; |
56bec294 CM |
2828 | atomic_inc(&ref->refs); |
2829 | ||
2830 | spin_unlock(&delayed_refs->lock); | |
8cc33e5c DS |
2831 | /* |
2832 | * Mutex was contended, block until it's | |
2833 | * released and try again | |
2834 | */ | |
56bec294 CM |
2835 | mutex_lock(&head->mutex); |
2836 | mutex_unlock(&head->mutex); | |
2837 | ||
2838 | btrfs_put_delayed_ref(ref); | |
1887be66 | 2839 | cond_resched(); |
56bec294 | 2840 | goto again; |
c46effa6 LB |
2841 | } else { |
2842 | WARN_ON(1); | |
56bec294 CM |
2843 | } |
2844 | node = rb_next(node); | |
2845 | } | |
2846 | spin_unlock(&delayed_refs->lock); | |
d7df2c79 | 2847 | cond_resched(); |
56bec294 | 2848 | goto again; |
5f39d397 | 2849 | } |
54aa1f4d | 2850 | out: |
fcebe456 JB |
2851 | ret = btrfs_delayed_qgroup_accounting(trans, root->fs_info); |
2852 | if (ret) | |
2853 | return ret; | |
edf39272 | 2854 | assert_qgroups_uptodate(trans); |
a28ec197 CM |
2855 | return 0; |
2856 | } | |
2857 | ||
5d4f98a2 YZ |
2858 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2859 | struct btrfs_root *root, | |
2860 | u64 bytenr, u64 num_bytes, u64 flags, | |
b1c79e09 | 2861 | int level, int is_data) |
5d4f98a2 YZ |
2862 | { |
2863 | struct btrfs_delayed_extent_op *extent_op; | |
2864 | int ret; | |
2865 | ||
78a6184a | 2866 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
2867 | if (!extent_op) |
2868 | return -ENOMEM; | |
2869 | ||
2870 | extent_op->flags_to_set = flags; | |
2871 | extent_op->update_flags = 1; | |
2872 | extent_op->update_key = 0; | |
2873 | extent_op->is_data = is_data ? 1 : 0; | |
b1c79e09 | 2874 | extent_op->level = level; |
5d4f98a2 | 2875 | |
66d7e7f0 AJ |
2876 | ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr, |
2877 | num_bytes, extent_op); | |
5d4f98a2 | 2878 | if (ret) |
78a6184a | 2879 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2880 | return ret; |
2881 | } | |
2882 | ||
2883 | static noinline int check_delayed_ref(struct btrfs_trans_handle *trans, | |
2884 | struct btrfs_root *root, | |
2885 | struct btrfs_path *path, | |
2886 | u64 objectid, u64 offset, u64 bytenr) | |
2887 | { | |
2888 | struct btrfs_delayed_ref_head *head; | |
2889 | struct btrfs_delayed_ref_node *ref; | |
2890 | struct btrfs_delayed_data_ref *data_ref; | |
2891 | struct btrfs_delayed_ref_root *delayed_refs; | |
2892 | struct rb_node *node; | |
2893 | int ret = 0; | |
2894 | ||
5d4f98a2 YZ |
2895 | delayed_refs = &trans->transaction->delayed_refs; |
2896 | spin_lock(&delayed_refs->lock); | |
2897 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
d7df2c79 JB |
2898 | if (!head) { |
2899 | spin_unlock(&delayed_refs->lock); | |
2900 | return 0; | |
2901 | } | |
5d4f98a2 YZ |
2902 | |
2903 | if (!mutex_trylock(&head->mutex)) { | |
2904 | atomic_inc(&head->node.refs); | |
2905 | spin_unlock(&delayed_refs->lock); | |
2906 | ||
b3b4aa74 | 2907 | btrfs_release_path(path); |
5d4f98a2 | 2908 | |
8cc33e5c DS |
2909 | /* |
2910 | * Mutex was contended, block until it's released and let | |
2911 | * caller try again | |
2912 | */ | |
5d4f98a2 YZ |
2913 | mutex_lock(&head->mutex); |
2914 | mutex_unlock(&head->mutex); | |
2915 | btrfs_put_delayed_ref(&head->node); | |
2916 | return -EAGAIN; | |
2917 | } | |
d7df2c79 | 2918 | spin_unlock(&delayed_refs->lock); |
5d4f98a2 | 2919 | |
d7df2c79 JB |
2920 | spin_lock(&head->lock); |
2921 | node = rb_first(&head->ref_root); | |
2922 | while (node) { | |
2923 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2924 | node = rb_next(node); | |
5d4f98a2 | 2925 | |
d7df2c79 JB |
2926 | /* If it's a shared ref we know a cross reference exists */ |
2927 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { | |
2928 | ret = 1; | |
2929 | break; | |
2930 | } | |
5d4f98a2 | 2931 | |
d7df2c79 | 2932 | data_ref = btrfs_delayed_node_to_data_ref(ref); |
5d4f98a2 | 2933 | |
d7df2c79 JB |
2934 | /* |
2935 | * If our ref doesn't match the one we're currently looking at | |
2936 | * then we have a cross reference. | |
2937 | */ | |
2938 | if (data_ref->root != root->root_key.objectid || | |
2939 | data_ref->objectid != objectid || | |
2940 | data_ref->offset != offset) { | |
2941 | ret = 1; | |
2942 | break; | |
2943 | } | |
5d4f98a2 | 2944 | } |
d7df2c79 | 2945 | spin_unlock(&head->lock); |
5d4f98a2 | 2946 | mutex_unlock(&head->mutex); |
5d4f98a2 YZ |
2947 | return ret; |
2948 | } | |
2949 | ||
2950 | static noinline int check_committed_ref(struct btrfs_trans_handle *trans, | |
2951 | struct btrfs_root *root, | |
2952 | struct btrfs_path *path, | |
2953 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d CM |
2954 | { |
2955 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
f321e491 | 2956 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2957 | struct btrfs_extent_data_ref *ref; |
2958 | struct btrfs_extent_inline_ref *iref; | |
2959 | struct btrfs_extent_item *ei; | |
f321e491 | 2960 | struct btrfs_key key; |
5d4f98a2 | 2961 | u32 item_size; |
be20aa9d | 2962 | int ret; |
925baedd | 2963 | |
be20aa9d | 2964 | key.objectid = bytenr; |
31840ae1 | 2965 | key.offset = (u64)-1; |
f321e491 | 2966 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 2967 | |
be20aa9d CM |
2968 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
2969 | if (ret < 0) | |
2970 | goto out; | |
79787eaa | 2971 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
2972 | |
2973 | ret = -ENOENT; | |
2974 | if (path->slots[0] == 0) | |
31840ae1 | 2975 | goto out; |
be20aa9d | 2976 | |
31840ae1 | 2977 | path->slots[0]--; |
f321e491 | 2978 | leaf = path->nodes[0]; |
5d4f98a2 | 2979 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 2980 | |
5d4f98a2 | 2981 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 2982 | goto out; |
f321e491 | 2983 | |
5d4f98a2 YZ |
2984 | ret = 1; |
2985 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2986 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2987 | if (item_size < sizeof(*ei)) { | |
2988 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
2989 | goto out; | |
2990 | } | |
2991 | #endif | |
2992 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 2993 | |
5d4f98a2 YZ |
2994 | if (item_size != sizeof(*ei) + |
2995 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
2996 | goto out; | |
be20aa9d | 2997 | |
5d4f98a2 YZ |
2998 | if (btrfs_extent_generation(leaf, ei) <= |
2999 | btrfs_root_last_snapshot(&root->root_item)) | |
3000 | goto out; | |
3001 | ||
3002 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
3003 | if (btrfs_extent_inline_ref_type(leaf, iref) != | |
3004 | BTRFS_EXTENT_DATA_REF_KEY) | |
3005 | goto out; | |
3006 | ||
3007 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
3008 | if (btrfs_extent_refs(leaf, ei) != | |
3009 | btrfs_extent_data_ref_count(leaf, ref) || | |
3010 | btrfs_extent_data_ref_root(leaf, ref) != | |
3011 | root->root_key.objectid || | |
3012 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
3013 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
3014 | goto out; | |
3015 | ||
3016 | ret = 0; | |
3017 | out: | |
3018 | return ret; | |
3019 | } | |
3020 | ||
3021 | int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, | |
3022 | struct btrfs_root *root, | |
3023 | u64 objectid, u64 offset, u64 bytenr) | |
3024 | { | |
3025 | struct btrfs_path *path; | |
3026 | int ret; | |
3027 | int ret2; | |
3028 | ||
3029 | path = btrfs_alloc_path(); | |
3030 | if (!path) | |
3031 | return -ENOENT; | |
3032 | ||
3033 | do { | |
3034 | ret = check_committed_ref(trans, root, path, objectid, | |
3035 | offset, bytenr); | |
3036 | if (ret && ret != -ENOENT) | |
f321e491 | 3037 | goto out; |
80ff3856 | 3038 | |
5d4f98a2 YZ |
3039 | ret2 = check_delayed_ref(trans, root, path, objectid, |
3040 | offset, bytenr); | |
3041 | } while (ret2 == -EAGAIN); | |
3042 | ||
3043 | if (ret2 && ret2 != -ENOENT) { | |
3044 | ret = ret2; | |
3045 | goto out; | |
f321e491 | 3046 | } |
5d4f98a2 YZ |
3047 | |
3048 | if (ret != -ENOENT || ret2 != -ENOENT) | |
3049 | ret = 0; | |
be20aa9d | 3050 | out: |
80ff3856 | 3051 | btrfs_free_path(path); |
f0486c68 YZ |
3052 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
3053 | WARN_ON(ret > 0); | |
f321e491 | 3054 | return ret; |
be20aa9d | 3055 | } |
c5739bba | 3056 | |
5d4f98a2 | 3057 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 3058 | struct btrfs_root *root, |
5d4f98a2 | 3059 | struct extent_buffer *buf, |
e339a6b0 | 3060 | int full_backref, int inc) |
31840ae1 ZY |
3061 | { |
3062 | u64 bytenr; | |
5d4f98a2 YZ |
3063 | u64 num_bytes; |
3064 | u64 parent; | |
31840ae1 | 3065 | u64 ref_root; |
31840ae1 | 3066 | u32 nritems; |
31840ae1 ZY |
3067 | struct btrfs_key key; |
3068 | struct btrfs_file_extent_item *fi; | |
3069 | int i; | |
3070 | int level; | |
3071 | int ret = 0; | |
31840ae1 | 3072 | int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, |
66d7e7f0 | 3073 | u64, u64, u64, u64, u64, u64, int); |
31840ae1 | 3074 | |
faa2dbf0 JB |
3075 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
3076 | if (unlikely(test_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state))) | |
3077 | return 0; | |
3078 | #endif | |
31840ae1 | 3079 | ref_root = btrfs_header_owner(buf); |
31840ae1 ZY |
3080 | nritems = btrfs_header_nritems(buf); |
3081 | level = btrfs_header_level(buf); | |
3082 | ||
27cdeb70 | 3083 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0) |
5d4f98a2 | 3084 | return 0; |
31840ae1 | 3085 | |
5d4f98a2 YZ |
3086 | if (inc) |
3087 | process_func = btrfs_inc_extent_ref; | |
3088 | else | |
3089 | process_func = btrfs_free_extent; | |
31840ae1 | 3090 | |
5d4f98a2 YZ |
3091 | if (full_backref) |
3092 | parent = buf->start; | |
3093 | else | |
3094 | parent = 0; | |
3095 | ||
3096 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 3097 | if (level == 0) { |
5d4f98a2 | 3098 | btrfs_item_key_to_cpu(buf, &key, i); |
31840ae1 ZY |
3099 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) |
3100 | continue; | |
5d4f98a2 | 3101 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
3102 | struct btrfs_file_extent_item); |
3103 | if (btrfs_file_extent_type(buf, fi) == | |
3104 | BTRFS_FILE_EXTENT_INLINE) | |
3105 | continue; | |
3106 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
3107 | if (bytenr == 0) | |
3108 | continue; | |
5d4f98a2 YZ |
3109 | |
3110 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
3111 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
3112 | ret = process_func(trans, root, bytenr, num_bytes, | |
3113 | parent, ref_root, key.objectid, | |
e339a6b0 | 3114 | key.offset, 1); |
31840ae1 ZY |
3115 | if (ret) |
3116 | goto fail; | |
3117 | } else { | |
5d4f98a2 YZ |
3118 | bytenr = btrfs_node_blockptr(buf, i); |
3119 | num_bytes = btrfs_level_size(root, level - 1); | |
3120 | ret = process_func(trans, root, bytenr, num_bytes, | |
66d7e7f0 | 3121 | parent, ref_root, level - 1, 0, |
e339a6b0 | 3122 | 1); |
31840ae1 ZY |
3123 | if (ret) |
3124 | goto fail; | |
3125 | } | |
3126 | } | |
3127 | return 0; | |
3128 | fail: | |
5d4f98a2 YZ |
3129 | return ret; |
3130 | } | |
3131 | ||
3132 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3133 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3134 | { |
e339a6b0 | 3135 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); |
5d4f98a2 YZ |
3136 | } |
3137 | ||
3138 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3139 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3140 | { |
e339a6b0 | 3141 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); |
31840ae1 ZY |
3142 | } |
3143 | ||
9078a3e1 CM |
3144 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
3145 | struct btrfs_root *root, | |
3146 | struct btrfs_path *path, | |
3147 | struct btrfs_block_group_cache *cache) | |
3148 | { | |
3149 | int ret; | |
9078a3e1 | 3150 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
5f39d397 CM |
3151 | unsigned long bi; |
3152 | struct extent_buffer *leaf; | |
9078a3e1 | 3153 | |
9078a3e1 | 3154 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
54aa1f4d CM |
3155 | if (ret < 0) |
3156 | goto fail; | |
79787eaa | 3157 | BUG_ON(ret); /* Corruption */ |
5f39d397 CM |
3158 | |
3159 | leaf = path->nodes[0]; | |
3160 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3161 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
3162 | btrfs_mark_buffer_dirty(leaf); | |
b3b4aa74 | 3163 | btrfs_release_path(path); |
54aa1f4d | 3164 | fail: |
79787eaa JM |
3165 | if (ret) { |
3166 | btrfs_abort_transaction(trans, root, ret); | |
9078a3e1 | 3167 | return ret; |
79787eaa | 3168 | } |
9078a3e1 CM |
3169 | return 0; |
3170 | ||
3171 | } | |
3172 | ||
4a8c9a62 YZ |
3173 | static struct btrfs_block_group_cache * |
3174 | next_block_group(struct btrfs_root *root, | |
3175 | struct btrfs_block_group_cache *cache) | |
3176 | { | |
3177 | struct rb_node *node; | |
3178 | spin_lock(&root->fs_info->block_group_cache_lock); | |
3179 | node = rb_next(&cache->cache_node); | |
3180 | btrfs_put_block_group(cache); | |
3181 | if (node) { | |
3182 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
3183 | cache_node); | |
11dfe35a | 3184 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
3185 | } else |
3186 | cache = NULL; | |
3187 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
3188 | return cache; | |
3189 | } | |
3190 | ||
0af3d00b JB |
3191 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
3192 | struct btrfs_trans_handle *trans, | |
3193 | struct btrfs_path *path) | |
3194 | { | |
3195 | struct btrfs_root *root = block_group->fs_info->tree_root; | |
3196 | struct inode *inode = NULL; | |
3197 | u64 alloc_hint = 0; | |
2b20982e | 3198 | int dcs = BTRFS_DC_ERROR; |
0af3d00b JB |
3199 | int num_pages = 0; |
3200 | int retries = 0; | |
3201 | int ret = 0; | |
3202 | ||
3203 | /* | |
3204 | * If this block group is smaller than 100 megs don't bother caching the | |
3205 | * block group. | |
3206 | */ | |
3207 | if (block_group->key.offset < (100 * 1024 * 1024)) { | |
3208 | spin_lock(&block_group->lock); | |
3209 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
3210 | spin_unlock(&block_group->lock); | |
3211 | return 0; | |
3212 | } | |
3213 | ||
3214 | again: | |
3215 | inode = lookup_free_space_inode(root, block_group, path); | |
3216 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { | |
3217 | ret = PTR_ERR(inode); | |
b3b4aa74 | 3218 | btrfs_release_path(path); |
0af3d00b JB |
3219 | goto out; |
3220 | } | |
3221 | ||
3222 | if (IS_ERR(inode)) { | |
3223 | BUG_ON(retries); | |
3224 | retries++; | |
3225 | ||
3226 | if (block_group->ro) | |
3227 | goto out_free; | |
3228 | ||
3229 | ret = create_free_space_inode(root, trans, block_group, path); | |
3230 | if (ret) | |
3231 | goto out_free; | |
3232 | goto again; | |
3233 | } | |
3234 | ||
5b0e95bf JB |
3235 | /* We've already setup this transaction, go ahead and exit */ |
3236 | if (block_group->cache_generation == trans->transid && | |
3237 | i_size_read(inode)) { | |
3238 | dcs = BTRFS_DC_SETUP; | |
3239 | goto out_put; | |
3240 | } | |
3241 | ||
0af3d00b JB |
3242 | /* |
3243 | * We want to set the generation to 0, that way if anything goes wrong | |
3244 | * from here on out we know not to trust this cache when we load up next | |
3245 | * time. | |
3246 | */ | |
3247 | BTRFS_I(inode)->generation = 0; | |
3248 | ret = btrfs_update_inode(trans, root, inode); | |
3249 | WARN_ON(ret); | |
3250 | ||
3251 | if (i_size_read(inode) > 0) { | |
7b61cd92 MX |
3252 | ret = btrfs_check_trunc_cache_free_space(root, |
3253 | &root->fs_info->global_block_rsv); | |
3254 | if (ret) | |
3255 | goto out_put; | |
3256 | ||
74514323 | 3257 | ret = btrfs_truncate_free_space_cache(root, trans, inode); |
0af3d00b JB |
3258 | if (ret) |
3259 | goto out_put; | |
3260 | } | |
3261 | ||
3262 | spin_lock(&block_group->lock); | |
cf7c1ef6 | 3263 | if (block_group->cached != BTRFS_CACHE_FINISHED || |
e570fd27 MX |
3264 | !btrfs_test_opt(root, SPACE_CACHE) || |
3265 | block_group->delalloc_bytes) { | |
cf7c1ef6 LB |
3266 | /* |
3267 | * don't bother trying to write stuff out _if_ | |
3268 | * a) we're not cached, | |
3269 | * b) we're with nospace_cache mount option. | |
3270 | */ | |
2b20982e | 3271 | dcs = BTRFS_DC_WRITTEN; |
0af3d00b JB |
3272 | spin_unlock(&block_group->lock); |
3273 | goto out_put; | |
3274 | } | |
3275 | spin_unlock(&block_group->lock); | |
3276 | ||
6fc823b1 JB |
3277 | /* |
3278 | * Try to preallocate enough space based on how big the block group is. | |
3279 | * Keep in mind this has to include any pinned space which could end up | |
3280 | * taking up quite a bit since it's not folded into the other space | |
3281 | * cache. | |
3282 | */ | |
3283 | num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024); | |
0af3d00b JB |
3284 | if (!num_pages) |
3285 | num_pages = 1; | |
3286 | ||
0af3d00b JB |
3287 | num_pages *= 16; |
3288 | num_pages *= PAGE_CACHE_SIZE; | |
3289 | ||
3290 | ret = btrfs_check_data_free_space(inode, num_pages); | |
3291 | if (ret) | |
3292 | goto out_put; | |
3293 | ||
3294 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
3295 | num_pages, num_pages, | |
3296 | &alloc_hint); | |
2b20982e JB |
3297 | if (!ret) |
3298 | dcs = BTRFS_DC_SETUP; | |
0af3d00b | 3299 | btrfs_free_reserved_data_space(inode, num_pages); |
c09544e0 | 3300 | |
0af3d00b JB |
3301 | out_put: |
3302 | iput(inode); | |
3303 | out_free: | |
b3b4aa74 | 3304 | btrfs_release_path(path); |
0af3d00b JB |
3305 | out: |
3306 | spin_lock(&block_group->lock); | |
e65cbb94 | 3307 | if (!ret && dcs == BTRFS_DC_SETUP) |
5b0e95bf | 3308 | block_group->cache_generation = trans->transid; |
2b20982e | 3309 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
3310 | spin_unlock(&block_group->lock); |
3311 | ||
3312 | return ret; | |
3313 | } | |
3314 | ||
96b5179d CM |
3315 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, |
3316 | struct btrfs_root *root) | |
9078a3e1 | 3317 | { |
4a8c9a62 | 3318 | struct btrfs_block_group_cache *cache; |
9078a3e1 | 3319 | int err = 0; |
9078a3e1 | 3320 | struct btrfs_path *path; |
96b5179d | 3321 | u64 last = 0; |
9078a3e1 CM |
3322 | |
3323 | path = btrfs_alloc_path(); | |
3324 | if (!path) | |
3325 | return -ENOMEM; | |
3326 | ||
0af3d00b JB |
3327 | again: |
3328 | while (1) { | |
3329 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
3330 | while (cache) { | |
3331 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
3332 | break; | |
3333 | cache = next_block_group(root, cache); | |
3334 | } | |
3335 | if (!cache) { | |
3336 | if (last == 0) | |
3337 | break; | |
3338 | last = 0; | |
3339 | continue; | |
3340 | } | |
3341 | err = cache_save_setup(cache, trans, path); | |
3342 | last = cache->key.objectid + cache->key.offset; | |
3343 | btrfs_put_block_group(cache); | |
3344 | } | |
3345 | ||
d397712b | 3346 | while (1) { |
4a8c9a62 YZ |
3347 | if (last == 0) { |
3348 | err = btrfs_run_delayed_refs(trans, root, | |
3349 | (unsigned long)-1); | |
79787eaa JM |
3350 | if (err) /* File system offline */ |
3351 | goto out; | |
0f9dd46c | 3352 | } |
54aa1f4d | 3353 | |
4a8c9a62 YZ |
3354 | cache = btrfs_lookup_first_block_group(root->fs_info, last); |
3355 | while (cache) { | |
0af3d00b JB |
3356 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) { |
3357 | btrfs_put_block_group(cache); | |
3358 | goto again; | |
3359 | } | |
3360 | ||
4a8c9a62 YZ |
3361 | if (cache->dirty) |
3362 | break; | |
3363 | cache = next_block_group(root, cache); | |
3364 | } | |
3365 | if (!cache) { | |
3366 | if (last == 0) | |
3367 | break; | |
3368 | last = 0; | |
3369 | continue; | |
3370 | } | |
0f9dd46c | 3371 | |
0cb59c99 JB |
3372 | if (cache->disk_cache_state == BTRFS_DC_SETUP) |
3373 | cache->disk_cache_state = BTRFS_DC_NEED_WRITE; | |
e8569813 | 3374 | cache->dirty = 0; |
4a8c9a62 | 3375 | last = cache->key.objectid + cache->key.offset; |
0f9dd46c | 3376 | |
4a8c9a62 | 3377 | err = write_one_cache_group(trans, root, path, cache); |
e84cc142 | 3378 | btrfs_put_block_group(cache); |
79787eaa JM |
3379 | if (err) /* File system offline */ |
3380 | goto out; | |
9078a3e1 | 3381 | } |
4a8c9a62 | 3382 | |
0cb59c99 JB |
3383 | while (1) { |
3384 | /* | |
3385 | * I don't think this is needed since we're just marking our | |
3386 | * preallocated extent as written, but just in case it can't | |
3387 | * hurt. | |
3388 | */ | |
3389 | if (last == 0) { | |
3390 | err = btrfs_run_delayed_refs(trans, root, | |
3391 | (unsigned long)-1); | |
79787eaa JM |
3392 | if (err) /* File system offline */ |
3393 | goto out; | |
0cb59c99 JB |
3394 | } |
3395 | ||
3396 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
3397 | while (cache) { | |
3398 | /* | |
3399 | * Really this shouldn't happen, but it could if we | |
3400 | * couldn't write the entire preallocated extent and | |
3401 | * splitting the extent resulted in a new block. | |
3402 | */ | |
3403 | if (cache->dirty) { | |
3404 | btrfs_put_block_group(cache); | |
3405 | goto again; | |
3406 | } | |
3407 | if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE) | |
3408 | break; | |
3409 | cache = next_block_group(root, cache); | |
3410 | } | |
3411 | if (!cache) { | |
3412 | if (last == 0) | |
3413 | break; | |
3414 | last = 0; | |
3415 | continue; | |
3416 | } | |
3417 | ||
79787eaa | 3418 | err = btrfs_write_out_cache(root, trans, cache, path); |
0cb59c99 JB |
3419 | |
3420 | /* | |
3421 | * If we didn't have an error then the cache state is still | |
3422 | * NEED_WRITE, so we can set it to WRITTEN. | |
3423 | */ | |
79787eaa | 3424 | if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE) |
0cb59c99 JB |
3425 | cache->disk_cache_state = BTRFS_DC_WRITTEN; |
3426 | last = cache->key.objectid + cache->key.offset; | |
3427 | btrfs_put_block_group(cache); | |
3428 | } | |
79787eaa | 3429 | out: |
0cb59c99 | 3430 | |
9078a3e1 | 3431 | btrfs_free_path(path); |
79787eaa | 3432 | return err; |
9078a3e1 CM |
3433 | } |
3434 | ||
d2fb3437 YZ |
3435 | int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr) |
3436 | { | |
3437 | struct btrfs_block_group_cache *block_group; | |
3438 | int readonly = 0; | |
3439 | ||
3440 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); | |
3441 | if (!block_group || block_group->ro) | |
3442 | readonly = 1; | |
3443 | if (block_group) | |
fa9c0d79 | 3444 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3445 | return readonly; |
3446 | } | |
3447 | ||
6ab0a202 JM |
3448 | static const char *alloc_name(u64 flags) |
3449 | { | |
3450 | switch (flags) { | |
3451 | case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA: | |
3452 | return "mixed"; | |
3453 | case BTRFS_BLOCK_GROUP_METADATA: | |
3454 | return "metadata"; | |
3455 | case BTRFS_BLOCK_GROUP_DATA: | |
3456 | return "data"; | |
3457 | case BTRFS_BLOCK_GROUP_SYSTEM: | |
3458 | return "system"; | |
3459 | default: | |
3460 | WARN_ON(1); | |
3461 | return "invalid-combination"; | |
3462 | }; | |
3463 | } | |
3464 | ||
593060d7 CM |
3465 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
3466 | u64 total_bytes, u64 bytes_used, | |
3467 | struct btrfs_space_info **space_info) | |
3468 | { | |
3469 | struct btrfs_space_info *found; | |
b742bb82 YZ |
3470 | int i; |
3471 | int factor; | |
b150a4f1 | 3472 | int ret; |
b742bb82 YZ |
3473 | |
3474 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
3475 | BTRFS_BLOCK_GROUP_RAID10)) | |
3476 | factor = 2; | |
3477 | else | |
3478 | factor = 1; | |
593060d7 CM |
3479 | |
3480 | found = __find_space_info(info, flags); | |
3481 | if (found) { | |
25179201 | 3482 | spin_lock(&found->lock); |
593060d7 | 3483 | found->total_bytes += total_bytes; |
89a55897 | 3484 | found->disk_total += total_bytes * factor; |
593060d7 | 3485 | found->bytes_used += bytes_used; |
b742bb82 | 3486 | found->disk_used += bytes_used * factor; |
8f18cf13 | 3487 | found->full = 0; |
25179201 | 3488 | spin_unlock(&found->lock); |
593060d7 CM |
3489 | *space_info = found; |
3490 | return 0; | |
3491 | } | |
c146afad | 3492 | found = kzalloc(sizeof(*found), GFP_NOFS); |
593060d7 CM |
3493 | if (!found) |
3494 | return -ENOMEM; | |
3495 | ||
b150a4f1 JB |
3496 | ret = percpu_counter_init(&found->total_bytes_pinned, 0); |
3497 | if (ret) { | |
3498 | kfree(found); | |
3499 | return ret; | |
3500 | } | |
3501 | ||
c1895442 | 3502 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) |
b742bb82 | 3503 | INIT_LIST_HEAD(&found->block_groups[i]); |
80eb234a | 3504 | init_rwsem(&found->groups_sem); |
0f9dd46c | 3505 | spin_lock_init(&found->lock); |
52ba6929 | 3506 | found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; |
593060d7 | 3507 | found->total_bytes = total_bytes; |
89a55897 | 3508 | found->disk_total = total_bytes * factor; |
593060d7 | 3509 | found->bytes_used = bytes_used; |
b742bb82 | 3510 | found->disk_used = bytes_used * factor; |
593060d7 | 3511 | found->bytes_pinned = 0; |
e8569813 | 3512 | found->bytes_reserved = 0; |
c146afad | 3513 | found->bytes_readonly = 0; |
f0486c68 | 3514 | found->bytes_may_use = 0; |
593060d7 | 3515 | found->full = 0; |
0e4f8f88 | 3516 | found->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 3517 | found->chunk_alloc = 0; |
fdb5effd JB |
3518 | found->flush = 0; |
3519 | init_waitqueue_head(&found->wait); | |
6ab0a202 JM |
3520 | |
3521 | ret = kobject_init_and_add(&found->kobj, &space_info_ktype, | |
3522 | info->space_info_kobj, "%s", | |
3523 | alloc_name(found->flags)); | |
3524 | if (ret) { | |
3525 | kfree(found); | |
3526 | return ret; | |
3527 | } | |
3528 | ||
593060d7 | 3529 | *space_info = found; |
4184ea7f | 3530 | list_add_rcu(&found->list, &info->space_info); |
b4d7c3c9 LZ |
3531 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
3532 | info->data_sinfo = found; | |
6ab0a202 JM |
3533 | |
3534 | return ret; | |
593060d7 CM |
3535 | } |
3536 | ||
8790d502 CM |
3537 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
3538 | { | |
899c81ea ID |
3539 | u64 extra_flags = chunk_to_extended(flags) & |
3540 | BTRFS_EXTENDED_PROFILE_MASK; | |
a46d11a8 | 3541 | |
de98ced9 | 3542 | write_seqlock(&fs_info->profiles_lock); |
a46d11a8 ID |
3543 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
3544 | fs_info->avail_data_alloc_bits |= extra_flags; | |
3545 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3546 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
3547 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3548 | fs_info->avail_system_alloc_bits |= extra_flags; | |
de98ced9 | 3549 | write_sequnlock(&fs_info->profiles_lock); |
8790d502 | 3550 | } |
593060d7 | 3551 | |
fc67c450 ID |
3552 | /* |
3553 | * returns target flags in extended format or 0 if restripe for this | |
3554 | * chunk_type is not in progress | |
c6664b42 ID |
3555 | * |
3556 | * should be called with either volume_mutex or balance_lock held | |
fc67c450 ID |
3557 | */ |
3558 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) | |
3559 | { | |
3560 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3561 | u64 target = 0; | |
3562 | ||
fc67c450 ID |
3563 | if (!bctl) |
3564 | return 0; | |
3565 | ||
3566 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
3567 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3568 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
3569 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
3570 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3571 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
3572 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
3573 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3574 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
3575 | } | |
3576 | ||
3577 | return target; | |
3578 | } | |
3579 | ||
a46d11a8 ID |
3580 | /* |
3581 | * @flags: available profiles in extended format (see ctree.h) | |
3582 | * | |
e4d8ec0f ID |
3583 | * Returns reduced profile in chunk format. If profile changing is in |
3584 | * progress (either running or paused) picks the target profile (if it's | |
3585 | * already available), otherwise falls back to plain reducing. | |
a46d11a8 | 3586 | */ |
48a3b636 | 3587 | static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) |
ec44a35c | 3588 | { |
cd02dca5 CM |
3589 | /* |
3590 | * we add in the count of missing devices because we want | |
3591 | * to make sure that any RAID levels on a degraded FS | |
3592 | * continue to be honored. | |
3593 | */ | |
3594 | u64 num_devices = root->fs_info->fs_devices->rw_devices + | |
3595 | root->fs_info->fs_devices->missing_devices; | |
fc67c450 | 3596 | u64 target; |
53b381b3 | 3597 | u64 tmp; |
a061fc8d | 3598 | |
fc67c450 ID |
3599 | /* |
3600 | * see if restripe for this chunk_type is in progress, if so | |
3601 | * try to reduce to the target profile | |
3602 | */ | |
e4d8ec0f | 3603 | spin_lock(&root->fs_info->balance_lock); |
fc67c450 ID |
3604 | target = get_restripe_target(root->fs_info, flags); |
3605 | if (target) { | |
3606 | /* pick target profile only if it's already available */ | |
3607 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { | |
e4d8ec0f | 3608 | spin_unlock(&root->fs_info->balance_lock); |
fc67c450 | 3609 | return extended_to_chunk(target); |
e4d8ec0f ID |
3610 | } |
3611 | } | |
3612 | spin_unlock(&root->fs_info->balance_lock); | |
3613 | ||
53b381b3 | 3614 | /* First, mask out the RAID levels which aren't possible */ |
a061fc8d | 3615 | if (num_devices == 1) |
53b381b3 DW |
3616 | flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 | |
3617 | BTRFS_BLOCK_GROUP_RAID5); | |
3618 | if (num_devices < 3) | |
3619 | flags &= ~BTRFS_BLOCK_GROUP_RAID6; | |
a061fc8d CM |
3620 | if (num_devices < 4) |
3621 | flags &= ~BTRFS_BLOCK_GROUP_RAID10; | |
3622 | ||
53b381b3 DW |
3623 | tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 | |
3624 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 | | |
3625 | BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10); | |
3626 | flags &= ~tmp; | |
ec44a35c | 3627 | |
53b381b3 DW |
3628 | if (tmp & BTRFS_BLOCK_GROUP_RAID6) |
3629 | tmp = BTRFS_BLOCK_GROUP_RAID6; | |
3630 | else if (tmp & BTRFS_BLOCK_GROUP_RAID5) | |
3631 | tmp = BTRFS_BLOCK_GROUP_RAID5; | |
3632 | else if (tmp & BTRFS_BLOCK_GROUP_RAID10) | |
3633 | tmp = BTRFS_BLOCK_GROUP_RAID10; | |
3634 | else if (tmp & BTRFS_BLOCK_GROUP_RAID1) | |
3635 | tmp = BTRFS_BLOCK_GROUP_RAID1; | |
3636 | else if (tmp & BTRFS_BLOCK_GROUP_RAID0) | |
3637 | tmp = BTRFS_BLOCK_GROUP_RAID0; | |
a46d11a8 | 3638 | |
53b381b3 | 3639 | return extended_to_chunk(flags | tmp); |
ec44a35c CM |
3640 | } |
3641 | ||
f8213bdc | 3642 | static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags) |
6a63209f | 3643 | { |
de98ced9 | 3644 | unsigned seq; |
f8213bdc | 3645 | u64 flags; |
de98ced9 MX |
3646 | |
3647 | do { | |
f8213bdc | 3648 | flags = orig_flags; |
de98ced9 MX |
3649 | seq = read_seqbegin(&root->fs_info->profiles_lock); |
3650 | ||
3651 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
3652 | flags |= root->fs_info->avail_data_alloc_bits; | |
3653 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3654 | flags |= root->fs_info->avail_system_alloc_bits; | |
3655 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3656 | flags |= root->fs_info->avail_metadata_alloc_bits; | |
3657 | } while (read_seqretry(&root->fs_info->profiles_lock, seq)); | |
6fef8df1 | 3658 | |
b742bb82 | 3659 | return btrfs_reduce_alloc_profile(root, flags); |
6a63209f JB |
3660 | } |
3661 | ||
6d07bcec | 3662 | u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data) |
9ed74f2d | 3663 | { |
b742bb82 | 3664 | u64 flags; |
53b381b3 | 3665 | u64 ret; |
9ed74f2d | 3666 | |
b742bb82 YZ |
3667 | if (data) |
3668 | flags = BTRFS_BLOCK_GROUP_DATA; | |
3669 | else if (root == root->fs_info->chunk_root) | |
3670 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
9ed74f2d | 3671 | else |
b742bb82 | 3672 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 3673 | |
53b381b3 DW |
3674 | ret = get_alloc_profile(root, flags); |
3675 | return ret; | |
6a63209f | 3676 | } |
9ed74f2d | 3677 | |
6a63209f | 3678 | /* |
6a63209f JB |
3679 | * This will check the space that the inode allocates from to make sure we have |
3680 | * enough space for bytes. | |
6a63209f | 3681 | */ |
0ca1f7ce | 3682 | int btrfs_check_data_free_space(struct inode *inode, u64 bytes) |
6a63209f | 3683 | { |
6a63209f | 3684 | struct btrfs_space_info *data_sinfo; |
0ca1f7ce | 3685 | struct btrfs_root *root = BTRFS_I(inode)->root; |
b4d7c3c9 | 3686 | struct btrfs_fs_info *fs_info = root->fs_info; |
ab6e2410 | 3687 | u64 used; |
0af3d00b | 3688 | int ret = 0, committed = 0, alloc_chunk = 1; |
6a63209f | 3689 | |
6a63209f | 3690 | /* make sure bytes are sectorsize aligned */ |
fda2832f | 3691 | bytes = ALIGN(bytes, root->sectorsize); |
6a63209f | 3692 | |
9dced186 | 3693 | if (btrfs_is_free_space_inode(inode)) { |
0af3d00b | 3694 | committed = 1; |
9dced186 | 3695 | ASSERT(current->journal_info); |
0af3d00b JB |
3696 | } |
3697 | ||
b4d7c3c9 | 3698 | data_sinfo = fs_info->data_sinfo; |
33b4d47f CM |
3699 | if (!data_sinfo) |
3700 | goto alloc; | |
9ed74f2d | 3701 | |
6a63209f JB |
3702 | again: |
3703 | /* make sure we have enough space to handle the data first */ | |
3704 | spin_lock(&data_sinfo->lock); | |
8929ecfa YZ |
3705 | used = data_sinfo->bytes_used + data_sinfo->bytes_reserved + |
3706 | data_sinfo->bytes_pinned + data_sinfo->bytes_readonly + | |
3707 | data_sinfo->bytes_may_use; | |
ab6e2410 JB |
3708 | |
3709 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 3710 | struct btrfs_trans_handle *trans; |
9ed74f2d | 3711 | |
6a63209f JB |
3712 | /* |
3713 | * if we don't have enough free bytes in this space then we need | |
3714 | * to alloc a new chunk. | |
3715 | */ | |
0af3d00b | 3716 | if (!data_sinfo->full && alloc_chunk) { |
6a63209f | 3717 | u64 alloc_target; |
9ed74f2d | 3718 | |
0e4f8f88 | 3719 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 3720 | spin_unlock(&data_sinfo->lock); |
33b4d47f | 3721 | alloc: |
6a63209f | 3722 | alloc_target = btrfs_get_alloc_profile(root, 1); |
9dced186 MX |
3723 | /* |
3724 | * It is ugly that we don't call nolock join | |
3725 | * transaction for the free space inode case here. | |
3726 | * But it is safe because we only do the data space | |
3727 | * reservation for the free space cache in the | |
3728 | * transaction context, the common join transaction | |
3729 | * just increase the counter of the current transaction | |
3730 | * handler, doesn't try to acquire the trans_lock of | |
3731 | * the fs. | |
3732 | */ | |
7a7eaa40 | 3733 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
3734 | if (IS_ERR(trans)) |
3735 | return PTR_ERR(trans); | |
9ed74f2d | 3736 | |
6a63209f | 3737 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
0e4f8f88 CM |
3738 | alloc_target, |
3739 | CHUNK_ALLOC_NO_FORCE); | |
6a63209f | 3740 | btrfs_end_transaction(trans, root); |
d52a5b5f MX |
3741 | if (ret < 0) { |
3742 | if (ret != -ENOSPC) | |
3743 | return ret; | |
3744 | else | |
3745 | goto commit_trans; | |
3746 | } | |
9ed74f2d | 3747 | |
b4d7c3c9 LZ |
3748 | if (!data_sinfo) |
3749 | data_sinfo = fs_info->data_sinfo; | |
3750 | ||
6a63209f JB |
3751 | goto again; |
3752 | } | |
f2bb8f5c JB |
3753 | |
3754 | /* | |
b150a4f1 JB |
3755 | * If we don't have enough pinned space to deal with this |
3756 | * allocation don't bother committing the transaction. | |
f2bb8f5c | 3757 | */ |
b150a4f1 JB |
3758 | if (percpu_counter_compare(&data_sinfo->total_bytes_pinned, |
3759 | bytes) < 0) | |
f2bb8f5c | 3760 | committed = 1; |
6a63209f | 3761 | spin_unlock(&data_sinfo->lock); |
6a63209f | 3762 | |
4e06bdd6 | 3763 | /* commit the current transaction and try again */ |
d52a5b5f | 3764 | commit_trans: |
a4abeea4 JB |
3765 | if (!committed && |
3766 | !atomic_read(&root->fs_info->open_ioctl_trans)) { | |
4e06bdd6 | 3767 | committed = 1; |
b150a4f1 | 3768 | |
7a7eaa40 | 3769 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
3770 | if (IS_ERR(trans)) |
3771 | return PTR_ERR(trans); | |
4e06bdd6 JB |
3772 | ret = btrfs_commit_transaction(trans, root); |
3773 | if (ret) | |
3774 | return ret; | |
3775 | goto again; | |
3776 | } | |
9ed74f2d | 3777 | |
cab45e22 JM |
3778 | trace_btrfs_space_reservation(root->fs_info, |
3779 | "space_info:enospc", | |
3780 | data_sinfo->flags, bytes, 1); | |
6a63209f JB |
3781 | return -ENOSPC; |
3782 | } | |
3783 | data_sinfo->bytes_may_use += bytes; | |
8c2a3ca2 | 3784 | trace_btrfs_space_reservation(root->fs_info, "space_info", |
2bcc0328 | 3785 | data_sinfo->flags, bytes, 1); |
6a63209f | 3786 | spin_unlock(&data_sinfo->lock); |
6a63209f | 3787 | |
9ed74f2d | 3788 | return 0; |
9ed74f2d | 3789 | } |
6a63209f | 3790 | |
6a63209f | 3791 | /* |
fb25e914 | 3792 | * Called if we need to clear a data reservation for this inode. |
6a63209f | 3793 | */ |
0ca1f7ce | 3794 | void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes) |
e3ccfa98 | 3795 | { |
0ca1f7ce | 3796 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6a63209f | 3797 | struct btrfs_space_info *data_sinfo; |
e3ccfa98 | 3798 | |
6a63209f | 3799 | /* make sure bytes are sectorsize aligned */ |
fda2832f | 3800 | bytes = ALIGN(bytes, root->sectorsize); |
e3ccfa98 | 3801 | |
b4d7c3c9 | 3802 | data_sinfo = root->fs_info->data_sinfo; |
6a63209f | 3803 | spin_lock(&data_sinfo->lock); |
7ee9e440 | 3804 | WARN_ON(data_sinfo->bytes_may_use < bytes); |
6a63209f | 3805 | data_sinfo->bytes_may_use -= bytes; |
8c2a3ca2 | 3806 | trace_btrfs_space_reservation(root->fs_info, "space_info", |
2bcc0328 | 3807 | data_sinfo->flags, bytes, 0); |
6a63209f | 3808 | spin_unlock(&data_sinfo->lock); |
e3ccfa98 JB |
3809 | } |
3810 | ||
97e728d4 | 3811 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 3812 | { |
97e728d4 JB |
3813 | struct list_head *head = &info->space_info; |
3814 | struct btrfs_space_info *found; | |
e3ccfa98 | 3815 | |
97e728d4 JB |
3816 | rcu_read_lock(); |
3817 | list_for_each_entry_rcu(found, head, list) { | |
3818 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 3819 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 3820 | } |
97e728d4 | 3821 | rcu_read_unlock(); |
e3ccfa98 JB |
3822 | } |
3823 | ||
3c76cd84 MX |
3824 | static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global) |
3825 | { | |
3826 | return (global->size << 1); | |
3827 | } | |
3828 | ||
e5bc2458 | 3829 | static int should_alloc_chunk(struct btrfs_root *root, |
698d0082 | 3830 | struct btrfs_space_info *sinfo, int force) |
32c00aff | 3831 | { |
fb25e914 | 3832 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
424499db | 3833 | u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly; |
0e4f8f88 | 3834 | u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved; |
e5bc2458 | 3835 | u64 thresh; |
e3ccfa98 | 3836 | |
0e4f8f88 CM |
3837 | if (force == CHUNK_ALLOC_FORCE) |
3838 | return 1; | |
3839 | ||
fb25e914 JB |
3840 | /* |
3841 | * We need to take into account the global rsv because for all intents | |
3842 | * and purposes it's used space. Don't worry about locking the | |
3843 | * global_rsv, it doesn't change except when the transaction commits. | |
3844 | */ | |
54338b5c | 3845 | if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA) |
3c76cd84 | 3846 | num_allocated += calc_global_rsv_need_space(global_rsv); |
fb25e914 | 3847 | |
0e4f8f88 CM |
3848 | /* |
3849 | * in limited mode, we want to have some free space up to | |
3850 | * about 1% of the FS size. | |
3851 | */ | |
3852 | if (force == CHUNK_ALLOC_LIMITED) { | |
6c41761f | 3853 | thresh = btrfs_super_total_bytes(root->fs_info->super_copy); |
0e4f8f88 CM |
3854 | thresh = max_t(u64, 64 * 1024 * 1024, |
3855 | div_factor_fine(thresh, 1)); | |
3856 | ||
3857 | if (num_bytes - num_allocated < thresh) | |
3858 | return 1; | |
3859 | } | |
0e4f8f88 | 3860 | |
698d0082 | 3861 | if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8)) |
14ed0ca6 | 3862 | return 0; |
424499db | 3863 | return 1; |
32c00aff JB |
3864 | } |
3865 | ||
15d1ff81 LB |
3866 | static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type) |
3867 | { | |
3868 | u64 num_dev; | |
3869 | ||
53b381b3 DW |
3870 | if (type & (BTRFS_BLOCK_GROUP_RAID10 | |
3871 | BTRFS_BLOCK_GROUP_RAID0 | | |
3872 | BTRFS_BLOCK_GROUP_RAID5 | | |
3873 | BTRFS_BLOCK_GROUP_RAID6)) | |
15d1ff81 LB |
3874 | num_dev = root->fs_info->fs_devices->rw_devices; |
3875 | else if (type & BTRFS_BLOCK_GROUP_RAID1) | |
3876 | num_dev = 2; | |
3877 | else | |
3878 | num_dev = 1; /* DUP or single */ | |
3879 | ||
3880 | /* metadata for updaing devices and chunk tree */ | |
3881 | return btrfs_calc_trans_metadata_size(root, num_dev + 1); | |
3882 | } | |
3883 | ||
3884 | static void check_system_chunk(struct btrfs_trans_handle *trans, | |
3885 | struct btrfs_root *root, u64 type) | |
3886 | { | |
3887 | struct btrfs_space_info *info; | |
3888 | u64 left; | |
3889 | u64 thresh; | |
3890 | ||
3891 | info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
3892 | spin_lock(&info->lock); | |
3893 | left = info->total_bytes - info->bytes_used - info->bytes_pinned - | |
3894 | info->bytes_reserved - info->bytes_readonly; | |
3895 | spin_unlock(&info->lock); | |
3896 | ||
3897 | thresh = get_system_chunk_thresh(root, type); | |
3898 | if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
c2cf52eb SK |
3899 | btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu", |
3900 | left, thresh, type); | |
15d1ff81 LB |
3901 | dump_space_info(info, 0, 0); |
3902 | } | |
3903 | ||
3904 | if (left < thresh) { | |
3905 | u64 flags; | |
3906 | ||
3907 | flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0); | |
3908 | btrfs_alloc_chunk(trans, root, flags); | |
3909 | } | |
3910 | } | |
3911 | ||
6324fbf3 | 3912 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
698d0082 | 3913 | struct btrfs_root *extent_root, u64 flags, int force) |
9ed74f2d | 3914 | { |
6324fbf3 | 3915 | struct btrfs_space_info *space_info; |
97e728d4 | 3916 | struct btrfs_fs_info *fs_info = extent_root->fs_info; |
6d74119f | 3917 | int wait_for_alloc = 0; |
9ed74f2d | 3918 | int ret = 0; |
9ed74f2d | 3919 | |
c6b305a8 JB |
3920 | /* Don't re-enter if we're already allocating a chunk */ |
3921 | if (trans->allocating_chunk) | |
3922 | return -ENOSPC; | |
3923 | ||
6324fbf3 | 3924 | space_info = __find_space_info(extent_root->fs_info, flags); |
593060d7 CM |
3925 | if (!space_info) { |
3926 | ret = update_space_info(extent_root->fs_info, flags, | |
3927 | 0, 0, &space_info); | |
79787eaa | 3928 | BUG_ON(ret); /* -ENOMEM */ |
9ed74f2d | 3929 | } |
79787eaa | 3930 | BUG_ON(!space_info); /* Logic error */ |
9ed74f2d | 3931 | |
6d74119f | 3932 | again: |
25179201 | 3933 | spin_lock(&space_info->lock); |
9e622d6b | 3934 | if (force < space_info->force_alloc) |
0e4f8f88 | 3935 | force = space_info->force_alloc; |
25179201 | 3936 | if (space_info->full) { |
09fb99a6 FDBM |
3937 | if (should_alloc_chunk(extent_root, space_info, force)) |
3938 | ret = -ENOSPC; | |
3939 | else | |
3940 | ret = 0; | |
25179201 | 3941 | spin_unlock(&space_info->lock); |
09fb99a6 | 3942 | return ret; |
9ed74f2d JB |
3943 | } |
3944 | ||
698d0082 | 3945 | if (!should_alloc_chunk(extent_root, space_info, force)) { |
25179201 | 3946 | spin_unlock(&space_info->lock); |
6d74119f JB |
3947 | return 0; |
3948 | } else if (space_info->chunk_alloc) { | |
3949 | wait_for_alloc = 1; | |
3950 | } else { | |
3951 | space_info->chunk_alloc = 1; | |
9ed74f2d | 3952 | } |
0e4f8f88 | 3953 | |
25179201 | 3954 | spin_unlock(&space_info->lock); |
9ed74f2d | 3955 | |
6d74119f JB |
3956 | mutex_lock(&fs_info->chunk_mutex); |
3957 | ||
3958 | /* | |
3959 | * The chunk_mutex is held throughout the entirety of a chunk | |
3960 | * allocation, so once we've acquired the chunk_mutex we know that the | |
3961 | * other guy is done and we need to recheck and see if we should | |
3962 | * allocate. | |
3963 | */ | |
3964 | if (wait_for_alloc) { | |
3965 | mutex_unlock(&fs_info->chunk_mutex); | |
3966 | wait_for_alloc = 0; | |
3967 | goto again; | |
3968 | } | |
3969 | ||
c6b305a8 JB |
3970 | trans->allocating_chunk = true; |
3971 | ||
67377734 JB |
3972 | /* |
3973 | * If we have mixed data/metadata chunks we want to make sure we keep | |
3974 | * allocating mixed chunks instead of individual chunks. | |
3975 | */ | |
3976 | if (btrfs_mixed_space_info(space_info)) | |
3977 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
3978 | ||
97e728d4 JB |
3979 | /* |
3980 | * if we're doing a data chunk, go ahead and make sure that | |
3981 | * we keep a reasonable number of metadata chunks allocated in the | |
3982 | * FS as well. | |
3983 | */ | |
9ed74f2d | 3984 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
3985 | fs_info->data_chunk_allocations++; |
3986 | if (!(fs_info->data_chunk_allocations % | |
3987 | fs_info->metadata_ratio)) | |
3988 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
3989 | } |
3990 | ||
15d1ff81 LB |
3991 | /* |
3992 | * Check if we have enough space in SYSTEM chunk because we may need | |
3993 | * to update devices. | |
3994 | */ | |
3995 | check_system_chunk(trans, extent_root, flags); | |
3996 | ||
2b82032c | 3997 | ret = btrfs_alloc_chunk(trans, extent_root, flags); |
c6b305a8 | 3998 | trans->allocating_chunk = false; |
92b8e897 | 3999 | |
9ed74f2d | 4000 | spin_lock(&space_info->lock); |
a81cb9a2 AO |
4001 | if (ret < 0 && ret != -ENOSPC) |
4002 | goto out; | |
9ed74f2d | 4003 | if (ret) |
6324fbf3 | 4004 | space_info->full = 1; |
424499db YZ |
4005 | else |
4006 | ret = 1; | |
6d74119f | 4007 | |
0e4f8f88 | 4008 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
a81cb9a2 | 4009 | out: |
6d74119f | 4010 | space_info->chunk_alloc = 0; |
9ed74f2d | 4011 | spin_unlock(&space_info->lock); |
a25c75d5 | 4012 | mutex_unlock(&fs_info->chunk_mutex); |
0f9dd46c | 4013 | return ret; |
6324fbf3 | 4014 | } |
9ed74f2d | 4015 | |
a80c8dcf JB |
4016 | static int can_overcommit(struct btrfs_root *root, |
4017 | struct btrfs_space_info *space_info, u64 bytes, | |
08e007d2 | 4018 | enum btrfs_reserve_flush_enum flush) |
a80c8dcf | 4019 | { |
96f1bb57 | 4020 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
a80c8dcf | 4021 | u64 profile = btrfs_get_alloc_profile(root, 0); |
3c76cd84 | 4022 | u64 space_size; |
a80c8dcf JB |
4023 | u64 avail; |
4024 | u64 used; | |
4025 | ||
4026 | used = space_info->bytes_used + space_info->bytes_reserved + | |
96f1bb57 JB |
4027 | space_info->bytes_pinned + space_info->bytes_readonly; |
4028 | ||
96f1bb57 JB |
4029 | /* |
4030 | * We only want to allow over committing if we have lots of actual space | |
4031 | * free, but if we don't have enough space to handle the global reserve | |
4032 | * space then we could end up having a real enospc problem when trying | |
4033 | * to allocate a chunk or some other such important allocation. | |
4034 | */ | |
3c76cd84 MX |
4035 | spin_lock(&global_rsv->lock); |
4036 | space_size = calc_global_rsv_need_space(global_rsv); | |
4037 | spin_unlock(&global_rsv->lock); | |
4038 | if (used + space_size >= space_info->total_bytes) | |
96f1bb57 JB |
4039 | return 0; |
4040 | ||
4041 | used += space_info->bytes_may_use; | |
a80c8dcf JB |
4042 | |
4043 | spin_lock(&root->fs_info->free_chunk_lock); | |
4044 | avail = root->fs_info->free_chunk_space; | |
4045 | spin_unlock(&root->fs_info->free_chunk_lock); | |
4046 | ||
4047 | /* | |
4048 | * If we have dup, raid1 or raid10 then only half of the free | |
53b381b3 DW |
4049 | * space is actually useable. For raid56, the space info used |
4050 | * doesn't include the parity drive, so we don't have to | |
4051 | * change the math | |
a80c8dcf JB |
4052 | */ |
4053 | if (profile & (BTRFS_BLOCK_GROUP_DUP | | |
4054 | BTRFS_BLOCK_GROUP_RAID1 | | |
4055 | BTRFS_BLOCK_GROUP_RAID10)) | |
4056 | avail >>= 1; | |
4057 | ||
4058 | /* | |
561c294d MX |
4059 | * If we aren't flushing all things, let us overcommit up to |
4060 | * 1/2th of the space. If we can flush, don't let us overcommit | |
4061 | * too much, let it overcommit up to 1/8 of the space. | |
a80c8dcf | 4062 | */ |
08e007d2 | 4063 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
14575aef | 4064 | avail >>= 3; |
a80c8dcf | 4065 | else |
14575aef | 4066 | avail >>= 1; |
a80c8dcf | 4067 | |
14575aef | 4068 | if (used + bytes < space_info->total_bytes + avail) |
a80c8dcf JB |
4069 | return 1; |
4070 | return 0; | |
4071 | } | |
4072 | ||
48a3b636 | 4073 | static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root, |
6c255e67 | 4074 | unsigned long nr_pages, int nr_items) |
da633a42 MX |
4075 | { |
4076 | struct super_block *sb = root->fs_info->sb; | |
da633a42 | 4077 | |
925a6efb JB |
4078 | if (down_read_trylock(&sb->s_umount)) { |
4079 | writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE); | |
4080 | up_read(&sb->s_umount); | |
4081 | } else { | |
da633a42 MX |
4082 | /* |
4083 | * We needn't worry the filesystem going from r/w to r/o though | |
4084 | * we don't acquire ->s_umount mutex, because the filesystem | |
4085 | * should guarantee the delalloc inodes list be empty after | |
4086 | * the filesystem is readonly(all dirty pages are written to | |
4087 | * the disk). | |
4088 | */ | |
6c255e67 | 4089 | btrfs_start_delalloc_roots(root->fs_info, 0, nr_items); |
98ad69cf | 4090 | if (!current->journal_info) |
6c255e67 | 4091 | btrfs_wait_ordered_roots(root->fs_info, nr_items); |
da633a42 MX |
4092 | } |
4093 | } | |
4094 | ||
18cd8ea6 MX |
4095 | static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim) |
4096 | { | |
4097 | u64 bytes; | |
4098 | int nr; | |
4099 | ||
4100 | bytes = btrfs_calc_trans_metadata_size(root, 1); | |
4101 | nr = (int)div64_u64(to_reclaim, bytes); | |
4102 | if (!nr) | |
4103 | nr = 1; | |
4104 | return nr; | |
4105 | } | |
4106 | ||
c61a16a7 MX |
4107 | #define EXTENT_SIZE_PER_ITEM (256 * 1024) |
4108 | ||
9ed74f2d | 4109 | /* |
5da9d01b | 4110 | * shrink metadata reservation for delalloc |
9ed74f2d | 4111 | */ |
f4c738c2 JB |
4112 | static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig, |
4113 | bool wait_ordered) | |
5da9d01b | 4114 | { |
0ca1f7ce | 4115 | struct btrfs_block_rsv *block_rsv; |
0019f10d | 4116 | struct btrfs_space_info *space_info; |
663350ac | 4117 | struct btrfs_trans_handle *trans; |
f4c738c2 | 4118 | u64 delalloc_bytes; |
5da9d01b | 4119 | u64 max_reclaim; |
b1953bce | 4120 | long time_left; |
d3ee29e3 MX |
4121 | unsigned long nr_pages; |
4122 | int loops; | |
b0244199 | 4123 | int items; |
08e007d2 | 4124 | enum btrfs_reserve_flush_enum flush; |
5da9d01b | 4125 | |
c61a16a7 | 4126 | /* Calc the number of the pages we need flush for space reservation */ |
b0244199 MX |
4127 | items = calc_reclaim_items_nr(root, to_reclaim); |
4128 | to_reclaim = items * EXTENT_SIZE_PER_ITEM; | |
c61a16a7 | 4129 | |
663350ac | 4130 | trans = (struct btrfs_trans_handle *)current->journal_info; |
0ca1f7ce | 4131 | block_rsv = &root->fs_info->delalloc_block_rsv; |
0019f10d | 4132 | space_info = block_rsv->space_info; |
bf9022e0 | 4133 | |
963d678b MX |
4134 | delalloc_bytes = percpu_counter_sum_positive( |
4135 | &root->fs_info->delalloc_bytes); | |
f4c738c2 | 4136 | if (delalloc_bytes == 0) { |
fdb5effd | 4137 | if (trans) |
f4c738c2 | 4138 | return; |
38c135af | 4139 | if (wait_ordered) |
b0244199 | 4140 | btrfs_wait_ordered_roots(root->fs_info, items); |
f4c738c2 | 4141 | return; |
fdb5effd JB |
4142 | } |
4143 | ||
d3ee29e3 | 4144 | loops = 0; |
f4c738c2 JB |
4145 | while (delalloc_bytes && loops < 3) { |
4146 | max_reclaim = min(delalloc_bytes, to_reclaim); | |
4147 | nr_pages = max_reclaim >> PAGE_CACHE_SHIFT; | |
6c255e67 | 4148 | btrfs_writeback_inodes_sb_nr(root, nr_pages, items); |
dea31f52 JB |
4149 | /* |
4150 | * We need to wait for the async pages to actually start before | |
4151 | * we do anything. | |
4152 | */ | |
9f3a074d MX |
4153 | max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages); |
4154 | if (!max_reclaim) | |
4155 | goto skip_async; | |
4156 | ||
4157 | if (max_reclaim <= nr_pages) | |
4158 | max_reclaim = 0; | |
4159 | else | |
4160 | max_reclaim -= nr_pages; | |
dea31f52 | 4161 | |
9f3a074d MX |
4162 | wait_event(root->fs_info->async_submit_wait, |
4163 | atomic_read(&root->fs_info->async_delalloc_pages) <= | |
4164 | (int)max_reclaim); | |
4165 | skip_async: | |
08e007d2 MX |
4166 | if (!trans) |
4167 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
4168 | else | |
4169 | flush = BTRFS_RESERVE_NO_FLUSH; | |
0019f10d | 4170 | spin_lock(&space_info->lock); |
08e007d2 | 4171 | if (can_overcommit(root, space_info, orig, flush)) { |
f4c738c2 JB |
4172 | spin_unlock(&space_info->lock); |
4173 | break; | |
4174 | } | |
0019f10d | 4175 | spin_unlock(&space_info->lock); |
5da9d01b | 4176 | |
36e39c40 | 4177 | loops++; |
f104d044 | 4178 | if (wait_ordered && !trans) { |
b0244199 | 4179 | btrfs_wait_ordered_roots(root->fs_info, items); |
f104d044 | 4180 | } else { |
f4c738c2 | 4181 | time_left = schedule_timeout_killable(1); |
f104d044 JB |
4182 | if (time_left) |
4183 | break; | |
4184 | } | |
963d678b MX |
4185 | delalloc_bytes = percpu_counter_sum_positive( |
4186 | &root->fs_info->delalloc_bytes); | |
5da9d01b | 4187 | } |
5da9d01b YZ |
4188 | } |
4189 | ||
663350ac JB |
4190 | /** |
4191 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
4192 | * @root - the root we're allocating for | |
4193 | * @bytes - the number of bytes we want to reserve | |
4194 | * @force - force the commit | |
8bb8ab2e | 4195 | * |
663350ac JB |
4196 | * This will check to make sure that committing the transaction will actually |
4197 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
4198 | * will return -ENOSPC. | |
8bb8ab2e | 4199 | */ |
663350ac JB |
4200 | static int may_commit_transaction(struct btrfs_root *root, |
4201 | struct btrfs_space_info *space_info, | |
4202 | u64 bytes, int force) | |
4203 | { | |
4204 | struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv; | |
4205 | struct btrfs_trans_handle *trans; | |
4206 | ||
4207 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
4208 | if (trans) | |
4209 | return -EAGAIN; | |
4210 | ||
4211 | if (force) | |
4212 | goto commit; | |
4213 | ||
4214 | /* See if there is enough pinned space to make this reservation */ | |
b150a4f1 | 4215 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
0424c548 | 4216 | bytes) >= 0) |
663350ac | 4217 | goto commit; |
663350ac JB |
4218 | |
4219 | /* | |
4220 | * See if there is some space in the delayed insertion reservation for | |
4221 | * this reservation. | |
4222 | */ | |
4223 | if (space_info != delayed_rsv->space_info) | |
4224 | return -ENOSPC; | |
4225 | ||
4226 | spin_lock(&delayed_rsv->lock); | |
b150a4f1 JB |
4227 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
4228 | bytes - delayed_rsv->size) >= 0) { | |
663350ac JB |
4229 | spin_unlock(&delayed_rsv->lock); |
4230 | return -ENOSPC; | |
4231 | } | |
4232 | spin_unlock(&delayed_rsv->lock); | |
4233 | ||
4234 | commit: | |
4235 | trans = btrfs_join_transaction(root); | |
4236 | if (IS_ERR(trans)) | |
4237 | return -ENOSPC; | |
4238 | ||
4239 | return btrfs_commit_transaction(trans, root); | |
4240 | } | |
4241 | ||
96c3f433 | 4242 | enum flush_state { |
67b0fd63 JB |
4243 | FLUSH_DELAYED_ITEMS_NR = 1, |
4244 | FLUSH_DELAYED_ITEMS = 2, | |
4245 | FLUSH_DELALLOC = 3, | |
4246 | FLUSH_DELALLOC_WAIT = 4, | |
ea658bad JB |
4247 | ALLOC_CHUNK = 5, |
4248 | COMMIT_TRANS = 6, | |
96c3f433 JB |
4249 | }; |
4250 | ||
4251 | static int flush_space(struct btrfs_root *root, | |
4252 | struct btrfs_space_info *space_info, u64 num_bytes, | |
4253 | u64 orig_bytes, int state) | |
4254 | { | |
4255 | struct btrfs_trans_handle *trans; | |
4256 | int nr; | |
f4c738c2 | 4257 | int ret = 0; |
96c3f433 JB |
4258 | |
4259 | switch (state) { | |
96c3f433 JB |
4260 | case FLUSH_DELAYED_ITEMS_NR: |
4261 | case FLUSH_DELAYED_ITEMS: | |
18cd8ea6 MX |
4262 | if (state == FLUSH_DELAYED_ITEMS_NR) |
4263 | nr = calc_reclaim_items_nr(root, num_bytes) * 2; | |
4264 | else | |
96c3f433 | 4265 | nr = -1; |
18cd8ea6 | 4266 | |
96c3f433 JB |
4267 | trans = btrfs_join_transaction(root); |
4268 | if (IS_ERR(trans)) { | |
4269 | ret = PTR_ERR(trans); | |
4270 | break; | |
4271 | } | |
4272 | ret = btrfs_run_delayed_items_nr(trans, root, nr); | |
4273 | btrfs_end_transaction(trans, root); | |
4274 | break; | |
67b0fd63 JB |
4275 | case FLUSH_DELALLOC: |
4276 | case FLUSH_DELALLOC_WAIT: | |
24af7dd1 | 4277 | shrink_delalloc(root, num_bytes * 2, orig_bytes, |
67b0fd63 JB |
4278 | state == FLUSH_DELALLOC_WAIT); |
4279 | break; | |
ea658bad JB |
4280 | case ALLOC_CHUNK: |
4281 | trans = btrfs_join_transaction(root); | |
4282 | if (IS_ERR(trans)) { | |
4283 | ret = PTR_ERR(trans); | |
4284 | break; | |
4285 | } | |
4286 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
ea658bad JB |
4287 | btrfs_get_alloc_profile(root, 0), |
4288 | CHUNK_ALLOC_NO_FORCE); | |
4289 | btrfs_end_transaction(trans, root); | |
4290 | if (ret == -ENOSPC) | |
4291 | ret = 0; | |
4292 | break; | |
96c3f433 JB |
4293 | case COMMIT_TRANS: |
4294 | ret = may_commit_transaction(root, space_info, orig_bytes, 0); | |
4295 | break; | |
4296 | default: | |
4297 | ret = -ENOSPC; | |
4298 | break; | |
4299 | } | |
4300 | ||
4301 | return ret; | |
4302 | } | |
21c7e756 MX |
4303 | |
4304 | static inline u64 | |
4305 | btrfs_calc_reclaim_metadata_size(struct btrfs_root *root, | |
4306 | struct btrfs_space_info *space_info) | |
4307 | { | |
4308 | u64 used; | |
4309 | u64 expected; | |
4310 | u64 to_reclaim; | |
4311 | ||
4312 | to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024, | |
4313 | 16 * 1024 * 1024); | |
4314 | spin_lock(&space_info->lock); | |
4315 | if (can_overcommit(root, space_info, to_reclaim, | |
4316 | BTRFS_RESERVE_FLUSH_ALL)) { | |
4317 | to_reclaim = 0; | |
4318 | goto out; | |
4319 | } | |
4320 | ||
4321 | used = space_info->bytes_used + space_info->bytes_reserved + | |
4322 | space_info->bytes_pinned + space_info->bytes_readonly + | |
4323 | space_info->bytes_may_use; | |
4324 | if (can_overcommit(root, space_info, 1024 * 1024, | |
4325 | BTRFS_RESERVE_FLUSH_ALL)) | |
4326 | expected = div_factor_fine(space_info->total_bytes, 95); | |
4327 | else | |
4328 | expected = div_factor_fine(space_info->total_bytes, 90); | |
4329 | ||
4330 | if (used > expected) | |
4331 | to_reclaim = used - expected; | |
4332 | else | |
4333 | to_reclaim = 0; | |
4334 | to_reclaim = min(to_reclaim, space_info->bytes_may_use + | |
4335 | space_info->bytes_reserved); | |
4336 | out: | |
4337 | spin_unlock(&space_info->lock); | |
4338 | ||
4339 | return to_reclaim; | |
4340 | } | |
4341 | ||
4342 | static inline int need_do_async_reclaim(struct btrfs_space_info *space_info, | |
4343 | struct btrfs_fs_info *fs_info, u64 used) | |
4344 | { | |
4345 | return (used >= div_factor_fine(space_info->total_bytes, 98) && | |
4346 | !btrfs_fs_closing(fs_info) && | |
4347 | !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state)); | |
4348 | } | |
4349 | ||
4350 | static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info, | |
4351 | struct btrfs_fs_info *fs_info) | |
4352 | { | |
4353 | u64 used; | |
4354 | ||
4355 | spin_lock(&space_info->lock); | |
4356 | used = space_info->bytes_used + space_info->bytes_reserved + | |
4357 | space_info->bytes_pinned + space_info->bytes_readonly + | |
4358 | space_info->bytes_may_use; | |
4359 | if (need_do_async_reclaim(space_info, fs_info, used)) { | |
4360 | spin_unlock(&space_info->lock); | |
4361 | return 1; | |
4362 | } | |
4363 | spin_unlock(&space_info->lock); | |
4364 | ||
4365 | return 0; | |
4366 | } | |
4367 | ||
4368 | static void btrfs_async_reclaim_metadata_space(struct work_struct *work) | |
4369 | { | |
4370 | struct btrfs_fs_info *fs_info; | |
4371 | struct btrfs_space_info *space_info; | |
4372 | u64 to_reclaim; | |
4373 | int flush_state; | |
4374 | ||
4375 | fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work); | |
4376 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
4377 | ||
4378 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root, | |
4379 | space_info); | |
4380 | if (!to_reclaim) | |
4381 | return; | |
4382 | ||
4383 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
4384 | do { | |
4385 | flush_space(fs_info->fs_root, space_info, to_reclaim, | |
4386 | to_reclaim, flush_state); | |
4387 | flush_state++; | |
4388 | if (!btrfs_need_do_async_reclaim(space_info, fs_info)) | |
4389 | return; | |
4390 | } while (flush_state <= COMMIT_TRANS); | |
4391 | ||
4392 | if (btrfs_need_do_async_reclaim(space_info, fs_info)) | |
4393 | queue_work(system_unbound_wq, work); | |
4394 | } | |
4395 | ||
4396 | void btrfs_init_async_reclaim_work(struct work_struct *work) | |
4397 | { | |
4398 | INIT_WORK(work, btrfs_async_reclaim_metadata_space); | |
4399 | } | |
4400 | ||
4a92b1b8 JB |
4401 | /** |
4402 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
4403 | * @root - the root we're allocating for | |
4404 | * @block_rsv - the block_rsv we're allocating for | |
4405 | * @orig_bytes - the number of bytes we want | |
48fc7f7e | 4406 | * @flush - whether or not we can flush to make our reservation |
8bb8ab2e | 4407 | * |
4a92b1b8 JB |
4408 | * This will reserve orgi_bytes number of bytes from the space info associated |
4409 | * with the block_rsv. If there is not enough space it will make an attempt to | |
4410 | * flush out space to make room. It will do this by flushing delalloc if | |
4411 | * possible or committing the transaction. If flush is 0 then no attempts to | |
4412 | * regain reservations will be made and this will fail if there is not enough | |
4413 | * space already. | |
8bb8ab2e | 4414 | */ |
4a92b1b8 | 4415 | static int reserve_metadata_bytes(struct btrfs_root *root, |
8bb8ab2e | 4416 | struct btrfs_block_rsv *block_rsv, |
08e007d2 MX |
4417 | u64 orig_bytes, |
4418 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 4419 | { |
f0486c68 | 4420 | struct btrfs_space_info *space_info = block_rsv->space_info; |
2bf64758 | 4421 | u64 used; |
8bb8ab2e | 4422 | u64 num_bytes = orig_bytes; |
67b0fd63 | 4423 | int flush_state = FLUSH_DELAYED_ITEMS_NR; |
8bb8ab2e | 4424 | int ret = 0; |
fdb5effd | 4425 | bool flushing = false; |
9ed74f2d | 4426 | |
8bb8ab2e | 4427 | again: |
fdb5effd | 4428 | ret = 0; |
8bb8ab2e | 4429 | spin_lock(&space_info->lock); |
fdb5effd | 4430 | /* |
08e007d2 MX |
4431 | * We only want to wait if somebody other than us is flushing and we |
4432 | * are actually allowed to flush all things. | |
fdb5effd | 4433 | */ |
08e007d2 MX |
4434 | while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing && |
4435 | space_info->flush) { | |
fdb5effd JB |
4436 | spin_unlock(&space_info->lock); |
4437 | /* | |
4438 | * If we have a trans handle we can't wait because the flusher | |
4439 | * may have to commit the transaction, which would mean we would | |
4440 | * deadlock since we are waiting for the flusher to finish, but | |
4441 | * hold the current transaction open. | |
4442 | */ | |
663350ac | 4443 | if (current->journal_info) |
fdb5effd | 4444 | return -EAGAIN; |
b9688bb8 AJ |
4445 | ret = wait_event_killable(space_info->wait, !space_info->flush); |
4446 | /* Must have been killed, return */ | |
4447 | if (ret) | |
fdb5effd JB |
4448 | return -EINTR; |
4449 | ||
4450 | spin_lock(&space_info->lock); | |
4451 | } | |
4452 | ||
4453 | ret = -ENOSPC; | |
2bf64758 JB |
4454 | used = space_info->bytes_used + space_info->bytes_reserved + |
4455 | space_info->bytes_pinned + space_info->bytes_readonly + | |
4456 | space_info->bytes_may_use; | |
9ed74f2d | 4457 | |
8bb8ab2e JB |
4458 | /* |
4459 | * The idea here is that we've not already over-reserved the block group | |
4460 | * then we can go ahead and save our reservation first and then start | |
4461 | * flushing if we need to. Otherwise if we've already overcommitted | |
4462 | * lets start flushing stuff first and then come back and try to make | |
4463 | * our reservation. | |
4464 | */ | |
2bf64758 JB |
4465 | if (used <= space_info->total_bytes) { |
4466 | if (used + orig_bytes <= space_info->total_bytes) { | |
fb25e914 | 4467 | space_info->bytes_may_use += orig_bytes; |
8c2a3ca2 | 4468 | trace_btrfs_space_reservation(root->fs_info, |
2bcc0328 | 4469 | "space_info", space_info->flags, orig_bytes, 1); |
8bb8ab2e JB |
4470 | ret = 0; |
4471 | } else { | |
4472 | /* | |
4473 | * Ok set num_bytes to orig_bytes since we aren't | |
4474 | * overocmmitted, this way we only try and reclaim what | |
4475 | * we need. | |
4476 | */ | |
4477 | num_bytes = orig_bytes; | |
4478 | } | |
4479 | } else { | |
4480 | /* | |
4481 | * Ok we're over committed, set num_bytes to the overcommitted | |
4482 | * amount plus the amount of bytes that we need for this | |
4483 | * reservation. | |
4484 | */ | |
2bf64758 | 4485 | num_bytes = used - space_info->total_bytes + |
96c3f433 | 4486 | (orig_bytes * 2); |
8bb8ab2e | 4487 | } |
9ed74f2d | 4488 | |
44734ed1 JB |
4489 | if (ret && can_overcommit(root, space_info, orig_bytes, flush)) { |
4490 | space_info->bytes_may_use += orig_bytes; | |
4491 | trace_btrfs_space_reservation(root->fs_info, "space_info", | |
4492 | space_info->flags, orig_bytes, | |
4493 | 1); | |
4494 | ret = 0; | |
2bf64758 JB |
4495 | } |
4496 | ||
8bb8ab2e JB |
4497 | /* |
4498 | * Couldn't make our reservation, save our place so while we're trying | |
4499 | * to reclaim space we can actually use it instead of somebody else | |
4500 | * stealing it from us. | |
08e007d2 MX |
4501 | * |
4502 | * We make the other tasks wait for the flush only when we can flush | |
4503 | * all things. | |
8bb8ab2e | 4504 | */ |
72bcd99d | 4505 | if (ret && flush != BTRFS_RESERVE_NO_FLUSH) { |
fdb5effd JB |
4506 | flushing = true; |
4507 | space_info->flush = 1; | |
21c7e756 MX |
4508 | } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
4509 | used += orig_bytes; | |
4510 | if (need_do_async_reclaim(space_info, root->fs_info, used) && | |
4511 | !work_busy(&root->fs_info->async_reclaim_work)) | |
4512 | queue_work(system_unbound_wq, | |
4513 | &root->fs_info->async_reclaim_work); | |
8bb8ab2e | 4514 | } |
f0486c68 | 4515 | spin_unlock(&space_info->lock); |
9ed74f2d | 4516 | |
08e007d2 | 4517 | if (!ret || flush == BTRFS_RESERVE_NO_FLUSH) |
8bb8ab2e | 4518 | goto out; |
f0486c68 | 4519 | |
96c3f433 JB |
4520 | ret = flush_space(root, space_info, num_bytes, orig_bytes, |
4521 | flush_state); | |
4522 | flush_state++; | |
08e007d2 MX |
4523 | |
4524 | /* | |
4525 | * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock | |
4526 | * would happen. So skip delalloc flush. | |
4527 | */ | |
4528 | if (flush == BTRFS_RESERVE_FLUSH_LIMIT && | |
4529 | (flush_state == FLUSH_DELALLOC || | |
4530 | flush_state == FLUSH_DELALLOC_WAIT)) | |
4531 | flush_state = ALLOC_CHUNK; | |
4532 | ||
96c3f433 | 4533 | if (!ret) |
8bb8ab2e | 4534 | goto again; |
08e007d2 MX |
4535 | else if (flush == BTRFS_RESERVE_FLUSH_LIMIT && |
4536 | flush_state < COMMIT_TRANS) | |
4537 | goto again; | |
4538 | else if (flush == BTRFS_RESERVE_FLUSH_ALL && | |
4539 | flush_state <= COMMIT_TRANS) | |
8bb8ab2e JB |
4540 | goto again; |
4541 | ||
4542 | out: | |
5d80366e JB |
4543 | if (ret == -ENOSPC && |
4544 | unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) { | |
4545 | struct btrfs_block_rsv *global_rsv = | |
4546 | &root->fs_info->global_block_rsv; | |
4547 | ||
4548 | if (block_rsv != global_rsv && | |
4549 | !block_rsv_use_bytes(global_rsv, orig_bytes)) | |
4550 | ret = 0; | |
4551 | } | |
cab45e22 JM |
4552 | if (ret == -ENOSPC) |
4553 | trace_btrfs_space_reservation(root->fs_info, | |
4554 | "space_info:enospc", | |
4555 | space_info->flags, orig_bytes, 1); | |
fdb5effd | 4556 | if (flushing) { |
8bb8ab2e | 4557 | spin_lock(&space_info->lock); |
fdb5effd JB |
4558 | space_info->flush = 0; |
4559 | wake_up_all(&space_info->wait); | |
8bb8ab2e | 4560 | spin_unlock(&space_info->lock); |
f0486c68 | 4561 | } |
f0486c68 YZ |
4562 | return ret; |
4563 | } | |
4564 | ||
79787eaa JM |
4565 | static struct btrfs_block_rsv *get_block_rsv( |
4566 | const struct btrfs_trans_handle *trans, | |
4567 | const struct btrfs_root *root) | |
f0486c68 | 4568 | { |
4c13d758 JB |
4569 | struct btrfs_block_rsv *block_rsv = NULL; |
4570 | ||
27cdeb70 | 4571 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
0e721106 JB |
4572 | block_rsv = trans->block_rsv; |
4573 | ||
4574 | if (root == root->fs_info->csum_root && trans->adding_csums) | |
f0486c68 | 4575 | block_rsv = trans->block_rsv; |
4c13d758 | 4576 | |
f7a81ea4 SB |
4577 | if (root == root->fs_info->uuid_root) |
4578 | block_rsv = trans->block_rsv; | |
4579 | ||
4c13d758 | 4580 | if (!block_rsv) |
f0486c68 YZ |
4581 | block_rsv = root->block_rsv; |
4582 | ||
4583 | if (!block_rsv) | |
4584 | block_rsv = &root->fs_info->empty_block_rsv; | |
4585 | ||
4586 | return block_rsv; | |
4587 | } | |
4588 | ||
4589 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
4590 | u64 num_bytes) | |
4591 | { | |
4592 | int ret = -ENOSPC; | |
4593 | spin_lock(&block_rsv->lock); | |
4594 | if (block_rsv->reserved >= num_bytes) { | |
4595 | block_rsv->reserved -= num_bytes; | |
4596 | if (block_rsv->reserved < block_rsv->size) | |
4597 | block_rsv->full = 0; | |
4598 | ret = 0; | |
4599 | } | |
4600 | spin_unlock(&block_rsv->lock); | |
4601 | return ret; | |
4602 | } | |
4603 | ||
4604 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
4605 | u64 num_bytes, int update_size) | |
4606 | { | |
4607 | spin_lock(&block_rsv->lock); | |
4608 | block_rsv->reserved += num_bytes; | |
4609 | if (update_size) | |
4610 | block_rsv->size += num_bytes; | |
4611 | else if (block_rsv->reserved >= block_rsv->size) | |
4612 | block_rsv->full = 1; | |
4613 | spin_unlock(&block_rsv->lock); | |
4614 | } | |
4615 | ||
d52be818 JB |
4616 | int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info, |
4617 | struct btrfs_block_rsv *dest, u64 num_bytes, | |
4618 | int min_factor) | |
4619 | { | |
4620 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
4621 | u64 min_bytes; | |
4622 | ||
4623 | if (global_rsv->space_info != dest->space_info) | |
4624 | return -ENOSPC; | |
4625 | ||
4626 | spin_lock(&global_rsv->lock); | |
4627 | min_bytes = div_factor(global_rsv->size, min_factor); | |
4628 | if (global_rsv->reserved < min_bytes + num_bytes) { | |
4629 | spin_unlock(&global_rsv->lock); | |
4630 | return -ENOSPC; | |
4631 | } | |
4632 | global_rsv->reserved -= num_bytes; | |
4633 | if (global_rsv->reserved < global_rsv->size) | |
4634 | global_rsv->full = 0; | |
4635 | spin_unlock(&global_rsv->lock); | |
4636 | ||
4637 | block_rsv_add_bytes(dest, num_bytes, 1); | |
4638 | return 0; | |
4639 | } | |
4640 | ||
8c2a3ca2 JB |
4641 | static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info, |
4642 | struct btrfs_block_rsv *block_rsv, | |
62a45b60 | 4643 | struct btrfs_block_rsv *dest, u64 num_bytes) |
f0486c68 YZ |
4644 | { |
4645 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
4646 | ||
4647 | spin_lock(&block_rsv->lock); | |
4648 | if (num_bytes == (u64)-1) | |
4649 | num_bytes = block_rsv->size; | |
4650 | block_rsv->size -= num_bytes; | |
4651 | if (block_rsv->reserved >= block_rsv->size) { | |
4652 | num_bytes = block_rsv->reserved - block_rsv->size; | |
4653 | block_rsv->reserved = block_rsv->size; | |
4654 | block_rsv->full = 1; | |
4655 | } else { | |
4656 | num_bytes = 0; | |
4657 | } | |
4658 | spin_unlock(&block_rsv->lock); | |
4659 | ||
4660 | if (num_bytes > 0) { | |
4661 | if (dest) { | |
e9e22899 JB |
4662 | spin_lock(&dest->lock); |
4663 | if (!dest->full) { | |
4664 | u64 bytes_to_add; | |
4665 | ||
4666 | bytes_to_add = dest->size - dest->reserved; | |
4667 | bytes_to_add = min(num_bytes, bytes_to_add); | |
4668 | dest->reserved += bytes_to_add; | |
4669 | if (dest->reserved >= dest->size) | |
4670 | dest->full = 1; | |
4671 | num_bytes -= bytes_to_add; | |
4672 | } | |
4673 | spin_unlock(&dest->lock); | |
4674 | } | |
4675 | if (num_bytes) { | |
f0486c68 | 4676 | spin_lock(&space_info->lock); |
fb25e914 | 4677 | space_info->bytes_may_use -= num_bytes; |
8c2a3ca2 | 4678 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4679 | space_info->flags, num_bytes, 0); |
f0486c68 | 4680 | spin_unlock(&space_info->lock); |
4e06bdd6 | 4681 | } |
9ed74f2d | 4682 | } |
f0486c68 | 4683 | } |
4e06bdd6 | 4684 | |
f0486c68 YZ |
4685 | static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src, |
4686 | struct btrfs_block_rsv *dst, u64 num_bytes) | |
4687 | { | |
4688 | int ret; | |
9ed74f2d | 4689 | |
f0486c68 YZ |
4690 | ret = block_rsv_use_bytes(src, num_bytes); |
4691 | if (ret) | |
4692 | return ret; | |
9ed74f2d | 4693 | |
f0486c68 | 4694 | block_rsv_add_bytes(dst, num_bytes, 1); |
9ed74f2d JB |
4695 | return 0; |
4696 | } | |
4697 | ||
66d8f3dd | 4698 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type) |
9ed74f2d | 4699 | { |
f0486c68 YZ |
4700 | memset(rsv, 0, sizeof(*rsv)); |
4701 | spin_lock_init(&rsv->lock); | |
66d8f3dd | 4702 | rsv->type = type; |
f0486c68 YZ |
4703 | } |
4704 | ||
66d8f3dd MX |
4705 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root, |
4706 | unsigned short type) | |
f0486c68 YZ |
4707 | { |
4708 | struct btrfs_block_rsv *block_rsv; | |
4709 | struct btrfs_fs_info *fs_info = root->fs_info; | |
9ed74f2d | 4710 | |
f0486c68 YZ |
4711 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
4712 | if (!block_rsv) | |
4713 | return NULL; | |
9ed74f2d | 4714 | |
66d8f3dd | 4715 | btrfs_init_block_rsv(block_rsv, type); |
f0486c68 YZ |
4716 | block_rsv->space_info = __find_space_info(fs_info, |
4717 | BTRFS_BLOCK_GROUP_METADATA); | |
f0486c68 YZ |
4718 | return block_rsv; |
4719 | } | |
9ed74f2d | 4720 | |
f0486c68 YZ |
4721 | void btrfs_free_block_rsv(struct btrfs_root *root, |
4722 | struct btrfs_block_rsv *rsv) | |
4723 | { | |
2aaa6655 JB |
4724 | if (!rsv) |
4725 | return; | |
dabdb640 JB |
4726 | btrfs_block_rsv_release(root, rsv, (u64)-1); |
4727 | kfree(rsv); | |
9ed74f2d JB |
4728 | } |
4729 | ||
08e007d2 MX |
4730 | int btrfs_block_rsv_add(struct btrfs_root *root, |
4731 | struct btrfs_block_rsv *block_rsv, u64 num_bytes, | |
4732 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 4733 | { |
f0486c68 | 4734 | int ret; |
9ed74f2d | 4735 | |
f0486c68 YZ |
4736 | if (num_bytes == 0) |
4737 | return 0; | |
8bb8ab2e | 4738 | |
61b520a9 | 4739 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
f0486c68 YZ |
4740 | if (!ret) { |
4741 | block_rsv_add_bytes(block_rsv, num_bytes, 1); | |
4742 | return 0; | |
4743 | } | |
9ed74f2d | 4744 | |
f0486c68 | 4745 | return ret; |
f0486c68 | 4746 | } |
9ed74f2d | 4747 | |
4a92b1b8 | 4748 | int btrfs_block_rsv_check(struct btrfs_root *root, |
36ba022a | 4749 | struct btrfs_block_rsv *block_rsv, int min_factor) |
f0486c68 YZ |
4750 | { |
4751 | u64 num_bytes = 0; | |
f0486c68 | 4752 | int ret = -ENOSPC; |
9ed74f2d | 4753 | |
f0486c68 YZ |
4754 | if (!block_rsv) |
4755 | return 0; | |
9ed74f2d | 4756 | |
f0486c68 | 4757 | spin_lock(&block_rsv->lock); |
36ba022a JB |
4758 | num_bytes = div_factor(block_rsv->size, min_factor); |
4759 | if (block_rsv->reserved >= num_bytes) | |
4760 | ret = 0; | |
4761 | spin_unlock(&block_rsv->lock); | |
9ed74f2d | 4762 | |
36ba022a JB |
4763 | return ret; |
4764 | } | |
4765 | ||
08e007d2 MX |
4766 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
4767 | struct btrfs_block_rsv *block_rsv, u64 min_reserved, | |
4768 | enum btrfs_reserve_flush_enum flush) | |
36ba022a JB |
4769 | { |
4770 | u64 num_bytes = 0; | |
4771 | int ret = -ENOSPC; | |
4772 | ||
4773 | if (!block_rsv) | |
4774 | return 0; | |
4775 | ||
4776 | spin_lock(&block_rsv->lock); | |
4777 | num_bytes = min_reserved; | |
13553e52 | 4778 | if (block_rsv->reserved >= num_bytes) |
f0486c68 | 4779 | ret = 0; |
13553e52 | 4780 | else |
f0486c68 | 4781 | num_bytes -= block_rsv->reserved; |
f0486c68 | 4782 | spin_unlock(&block_rsv->lock); |
13553e52 | 4783 | |
f0486c68 YZ |
4784 | if (!ret) |
4785 | return 0; | |
4786 | ||
aa38a711 | 4787 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
dabdb640 JB |
4788 | if (!ret) { |
4789 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
f0486c68 | 4790 | return 0; |
6a63209f | 4791 | } |
9ed74f2d | 4792 | |
13553e52 | 4793 | return ret; |
f0486c68 YZ |
4794 | } |
4795 | ||
4796 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv, | |
4797 | struct btrfs_block_rsv *dst_rsv, | |
4798 | u64 num_bytes) | |
4799 | { | |
4800 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
4801 | } | |
4802 | ||
4803 | void btrfs_block_rsv_release(struct btrfs_root *root, | |
4804 | struct btrfs_block_rsv *block_rsv, | |
4805 | u64 num_bytes) | |
4806 | { | |
4807 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; | |
17504584 | 4808 | if (global_rsv == block_rsv || |
f0486c68 YZ |
4809 | block_rsv->space_info != global_rsv->space_info) |
4810 | global_rsv = NULL; | |
8c2a3ca2 JB |
4811 | block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv, |
4812 | num_bytes); | |
6a63209f JB |
4813 | } |
4814 | ||
4815 | /* | |
8929ecfa YZ |
4816 | * helper to calculate size of global block reservation. |
4817 | * the desired value is sum of space used by extent tree, | |
4818 | * checksum tree and root tree | |
6a63209f | 4819 | */ |
8929ecfa | 4820 | static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info) |
6a63209f | 4821 | { |
8929ecfa YZ |
4822 | struct btrfs_space_info *sinfo; |
4823 | u64 num_bytes; | |
4824 | u64 meta_used; | |
4825 | u64 data_used; | |
6c41761f | 4826 | int csum_size = btrfs_super_csum_size(fs_info->super_copy); |
6a63209f | 4827 | |
8929ecfa YZ |
4828 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA); |
4829 | spin_lock(&sinfo->lock); | |
4830 | data_used = sinfo->bytes_used; | |
4831 | spin_unlock(&sinfo->lock); | |
33b4d47f | 4832 | |
8929ecfa YZ |
4833 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
4834 | spin_lock(&sinfo->lock); | |
6d48755d JB |
4835 | if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) |
4836 | data_used = 0; | |
8929ecfa YZ |
4837 | meta_used = sinfo->bytes_used; |
4838 | spin_unlock(&sinfo->lock); | |
ab6e2410 | 4839 | |
8929ecfa YZ |
4840 | num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) * |
4841 | csum_size * 2; | |
4842 | num_bytes += div64_u64(data_used + meta_used, 50); | |
4e06bdd6 | 4843 | |
8929ecfa | 4844 | if (num_bytes * 3 > meta_used) |
8e62c2de | 4845 | num_bytes = div64_u64(meta_used, 3); |
ab6e2410 | 4846 | |
8929ecfa YZ |
4847 | return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10); |
4848 | } | |
6a63209f | 4849 | |
8929ecfa YZ |
4850 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
4851 | { | |
4852 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
4853 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
4854 | u64 num_bytes; | |
6a63209f | 4855 | |
8929ecfa | 4856 | num_bytes = calc_global_metadata_size(fs_info); |
33b4d47f | 4857 | |
8929ecfa | 4858 | spin_lock(&sinfo->lock); |
1f699d38 | 4859 | spin_lock(&block_rsv->lock); |
4e06bdd6 | 4860 | |
fdf30d1c | 4861 | block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024); |
4e06bdd6 | 4862 | |
8929ecfa | 4863 | num_bytes = sinfo->bytes_used + sinfo->bytes_pinned + |
6d48755d JB |
4864 | sinfo->bytes_reserved + sinfo->bytes_readonly + |
4865 | sinfo->bytes_may_use; | |
8929ecfa YZ |
4866 | |
4867 | if (sinfo->total_bytes > num_bytes) { | |
4868 | num_bytes = sinfo->total_bytes - num_bytes; | |
4869 | block_rsv->reserved += num_bytes; | |
fb25e914 | 4870 | sinfo->bytes_may_use += num_bytes; |
8c2a3ca2 | 4871 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4872 | sinfo->flags, num_bytes, 1); |
6a63209f | 4873 | } |
6a63209f | 4874 | |
8929ecfa YZ |
4875 | if (block_rsv->reserved >= block_rsv->size) { |
4876 | num_bytes = block_rsv->reserved - block_rsv->size; | |
fb25e914 | 4877 | sinfo->bytes_may_use -= num_bytes; |
8c2a3ca2 | 4878 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4879 | sinfo->flags, num_bytes, 0); |
8929ecfa YZ |
4880 | block_rsv->reserved = block_rsv->size; |
4881 | block_rsv->full = 1; | |
4882 | } | |
182608c8 | 4883 | |
8929ecfa | 4884 | spin_unlock(&block_rsv->lock); |
1f699d38 | 4885 | spin_unlock(&sinfo->lock); |
6a63209f JB |
4886 | } |
4887 | ||
f0486c68 | 4888 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 4889 | { |
f0486c68 | 4890 | struct btrfs_space_info *space_info; |
6a63209f | 4891 | |
f0486c68 YZ |
4892 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
4893 | fs_info->chunk_block_rsv.space_info = space_info; | |
6a63209f | 4894 | |
f0486c68 | 4895 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa | 4896 | fs_info->global_block_rsv.space_info = space_info; |
8929ecfa | 4897 | fs_info->delalloc_block_rsv.space_info = space_info; |
f0486c68 YZ |
4898 | fs_info->trans_block_rsv.space_info = space_info; |
4899 | fs_info->empty_block_rsv.space_info = space_info; | |
6d668dda | 4900 | fs_info->delayed_block_rsv.space_info = space_info; |
f0486c68 | 4901 | |
8929ecfa YZ |
4902 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
4903 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
4904 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
4905 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
3a6cad90 SB |
4906 | if (fs_info->quota_root) |
4907 | fs_info->quota_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 4908 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa | 4909 | |
8929ecfa | 4910 | update_global_block_rsv(fs_info); |
6a63209f JB |
4911 | } |
4912 | ||
8929ecfa | 4913 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 4914 | { |
8c2a3ca2 JB |
4915 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
4916 | (u64)-1); | |
8929ecfa YZ |
4917 | WARN_ON(fs_info->delalloc_block_rsv.size > 0); |
4918 | WARN_ON(fs_info->delalloc_block_rsv.reserved > 0); | |
4919 | WARN_ON(fs_info->trans_block_rsv.size > 0); | |
4920 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
4921 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
4922 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
6d668dda JB |
4923 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
4924 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
fcb80c2a JB |
4925 | } |
4926 | ||
a22285a6 YZ |
4927 | void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans, |
4928 | struct btrfs_root *root) | |
6a63209f | 4929 | { |
0e721106 JB |
4930 | if (!trans->block_rsv) |
4931 | return; | |
4932 | ||
a22285a6 YZ |
4933 | if (!trans->bytes_reserved) |
4934 | return; | |
6a63209f | 4935 | |
e77266e4 | 4936 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
2bcc0328 | 4937 | trans->transid, trans->bytes_reserved, 0); |
b24e03db | 4938 | btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved); |
a22285a6 YZ |
4939 | trans->bytes_reserved = 0; |
4940 | } | |
6a63209f | 4941 | |
79787eaa | 4942 | /* Can only return 0 or -ENOSPC */ |
d68fc57b YZ |
4943 | int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans, |
4944 | struct inode *inode) | |
4945 | { | |
4946 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4947 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
4948 | struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv; | |
4949 | ||
4950 | /* | |
fcb80c2a JB |
4951 | * We need to hold space in order to delete our orphan item once we've |
4952 | * added it, so this takes the reservation so we can release it later | |
4953 | * when we are truly done with the orphan item. | |
d68fc57b | 4954 | */ |
ff5714cc | 4955 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
4956 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
4957 | btrfs_ino(inode), num_bytes, 1); | |
d68fc57b | 4958 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); |
6a63209f JB |
4959 | } |
4960 | ||
d68fc57b | 4961 | void btrfs_orphan_release_metadata(struct inode *inode) |
97e728d4 | 4962 | { |
d68fc57b | 4963 | struct btrfs_root *root = BTRFS_I(inode)->root; |
ff5714cc | 4964 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
4965 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
4966 | btrfs_ino(inode), num_bytes, 0); | |
d68fc57b YZ |
4967 | btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes); |
4968 | } | |
97e728d4 | 4969 | |
d5c12070 MX |
4970 | /* |
4971 | * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation | |
4972 | * root: the root of the parent directory | |
4973 | * rsv: block reservation | |
4974 | * items: the number of items that we need do reservation | |
4975 | * qgroup_reserved: used to return the reserved size in qgroup | |
4976 | * | |
4977 | * This function is used to reserve the space for snapshot/subvolume | |
4978 | * creation and deletion. Those operations are different with the | |
4979 | * common file/directory operations, they change two fs/file trees | |
4980 | * and root tree, the number of items that the qgroup reserves is | |
4981 | * different with the free space reservation. So we can not use | |
4982 | * the space reseravtion mechanism in start_transaction(). | |
4983 | */ | |
4984 | int btrfs_subvolume_reserve_metadata(struct btrfs_root *root, | |
4985 | struct btrfs_block_rsv *rsv, | |
4986 | int items, | |
ee3441b4 JM |
4987 | u64 *qgroup_reserved, |
4988 | bool use_global_rsv) | |
a22285a6 | 4989 | { |
d5c12070 MX |
4990 | u64 num_bytes; |
4991 | int ret; | |
ee3441b4 | 4992 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
d5c12070 MX |
4993 | |
4994 | if (root->fs_info->quota_enabled) { | |
4995 | /* One for parent inode, two for dir entries */ | |
4996 | num_bytes = 3 * root->leafsize; | |
4997 | ret = btrfs_qgroup_reserve(root, num_bytes); | |
4998 | if (ret) | |
4999 | return ret; | |
5000 | } else { | |
5001 | num_bytes = 0; | |
5002 | } | |
5003 | ||
5004 | *qgroup_reserved = num_bytes; | |
5005 | ||
5006 | num_bytes = btrfs_calc_trans_metadata_size(root, items); | |
5007 | rsv->space_info = __find_space_info(root->fs_info, | |
5008 | BTRFS_BLOCK_GROUP_METADATA); | |
5009 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, | |
5010 | BTRFS_RESERVE_FLUSH_ALL); | |
ee3441b4 JM |
5011 | |
5012 | if (ret == -ENOSPC && use_global_rsv) | |
5013 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes); | |
5014 | ||
d5c12070 MX |
5015 | if (ret) { |
5016 | if (*qgroup_reserved) | |
5017 | btrfs_qgroup_free(root, *qgroup_reserved); | |
5018 | } | |
5019 | ||
5020 | return ret; | |
5021 | } | |
5022 | ||
5023 | void btrfs_subvolume_release_metadata(struct btrfs_root *root, | |
5024 | struct btrfs_block_rsv *rsv, | |
5025 | u64 qgroup_reserved) | |
5026 | { | |
5027 | btrfs_block_rsv_release(root, rsv, (u64)-1); | |
5028 | if (qgroup_reserved) | |
5029 | btrfs_qgroup_free(root, qgroup_reserved); | |
97e728d4 JB |
5030 | } |
5031 | ||
7709cde3 JB |
5032 | /** |
5033 | * drop_outstanding_extent - drop an outstanding extent | |
5034 | * @inode: the inode we're dropping the extent for | |
5035 | * | |
5036 | * This is called when we are freeing up an outstanding extent, either called | |
5037 | * after an error or after an extent is written. This will return the number of | |
5038 | * reserved extents that need to be freed. This must be called with | |
5039 | * BTRFS_I(inode)->lock held. | |
5040 | */ | |
9e0baf60 JB |
5041 | static unsigned drop_outstanding_extent(struct inode *inode) |
5042 | { | |
7fd2ae21 | 5043 | unsigned drop_inode_space = 0; |
9e0baf60 JB |
5044 | unsigned dropped_extents = 0; |
5045 | ||
9e0baf60 JB |
5046 | BUG_ON(!BTRFS_I(inode)->outstanding_extents); |
5047 | BTRFS_I(inode)->outstanding_extents--; | |
5048 | ||
7fd2ae21 | 5049 | if (BTRFS_I(inode)->outstanding_extents == 0 && |
72ac3c0d JB |
5050 | test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
5051 | &BTRFS_I(inode)->runtime_flags)) | |
7fd2ae21 | 5052 | drop_inode_space = 1; |
7fd2ae21 | 5053 | |
9e0baf60 JB |
5054 | /* |
5055 | * If we have more or the same amount of outsanding extents than we have | |
5056 | * reserved then we need to leave the reserved extents count alone. | |
5057 | */ | |
5058 | if (BTRFS_I(inode)->outstanding_extents >= | |
5059 | BTRFS_I(inode)->reserved_extents) | |
7fd2ae21 | 5060 | return drop_inode_space; |
9e0baf60 JB |
5061 | |
5062 | dropped_extents = BTRFS_I(inode)->reserved_extents - | |
5063 | BTRFS_I(inode)->outstanding_extents; | |
5064 | BTRFS_I(inode)->reserved_extents -= dropped_extents; | |
7fd2ae21 | 5065 | return dropped_extents + drop_inode_space; |
9e0baf60 JB |
5066 | } |
5067 | ||
7709cde3 JB |
5068 | /** |
5069 | * calc_csum_metadata_size - return the amount of metada space that must be | |
5070 | * reserved/free'd for the given bytes. | |
5071 | * @inode: the inode we're manipulating | |
5072 | * @num_bytes: the number of bytes in question | |
5073 | * @reserve: 1 if we are reserving space, 0 if we are freeing space | |
5074 | * | |
5075 | * This adjusts the number of csum_bytes in the inode and then returns the | |
5076 | * correct amount of metadata that must either be reserved or freed. We | |
5077 | * calculate how many checksums we can fit into one leaf and then divide the | |
5078 | * number of bytes that will need to be checksumed by this value to figure out | |
5079 | * how many checksums will be required. If we are adding bytes then the number | |
5080 | * may go up and we will return the number of additional bytes that must be | |
5081 | * reserved. If it is going down we will return the number of bytes that must | |
5082 | * be freed. | |
5083 | * | |
5084 | * This must be called with BTRFS_I(inode)->lock held. | |
5085 | */ | |
5086 | static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes, | |
5087 | int reserve) | |
6324fbf3 | 5088 | { |
7709cde3 JB |
5089 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5090 | u64 csum_size; | |
5091 | int num_csums_per_leaf; | |
5092 | int num_csums; | |
5093 | int old_csums; | |
5094 | ||
5095 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM && | |
5096 | BTRFS_I(inode)->csum_bytes == 0) | |
5097 | return 0; | |
5098 | ||
5099 | old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize); | |
5100 | if (reserve) | |
5101 | BTRFS_I(inode)->csum_bytes += num_bytes; | |
5102 | else | |
5103 | BTRFS_I(inode)->csum_bytes -= num_bytes; | |
5104 | csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item); | |
5105 | num_csums_per_leaf = (int)div64_u64(csum_size, | |
5106 | sizeof(struct btrfs_csum_item) + | |
5107 | sizeof(struct btrfs_disk_key)); | |
5108 | num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize); | |
5109 | num_csums = num_csums + num_csums_per_leaf - 1; | |
5110 | num_csums = num_csums / num_csums_per_leaf; | |
5111 | ||
5112 | old_csums = old_csums + num_csums_per_leaf - 1; | |
5113 | old_csums = old_csums / num_csums_per_leaf; | |
5114 | ||
5115 | /* No change, no need to reserve more */ | |
5116 | if (old_csums == num_csums) | |
5117 | return 0; | |
5118 | ||
5119 | if (reserve) | |
5120 | return btrfs_calc_trans_metadata_size(root, | |
5121 | num_csums - old_csums); | |
5122 | ||
5123 | return btrfs_calc_trans_metadata_size(root, old_csums - num_csums); | |
0ca1f7ce | 5124 | } |
c146afad | 5125 | |
0ca1f7ce YZ |
5126 | int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes) |
5127 | { | |
5128 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5129 | struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv; | |
9e0baf60 | 5130 | u64 to_reserve = 0; |
660d3f6c | 5131 | u64 csum_bytes; |
9e0baf60 | 5132 | unsigned nr_extents = 0; |
660d3f6c | 5133 | int extra_reserve = 0; |
08e007d2 | 5134 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL; |
eb6b88d9 | 5135 | int ret = 0; |
c64c2bd8 | 5136 | bool delalloc_lock = true; |
88e081bf WS |
5137 | u64 to_free = 0; |
5138 | unsigned dropped; | |
6324fbf3 | 5139 | |
c64c2bd8 JB |
5140 | /* If we are a free space inode we need to not flush since we will be in |
5141 | * the middle of a transaction commit. We also don't need the delalloc | |
5142 | * mutex since we won't race with anybody. We need this mostly to make | |
5143 | * lockdep shut its filthy mouth. | |
5144 | */ | |
5145 | if (btrfs_is_free_space_inode(inode)) { | |
08e007d2 | 5146 | flush = BTRFS_RESERVE_NO_FLUSH; |
c64c2bd8 JB |
5147 | delalloc_lock = false; |
5148 | } | |
c09544e0 | 5149 | |
08e007d2 MX |
5150 | if (flush != BTRFS_RESERVE_NO_FLUSH && |
5151 | btrfs_transaction_in_commit(root->fs_info)) | |
0ca1f7ce | 5152 | schedule_timeout(1); |
ec44a35c | 5153 | |
c64c2bd8 JB |
5154 | if (delalloc_lock) |
5155 | mutex_lock(&BTRFS_I(inode)->delalloc_mutex); | |
5156 | ||
0ca1f7ce | 5157 | num_bytes = ALIGN(num_bytes, root->sectorsize); |
8bb8ab2e | 5158 | |
9e0baf60 JB |
5159 | spin_lock(&BTRFS_I(inode)->lock); |
5160 | BTRFS_I(inode)->outstanding_extents++; | |
5161 | ||
5162 | if (BTRFS_I(inode)->outstanding_extents > | |
660d3f6c | 5163 | BTRFS_I(inode)->reserved_extents) |
9e0baf60 JB |
5164 | nr_extents = BTRFS_I(inode)->outstanding_extents - |
5165 | BTRFS_I(inode)->reserved_extents; | |
57a45ced | 5166 | |
7fd2ae21 JB |
5167 | /* |
5168 | * Add an item to reserve for updating the inode when we complete the | |
5169 | * delalloc io. | |
5170 | */ | |
72ac3c0d JB |
5171 | if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
5172 | &BTRFS_I(inode)->runtime_flags)) { | |
7fd2ae21 | 5173 | nr_extents++; |
660d3f6c | 5174 | extra_reserve = 1; |
593060d7 | 5175 | } |
7fd2ae21 JB |
5176 | |
5177 | to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents); | |
7709cde3 | 5178 | to_reserve += calc_csum_metadata_size(inode, num_bytes, 1); |
660d3f6c | 5179 | csum_bytes = BTRFS_I(inode)->csum_bytes; |
9e0baf60 | 5180 | spin_unlock(&BTRFS_I(inode)->lock); |
57a45ced | 5181 | |
88e081bf | 5182 | if (root->fs_info->quota_enabled) { |
c5567237 AJ |
5183 | ret = btrfs_qgroup_reserve(root, num_bytes + |
5184 | nr_extents * root->leafsize); | |
88e081bf WS |
5185 | if (ret) |
5186 | goto out_fail; | |
5187 | } | |
c5567237 | 5188 | |
88e081bf WS |
5189 | ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush); |
5190 | if (unlikely(ret)) { | |
5191 | if (root->fs_info->quota_enabled) | |
4b5829a8 MX |
5192 | btrfs_qgroup_free(root, num_bytes + |
5193 | nr_extents * root->leafsize); | |
88e081bf | 5194 | goto out_fail; |
9e0baf60 | 5195 | } |
25179201 | 5196 | |
660d3f6c JB |
5197 | spin_lock(&BTRFS_I(inode)->lock); |
5198 | if (extra_reserve) { | |
72ac3c0d JB |
5199 | set_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
5200 | &BTRFS_I(inode)->runtime_flags); | |
660d3f6c JB |
5201 | nr_extents--; |
5202 | } | |
5203 | BTRFS_I(inode)->reserved_extents += nr_extents; | |
5204 | spin_unlock(&BTRFS_I(inode)->lock); | |
c64c2bd8 JB |
5205 | |
5206 | if (delalloc_lock) | |
5207 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); | |
660d3f6c | 5208 | |
8c2a3ca2 | 5209 | if (to_reserve) |
67871254 | 5210 | trace_btrfs_space_reservation(root->fs_info, "delalloc", |
8c2a3ca2 | 5211 | btrfs_ino(inode), to_reserve, 1); |
0ca1f7ce YZ |
5212 | block_rsv_add_bytes(block_rsv, to_reserve, 1); |
5213 | ||
0ca1f7ce | 5214 | return 0; |
88e081bf WS |
5215 | |
5216 | out_fail: | |
5217 | spin_lock(&BTRFS_I(inode)->lock); | |
5218 | dropped = drop_outstanding_extent(inode); | |
5219 | /* | |
5220 | * If the inodes csum_bytes is the same as the original | |
5221 | * csum_bytes then we know we haven't raced with any free()ers | |
5222 | * so we can just reduce our inodes csum bytes and carry on. | |
88e081bf | 5223 | */ |
f4881bc7 | 5224 | if (BTRFS_I(inode)->csum_bytes == csum_bytes) { |
88e081bf | 5225 | calc_csum_metadata_size(inode, num_bytes, 0); |
f4881bc7 JB |
5226 | } else { |
5227 | u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes; | |
5228 | u64 bytes; | |
5229 | ||
5230 | /* | |
5231 | * This is tricky, but first we need to figure out how much we | |
5232 | * free'd from any free-ers that occured during this | |
5233 | * reservation, so we reset ->csum_bytes to the csum_bytes | |
5234 | * before we dropped our lock, and then call the free for the | |
5235 | * number of bytes that were freed while we were trying our | |
5236 | * reservation. | |
5237 | */ | |
5238 | bytes = csum_bytes - BTRFS_I(inode)->csum_bytes; | |
5239 | BTRFS_I(inode)->csum_bytes = csum_bytes; | |
5240 | to_free = calc_csum_metadata_size(inode, bytes, 0); | |
5241 | ||
5242 | ||
5243 | /* | |
5244 | * Now we need to see how much we would have freed had we not | |
5245 | * been making this reservation and our ->csum_bytes were not | |
5246 | * artificially inflated. | |
5247 | */ | |
5248 | BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes; | |
5249 | bytes = csum_bytes - orig_csum_bytes; | |
5250 | bytes = calc_csum_metadata_size(inode, bytes, 0); | |
5251 | ||
5252 | /* | |
5253 | * Now reset ->csum_bytes to what it should be. If bytes is | |
5254 | * more than to_free then we would have free'd more space had we | |
5255 | * not had an artificially high ->csum_bytes, so we need to free | |
5256 | * the remainder. If bytes is the same or less then we don't | |
5257 | * need to do anything, the other free-ers did the correct | |
5258 | * thing. | |
5259 | */ | |
5260 | BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes; | |
5261 | if (bytes > to_free) | |
5262 | to_free = bytes - to_free; | |
5263 | else | |
5264 | to_free = 0; | |
5265 | } | |
88e081bf WS |
5266 | spin_unlock(&BTRFS_I(inode)->lock); |
5267 | if (dropped) | |
5268 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
5269 | ||
5270 | if (to_free) { | |
5271 | btrfs_block_rsv_release(root, block_rsv, to_free); | |
5272 | trace_btrfs_space_reservation(root->fs_info, "delalloc", | |
5273 | btrfs_ino(inode), to_free, 0); | |
5274 | } | |
5275 | if (delalloc_lock) | |
5276 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); | |
5277 | return ret; | |
0ca1f7ce YZ |
5278 | } |
5279 | ||
7709cde3 JB |
5280 | /** |
5281 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
5282 | * @inode: the inode to release the reservation for | |
5283 | * @num_bytes: the number of bytes we're releasing | |
5284 | * | |
5285 | * This will release the metadata reservation for an inode. This can be called | |
5286 | * once we complete IO for a given set of bytes to release their metadata | |
5287 | * reservations. | |
5288 | */ | |
0ca1f7ce YZ |
5289 | void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes) |
5290 | { | |
5291 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
9e0baf60 JB |
5292 | u64 to_free = 0; |
5293 | unsigned dropped; | |
0ca1f7ce YZ |
5294 | |
5295 | num_bytes = ALIGN(num_bytes, root->sectorsize); | |
7709cde3 | 5296 | spin_lock(&BTRFS_I(inode)->lock); |
9e0baf60 | 5297 | dropped = drop_outstanding_extent(inode); |
97e728d4 | 5298 | |
0934856d MX |
5299 | if (num_bytes) |
5300 | to_free = calc_csum_metadata_size(inode, num_bytes, 0); | |
7709cde3 | 5301 | spin_unlock(&BTRFS_I(inode)->lock); |
9e0baf60 JB |
5302 | if (dropped > 0) |
5303 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
0ca1f7ce | 5304 | |
8c2a3ca2 JB |
5305 | trace_btrfs_space_reservation(root->fs_info, "delalloc", |
5306 | btrfs_ino(inode), to_free, 0); | |
c5567237 AJ |
5307 | if (root->fs_info->quota_enabled) { |
5308 | btrfs_qgroup_free(root, num_bytes + | |
5309 | dropped * root->leafsize); | |
5310 | } | |
5311 | ||
0ca1f7ce YZ |
5312 | btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv, |
5313 | to_free); | |
5314 | } | |
5315 | ||
7709cde3 JB |
5316 | /** |
5317 | * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc | |
5318 | * @inode: inode we're writing to | |
5319 | * @num_bytes: the number of bytes we want to allocate | |
5320 | * | |
5321 | * This will do the following things | |
5322 | * | |
5323 | * o reserve space in the data space info for num_bytes | |
5324 | * o reserve space in the metadata space info based on number of outstanding | |
5325 | * extents and how much csums will be needed | |
5326 | * o add to the inodes ->delalloc_bytes | |
5327 | * o add it to the fs_info's delalloc inodes list. | |
5328 | * | |
5329 | * This will return 0 for success and -ENOSPC if there is no space left. | |
5330 | */ | |
0ca1f7ce YZ |
5331 | int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes) |
5332 | { | |
5333 | int ret; | |
5334 | ||
5335 | ret = btrfs_check_data_free_space(inode, num_bytes); | |
d397712b | 5336 | if (ret) |
0ca1f7ce YZ |
5337 | return ret; |
5338 | ||
5339 | ret = btrfs_delalloc_reserve_metadata(inode, num_bytes); | |
5340 | if (ret) { | |
5341 | btrfs_free_reserved_data_space(inode, num_bytes); | |
5342 | return ret; | |
5343 | } | |
5344 | ||
5345 | return 0; | |
5346 | } | |
5347 | ||
7709cde3 JB |
5348 | /** |
5349 | * btrfs_delalloc_release_space - release data and metadata space for delalloc | |
5350 | * @inode: inode we're releasing space for | |
5351 | * @num_bytes: the number of bytes we want to free up | |
5352 | * | |
5353 | * This must be matched with a call to btrfs_delalloc_reserve_space. This is | |
5354 | * called in the case that we don't need the metadata AND data reservations | |
5355 | * anymore. So if there is an error or we insert an inline extent. | |
5356 | * | |
5357 | * This function will release the metadata space that was not used and will | |
5358 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
5359 | * list if there are no delalloc bytes left. | |
5360 | */ | |
0ca1f7ce YZ |
5361 | void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes) |
5362 | { | |
5363 | btrfs_delalloc_release_metadata(inode, num_bytes); | |
5364 | btrfs_free_reserved_data_space(inode, num_bytes); | |
6324fbf3 CM |
5365 | } |
5366 | ||
c53d613e | 5367 | static int update_block_group(struct btrfs_root *root, |
f0486c68 | 5368 | u64 bytenr, u64 num_bytes, int alloc) |
9078a3e1 | 5369 | { |
0af3d00b | 5370 | struct btrfs_block_group_cache *cache = NULL; |
9078a3e1 | 5371 | struct btrfs_fs_info *info = root->fs_info; |
db94535d | 5372 | u64 total = num_bytes; |
9078a3e1 | 5373 | u64 old_val; |
db94535d | 5374 | u64 byte_in_group; |
0af3d00b | 5375 | int factor; |
3e1ad54f | 5376 | |
5d4f98a2 | 5377 | /* block accounting for super block */ |
eb73c1b7 | 5378 | spin_lock(&info->delalloc_root_lock); |
6c41761f | 5379 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
5380 | if (alloc) |
5381 | old_val += num_bytes; | |
5382 | else | |
5383 | old_val -= num_bytes; | |
6c41761f | 5384 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
eb73c1b7 | 5385 | spin_unlock(&info->delalloc_root_lock); |
5d4f98a2 | 5386 | |
d397712b | 5387 | while (total) { |
db94535d | 5388 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 5389 | if (!cache) |
79787eaa | 5390 | return -ENOENT; |
b742bb82 YZ |
5391 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | |
5392 | BTRFS_BLOCK_GROUP_RAID1 | | |
5393 | BTRFS_BLOCK_GROUP_RAID10)) | |
5394 | factor = 2; | |
5395 | else | |
5396 | factor = 1; | |
9d66e233 JB |
5397 | /* |
5398 | * If this block group has free space cache written out, we | |
5399 | * need to make sure to load it if we are removing space. This | |
5400 | * is because we need the unpinning stage to actually add the | |
5401 | * space back to the block group, otherwise we will leak space. | |
5402 | */ | |
5403 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
f6373bf3 | 5404 | cache_block_group(cache, 1); |
0af3d00b | 5405 | |
db94535d CM |
5406 | byte_in_group = bytenr - cache->key.objectid; |
5407 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 5408 | |
25179201 | 5409 | spin_lock(&cache->space_info->lock); |
c286ac48 | 5410 | spin_lock(&cache->lock); |
0af3d00b | 5411 | |
73bc1876 | 5412 | if (btrfs_test_opt(root, SPACE_CACHE) && |
0af3d00b JB |
5413 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
5414 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
5415 | ||
0f9dd46c | 5416 | cache->dirty = 1; |
9078a3e1 | 5417 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 5418 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 5419 | if (alloc) { |
db94535d | 5420 | old_val += num_bytes; |
11833d66 YZ |
5421 | btrfs_set_block_group_used(&cache->item, old_val); |
5422 | cache->reserved -= num_bytes; | |
11833d66 | 5423 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
5424 | cache->space_info->bytes_used += num_bytes; |
5425 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 5426 | spin_unlock(&cache->lock); |
25179201 | 5427 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 5428 | } else { |
db94535d | 5429 | old_val -= num_bytes; |
c286ac48 | 5430 | btrfs_set_block_group_used(&cache->item, old_val); |
f0486c68 YZ |
5431 | cache->pinned += num_bytes; |
5432 | cache->space_info->bytes_pinned += num_bytes; | |
6324fbf3 | 5433 | cache->space_info->bytes_used -= num_bytes; |
b742bb82 | 5434 | cache->space_info->disk_used -= num_bytes * factor; |
c286ac48 | 5435 | spin_unlock(&cache->lock); |
25179201 | 5436 | spin_unlock(&cache->space_info->lock); |
1f3c79a2 | 5437 | |
f0486c68 YZ |
5438 | set_extent_dirty(info->pinned_extents, |
5439 | bytenr, bytenr + num_bytes - 1, | |
5440 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 5441 | } |
fa9c0d79 | 5442 | btrfs_put_block_group(cache); |
db94535d CM |
5443 | total -= num_bytes; |
5444 | bytenr += num_bytes; | |
9078a3e1 CM |
5445 | } |
5446 | return 0; | |
5447 | } | |
6324fbf3 | 5448 | |
a061fc8d CM |
5449 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) |
5450 | { | |
0f9dd46c | 5451 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 5452 | u64 bytenr; |
0f9dd46c | 5453 | |
a1897fdd LB |
5454 | spin_lock(&root->fs_info->block_group_cache_lock); |
5455 | bytenr = root->fs_info->first_logical_byte; | |
5456 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
5457 | ||
5458 | if (bytenr < (u64)-1) | |
5459 | return bytenr; | |
5460 | ||
0f9dd46c JB |
5461 | cache = btrfs_lookup_first_block_group(root->fs_info, search_start); |
5462 | if (!cache) | |
a061fc8d | 5463 | return 0; |
0f9dd46c | 5464 | |
d2fb3437 | 5465 | bytenr = cache->key.objectid; |
fa9c0d79 | 5466 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
5467 | |
5468 | return bytenr; | |
a061fc8d CM |
5469 | } |
5470 | ||
f0486c68 YZ |
5471 | static int pin_down_extent(struct btrfs_root *root, |
5472 | struct btrfs_block_group_cache *cache, | |
5473 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 5474 | { |
11833d66 YZ |
5475 | spin_lock(&cache->space_info->lock); |
5476 | spin_lock(&cache->lock); | |
5477 | cache->pinned += num_bytes; | |
5478 | cache->space_info->bytes_pinned += num_bytes; | |
5479 | if (reserved) { | |
5480 | cache->reserved -= num_bytes; | |
5481 | cache->space_info->bytes_reserved -= num_bytes; | |
5482 | } | |
5483 | spin_unlock(&cache->lock); | |
5484 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 5485 | |
f0486c68 YZ |
5486 | set_extent_dirty(root->fs_info->pinned_extents, bytenr, |
5487 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); | |
0be5dc67 JB |
5488 | if (reserved) |
5489 | trace_btrfs_reserved_extent_free(root, bytenr, num_bytes); | |
f0486c68 YZ |
5490 | return 0; |
5491 | } | |
68b38550 | 5492 | |
f0486c68 YZ |
5493 | /* |
5494 | * this function must be called within transaction | |
5495 | */ | |
5496 | int btrfs_pin_extent(struct btrfs_root *root, | |
5497 | u64 bytenr, u64 num_bytes, int reserved) | |
5498 | { | |
5499 | struct btrfs_block_group_cache *cache; | |
68b38550 | 5500 | |
f0486c68 | 5501 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); |
79787eaa | 5502 | BUG_ON(!cache); /* Logic error */ |
f0486c68 YZ |
5503 | |
5504 | pin_down_extent(root, cache, bytenr, num_bytes, reserved); | |
5505 | ||
5506 | btrfs_put_block_group(cache); | |
11833d66 YZ |
5507 | return 0; |
5508 | } | |
5509 | ||
f0486c68 | 5510 | /* |
e688b725 CM |
5511 | * this function must be called within transaction |
5512 | */ | |
dcfac415 | 5513 | int btrfs_pin_extent_for_log_replay(struct btrfs_root *root, |
e688b725 CM |
5514 | u64 bytenr, u64 num_bytes) |
5515 | { | |
5516 | struct btrfs_block_group_cache *cache; | |
b50c6e25 | 5517 | int ret; |
e688b725 CM |
5518 | |
5519 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); | |
b50c6e25 JB |
5520 | if (!cache) |
5521 | return -EINVAL; | |
e688b725 CM |
5522 | |
5523 | /* | |
5524 | * pull in the free space cache (if any) so that our pin | |
5525 | * removes the free space from the cache. We have load_only set | |
5526 | * to one because the slow code to read in the free extents does check | |
5527 | * the pinned extents. | |
5528 | */ | |
f6373bf3 | 5529 | cache_block_group(cache, 1); |
e688b725 CM |
5530 | |
5531 | pin_down_extent(root, cache, bytenr, num_bytes, 0); | |
5532 | ||
5533 | /* remove us from the free space cache (if we're there at all) */ | |
b50c6e25 | 5534 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
e688b725 | 5535 | btrfs_put_block_group(cache); |
b50c6e25 | 5536 | return ret; |
e688b725 CM |
5537 | } |
5538 | ||
8c2a1a30 JB |
5539 | static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes) |
5540 | { | |
5541 | int ret; | |
5542 | struct btrfs_block_group_cache *block_group; | |
5543 | struct btrfs_caching_control *caching_ctl; | |
5544 | ||
5545 | block_group = btrfs_lookup_block_group(root->fs_info, start); | |
5546 | if (!block_group) | |
5547 | return -EINVAL; | |
5548 | ||
5549 | cache_block_group(block_group, 0); | |
5550 | caching_ctl = get_caching_control(block_group); | |
5551 | ||
5552 | if (!caching_ctl) { | |
5553 | /* Logic error */ | |
5554 | BUG_ON(!block_group_cache_done(block_group)); | |
5555 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
5556 | } else { | |
5557 | mutex_lock(&caching_ctl->mutex); | |
5558 | ||
5559 | if (start >= caching_ctl->progress) { | |
5560 | ret = add_excluded_extent(root, start, num_bytes); | |
5561 | } else if (start + num_bytes <= caching_ctl->progress) { | |
5562 | ret = btrfs_remove_free_space(block_group, | |
5563 | start, num_bytes); | |
5564 | } else { | |
5565 | num_bytes = caching_ctl->progress - start; | |
5566 | ret = btrfs_remove_free_space(block_group, | |
5567 | start, num_bytes); | |
5568 | if (ret) | |
5569 | goto out_lock; | |
5570 | ||
5571 | num_bytes = (start + num_bytes) - | |
5572 | caching_ctl->progress; | |
5573 | start = caching_ctl->progress; | |
5574 | ret = add_excluded_extent(root, start, num_bytes); | |
5575 | } | |
5576 | out_lock: | |
5577 | mutex_unlock(&caching_ctl->mutex); | |
5578 | put_caching_control(caching_ctl); | |
5579 | } | |
5580 | btrfs_put_block_group(block_group); | |
5581 | return ret; | |
5582 | } | |
5583 | ||
5584 | int btrfs_exclude_logged_extents(struct btrfs_root *log, | |
5585 | struct extent_buffer *eb) | |
5586 | { | |
5587 | struct btrfs_file_extent_item *item; | |
5588 | struct btrfs_key key; | |
5589 | int found_type; | |
5590 | int i; | |
5591 | ||
5592 | if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS)) | |
5593 | return 0; | |
5594 | ||
5595 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
5596 | btrfs_item_key_to_cpu(eb, &key, i); | |
5597 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
5598 | continue; | |
5599 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
5600 | found_type = btrfs_file_extent_type(eb, item); | |
5601 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
5602 | continue; | |
5603 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
5604 | continue; | |
5605 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
5606 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
5607 | __exclude_logged_extent(log, key.objectid, key.offset); | |
5608 | } | |
5609 | ||
5610 | return 0; | |
5611 | } | |
5612 | ||
fb25e914 JB |
5613 | /** |
5614 | * btrfs_update_reserved_bytes - update the block_group and space info counters | |
5615 | * @cache: The cache we are manipulating | |
5616 | * @num_bytes: The number of bytes in question | |
5617 | * @reserve: One of the reservation enums | |
e570fd27 | 5618 | * @delalloc: The blocks are allocated for the delalloc write |
fb25e914 JB |
5619 | * |
5620 | * This is called by the allocator when it reserves space, or by somebody who is | |
5621 | * freeing space that was never actually used on disk. For example if you | |
5622 | * reserve some space for a new leaf in transaction A and before transaction A | |
5623 | * commits you free that leaf, you call this with reserve set to 0 in order to | |
5624 | * clear the reservation. | |
5625 | * | |
5626 | * Metadata reservations should be called with RESERVE_ALLOC so we do the proper | |
5627 | * ENOSPC accounting. For data we handle the reservation through clearing the | |
5628 | * delalloc bits in the io_tree. We have to do this since we could end up | |
5629 | * allocating less disk space for the amount of data we have reserved in the | |
5630 | * case of compression. | |
5631 | * | |
5632 | * If this is a reservation and the block group has become read only we cannot | |
5633 | * make the reservation and return -EAGAIN, otherwise this function always | |
5634 | * succeeds. | |
f0486c68 | 5635 | */ |
fb25e914 | 5636 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
e570fd27 | 5637 | u64 num_bytes, int reserve, int delalloc) |
11833d66 | 5638 | { |
fb25e914 | 5639 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 5640 | int ret = 0; |
79787eaa | 5641 | |
fb25e914 JB |
5642 | spin_lock(&space_info->lock); |
5643 | spin_lock(&cache->lock); | |
5644 | if (reserve != RESERVE_FREE) { | |
f0486c68 YZ |
5645 | if (cache->ro) { |
5646 | ret = -EAGAIN; | |
5647 | } else { | |
fb25e914 JB |
5648 | cache->reserved += num_bytes; |
5649 | space_info->bytes_reserved += num_bytes; | |
5650 | if (reserve == RESERVE_ALLOC) { | |
8c2a3ca2 | 5651 | trace_btrfs_space_reservation(cache->fs_info, |
2bcc0328 LB |
5652 | "space_info", space_info->flags, |
5653 | num_bytes, 0); | |
fb25e914 JB |
5654 | space_info->bytes_may_use -= num_bytes; |
5655 | } | |
e570fd27 MX |
5656 | |
5657 | if (delalloc) | |
5658 | cache->delalloc_bytes += num_bytes; | |
f0486c68 | 5659 | } |
fb25e914 JB |
5660 | } else { |
5661 | if (cache->ro) | |
5662 | space_info->bytes_readonly += num_bytes; | |
5663 | cache->reserved -= num_bytes; | |
5664 | space_info->bytes_reserved -= num_bytes; | |
e570fd27 MX |
5665 | |
5666 | if (delalloc) | |
5667 | cache->delalloc_bytes -= num_bytes; | |
324ae4df | 5668 | } |
fb25e914 JB |
5669 | spin_unlock(&cache->lock); |
5670 | spin_unlock(&space_info->lock); | |
f0486c68 | 5671 | return ret; |
324ae4df | 5672 | } |
9078a3e1 | 5673 | |
143bede5 | 5674 | void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans, |
11833d66 | 5675 | struct btrfs_root *root) |
e8569813 | 5676 | { |
e8569813 | 5677 | struct btrfs_fs_info *fs_info = root->fs_info; |
11833d66 YZ |
5678 | struct btrfs_caching_control *next; |
5679 | struct btrfs_caching_control *caching_ctl; | |
5680 | struct btrfs_block_group_cache *cache; | |
e8569813 | 5681 | |
9e351cc8 | 5682 | down_write(&fs_info->commit_root_sem); |
25179201 | 5683 | |
11833d66 YZ |
5684 | list_for_each_entry_safe(caching_ctl, next, |
5685 | &fs_info->caching_block_groups, list) { | |
5686 | cache = caching_ctl->block_group; | |
5687 | if (block_group_cache_done(cache)) { | |
5688 | cache->last_byte_to_unpin = (u64)-1; | |
5689 | list_del_init(&caching_ctl->list); | |
5690 | put_caching_control(caching_ctl); | |
e8569813 | 5691 | } else { |
11833d66 | 5692 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 5693 | } |
e8569813 | 5694 | } |
11833d66 YZ |
5695 | |
5696 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
5697 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
5698 | else | |
5699 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
5700 | ||
9e351cc8 | 5701 | up_write(&fs_info->commit_root_sem); |
8929ecfa YZ |
5702 | |
5703 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
5704 | } |
5705 | ||
11833d66 | 5706 | static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
ccd467d6 | 5707 | { |
11833d66 YZ |
5708 | struct btrfs_fs_info *fs_info = root->fs_info; |
5709 | struct btrfs_block_group_cache *cache = NULL; | |
7b398f8e JB |
5710 | struct btrfs_space_info *space_info; |
5711 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
11833d66 | 5712 | u64 len; |
7b398f8e | 5713 | bool readonly; |
ccd467d6 | 5714 | |
11833d66 | 5715 | while (start <= end) { |
7b398f8e | 5716 | readonly = false; |
11833d66 YZ |
5717 | if (!cache || |
5718 | start >= cache->key.objectid + cache->key.offset) { | |
5719 | if (cache) | |
5720 | btrfs_put_block_group(cache); | |
5721 | cache = btrfs_lookup_block_group(fs_info, start); | |
79787eaa | 5722 | BUG_ON(!cache); /* Logic error */ |
11833d66 YZ |
5723 | } |
5724 | ||
5725 | len = cache->key.objectid + cache->key.offset - start; | |
5726 | len = min(len, end + 1 - start); | |
5727 | ||
5728 | if (start < cache->last_byte_to_unpin) { | |
5729 | len = min(len, cache->last_byte_to_unpin - start); | |
5730 | btrfs_add_free_space(cache, start, len); | |
5731 | } | |
5732 | ||
f0486c68 | 5733 | start += len; |
7b398f8e | 5734 | space_info = cache->space_info; |
f0486c68 | 5735 | |
7b398f8e | 5736 | spin_lock(&space_info->lock); |
11833d66 YZ |
5737 | spin_lock(&cache->lock); |
5738 | cache->pinned -= len; | |
7b398f8e | 5739 | space_info->bytes_pinned -= len; |
d288db5d | 5740 | percpu_counter_add(&space_info->total_bytes_pinned, -len); |
7b398f8e JB |
5741 | if (cache->ro) { |
5742 | space_info->bytes_readonly += len; | |
5743 | readonly = true; | |
5744 | } | |
11833d66 | 5745 | spin_unlock(&cache->lock); |
7b398f8e JB |
5746 | if (!readonly && global_rsv->space_info == space_info) { |
5747 | spin_lock(&global_rsv->lock); | |
5748 | if (!global_rsv->full) { | |
5749 | len = min(len, global_rsv->size - | |
5750 | global_rsv->reserved); | |
5751 | global_rsv->reserved += len; | |
5752 | space_info->bytes_may_use += len; | |
5753 | if (global_rsv->reserved >= global_rsv->size) | |
5754 | global_rsv->full = 1; | |
5755 | } | |
5756 | spin_unlock(&global_rsv->lock); | |
5757 | } | |
5758 | spin_unlock(&space_info->lock); | |
ccd467d6 | 5759 | } |
11833d66 YZ |
5760 | |
5761 | if (cache) | |
5762 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
5763 | return 0; |
5764 | } | |
5765 | ||
5766 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
11833d66 | 5767 | struct btrfs_root *root) |
a28ec197 | 5768 | { |
11833d66 YZ |
5769 | struct btrfs_fs_info *fs_info = root->fs_info; |
5770 | struct extent_io_tree *unpin; | |
1a5bc167 CM |
5771 | u64 start; |
5772 | u64 end; | |
a28ec197 | 5773 | int ret; |
a28ec197 | 5774 | |
79787eaa JM |
5775 | if (trans->aborted) |
5776 | return 0; | |
5777 | ||
11833d66 YZ |
5778 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
5779 | unpin = &fs_info->freed_extents[1]; | |
5780 | else | |
5781 | unpin = &fs_info->freed_extents[0]; | |
5782 | ||
d397712b | 5783 | while (1) { |
1a5bc167 | 5784 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
e6138876 | 5785 | EXTENT_DIRTY, NULL); |
1a5bc167 | 5786 | if (ret) |
a28ec197 | 5787 | break; |
1f3c79a2 | 5788 | |
5378e607 LD |
5789 | if (btrfs_test_opt(root, DISCARD)) |
5790 | ret = btrfs_discard_extent(root, start, | |
5791 | end + 1 - start, NULL); | |
1f3c79a2 | 5792 | |
1a5bc167 | 5793 | clear_extent_dirty(unpin, start, end, GFP_NOFS); |
11833d66 | 5794 | unpin_extent_range(root, start, end); |
b9473439 | 5795 | cond_resched(); |
a28ec197 | 5796 | } |
817d52f8 | 5797 | |
e20d96d6 CM |
5798 | return 0; |
5799 | } | |
5800 | ||
b150a4f1 JB |
5801 | static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes, |
5802 | u64 owner, u64 root_objectid) | |
5803 | { | |
5804 | struct btrfs_space_info *space_info; | |
5805 | u64 flags; | |
5806 | ||
5807 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
5808 | if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID) | |
5809 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
5810 | else | |
5811 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
5812 | } else { | |
5813 | flags = BTRFS_BLOCK_GROUP_DATA; | |
5814 | } | |
5815 | ||
5816 | space_info = __find_space_info(fs_info, flags); | |
5817 | BUG_ON(!space_info); /* Logic bug */ | |
5818 | percpu_counter_add(&space_info->total_bytes_pinned, num_bytes); | |
5819 | } | |
5820 | ||
5821 | ||
5d4f98a2 YZ |
5822 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
5823 | struct btrfs_root *root, | |
5824 | u64 bytenr, u64 num_bytes, u64 parent, | |
5825 | u64 root_objectid, u64 owner_objectid, | |
5826 | u64 owner_offset, int refs_to_drop, | |
fcebe456 JB |
5827 | struct btrfs_delayed_extent_op *extent_op, |
5828 | int no_quota) | |
a28ec197 | 5829 | { |
e2fa7227 | 5830 | struct btrfs_key key; |
5d4f98a2 | 5831 | struct btrfs_path *path; |
1261ec42 CM |
5832 | struct btrfs_fs_info *info = root->fs_info; |
5833 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 5834 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
5835 | struct btrfs_extent_item *ei; |
5836 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 5837 | int ret; |
5d4f98a2 | 5838 | int is_data; |
952fccac CM |
5839 | int extent_slot = 0; |
5840 | int found_extent = 0; | |
5841 | int num_to_del = 1; | |
5d4f98a2 YZ |
5842 | u32 item_size; |
5843 | u64 refs; | |
fcebe456 JB |
5844 | int last_ref = 0; |
5845 | enum btrfs_qgroup_operation_type type = BTRFS_QGROUP_OPER_SUB_EXCL; | |
3173a18f JB |
5846 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
5847 | SKINNY_METADATA); | |
037e6390 | 5848 | |
fcebe456 JB |
5849 | if (!info->quota_enabled || !is_fstree(root_objectid)) |
5850 | no_quota = 1; | |
5851 | ||
5caf2a00 | 5852 | path = btrfs_alloc_path(); |
54aa1f4d CM |
5853 | if (!path) |
5854 | return -ENOMEM; | |
5f26f772 | 5855 | |
3c12ac72 | 5856 | path->reada = 1; |
b9473439 | 5857 | path->leave_spinning = 1; |
5d4f98a2 YZ |
5858 | |
5859 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
5860 | BUG_ON(!is_data && refs_to_drop != 1); | |
5861 | ||
3173a18f JB |
5862 | if (is_data) |
5863 | skinny_metadata = 0; | |
5864 | ||
5d4f98a2 YZ |
5865 | ret = lookup_extent_backref(trans, extent_root, path, &iref, |
5866 | bytenr, num_bytes, parent, | |
5867 | root_objectid, owner_objectid, | |
5868 | owner_offset); | |
7bb86316 | 5869 | if (ret == 0) { |
952fccac | 5870 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
5871 | while (extent_slot >= 0) { |
5872 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 5873 | extent_slot); |
5d4f98a2 | 5874 | if (key.objectid != bytenr) |
952fccac | 5875 | break; |
5d4f98a2 YZ |
5876 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
5877 | key.offset == num_bytes) { | |
952fccac CM |
5878 | found_extent = 1; |
5879 | break; | |
5880 | } | |
3173a18f JB |
5881 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
5882 | key.offset == owner_objectid) { | |
5883 | found_extent = 1; | |
5884 | break; | |
5885 | } | |
952fccac CM |
5886 | if (path->slots[0] - extent_slot > 5) |
5887 | break; | |
5d4f98a2 | 5888 | extent_slot--; |
952fccac | 5889 | } |
5d4f98a2 YZ |
5890 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
5891 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
5892 | if (found_extent && item_size < sizeof(*ei)) | |
5893 | found_extent = 0; | |
5894 | #endif | |
31840ae1 | 5895 | if (!found_extent) { |
5d4f98a2 | 5896 | BUG_ON(iref); |
56bec294 | 5897 | ret = remove_extent_backref(trans, extent_root, path, |
5d4f98a2 | 5898 | NULL, refs_to_drop, |
fcebe456 | 5899 | is_data, &last_ref); |
005d6427 DS |
5900 | if (ret) { |
5901 | btrfs_abort_transaction(trans, extent_root, ret); | |
5902 | goto out; | |
5903 | } | |
b3b4aa74 | 5904 | btrfs_release_path(path); |
b9473439 | 5905 | path->leave_spinning = 1; |
5d4f98a2 YZ |
5906 | |
5907 | key.objectid = bytenr; | |
5908 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
5909 | key.offset = num_bytes; | |
5910 | ||
3173a18f JB |
5911 | if (!is_data && skinny_metadata) { |
5912 | key.type = BTRFS_METADATA_ITEM_KEY; | |
5913 | key.offset = owner_objectid; | |
5914 | } | |
5915 | ||
31840ae1 ZY |
5916 | ret = btrfs_search_slot(trans, extent_root, |
5917 | &key, path, -1, 1); | |
3173a18f JB |
5918 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
5919 | /* | |
5920 | * Couldn't find our skinny metadata item, | |
5921 | * see if we have ye olde extent item. | |
5922 | */ | |
5923 | path->slots[0]--; | |
5924 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
5925 | path->slots[0]); | |
5926 | if (key.objectid == bytenr && | |
5927 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
5928 | key.offset == num_bytes) | |
5929 | ret = 0; | |
5930 | } | |
5931 | ||
5932 | if (ret > 0 && skinny_metadata) { | |
5933 | skinny_metadata = false; | |
9ce49a0b | 5934 | key.objectid = bytenr; |
3173a18f JB |
5935 | key.type = BTRFS_EXTENT_ITEM_KEY; |
5936 | key.offset = num_bytes; | |
5937 | btrfs_release_path(path); | |
5938 | ret = btrfs_search_slot(trans, extent_root, | |
5939 | &key, path, -1, 1); | |
5940 | } | |
5941 | ||
f3465ca4 | 5942 | if (ret) { |
c2cf52eb | 5943 | btrfs_err(info, "umm, got %d back from search, was looking for %llu", |
c1c9ff7c | 5944 | ret, bytenr); |
b783e62d JB |
5945 | if (ret > 0) |
5946 | btrfs_print_leaf(extent_root, | |
5947 | path->nodes[0]); | |
f3465ca4 | 5948 | } |
005d6427 DS |
5949 | if (ret < 0) { |
5950 | btrfs_abort_transaction(trans, extent_root, ret); | |
5951 | goto out; | |
5952 | } | |
31840ae1 ZY |
5953 | extent_slot = path->slots[0]; |
5954 | } | |
fae7f21c | 5955 | } else if (WARN_ON(ret == -ENOENT)) { |
7bb86316 | 5956 | btrfs_print_leaf(extent_root, path->nodes[0]); |
c2cf52eb SK |
5957 | btrfs_err(info, |
5958 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
c1c9ff7c GU |
5959 | bytenr, parent, root_objectid, owner_objectid, |
5960 | owner_offset); | |
c4a050bb JB |
5961 | btrfs_abort_transaction(trans, extent_root, ret); |
5962 | goto out; | |
79787eaa | 5963 | } else { |
005d6427 DS |
5964 | btrfs_abort_transaction(trans, extent_root, ret); |
5965 | goto out; | |
7bb86316 | 5966 | } |
5f39d397 CM |
5967 | |
5968 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
5969 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
5970 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
5971 | if (item_size < sizeof(*ei)) { | |
5972 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
5973 | ret = convert_extent_item_v0(trans, extent_root, path, | |
5974 | owner_objectid, 0); | |
005d6427 DS |
5975 | if (ret < 0) { |
5976 | btrfs_abort_transaction(trans, extent_root, ret); | |
5977 | goto out; | |
5978 | } | |
5d4f98a2 | 5979 | |
b3b4aa74 | 5980 | btrfs_release_path(path); |
5d4f98a2 YZ |
5981 | path->leave_spinning = 1; |
5982 | ||
5983 | key.objectid = bytenr; | |
5984 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
5985 | key.offset = num_bytes; | |
5986 | ||
5987 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
5988 | -1, 1); | |
5989 | if (ret) { | |
c2cf52eb | 5990 | btrfs_err(info, "umm, got %d back from search, was looking for %llu", |
c1c9ff7c | 5991 | ret, bytenr); |
5d4f98a2 YZ |
5992 | btrfs_print_leaf(extent_root, path->nodes[0]); |
5993 | } | |
005d6427 DS |
5994 | if (ret < 0) { |
5995 | btrfs_abort_transaction(trans, extent_root, ret); | |
5996 | goto out; | |
5997 | } | |
5998 | ||
5d4f98a2 YZ |
5999 | extent_slot = path->slots[0]; |
6000 | leaf = path->nodes[0]; | |
6001 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
6002 | } | |
6003 | #endif | |
6004 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 6005 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 6006 | struct btrfs_extent_item); |
3173a18f JB |
6007 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
6008 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 YZ |
6009 | struct btrfs_tree_block_info *bi; |
6010 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
6011 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
6012 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
6013 | } | |
56bec294 | 6014 | |
5d4f98a2 | 6015 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 JB |
6016 | if (refs < refs_to_drop) { |
6017 | btrfs_err(info, "trying to drop %d refs but we only have %Lu " | |
351fd353 | 6018 | "for bytenr %Lu", refs_to_drop, refs, bytenr); |
32b02538 JB |
6019 | ret = -EINVAL; |
6020 | btrfs_abort_transaction(trans, extent_root, ret); | |
6021 | goto out; | |
6022 | } | |
56bec294 | 6023 | refs -= refs_to_drop; |
5f39d397 | 6024 | |
5d4f98a2 | 6025 | if (refs > 0) { |
fcebe456 | 6026 | type = BTRFS_QGROUP_OPER_SUB_SHARED; |
5d4f98a2 YZ |
6027 | if (extent_op) |
6028 | __run_delayed_extent_op(extent_op, leaf, ei); | |
6029 | /* | |
6030 | * In the case of inline back ref, reference count will | |
6031 | * be updated by remove_extent_backref | |
952fccac | 6032 | */ |
5d4f98a2 YZ |
6033 | if (iref) { |
6034 | BUG_ON(!found_extent); | |
6035 | } else { | |
6036 | btrfs_set_extent_refs(leaf, ei, refs); | |
6037 | btrfs_mark_buffer_dirty(leaf); | |
6038 | } | |
6039 | if (found_extent) { | |
6040 | ret = remove_extent_backref(trans, extent_root, path, | |
6041 | iref, refs_to_drop, | |
fcebe456 | 6042 | is_data, &last_ref); |
005d6427 DS |
6043 | if (ret) { |
6044 | btrfs_abort_transaction(trans, extent_root, ret); | |
6045 | goto out; | |
6046 | } | |
952fccac | 6047 | } |
b150a4f1 JB |
6048 | add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid, |
6049 | root_objectid); | |
5d4f98a2 | 6050 | } else { |
5d4f98a2 YZ |
6051 | if (found_extent) { |
6052 | BUG_ON(is_data && refs_to_drop != | |
6053 | extent_data_ref_count(root, path, iref)); | |
6054 | if (iref) { | |
6055 | BUG_ON(path->slots[0] != extent_slot); | |
6056 | } else { | |
6057 | BUG_ON(path->slots[0] != extent_slot + 1); | |
6058 | path->slots[0] = extent_slot; | |
6059 | num_to_del = 2; | |
6060 | } | |
78fae27e | 6061 | } |
b9473439 | 6062 | |
fcebe456 | 6063 | last_ref = 1; |
952fccac CM |
6064 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
6065 | num_to_del); | |
005d6427 DS |
6066 | if (ret) { |
6067 | btrfs_abort_transaction(trans, extent_root, ret); | |
6068 | goto out; | |
6069 | } | |
b3b4aa74 | 6070 | btrfs_release_path(path); |
21af804c | 6071 | |
5d4f98a2 | 6072 | if (is_data) { |
459931ec | 6073 | ret = btrfs_del_csums(trans, root, bytenr, num_bytes); |
005d6427 DS |
6074 | if (ret) { |
6075 | btrfs_abort_transaction(trans, extent_root, ret); | |
6076 | goto out; | |
6077 | } | |
459931ec CM |
6078 | } |
6079 | ||
c53d613e | 6080 | ret = update_block_group(root, bytenr, num_bytes, 0); |
005d6427 DS |
6081 | if (ret) { |
6082 | btrfs_abort_transaction(trans, extent_root, ret); | |
6083 | goto out; | |
6084 | } | |
a28ec197 | 6085 | } |
fcebe456 JB |
6086 | btrfs_release_path(path); |
6087 | ||
6088 | /* Deal with the quota accounting */ | |
6089 | if (!ret && last_ref && !no_quota) { | |
6090 | int mod_seq = 0; | |
6091 | ||
6092 | if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID && | |
6093 | type == BTRFS_QGROUP_OPER_SUB_SHARED) | |
6094 | mod_seq = 1; | |
6095 | ||
6096 | ret = btrfs_qgroup_record_ref(trans, info, root_objectid, | |
6097 | bytenr, num_bytes, type, | |
6098 | mod_seq); | |
6099 | } | |
79787eaa | 6100 | out: |
5caf2a00 | 6101 | btrfs_free_path(path); |
a28ec197 CM |
6102 | return ret; |
6103 | } | |
6104 | ||
1887be66 | 6105 | /* |
f0486c68 | 6106 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
6107 | * delayed ref for that extent as well. This searches the delayed ref tree for |
6108 | * a given extent, and if there are no other delayed refs to be processed, it | |
6109 | * removes it from the tree. | |
6110 | */ | |
6111 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
6112 | struct btrfs_root *root, u64 bytenr) | |
6113 | { | |
6114 | struct btrfs_delayed_ref_head *head; | |
6115 | struct btrfs_delayed_ref_root *delayed_refs; | |
f0486c68 | 6116 | int ret = 0; |
1887be66 CM |
6117 | |
6118 | delayed_refs = &trans->transaction->delayed_refs; | |
6119 | spin_lock(&delayed_refs->lock); | |
6120 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
6121 | if (!head) | |
cf93da7b | 6122 | goto out_delayed_unlock; |
1887be66 | 6123 | |
d7df2c79 JB |
6124 | spin_lock(&head->lock); |
6125 | if (rb_first(&head->ref_root)) | |
1887be66 CM |
6126 | goto out; |
6127 | ||
5d4f98a2 YZ |
6128 | if (head->extent_op) { |
6129 | if (!head->must_insert_reserved) | |
6130 | goto out; | |
78a6184a | 6131 | btrfs_free_delayed_extent_op(head->extent_op); |
5d4f98a2 YZ |
6132 | head->extent_op = NULL; |
6133 | } | |
6134 | ||
1887be66 CM |
6135 | /* |
6136 | * waiting for the lock here would deadlock. If someone else has it | |
6137 | * locked they are already in the process of dropping it anyway | |
6138 | */ | |
6139 | if (!mutex_trylock(&head->mutex)) | |
6140 | goto out; | |
6141 | ||
6142 | /* | |
6143 | * at this point we have a head with no other entries. Go | |
6144 | * ahead and process it. | |
6145 | */ | |
6146 | head->node.in_tree = 0; | |
c46effa6 | 6147 | rb_erase(&head->href_node, &delayed_refs->href_root); |
c3e69d58 | 6148 | |
d7df2c79 | 6149 | atomic_dec(&delayed_refs->num_entries); |
1887be66 CM |
6150 | |
6151 | /* | |
6152 | * we don't take a ref on the node because we're removing it from the | |
6153 | * tree, so we just steal the ref the tree was holding. | |
6154 | */ | |
c3e69d58 | 6155 | delayed_refs->num_heads--; |
d7df2c79 | 6156 | if (head->processing == 0) |
c3e69d58 | 6157 | delayed_refs->num_heads_ready--; |
d7df2c79 JB |
6158 | head->processing = 0; |
6159 | spin_unlock(&head->lock); | |
1887be66 CM |
6160 | spin_unlock(&delayed_refs->lock); |
6161 | ||
f0486c68 YZ |
6162 | BUG_ON(head->extent_op); |
6163 | if (head->must_insert_reserved) | |
6164 | ret = 1; | |
6165 | ||
6166 | mutex_unlock(&head->mutex); | |
1887be66 | 6167 | btrfs_put_delayed_ref(&head->node); |
f0486c68 | 6168 | return ret; |
1887be66 | 6169 | out: |
d7df2c79 | 6170 | spin_unlock(&head->lock); |
cf93da7b CM |
6171 | |
6172 | out_delayed_unlock: | |
1887be66 CM |
6173 | spin_unlock(&delayed_refs->lock); |
6174 | return 0; | |
6175 | } | |
6176 | ||
f0486c68 YZ |
6177 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
6178 | struct btrfs_root *root, | |
6179 | struct extent_buffer *buf, | |
5581a51a | 6180 | u64 parent, int last_ref) |
f0486c68 | 6181 | { |
f0486c68 | 6182 | struct btrfs_block_group_cache *cache = NULL; |
b150a4f1 | 6183 | int pin = 1; |
f0486c68 YZ |
6184 | int ret; |
6185 | ||
6186 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
66d7e7f0 AJ |
6187 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
6188 | buf->start, buf->len, | |
6189 | parent, root->root_key.objectid, | |
6190 | btrfs_header_level(buf), | |
5581a51a | 6191 | BTRFS_DROP_DELAYED_REF, NULL, 0); |
79787eaa | 6192 | BUG_ON(ret); /* -ENOMEM */ |
f0486c68 YZ |
6193 | } |
6194 | ||
6195 | if (!last_ref) | |
6196 | return; | |
6197 | ||
f0486c68 | 6198 | cache = btrfs_lookup_block_group(root->fs_info, buf->start); |
f0486c68 YZ |
6199 | |
6200 | if (btrfs_header_generation(buf) == trans->transid) { | |
6201 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
6202 | ret = check_ref_cleanup(trans, root, buf->start); | |
6203 | if (!ret) | |
37be25bc | 6204 | goto out; |
f0486c68 YZ |
6205 | } |
6206 | ||
6207 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { | |
6208 | pin_down_extent(root, cache, buf->start, buf->len, 1); | |
37be25bc | 6209 | goto out; |
f0486c68 YZ |
6210 | } |
6211 | ||
6212 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
6213 | ||
6214 | btrfs_add_free_space(cache, buf->start, buf->len); | |
e570fd27 | 6215 | btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0); |
0be5dc67 | 6216 | trace_btrfs_reserved_extent_free(root, buf->start, buf->len); |
b150a4f1 | 6217 | pin = 0; |
f0486c68 YZ |
6218 | } |
6219 | out: | |
b150a4f1 JB |
6220 | if (pin) |
6221 | add_pinned_bytes(root->fs_info, buf->len, | |
6222 | btrfs_header_level(buf), | |
6223 | root->root_key.objectid); | |
6224 | ||
a826d6dc JB |
6225 | /* |
6226 | * Deleting the buffer, clear the corrupt flag since it doesn't matter | |
6227 | * anymore. | |
6228 | */ | |
6229 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
f0486c68 YZ |
6230 | btrfs_put_block_group(cache); |
6231 | } | |
6232 | ||
79787eaa | 6233 | /* Can return -ENOMEM */ |
66d7e7f0 AJ |
6234 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
6235 | u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, | |
fcebe456 | 6236 | u64 owner, u64 offset, int no_quota) |
925baedd CM |
6237 | { |
6238 | int ret; | |
66d7e7f0 | 6239 | struct btrfs_fs_info *fs_info = root->fs_info; |
925baedd | 6240 | |
faa2dbf0 JB |
6241 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
6242 | if (unlikely(test_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state))) | |
6243 | return 0; | |
6244 | #endif | |
b150a4f1 JB |
6245 | add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid); |
6246 | ||
56bec294 CM |
6247 | /* |
6248 | * tree log blocks never actually go into the extent allocation | |
6249 | * tree, just update pinning info and exit early. | |
56bec294 | 6250 | */ |
5d4f98a2 YZ |
6251 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
6252 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 6253 | /* unlocks the pinned mutex */ |
11833d66 | 6254 | btrfs_pin_extent(root, bytenr, num_bytes, 1); |
56bec294 | 6255 | ret = 0; |
5d4f98a2 | 6256 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
66d7e7f0 AJ |
6257 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
6258 | num_bytes, | |
5d4f98a2 | 6259 | parent, root_objectid, (int)owner, |
fcebe456 | 6260 | BTRFS_DROP_DELAYED_REF, NULL, no_quota); |
5d4f98a2 | 6261 | } else { |
66d7e7f0 AJ |
6262 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
6263 | num_bytes, | |
6264 | parent, root_objectid, owner, | |
6265 | offset, BTRFS_DROP_DELAYED_REF, | |
fcebe456 | 6266 | NULL, no_quota); |
56bec294 | 6267 | } |
925baedd CM |
6268 | return ret; |
6269 | } | |
6270 | ||
53b381b3 DW |
6271 | static u64 stripe_align(struct btrfs_root *root, |
6272 | struct btrfs_block_group_cache *cache, | |
6273 | u64 val, u64 num_bytes) | |
87ee04eb | 6274 | { |
fda2832f | 6275 | u64 ret = ALIGN(val, root->stripesize); |
87ee04eb CM |
6276 | return ret; |
6277 | } | |
6278 | ||
817d52f8 JB |
6279 | /* |
6280 | * when we wait for progress in the block group caching, its because | |
6281 | * our allocation attempt failed at least once. So, we must sleep | |
6282 | * and let some progress happen before we try again. | |
6283 | * | |
6284 | * This function will sleep at least once waiting for new free space to | |
6285 | * show up, and then it will check the block group free space numbers | |
6286 | * for our min num_bytes. Another option is to have it go ahead | |
6287 | * and look in the rbtree for a free extent of a given size, but this | |
6288 | * is a good start. | |
36cce922 JB |
6289 | * |
6290 | * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using | |
6291 | * any of the information in this block group. | |
817d52f8 | 6292 | */ |
36cce922 | 6293 | static noinline void |
817d52f8 JB |
6294 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, |
6295 | u64 num_bytes) | |
6296 | { | |
11833d66 | 6297 | struct btrfs_caching_control *caching_ctl; |
817d52f8 | 6298 | |
11833d66 YZ |
6299 | caching_ctl = get_caching_control(cache); |
6300 | if (!caching_ctl) | |
36cce922 | 6301 | return; |
817d52f8 | 6302 | |
11833d66 | 6303 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 6304 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
6305 | |
6306 | put_caching_control(caching_ctl); | |
11833d66 YZ |
6307 | } |
6308 | ||
6309 | static noinline int | |
6310 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
6311 | { | |
6312 | struct btrfs_caching_control *caching_ctl; | |
36cce922 | 6313 | int ret = 0; |
11833d66 YZ |
6314 | |
6315 | caching_ctl = get_caching_control(cache); | |
6316 | if (!caching_ctl) | |
36cce922 | 6317 | return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0; |
11833d66 YZ |
6318 | |
6319 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
36cce922 JB |
6320 | if (cache->cached == BTRFS_CACHE_ERROR) |
6321 | ret = -EIO; | |
11833d66 | 6322 | put_caching_control(caching_ctl); |
36cce922 | 6323 | return ret; |
817d52f8 JB |
6324 | } |
6325 | ||
31e50229 | 6326 | int __get_raid_index(u64 flags) |
b742bb82 | 6327 | { |
7738a53a | 6328 | if (flags & BTRFS_BLOCK_GROUP_RAID10) |
e6ec716f | 6329 | return BTRFS_RAID_RAID10; |
7738a53a | 6330 | else if (flags & BTRFS_BLOCK_GROUP_RAID1) |
e6ec716f | 6331 | return BTRFS_RAID_RAID1; |
7738a53a | 6332 | else if (flags & BTRFS_BLOCK_GROUP_DUP) |
e6ec716f | 6333 | return BTRFS_RAID_DUP; |
7738a53a | 6334 | else if (flags & BTRFS_BLOCK_GROUP_RAID0) |
e6ec716f | 6335 | return BTRFS_RAID_RAID0; |
53b381b3 | 6336 | else if (flags & BTRFS_BLOCK_GROUP_RAID5) |
e942f883 | 6337 | return BTRFS_RAID_RAID5; |
53b381b3 | 6338 | else if (flags & BTRFS_BLOCK_GROUP_RAID6) |
e942f883 | 6339 | return BTRFS_RAID_RAID6; |
7738a53a | 6340 | |
e942f883 | 6341 | return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */ |
b742bb82 YZ |
6342 | } |
6343 | ||
6ab0a202 | 6344 | int get_block_group_index(struct btrfs_block_group_cache *cache) |
7738a53a | 6345 | { |
31e50229 | 6346 | return __get_raid_index(cache->flags); |
7738a53a ID |
6347 | } |
6348 | ||
6ab0a202 JM |
6349 | static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = { |
6350 | [BTRFS_RAID_RAID10] = "raid10", | |
6351 | [BTRFS_RAID_RAID1] = "raid1", | |
6352 | [BTRFS_RAID_DUP] = "dup", | |
6353 | [BTRFS_RAID_RAID0] = "raid0", | |
6354 | [BTRFS_RAID_SINGLE] = "single", | |
6355 | [BTRFS_RAID_RAID5] = "raid5", | |
6356 | [BTRFS_RAID_RAID6] = "raid6", | |
6357 | }; | |
6358 | ||
1b8e5df6 | 6359 | static const char *get_raid_name(enum btrfs_raid_types type) |
6ab0a202 JM |
6360 | { |
6361 | if (type >= BTRFS_NR_RAID_TYPES) | |
6362 | return NULL; | |
6363 | ||
6364 | return btrfs_raid_type_names[type]; | |
6365 | } | |
6366 | ||
817d52f8 | 6367 | enum btrfs_loop_type { |
285ff5af JB |
6368 | LOOP_CACHING_NOWAIT = 0, |
6369 | LOOP_CACHING_WAIT = 1, | |
6370 | LOOP_ALLOC_CHUNK = 2, | |
6371 | LOOP_NO_EMPTY_SIZE = 3, | |
817d52f8 JB |
6372 | }; |
6373 | ||
e570fd27 MX |
6374 | static inline void |
6375 | btrfs_lock_block_group(struct btrfs_block_group_cache *cache, | |
6376 | int delalloc) | |
6377 | { | |
6378 | if (delalloc) | |
6379 | down_read(&cache->data_rwsem); | |
6380 | } | |
6381 | ||
6382 | static inline void | |
6383 | btrfs_grab_block_group(struct btrfs_block_group_cache *cache, | |
6384 | int delalloc) | |
6385 | { | |
6386 | btrfs_get_block_group(cache); | |
6387 | if (delalloc) | |
6388 | down_read(&cache->data_rwsem); | |
6389 | } | |
6390 | ||
6391 | static struct btrfs_block_group_cache * | |
6392 | btrfs_lock_cluster(struct btrfs_block_group_cache *block_group, | |
6393 | struct btrfs_free_cluster *cluster, | |
6394 | int delalloc) | |
6395 | { | |
6396 | struct btrfs_block_group_cache *used_bg; | |
6397 | bool locked = false; | |
6398 | again: | |
6399 | spin_lock(&cluster->refill_lock); | |
6400 | if (locked) { | |
6401 | if (used_bg == cluster->block_group) | |
6402 | return used_bg; | |
6403 | ||
6404 | up_read(&used_bg->data_rwsem); | |
6405 | btrfs_put_block_group(used_bg); | |
6406 | } | |
6407 | ||
6408 | used_bg = cluster->block_group; | |
6409 | if (!used_bg) | |
6410 | return NULL; | |
6411 | ||
6412 | if (used_bg == block_group) | |
6413 | return used_bg; | |
6414 | ||
6415 | btrfs_get_block_group(used_bg); | |
6416 | ||
6417 | if (!delalloc) | |
6418 | return used_bg; | |
6419 | ||
6420 | if (down_read_trylock(&used_bg->data_rwsem)) | |
6421 | return used_bg; | |
6422 | ||
6423 | spin_unlock(&cluster->refill_lock); | |
6424 | down_read(&used_bg->data_rwsem); | |
6425 | locked = true; | |
6426 | goto again; | |
6427 | } | |
6428 | ||
6429 | static inline void | |
6430 | btrfs_release_block_group(struct btrfs_block_group_cache *cache, | |
6431 | int delalloc) | |
6432 | { | |
6433 | if (delalloc) | |
6434 | up_read(&cache->data_rwsem); | |
6435 | btrfs_put_block_group(cache); | |
6436 | } | |
6437 | ||
fec577fb CM |
6438 | /* |
6439 | * walks the btree of allocated extents and find a hole of a given size. | |
6440 | * The key ins is changed to record the hole: | |
a4820398 | 6441 | * ins->objectid == start position |
62e2749e | 6442 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
a4820398 | 6443 | * ins->offset == the size of the hole. |
fec577fb | 6444 | * Any available blocks before search_start are skipped. |
a4820398 MX |
6445 | * |
6446 | * If there is no suitable free space, we will record the max size of | |
6447 | * the free space extent currently. | |
fec577fb | 6448 | */ |
00361589 | 6449 | static noinline int find_free_extent(struct btrfs_root *orig_root, |
98ed5174 | 6450 | u64 num_bytes, u64 empty_size, |
98ed5174 | 6451 | u64 hint_byte, struct btrfs_key *ins, |
e570fd27 | 6452 | u64 flags, int delalloc) |
fec577fb | 6453 | { |
80eb234a | 6454 | int ret = 0; |
d397712b | 6455 | struct btrfs_root *root = orig_root->fs_info->extent_root; |
fa9c0d79 | 6456 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 6457 | struct btrfs_block_group_cache *block_group = NULL; |
81c9ad23 | 6458 | u64 search_start = 0; |
a4820398 | 6459 | u64 max_extent_size = 0; |
239b14b3 | 6460 | int empty_cluster = 2 * 1024 * 1024; |
80eb234a | 6461 | struct btrfs_space_info *space_info; |
fa9c0d79 | 6462 | int loop = 0; |
b6919a58 DS |
6463 | int index = __get_raid_index(flags); |
6464 | int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ? | |
fb25e914 | 6465 | RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC; |
0a24325e | 6466 | bool failed_cluster_refill = false; |
1cdda9b8 | 6467 | bool failed_alloc = false; |
67377734 | 6468 | bool use_cluster = true; |
60d2adbb | 6469 | bool have_caching_bg = false; |
fec577fb | 6470 | |
db94535d | 6471 | WARN_ON(num_bytes < root->sectorsize); |
b1a4d965 | 6472 | btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); |
80eb234a JB |
6473 | ins->objectid = 0; |
6474 | ins->offset = 0; | |
b1a4d965 | 6475 | |
b6919a58 | 6476 | trace_find_free_extent(orig_root, num_bytes, empty_size, flags); |
3f7de037 | 6477 | |
b6919a58 | 6478 | space_info = __find_space_info(root->fs_info, flags); |
1b1d1f66 | 6479 | if (!space_info) { |
b6919a58 | 6480 | btrfs_err(root->fs_info, "No space info for %llu", flags); |
1b1d1f66 JB |
6481 | return -ENOSPC; |
6482 | } | |
2552d17e | 6483 | |
67377734 JB |
6484 | /* |
6485 | * If the space info is for both data and metadata it means we have a | |
6486 | * small filesystem and we can't use the clustering stuff. | |
6487 | */ | |
6488 | if (btrfs_mixed_space_info(space_info)) | |
6489 | use_cluster = false; | |
6490 | ||
b6919a58 | 6491 | if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) { |
fa9c0d79 | 6492 | last_ptr = &root->fs_info->meta_alloc_cluster; |
536ac8ae CM |
6493 | if (!btrfs_test_opt(root, SSD)) |
6494 | empty_cluster = 64 * 1024; | |
239b14b3 CM |
6495 | } |
6496 | ||
b6919a58 | 6497 | if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster && |
67377734 | 6498 | btrfs_test_opt(root, SSD)) { |
fa9c0d79 CM |
6499 | last_ptr = &root->fs_info->data_alloc_cluster; |
6500 | } | |
0f9dd46c | 6501 | |
239b14b3 | 6502 | if (last_ptr) { |
fa9c0d79 CM |
6503 | spin_lock(&last_ptr->lock); |
6504 | if (last_ptr->block_group) | |
6505 | hint_byte = last_ptr->window_start; | |
6506 | spin_unlock(&last_ptr->lock); | |
239b14b3 | 6507 | } |
fa9c0d79 | 6508 | |
a061fc8d | 6509 | search_start = max(search_start, first_logical_byte(root, 0)); |
239b14b3 | 6510 | search_start = max(search_start, hint_byte); |
0b86a832 | 6511 | |
817d52f8 | 6512 | if (!last_ptr) |
fa9c0d79 | 6513 | empty_cluster = 0; |
fa9c0d79 | 6514 | |
2552d17e | 6515 | if (search_start == hint_byte) { |
2552d17e JB |
6516 | block_group = btrfs_lookup_block_group(root->fs_info, |
6517 | search_start); | |
817d52f8 JB |
6518 | /* |
6519 | * we don't want to use the block group if it doesn't match our | |
6520 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
6521 | * |
6522 | * However if we are re-searching with an ideal block group | |
6523 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 6524 | */ |
b6919a58 | 6525 | if (block_group && block_group_bits(block_group, flags) && |
285ff5af | 6526 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 6527 | down_read(&space_info->groups_sem); |
44fb5511 CM |
6528 | if (list_empty(&block_group->list) || |
6529 | block_group->ro) { | |
6530 | /* | |
6531 | * someone is removing this block group, | |
6532 | * we can't jump into the have_block_group | |
6533 | * target because our list pointers are not | |
6534 | * valid | |
6535 | */ | |
6536 | btrfs_put_block_group(block_group); | |
6537 | up_read(&space_info->groups_sem); | |
ccf0e725 | 6538 | } else { |
b742bb82 | 6539 | index = get_block_group_index(block_group); |
e570fd27 | 6540 | btrfs_lock_block_group(block_group, delalloc); |
44fb5511 | 6541 | goto have_block_group; |
ccf0e725 | 6542 | } |
2552d17e | 6543 | } else if (block_group) { |
fa9c0d79 | 6544 | btrfs_put_block_group(block_group); |
2552d17e | 6545 | } |
42e70e7a | 6546 | } |
2552d17e | 6547 | search: |
60d2adbb | 6548 | have_caching_bg = false; |
80eb234a | 6549 | down_read(&space_info->groups_sem); |
b742bb82 YZ |
6550 | list_for_each_entry(block_group, &space_info->block_groups[index], |
6551 | list) { | |
6226cb0a | 6552 | u64 offset; |
817d52f8 | 6553 | int cached; |
8a1413a2 | 6554 | |
e570fd27 | 6555 | btrfs_grab_block_group(block_group, delalloc); |
2552d17e | 6556 | search_start = block_group->key.objectid; |
42e70e7a | 6557 | |
83a50de9 CM |
6558 | /* |
6559 | * this can happen if we end up cycling through all the | |
6560 | * raid types, but we want to make sure we only allocate | |
6561 | * for the proper type. | |
6562 | */ | |
b6919a58 | 6563 | if (!block_group_bits(block_group, flags)) { |
83a50de9 CM |
6564 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
6565 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
6566 | BTRFS_BLOCK_GROUP_RAID5 | |
6567 | BTRFS_BLOCK_GROUP_RAID6 | | |
83a50de9 CM |
6568 | BTRFS_BLOCK_GROUP_RAID10; |
6569 | ||
6570 | /* | |
6571 | * if they asked for extra copies and this block group | |
6572 | * doesn't provide them, bail. This does allow us to | |
6573 | * fill raid0 from raid1. | |
6574 | */ | |
b6919a58 | 6575 | if ((flags & extra) && !(block_group->flags & extra)) |
83a50de9 CM |
6576 | goto loop; |
6577 | } | |
6578 | ||
2552d17e | 6579 | have_block_group: |
291c7d2f JB |
6580 | cached = block_group_cache_done(block_group); |
6581 | if (unlikely(!cached)) { | |
f6373bf3 | 6582 | ret = cache_block_group(block_group, 0); |
1d4284bd CM |
6583 | BUG_ON(ret < 0); |
6584 | ret = 0; | |
817d52f8 JB |
6585 | } |
6586 | ||
36cce922 JB |
6587 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) |
6588 | goto loop; | |
ea6a478e | 6589 | if (unlikely(block_group->ro)) |
2552d17e | 6590 | goto loop; |
0f9dd46c | 6591 | |
0a24325e | 6592 | /* |
062c05c4 AO |
6593 | * Ok we want to try and use the cluster allocator, so |
6594 | * lets look there | |
0a24325e | 6595 | */ |
062c05c4 | 6596 | if (last_ptr) { |
215a63d1 | 6597 | struct btrfs_block_group_cache *used_block_group; |
8de972b4 | 6598 | unsigned long aligned_cluster; |
fa9c0d79 CM |
6599 | /* |
6600 | * the refill lock keeps out other | |
6601 | * people trying to start a new cluster | |
6602 | */ | |
e570fd27 MX |
6603 | used_block_group = btrfs_lock_cluster(block_group, |
6604 | last_ptr, | |
6605 | delalloc); | |
6606 | if (!used_block_group) | |
44fb5511 | 6607 | goto refill_cluster; |
274bd4fb | 6608 | |
e570fd27 MX |
6609 | if (used_block_group != block_group && |
6610 | (used_block_group->ro || | |
6611 | !block_group_bits(used_block_group, flags))) | |
6612 | goto release_cluster; | |
44fb5511 | 6613 | |
274bd4fb | 6614 | offset = btrfs_alloc_from_cluster(used_block_group, |
a4820398 MX |
6615 | last_ptr, |
6616 | num_bytes, | |
6617 | used_block_group->key.objectid, | |
6618 | &max_extent_size); | |
fa9c0d79 CM |
6619 | if (offset) { |
6620 | /* we have a block, we're done */ | |
6621 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 | 6622 | trace_btrfs_reserve_extent_cluster(root, |
89d4346a MX |
6623 | used_block_group, |
6624 | search_start, num_bytes); | |
215a63d1 | 6625 | if (used_block_group != block_group) { |
e570fd27 MX |
6626 | btrfs_release_block_group(block_group, |
6627 | delalloc); | |
215a63d1 MX |
6628 | block_group = used_block_group; |
6629 | } | |
fa9c0d79 CM |
6630 | goto checks; |
6631 | } | |
6632 | ||
274bd4fb | 6633 | WARN_ON(last_ptr->block_group != used_block_group); |
e570fd27 | 6634 | release_cluster: |
062c05c4 AO |
6635 | /* If we are on LOOP_NO_EMPTY_SIZE, we can't |
6636 | * set up a new clusters, so lets just skip it | |
6637 | * and let the allocator find whatever block | |
6638 | * it can find. If we reach this point, we | |
6639 | * will have tried the cluster allocator | |
6640 | * plenty of times and not have found | |
6641 | * anything, so we are likely way too | |
6642 | * fragmented for the clustering stuff to find | |
a5f6f719 AO |
6643 | * anything. |
6644 | * | |
6645 | * However, if the cluster is taken from the | |
6646 | * current block group, release the cluster | |
6647 | * first, so that we stand a better chance of | |
6648 | * succeeding in the unclustered | |
6649 | * allocation. */ | |
6650 | if (loop >= LOOP_NO_EMPTY_SIZE && | |
e570fd27 | 6651 | used_block_group != block_group) { |
062c05c4 | 6652 | spin_unlock(&last_ptr->refill_lock); |
e570fd27 MX |
6653 | btrfs_release_block_group(used_block_group, |
6654 | delalloc); | |
062c05c4 AO |
6655 | goto unclustered_alloc; |
6656 | } | |
6657 | ||
fa9c0d79 CM |
6658 | /* |
6659 | * this cluster didn't work out, free it and | |
6660 | * start over | |
6661 | */ | |
6662 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
6663 | ||
e570fd27 MX |
6664 | if (used_block_group != block_group) |
6665 | btrfs_release_block_group(used_block_group, | |
6666 | delalloc); | |
6667 | refill_cluster: | |
a5f6f719 AO |
6668 | if (loop >= LOOP_NO_EMPTY_SIZE) { |
6669 | spin_unlock(&last_ptr->refill_lock); | |
6670 | goto unclustered_alloc; | |
6671 | } | |
6672 | ||
8de972b4 CM |
6673 | aligned_cluster = max_t(unsigned long, |
6674 | empty_cluster + empty_size, | |
6675 | block_group->full_stripe_len); | |
6676 | ||
fa9c0d79 | 6677 | /* allocate a cluster in this block group */ |
00361589 JB |
6678 | ret = btrfs_find_space_cluster(root, block_group, |
6679 | last_ptr, search_start, | |
6680 | num_bytes, | |
6681 | aligned_cluster); | |
fa9c0d79 CM |
6682 | if (ret == 0) { |
6683 | /* | |
6684 | * now pull our allocation out of this | |
6685 | * cluster | |
6686 | */ | |
6687 | offset = btrfs_alloc_from_cluster(block_group, | |
a4820398 MX |
6688 | last_ptr, |
6689 | num_bytes, | |
6690 | search_start, | |
6691 | &max_extent_size); | |
fa9c0d79 CM |
6692 | if (offset) { |
6693 | /* we found one, proceed */ | |
6694 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 JB |
6695 | trace_btrfs_reserve_extent_cluster(root, |
6696 | block_group, search_start, | |
6697 | num_bytes); | |
fa9c0d79 CM |
6698 | goto checks; |
6699 | } | |
0a24325e JB |
6700 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
6701 | && !failed_cluster_refill) { | |
817d52f8 JB |
6702 | spin_unlock(&last_ptr->refill_lock); |
6703 | ||
0a24325e | 6704 | failed_cluster_refill = true; |
817d52f8 JB |
6705 | wait_block_group_cache_progress(block_group, |
6706 | num_bytes + empty_cluster + empty_size); | |
6707 | goto have_block_group; | |
fa9c0d79 | 6708 | } |
817d52f8 | 6709 | |
fa9c0d79 CM |
6710 | /* |
6711 | * at this point we either didn't find a cluster | |
6712 | * or we weren't able to allocate a block from our | |
6713 | * cluster. Free the cluster we've been trying | |
6714 | * to use, and go to the next block group | |
6715 | */ | |
0a24325e | 6716 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 6717 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 6718 | goto loop; |
fa9c0d79 CM |
6719 | } |
6720 | ||
062c05c4 | 6721 | unclustered_alloc: |
a5f6f719 AO |
6722 | spin_lock(&block_group->free_space_ctl->tree_lock); |
6723 | if (cached && | |
6724 | block_group->free_space_ctl->free_space < | |
6725 | num_bytes + empty_cluster + empty_size) { | |
a4820398 MX |
6726 | if (block_group->free_space_ctl->free_space > |
6727 | max_extent_size) | |
6728 | max_extent_size = | |
6729 | block_group->free_space_ctl->free_space; | |
a5f6f719 AO |
6730 | spin_unlock(&block_group->free_space_ctl->tree_lock); |
6731 | goto loop; | |
6732 | } | |
6733 | spin_unlock(&block_group->free_space_ctl->tree_lock); | |
6734 | ||
6226cb0a | 6735 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
a4820398 MX |
6736 | num_bytes, empty_size, |
6737 | &max_extent_size); | |
1cdda9b8 JB |
6738 | /* |
6739 | * If we didn't find a chunk, and we haven't failed on this | |
6740 | * block group before, and this block group is in the middle of | |
6741 | * caching and we are ok with waiting, then go ahead and wait | |
6742 | * for progress to be made, and set failed_alloc to true. | |
6743 | * | |
6744 | * If failed_alloc is true then we've already waited on this | |
6745 | * block group once and should move on to the next block group. | |
6746 | */ | |
6747 | if (!offset && !failed_alloc && !cached && | |
6748 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 6749 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
6750 | num_bytes + empty_size); |
6751 | failed_alloc = true; | |
817d52f8 | 6752 | goto have_block_group; |
1cdda9b8 | 6753 | } else if (!offset) { |
60d2adbb MX |
6754 | if (!cached) |
6755 | have_caching_bg = true; | |
1cdda9b8 | 6756 | goto loop; |
817d52f8 | 6757 | } |
fa9c0d79 | 6758 | checks: |
215a63d1 | 6759 | search_start = stripe_align(root, block_group, |
53b381b3 | 6760 | offset, num_bytes); |
25179201 | 6761 | |
2552d17e JB |
6762 | /* move on to the next group */ |
6763 | if (search_start + num_bytes > | |
215a63d1 MX |
6764 | block_group->key.objectid + block_group->key.offset) { |
6765 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 6766 | goto loop; |
6226cb0a | 6767 | } |
f5a31e16 | 6768 | |
f0486c68 | 6769 | if (offset < search_start) |
215a63d1 | 6770 | btrfs_add_free_space(block_group, offset, |
f0486c68 YZ |
6771 | search_start - offset); |
6772 | BUG_ON(offset > search_start); | |
2552d17e | 6773 | |
215a63d1 | 6774 | ret = btrfs_update_reserved_bytes(block_group, num_bytes, |
e570fd27 | 6775 | alloc_type, delalloc); |
f0486c68 | 6776 | if (ret == -EAGAIN) { |
215a63d1 | 6777 | btrfs_add_free_space(block_group, offset, num_bytes); |
2552d17e | 6778 | goto loop; |
0f9dd46c | 6779 | } |
0b86a832 | 6780 | |
f0486c68 | 6781 | /* we are all good, lets return */ |
2552d17e JB |
6782 | ins->objectid = search_start; |
6783 | ins->offset = num_bytes; | |
d2fb3437 | 6784 | |
3f7de037 JB |
6785 | trace_btrfs_reserve_extent(orig_root, block_group, |
6786 | search_start, num_bytes); | |
e570fd27 | 6787 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
6788 | break; |
6789 | loop: | |
0a24325e | 6790 | failed_cluster_refill = false; |
1cdda9b8 | 6791 | failed_alloc = false; |
b742bb82 | 6792 | BUG_ON(index != get_block_group_index(block_group)); |
e570fd27 | 6793 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
6794 | } |
6795 | up_read(&space_info->groups_sem); | |
6796 | ||
60d2adbb MX |
6797 | if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg) |
6798 | goto search; | |
6799 | ||
b742bb82 YZ |
6800 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
6801 | goto search; | |
6802 | ||
285ff5af | 6803 | /* |
ccf0e725 JB |
6804 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking |
6805 | * caching kthreads as we move along | |
817d52f8 JB |
6806 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
6807 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
6808 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
6809 | * again | |
fa9c0d79 | 6810 | */ |
723bda20 | 6811 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { |
b742bb82 | 6812 | index = 0; |
723bda20 | 6813 | loop++; |
817d52f8 | 6814 | if (loop == LOOP_ALLOC_CHUNK) { |
00361589 | 6815 | struct btrfs_trans_handle *trans; |
f017f15f WS |
6816 | int exist = 0; |
6817 | ||
6818 | trans = current->journal_info; | |
6819 | if (trans) | |
6820 | exist = 1; | |
6821 | else | |
6822 | trans = btrfs_join_transaction(root); | |
00361589 | 6823 | |
00361589 JB |
6824 | if (IS_ERR(trans)) { |
6825 | ret = PTR_ERR(trans); | |
6826 | goto out; | |
6827 | } | |
6828 | ||
b6919a58 | 6829 | ret = do_chunk_alloc(trans, root, flags, |
ea658bad JB |
6830 | CHUNK_ALLOC_FORCE); |
6831 | /* | |
6832 | * Do not bail out on ENOSPC since we | |
6833 | * can do more things. | |
6834 | */ | |
00361589 | 6835 | if (ret < 0 && ret != -ENOSPC) |
ea658bad JB |
6836 | btrfs_abort_transaction(trans, |
6837 | root, ret); | |
00361589 JB |
6838 | else |
6839 | ret = 0; | |
f017f15f WS |
6840 | if (!exist) |
6841 | btrfs_end_transaction(trans, root); | |
00361589 | 6842 | if (ret) |
ea658bad | 6843 | goto out; |
2552d17e JB |
6844 | } |
6845 | ||
723bda20 JB |
6846 | if (loop == LOOP_NO_EMPTY_SIZE) { |
6847 | empty_size = 0; | |
6848 | empty_cluster = 0; | |
fa9c0d79 | 6849 | } |
723bda20 JB |
6850 | |
6851 | goto search; | |
2552d17e JB |
6852 | } else if (!ins->objectid) { |
6853 | ret = -ENOSPC; | |
d82a6f1d | 6854 | } else if (ins->objectid) { |
80eb234a | 6855 | ret = 0; |
be744175 | 6856 | } |
79787eaa | 6857 | out: |
a4820398 MX |
6858 | if (ret == -ENOSPC) |
6859 | ins->offset = max_extent_size; | |
0f70abe2 | 6860 | return ret; |
fec577fb | 6861 | } |
ec44a35c | 6862 | |
9ed74f2d JB |
6863 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
6864 | int dump_block_groups) | |
0f9dd46c JB |
6865 | { |
6866 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 6867 | int index = 0; |
0f9dd46c | 6868 | |
9ed74f2d | 6869 | spin_lock(&info->lock); |
efe120a0 | 6870 | printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n", |
c1c9ff7c GU |
6871 | info->flags, |
6872 | info->total_bytes - info->bytes_used - info->bytes_pinned - | |
6873 | info->bytes_reserved - info->bytes_readonly, | |
d397712b | 6874 | (info->full) ? "" : "not "); |
efe120a0 | 6875 | printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, " |
8929ecfa | 6876 | "reserved=%llu, may_use=%llu, readonly=%llu\n", |
c1c9ff7c GU |
6877 | info->total_bytes, info->bytes_used, info->bytes_pinned, |
6878 | info->bytes_reserved, info->bytes_may_use, | |
6879 | info->bytes_readonly); | |
9ed74f2d JB |
6880 | spin_unlock(&info->lock); |
6881 | ||
6882 | if (!dump_block_groups) | |
6883 | return; | |
0f9dd46c | 6884 | |
80eb234a | 6885 | down_read(&info->groups_sem); |
b742bb82 YZ |
6886 | again: |
6887 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 6888 | spin_lock(&cache->lock); |
efe120a0 FH |
6889 | printk(KERN_INFO "BTRFS: " |
6890 | "block group %llu has %llu bytes, " | |
6891 | "%llu used %llu pinned %llu reserved %s\n", | |
c1c9ff7c GU |
6892 | cache->key.objectid, cache->key.offset, |
6893 | btrfs_block_group_used(&cache->item), cache->pinned, | |
6894 | cache->reserved, cache->ro ? "[readonly]" : ""); | |
0f9dd46c JB |
6895 | btrfs_dump_free_space(cache, bytes); |
6896 | spin_unlock(&cache->lock); | |
6897 | } | |
b742bb82 YZ |
6898 | if (++index < BTRFS_NR_RAID_TYPES) |
6899 | goto again; | |
80eb234a | 6900 | up_read(&info->groups_sem); |
0f9dd46c | 6901 | } |
e8569813 | 6902 | |
00361589 | 6903 | int btrfs_reserve_extent(struct btrfs_root *root, |
11833d66 YZ |
6904 | u64 num_bytes, u64 min_alloc_size, |
6905 | u64 empty_size, u64 hint_byte, | |
e570fd27 | 6906 | struct btrfs_key *ins, int is_data, int delalloc) |
fec577fb | 6907 | { |
9e622d6b | 6908 | bool final_tried = false; |
b6919a58 | 6909 | u64 flags; |
fec577fb | 6910 | int ret; |
925baedd | 6911 | |
b6919a58 | 6912 | flags = btrfs_get_alloc_profile(root, is_data); |
98d20f67 | 6913 | again: |
db94535d | 6914 | WARN_ON(num_bytes < root->sectorsize); |
00361589 | 6915 | ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins, |
e570fd27 | 6916 | flags, delalloc); |
3b951516 | 6917 | |
9e622d6b | 6918 | if (ret == -ENOSPC) { |
a4820398 MX |
6919 | if (!final_tried && ins->offset) { |
6920 | num_bytes = min(num_bytes >> 1, ins->offset); | |
24542bf7 | 6921 | num_bytes = round_down(num_bytes, root->sectorsize); |
9e622d6b | 6922 | num_bytes = max(num_bytes, min_alloc_size); |
9e622d6b MX |
6923 | if (num_bytes == min_alloc_size) |
6924 | final_tried = true; | |
6925 | goto again; | |
6926 | } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
6927 | struct btrfs_space_info *sinfo; | |
6928 | ||
b6919a58 | 6929 | sinfo = __find_space_info(root->fs_info, flags); |
c2cf52eb | 6930 | btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu", |
c1c9ff7c | 6931 | flags, num_bytes); |
53804280 JM |
6932 | if (sinfo) |
6933 | dump_space_info(sinfo, num_bytes, 1); | |
9e622d6b | 6934 | } |
925baedd | 6935 | } |
0f9dd46c JB |
6936 | |
6937 | return ret; | |
e6dcd2dc CM |
6938 | } |
6939 | ||
e688b725 | 6940 | static int __btrfs_free_reserved_extent(struct btrfs_root *root, |
e570fd27 MX |
6941 | u64 start, u64 len, |
6942 | int pin, int delalloc) | |
65b51a00 | 6943 | { |
0f9dd46c | 6944 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 6945 | int ret = 0; |
0f9dd46c | 6946 | |
0f9dd46c JB |
6947 | cache = btrfs_lookup_block_group(root->fs_info, start); |
6948 | if (!cache) { | |
c2cf52eb | 6949 | btrfs_err(root->fs_info, "Unable to find block group for %llu", |
c1c9ff7c | 6950 | start); |
0f9dd46c JB |
6951 | return -ENOSPC; |
6952 | } | |
1f3c79a2 | 6953 | |
5378e607 LD |
6954 | if (btrfs_test_opt(root, DISCARD)) |
6955 | ret = btrfs_discard_extent(root, start, len, NULL); | |
1f3c79a2 | 6956 | |
e688b725 CM |
6957 | if (pin) |
6958 | pin_down_extent(root, cache, start, len, 1); | |
6959 | else { | |
6960 | btrfs_add_free_space(cache, start, len); | |
e570fd27 | 6961 | btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc); |
e688b725 | 6962 | } |
fa9c0d79 | 6963 | btrfs_put_block_group(cache); |
817d52f8 | 6964 | |
1abe9b8a | 6965 | trace_btrfs_reserved_extent_free(root, start, len); |
6966 | ||
e6dcd2dc CM |
6967 | return ret; |
6968 | } | |
6969 | ||
e688b725 | 6970 | int btrfs_free_reserved_extent(struct btrfs_root *root, |
e570fd27 | 6971 | u64 start, u64 len, int delalloc) |
e688b725 | 6972 | { |
e570fd27 | 6973 | return __btrfs_free_reserved_extent(root, start, len, 0, delalloc); |
e688b725 CM |
6974 | } |
6975 | ||
6976 | int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root, | |
6977 | u64 start, u64 len) | |
6978 | { | |
e570fd27 | 6979 | return __btrfs_free_reserved_extent(root, start, len, 1, 0); |
e688b725 CM |
6980 | } |
6981 | ||
5d4f98a2 YZ |
6982 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
6983 | struct btrfs_root *root, | |
6984 | u64 parent, u64 root_objectid, | |
6985 | u64 flags, u64 owner, u64 offset, | |
6986 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
6987 | { |
6988 | int ret; | |
5d4f98a2 | 6989 | struct btrfs_fs_info *fs_info = root->fs_info; |
e6dcd2dc | 6990 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 6991 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 6992 | struct btrfs_path *path; |
5d4f98a2 YZ |
6993 | struct extent_buffer *leaf; |
6994 | int type; | |
6995 | u32 size; | |
26b8003f | 6996 | |
5d4f98a2 YZ |
6997 | if (parent > 0) |
6998 | type = BTRFS_SHARED_DATA_REF_KEY; | |
6999 | else | |
7000 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 7001 | |
5d4f98a2 | 7002 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
7003 | |
7004 | path = btrfs_alloc_path(); | |
db5b493a TI |
7005 | if (!path) |
7006 | return -ENOMEM; | |
47e4bb98 | 7007 | |
b9473439 | 7008 | path->leave_spinning = 1; |
5d4f98a2 YZ |
7009 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
7010 | ins, size); | |
79787eaa JM |
7011 | if (ret) { |
7012 | btrfs_free_path(path); | |
7013 | return ret; | |
7014 | } | |
0f9dd46c | 7015 | |
5d4f98a2 YZ |
7016 | leaf = path->nodes[0]; |
7017 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 7018 | struct btrfs_extent_item); |
5d4f98a2 YZ |
7019 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
7020 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
7021 | btrfs_set_extent_flags(leaf, extent_item, | |
7022 | flags | BTRFS_EXTENT_FLAG_DATA); | |
7023 | ||
7024 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
7025 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
7026 | if (parent > 0) { | |
7027 | struct btrfs_shared_data_ref *ref; | |
7028 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
7029 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
7030 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
7031 | } else { | |
7032 | struct btrfs_extent_data_ref *ref; | |
7033 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
7034 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
7035 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
7036 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
7037 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
7038 | } | |
47e4bb98 CM |
7039 | |
7040 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 7041 | btrfs_free_path(path); |
f510cfec | 7042 | |
fcebe456 JB |
7043 | /* Always set parent to 0 here since its exclusive anyway. */ |
7044 | ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid, | |
7045 | ins->objectid, ins->offset, | |
7046 | BTRFS_QGROUP_OPER_ADD_EXCL, 0); | |
7047 | if (ret) | |
7048 | return ret; | |
7049 | ||
c53d613e | 7050 | ret = update_block_group(root, ins->objectid, ins->offset, 1); |
79787eaa | 7051 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 7052 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 7053 | ins->objectid, ins->offset); |
f5947066 CM |
7054 | BUG(); |
7055 | } | |
0be5dc67 | 7056 | trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset); |
e6dcd2dc CM |
7057 | return ret; |
7058 | } | |
7059 | ||
5d4f98a2 YZ |
7060 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
7061 | struct btrfs_root *root, | |
7062 | u64 parent, u64 root_objectid, | |
7063 | u64 flags, struct btrfs_disk_key *key, | |
fcebe456 JB |
7064 | int level, struct btrfs_key *ins, |
7065 | int no_quota) | |
e6dcd2dc CM |
7066 | { |
7067 | int ret; | |
5d4f98a2 YZ |
7068 | struct btrfs_fs_info *fs_info = root->fs_info; |
7069 | struct btrfs_extent_item *extent_item; | |
7070 | struct btrfs_tree_block_info *block_info; | |
7071 | struct btrfs_extent_inline_ref *iref; | |
7072 | struct btrfs_path *path; | |
7073 | struct extent_buffer *leaf; | |
3173a18f | 7074 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
fcebe456 | 7075 | u64 num_bytes = ins->offset; |
3173a18f JB |
7076 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
7077 | SKINNY_METADATA); | |
7078 | ||
7079 | if (!skinny_metadata) | |
7080 | size += sizeof(*block_info); | |
1c2308f8 | 7081 | |
5d4f98a2 | 7082 | path = btrfs_alloc_path(); |
857cc2fc JB |
7083 | if (!path) { |
7084 | btrfs_free_and_pin_reserved_extent(root, ins->objectid, | |
7085 | root->leafsize); | |
d8926bb3 | 7086 | return -ENOMEM; |
857cc2fc | 7087 | } |
56bec294 | 7088 | |
5d4f98a2 YZ |
7089 | path->leave_spinning = 1; |
7090 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
7091 | ins, size); | |
79787eaa | 7092 | if (ret) { |
857cc2fc JB |
7093 | btrfs_free_and_pin_reserved_extent(root, ins->objectid, |
7094 | root->leafsize); | |
79787eaa JM |
7095 | btrfs_free_path(path); |
7096 | return ret; | |
7097 | } | |
5d4f98a2 YZ |
7098 | |
7099 | leaf = path->nodes[0]; | |
7100 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
7101 | struct btrfs_extent_item); | |
7102 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
7103 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
7104 | btrfs_set_extent_flags(leaf, extent_item, | |
7105 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 7106 | |
3173a18f JB |
7107 | if (skinny_metadata) { |
7108 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
fcebe456 | 7109 | num_bytes = root->leafsize; |
3173a18f JB |
7110 | } else { |
7111 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
7112 | btrfs_set_tree_block_key(leaf, block_info, key); | |
7113 | btrfs_set_tree_block_level(leaf, block_info, level); | |
7114 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
7115 | } | |
5d4f98a2 | 7116 | |
5d4f98a2 YZ |
7117 | if (parent > 0) { |
7118 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
7119 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
7120 | BTRFS_SHARED_BLOCK_REF_KEY); | |
7121 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
7122 | } else { | |
7123 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
7124 | BTRFS_TREE_BLOCK_REF_KEY); | |
7125 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
7126 | } | |
7127 | ||
7128 | btrfs_mark_buffer_dirty(leaf); | |
7129 | btrfs_free_path(path); | |
7130 | ||
fcebe456 JB |
7131 | if (!no_quota) { |
7132 | ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid, | |
7133 | ins->objectid, num_bytes, | |
7134 | BTRFS_QGROUP_OPER_ADD_EXCL, 0); | |
7135 | if (ret) | |
7136 | return ret; | |
7137 | } | |
7138 | ||
3173a18f | 7139 | ret = update_block_group(root, ins->objectid, root->leafsize, 1); |
79787eaa | 7140 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 7141 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 7142 | ins->objectid, ins->offset); |
5d4f98a2 YZ |
7143 | BUG(); |
7144 | } | |
0be5dc67 JB |
7145 | |
7146 | trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->leafsize); | |
5d4f98a2 YZ |
7147 | return ret; |
7148 | } | |
7149 | ||
7150 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
7151 | struct btrfs_root *root, | |
7152 | u64 root_objectid, u64 owner, | |
7153 | u64 offset, struct btrfs_key *ins) | |
7154 | { | |
7155 | int ret; | |
7156 | ||
7157 | BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
7158 | ||
66d7e7f0 AJ |
7159 | ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid, |
7160 | ins->offset, 0, | |
7161 | root_objectid, owner, offset, | |
7162 | BTRFS_ADD_DELAYED_EXTENT, NULL, 0); | |
e6dcd2dc CM |
7163 | return ret; |
7164 | } | |
e02119d5 CM |
7165 | |
7166 | /* | |
7167 | * this is used by the tree logging recovery code. It records that | |
7168 | * an extent has been allocated and makes sure to clear the free | |
7169 | * space cache bits as well | |
7170 | */ | |
5d4f98a2 YZ |
7171 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
7172 | struct btrfs_root *root, | |
7173 | u64 root_objectid, u64 owner, u64 offset, | |
7174 | struct btrfs_key *ins) | |
e02119d5 CM |
7175 | { |
7176 | int ret; | |
7177 | struct btrfs_block_group_cache *block_group; | |
11833d66 | 7178 | |
8c2a1a30 JB |
7179 | /* |
7180 | * Mixed block groups will exclude before processing the log so we only | |
7181 | * need to do the exlude dance if this fs isn't mixed. | |
7182 | */ | |
7183 | if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) { | |
7184 | ret = __exclude_logged_extent(root, ins->objectid, ins->offset); | |
b50c6e25 | 7185 | if (ret) |
8c2a1a30 | 7186 | return ret; |
11833d66 YZ |
7187 | } |
7188 | ||
8c2a1a30 JB |
7189 | block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); |
7190 | if (!block_group) | |
7191 | return -EINVAL; | |
7192 | ||
fb25e914 | 7193 | ret = btrfs_update_reserved_bytes(block_group, ins->offset, |
e570fd27 | 7194 | RESERVE_ALLOC_NO_ACCOUNT, 0); |
79787eaa | 7195 | BUG_ON(ret); /* logic error */ |
5d4f98a2 YZ |
7196 | ret = alloc_reserved_file_extent(trans, root, 0, root_objectid, |
7197 | 0, owner, offset, ins, 1); | |
b50c6e25 | 7198 | btrfs_put_block_group(block_group); |
e02119d5 CM |
7199 | return ret; |
7200 | } | |
7201 | ||
48a3b636 ES |
7202 | static struct extent_buffer * |
7203 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
7204 | u64 bytenr, u32 blocksize, int level) | |
65b51a00 CM |
7205 | { |
7206 | struct extent_buffer *buf; | |
7207 | ||
7208 | buf = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
7209 | if (!buf) | |
7210 | return ERR_PTR(-ENOMEM); | |
7211 | btrfs_set_header_generation(buf, trans->transid); | |
85d4e461 | 7212 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 CM |
7213 | btrfs_tree_lock(buf); |
7214 | clean_tree_block(trans, root, buf); | |
3083ee2e | 7215 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de CM |
7216 | |
7217 | btrfs_set_lock_blocking(buf); | |
65b51a00 | 7218 | btrfs_set_buffer_uptodate(buf); |
b4ce94de | 7219 | |
d0c803c4 | 7220 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
8cef4e16 YZ |
7221 | /* |
7222 | * we allow two log transactions at a time, use different | |
7223 | * EXENT bit to differentiate dirty pages. | |
7224 | */ | |
7225 | if (root->log_transid % 2 == 0) | |
7226 | set_extent_dirty(&root->dirty_log_pages, buf->start, | |
7227 | buf->start + buf->len - 1, GFP_NOFS); | |
7228 | else | |
7229 | set_extent_new(&root->dirty_log_pages, buf->start, | |
7230 | buf->start + buf->len - 1, GFP_NOFS); | |
d0c803c4 CM |
7231 | } else { |
7232 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, | |
65b51a00 | 7233 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 7234 | } |
65b51a00 | 7235 | trans->blocks_used++; |
b4ce94de | 7236 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
7237 | return buf; |
7238 | } | |
7239 | ||
f0486c68 YZ |
7240 | static struct btrfs_block_rsv * |
7241 | use_block_rsv(struct btrfs_trans_handle *trans, | |
7242 | struct btrfs_root *root, u32 blocksize) | |
7243 | { | |
7244 | struct btrfs_block_rsv *block_rsv; | |
68a82277 | 7245 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
f0486c68 | 7246 | int ret; |
d88033db | 7247 | bool global_updated = false; |
f0486c68 YZ |
7248 | |
7249 | block_rsv = get_block_rsv(trans, root); | |
7250 | ||
b586b323 MX |
7251 | if (unlikely(block_rsv->size == 0)) |
7252 | goto try_reserve; | |
d88033db | 7253 | again: |
f0486c68 YZ |
7254 | ret = block_rsv_use_bytes(block_rsv, blocksize); |
7255 | if (!ret) | |
7256 | return block_rsv; | |
7257 | ||
b586b323 MX |
7258 | if (block_rsv->failfast) |
7259 | return ERR_PTR(ret); | |
7260 | ||
d88033db MX |
7261 | if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) { |
7262 | global_updated = true; | |
7263 | update_global_block_rsv(root->fs_info); | |
7264 | goto again; | |
7265 | } | |
7266 | ||
b586b323 MX |
7267 | if (btrfs_test_opt(root, ENOSPC_DEBUG)) { |
7268 | static DEFINE_RATELIMIT_STATE(_rs, | |
7269 | DEFAULT_RATELIMIT_INTERVAL * 10, | |
7270 | /*DEFAULT_RATELIMIT_BURST*/ 1); | |
7271 | if (__ratelimit(&_rs)) | |
7272 | WARN(1, KERN_DEBUG | |
efe120a0 | 7273 | "BTRFS: block rsv returned %d\n", ret); |
b586b323 MX |
7274 | } |
7275 | try_reserve: | |
7276 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, | |
7277 | BTRFS_RESERVE_NO_FLUSH); | |
7278 | if (!ret) | |
7279 | return block_rsv; | |
7280 | /* | |
7281 | * If we couldn't reserve metadata bytes try and use some from | |
5881cfc9 MX |
7282 | * the global reserve if its space type is the same as the global |
7283 | * reservation. | |
b586b323 | 7284 | */ |
5881cfc9 MX |
7285 | if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL && |
7286 | block_rsv->space_info == global_rsv->space_info) { | |
b586b323 MX |
7287 | ret = block_rsv_use_bytes(global_rsv, blocksize); |
7288 | if (!ret) | |
7289 | return global_rsv; | |
7290 | } | |
7291 | return ERR_PTR(ret); | |
f0486c68 YZ |
7292 | } |
7293 | ||
8c2a3ca2 JB |
7294 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
7295 | struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
f0486c68 YZ |
7296 | { |
7297 | block_rsv_add_bytes(block_rsv, blocksize, 0); | |
8c2a3ca2 | 7298 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0); |
f0486c68 YZ |
7299 | } |
7300 | ||
fec577fb | 7301 | /* |
f0486c68 YZ |
7302 | * finds a free extent and does all the dirty work required for allocation |
7303 | * returns the key for the extent through ins, and a tree buffer for | |
7304 | * the first block of the extent through buf. | |
7305 | * | |
fec577fb CM |
7306 | * returns the tree buffer or NULL. |
7307 | */ | |
5f39d397 | 7308 | struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
7309 | struct btrfs_root *root, u32 blocksize, |
7310 | u64 parent, u64 root_objectid, | |
7311 | struct btrfs_disk_key *key, int level, | |
5581a51a | 7312 | u64 hint, u64 empty_size) |
fec577fb | 7313 | { |
e2fa7227 | 7314 | struct btrfs_key ins; |
f0486c68 | 7315 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 7316 | struct extent_buffer *buf; |
f0486c68 YZ |
7317 | u64 flags = 0; |
7318 | int ret; | |
3173a18f JB |
7319 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
7320 | SKINNY_METADATA); | |
fec577fb | 7321 | |
faa2dbf0 JB |
7322 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
7323 | if (unlikely(test_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state))) { | |
7324 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, | |
7325 | blocksize, level); | |
7326 | if (!IS_ERR(buf)) | |
7327 | root->alloc_bytenr += blocksize; | |
7328 | return buf; | |
7329 | } | |
7330 | #endif | |
f0486c68 YZ |
7331 | block_rsv = use_block_rsv(trans, root, blocksize); |
7332 | if (IS_ERR(block_rsv)) | |
7333 | return ERR_CAST(block_rsv); | |
7334 | ||
00361589 | 7335 | ret = btrfs_reserve_extent(root, blocksize, blocksize, |
e570fd27 | 7336 | empty_size, hint, &ins, 0, 0); |
fec577fb | 7337 | if (ret) { |
8c2a3ca2 | 7338 | unuse_block_rsv(root->fs_info, block_rsv, blocksize); |
54aa1f4d | 7339 | return ERR_PTR(ret); |
fec577fb | 7340 | } |
55c69072 | 7341 | |
4008c04a CM |
7342 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, |
7343 | blocksize, level); | |
79787eaa | 7344 | BUG_ON(IS_ERR(buf)); /* -ENOMEM */ |
f0486c68 YZ |
7345 | |
7346 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
7347 | if (parent == 0) | |
7348 | parent = ins.objectid; | |
7349 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
7350 | } else | |
7351 | BUG_ON(parent > 0); | |
7352 | ||
7353 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
7354 | struct btrfs_delayed_extent_op *extent_op; | |
78a6184a | 7355 | extent_op = btrfs_alloc_delayed_extent_op(); |
79787eaa | 7356 | BUG_ON(!extent_op); /* -ENOMEM */ |
f0486c68 YZ |
7357 | if (key) |
7358 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
7359 | else | |
7360 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
7361 | extent_op->flags_to_set = flags; | |
3173a18f JB |
7362 | if (skinny_metadata) |
7363 | extent_op->update_key = 0; | |
7364 | else | |
7365 | extent_op->update_key = 1; | |
f0486c68 YZ |
7366 | extent_op->update_flags = 1; |
7367 | extent_op->is_data = 0; | |
b1c79e09 | 7368 | extent_op->level = level; |
f0486c68 | 7369 | |
66d7e7f0 AJ |
7370 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
7371 | ins.objectid, | |
f0486c68 YZ |
7372 | ins.offset, parent, root_objectid, |
7373 | level, BTRFS_ADD_DELAYED_EXTENT, | |
5581a51a | 7374 | extent_op, 0); |
79787eaa | 7375 | BUG_ON(ret); /* -ENOMEM */ |
f0486c68 | 7376 | } |
fec577fb CM |
7377 | return buf; |
7378 | } | |
a28ec197 | 7379 | |
2c47e605 YZ |
7380 | struct walk_control { |
7381 | u64 refs[BTRFS_MAX_LEVEL]; | |
7382 | u64 flags[BTRFS_MAX_LEVEL]; | |
7383 | struct btrfs_key update_progress; | |
7384 | int stage; | |
7385 | int level; | |
7386 | int shared_level; | |
7387 | int update_ref; | |
7388 | int keep_locks; | |
1c4850e2 YZ |
7389 | int reada_slot; |
7390 | int reada_count; | |
66d7e7f0 | 7391 | int for_reloc; |
2c47e605 YZ |
7392 | }; |
7393 | ||
7394 | #define DROP_REFERENCE 1 | |
7395 | #define UPDATE_BACKREF 2 | |
7396 | ||
1c4850e2 YZ |
7397 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
7398 | struct btrfs_root *root, | |
7399 | struct walk_control *wc, | |
7400 | struct btrfs_path *path) | |
6407bf6d | 7401 | { |
1c4850e2 YZ |
7402 | u64 bytenr; |
7403 | u64 generation; | |
7404 | u64 refs; | |
94fcca9f | 7405 | u64 flags; |
5d4f98a2 | 7406 | u32 nritems; |
1c4850e2 YZ |
7407 | u32 blocksize; |
7408 | struct btrfs_key key; | |
7409 | struct extent_buffer *eb; | |
6407bf6d | 7410 | int ret; |
1c4850e2 YZ |
7411 | int slot; |
7412 | int nread = 0; | |
6407bf6d | 7413 | |
1c4850e2 YZ |
7414 | if (path->slots[wc->level] < wc->reada_slot) { |
7415 | wc->reada_count = wc->reada_count * 2 / 3; | |
7416 | wc->reada_count = max(wc->reada_count, 2); | |
7417 | } else { | |
7418 | wc->reada_count = wc->reada_count * 3 / 2; | |
7419 | wc->reada_count = min_t(int, wc->reada_count, | |
7420 | BTRFS_NODEPTRS_PER_BLOCK(root)); | |
7421 | } | |
7bb86316 | 7422 | |
1c4850e2 YZ |
7423 | eb = path->nodes[wc->level]; |
7424 | nritems = btrfs_header_nritems(eb); | |
7425 | blocksize = btrfs_level_size(root, wc->level - 1); | |
bd56b302 | 7426 | |
1c4850e2 YZ |
7427 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
7428 | if (nread >= wc->reada_count) | |
7429 | break; | |
bd56b302 | 7430 | |
2dd3e67b | 7431 | cond_resched(); |
1c4850e2 YZ |
7432 | bytenr = btrfs_node_blockptr(eb, slot); |
7433 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 7434 | |
1c4850e2 YZ |
7435 | if (slot == path->slots[wc->level]) |
7436 | goto reada; | |
5d4f98a2 | 7437 | |
1c4850e2 YZ |
7438 | if (wc->stage == UPDATE_BACKREF && |
7439 | generation <= root->root_key.offset) | |
bd56b302 CM |
7440 | continue; |
7441 | ||
94fcca9f | 7442 | /* We don't lock the tree block, it's OK to be racy here */ |
3173a18f JB |
7443 | ret = btrfs_lookup_extent_info(trans, root, bytenr, |
7444 | wc->level - 1, 1, &refs, | |
7445 | &flags); | |
79787eaa JM |
7446 | /* We don't care about errors in readahead. */ |
7447 | if (ret < 0) | |
7448 | continue; | |
94fcca9f YZ |
7449 | BUG_ON(refs == 0); |
7450 | ||
1c4850e2 | 7451 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
7452 | if (refs == 1) |
7453 | goto reada; | |
bd56b302 | 7454 | |
94fcca9f YZ |
7455 | if (wc->level == 1 && |
7456 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
7457 | continue; | |
1c4850e2 YZ |
7458 | if (!wc->update_ref || |
7459 | generation <= root->root_key.offset) | |
7460 | continue; | |
7461 | btrfs_node_key_to_cpu(eb, &key, slot); | |
7462 | ret = btrfs_comp_cpu_keys(&key, | |
7463 | &wc->update_progress); | |
7464 | if (ret < 0) | |
7465 | continue; | |
94fcca9f YZ |
7466 | } else { |
7467 | if (wc->level == 1 && | |
7468 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
7469 | continue; | |
6407bf6d | 7470 | } |
1c4850e2 YZ |
7471 | reada: |
7472 | ret = readahead_tree_block(root, bytenr, blocksize, | |
7473 | generation); | |
7474 | if (ret) | |
bd56b302 | 7475 | break; |
1c4850e2 | 7476 | nread++; |
20524f02 | 7477 | } |
1c4850e2 | 7478 | wc->reada_slot = slot; |
20524f02 | 7479 | } |
2c47e605 | 7480 | |
1152651a MF |
7481 | static int account_leaf_items(struct btrfs_trans_handle *trans, |
7482 | struct btrfs_root *root, | |
7483 | struct extent_buffer *eb) | |
7484 | { | |
7485 | int nr = btrfs_header_nritems(eb); | |
7486 | int i, extent_type, ret; | |
7487 | struct btrfs_key key; | |
7488 | struct btrfs_file_extent_item *fi; | |
7489 | u64 bytenr, num_bytes; | |
7490 | ||
7491 | for (i = 0; i < nr; i++) { | |
7492 | btrfs_item_key_to_cpu(eb, &key, i); | |
7493 | ||
7494 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
7495 | continue; | |
7496 | ||
7497 | fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
7498 | /* filter out non qgroup-accountable extents */ | |
7499 | extent_type = btrfs_file_extent_type(eb, fi); | |
7500 | ||
7501 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) | |
7502 | continue; | |
7503 | ||
7504 | bytenr = btrfs_file_extent_disk_bytenr(eb, fi); | |
7505 | if (!bytenr) | |
7506 | continue; | |
7507 | ||
7508 | num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi); | |
7509 | ||
7510 | ret = btrfs_qgroup_record_ref(trans, root->fs_info, | |
7511 | root->objectid, | |
7512 | bytenr, num_bytes, | |
7513 | BTRFS_QGROUP_OPER_SUB_SUBTREE, 0); | |
7514 | if (ret) | |
7515 | return ret; | |
7516 | } | |
7517 | return 0; | |
7518 | } | |
7519 | ||
7520 | /* | |
7521 | * Walk up the tree from the bottom, freeing leaves and any interior | |
7522 | * nodes which have had all slots visited. If a node (leaf or | |
7523 | * interior) is freed, the node above it will have it's slot | |
7524 | * incremented. The root node will never be freed. | |
7525 | * | |
7526 | * At the end of this function, we should have a path which has all | |
7527 | * slots incremented to the next position for a search. If we need to | |
7528 | * read a new node it will be NULL and the node above it will have the | |
7529 | * correct slot selected for a later read. | |
7530 | * | |
7531 | * If we increment the root nodes slot counter past the number of | |
7532 | * elements, 1 is returned to signal completion of the search. | |
7533 | */ | |
7534 | static int adjust_slots_upwards(struct btrfs_root *root, | |
7535 | struct btrfs_path *path, int root_level) | |
7536 | { | |
7537 | int level = 0; | |
7538 | int nr, slot; | |
7539 | struct extent_buffer *eb; | |
7540 | ||
7541 | if (root_level == 0) | |
7542 | return 1; | |
7543 | ||
7544 | while (level <= root_level) { | |
7545 | eb = path->nodes[level]; | |
7546 | nr = btrfs_header_nritems(eb); | |
7547 | path->slots[level]++; | |
7548 | slot = path->slots[level]; | |
7549 | if (slot >= nr || level == 0) { | |
7550 | /* | |
7551 | * Don't free the root - we will detect this | |
7552 | * condition after our loop and return a | |
7553 | * positive value for caller to stop walking the tree. | |
7554 | */ | |
7555 | if (level != root_level) { | |
7556 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
7557 | path->locks[level] = 0; | |
7558 | ||
7559 | free_extent_buffer(eb); | |
7560 | path->nodes[level] = NULL; | |
7561 | path->slots[level] = 0; | |
7562 | } | |
7563 | } else { | |
7564 | /* | |
7565 | * We have a valid slot to walk back down | |
7566 | * from. Stop here so caller can process these | |
7567 | * new nodes. | |
7568 | */ | |
7569 | break; | |
7570 | } | |
7571 | ||
7572 | level++; | |
7573 | } | |
7574 | ||
7575 | eb = path->nodes[root_level]; | |
7576 | if (path->slots[root_level] >= btrfs_header_nritems(eb)) | |
7577 | return 1; | |
7578 | ||
7579 | return 0; | |
7580 | } | |
7581 | ||
7582 | /* | |
7583 | * root_eb is the subtree root and is locked before this function is called. | |
7584 | */ | |
7585 | static int account_shared_subtree(struct btrfs_trans_handle *trans, | |
7586 | struct btrfs_root *root, | |
7587 | struct extent_buffer *root_eb, | |
7588 | u64 root_gen, | |
7589 | int root_level) | |
7590 | { | |
7591 | int ret = 0; | |
7592 | int level; | |
7593 | struct extent_buffer *eb = root_eb; | |
7594 | struct btrfs_path *path = NULL; | |
7595 | ||
7596 | BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL); | |
7597 | BUG_ON(root_eb == NULL); | |
7598 | ||
7599 | if (!root->fs_info->quota_enabled) | |
7600 | return 0; | |
7601 | ||
7602 | if (!extent_buffer_uptodate(root_eb)) { | |
7603 | ret = btrfs_read_buffer(root_eb, root_gen); | |
7604 | if (ret) | |
7605 | goto out; | |
7606 | } | |
7607 | ||
7608 | if (root_level == 0) { | |
7609 | ret = account_leaf_items(trans, root, root_eb); | |
7610 | goto out; | |
7611 | } | |
7612 | ||
7613 | path = btrfs_alloc_path(); | |
7614 | if (!path) | |
7615 | return -ENOMEM; | |
7616 | ||
7617 | /* | |
7618 | * Walk down the tree. Missing extent blocks are filled in as | |
7619 | * we go. Metadata is accounted every time we read a new | |
7620 | * extent block. | |
7621 | * | |
7622 | * When we reach a leaf, we account for file extent items in it, | |
7623 | * walk back up the tree (adjusting slot pointers as we go) | |
7624 | * and restart the search process. | |
7625 | */ | |
7626 | extent_buffer_get(root_eb); /* For path */ | |
7627 | path->nodes[root_level] = root_eb; | |
7628 | path->slots[root_level] = 0; | |
7629 | path->locks[root_level] = 0; /* so release_path doesn't try to unlock */ | |
7630 | walk_down: | |
7631 | level = root_level; | |
7632 | while (level >= 0) { | |
7633 | if (path->nodes[level] == NULL) { | |
7634 | int child_bsize = root->nodesize; | |
7635 | int parent_slot; | |
7636 | u64 child_gen; | |
7637 | u64 child_bytenr; | |
7638 | ||
7639 | /* We need to get child blockptr/gen from | |
7640 | * parent before we can read it. */ | |
7641 | eb = path->nodes[level + 1]; | |
7642 | parent_slot = path->slots[level + 1]; | |
7643 | child_bytenr = btrfs_node_blockptr(eb, parent_slot); | |
7644 | child_gen = btrfs_node_ptr_generation(eb, parent_slot); | |
7645 | ||
7646 | eb = read_tree_block(root, child_bytenr, child_bsize, | |
7647 | child_gen); | |
7648 | if (!eb || !extent_buffer_uptodate(eb)) { | |
7649 | ret = -EIO; | |
7650 | goto out; | |
7651 | } | |
7652 | ||
7653 | path->nodes[level] = eb; | |
7654 | path->slots[level] = 0; | |
7655 | ||
7656 | btrfs_tree_read_lock(eb); | |
7657 | btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); | |
7658 | path->locks[level] = BTRFS_READ_LOCK_BLOCKING; | |
7659 | ||
7660 | ret = btrfs_qgroup_record_ref(trans, root->fs_info, | |
7661 | root->objectid, | |
7662 | child_bytenr, | |
7663 | child_bsize, | |
7664 | BTRFS_QGROUP_OPER_SUB_SUBTREE, | |
7665 | 0); | |
7666 | if (ret) | |
7667 | goto out; | |
7668 | ||
7669 | } | |
7670 | ||
7671 | if (level == 0) { | |
7672 | ret = account_leaf_items(trans, root, path->nodes[level]); | |
7673 | if (ret) | |
7674 | goto out; | |
7675 | ||
7676 | /* Nonzero return here means we completed our search */ | |
7677 | ret = adjust_slots_upwards(root, path, root_level); | |
7678 | if (ret) | |
7679 | break; | |
7680 | ||
7681 | /* Restart search with new slots */ | |
7682 | goto walk_down; | |
7683 | } | |
7684 | ||
7685 | level--; | |
7686 | } | |
7687 | ||
7688 | ret = 0; | |
7689 | out: | |
7690 | btrfs_free_path(path); | |
7691 | ||
7692 | return ret; | |
7693 | } | |
7694 | ||
f82d02d9 | 7695 | /* |
2c016dc2 | 7696 | * helper to process tree block while walking down the tree. |
2c47e605 | 7697 | * |
2c47e605 YZ |
7698 | * when wc->stage == UPDATE_BACKREF, this function updates |
7699 | * back refs for pointers in the block. | |
7700 | * | |
7701 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 7702 | */ |
2c47e605 | 7703 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 7704 | struct btrfs_root *root, |
2c47e605 | 7705 | struct btrfs_path *path, |
94fcca9f | 7706 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 7707 | { |
2c47e605 YZ |
7708 | int level = wc->level; |
7709 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 7710 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
7711 | int ret; |
7712 | ||
2c47e605 YZ |
7713 | if (wc->stage == UPDATE_BACKREF && |
7714 | btrfs_header_owner(eb) != root->root_key.objectid) | |
7715 | return 1; | |
f82d02d9 | 7716 | |
2c47e605 YZ |
7717 | /* |
7718 | * when reference count of tree block is 1, it won't increase | |
7719 | * again. once full backref flag is set, we never clear it. | |
7720 | */ | |
94fcca9f YZ |
7721 | if (lookup_info && |
7722 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
7723 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 YZ |
7724 | BUG_ON(!path->locks[level]); |
7725 | ret = btrfs_lookup_extent_info(trans, root, | |
3173a18f | 7726 | eb->start, level, 1, |
2c47e605 YZ |
7727 | &wc->refs[level], |
7728 | &wc->flags[level]); | |
79787eaa JM |
7729 | BUG_ON(ret == -ENOMEM); |
7730 | if (ret) | |
7731 | return ret; | |
2c47e605 YZ |
7732 | BUG_ON(wc->refs[level] == 0); |
7733 | } | |
5d4f98a2 | 7734 | |
2c47e605 YZ |
7735 | if (wc->stage == DROP_REFERENCE) { |
7736 | if (wc->refs[level] > 1) | |
7737 | return 1; | |
f82d02d9 | 7738 | |
2c47e605 | 7739 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 7740 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
7741 | path->locks[level] = 0; |
7742 | } | |
7743 | return 0; | |
7744 | } | |
f82d02d9 | 7745 | |
2c47e605 YZ |
7746 | /* wc->stage == UPDATE_BACKREF */ |
7747 | if (!(wc->flags[level] & flag)) { | |
7748 | BUG_ON(!path->locks[level]); | |
e339a6b0 | 7749 | ret = btrfs_inc_ref(trans, root, eb, 1); |
79787eaa | 7750 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 7751 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 7752 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 | 7753 | ret = btrfs_set_disk_extent_flags(trans, root, eb->start, |
b1c79e09 JB |
7754 | eb->len, flag, |
7755 | btrfs_header_level(eb), 0); | |
79787eaa | 7756 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
7757 | wc->flags[level] |= flag; |
7758 | } | |
7759 | ||
7760 | /* | |
7761 | * the block is shared by multiple trees, so it's not good to | |
7762 | * keep the tree lock | |
7763 | */ | |
7764 | if (path->locks[level] && level > 0) { | |
bd681513 | 7765 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
7766 | path->locks[level] = 0; |
7767 | } | |
7768 | return 0; | |
7769 | } | |
7770 | ||
1c4850e2 | 7771 | /* |
2c016dc2 | 7772 | * helper to process tree block pointer. |
1c4850e2 YZ |
7773 | * |
7774 | * when wc->stage == DROP_REFERENCE, this function checks | |
7775 | * reference count of the block pointed to. if the block | |
7776 | * is shared and we need update back refs for the subtree | |
7777 | * rooted at the block, this function changes wc->stage to | |
7778 | * UPDATE_BACKREF. if the block is shared and there is no | |
7779 | * need to update back, this function drops the reference | |
7780 | * to the block. | |
7781 | * | |
7782 | * NOTE: return value 1 means we should stop walking down. | |
7783 | */ | |
7784 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
7785 | struct btrfs_root *root, | |
7786 | struct btrfs_path *path, | |
94fcca9f | 7787 | struct walk_control *wc, int *lookup_info) |
1c4850e2 YZ |
7788 | { |
7789 | u64 bytenr; | |
7790 | u64 generation; | |
7791 | u64 parent; | |
7792 | u32 blocksize; | |
7793 | struct btrfs_key key; | |
7794 | struct extent_buffer *next; | |
7795 | int level = wc->level; | |
7796 | int reada = 0; | |
7797 | int ret = 0; | |
1152651a | 7798 | bool need_account = false; |
1c4850e2 YZ |
7799 | |
7800 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
7801 | path->slots[level]); | |
7802 | /* | |
7803 | * if the lower level block was created before the snapshot | |
7804 | * was created, we know there is no need to update back refs | |
7805 | * for the subtree | |
7806 | */ | |
7807 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
7808 | generation <= root->root_key.offset) { |
7809 | *lookup_info = 1; | |
1c4850e2 | 7810 | return 1; |
94fcca9f | 7811 | } |
1c4850e2 YZ |
7812 | |
7813 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
7814 | blocksize = btrfs_level_size(root, level - 1); | |
7815 | ||
7816 | next = btrfs_find_tree_block(root, bytenr, blocksize); | |
7817 | if (!next) { | |
7818 | next = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
90d2c51d MX |
7819 | if (!next) |
7820 | return -ENOMEM; | |
b2aaaa3b JB |
7821 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, next, |
7822 | level - 1); | |
1c4850e2 YZ |
7823 | reada = 1; |
7824 | } | |
7825 | btrfs_tree_lock(next); | |
7826 | btrfs_set_lock_blocking(next); | |
7827 | ||
3173a18f | 7828 | ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1, |
94fcca9f YZ |
7829 | &wc->refs[level - 1], |
7830 | &wc->flags[level - 1]); | |
79787eaa JM |
7831 | if (ret < 0) { |
7832 | btrfs_tree_unlock(next); | |
7833 | return ret; | |
7834 | } | |
7835 | ||
c2cf52eb SK |
7836 | if (unlikely(wc->refs[level - 1] == 0)) { |
7837 | btrfs_err(root->fs_info, "Missing references."); | |
7838 | BUG(); | |
7839 | } | |
94fcca9f | 7840 | *lookup_info = 0; |
1c4850e2 | 7841 | |
94fcca9f | 7842 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 7843 | if (wc->refs[level - 1] > 1) { |
1152651a | 7844 | need_account = true; |
94fcca9f YZ |
7845 | if (level == 1 && |
7846 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
7847 | goto skip; | |
7848 | ||
1c4850e2 YZ |
7849 | if (!wc->update_ref || |
7850 | generation <= root->root_key.offset) | |
7851 | goto skip; | |
7852 | ||
7853 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
7854 | path->slots[level]); | |
7855 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
7856 | if (ret < 0) | |
7857 | goto skip; | |
7858 | ||
7859 | wc->stage = UPDATE_BACKREF; | |
7860 | wc->shared_level = level - 1; | |
7861 | } | |
94fcca9f YZ |
7862 | } else { |
7863 | if (level == 1 && | |
7864 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
7865 | goto skip; | |
1c4850e2 YZ |
7866 | } |
7867 | ||
b9fab919 | 7868 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
7869 | btrfs_tree_unlock(next); |
7870 | free_extent_buffer(next); | |
7871 | next = NULL; | |
94fcca9f | 7872 | *lookup_info = 1; |
1c4850e2 YZ |
7873 | } |
7874 | ||
7875 | if (!next) { | |
7876 | if (reada && level == 1) | |
7877 | reada_walk_down(trans, root, wc, path); | |
7878 | next = read_tree_block(root, bytenr, blocksize, generation); | |
416bc658 JB |
7879 | if (!next || !extent_buffer_uptodate(next)) { |
7880 | free_extent_buffer(next); | |
97d9a8a4 | 7881 | return -EIO; |
416bc658 | 7882 | } |
1c4850e2 YZ |
7883 | btrfs_tree_lock(next); |
7884 | btrfs_set_lock_blocking(next); | |
7885 | } | |
7886 | ||
7887 | level--; | |
7888 | BUG_ON(level != btrfs_header_level(next)); | |
7889 | path->nodes[level] = next; | |
7890 | path->slots[level] = 0; | |
bd681513 | 7891 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
7892 | wc->level = level; |
7893 | if (wc->level == 1) | |
7894 | wc->reada_slot = 0; | |
7895 | return 0; | |
7896 | skip: | |
7897 | wc->refs[level - 1] = 0; | |
7898 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
7899 | if (wc->stage == DROP_REFERENCE) { |
7900 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
7901 | parent = path->nodes[level]->start; | |
7902 | } else { | |
7903 | BUG_ON(root->root_key.objectid != | |
7904 | btrfs_header_owner(path->nodes[level])); | |
7905 | parent = 0; | |
7906 | } | |
1c4850e2 | 7907 | |
1152651a MF |
7908 | if (need_account) { |
7909 | ret = account_shared_subtree(trans, root, next, | |
7910 | generation, level - 1); | |
7911 | if (ret) { | |
7912 | printk_ratelimited(KERN_ERR "BTRFS: %s Error " | |
7913 | "%d accounting shared subtree. Quota " | |
7914 | "is out of sync, rescan required.\n", | |
7915 | root->fs_info->sb->s_id, ret); | |
7916 | } | |
7917 | } | |
94fcca9f | 7918 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent, |
66d7e7f0 | 7919 | root->root_key.objectid, level - 1, 0, 0); |
79787eaa | 7920 | BUG_ON(ret); /* -ENOMEM */ |
1c4850e2 | 7921 | } |
1c4850e2 YZ |
7922 | btrfs_tree_unlock(next); |
7923 | free_extent_buffer(next); | |
94fcca9f | 7924 | *lookup_info = 1; |
1c4850e2 YZ |
7925 | return 1; |
7926 | } | |
7927 | ||
2c47e605 | 7928 | /* |
2c016dc2 | 7929 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
7930 | * |
7931 | * when wc->stage == DROP_REFERENCE, this function drops | |
7932 | * reference count on the block. | |
7933 | * | |
7934 | * when wc->stage == UPDATE_BACKREF, this function changes | |
7935 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
7936 | * to UPDATE_BACKREF previously while processing the block. | |
7937 | * | |
7938 | * NOTE: return value 1 means we should stop walking up. | |
7939 | */ | |
7940 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
7941 | struct btrfs_root *root, | |
7942 | struct btrfs_path *path, | |
7943 | struct walk_control *wc) | |
7944 | { | |
f0486c68 | 7945 | int ret; |
2c47e605 YZ |
7946 | int level = wc->level; |
7947 | struct extent_buffer *eb = path->nodes[level]; | |
7948 | u64 parent = 0; | |
7949 | ||
7950 | if (wc->stage == UPDATE_BACKREF) { | |
7951 | BUG_ON(wc->shared_level < level); | |
7952 | if (level < wc->shared_level) | |
7953 | goto out; | |
7954 | ||
2c47e605 YZ |
7955 | ret = find_next_key(path, level + 1, &wc->update_progress); |
7956 | if (ret > 0) | |
7957 | wc->update_ref = 0; | |
7958 | ||
7959 | wc->stage = DROP_REFERENCE; | |
7960 | wc->shared_level = -1; | |
7961 | path->slots[level] = 0; | |
7962 | ||
7963 | /* | |
7964 | * check reference count again if the block isn't locked. | |
7965 | * we should start walking down the tree again if reference | |
7966 | * count is one. | |
7967 | */ | |
7968 | if (!path->locks[level]) { | |
7969 | BUG_ON(level == 0); | |
7970 | btrfs_tree_lock(eb); | |
7971 | btrfs_set_lock_blocking(eb); | |
bd681513 | 7972 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
7973 | |
7974 | ret = btrfs_lookup_extent_info(trans, root, | |
3173a18f | 7975 | eb->start, level, 1, |
2c47e605 YZ |
7976 | &wc->refs[level], |
7977 | &wc->flags[level]); | |
79787eaa JM |
7978 | if (ret < 0) { |
7979 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 7980 | path->locks[level] = 0; |
79787eaa JM |
7981 | return ret; |
7982 | } | |
2c47e605 YZ |
7983 | BUG_ON(wc->refs[level] == 0); |
7984 | if (wc->refs[level] == 1) { | |
bd681513 | 7985 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 7986 | path->locks[level] = 0; |
2c47e605 YZ |
7987 | return 1; |
7988 | } | |
f82d02d9 | 7989 | } |
2c47e605 | 7990 | } |
f82d02d9 | 7991 | |
2c47e605 YZ |
7992 | /* wc->stage == DROP_REFERENCE */ |
7993 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 7994 | |
2c47e605 YZ |
7995 | if (wc->refs[level] == 1) { |
7996 | if (level == 0) { | |
7997 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
e339a6b0 | 7998 | ret = btrfs_dec_ref(trans, root, eb, 1); |
2c47e605 | 7999 | else |
e339a6b0 | 8000 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8001 | BUG_ON(ret); /* -ENOMEM */ |
1152651a MF |
8002 | ret = account_leaf_items(trans, root, eb); |
8003 | if (ret) { | |
8004 | printk_ratelimited(KERN_ERR "BTRFS: %s Error " | |
8005 | "%d accounting leaf items. Quota " | |
8006 | "is out of sync, rescan required.\n", | |
8007 | root->fs_info->sb->s_id, ret); | |
8008 | } | |
2c47e605 YZ |
8009 | } |
8010 | /* make block locked assertion in clean_tree_block happy */ | |
8011 | if (!path->locks[level] && | |
8012 | btrfs_header_generation(eb) == trans->transid) { | |
8013 | btrfs_tree_lock(eb); | |
8014 | btrfs_set_lock_blocking(eb); | |
bd681513 | 8015 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
8016 | } |
8017 | clean_tree_block(trans, root, eb); | |
8018 | } | |
8019 | ||
8020 | if (eb == root->node) { | |
8021 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
8022 | parent = eb->start; | |
8023 | else | |
8024 | BUG_ON(root->root_key.objectid != | |
8025 | btrfs_header_owner(eb)); | |
8026 | } else { | |
8027 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
8028 | parent = path->nodes[level + 1]->start; | |
8029 | else | |
8030 | BUG_ON(root->root_key.objectid != | |
8031 | btrfs_header_owner(path->nodes[level + 1])); | |
f82d02d9 | 8032 | } |
f82d02d9 | 8033 | |
5581a51a | 8034 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
8035 | out: |
8036 | wc->refs[level] = 0; | |
8037 | wc->flags[level] = 0; | |
f0486c68 | 8038 | return 0; |
2c47e605 YZ |
8039 | } |
8040 | ||
8041 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
8042 | struct btrfs_root *root, | |
8043 | struct btrfs_path *path, | |
8044 | struct walk_control *wc) | |
8045 | { | |
2c47e605 | 8046 | int level = wc->level; |
94fcca9f | 8047 | int lookup_info = 1; |
2c47e605 YZ |
8048 | int ret; |
8049 | ||
8050 | while (level >= 0) { | |
94fcca9f | 8051 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
8052 | if (ret > 0) |
8053 | break; | |
8054 | ||
8055 | if (level == 0) | |
8056 | break; | |
8057 | ||
7a7965f8 YZ |
8058 | if (path->slots[level] >= |
8059 | btrfs_header_nritems(path->nodes[level])) | |
8060 | break; | |
8061 | ||
94fcca9f | 8062 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
8063 | if (ret > 0) { |
8064 | path->slots[level]++; | |
8065 | continue; | |
90d2c51d MX |
8066 | } else if (ret < 0) |
8067 | return ret; | |
1c4850e2 | 8068 | level = wc->level; |
f82d02d9 | 8069 | } |
f82d02d9 YZ |
8070 | return 0; |
8071 | } | |
8072 | ||
d397712b | 8073 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 8074 | struct btrfs_root *root, |
f82d02d9 | 8075 | struct btrfs_path *path, |
2c47e605 | 8076 | struct walk_control *wc, int max_level) |
20524f02 | 8077 | { |
2c47e605 | 8078 | int level = wc->level; |
20524f02 | 8079 | int ret; |
9f3a7427 | 8080 | |
2c47e605 YZ |
8081 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
8082 | while (level < max_level && path->nodes[level]) { | |
8083 | wc->level = level; | |
8084 | if (path->slots[level] + 1 < | |
8085 | btrfs_header_nritems(path->nodes[level])) { | |
8086 | path->slots[level]++; | |
20524f02 CM |
8087 | return 0; |
8088 | } else { | |
2c47e605 YZ |
8089 | ret = walk_up_proc(trans, root, path, wc); |
8090 | if (ret > 0) | |
8091 | return 0; | |
bd56b302 | 8092 | |
2c47e605 | 8093 | if (path->locks[level]) { |
bd681513 CM |
8094 | btrfs_tree_unlock_rw(path->nodes[level], |
8095 | path->locks[level]); | |
2c47e605 | 8096 | path->locks[level] = 0; |
f82d02d9 | 8097 | } |
2c47e605 YZ |
8098 | free_extent_buffer(path->nodes[level]); |
8099 | path->nodes[level] = NULL; | |
8100 | level++; | |
20524f02 CM |
8101 | } |
8102 | } | |
8103 | return 1; | |
8104 | } | |
8105 | ||
9aca1d51 | 8106 | /* |
2c47e605 YZ |
8107 | * drop a subvolume tree. |
8108 | * | |
8109 | * this function traverses the tree freeing any blocks that only | |
8110 | * referenced by the tree. | |
8111 | * | |
8112 | * when a shared tree block is found. this function decreases its | |
8113 | * reference count by one. if update_ref is true, this function | |
8114 | * also make sure backrefs for the shared block and all lower level | |
8115 | * blocks are properly updated. | |
9d1a2a3a DS |
8116 | * |
8117 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 8118 | */ |
2c536799 | 8119 | int btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
8120 | struct btrfs_block_rsv *block_rsv, int update_ref, |
8121 | int for_reloc) | |
20524f02 | 8122 | { |
5caf2a00 | 8123 | struct btrfs_path *path; |
2c47e605 YZ |
8124 | struct btrfs_trans_handle *trans; |
8125 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
9f3a7427 | 8126 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
8127 | struct walk_control *wc; |
8128 | struct btrfs_key key; | |
8129 | int err = 0; | |
8130 | int ret; | |
8131 | int level; | |
d29a9f62 | 8132 | bool root_dropped = false; |
20524f02 | 8133 | |
1152651a MF |
8134 | btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid); |
8135 | ||
5caf2a00 | 8136 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
8137 | if (!path) { |
8138 | err = -ENOMEM; | |
8139 | goto out; | |
8140 | } | |
20524f02 | 8141 | |
2c47e605 | 8142 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
8143 | if (!wc) { |
8144 | btrfs_free_path(path); | |
cb1b69f4 TI |
8145 | err = -ENOMEM; |
8146 | goto out; | |
38a1a919 | 8147 | } |
2c47e605 | 8148 | |
a22285a6 | 8149 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
8150 | if (IS_ERR(trans)) { |
8151 | err = PTR_ERR(trans); | |
8152 | goto out_free; | |
8153 | } | |
98d5dc13 | 8154 | |
3fd0a558 YZ |
8155 | if (block_rsv) |
8156 | trans->block_rsv = block_rsv; | |
2c47e605 | 8157 | |
9f3a7427 | 8158 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 8159 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
8160 | path->nodes[level] = btrfs_lock_root_node(root); |
8161 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 8162 | path->slots[level] = 0; |
bd681513 | 8163 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
8164 | memset(&wc->update_progress, 0, |
8165 | sizeof(wc->update_progress)); | |
9f3a7427 | 8166 | } else { |
9f3a7427 | 8167 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
8168 | memcpy(&wc->update_progress, &key, |
8169 | sizeof(wc->update_progress)); | |
8170 | ||
6702ed49 | 8171 | level = root_item->drop_level; |
2c47e605 | 8172 | BUG_ON(level == 0); |
6702ed49 | 8173 | path->lowest_level = level; |
2c47e605 YZ |
8174 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
8175 | path->lowest_level = 0; | |
8176 | if (ret < 0) { | |
8177 | err = ret; | |
79787eaa | 8178 | goto out_end_trans; |
9f3a7427 | 8179 | } |
1c4850e2 | 8180 | WARN_ON(ret > 0); |
2c47e605 | 8181 | |
7d9eb12c CM |
8182 | /* |
8183 | * unlock our path, this is safe because only this | |
8184 | * function is allowed to delete this snapshot | |
8185 | */ | |
5d4f98a2 | 8186 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
8187 | |
8188 | level = btrfs_header_level(root->node); | |
8189 | while (1) { | |
8190 | btrfs_tree_lock(path->nodes[level]); | |
8191 | btrfs_set_lock_blocking(path->nodes[level]); | |
fec386ac | 8192 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
8193 | |
8194 | ret = btrfs_lookup_extent_info(trans, root, | |
8195 | path->nodes[level]->start, | |
3173a18f | 8196 | level, 1, &wc->refs[level], |
2c47e605 | 8197 | &wc->flags[level]); |
79787eaa JM |
8198 | if (ret < 0) { |
8199 | err = ret; | |
8200 | goto out_end_trans; | |
8201 | } | |
2c47e605 YZ |
8202 | BUG_ON(wc->refs[level] == 0); |
8203 | ||
8204 | if (level == root_item->drop_level) | |
8205 | break; | |
8206 | ||
8207 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 8208 | path->locks[level] = 0; |
2c47e605 YZ |
8209 | WARN_ON(wc->refs[level] != 1); |
8210 | level--; | |
8211 | } | |
9f3a7427 | 8212 | } |
2c47e605 YZ |
8213 | |
8214 | wc->level = level; | |
8215 | wc->shared_level = -1; | |
8216 | wc->stage = DROP_REFERENCE; | |
8217 | wc->update_ref = update_ref; | |
8218 | wc->keep_locks = 0; | |
66d7e7f0 | 8219 | wc->for_reloc = for_reloc; |
1c4850e2 | 8220 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
2c47e605 | 8221 | |
d397712b | 8222 | while (1) { |
9d1a2a3a | 8223 | |
2c47e605 YZ |
8224 | ret = walk_down_tree(trans, root, path, wc); |
8225 | if (ret < 0) { | |
8226 | err = ret; | |
20524f02 | 8227 | break; |
2c47e605 | 8228 | } |
9aca1d51 | 8229 | |
2c47e605 YZ |
8230 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
8231 | if (ret < 0) { | |
8232 | err = ret; | |
20524f02 | 8233 | break; |
2c47e605 YZ |
8234 | } |
8235 | ||
8236 | if (ret > 0) { | |
8237 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
8238 | break; |
8239 | } | |
2c47e605 YZ |
8240 | |
8241 | if (wc->stage == DROP_REFERENCE) { | |
8242 | level = wc->level; | |
8243 | btrfs_node_key(path->nodes[level], | |
8244 | &root_item->drop_progress, | |
8245 | path->slots[level]); | |
8246 | root_item->drop_level = level; | |
8247 | } | |
8248 | ||
8249 | BUG_ON(wc->level == 0); | |
3c8f2422 JB |
8250 | if (btrfs_should_end_transaction(trans, tree_root) || |
8251 | (!for_reloc && btrfs_need_cleaner_sleep(root))) { | |
2c47e605 YZ |
8252 | ret = btrfs_update_root(trans, tree_root, |
8253 | &root->root_key, | |
8254 | root_item); | |
79787eaa JM |
8255 | if (ret) { |
8256 | btrfs_abort_transaction(trans, tree_root, ret); | |
8257 | err = ret; | |
8258 | goto out_end_trans; | |
8259 | } | |
2c47e605 | 8260 | |
1152651a MF |
8261 | /* |
8262 | * Qgroup update accounting is run from | |
8263 | * delayed ref handling. This usually works | |
8264 | * out because delayed refs are normally the | |
8265 | * only way qgroup updates are added. However, | |
8266 | * we may have added updates during our tree | |
8267 | * walk so run qgroups here to make sure we | |
8268 | * don't lose any updates. | |
8269 | */ | |
8270 | ret = btrfs_delayed_qgroup_accounting(trans, | |
8271 | root->fs_info); | |
8272 | if (ret) | |
8273 | printk_ratelimited(KERN_ERR "BTRFS: Failure %d " | |
8274 | "running qgroup updates " | |
8275 | "during snapshot delete. " | |
8276 | "Quota is out of sync, " | |
8277 | "rescan required.\n", ret); | |
8278 | ||
3fd0a558 | 8279 | btrfs_end_transaction_throttle(trans, tree_root); |
3c8f2422 | 8280 | if (!for_reloc && btrfs_need_cleaner_sleep(root)) { |
efe120a0 | 8281 | pr_debug("BTRFS: drop snapshot early exit\n"); |
3c8f2422 JB |
8282 | err = -EAGAIN; |
8283 | goto out_free; | |
8284 | } | |
8285 | ||
a22285a6 | 8286 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
8287 | if (IS_ERR(trans)) { |
8288 | err = PTR_ERR(trans); | |
8289 | goto out_free; | |
8290 | } | |
3fd0a558 YZ |
8291 | if (block_rsv) |
8292 | trans->block_rsv = block_rsv; | |
c3e69d58 | 8293 | } |
20524f02 | 8294 | } |
b3b4aa74 | 8295 | btrfs_release_path(path); |
79787eaa JM |
8296 | if (err) |
8297 | goto out_end_trans; | |
2c47e605 YZ |
8298 | |
8299 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
79787eaa JM |
8300 | if (ret) { |
8301 | btrfs_abort_transaction(trans, tree_root, ret); | |
8302 | goto out_end_trans; | |
8303 | } | |
2c47e605 | 8304 | |
76dda93c | 8305 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
cb517eab MX |
8306 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
8307 | NULL, NULL); | |
79787eaa JM |
8308 | if (ret < 0) { |
8309 | btrfs_abort_transaction(trans, tree_root, ret); | |
8310 | err = ret; | |
8311 | goto out_end_trans; | |
8312 | } else if (ret > 0) { | |
84cd948c JB |
8313 | /* if we fail to delete the orphan item this time |
8314 | * around, it'll get picked up the next time. | |
8315 | * | |
8316 | * The most common failure here is just -ENOENT. | |
8317 | */ | |
8318 | btrfs_del_orphan_item(trans, tree_root, | |
8319 | root->root_key.objectid); | |
76dda93c YZ |
8320 | } |
8321 | } | |
8322 | ||
27cdeb70 | 8323 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) { |
cb517eab | 8324 | btrfs_drop_and_free_fs_root(tree_root->fs_info, root); |
76dda93c YZ |
8325 | } else { |
8326 | free_extent_buffer(root->node); | |
8327 | free_extent_buffer(root->commit_root); | |
b0feb9d9 | 8328 | btrfs_put_fs_root(root); |
76dda93c | 8329 | } |
d29a9f62 | 8330 | root_dropped = true; |
79787eaa | 8331 | out_end_trans: |
1152651a MF |
8332 | ret = btrfs_delayed_qgroup_accounting(trans, tree_root->fs_info); |
8333 | if (ret) | |
8334 | printk_ratelimited(KERN_ERR "BTRFS: Failure %d " | |
8335 | "running qgroup updates " | |
8336 | "during snapshot delete. " | |
8337 | "Quota is out of sync, " | |
8338 | "rescan required.\n", ret); | |
8339 | ||
3fd0a558 | 8340 | btrfs_end_transaction_throttle(trans, tree_root); |
79787eaa | 8341 | out_free: |
2c47e605 | 8342 | kfree(wc); |
5caf2a00 | 8343 | btrfs_free_path(path); |
cb1b69f4 | 8344 | out: |
d29a9f62 JB |
8345 | /* |
8346 | * So if we need to stop dropping the snapshot for whatever reason we | |
8347 | * need to make sure to add it back to the dead root list so that we | |
8348 | * keep trying to do the work later. This also cleans up roots if we | |
8349 | * don't have it in the radix (like when we recover after a power fail | |
8350 | * or unmount) so we don't leak memory. | |
8351 | */ | |
b37b39cd | 8352 | if (!for_reloc && root_dropped == false) |
d29a9f62 | 8353 | btrfs_add_dead_root(root); |
90515e7f | 8354 | if (err && err != -EAGAIN) |
cb1b69f4 | 8355 | btrfs_std_error(root->fs_info, err); |
2c536799 | 8356 | return err; |
20524f02 | 8357 | } |
9078a3e1 | 8358 | |
2c47e605 YZ |
8359 | /* |
8360 | * drop subtree rooted at tree block 'node'. | |
8361 | * | |
8362 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 8363 | * only used by relocation code |
2c47e605 | 8364 | */ |
f82d02d9 YZ |
8365 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
8366 | struct btrfs_root *root, | |
8367 | struct extent_buffer *node, | |
8368 | struct extent_buffer *parent) | |
8369 | { | |
8370 | struct btrfs_path *path; | |
2c47e605 | 8371 | struct walk_control *wc; |
f82d02d9 YZ |
8372 | int level; |
8373 | int parent_level; | |
8374 | int ret = 0; | |
8375 | int wret; | |
8376 | ||
2c47e605 YZ |
8377 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
8378 | ||
f82d02d9 | 8379 | path = btrfs_alloc_path(); |
db5b493a TI |
8380 | if (!path) |
8381 | return -ENOMEM; | |
f82d02d9 | 8382 | |
2c47e605 | 8383 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
8384 | if (!wc) { |
8385 | btrfs_free_path(path); | |
8386 | return -ENOMEM; | |
8387 | } | |
2c47e605 | 8388 | |
b9447ef8 | 8389 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
8390 | parent_level = btrfs_header_level(parent); |
8391 | extent_buffer_get(parent); | |
8392 | path->nodes[parent_level] = parent; | |
8393 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
8394 | ||
b9447ef8 | 8395 | btrfs_assert_tree_locked(node); |
f82d02d9 | 8396 | level = btrfs_header_level(node); |
f82d02d9 YZ |
8397 | path->nodes[level] = node; |
8398 | path->slots[level] = 0; | |
bd681513 | 8399 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
8400 | |
8401 | wc->refs[parent_level] = 1; | |
8402 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
8403 | wc->level = level; | |
8404 | wc->shared_level = -1; | |
8405 | wc->stage = DROP_REFERENCE; | |
8406 | wc->update_ref = 0; | |
8407 | wc->keep_locks = 1; | |
66d7e7f0 | 8408 | wc->for_reloc = 1; |
1c4850e2 | 8409 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
f82d02d9 YZ |
8410 | |
8411 | while (1) { | |
2c47e605 YZ |
8412 | wret = walk_down_tree(trans, root, path, wc); |
8413 | if (wret < 0) { | |
f82d02d9 | 8414 | ret = wret; |
f82d02d9 | 8415 | break; |
2c47e605 | 8416 | } |
f82d02d9 | 8417 | |
2c47e605 | 8418 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
8419 | if (wret < 0) |
8420 | ret = wret; | |
8421 | if (wret != 0) | |
8422 | break; | |
8423 | } | |
8424 | ||
2c47e605 | 8425 | kfree(wc); |
f82d02d9 YZ |
8426 | btrfs_free_path(path); |
8427 | return ret; | |
8428 | } | |
8429 | ||
ec44a35c CM |
8430 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) |
8431 | { | |
8432 | u64 num_devices; | |
fc67c450 | 8433 | u64 stripped; |
e4d8ec0f | 8434 | |
fc67c450 ID |
8435 | /* |
8436 | * if restripe for this chunk_type is on pick target profile and | |
8437 | * return, otherwise do the usual balance | |
8438 | */ | |
8439 | stripped = get_restripe_target(root->fs_info, flags); | |
8440 | if (stripped) | |
8441 | return extended_to_chunk(stripped); | |
e4d8ec0f | 8442 | |
cd02dca5 CM |
8443 | /* |
8444 | * we add in the count of missing devices because we want | |
8445 | * to make sure that any RAID levels on a degraded FS | |
8446 | * continue to be honored. | |
8447 | */ | |
8448 | num_devices = root->fs_info->fs_devices->rw_devices + | |
8449 | root->fs_info->fs_devices->missing_devices; | |
8450 | ||
fc67c450 | 8451 | stripped = BTRFS_BLOCK_GROUP_RAID0 | |
53b381b3 | 8452 | BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 | |
fc67c450 ID |
8453 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; |
8454 | ||
ec44a35c CM |
8455 | if (num_devices == 1) { |
8456 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
8457 | stripped = flags & ~stripped; | |
8458 | ||
8459 | /* turn raid0 into single device chunks */ | |
8460 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
8461 | return stripped; | |
8462 | ||
8463 | /* turn mirroring into duplication */ | |
8464 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
8465 | BTRFS_BLOCK_GROUP_RAID10)) | |
8466 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
ec44a35c CM |
8467 | } else { |
8468 | /* they already had raid on here, just return */ | |
ec44a35c CM |
8469 | if (flags & stripped) |
8470 | return flags; | |
8471 | ||
8472 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
8473 | stripped = flags & ~stripped; | |
8474 | ||
8475 | /* switch duplicated blocks with raid1 */ | |
8476 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
8477 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
8478 | ||
e3176ca2 | 8479 | /* this is drive concat, leave it alone */ |
ec44a35c | 8480 | } |
e3176ca2 | 8481 | |
ec44a35c CM |
8482 | return flags; |
8483 | } | |
8484 | ||
199c36ea | 8485 | static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 8486 | { |
f0486c68 YZ |
8487 | struct btrfs_space_info *sinfo = cache->space_info; |
8488 | u64 num_bytes; | |
199c36ea | 8489 | u64 min_allocable_bytes; |
f0486c68 | 8490 | int ret = -ENOSPC; |
0ef3e66b | 8491 | |
c286ac48 | 8492 | |
199c36ea MX |
8493 | /* |
8494 | * We need some metadata space and system metadata space for | |
8495 | * allocating chunks in some corner cases until we force to set | |
8496 | * it to be readonly. | |
8497 | */ | |
8498 | if ((sinfo->flags & | |
8499 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
8500 | !force) | |
8501 | min_allocable_bytes = 1 * 1024 * 1024; | |
8502 | else | |
8503 | min_allocable_bytes = 0; | |
8504 | ||
f0486c68 YZ |
8505 | spin_lock(&sinfo->lock); |
8506 | spin_lock(&cache->lock); | |
61cfea9b W |
8507 | |
8508 | if (cache->ro) { | |
8509 | ret = 0; | |
8510 | goto out; | |
8511 | } | |
8512 | ||
f0486c68 YZ |
8513 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
8514 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
8515 | ||
8516 | if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned + | |
37be25bc JB |
8517 | sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes + |
8518 | min_allocable_bytes <= sinfo->total_bytes) { | |
f0486c68 | 8519 | sinfo->bytes_readonly += num_bytes; |
f0486c68 YZ |
8520 | cache->ro = 1; |
8521 | ret = 0; | |
8522 | } | |
61cfea9b | 8523 | out: |
f0486c68 YZ |
8524 | spin_unlock(&cache->lock); |
8525 | spin_unlock(&sinfo->lock); | |
8526 | return ret; | |
8527 | } | |
7d9eb12c | 8528 | |
f0486c68 YZ |
8529 | int btrfs_set_block_group_ro(struct btrfs_root *root, |
8530 | struct btrfs_block_group_cache *cache) | |
c286ac48 | 8531 | |
f0486c68 YZ |
8532 | { |
8533 | struct btrfs_trans_handle *trans; | |
8534 | u64 alloc_flags; | |
8535 | int ret; | |
7d9eb12c | 8536 | |
f0486c68 | 8537 | BUG_ON(cache->ro); |
0ef3e66b | 8538 | |
ff5714cc | 8539 | trans = btrfs_join_transaction(root); |
79787eaa JM |
8540 | if (IS_ERR(trans)) |
8541 | return PTR_ERR(trans); | |
5d4f98a2 | 8542 | |
f0486c68 | 8543 | alloc_flags = update_block_group_flags(root, cache->flags); |
79787eaa | 8544 | if (alloc_flags != cache->flags) { |
698d0082 | 8545 | ret = do_chunk_alloc(trans, root, alloc_flags, |
79787eaa JM |
8546 | CHUNK_ALLOC_FORCE); |
8547 | if (ret < 0) | |
8548 | goto out; | |
8549 | } | |
5d4f98a2 | 8550 | |
199c36ea | 8551 | ret = set_block_group_ro(cache, 0); |
f0486c68 YZ |
8552 | if (!ret) |
8553 | goto out; | |
8554 | alloc_flags = get_alloc_profile(root, cache->space_info->flags); | |
698d0082 | 8555 | ret = do_chunk_alloc(trans, root, alloc_flags, |
0e4f8f88 | 8556 | CHUNK_ALLOC_FORCE); |
f0486c68 YZ |
8557 | if (ret < 0) |
8558 | goto out; | |
199c36ea | 8559 | ret = set_block_group_ro(cache, 0); |
f0486c68 YZ |
8560 | out: |
8561 | btrfs_end_transaction(trans, root); | |
8562 | return ret; | |
8563 | } | |
5d4f98a2 | 8564 | |
c87f08ca CM |
8565 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, |
8566 | struct btrfs_root *root, u64 type) | |
8567 | { | |
8568 | u64 alloc_flags = get_alloc_profile(root, type); | |
698d0082 | 8569 | return do_chunk_alloc(trans, root, alloc_flags, |
0e4f8f88 | 8570 | CHUNK_ALLOC_FORCE); |
c87f08ca CM |
8571 | } |
8572 | ||
6d07bcec MX |
8573 | /* |
8574 | * helper to account the unused space of all the readonly block group in the | |
8575 | * list. takes mirrors into account. | |
8576 | */ | |
8577 | static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list) | |
8578 | { | |
8579 | struct btrfs_block_group_cache *block_group; | |
8580 | u64 free_bytes = 0; | |
8581 | int factor; | |
8582 | ||
8583 | list_for_each_entry(block_group, groups_list, list) { | |
8584 | spin_lock(&block_group->lock); | |
8585 | ||
8586 | if (!block_group->ro) { | |
8587 | spin_unlock(&block_group->lock); | |
8588 | continue; | |
8589 | } | |
8590 | ||
8591 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
8592 | BTRFS_BLOCK_GROUP_RAID10 | | |
8593 | BTRFS_BLOCK_GROUP_DUP)) | |
8594 | factor = 2; | |
8595 | else | |
8596 | factor = 1; | |
8597 | ||
8598 | free_bytes += (block_group->key.offset - | |
8599 | btrfs_block_group_used(&block_group->item)) * | |
8600 | factor; | |
8601 | ||
8602 | spin_unlock(&block_group->lock); | |
8603 | } | |
8604 | ||
8605 | return free_bytes; | |
8606 | } | |
8607 | ||
8608 | /* | |
8609 | * helper to account the unused space of all the readonly block group in the | |
8610 | * space_info. takes mirrors into account. | |
8611 | */ | |
8612 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) | |
8613 | { | |
8614 | int i; | |
8615 | u64 free_bytes = 0; | |
8616 | ||
8617 | spin_lock(&sinfo->lock); | |
8618 | ||
67871254 | 8619 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) |
6d07bcec MX |
8620 | if (!list_empty(&sinfo->block_groups[i])) |
8621 | free_bytes += __btrfs_get_ro_block_group_free_space( | |
8622 | &sinfo->block_groups[i]); | |
8623 | ||
8624 | spin_unlock(&sinfo->lock); | |
8625 | ||
8626 | return free_bytes; | |
8627 | } | |
8628 | ||
143bede5 | 8629 | void btrfs_set_block_group_rw(struct btrfs_root *root, |
f0486c68 | 8630 | struct btrfs_block_group_cache *cache) |
5d4f98a2 | 8631 | { |
f0486c68 YZ |
8632 | struct btrfs_space_info *sinfo = cache->space_info; |
8633 | u64 num_bytes; | |
8634 | ||
8635 | BUG_ON(!cache->ro); | |
8636 | ||
8637 | spin_lock(&sinfo->lock); | |
8638 | spin_lock(&cache->lock); | |
8639 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - | |
8640 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
8641 | sinfo->bytes_readonly -= num_bytes; | |
8642 | cache->ro = 0; | |
8643 | spin_unlock(&cache->lock); | |
8644 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
8645 | } |
8646 | ||
ba1bf481 JB |
8647 | /* |
8648 | * checks to see if its even possible to relocate this block group. | |
8649 | * | |
8650 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
8651 | * ok to go ahead and try. | |
8652 | */ | |
8653 | int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr) | |
1a40e23b | 8654 | { |
ba1bf481 JB |
8655 | struct btrfs_block_group_cache *block_group; |
8656 | struct btrfs_space_info *space_info; | |
8657 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
8658 | struct btrfs_device *device; | |
6df9a95e | 8659 | struct btrfs_trans_handle *trans; |
cdcb725c | 8660 | u64 min_free; |
6719db6a JB |
8661 | u64 dev_min = 1; |
8662 | u64 dev_nr = 0; | |
4a5e98f5 | 8663 | u64 target; |
cdcb725c | 8664 | int index; |
ba1bf481 JB |
8665 | int full = 0; |
8666 | int ret = 0; | |
1a40e23b | 8667 | |
ba1bf481 | 8668 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); |
1a40e23b | 8669 | |
ba1bf481 JB |
8670 | /* odd, couldn't find the block group, leave it alone */ |
8671 | if (!block_group) | |
8672 | return -1; | |
1a40e23b | 8673 | |
cdcb725c | 8674 | min_free = btrfs_block_group_used(&block_group->item); |
8675 | ||
ba1bf481 | 8676 | /* no bytes used, we're good */ |
cdcb725c | 8677 | if (!min_free) |
1a40e23b ZY |
8678 | goto out; |
8679 | ||
ba1bf481 JB |
8680 | space_info = block_group->space_info; |
8681 | spin_lock(&space_info->lock); | |
17d217fe | 8682 | |
ba1bf481 | 8683 | full = space_info->full; |
17d217fe | 8684 | |
ba1bf481 JB |
8685 | /* |
8686 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
8687 | * relocate it unless we're able to allocate a new chunk below. |
8688 | * | |
8689 | * Otherwise, we need to make sure we have room in the space to handle | |
8690 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 8691 | */ |
7ce618db | 8692 | if ((space_info->total_bytes != block_group->key.offset) && |
cdcb725c | 8693 | (space_info->bytes_used + space_info->bytes_reserved + |
8694 | space_info->bytes_pinned + space_info->bytes_readonly + | |
8695 | min_free < space_info->total_bytes)) { | |
ba1bf481 JB |
8696 | spin_unlock(&space_info->lock); |
8697 | goto out; | |
17d217fe | 8698 | } |
ba1bf481 | 8699 | spin_unlock(&space_info->lock); |
ea8c2819 | 8700 | |
ba1bf481 JB |
8701 | /* |
8702 | * ok we don't have enough space, but maybe we have free space on our | |
8703 | * devices to allocate new chunks for relocation, so loop through our | |
4a5e98f5 ID |
8704 | * alloc devices and guess if we have enough space. if this block |
8705 | * group is going to be restriped, run checks against the target | |
8706 | * profile instead of the current one. | |
ba1bf481 JB |
8707 | */ |
8708 | ret = -1; | |
ea8c2819 | 8709 | |
cdcb725c | 8710 | /* |
8711 | * index: | |
8712 | * 0: raid10 | |
8713 | * 1: raid1 | |
8714 | * 2: dup | |
8715 | * 3: raid0 | |
8716 | * 4: single | |
8717 | */ | |
4a5e98f5 ID |
8718 | target = get_restripe_target(root->fs_info, block_group->flags); |
8719 | if (target) { | |
31e50229 | 8720 | index = __get_raid_index(extended_to_chunk(target)); |
4a5e98f5 ID |
8721 | } else { |
8722 | /* | |
8723 | * this is just a balance, so if we were marked as full | |
8724 | * we know there is no space for a new chunk | |
8725 | */ | |
8726 | if (full) | |
8727 | goto out; | |
8728 | ||
8729 | index = get_block_group_index(block_group); | |
8730 | } | |
8731 | ||
e6ec716f | 8732 | if (index == BTRFS_RAID_RAID10) { |
cdcb725c | 8733 | dev_min = 4; |
6719db6a JB |
8734 | /* Divide by 2 */ |
8735 | min_free >>= 1; | |
e6ec716f | 8736 | } else if (index == BTRFS_RAID_RAID1) { |
cdcb725c | 8737 | dev_min = 2; |
e6ec716f | 8738 | } else if (index == BTRFS_RAID_DUP) { |
6719db6a JB |
8739 | /* Multiply by 2 */ |
8740 | min_free <<= 1; | |
e6ec716f | 8741 | } else if (index == BTRFS_RAID_RAID0) { |
cdcb725c | 8742 | dev_min = fs_devices->rw_devices; |
6719db6a | 8743 | do_div(min_free, dev_min); |
cdcb725c | 8744 | } |
8745 | ||
6df9a95e JB |
8746 | /* We need to do this so that we can look at pending chunks */ |
8747 | trans = btrfs_join_transaction(root); | |
8748 | if (IS_ERR(trans)) { | |
8749 | ret = PTR_ERR(trans); | |
8750 | goto out; | |
8751 | } | |
8752 | ||
ba1bf481 JB |
8753 | mutex_lock(&root->fs_info->chunk_mutex); |
8754 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { | |
7bfc837d | 8755 | u64 dev_offset; |
56bec294 | 8756 | |
ba1bf481 JB |
8757 | /* |
8758 | * check to make sure we can actually find a chunk with enough | |
8759 | * space to fit our block group in. | |
8760 | */ | |
63a212ab SB |
8761 | if (device->total_bytes > device->bytes_used + min_free && |
8762 | !device->is_tgtdev_for_dev_replace) { | |
6df9a95e | 8763 | ret = find_free_dev_extent(trans, device, min_free, |
7bfc837d | 8764 | &dev_offset, NULL); |
ba1bf481 | 8765 | if (!ret) |
cdcb725c | 8766 | dev_nr++; |
8767 | ||
8768 | if (dev_nr >= dev_min) | |
73e48b27 | 8769 | break; |
cdcb725c | 8770 | |
ba1bf481 | 8771 | ret = -1; |
725c8463 | 8772 | } |
edbd8d4e | 8773 | } |
ba1bf481 | 8774 | mutex_unlock(&root->fs_info->chunk_mutex); |
6df9a95e | 8775 | btrfs_end_transaction(trans, root); |
edbd8d4e | 8776 | out: |
ba1bf481 | 8777 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
8778 | return ret; |
8779 | } | |
8780 | ||
b2950863 CH |
8781 | static int find_first_block_group(struct btrfs_root *root, |
8782 | struct btrfs_path *path, struct btrfs_key *key) | |
0b86a832 | 8783 | { |
925baedd | 8784 | int ret = 0; |
0b86a832 CM |
8785 | struct btrfs_key found_key; |
8786 | struct extent_buffer *leaf; | |
8787 | int slot; | |
edbd8d4e | 8788 | |
0b86a832 CM |
8789 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
8790 | if (ret < 0) | |
925baedd CM |
8791 | goto out; |
8792 | ||
d397712b | 8793 | while (1) { |
0b86a832 | 8794 | slot = path->slots[0]; |
edbd8d4e | 8795 | leaf = path->nodes[0]; |
0b86a832 CM |
8796 | if (slot >= btrfs_header_nritems(leaf)) { |
8797 | ret = btrfs_next_leaf(root, path); | |
8798 | if (ret == 0) | |
8799 | continue; | |
8800 | if (ret < 0) | |
925baedd | 8801 | goto out; |
0b86a832 | 8802 | break; |
edbd8d4e | 8803 | } |
0b86a832 | 8804 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 8805 | |
0b86a832 | 8806 | if (found_key.objectid >= key->objectid && |
925baedd CM |
8807 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
8808 | ret = 0; | |
8809 | goto out; | |
8810 | } | |
0b86a832 | 8811 | path->slots[0]++; |
edbd8d4e | 8812 | } |
925baedd | 8813 | out: |
0b86a832 | 8814 | return ret; |
edbd8d4e CM |
8815 | } |
8816 | ||
0af3d00b JB |
8817 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
8818 | { | |
8819 | struct btrfs_block_group_cache *block_group; | |
8820 | u64 last = 0; | |
8821 | ||
8822 | while (1) { | |
8823 | struct inode *inode; | |
8824 | ||
8825 | block_group = btrfs_lookup_first_block_group(info, last); | |
8826 | while (block_group) { | |
8827 | spin_lock(&block_group->lock); | |
8828 | if (block_group->iref) | |
8829 | break; | |
8830 | spin_unlock(&block_group->lock); | |
8831 | block_group = next_block_group(info->tree_root, | |
8832 | block_group); | |
8833 | } | |
8834 | if (!block_group) { | |
8835 | if (last == 0) | |
8836 | break; | |
8837 | last = 0; | |
8838 | continue; | |
8839 | } | |
8840 | ||
8841 | inode = block_group->inode; | |
8842 | block_group->iref = 0; | |
8843 | block_group->inode = NULL; | |
8844 | spin_unlock(&block_group->lock); | |
8845 | iput(inode); | |
8846 | last = block_group->key.objectid + block_group->key.offset; | |
8847 | btrfs_put_block_group(block_group); | |
8848 | } | |
8849 | } | |
8850 | ||
1a40e23b ZY |
8851 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
8852 | { | |
8853 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 8854 | struct btrfs_space_info *space_info; |
11833d66 | 8855 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
8856 | struct rb_node *n; |
8857 | ||
9e351cc8 | 8858 | down_write(&info->commit_root_sem); |
11833d66 YZ |
8859 | while (!list_empty(&info->caching_block_groups)) { |
8860 | caching_ctl = list_entry(info->caching_block_groups.next, | |
8861 | struct btrfs_caching_control, list); | |
8862 | list_del(&caching_ctl->list); | |
8863 | put_caching_control(caching_ctl); | |
8864 | } | |
9e351cc8 | 8865 | up_write(&info->commit_root_sem); |
11833d66 | 8866 | |
1a40e23b ZY |
8867 | spin_lock(&info->block_group_cache_lock); |
8868 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
8869 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
8870 | cache_node); | |
1a40e23b ZY |
8871 | rb_erase(&block_group->cache_node, |
8872 | &info->block_group_cache_tree); | |
d899e052 YZ |
8873 | spin_unlock(&info->block_group_cache_lock); |
8874 | ||
80eb234a | 8875 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 8876 | list_del(&block_group->list); |
80eb234a | 8877 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 8878 | |
817d52f8 | 8879 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 8880 | wait_block_group_cache_done(block_group); |
817d52f8 | 8881 | |
3c14874a JB |
8882 | /* |
8883 | * We haven't cached this block group, which means we could | |
8884 | * possibly have excluded extents on this block group. | |
8885 | */ | |
36cce922 JB |
8886 | if (block_group->cached == BTRFS_CACHE_NO || |
8887 | block_group->cached == BTRFS_CACHE_ERROR) | |
3c14874a JB |
8888 | free_excluded_extents(info->extent_root, block_group); |
8889 | ||
817d52f8 | 8890 | btrfs_remove_free_space_cache(block_group); |
11dfe35a | 8891 | btrfs_put_block_group(block_group); |
d899e052 YZ |
8892 | |
8893 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
8894 | } |
8895 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
8896 | |
8897 | /* now that all the block groups are freed, go through and | |
8898 | * free all the space_info structs. This is only called during | |
8899 | * the final stages of unmount, and so we know nobody is | |
8900 | * using them. We call synchronize_rcu() once before we start, | |
8901 | * just to be on the safe side. | |
8902 | */ | |
8903 | synchronize_rcu(); | |
8904 | ||
8929ecfa YZ |
8905 | release_global_block_rsv(info); |
8906 | ||
67871254 | 8907 | while (!list_empty(&info->space_info)) { |
6ab0a202 JM |
8908 | int i; |
8909 | ||
4184ea7f CM |
8910 | space_info = list_entry(info->space_info.next, |
8911 | struct btrfs_space_info, | |
8912 | list); | |
b069e0c3 | 8913 | if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) { |
fae7f21c | 8914 | if (WARN_ON(space_info->bytes_pinned > 0 || |
b069e0c3 | 8915 | space_info->bytes_reserved > 0 || |
fae7f21c | 8916 | space_info->bytes_may_use > 0)) { |
b069e0c3 DS |
8917 | dump_space_info(space_info, 0, 0); |
8918 | } | |
f0486c68 | 8919 | } |
4184ea7f | 8920 | list_del(&space_info->list); |
6ab0a202 JM |
8921 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
8922 | struct kobject *kobj; | |
c1895442 JM |
8923 | kobj = space_info->block_group_kobjs[i]; |
8924 | space_info->block_group_kobjs[i] = NULL; | |
8925 | if (kobj) { | |
6ab0a202 JM |
8926 | kobject_del(kobj); |
8927 | kobject_put(kobj); | |
8928 | } | |
8929 | } | |
8930 | kobject_del(&space_info->kobj); | |
8931 | kobject_put(&space_info->kobj); | |
4184ea7f | 8932 | } |
1a40e23b ZY |
8933 | return 0; |
8934 | } | |
8935 | ||
b742bb82 YZ |
8936 | static void __link_block_group(struct btrfs_space_info *space_info, |
8937 | struct btrfs_block_group_cache *cache) | |
8938 | { | |
8939 | int index = get_block_group_index(cache); | |
ed55b6ac | 8940 | bool first = false; |
b742bb82 YZ |
8941 | |
8942 | down_write(&space_info->groups_sem); | |
ed55b6ac JM |
8943 | if (list_empty(&space_info->block_groups[index])) |
8944 | first = true; | |
8945 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
8946 | up_write(&space_info->groups_sem); | |
8947 | ||
8948 | if (first) { | |
c1895442 | 8949 | struct raid_kobject *rkobj; |
6ab0a202 JM |
8950 | int ret; |
8951 | ||
c1895442 JM |
8952 | rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS); |
8953 | if (!rkobj) | |
8954 | goto out_err; | |
8955 | rkobj->raid_type = index; | |
8956 | kobject_init(&rkobj->kobj, &btrfs_raid_ktype); | |
8957 | ret = kobject_add(&rkobj->kobj, &space_info->kobj, | |
8958 | "%s", get_raid_name(index)); | |
6ab0a202 | 8959 | if (ret) { |
c1895442 JM |
8960 | kobject_put(&rkobj->kobj); |
8961 | goto out_err; | |
6ab0a202 | 8962 | } |
c1895442 | 8963 | space_info->block_group_kobjs[index] = &rkobj->kobj; |
6ab0a202 | 8964 | } |
c1895442 JM |
8965 | |
8966 | return; | |
8967 | out_err: | |
8968 | pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n"); | |
b742bb82 YZ |
8969 | } |
8970 | ||
920e4a58 MX |
8971 | static struct btrfs_block_group_cache * |
8972 | btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size) | |
8973 | { | |
8974 | struct btrfs_block_group_cache *cache; | |
8975 | ||
8976 | cache = kzalloc(sizeof(*cache), GFP_NOFS); | |
8977 | if (!cache) | |
8978 | return NULL; | |
8979 | ||
8980 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), | |
8981 | GFP_NOFS); | |
8982 | if (!cache->free_space_ctl) { | |
8983 | kfree(cache); | |
8984 | return NULL; | |
8985 | } | |
8986 | ||
8987 | cache->key.objectid = start; | |
8988 | cache->key.offset = size; | |
8989 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
8990 | ||
8991 | cache->sectorsize = root->sectorsize; | |
8992 | cache->fs_info = root->fs_info; | |
8993 | cache->full_stripe_len = btrfs_full_stripe_len(root, | |
8994 | &root->fs_info->mapping_tree, | |
8995 | start); | |
8996 | atomic_set(&cache->count, 1); | |
8997 | spin_lock_init(&cache->lock); | |
e570fd27 | 8998 | init_rwsem(&cache->data_rwsem); |
920e4a58 MX |
8999 | INIT_LIST_HEAD(&cache->list); |
9000 | INIT_LIST_HEAD(&cache->cluster_list); | |
9001 | INIT_LIST_HEAD(&cache->new_bg_list); | |
9002 | btrfs_init_free_space_ctl(cache); | |
9003 | ||
9004 | return cache; | |
9005 | } | |
9006 | ||
9078a3e1 CM |
9007 | int btrfs_read_block_groups(struct btrfs_root *root) |
9008 | { | |
9009 | struct btrfs_path *path; | |
9010 | int ret; | |
9078a3e1 | 9011 | struct btrfs_block_group_cache *cache; |
be744175 | 9012 | struct btrfs_fs_info *info = root->fs_info; |
6324fbf3 | 9013 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
9014 | struct btrfs_key key; |
9015 | struct btrfs_key found_key; | |
5f39d397 | 9016 | struct extent_buffer *leaf; |
0af3d00b JB |
9017 | int need_clear = 0; |
9018 | u64 cache_gen; | |
96b5179d | 9019 | |
be744175 | 9020 | root = info->extent_root; |
9078a3e1 | 9021 | key.objectid = 0; |
0b86a832 | 9022 | key.offset = 0; |
9078a3e1 | 9023 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
9078a3e1 CM |
9024 | path = btrfs_alloc_path(); |
9025 | if (!path) | |
9026 | return -ENOMEM; | |
026fd317 | 9027 | path->reada = 1; |
9078a3e1 | 9028 | |
6c41761f | 9029 | cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy); |
73bc1876 | 9030 | if (btrfs_test_opt(root, SPACE_CACHE) && |
6c41761f | 9031 | btrfs_super_generation(root->fs_info->super_copy) != cache_gen) |
0af3d00b | 9032 | need_clear = 1; |
88c2ba3b JB |
9033 | if (btrfs_test_opt(root, CLEAR_CACHE)) |
9034 | need_clear = 1; | |
0af3d00b | 9035 | |
d397712b | 9036 | while (1) { |
0b86a832 | 9037 | ret = find_first_block_group(root, path, &key); |
b742bb82 YZ |
9038 | if (ret > 0) |
9039 | break; | |
0b86a832 CM |
9040 | if (ret != 0) |
9041 | goto error; | |
920e4a58 | 9042 | |
5f39d397 CM |
9043 | leaf = path->nodes[0]; |
9044 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
920e4a58 MX |
9045 | |
9046 | cache = btrfs_create_block_group_cache(root, found_key.objectid, | |
9047 | found_key.offset); | |
9078a3e1 | 9048 | if (!cache) { |
0b86a832 | 9049 | ret = -ENOMEM; |
f0486c68 | 9050 | goto error; |
9078a3e1 | 9051 | } |
96303081 | 9052 | |
cf7c1ef6 LB |
9053 | if (need_clear) { |
9054 | /* | |
9055 | * When we mount with old space cache, we need to | |
9056 | * set BTRFS_DC_CLEAR and set dirty flag. | |
9057 | * | |
9058 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
9059 | * truncate the old free space cache inode and | |
9060 | * setup a new one. | |
9061 | * b) Setting 'dirty flag' makes sure that we flush | |
9062 | * the new space cache info onto disk. | |
9063 | */ | |
0af3d00b | 9064 | cache->disk_cache_state = BTRFS_DC_CLEAR; |
cf7c1ef6 LB |
9065 | if (btrfs_test_opt(root, SPACE_CACHE)) |
9066 | cache->dirty = 1; | |
9067 | } | |
0af3d00b | 9068 | |
5f39d397 CM |
9069 | read_extent_buffer(leaf, &cache->item, |
9070 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
9071 | sizeof(cache->item)); | |
920e4a58 | 9072 | cache->flags = btrfs_block_group_flags(&cache->item); |
0b86a832 | 9073 | |
9078a3e1 | 9074 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 9075 | btrfs_release_path(path); |
34d52cb6 | 9076 | |
3c14874a JB |
9077 | /* |
9078 | * We need to exclude the super stripes now so that the space | |
9079 | * info has super bytes accounted for, otherwise we'll think | |
9080 | * we have more space than we actually do. | |
9081 | */ | |
835d974f JB |
9082 | ret = exclude_super_stripes(root, cache); |
9083 | if (ret) { | |
9084 | /* | |
9085 | * We may have excluded something, so call this just in | |
9086 | * case. | |
9087 | */ | |
9088 | free_excluded_extents(root, cache); | |
920e4a58 | 9089 | btrfs_put_block_group(cache); |
835d974f JB |
9090 | goto error; |
9091 | } | |
3c14874a | 9092 | |
817d52f8 JB |
9093 | /* |
9094 | * check for two cases, either we are full, and therefore | |
9095 | * don't need to bother with the caching work since we won't | |
9096 | * find any space, or we are empty, and we can just add all | |
9097 | * the space in and be done with it. This saves us _alot_ of | |
9098 | * time, particularly in the full case. | |
9099 | */ | |
9100 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 9101 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 9102 | cache->cached = BTRFS_CACHE_FINISHED; |
1b2da372 | 9103 | free_excluded_extents(root, cache); |
817d52f8 | 9104 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 9105 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 JB |
9106 | cache->cached = BTRFS_CACHE_FINISHED; |
9107 | add_new_free_space(cache, root->fs_info, | |
9108 | found_key.objectid, | |
9109 | found_key.objectid + | |
9110 | found_key.offset); | |
11833d66 | 9111 | free_excluded_extents(root, cache); |
817d52f8 | 9112 | } |
96b5179d | 9113 | |
8c579fe7 JB |
9114 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
9115 | if (ret) { | |
9116 | btrfs_remove_free_space_cache(cache); | |
9117 | btrfs_put_block_group(cache); | |
9118 | goto error; | |
9119 | } | |
9120 | ||
6324fbf3 CM |
9121 | ret = update_space_info(info, cache->flags, found_key.offset, |
9122 | btrfs_block_group_used(&cache->item), | |
9123 | &space_info); | |
8c579fe7 JB |
9124 | if (ret) { |
9125 | btrfs_remove_free_space_cache(cache); | |
9126 | spin_lock(&info->block_group_cache_lock); | |
9127 | rb_erase(&cache->cache_node, | |
9128 | &info->block_group_cache_tree); | |
9129 | spin_unlock(&info->block_group_cache_lock); | |
9130 | btrfs_put_block_group(cache); | |
9131 | goto error; | |
9132 | } | |
9133 | ||
6324fbf3 | 9134 | cache->space_info = space_info; |
1b2da372 | 9135 | spin_lock(&cache->space_info->lock); |
f0486c68 | 9136 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
9137 | spin_unlock(&cache->space_info->lock); |
9138 | ||
b742bb82 | 9139 | __link_block_group(space_info, cache); |
0f9dd46c | 9140 | |
75ccf47d | 9141 | set_avail_alloc_bits(root->fs_info, cache->flags); |
2b82032c | 9142 | if (btrfs_chunk_readonly(root, cache->key.objectid)) |
199c36ea | 9143 | set_block_group_ro(cache, 1); |
9078a3e1 | 9144 | } |
b742bb82 YZ |
9145 | |
9146 | list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) { | |
9147 | if (!(get_alloc_profile(root, space_info->flags) & | |
9148 | (BTRFS_BLOCK_GROUP_RAID10 | | |
9149 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
9150 | BTRFS_BLOCK_GROUP_RAID5 | |
9151 | BTRFS_BLOCK_GROUP_RAID6 | | |
b742bb82 YZ |
9152 | BTRFS_BLOCK_GROUP_DUP))) |
9153 | continue; | |
9154 | /* | |
9155 | * avoid allocating from un-mirrored block group if there are | |
9156 | * mirrored block groups. | |
9157 | */ | |
1095cc0d | 9158 | list_for_each_entry(cache, |
9159 | &space_info->block_groups[BTRFS_RAID_RAID0], | |
9160 | list) | |
199c36ea | 9161 | set_block_group_ro(cache, 1); |
1095cc0d | 9162 | list_for_each_entry(cache, |
9163 | &space_info->block_groups[BTRFS_RAID_SINGLE], | |
9164 | list) | |
199c36ea | 9165 | set_block_group_ro(cache, 1); |
9078a3e1 | 9166 | } |
f0486c68 YZ |
9167 | |
9168 | init_global_block_rsv(info); | |
0b86a832 CM |
9169 | ret = 0; |
9170 | error: | |
9078a3e1 | 9171 | btrfs_free_path(path); |
0b86a832 | 9172 | return ret; |
9078a3e1 | 9173 | } |
6324fbf3 | 9174 | |
ea658bad JB |
9175 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans, |
9176 | struct btrfs_root *root) | |
9177 | { | |
9178 | struct btrfs_block_group_cache *block_group, *tmp; | |
9179 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
9180 | struct btrfs_block_group_item item; | |
9181 | struct btrfs_key key; | |
9182 | int ret = 0; | |
9183 | ||
9184 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, | |
9185 | new_bg_list) { | |
9186 | list_del_init(&block_group->new_bg_list); | |
9187 | ||
9188 | if (ret) | |
9189 | continue; | |
9190 | ||
9191 | spin_lock(&block_group->lock); | |
9192 | memcpy(&item, &block_group->item, sizeof(item)); | |
9193 | memcpy(&key, &block_group->key, sizeof(key)); | |
9194 | spin_unlock(&block_group->lock); | |
9195 | ||
9196 | ret = btrfs_insert_item(trans, extent_root, &key, &item, | |
9197 | sizeof(item)); | |
9198 | if (ret) | |
9199 | btrfs_abort_transaction(trans, extent_root, ret); | |
6df9a95e JB |
9200 | ret = btrfs_finish_chunk_alloc(trans, extent_root, |
9201 | key.objectid, key.offset); | |
9202 | if (ret) | |
9203 | btrfs_abort_transaction(trans, extent_root, ret); | |
ea658bad JB |
9204 | } |
9205 | } | |
9206 | ||
6324fbf3 CM |
9207 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, |
9208 | struct btrfs_root *root, u64 bytes_used, | |
e17cade2 | 9209 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
9210 | u64 size) |
9211 | { | |
9212 | int ret; | |
6324fbf3 CM |
9213 | struct btrfs_root *extent_root; |
9214 | struct btrfs_block_group_cache *cache; | |
6324fbf3 CM |
9215 | |
9216 | extent_root = root->fs_info->extent_root; | |
6324fbf3 | 9217 | |
995946dd | 9218 | btrfs_set_log_full_commit(root->fs_info, trans); |
e02119d5 | 9219 | |
920e4a58 | 9220 | cache = btrfs_create_block_group_cache(root, chunk_offset, size); |
0f9dd46c JB |
9221 | if (!cache) |
9222 | return -ENOMEM; | |
34d52cb6 | 9223 | |
6324fbf3 | 9224 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 | 9225 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
6324fbf3 CM |
9226 | btrfs_set_block_group_flags(&cache->item, type); |
9227 | ||
920e4a58 | 9228 | cache->flags = type; |
11833d66 | 9229 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 9230 | cache->cached = BTRFS_CACHE_FINISHED; |
835d974f JB |
9231 | ret = exclude_super_stripes(root, cache); |
9232 | if (ret) { | |
9233 | /* | |
9234 | * We may have excluded something, so call this just in | |
9235 | * case. | |
9236 | */ | |
9237 | free_excluded_extents(root, cache); | |
920e4a58 | 9238 | btrfs_put_block_group(cache); |
835d974f JB |
9239 | return ret; |
9240 | } | |
96303081 | 9241 | |
817d52f8 JB |
9242 | add_new_free_space(cache, root->fs_info, chunk_offset, |
9243 | chunk_offset + size); | |
9244 | ||
11833d66 YZ |
9245 | free_excluded_extents(root, cache); |
9246 | ||
8c579fe7 JB |
9247 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
9248 | if (ret) { | |
9249 | btrfs_remove_free_space_cache(cache); | |
9250 | btrfs_put_block_group(cache); | |
9251 | return ret; | |
9252 | } | |
9253 | ||
6324fbf3 CM |
9254 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, |
9255 | &cache->space_info); | |
8c579fe7 JB |
9256 | if (ret) { |
9257 | btrfs_remove_free_space_cache(cache); | |
9258 | spin_lock(&root->fs_info->block_group_cache_lock); | |
9259 | rb_erase(&cache->cache_node, | |
9260 | &root->fs_info->block_group_cache_tree); | |
9261 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
9262 | btrfs_put_block_group(cache); | |
9263 | return ret; | |
9264 | } | |
c7c144db | 9265 | update_global_block_rsv(root->fs_info); |
1b2da372 JB |
9266 | |
9267 | spin_lock(&cache->space_info->lock); | |
f0486c68 | 9268 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
9269 | spin_unlock(&cache->space_info->lock); |
9270 | ||
b742bb82 | 9271 | __link_block_group(cache->space_info, cache); |
6324fbf3 | 9272 | |
ea658bad | 9273 | list_add_tail(&cache->new_bg_list, &trans->new_bgs); |
6324fbf3 | 9274 | |
d18a2c44 | 9275 | set_avail_alloc_bits(extent_root->fs_info, type); |
925baedd | 9276 | |
6324fbf3 CM |
9277 | return 0; |
9278 | } | |
1a40e23b | 9279 | |
10ea00f5 ID |
9280 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
9281 | { | |
899c81ea ID |
9282 | u64 extra_flags = chunk_to_extended(flags) & |
9283 | BTRFS_EXTENDED_PROFILE_MASK; | |
10ea00f5 | 9284 | |
de98ced9 | 9285 | write_seqlock(&fs_info->profiles_lock); |
10ea00f5 ID |
9286 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
9287 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
9288 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
9289 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
9290 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
9291 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
de98ced9 | 9292 | write_sequnlock(&fs_info->profiles_lock); |
10ea00f5 ID |
9293 | } |
9294 | ||
1a40e23b ZY |
9295 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
9296 | struct btrfs_root *root, u64 group_start) | |
9297 | { | |
9298 | struct btrfs_path *path; | |
9299 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 9300 | struct btrfs_free_cluster *cluster; |
0af3d00b | 9301 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
1a40e23b | 9302 | struct btrfs_key key; |
0af3d00b | 9303 | struct inode *inode; |
c1895442 | 9304 | struct kobject *kobj = NULL; |
1a40e23b | 9305 | int ret; |
10ea00f5 | 9306 | int index; |
89a55897 | 9307 | int factor; |
1a40e23b | 9308 | |
1a40e23b ZY |
9309 | root = root->fs_info->extent_root; |
9310 | ||
9311 | block_group = btrfs_lookup_block_group(root->fs_info, group_start); | |
9312 | BUG_ON(!block_group); | |
c146afad | 9313 | BUG_ON(!block_group->ro); |
1a40e23b | 9314 | |
9f7c43c9 | 9315 | /* |
9316 | * Free the reserved super bytes from this block group before | |
9317 | * remove it. | |
9318 | */ | |
9319 | free_excluded_extents(root, block_group); | |
9320 | ||
1a40e23b | 9321 | memcpy(&key, &block_group->key, sizeof(key)); |
10ea00f5 | 9322 | index = get_block_group_index(block_group); |
89a55897 JB |
9323 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | |
9324 | BTRFS_BLOCK_GROUP_RAID1 | | |
9325 | BTRFS_BLOCK_GROUP_RAID10)) | |
9326 | factor = 2; | |
9327 | else | |
9328 | factor = 1; | |
1a40e23b | 9329 | |
44fb5511 CM |
9330 | /* make sure this block group isn't part of an allocation cluster */ |
9331 | cluster = &root->fs_info->data_alloc_cluster; | |
9332 | spin_lock(&cluster->refill_lock); | |
9333 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
9334 | spin_unlock(&cluster->refill_lock); | |
9335 | ||
9336 | /* | |
9337 | * make sure this block group isn't part of a metadata | |
9338 | * allocation cluster | |
9339 | */ | |
9340 | cluster = &root->fs_info->meta_alloc_cluster; | |
9341 | spin_lock(&cluster->refill_lock); | |
9342 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
9343 | spin_unlock(&cluster->refill_lock); | |
9344 | ||
1a40e23b | 9345 | path = btrfs_alloc_path(); |
d8926bb3 MF |
9346 | if (!path) { |
9347 | ret = -ENOMEM; | |
9348 | goto out; | |
9349 | } | |
1a40e23b | 9350 | |
10b2f34d | 9351 | inode = lookup_free_space_inode(tree_root, block_group, path); |
0af3d00b | 9352 | if (!IS_ERR(inode)) { |
b532402e | 9353 | ret = btrfs_orphan_add(trans, inode); |
79787eaa JM |
9354 | if (ret) { |
9355 | btrfs_add_delayed_iput(inode); | |
9356 | goto out; | |
9357 | } | |
0af3d00b JB |
9358 | clear_nlink(inode); |
9359 | /* One for the block groups ref */ | |
9360 | spin_lock(&block_group->lock); | |
9361 | if (block_group->iref) { | |
9362 | block_group->iref = 0; | |
9363 | block_group->inode = NULL; | |
9364 | spin_unlock(&block_group->lock); | |
9365 | iput(inode); | |
9366 | } else { | |
9367 | spin_unlock(&block_group->lock); | |
9368 | } | |
9369 | /* One for our lookup ref */ | |
455757c3 | 9370 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
9371 | } |
9372 | ||
9373 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
9374 | key.offset = block_group->key.objectid; | |
9375 | key.type = 0; | |
9376 | ||
9377 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
9378 | if (ret < 0) | |
9379 | goto out; | |
9380 | if (ret > 0) | |
b3b4aa74 | 9381 | btrfs_release_path(path); |
0af3d00b JB |
9382 | if (ret == 0) { |
9383 | ret = btrfs_del_item(trans, tree_root, path); | |
9384 | if (ret) | |
9385 | goto out; | |
b3b4aa74 | 9386 | btrfs_release_path(path); |
0af3d00b JB |
9387 | } |
9388 | ||
3dfdb934 | 9389 | spin_lock(&root->fs_info->block_group_cache_lock); |
1a40e23b ZY |
9390 | rb_erase(&block_group->cache_node, |
9391 | &root->fs_info->block_group_cache_tree); | |
a1897fdd LB |
9392 | |
9393 | if (root->fs_info->first_logical_byte == block_group->key.objectid) | |
9394 | root->fs_info->first_logical_byte = (u64)-1; | |
3dfdb934 | 9395 | spin_unlock(&root->fs_info->block_group_cache_lock); |
817d52f8 | 9396 | |
80eb234a | 9397 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
9398 | /* |
9399 | * we must use list_del_init so people can check to see if they | |
9400 | * are still on the list after taking the semaphore | |
9401 | */ | |
9402 | list_del_init(&block_group->list); | |
6ab0a202 | 9403 | if (list_empty(&block_group->space_info->block_groups[index])) { |
c1895442 JM |
9404 | kobj = block_group->space_info->block_group_kobjs[index]; |
9405 | block_group->space_info->block_group_kobjs[index] = NULL; | |
10ea00f5 | 9406 | clear_avail_alloc_bits(root->fs_info, block_group->flags); |
6ab0a202 | 9407 | } |
80eb234a | 9408 | up_write(&block_group->space_info->groups_sem); |
c1895442 JM |
9409 | if (kobj) { |
9410 | kobject_del(kobj); | |
9411 | kobject_put(kobj); | |
9412 | } | |
1a40e23b | 9413 | |
817d52f8 | 9414 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 9415 | wait_block_group_cache_done(block_group); |
817d52f8 JB |
9416 | |
9417 | btrfs_remove_free_space_cache(block_group); | |
9418 | ||
c146afad YZ |
9419 | spin_lock(&block_group->space_info->lock); |
9420 | block_group->space_info->total_bytes -= block_group->key.offset; | |
9421 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 9422 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
c146afad | 9423 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 9424 | |
0af3d00b JB |
9425 | memcpy(&key, &block_group->key, sizeof(key)); |
9426 | ||
283bb197 | 9427 | btrfs_clear_space_info_full(root->fs_info); |
c146afad | 9428 | |
fa9c0d79 CM |
9429 | btrfs_put_block_group(block_group); |
9430 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
9431 | |
9432 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
9433 | if (ret > 0) | |
9434 | ret = -EIO; | |
9435 | if (ret < 0) | |
9436 | goto out; | |
9437 | ||
9438 | ret = btrfs_del_item(trans, root, path); | |
9439 | out: | |
9440 | btrfs_free_path(path); | |
9441 | return ret; | |
9442 | } | |
acce952b | 9443 | |
c59021f8 | 9444 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
9445 | { | |
9446 | struct btrfs_space_info *space_info; | |
1aba86d6 | 9447 | struct btrfs_super_block *disk_super; |
9448 | u64 features; | |
9449 | u64 flags; | |
9450 | int mixed = 0; | |
c59021f8 | 9451 | int ret; |
9452 | ||
6c41761f | 9453 | disk_super = fs_info->super_copy; |
1aba86d6 | 9454 | if (!btrfs_super_root(disk_super)) |
9455 | return 1; | |
c59021f8 | 9456 | |
1aba86d6 | 9457 | features = btrfs_super_incompat_flags(disk_super); |
9458 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
9459 | mixed = 1; | |
c59021f8 | 9460 | |
1aba86d6 | 9461 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
9462 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
c59021f8 | 9463 | if (ret) |
1aba86d6 | 9464 | goto out; |
c59021f8 | 9465 | |
1aba86d6 | 9466 | if (mixed) { |
9467 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
9468 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
9469 | } else { | |
9470 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
9471 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
9472 | if (ret) | |
9473 | goto out; | |
9474 | ||
9475 | flags = BTRFS_BLOCK_GROUP_DATA; | |
9476 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
9477 | } | |
9478 | out: | |
c59021f8 | 9479 | return ret; |
9480 | } | |
9481 | ||
acce952b | 9482 | int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
9483 | { | |
9484 | return unpin_extent_range(root, start, end); | |
9485 | } | |
9486 | ||
9487 | int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr, | |
5378e607 | 9488 | u64 num_bytes, u64 *actual_bytes) |
acce952b | 9489 | { |
5378e607 | 9490 | return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes); |
acce952b | 9491 | } |
f7039b1d LD |
9492 | |
9493 | int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range) | |
9494 | { | |
9495 | struct btrfs_fs_info *fs_info = root->fs_info; | |
9496 | struct btrfs_block_group_cache *cache = NULL; | |
9497 | u64 group_trimmed; | |
9498 | u64 start; | |
9499 | u64 end; | |
9500 | u64 trimmed = 0; | |
2cac13e4 | 9501 | u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
f7039b1d LD |
9502 | int ret = 0; |
9503 | ||
2cac13e4 LB |
9504 | /* |
9505 | * try to trim all FS space, our block group may start from non-zero. | |
9506 | */ | |
9507 | if (range->len == total_bytes) | |
9508 | cache = btrfs_lookup_first_block_group(fs_info, range->start); | |
9509 | else | |
9510 | cache = btrfs_lookup_block_group(fs_info, range->start); | |
f7039b1d LD |
9511 | |
9512 | while (cache) { | |
9513 | if (cache->key.objectid >= (range->start + range->len)) { | |
9514 | btrfs_put_block_group(cache); | |
9515 | break; | |
9516 | } | |
9517 | ||
9518 | start = max(range->start, cache->key.objectid); | |
9519 | end = min(range->start + range->len, | |
9520 | cache->key.objectid + cache->key.offset); | |
9521 | ||
9522 | if (end - start >= range->minlen) { | |
9523 | if (!block_group_cache_done(cache)) { | |
f6373bf3 | 9524 | ret = cache_block_group(cache, 0); |
1be41b78 JB |
9525 | if (ret) { |
9526 | btrfs_put_block_group(cache); | |
9527 | break; | |
9528 | } | |
9529 | ret = wait_block_group_cache_done(cache); | |
9530 | if (ret) { | |
9531 | btrfs_put_block_group(cache); | |
9532 | break; | |
9533 | } | |
f7039b1d LD |
9534 | } |
9535 | ret = btrfs_trim_block_group(cache, | |
9536 | &group_trimmed, | |
9537 | start, | |
9538 | end, | |
9539 | range->minlen); | |
9540 | ||
9541 | trimmed += group_trimmed; | |
9542 | if (ret) { | |
9543 | btrfs_put_block_group(cache); | |
9544 | break; | |
9545 | } | |
9546 | } | |
9547 | ||
9548 | cache = next_block_group(fs_info->tree_root, cache); | |
9549 | } | |
9550 | ||
9551 | range->len = trimmed; | |
9552 | return ret; | |
9553 | } | |
8257b2dc MX |
9554 | |
9555 | /* | |
9556 | * btrfs_{start,end}_write() is similar to mnt_{want, drop}_write(), | |
9557 | * they are used to prevent the some tasks writing data into the page cache | |
9558 | * by nocow before the subvolume is snapshoted, but flush the data into | |
9559 | * the disk after the snapshot creation. | |
9560 | */ | |
9561 | void btrfs_end_nocow_write(struct btrfs_root *root) | |
9562 | { | |
9563 | percpu_counter_dec(&root->subv_writers->counter); | |
9564 | /* | |
9565 | * Make sure counter is updated before we wake up | |
9566 | * waiters. | |
9567 | */ | |
9568 | smp_mb(); | |
9569 | if (waitqueue_active(&root->subv_writers->wait)) | |
9570 | wake_up(&root->subv_writers->wait); | |
9571 | } | |
9572 | ||
9573 | int btrfs_start_nocow_write(struct btrfs_root *root) | |
9574 | { | |
9575 | if (unlikely(atomic_read(&root->will_be_snapshoted))) | |
9576 | return 0; | |
9577 | ||
9578 | percpu_counter_inc(&root->subv_writers->counter); | |
9579 | /* | |
9580 | * Make sure counter is updated before we check for snapshot creation. | |
9581 | */ | |
9582 | smp_mb(); | |
9583 | if (unlikely(atomic_read(&root->will_be_snapshoted))) { | |
9584 | btrfs_end_nocow_write(root); | |
9585 | return 0; | |
9586 | } | |
9587 | return 1; | |
9588 | } |